xref: /kernel/linux/linux-5.10/drivers/scsi/hpsa.c (revision 8c2ecf20)
1/*
2 *    Disk Array driver for HP Smart Array SAS controllers
3 *    Copyright (c) 2019-2020 Microchip Technology Inc. and its subsidiaries
4 *    Copyright 2016 Microsemi Corporation
5 *    Copyright 2014-2015 PMC-Sierra, Inc.
6 *    Copyright 2000,2009-2015 Hewlett-Packard Development Company, L.P.
7 *
8 *    This program is free software; you can redistribute it and/or modify
9 *    it under the terms of the GNU General Public License as published by
10 *    the Free Software Foundation; version 2 of the License.
11 *
12 *    This program is distributed in the hope that it will be useful,
13 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
14 *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
15 *    NON INFRINGEMENT.  See the GNU General Public License for more details.
16 *
17 *    Questions/Comments/Bugfixes to esc.storagedev@microsemi.com
18 *
19 */
20
21#include <linux/module.h>
22#include <linux/interrupt.h>
23#include <linux/types.h>
24#include <linux/pci.h>
25#include <linux/kernel.h>
26#include <linux/slab.h>
27#include <linux/delay.h>
28#include <linux/fs.h>
29#include <linux/timer.h>
30#include <linux/init.h>
31#include <linux/spinlock.h>
32#include <linux/compat.h>
33#include <linux/blktrace_api.h>
34#include <linux/uaccess.h>
35#include <linux/io.h>
36#include <linux/dma-mapping.h>
37#include <linux/completion.h>
38#include <linux/moduleparam.h>
39#include <scsi/scsi.h>
40#include <scsi/scsi_cmnd.h>
41#include <scsi/scsi_device.h>
42#include <scsi/scsi_host.h>
43#include <scsi/scsi_tcq.h>
44#include <scsi/scsi_eh.h>
45#include <scsi/scsi_transport_sas.h>
46#include <scsi/scsi_dbg.h>
47#include <linux/cciss_ioctl.h>
48#include <linux/string.h>
49#include <linux/bitmap.h>
50#include <linux/atomic.h>
51#include <linux/jiffies.h>
52#include <linux/percpu-defs.h>
53#include <linux/percpu.h>
54#include <asm/unaligned.h>
55#include <asm/div64.h>
56#include "hpsa_cmd.h"
57#include "hpsa.h"
58
59/*
60 * HPSA_DRIVER_VERSION must be 3 byte values (0-255) separated by '.'
61 * with an optional trailing '-' followed by a byte value (0-255).
62 */
63#define HPSA_DRIVER_VERSION "3.4.20-200"
64#define DRIVER_NAME "HP HPSA Driver (v " HPSA_DRIVER_VERSION ")"
65#define HPSA "hpsa"
66
67/* How long to wait for CISS doorbell communication */
68#define CLEAR_EVENT_WAIT_INTERVAL 20	/* ms for each msleep() call */
69#define MODE_CHANGE_WAIT_INTERVAL 10	/* ms for each msleep() call */
70#define MAX_CLEAR_EVENT_WAIT 30000	/* times 20 ms = 600 s */
71#define MAX_MODE_CHANGE_WAIT 2000	/* times 10 ms = 20 s */
72#define MAX_IOCTL_CONFIG_WAIT 1000
73
74/*define how many times we will try a command because of bus resets */
75#define MAX_CMD_RETRIES 3
76/* How long to wait before giving up on a command */
77#define HPSA_EH_PTRAID_TIMEOUT (240 * HZ)
78
79/* Embedded module documentation macros - see modules.h */
80MODULE_AUTHOR("Hewlett-Packard Company");
81MODULE_DESCRIPTION("Driver for HP Smart Array Controller version " \
82	HPSA_DRIVER_VERSION);
83MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
84MODULE_VERSION(HPSA_DRIVER_VERSION);
85MODULE_LICENSE("GPL");
86MODULE_ALIAS("cciss");
87
88static int hpsa_simple_mode;
89module_param(hpsa_simple_mode, int, S_IRUGO|S_IWUSR);
90MODULE_PARM_DESC(hpsa_simple_mode,
91	"Use 'simple mode' rather than 'performant mode'");
92
93/* define the PCI info for the cards we can control */
94static const struct pci_device_id hpsa_pci_device_id[] = {
95	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
96	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
97	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
98	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
99	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
100	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
101	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
102	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3233},
103	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3350},
104	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3351},
105	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3352},
106	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3353},
107	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3354},
108	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3355},
109	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3356},
110	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103c, 0x1920},
111	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103C, 0x1921},
112	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103C, 0x1922},
113	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103C, 0x1923},
114	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103C, 0x1924},
115	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103c, 0x1925},
116	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103C, 0x1926},
117	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103C, 0x1928},
118	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSH,     0x103C, 0x1929},
119	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21BD},
120	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21BE},
121	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21BF},
122	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C0},
123	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C1},
124	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C2},
125	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C3},
126	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C4},
127	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C5},
128	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C6},
129	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C7},
130	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C8},
131	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21C9},
132	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21CA},
133	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21CB},
134	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21CC},
135	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21CD},
136	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSI,     0x103C, 0x21CE},
137	{PCI_VENDOR_ID_ADAPTEC2, 0x0290, 0x9005, 0x0580},
138	{PCI_VENDOR_ID_ADAPTEC2, 0x0290, 0x9005, 0x0581},
139	{PCI_VENDOR_ID_ADAPTEC2, 0x0290, 0x9005, 0x0582},
140	{PCI_VENDOR_ID_ADAPTEC2, 0x0290, 0x9005, 0x0583},
141	{PCI_VENDOR_ID_ADAPTEC2, 0x0290, 0x9005, 0x0584},
142	{PCI_VENDOR_ID_ADAPTEC2, 0x0290, 0x9005, 0x0585},
143	{PCI_VENDOR_ID_HP_3PAR, 0x0075, 0x1590, 0x0076},
144	{PCI_VENDOR_ID_HP_3PAR, 0x0075, 0x1590, 0x0087},
145	{PCI_VENDOR_ID_HP_3PAR, 0x0075, 0x1590, 0x007D},
146	{PCI_VENDOR_ID_HP_3PAR, 0x0075, 0x1590, 0x0088},
147	{PCI_VENDOR_ID_HP, 0x333f, 0x103c, 0x333f},
148	{PCI_VENDOR_ID_HP,     PCI_ANY_ID,	PCI_ANY_ID, PCI_ANY_ID,
149		PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
150	{PCI_VENDOR_ID_COMPAQ,     PCI_ANY_ID,	PCI_ANY_ID, PCI_ANY_ID,
151		PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
152	{0,}
153};
154
155MODULE_DEVICE_TABLE(pci, hpsa_pci_device_id);
156
157/*  board_id = Subsystem Device ID & Vendor ID
158 *  product = Marketing Name for the board
159 *  access = Address of the struct of function pointers
160 */
161static struct board_type products[] = {
162	{0x40700E11, "Smart Array 5300", &SA5A_access},
163	{0x40800E11, "Smart Array 5i", &SA5B_access},
164	{0x40820E11, "Smart Array 532", &SA5B_access},
165	{0x40830E11, "Smart Array 5312", &SA5B_access},
166	{0x409A0E11, "Smart Array 641", &SA5A_access},
167	{0x409B0E11, "Smart Array 642", &SA5A_access},
168	{0x409C0E11, "Smart Array 6400", &SA5A_access},
169	{0x409D0E11, "Smart Array 6400 EM", &SA5A_access},
170	{0x40910E11, "Smart Array 6i", &SA5A_access},
171	{0x3225103C, "Smart Array P600", &SA5A_access},
172	{0x3223103C, "Smart Array P800", &SA5A_access},
173	{0x3234103C, "Smart Array P400", &SA5A_access},
174	{0x3235103C, "Smart Array P400i", &SA5A_access},
175	{0x3211103C, "Smart Array E200i", &SA5A_access},
176	{0x3212103C, "Smart Array E200", &SA5A_access},
177	{0x3213103C, "Smart Array E200i", &SA5A_access},
178	{0x3214103C, "Smart Array E200i", &SA5A_access},
179	{0x3215103C, "Smart Array E200i", &SA5A_access},
180	{0x3237103C, "Smart Array E500", &SA5A_access},
181	{0x323D103C, "Smart Array P700m", &SA5A_access},
182	{0x3241103C, "Smart Array P212", &SA5_access},
183	{0x3243103C, "Smart Array P410", &SA5_access},
184	{0x3245103C, "Smart Array P410i", &SA5_access},
185	{0x3247103C, "Smart Array P411", &SA5_access},
186	{0x3249103C, "Smart Array P812", &SA5_access},
187	{0x324A103C, "Smart Array P712m", &SA5_access},
188	{0x324B103C, "Smart Array P711m", &SA5_access},
189	{0x3233103C, "HP StorageWorks 1210m", &SA5_access}, /* alias of 333f */
190	{0x3350103C, "Smart Array P222", &SA5_access},
191	{0x3351103C, "Smart Array P420", &SA5_access},
192	{0x3352103C, "Smart Array P421", &SA5_access},
193	{0x3353103C, "Smart Array P822", &SA5_access},
194	{0x3354103C, "Smart Array P420i", &SA5_access},
195	{0x3355103C, "Smart Array P220i", &SA5_access},
196	{0x3356103C, "Smart Array P721m", &SA5_access},
197	{0x1920103C, "Smart Array P430i", &SA5_access},
198	{0x1921103C, "Smart Array P830i", &SA5_access},
199	{0x1922103C, "Smart Array P430", &SA5_access},
200	{0x1923103C, "Smart Array P431", &SA5_access},
201	{0x1924103C, "Smart Array P830", &SA5_access},
202	{0x1925103C, "Smart Array P831", &SA5_access},
203	{0x1926103C, "Smart Array P731m", &SA5_access},
204	{0x1928103C, "Smart Array P230i", &SA5_access},
205	{0x1929103C, "Smart Array P530", &SA5_access},
206	{0x21BD103C, "Smart Array P244br", &SA5_access},
207	{0x21BE103C, "Smart Array P741m", &SA5_access},
208	{0x21BF103C, "Smart HBA H240ar", &SA5_access},
209	{0x21C0103C, "Smart Array P440ar", &SA5_access},
210	{0x21C1103C, "Smart Array P840ar", &SA5_access},
211	{0x21C2103C, "Smart Array P440", &SA5_access},
212	{0x21C3103C, "Smart Array P441", &SA5_access},
213	{0x21C4103C, "Smart Array", &SA5_access},
214	{0x21C5103C, "Smart Array P841", &SA5_access},
215	{0x21C6103C, "Smart HBA H244br", &SA5_access},
216	{0x21C7103C, "Smart HBA H240", &SA5_access},
217	{0x21C8103C, "Smart HBA H241", &SA5_access},
218	{0x21C9103C, "Smart Array", &SA5_access},
219	{0x21CA103C, "Smart Array P246br", &SA5_access},
220	{0x21CB103C, "Smart Array P840", &SA5_access},
221	{0x21CC103C, "Smart Array", &SA5_access},
222	{0x21CD103C, "Smart Array", &SA5_access},
223	{0x21CE103C, "Smart HBA", &SA5_access},
224	{0x05809005, "SmartHBA-SA", &SA5_access},
225	{0x05819005, "SmartHBA-SA 8i", &SA5_access},
226	{0x05829005, "SmartHBA-SA 8i8e", &SA5_access},
227	{0x05839005, "SmartHBA-SA 8e", &SA5_access},
228	{0x05849005, "SmartHBA-SA 16i", &SA5_access},
229	{0x05859005, "SmartHBA-SA 4i4e", &SA5_access},
230	{0x00761590, "HP Storage P1224 Array Controller", &SA5_access},
231	{0x00871590, "HP Storage P1224e Array Controller", &SA5_access},
232	{0x007D1590, "HP Storage P1228 Array Controller", &SA5_access},
233	{0x00881590, "HP Storage P1228e Array Controller", &SA5_access},
234	{0x333f103c, "HP StorageWorks 1210m Array Controller", &SA5_access},
235	{0xFFFF103C, "Unknown Smart Array", &SA5_access},
236};
237
238static struct scsi_transport_template *hpsa_sas_transport_template;
239static int hpsa_add_sas_host(struct ctlr_info *h);
240static void hpsa_delete_sas_host(struct ctlr_info *h);
241static int hpsa_add_sas_device(struct hpsa_sas_node *hpsa_sas_node,
242			struct hpsa_scsi_dev_t *device);
243static void hpsa_remove_sas_device(struct hpsa_scsi_dev_t *device);
244static struct hpsa_scsi_dev_t
245	*hpsa_find_device_by_sas_rphy(struct ctlr_info *h,
246		struct sas_rphy *rphy);
247
248#define SCSI_CMD_BUSY ((struct scsi_cmnd *)&hpsa_cmd_busy)
249static const struct scsi_cmnd hpsa_cmd_busy;
250#define SCSI_CMD_IDLE ((struct scsi_cmnd *)&hpsa_cmd_idle)
251static const struct scsi_cmnd hpsa_cmd_idle;
252static int number_of_controllers;
253
254static irqreturn_t do_hpsa_intr_intx(int irq, void *dev_id);
255static irqreturn_t do_hpsa_intr_msi(int irq, void *dev_id);
256static int hpsa_ioctl(struct scsi_device *dev, unsigned int cmd,
257		      void __user *arg);
258static int hpsa_passthru_ioctl(struct ctlr_info *h,
259			       IOCTL_Command_struct *iocommand);
260static int hpsa_big_passthru_ioctl(struct ctlr_info *h,
261				   BIG_IOCTL_Command_struct *ioc);
262
263#ifdef CONFIG_COMPAT
264static int hpsa_compat_ioctl(struct scsi_device *dev, unsigned int cmd,
265	void __user *arg);
266#endif
267
268static void cmd_free(struct ctlr_info *h, struct CommandList *c);
269static struct CommandList *cmd_alloc(struct ctlr_info *h);
270static void cmd_tagged_free(struct ctlr_info *h, struct CommandList *c);
271static struct CommandList *cmd_tagged_alloc(struct ctlr_info *h,
272					    struct scsi_cmnd *scmd);
273static int fill_cmd(struct CommandList *c, u8 cmd, struct ctlr_info *h,
274	void *buff, size_t size, u16 page_code, unsigned char *scsi3addr,
275	int cmd_type);
276static void hpsa_free_cmd_pool(struct ctlr_info *h);
277#define VPD_PAGE (1 << 8)
278#define HPSA_SIMPLE_ERROR_BITS 0x03
279
280static int hpsa_scsi_queue_command(struct Scsi_Host *h, struct scsi_cmnd *cmd);
281static void hpsa_scan_start(struct Scsi_Host *);
282static int hpsa_scan_finished(struct Scsi_Host *sh,
283	unsigned long elapsed_time);
284static int hpsa_change_queue_depth(struct scsi_device *sdev, int qdepth);
285
286static int hpsa_eh_device_reset_handler(struct scsi_cmnd *scsicmd);
287static int hpsa_slave_alloc(struct scsi_device *sdev);
288static int hpsa_slave_configure(struct scsi_device *sdev);
289static void hpsa_slave_destroy(struct scsi_device *sdev);
290
291static void hpsa_update_scsi_devices(struct ctlr_info *h);
292static int check_for_unit_attention(struct ctlr_info *h,
293	struct CommandList *c);
294static void check_ioctl_unit_attention(struct ctlr_info *h,
295	struct CommandList *c);
296/* performant mode helper functions */
297static void calc_bucket_map(int *bucket, int num_buckets,
298	int nsgs, int min_blocks, u32 *bucket_map);
299static void hpsa_free_performant_mode(struct ctlr_info *h);
300static int hpsa_put_ctlr_into_performant_mode(struct ctlr_info *h);
301static inline u32 next_command(struct ctlr_info *h, u8 q);
302static int hpsa_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
303			       u32 *cfg_base_addr, u64 *cfg_base_addr_index,
304			       u64 *cfg_offset);
305static int hpsa_pci_find_memory_BAR(struct pci_dev *pdev,
306				    unsigned long *memory_bar);
307static int hpsa_lookup_board_id(struct pci_dev *pdev, u32 *board_id,
308				bool *legacy_board);
309static int wait_for_device_to_become_ready(struct ctlr_info *h,
310					   unsigned char lunaddr[],
311					   int reply_queue);
312static int hpsa_wait_for_board_state(struct pci_dev *pdev, void __iomem *vaddr,
313				     int wait_for_ready);
314static inline void finish_cmd(struct CommandList *c);
315static int hpsa_wait_for_mode_change_ack(struct ctlr_info *h);
316#define BOARD_NOT_READY 0
317#define BOARD_READY 1
318static void hpsa_drain_accel_commands(struct ctlr_info *h);
319static void hpsa_flush_cache(struct ctlr_info *h);
320static int hpsa_scsi_ioaccel_queue_command(struct ctlr_info *h,
321	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
322	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk);
323static void hpsa_command_resubmit_worker(struct work_struct *work);
324static u32 lockup_detected(struct ctlr_info *h);
325static int detect_controller_lockup(struct ctlr_info *h);
326static void hpsa_disable_rld_caching(struct ctlr_info *h);
327static inline int hpsa_scsi_do_report_phys_luns(struct ctlr_info *h,
328	struct ReportExtendedLUNdata *buf, int bufsize);
329static bool hpsa_vpd_page_supported(struct ctlr_info *h,
330	unsigned char scsi3addr[], u8 page);
331static int hpsa_luns_changed(struct ctlr_info *h);
332static bool hpsa_cmd_dev_match(struct ctlr_info *h, struct CommandList *c,
333			       struct hpsa_scsi_dev_t *dev,
334			       unsigned char *scsi3addr);
335
336static inline struct ctlr_info *sdev_to_hba(struct scsi_device *sdev)
337{
338	unsigned long *priv = shost_priv(sdev->host);
339	return (struct ctlr_info *) *priv;
340}
341
342static inline struct ctlr_info *shost_to_hba(struct Scsi_Host *sh)
343{
344	unsigned long *priv = shost_priv(sh);
345	return (struct ctlr_info *) *priv;
346}
347
348static inline bool hpsa_is_cmd_idle(struct CommandList *c)
349{
350	return c->scsi_cmd == SCSI_CMD_IDLE;
351}
352
353/* extract sense key, asc, and ascq from sense data.  -1 means invalid. */
354static void decode_sense_data(const u8 *sense_data, int sense_data_len,
355			u8 *sense_key, u8 *asc, u8 *ascq)
356{
357	struct scsi_sense_hdr sshdr;
358	bool rc;
359
360	*sense_key = -1;
361	*asc = -1;
362	*ascq = -1;
363
364	if (sense_data_len < 1)
365		return;
366
367	rc = scsi_normalize_sense(sense_data, sense_data_len, &sshdr);
368	if (rc) {
369		*sense_key = sshdr.sense_key;
370		*asc = sshdr.asc;
371		*ascq = sshdr.ascq;
372	}
373}
374
375static int check_for_unit_attention(struct ctlr_info *h,
376	struct CommandList *c)
377{
378	u8 sense_key, asc, ascq;
379	int sense_len;
380
381	if (c->err_info->SenseLen > sizeof(c->err_info->SenseInfo))
382		sense_len = sizeof(c->err_info->SenseInfo);
383	else
384		sense_len = c->err_info->SenseLen;
385
386	decode_sense_data(c->err_info->SenseInfo, sense_len,
387				&sense_key, &asc, &ascq);
388	if (sense_key != UNIT_ATTENTION || asc == 0xff)
389		return 0;
390
391	switch (asc) {
392	case STATE_CHANGED:
393		dev_warn(&h->pdev->dev,
394			"%s: a state change detected, command retried\n",
395			h->devname);
396		break;
397	case LUN_FAILED:
398		dev_warn(&h->pdev->dev,
399			"%s: LUN failure detected\n", h->devname);
400		break;
401	case REPORT_LUNS_CHANGED:
402		dev_warn(&h->pdev->dev,
403			"%s: report LUN data changed\n", h->devname);
404	/*
405	 * Note: this REPORT_LUNS_CHANGED condition only occurs on the external
406	 * target (array) devices.
407	 */
408		break;
409	case POWER_OR_RESET:
410		dev_warn(&h->pdev->dev,
411			"%s: a power on or device reset detected\n",
412			h->devname);
413		break;
414	case UNIT_ATTENTION_CLEARED:
415		dev_warn(&h->pdev->dev,
416			"%s: unit attention cleared by another initiator\n",
417			h->devname);
418		break;
419	default:
420		dev_warn(&h->pdev->dev,
421			"%s: unknown unit attention detected\n",
422			h->devname);
423		break;
424	}
425	return 1;
426}
427
428static int check_for_busy(struct ctlr_info *h, struct CommandList *c)
429{
430	if (c->err_info->CommandStatus != CMD_TARGET_STATUS ||
431		(c->err_info->ScsiStatus != SAM_STAT_BUSY &&
432		 c->err_info->ScsiStatus != SAM_STAT_TASK_SET_FULL))
433		return 0;
434	dev_warn(&h->pdev->dev, HPSA "device busy");
435	return 1;
436}
437
438static u32 lockup_detected(struct ctlr_info *h);
439static ssize_t host_show_lockup_detected(struct device *dev,
440		struct device_attribute *attr, char *buf)
441{
442	int ld;
443	struct ctlr_info *h;
444	struct Scsi_Host *shost = class_to_shost(dev);
445
446	h = shost_to_hba(shost);
447	ld = lockup_detected(h);
448
449	return sprintf(buf, "ld=%d\n", ld);
450}
451
452static ssize_t host_store_hp_ssd_smart_path_status(struct device *dev,
453					 struct device_attribute *attr,
454					 const char *buf, size_t count)
455{
456	int status, len;
457	struct ctlr_info *h;
458	struct Scsi_Host *shost = class_to_shost(dev);
459	char tmpbuf[10];
460
461	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
462		return -EACCES;
463	len = count > sizeof(tmpbuf) - 1 ? sizeof(tmpbuf) - 1 : count;
464	strncpy(tmpbuf, buf, len);
465	tmpbuf[len] = '\0';
466	if (sscanf(tmpbuf, "%d", &status) != 1)
467		return -EINVAL;
468	h = shost_to_hba(shost);
469	h->acciopath_status = !!status;
470	dev_warn(&h->pdev->dev,
471		"hpsa: HP SSD Smart Path %s via sysfs update.\n",
472		h->acciopath_status ? "enabled" : "disabled");
473	return count;
474}
475
476static ssize_t host_store_raid_offload_debug(struct device *dev,
477					 struct device_attribute *attr,
478					 const char *buf, size_t count)
479{
480	int debug_level, len;
481	struct ctlr_info *h;
482	struct Scsi_Host *shost = class_to_shost(dev);
483	char tmpbuf[10];
484
485	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
486		return -EACCES;
487	len = count > sizeof(tmpbuf) - 1 ? sizeof(tmpbuf) - 1 : count;
488	strncpy(tmpbuf, buf, len);
489	tmpbuf[len] = '\0';
490	if (sscanf(tmpbuf, "%d", &debug_level) != 1)
491		return -EINVAL;
492	if (debug_level < 0)
493		debug_level = 0;
494	h = shost_to_hba(shost);
495	h->raid_offload_debug = debug_level;
496	dev_warn(&h->pdev->dev, "hpsa: Set raid_offload_debug level = %d\n",
497		h->raid_offload_debug);
498	return count;
499}
500
501static ssize_t host_store_rescan(struct device *dev,
502				 struct device_attribute *attr,
503				 const char *buf, size_t count)
504{
505	struct ctlr_info *h;
506	struct Scsi_Host *shost = class_to_shost(dev);
507	h = shost_to_hba(shost);
508	hpsa_scan_start(h->scsi_host);
509	return count;
510}
511
512static void hpsa_turn_off_ioaccel_for_device(struct hpsa_scsi_dev_t *device)
513{
514	device->offload_enabled = 0;
515	device->offload_to_be_enabled = 0;
516}
517
518static ssize_t host_show_firmware_revision(struct device *dev,
519	     struct device_attribute *attr, char *buf)
520{
521	struct ctlr_info *h;
522	struct Scsi_Host *shost = class_to_shost(dev);
523	unsigned char *fwrev;
524
525	h = shost_to_hba(shost);
526	if (!h->hba_inquiry_data)
527		return 0;
528	fwrev = &h->hba_inquiry_data[32];
529	return snprintf(buf, 20, "%c%c%c%c\n",
530		fwrev[0], fwrev[1], fwrev[2], fwrev[3]);
531}
532
533static ssize_t host_show_commands_outstanding(struct device *dev,
534	     struct device_attribute *attr, char *buf)
535{
536	struct Scsi_Host *shost = class_to_shost(dev);
537	struct ctlr_info *h = shost_to_hba(shost);
538
539	return snprintf(buf, 20, "%d\n",
540			atomic_read(&h->commands_outstanding));
541}
542
543static ssize_t host_show_transport_mode(struct device *dev,
544	struct device_attribute *attr, char *buf)
545{
546	struct ctlr_info *h;
547	struct Scsi_Host *shost = class_to_shost(dev);
548
549	h = shost_to_hba(shost);
550	return snprintf(buf, 20, "%s\n",
551		h->transMethod & CFGTBL_Trans_Performant ?
552			"performant" : "simple");
553}
554
555static ssize_t host_show_hp_ssd_smart_path_status(struct device *dev,
556	struct device_attribute *attr, char *buf)
557{
558	struct ctlr_info *h;
559	struct Scsi_Host *shost = class_to_shost(dev);
560
561	h = shost_to_hba(shost);
562	return snprintf(buf, 30, "HP SSD Smart Path %s\n",
563		(h->acciopath_status == 1) ?  "enabled" : "disabled");
564}
565
566/* List of controllers which cannot be hard reset on kexec with reset_devices */
567static u32 unresettable_controller[] = {
568	0x324a103C, /* Smart Array P712m */
569	0x324b103C, /* Smart Array P711m */
570	0x3223103C, /* Smart Array P800 */
571	0x3234103C, /* Smart Array P400 */
572	0x3235103C, /* Smart Array P400i */
573	0x3211103C, /* Smart Array E200i */
574	0x3212103C, /* Smart Array E200 */
575	0x3213103C, /* Smart Array E200i */
576	0x3214103C, /* Smart Array E200i */
577	0x3215103C, /* Smart Array E200i */
578	0x3237103C, /* Smart Array E500 */
579	0x323D103C, /* Smart Array P700m */
580	0x40800E11, /* Smart Array 5i */
581	0x409C0E11, /* Smart Array 6400 */
582	0x409D0E11, /* Smart Array 6400 EM */
583	0x40700E11, /* Smart Array 5300 */
584	0x40820E11, /* Smart Array 532 */
585	0x40830E11, /* Smart Array 5312 */
586	0x409A0E11, /* Smart Array 641 */
587	0x409B0E11, /* Smart Array 642 */
588	0x40910E11, /* Smart Array 6i */
589};
590
591/* List of controllers which cannot even be soft reset */
592static u32 soft_unresettable_controller[] = {
593	0x40800E11, /* Smart Array 5i */
594	0x40700E11, /* Smart Array 5300 */
595	0x40820E11, /* Smart Array 532 */
596	0x40830E11, /* Smart Array 5312 */
597	0x409A0E11, /* Smart Array 641 */
598	0x409B0E11, /* Smart Array 642 */
599	0x40910E11, /* Smart Array 6i */
600	/* Exclude 640x boards.  These are two pci devices in one slot
601	 * which share a battery backed cache module.  One controls the
602	 * cache, the other accesses the cache through the one that controls
603	 * it.  If we reset the one controlling the cache, the other will
604	 * likely not be happy.  Just forbid resetting this conjoined mess.
605	 * The 640x isn't really supported by hpsa anyway.
606	 */
607	0x409C0E11, /* Smart Array 6400 */
608	0x409D0E11, /* Smart Array 6400 EM */
609};
610
611static int board_id_in_array(u32 a[], int nelems, u32 board_id)
612{
613	int i;
614
615	for (i = 0; i < nelems; i++)
616		if (a[i] == board_id)
617			return 1;
618	return 0;
619}
620
621static int ctlr_is_hard_resettable(u32 board_id)
622{
623	return !board_id_in_array(unresettable_controller,
624			ARRAY_SIZE(unresettable_controller), board_id);
625}
626
627static int ctlr_is_soft_resettable(u32 board_id)
628{
629	return !board_id_in_array(soft_unresettable_controller,
630			ARRAY_SIZE(soft_unresettable_controller), board_id);
631}
632
633static int ctlr_is_resettable(u32 board_id)
634{
635	return ctlr_is_hard_resettable(board_id) ||
636		ctlr_is_soft_resettable(board_id);
637}
638
639static ssize_t host_show_resettable(struct device *dev,
640	struct device_attribute *attr, char *buf)
641{
642	struct ctlr_info *h;
643	struct Scsi_Host *shost = class_to_shost(dev);
644
645	h = shost_to_hba(shost);
646	return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h->board_id));
647}
648
649static inline int is_logical_dev_addr_mode(unsigned char scsi3addr[])
650{
651	return (scsi3addr[3] & 0xC0) == 0x40;
652}
653
654static const char * const raid_label[] = { "0", "4", "1(+0)", "5", "5+1", "6",
655	"1(+0)ADM", "UNKNOWN", "PHYS DRV"
656};
657#define HPSA_RAID_0	0
658#define HPSA_RAID_4	1
659#define HPSA_RAID_1	2	/* also used for RAID 10 */
660#define HPSA_RAID_5	3	/* also used for RAID 50 */
661#define HPSA_RAID_51	4
662#define HPSA_RAID_6	5	/* also used for RAID 60 */
663#define HPSA_RAID_ADM	6	/* also used for RAID 1+0 ADM */
664#define RAID_UNKNOWN (ARRAY_SIZE(raid_label) - 2)
665#define PHYSICAL_DRIVE (ARRAY_SIZE(raid_label) - 1)
666
667static inline bool is_logical_device(struct hpsa_scsi_dev_t *device)
668{
669	return !device->physical_device;
670}
671
672static ssize_t raid_level_show(struct device *dev,
673	     struct device_attribute *attr, char *buf)
674{
675	ssize_t l = 0;
676	unsigned char rlevel;
677	struct ctlr_info *h;
678	struct scsi_device *sdev;
679	struct hpsa_scsi_dev_t *hdev;
680	unsigned long flags;
681
682	sdev = to_scsi_device(dev);
683	h = sdev_to_hba(sdev);
684	spin_lock_irqsave(&h->lock, flags);
685	hdev = sdev->hostdata;
686	if (!hdev) {
687		spin_unlock_irqrestore(&h->lock, flags);
688		return -ENODEV;
689	}
690
691	/* Is this even a logical drive? */
692	if (!is_logical_device(hdev)) {
693		spin_unlock_irqrestore(&h->lock, flags);
694		l = snprintf(buf, PAGE_SIZE, "N/A\n");
695		return l;
696	}
697
698	rlevel = hdev->raid_level;
699	spin_unlock_irqrestore(&h->lock, flags);
700	if (rlevel > RAID_UNKNOWN)
701		rlevel = RAID_UNKNOWN;
702	l = snprintf(buf, PAGE_SIZE, "RAID %s\n", raid_label[rlevel]);
703	return l;
704}
705
706static ssize_t lunid_show(struct device *dev,
707	     struct device_attribute *attr, char *buf)
708{
709	struct ctlr_info *h;
710	struct scsi_device *sdev;
711	struct hpsa_scsi_dev_t *hdev;
712	unsigned long flags;
713	unsigned char lunid[8];
714
715	sdev = to_scsi_device(dev);
716	h = sdev_to_hba(sdev);
717	spin_lock_irqsave(&h->lock, flags);
718	hdev = sdev->hostdata;
719	if (!hdev) {
720		spin_unlock_irqrestore(&h->lock, flags);
721		return -ENODEV;
722	}
723	memcpy(lunid, hdev->scsi3addr, sizeof(lunid));
724	spin_unlock_irqrestore(&h->lock, flags);
725	return snprintf(buf, 20, "0x%8phN\n", lunid);
726}
727
728static ssize_t unique_id_show(struct device *dev,
729	     struct device_attribute *attr, char *buf)
730{
731	struct ctlr_info *h;
732	struct scsi_device *sdev;
733	struct hpsa_scsi_dev_t *hdev;
734	unsigned long flags;
735	unsigned char sn[16];
736
737	sdev = to_scsi_device(dev);
738	h = sdev_to_hba(sdev);
739	spin_lock_irqsave(&h->lock, flags);
740	hdev = sdev->hostdata;
741	if (!hdev) {
742		spin_unlock_irqrestore(&h->lock, flags);
743		return -ENODEV;
744	}
745	memcpy(sn, hdev->device_id, sizeof(sn));
746	spin_unlock_irqrestore(&h->lock, flags);
747	return snprintf(buf, 16 * 2 + 2,
748			"%02X%02X%02X%02X%02X%02X%02X%02X"
749			"%02X%02X%02X%02X%02X%02X%02X%02X\n",
750			sn[0], sn[1], sn[2], sn[3],
751			sn[4], sn[5], sn[6], sn[7],
752			sn[8], sn[9], sn[10], sn[11],
753			sn[12], sn[13], sn[14], sn[15]);
754}
755
756static ssize_t sas_address_show(struct device *dev,
757	      struct device_attribute *attr, char *buf)
758{
759	struct ctlr_info *h;
760	struct scsi_device *sdev;
761	struct hpsa_scsi_dev_t *hdev;
762	unsigned long flags;
763	u64 sas_address;
764
765	sdev = to_scsi_device(dev);
766	h = sdev_to_hba(sdev);
767	spin_lock_irqsave(&h->lock, flags);
768	hdev = sdev->hostdata;
769	if (!hdev || is_logical_device(hdev) || !hdev->expose_device) {
770		spin_unlock_irqrestore(&h->lock, flags);
771		return -ENODEV;
772	}
773	sas_address = hdev->sas_address;
774	spin_unlock_irqrestore(&h->lock, flags);
775
776	return snprintf(buf, PAGE_SIZE, "0x%016llx\n", sas_address);
777}
778
779static ssize_t host_show_hp_ssd_smart_path_enabled(struct device *dev,
780	     struct device_attribute *attr, char *buf)
781{
782	struct ctlr_info *h;
783	struct scsi_device *sdev;
784	struct hpsa_scsi_dev_t *hdev;
785	unsigned long flags;
786	int offload_enabled;
787
788	sdev = to_scsi_device(dev);
789	h = sdev_to_hba(sdev);
790	spin_lock_irqsave(&h->lock, flags);
791	hdev = sdev->hostdata;
792	if (!hdev) {
793		spin_unlock_irqrestore(&h->lock, flags);
794		return -ENODEV;
795	}
796	offload_enabled = hdev->offload_enabled;
797	spin_unlock_irqrestore(&h->lock, flags);
798
799	if (hdev->devtype == TYPE_DISK || hdev->devtype == TYPE_ZBC)
800		return snprintf(buf, 20, "%d\n", offload_enabled);
801	else
802		return snprintf(buf, 40, "%s\n",
803				"Not applicable for a controller");
804}
805
806#define MAX_PATHS 8
807static ssize_t path_info_show(struct device *dev,
808	     struct device_attribute *attr, char *buf)
809{
810	struct ctlr_info *h;
811	struct scsi_device *sdev;
812	struct hpsa_scsi_dev_t *hdev;
813	unsigned long flags;
814	int i;
815	int output_len = 0;
816	u8 box;
817	u8 bay;
818	u8 path_map_index = 0;
819	char *active;
820	unsigned char phys_connector[2];
821
822	sdev = to_scsi_device(dev);
823	h = sdev_to_hba(sdev);
824	spin_lock_irqsave(&h->devlock, flags);
825	hdev = sdev->hostdata;
826	if (!hdev) {
827		spin_unlock_irqrestore(&h->devlock, flags);
828		return -ENODEV;
829	}
830
831	bay = hdev->bay;
832	for (i = 0; i < MAX_PATHS; i++) {
833		path_map_index = 1<<i;
834		if (i == hdev->active_path_index)
835			active = "Active";
836		else if (hdev->path_map & path_map_index)
837			active = "Inactive";
838		else
839			continue;
840
841		output_len += scnprintf(buf + output_len,
842				PAGE_SIZE - output_len,
843				"[%d:%d:%d:%d] %20.20s ",
844				h->scsi_host->host_no,
845				hdev->bus, hdev->target, hdev->lun,
846				scsi_device_type(hdev->devtype));
847
848		if (hdev->devtype == TYPE_RAID || is_logical_device(hdev)) {
849			output_len += scnprintf(buf + output_len,
850						PAGE_SIZE - output_len,
851						"%s\n", active);
852			continue;
853		}
854
855		box = hdev->box[i];
856		memcpy(&phys_connector, &hdev->phys_connector[i],
857			sizeof(phys_connector));
858		if (phys_connector[0] < '0')
859			phys_connector[0] = '0';
860		if (phys_connector[1] < '0')
861			phys_connector[1] = '0';
862		output_len += scnprintf(buf + output_len,
863				PAGE_SIZE - output_len,
864				"PORT: %.2s ",
865				phys_connector);
866		if ((hdev->devtype == TYPE_DISK || hdev->devtype == TYPE_ZBC) &&
867			hdev->expose_device) {
868			if (box == 0 || box == 0xFF) {
869				output_len += scnprintf(buf + output_len,
870					PAGE_SIZE - output_len,
871					"BAY: %hhu %s\n",
872					bay, active);
873			} else {
874				output_len += scnprintf(buf + output_len,
875					PAGE_SIZE - output_len,
876					"BOX: %hhu BAY: %hhu %s\n",
877					box, bay, active);
878			}
879		} else if (box != 0 && box != 0xFF) {
880			output_len += scnprintf(buf + output_len,
881				PAGE_SIZE - output_len, "BOX: %hhu %s\n",
882				box, active);
883		} else
884			output_len += scnprintf(buf + output_len,
885				PAGE_SIZE - output_len, "%s\n", active);
886	}
887
888	spin_unlock_irqrestore(&h->devlock, flags);
889	return output_len;
890}
891
892static ssize_t host_show_ctlr_num(struct device *dev,
893	struct device_attribute *attr, char *buf)
894{
895	struct ctlr_info *h;
896	struct Scsi_Host *shost = class_to_shost(dev);
897
898	h = shost_to_hba(shost);
899	return snprintf(buf, 20, "%d\n", h->ctlr);
900}
901
902static ssize_t host_show_legacy_board(struct device *dev,
903	struct device_attribute *attr, char *buf)
904{
905	struct ctlr_info *h;
906	struct Scsi_Host *shost = class_to_shost(dev);
907
908	h = shost_to_hba(shost);
909	return snprintf(buf, 20, "%d\n", h->legacy_board ? 1 : 0);
910}
911
912static DEVICE_ATTR_RO(raid_level);
913static DEVICE_ATTR_RO(lunid);
914static DEVICE_ATTR_RO(unique_id);
915static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
916static DEVICE_ATTR_RO(sas_address);
917static DEVICE_ATTR(hp_ssd_smart_path_enabled, S_IRUGO,
918			host_show_hp_ssd_smart_path_enabled, NULL);
919static DEVICE_ATTR_RO(path_info);
920static DEVICE_ATTR(hp_ssd_smart_path_status, S_IWUSR|S_IRUGO|S_IROTH,
921		host_show_hp_ssd_smart_path_status,
922		host_store_hp_ssd_smart_path_status);
923static DEVICE_ATTR(raid_offload_debug, S_IWUSR, NULL,
924			host_store_raid_offload_debug);
925static DEVICE_ATTR(firmware_revision, S_IRUGO,
926	host_show_firmware_revision, NULL);
927static DEVICE_ATTR(commands_outstanding, S_IRUGO,
928	host_show_commands_outstanding, NULL);
929static DEVICE_ATTR(transport_mode, S_IRUGO,
930	host_show_transport_mode, NULL);
931static DEVICE_ATTR(resettable, S_IRUGO,
932	host_show_resettable, NULL);
933static DEVICE_ATTR(lockup_detected, S_IRUGO,
934	host_show_lockup_detected, NULL);
935static DEVICE_ATTR(ctlr_num, S_IRUGO,
936	host_show_ctlr_num, NULL);
937static DEVICE_ATTR(legacy_board, S_IRUGO,
938	host_show_legacy_board, NULL);
939
940static struct device_attribute *hpsa_sdev_attrs[] = {
941	&dev_attr_raid_level,
942	&dev_attr_lunid,
943	&dev_attr_unique_id,
944	&dev_attr_hp_ssd_smart_path_enabled,
945	&dev_attr_path_info,
946	&dev_attr_sas_address,
947	NULL,
948};
949
950static struct device_attribute *hpsa_shost_attrs[] = {
951	&dev_attr_rescan,
952	&dev_attr_firmware_revision,
953	&dev_attr_commands_outstanding,
954	&dev_attr_transport_mode,
955	&dev_attr_resettable,
956	&dev_attr_hp_ssd_smart_path_status,
957	&dev_attr_raid_offload_debug,
958	&dev_attr_lockup_detected,
959	&dev_attr_ctlr_num,
960	&dev_attr_legacy_board,
961	NULL,
962};
963
964#define HPSA_NRESERVED_CMDS	(HPSA_CMDS_RESERVED_FOR_DRIVER +\
965				 HPSA_MAX_CONCURRENT_PASSTHRUS)
966
967static struct scsi_host_template hpsa_driver_template = {
968	.module			= THIS_MODULE,
969	.name			= HPSA,
970	.proc_name		= HPSA,
971	.queuecommand		= hpsa_scsi_queue_command,
972	.scan_start		= hpsa_scan_start,
973	.scan_finished		= hpsa_scan_finished,
974	.change_queue_depth	= hpsa_change_queue_depth,
975	.this_id		= -1,
976	.eh_device_reset_handler = hpsa_eh_device_reset_handler,
977	.ioctl			= hpsa_ioctl,
978	.slave_alloc		= hpsa_slave_alloc,
979	.slave_configure	= hpsa_slave_configure,
980	.slave_destroy		= hpsa_slave_destroy,
981#ifdef CONFIG_COMPAT
982	.compat_ioctl		= hpsa_compat_ioctl,
983#endif
984	.sdev_attrs = hpsa_sdev_attrs,
985	.shost_attrs = hpsa_shost_attrs,
986	.max_sectors = 2048,
987	.no_write_same = 1,
988};
989
990static inline u32 next_command(struct ctlr_info *h, u8 q)
991{
992	u32 a;
993	struct reply_queue_buffer *rq = &h->reply_queue[q];
994
995	if (h->transMethod & CFGTBL_Trans_io_accel1)
996		return h->access.command_completed(h, q);
997
998	if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
999		return h->access.command_completed(h, q);
1000
1001	if ((rq->head[rq->current_entry] & 1) == rq->wraparound) {
1002		a = rq->head[rq->current_entry];
1003		rq->current_entry++;
1004		atomic_dec(&h->commands_outstanding);
1005	} else {
1006		a = FIFO_EMPTY;
1007	}
1008	/* Check for wraparound */
1009	if (rq->current_entry == h->max_commands) {
1010		rq->current_entry = 0;
1011		rq->wraparound ^= 1;
1012	}
1013	return a;
1014}
1015
1016/*
1017 * There are some special bits in the bus address of the
1018 * command that we have to set for the controller to know
1019 * how to process the command:
1020 *
1021 * Normal performant mode:
1022 * bit 0: 1 means performant mode, 0 means simple mode.
1023 * bits 1-3 = block fetch table entry
1024 * bits 4-6 = command type (== 0)
1025 *
1026 * ioaccel1 mode:
1027 * bit 0 = "performant mode" bit.
1028 * bits 1-3 = block fetch table entry
1029 * bits 4-6 = command type (== 110)
1030 * (command type is needed because ioaccel1 mode
1031 * commands are submitted through the same register as normal
1032 * mode commands, so this is how the controller knows whether
1033 * the command is normal mode or ioaccel1 mode.)
1034 *
1035 * ioaccel2 mode:
1036 * bit 0 = "performant mode" bit.
1037 * bits 1-4 = block fetch table entry (note extra bit)
1038 * bits 4-6 = not needed, because ioaccel2 mode has
1039 * a separate special register for submitting commands.
1040 */
1041
1042/*
1043 * set_performant_mode: Modify the tag for cciss performant
1044 * set bit 0 for pull model, bits 3-1 for block fetch
1045 * register number
1046 */
1047#define DEFAULT_REPLY_QUEUE (-1)
1048static void set_performant_mode(struct ctlr_info *h, struct CommandList *c,
1049					int reply_queue)
1050{
1051	if (likely(h->transMethod & CFGTBL_Trans_Performant)) {
1052		c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
1053		if (unlikely(!h->msix_vectors))
1054			return;
1055		c->Header.ReplyQueue = reply_queue;
1056	}
1057}
1058
1059static void set_ioaccel1_performant_mode(struct ctlr_info *h,
1060						struct CommandList *c,
1061						int reply_queue)
1062{
1063	struct io_accel1_cmd *cp = &h->ioaccel_cmd_pool[c->cmdindex];
1064
1065	/*
1066	 * Tell the controller to post the reply to the queue for this
1067	 * processor.  This seems to give the best I/O throughput.
1068	 */
1069	cp->ReplyQueue = reply_queue;
1070	/*
1071	 * Set the bits in the address sent down to include:
1072	 *  - performant mode bit (bit 0)
1073	 *  - pull count (bits 1-3)
1074	 *  - command type (bits 4-6)
1075	 */
1076	c->busaddr |= 1 | (h->ioaccel1_blockFetchTable[c->Header.SGList] << 1) |
1077					IOACCEL1_BUSADDR_CMDTYPE;
1078}
1079
1080static void set_ioaccel2_tmf_performant_mode(struct ctlr_info *h,
1081						struct CommandList *c,
1082						int reply_queue)
1083{
1084	struct hpsa_tmf_struct *cp = (struct hpsa_tmf_struct *)
1085		&h->ioaccel2_cmd_pool[c->cmdindex];
1086
1087	/* Tell the controller to post the reply to the queue for this
1088	 * processor.  This seems to give the best I/O throughput.
1089	 */
1090	cp->reply_queue = reply_queue;
1091	/* Set the bits in the address sent down to include:
1092	 *  - performant mode bit not used in ioaccel mode 2
1093	 *  - pull count (bits 0-3)
1094	 *  - command type isn't needed for ioaccel2
1095	 */
1096	c->busaddr |= h->ioaccel2_blockFetchTable[0];
1097}
1098
1099static void set_ioaccel2_performant_mode(struct ctlr_info *h,
1100						struct CommandList *c,
1101						int reply_queue)
1102{
1103	struct io_accel2_cmd *cp = &h->ioaccel2_cmd_pool[c->cmdindex];
1104
1105	/*
1106	 * Tell the controller to post the reply to the queue for this
1107	 * processor.  This seems to give the best I/O throughput.
1108	 */
1109	cp->reply_queue = reply_queue;
1110	/*
1111	 * Set the bits in the address sent down to include:
1112	 *  - performant mode bit not used in ioaccel mode 2
1113	 *  - pull count (bits 0-3)
1114	 *  - command type isn't needed for ioaccel2
1115	 */
1116	c->busaddr |= (h->ioaccel2_blockFetchTable[cp->sg_count]);
1117}
1118
1119static int is_firmware_flash_cmd(u8 *cdb)
1120{
1121	return cdb[0] == BMIC_WRITE && cdb[6] == BMIC_FLASH_FIRMWARE;
1122}
1123
1124/*
1125 * During firmware flash, the heartbeat register may not update as frequently
1126 * as it should.  So we dial down lockup detection during firmware flash. and
1127 * dial it back up when firmware flash completes.
1128 */
1129#define HEARTBEAT_SAMPLE_INTERVAL_DURING_FLASH (240 * HZ)
1130#define HEARTBEAT_SAMPLE_INTERVAL (30 * HZ)
1131#define HPSA_EVENT_MONITOR_INTERVAL (15 * HZ)
1132static void dial_down_lockup_detection_during_fw_flash(struct ctlr_info *h,
1133		struct CommandList *c)
1134{
1135	if (!is_firmware_flash_cmd(c->Request.CDB))
1136		return;
1137	atomic_inc(&h->firmware_flash_in_progress);
1138	h->heartbeat_sample_interval = HEARTBEAT_SAMPLE_INTERVAL_DURING_FLASH;
1139}
1140
1141static void dial_up_lockup_detection_on_fw_flash_complete(struct ctlr_info *h,
1142		struct CommandList *c)
1143{
1144	if (is_firmware_flash_cmd(c->Request.CDB) &&
1145		atomic_dec_and_test(&h->firmware_flash_in_progress))
1146		h->heartbeat_sample_interval = HEARTBEAT_SAMPLE_INTERVAL;
1147}
1148
1149static void __enqueue_cmd_and_start_io(struct ctlr_info *h,
1150	struct CommandList *c, int reply_queue)
1151{
1152	dial_down_lockup_detection_during_fw_flash(h, c);
1153	atomic_inc(&h->commands_outstanding);
1154	if (c->device)
1155		atomic_inc(&c->device->commands_outstanding);
1156
1157	reply_queue = h->reply_map[raw_smp_processor_id()];
1158	switch (c->cmd_type) {
1159	case CMD_IOACCEL1:
1160		set_ioaccel1_performant_mode(h, c, reply_queue);
1161		writel(c->busaddr, h->vaddr + SA5_REQUEST_PORT_OFFSET);
1162		break;
1163	case CMD_IOACCEL2:
1164		set_ioaccel2_performant_mode(h, c, reply_queue);
1165		writel(c->busaddr, h->vaddr + IOACCEL2_INBOUND_POSTQ_32);
1166		break;
1167	case IOACCEL2_TMF:
1168		set_ioaccel2_tmf_performant_mode(h, c, reply_queue);
1169		writel(c->busaddr, h->vaddr + IOACCEL2_INBOUND_POSTQ_32);
1170		break;
1171	default:
1172		set_performant_mode(h, c, reply_queue);
1173		h->access.submit_command(h, c);
1174	}
1175}
1176
1177static void enqueue_cmd_and_start_io(struct ctlr_info *h, struct CommandList *c)
1178{
1179	__enqueue_cmd_and_start_io(h, c, DEFAULT_REPLY_QUEUE);
1180}
1181
1182static inline int is_hba_lunid(unsigned char scsi3addr[])
1183{
1184	return memcmp(scsi3addr, RAID_CTLR_LUNID, 8) == 0;
1185}
1186
1187static inline int is_scsi_rev_5(struct ctlr_info *h)
1188{
1189	if (!h->hba_inquiry_data)
1190		return 0;
1191	if ((h->hba_inquiry_data[2] & 0x07) == 5)
1192		return 1;
1193	return 0;
1194}
1195
1196static int hpsa_find_target_lun(struct ctlr_info *h,
1197	unsigned char scsi3addr[], int bus, int *target, int *lun)
1198{
1199	/* finds an unused bus, target, lun for a new physical device
1200	 * assumes h->devlock is held
1201	 */
1202	int i, found = 0;
1203	DECLARE_BITMAP(lun_taken, HPSA_MAX_DEVICES);
1204
1205	bitmap_zero(lun_taken, HPSA_MAX_DEVICES);
1206
1207	for (i = 0; i < h->ndevices; i++) {
1208		if (h->dev[i]->bus == bus && h->dev[i]->target != -1)
1209			__set_bit(h->dev[i]->target, lun_taken);
1210	}
1211
1212	i = find_first_zero_bit(lun_taken, HPSA_MAX_DEVICES);
1213	if (i < HPSA_MAX_DEVICES) {
1214		/* *bus = 1; */
1215		*target = i;
1216		*lun = 0;
1217		found = 1;
1218	}
1219	return !found;
1220}
1221
1222static void hpsa_show_dev_msg(const char *level, struct ctlr_info *h,
1223	struct hpsa_scsi_dev_t *dev, char *description)
1224{
1225#define LABEL_SIZE 25
1226	char label[LABEL_SIZE];
1227
1228	if (h == NULL || h->pdev == NULL || h->scsi_host == NULL)
1229		return;
1230
1231	switch (dev->devtype) {
1232	case TYPE_RAID:
1233		snprintf(label, LABEL_SIZE, "controller");
1234		break;
1235	case TYPE_ENCLOSURE:
1236		snprintf(label, LABEL_SIZE, "enclosure");
1237		break;
1238	case TYPE_DISK:
1239	case TYPE_ZBC:
1240		if (dev->external)
1241			snprintf(label, LABEL_SIZE, "external");
1242		else if (!is_logical_dev_addr_mode(dev->scsi3addr))
1243			snprintf(label, LABEL_SIZE, "%s",
1244				raid_label[PHYSICAL_DRIVE]);
1245		else
1246			snprintf(label, LABEL_SIZE, "RAID-%s",
1247				dev->raid_level > RAID_UNKNOWN ? "?" :
1248				raid_label[dev->raid_level]);
1249		break;
1250	case TYPE_ROM:
1251		snprintf(label, LABEL_SIZE, "rom");
1252		break;
1253	case TYPE_TAPE:
1254		snprintf(label, LABEL_SIZE, "tape");
1255		break;
1256	case TYPE_MEDIUM_CHANGER:
1257		snprintf(label, LABEL_SIZE, "changer");
1258		break;
1259	default:
1260		snprintf(label, LABEL_SIZE, "UNKNOWN");
1261		break;
1262	}
1263
1264	dev_printk(level, &h->pdev->dev,
1265			"scsi %d:%d:%d:%d: %s %s %.8s %.16s %s SSDSmartPathCap%c En%c Exp=%d\n",
1266			h->scsi_host->host_no, dev->bus, dev->target, dev->lun,
1267			description,
1268			scsi_device_type(dev->devtype),
1269			dev->vendor,
1270			dev->model,
1271			label,
1272			dev->offload_config ? '+' : '-',
1273			dev->offload_to_be_enabled ? '+' : '-',
1274			dev->expose_device);
1275}
1276
1277/* Add an entry into h->dev[] array. */
1278static int hpsa_scsi_add_entry(struct ctlr_info *h,
1279		struct hpsa_scsi_dev_t *device,
1280		struct hpsa_scsi_dev_t *added[], int *nadded)
1281{
1282	/* assumes h->devlock is held */
1283	int n = h->ndevices;
1284	int i;
1285	unsigned char addr1[8], addr2[8];
1286	struct hpsa_scsi_dev_t *sd;
1287
1288	if (n >= HPSA_MAX_DEVICES) {
1289		dev_err(&h->pdev->dev, "too many devices, some will be "
1290			"inaccessible.\n");
1291		return -1;
1292	}
1293
1294	/* physical devices do not have lun or target assigned until now. */
1295	if (device->lun != -1)
1296		/* Logical device, lun is already assigned. */
1297		goto lun_assigned;
1298
1299	/* If this device a non-zero lun of a multi-lun device
1300	 * byte 4 of the 8-byte LUN addr will contain the logical
1301	 * unit no, zero otherwise.
1302	 */
1303	if (device->scsi3addr[4] == 0) {
1304		/* This is not a non-zero lun of a multi-lun device */
1305		if (hpsa_find_target_lun(h, device->scsi3addr,
1306			device->bus, &device->target, &device->lun) != 0)
1307			return -1;
1308		goto lun_assigned;
1309	}
1310
1311	/* This is a non-zero lun of a multi-lun device.
1312	 * Search through our list and find the device which
1313	 * has the same 8 byte LUN address, excepting byte 4 and 5.
1314	 * Assign the same bus and target for this new LUN.
1315	 * Use the logical unit number from the firmware.
1316	 */
1317	memcpy(addr1, device->scsi3addr, 8);
1318	addr1[4] = 0;
1319	addr1[5] = 0;
1320	for (i = 0; i < n; i++) {
1321		sd = h->dev[i];
1322		memcpy(addr2, sd->scsi3addr, 8);
1323		addr2[4] = 0;
1324		addr2[5] = 0;
1325		/* differ only in byte 4 and 5? */
1326		if (memcmp(addr1, addr2, 8) == 0) {
1327			device->bus = sd->bus;
1328			device->target = sd->target;
1329			device->lun = device->scsi3addr[4];
1330			break;
1331		}
1332	}
1333	if (device->lun == -1) {
1334		dev_warn(&h->pdev->dev, "physical device with no LUN=0,"
1335			" suspect firmware bug or unsupported hardware "
1336			"configuration.\n");
1337		return -1;
1338	}
1339
1340lun_assigned:
1341
1342	h->dev[n] = device;
1343	h->ndevices++;
1344	added[*nadded] = device;
1345	(*nadded)++;
1346	hpsa_show_dev_msg(KERN_INFO, h, device,
1347		device->expose_device ? "added" : "masked");
1348	return 0;
1349}
1350
1351/*
1352 * Called during a scan operation.
1353 *
1354 * Update an entry in h->dev[] array.
1355 */
1356static void hpsa_scsi_update_entry(struct ctlr_info *h,
1357	int entry, struct hpsa_scsi_dev_t *new_entry)
1358{
1359	/* assumes h->devlock is held */
1360	BUG_ON(entry < 0 || entry >= HPSA_MAX_DEVICES);
1361
1362	/* Raid level changed. */
1363	h->dev[entry]->raid_level = new_entry->raid_level;
1364
1365	/*
1366	 * ioacccel_handle may have changed for a dual domain disk
1367	 */
1368	h->dev[entry]->ioaccel_handle = new_entry->ioaccel_handle;
1369
1370	/* Raid offload parameters changed.  Careful about the ordering. */
1371	if (new_entry->offload_config && new_entry->offload_to_be_enabled) {
1372		/*
1373		 * if drive is newly offload_enabled, we want to copy the
1374		 * raid map data first.  If previously offload_enabled and
1375		 * offload_config were set, raid map data had better be
1376		 * the same as it was before. If raid map data has changed
1377		 * then it had better be the case that
1378		 * h->dev[entry]->offload_enabled is currently 0.
1379		 */
1380		h->dev[entry]->raid_map = new_entry->raid_map;
1381		h->dev[entry]->ioaccel_handle = new_entry->ioaccel_handle;
1382	}
1383	if (new_entry->offload_to_be_enabled) {
1384		h->dev[entry]->ioaccel_handle = new_entry->ioaccel_handle;
1385		wmb(); /* set ioaccel_handle *before* hba_ioaccel_enabled */
1386	}
1387	h->dev[entry]->hba_ioaccel_enabled = new_entry->hba_ioaccel_enabled;
1388	h->dev[entry]->offload_config = new_entry->offload_config;
1389	h->dev[entry]->offload_to_mirror = new_entry->offload_to_mirror;
1390	h->dev[entry]->queue_depth = new_entry->queue_depth;
1391
1392	/*
1393	 * We can turn off ioaccel offload now, but need to delay turning
1394	 * ioaccel on until we can update h->dev[entry]->phys_disk[], but we
1395	 * can't do that until all the devices are updated.
1396	 */
1397	h->dev[entry]->offload_to_be_enabled = new_entry->offload_to_be_enabled;
1398
1399	/*
1400	 * turn ioaccel off immediately if told to do so.
1401	 */
1402	if (!new_entry->offload_to_be_enabled)
1403		h->dev[entry]->offload_enabled = 0;
1404
1405	hpsa_show_dev_msg(KERN_INFO, h, h->dev[entry], "updated");
1406}
1407
1408/* Replace an entry from h->dev[] array. */
1409static void hpsa_scsi_replace_entry(struct ctlr_info *h,
1410	int entry, struct hpsa_scsi_dev_t *new_entry,
1411	struct hpsa_scsi_dev_t *added[], int *nadded,
1412	struct hpsa_scsi_dev_t *removed[], int *nremoved)
1413{
1414	/* assumes h->devlock is held */
1415	BUG_ON(entry < 0 || entry >= HPSA_MAX_DEVICES);
1416	removed[*nremoved] = h->dev[entry];
1417	(*nremoved)++;
1418
1419	/*
1420	 * New physical devices won't have target/lun assigned yet
1421	 * so we need to preserve the values in the slot we are replacing.
1422	 */
1423	if (new_entry->target == -1) {
1424		new_entry->target = h->dev[entry]->target;
1425		new_entry->lun = h->dev[entry]->lun;
1426	}
1427
1428	h->dev[entry] = new_entry;
1429	added[*nadded] = new_entry;
1430	(*nadded)++;
1431
1432	hpsa_show_dev_msg(KERN_INFO, h, new_entry, "replaced");
1433}
1434
1435/* Remove an entry from h->dev[] array. */
1436static void hpsa_scsi_remove_entry(struct ctlr_info *h, int entry,
1437	struct hpsa_scsi_dev_t *removed[], int *nremoved)
1438{
1439	/* assumes h->devlock is held */
1440	int i;
1441	struct hpsa_scsi_dev_t *sd;
1442
1443	BUG_ON(entry < 0 || entry >= HPSA_MAX_DEVICES);
1444
1445	sd = h->dev[entry];
1446	removed[*nremoved] = h->dev[entry];
1447	(*nremoved)++;
1448
1449	for (i = entry; i < h->ndevices-1; i++)
1450		h->dev[i] = h->dev[i+1];
1451	h->ndevices--;
1452	hpsa_show_dev_msg(KERN_INFO, h, sd, "removed");
1453}
1454
1455#define SCSI3ADDR_EQ(a, b) ( \
1456	(a)[7] == (b)[7] && \
1457	(a)[6] == (b)[6] && \
1458	(a)[5] == (b)[5] && \
1459	(a)[4] == (b)[4] && \
1460	(a)[3] == (b)[3] && \
1461	(a)[2] == (b)[2] && \
1462	(a)[1] == (b)[1] && \
1463	(a)[0] == (b)[0])
1464
1465static void fixup_botched_add(struct ctlr_info *h,
1466	struct hpsa_scsi_dev_t *added)
1467{
1468	/* called when scsi_add_device fails in order to re-adjust
1469	 * h->dev[] to match the mid layer's view.
1470	 */
1471	unsigned long flags;
1472	int i, j;
1473
1474	spin_lock_irqsave(&h->lock, flags);
1475	for (i = 0; i < h->ndevices; i++) {
1476		if (h->dev[i] == added) {
1477			for (j = i; j < h->ndevices-1; j++)
1478				h->dev[j] = h->dev[j+1];
1479			h->ndevices--;
1480			break;
1481		}
1482	}
1483	spin_unlock_irqrestore(&h->lock, flags);
1484	kfree(added);
1485}
1486
1487static inline int device_is_the_same(struct hpsa_scsi_dev_t *dev1,
1488	struct hpsa_scsi_dev_t *dev2)
1489{
1490	/* we compare everything except lun and target as these
1491	 * are not yet assigned.  Compare parts likely
1492	 * to differ first
1493	 */
1494	if (memcmp(dev1->scsi3addr, dev2->scsi3addr,
1495		sizeof(dev1->scsi3addr)) != 0)
1496		return 0;
1497	if (memcmp(dev1->device_id, dev2->device_id,
1498		sizeof(dev1->device_id)) != 0)
1499		return 0;
1500	if (memcmp(dev1->model, dev2->model, sizeof(dev1->model)) != 0)
1501		return 0;
1502	if (memcmp(dev1->vendor, dev2->vendor, sizeof(dev1->vendor)) != 0)
1503		return 0;
1504	if (dev1->devtype != dev2->devtype)
1505		return 0;
1506	if (dev1->bus != dev2->bus)
1507		return 0;
1508	return 1;
1509}
1510
1511static inline int device_updated(struct hpsa_scsi_dev_t *dev1,
1512	struct hpsa_scsi_dev_t *dev2)
1513{
1514	/* Device attributes that can change, but don't mean
1515	 * that the device is a different device, nor that the OS
1516	 * needs to be told anything about the change.
1517	 */
1518	if (dev1->raid_level != dev2->raid_level)
1519		return 1;
1520	if (dev1->offload_config != dev2->offload_config)
1521		return 1;
1522	if (dev1->offload_to_be_enabled != dev2->offload_to_be_enabled)
1523		return 1;
1524	if (!is_logical_dev_addr_mode(dev1->scsi3addr))
1525		if (dev1->queue_depth != dev2->queue_depth)
1526			return 1;
1527	/*
1528	 * This can happen for dual domain devices. An active
1529	 * path change causes the ioaccel handle to change
1530	 *
1531	 * for example note the handle differences between p0 and p1
1532	 * Device                    WWN               ,WWN hash,Handle
1533	 * D016 p0|0x3 [02]P2E:01:01,0x5000C5005FC4DACA,0x9B5616,0x01030003
1534	 *	p1                   0x5000C5005FC4DAC9,0x6798C0,0x00040004
1535	 */
1536	if (dev1->ioaccel_handle != dev2->ioaccel_handle)
1537		return 1;
1538	return 0;
1539}
1540
1541/* Find needle in haystack.  If exact match found, return DEVICE_SAME,
1542 * and return needle location in *index.  If scsi3addr matches, but not
1543 * vendor, model, serial num, etc. return DEVICE_CHANGED, and return needle
1544 * location in *index.
1545 * In the case of a minor device attribute change, such as RAID level, just
1546 * return DEVICE_UPDATED, along with the updated device's location in index.
1547 * If needle not found, return DEVICE_NOT_FOUND.
1548 */
1549static int hpsa_scsi_find_entry(struct hpsa_scsi_dev_t *needle,
1550	struct hpsa_scsi_dev_t *haystack[], int haystack_size,
1551	int *index)
1552{
1553	int i;
1554#define DEVICE_NOT_FOUND 0
1555#define DEVICE_CHANGED 1
1556#define DEVICE_SAME 2
1557#define DEVICE_UPDATED 3
1558	if (needle == NULL)
1559		return DEVICE_NOT_FOUND;
1560
1561	for (i = 0; i < haystack_size; i++) {
1562		if (haystack[i] == NULL) /* previously removed. */
1563			continue;
1564		if (SCSI3ADDR_EQ(needle->scsi3addr, haystack[i]->scsi3addr)) {
1565			*index = i;
1566			if (device_is_the_same(needle, haystack[i])) {
1567				if (device_updated(needle, haystack[i]))
1568					return DEVICE_UPDATED;
1569				return DEVICE_SAME;
1570			} else {
1571				/* Keep offline devices offline */
1572				if (needle->volume_offline)
1573					return DEVICE_NOT_FOUND;
1574				return DEVICE_CHANGED;
1575			}
1576		}
1577	}
1578	*index = -1;
1579	return DEVICE_NOT_FOUND;
1580}
1581
1582static void hpsa_monitor_offline_device(struct ctlr_info *h,
1583					unsigned char scsi3addr[])
1584{
1585	struct offline_device_entry *device;
1586	unsigned long flags;
1587
1588	/* Check to see if device is already on the list */
1589	spin_lock_irqsave(&h->offline_device_lock, flags);
1590	list_for_each_entry(device, &h->offline_device_list, offline_list) {
1591		if (memcmp(device->scsi3addr, scsi3addr,
1592			sizeof(device->scsi3addr)) == 0) {
1593			spin_unlock_irqrestore(&h->offline_device_lock, flags);
1594			return;
1595		}
1596	}
1597	spin_unlock_irqrestore(&h->offline_device_lock, flags);
1598
1599	/* Device is not on the list, add it. */
1600	device = kmalloc(sizeof(*device), GFP_KERNEL);
1601	if (!device)
1602		return;
1603
1604	memcpy(device->scsi3addr, scsi3addr, sizeof(device->scsi3addr));
1605	spin_lock_irqsave(&h->offline_device_lock, flags);
1606	list_add_tail(&device->offline_list, &h->offline_device_list);
1607	spin_unlock_irqrestore(&h->offline_device_lock, flags);
1608}
1609
1610/* Print a message explaining various offline volume states */
1611static void hpsa_show_volume_status(struct ctlr_info *h,
1612	struct hpsa_scsi_dev_t *sd)
1613{
1614	if (sd->volume_offline == HPSA_VPD_LV_STATUS_UNSUPPORTED)
1615		dev_info(&h->pdev->dev,
1616			"C%d:B%d:T%d:L%d Volume status is not available through vital product data pages.\n",
1617			h->scsi_host->host_no,
1618			sd->bus, sd->target, sd->lun);
1619	switch (sd->volume_offline) {
1620	case HPSA_LV_OK:
1621		break;
1622	case HPSA_LV_UNDERGOING_ERASE:
1623		dev_info(&h->pdev->dev,
1624			"C%d:B%d:T%d:L%d Volume is undergoing background erase process.\n",
1625			h->scsi_host->host_no,
1626			sd->bus, sd->target, sd->lun);
1627		break;
1628	case HPSA_LV_NOT_AVAILABLE:
1629		dev_info(&h->pdev->dev,
1630			"C%d:B%d:T%d:L%d Volume is waiting for transforming volume.\n",
1631			h->scsi_host->host_no,
1632			sd->bus, sd->target, sd->lun);
1633		break;
1634	case HPSA_LV_UNDERGOING_RPI:
1635		dev_info(&h->pdev->dev,
1636			"C%d:B%d:T%d:L%d Volume is undergoing rapid parity init.\n",
1637			h->scsi_host->host_no,
1638			sd->bus, sd->target, sd->lun);
1639		break;
1640	case HPSA_LV_PENDING_RPI:
1641		dev_info(&h->pdev->dev,
1642			"C%d:B%d:T%d:L%d Volume is queued for rapid parity initialization process.\n",
1643			h->scsi_host->host_no,
1644			sd->bus, sd->target, sd->lun);
1645		break;
1646	case HPSA_LV_ENCRYPTED_NO_KEY:
1647		dev_info(&h->pdev->dev,
1648			"C%d:B%d:T%d:L%d Volume is encrypted and cannot be accessed because key is not present.\n",
1649			h->scsi_host->host_no,
1650			sd->bus, sd->target, sd->lun);
1651		break;
1652	case HPSA_LV_PLAINTEXT_IN_ENCRYPT_ONLY_CONTROLLER:
1653		dev_info(&h->pdev->dev,
1654			"C%d:B%d:T%d:L%d Volume is not encrypted and cannot be accessed because controller is in encryption-only mode.\n",
1655			h->scsi_host->host_no,
1656			sd->bus, sd->target, sd->lun);
1657		break;
1658	case HPSA_LV_UNDERGOING_ENCRYPTION:
1659		dev_info(&h->pdev->dev,
1660			"C%d:B%d:T%d:L%d Volume is undergoing encryption process.\n",
1661			h->scsi_host->host_no,
1662			sd->bus, sd->target, sd->lun);
1663		break;
1664	case HPSA_LV_UNDERGOING_ENCRYPTION_REKEYING:
1665		dev_info(&h->pdev->dev,
1666			"C%d:B%d:T%d:L%d Volume is undergoing encryption re-keying process.\n",
1667			h->scsi_host->host_no,
1668			sd->bus, sd->target, sd->lun);
1669		break;
1670	case HPSA_LV_ENCRYPTED_IN_NON_ENCRYPTED_CONTROLLER:
1671		dev_info(&h->pdev->dev,
1672			"C%d:B%d:T%d:L%d Volume is encrypted and cannot be accessed because controller does not have encryption enabled.\n",
1673			h->scsi_host->host_no,
1674			sd->bus, sd->target, sd->lun);
1675		break;
1676	case HPSA_LV_PENDING_ENCRYPTION:
1677		dev_info(&h->pdev->dev,
1678			"C%d:B%d:T%d:L%d Volume is pending migration to encrypted state, but process has not started.\n",
1679			h->scsi_host->host_no,
1680			sd->bus, sd->target, sd->lun);
1681		break;
1682	case HPSA_LV_PENDING_ENCRYPTION_REKEYING:
1683		dev_info(&h->pdev->dev,
1684			"C%d:B%d:T%d:L%d Volume is encrypted and is pending encryption rekeying.\n",
1685			h->scsi_host->host_no,
1686			sd->bus, sd->target, sd->lun);
1687		break;
1688	}
1689}
1690
1691/*
1692 * Figure the list of physical drive pointers for a logical drive with
1693 * raid offload configured.
1694 */
1695static void hpsa_figure_phys_disk_ptrs(struct ctlr_info *h,
1696				struct hpsa_scsi_dev_t *dev[], int ndevices,
1697				struct hpsa_scsi_dev_t *logical_drive)
1698{
1699	struct raid_map_data *map = &logical_drive->raid_map;
1700	struct raid_map_disk_data *dd = &map->data[0];
1701	int i, j;
1702	int total_disks_per_row = le16_to_cpu(map->data_disks_per_row) +
1703				le16_to_cpu(map->metadata_disks_per_row);
1704	int nraid_map_entries = le16_to_cpu(map->row_cnt) *
1705				le16_to_cpu(map->layout_map_count) *
1706				total_disks_per_row;
1707	int nphys_disk = le16_to_cpu(map->layout_map_count) *
1708				total_disks_per_row;
1709	int qdepth;
1710
1711	if (nraid_map_entries > RAID_MAP_MAX_ENTRIES)
1712		nraid_map_entries = RAID_MAP_MAX_ENTRIES;
1713
1714	logical_drive->nphysical_disks = nraid_map_entries;
1715
1716	qdepth = 0;
1717	for (i = 0; i < nraid_map_entries; i++) {
1718		logical_drive->phys_disk[i] = NULL;
1719		if (!logical_drive->offload_config)
1720			continue;
1721		for (j = 0; j < ndevices; j++) {
1722			if (dev[j] == NULL)
1723				continue;
1724			if (dev[j]->devtype != TYPE_DISK &&
1725			    dev[j]->devtype != TYPE_ZBC)
1726				continue;
1727			if (is_logical_device(dev[j]))
1728				continue;
1729			if (dev[j]->ioaccel_handle != dd[i].ioaccel_handle)
1730				continue;
1731
1732			logical_drive->phys_disk[i] = dev[j];
1733			if (i < nphys_disk)
1734				qdepth = min(h->nr_cmds, qdepth +
1735				    logical_drive->phys_disk[i]->queue_depth);
1736			break;
1737		}
1738
1739		/*
1740		 * This can happen if a physical drive is removed and
1741		 * the logical drive is degraded.  In that case, the RAID
1742		 * map data will refer to a physical disk which isn't actually
1743		 * present.  And in that case offload_enabled should already
1744		 * be 0, but we'll turn it off here just in case
1745		 */
1746		if (!logical_drive->phys_disk[i]) {
1747			dev_warn(&h->pdev->dev,
1748				"%s: [%d:%d:%d:%d] A phys disk component of LV is missing, turning off offload_enabled for LV.\n",
1749				__func__,
1750				h->scsi_host->host_no, logical_drive->bus,
1751				logical_drive->target, logical_drive->lun);
1752			hpsa_turn_off_ioaccel_for_device(logical_drive);
1753			logical_drive->queue_depth = 8;
1754		}
1755	}
1756	if (nraid_map_entries)
1757		/*
1758		 * This is correct for reads, too high for full stripe writes,
1759		 * way too high for partial stripe writes
1760		 */
1761		logical_drive->queue_depth = qdepth;
1762	else {
1763		if (logical_drive->external)
1764			logical_drive->queue_depth = EXTERNAL_QD;
1765		else
1766			logical_drive->queue_depth = h->nr_cmds;
1767	}
1768}
1769
1770static void hpsa_update_log_drive_phys_drive_ptrs(struct ctlr_info *h,
1771				struct hpsa_scsi_dev_t *dev[], int ndevices)
1772{
1773	int i;
1774
1775	for (i = 0; i < ndevices; i++) {
1776		if (dev[i] == NULL)
1777			continue;
1778		if (dev[i]->devtype != TYPE_DISK &&
1779		    dev[i]->devtype != TYPE_ZBC)
1780			continue;
1781		if (!is_logical_device(dev[i]))
1782			continue;
1783
1784		/*
1785		 * If offload is currently enabled, the RAID map and
1786		 * phys_disk[] assignment *better* not be changing
1787		 * because we would be changing ioaccel phsy_disk[] pointers
1788		 * on a ioaccel volume processing I/O requests.
1789		 *
1790		 * If an ioaccel volume status changed, initially because it was
1791		 * re-configured and thus underwent a transformation, or
1792		 * a drive failed, we would have received a state change
1793		 * request and ioaccel should have been turned off. When the
1794		 * transformation completes, we get another state change
1795		 * request to turn ioaccel back on. In this case, we need
1796		 * to update the ioaccel information.
1797		 *
1798		 * Thus: If it is not currently enabled, but will be after
1799		 * the scan completes, make sure the ioaccel pointers
1800		 * are up to date.
1801		 */
1802
1803		if (!dev[i]->offload_enabled && dev[i]->offload_to_be_enabled)
1804			hpsa_figure_phys_disk_ptrs(h, dev, ndevices, dev[i]);
1805	}
1806}
1807
1808static int hpsa_add_device(struct ctlr_info *h, struct hpsa_scsi_dev_t *device)
1809{
1810	int rc = 0;
1811
1812	if (!h->scsi_host)
1813		return 1;
1814
1815	if (is_logical_device(device)) /* RAID */
1816		rc = scsi_add_device(h->scsi_host, device->bus,
1817					device->target, device->lun);
1818	else /* HBA */
1819		rc = hpsa_add_sas_device(h->sas_host, device);
1820
1821	return rc;
1822}
1823
1824static int hpsa_find_outstanding_commands_for_dev(struct ctlr_info *h,
1825						struct hpsa_scsi_dev_t *dev)
1826{
1827	int i;
1828	int count = 0;
1829
1830	for (i = 0; i < h->nr_cmds; i++) {
1831		struct CommandList *c = h->cmd_pool + i;
1832		int refcount = atomic_inc_return(&c->refcount);
1833
1834		if (refcount > 1 && hpsa_cmd_dev_match(h, c, dev,
1835				dev->scsi3addr)) {
1836			unsigned long flags;
1837
1838			spin_lock_irqsave(&h->lock, flags);	/* Implied MB */
1839			if (!hpsa_is_cmd_idle(c))
1840				++count;
1841			spin_unlock_irqrestore(&h->lock, flags);
1842		}
1843
1844		cmd_free(h, c);
1845	}
1846
1847	return count;
1848}
1849
1850#define NUM_WAIT 20
1851static void hpsa_wait_for_outstanding_commands_for_dev(struct ctlr_info *h,
1852						struct hpsa_scsi_dev_t *device)
1853{
1854	int cmds = 0;
1855	int waits = 0;
1856	int num_wait = NUM_WAIT;
1857
1858	if (device->external)
1859		num_wait = HPSA_EH_PTRAID_TIMEOUT;
1860
1861	while (1) {
1862		cmds = hpsa_find_outstanding_commands_for_dev(h, device);
1863		if (cmds == 0)
1864			break;
1865		if (++waits > num_wait)
1866			break;
1867		msleep(1000);
1868	}
1869
1870	if (waits > num_wait) {
1871		dev_warn(&h->pdev->dev,
1872			"%s: removing device [%d:%d:%d:%d] with %d outstanding commands!\n",
1873			__func__,
1874			h->scsi_host->host_no,
1875			device->bus, device->target, device->lun, cmds);
1876	}
1877}
1878
1879static void hpsa_remove_device(struct ctlr_info *h,
1880			struct hpsa_scsi_dev_t *device)
1881{
1882	struct scsi_device *sdev = NULL;
1883
1884	if (!h->scsi_host)
1885		return;
1886
1887	/*
1888	 * Allow for commands to drain
1889	 */
1890	device->removed = 1;
1891	hpsa_wait_for_outstanding_commands_for_dev(h, device);
1892
1893	if (is_logical_device(device)) { /* RAID */
1894		sdev = scsi_device_lookup(h->scsi_host, device->bus,
1895						device->target, device->lun);
1896		if (sdev) {
1897			scsi_remove_device(sdev);
1898			scsi_device_put(sdev);
1899		} else {
1900			/*
1901			 * We don't expect to get here.  Future commands
1902			 * to this device will get a selection timeout as
1903			 * if the device were gone.
1904			 */
1905			hpsa_show_dev_msg(KERN_WARNING, h, device,
1906					"didn't find device for removal.");
1907		}
1908	} else { /* HBA */
1909
1910		hpsa_remove_sas_device(device);
1911	}
1912}
1913
1914static void adjust_hpsa_scsi_table(struct ctlr_info *h,
1915	struct hpsa_scsi_dev_t *sd[], int nsds)
1916{
1917	/* sd contains scsi3 addresses and devtypes, and inquiry
1918	 * data.  This function takes what's in sd to be the current
1919	 * reality and updates h->dev[] to reflect that reality.
1920	 */
1921	int i, entry, device_change, changes = 0;
1922	struct hpsa_scsi_dev_t *csd;
1923	unsigned long flags;
1924	struct hpsa_scsi_dev_t **added, **removed;
1925	int nadded, nremoved;
1926
1927	/*
1928	 * A reset can cause a device status to change
1929	 * re-schedule the scan to see what happened.
1930	 */
1931	spin_lock_irqsave(&h->reset_lock, flags);
1932	if (h->reset_in_progress) {
1933		h->drv_req_rescan = 1;
1934		spin_unlock_irqrestore(&h->reset_lock, flags);
1935		return;
1936	}
1937	spin_unlock_irqrestore(&h->reset_lock, flags);
1938
1939	added = kcalloc(HPSA_MAX_DEVICES, sizeof(*added), GFP_KERNEL);
1940	removed = kcalloc(HPSA_MAX_DEVICES, sizeof(*removed), GFP_KERNEL);
1941
1942	if (!added || !removed) {
1943		dev_warn(&h->pdev->dev, "out of memory in "
1944			"adjust_hpsa_scsi_table\n");
1945		goto free_and_out;
1946	}
1947
1948	spin_lock_irqsave(&h->devlock, flags);
1949
1950	/* find any devices in h->dev[] that are not in
1951	 * sd[] and remove them from h->dev[], and for any
1952	 * devices which have changed, remove the old device
1953	 * info and add the new device info.
1954	 * If minor device attributes change, just update
1955	 * the existing device structure.
1956	 */
1957	i = 0;
1958	nremoved = 0;
1959	nadded = 0;
1960	while (i < h->ndevices) {
1961		csd = h->dev[i];
1962		device_change = hpsa_scsi_find_entry(csd, sd, nsds, &entry);
1963		if (device_change == DEVICE_NOT_FOUND) {
1964			changes++;
1965			hpsa_scsi_remove_entry(h, i, removed, &nremoved);
1966			continue; /* remove ^^^, hence i not incremented */
1967		} else if (device_change == DEVICE_CHANGED) {
1968			changes++;
1969			hpsa_scsi_replace_entry(h, i, sd[entry],
1970				added, &nadded, removed, &nremoved);
1971			/* Set it to NULL to prevent it from being freed
1972			 * at the bottom of hpsa_update_scsi_devices()
1973			 */
1974			sd[entry] = NULL;
1975		} else if (device_change == DEVICE_UPDATED) {
1976			hpsa_scsi_update_entry(h, i, sd[entry]);
1977		}
1978		i++;
1979	}
1980
1981	/* Now, make sure every device listed in sd[] is also
1982	 * listed in h->dev[], adding them if they aren't found
1983	 */
1984
1985	for (i = 0; i < nsds; i++) {
1986		if (!sd[i]) /* if already added above. */
1987			continue;
1988
1989		/* Don't add devices which are NOT READY, FORMAT IN PROGRESS
1990		 * as the SCSI mid-layer does not handle such devices well.
1991		 * It relentlessly loops sending TUR at 3Hz, then READ(10)
1992		 * at 160Hz, and prevents the system from coming up.
1993		 */
1994		if (sd[i]->volume_offline) {
1995			hpsa_show_volume_status(h, sd[i]);
1996			hpsa_show_dev_msg(KERN_INFO, h, sd[i], "offline");
1997			continue;
1998		}
1999
2000		device_change = hpsa_scsi_find_entry(sd[i], h->dev,
2001					h->ndevices, &entry);
2002		if (device_change == DEVICE_NOT_FOUND) {
2003			changes++;
2004			if (hpsa_scsi_add_entry(h, sd[i], added, &nadded) != 0)
2005				break;
2006			sd[i] = NULL; /* prevent from being freed later. */
2007		} else if (device_change == DEVICE_CHANGED) {
2008			/* should never happen... */
2009			changes++;
2010			dev_warn(&h->pdev->dev,
2011				"device unexpectedly changed.\n");
2012			/* but if it does happen, we just ignore that device */
2013		}
2014	}
2015	hpsa_update_log_drive_phys_drive_ptrs(h, h->dev, h->ndevices);
2016
2017	/*
2018	 * Now that h->dev[]->phys_disk[] is coherent, we can enable
2019	 * any logical drives that need it enabled.
2020	 *
2021	 * The raid map should be current by now.
2022	 *
2023	 * We are updating the device list used for I/O requests.
2024	 */
2025	for (i = 0; i < h->ndevices; i++) {
2026		if (h->dev[i] == NULL)
2027			continue;
2028		h->dev[i]->offload_enabled = h->dev[i]->offload_to_be_enabled;
2029	}
2030
2031	spin_unlock_irqrestore(&h->devlock, flags);
2032
2033	/* Monitor devices which are in one of several NOT READY states to be
2034	 * brought online later. This must be done without holding h->devlock,
2035	 * so don't touch h->dev[]
2036	 */
2037	for (i = 0; i < nsds; i++) {
2038		if (!sd[i]) /* if already added above. */
2039			continue;
2040		if (sd[i]->volume_offline)
2041			hpsa_monitor_offline_device(h, sd[i]->scsi3addr);
2042	}
2043
2044	/* Don't notify scsi mid layer of any changes the first time through
2045	 * (or if there are no changes) scsi_scan_host will do it later the
2046	 * first time through.
2047	 */
2048	if (!changes)
2049		goto free_and_out;
2050
2051	/* Notify scsi mid layer of any removed devices */
2052	for (i = 0; i < nremoved; i++) {
2053		if (removed[i] == NULL)
2054			continue;
2055		if (removed[i]->expose_device)
2056			hpsa_remove_device(h, removed[i]);
2057		kfree(removed[i]);
2058		removed[i] = NULL;
2059	}
2060
2061	/* Notify scsi mid layer of any added devices */
2062	for (i = 0; i < nadded; i++) {
2063		int rc = 0;
2064
2065		if (added[i] == NULL)
2066			continue;
2067		if (!(added[i]->expose_device))
2068			continue;
2069		rc = hpsa_add_device(h, added[i]);
2070		if (!rc)
2071			continue;
2072		dev_warn(&h->pdev->dev,
2073			"addition failed %d, device not added.", rc);
2074		/* now we have to remove it from h->dev,
2075		 * since it didn't get added to scsi mid layer
2076		 */
2077		fixup_botched_add(h, added[i]);
2078		h->drv_req_rescan = 1;
2079	}
2080
2081free_and_out:
2082	kfree(added);
2083	kfree(removed);
2084}
2085
2086/*
2087 * Lookup bus/target/lun and return corresponding struct hpsa_scsi_dev_t *
2088 * Assume's h->devlock is held.
2089 */
2090static struct hpsa_scsi_dev_t *lookup_hpsa_scsi_dev(struct ctlr_info *h,
2091	int bus, int target, int lun)
2092{
2093	int i;
2094	struct hpsa_scsi_dev_t *sd;
2095
2096	for (i = 0; i < h->ndevices; i++) {
2097		sd = h->dev[i];
2098		if (sd->bus == bus && sd->target == target && sd->lun == lun)
2099			return sd;
2100	}
2101	return NULL;
2102}
2103
2104static int hpsa_slave_alloc(struct scsi_device *sdev)
2105{
2106	struct hpsa_scsi_dev_t *sd = NULL;
2107	unsigned long flags;
2108	struct ctlr_info *h;
2109
2110	h = sdev_to_hba(sdev);
2111	spin_lock_irqsave(&h->devlock, flags);
2112	if (sdev_channel(sdev) == HPSA_PHYSICAL_DEVICE_BUS) {
2113		struct scsi_target *starget;
2114		struct sas_rphy *rphy;
2115
2116		starget = scsi_target(sdev);
2117		rphy = target_to_rphy(starget);
2118		sd = hpsa_find_device_by_sas_rphy(h, rphy);
2119		if (sd) {
2120			sd->target = sdev_id(sdev);
2121			sd->lun = sdev->lun;
2122		}
2123	}
2124	if (!sd)
2125		sd = lookup_hpsa_scsi_dev(h, sdev_channel(sdev),
2126					sdev_id(sdev), sdev->lun);
2127
2128	if (sd && sd->expose_device) {
2129		atomic_set(&sd->ioaccel_cmds_out, 0);
2130		sdev->hostdata = sd;
2131	} else
2132		sdev->hostdata = NULL;
2133	spin_unlock_irqrestore(&h->devlock, flags);
2134	return 0;
2135}
2136
2137/* configure scsi device based on internal per-device structure */
2138#define CTLR_TIMEOUT (120 * HZ)
2139static int hpsa_slave_configure(struct scsi_device *sdev)
2140{
2141	struct hpsa_scsi_dev_t *sd;
2142	int queue_depth;
2143
2144	sd = sdev->hostdata;
2145	sdev->no_uld_attach = !sd || !sd->expose_device;
2146
2147	if (sd) {
2148		sd->was_removed = 0;
2149		queue_depth = sd->queue_depth != 0 ?
2150				sd->queue_depth : sdev->host->can_queue;
2151		if (sd->external) {
2152			queue_depth = EXTERNAL_QD;
2153			sdev->eh_timeout = HPSA_EH_PTRAID_TIMEOUT;
2154			blk_queue_rq_timeout(sdev->request_queue,
2155						HPSA_EH_PTRAID_TIMEOUT);
2156		}
2157		if (is_hba_lunid(sd->scsi3addr)) {
2158			sdev->eh_timeout = CTLR_TIMEOUT;
2159			blk_queue_rq_timeout(sdev->request_queue, CTLR_TIMEOUT);
2160		}
2161	} else {
2162		queue_depth = sdev->host->can_queue;
2163	}
2164
2165	scsi_change_queue_depth(sdev, queue_depth);
2166
2167	return 0;
2168}
2169
2170static void hpsa_slave_destroy(struct scsi_device *sdev)
2171{
2172	struct hpsa_scsi_dev_t *hdev = NULL;
2173
2174	hdev = sdev->hostdata;
2175
2176	if (hdev)
2177		hdev->was_removed = 1;
2178}
2179
2180static void hpsa_free_ioaccel2_sg_chain_blocks(struct ctlr_info *h)
2181{
2182	int i;
2183
2184	if (!h->ioaccel2_cmd_sg_list)
2185		return;
2186	for (i = 0; i < h->nr_cmds; i++) {
2187		kfree(h->ioaccel2_cmd_sg_list[i]);
2188		h->ioaccel2_cmd_sg_list[i] = NULL;
2189	}
2190	kfree(h->ioaccel2_cmd_sg_list);
2191	h->ioaccel2_cmd_sg_list = NULL;
2192}
2193
2194static int hpsa_allocate_ioaccel2_sg_chain_blocks(struct ctlr_info *h)
2195{
2196	int i;
2197
2198	if (h->chainsize <= 0)
2199		return 0;
2200
2201	h->ioaccel2_cmd_sg_list =
2202		kcalloc(h->nr_cmds, sizeof(*h->ioaccel2_cmd_sg_list),
2203					GFP_KERNEL);
2204	if (!h->ioaccel2_cmd_sg_list)
2205		return -ENOMEM;
2206	for (i = 0; i < h->nr_cmds; i++) {
2207		h->ioaccel2_cmd_sg_list[i] =
2208			kmalloc_array(h->maxsgentries,
2209				      sizeof(*h->ioaccel2_cmd_sg_list[i]),
2210				      GFP_KERNEL);
2211		if (!h->ioaccel2_cmd_sg_list[i])
2212			goto clean;
2213	}
2214	return 0;
2215
2216clean:
2217	hpsa_free_ioaccel2_sg_chain_blocks(h);
2218	return -ENOMEM;
2219}
2220
2221static void hpsa_free_sg_chain_blocks(struct ctlr_info *h)
2222{
2223	int i;
2224
2225	if (!h->cmd_sg_list)
2226		return;
2227	for (i = 0; i < h->nr_cmds; i++) {
2228		kfree(h->cmd_sg_list[i]);
2229		h->cmd_sg_list[i] = NULL;
2230	}
2231	kfree(h->cmd_sg_list);
2232	h->cmd_sg_list = NULL;
2233}
2234
2235static int hpsa_alloc_sg_chain_blocks(struct ctlr_info *h)
2236{
2237	int i;
2238
2239	if (h->chainsize <= 0)
2240		return 0;
2241
2242	h->cmd_sg_list = kcalloc(h->nr_cmds, sizeof(*h->cmd_sg_list),
2243				 GFP_KERNEL);
2244	if (!h->cmd_sg_list)
2245		return -ENOMEM;
2246
2247	for (i = 0; i < h->nr_cmds; i++) {
2248		h->cmd_sg_list[i] = kmalloc_array(h->chainsize,
2249						  sizeof(*h->cmd_sg_list[i]),
2250						  GFP_KERNEL);
2251		if (!h->cmd_sg_list[i])
2252			goto clean;
2253
2254	}
2255	return 0;
2256
2257clean:
2258	hpsa_free_sg_chain_blocks(h);
2259	return -ENOMEM;
2260}
2261
2262static int hpsa_map_ioaccel2_sg_chain_block(struct ctlr_info *h,
2263	struct io_accel2_cmd *cp, struct CommandList *c)
2264{
2265	struct ioaccel2_sg_element *chain_block;
2266	u64 temp64;
2267	u32 chain_size;
2268
2269	chain_block = h->ioaccel2_cmd_sg_list[c->cmdindex];
2270	chain_size = le32_to_cpu(cp->sg[0].length);
2271	temp64 = dma_map_single(&h->pdev->dev, chain_block, chain_size,
2272				DMA_TO_DEVICE);
2273	if (dma_mapping_error(&h->pdev->dev, temp64)) {
2274		/* prevent subsequent unmapping */
2275		cp->sg->address = 0;
2276		return -1;
2277	}
2278	cp->sg->address = cpu_to_le64(temp64);
2279	return 0;
2280}
2281
2282static void hpsa_unmap_ioaccel2_sg_chain_block(struct ctlr_info *h,
2283	struct io_accel2_cmd *cp)
2284{
2285	struct ioaccel2_sg_element *chain_sg;
2286	u64 temp64;
2287	u32 chain_size;
2288
2289	chain_sg = cp->sg;
2290	temp64 = le64_to_cpu(chain_sg->address);
2291	chain_size = le32_to_cpu(cp->sg[0].length);
2292	dma_unmap_single(&h->pdev->dev, temp64, chain_size, DMA_TO_DEVICE);
2293}
2294
2295static int hpsa_map_sg_chain_block(struct ctlr_info *h,
2296	struct CommandList *c)
2297{
2298	struct SGDescriptor *chain_sg, *chain_block;
2299	u64 temp64;
2300	u32 chain_len;
2301
2302	chain_sg = &c->SG[h->max_cmd_sg_entries - 1];
2303	chain_block = h->cmd_sg_list[c->cmdindex];
2304	chain_sg->Ext = cpu_to_le32(HPSA_SG_CHAIN);
2305	chain_len = sizeof(*chain_sg) *
2306		(le16_to_cpu(c->Header.SGTotal) - h->max_cmd_sg_entries);
2307	chain_sg->Len = cpu_to_le32(chain_len);
2308	temp64 = dma_map_single(&h->pdev->dev, chain_block, chain_len,
2309				DMA_TO_DEVICE);
2310	if (dma_mapping_error(&h->pdev->dev, temp64)) {
2311		/* prevent subsequent unmapping */
2312		chain_sg->Addr = cpu_to_le64(0);
2313		return -1;
2314	}
2315	chain_sg->Addr = cpu_to_le64(temp64);
2316	return 0;
2317}
2318
2319static void hpsa_unmap_sg_chain_block(struct ctlr_info *h,
2320	struct CommandList *c)
2321{
2322	struct SGDescriptor *chain_sg;
2323
2324	if (le16_to_cpu(c->Header.SGTotal) <= h->max_cmd_sg_entries)
2325		return;
2326
2327	chain_sg = &c->SG[h->max_cmd_sg_entries - 1];
2328	dma_unmap_single(&h->pdev->dev, le64_to_cpu(chain_sg->Addr),
2329			le32_to_cpu(chain_sg->Len), DMA_TO_DEVICE);
2330}
2331
2332
2333/* Decode the various types of errors on ioaccel2 path.
2334 * Return 1 for any error that should generate a RAID path retry.
2335 * Return 0 for errors that don't require a RAID path retry.
2336 */
2337static int handle_ioaccel_mode2_error(struct ctlr_info *h,
2338					struct CommandList *c,
2339					struct scsi_cmnd *cmd,
2340					struct io_accel2_cmd *c2,
2341					struct hpsa_scsi_dev_t *dev)
2342{
2343	int data_len;
2344	int retry = 0;
2345	u32 ioaccel2_resid = 0;
2346
2347	switch (c2->error_data.serv_response) {
2348	case IOACCEL2_SERV_RESPONSE_COMPLETE:
2349		switch (c2->error_data.status) {
2350		case IOACCEL2_STATUS_SR_TASK_COMP_GOOD:
2351			if (cmd)
2352				cmd->result = 0;
2353			break;
2354		case IOACCEL2_STATUS_SR_TASK_COMP_CHK_COND:
2355			cmd->result |= SAM_STAT_CHECK_CONDITION;
2356			if (c2->error_data.data_present !=
2357					IOACCEL2_SENSE_DATA_PRESENT) {
2358				memset(cmd->sense_buffer, 0,
2359					SCSI_SENSE_BUFFERSIZE);
2360				break;
2361			}
2362			/* copy the sense data */
2363			data_len = c2->error_data.sense_data_len;
2364			if (data_len > SCSI_SENSE_BUFFERSIZE)
2365				data_len = SCSI_SENSE_BUFFERSIZE;
2366			if (data_len > sizeof(c2->error_data.sense_data_buff))
2367				data_len =
2368					sizeof(c2->error_data.sense_data_buff);
2369			memcpy(cmd->sense_buffer,
2370				c2->error_data.sense_data_buff, data_len);
2371			retry = 1;
2372			break;
2373		case IOACCEL2_STATUS_SR_TASK_COMP_BUSY:
2374			retry = 1;
2375			break;
2376		case IOACCEL2_STATUS_SR_TASK_COMP_RES_CON:
2377			retry = 1;
2378			break;
2379		case IOACCEL2_STATUS_SR_TASK_COMP_SET_FULL:
2380			retry = 1;
2381			break;
2382		case IOACCEL2_STATUS_SR_TASK_COMP_ABORTED:
2383			retry = 1;
2384			break;
2385		default:
2386			retry = 1;
2387			break;
2388		}
2389		break;
2390	case IOACCEL2_SERV_RESPONSE_FAILURE:
2391		switch (c2->error_data.status) {
2392		case IOACCEL2_STATUS_SR_IO_ERROR:
2393		case IOACCEL2_STATUS_SR_IO_ABORTED:
2394		case IOACCEL2_STATUS_SR_OVERRUN:
2395			retry = 1;
2396			break;
2397		case IOACCEL2_STATUS_SR_UNDERRUN:
2398			cmd->result = (DID_OK << 16);		/* host byte */
2399			cmd->result |= (COMMAND_COMPLETE << 8);	/* msg byte */
2400			ioaccel2_resid = get_unaligned_le32(
2401						&c2->error_data.resid_cnt[0]);
2402			scsi_set_resid(cmd, ioaccel2_resid);
2403			break;
2404		case IOACCEL2_STATUS_SR_NO_PATH_TO_DEVICE:
2405		case IOACCEL2_STATUS_SR_INVALID_DEVICE:
2406		case IOACCEL2_STATUS_SR_IOACCEL_DISABLED:
2407			/*
2408			 * Did an HBA disk disappear? We will eventually
2409			 * get a state change event from the controller but
2410			 * in the meantime, we need to tell the OS that the
2411			 * HBA disk is no longer there and stop I/O
2412			 * from going down. This allows the potential re-insert
2413			 * of the disk to get the same device node.
2414			 */
2415			if (dev->physical_device && dev->expose_device) {
2416				cmd->result = DID_NO_CONNECT << 16;
2417				dev->removed = 1;
2418				h->drv_req_rescan = 1;
2419				dev_warn(&h->pdev->dev,
2420					"%s: device is gone!\n", __func__);
2421			} else
2422				/*
2423				 * Retry by sending down the RAID path.
2424				 * We will get an event from ctlr to
2425				 * trigger rescan regardless.
2426				 */
2427				retry = 1;
2428			break;
2429		default:
2430			retry = 1;
2431		}
2432		break;
2433	case IOACCEL2_SERV_RESPONSE_TMF_COMPLETE:
2434		break;
2435	case IOACCEL2_SERV_RESPONSE_TMF_SUCCESS:
2436		break;
2437	case IOACCEL2_SERV_RESPONSE_TMF_REJECTED:
2438		retry = 1;
2439		break;
2440	case IOACCEL2_SERV_RESPONSE_TMF_WRONG_LUN:
2441		break;
2442	default:
2443		retry = 1;
2444		break;
2445	}
2446
2447	if (dev->in_reset)
2448		retry = 0;
2449
2450	return retry;	/* retry on raid path? */
2451}
2452
2453static void hpsa_cmd_resolve_events(struct ctlr_info *h,
2454		struct CommandList *c)
2455{
2456	struct hpsa_scsi_dev_t *dev = c->device;
2457
2458	/*
2459	 * Reset c->scsi_cmd here so that the reset handler will know
2460	 * this command has completed.  Then, check to see if the handler is
2461	 * waiting for this command, and, if so, wake it.
2462	 */
2463	c->scsi_cmd = SCSI_CMD_IDLE;
2464	mb();	/* Declare command idle before checking for pending events. */
2465	if (dev) {
2466		atomic_dec(&dev->commands_outstanding);
2467		if (dev->in_reset &&
2468			atomic_read(&dev->commands_outstanding) <= 0)
2469			wake_up_all(&h->event_sync_wait_queue);
2470	}
2471}
2472
2473static void hpsa_cmd_resolve_and_free(struct ctlr_info *h,
2474				      struct CommandList *c)
2475{
2476	hpsa_cmd_resolve_events(h, c);
2477	cmd_tagged_free(h, c);
2478}
2479
2480static void hpsa_cmd_free_and_done(struct ctlr_info *h,
2481		struct CommandList *c, struct scsi_cmnd *cmd)
2482{
2483	hpsa_cmd_resolve_and_free(h, c);
2484	if (cmd && cmd->scsi_done)
2485		cmd->scsi_done(cmd);
2486}
2487
2488static void hpsa_retry_cmd(struct ctlr_info *h, struct CommandList *c)
2489{
2490	INIT_WORK(&c->work, hpsa_command_resubmit_worker);
2491	queue_work_on(raw_smp_processor_id(), h->resubmit_wq, &c->work);
2492}
2493
2494static void process_ioaccel2_completion(struct ctlr_info *h,
2495		struct CommandList *c, struct scsi_cmnd *cmd,
2496		struct hpsa_scsi_dev_t *dev)
2497{
2498	struct io_accel2_cmd *c2 = &h->ioaccel2_cmd_pool[c->cmdindex];
2499
2500	/* check for good status */
2501	if (likely(c2->error_data.serv_response == 0 &&
2502			c2->error_data.status == 0)) {
2503		cmd->result = 0;
2504		return hpsa_cmd_free_and_done(h, c, cmd);
2505	}
2506
2507	/*
2508	 * Any RAID offload error results in retry which will use
2509	 * the normal I/O path so the controller can handle whatever is
2510	 * wrong.
2511	 */
2512	if (is_logical_device(dev) &&
2513		c2->error_data.serv_response ==
2514			IOACCEL2_SERV_RESPONSE_FAILURE) {
2515		if (c2->error_data.status ==
2516			IOACCEL2_STATUS_SR_IOACCEL_DISABLED) {
2517			hpsa_turn_off_ioaccel_for_device(dev);
2518		}
2519
2520		if (dev->in_reset) {
2521			cmd->result = DID_RESET << 16;
2522			return hpsa_cmd_free_and_done(h, c, cmd);
2523		}
2524
2525		return hpsa_retry_cmd(h, c);
2526	}
2527
2528	if (handle_ioaccel_mode2_error(h, c, cmd, c2, dev))
2529		return hpsa_retry_cmd(h, c);
2530
2531	return hpsa_cmd_free_and_done(h, c, cmd);
2532}
2533
2534/* Returns 0 on success, < 0 otherwise. */
2535static int hpsa_evaluate_tmf_status(struct ctlr_info *h,
2536					struct CommandList *cp)
2537{
2538	u8 tmf_status = cp->err_info->ScsiStatus;
2539
2540	switch (tmf_status) {
2541	case CISS_TMF_COMPLETE:
2542		/*
2543		 * CISS_TMF_COMPLETE never happens, instead,
2544		 * ei->CommandStatus == 0 for this case.
2545		 */
2546	case CISS_TMF_SUCCESS:
2547		return 0;
2548	case CISS_TMF_INVALID_FRAME:
2549	case CISS_TMF_NOT_SUPPORTED:
2550	case CISS_TMF_FAILED:
2551	case CISS_TMF_WRONG_LUN:
2552	case CISS_TMF_OVERLAPPED_TAG:
2553		break;
2554	default:
2555		dev_warn(&h->pdev->dev, "Unknown TMF status: 0x%02x\n",
2556				tmf_status);
2557		break;
2558	}
2559	return -tmf_status;
2560}
2561
2562static void complete_scsi_command(struct CommandList *cp)
2563{
2564	struct scsi_cmnd *cmd;
2565	struct ctlr_info *h;
2566	struct ErrorInfo *ei;
2567	struct hpsa_scsi_dev_t *dev;
2568	struct io_accel2_cmd *c2;
2569
2570	u8 sense_key;
2571	u8 asc;      /* additional sense code */
2572	u8 ascq;     /* additional sense code qualifier */
2573	unsigned long sense_data_size;
2574
2575	ei = cp->err_info;
2576	cmd = cp->scsi_cmd;
2577	h = cp->h;
2578
2579	if (!cmd->device) {
2580		cmd->result = DID_NO_CONNECT << 16;
2581		return hpsa_cmd_free_and_done(h, cp, cmd);
2582	}
2583
2584	dev = cmd->device->hostdata;
2585	if (!dev) {
2586		cmd->result = DID_NO_CONNECT << 16;
2587		return hpsa_cmd_free_and_done(h, cp, cmd);
2588	}
2589	c2 = &h->ioaccel2_cmd_pool[cp->cmdindex];
2590
2591	scsi_dma_unmap(cmd); /* undo the DMA mappings */
2592	if ((cp->cmd_type == CMD_SCSI) &&
2593		(le16_to_cpu(cp->Header.SGTotal) > h->max_cmd_sg_entries))
2594		hpsa_unmap_sg_chain_block(h, cp);
2595
2596	if ((cp->cmd_type == CMD_IOACCEL2) &&
2597		(c2->sg[0].chain_indicator == IOACCEL2_CHAIN))
2598		hpsa_unmap_ioaccel2_sg_chain_block(h, c2);
2599
2600	cmd->result = (DID_OK << 16); 		/* host byte */
2601	cmd->result |= (COMMAND_COMPLETE << 8);	/* msg byte */
2602
2603	/* SCSI command has already been cleaned up in SML */
2604	if (dev->was_removed) {
2605		hpsa_cmd_resolve_and_free(h, cp);
2606		return;
2607	}
2608
2609	if (cp->cmd_type == CMD_IOACCEL2 || cp->cmd_type == CMD_IOACCEL1) {
2610		if (dev->physical_device && dev->expose_device &&
2611			dev->removed) {
2612			cmd->result = DID_NO_CONNECT << 16;
2613			return hpsa_cmd_free_and_done(h, cp, cmd);
2614		}
2615		if (likely(cp->phys_disk != NULL))
2616			atomic_dec(&cp->phys_disk->ioaccel_cmds_out);
2617	}
2618
2619	/*
2620	 * We check for lockup status here as it may be set for
2621	 * CMD_SCSI, CMD_IOACCEL1 and CMD_IOACCEL2 commands by
2622	 * fail_all_oustanding_cmds()
2623	 */
2624	if (unlikely(ei->CommandStatus == CMD_CTLR_LOCKUP)) {
2625		/* DID_NO_CONNECT will prevent a retry */
2626		cmd->result = DID_NO_CONNECT << 16;
2627		return hpsa_cmd_free_and_done(h, cp, cmd);
2628	}
2629
2630	if (cp->cmd_type == CMD_IOACCEL2)
2631		return process_ioaccel2_completion(h, cp, cmd, dev);
2632
2633	scsi_set_resid(cmd, ei->ResidualCnt);
2634	if (ei->CommandStatus == 0)
2635		return hpsa_cmd_free_and_done(h, cp, cmd);
2636
2637	/* For I/O accelerator commands, copy over some fields to the normal
2638	 * CISS header used below for error handling.
2639	 */
2640	if (cp->cmd_type == CMD_IOACCEL1) {
2641		struct io_accel1_cmd *c = &h->ioaccel_cmd_pool[cp->cmdindex];
2642		cp->Header.SGList = scsi_sg_count(cmd);
2643		cp->Header.SGTotal = cpu_to_le16(cp->Header.SGList);
2644		cp->Request.CDBLen = le16_to_cpu(c->io_flags) &
2645			IOACCEL1_IOFLAGS_CDBLEN_MASK;
2646		cp->Header.tag = c->tag;
2647		memcpy(cp->Header.LUN.LunAddrBytes, c->CISS_LUN, 8);
2648		memcpy(cp->Request.CDB, c->CDB, cp->Request.CDBLen);
2649
2650		/* Any RAID offload error results in retry which will use
2651		 * the normal I/O path so the controller can handle whatever's
2652		 * wrong.
2653		 */
2654		if (is_logical_device(dev)) {
2655			if (ei->CommandStatus == CMD_IOACCEL_DISABLED)
2656				dev->offload_enabled = 0;
2657			return hpsa_retry_cmd(h, cp);
2658		}
2659	}
2660
2661	/* an error has occurred */
2662	switch (ei->CommandStatus) {
2663
2664	case CMD_TARGET_STATUS:
2665		cmd->result |= ei->ScsiStatus;
2666		/* copy the sense data */
2667		if (SCSI_SENSE_BUFFERSIZE < sizeof(ei->SenseInfo))
2668			sense_data_size = SCSI_SENSE_BUFFERSIZE;
2669		else
2670			sense_data_size = sizeof(ei->SenseInfo);
2671		if (ei->SenseLen < sense_data_size)
2672			sense_data_size = ei->SenseLen;
2673		memcpy(cmd->sense_buffer, ei->SenseInfo, sense_data_size);
2674		if (ei->ScsiStatus)
2675			decode_sense_data(ei->SenseInfo, sense_data_size,
2676				&sense_key, &asc, &ascq);
2677		if (ei->ScsiStatus == SAM_STAT_CHECK_CONDITION) {
2678			switch (sense_key) {
2679			case ABORTED_COMMAND:
2680				cmd->result |= DID_SOFT_ERROR << 16;
2681				break;
2682			case UNIT_ATTENTION:
2683				if (asc == 0x3F && ascq == 0x0E)
2684					h->drv_req_rescan = 1;
2685				break;
2686			case ILLEGAL_REQUEST:
2687				if (asc == 0x25 && ascq == 0x00) {
2688					dev->removed = 1;
2689					cmd->result = DID_NO_CONNECT << 16;
2690				}
2691				break;
2692			}
2693			break;
2694		}
2695		/* Problem was not a check condition
2696		 * Pass it up to the upper layers...
2697		 */
2698		if (ei->ScsiStatus) {
2699			dev_warn(&h->pdev->dev, "cp %p has status 0x%x "
2700				"Sense: 0x%x, ASC: 0x%x, ASCQ: 0x%x, "
2701				"Returning result: 0x%x\n",
2702				cp, ei->ScsiStatus,
2703				sense_key, asc, ascq,
2704				cmd->result);
2705		} else {  /* scsi status is zero??? How??? */
2706			dev_warn(&h->pdev->dev, "cp %p SCSI status was 0. "
2707				"Returning no connection.\n", cp),
2708
2709			/* Ordinarily, this case should never happen,
2710			 * but there is a bug in some released firmware
2711			 * revisions that allows it to happen if, for
2712			 * example, a 4100 backplane loses power and
2713			 * the tape drive is in it.  We assume that
2714			 * it's a fatal error of some kind because we
2715			 * can't show that it wasn't. We will make it
2716			 * look like selection timeout since that is
2717			 * the most common reason for this to occur,
2718			 * and it's severe enough.
2719			 */
2720
2721			cmd->result = DID_NO_CONNECT << 16;
2722		}
2723		break;
2724
2725	case CMD_DATA_UNDERRUN: /* let mid layer handle it. */
2726		break;
2727	case CMD_DATA_OVERRUN:
2728		dev_warn(&h->pdev->dev,
2729			"CDB %16phN data overrun\n", cp->Request.CDB);
2730		break;
2731	case CMD_INVALID: {
2732		/* print_bytes(cp, sizeof(*cp), 1, 0);
2733		print_cmd(cp); */
2734		/* We get CMD_INVALID if you address a non-existent device
2735		 * instead of a selection timeout (no response).  You will
2736		 * see this if you yank out a drive, then try to access it.
2737		 * This is kind of a shame because it means that any other
2738		 * CMD_INVALID (e.g. driver bug) will get interpreted as a
2739		 * missing target. */
2740		cmd->result = DID_NO_CONNECT << 16;
2741	}
2742		break;
2743	case CMD_PROTOCOL_ERR:
2744		cmd->result = DID_ERROR << 16;
2745		dev_warn(&h->pdev->dev, "CDB %16phN : protocol error\n",
2746				cp->Request.CDB);
2747		break;
2748	case CMD_HARDWARE_ERR:
2749		cmd->result = DID_ERROR << 16;
2750		dev_warn(&h->pdev->dev, "CDB %16phN : hardware error\n",
2751			cp->Request.CDB);
2752		break;
2753	case CMD_CONNECTION_LOST:
2754		cmd->result = DID_ERROR << 16;
2755		dev_warn(&h->pdev->dev, "CDB %16phN : connection lost\n",
2756			cp->Request.CDB);
2757		break;
2758	case CMD_ABORTED:
2759		cmd->result = DID_ABORT << 16;
2760		break;
2761	case CMD_ABORT_FAILED:
2762		cmd->result = DID_ERROR << 16;
2763		dev_warn(&h->pdev->dev, "CDB %16phN : abort failed\n",
2764			cp->Request.CDB);
2765		break;
2766	case CMD_UNSOLICITED_ABORT:
2767		cmd->result = DID_SOFT_ERROR << 16; /* retry the command */
2768		dev_warn(&h->pdev->dev, "CDB %16phN : unsolicited abort\n",
2769			cp->Request.CDB);
2770		break;
2771	case CMD_TIMEOUT:
2772		cmd->result = DID_TIME_OUT << 16;
2773		dev_warn(&h->pdev->dev, "CDB %16phN timed out\n",
2774			cp->Request.CDB);
2775		break;
2776	case CMD_UNABORTABLE:
2777		cmd->result = DID_ERROR << 16;
2778		dev_warn(&h->pdev->dev, "Command unabortable\n");
2779		break;
2780	case CMD_TMF_STATUS:
2781		if (hpsa_evaluate_tmf_status(h, cp)) /* TMF failed? */
2782			cmd->result = DID_ERROR << 16;
2783		break;
2784	case CMD_IOACCEL_DISABLED:
2785		/* This only handles the direct pass-through case since RAID
2786		 * offload is handled above.  Just attempt a retry.
2787		 */
2788		cmd->result = DID_SOFT_ERROR << 16;
2789		dev_warn(&h->pdev->dev,
2790				"cp %p had HP SSD Smart Path error\n", cp);
2791		break;
2792	default:
2793		cmd->result = DID_ERROR << 16;
2794		dev_warn(&h->pdev->dev, "cp %p returned unknown status %x\n",
2795				cp, ei->CommandStatus);
2796	}
2797
2798	return hpsa_cmd_free_and_done(h, cp, cmd);
2799}
2800
2801static void hpsa_pci_unmap(struct pci_dev *pdev, struct CommandList *c,
2802		int sg_used, enum dma_data_direction data_direction)
2803{
2804	int i;
2805
2806	for (i = 0; i < sg_used; i++)
2807		dma_unmap_single(&pdev->dev, le64_to_cpu(c->SG[i].Addr),
2808				le32_to_cpu(c->SG[i].Len),
2809				data_direction);
2810}
2811
2812static int hpsa_map_one(struct pci_dev *pdev,
2813		struct CommandList *cp,
2814		unsigned char *buf,
2815		size_t buflen,
2816		enum dma_data_direction data_direction)
2817{
2818	u64 addr64;
2819
2820	if (buflen == 0 || data_direction == DMA_NONE) {
2821		cp->Header.SGList = 0;
2822		cp->Header.SGTotal = cpu_to_le16(0);
2823		return 0;
2824	}
2825
2826	addr64 = dma_map_single(&pdev->dev, buf, buflen, data_direction);
2827	if (dma_mapping_error(&pdev->dev, addr64)) {
2828		/* Prevent subsequent unmap of something never mapped */
2829		cp->Header.SGList = 0;
2830		cp->Header.SGTotal = cpu_to_le16(0);
2831		return -1;
2832	}
2833	cp->SG[0].Addr = cpu_to_le64(addr64);
2834	cp->SG[0].Len = cpu_to_le32(buflen);
2835	cp->SG[0].Ext = cpu_to_le32(HPSA_SG_LAST); /* we are not chaining */
2836	cp->Header.SGList = 1;   /* no. SGs contig in this cmd */
2837	cp->Header.SGTotal = cpu_to_le16(1); /* total sgs in cmd list */
2838	return 0;
2839}
2840
2841#define NO_TIMEOUT ((unsigned long) -1)
2842#define DEFAULT_TIMEOUT 30000 /* milliseconds */
2843static int hpsa_scsi_do_simple_cmd_core(struct ctlr_info *h,
2844	struct CommandList *c, int reply_queue, unsigned long timeout_msecs)
2845{
2846	DECLARE_COMPLETION_ONSTACK(wait);
2847
2848	c->waiting = &wait;
2849	__enqueue_cmd_and_start_io(h, c, reply_queue);
2850	if (timeout_msecs == NO_TIMEOUT) {
2851		/* TODO: get rid of this no-timeout thing */
2852		wait_for_completion_io(&wait);
2853		return IO_OK;
2854	}
2855	if (!wait_for_completion_io_timeout(&wait,
2856					msecs_to_jiffies(timeout_msecs))) {
2857		dev_warn(&h->pdev->dev, "Command timed out.\n");
2858		return -ETIMEDOUT;
2859	}
2860	return IO_OK;
2861}
2862
2863static int hpsa_scsi_do_simple_cmd(struct ctlr_info *h, struct CommandList *c,
2864				   int reply_queue, unsigned long timeout_msecs)
2865{
2866	if (unlikely(lockup_detected(h))) {
2867		c->err_info->CommandStatus = CMD_CTLR_LOCKUP;
2868		return IO_OK;
2869	}
2870	return hpsa_scsi_do_simple_cmd_core(h, c, reply_queue, timeout_msecs);
2871}
2872
2873static u32 lockup_detected(struct ctlr_info *h)
2874{
2875	int cpu;
2876	u32 rc, *lockup_detected;
2877
2878	cpu = get_cpu();
2879	lockup_detected = per_cpu_ptr(h->lockup_detected, cpu);
2880	rc = *lockup_detected;
2881	put_cpu();
2882	return rc;
2883}
2884
2885#define MAX_DRIVER_CMD_RETRIES 25
2886static int hpsa_scsi_do_simple_cmd_with_retry(struct ctlr_info *h,
2887		struct CommandList *c, enum dma_data_direction data_direction,
2888		unsigned long timeout_msecs)
2889{
2890	int backoff_time = 10, retry_count = 0;
2891	int rc;
2892
2893	do {
2894		memset(c->err_info, 0, sizeof(*c->err_info));
2895		rc = hpsa_scsi_do_simple_cmd(h, c, DEFAULT_REPLY_QUEUE,
2896						  timeout_msecs);
2897		if (rc)
2898			break;
2899		retry_count++;
2900		if (retry_count > 3) {
2901			msleep(backoff_time);
2902			if (backoff_time < 1000)
2903				backoff_time *= 2;
2904		}
2905	} while ((check_for_unit_attention(h, c) ||
2906			check_for_busy(h, c)) &&
2907			retry_count <= MAX_DRIVER_CMD_RETRIES);
2908	hpsa_pci_unmap(h->pdev, c, 1, data_direction);
2909	if (retry_count > MAX_DRIVER_CMD_RETRIES)
2910		rc = -EIO;
2911	return rc;
2912}
2913
2914static void hpsa_print_cmd(struct ctlr_info *h, char *txt,
2915				struct CommandList *c)
2916{
2917	const u8 *cdb = c->Request.CDB;
2918	const u8 *lun = c->Header.LUN.LunAddrBytes;
2919
2920	dev_warn(&h->pdev->dev, "%s: LUN:%8phN CDB:%16phN\n",
2921		 txt, lun, cdb);
2922}
2923
2924static void hpsa_scsi_interpret_error(struct ctlr_info *h,
2925			struct CommandList *cp)
2926{
2927	const struct ErrorInfo *ei = cp->err_info;
2928	struct device *d = &cp->h->pdev->dev;
2929	u8 sense_key, asc, ascq;
2930	int sense_len;
2931
2932	switch (ei->CommandStatus) {
2933	case CMD_TARGET_STATUS:
2934		if (ei->SenseLen > sizeof(ei->SenseInfo))
2935			sense_len = sizeof(ei->SenseInfo);
2936		else
2937			sense_len = ei->SenseLen;
2938		decode_sense_data(ei->SenseInfo, sense_len,
2939					&sense_key, &asc, &ascq);
2940		hpsa_print_cmd(h, "SCSI status", cp);
2941		if (ei->ScsiStatus == SAM_STAT_CHECK_CONDITION)
2942			dev_warn(d, "SCSI Status = 02, Sense key = 0x%02x, ASC = 0x%02x, ASCQ = 0x%02x\n",
2943				sense_key, asc, ascq);
2944		else
2945			dev_warn(d, "SCSI Status = 0x%02x\n", ei->ScsiStatus);
2946		if (ei->ScsiStatus == 0)
2947			dev_warn(d, "SCSI status is abnormally zero.  "
2948			"(probably indicates selection timeout "
2949			"reported incorrectly due to a known "
2950			"firmware bug, circa July, 2001.)\n");
2951		break;
2952	case CMD_DATA_UNDERRUN: /* let mid layer handle it. */
2953		break;
2954	case CMD_DATA_OVERRUN:
2955		hpsa_print_cmd(h, "overrun condition", cp);
2956		break;
2957	case CMD_INVALID: {
2958		/* controller unfortunately reports SCSI passthru's
2959		 * to non-existent targets as invalid commands.
2960		 */
2961		hpsa_print_cmd(h, "invalid command", cp);
2962		dev_warn(d, "probably means device no longer present\n");
2963		}
2964		break;
2965	case CMD_PROTOCOL_ERR:
2966		hpsa_print_cmd(h, "protocol error", cp);
2967		break;
2968	case CMD_HARDWARE_ERR:
2969		hpsa_print_cmd(h, "hardware error", cp);
2970		break;
2971	case CMD_CONNECTION_LOST:
2972		hpsa_print_cmd(h, "connection lost", cp);
2973		break;
2974	case CMD_ABORTED:
2975		hpsa_print_cmd(h, "aborted", cp);
2976		break;
2977	case CMD_ABORT_FAILED:
2978		hpsa_print_cmd(h, "abort failed", cp);
2979		break;
2980	case CMD_UNSOLICITED_ABORT:
2981		hpsa_print_cmd(h, "unsolicited abort", cp);
2982		break;
2983	case CMD_TIMEOUT:
2984		hpsa_print_cmd(h, "timed out", cp);
2985		break;
2986	case CMD_UNABORTABLE:
2987		hpsa_print_cmd(h, "unabortable", cp);
2988		break;
2989	case CMD_CTLR_LOCKUP:
2990		hpsa_print_cmd(h, "controller lockup detected", cp);
2991		break;
2992	default:
2993		hpsa_print_cmd(h, "unknown status", cp);
2994		dev_warn(d, "Unknown command status %x\n",
2995				ei->CommandStatus);
2996	}
2997}
2998
2999static int hpsa_do_receive_diagnostic(struct ctlr_info *h, u8 *scsi3addr,
3000					u8 page, u8 *buf, size_t bufsize)
3001{
3002	int rc = IO_OK;
3003	struct CommandList *c;
3004	struct ErrorInfo *ei;
3005
3006	c = cmd_alloc(h);
3007	if (fill_cmd(c, RECEIVE_DIAGNOSTIC, h, buf, bufsize,
3008			page, scsi3addr, TYPE_CMD)) {
3009		rc = -1;
3010		goto out;
3011	}
3012	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3013			NO_TIMEOUT);
3014	if (rc)
3015		goto out;
3016	ei = c->err_info;
3017	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
3018		hpsa_scsi_interpret_error(h, c);
3019		rc = -1;
3020	}
3021out:
3022	cmd_free(h, c);
3023	return rc;
3024}
3025
3026static u64 hpsa_get_enclosure_logical_identifier(struct ctlr_info *h,
3027						u8 *scsi3addr)
3028{
3029	u8 *buf;
3030	u64 sa = 0;
3031	int rc = 0;
3032
3033	buf = kzalloc(1024, GFP_KERNEL);
3034	if (!buf)
3035		return 0;
3036
3037	rc = hpsa_do_receive_diagnostic(h, scsi3addr, RECEIVE_DIAGNOSTIC,
3038					buf, 1024);
3039
3040	if (rc)
3041		goto out;
3042
3043	sa = get_unaligned_be64(buf+12);
3044
3045out:
3046	kfree(buf);
3047	return sa;
3048}
3049
3050static int hpsa_scsi_do_inquiry(struct ctlr_info *h, unsigned char *scsi3addr,
3051			u16 page, unsigned char *buf,
3052			unsigned char bufsize)
3053{
3054	int rc = IO_OK;
3055	struct CommandList *c;
3056	struct ErrorInfo *ei;
3057
3058	c = cmd_alloc(h);
3059
3060	if (fill_cmd(c, HPSA_INQUIRY, h, buf, bufsize,
3061			page, scsi3addr, TYPE_CMD)) {
3062		rc = -1;
3063		goto out;
3064	}
3065	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3066			NO_TIMEOUT);
3067	if (rc)
3068		goto out;
3069	ei = c->err_info;
3070	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
3071		hpsa_scsi_interpret_error(h, c);
3072		rc = -1;
3073	}
3074out:
3075	cmd_free(h, c);
3076	return rc;
3077}
3078
3079static int hpsa_send_reset(struct ctlr_info *h, struct hpsa_scsi_dev_t *dev,
3080	u8 reset_type, int reply_queue)
3081{
3082	int rc = IO_OK;
3083	struct CommandList *c;
3084	struct ErrorInfo *ei;
3085
3086	c = cmd_alloc(h);
3087	c->device = dev;
3088
3089	/* fill_cmd can't fail here, no data buffer to map. */
3090	(void) fill_cmd(c, reset_type, h, NULL, 0, 0, dev->scsi3addr, TYPE_MSG);
3091	rc = hpsa_scsi_do_simple_cmd(h, c, reply_queue, NO_TIMEOUT);
3092	if (rc) {
3093		dev_warn(&h->pdev->dev, "Failed to send reset command\n");
3094		goto out;
3095	}
3096	/* no unmap needed here because no data xfer. */
3097
3098	ei = c->err_info;
3099	if (ei->CommandStatus != 0) {
3100		hpsa_scsi_interpret_error(h, c);
3101		rc = -1;
3102	}
3103out:
3104	cmd_free(h, c);
3105	return rc;
3106}
3107
3108static bool hpsa_cmd_dev_match(struct ctlr_info *h, struct CommandList *c,
3109			       struct hpsa_scsi_dev_t *dev,
3110			       unsigned char *scsi3addr)
3111{
3112	int i;
3113	bool match = false;
3114	struct io_accel2_cmd *c2 = &h->ioaccel2_cmd_pool[c->cmdindex];
3115	struct hpsa_tmf_struct *ac = (struct hpsa_tmf_struct *) c2;
3116
3117	if (hpsa_is_cmd_idle(c))
3118		return false;
3119
3120	switch (c->cmd_type) {
3121	case CMD_SCSI:
3122	case CMD_IOCTL_PEND:
3123		match = !memcmp(scsi3addr, &c->Header.LUN.LunAddrBytes,
3124				sizeof(c->Header.LUN.LunAddrBytes));
3125		break;
3126
3127	case CMD_IOACCEL1:
3128	case CMD_IOACCEL2:
3129		if (c->phys_disk == dev) {
3130			/* HBA mode match */
3131			match = true;
3132		} else {
3133			/* Possible RAID mode -- check each phys dev. */
3134			/* FIXME:  Do we need to take out a lock here?  If
3135			 * so, we could just call hpsa_get_pdisk_of_ioaccel2()
3136			 * instead. */
3137			for (i = 0; i < dev->nphysical_disks && !match; i++) {
3138				/* FIXME: an alternate test might be
3139				 *
3140				 * match = dev->phys_disk[i]->ioaccel_handle
3141				 *              == c2->scsi_nexus;      */
3142				match = dev->phys_disk[i] == c->phys_disk;
3143			}
3144		}
3145		break;
3146
3147	case IOACCEL2_TMF:
3148		for (i = 0; i < dev->nphysical_disks && !match; i++) {
3149			match = dev->phys_disk[i]->ioaccel_handle ==
3150					le32_to_cpu(ac->it_nexus);
3151		}
3152		break;
3153
3154	case 0:		/* The command is in the middle of being initialized. */
3155		match = false;
3156		break;
3157
3158	default:
3159		dev_err(&h->pdev->dev, "unexpected cmd_type: %d\n",
3160			c->cmd_type);
3161		BUG();
3162	}
3163
3164	return match;
3165}
3166
3167static int hpsa_do_reset(struct ctlr_info *h, struct hpsa_scsi_dev_t *dev,
3168	u8 reset_type, int reply_queue)
3169{
3170	int rc = 0;
3171
3172	/* We can really only handle one reset at a time */
3173	if (mutex_lock_interruptible(&h->reset_mutex) == -EINTR) {
3174		dev_warn(&h->pdev->dev, "concurrent reset wait interrupted.\n");
3175		return -EINTR;
3176	}
3177
3178	rc = hpsa_send_reset(h, dev, reset_type, reply_queue);
3179	if (!rc) {
3180		/* incremented by sending the reset request */
3181		atomic_dec(&dev->commands_outstanding);
3182		wait_event(h->event_sync_wait_queue,
3183			atomic_read(&dev->commands_outstanding) <= 0 ||
3184			lockup_detected(h));
3185	}
3186
3187	if (unlikely(lockup_detected(h))) {
3188		dev_warn(&h->pdev->dev,
3189			 "Controller lockup detected during reset wait\n");
3190		rc = -ENODEV;
3191	}
3192
3193	if (!rc)
3194		rc = wait_for_device_to_become_ready(h, dev->scsi3addr, 0);
3195
3196	mutex_unlock(&h->reset_mutex);
3197	return rc;
3198}
3199
3200static void hpsa_get_raid_level(struct ctlr_info *h,
3201	unsigned char *scsi3addr, unsigned char *raid_level)
3202{
3203	int rc;
3204	unsigned char *buf;
3205
3206	*raid_level = RAID_UNKNOWN;
3207	buf = kzalloc(64, GFP_KERNEL);
3208	if (!buf)
3209		return;
3210
3211	if (!hpsa_vpd_page_supported(h, scsi3addr,
3212		HPSA_VPD_LV_DEVICE_GEOMETRY))
3213		goto exit;
3214
3215	rc = hpsa_scsi_do_inquiry(h, scsi3addr, VPD_PAGE |
3216		HPSA_VPD_LV_DEVICE_GEOMETRY, buf, 64);
3217
3218	if (rc == 0)
3219		*raid_level = buf[8];
3220	if (*raid_level > RAID_UNKNOWN)
3221		*raid_level = RAID_UNKNOWN;
3222exit:
3223	kfree(buf);
3224	return;
3225}
3226
3227#define HPSA_MAP_DEBUG
3228#ifdef HPSA_MAP_DEBUG
3229static void hpsa_debug_map_buff(struct ctlr_info *h, int rc,
3230				struct raid_map_data *map_buff)
3231{
3232	struct raid_map_disk_data *dd = &map_buff->data[0];
3233	int map, row, col;
3234	u16 map_cnt, row_cnt, disks_per_row;
3235
3236	if (rc != 0)
3237		return;
3238
3239	/* Show details only if debugging has been activated. */
3240	if (h->raid_offload_debug < 2)
3241		return;
3242
3243	dev_info(&h->pdev->dev, "structure_size = %u\n",
3244				le32_to_cpu(map_buff->structure_size));
3245	dev_info(&h->pdev->dev, "volume_blk_size = %u\n",
3246			le32_to_cpu(map_buff->volume_blk_size));
3247	dev_info(&h->pdev->dev, "volume_blk_cnt = 0x%llx\n",
3248			le64_to_cpu(map_buff->volume_blk_cnt));
3249	dev_info(&h->pdev->dev, "physicalBlockShift = %u\n",
3250			map_buff->phys_blk_shift);
3251	dev_info(&h->pdev->dev, "parity_rotation_shift = %u\n",
3252			map_buff->parity_rotation_shift);
3253	dev_info(&h->pdev->dev, "strip_size = %u\n",
3254			le16_to_cpu(map_buff->strip_size));
3255	dev_info(&h->pdev->dev, "disk_starting_blk = 0x%llx\n",
3256			le64_to_cpu(map_buff->disk_starting_blk));
3257	dev_info(&h->pdev->dev, "disk_blk_cnt = 0x%llx\n",
3258			le64_to_cpu(map_buff->disk_blk_cnt));
3259	dev_info(&h->pdev->dev, "data_disks_per_row = %u\n",
3260			le16_to_cpu(map_buff->data_disks_per_row));
3261	dev_info(&h->pdev->dev, "metadata_disks_per_row = %u\n",
3262			le16_to_cpu(map_buff->metadata_disks_per_row));
3263	dev_info(&h->pdev->dev, "row_cnt = %u\n",
3264			le16_to_cpu(map_buff->row_cnt));
3265	dev_info(&h->pdev->dev, "layout_map_count = %u\n",
3266			le16_to_cpu(map_buff->layout_map_count));
3267	dev_info(&h->pdev->dev, "flags = 0x%x\n",
3268			le16_to_cpu(map_buff->flags));
3269	dev_info(&h->pdev->dev, "encryption = %s\n",
3270			le16_to_cpu(map_buff->flags) &
3271			RAID_MAP_FLAG_ENCRYPT_ON ?  "ON" : "OFF");
3272	dev_info(&h->pdev->dev, "dekindex = %u\n",
3273			le16_to_cpu(map_buff->dekindex));
3274	map_cnt = le16_to_cpu(map_buff->layout_map_count);
3275	for (map = 0; map < map_cnt; map++) {
3276		dev_info(&h->pdev->dev, "Map%u:\n", map);
3277		row_cnt = le16_to_cpu(map_buff->row_cnt);
3278		for (row = 0; row < row_cnt; row++) {
3279			dev_info(&h->pdev->dev, "  Row%u:\n", row);
3280			disks_per_row =
3281				le16_to_cpu(map_buff->data_disks_per_row);
3282			for (col = 0; col < disks_per_row; col++, dd++)
3283				dev_info(&h->pdev->dev,
3284					"    D%02u: h=0x%04x xor=%u,%u\n",
3285					col, dd->ioaccel_handle,
3286					dd->xor_mult[0], dd->xor_mult[1]);
3287			disks_per_row =
3288				le16_to_cpu(map_buff->metadata_disks_per_row);
3289			for (col = 0; col < disks_per_row; col++, dd++)
3290				dev_info(&h->pdev->dev,
3291					"    M%02u: h=0x%04x xor=%u,%u\n",
3292					col, dd->ioaccel_handle,
3293					dd->xor_mult[0], dd->xor_mult[1]);
3294		}
3295	}
3296}
3297#else
3298static void hpsa_debug_map_buff(__attribute__((unused)) struct ctlr_info *h,
3299			__attribute__((unused)) int rc,
3300			__attribute__((unused)) struct raid_map_data *map_buff)
3301{
3302}
3303#endif
3304
3305static int hpsa_get_raid_map(struct ctlr_info *h,
3306	unsigned char *scsi3addr, struct hpsa_scsi_dev_t *this_device)
3307{
3308	int rc = 0;
3309	struct CommandList *c;
3310	struct ErrorInfo *ei;
3311
3312	c = cmd_alloc(h);
3313
3314	if (fill_cmd(c, HPSA_GET_RAID_MAP, h, &this_device->raid_map,
3315			sizeof(this_device->raid_map), 0,
3316			scsi3addr, TYPE_CMD)) {
3317		dev_warn(&h->pdev->dev, "hpsa_get_raid_map fill_cmd failed\n");
3318		cmd_free(h, c);
3319		return -1;
3320	}
3321	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3322			NO_TIMEOUT);
3323	if (rc)
3324		goto out;
3325	ei = c->err_info;
3326	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
3327		hpsa_scsi_interpret_error(h, c);
3328		rc = -1;
3329		goto out;
3330	}
3331	cmd_free(h, c);
3332
3333	/* @todo in the future, dynamically allocate RAID map memory */
3334	if (le32_to_cpu(this_device->raid_map.structure_size) >
3335				sizeof(this_device->raid_map)) {
3336		dev_warn(&h->pdev->dev, "RAID map size is too large!\n");
3337		rc = -1;
3338	}
3339	hpsa_debug_map_buff(h, rc, &this_device->raid_map);
3340	return rc;
3341out:
3342	cmd_free(h, c);
3343	return rc;
3344}
3345
3346static int hpsa_bmic_sense_subsystem_information(struct ctlr_info *h,
3347		unsigned char scsi3addr[], u16 bmic_device_index,
3348		struct bmic_sense_subsystem_info *buf, size_t bufsize)
3349{
3350	int rc = IO_OK;
3351	struct CommandList *c;
3352	struct ErrorInfo *ei;
3353
3354	c = cmd_alloc(h);
3355
3356	rc = fill_cmd(c, BMIC_SENSE_SUBSYSTEM_INFORMATION, h, buf, bufsize,
3357		0, RAID_CTLR_LUNID, TYPE_CMD);
3358	if (rc)
3359		goto out;
3360
3361	c->Request.CDB[2] = bmic_device_index & 0xff;
3362	c->Request.CDB[9] = (bmic_device_index >> 8) & 0xff;
3363
3364	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3365			NO_TIMEOUT);
3366	if (rc)
3367		goto out;
3368	ei = c->err_info;
3369	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
3370		hpsa_scsi_interpret_error(h, c);
3371		rc = -1;
3372	}
3373out:
3374	cmd_free(h, c);
3375	return rc;
3376}
3377
3378static int hpsa_bmic_id_controller(struct ctlr_info *h,
3379	struct bmic_identify_controller *buf, size_t bufsize)
3380{
3381	int rc = IO_OK;
3382	struct CommandList *c;
3383	struct ErrorInfo *ei;
3384
3385	c = cmd_alloc(h);
3386
3387	rc = fill_cmd(c, BMIC_IDENTIFY_CONTROLLER, h, buf, bufsize,
3388		0, RAID_CTLR_LUNID, TYPE_CMD);
3389	if (rc)
3390		goto out;
3391
3392	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3393			NO_TIMEOUT);
3394	if (rc)
3395		goto out;
3396	ei = c->err_info;
3397	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
3398		hpsa_scsi_interpret_error(h, c);
3399		rc = -1;
3400	}
3401out:
3402	cmd_free(h, c);
3403	return rc;
3404}
3405
3406static int hpsa_bmic_id_physical_device(struct ctlr_info *h,
3407		unsigned char scsi3addr[], u16 bmic_device_index,
3408		struct bmic_identify_physical_device *buf, size_t bufsize)
3409{
3410	int rc = IO_OK;
3411	struct CommandList *c;
3412	struct ErrorInfo *ei;
3413
3414	c = cmd_alloc(h);
3415	rc = fill_cmd(c, BMIC_IDENTIFY_PHYSICAL_DEVICE, h, buf, bufsize,
3416		0, RAID_CTLR_LUNID, TYPE_CMD);
3417	if (rc)
3418		goto out;
3419
3420	c->Request.CDB[2] = bmic_device_index & 0xff;
3421	c->Request.CDB[9] = (bmic_device_index >> 8) & 0xff;
3422
3423	hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3424						NO_TIMEOUT);
3425	ei = c->err_info;
3426	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
3427		hpsa_scsi_interpret_error(h, c);
3428		rc = -1;
3429	}
3430out:
3431	cmd_free(h, c);
3432
3433	return rc;
3434}
3435
3436/*
3437 * get enclosure information
3438 * struct ReportExtendedLUNdata *rlep - Used for BMIC drive number
3439 * struct hpsa_scsi_dev_t *encl_dev - device entry for enclosure
3440 * Uses id_physical_device to determine the box_index.
3441 */
3442static void hpsa_get_enclosure_info(struct ctlr_info *h,
3443			unsigned char *scsi3addr,
3444			struct ReportExtendedLUNdata *rlep, int rle_index,
3445			struct hpsa_scsi_dev_t *encl_dev)
3446{
3447	int rc = -1;
3448	struct CommandList *c = NULL;
3449	struct ErrorInfo *ei = NULL;
3450	struct bmic_sense_storage_box_params *bssbp = NULL;
3451	struct bmic_identify_physical_device *id_phys = NULL;
3452	struct ext_report_lun_entry *rle;
3453	u16 bmic_device_index = 0;
3454
3455	if (rle_index < 0 || rle_index >= HPSA_MAX_PHYS_LUN)
3456		return;
3457
3458	rle = &rlep->LUN[rle_index];
3459
3460	encl_dev->eli =
3461		hpsa_get_enclosure_logical_identifier(h, scsi3addr);
3462
3463	bmic_device_index = GET_BMIC_DRIVE_NUMBER(&rle->lunid[0]);
3464
3465	if (encl_dev->target == -1 || encl_dev->lun == -1) {
3466		rc = IO_OK;
3467		goto out;
3468	}
3469
3470	if (bmic_device_index == 0xFF00 || MASKED_DEVICE(&rle->lunid[0])) {
3471		rc = IO_OK;
3472		goto out;
3473	}
3474
3475	bssbp = kzalloc(sizeof(*bssbp), GFP_KERNEL);
3476	if (!bssbp)
3477		goto out;
3478
3479	id_phys = kzalloc(sizeof(*id_phys), GFP_KERNEL);
3480	if (!id_phys)
3481		goto out;
3482
3483	rc = hpsa_bmic_id_physical_device(h, scsi3addr, bmic_device_index,
3484						id_phys, sizeof(*id_phys));
3485	if (rc) {
3486		dev_warn(&h->pdev->dev, "%s: id_phys failed %d bdi[0x%x]\n",
3487			__func__, encl_dev->external, bmic_device_index);
3488		goto out;
3489	}
3490
3491	c = cmd_alloc(h);
3492
3493	rc = fill_cmd(c, BMIC_SENSE_STORAGE_BOX_PARAMS, h, bssbp,
3494			sizeof(*bssbp), 0, RAID_CTLR_LUNID, TYPE_CMD);
3495
3496	if (rc)
3497		goto out;
3498
3499	if (id_phys->phys_connector[1] == 'E')
3500		c->Request.CDB[5] = id_phys->box_index;
3501	else
3502		c->Request.CDB[5] = 0;
3503
3504	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3505						NO_TIMEOUT);
3506	if (rc)
3507		goto out;
3508
3509	ei = c->err_info;
3510	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
3511		rc = -1;
3512		goto out;
3513	}
3514
3515	encl_dev->box[id_phys->active_path_number] = bssbp->phys_box_on_port;
3516	memcpy(&encl_dev->phys_connector[id_phys->active_path_number],
3517		bssbp->phys_connector, sizeof(bssbp->phys_connector));
3518
3519	rc = IO_OK;
3520out:
3521	kfree(bssbp);
3522	kfree(id_phys);
3523
3524	if (c)
3525		cmd_free(h, c);
3526
3527	if (rc != IO_OK)
3528		hpsa_show_dev_msg(KERN_INFO, h, encl_dev,
3529			"Error, could not get enclosure information");
3530}
3531
3532static u64 hpsa_get_sas_address_from_report_physical(struct ctlr_info *h,
3533						unsigned char *scsi3addr)
3534{
3535	struct ReportExtendedLUNdata *physdev;
3536	u32 nphysicals;
3537	u64 sa = 0;
3538	int i;
3539
3540	physdev = kzalloc(sizeof(*physdev), GFP_KERNEL);
3541	if (!physdev)
3542		return 0;
3543
3544	if (hpsa_scsi_do_report_phys_luns(h, physdev, sizeof(*physdev))) {
3545		dev_err(&h->pdev->dev, "report physical LUNs failed.\n");
3546		kfree(physdev);
3547		return 0;
3548	}
3549	nphysicals = get_unaligned_be32(physdev->LUNListLength) / 24;
3550
3551	for (i = 0; i < nphysicals; i++)
3552		if (!memcmp(&physdev->LUN[i].lunid[0], scsi3addr, 8)) {
3553			sa = get_unaligned_be64(&physdev->LUN[i].wwid[0]);
3554			break;
3555		}
3556
3557	kfree(physdev);
3558
3559	return sa;
3560}
3561
3562static void hpsa_get_sas_address(struct ctlr_info *h, unsigned char *scsi3addr,
3563					struct hpsa_scsi_dev_t *dev)
3564{
3565	int rc;
3566	u64 sa = 0;
3567
3568	if (is_hba_lunid(scsi3addr)) {
3569		struct bmic_sense_subsystem_info *ssi;
3570
3571		ssi = kzalloc(sizeof(*ssi), GFP_KERNEL);
3572		if (!ssi)
3573			return;
3574
3575		rc = hpsa_bmic_sense_subsystem_information(h,
3576					scsi3addr, 0, ssi, sizeof(*ssi));
3577		if (rc == 0) {
3578			sa = get_unaligned_be64(ssi->primary_world_wide_id);
3579			h->sas_address = sa;
3580		}
3581
3582		kfree(ssi);
3583	} else
3584		sa = hpsa_get_sas_address_from_report_physical(h, scsi3addr);
3585
3586	dev->sas_address = sa;
3587}
3588
3589static void hpsa_ext_ctrl_present(struct ctlr_info *h,
3590	struct ReportExtendedLUNdata *physdev)
3591{
3592	u32 nphysicals;
3593	int i;
3594
3595	if (h->discovery_polling)
3596		return;
3597
3598	nphysicals = (get_unaligned_be32(physdev->LUNListLength) / 24) + 1;
3599
3600	for (i = 0; i < nphysicals; i++) {
3601		if (physdev->LUN[i].device_type ==
3602			BMIC_DEVICE_TYPE_CONTROLLER
3603			&& !is_hba_lunid(physdev->LUN[i].lunid)) {
3604			dev_info(&h->pdev->dev,
3605				"External controller present, activate discovery polling and disable rld caching\n");
3606			hpsa_disable_rld_caching(h);
3607			h->discovery_polling = 1;
3608			break;
3609		}
3610	}
3611}
3612
3613/* Get a device id from inquiry page 0x83 */
3614static bool hpsa_vpd_page_supported(struct ctlr_info *h,
3615	unsigned char scsi3addr[], u8 page)
3616{
3617	int rc;
3618	int i;
3619	int pages;
3620	unsigned char *buf, bufsize;
3621
3622	buf = kzalloc(256, GFP_KERNEL);
3623	if (!buf)
3624		return false;
3625
3626	/* Get the size of the page list first */
3627	rc = hpsa_scsi_do_inquiry(h, scsi3addr,
3628				VPD_PAGE | HPSA_VPD_SUPPORTED_PAGES,
3629				buf, HPSA_VPD_HEADER_SZ);
3630	if (rc != 0)
3631		goto exit_unsupported;
3632	pages = buf[3];
3633	if ((pages + HPSA_VPD_HEADER_SZ) <= 255)
3634		bufsize = pages + HPSA_VPD_HEADER_SZ;
3635	else
3636		bufsize = 255;
3637
3638	/* Get the whole VPD page list */
3639	rc = hpsa_scsi_do_inquiry(h, scsi3addr,
3640				VPD_PAGE | HPSA_VPD_SUPPORTED_PAGES,
3641				buf, bufsize);
3642	if (rc != 0)
3643		goto exit_unsupported;
3644
3645	pages = buf[3];
3646	for (i = 1; i <= pages; i++)
3647		if (buf[3 + i] == page)
3648			goto exit_supported;
3649exit_unsupported:
3650	kfree(buf);
3651	return false;
3652exit_supported:
3653	kfree(buf);
3654	return true;
3655}
3656
3657/*
3658 * Called during a scan operation.
3659 * Sets ioaccel status on the new device list, not the existing device list
3660 *
3661 * The device list used during I/O will be updated later in
3662 * adjust_hpsa_scsi_table.
3663 */
3664static void hpsa_get_ioaccel_status(struct ctlr_info *h,
3665	unsigned char *scsi3addr, struct hpsa_scsi_dev_t *this_device)
3666{
3667	int rc;
3668	unsigned char *buf;
3669	u8 ioaccel_status;
3670
3671	this_device->offload_config = 0;
3672	this_device->offload_enabled = 0;
3673	this_device->offload_to_be_enabled = 0;
3674
3675	buf = kzalloc(64, GFP_KERNEL);
3676	if (!buf)
3677		return;
3678	if (!hpsa_vpd_page_supported(h, scsi3addr, HPSA_VPD_LV_IOACCEL_STATUS))
3679		goto out;
3680	rc = hpsa_scsi_do_inquiry(h, scsi3addr,
3681			VPD_PAGE | HPSA_VPD_LV_IOACCEL_STATUS, buf, 64);
3682	if (rc != 0)
3683		goto out;
3684
3685#define IOACCEL_STATUS_BYTE 4
3686#define OFFLOAD_CONFIGURED_BIT 0x01
3687#define OFFLOAD_ENABLED_BIT 0x02
3688	ioaccel_status = buf[IOACCEL_STATUS_BYTE];
3689	this_device->offload_config =
3690		!!(ioaccel_status & OFFLOAD_CONFIGURED_BIT);
3691	if (this_device->offload_config) {
3692		bool offload_enabled =
3693			!!(ioaccel_status & OFFLOAD_ENABLED_BIT);
3694		/*
3695		 * Check to see if offload can be enabled.
3696		 */
3697		if (offload_enabled) {
3698			rc = hpsa_get_raid_map(h, scsi3addr, this_device);
3699			if (rc) /* could not load raid_map */
3700				goto out;
3701			this_device->offload_to_be_enabled = 1;
3702		}
3703	}
3704
3705out:
3706	kfree(buf);
3707	return;
3708}
3709
3710/* Get the device id from inquiry page 0x83 */
3711static int hpsa_get_device_id(struct ctlr_info *h, unsigned char *scsi3addr,
3712	unsigned char *device_id, int index, int buflen)
3713{
3714	int rc;
3715	unsigned char *buf;
3716
3717	/* Does controller have VPD for device id? */
3718	if (!hpsa_vpd_page_supported(h, scsi3addr, HPSA_VPD_LV_DEVICE_ID))
3719		return 1; /* not supported */
3720
3721	buf = kzalloc(64, GFP_KERNEL);
3722	if (!buf)
3723		return -ENOMEM;
3724
3725	rc = hpsa_scsi_do_inquiry(h, scsi3addr, VPD_PAGE |
3726					HPSA_VPD_LV_DEVICE_ID, buf, 64);
3727	if (rc == 0) {
3728		if (buflen > 16)
3729			buflen = 16;
3730		memcpy(device_id, &buf[8], buflen);
3731	}
3732
3733	kfree(buf);
3734
3735	return rc; /*0 - got id,  otherwise, didn't */
3736}
3737
3738static int hpsa_scsi_do_report_luns(struct ctlr_info *h, int logical,
3739		void *buf, int bufsize,
3740		int extended_response)
3741{
3742	int rc = IO_OK;
3743	struct CommandList *c;
3744	unsigned char scsi3addr[8];
3745	struct ErrorInfo *ei;
3746
3747	c = cmd_alloc(h);
3748
3749	/* address the controller */
3750	memset(scsi3addr, 0, sizeof(scsi3addr));
3751	if (fill_cmd(c, logical ? HPSA_REPORT_LOG : HPSA_REPORT_PHYS, h,
3752		buf, bufsize, 0, scsi3addr, TYPE_CMD)) {
3753		rc = -EAGAIN;
3754		goto out;
3755	}
3756	if (extended_response)
3757		c->Request.CDB[1] = extended_response;
3758	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
3759			NO_TIMEOUT);
3760	if (rc)
3761		goto out;
3762	ei = c->err_info;
3763	if (ei->CommandStatus != 0 &&
3764	    ei->CommandStatus != CMD_DATA_UNDERRUN) {
3765		hpsa_scsi_interpret_error(h, c);
3766		rc = -EIO;
3767	} else {
3768		struct ReportLUNdata *rld = buf;
3769
3770		if (rld->extended_response_flag != extended_response) {
3771			if (!h->legacy_board) {
3772				dev_err(&h->pdev->dev,
3773					"report luns requested format %u, got %u\n",
3774					extended_response,
3775					rld->extended_response_flag);
3776				rc = -EINVAL;
3777			} else
3778				rc = -EOPNOTSUPP;
3779		}
3780	}
3781out:
3782	cmd_free(h, c);
3783	return rc;
3784}
3785
3786static inline int hpsa_scsi_do_report_phys_luns(struct ctlr_info *h,
3787		struct ReportExtendedLUNdata *buf, int bufsize)
3788{
3789	int rc;
3790	struct ReportLUNdata *lbuf;
3791
3792	rc = hpsa_scsi_do_report_luns(h, 0, buf, bufsize,
3793				      HPSA_REPORT_PHYS_EXTENDED);
3794	if (!rc || rc != -EOPNOTSUPP)
3795		return rc;
3796
3797	/* REPORT PHYS EXTENDED is not supported */
3798	lbuf = kzalloc(sizeof(*lbuf), GFP_KERNEL);
3799	if (!lbuf)
3800		return -ENOMEM;
3801
3802	rc = hpsa_scsi_do_report_luns(h, 0, lbuf, sizeof(*lbuf), 0);
3803	if (!rc) {
3804		int i;
3805		u32 nphys;
3806
3807		/* Copy ReportLUNdata header */
3808		memcpy(buf, lbuf, 8);
3809		nphys = be32_to_cpu(*((__be32 *)lbuf->LUNListLength)) / 8;
3810		for (i = 0; i < nphys; i++)
3811			memcpy(buf->LUN[i].lunid, lbuf->LUN[i], 8);
3812	}
3813	kfree(lbuf);
3814	return rc;
3815}
3816
3817static inline int hpsa_scsi_do_report_log_luns(struct ctlr_info *h,
3818		struct ReportLUNdata *buf, int bufsize)
3819{
3820	return hpsa_scsi_do_report_luns(h, 1, buf, bufsize, 0);
3821}
3822
3823static inline void hpsa_set_bus_target_lun(struct hpsa_scsi_dev_t *device,
3824	int bus, int target, int lun)
3825{
3826	device->bus = bus;
3827	device->target = target;
3828	device->lun = lun;
3829}
3830
3831/* Use VPD inquiry to get details of volume status */
3832static int hpsa_get_volume_status(struct ctlr_info *h,
3833					unsigned char scsi3addr[])
3834{
3835	int rc;
3836	int status;
3837	int size;
3838	unsigned char *buf;
3839
3840	buf = kzalloc(64, GFP_KERNEL);
3841	if (!buf)
3842		return HPSA_VPD_LV_STATUS_UNSUPPORTED;
3843
3844	/* Does controller have VPD for logical volume status? */
3845	if (!hpsa_vpd_page_supported(h, scsi3addr, HPSA_VPD_LV_STATUS))
3846		goto exit_failed;
3847
3848	/* Get the size of the VPD return buffer */
3849	rc = hpsa_scsi_do_inquiry(h, scsi3addr, VPD_PAGE | HPSA_VPD_LV_STATUS,
3850					buf, HPSA_VPD_HEADER_SZ);
3851	if (rc != 0)
3852		goto exit_failed;
3853	size = buf[3];
3854
3855	/* Now get the whole VPD buffer */
3856	rc = hpsa_scsi_do_inquiry(h, scsi3addr, VPD_PAGE | HPSA_VPD_LV_STATUS,
3857					buf, size + HPSA_VPD_HEADER_SZ);
3858	if (rc != 0)
3859		goto exit_failed;
3860	status = buf[4]; /* status byte */
3861
3862	kfree(buf);
3863	return status;
3864exit_failed:
3865	kfree(buf);
3866	return HPSA_VPD_LV_STATUS_UNSUPPORTED;
3867}
3868
3869/* Determine offline status of a volume.
3870 * Return either:
3871 *  0 (not offline)
3872 *  0xff (offline for unknown reasons)
3873 *  # (integer code indicating one of several NOT READY states
3874 *     describing why a volume is to be kept offline)
3875 */
3876static unsigned char hpsa_volume_offline(struct ctlr_info *h,
3877					unsigned char scsi3addr[])
3878{
3879	struct CommandList *c;
3880	unsigned char *sense;
3881	u8 sense_key, asc, ascq;
3882	int sense_len;
3883	int rc, ldstat = 0;
3884	u16 cmd_status;
3885	u8 scsi_status;
3886#define ASC_LUN_NOT_READY 0x04
3887#define ASCQ_LUN_NOT_READY_FORMAT_IN_PROGRESS 0x04
3888#define ASCQ_LUN_NOT_READY_INITIALIZING_CMD_REQ 0x02
3889
3890	c = cmd_alloc(h);
3891
3892	(void) fill_cmd(c, TEST_UNIT_READY, h, NULL, 0, 0, scsi3addr, TYPE_CMD);
3893	rc = hpsa_scsi_do_simple_cmd(h, c, DEFAULT_REPLY_QUEUE,
3894					NO_TIMEOUT);
3895	if (rc) {
3896		cmd_free(h, c);
3897		return HPSA_VPD_LV_STATUS_UNSUPPORTED;
3898	}
3899	sense = c->err_info->SenseInfo;
3900	if (c->err_info->SenseLen > sizeof(c->err_info->SenseInfo))
3901		sense_len = sizeof(c->err_info->SenseInfo);
3902	else
3903		sense_len = c->err_info->SenseLen;
3904	decode_sense_data(sense, sense_len, &sense_key, &asc, &ascq);
3905	cmd_status = c->err_info->CommandStatus;
3906	scsi_status = c->err_info->ScsiStatus;
3907	cmd_free(h, c);
3908
3909	/* Determine the reason for not ready state */
3910	ldstat = hpsa_get_volume_status(h, scsi3addr);
3911
3912	/* Keep volume offline in certain cases: */
3913	switch (ldstat) {
3914	case HPSA_LV_FAILED:
3915	case HPSA_LV_UNDERGOING_ERASE:
3916	case HPSA_LV_NOT_AVAILABLE:
3917	case HPSA_LV_UNDERGOING_RPI:
3918	case HPSA_LV_PENDING_RPI:
3919	case HPSA_LV_ENCRYPTED_NO_KEY:
3920	case HPSA_LV_PLAINTEXT_IN_ENCRYPT_ONLY_CONTROLLER:
3921	case HPSA_LV_UNDERGOING_ENCRYPTION:
3922	case HPSA_LV_UNDERGOING_ENCRYPTION_REKEYING:
3923	case HPSA_LV_ENCRYPTED_IN_NON_ENCRYPTED_CONTROLLER:
3924		return ldstat;
3925	case HPSA_VPD_LV_STATUS_UNSUPPORTED:
3926		/* If VPD status page isn't available,
3927		 * use ASC/ASCQ to determine state
3928		 */
3929		if ((ascq == ASCQ_LUN_NOT_READY_FORMAT_IN_PROGRESS) ||
3930			(ascq == ASCQ_LUN_NOT_READY_INITIALIZING_CMD_REQ))
3931			return ldstat;
3932		break;
3933	default:
3934		break;
3935	}
3936	return HPSA_LV_OK;
3937}
3938
3939static int hpsa_update_device_info(struct ctlr_info *h,
3940	unsigned char scsi3addr[], struct hpsa_scsi_dev_t *this_device,
3941	unsigned char *is_OBDR_device)
3942{
3943
3944#define OBDR_SIG_OFFSET 43
3945#define OBDR_TAPE_SIG "$DR-10"
3946#define OBDR_SIG_LEN (sizeof(OBDR_TAPE_SIG) - 1)
3947#define OBDR_TAPE_INQ_SIZE (OBDR_SIG_OFFSET + OBDR_SIG_LEN)
3948
3949	unsigned char *inq_buff;
3950	unsigned char *obdr_sig;
3951	int rc = 0;
3952
3953	inq_buff = kzalloc(OBDR_TAPE_INQ_SIZE, GFP_KERNEL);
3954	if (!inq_buff) {
3955		rc = -ENOMEM;
3956		goto bail_out;
3957	}
3958
3959	/* Do an inquiry to the device to see what it is. */
3960	if (hpsa_scsi_do_inquiry(h, scsi3addr, 0, inq_buff,
3961		(unsigned char) OBDR_TAPE_INQ_SIZE) != 0) {
3962		dev_err(&h->pdev->dev,
3963			"%s: inquiry failed, device will be skipped.\n",
3964			__func__);
3965		rc = HPSA_INQUIRY_FAILED;
3966		goto bail_out;
3967	}
3968
3969	scsi_sanitize_inquiry_string(&inq_buff[8], 8);
3970	scsi_sanitize_inquiry_string(&inq_buff[16], 16);
3971
3972	this_device->devtype = (inq_buff[0] & 0x1f);
3973	memcpy(this_device->scsi3addr, scsi3addr, 8);
3974	memcpy(this_device->vendor, &inq_buff[8],
3975		sizeof(this_device->vendor));
3976	memcpy(this_device->model, &inq_buff[16],
3977		sizeof(this_device->model));
3978	this_device->rev = inq_buff[2];
3979	memset(this_device->device_id, 0,
3980		sizeof(this_device->device_id));
3981	if (hpsa_get_device_id(h, scsi3addr, this_device->device_id, 8,
3982		sizeof(this_device->device_id)) < 0) {
3983		dev_err(&h->pdev->dev,
3984			"hpsa%d: %s: can't get device id for [%d:%d:%d:%d]\t%s\t%.16s\n",
3985			h->ctlr, __func__,
3986			h->scsi_host->host_no,
3987			this_device->bus, this_device->target,
3988			this_device->lun,
3989			scsi_device_type(this_device->devtype),
3990			this_device->model);
3991		rc = HPSA_LV_FAILED;
3992		goto bail_out;
3993	}
3994
3995	if ((this_device->devtype == TYPE_DISK ||
3996		this_device->devtype == TYPE_ZBC) &&
3997		is_logical_dev_addr_mode(scsi3addr)) {
3998		unsigned char volume_offline;
3999
4000		hpsa_get_raid_level(h, scsi3addr, &this_device->raid_level);
4001		if (h->fw_support & MISC_FW_RAID_OFFLOAD_BASIC)
4002			hpsa_get_ioaccel_status(h, scsi3addr, this_device);
4003		volume_offline = hpsa_volume_offline(h, scsi3addr);
4004		if (volume_offline == HPSA_VPD_LV_STATUS_UNSUPPORTED &&
4005		    h->legacy_board) {
4006			/*
4007			 * Legacy boards might not support volume status
4008			 */
4009			dev_info(&h->pdev->dev,
4010				 "C0:T%d:L%d Volume status not available, assuming online.\n",
4011				 this_device->target, this_device->lun);
4012			volume_offline = 0;
4013		}
4014		this_device->volume_offline = volume_offline;
4015		if (volume_offline == HPSA_LV_FAILED) {
4016			rc = HPSA_LV_FAILED;
4017			dev_err(&h->pdev->dev,
4018				"%s: LV failed, device will be skipped.\n",
4019				__func__);
4020			goto bail_out;
4021		}
4022	} else {
4023		this_device->raid_level = RAID_UNKNOWN;
4024		this_device->offload_config = 0;
4025		hpsa_turn_off_ioaccel_for_device(this_device);
4026		this_device->hba_ioaccel_enabled = 0;
4027		this_device->volume_offline = 0;
4028		this_device->queue_depth = h->nr_cmds;
4029	}
4030
4031	if (this_device->external)
4032		this_device->queue_depth = EXTERNAL_QD;
4033
4034	if (is_OBDR_device) {
4035		/* See if this is a One-Button-Disaster-Recovery device
4036		 * by looking for "$DR-10" at offset 43 in inquiry data.
4037		 */
4038		obdr_sig = &inq_buff[OBDR_SIG_OFFSET];
4039		*is_OBDR_device = (this_device->devtype == TYPE_ROM &&
4040					strncmp(obdr_sig, OBDR_TAPE_SIG,
4041						OBDR_SIG_LEN) == 0);
4042	}
4043	kfree(inq_buff);
4044	return 0;
4045
4046bail_out:
4047	kfree(inq_buff);
4048	return rc;
4049}
4050
4051/*
4052 * Helper function to assign bus, target, lun mapping of devices.
4053 * Logical drive target and lun are assigned at this time, but
4054 * physical device lun and target assignment are deferred (assigned
4055 * in hpsa_find_target_lun, called by hpsa_scsi_add_entry.)
4056*/
4057static void figure_bus_target_lun(struct ctlr_info *h,
4058	u8 *lunaddrbytes, struct hpsa_scsi_dev_t *device)
4059{
4060	u32 lunid = get_unaligned_le32(lunaddrbytes);
4061
4062	if (!is_logical_dev_addr_mode(lunaddrbytes)) {
4063		/* physical device, target and lun filled in later */
4064		if (is_hba_lunid(lunaddrbytes)) {
4065			int bus = HPSA_HBA_BUS;
4066
4067			if (!device->rev)
4068				bus = HPSA_LEGACY_HBA_BUS;
4069			hpsa_set_bus_target_lun(device,
4070					bus, 0, lunid & 0x3fff);
4071		} else
4072			/* defer target, lun assignment for physical devices */
4073			hpsa_set_bus_target_lun(device,
4074					HPSA_PHYSICAL_DEVICE_BUS, -1, -1);
4075		return;
4076	}
4077	/* It's a logical device */
4078	if (device->external) {
4079		hpsa_set_bus_target_lun(device,
4080			HPSA_EXTERNAL_RAID_VOLUME_BUS, (lunid >> 16) & 0x3fff,
4081			lunid & 0x00ff);
4082		return;
4083	}
4084	hpsa_set_bus_target_lun(device, HPSA_RAID_VOLUME_BUS,
4085				0, lunid & 0x3fff);
4086}
4087
4088static int  figure_external_status(struct ctlr_info *h, int raid_ctlr_position,
4089	int i, int nphysicals, int nlocal_logicals)
4090{
4091	/* In report logicals, local logicals are listed first,
4092	* then any externals.
4093	*/
4094	int logicals_start = nphysicals + (raid_ctlr_position == 0);
4095
4096	if (i == raid_ctlr_position)
4097		return 0;
4098
4099	if (i < logicals_start)
4100		return 0;
4101
4102	/* i is in logicals range, but still within local logicals */
4103	if ((i - nphysicals - (raid_ctlr_position == 0)) < nlocal_logicals)
4104		return 0;
4105
4106	return 1; /* it's an external lun */
4107}
4108
4109/*
4110 * Do CISS_REPORT_PHYS and CISS_REPORT_LOG.  Data is returned in physdev,
4111 * logdev.  The number of luns in physdev and logdev are returned in
4112 * *nphysicals and *nlogicals, respectively.
4113 * Returns 0 on success, -1 otherwise.
4114 */
4115static int hpsa_gather_lun_info(struct ctlr_info *h,
4116	struct ReportExtendedLUNdata *physdev, u32 *nphysicals,
4117	struct ReportLUNdata *logdev, u32 *nlogicals)
4118{
4119	if (hpsa_scsi_do_report_phys_luns(h, physdev, sizeof(*physdev))) {
4120		dev_err(&h->pdev->dev, "report physical LUNs failed.\n");
4121		return -1;
4122	}
4123	*nphysicals = be32_to_cpu(*((__be32 *)physdev->LUNListLength)) / 24;
4124	if (*nphysicals > HPSA_MAX_PHYS_LUN) {
4125		dev_warn(&h->pdev->dev, "maximum physical LUNs (%d) exceeded. %d LUNs ignored.\n",
4126			HPSA_MAX_PHYS_LUN, *nphysicals - HPSA_MAX_PHYS_LUN);
4127		*nphysicals = HPSA_MAX_PHYS_LUN;
4128	}
4129	if (hpsa_scsi_do_report_log_luns(h, logdev, sizeof(*logdev))) {
4130		dev_err(&h->pdev->dev, "report logical LUNs failed.\n");
4131		return -1;
4132	}
4133	*nlogicals = be32_to_cpu(*((__be32 *) logdev->LUNListLength)) / 8;
4134	/* Reject Logicals in excess of our max capability. */
4135	if (*nlogicals > HPSA_MAX_LUN) {
4136		dev_warn(&h->pdev->dev,
4137			"maximum logical LUNs (%d) exceeded.  "
4138			"%d LUNs ignored.\n", HPSA_MAX_LUN,
4139			*nlogicals - HPSA_MAX_LUN);
4140		*nlogicals = HPSA_MAX_LUN;
4141	}
4142	if (*nlogicals + *nphysicals > HPSA_MAX_PHYS_LUN) {
4143		dev_warn(&h->pdev->dev,
4144			"maximum logical + physical LUNs (%d) exceeded. "
4145			"%d LUNs ignored.\n", HPSA_MAX_PHYS_LUN,
4146			*nphysicals + *nlogicals - HPSA_MAX_PHYS_LUN);
4147		*nlogicals = HPSA_MAX_PHYS_LUN - *nphysicals;
4148	}
4149	return 0;
4150}
4151
4152static u8 *figure_lunaddrbytes(struct ctlr_info *h, int raid_ctlr_position,
4153	int i, int nphysicals, int nlogicals,
4154	struct ReportExtendedLUNdata *physdev_list,
4155	struct ReportLUNdata *logdev_list)
4156{
4157	/* Helper function, figure out where the LUN ID info is coming from
4158	 * given index i, lists of physical and logical devices, where in
4159	 * the list the raid controller is supposed to appear (first or last)
4160	 */
4161
4162	int logicals_start = nphysicals + (raid_ctlr_position == 0);
4163	int last_device = nphysicals + nlogicals + (raid_ctlr_position == 0);
4164
4165	if (i == raid_ctlr_position)
4166		return RAID_CTLR_LUNID;
4167
4168	if (i < logicals_start)
4169		return &physdev_list->LUN[i -
4170				(raid_ctlr_position == 0)].lunid[0];
4171
4172	if (i < last_device)
4173		return &logdev_list->LUN[i - nphysicals -
4174			(raid_ctlr_position == 0)][0];
4175	BUG();
4176	return NULL;
4177}
4178
4179/* get physical drive ioaccel handle and queue depth */
4180static void hpsa_get_ioaccel_drive_info(struct ctlr_info *h,
4181		struct hpsa_scsi_dev_t *dev,
4182		struct ReportExtendedLUNdata *rlep, int rle_index,
4183		struct bmic_identify_physical_device *id_phys)
4184{
4185	int rc;
4186	struct ext_report_lun_entry *rle;
4187
4188	if (rle_index < 0 || rle_index >= HPSA_MAX_PHYS_LUN)
4189		return;
4190
4191	rle = &rlep->LUN[rle_index];
4192
4193	dev->ioaccel_handle = rle->ioaccel_handle;
4194	if ((rle->device_flags & 0x08) && dev->ioaccel_handle)
4195		dev->hba_ioaccel_enabled = 1;
4196	memset(id_phys, 0, sizeof(*id_phys));
4197	rc = hpsa_bmic_id_physical_device(h, &rle->lunid[0],
4198			GET_BMIC_DRIVE_NUMBER(&rle->lunid[0]), id_phys,
4199			sizeof(*id_phys));
4200	if (!rc)
4201		/* Reserve space for FW operations */
4202#define DRIVE_CMDS_RESERVED_FOR_FW 2
4203#define DRIVE_QUEUE_DEPTH 7
4204		dev->queue_depth =
4205			le16_to_cpu(id_phys->current_queue_depth_limit) -
4206				DRIVE_CMDS_RESERVED_FOR_FW;
4207	else
4208		dev->queue_depth = DRIVE_QUEUE_DEPTH; /* conservative */
4209}
4210
4211static void hpsa_get_path_info(struct hpsa_scsi_dev_t *this_device,
4212	struct ReportExtendedLUNdata *rlep, int rle_index,
4213	struct bmic_identify_physical_device *id_phys)
4214{
4215	struct ext_report_lun_entry *rle;
4216
4217	if (rle_index < 0 || rle_index >= HPSA_MAX_PHYS_LUN)
4218		return;
4219
4220	rle = &rlep->LUN[rle_index];
4221
4222	if ((rle->device_flags & 0x08) && this_device->ioaccel_handle)
4223		this_device->hba_ioaccel_enabled = 1;
4224
4225	memcpy(&this_device->active_path_index,
4226		&id_phys->active_path_number,
4227		sizeof(this_device->active_path_index));
4228	memcpy(&this_device->path_map,
4229		&id_phys->redundant_path_present_map,
4230		sizeof(this_device->path_map));
4231	memcpy(&this_device->box,
4232		&id_phys->alternate_paths_phys_box_on_port,
4233		sizeof(this_device->box));
4234	memcpy(&this_device->phys_connector,
4235		&id_phys->alternate_paths_phys_connector,
4236		sizeof(this_device->phys_connector));
4237	memcpy(&this_device->bay,
4238		&id_phys->phys_bay_in_box,
4239		sizeof(this_device->bay));
4240}
4241
4242/* get number of local logical disks. */
4243static int hpsa_set_local_logical_count(struct ctlr_info *h,
4244	struct bmic_identify_controller *id_ctlr,
4245	u32 *nlocals)
4246{
4247	int rc;
4248
4249	if (!id_ctlr) {
4250		dev_warn(&h->pdev->dev, "%s: id_ctlr buffer is NULL.\n",
4251			__func__);
4252		return -ENOMEM;
4253	}
4254	memset(id_ctlr, 0, sizeof(*id_ctlr));
4255	rc = hpsa_bmic_id_controller(h, id_ctlr, sizeof(*id_ctlr));
4256	if (!rc)
4257		if (id_ctlr->configured_logical_drive_count < 255)
4258			*nlocals = id_ctlr->configured_logical_drive_count;
4259		else
4260			*nlocals = le16_to_cpu(
4261					id_ctlr->extended_logical_unit_count);
4262	else
4263		*nlocals = -1;
4264	return rc;
4265}
4266
4267static bool hpsa_is_disk_spare(struct ctlr_info *h, u8 *lunaddrbytes)
4268{
4269	struct bmic_identify_physical_device *id_phys;
4270	bool is_spare = false;
4271	int rc;
4272
4273	id_phys = kzalloc(sizeof(*id_phys), GFP_KERNEL);
4274	if (!id_phys)
4275		return false;
4276
4277	rc = hpsa_bmic_id_physical_device(h,
4278					lunaddrbytes,
4279					GET_BMIC_DRIVE_NUMBER(lunaddrbytes),
4280					id_phys, sizeof(*id_phys));
4281	if (rc == 0)
4282		is_spare = (id_phys->more_flags >> 6) & 0x01;
4283
4284	kfree(id_phys);
4285	return is_spare;
4286}
4287
4288#define RPL_DEV_FLAG_NON_DISK                           0x1
4289#define RPL_DEV_FLAG_UNCONFIG_DISK_REPORTING_SUPPORTED  0x2
4290#define RPL_DEV_FLAG_UNCONFIG_DISK                      0x4
4291
4292#define BMIC_DEVICE_TYPE_ENCLOSURE  6
4293
4294static bool hpsa_skip_device(struct ctlr_info *h, u8 *lunaddrbytes,
4295				struct ext_report_lun_entry *rle)
4296{
4297	u8 device_flags;
4298	u8 device_type;
4299
4300	if (!MASKED_DEVICE(lunaddrbytes))
4301		return false;
4302
4303	device_flags = rle->device_flags;
4304	device_type = rle->device_type;
4305
4306	if (device_flags & RPL_DEV_FLAG_NON_DISK) {
4307		if (device_type == BMIC_DEVICE_TYPE_ENCLOSURE)
4308			return false;
4309		return true;
4310	}
4311
4312	if (!(device_flags & RPL_DEV_FLAG_UNCONFIG_DISK_REPORTING_SUPPORTED))
4313		return false;
4314
4315	if (device_flags & RPL_DEV_FLAG_UNCONFIG_DISK)
4316		return false;
4317
4318	/*
4319	 * Spares may be spun down, we do not want to
4320	 * do an Inquiry to a RAID set spare drive as
4321	 * that would have them spun up, that is a
4322	 * performance hit because I/O to the RAID device
4323	 * stops while the spin up occurs which can take
4324	 * over 50 seconds.
4325	 */
4326	if (hpsa_is_disk_spare(h, lunaddrbytes))
4327		return true;
4328
4329	return false;
4330}
4331
4332static void hpsa_update_scsi_devices(struct ctlr_info *h)
4333{
4334	/* the idea here is we could get notified
4335	 * that some devices have changed, so we do a report
4336	 * physical luns and report logical luns cmd, and adjust
4337	 * our list of devices accordingly.
4338	 *
4339	 * The scsi3addr's of devices won't change so long as the
4340	 * adapter is not reset.  That means we can rescan and
4341	 * tell which devices we already know about, vs. new
4342	 * devices, vs.  disappearing devices.
4343	 */
4344	struct ReportExtendedLUNdata *physdev_list = NULL;
4345	struct ReportLUNdata *logdev_list = NULL;
4346	struct bmic_identify_physical_device *id_phys = NULL;
4347	struct bmic_identify_controller *id_ctlr = NULL;
4348	u32 nphysicals = 0;
4349	u32 nlogicals = 0;
4350	u32 nlocal_logicals = 0;
4351	u32 ndev_allocated = 0;
4352	struct hpsa_scsi_dev_t **currentsd, *this_device, *tmpdevice;
4353	int ncurrent = 0;
4354	int i, n_ext_target_devs, ndevs_to_allocate;
4355	int raid_ctlr_position;
4356	bool physical_device;
4357	DECLARE_BITMAP(lunzerobits, MAX_EXT_TARGETS);
4358
4359	currentsd = kcalloc(HPSA_MAX_DEVICES, sizeof(*currentsd), GFP_KERNEL);
4360	physdev_list = kzalloc(sizeof(*physdev_list), GFP_KERNEL);
4361	logdev_list = kzalloc(sizeof(*logdev_list), GFP_KERNEL);
4362	tmpdevice = kzalloc(sizeof(*tmpdevice), GFP_KERNEL);
4363	id_phys = kzalloc(sizeof(*id_phys), GFP_KERNEL);
4364	id_ctlr = kzalloc(sizeof(*id_ctlr), GFP_KERNEL);
4365
4366	if (!currentsd || !physdev_list || !logdev_list ||
4367		!tmpdevice || !id_phys || !id_ctlr) {
4368		dev_err(&h->pdev->dev, "out of memory\n");
4369		goto out;
4370	}
4371	memset(lunzerobits, 0, sizeof(lunzerobits));
4372
4373	h->drv_req_rescan = 0; /* cancel scheduled rescan - we're doing it. */
4374
4375	if (hpsa_gather_lun_info(h, physdev_list, &nphysicals,
4376			logdev_list, &nlogicals)) {
4377		h->drv_req_rescan = 1;
4378		goto out;
4379	}
4380
4381	/* Set number of local logicals (non PTRAID) */
4382	if (hpsa_set_local_logical_count(h, id_ctlr, &nlocal_logicals)) {
4383		dev_warn(&h->pdev->dev,
4384			"%s: Can't determine number of local logical devices.\n",
4385			__func__);
4386	}
4387
4388	/* We might see up to the maximum number of logical and physical disks
4389	 * plus external target devices, and a device for the local RAID
4390	 * controller.
4391	 */
4392	ndevs_to_allocate = nphysicals + nlogicals + MAX_EXT_TARGETS + 1;
4393
4394	hpsa_ext_ctrl_present(h, physdev_list);
4395
4396	/* Allocate the per device structures */
4397	for (i = 0; i < ndevs_to_allocate; i++) {
4398		if (i >= HPSA_MAX_DEVICES) {
4399			dev_warn(&h->pdev->dev, "maximum devices (%d) exceeded."
4400				"  %d devices ignored.\n", HPSA_MAX_DEVICES,
4401				ndevs_to_allocate - HPSA_MAX_DEVICES);
4402			break;
4403		}
4404
4405		currentsd[i] = kzalloc(sizeof(*currentsd[i]), GFP_KERNEL);
4406		if (!currentsd[i]) {
4407			h->drv_req_rescan = 1;
4408			goto out;
4409		}
4410		ndev_allocated++;
4411	}
4412
4413	if (is_scsi_rev_5(h))
4414		raid_ctlr_position = 0;
4415	else
4416		raid_ctlr_position = nphysicals + nlogicals;
4417
4418	/* adjust our table of devices */
4419	n_ext_target_devs = 0;
4420	for (i = 0; i < nphysicals + nlogicals + 1; i++) {
4421		u8 *lunaddrbytes, is_OBDR = 0;
4422		int rc = 0;
4423		int phys_dev_index = i - (raid_ctlr_position == 0);
4424		bool skip_device = false;
4425
4426		memset(tmpdevice, 0, sizeof(*tmpdevice));
4427
4428		physical_device = i < nphysicals + (raid_ctlr_position == 0);
4429
4430		/* Figure out where the LUN ID info is coming from */
4431		lunaddrbytes = figure_lunaddrbytes(h, raid_ctlr_position,
4432			i, nphysicals, nlogicals, physdev_list, logdev_list);
4433
4434		/* Determine if this is a lun from an external target array */
4435		tmpdevice->external =
4436			figure_external_status(h, raid_ctlr_position, i,
4437						nphysicals, nlocal_logicals);
4438
4439		/*
4440		 * Skip over some devices such as a spare.
4441		 */
4442		if (phys_dev_index >= 0 && !tmpdevice->external &&
4443			physical_device) {
4444			skip_device = hpsa_skip_device(h, lunaddrbytes,
4445					&physdev_list->LUN[phys_dev_index]);
4446			if (skip_device)
4447				continue;
4448		}
4449
4450		/* Get device type, vendor, model, device id, raid_map */
4451		rc = hpsa_update_device_info(h, lunaddrbytes, tmpdevice,
4452							&is_OBDR);
4453		if (rc == -ENOMEM) {
4454			dev_warn(&h->pdev->dev,
4455				"Out of memory, rescan deferred.\n");
4456			h->drv_req_rescan = 1;
4457			goto out;
4458		}
4459		if (rc) {
4460			h->drv_req_rescan = 1;
4461			continue;
4462		}
4463
4464		figure_bus_target_lun(h, lunaddrbytes, tmpdevice);
4465		this_device = currentsd[ncurrent];
4466
4467		*this_device = *tmpdevice;
4468		this_device->physical_device = physical_device;
4469
4470		/*
4471		 * Expose all devices except for physical devices that
4472		 * are masked.
4473		 */
4474		if (MASKED_DEVICE(lunaddrbytes) && this_device->physical_device)
4475			this_device->expose_device = 0;
4476		else
4477			this_device->expose_device = 1;
4478
4479
4480		/*
4481		 * Get the SAS address for physical devices that are exposed.
4482		 */
4483		if (this_device->physical_device && this_device->expose_device)
4484			hpsa_get_sas_address(h, lunaddrbytes, this_device);
4485
4486		switch (this_device->devtype) {
4487		case TYPE_ROM:
4488			/* We don't *really* support actual CD-ROM devices,
4489			 * just "One Button Disaster Recovery" tape drive
4490			 * which temporarily pretends to be a CD-ROM drive.
4491			 * So we check that the device is really an OBDR tape
4492			 * device by checking for "$DR-10" in bytes 43-48 of
4493			 * the inquiry data.
4494			 */
4495			if (is_OBDR)
4496				ncurrent++;
4497			break;
4498		case TYPE_DISK:
4499		case TYPE_ZBC:
4500			if (this_device->physical_device) {
4501				/* The disk is in HBA mode. */
4502				/* Never use RAID mapper in HBA mode. */
4503				this_device->offload_enabled = 0;
4504				hpsa_get_ioaccel_drive_info(h, this_device,
4505					physdev_list, phys_dev_index, id_phys);
4506				hpsa_get_path_info(this_device,
4507					physdev_list, phys_dev_index, id_phys);
4508			}
4509			ncurrent++;
4510			break;
4511		case TYPE_TAPE:
4512		case TYPE_MEDIUM_CHANGER:
4513			ncurrent++;
4514			break;
4515		case TYPE_ENCLOSURE:
4516			if (!this_device->external)
4517				hpsa_get_enclosure_info(h, lunaddrbytes,
4518						physdev_list, phys_dev_index,
4519						this_device);
4520			ncurrent++;
4521			break;
4522		case TYPE_RAID:
4523			/* Only present the Smartarray HBA as a RAID controller.
4524			 * If it's a RAID controller other than the HBA itself
4525			 * (an external RAID controller, MSA500 or similar)
4526			 * don't present it.
4527			 */
4528			if (!is_hba_lunid(lunaddrbytes))
4529				break;
4530			ncurrent++;
4531			break;
4532		default:
4533			break;
4534		}
4535		if (ncurrent >= HPSA_MAX_DEVICES)
4536			break;
4537	}
4538
4539	if (h->sas_host == NULL) {
4540		int rc = 0;
4541
4542		rc = hpsa_add_sas_host(h);
4543		if (rc) {
4544			dev_warn(&h->pdev->dev,
4545				"Could not add sas host %d\n", rc);
4546			goto out;
4547		}
4548	}
4549
4550	adjust_hpsa_scsi_table(h, currentsd, ncurrent);
4551out:
4552	kfree(tmpdevice);
4553	for (i = 0; i < ndev_allocated; i++)
4554		kfree(currentsd[i]);
4555	kfree(currentsd);
4556	kfree(physdev_list);
4557	kfree(logdev_list);
4558	kfree(id_ctlr);
4559	kfree(id_phys);
4560}
4561
4562static void hpsa_set_sg_descriptor(struct SGDescriptor *desc,
4563				   struct scatterlist *sg)
4564{
4565	u64 addr64 = (u64) sg_dma_address(sg);
4566	unsigned int len = sg_dma_len(sg);
4567
4568	desc->Addr = cpu_to_le64(addr64);
4569	desc->Len = cpu_to_le32(len);
4570	desc->Ext = 0;
4571}
4572
4573/*
4574 * hpsa_scatter_gather takes a struct scsi_cmnd, (cmd), and does the pci
4575 * dma mapping  and fills in the scatter gather entries of the
4576 * hpsa command, cp.
4577 */
4578static int hpsa_scatter_gather(struct ctlr_info *h,
4579		struct CommandList *cp,
4580		struct scsi_cmnd *cmd)
4581{
4582	struct scatterlist *sg;
4583	int use_sg, i, sg_limit, chained, last_sg;
4584	struct SGDescriptor *curr_sg;
4585
4586	BUG_ON(scsi_sg_count(cmd) > h->maxsgentries);
4587
4588	use_sg = scsi_dma_map(cmd);
4589	if (use_sg < 0)
4590		return use_sg;
4591
4592	if (!use_sg)
4593		goto sglist_finished;
4594
4595	/*
4596	 * If the number of entries is greater than the max for a single list,
4597	 * then we have a chained list; we will set up all but one entry in the
4598	 * first list (the last entry is saved for link information);
4599	 * otherwise, we don't have a chained list and we'll set up at each of
4600	 * the entries in the one list.
4601	 */
4602	curr_sg = cp->SG;
4603	chained = use_sg > h->max_cmd_sg_entries;
4604	sg_limit = chained ? h->max_cmd_sg_entries - 1 : use_sg;
4605	last_sg = scsi_sg_count(cmd) - 1;
4606	scsi_for_each_sg(cmd, sg, sg_limit, i) {
4607		hpsa_set_sg_descriptor(curr_sg, sg);
4608		curr_sg++;
4609	}
4610
4611	if (chained) {
4612		/*
4613		 * Continue with the chained list.  Set curr_sg to the chained
4614		 * list.  Modify the limit to the total count less the entries
4615		 * we've already set up.  Resume the scan at the list entry
4616		 * where the previous loop left off.
4617		 */
4618		curr_sg = h->cmd_sg_list[cp->cmdindex];
4619		sg_limit = use_sg - sg_limit;
4620		for_each_sg(sg, sg, sg_limit, i) {
4621			hpsa_set_sg_descriptor(curr_sg, sg);
4622			curr_sg++;
4623		}
4624	}
4625
4626	/* Back the pointer up to the last entry and mark it as "last". */
4627	(curr_sg - 1)->Ext = cpu_to_le32(HPSA_SG_LAST);
4628
4629	if (use_sg + chained > h->maxSG)
4630		h->maxSG = use_sg + chained;
4631
4632	if (chained) {
4633		cp->Header.SGList = h->max_cmd_sg_entries;
4634		cp->Header.SGTotal = cpu_to_le16(use_sg + 1);
4635		if (hpsa_map_sg_chain_block(h, cp)) {
4636			scsi_dma_unmap(cmd);
4637			return -1;
4638		}
4639		return 0;
4640	}
4641
4642sglist_finished:
4643
4644	cp->Header.SGList = (u8) use_sg;   /* no. SGs contig in this cmd */
4645	cp->Header.SGTotal = cpu_to_le16(use_sg); /* total sgs in cmd list */
4646	return 0;
4647}
4648
4649static inline void warn_zero_length_transfer(struct ctlr_info *h,
4650						u8 *cdb, int cdb_len,
4651						const char *func)
4652{
4653	dev_warn(&h->pdev->dev,
4654		 "%s: Blocking zero-length request: CDB:%*phN\n",
4655		 func, cdb_len, cdb);
4656}
4657
4658#define IO_ACCEL_INELIGIBLE 1
4659/* zero-length transfers trigger hardware errors. */
4660static bool is_zero_length_transfer(u8 *cdb)
4661{
4662	u32 block_cnt;
4663
4664	/* Block zero-length transfer sizes on certain commands. */
4665	switch (cdb[0]) {
4666	case READ_10:
4667	case WRITE_10:
4668	case VERIFY:		/* 0x2F */
4669	case WRITE_VERIFY:	/* 0x2E */
4670		block_cnt = get_unaligned_be16(&cdb[7]);
4671		break;
4672	case READ_12:
4673	case WRITE_12:
4674	case VERIFY_12: /* 0xAF */
4675	case WRITE_VERIFY_12:	/* 0xAE */
4676		block_cnt = get_unaligned_be32(&cdb[6]);
4677		break;
4678	case READ_16:
4679	case WRITE_16:
4680	case VERIFY_16:		/* 0x8F */
4681		block_cnt = get_unaligned_be32(&cdb[10]);
4682		break;
4683	default:
4684		return false;
4685	}
4686
4687	return block_cnt == 0;
4688}
4689
4690static int fixup_ioaccel_cdb(u8 *cdb, int *cdb_len)
4691{
4692	int is_write = 0;
4693	u32 block;
4694	u32 block_cnt;
4695
4696	/* Perform some CDB fixups if needed using 10 byte reads/writes only */
4697	switch (cdb[0]) {
4698	case WRITE_6:
4699	case WRITE_12:
4700		is_write = 1;
4701		fallthrough;
4702	case READ_6:
4703	case READ_12:
4704		if (*cdb_len == 6) {
4705			block = (((cdb[1] & 0x1F) << 16) |
4706				(cdb[2] << 8) |
4707				cdb[3]);
4708			block_cnt = cdb[4];
4709			if (block_cnt == 0)
4710				block_cnt = 256;
4711		} else {
4712			BUG_ON(*cdb_len != 12);
4713			block = get_unaligned_be32(&cdb[2]);
4714			block_cnt = get_unaligned_be32(&cdb[6]);
4715		}
4716		if (block_cnt > 0xffff)
4717			return IO_ACCEL_INELIGIBLE;
4718
4719		cdb[0] = is_write ? WRITE_10 : READ_10;
4720		cdb[1] = 0;
4721		cdb[2] = (u8) (block >> 24);
4722		cdb[3] = (u8) (block >> 16);
4723		cdb[4] = (u8) (block >> 8);
4724		cdb[5] = (u8) (block);
4725		cdb[6] = 0;
4726		cdb[7] = (u8) (block_cnt >> 8);
4727		cdb[8] = (u8) (block_cnt);
4728		cdb[9] = 0;
4729		*cdb_len = 10;
4730		break;
4731	}
4732	return 0;
4733}
4734
4735static int hpsa_scsi_ioaccel1_queue_command(struct ctlr_info *h,
4736	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
4737	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk)
4738{
4739	struct scsi_cmnd *cmd = c->scsi_cmd;
4740	struct io_accel1_cmd *cp = &h->ioaccel_cmd_pool[c->cmdindex];
4741	unsigned int len;
4742	unsigned int total_len = 0;
4743	struct scatterlist *sg;
4744	u64 addr64;
4745	int use_sg, i;
4746	struct SGDescriptor *curr_sg;
4747	u32 control = IOACCEL1_CONTROL_SIMPLEQUEUE;
4748
4749	/* TODO: implement chaining support */
4750	if (scsi_sg_count(cmd) > h->ioaccel_maxsg) {
4751		atomic_dec(&phys_disk->ioaccel_cmds_out);
4752		return IO_ACCEL_INELIGIBLE;
4753	}
4754
4755	BUG_ON(cmd->cmd_len > IOACCEL1_IOFLAGS_CDBLEN_MAX);
4756
4757	if (is_zero_length_transfer(cdb)) {
4758		warn_zero_length_transfer(h, cdb, cdb_len, __func__);
4759		atomic_dec(&phys_disk->ioaccel_cmds_out);
4760		return IO_ACCEL_INELIGIBLE;
4761	}
4762
4763	if (fixup_ioaccel_cdb(cdb, &cdb_len)) {
4764		atomic_dec(&phys_disk->ioaccel_cmds_out);
4765		return IO_ACCEL_INELIGIBLE;
4766	}
4767
4768	c->cmd_type = CMD_IOACCEL1;
4769
4770	/* Adjust the DMA address to point to the accelerated command buffer */
4771	c->busaddr = (u32) h->ioaccel_cmd_pool_dhandle +
4772				(c->cmdindex * sizeof(*cp));
4773	BUG_ON(c->busaddr & 0x0000007F);
4774
4775	use_sg = scsi_dma_map(cmd);
4776	if (use_sg < 0) {
4777		atomic_dec(&phys_disk->ioaccel_cmds_out);
4778		return use_sg;
4779	}
4780
4781	if (use_sg) {
4782		curr_sg = cp->SG;
4783		scsi_for_each_sg(cmd, sg, use_sg, i) {
4784			addr64 = (u64) sg_dma_address(sg);
4785			len  = sg_dma_len(sg);
4786			total_len += len;
4787			curr_sg->Addr = cpu_to_le64(addr64);
4788			curr_sg->Len = cpu_to_le32(len);
4789			curr_sg->Ext = cpu_to_le32(0);
4790			curr_sg++;
4791		}
4792		(--curr_sg)->Ext = cpu_to_le32(HPSA_SG_LAST);
4793
4794		switch (cmd->sc_data_direction) {
4795		case DMA_TO_DEVICE:
4796			control |= IOACCEL1_CONTROL_DATA_OUT;
4797			break;
4798		case DMA_FROM_DEVICE:
4799			control |= IOACCEL1_CONTROL_DATA_IN;
4800			break;
4801		case DMA_NONE:
4802			control |= IOACCEL1_CONTROL_NODATAXFER;
4803			break;
4804		default:
4805			dev_err(&h->pdev->dev, "unknown data direction: %d\n",
4806			cmd->sc_data_direction);
4807			BUG();
4808			break;
4809		}
4810	} else {
4811		control |= IOACCEL1_CONTROL_NODATAXFER;
4812	}
4813
4814	c->Header.SGList = use_sg;
4815	/* Fill out the command structure to submit */
4816	cp->dev_handle = cpu_to_le16(ioaccel_handle & 0xFFFF);
4817	cp->transfer_len = cpu_to_le32(total_len);
4818	cp->io_flags = cpu_to_le16(IOACCEL1_IOFLAGS_IO_REQ |
4819			(cdb_len & IOACCEL1_IOFLAGS_CDBLEN_MASK));
4820	cp->control = cpu_to_le32(control);
4821	memcpy(cp->CDB, cdb, cdb_len);
4822	memcpy(cp->CISS_LUN, scsi3addr, 8);
4823	/* Tag was already set at init time. */
4824	enqueue_cmd_and_start_io(h, c);
4825	return 0;
4826}
4827
4828/*
4829 * Queue a command directly to a device behind the controller using the
4830 * I/O accelerator path.
4831 */
4832static int hpsa_scsi_ioaccel_direct_map(struct ctlr_info *h,
4833	struct CommandList *c)
4834{
4835	struct scsi_cmnd *cmd = c->scsi_cmd;
4836	struct hpsa_scsi_dev_t *dev = cmd->device->hostdata;
4837
4838	if (!dev)
4839		return -1;
4840
4841	c->phys_disk = dev;
4842
4843	if (dev->in_reset)
4844		return -1;
4845
4846	return hpsa_scsi_ioaccel_queue_command(h, c, dev->ioaccel_handle,
4847		cmd->cmnd, cmd->cmd_len, dev->scsi3addr, dev);
4848}
4849
4850/*
4851 * Set encryption parameters for the ioaccel2 request
4852 */
4853static void set_encrypt_ioaccel2(struct ctlr_info *h,
4854	struct CommandList *c, struct io_accel2_cmd *cp)
4855{
4856	struct scsi_cmnd *cmd = c->scsi_cmd;
4857	struct hpsa_scsi_dev_t *dev = cmd->device->hostdata;
4858	struct raid_map_data *map = &dev->raid_map;
4859	u64 first_block;
4860
4861	/* Are we doing encryption on this device */
4862	if (!(le16_to_cpu(map->flags) & RAID_MAP_FLAG_ENCRYPT_ON))
4863		return;
4864	/* Set the data encryption key index. */
4865	cp->dekindex = map->dekindex;
4866
4867	/* Set the encryption enable flag, encoded into direction field. */
4868	cp->direction |= IOACCEL2_DIRECTION_ENCRYPT_MASK;
4869
4870	/* Set encryption tweak values based on logical block address
4871	 * If block size is 512, tweak value is LBA.
4872	 * For other block sizes, tweak is (LBA * block size)/ 512)
4873	 */
4874	switch (cmd->cmnd[0]) {
4875	/* Required? 6-byte cdbs eliminated by fixup_ioaccel_cdb */
4876	case READ_6:
4877	case WRITE_6:
4878		first_block = (((cmd->cmnd[1] & 0x1F) << 16) |
4879				(cmd->cmnd[2] << 8) |
4880				cmd->cmnd[3]);
4881		break;
4882	case WRITE_10:
4883	case READ_10:
4884	/* Required? 12-byte cdbs eliminated by fixup_ioaccel_cdb */
4885	case WRITE_12:
4886	case READ_12:
4887		first_block = get_unaligned_be32(&cmd->cmnd[2]);
4888		break;
4889	case WRITE_16:
4890	case READ_16:
4891		first_block = get_unaligned_be64(&cmd->cmnd[2]);
4892		break;
4893	default:
4894		dev_err(&h->pdev->dev,
4895			"ERROR: %s: size (0x%x) not supported for encryption\n",
4896			__func__, cmd->cmnd[0]);
4897		BUG();
4898		break;
4899	}
4900
4901	if (le32_to_cpu(map->volume_blk_size) != 512)
4902		first_block = first_block *
4903				le32_to_cpu(map->volume_blk_size)/512;
4904
4905	cp->tweak_lower = cpu_to_le32(first_block);
4906	cp->tweak_upper = cpu_to_le32(first_block >> 32);
4907}
4908
4909static int hpsa_scsi_ioaccel2_queue_command(struct ctlr_info *h,
4910	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
4911	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk)
4912{
4913	struct scsi_cmnd *cmd = c->scsi_cmd;
4914	struct io_accel2_cmd *cp = &h->ioaccel2_cmd_pool[c->cmdindex];
4915	struct ioaccel2_sg_element *curr_sg;
4916	int use_sg, i;
4917	struct scatterlist *sg;
4918	u64 addr64;
4919	u32 len;
4920	u32 total_len = 0;
4921
4922	if (!cmd->device)
4923		return -1;
4924
4925	if (!cmd->device->hostdata)
4926		return -1;
4927
4928	BUG_ON(scsi_sg_count(cmd) > h->maxsgentries);
4929
4930	if (is_zero_length_transfer(cdb)) {
4931		warn_zero_length_transfer(h, cdb, cdb_len, __func__);
4932		atomic_dec(&phys_disk->ioaccel_cmds_out);
4933		return IO_ACCEL_INELIGIBLE;
4934	}
4935
4936	if (fixup_ioaccel_cdb(cdb, &cdb_len)) {
4937		atomic_dec(&phys_disk->ioaccel_cmds_out);
4938		return IO_ACCEL_INELIGIBLE;
4939	}
4940
4941	c->cmd_type = CMD_IOACCEL2;
4942	/* Adjust the DMA address to point to the accelerated command buffer */
4943	c->busaddr = (u32) h->ioaccel2_cmd_pool_dhandle +
4944				(c->cmdindex * sizeof(*cp));
4945	BUG_ON(c->busaddr & 0x0000007F);
4946
4947	memset(cp, 0, sizeof(*cp));
4948	cp->IU_type = IOACCEL2_IU_TYPE;
4949
4950	use_sg = scsi_dma_map(cmd);
4951	if (use_sg < 0) {
4952		atomic_dec(&phys_disk->ioaccel_cmds_out);
4953		return use_sg;
4954	}
4955
4956	if (use_sg) {
4957		curr_sg = cp->sg;
4958		if (use_sg > h->ioaccel_maxsg) {
4959			addr64 = le64_to_cpu(
4960				h->ioaccel2_cmd_sg_list[c->cmdindex]->address);
4961			curr_sg->address = cpu_to_le64(addr64);
4962			curr_sg->length = 0;
4963			curr_sg->reserved[0] = 0;
4964			curr_sg->reserved[1] = 0;
4965			curr_sg->reserved[2] = 0;
4966			curr_sg->chain_indicator = IOACCEL2_CHAIN;
4967
4968			curr_sg = h->ioaccel2_cmd_sg_list[c->cmdindex];
4969		}
4970		scsi_for_each_sg(cmd, sg, use_sg, i) {
4971			addr64 = (u64) sg_dma_address(sg);
4972			len  = sg_dma_len(sg);
4973			total_len += len;
4974			curr_sg->address = cpu_to_le64(addr64);
4975			curr_sg->length = cpu_to_le32(len);
4976			curr_sg->reserved[0] = 0;
4977			curr_sg->reserved[1] = 0;
4978			curr_sg->reserved[2] = 0;
4979			curr_sg->chain_indicator = 0;
4980			curr_sg++;
4981		}
4982
4983		/*
4984		 * Set the last s/g element bit
4985		 */
4986		(curr_sg - 1)->chain_indicator = IOACCEL2_LAST_SG;
4987
4988		switch (cmd->sc_data_direction) {
4989		case DMA_TO_DEVICE:
4990			cp->direction &= ~IOACCEL2_DIRECTION_MASK;
4991			cp->direction |= IOACCEL2_DIR_DATA_OUT;
4992			break;
4993		case DMA_FROM_DEVICE:
4994			cp->direction &= ~IOACCEL2_DIRECTION_MASK;
4995			cp->direction |= IOACCEL2_DIR_DATA_IN;
4996			break;
4997		case DMA_NONE:
4998			cp->direction &= ~IOACCEL2_DIRECTION_MASK;
4999			cp->direction |= IOACCEL2_DIR_NO_DATA;
5000			break;
5001		default:
5002			dev_err(&h->pdev->dev, "unknown data direction: %d\n",
5003				cmd->sc_data_direction);
5004			BUG();
5005			break;
5006		}
5007	} else {
5008		cp->direction &= ~IOACCEL2_DIRECTION_MASK;
5009		cp->direction |= IOACCEL2_DIR_NO_DATA;
5010	}
5011
5012	/* Set encryption parameters, if necessary */
5013	set_encrypt_ioaccel2(h, c, cp);
5014
5015	cp->scsi_nexus = cpu_to_le32(ioaccel_handle);
5016	cp->Tag = cpu_to_le32(c->cmdindex << DIRECT_LOOKUP_SHIFT);
5017	memcpy(cp->cdb, cdb, sizeof(cp->cdb));
5018
5019	cp->data_len = cpu_to_le32(total_len);
5020	cp->err_ptr = cpu_to_le64(c->busaddr +
5021			offsetof(struct io_accel2_cmd, error_data));
5022	cp->err_len = cpu_to_le32(sizeof(cp->error_data));
5023
5024	/* fill in sg elements */
5025	if (use_sg > h->ioaccel_maxsg) {
5026		cp->sg_count = 1;
5027		cp->sg[0].length = cpu_to_le32(use_sg * sizeof(cp->sg[0]));
5028		if (hpsa_map_ioaccel2_sg_chain_block(h, cp, c)) {
5029			atomic_dec(&phys_disk->ioaccel_cmds_out);
5030			scsi_dma_unmap(cmd);
5031			return -1;
5032		}
5033	} else
5034		cp->sg_count = (u8) use_sg;
5035
5036	if (phys_disk->in_reset) {
5037		cmd->result = DID_RESET << 16;
5038		return -1;
5039	}
5040
5041	enqueue_cmd_and_start_io(h, c);
5042	return 0;
5043}
5044
5045/*
5046 * Queue a command to the correct I/O accelerator path.
5047 */
5048static int hpsa_scsi_ioaccel_queue_command(struct ctlr_info *h,
5049	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
5050	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk)
5051{
5052	if (!c->scsi_cmd->device)
5053		return -1;
5054
5055	if (!c->scsi_cmd->device->hostdata)
5056		return -1;
5057
5058	if (phys_disk->in_reset)
5059		return -1;
5060
5061	/* Try to honor the device's queue depth */
5062	if (atomic_inc_return(&phys_disk->ioaccel_cmds_out) >
5063					phys_disk->queue_depth) {
5064		atomic_dec(&phys_disk->ioaccel_cmds_out);
5065		return IO_ACCEL_INELIGIBLE;
5066	}
5067	if (h->transMethod & CFGTBL_Trans_io_accel1)
5068		return hpsa_scsi_ioaccel1_queue_command(h, c, ioaccel_handle,
5069						cdb, cdb_len, scsi3addr,
5070						phys_disk);
5071	else
5072		return hpsa_scsi_ioaccel2_queue_command(h, c, ioaccel_handle,
5073						cdb, cdb_len, scsi3addr,
5074						phys_disk);
5075}
5076
5077static void raid_map_helper(struct raid_map_data *map,
5078		int offload_to_mirror, u32 *map_index, u32 *current_group)
5079{
5080	if (offload_to_mirror == 0)  {
5081		/* use physical disk in the first mirrored group. */
5082		*map_index %= le16_to_cpu(map->data_disks_per_row);
5083		return;
5084	}
5085	do {
5086		/* determine mirror group that *map_index indicates */
5087		*current_group = *map_index /
5088			le16_to_cpu(map->data_disks_per_row);
5089		if (offload_to_mirror == *current_group)
5090			continue;
5091		if (*current_group < le16_to_cpu(map->layout_map_count) - 1) {
5092			/* select map index from next group */
5093			*map_index += le16_to_cpu(map->data_disks_per_row);
5094			(*current_group)++;
5095		} else {
5096			/* select map index from first group */
5097			*map_index %= le16_to_cpu(map->data_disks_per_row);
5098			*current_group = 0;
5099		}
5100	} while (offload_to_mirror != *current_group);
5101}
5102
5103/*
5104 * Attempt to perform offload RAID mapping for a logical volume I/O.
5105 */
5106static int hpsa_scsi_ioaccel_raid_map(struct ctlr_info *h,
5107	struct CommandList *c)
5108{
5109	struct scsi_cmnd *cmd = c->scsi_cmd;
5110	struct hpsa_scsi_dev_t *dev = cmd->device->hostdata;
5111	struct raid_map_data *map = &dev->raid_map;
5112	struct raid_map_disk_data *dd = &map->data[0];
5113	int is_write = 0;
5114	u32 map_index;
5115	u64 first_block, last_block;
5116	u32 block_cnt;
5117	u32 blocks_per_row;
5118	u64 first_row, last_row;
5119	u32 first_row_offset, last_row_offset;
5120	u32 first_column, last_column;
5121	u64 r0_first_row, r0_last_row;
5122	u32 r5or6_blocks_per_row;
5123	u64 r5or6_first_row, r5or6_last_row;
5124	u32 r5or6_first_row_offset, r5or6_last_row_offset;
5125	u32 r5or6_first_column, r5or6_last_column;
5126	u32 total_disks_per_row;
5127	u32 stripesize;
5128	u32 first_group, last_group, current_group;
5129	u32 map_row;
5130	u32 disk_handle;
5131	u64 disk_block;
5132	u32 disk_block_cnt;
5133	u8 cdb[16];
5134	u8 cdb_len;
5135	u16 strip_size;
5136#if BITS_PER_LONG == 32
5137	u64 tmpdiv;
5138#endif
5139	int offload_to_mirror;
5140
5141	if (!dev)
5142		return -1;
5143
5144	if (dev->in_reset)
5145		return -1;
5146
5147	/* check for valid opcode, get LBA and block count */
5148	switch (cmd->cmnd[0]) {
5149	case WRITE_6:
5150		is_write = 1;
5151		fallthrough;
5152	case READ_6:
5153		first_block = (((cmd->cmnd[1] & 0x1F) << 16) |
5154				(cmd->cmnd[2] << 8) |
5155				cmd->cmnd[3]);
5156		block_cnt = cmd->cmnd[4];
5157		if (block_cnt == 0)
5158			block_cnt = 256;
5159		break;
5160	case WRITE_10:
5161		is_write = 1;
5162		fallthrough;
5163	case READ_10:
5164		first_block =
5165			(((u64) cmd->cmnd[2]) << 24) |
5166			(((u64) cmd->cmnd[3]) << 16) |
5167			(((u64) cmd->cmnd[4]) << 8) |
5168			cmd->cmnd[5];
5169		block_cnt =
5170			(((u32) cmd->cmnd[7]) << 8) |
5171			cmd->cmnd[8];
5172		break;
5173	case WRITE_12:
5174		is_write = 1;
5175		fallthrough;
5176	case READ_12:
5177		first_block =
5178			(((u64) cmd->cmnd[2]) << 24) |
5179			(((u64) cmd->cmnd[3]) << 16) |
5180			(((u64) cmd->cmnd[4]) << 8) |
5181			cmd->cmnd[5];
5182		block_cnt =
5183			(((u32) cmd->cmnd[6]) << 24) |
5184			(((u32) cmd->cmnd[7]) << 16) |
5185			(((u32) cmd->cmnd[8]) << 8) |
5186		cmd->cmnd[9];
5187		break;
5188	case WRITE_16:
5189		is_write = 1;
5190		fallthrough;
5191	case READ_16:
5192		first_block =
5193			(((u64) cmd->cmnd[2]) << 56) |
5194			(((u64) cmd->cmnd[3]) << 48) |
5195			(((u64) cmd->cmnd[4]) << 40) |
5196			(((u64) cmd->cmnd[5]) << 32) |
5197			(((u64) cmd->cmnd[6]) << 24) |
5198			(((u64) cmd->cmnd[7]) << 16) |
5199			(((u64) cmd->cmnd[8]) << 8) |
5200			cmd->cmnd[9];
5201		block_cnt =
5202			(((u32) cmd->cmnd[10]) << 24) |
5203			(((u32) cmd->cmnd[11]) << 16) |
5204			(((u32) cmd->cmnd[12]) << 8) |
5205			cmd->cmnd[13];
5206		break;
5207	default:
5208		return IO_ACCEL_INELIGIBLE; /* process via normal I/O path */
5209	}
5210	last_block = first_block + block_cnt - 1;
5211
5212	/* check for write to non-RAID-0 */
5213	if (is_write && dev->raid_level != 0)
5214		return IO_ACCEL_INELIGIBLE;
5215
5216	/* check for invalid block or wraparound */
5217	if (last_block >= le64_to_cpu(map->volume_blk_cnt) ||
5218		last_block < first_block)
5219		return IO_ACCEL_INELIGIBLE;
5220
5221	/* calculate stripe information for the request */
5222	blocks_per_row = le16_to_cpu(map->data_disks_per_row) *
5223				le16_to_cpu(map->strip_size);
5224	strip_size = le16_to_cpu(map->strip_size);
5225#if BITS_PER_LONG == 32
5226	tmpdiv = first_block;
5227	(void) do_div(tmpdiv, blocks_per_row);
5228	first_row = tmpdiv;
5229	tmpdiv = last_block;
5230	(void) do_div(tmpdiv, blocks_per_row);
5231	last_row = tmpdiv;
5232	first_row_offset = (u32) (first_block - (first_row * blocks_per_row));
5233	last_row_offset = (u32) (last_block - (last_row * blocks_per_row));
5234	tmpdiv = first_row_offset;
5235	(void) do_div(tmpdiv, strip_size);
5236	first_column = tmpdiv;
5237	tmpdiv = last_row_offset;
5238	(void) do_div(tmpdiv, strip_size);
5239	last_column = tmpdiv;
5240#else
5241	first_row = first_block / blocks_per_row;
5242	last_row = last_block / blocks_per_row;
5243	first_row_offset = (u32) (first_block - (first_row * blocks_per_row));
5244	last_row_offset = (u32) (last_block - (last_row * blocks_per_row));
5245	first_column = first_row_offset / strip_size;
5246	last_column = last_row_offset / strip_size;
5247#endif
5248
5249	/* if this isn't a single row/column then give to the controller */
5250	if ((first_row != last_row) || (first_column != last_column))
5251		return IO_ACCEL_INELIGIBLE;
5252
5253	/* proceeding with driver mapping */
5254	total_disks_per_row = le16_to_cpu(map->data_disks_per_row) +
5255				le16_to_cpu(map->metadata_disks_per_row);
5256	map_row = ((u32)(first_row >> map->parity_rotation_shift)) %
5257				le16_to_cpu(map->row_cnt);
5258	map_index = (map_row * total_disks_per_row) + first_column;
5259
5260	switch (dev->raid_level) {
5261	case HPSA_RAID_0:
5262		break; /* nothing special to do */
5263	case HPSA_RAID_1:
5264		/* Handles load balance across RAID 1 members.
5265		 * (2-drive R1 and R10 with even # of drives.)
5266		 * Appropriate for SSDs, not optimal for HDDs
5267		 * Ensure we have the correct raid_map.
5268		 */
5269		if (le16_to_cpu(map->layout_map_count) != 2) {
5270			hpsa_turn_off_ioaccel_for_device(dev);
5271			return IO_ACCEL_INELIGIBLE;
5272		}
5273		if (dev->offload_to_mirror)
5274			map_index += le16_to_cpu(map->data_disks_per_row);
5275		dev->offload_to_mirror = !dev->offload_to_mirror;
5276		break;
5277	case HPSA_RAID_ADM:
5278		/* Handles N-way mirrors  (R1-ADM)
5279		 * and R10 with # of drives divisible by 3.)
5280		 * Ensure we have the correct raid_map.
5281		 */
5282		if (le16_to_cpu(map->layout_map_count) != 3) {
5283			hpsa_turn_off_ioaccel_for_device(dev);
5284			return IO_ACCEL_INELIGIBLE;
5285		}
5286
5287		offload_to_mirror = dev->offload_to_mirror;
5288		raid_map_helper(map, offload_to_mirror,
5289				&map_index, &current_group);
5290		/* set mirror group to use next time */
5291		offload_to_mirror =
5292			(offload_to_mirror >=
5293			le16_to_cpu(map->layout_map_count) - 1)
5294			? 0 : offload_to_mirror + 1;
5295		dev->offload_to_mirror = offload_to_mirror;
5296		/* Avoid direct use of dev->offload_to_mirror within this
5297		 * function since multiple threads might simultaneously
5298		 * increment it beyond the range of dev->layout_map_count -1.
5299		 */
5300		break;
5301	case HPSA_RAID_5:
5302	case HPSA_RAID_6:
5303		if (le16_to_cpu(map->layout_map_count) <= 1)
5304			break;
5305
5306		/* Verify first and last block are in same RAID group */
5307		r5or6_blocks_per_row =
5308			le16_to_cpu(map->strip_size) *
5309			le16_to_cpu(map->data_disks_per_row);
5310		if (r5or6_blocks_per_row == 0) {
5311			hpsa_turn_off_ioaccel_for_device(dev);
5312			return IO_ACCEL_INELIGIBLE;
5313		}
5314		stripesize = r5or6_blocks_per_row *
5315			le16_to_cpu(map->layout_map_count);
5316#if BITS_PER_LONG == 32
5317		tmpdiv = first_block;
5318		first_group = do_div(tmpdiv, stripesize);
5319		tmpdiv = first_group;
5320		(void) do_div(tmpdiv, r5or6_blocks_per_row);
5321		first_group = tmpdiv;
5322		tmpdiv = last_block;
5323		last_group = do_div(tmpdiv, stripesize);
5324		tmpdiv = last_group;
5325		(void) do_div(tmpdiv, r5or6_blocks_per_row);
5326		last_group = tmpdiv;
5327#else
5328		first_group = (first_block % stripesize) / r5or6_blocks_per_row;
5329		last_group = (last_block % stripesize) / r5or6_blocks_per_row;
5330#endif
5331		if (first_group != last_group)
5332			return IO_ACCEL_INELIGIBLE;
5333
5334		/* Verify request is in a single row of RAID 5/6 */
5335#if BITS_PER_LONG == 32
5336		tmpdiv = first_block;
5337		(void) do_div(tmpdiv, stripesize);
5338		first_row = r5or6_first_row = r0_first_row = tmpdiv;
5339		tmpdiv = last_block;
5340		(void) do_div(tmpdiv, stripesize);
5341		r5or6_last_row = r0_last_row = tmpdiv;
5342#else
5343		first_row = r5or6_first_row = r0_first_row =
5344						first_block / stripesize;
5345		r5or6_last_row = r0_last_row = last_block / stripesize;
5346#endif
5347		if (r5or6_first_row != r5or6_last_row)
5348			return IO_ACCEL_INELIGIBLE;
5349
5350
5351		/* Verify request is in a single column */
5352#if BITS_PER_LONG == 32
5353		tmpdiv = first_block;
5354		first_row_offset = do_div(tmpdiv, stripesize);
5355		tmpdiv = first_row_offset;
5356		first_row_offset = (u32) do_div(tmpdiv, r5or6_blocks_per_row);
5357		r5or6_first_row_offset = first_row_offset;
5358		tmpdiv = last_block;
5359		r5or6_last_row_offset = do_div(tmpdiv, stripesize);
5360		tmpdiv = r5or6_last_row_offset;
5361		r5or6_last_row_offset = do_div(tmpdiv, r5or6_blocks_per_row);
5362		tmpdiv = r5or6_first_row_offset;
5363		(void) do_div(tmpdiv, map->strip_size);
5364		first_column = r5or6_first_column = tmpdiv;
5365		tmpdiv = r5or6_last_row_offset;
5366		(void) do_div(tmpdiv, map->strip_size);
5367		r5or6_last_column = tmpdiv;
5368#else
5369		first_row_offset = r5or6_first_row_offset =
5370			(u32)((first_block % stripesize) %
5371						r5or6_blocks_per_row);
5372
5373		r5or6_last_row_offset =
5374			(u32)((last_block % stripesize) %
5375						r5or6_blocks_per_row);
5376
5377		first_column = r5or6_first_column =
5378			r5or6_first_row_offset / le16_to_cpu(map->strip_size);
5379		r5or6_last_column =
5380			r5or6_last_row_offset / le16_to_cpu(map->strip_size);
5381#endif
5382		if (r5or6_first_column != r5or6_last_column)
5383			return IO_ACCEL_INELIGIBLE;
5384
5385		/* Request is eligible */
5386		map_row = ((u32)(first_row >> map->parity_rotation_shift)) %
5387			le16_to_cpu(map->row_cnt);
5388
5389		map_index = (first_group *
5390			(le16_to_cpu(map->row_cnt) * total_disks_per_row)) +
5391			(map_row * total_disks_per_row) + first_column;
5392		break;
5393	default:
5394		return IO_ACCEL_INELIGIBLE;
5395	}
5396
5397	if (unlikely(map_index >= RAID_MAP_MAX_ENTRIES))
5398		return IO_ACCEL_INELIGIBLE;
5399
5400	c->phys_disk = dev->phys_disk[map_index];
5401	if (!c->phys_disk)
5402		return IO_ACCEL_INELIGIBLE;
5403
5404	disk_handle = dd[map_index].ioaccel_handle;
5405	disk_block = le64_to_cpu(map->disk_starting_blk) +
5406			first_row * le16_to_cpu(map->strip_size) +
5407			(first_row_offset - first_column *
5408			le16_to_cpu(map->strip_size));
5409	disk_block_cnt = block_cnt;
5410
5411	/* handle differing logical/physical block sizes */
5412	if (map->phys_blk_shift) {
5413		disk_block <<= map->phys_blk_shift;
5414		disk_block_cnt <<= map->phys_blk_shift;
5415	}
5416	BUG_ON(disk_block_cnt > 0xffff);
5417
5418	/* build the new CDB for the physical disk I/O */
5419	if (disk_block > 0xffffffff) {
5420		cdb[0] = is_write ? WRITE_16 : READ_16;
5421		cdb[1] = 0;
5422		cdb[2] = (u8) (disk_block >> 56);
5423		cdb[3] = (u8) (disk_block >> 48);
5424		cdb[4] = (u8) (disk_block >> 40);
5425		cdb[5] = (u8) (disk_block >> 32);
5426		cdb[6] = (u8) (disk_block >> 24);
5427		cdb[7] = (u8) (disk_block >> 16);
5428		cdb[8] = (u8) (disk_block >> 8);
5429		cdb[9] = (u8) (disk_block);
5430		cdb[10] = (u8) (disk_block_cnt >> 24);
5431		cdb[11] = (u8) (disk_block_cnt >> 16);
5432		cdb[12] = (u8) (disk_block_cnt >> 8);
5433		cdb[13] = (u8) (disk_block_cnt);
5434		cdb[14] = 0;
5435		cdb[15] = 0;
5436		cdb_len = 16;
5437	} else {
5438		cdb[0] = is_write ? WRITE_10 : READ_10;
5439		cdb[1] = 0;
5440		cdb[2] = (u8) (disk_block >> 24);
5441		cdb[3] = (u8) (disk_block >> 16);
5442		cdb[4] = (u8) (disk_block >> 8);
5443		cdb[5] = (u8) (disk_block);
5444		cdb[6] = 0;
5445		cdb[7] = (u8) (disk_block_cnt >> 8);
5446		cdb[8] = (u8) (disk_block_cnt);
5447		cdb[9] = 0;
5448		cdb_len = 10;
5449	}
5450	return hpsa_scsi_ioaccel_queue_command(h, c, disk_handle, cdb, cdb_len,
5451						dev->scsi3addr,
5452						dev->phys_disk[map_index]);
5453}
5454
5455/*
5456 * Submit commands down the "normal" RAID stack path
5457 * All callers to hpsa_ciss_submit must check lockup_detected
5458 * beforehand, before (opt.) and after calling cmd_alloc
5459 */
5460static int hpsa_ciss_submit(struct ctlr_info *h,
5461	struct CommandList *c, struct scsi_cmnd *cmd,
5462	struct hpsa_scsi_dev_t *dev)
5463{
5464	cmd->host_scribble = (unsigned char *) c;
5465	c->cmd_type = CMD_SCSI;
5466	c->scsi_cmd = cmd;
5467	c->Header.ReplyQueue = 0;  /* unused in simple mode */
5468	memcpy(&c->Header.LUN.LunAddrBytes[0], &dev->scsi3addr[0], 8);
5469	c->Header.tag = cpu_to_le64((c->cmdindex << DIRECT_LOOKUP_SHIFT));
5470
5471	/* Fill in the request block... */
5472
5473	c->Request.Timeout = 0;
5474	BUG_ON(cmd->cmd_len > sizeof(c->Request.CDB));
5475	c->Request.CDBLen = cmd->cmd_len;
5476	memcpy(c->Request.CDB, cmd->cmnd, cmd->cmd_len);
5477	switch (cmd->sc_data_direction) {
5478	case DMA_TO_DEVICE:
5479		c->Request.type_attr_dir =
5480			TYPE_ATTR_DIR(TYPE_CMD, ATTR_SIMPLE, XFER_WRITE);
5481		break;
5482	case DMA_FROM_DEVICE:
5483		c->Request.type_attr_dir =
5484			TYPE_ATTR_DIR(TYPE_CMD, ATTR_SIMPLE, XFER_READ);
5485		break;
5486	case DMA_NONE:
5487		c->Request.type_attr_dir =
5488			TYPE_ATTR_DIR(TYPE_CMD, ATTR_SIMPLE, XFER_NONE);
5489		break;
5490	case DMA_BIDIRECTIONAL:
5491		/* This can happen if a buggy application does a scsi passthru
5492		 * and sets both inlen and outlen to non-zero. ( see
5493		 * ../scsi/scsi_ioctl.c:scsi_ioctl_send_command() )
5494		 */
5495
5496		c->Request.type_attr_dir =
5497			TYPE_ATTR_DIR(TYPE_CMD, ATTR_SIMPLE, XFER_RSVD);
5498		/* This is technically wrong, and hpsa controllers should
5499		 * reject it with CMD_INVALID, which is the most correct
5500		 * response, but non-fibre backends appear to let it
5501		 * slide by, and give the same results as if this field
5502		 * were set correctly.  Either way is acceptable for
5503		 * our purposes here.
5504		 */
5505
5506		break;
5507
5508	default:
5509		dev_err(&h->pdev->dev, "unknown data direction: %d\n",
5510			cmd->sc_data_direction);
5511		BUG();
5512		break;
5513	}
5514
5515	if (hpsa_scatter_gather(h, c, cmd) < 0) { /* Fill SG list */
5516		hpsa_cmd_resolve_and_free(h, c);
5517		return SCSI_MLQUEUE_HOST_BUSY;
5518	}
5519
5520	if (dev->in_reset) {
5521		hpsa_cmd_resolve_and_free(h, c);
5522		return SCSI_MLQUEUE_HOST_BUSY;
5523	}
5524
5525	c->device = dev;
5526
5527	enqueue_cmd_and_start_io(h, c);
5528	/* the cmd'll come back via intr handler in complete_scsi_command()  */
5529	return 0;
5530}
5531
5532static void hpsa_cmd_init(struct ctlr_info *h, int index,
5533				struct CommandList *c)
5534{
5535	dma_addr_t cmd_dma_handle, err_dma_handle;
5536
5537	/* Zero out all of commandlist except the last field, refcount */
5538	memset(c, 0, offsetof(struct CommandList, refcount));
5539	c->Header.tag = cpu_to_le64((u64) (index << DIRECT_LOOKUP_SHIFT));
5540	cmd_dma_handle = h->cmd_pool_dhandle + index * sizeof(*c);
5541	c->err_info = h->errinfo_pool + index;
5542	memset(c->err_info, 0, sizeof(*c->err_info));
5543	err_dma_handle = h->errinfo_pool_dhandle
5544	    + index * sizeof(*c->err_info);
5545	c->cmdindex = index;
5546	c->busaddr = (u32) cmd_dma_handle;
5547	c->ErrDesc.Addr = cpu_to_le64((u64) err_dma_handle);
5548	c->ErrDesc.Len = cpu_to_le32((u32) sizeof(*c->err_info));
5549	c->h = h;
5550	c->scsi_cmd = SCSI_CMD_IDLE;
5551}
5552
5553static void hpsa_preinitialize_commands(struct ctlr_info *h)
5554{
5555	int i;
5556
5557	for (i = 0; i < h->nr_cmds; i++) {
5558		struct CommandList *c = h->cmd_pool + i;
5559
5560		hpsa_cmd_init(h, i, c);
5561		atomic_set(&c->refcount, 0);
5562	}
5563}
5564
5565static inline void hpsa_cmd_partial_init(struct ctlr_info *h, int index,
5566				struct CommandList *c)
5567{
5568	dma_addr_t cmd_dma_handle = h->cmd_pool_dhandle + index * sizeof(*c);
5569
5570	BUG_ON(c->cmdindex != index);
5571
5572	memset(c->Request.CDB, 0, sizeof(c->Request.CDB));
5573	memset(c->err_info, 0, sizeof(*c->err_info));
5574	c->busaddr = (u32) cmd_dma_handle;
5575}
5576
5577static int hpsa_ioaccel_submit(struct ctlr_info *h,
5578		struct CommandList *c, struct scsi_cmnd *cmd)
5579{
5580	struct hpsa_scsi_dev_t *dev = cmd->device->hostdata;
5581	int rc = IO_ACCEL_INELIGIBLE;
5582
5583	if (!dev)
5584		return SCSI_MLQUEUE_HOST_BUSY;
5585
5586	if (dev->in_reset)
5587		return SCSI_MLQUEUE_HOST_BUSY;
5588
5589	if (hpsa_simple_mode)
5590		return IO_ACCEL_INELIGIBLE;
5591
5592	cmd->host_scribble = (unsigned char *) c;
5593
5594	if (dev->offload_enabled) {
5595		hpsa_cmd_init(h, c->cmdindex, c);
5596		c->cmd_type = CMD_SCSI;
5597		c->scsi_cmd = cmd;
5598		c->device = dev;
5599		rc = hpsa_scsi_ioaccel_raid_map(h, c);
5600		if (rc < 0)     /* scsi_dma_map failed. */
5601			rc = SCSI_MLQUEUE_HOST_BUSY;
5602	} else if (dev->hba_ioaccel_enabled) {
5603		hpsa_cmd_init(h, c->cmdindex, c);
5604		c->cmd_type = CMD_SCSI;
5605		c->scsi_cmd = cmd;
5606		c->device = dev;
5607		rc = hpsa_scsi_ioaccel_direct_map(h, c);
5608		if (rc < 0)     /* scsi_dma_map failed. */
5609			rc = SCSI_MLQUEUE_HOST_BUSY;
5610	}
5611	return rc;
5612}
5613
5614static void hpsa_command_resubmit_worker(struct work_struct *work)
5615{
5616	struct scsi_cmnd *cmd;
5617	struct hpsa_scsi_dev_t *dev;
5618	struct CommandList *c = container_of(work, struct CommandList, work);
5619
5620	cmd = c->scsi_cmd;
5621	dev = cmd->device->hostdata;
5622	if (!dev) {
5623		cmd->result = DID_NO_CONNECT << 16;
5624		return hpsa_cmd_free_and_done(c->h, c, cmd);
5625	}
5626
5627	if (dev->in_reset) {
5628		cmd->result = DID_RESET << 16;
5629		return hpsa_cmd_free_and_done(c->h, c, cmd);
5630	}
5631
5632	if (c->cmd_type == CMD_IOACCEL2) {
5633		struct ctlr_info *h = c->h;
5634		struct io_accel2_cmd *c2 = &h->ioaccel2_cmd_pool[c->cmdindex];
5635		int rc;
5636
5637		if (c2->error_data.serv_response ==
5638				IOACCEL2_STATUS_SR_TASK_COMP_SET_FULL) {
5639			rc = hpsa_ioaccel_submit(h, c, cmd);
5640			if (rc == 0)
5641				return;
5642			if (rc == SCSI_MLQUEUE_HOST_BUSY) {
5643				/*
5644				 * If we get here, it means dma mapping failed.
5645				 * Try again via scsi mid layer, which will
5646				 * then get SCSI_MLQUEUE_HOST_BUSY.
5647				 */
5648				cmd->result = DID_IMM_RETRY << 16;
5649				return hpsa_cmd_free_and_done(h, c, cmd);
5650			}
5651			/* else, fall thru and resubmit down CISS path */
5652		}
5653	}
5654	hpsa_cmd_partial_init(c->h, c->cmdindex, c);
5655	if (hpsa_ciss_submit(c->h, c, cmd, dev)) {
5656		/*
5657		 * If we get here, it means dma mapping failed. Try
5658		 * again via scsi mid layer, which will then get
5659		 * SCSI_MLQUEUE_HOST_BUSY.
5660		 *
5661		 * hpsa_ciss_submit will have already freed c
5662		 * if it encountered a dma mapping failure.
5663		 */
5664		cmd->result = DID_IMM_RETRY << 16;
5665		cmd->scsi_done(cmd);
5666	}
5667}
5668
5669/* Running in struct Scsi_Host->host_lock less mode */
5670static int hpsa_scsi_queue_command(struct Scsi_Host *sh, struct scsi_cmnd *cmd)
5671{
5672	struct ctlr_info *h;
5673	struct hpsa_scsi_dev_t *dev;
5674	struct CommandList *c;
5675	int rc = 0;
5676
5677	/* Get the ptr to our adapter structure out of cmd->host. */
5678	h = sdev_to_hba(cmd->device);
5679
5680	BUG_ON(cmd->request->tag < 0);
5681
5682	dev = cmd->device->hostdata;
5683	if (!dev) {
5684		cmd->result = DID_NO_CONNECT << 16;
5685		cmd->scsi_done(cmd);
5686		return 0;
5687	}
5688
5689	if (dev->removed) {
5690		cmd->result = DID_NO_CONNECT << 16;
5691		cmd->scsi_done(cmd);
5692		return 0;
5693	}
5694
5695	if (unlikely(lockup_detected(h))) {
5696		cmd->result = DID_NO_CONNECT << 16;
5697		cmd->scsi_done(cmd);
5698		return 0;
5699	}
5700
5701	if (dev->in_reset)
5702		return SCSI_MLQUEUE_DEVICE_BUSY;
5703
5704	c = cmd_tagged_alloc(h, cmd);
5705	if (c == NULL)
5706		return SCSI_MLQUEUE_DEVICE_BUSY;
5707
5708	/*
5709	 * This is necessary because the SML doesn't zero out this field during
5710	 * error recovery.
5711	 */
5712	cmd->result = 0;
5713
5714	/*
5715	 * Call alternate submit routine for I/O accelerated commands.
5716	 * Retries always go down the normal I/O path.
5717	 */
5718	if (likely(cmd->retries == 0 &&
5719			!blk_rq_is_passthrough(cmd->request) &&
5720			h->acciopath_status)) {
5721		rc = hpsa_ioaccel_submit(h, c, cmd);
5722		if (rc == 0)
5723			return 0;
5724		if (rc == SCSI_MLQUEUE_HOST_BUSY) {
5725			hpsa_cmd_resolve_and_free(h, c);
5726			return SCSI_MLQUEUE_HOST_BUSY;
5727		}
5728	}
5729	return hpsa_ciss_submit(h, c, cmd, dev);
5730}
5731
5732static void hpsa_scan_complete(struct ctlr_info *h)
5733{
5734	unsigned long flags;
5735
5736	spin_lock_irqsave(&h->scan_lock, flags);
5737	h->scan_finished = 1;
5738	wake_up(&h->scan_wait_queue);
5739	spin_unlock_irqrestore(&h->scan_lock, flags);
5740}
5741
5742static void hpsa_scan_start(struct Scsi_Host *sh)
5743{
5744	struct ctlr_info *h = shost_to_hba(sh);
5745	unsigned long flags;
5746
5747	/*
5748	 * Don't let rescans be initiated on a controller known to be locked
5749	 * up.  If the controller locks up *during* a rescan, that thread is
5750	 * probably hosed, but at least we can prevent new rescan threads from
5751	 * piling up on a locked up controller.
5752	 */
5753	if (unlikely(lockup_detected(h)))
5754		return hpsa_scan_complete(h);
5755
5756	/*
5757	 * If a scan is already waiting to run, no need to add another
5758	 */
5759	spin_lock_irqsave(&h->scan_lock, flags);
5760	if (h->scan_waiting) {
5761		spin_unlock_irqrestore(&h->scan_lock, flags);
5762		return;
5763	}
5764
5765	spin_unlock_irqrestore(&h->scan_lock, flags);
5766
5767	/* wait until any scan already in progress is finished. */
5768	while (1) {
5769		spin_lock_irqsave(&h->scan_lock, flags);
5770		if (h->scan_finished)
5771			break;
5772		h->scan_waiting = 1;
5773		spin_unlock_irqrestore(&h->scan_lock, flags);
5774		wait_event(h->scan_wait_queue, h->scan_finished);
5775		/* Note: We don't need to worry about a race between this
5776		 * thread and driver unload because the midlayer will
5777		 * have incremented the reference count, so unload won't
5778		 * happen if we're in here.
5779		 */
5780	}
5781	h->scan_finished = 0; /* mark scan as in progress */
5782	h->scan_waiting = 0;
5783	spin_unlock_irqrestore(&h->scan_lock, flags);
5784
5785	if (unlikely(lockup_detected(h)))
5786		return hpsa_scan_complete(h);
5787
5788	/*
5789	 * Do the scan after a reset completion
5790	 */
5791	spin_lock_irqsave(&h->reset_lock, flags);
5792	if (h->reset_in_progress) {
5793		h->drv_req_rescan = 1;
5794		spin_unlock_irqrestore(&h->reset_lock, flags);
5795		hpsa_scan_complete(h);
5796		return;
5797	}
5798	spin_unlock_irqrestore(&h->reset_lock, flags);
5799
5800	hpsa_update_scsi_devices(h);
5801
5802	hpsa_scan_complete(h);
5803}
5804
5805static int hpsa_change_queue_depth(struct scsi_device *sdev, int qdepth)
5806{
5807	struct hpsa_scsi_dev_t *logical_drive = sdev->hostdata;
5808
5809	if (!logical_drive)
5810		return -ENODEV;
5811
5812	if (qdepth < 1)
5813		qdepth = 1;
5814	else if (qdepth > logical_drive->queue_depth)
5815		qdepth = logical_drive->queue_depth;
5816
5817	return scsi_change_queue_depth(sdev, qdepth);
5818}
5819
5820static int hpsa_scan_finished(struct Scsi_Host *sh,
5821	unsigned long elapsed_time)
5822{
5823	struct ctlr_info *h = shost_to_hba(sh);
5824	unsigned long flags;
5825	int finished;
5826
5827	spin_lock_irqsave(&h->scan_lock, flags);
5828	finished = h->scan_finished;
5829	spin_unlock_irqrestore(&h->scan_lock, flags);
5830	return finished;
5831}
5832
5833static int hpsa_scsi_host_alloc(struct ctlr_info *h)
5834{
5835	struct Scsi_Host *sh;
5836
5837	sh = scsi_host_alloc(&hpsa_driver_template, sizeof(struct ctlr_info));
5838	if (sh == NULL) {
5839		dev_err(&h->pdev->dev, "scsi_host_alloc failed\n");
5840		return -ENOMEM;
5841	}
5842
5843	sh->io_port = 0;
5844	sh->n_io_port = 0;
5845	sh->this_id = -1;
5846	sh->max_channel = 3;
5847	sh->max_cmd_len = MAX_COMMAND_SIZE;
5848	sh->max_lun = HPSA_MAX_LUN;
5849	sh->max_id = HPSA_MAX_LUN;
5850	sh->can_queue = h->nr_cmds - HPSA_NRESERVED_CMDS;
5851	sh->cmd_per_lun = sh->can_queue;
5852	sh->sg_tablesize = h->maxsgentries;
5853	sh->transportt = hpsa_sas_transport_template;
5854	sh->hostdata[0] = (unsigned long) h;
5855	sh->irq = pci_irq_vector(h->pdev, 0);
5856	sh->unique_id = sh->irq;
5857
5858	h->scsi_host = sh;
5859	return 0;
5860}
5861
5862static int hpsa_scsi_add_host(struct ctlr_info *h)
5863{
5864	int rv;
5865
5866	rv = scsi_add_host(h->scsi_host, &h->pdev->dev);
5867	if (rv) {
5868		dev_err(&h->pdev->dev, "scsi_add_host failed\n");
5869		return rv;
5870	}
5871	scsi_scan_host(h->scsi_host);
5872	return 0;
5873}
5874
5875/*
5876 * The block layer has already gone to the trouble of picking out a unique,
5877 * small-integer tag for this request.  We use an offset from that value as
5878 * an index to select our command block.  (The offset allows us to reserve the
5879 * low-numbered entries for our own uses.)
5880 */
5881static int hpsa_get_cmd_index(struct scsi_cmnd *scmd)
5882{
5883	int idx = scmd->request->tag;
5884
5885	if (idx < 0)
5886		return idx;
5887
5888	/* Offset to leave space for internal cmds. */
5889	return idx += HPSA_NRESERVED_CMDS;
5890}
5891
5892/*
5893 * Send a TEST_UNIT_READY command to the specified LUN using the specified
5894 * reply queue; returns zero if the unit is ready, and non-zero otherwise.
5895 */
5896static int hpsa_send_test_unit_ready(struct ctlr_info *h,
5897				struct CommandList *c, unsigned char lunaddr[],
5898				int reply_queue)
5899{
5900	int rc;
5901
5902	/* Send the Test Unit Ready, fill_cmd can't fail, no mapping */
5903	(void) fill_cmd(c, TEST_UNIT_READY, h,
5904			NULL, 0, 0, lunaddr, TYPE_CMD);
5905	rc = hpsa_scsi_do_simple_cmd(h, c, reply_queue, NO_TIMEOUT);
5906	if (rc)
5907		return rc;
5908	/* no unmap needed here because no data xfer. */
5909
5910	/* Check if the unit is already ready. */
5911	if (c->err_info->CommandStatus == CMD_SUCCESS)
5912		return 0;
5913
5914	/*
5915	 * The first command sent after reset will receive "unit attention" to
5916	 * indicate that the LUN has been reset...this is actually what we're
5917	 * looking for (but, success is good too).
5918	 */
5919	if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
5920		c->err_info->ScsiStatus == SAM_STAT_CHECK_CONDITION &&
5921			(c->err_info->SenseInfo[2] == NO_SENSE ||
5922			 c->err_info->SenseInfo[2] == UNIT_ATTENTION))
5923		return 0;
5924
5925	return 1;
5926}
5927
5928/*
5929 * Wait for a TEST_UNIT_READY command to complete, retrying as necessary;
5930 * returns zero when the unit is ready, and non-zero when giving up.
5931 */
5932static int hpsa_wait_for_test_unit_ready(struct ctlr_info *h,
5933				struct CommandList *c,
5934				unsigned char lunaddr[], int reply_queue)
5935{
5936	int rc;
5937	int count = 0;
5938	int waittime = 1; /* seconds */
5939
5940	/* Send test unit ready until device ready, or give up. */
5941	for (count = 0; count < HPSA_TUR_RETRY_LIMIT; count++) {
5942
5943		/*
5944		 * Wait for a bit.  do this first, because if we send
5945		 * the TUR right away, the reset will just abort it.
5946		 */
5947		msleep(1000 * waittime);
5948
5949		rc = hpsa_send_test_unit_ready(h, c, lunaddr, reply_queue);
5950		if (!rc)
5951			break;
5952
5953		/* Increase wait time with each try, up to a point. */
5954		if (waittime < HPSA_MAX_WAIT_INTERVAL_SECS)
5955			waittime *= 2;
5956
5957		dev_warn(&h->pdev->dev,
5958			 "waiting %d secs for device to become ready.\n",
5959			 waittime);
5960	}
5961
5962	return rc;
5963}
5964
5965static int wait_for_device_to_become_ready(struct ctlr_info *h,
5966					   unsigned char lunaddr[],
5967					   int reply_queue)
5968{
5969	int first_queue;
5970	int last_queue;
5971	int rq;
5972	int rc = 0;
5973	struct CommandList *c;
5974
5975	c = cmd_alloc(h);
5976
5977	/*
5978	 * If no specific reply queue was requested, then send the TUR
5979	 * repeatedly, requesting a reply on each reply queue; otherwise execute
5980	 * the loop exactly once using only the specified queue.
5981	 */
5982	if (reply_queue == DEFAULT_REPLY_QUEUE) {
5983		first_queue = 0;
5984		last_queue = h->nreply_queues - 1;
5985	} else {
5986		first_queue = reply_queue;
5987		last_queue = reply_queue;
5988	}
5989
5990	for (rq = first_queue; rq <= last_queue; rq++) {
5991		rc = hpsa_wait_for_test_unit_ready(h, c, lunaddr, rq);
5992		if (rc)
5993			break;
5994	}
5995
5996	if (rc)
5997		dev_warn(&h->pdev->dev, "giving up on device.\n");
5998	else
5999		dev_warn(&h->pdev->dev, "device is ready.\n");
6000
6001	cmd_free(h, c);
6002	return rc;
6003}
6004
6005/* Need at least one of these error handlers to keep ../scsi/hosts.c from
6006 * complaining.  Doing a host- or bus-reset can't do anything good here.
6007 */
6008static int hpsa_eh_device_reset_handler(struct scsi_cmnd *scsicmd)
6009{
6010	int rc = SUCCESS;
6011	int i;
6012	struct ctlr_info *h;
6013	struct hpsa_scsi_dev_t *dev = NULL;
6014	u8 reset_type;
6015	char msg[48];
6016	unsigned long flags;
6017
6018	/* find the controller to which the command to be aborted was sent */
6019	h = sdev_to_hba(scsicmd->device);
6020	if (h == NULL) /* paranoia */
6021		return FAILED;
6022
6023	spin_lock_irqsave(&h->reset_lock, flags);
6024	h->reset_in_progress = 1;
6025	spin_unlock_irqrestore(&h->reset_lock, flags);
6026
6027	if (lockup_detected(h)) {
6028		rc = FAILED;
6029		goto return_reset_status;
6030	}
6031
6032	dev = scsicmd->device->hostdata;
6033	if (!dev) {
6034		dev_err(&h->pdev->dev, "%s: device lookup failed\n", __func__);
6035		rc = FAILED;
6036		goto return_reset_status;
6037	}
6038
6039	if (dev->devtype == TYPE_ENCLOSURE) {
6040		rc = SUCCESS;
6041		goto return_reset_status;
6042	}
6043
6044	/* if controller locked up, we can guarantee command won't complete */
6045	if (lockup_detected(h)) {
6046		snprintf(msg, sizeof(msg),
6047			 "cmd %d RESET FAILED, lockup detected",
6048			 hpsa_get_cmd_index(scsicmd));
6049		hpsa_show_dev_msg(KERN_WARNING, h, dev, msg);
6050		rc = FAILED;
6051		goto return_reset_status;
6052	}
6053
6054	/* this reset request might be the result of a lockup; check */
6055	if (detect_controller_lockup(h)) {
6056		snprintf(msg, sizeof(msg),
6057			 "cmd %d RESET FAILED, new lockup detected",
6058			 hpsa_get_cmd_index(scsicmd));
6059		hpsa_show_dev_msg(KERN_WARNING, h, dev, msg);
6060		rc = FAILED;
6061		goto return_reset_status;
6062	}
6063
6064	/* Do not attempt on controller */
6065	if (is_hba_lunid(dev->scsi3addr)) {
6066		rc = SUCCESS;
6067		goto return_reset_status;
6068	}
6069
6070	if (is_logical_dev_addr_mode(dev->scsi3addr))
6071		reset_type = HPSA_DEVICE_RESET_MSG;
6072	else
6073		reset_type = HPSA_PHYS_TARGET_RESET;
6074
6075	sprintf(msg, "resetting %s",
6076		reset_type == HPSA_DEVICE_RESET_MSG ? "logical " : "physical ");
6077	hpsa_show_dev_msg(KERN_WARNING, h, dev, msg);
6078
6079	/*
6080	 * wait to see if any commands will complete before sending reset
6081	 */
6082	dev->in_reset = true; /* block any new cmds from OS for this device */
6083	for (i = 0; i < 10; i++) {
6084		if (atomic_read(&dev->commands_outstanding) > 0)
6085			msleep(1000);
6086		else
6087			break;
6088	}
6089
6090	/* send a reset to the SCSI LUN which the command was sent to */
6091	rc = hpsa_do_reset(h, dev, reset_type, DEFAULT_REPLY_QUEUE);
6092	if (rc == 0)
6093		rc = SUCCESS;
6094	else
6095		rc = FAILED;
6096
6097	sprintf(msg, "reset %s %s",
6098		reset_type == HPSA_DEVICE_RESET_MSG ? "logical " : "physical ",
6099		rc == SUCCESS ? "completed successfully" : "failed");
6100	hpsa_show_dev_msg(KERN_WARNING, h, dev, msg);
6101
6102return_reset_status:
6103	spin_lock_irqsave(&h->reset_lock, flags);
6104	h->reset_in_progress = 0;
6105	if (dev)
6106		dev->in_reset = false;
6107	spin_unlock_irqrestore(&h->reset_lock, flags);
6108	return rc;
6109}
6110
6111/*
6112 * For operations with an associated SCSI command, a command block is allocated
6113 * at init, and managed by cmd_tagged_alloc() and cmd_tagged_free() using the
6114 * block request tag as an index into a table of entries.  cmd_tagged_free() is
6115 * the complement, although cmd_free() may be called instead.
6116 */
6117static struct CommandList *cmd_tagged_alloc(struct ctlr_info *h,
6118					    struct scsi_cmnd *scmd)
6119{
6120	int idx = hpsa_get_cmd_index(scmd);
6121	struct CommandList *c = h->cmd_pool + idx;
6122
6123	if (idx < HPSA_NRESERVED_CMDS || idx >= h->nr_cmds) {
6124		dev_err(&h->pdev->dev, "Bad block tag: %d not in [%d..%d]\n",
6125			idx, HPSA_NRESERVED_CMDS, h->nr_cmds - 1);
6126		/* The index value comes from the block layer, so if it's out of
6127		 * bounds, it's probably not our bug.
6128		 */
6129		BUG();
6130	}
6131
6132	if (unlikely(!hpsa_is_cmd_idle(c))) {
6133		/*
6134		 * We expect that the SCSI layer will hand us a unique tag
6135		 * value.  Thus, there should never be a collision here between
6136		 * two requests...because if the selected command isn't idle
6137		 * then someone is going to be very disappointed.
6138		 */
6139		if (idx != h->last_collision_tag) { /* Print once per tag */
6140			dev_warn(&h->pdev->dev,
6141				"%s: tag collision (tag=%d)\n", __func__, idx);
6142			if (scmd)
6143				scsi_print_command(scmd);
6144			h->last_collision_tag = idx;
6145		}
6146		return NULL;
6147	}
6148
6149	atomic_inc(&c->refcount);
6150
6151	hpsa_cmd_partial_init(h, idx, c);
6152	return c;
6153}
6154
6155static void cmd_tagged_free(struct ctlr_info *h, struct CommandList *c)
6156{
6157	/*
6158	 * Release our reference to the block.  We don't need to do anything
6159	 * else to free it, because it is accessed by index.
6160	 */
6161	(void)atomic_dec(&c->refcount);
6162}
6163
6164/*
6165 * For operations that cannot sleep, a command block is allocated at init,
6166 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
6167 * which ones are free or in use.  Lock must be held when calling this.
6168 * cmd_free() is the complement.
6169 * This function never gives up and returns NULL.  If it hangs,
6170 * another thread must call cmd_free() to free some tags.
6171 */
6172
6173static struct CommandList *cmd_alloc(struct ctlr_info *h)
6174{
6175	struct CommandList *c;
6176	int refcount, i;
6177	int offset = 0;
6178
6179	/*
6180	 * There is some *extremely* small but non-zero chance that that
6181	 * multiple threads could get in here, and one thread could
6182	 * be scanning through the list of bits looking for a free
6183	 * one, but the free ones are always behind him, and other
6184	 * threads sneak in behind him and eat them before he can
6185	 * get to them, so that while there is always a free one, a
6186	 * very unlucky thread might be starved anyway, never able to
6187	 * beat the other threads.  In reality, this happens so
6188	 * infrequently as to be indistinguishable from never.
6189	 *
6190	 * Note that we start allocating commands before the SCSI host structure
6191	 * is initialized.  Since the search starts at bit zero, this
6192	 * all works, since we have at least one command structure available;
6193	 * however, it means that the structures with the low indexes have to be
6194	 * reserved for driver-initiated requests, while requests from the block
6195	 * layer will use the higher indexes.
6196	 */
6197
6198	for (;;) {
6199		i = find_next_zero_bit(h->cmd_pool_bits,
6200					HPSA_NRESERVED_CMDS,
6201					offset);
6202		if (unlikely(i >= HPSA_NRESERVED_CMDS)) {
6203			offset = 0;
6204			continue;
6205		}
6206		c = h->cmd_pool + i;
6207		refcount = atomic_inc_return(&c->refcount);
6208		if (unlikely(refcount > 1)) {
6209			cmd_free(h, c); /* already in use */
6210			offset = (i + 1) % HPSA_NRESERVED_CMDS;
6211			continue;
6212		}
6213		set_bit(i & (BITS_PER_LONG - 1),
6214			h->cmd_pool_bits + (i / BITS_PER_LONG));
6215		break; /* it's ours now. */
6216	}
6217	hpsa_cmd_partial_init(h, i, c);
6218	c->device = NULL;
6219	return c;
6220}
6221
6222/*
6223 * This is the complementary operation to cmd_alloc().  Note, however, in some
6224 * corner cases it may also be used to free blocks allocated by
6225 * cmd_tagged_alloc() in which case the ref-count decrement does the trick and
6226 * the clear-bit is harmless.
6227 */
6228static void cmd_free(struct ctlr_info *h, struct CommandList *c)
6229{
6230	if (atomic_dec_and_test(&c->refcount)) {
6231		int i;
6232
6233		i = c - h->cmd_pool;
6234		clear_bit(i & (BITS_PER_LONG - 1),
6235			  h->cmd_pool_bits + (i / BITS_PER_LONG));
6236	}
6237}
6238
6239#ifdef CONFIG_COMPAT
6240
6241static int hpsa_ioctl32_passthru(struct scsi_device *dev, unsigned int cmd,
6242	void __user *arg)
6243{
6244	struct ctlr_info *h = sdev_to_hba(dev);
6245	IOCTL32_Command_struct __user *arg32 = arg;
6246	IOCTL_Command_struct arg64;
6247	int err;
6248	u32 cp;
6249
6250	if (!arg)
6251		return -EINVAL;
6252
6253	memset(&arg64, 0, sizeof(arg64));
6254	if (copy_from_user(&arg64, arg32, offsetof(IOCTL_Command_struct, buf)))
6255		return -EFAULT;
6256	if (get_user(cp, &arg32->buf))
6257		return -EFAULT;
6258	arg64.buf = compat_ptr(cp);
6259
6260	if (atomic_dec_if_positive(&h->passthru_cmds_avail) < 0)
6261		return -EAGAIN;
6262	err = hpsa_passthru_ioctl(h, &arg64);
6263	atomic_inc(&h->passthru_cmds_avail);
6264	if (err)
6265		return err;
6266	if (copy_to_user(&arg32->error_info, &arg64.error_info,
6267			 sizeof(arg32->error_info)))
6268		return -EFAULT;
6269	return 0;
6270}
6271
6272static int hpsa_ioctl32_big_passthru(struct scsi_device *dev,
6273	unsigned int cmd, void __user *arg)
6274{
6275	struct ctlr_info *h = sdev_to_hba(dev);
6276	BIG_IOCTL32_Command_struct __user *arg32 = arg;
6277	BIG_IOCTL_Command_struct arg64;
6278	int err;
6279	u32 cp;
6280
6281	if (!arg)
6282		return -EINVAL;
6283	memset(&arg64, 0, sizeof(arg64));
6284	if (copy_from_user(&arg64, arg32,
6285			   offsetof(BIG_IOCTL32_Command_struct, buf)))
6286		return -EFAULT;
6287	if (get_user(cp, &arg32->buf))
6288		return -EFAULT;
6289	arg64.buf = compat_ptr(cp);
6290
6291	if (atomic_dec_if_positive(&h->passthru_cmds_avail) < 0)
6292		return -EAGAIN;
6293	err = hpsa_big_passthru_ioctl(h, &arg64);
6294	atomic_inc(&h->passthru_cmds_avail);
6295	if (err)
6296		return err;
6297	if (copy_to_user(&arg32->error_info, &arg64.error_info,
6298			 sizeof(arg32->error_info)))
6299		return -EFAULT;
6300	return 0;
6301}
6302
6303static int hpsa_compat_ioctl(struct scsi_device *dev, unsigned int cmd,
6304			     void __user *arg)
6305{
6306	switch (cmd) {
6307	case CCISS_GETPCIINFO:
6308	case CCISS_GETINTINFO:
6309	case CCISS_SETINTINFO:
6310	case CCISS_GETNODENAME:
6311	case CCISS_SETNODENAME:
6312	case CCISS_GETHEARTBEAT:
6313	case CCISS_GETBUSTYPES:
6314	case CCISS_GETFIRMVER:
6315	case CCISS_GETDRIVVER:
6316	case CCISS_REVALIDVOLS:
6317	case CCISS_DEREGDISK:
6318	case CCISS_REGNEWDISK:
6319	case CCISS_REGNEWD:
6320	case CCISS_RESCANDISK:
6321	case CCISS_GETLUNINFO:
6322		return hpsa_ioctl(dev, cmd, arg);
6323
6324	case CCISS_PASSTHRU32:
6325		return hpsa_ioctl32_passthru(dev, cmd, arg);
6326	case CCISS_BIG_PASSTHRU32:
6327		return hpsa_ioctl32_big_passthru(dev, cmd, arg);
6328
6329	default:
6330		return -ENOIOCTLCMD;
6331	}
6332}
6333#endif
6334
6335static int hpsa_getpciinfo_ioctl(struct ctlr_info *h, void __user *argp)
6336{
6337	struct hpsa_pci_info pciinfo;
6338
6339	if (!argp)
6340		return -EINVAL;
6341	pciinfo.domain = pci_domain_nr(h->pdev->bus);
6342	pciinfo.bus = h->pdev->bus->number;
6343	pciinfo.dev_fn = h->pdev->devfn;
6344	pciinfo.board_id = h->board_id;
6345	if (copy_to_user(argp, &pciinfo, sizeof(pciinfo)))
6346		return -EFAULT;
6347	return 0;
6348}
6349
6350static int hpsa_getdrivver_ioctl(struct ctlr_info *h, void __user *argp)
6351{
6352	DriverVer_type DriverVer;
6353	unsigned char vmaj, vmin, vsubmin;
6354	int rc;
6355
6356	rc = sscanf(HPSA_DRIVER_VERSION, "%hhu.%hhu.%hhu",
6357		&vmaj, &vmin, &vsubmin);
6358	if (rc != 3) {
6359		dev_info(&h->pdev->dev, "driver version string '%s' "
6360			"unrecognized.", HPSA_DRIVER_VERSION);
6361		vmaj = 0;
6362		vmin = 0;
6363		vsubmin = 0;
6364	}
6365	DriverVer = (vmaj << 16) | (vmin << 8) | vsubmin;
6366	if (!argp)
6367		return -EINVAL;
6368	if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
6369		return -EFAULT;
6370	return 0;
6371}
6372
6373static int hpsa_passthru_ioctl(struct ctlr_info *h,
6374			       IOCTL_Command_struct *iocommand)
6375{
6376	struct CommandList *c;
6377	char *buff = NULL;
6378	u64 temp64;
6379	int rc = 0;
6380
6381	if (!capable(CAP_SYS_RAWIO))
6382		return -EPERM;
6383	if ((iocommand->buf_size < 1) &&
6384	    (iocommand->Request.Type.Direction != XFER_NONE)) {
6385		return -EINVAL;
6386	}
6387	if (iocommand->buf_size > 0) {
6388		buff = kmalloc(iocommand->buf_size, GFP_KERNEL);
6389		if (buff == NULL)
6390			return -ENOMEM;
6391		if (iocommand->Request.Type.Direction & XFER_WRITE) {
6392			/* Copy the data into the buffer we created */
6393			if (copy_from_user(buff, iocommand->buf,
6394				iocommand->buf_size)) {
6395				rc = -EFAULT;
6396				goto out_kfree;
6397			}
6398		} else {
6399			memset(buff, 0, iocommand->buf_size);
6400		}
6401	}
6402	c = cmd_alloc(h);
6403
6404	/* Fill in the command type */
6405	c->cmd_type = CMD_IOCTL_PEND;
6406	c->scsi_cmd = SCSI_CMD_BUSY;
6407	/* Fill in Command Header */
6408	c->Header.ReplyQueue = 0; /* unused in simple mode */
6409	if (iocommand->buf_size > 0) {	/* buffer to fill */
6410		c->Header.SGList = 1;
6411		c->Header.SGTotal = cpu_to_le16(1);
6412	} else	{ /* no buffers to fill */
6413		c->Header.SGList = 0;
6414		c->Header.SGTotal = cpu_to_le16(0);
6415	}
6416	memcpy(&c->Header.LUN, &iocommand->LUN_info, sizeof(c->Header.LUN));
6417
6418	/* Fill in Request block */
6419	memcpy(&c->Request, &iocommand->Request,
6420		sizeof(c->Request));
6421
6422	/* Fill in the scatter gather information */
6423	if (iocommand->buf_size > 0) {
6424		temp64 = dma_map_single(&h->pdev->dev, buff,
6425			iocommand->buf_size, DMA_BIDIRECTIONAL);
6426		if (dma_mapping_error(&h->pdev->dev, (dma_addr_t) temp64)) {
6427			c->SG[0].Addr = cpu_to_le64(0);
6428			c->SG[0].Len = cpu_to_le32(0);
6429			rc = -ENOMEM;
6430			goto out;
6431		}
6432		c->SG[0].Addr = cpu_to_le64(temp64);
6433		c->SG[0].Len = cpu_to_le32(iocommand->buf_size);
6434		c->SG[0].Ext = cpu_to_le32(HPSA_SG_LAST); /* not chaining */
6435	}
6436	rc = hpsa_scsi_do_simple_cmd(h, c, DEFAULT_REPLY_QUEUE,
6437					NO_TIMEOUT);
6438	if (iocommand->buf_size > 0)
6439		hpsa_pci_unmap(h->pdev, c, 1, DMA_BIDIRECTIONAL);
6440	check_ioctl_unit_attention(h, c);
6441	if (rc) {
6442		rc = -EIO;
6443		goto out;
6444	}
6445
6446	/* Copy the error information out */
6447	memcpy(&iocommand->error_info, c->err_info,
6448		sizeof(iocommand->error_info));
6449	if ((iocommand->Request.Type.Direction & XFER_READ) &&
6450		iocommand->buf_size > 0) {
6451		/* Copy the data out of the buffer we created */
6452		if (copy_to_user(iocommand->buf, buff, iocommand->buf_size)) {
6453			rc = -EFAULT;
6454			goto out;
6455		}
6456	}
6457out:
6458	cmd_free(h, c);
6459out_kfree:
6460	kfree(buff);
6461	return rc;
6462}
6463
6464static int hpsa_big_passthru_ioctl(struct ctlr_info *h,
6465				   BIG_IOCTL_Command_struct *ioc)
6466{
6467	struct CommandList *c;
6468	unsigned char **buff = NULL;
6469	int *buff_size = NULL;
6470	u64 temp64;
6471	BYTE sg_used = 0;
6472	int status = 0;
6473	u32 left;
6474	u32 sz;
6475	BYTE __user *data_ptr;
6476
6477	if (!capable(CAP_SYS_RAWIO))
6478		return -EPERM;
6479
6480	if ((ioc->buf_size < 1) &&
6481	    (ioc->Request.Type.Direction != XFER_NONE))
6482		return -EINVAL;
6483	/* Check kmalloc limits  using all SGs */
6484	if (ioc->malloc_size > MAX_KMALLOC_SIZE)
6485		return -EINVAL;
6486	if (ioc->buf_size > ioc->malloc_size * SG_ENTRIES_IN_CMD)
6487		return -EINVAL;
6488	buff = kcalloc(SG_ENTRIES_IN_CMD, sizeof(char *), GFP_KERNEL);
6489	if (!buff) {
6490		status = -ENOMEM;
6491		goto cleanup1;
6492	}
6493	buff_size = kmalloc_array(SG_ENTRIES_IN_CMD, sizeof(int), GFP_KERNEL);
6494	if (!buff_size) {
6495		status = -ENOMEM;
6496		goto cleanup1;
6497	}
6498	left = ioc->buf_size;
6499	data_ptr = ioc->buf;
6500	while (left) {
6501		sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
6502		buff_size[sg_used] = sz;
6503		buff[sg_used] = kmalloc(sz, GFP_KERNEL);
6504		if (buff[sg_used] == NULL) {
6505			status = -ENOMEM;
6506			goto cleanup1;
6507		}
6508		if (ioc->Request.Type.Direction & XFER_WRITE) {
6509			if (copy_from_user(buff[sg_used], data_ptr, sz)) {
6510				status = -EFAULT;
6511				goto cleanup1;
6512			}
6513		} else
6514			memset(buff[sg_used], 0, sz);
6515		left -= sz;
6516		data_ptr += sz;
6517		sg_used++;
6518	}
6519	c = cmd_alloc(h);
6520
6521	c->cmd_type = CMD_IOCTL_PEND;
6522	c->scsi_cmd = SCSI_CMD_BUSY;
6523	c->Header.ReplyQueue = 0;
6524	c->Header.SGList = (u8) sg_used;
6525	c->Header.SGTotal = cpu_to_le16(sg_used);
6526	memcpy(&c->Header.LUN, &ioc->LUN_info, sizeof(c->Header.LUN));
6527	memcpy(&c->Request, &ioc->Request, sizeof(c->Request));
6528	if (ioc->buf_size > 0) {
6529		int i;
6530		for (i = 0; i < sg_used; i++) {
6531			temp64 = dma_map_single(&h->pdev->dev, buff[i],
6532				    buff_size[i], DMA_BIDIRECTIONAL);
6533			if (dma_mapping_error(&h->pdev->dev,
6534							(dma_addr_t) temp64)) {
6535				c->SG[i].Addr = cpu_to_le64(0);
6536				c->SG[i].Len = cpu_to_le32(0);
6537				hpsa_pci_unmap(h->pdev, c, i,
6538					DMA_BIDIRECTIONAL);
6539				status = -ENOMEM;
6540				goto cleanup0;
6541			}
6542			c->SG[i].Addr = cpu_to_le64(temp64);
6543			c->SG[i].Len = cpu_to_le32(buff_size[i]);
6544			c->SG[i].Ext = cpu_to_le32(0);
6545		}
6546		c->SG[--i].Ext = cpu_to_le32(HPSA_SG_LAST);
6547	}
6548	status = hpsa_scsi_do_simple_cmd(h, c, DEFAULT_REPLY_QUEUE,
6549						NO_TIMEOUT);
6550	if (sg_used)
6551		hpsa_pci_unmap(h->pdev, c, sg_used, DMA_BIDIRECTIONAL);
6552	check_ioctl_unit_attention(h, c);
6553	if (status) {
6554		status = -EIO;
6555		goto cleanup0;
6556	}
6557
6558	/* Copy the error information out */
6559	memcpy(&ioc->error_info, c->err_info, sizeof(ioc->error_info));
6560	if ((ioc->Request.Type.Direction & XFER_READ) && ioc->buf_size > 0) {
6561		int i;
6562
6563		/* Copy the data out of the buffer we created */
6564		BYTE __user *ptr = ioc->buf;
6565		for (i = 0; i < sg_used; i++) {
6566			if (copy_to_user(ptr, buff[i], buff_size[i])) {
6567				status = -EFAULT;
6568				goto cleanup0;
6569			}
6570			ptr += buff_size[i];
6571		}
6572	}
6573	status = 0;
6574cleanup0:
6575	cmd_free(h, c);
6576cleanup1:
6577	if (buff) {
6578		int i;
6579
6580		for (i = 0; i < sg_used; i++)
6581			kfree(buff[i]);
6582		kfree(buff);
6583	}
6584	kfree(buff_size);
6585	return status;
6586}
6587
6588static void check_ioctl_unit_attention(struct ctlr_info *h,
6589	struct CommandList *c)
6590{
6591	if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
6592			c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
6593		(void) check_for_unit_attention(h, c);
6594}
6595
6596/*
6597 * ioctl
6598 */
6599static int hpsa_ioctl(struct scsi_device *dev, unsigned int cmd,
6600		      void __user *argp)
6601{
6602	struct ctlr_info *h = sdev_to_hba(dev);
6603	int rc;
6604
6605	switch (cmd) {
6606	case CCISS_DEREGDISK:
6607	case CCISS_REGNEWDISK:
6608	case CCISS_REGNEWD:
6609		hpsa_scan_start(h->scsi_host);
6610		return 0;
6611	case CCISS_GETPCIINFO:
6612		return hpsa_getpciinfo_ioctl(h, argp);
6613	case CCISS_GETDRIVVER:
6614		return hpsa_getdrivver_ioctl(h, argp);
6615	case CCISS_PASSTHRU: {
6616		IOCTL_Command_struct iocommand;
6617
6618		if (!argp)
6619			return -EINVAL;
6620		if (copy_from_user(&iocommand, argp, sizeof(iocommand)))
6621			return -EFAULT;
6622		if (atomic_dec_if_positive(&h->passthru_cmds_avail) < 0)
6623			return -EAGAIN;
6624		rc = hpsa_passthru_ioctl(h, &iocommand);
6625		atomic_inc(&h->passthru_cmds_avail);
6626		if (!rc && copy_to_user(argp, &iocommand, sizeof(iocommand)))
6627			rc = -EFAULT;
6628		return rc;
6629	}
6630	case CCISS_BIG_PASSTHRU: {
6631		BIG_IOCTL_Command_struct ioc;
6632		if (!argp)
6633			return -EINVAL;
6634		if (copy_from_user(&ioc, argp, sizeof(ioc)))
6635			return -EFAULT;
6636		if (atomic_dec_if_positive(&h->passthru_cmds_avail) < 0)
6637			return -EAGAIN;
6638		rc = hpsa_big_passthru_ioctl(h, &ioc);
6639		atomic_inc(&h->passthru_cmds_avail);
6640		if (!rc && copy_to_user(argp, &ioc, sizeof(ioc)))
6641			rc = -EFAULT;
6642		return rc;
6643	}
6644	default:
6645		return -ENOTTY;
6646	}
6647}
6648
6649static void hpsa_send_host_reset(struct ctlr_info *h, u8 reset_type)
6650{
6651	struct CommandList *c;
6652
6653	c = cmd_alloc(h);
6654
6655	/* fill_cmd can't fail here, no data buffer to map */
6656	(void) fill_cmd(c, HPSA_DEVICE_RESET_MSG, h, NULL, 0, 0,
6657		RAID_CTLR_LUNID, TYPE_MSG);
6658	c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
6659	c->waiting = NULL;
6660	enqueue_cmd_and_start_io(h, c);
6661	/* Don't wait for completion, the reset won't complete.  Don't free
6662	 * the command either.  This is the last command we will send before
6663	 * re-initializing everything, so it doesn't matter and won't leak.
6664	 */
6665	return;
6666}
6667
6668static int fill_cmd(struct CommandList *c, u8 cmd, struct ctlr_info *h,
6669	void *buff, size_t size, u16 page_code, unsigned char *scsi3addr,
6670	int cmd_type)
6671{
6672	enum dma_data_direction dir = DMA_NONE;
6673
6674	c->cmd_type = CMD_IOCTL_PEND;
6675	c->scsi_cmd = SCSI_CMD_BUSY;
6676	c->Header.ReplyQueue = 0;
6677	if (buff != NULL && size > 0) {
6678		c->Header.SGList = 1;
6679		c->Header.SGTotal = cpu_to_le16(1);
6680	} else {
6681		c->Header.SGList = 0;
6682		c->Header.SGTotal = cpu_to_le16(0);
6683	}
6684	memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
6685
6686	if (cmd_type == TYPE_CMD) {
6687		switch (cmd) {
6688		case HPSA_INQUIRY:
6689			/* are we trying to read a vital product page */
6690			if (page_code & VPD_PAGE) {
6691				c->Request.CDB[1] = 0x01;
6692				c->Request.CDB[2] = (page_code & 0xff);
6693			}
6694			c->Request.CDBLen = 6;
6695			c->Request.type_attr_dir =
6696				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6697			c->Request.Timeout = 0;
6698			c->Request.CDB[0] = HPSA_INQUIRY;
6699			c->Request.CDB[4] = size & 0xFF;
6700			break;
6701		case RECEIVE_DIAGNOSTIC:
6702			c->Request.CDBLen = 6;
6703			c->Request.type_attr_dir =
6704				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6705			c->Request.Timeout = 0;
6706			c->Request.CDB[0] = cmd;
6707			c->Request.CDB[1] = 1;
6708			c->Request.CDB[2] = 1;
6709			c->Request.CDB[3] = (size >> 8) & 0xFF;
6710			c->Request.CDB[4] = size & 0xFF;
6711			break;
6712		case HPSA_REPORT_LOG:
6713		case HPSA_REPORT_PHYS:
6714			/* Talking to controller so It's a physical command
6715			   mode = 00 target = 0.  Nothing to write.
6716			 */
6717			c->Request.CDBLen = 12;
6718			c->Request.type_attr_dir =
6719				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6720			c->Request.Timeout = 0;
6721			c->Request.CDB[0] = cmd;
6722			c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
6723			c->Request.CDB[7] = (size >> 16) & 0xFF;
6724			c->Request.CDB[8] = (size >> 8) & 0xFF;
6725			c->Request.CDB[9] = size & 0xFF;
6726			break;
6727		case BMIC_SENSE_DIAG_OPTIONS:
6728			c->Request.CDBLen = 16;
6729			c->Request.type_attr_dir =
6730				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6731			c->Request.Timeout = 0;
6732			/* Spec says this should be BMIC_WRITE */
6733			c->Request.CDB[0] = BMIC_READ;
6734			c->Request.CDB[6] = BMIC_SENSE_DIAG_OPTIONS;
6735			break;
6736		case BMIC_SET_DIAG_OPTIONS:
6737			c->Request.CDBLen = 16;
6738			c->Request.type_attr_dir =
6739					TYPE_ATTR_DIR(cmd_type,
6740						ATTR_SIMPLE, XFER_WRITE);
6741			c->Request.Timeout = 0;
6742			c->Request.CDB[0] = BMIC_WRITE;
6743			c->Request.CDB[6] = BMIC_SET_DIAG_OPTIONS;
6744			break;
6745		case HPSA_CACHE_FLUSH:
6746			c->Request.CDBLen = 12;
6747			c->Request.type_attr_dir =
6748					TYPE_ATTR_DIR(cmd_type,
6749						ATTR_SIMPLE, XFER_WRITE);
6750			c->Request.Timeout = 0;
6751			c->Request.CDB[0] = BMIC_WRITE;
6752			c->Request.CDB[6] = BMIC_CACHE_FLUSH;
6753			c->Request.CDB[7] = (size >> 8) & 0xFF;
6754			c->Request.CDB[8] = size & 0xFF;
6755			break;
6756		case TEST_UNIT_READY:
6757			c->Request.CDBLen = 6;
6758			c->Request.type_attr_dir =
6759				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_NONE);
6760			c->Request.Timeout = 0;
6761			break;
6762		case HPSA_GET_RAID_MAP:
6763			c->Request.CDBLen = 12;
6764			c->Request.type_attr_dir =
6765				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6766			c->Request.Timeout = 0;
6767			c->Request.CDB[0] = HPSA_CISS_READ;
6768			c->Request.CDB[1] = cmd;
6769			c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
6770			c->Request.CDB[7] = (size >> 16) & 0xFF;
6771			c->Request.CDB[8] = (size >> 8) & 0xFF;
6772			c->Request.CDB[9] = size & 0xFF;
6773			break;
6774		case BMIC_SENSE_CONTROLLER_PARAMETERS:
6775			c->Request.CDBLen = 10;
6776			c->Request.type_attr_dir =
6777				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6778			c->Request.Timeout = 0;
6779			c->Request.CDB[0] = BMIC_READ;
6780			c->Request.CDB[6] = BMIC_SENSE_CONTROLLER_PARAMETERS;
6781			c->Request.CDB[7] = (size >> 16) & 0xFF;
6782			c->Request.CDB[8] = (size >> 8) & 0xFF;
6783			break;
6784		case BMIC_IDENTIFY_PHYSICAL_DEVICE:
6785			c->Request.CDBLen = 10;
6786			c->Request.type_attr_dir =
6787				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6788			c->Request.Timeout = 0;
6789			c->Request.CDB[0] = BMIC_READ;
6790			c->Request.CDB[6] = BMIC_IDENTIFY_PHYSICAL_DEVICE;
6791			c->Request.CDB[7] = (size >> 16) & 0xFF;
6792			c->Request.CDB[8] = (size >> 8) & 0XFF;
6793			break;
6794		case BMIC_SENSE_SUBSYSTEM_INFORMATION:
6795			c->Request.CDBLen = 10;
6796			c->Request.type_attr_dir =
6797				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6798			c->Request.Timeout = 0;
6799			c->Request.CDB[0] = BMIC_READ;
6800			c->Request.CDB[6] = BMIC_SENSE_SUBSYSTEM_INFORMATION;
6801			c->Request.CDB[7] = (size >> 16) & 0xFF;
6802			c->Request.CDB[8] = (size >> 8) & 0XFF;
6803			break;
6804		case BMIC_SENSE_STORAGE_BOX_PARAMS:
6805			c->Request.CDBLen = 10;
6806			c->Request.type_attr_dir =
6807				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6808			c->Request.Timeout = 0;
6809			c->Request.CDB[0] = BMIC_READ;
6810			c->Request.CDB[6] = BMIC_SENSE_STORAGE_BOX_PARAMS;
6811			c->Request.CDB[7] = (size >> 16) & 0xFF;
6812			c->Request.CDB[8] = (size >> 8) & 0XFF;
6813			break;
6814		case BMIC_IDENTIFY_CONTROLLER:
6815			c->Request.CDBLen = 10;
6816			c->Request.type_attr_dir =
6817				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
6818			c->Request.Timeout = 0;
6819			c->Request.CDB[0] = BMIC_READ;
6820			c->Request.CDB[1] = 0;
6821			c->Request.CDB[2] = 0;
6822			c->Request.CDB[3] = 0;
6823			c->Request.CDB[4] = 0;
6824			c->Request.CDB[5] = 0;
6825			c->Request.CDB[6] = BMIC_IDENTIFY_CONTROLLER;
6826			c->Request.CDB[7] = (size >> 16) & 0xFF;
6827			c->Request.CDB[8] = (size >> 8) & 0XFF;
6828			c->Request.CDB[9] = 0;
6829			break;
6830		default:
6831			dev_warn(&h->pdev->dev, "unknown command 0x%c\n", cmd);
6832			BUG();
6833		}
6834	} else if (cmd_type == TYPE_MSG) {
6835		switch (cmd) {
6836
6837		case  HPSA_PHYS_TARGET_RESET:
6838			c->Request.CDBLen = 16;
6839			c->Request.type_attr_dir =
6840				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_NONE);
6841			c->Request.Timeout = 0; /* Don't time out */
6842			memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
6843			c->Request.CDB[0] = HPSA_RESET;
6844			c->Request.CDB[1] = HPSA_TARGET_RESET_TYPE;
6845			/* Physical target reset needs no control bytes 4-7*/
6846			c->Request.CDB[4] = 0x00;
6847			c->Request.CDB[5] = 0x00;
6848			c->Request.CDB[6] = 0x00;
6849			c->Request.CDB[7] = 0x00;
6850			break;
6851		case  HPSA_DEVICE_RESET_MSG:
6852			c->Request.CDBLen = 16;
6853			c->Request.type_attr_dir =
6854				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_NONE);
6855			c->Request.Timeout = 0; /* Don't time out */
6856			memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
6857			c->Request.CDB[0] =  cmd;
6858			c->Request.CDB[1] = HPSA_RESET_TYPE_LUN;
6859			/* If bytes 4-7 are zero, it means reset the */
6860			/* LunID device */
6861			c->Request.CDB[4] = 0x00;
6862			c->Request.CDB[5] = 0x00;
6863			c->Request.CDB[6] = 0x00;
6864			c->Request.CDB[7] = 0x00;
6865			break;
6866		default:
6867			dev_warn(&h->pdev->dev, "unknown message type %d\n",
6868				cmd);
6869			BUG();
6870		}
6871	} else {
6872		dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
6873		BUG();
6874	}
6875
6876	switch (GET_DIR(c->Request.type_attr_dir)) {
6877	case XFER_READ:
6878		dir = DMA_FROM_DEVICE;
6879		break;
6880	case XFER_WRITE:
6881		dir = DMA_TO_DEVICE;
6882		break;
6883	case XFER_NONE:
6884		dir = DMA_NONE;
6885		break;
6886	default:
6887		dir = DMA_BIDIRECTIONAL;
6888	}
6889	if (hpsa_map_one(h->pdev, c, buff, size, dir))
6890		return -1;
6891	return 0;
6892}
6893
6894/*
6895 * Map (physical) PCI mem into (virtual) kernel space
6896 */
6897static void __iomem *remap_pci_mem(ulong base, ulong size)
6898{
6899	ulong page_base = ((ulong) base) & PAGE_MASK;
6900	ulong page_offs = ((ulong) base) - page_base;
6901	void __iomem *page_remapped = ioremap(page_base,
6902		page_offs + size);
6903
6904	return page_remapped ? (page_remapped + page_offs) : NULL;
6905}
6906
6907static inline unsigned long get_next_completion(struct ctlr_info *h, u8 q)
6908{
6909	return h->access.command_completed(h, q);
6910}
6911
6912static inline bool interrupt_pending(struct ctlr_info *h)
6913{
6914	return h->access.intr_pending(h);
6915}
6916
6917static inline long interrupt_not_for_us(struct ctlr_info *h)
6918{
6919	return (h->access.intr_pending(h) == 0) ||
6920		(h->interrupts_enabled == 0);
6921}
6922
6923static inline int bad_tag(struct ctlr_info *h, u32 tag_index,
6924	u32 raw_tag)
6925{
6926	if (unlikely(tag_index >= h->nr_cmds)) {
6927		dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
6928		return 1;
6929	}
6930	return 0;
6931}
6932
6933static inline void finish_cmd(struct CommandList *c)
6934{
6935	dial_up_lockup_detection_on_fw_flash_complete(c->h, c);
6936	if (likely(c->cmd_type == CMD_IOACCEL1 || c->cmd_type == CMD_SCSI
6937			|| c->cmd_type == CMD_IOACCEL2))
6938		complete_scsi_command(c);
6939	else if (c->cmd_type == CMD_IOCTL_PEND || c->cmd_type == IOACCEL2_TMF)
6940		complete(c->waiting);
6941}
6942
6943/* process completion of an indexed ("direct lookup") command */
6944static inline void process_indexed_cmd(struct ctlr_info *h,
6945	u32 raw_tag)
6946{
6947	u32 tag_index;
6948	struct CommandList *c;
6949
6950	tag_index = raw_tag >> DIRECT_LOOKUP_SHIFT;
6951	if (!bad_tag(h, tag_index, raw_tag)) {
6952		c = h->cmd_pool + tag_index;
6953		finish_cmd(c);
6954	}
6955}
6956
6957/* Some controllers, like p400, will give us one interrupt
6958 * after a soft reset, even if we turned interrupts off.
6959 * Only need to check for this in the hpsa_xxx_discard_completions
6960 * functions.
6961 */
6962static int ignore_bogus_interrupt(struct ctlr_info *h)
6963{
6964	if (likely(!reset_devices))
6965		return 0;
6966
6967	if (likely(h->interrupts_enabled))
6968		return 0;
6969
6970	dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
6971		"(known firmware bug.)  Ignoring.\n");
6972
6973	return 1;
6974}
6975
6976/*
6977 * Convert &h->q[x] (passed to interrupt handlers) back to h.
6978 * Relies on (h-q[x] == x) being true for x such that
6979 * 0 <= x < MAX_REPLY_QUEUES.
6980 */
6981static struct ctlr_info *queue_to_hba(u8 *queue)
6982{
6983	return container_of((queue - *queue), struct ctlr_info, q[0]);
6984}
6985
6986static irqreturn_t hpsa_intx_discard_completions(int irq, void *queue)
6987{
6988	struct ctlr_info *h = queue_to_hba(queue);
6989	u8 q = *(u8 *) queue;
6990	u32 raw_tag;
6991
6992	if (ignore_bogus_interrupt(h))
6993		return IRQ_NONE;
6994
6995	if (interrupt_not_for_us(h))
6996		return IRQ_NONE;
6997	h->last_intr_timestamp = get_jiffies_64();
6998	while (interrupt_pending(h)) {
6999		raw_tag = get_next_completion(h, q);
7000		while (raw_tag != FIFO_EMPTY)
7001			raw_tag = next_command(h, q);
7002	}
7003	return IRQ_HANDLED;
7004}
7005
7006static irqreturn_t hpsa_msix_discard_completions(int irq, void *queue)
7007{
7008	struct ctlr_info *h = queue_to_hba(queue);
7009	u32 raw_tag;
7010	u8 q = *(u8 *) queue;
7011
7012	if (ignore_bogus_interrupt(h))
7013		return IRQ_NONE;
7014
7015	h->last_intr_timestamp = get_jiffies_64();
7016	raw_tag = get_next_completion(h, q);
7017	while (raw_tag != FIFO_EMPTY)
7018		raw_tag = next_command(h, q);
7019	return IRQ_HANDLED;
7020}
7021
7022static irqreturn_t do_hpsa_intr_intx(int irq, void *queue)
7023{
7024	struct ctlr_info *h = queue_to_hba((u8 *) queue);
7025	u32 raw_tag;
7026	u8 q = *(u8 *) queue;
7027
7028	if (interrupt_not_for_us(h))
7029		return IRQ_NONE;
7030	h->last_intr_timestamp = get_jiffies_64();
7031	while (interrupt_pending(h)) {
7032		raw_tag = get_next_completion(h, q);
7033		while (raw_tag != FIFO_EMPTY) {
7034			process_indexed_cmd(h, raw_tag);
7035			raw_tag = next_command(h, q);
7036		}
7037	}
7038	return IRQ_HANDLED;
7039}
7040
7041static irqreturn_t do_hpsa_intr_msi(int irq, void *queue)
7042{
7043	struct ctlr_info *h = queue_to_hba(queue);
7044	u32 raw_tag;
7045	u8 q = *(u8 *) queue;
7046
7047	h->last_intr_timestamp = get_jiffies_64();
7048	raw_tag = get_next_completion(h, q);
7049	while (raw_tag != FIFO_EMPTY) {
7050		process_indexed_cmd(h, raw_tag);
7051		raw_tag = next_command(h, q);
7052	}
7053	return IRQ_HANDLED;
7054}
7055
7056/* Send a message CDB to the firmware. Careful, this only works
7057 * in simple mode, not performant mode due to the tag lookup.
7058 * We only ever use this immediately after a controller reset.
7059 */
7060static int hpsa_message(struct pci_dev *pdev, unsigned char opcode,
7061			unsigned char type)
7062{
7063	struct Command {
7064		struct CommandListHeader CommandHeader;
7065		struct RequestBlock Request;
7066		struct ErrDescriptor ErrorDescriptor;
7067	};
7068	struct Command *cmd;
7069	static const size_t cmd_sz = sizeof(*cmd) +
7070					sizeof(cmd->ErrorDescriptor);
7071	dma_addr_t paddr64;
7072	__le32 paddr32;
7073	u32 tag;
7074	void __iomem *vaddr;
7075	int i, err;
7076
7077	vaddr = pci_ioremap_bar(pdev, 0);
7078	if (vaddr == NULL)
7079		return -ENOMEM;
7080
7081	/* The Inbound Post Queue only accepts 32-bit physical addresses for the
7082	 * CCISS commands, so they must be allocated from the lower 4GiB of
7083	 * memory.
7084	 */
7085	err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
7086	if (err) {
7087		iounmap(vaddr);
7088		return err;
7089	}
7090
7091	cmd = dma_alloc_coherent(&pdev->dev, cmd_sz, &paddr64, GFP_KERNEL);
7092	if (cmd == NULL) {
7093		iounmap(vaddr);
7094		return -ENOMEM;
7095	}
7096
7097	/* This must fit, because of the 32-bit consistent DMA mask.  Also,
7098	 * although there's no guarantee, we assume that the address is at
7099	 * least 4-byte aligned (most likely, it's page-aligned).
7100	 */
7101	paddr32 = cpu_to_le32(paddr64);
7102
7103	cmd->CommandHeader.ReplyQueue = 0;
7104	cmd->CommandHeader.SGList = 0;
7105	cmd->CommandHeader.SGTotal = cpu_to_le16(0);
7106	cmd->CommandHeader.tag = cpu_to_le64(paddr64);
7107	memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
7108
7109	cmd->Request.CDBLen = 16;
7110	cmd->Request.type_attr_dir =
7111			TYPE_ATTR_DIR(TYPE_MSG, ATTR_HEADOFQUEUE, XFER_NONE);
7112	cmd->Request.Timeout = 0; /* Don't time out */
7113	cmd->Request.CDB[0] = opcode;
7114	cmd->Request.CDB[1] = type;
7115	memset(&cmd->Request.CDB[2], 0, 14); /* rest of the CDB is reserved */
7116	cmd->ErrorDescriptor.Addr =
7117			cpu_to_le64((le32_to_cpu(paddr32) + sizeof(*cmd)));
7118	cmd->ErrorDescriptor.Len = cpu_to_le32(sizeof(struct ErrorInfo));
7119
7120	writel(le32_to_cpu(paddr32), vaddr + SA5_REQUEST_PORT_OFFSET);
7121
7122	for (i = 0; i < HPSA_MSG_SEND_RETRY_LIMIT; i++) {
7123		tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
7124		if ((tag & ~HPSA_SIMPLE_ERROR_BITS) == paddr64)
7125			break;
7126		msleep(HPSA_MSG_SEND_RETRY_INTERVAL_MSECS);
7127	}
7128
7129	iounmap(vaddr);
7130
7131	/* we leak the DMA buffer here ... no choice since the controller could
7132	 *  still complete the command.
7133	 */
7134	if (i == HPSA_MSG_SEND_RETRY_LIMIT) {
7135		dev_err(&pdev->dev, "controller message %02x:%02x timed out\n",
7136			opcode, type);
7137		return -ETIMEDOUT;
7138	}
7139
7140	dma_free_coherent(&pdev->dev, cmd_sz, cmd, paddr64);
7141
7142	if (tag & HPSA_ERROR_BIT) {
7143		dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
7144			opcode, type);
7145		return -EIO;
7146	}
7147
7148	dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
7149		opcode, type);
7150	return 0;
7151}
7152
7153#define hpsa_noop(p) hpsa_message(p, 3, 0)
7154
7155static int hpsa_controller_hard_reset(struct pci_dev *pdev,
7156	void __iomem *vaddr, u32 use_doorbell)
7157{
7158
7159	if (use_doorbell) {
7160		/* For everything after the P600, the PCI power state method
7161		 * of resetting the controller doesn't work, so we have this
7162		 * other way using the doorbell register.
7163		 */
7164		dev_info(&pdev->dev, "using doorbell to reset controller\n");
7165		writel(use_doorbell, vaddr + SA5_DOORBELL);
7166
7167		/* PMC hardware guys tell us we need a 10 second delay after
7168		 * doorbell reset and before any attempt to talk to the board
7169		 * at all to ensure that this actually works and doesn't fall
7170		 * over in some weird corner cases.
7171		 */
7172		msleep(10000);
7173	} else { /* Try to do it the PCI power state way */
7174
7175		/* Quoting from the Open CISS Specification: "The Power
7176		 * Management Control/Status Register (CSR) controls the power
7177		 * state of the device.  The normal operating state is D0,
7178		 * CSR=00h.  The software off state is D3, CSR=03h.  To reset
7179		 * the controller, place the interface device in D3 then to D0,
7180		 * this causes a secondary PCI reset which will reset the
7181		 * controller." */
7182
7183		int rc = 0;
7184
7185		dev_info(&pdev->dev, "using PCI PM to reset controller\n");
7186
7187		/* enter the D3hot power management state */
7188		rc = pci_set_power_state(pdev, PCI_D3hot);
7189		if (rc)
7190			return rc;
7191
7192		msleep(500);
7193
7194		/* enter the D0 power management state */
7195		rc = pci_set_power_state(pdev, PCI_D0);
7196		if (rc)
7197			return rc;
7198
7199		/*
7200		 * The P600 requires a small delay when changing states.
7201		 * Otherwise we may think the board did not reset and we bail.
7202		 * This for kdump only and is particular to the P600.
7203		 */
7204		msleep(500);
7205	}
7206	return 0;
7207}
7208
7209static void init_driver_version(char *driver_version, int len)
7210{
7211	memset(driver_version, 0, len);
7212	strncpy(driver_version, HPSA " " HPSA_DRIVER_VERSION, len - 1);
7213}
7214
7215static int write_driver_ver_to_cfgtable(struct CfgTable __iomem *cfgtable)
7216{
7217	char *driver_version;
7218	int i, size = sizeof(cfgtable->driver_version);
7219
7220	driver_version = kmalloc(size, GFP_KERNEL);
7221	if (!driver_version)
7222		return -ENOMEM;
7223
7224	init_driver_version(driver_version, size);
7225	for (i = 0; i < size; i++)
7226		writeb(driver_version[i], &cfgtable->driver_version[i]);
7227	kfree(driver_version);
7228	return 0;
7229}
7230
7231static void read_driver_ver_from_cfgtable(struct CfgTable __iomem *cfgtable,
7232					  unsigned char *driver_ver)
7233{
7234	int i;
7235
7236	for (i = 0; i < sizeof(cfgtable->driver_version); i++)
7237		driver_ver[i] = readb(&cfgtable->driver_version[i]);
7238}
7239
7240static int controller_reset_failed(struct CfgTable __iomem *cfgtable)
7241{
7242
7243	char *driver_ver, *old_driver_ver;
7244	int rc, size = sizeof(cfgtable->driver_version);
7245
7246	old_driver_ver = kmalloc_array(2, size, GFP_KERNEL);
7247	if (!old_driver_ver)
7248		return -ENOMEM;
7249	driver_ver = old_driver_ver + size;
7250
7251	/* After a reset, the 32 bytes of "driver version" in the cfgtable
7252	 * should have been changed, otherwise we know the reset failed.
7253	 */
7254	init_driver_version(old_driver_ver, size);
7255	read_driver_ver_from_cfgtable(cfgtable, driver_ver);
7256	rc = !memcmp(driver_ver, old_driver_ver, size);
7257	kfree(old_driver_ver);
7258	return rc;
7259}
7260/* This does a hard reset of the controller using PCI power management
7261 * states or the using the doorbell register.
7262 */
7263static int hpsa_kdump_hard_reset_controller(struct pci_dev *pdev, u32 board_id)
7264{
7265	u64 cfg_offset;
7266	u32 cfg_base_addr;
7267	u64 cfg_base_addr_index;
7268	void __iomem *vaddr;
7269	unsigned long paddr;
7270	u32 misc_fw_support;
7271	int rc;
7272	struct CfgTable __iomem *cfgtable;
7273	u32 use_doorbell;
7274	u16 command_register;
7275
7276	/* For controllers as old as the P600, this is very nearly
7277	 * the same thing as
7278	 *
7279	 * pci_save_state(pci_dev);
7280	 * pci_set_power_state(pci_dev, PCI_D3hot);
7281	 * pci_set_power_state(pci_dev, PCI_D0);
7282	 * pci_restore_state(pci_dev);
7283	 *
7284	 * For controllers newer than the P600, the pci power state
7285	 * method of resetting doesn't work so we have another way
7286	 * using the doorbell register.
7287	 */
7288
7289	if (!ctlr_is_resettable(board_id)) {
7290		dev_warn(&pdev->dev, "Controller not resettable\n");
7291		return -ENODEV;
7292	}
7293
7294	/* if controller is soft- but not hard resettable... */
7295	if (!ctlr_is_hard_resettable(board_id))
7296		return -ENOTSUPP; /* try soft reset later. */
7297
7298	/* Save the PCI command register */
7299	pci_read_config_word(pdev, 4, &command_register);
7300	pci_save_state(pdev);
7301
7302	/* find the first memory BAR, so we can find the cfg table */
7303	rc = hpsa_pci_find_memory_BAR(pdev, &paddr);
7304	if (rc)
7305		return rc;
7306	vaddr = remap_pci_mem(paddr, 0x250);
7307	if (!vaddr)
7308		return -ENOMEM;
7309
7310	/* find cfgtable in order to check if reset via doorbell is supported */
7311	rc = hpsa_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
7312					&cfg_base_addr_index, &cfg_offset);
7313	if (rc)
7314		goto unmap_vaddr;
7315	cfgtable = remap_pci_mem(pci_resource_start(pdev,
7316		       cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
7317	if (!cfgtable) {
7318		rc = -ENOMEM;
7319		goto unmap_vaddr;
7320	}
7321	rc = write_driver_ver_to_cfgtable(cfgtable);
7322	if (rc)
7323		goto unmap_cfgtable;
7324
7325	/* If reset via doorbell register is supported, use that.
7326	 * There are two such methods.  Favor the newest method.
7327	 */
7328	misc_fw_support = readl(&cfgtable->misc_fw_support);
7329	use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
7330	if (use_doorbell) {
7331		use_doorbell = DOORBELL_CTLR_RESET2;
7332	} else {
7333		use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
7334		if (use_doorbell) {
7335			dev_warn(&pdev->dev,
7336				"Soft reset not supported. Firmware update is required.\n");
7337			rc = -ENOTSUPP; /* try soft reset */
7338			goto unmap_cfgtable;
7339		}
7340	}
7341
7342	rc = hpsa_controller_hard_reset(pdev, vaddr, use_doorbell);
7343	if (rc)
7344		goto unmap_cfgtable;
7345
7346	pci_restore_state(pdev);
7347	pci_write_config_word(pdev, 4, command_register);
7348
7349	/* Some devices (notably the HP Smart Array 5i Controller)
7350	   need a little pause here */
7351	msleep(HPSA_POST_RESET_PAUSE_MSECS);
7352
7353	rc = hpsa_wait_for_board_state(pdev, vaddr, BOARD_READY);
7354	if (rc) {
7355		dev_warn(&pdev->dev,
7356			"Failed waiting for board to become ready after hard reset\n");
7357		goto unmap_cfgtable;
7358	}
7359
7360	rc = controller_reset_failed(vaddr);
7361	if (rc < 0)
7362		goto unmap_cfgtable;
7363	if (rc) {
7364		dev_warn(&pdev->dev, "Unable to successfully reset "
7365			"controller. Will try soft reset.\n");
7366		rc = -ENOTSUPP;
7367	} else {
7368		dev_info(&pdev->dev, "board ready after hard reset.\n");
7369	}
7370
7371unmap_cfgtable:
7372	iounmap(cfgtable);
7373
7374unmap_vaddr:
7375	iounmap(vaddr);
7376	return rc;
7377}
7378
7379/*
7380 *  We cannot read the structure directly, for portability we must use
7381 *   the io functions.
7382 *   This is for debug only.
7383 */
7384static void print_cfg_table(struct device *dev, struct CfgTable __iomem *tb)
7385{
7386#ifdef HPSA_DEBUG
7387	int i;
7388	char temp_name[17];
7389
7390	dev_info(dev, "Controller Configuration information\n");
7391	dev_info(dev, "------------------------------------\n");
7392	for (i = 0; i < 4; i++)
7393		temp_name[i] = readb(&(tb->Signature[i]));
7394	temp_name[4] = '\0';
7395	dev_info(dev, "   Signature = %s\n", temp_name);
7396	dev_info(dev, "   Spec Number = %d\n", readl(&(tb->SpecValence)));
7397	dev_info(dev, "   Transport methods supported = 0x%x\n",
7398	       readl(&(tb->TransportSupport)));
7399	dev_info(dev, "   Transport methods active = 0x%x\n",
7400	       readl(&(tb->TransportActive)));
7401	dev_info(dev, "   Requested transport Method = 0x%x\n",
7402	       readl(&(tb->HostWrite.TransportRequest)));
7403	dev_info(dev, "   Coalesce Interrupt Delay = 0x%x\n",
7404	       readl(&(tb->HostWrite.CoalIntDelay)));
7405	dev_info(dev, "   Coalesce Interrupt Count = 0x%x\n",
7406	       readl(&(tb->HostWrite.CoalIntCount)));
7407	dev_info(dev, "   Max outstanding commands = %d\n",
7408	       readl(&(tb->CmdsOutMax)));
7409	dev_info(dev, "   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
7410	for (i = 0; i < 16; i++)
7411		temp_name[i] = readb(&(tb->ServerName[i]));
7412	temp_name[16] = '\0';
7413	dev_info(dev, "   Server Name = %s\n", temp_name);
7414	dev_info(dev, "   Heartbeat Counter = 0x%x\n\n\n",
7415		readl(&(tb->HeartBeat)));
7416#endif				/* HPSA_DEBUG */
7417}
7418
7419static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
7420{
7421	int i, offset, mem_type, bar_type;
7422
7423	if (pci_bar_addr == PCI_BASE_ADDRESS_0)	/* looking for BAR zero? */
7424		return 0;
7425	offset = 0;
7426	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
7427		bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
7428		if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
7429			offset += 4;
7430		else {
7431			mem_type = pci_resource_flags(pdev, i) &
7432			    PCI_BASE_ADDRESS_MEM_TYPE_MASK;
7433			switch (mem_type) {
7434			case PCI_BASE_ADDRESS_MEM_TYPE_32:
7435			case PCI_BASE_ADDRESS_MEM_TYPE_1M:
7436				offset += 4;	/* 32 bit */
7437				break;
7438			case PCI_BASE_ADDRESS_MEM_TYPE_64:
7439				offset += 8;
7440				break;
7441			default:	/* reserved in PCI 2.2 */
7442				dev_warn(&pdev->dev,
7443				       "base address is invalid\n");
7444				return -1;
7445				break;
7446			}
7447		}
7448		if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
7449			return i + 1;
7450	}
7451	return -1;
7452}
7453
7454static void hpsa_disable_interrupt_mode(struct ctlr_info *h)
7455{
7456	pci_free_irq_vectors(h->pdev);
7457	h->msix_vectors = 0;
7458}
7459
7460static void hpsa_setup_reply_map(struct ctlr_info *h)
7461{
7462	const struct cpumask *mask;
7463	unsigned int queue, cpu;
7464
7465	for (queue = 0; queue < h->msix_vectors; queue++) {
7466		mask = pci_irq_get_affinity(h->pdev, queue);
7467		if (!mask)
7468			goto fallback;
7469
7470		for_each_cpu(cpu, mask)
7471			h->reply_map[cpu] = queue;
7472	}
7473	return;
7474
7475fallback:
7476	for_each_possible_cpu(cpu)
7477		h->reply_map[cpu] = 0;
7478}
7479
7480/* If MSI/MSI-X is supported by the kernel we will try to enable it on
7481 * controllers that are capable. If not, we use legacy INTx mode.
7482 */
7483static int hpsa_interrupt_mode(struct ctlr_info *h)
7484{
7485	unsigned int flags = PCI_IRQ_LEGACY;
7486	int ret;
7487
7488	/* Some boards advertise MSI but don't really support it */
7489	switch (h->board_id) {
7490	case 0x40700E11:
7491	case 0x40800E11:
7492	case 0x40820E11:
7493	case 0x40830E11:
7494		break;
7495	default:
7496		ret = pci_alloc_irq_vectors(h->pdev, 1, MAX_REPLY_QUEUES,
7497				PCI_IRQ_MSIX | PCI_IRQ_AFFINITY);
7498		if (ret > 0) {
7499			h->msix_vectors = ret;
7500			return 0;
7501		}
7502
7503		flags |= PCI_IRQ_MSI;
7504		break;
7505	}
7506
7507	ret = pci_alloc_irq_vectors(h->pdev, 1, 1, flags);
7508	if (ret < 0)
7509		return ret;
7510	return 0;
7511}
7512
7513static int hpsa_lookup_board_id(struct pci_dev *pdev, u32 *board_id,
7514				bool *legacy_board)
7515{
7516	int i;
7517	u32 subsystem_vendor_id, subsystem_device_id;
7518
7519	subsystem_vendor_id = pdev->subsystem_vendor;
7520	subsystem_device_id = pdev->subsystem_device;
7521	*board_id = ((subsystem_device_id << 16) & 0xffff0000) |
7522		    subsystem_vendor_id;
7523
7524	if (legacy_board)
7525		*legacy_board = false;
7526	for (i = 0; i < ARRAY_SIZE(products); i++)
7527		if (*board_id == products[i].board_id) {
7528			if (products[i].access != &SA5A_access &&
7529			    products[i].access != &SA5B_access)
7530				return i;
7531			dev_warn(&pdev->dev,
7532				 "legacy board ID: 0x%08x\n",
7533				 *board_id);
7534			if (legacy_board)
7535			    *legacy_board = true;
7536			return i;
7537		}
7538
7539	dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x\n", *board_id);
7540	if (legacy_board)
7541		*legacy_board = true;
7542	return ARRAY_SIZE(products) - 1; /* generic unknown smart array */
7543}
7544
7545static int hpsa_pci_find_memory_BAR(struct pci_dev *pdev,
7546				    unsigned long *memory_bar)
7547{
7548	int i;
7549
7550	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
7551		if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
7552			/* addressing mode bits already removed */
7553			*memory_bar = pci_resource_start(pdev, i);
7554			dev_dbg(&pdev->dev, "memory BAR = %lx\n",
7555				*memory_bar);
7556			return 0;
7557		}
7558	dev_warn(&pdev->dev, "no memory BAR found\n");
7559	return -ENODEV;
7560}
7561
7562static int hpsa_wait_for_board_state(struct pci_dev *pdev, void __iomem *vaddr,
7563				     int wait_for_ready)
7564{
7565	int i, iterations;
7566	u32 scratchpad;
7567	if (wait_for_ready)
7568		iterations = HPSA_BOARD_READY_ITERATIONS;
7569	else
7570		iterations = HPSA_BOARD_NOT_READY_ITERATIONS;
7571
7572	for (i = 0; i < iterations; i++) {
7573		scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
7574		if (wait_for_ready) {
7575			if (scratchpad == HPSA_FIRMWARE_READY)
7576				return 0;
7577		} else {
7578			if (scratchpad != HPSA_FIRMWARE_READY)
7579				return 0;
7580		}
7581		msleep(HPSA_BOARD_READY_POLL_INTERVAL_MSECS);
7582	}
7583	dev_warn(&pdev->dev, "board not ready, timed out.\n");
7584	return -ENODEV;
7585}
7586
7587static int hpsa_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
7588			       u32 *cfg_base_addr, u64 *cfg_base_addr_index,
7589			       u64 *cfg_offset)
7590{
7591	*cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
7592	*cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
7593	*cfg_base_addr &= (u32) 0x0000ffff;
7594	*cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
7595	if (*cfg_base_addr_index == -1) {
7596		dev_warn(&pdev->dev, "cannot find cfg_base_addr_index\n");
7597		return -ENODEV;
7598	}
7599	return 0;
7600}
7601
7602static void hpsa_free_cfgtables(struct ctlr_info *h)
7603{
7604	if (h->transtable) {
7605		iounmap(h->transtable);
7606		h->transtable = NULL;
7607	}
7608	if (h->cfgtable) {
7609		iounmap(h->cfgtable);
7610		h->cfgtable = NULL;
7611	}
7612}
7613
7614/* Find and map CISS config table and transfer table
7615+ * several items must be unmapped (freed) later
7616+ * */
7617static int hpsa_find_cfgtables(struct ctlr_info *h)
7618{
7619	u64 cfg_offset;
7620	u32 cfg_base_addr;
7621	u64 cfg_base_addr_index;
7622	u32 trans_offset;
7623	int rc;
7624
7625	rc = hpsa_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
7626		&cfg_base_addr_index, &cfg_offset);
7627	if (rc)
7628		return rc;
7629	h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
7630		       cfg_base_addr_index) + cfg_offset, sizeof(*h->cfgtable));
7631	if (!h->cfgtable) {
7632		dev_err(&h->pdev->dev, "Failed mapping cfgtable\n");
7633		return -ENOMEM;
7634	}
7635	rc = write_driver_ver_to_cfgtable(h->cfgtable);
7636	if (rc)
7637		return rc;
7638	/* Find performant mode table. */
7639	trans_offset = readl(&h->cfgtable->TransMethodOffset);
7640	h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
7641				cfg_base_addr_index)+cfg_offset+trans_offset,
7642				sizeof(*h->transtable));
7643	if (!h->transtable) {
7644		dev_err(&h->pdev->dev, "Failed mapping transfer table\n");
7645		hpsa_free_cfgtables(h);
7646		return -ENOMEM;
7647	}
7648	return 0;
7649}
7650
7651static void hpsa_get_max_perf_mode_cmds(struct ctlr_info *h)
7652{
7653#define MIN_MAX_COMMANDS 16
7654	BUILD_BUG_ON(MIN_MAX_COMMANDS <= HPSA_NRESERVED_CMDS);
7655
7656	h->max_commands = readl(&h->cfgtable->MaxPerformantModeCommands);
7657
7658	/* Limit commands in memory limited kdump scenario. */
7659	if (reset_devices && h->max_commands > 32)
7660		h->max_commands = 32;
7661
7662	if (h->max_commands < MIN_MAX_COMMANDS) {
7663		dev_warn(&h->pdev->dev,
7664			"Controller reports max supported commands of %d Using %d instead. Ensure that firmware is up to date.\n",
7665			h->max_commands,
7666			MIN_MAX_COMMANDS);
7667		h->max_commands = MIN_MAX_COMMANDS;
7668	}
7669}
7670
7671/* If the controller reports that the total max sg entries is greater than 512,
7672 * then we know that chained SG blocks work.  (Original smart arrays did not
7673 * support chained SG blocks and would return zero for max sg entries.)
7674 */
7675static int hpsa_supports_chained_sg_blocks(struct ctlr_info *h)
7676{
7677	return h->maxsgentries > 512;
7678}
7679
7680/* Interrogate the hardware for some limits:
7681 * max commands, max SG elements without chaining, and with chaining,
7682 * SG chain block size, etc.
7683 */
7684static void hpsa_find_board_params(struct ctlr_info *h)
7685{
7686	hpsa_get_max_perf_mode_cmds(h);
7687	h->nr_cmds = h->max_commands;
7688	h->maxsgentries = readl(&(h->cfgtable->MaxScatterGatherElements));
7689	h->fw_support = readl(&(h->cfgtable->misc_fw_support));
7690	if (hpsa_supports_chained_sg_blocks(h)) {
7691		/* Limit in-command s/g elements to 32 save dma'able memory. */
7692		h->max_cmd_sg_entries = 32;
7693		h->chainsize = h->maxsgentries - h->max_cmd_sg_entries;
7694		h->maxsgentries--; /* save one for chain pointer */
7695	} else {
7696		/*
7697		 * Original smart arrays supported at most 31 s/g entries
7698		 * embedded inline in the command (trying to use more
7699		 * would lock up the controller)
7700		 */
7701		h->max_cmd_sg_entries = 31;
7702		h->maxsgentries = 31; /* default to traditional values */
7703		h->chainsize = 0;
7704	}
7705
7706	/* Find out what task management functions are supported and cache */
7707	h->TMFSupportFlags = readl(&(h->cfgtable->TMFSupportFlags));
7708	if (!(HPSATMF_PHYS_TASK_ABORT & h->TMFSupportFlags))
7709		dev_warn(&h->pdev->dev, "Physical aborts not supported\n");
7710	if (!(HPSATMF_LOG_TASK_ABORT & h->TMFSupportFlags))
7711		dev_warn(&h->pdev->dev, "Logical aborts not supported\n");
7712	if (!(HPSATMF_IOACCEL_ENABLED & h->TMFSupportFlags))
7713		dev_warn(&h->pdev->dev, "HP SSD Smart Path aborts not supported\n");
7714}
7715
7716static inline bool hpsa_CISS_signature_present(struct ctlr_info *h)
7717{
7718	if (!check_signature(h->cfgtable->Signature, "CISS", 4)) {
7719		dev_err(&h->pdev->dev, "not a valid CISS config table\n");
7720		return false;
7721	}
7722	return true;
7723}
7724
7725static inline void hpsa_set_driver_support_bits(struct ctlr_info *h)
7726{
7727	u32 driver_support;
7728
7729	driver_support = readl(&(h->cfgtable->driver_support));
7730	/* Need to enable prefetch in the SCSI core for 6400 in x86 */
7731#ifdef CONFIG_X86
7732	driver_support |= ENABLE_SCSI_PREFETCH;
7733#endif
7734	driver_support |= ENABLE_UNIT_ATTN;
7735	writel(driver_support, &(h->cfgtable->driver_support));
7736}
7737
7738/* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
7739 * in a prefetch beyond physical memory.
7740 */
7741static inline void hpsa_p600_dma_prefetch_quirk(struct ctlr_info *h)
7742{
7743	u32 dma_prefetch;
7744
7745	if (h->board_id != 0x3225103C)
7746		return;
7747	dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
7748	dma_prefetch |= 0x8000;
7749	writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
7750}
7751
7752static int hpsa_wait_for_clear_event_notify_ack(struct ctlr_info *h)
7753{
7754	int i;
7755	u32 doorbell_value;
7756	unsigned long flags;
7757	/* wait until the clear_event_notify bit 6 is cleared by controller. */
7758	for (i = 0; i < MAX_CLEAR_EVENT_WAIT; i++) {
7759		spin_lock_irqsave(&h->lock, flags);
7760		doorbell_value = readl(h->vaddr + SA5_DOORBELL);
7761		spin_unlock_irqrestore(&h->lock, flags);
7762		if (!(doorbell_value & DOORBELL_CLEAR_EVENTS))
7763			goto done;
7764		/* delay and try again */
7765		msleep(CLEAR_EVENT_WAIT_INTERVAL);
7766	}
7767	return -ENODEV;
7768done:
7769	return 0;
7770}
7771
7772static int hpsa_wait_for_mode_change_ack(struct ctlr_info *h)
7773{
7774	int i;
7775	u32 doorbell_value;
7776	unsigned long flags;
7777
7778	/* under certain very rare conditions, this can take awhile.
7779	 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
7780	 * as we enter this code.)
7781	 */
7782	for (i = 0; i < MAX_MODE_CHANGE_WAIT; i++) {
7783		if (h->remove_in_progress)
7784			goto done;
7785		spin_lock_irqsave(&h->lock, flags);
7786		doorbell_value = readl(h->vaddr + SA5_DOORBELL);
7787		spin_unlock_irqrestore(&h->lock, flags);
7788		if (!(doorbell_value & CFGTBL_ChangeReq))
7789			goto done;
7790		/* delay and try again */
7791		msleep(MODE_CHANGE_WAIT_INTERVAL);
7792	}
7793	return -ENODEV;
7794done:
7795	return 0;
7796}
7797
7798/* return -ENODEV or other reason on error, 0 on success */
7799static int hpsa_enter_simple_mode(struct ctlr_info *h)
7800{
7801	u32 trans_support;
7802
7803	trans_support = readl(&(h->cfgtable->TransportSupport));
7804	if (!(trans_support & SIMPLE_MODE))
7805		return -ENOTSUPP;
7806
7807	h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
7808
7809	/* Update the field, and then ring the doorbell */
7810	writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
7811	writel(0, &h->cfgtable->HostWrite.command_pool_addr_hi);
7812	writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
7813	if (hpsa_wait_for_mode_change_ack(h))
7814		goto error;
7815	print_cfg_table(&h->pdev->dev, h->cfgtable);
7816	if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple))
7817		goto error;
7818	h->transMethod = CFGTBL_Trans_Simple;
7819	return 0;
7820error:
7821	dev_err(&h->pdev->dev, "failed to enter simple mode\n");
7822	return -ENODEV;
7823}
7824
7825/* free items allocated or mapped by hpsa_pci_init */
7826static void hpsa_free_pci_init(struct ctlr_info *h)
7827{
7828	hpsa_free_cfgtables(h);			/* pci_init 4 */
7829	iounmap(h->vaddr);			/* pci_init 3 */
7830	h->vaddr = NULL;
7831	hpsa_disable_interrupt_mode(h);		/* pci_init 2 */
7832	/*
7833	 * call pci_disable_device before pci_release_regions per
7834	 * Documentation/driver-api/pci/pci.rst
7835	 */
7836	pci_disable_device(h->pdev);		/* pci_init 1 */
7837	pci_release_regions(h->pdev);		/* pci_init 2 */
7838}
7839
7840/* several items must be freed later */
7841static int hpsa_pci_init(struct ctlr_info *h)
7842{
7843	int prod_index, err;
7844	bool legacy_board;
7845
7846	prod_index = hpsa_lookup_board_id(h->pdev, &h->board_id, &legacy_board);
7847	if (prod_index < 0)
7848		return prod_index;
7849	h->product_name = products[prod_index].product_name;
7850	h->access = *(products[prod_index].access);
7851	h->legacy_board = legacy_board;
7852	pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
7853			       PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
7854
7855	err = pci_enable_device(h->pdev);
7856	if (err) {
7857		dev_err(&h->pdev->dev, "failed to enable PCI device\n");
7858		pci_disable_device(h->pdev);
7859		return err;
7860	}
7861
7862	err = pci_request_regions(h->pdev, HPSA);
7863	if (err) {
7864		dev_err(&h->pdev->dev,
7865			"failed to obtain PCI resources\n");
7866		pci_disable_device(h->pdev);
7867		return err;
7868	}
7869
7870	pci_set_master(h->pdev);
7871
7872	err = hpsa_interrupt_mode(h);
7873	if (err)
7874		goto clean1;
7875
7876	/* setup mapping between CPU and reply queue */
7877	hpsa_setup_reply_map(h);
7878
7879	err = hpsa_pci_find_memory_BAR(h->pdev, &h->paddr);
7880	if (err)
7881		goto clean2;	/* intmode+region, pci */
7882	h->vaddr = remap_pci_mem(h->paddr, 0x250);
7883	if (!h->vaddr) {
7884		dev_err(&h->pdev->dev, "failed to remap PCI mem\n");
7885		err = -ENOMEM;
7886		goto clean2;	/* intmode+region, pci */
7887	}
7888	err = hpsa_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
7889	if (err)
7890		goto clean3;	/* vaddr, intmode+region, pci */
7891	err = hpsa_find_cfgtables(h);
7892	if (err)
7893		goto clean3;	/* vaddr, intmode+region, pci */
7894	hpsa_find_board_params(h);
7895
7896	if (!hpsa_CISS_signature_present(h)) {
7897		err = -ENODEV;
7898		goto clean4;	/* cfgtables, vaddr, intmode+region, pci */
7899	}
7900	hpsa_set_driver_support_bits(h);
7901	hpsa_p600_dma_prefetch_quirk(h);
7902	err = hpsa_enter_simple_mode(h);
7903	if (err)
7904		goto clean4;	/* cfgtables, vaddr, intmode+region, pci */
7905	return 0;
7906
7907clean4:	/* cfgtables, vaddr, intmode+region, pci */
7908	hpsa_free_cfgtables(h);
7909clean3:	/* vaddr, intmode+region, pci */
7910	iounmap(h->vaddr);
7911	h->vaddr = NULL;
7912clean2:	/* intmode+region, pci */
7913	hpsa_disable_interrupt_mode(h);
7914clean1:
7915	/*
7916	 * call pci_disable_device before pci_release_regions per
7917	 * Documentation/driver-api/pci/pci.rst
7918	 */
7919	pci_disable_device(h->pdev);
7920	pci_release_regions(h->pdev);
7921	return err;
7922}
7923
7924static void hpsa_hba_inquiry(struct ctlr_info *h)
7925{
7926	int rc;
7927
7928#define HBA_INQUIRY_BYTE_COUNT 64
7929	h->hba_inquiry_data = kmalloc(HBA_INQUIRY_BYTE_COUNT, GFP_KERNEL);
7930	if (!h->hba_inquiry_data)
7931		return;
7932	rc = hpsa_scsi_do_inquiry(h, RAID_CTLR_LUNID, 0,
7933		h->hba_inquiry_data, HBA_INQUIRY_BYTE_COUNT);
7934	if (rc != 0) {
7935		kfree(h->hba_inquiry_data);
7936		h->hba_inquiry_data = NULL;
7937	}
7938}
7939
7940static int hpsa_init_reset_devices(struct pci_dev *pdev, u32 board_id)
7941{
7942	int rc, i;
7943	void __iomem *vaddr;
7944
7945	if (!reset_devices)
7946		return 0;
7947
7948	/* kdump kernel is loading, we don't know in which state is
7949	 * the pci interface. The dev->enable_cnt is equal zero
7950	 * so we call enable+disable, wait a while and switch it on.
7951	 */
7952	rc = pci_enable_device(pdev);
7953	if (rc) {
7954		dev_warn(&pdev->dev, "Failed to enable PCI device\n");
7955		return -ENODEV;
7956	}
7957	pci_disable_device(pdev);
7958	msleep(260);			/* a randomly chosen number */
7959	rc = pci_enable_device(pdev);
7960	if (rc) {
7961		dev_warn(&pdev->dev, "failed to enable device.\n");
7962		return -ENODEV;
7963	}
7964
7965	pci_set_master(pdev);
7966
7967	vaddr = pci_ioremap_bar(pdev, 0);
7968	if (vaddr == NULL) {
7969		rc = -ENOMEM;
7970		goto out_disable;
7971	}
7972	writel(SA5_INTR_OFF, vaddr + SA5_REPLY_INTR_MASK_OFFSET);
7973	iounmap(vaddr);
7974
7975	/* Reset the controller with a PCI power-cycle or via doorbell */
7976	rc = hpsa_kdump_hard_reset_controller(pdev, board_id);
7977
7978	/* -ENOTSUPP here means we cannot reset the controller
7979	 * but it's already (and still) up and running in
7980	 * "performant mode".  Or, it might be 640x, which can't reset
7981	 * due to concerns about shared bbwc between 6402/6404 pair.
7982	 */
7983	if (rc)
7984		goto out_disable;
7985
7986	/* Now try to get the controller to respond to a no-op */
7987	dev_info(&pdev->dev, "Waiting for controller to respond to no-op\n");
7988	for (i = 0; i < HPSA_POST_RESET_NOOP_RETRIES; i++) {
7989		if (hpsa_noop(pdev) == 0)
7990			break;
7991		else
7992			dev_warn(&pdev->dev, "no-op failed%s\n",
7993					(i < 11 ? "; re-trying" : ""));
7994	}
7995
7996out_disable:
7997
7998	pci_disable_device(pdev);
7999	return rc;
8000}
8001
8002static void hpsa_free_cmd_pool(struct ctlr_info *h)
8003{
8004	kfree(h->cmd_pool_bits);
8005	h->cmd_pool_bits = NULL;
8006	if (h->cmd_pool) {
8007		dma_free_coherent(&h->pdev->dev,
8008				h->nr_cmds * sizeof(struct CommandList),
8009				h->cmd_pool,
8010				h->cmd_pool_dhandle);
8011		h->cmd_pool = NULL;
8012		h->cmd_pool_dhandle = 0;
8013	}
8014	if (h->errinfo_pool) {
8015		dma_free_coherent(&h->pdev->dev,
8016				h->nr_cmds * sizeof(struct ErrorInfo),
8017				h->errinfo_pool,
8018				h->errinfo_pool_dhandle);
8019		h->errinfo_pool = NULL;
8020		h->errinfo_pool_dhandle = 0;
8021	}
8022}
8023
8024static int hpsa_alloc_cmd_pool(struct ctlr_info *h)
8025{
8026	h->cmd_pool_bits = kcalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG),
8027				   sizeof(unsigned long),
8028				   GFP_KERNEL);
8029	h->cmd_pool = dma_alloc_coherent(&h->pdev->dev,
8030		    h->nr_cmds * sizeof(*h->cmd_pool),
8031		    &h->cmd_pool_dhandle, GFP_KERNEL);
8032	h->errinfo_pool = dma_alloc_coherent(&h->pdev->dev,
8033		    h->nr_cmds * sizeof(*h->errinfo_pool),
8034		    &h->errinfo_pool_dhandle, GFP_KERNEL);
8035	if ((h->cmd_pool_bits == NULL)
8036	    || (h->cmd_pool == NULL)
8037	    || (h->errinfo_pool == NULL)) {
8038		dev_err(&h->pdev->dev, "out of memory in %s", __func__);
8039		goto clean_up;
8040	}
8041	hpsa_preinitialize_commands(h);
8042	return 0;
8043clean_up:
8044	hpsa_free_cmd_pool(h);
8045	return -ENOMEM;
8046}
8047
8048/* clear affinity hints and free MSI-X, MSI, or legacy INTx vectors */
8049static void hpsa_free_irqs(struct ctlr_info *h)
8050{
8051	int i;
8052	int irq_vector = 0;
8053
8054	if (hpsa_simple_mode)
8055		irq_vector = h->intr_mode;
8056
8057	if (!h->msix_vectors || h->intr_mode != PERF_MODE_INT) {
8058		/* Single reply queue, only one irq to free */
8059		free_irq(pci_irq_vector(h->pdev, irq_vector),
8060				&h->q[h->intr_mode]);
8061		h->q[h->intr_mode] = 0;
8062		return;
8063	}
8064
8065	for (i = 0; i < h->msix_vectors; i++) {
8066		free_irq(pci_irq_vector(h->pdev, i), &h->q[i]);
8067		h->q[i] = 0;
8068	}
8069	for (; i < MAX_REPLY_QUEUES; i++)
8070		h->q[i] = 0;
8071}
8072
8073/* returns 0 on success; cleans up and returns -Enn on error */
8074static int hpsa_request_irqs(struct ctlr_info *h,
8075	irqreturn_t (*msixhandler)(int, void *),
8076	irqreturn_t (*intxhandler)(int, void *))
8077{
8078	int rc, i;
8079	int irq_vector = 0;
8080
8081	if (hpsa_simple_mode)
8082		irq_vector = h->intr_mode;
8083
8084	/*
8085	 * initialize h->q[x] = x so that interrupt handlers know which
8086	 * queue to process.
8087	 */
8088	for (i = 0; i < MAX_REPLY_QUEUES; i++)
8089		h->q[i] = (u8) i;
8090
8091	if (h->intr_mode == PERF_MODE_INT && h->msix_vectors > 0) {
8092		/* If performant mode and MSI-X, use multiple reply queues */
8093		for (i = 0; i < h->msix_vectors; i++) {
8094			sprintf(h->intrname[i], "%s-msix%d", h->devname, i);
8095			rc = request_irq(pci_irq_vector(h->pdev, i), msixhandler,
8096					0, h->intrname[i],
8097					&h->q[i]);
8098			if (rc) {
8099				int j;
8100
8101				dev_err(&h->pdev->dev,
8102					"failed to get irq %d for %s\n",
8103				       pci_irq_vector(h->pdev, i), h->devname);
8104				for (j = 0; j < i; j++) {
8105					free_irq(pci_irq_vector(h->pdev, j), &h->q[j]);
8106					h->q[j] = 0;
8107				}
8108				for (; j < MAX_REPLY_QUEUES; j++)
8109					h->q[j] = 0;
8110				return rc;
8111			}
8112		}
8113	} else {
8114		/* Use single reply pool */
8115		if (h->msix_vectors > 0 || h->pdev->msi_enabled) {
8116			sprintf(h->intrname[0], "%s-msi%s", h->devname,
8117				h->msix_vectors ? "x" : "");
8118			rc = request_irq(pci_irq_vector(h->pdev, irq_vector),
8119				msixhandler, 0,
8120				h->intrname[0],
8121				&h->q[h->intr_mode]);
8122		} else {
8123			sprintf(h->intrname[h->intr_mode],
8124				"%s-intx", h->devname);
8125			rc = request_irq(pci_irq_vector(h->pdev, irq_vector),
8126				intxhandler, IRQF_SHARED,
8127				h->intrname[0],
8128				&h->q[h->intr_mode]);
8129		}
8130	}
8131	if (rc) {
8132		dev_err(&h->pdev->dev, "failed to get irq %d for %s\n",
8133		       pci_irq_vector(h->pdev, irq_vector), h->devname);
8134		hpsa_free_irqs(h);
8135		return -ENODEV;
8136	}
8137	return 0;
8138}
8139
8140static int hpsa_kdump_soft_reset(struct ctlr_info *h)
8141{
8142	int rc;
8143	hpsa_send_host_reset(h, HPSA_RESET_TYPE_CONTROLLER);
8144
8145	dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
8146	rc = hpsa_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY);
8147	if (rc) {
8148		dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
8149		return rc;
8150	}
8151
8152	dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
8153	rc = hpsa_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
8154	if (rc) {
8155		dev_warn(&h->pdev->dev, "Board failed to become ready "
8156			"after soft reset.\n");
8157		return rc;
8158	}
8159
8160	return 0;
8161}
8162
8163static void hpsa_free_reply_queues(struct ctlr_info *h)
8164{
8165	int i;
8166
8167	for (i = 0; i < h->nreply_queues; i++) {
8168		if (!h->reply_queue[i].head)
8169			continue;
8170		dma_free_coherent(&h->pdev->dev,
8171					h->reply_queue_size,
8172					h->reply_queue[i].head,
8173					h->reply_queue[i].busaddr);
8174		h->reply_queue[i].head = NULL;
8175		h->reply_queue[i].busaddr = 0;
8176	}
8177	h->reply_queue_size = 0;
8178}
8179
8180static void hpsa_undo_allocations_after_kdump_soft_reset(struct ctlr_info *h)
8181{
8182	hpsa_free_performant_mode(h);		/* init_one 7 */
8183	hpsa_free_sg_chain_blocks(h);		/* init_one 6 */
8184	hpsa_free_cmd_pool(h);			/* init_one 5 */
8185	hpsa_free_irqs(h);			/* init_one 4 */
8186	scsi_host_put(h->scsi_host);		/* init_one 3 */
8187	h->scsi_host = NULL;			/* init_one 3 */
8188	hpsa_free_pci_init(h);			/* init_one 2_5 */
8189	free_percpu(h->lockup_detected);	/* init_one 2 */
8190	h->lockup_detected = NULL;		/* init_one 2 */
8191	if (h->resubmit_wq) {
8192		destroy_workqueue(h->resubmit_wq);	/* init_one 1 */
8193		h->resubmit_wq = NULL;
8194	}
8195	if (h->rescan_ctlr_wq) {
8196		destroy_workqueue(h->rescan_ctlr_wq);
8197		h->rescan_ctlr_wq = NULL;
8198	}
8199	if (h->monitor_ctlr_wq) {
8200		destroy_workqueue(h->monitor_ctlr_wq);
8201		h->monitor_ctlr_wq = NULL;
8202	}
8203
8204	kfree(h);				/* init_one 1 */
8205}
8206
8207/* Called when controller lockup detected. */
8208static void fail_all_outstanding_cmds(struct ctlr_info *h)
8209{
8210	int i, refcount;
8211	struct CommandList *c;
8212	int failcount = 0;
8213
8214	flush_workqueue(h->resubmit_wq); /* ensure all cmds are fully built */
8215	for (i = 0; i < h->nr_cmds; i++) {
8216		c = h->cmd_pool + i;
8217		refcount = atomic_inc_return(&c->refcount);
8218		if (refcount > 1) {
8219			c->err_info->CommandStatus = CMD_CTLR_LOCKUP;
8220			finish_cmd(c);
8221			atomic_dec(&h->commands_outstanding);
8222			failcount++;
8223		}
8224		cmd_free(h, c);
8225	}
8226	dev_warn(&h->pdev->dev,
8227		"failed %d commands in fail_all\n", failcount);
8228}
8229
8230static void set_lockup_detected_for_all_cpus(struct ctlr_info *h, u32 value)
8231{
8232	int cpu;
8233
8234	for_each_online_cpu(cpu) {
8235		u32 *lockup_detected;
8236		lockup_detected = per_cpu_ptr(h->lockup_detected, cpu);
8237		*lockup_detected = value;
8238	}
8239	wmb(); /* be sure the per-cpu variables are out to memory */
8240}
8241
8242static void controller_lockup_detected(struct ctlr_info *h)
8243{
8244	unsigned long flags;
8245	u32 lockup_detected;
8246
8247	h->access.set_intr_mask(h, HPSA_INTR_OFF);
8248	spin_lock_irqsave(&h->lock, flags);
8249	lockup_detected = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
8250	if (!lockup_detected) {
8251		/* no heartbeat, but controller gave us a zero. */
8252		dev_warn(&h->pdev->dev,
8253			"lockup detected after %d but scratchpad register is zero\n",
8254			h->heartbeat_sample_interval / HZ);
8255		lockup_detected = 0xffffffff;
8256	}
8257	set_lockup_detected_for_all_cpus(h, lockup_detected);
8258	spin_unlock_irqrestore(&h->lock, flags);
8259	dev_warn(&h->pdev->dev, "Controller lockup detected: 0x%08x after %d\n",
8260			lockup_detected, h->heartbeat_sample_interval / HZ);
8261	if (lockup_detected == 0xffff0000) {
8262		dev_warn(&h->pdev->dev, "Telling controller to do a CHKPT\n");
8263		writel(DOORBELL_GENERATE_CHKPT, h->vaddr + SA5_DOORBELL);
8264	}
8265	pci_disable_device(h->pdev);
8266	fail_all_outstanding_cmds(h);
8267}
8268
8269static int detect_controller_lockup(struct ctlr_info *h)
8270{
8271	u64 now;
8272	u32 heartbeat;
8273	unsigned long flags;
8274
8275	now = get_jiffies_64();
8276	/* If we've received an interrupt recently, we're ok. */
8277	if (time_after64(h->last_intr_timestamp +
8278				(h->heartbeat_sample_interval), now))
8279		return false;
8280
8281	/*
8282	 * If we've already checked the heartbeat recently, we're ok.
8283	 * This could happen if someone sends us a signal. We
8284	 * otherwise don't care about signals in this thread.
8285	 */
8286	if (time_after64(h->last_heartbeat_timestamp +
8287				(h->heartbeat_sample_interval), now))
8288		return false;
8289
8290	/* If heartbeat has not changed since we last looked, we're not ok. */
8291	spin_lock_irqsave(&h->lock, flags);
8292	heartbeat = readl(&h->cfgtable->HeartBeat);
8293	spin_unlock_irqrestore(&h->lock, flags);
8294	if (h->last_heartbeat == heartbeat) {
8295		controller_lockup_detected(h);
8296		return true;
8297	}
8298
8299	/* We're ok. */
8300	h->last_heartbeat = heartbeat;
8301	h->last_heartbeat_timestamp = now;
8302	return false;
8303}
8304
8305/*
8306 * Set ioaccel status for all ioaccel volumes.
8307 *
8308 * Called from monitor controller worker (hpsa_event_monitor_worker)
8309 *
8310 * A Volume (or Volumes that comprise an Array set) may be undergoing a
8311 * transformation, so we will be turning off ioaccel for all volumes that
8312 * make up the Array.
8313 */
8314static void hpsa_set_ioaccel_status(struct ctlr_info *h)
8315{
8316	int rc;
8317	int i;
8318	u8 ioaccel_status;
8319	unsigned char *buf;
8320	struct hpsa_scsi_dev_t *device;
8321
8322	if (!h)
8323		return;
8324
8325	buf = kmalloc(64, GFP_KERNEL);
8326	if (!buf)
8327		return;
8328
8329	/*
8330	 * Run through current device list used during I/O requests.
8331	 */
8332	for (i = 0; i < h->ndevices; i++) {
8333		int offload_to_be_enabled = 0;
8334		int offload_config = 0;
8335
8336		device = h->dev[i];
8337
8338		if (!device)
8339			continue;
8340		if (!hpsa_vpd_page_supported(h, device->scsi3addr,
8341						HPSA_VPD_LV_IOACCEL_STATUS))
8342			continue;
8343
8344		memset(buf, 0, 64);
8345
8346		rc = hpsa_scsi_do_inquiry(h, device->scsi3addr,
8347					VPD_PAGE | HPSA_VPD_LV_IOACCEL_STATUS,
8348					buf, 64);
8349		if (rc != 0)
8350			continue;
8351
8352		ioaccel_status = buf[IOACCEL_STATUS_BYTE];
8353
8354		/*
8355		 * Check if offload is still configured on
8356		 */
8357		offload_config =
8358				!!(ioaccel_status & OFFLOAD_CONFIGURED_BIT);
8359		/*
8360		 * If offload is configured on, check to see if ioaccel
8361		 * needs to be enabled.
8362		 */
8363		if (offload_config)
8364			offload_to_be_enabled =
8365				!!(ioaccel_status & OFFLOAD_ENABLED_BIT);
8366
8367		/*
8368		 * If ioaccel is to be re-enabled, re-enable later during the
8369		 * scan operation so the driver can get a fresh raidmap
8370		 * before turning ioaccel back on.
8371		 */
8372		if (offload_to_be_enabled)
8373			continue;
8374
8375		/*
8376		 * Immediately turn off ioaccel for any volume the
8377		 * controller tells us to. Some of the reasons could be:
8378		 *    transformation - change to the LVs of an Array.
8379		 *    degraded volume - component failure
8380		 */
8381		hpsa_turn_off_ioaccel_for_device(device);
8382	}
8383
8384	kfree(buf);
8385}
8386
8387static void hpsa_ack_ctlr_events(struct ctlr_info *h)
8388{
8389	char *event_type;
8390
8391	if (!(h->fw_support & MISC_FW_EVENT_NOTIFY))
8392		return;
8393
8394	/* Ask the controller to clear the events we're handling. */
8395	if ((h->transMethod & (CFGTBL_Trans_io_accel1
8396			| CFGTBL_Trans_io_accel2)) &&
8397		(h->events & HPSA_EVENT_NOTIFY_ACCEL_IO_PATH_STATE_CHANGE ||
8398		 h->events & HPSA_EVENT_NOTIFY_ACCEL_IO_PATH_CONFIG_CHANGE)) {
8399
8400		if (h->events & HPSA_EVENT_NOTIFY_ACCEL_IO_PATH_STATE_CHANGE)
8401			event_type = "state change";
8402		if (h->events & HPSA_EVENT_NOTIFY_ACCEL_IO_PATH_CONFIG_CHANGE)
8403			event_type = "configuration change";
8404		/* Stop sending new RAID offload reqs via the IO accelerator */
8405		scsi_block_requests(h->scsi_host);
8406		hpsa_set_ioaccel_status(h);
8407		hpsa_drain_accel_commands(h);
8408		/* Set 'accelerator path config change' bit */
8409		dev_warn(&h->pdev->dev,
8410			"Acknowledging event: 0x%08x (HP SSD Smart Path %s)\n",
8411			h->events, event_type);
8412		writel(h->events, &(h->cfgtable->clear_event_notify));
8413		/* Set the "clear event notify field update" bit 6 */
8414		writel(DOORBELL_CLEAR_EVENTS, h->vaddr + SA5_DOORBELL);
8415		/* Wait until ctlr clears 'clear event notify field', bit 6 */
8416		hpsa_wait_for_clear_event_notify_ack(h);
8417		scsi_unblock_requests(h->scsi_host);
8418	} else {
8419		/* Acknowledge controller notification events. */
8420		writel(h->events, &(h->cfgtable->clear_event_notify));
8421		writel(DOORBELL_CLEAR_EVENTS, h->vaddr + SA5_DOORBELL);
8422		hpsa_wait_for_clear_event_notify_ack(h);
8423	}
8424	return;
8425}
8426
8427/* Check a register on the controller to see if there are configuration
8428 * changes (added/changed/removed logical drives, etc.) which mean that
8429 * we should rescan the controller for devices.
8430 * Also check flag for driver-initiated rescan.
8431 */
8432static int hpsa_ctlr_needs_rescan(struct ctlr_info *h)
8433{
8434	if (h->drv_req_rescan) {
8435		h->drv_req_rescan = 0;
8436		return 1;
8437	}
8438
8439	if (!(h->fw_support & MISC_FW_EVENT_NOTIFY))
8440		return 0;
8441
8442	h->events = readl(&(h->cfgtable->event_notify));
8443	return h->events & RESCAN_REQUIRED_EVENT_BITS;
8444}
8445
8446/*
8447 * Check if any of the offline devices have become ready
8448 */
8449static int hpsa_offline_devices_ready(struct ctlr_info *h)
8450{
8451	unsigned long flags;
8452	struct offline_device_entry *d;
8453	struct list_head *this, *tmp;
8454
8455	spin_lock_irqsave(&h->offline_device_lock, flags);
8456	list_for_each_safe(this, tmp, &h->offline_device_list) {
8457		d = list_entry(this, struct offline_device_entry,
8458				offline_list);
8459		spin_unlock_irqrestore(&h->offline_device_lock, flags);
8460		if (!hpsa_volume_offline(h, d->scsi3addr)) {
8461			spin_lock_irqsave(&h->offline_device_lock, flags);
8462			list_del(&d->offline_list);
8463			spin_unlock_irqrestore(&h->offline_device_lock, flags);
8464			return 1;
8465		}
8466		spin_lock_irqsave(&h->offline_device_lock, flags);
8467	}
8468	spin_unlock_irqrestore(&h->offline_device_lock, flags);
8469	return 0;
8470}
8471
8472static int hpsa_luns_changed(struct ctlr_info *h)
8473{
8474	int rc = 1; /* assume there are changes */
8475	struct ReportLUNdata *logdev = NULL;
8476
8477	/* if we can't find out if lun data has changed,
8478	 * assume that it has.
8479	 */
8480
8481	if (!h->lastlogicals)
8482		return rc;
8483
8484	logdev = kzalloc(sizeof(*logdev), GFP_KERNEL);
8485	if (!logdev)
8486		return rc;
8487
8488	if (hpsa_scsi_do_report_luns(h, 1, logdev, sizeof(*logdev), 0)) {
8489		dev_warn(&h->pdev->dev,
8490			"report luns failed, can't track lun changes.\n");
8491		goto out;
8492	}
8493	if (memcmp(logdev, h->lastlogicals, sizeof(*logdev))) {
8494		dev_info(&h->pdev->dev,
8495			"Lun changes detected.\n");
8496		memcpy(h->lastlogicals, logdev, sizeof(*logdev));
8497		goto out;
8498	} else
8499		rc = 0; /* no changes detected. */
8500out:
8501	kfree(logdev);
8502	return rc;
8503}
8504
8505static void hpsa_perform_rescan(struct ctlr_info *h)
8506{
8507	struct Scsi_Host *sh = NULL;
8508	unsigned long flags;
8509
8510	/*
8511	 * Do the scan after the reset
8512	 */
8513	spin_lock_irqsave(&h->reset_lock, flags);
8514	if (h->reset_in_progress) {
8515		h->drv_req_rescan = 1;
8516		spin_unlock_irqrestore(&h->reset_lock, flags);
8517		return;
8518	}
8519	spin_unlock_irqrestore(&h->reset_lock, flags);
8520
8521	sh = scsi_host_get(h->scsi_host);
8522	if (sh != NULL) {
8523		hpsa_scan_start(sh);
8524		scsi_host_put(sh);
8525		h->drv_req_rescan = 0;
8526	}
8527}
8528
8529/*
8530 * watch for controller events
8531 */
8532static void hpsa_event_monitor_worker(struct work_struct *work)
8533{
8534	struct ctlr_info *h = container_of(to_delayed_work(work),
8535					struct ctlr_info, event_monitor_work);
8536	unsigned long flags;
8537
8538	spin_lock_irqsave(&h->lock, flags);
8539	if (h->remove_in_progress) {
8540		spin_unlock_irqrestore(&h->lock, flags);
8541		return;
8542	}
8543	spin_unlock_irqrestore(&h->lock, flags);
8544
8545	if (hpsa_ctlr_needs_rescan(h)) {
8546		hpsa_ack_ctlr_events(h);
8547		hpsa_perform_rescan(h);
8548	}
8549
8550	spin_lock_irqsave(&h->lock, flags);
8551	if (!h->remove_in_progress)
8552		queue_delayed_work(h->monitor_ctlr_wq, &h->event_monitor_work,
8553				HPSA_EVENT_MONITOR_INTERVAL);
8554	spin_unlock_irqrestore(&h->lock, flags);
8555}
8556
8557static void hpsa_rescan_ctlr_worker(struct work_struct *work)
8558{
8559	unsigned long flags;
8560	struct ctlr_info *h = container_of(to_delayed_work(work),
8561					struct ctlr_info, rescan_ctlr_work);
8562
8563	spin_lock_irqsave(&h->lock, flags);
8564	if (h->remove_in_progress) {
8565		spin_unlock_irqrestore(&h->lock, flags);
8566		return;
8567	}
8568	spin_unlock_irqrestore(&h->lock, flags);
8569
8570	if (h->drv_req_rescan || hpsa_offline_devices_ready(h)) {
8571		hpsa_perform_rescan(h);
8572	} else if (h->discovery_polling) {
8573		if (hpsa_luns_changed(h)) {
8574			dev_info(&h->pdev->dev,
8575				"driver discovery polling rescan.\n");
8576			hpsa_perform_rescan(h);
8577		}
8578	}
8579	spin_lock_irqsave(&h->lock, flags);
8580	if (!h->remove_in_progress)
8581		queue_delayed_work(h->rescan_ctlr_wq, &h->rescan_ctlr_work,
8582				h->heartbeat_sample_interval);
8583	spin_unlock_irqrestore(&h->lock, flags);
8584}
8585
8586static void hpsa_monitor_ctlr_worker(struct work_struct *work)
8587{
8588	unsigned long flags;
8589	struct ctlr_info *h = container_of(to_delayed_work(work),
8590					struct ctlr_info, monitor_ctlr_work);
8591
8592	detect_controller_lockup(h);
8593	if (lockup_detected(h))
8594		return;
8595
8596	spin_lock_irqsave(&h->lock, flags);
8597	if (!h->remove_in_progress)
8598		queue_delayed_work(h->monitor_ctlr_wq, &h->monitor_ctlr_work,
8599				h->heartbeat_sample_interval);
8600	spin_unlock_irqrestore(&h->lock, flags);
8601}
8602
8603static struct workqueue_struct *hpsa_create_controller_wq(struct ctlr_info *h,
8604						char *name)
8605{
8606	struct workqueue_struct *wq = NULL;
8607
8608	wq = alloc_ordered_workqueue("%s_%d_hpsa", 0, name, h->ctlr);
8609	if (!wq)
8610		dev_err(&h->pdev->dev, "failed to create %s workqueue\n", name);
8611
8612	return wq;
8613}
8614
8615static void hpda_free_ctlr_info(struct ctlr_info *h)
8616{
8617	kfree(h->reply_map);
8618	kfree(h);
8619}
8620
8621static struct ctlr_info *hpda_alloc_ctlr_info(void)
8622{
8623	struct ctlr_info *h;
8624
8625	h = kzalloc(sizeof(*h), GFP_KERNEL);
8626	if (!h)
8627		return NULL;
8628
8629	h->reply_map = kcalloc(nr_cpu_ids, sizeof(*h->reply_map), GFP_KERNEL);
8630	if (!h->reply_map) {
8631		kfree(h);
8632		return NULL;
8633	}
8634	return h;
8635}
8636
8637static int hpsa_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
8638{
8639	int dac, rc;
8640	struct ctlr_info *h;
8641	int try_soft_reset = 0;
8642	unsigned long flags;
8643	u32 board_id;
8644
8645	if (number_of_controllers == 0)
8646		printk(KERN_INFO DRIVER_NAME "\n");
8647
8648	rc = hpsa_lookup_board_id(pdev, &board_id, NULL);
8649	if (rc < 0) {
8650		dev_warn(&pdev->dev, "Board ID not found\n");
8651		return rc;
8652	}
8653
8654	rc = hpsa_init_reset_devices(pdev, board_id);
8655	if (rc) {
8656		if (rc != -ENOTSUPP)
8657			return rc;
8658		/* If the reset fails in a particular way (it has no way to do
8659		 * a proper hard reset, so returns -ENOTSUPP) we can try to do
8660		 * a soft reset once we get the controller configured up to the
8661		 * point that it can accept a command.
8662		 */
8663		try_soft_reset = 1;
8664		rc = 0;
8665	}
8666
8667reinit_after_soft_reset:
8668
8669	/* Command structures must be aligned on a 32-byte boundary because
8670	 * the 5 lower bits of the address are used by the hardware. and by
8671	 * the driver.  See comments in hpsa.h for more info.
8672	 */
8673	BUILD_BUG_ON(sizeof(struct CommandList) % COMMANDLIST_ALIGNMENT);
8674	h = hpda_alloc_ctlr_info();
8675	if (!h) {
8676		dev_err(&pdev->dev, "Failed to allocate controller head\n");
8677		return -ENOMEM;
8678	}
8679
8680	h->pdev = pdev;
8681
8682	h->intr_mode = hpsa_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
8683	INIT_LIST_HEAD(&h->offline_device_list);
8684	spin_lock_init(&h->lock);
8685	spin_lock_init(&h->offline_device_lock);
8686	spin_lock_init(&h->scan_lock);
8687	spin_lock_init(&h->reset_lock);
8688	atomic_set(&h->passthru_cmds_avail, HPSA_MAX_CONCURRENT_PASSTHRUS);
8689
8690	/* Allocate and clear per-cpu variable lockup_detected */
8691	h->lockup_detected = alloc_percpu(u32);
8692	if (!h->lockup_detected) {
8693		dev_err(&h->pdev->dev, "Failed to allocate lockup detector\n");
8694		rc = -ENOMEM;
8695		goto clean1;	/* aer/h */
8696	}
8697	set_lockup_detected_for_all_cpus(h, 0);
8698
8699	rc = hpsa_pci_init(h);
8700	if (rc)
8701		goto clean2;	/* lu, aer/h */
8702
8703	/* relies on h-> settings made by hpsa_pci_init, including
8704	 * interrupt_mode h->intr */
8705	rc = hpsa_scsi_host_alloc(h);
8706	if (rc)
8707		goto clean2_5;	/* pci, lu, aer/h */
8708
8709	sprintf(h->devname, HPSA "%d", h->scsi_host->host_no);
8710	h->ctlr = number_of_controllers;
8711	number_of_controllers++;
8712
8713	/* configure PCI DMA stuff */
8714	rc = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
8715	if (rc == 0) {
8716		dac = 1;
8717	} else {
8718		rc = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
8719		if (rc == 0) {
8720			dac = 0;
8721		} else {
8722			dev_err(&pdev->dev, "no suitable DMA available\n");
8723			goto clean3;	/* shost, pci, lu, aer/h */
8724		}
8725	}
8726
8727	/* make sure the board interrupts are off */
8728	h->access.set_intr_mask(h, HPSA_INTR_OFF);
8729
8730	rc = hpsa_request_irqs(h, do_hpsa_intr_msi, do_hpsa_intr_intx);
8731	if (rc)
8732		goto clean3;	/* shost, pci, lu, aer/h */
8733	rc = hpsa_alloc_cmd_pool(h);
8734	if (rc)
8735		goto clean4;	/* irq, shost, pci, lu, aer/h */
8736	rc = hpsa_alloc_sg_chain_blocks(h);
8737	if (rc)
8738		goto clean5;	/* cmd, irq, shost, pci, lu, aer/h */
8739	init_waitqueue_head(&h->scan_wait_queue);
8740	init_waitqueue_head(&h->event_sync_wait_queue);
8741	mutex_init(&h->reset_mutex);
8742	h->scan_finished = 1; /* no scan currently in progress */
8743	h->scan_waiting = 0;
8744
8745	pci_set_drvdata(pdev, h);
8746	h->ndevices = 0;
8747
8748	spin_lock_init(&h->devlock);
8749	rc = hpsa_put_ctlr_into_performant_mode(h);
8750	if (rc)
8751		goto clean6; /* sg, cmd, irq, shost, pci, lu, aer/h */
8752
8753	/* create the resubmit workqueue */
8754	h->rescan_ctlr_wq = hpsa_create_controller_wq(h, "rescan");
8755	if (!h->rescan_ctlr_wq) {
8756		rc = -ENOMEM;
8757		goto clean7;
8758	}
8759
8760	h->resubmit_wq = hpsa_create_controller_wq(h, "resubmit");
8761	if (!h->resubmit_wq) {
8762		rc = -ENOMEM;
8763		goto clean7;	/* aer/h */
8764	}
8765
8766	h->monitor_ctlr_wq = hpsa_create_controller_wq(h, "monitor");
8767	if (!h->monitor_ctlr_wq) {
8768		rc = -ENOMEM;
8769		goto clean7;
8770	}
8771
8772	/*
8773	 * At this point, the controller is ready to take commands.
8774	 * Now, if reset_devices and the hard reset didn't work, try
8775	 * the soft reset and see if that works.
8776	 */
8777	if (try_soft_reset) {
8778
8779		/* This is kind of gross.  We may or may not get a completion
8780		 * from the soft reset command, and if we do, then the value
8781		 * from the fifo may or may not be valid.  So, we wait 10 secs
8782		 * after the reset throwing away any completions we get during
8783		 * that time.  Unregister the interrupt handler and register
8784		 * fake ones to scoop up any residual completions.
8785		 */
8786		spin_lock_irqsave(&h->lock, flags);
8787		h->access.set_intr_mask(h, HPSA_INTR_OFF);
8788		spin_unlock_irqrestore(&h->lock, flags);
8789		hpsa_free_irqs(h);
8790		rc = hpsa_request_irqs(h, hpsa_msix_discard_completions,
8791					hpsa_intx_discard_completions);
8792		if (rc) {
8793			dev_warn(&h->pdev->dev,
8794				"Failed to request_irq after soft reset.\n");
8795			/*
8796			 * cannot goto clean7 or free_irqs will be called
8797			 * again. Instead, do its work
8798			 */
8799			hpsa_free_performant_mode(h);	/* clean7 */
8800			hpsa_free_sg_chain_blocks(h);	/* clean6 */
8801			hpsa_free_cmd_pool(h);		/* clean5 */
8802			/*
8803			 * skip hpsa_free_irqs(h) clean4 since that
8804			 * was just called before request_irqs failed
8805			 */
8806			goto clean3;
8807		}
8808
8809		rc = hpsa_kdump_soft_reset(h);
8810		if (rc)
8811			/* Neither hard nor soft reset worked, we're hosed. */
8812			goto clean7;
8813
8814		dev_info(&h->pdev->dev, "Board READY.\n");
8815		dev_info(&h->pdev->dev,
8816			"Waiting for stale completions to drain.\n");
8817		h->access.set_intr_mask(h, HPSA_INTR_ON);
8818		msleep(10000);
8819		h->access.set_intr_mask(h, HPSA_INTR_OFF);
8820
8821		rc = controller_reset_failed(h->cfgtable);
8822		if (rc)
8823			dev_info(&h->pdev->dev,
8824				"Soft reset appears to have failed.\n");
8825
8826		/* since the controller's reset, we have to go back and re-init
8827		 * everything.  Easiest to just forget what we've done and do it
8828		 * all over again.
8829		 */
8830		hpsa_undo_allocations_after_kdump_soft_reset(h);
8831		try_soft_reset = 0;
8832		if (rc)
8833			/* don't goto clean, we already unallocated */
8834			return -ENODEV;
8835
8836		goto reinit_after_soft_reset;
8837	}
8838
8839	/* Enable Accelerated IO path at driver layer */
8840	h->acciopath_status = 1;
8841	/* Disable discovery polling.*/
8842	h->discovery_polling = 0;
8843
8844
8845	/* Turn the interrupts on so we can service requests */
8846	h->access.set_intr_mask(h, HPSA_INTR_ON);
8847
8848	hpsa_hba_inquiry(h);
8849
8850	h->lastlogicals = kzalloc(sizeof(*(h->lastlogicals)), GFP_KERNEL);
8851	if (!h->lastlogicals)
8852		dev_info(&h->pdev->dev,
8853			"Can't track change to report lun data\n");
8854
8855	/* hook into SCSI subsystem */
8856	rc = hpsa_scsi_add_host(h);
8857	if (rc)
8858		goto clean8; /* lastlogicals, perf, sg, cmd, irq, shost, pci, lu, aer/h */
8859
8860	/* Monitor the controller for firmware lockups */
8861	h->heartbeat_sample_interval = HEARTBEAT_SAMPLE_INTERVAL;
8862	INIT_DELAYED_WORK(&h->monitor_ctlr_work, hpsa_monitor_ctlr_worker);
8863	schedule_delayed_work(&h->monitor_ctlr_work,
8864				h->heartbeat_sample_interval);
8865	INIT_DELAYED_WORK(&h->rescan_ctlr_work, hpsa_rescan_ctlr_worker);
8866	queue_delayed_work(h->rescan_ctlr_wq, &h->rescan_ctlr_work,
8867				h->heartbeat_sample_interval);
8868	INIT_DELAYED_WORK(&h->event_monitor_work, hpsa_event_monitor_worker);
8869	schedule_delayed_work(&h->event_monitor_work,
8870				HPSA_EVENT_MONITOR_INTERVAL);
8871	return 0;
8872
8873clean8: /* lastlogicals, perf, sg, cmd, irq, shost, pci, lu, aer/h */
8874	kfree(h->lastlogicals);
8875clean7: /* perf, sg, cmd, irq, shost, pci, lu, aer/h */
8876	hpsa_free_performant_mode(h);
8877	h->access.set_intr_mask(h, HPSA_INTR_OFF);
8878clean6: /* sg, cmd, irq, pci, lockup, wq/aer/h */
8879	hpsa_free_sg_chain_blocks(h);
8880clean5: /* cmd, irq, shost, pci, lu, aer/h */
8881	hpsa_free_cmd_pool(h);
8882clean4: /* irq, shost, pci, lu, aer/h */
8883	hpsa_free_irqs(h);
8884clean3: /* shost, pci, lu, aer/h */
8885	scsi_host_put(h->scsi_host);
8886	h->scsi_host = NULL;
8887clean2_5: /* pci, lu, aer/h */
8888	hpsa_free_pci_init(h);
8889clean2: /* lu, aer/h */
8890	if (h->lockup_detected) {
8891		free_percpu(h->lockup_detected);
8892		h->lockup_detected = NULL;
8893	}
8894clean1:	/* wq/aer/h */
8895	if (h->resubmit_wq) {
8896		destroy_workqueue(h->resubmit_wq);
8897		h->resubmit_wq = NULL;
8898	}
8899	if (h->rescan_ctlr_wq) {
8900		destroy_workqueue(h->rescan_ctlr_wq);
8901		h->rescan_ctlr_wq = NULL;
8902	}
8903	if (h->monitor_ctlr_wq) {
8904		destroy_workqueue(h->monitor_ctlr_wq);
8905		h->monitor_ctlr_wq = NULL;
8906	}
8907	hpda_free_ctlr_info(h);
8908	return rc;
8909}
8910
8911static void hpsa_flush_cache(struct ctlr_info *h)
8912{
8913	char *flush_buf;
8914	struct CommandList *c;
8915	int rc;
8916
8917	if (unlikely(lockup_detected(h)))
8918		return;
8919	flush_buf = kzalloc(4, GFP_KERNEL);
8920	if (!flush_buf)
8921		return;
8922
8923	c = cmd_alloc(h);
8924
8925	if (fill_cmd(c, HPSA_CACHE_FLUSH, h, flush_buf, 4, 0,
8926		RAID_CTLR_LUNID, TYPE_CMD)) {
8927		goto out;
8928	}
8929	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_TO_DEVICE,
8930			DEFAULT_TIMEOUT);
8931	if (rc)
8932		goto out;
8933	if (c->err_info->CommandStatus != 0)
8934out:
8935		dev_warn(&h->pdev->dev,
8936			"error flushing cache on controller\n");
8937	cmd_free(h, c);
8938	kfree(flush_buf);
8939}
8940
8941/* Make controller gather fresh report lun data each time we
8942 * send down a report luns request
8943 */
8944static void hpsa_disable_rld_caching(struct ctlr_info *h)
8945{
8946	u32 *options;
8947	struct CommandList *c;
8948	int rc;
8949
8950	/* Don't bother trying to set diag options if locked up */
8951	if (unlikely(h->lockup_detected))
8952		return;
8953
8954	options = kzalloc(sizeof(*options), GFP_KERNEL);
8955	if (!options)
8956		return;
8957
8958	c = cmd_alloc(h);
8959
8960	/* first, get the current diag options settings */
8961	if (fill_cmd(c, BMIC_SENSE_DIAG_OPTIONS, h, options, 4, 0,
8962		RAID_CTLR_LUNID, TYPE_CMD))
8963		goto errout;
8964
8965	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
8966			NO_TIMEOUT);
8967	if ((rc != 0) || (c->err_info->CommandStatus != 0))
8968		goto errout;
8969
8970	/* Now, set the bit for disabling the RLD caching */
8971	*options |= HPSA_DIAG_OPTS_DISABLE_RLD_CACHING;
8972
8973	if (fill_cmd(c, BMIC_SET_DIAG_OPTIONS, h, options, 4, 0,
8974		RAID_CTLR_LUNID, TYPE_CMD))
8975		goto errout;
8976
8977	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_TO_DEVICE,
8978			NO_TIMEOUT);
8979	if ((rc != 0)  || (c->err_info->CommandStatus != 0))
8980		goto errout;
8981
8982	/* Now verify that it got set: */
8983	if (fill_cmd(c, BMIC_SENSE_DIAG_OPTIONS, h, options, 4, 0,
8984		RAID_CTLR_LUNID, TYPE_CMD))
8985		goto errout;
8986
8987	rc = hpsa_scsi_do_simple_cmd_with_retry(h, c, DMA_FROM_DEVICE,
8988			NO_TIMEOUT);
8989	if ((rc != 0)  || (c->err_info->CommandStatus != 0))
8990		goto errout;
8991
8992	if (*options & HPSA_DIAG_OPTS_DISABLE_RLD_CACHING)
8993		goto out;
8994
8995errout:
8996	dev_err(&h->pdev->dev,
8997			"Error: failed to disable report lun data caching.\n");
8998out:
8999	cmd_free(h, c);
9000	kfree(options);
9001}
9002
9003static void __hpsa_shutdown(struct pci_dev *pdev)
9004{
9005	struct ctlr_info *h;
9006
9007	h = pci_get_drvdata(pdev);
9008	/* Turn board interrupts off  and send the flush cache command
9009	 * sendcmd will turn off interrupt, and send the flush...
9010	 * To write all data in the battery backed cache to disks
9011	 */
9012	hpsa_flush_cache(h);
9013	h->access.set_intr_mask(h, HPSA_INTR_OFF);
9014	hpsa_free_irqs(h);			/* init_one 4 */
9015	hpsa_disable_interrupt_mode(h);		/* pci_init 2 */
9016}
9017
9018static void hpsa_shutdown(struct pci_dev *pdev)
9019{
9020	__hpsa_shutdown(pdev);
9021	pci_disable_device(pdev);
9022}
9023
9024static void hpsa_free_device_info(struct ctlr_info *h)
9025{
9026	int i;
9027
9028	for (i = 0; i < h->ndevices; i++) {
9029		kfree(h->dev[i]);
9030		h->dev[i] = NULL;
9031	}
9032}
9033
9034static void hpsa_remove_one(struct pci_dev *pdev)
9035{
9036	struct ctlr_info *h;
9037	unsigned long flags;
9038
9039	if (pci_get_drvdata(pdev) == NULL) {
9040		dev_err(&pdev->dev, "unable to remove device\n");
9041		return;
9042	}
9043	h = pci_get_drvdata(pdev);
9044
9045	/* Get rid of any controller monitoring work items */
9046	spin_lock_irqsave(&h->lock, flags);
9047	h->remove_in_progress = 1;
9048	spin_unlock_irqrestore(&h->lock, flags);
9049	cancel_delayed_work_sync(&h->monitor_ctlr_work);
9050	cancel_delayed_work_sync(&h->rescan_ctlr_work);
9051	cancel_delayed_work_sync(&h->event_monitor_work);
9052	destroy_workqueue(h->rescan_ctlr_wq);
9053	destroy_workqueue(h->resubmit_wq);
9054	destroy_workqueue(h->monitor_ctlr_wq);
9055
9056	hpsa_delete_sas_host(h);
9057
9058	/*
9059	 * Call before disabling interrupts.
9060	 * scsi_remove_host can trigger I/O operations especially
9061	 * when multipath is enabled. There can be SYNCHRONIZE CACHE
9062	 * operations which cannot complete and will hang the system.
9063	 */
9064	if (h->scsi_host)
9065		scsi_remove_host(h->scsi_host);		/* init_one 8 */
9066	/* includes hpsa_free_irqs - init_one 4 */
9067	/* includes hpsa_disable_interrupt_mode - pci_init 2 */
9068	__hpsa_shutdown(pdev);
9069
9070	hpsa_free_device_info(h);		/* scan */
9071
9072	kfree(h->hba_inquiry_data);			/* init_one 10 */
9073	h->hba_inquiry_data = NULL;			/* init_one 10 */
9074	hpsa_free_ioaccel2_sg_chain_blocks(h);
9075	hpsa_free_performant_mode(h);			/* init_one 7 */
9076	hpsa_free_sg_chain_blocks(h);			/* init_one 6 */
9077	hpsa_free_cmd_pool(h);				/* init_one 5 */
9078	kfree(h->lastlogicals);
9079
9080	/* hpsa_free_irqs already called via hpsa_shutdown init_one 4 */
9081
9082	scsi_host_put(h->scsi_host);			/* init_one 3 */
9083	h->scsi_host = NULL;				/* init_one 3 */
9084
9085	/* includes hpsa_disable_interrupt_mode - pci_init 2 */
9086	hpsa_free_pci_init(h);				/* init_one 2.5 */
9087
9088	free_percpu(h->lockup_detected);		/* init_one 2 */
9089	h->lockup_detected = NULL;			/* init_one 2 */
9090	/* (void) pci_disable_pcie_error_reporting(pdev); */	/* init_one 1 */
9091
9092	hpda_free_ctlr_info(h);				/* init_one 1 */
9093}
9094
9095static int hpsa_suspend(__attribute__((unused)) struct pci_dev *pdev,
9096	__attribute__((unused)) pm_message_t state)
9097{
9098	return -ENOSYS;
9099}
9100
9101static int hpsa_resume(__attribute__((unused)) struct pci_dev *pdev)
9102{
9103	return -ENOSYS;
9104}
9105
9106static struct pci_driver hpsa_pci_driver = {
9107	.name = HPSA,
9108	.probe = hpsa_init_one,
9109	.remove = hpsa_remove_one,
9110	.id_table = hpsa_pci_device_id,	/* id_table */
9111	.shutdown = hpsa_shutdown,
9112	.suspend = hpsa_suspend,
9113	.resume = hpsa_resume,
9114};
9115
9116/* Fill in bucket_map[], given nsgs (the max number of
9117 * scatter gather elements supported) and bucket[],
9118 * which is an array of 8 integers.  The bucket[] array
9119 * contains 8 different DMA transfer sizes (in 16
9120 * byte increments) which the controller uses to fetch
9121 * commands.  This function fills in bucket_map[], which
9122 * maps a given number of scatter gather elements to one of
9123 * the 8 DMA transfer sizes.  The point of it is to allow the
9124 * controller to only do as much DMA as needed to fetch the
9125 * command, with the DMA transfer size encoded in the lower
9126 * bits of the command address.
9127 */
9128static void  calc_bucket_map(int bucket[], int num_buckets,
9129	int nsgs, int min_blocks, u32 *bucket_map)
9130{
9131	int i, j, b, size;
9132
9133	/* Note, bucket_map must have nsgs+1 entries. */
9134	for (i = 0; i <= nsgs; i++) {
9135		/* Compute size of a command with i SG entries */
9136		size = i + min_blocks;
9137		b = num_buckets; /* Assume the biggest bucket */
9138		/* Find the bucket that is just big enough */
9139		for (j = 0; j < num_buckets; j++) {
9140			if (bucket[j] >= size) {
9141				b = j;
9142				break;
9143			}
9144		}
9145		/* for a command with i SG entries, use bucket b. */
9146		bucket_map[i] = b;
9147	}
9148}
9149
9150/*
9151 * return -ENODEV on err, 0 on success (or no action)
9152 * allocates numerous items that must be freed later
9153 */
9154static int hpsa_enter_performant_mode(struct ctlr_info *h, u32 trans_support)
9155{
9156	int i;
9157	unsigned long register_value;
9158	unsigned long transMethod = CFGTBL_Trans_Performant |
9159			(trans_support & CFGTBL_Trans_use_short_tags) |
9160				CFGTBL_Trans_enable_directed_msix |
9161			(trans_support & (CFGTBL_Trans_io_accel1 |
9162				CFGTBL_Trans_io_accel2));
9163	struct access_method access = SA5_performant_access;
9164
9165	/* This is a bit complicated.  There are 8 registers on
9166	 * the controller which we write to to tell it 8 different
9167	 * sizes of commands which there may be.  It's a way of
9168	 * reducing the DMA done to fetch each command.  Encoded into
9169	 * each command's tag are 3 bits which communicate to the controller
9170	 * which of the eight sizes that command fits within.  The size of
9171	 * each command depends on how many scatter gather entries there are.
9172	 * Each SG entry requires 16 bytes.  The eight registers are programmed
9173	 * with the number of 16-byte blocks a command of that size requires.
9174	 * The smallest command possible requires 5 such 16 byte blocks.
9175	 * the largest command possible requires SG_ENTRIES_IN_CMD + 4 16-byte
9176	 * blocks.  Note, this only extends to the SG entries contained
9177	 * within the command block, and does not extend to chained blocks
9178	 * of SG elements.   bft[] contains the eight values we write to
9179	 * the registers.  They are not evenly distributed, but have more
9180	 * sizes for small commands, and fewer sizes for larger commands.
9181	 */
9182	int bft[8] = {5, 6, 8, 10, 12, 20, 28, SG_ENTRIES_IN_CMD + 4};
9183#define MIN_IOACCEL2_BFT_ENTRY 5
9184#define HPSA_IOACCEL2_HEADER_SZ 4
9185	int bft2[16] = {MIN_IOACCEL2_BFT_ENTRY, 6, 7, 8, 9, 10, 11, 12,
9186			13, 14, 15, 16, 17, 18, 19,
9187			HPSA_IOACCEL2_HEADER_SZ + IOACCEL2_MAXSGENTRIES};
9188	BUILD_BUG_ON(ARRAY_SIZE(bft2) != 16);
9189	BUILD_BUG_ON(ARRAY_SIZE(bft) != 8);
9190	BUILD_BUG_ON(offsetof(struct io_accel2_cmd, sg) >
9191				 16 * MIN_IOACCEL2_BFT_ENTRY);
9192	BUILD_BUG_ON(sizeof(struct ioaccel2_sg_element) != 16);
9193	BUILD_BUG_ON(28 > SG_ENTRIES_IN_CMD + 4);
9194	/*  5 = 1 s/g entry or 4k
9195	 *  6 = 2 s/g entry or 8k
9196	 *  8 = 4 s/g entry or 16k
9197	 * 10 = 6 s/g entry or 24k
9198	 */
9199
9200	/* If the controller supports either ioaccel method then
9201	 * we can also use the RAID stack submit path that does not
9202	 * perform the superfluous readl() after each command submission.
9203	 */
9204	if (trans_support & (CFGTBL_Trans_io_accel1 | CFGTBL_Trans_io_accel2))
9205		access = SA5_performant_access_no_read;
9206
9207	/* Controller spec: zero out this buffer. */
9208	for (i = 0; i < h->nreply_queues; i++)
9209		memset(h->reply_queue[i].head, 0, h->reply_queue_size);
9210
9211	bft[7] = SG_ENTRIES_IN_CMD + 4;
9212	calc_bucket_map(bft, ARRAY_SIZE(bft),
9213				SG_ENTRIES_IN_CMD, 4, h->blockFetchTable);
9214	for (i = 0; i < 8; i++)
9215		writel(bft[i], &h->transtable->BlockFetch[i]);
9216
9217	/* size of controller ring buffer */
9218	writel(h->max_commands, &h->transtable->RepQSize);
9219	writel(h->nreply_queues, &h->transtable->RepQCount);
9220	writel(0, &h->transtable->RepQCtrAddrLow32);
9221	writel(0, &h->transtable->RepQCtrAddrHigh32);
9222
9223	for (i = 0; i < h->nreply_queues; i++) {
9224		writel(0, &h->transtable->RepQAddr[i].upper);
9225		writel(h->reply_queue[i].busaddr,
9226			&h->transtable->RepQAddr[i].lower);
9227	}
9228
9229	writel(0, &h->cfgtable->HostWrite.command_pool_addr_hi);
9230	writel(transMethod, &(h->cfgtable->HostWrite.TransportRequest));
9231	/*
9232	 * enable outbound interrupt coalescing in accelerator mode;
9233	 */
9234	if (trans_support & CFGTBL_Trans_io_accel1) {
9235		access = SA5_ioaccel_mode1_access;
9236		writel(10, &h->cfgtable->HostWrite.CoalIntDelay);
9237		writel(4, &h->cfgtable->HostWrite.CoalIntCount);
9238	} else
9239		if (trans_support & CFGTBL_Trans_io_accel2)
9240			access = SA5_ioaccel_mode2_access;
9241	writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
9242	if (hpsa_wait_for_mode_change_ack(h)) {
9243		dev_err(&h->pdev->dev,
9244			"performant mode problem - doorbell timeout\n");
9245		return -ENODEV;
9246	}
9247	register_value = readl(&(h->cfgtable->TransportActive));
9248	if (!(register_value & CFGTBL_Trans_Performant)) {
9249		dev_err(&h->pdev->dev,
9250			"performant mode problem - transport not active\n");
9251		return -ENODEV;
9252	}
9253	/* Change the access methods to the performant access methods */
9254	h->access = access;
9255	h->transMethod = transMethod;
9256
9257	if (!((trans_support & CFGTBL_Trans_io_accel1) ||
9258		(trans_support & CFGTBL_Trans_io_accel2)))
9259		return 0;
9260
9261	if (trans_support & CFGTBL_Trans_io_accel1) {
9262		/* Set up I/O accelerator mode */
9263		for (i = 0; i < h->nreply_queues; i++) {
9264			writel(i, h->vaddr + IOACCEL_MODE1_REPLY_QUEUE_INDEX);
9265			h->reply_queue[i].current_entry =
9266				readl(h->vaddr + IOACCEL_MODE1_PRODUCER_INDEX);
9267		}
9268		bft[7] = h->ioaccel_maxsg + 8;
9269		calc_bucket_map(bft, ARRAY_SIZE(bft), h->ioaccel_maxsg, 8,
9270				h->ioaccel1_blockFetchTable);
9271
9272		/* initialize all reply queue entries to unused */
9273		for (i = 0; i < h->nreply_queues; i++)
9274			memset(h->reply_queue[i].head,
9275				(u8) IOACCEL_MODE1_REPLY_UNUSED,
9276				h->reply_queue_size);
9277
9278		/* set all the constant fields in the accelerator command
9279		 * frames once at init time to save CPU cycles later.
9280		 */
9281		for (i = 0; i < h->nr_cmds; i++) {
9282			struct io_accel1_cmd *cp = &h->ioaccel_cmd_pool[i];
9283
9284			cp->function = IOACCEL1_FUNCTION_SCSIIO;
9285			cp->err_info = (u32) (h->errinfo_pool_dhandle +
9286					(i * sizeof(struct ErrorInfo)));
9287			cp->err_info_len = sizeof(struct ErrorInfo);
9288			cp->sgl_offset = IOACCEL1_SGLOFFSET;
9289			cp->host_context_flags =
9290				cpu_to_le16(IOACCEL1_HCFLAGS_CISS_FORMAT);
9291			cp->timeout_sec = 0;
9292			cp->ReplyQueue = 0;
9293			cp->tag =
9294				cpu_to_le64((i << DIRECT_LOOKUP_SHIFT));
9295			cp->host_addr =
9296				cpu_to_le64(h->ioaccel_cmd_pool_dhandle +
9297					(i * sizeof(struct io_accel1_cmd)));
9298		}
9299	} else if (trans_support & CFGTBL_Trans_io_accel2) {
9300		u64 cfg_offset, cfg_base_addr_index;
9301		u32 bft2_offset, cfg_base_addr;
9302		int rc;
9303
9304		rc = hpsa_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
9305			&cfg_base_addr_index, &cfg_offset);
9306		BUILD_BUG_ON(offsetof(struct io_accel2_cmd, sg) != 64);
9307		bft2[15] = h->ioaccel_maxsg + HPSA_IOACCEL2_HEADER_SZ;
9308		calc_bucket_map(bft2, ARRAY_SIZE(bft2), h->ioaccel_maxsg,
9309				4, h->ioaccel2_blockFetchTable);
9310		bft2_offset = readl(&h->cfgtable->io_accel_request_size_offset);
9311		BUILD_BUG_ON(offsetof(struct CfgTable,
9312				io_accel_request_size_offset) != 0xb8);
9313		h->ioaccel2_bft2_regs =
9314			remap_pci_mem(pci_resource_start(h->pdev,
9315					cfg_base_addr_index) +
9316					cfg_offset + bft2_offset,
9317					ARRAY_SIZE(bft2) *
9318					sizeof(*h->ioaccel2_bft2_regs));
9319		for (i = 0; i < ARRAY_SIZE(bft2); i++)
9320			writel(bft2[i], &h->ioaccel2_bft2_regs[i]);
9321	}
9322	writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
9323	if (hpsa_wait_for_mode_change_ack(h)) {
9324		dev_err(&h->pdev->dev,
9325			"performant mode problem - enabling ioaccel mode\n");
9326		return -ENODEV;
9327	}
9328	return 0;
9329}
9330
9331/* Free ioaccel1 mode command blocks and block fetch table */
9332static void hpsa_free_ioaccel1_cmd_and_bft(struct ctlr_info *h)
9333{
9334	if (h->ioaccel_cmd_pool) {
9335		dma_free_coherent(&h->pdev->dev,
9336				  h->nr_cmds * sizeof(*h->ioaccel_cmd_pool),
9337				  h->ioaccel_cmd_pool,
9338				  h->ioaccel_cmd_pool_dhandle);
9339		h->ioaccel_cmd_pool = NULL;
9340		h->ioaccel_cmd_pool_dhandle = 0;
9341	}
9342	kfree(h->ioaccel1_blockFetchTable);
9343	h->ioaccel1_blockFetchTable = NULL;
9344}
9345
9346/* Allocate ioaccel1 mode command blocks and block fetch table */
9347static int hpsa_alloc_ioaccel1_cmd_and_bft(struct ctlr_info *h)
9348{
9349	h->ioaccel_maxsg =
9350		readl(&(h->cfgtable->io_accel_max_embedded_sg_count));
9351	if (h->ioaccel_maxsg > IOACCEL1_MAXSGENTRIES)
9352		h->ioaccel_maxsg = IOACCEL1_MAXSGENTRIES;
9353
9354	/* Command structures must be aligned on a 128-byte boundary
9355	 * because the 7 lower bits of the address are used by the
9356	 * hardware.
9357	 */
9358	BUILD_BUG_ON(sizeof(struct io_accel1_cmd) %
9359			IOACCEL1_COMMANDLIST_ALIGNMENT);
9360	h->ioaccel_cmd_pool =
9361		dma_alloc_coherent(&h->pdev->dev,
9362			h->nr_cmds * sizeof(*h->ioaccel_cmd_pool),
9363			&h->ioaccel_cmd_pool_dhandle, GFP_KERNEL);
9364
9365	h->ioaccel1_blockFetchTable =
9366		kmalloc(((h->ioaccel_maxsg + 1) *
9367				sizeof(u32)), GFP_KERNEL);
9368
9369	if ((h->ioaccel_cmd_pool == NULL) ||
9370		(h->ioaccel1_blockFetchTable == NULL))
9371		goto clean_up;
9372
9373	memset(h->ioaccel_cmd_pool, 0,
9374		h->nr_cmds * sizeof(*h->ioaccel_cmd_pool));
9375	return 0;
9376
9377clean_up:
9378	hpsa_free_ioaccel1_cmd_and_bft(h);
9379	return -ENOMEM;
9380}
9381
9382/* Free ioaccel2 mode command blocks and block fetch table */
9383static void hpsa_free_ioaccel2_cmd_and_bft(struct ctlr_info *h)
9384{
9385	hpsa_free_ioaccel2_sg_chain_blocks(h);
9386
9387	if (h->ioaccel2_cmd_pool) {
9388		dma_free_coherent(&h->pdev->dev,
9389				  h->nr_cmds * sizeof(*h->ioaccel2_cmd_pool),
9390				  h->ioaccel2_cmd_pool,
9391				  h->ioaccel2_cmd_pool_dhandle);
9392		h->ioaccel2_cmd_pool = NULL;
9393		h->ioaccel2_cmd_pool_dhandle = 0;
9394	}
9395	kfree(h->ioaccel2_blockFetchTable);
9396	h->ioaccel2_blockFetchTable = NULL;
9397}
9398
9399/* Allocate ioaccel2 mode command blocks and block fetch table */
9400static int hpsa_alloc_ioaccel2_cmd_and_bft(struct ctlr_info *h)
9401{
9402	int rc;
9403
9404	/* Allocate ioaccel2 mode command blocks and block fetch table */
9405
9406	h->ioaccel_maxsg =
9407		readl(&(h->cfgtable->io_accel_max_embedded_sg_count));
9408	if (h->ioaccel_maxsg > IOACCEL2_MAXSGENTRIES)
9409		h->ioaccel_maxsg = IOACCEL2_MAXSGENTRIES;
9410
9411	BUILD_BUG_ON(sizeof(struct io_accel2_cmd) %
9412			IOACCEL2_COMMANDLIST_ALIGNMENT);
9413	h->ioaccel2_cmd_pool =
9414		dma_alloc_coherent(&h->pdev->dev,
9415			h->nr_cmds * sizeof(*h->ioaccel2_cmd_pool),
9416			&h->ioaccel2_cmd_pool_dhandle, GFP_KERNEL);
9417
9418	h->ioaccel2_blockFetchTable =
9419		kmalloc(((h->ioaccel_maxsg + 1) *
9420				sizeof(u32)), GFP_KERNEL);
9421
9422	if ((h->ioaccel2_cmd_pool == NULL) ||
9423		(h->ioaccel2_blockFetchTable == NULL)) {
9424		rc = -ENOMEM;
9425		goto clean_up;
9426	}
9427
9428	rc = hpsa_allocate_ioaccel2_sg_chain_blocks(h);
9429	if (rc)
9430		goto clean_up;
9431
9432	memset(h->ioaccel2_cmd_pool, 0,
9433		h->nr_cmds * sizeof(*h->ioaccel2_cmd_pool));
9434	return 0;
9435
9436clean_up:
9437	hpsa_free_ioaccel2_cmd_and_bft(h);
9438	return rc;
9439}
9440
9441/* Free items allocated by hpsa_put_ctlr_into_performant_mode */
9442static void hpsa_free_performant_mode(struct ctlr_info *h)
9443{
9444	kfree(h->blockFetchTable);
9445	h->blockFetchTable = NULL;
9446	hpsa_free_reply_queues(h);
9447	hpsa_free_ioaccel1_cmd_and_bft(h);
9448	hpsa_free_ioaccel2_cmd_and_bft(h);
9449}
9450
9451/* return -ENODEV on error, 0 on success (or no action)
9452 * allocates numerous items that must be freed later
9453 */
9454static int hpsa_put_ctlr_into_performant_mode(struct ctlr_info *h)
9455{
9456	u32 trans_support;
9457	unsigned long transMethod = CFGTBL_Trans_Performant |
9458					CFGTBL_Trans_use_short_tags;
9459	int i, rc;
9460
9461	if (hpsa_simple_mode)
9462		return 0;
9463
9464	trans_support = readl(&(h->cfgtable->TransportSupport));
9465	if (!(trans_support & PERFORMANT_MODE))
9466		return 0;
9467
9468	/* Check for I/O accelerator mode support */
9469	if (trans_support & CFGTBL_Trans_io_accel1) {
9470		transMethod |= CFGTBL_Trans_io_accel1 |
9471				CFGTBL_Trans_enable_directed_msix;
9472		rc = hpsa_alloc_ioaccel1_cmd_and_bft(h);
9473		if (rc)
9474			return rc;
9475	} else if (trans_support & CFGTBL_Trans_io_accel2) {
9476		transMethod |= CFGTBL_Trans_io_accel2 |
9477				CFGTBL_Trans_enable_directed_msix;
9478		rc = hpsa_alloc_ioaccel2_cmd_and_bft(h);
9479		if (rc)
9480			return rc;
9481	}
9482
9483	h->nreply_queues = h->msix_vectors > 0 ? h->msix_vectors : 1;
9484	hpsa_get_max_perf_mode_cmds(h);
9485	/* Performant mode ring buffer and supporting data structures */
9486	h->reply_queue_size = h->max_commands * sizeof(u64);
9487
9488	for (i = 0; i < h->nreply_queues; i++) {
9489		h->reply_queue[i].head = dma_alloc_coherent(&h->pdev->dev,
9490						h->reply_queue_size,
9491						&h->reply_queue[i].busaddr,
9492						GFP_KERNEL);
9493		if (!h->reply_queue[i].head) {
9494			rc = -ENOMEM;
9495			goto clean1;	/* rq, ioaccel */
9496		}
9497		h->reply_queue[i].size = h->max_commands;
9498		h->reply_queue[i].wraparound = 1;  /* spec: init to 1 */
9499		h->reply_queue[i].current_entry = 0;
9500	}
9501
9502	/* Need a block fetch table for performant mode */
9503	h->blockFetchTable = kmalloc(((SG_ENTRIES_IN_CMD + 1) *
9504				sizeof(u32)), GFP_KERNEL);
9505	if (!h->blockFetchTable) {
9506		rc = -ENOMEM;
9507		goto clean1;	/* rq, ioaccel */
9508	}
9509
9510	rc = hpsa_enter_performant_mode(h, trans_support);
9511	if (rc)
9512		goto clean2;	/* bft, rq, ioaccel */
9513	return 0;
9514
9515clean2:	/* bft, rq, ioaccel */
9516	kfree(h->blockFetchTable);
9517	h->blockFetchTable = NULL;
9518clean1:	/* rq, ioaccel */
9519	hpsa_free_reply_queues(h);
9520	hpsa_free_ioaccel1_cmd_and_bft(h);
9521	hpsa_free_ioaccel2_cmd_and_bft(h);
9522	return rc;
9523}
9524
9525static int is_accelerated_cmd(struct CommandList *c)
9526{
9527	return c->cmd_type == CMD_IOACCEL1 || c->cmd_type == CMD_IOACCEL2;
9528}
9529
9530static void hpsa_drain_accel_commands(struct ctlr_info *h)
9531{
9532	struct CommandList *c = NULL;
9533	int i, accel_cmds_out;
9534	int refcount;
9535
9536	do { /* wait for all outstanding ioaccel commands to drain out */
9537		accel_cmds_out = 0;
9538		for (i = 0; i < h->nr_cmds; i++) {
9539			c = h->cmd_pool + i;
9540			refcount = atomic_inc_return(&c->refcount);
9541			if (refcount > 1) /* Command is allocated */
9542				accel_cmds_out += is_accelerated_cmd(c);
9543			cmd_free(h, c);
9544		}
9545		if (accel_cmds_out <= 0)
9546			break;
9547		msleep(100);
9548	} while (1);
9549}
9550
9551static struct hpsa_sas_phy *hpsa_alloc_sas_phy(
9552				struct hpsa_sas_port *hpsa_sas_port)
9553{
9554	struct hpsa_sas_phy *hpsa_sas_phy;
9555	struct sas_phy *phy;
9556
9557	hpsa_sas_phy = kzalloc(sizeof(*hpsa_sas_phy), GFP_KERNEL);
9558	if (!hpsa_sas_phy)
9559		return NULL;
9560
9561	phy = sas_phy_alloc(hpsa_sas_port->parent_node->parent_dev,
9562		hpsa_sas_port->next_phy_index);
9563	if (!phy) {
9564		kfree(hpsa_sas_phy);
9565		return NULL;
9566	}
9567
9568	hpsa_sas_port->next_phy_index++;
9569	hpsa_sas_phy->phy = phy;
9570	hpsa_sas_phy->parent_port = hpsa_sas_port;
9571
9572	return hpsa_sas_phy;
9573}
9574
9575static void hpsa_free_sas_phy(struct hpsa_sas_phy *hpsa_sas_phy)
9576{
9577	struct sas_phy *phy = hpsa_sas_phy->phy;
9578
9579	sas_port_delete_phy(hpsa_sas_phy->parent_port->port, phy);
9580	if (hpsa_sas_phy->added_to_port)
9581		list_del(&hpsa_sas_phy->phy_list_entry);
9582	sas_phy_delete(phy);
9583	kfree(hpsa_sas_phy);
9584}
9585
9586static int hpsa_sas_port_add_phy(struct hpsa_sas_phy *hpsa_sas_phy)
9587{
9588	int rc;
9589	struct hpsa_sas_port *hpsa_sas_port;
9590	struct sas_phy *phy;
9591	struct sas_identify *identify;
9592
9593	hpsa_sas_port = hpsa_sas_phy->parent_port;
9594	phy = hpsa_sas_phy->phy;
9595
9596	identify = &phy->identify;
9597	memset(identify, 0, sizeof(*identify));
9598	identify->sas_address = hpsa_sas_port->sas_address;
9599	identify->device_type = SAS_END_DEVICE;
9600	identify->initiator_port_protocols = SAS_PROTOCOL_STP;
9601	identify->target_port_protocols = SAS_PROTOCOL_STP;
9602	phy->minimum_linkrate_hw = SAS_LINK_RATE_UNKNOWN;
9603	phy->maximum_linkrate_hw = SAS_LINK_RATE_UNKNOWN;
9604	phy->minimum_linkrate = SAS_LINK_RATE_UNKNOWN;
9605	phy->maximum_linkrate = SAS_LINK_RATE_UNKNOWN;
9606	phy->negotiated_linkrate = SAS_LINK_RATE_UNKNOWN;
9607
9608	rc = sas_phy_add(hpsa_sas_phy->phy);
9609	if (rc)
9610		return rc;
9611
9612	sas_port_add_phy(hpsa_sas_port->port, hpsa_sas_phy->phy);
9613	list_add_tail(&hpsa_sas_phy->phy_list_entry,
9614			&hpsa_sas_port->phy_list_head);
9615	hpsa_sas_phy->added_to_port = true;
9616
9617	return 0;
9618}
9619
9620static int
9621	hpsa_sas_port_add_rphy(struct hpsa_sas_port *hpsa_sas_port,
9622				struct sas_rphy *rphy)
9623{
9624	struct sas_identify *identify;
9625
9626	identify = &rphy->identify;
9627	identify->sas_address = hpsa_sas_port->sas_address;
9628	identify->initiator_port_protocols = SAS_PROTOCOL_STP;
9629	identify->target_port_protocols = SAS_PROTOCOL_STP;
9630
9631	return sas_rphy_add(rphy);
9632}
9633
9634static struct hpsa_sas_port
9635	*hpsa_alloc_sas_port(struct hpsa_sas_node *hpsa_sas_node,
9636				u64 sas_address)
9637{
9638	int rc;
9639	struct hpsa_sas_port *hpsa_sas_port;
9640	struct sas_port *port;
9641
9642	hpsa_sas_port = kzalloc(sizeof(*hpsa_sas_port), GFP_KERNEL);
9643	if (!hpsa_sas_port)
9644		return NULL;
9645
9646	INIT_LIST_HEAD(&hpsa_sas_port->phy_list_head);
9647	hpsa_sas_port->parent_node = hpsa_sas_node;
9648
9649	port = sas_port_alloc_num(hpsa_sas_node->parent_dev);
9650	if (!port)
9651		goto free_hpsa_port;
9652
9653	rc = sas_port_add(port);
9654	if (rc)
9655		goto free_sas_port;
9656
9657	hpsa_sas_port->port = port;
9658	hpsa_sas_port->sas_address = sas_address;
9659	list_add_tail(&hpsa_sas_port->port_list_entry,
9660			&hpsa_sas_node->port_list_head);
9661
9662	return hpsa_sas_port;
9663
9664free_sas_port:
9665	sas_port_free(port);
9666free_hpsa_port:
9667	kfree(hpsa_sas_port);
9668
9669	return NULL;
9670}
9671
9672static void hpsa_free_sas_port(struct hpsa_sas_port *hpsa_sas_port)
9673{
9674	struct hpsa_sas_phy *hpsa_sas_phy;
9675	struct hpsa_sas_phy *next;
9676
9677	list_for_each_entry_safe(hpsa_sas_phy, next,
9678			&hpsa_sas_port->phy_list_head, phy_list_entry)
9679		hpsa_free_sas_phy(hpsa_sas_phy);
9680
9681	sas_port_delete(hpsa_sas_port->port);
9682	list_del(&hpsa_sas_port->port_list_entry);
9683	kfree(hpsa_sas_port);
9684}
9685
9686static struct hpsa_sas_node *hpsa_alloc_sas_node(struct device *parent_dev)
9687{
9688	struct hpsa_sas_node *hpsa_sas_node;
9689
9690	hpsa_sas_node = kzalloc(sizeof(*hpsa_sas_node), GFP_KERNEL);
9691	if (hpsa_sas_node) {
9692		hpsa_sas_node->parent_dev = parent_dev;
9693		INIT_LIST_HEAD(&hpsa_sas_node->port_list_head);
9694	}
9695
9696	return hpsa_sas_node;
9697}
9698
9699static void hpsa_free_sas_node(struct hpsa_sas_node *hpsa_sas_node)
9700{
9701	struct hpsa_sas_port *hpsa_sas_port;
9702	struct hpsa_sas_port *next;
9703
9704	if (!hpsa_sas_node)
9705		return;
9706
9707	list_for_each_entry_safe(hpsa_sas_port, next,
9708			&hpsa_sas_node->port_list_head, port_list_entry)
9709		hpsa_free_sas_port(hpsa_sas_port);
9710
9711	kfree(hpsa_sas_node);
9712}
9713
9714static struct hpsa_scsi_dev_t
9715	*hpsa_find_device_by_sas_rphy(struct ctlr_info *h,
9716					struct sas_rphy *rphy)
9717{
9718	int i;
9719	struct hpsa_scsi_dev_t *device;
9720
9721	for (i = 0; i < h->ndevices; i++) {
9722		device = h->dev[i];
9723		if (!device->sas_port)
9724			continue;
9725		if (device->sas_port->rphy == rphy)
9726			return device;
9727	}
9728
9729	return NULL;
9730}
9731
9732static int hpsa_add_sas_host(struct ctlr_info *h)
9733{
9734	int rc;
9735	struct device *parent_dev;
9736	struct hpsa_sas_node *hpsa_sas_node;
9737	struct hpsa_sas_port *hpsa_sas_port;
9738	struct hpsa_sas_phy *hpsa_sas_phy;
9739
9740	parent_dev = &h->scsi_host->shost_dev;
9741
9742	hpsa_sas_node = hpsa_alloc_sas_node(parent_dev);
9743	if (!hpsa_sas_node)
9744		return -ENOMEM;
9745
9746	hpsa_sas_port = hpsa_alloc_sas_port(hpsa_sas_node, h->sas_address);
9747	if (!hpsa_sas_port) {
9748		rc = -ENODEV;
9749		goto free_sas_node;
9750	}
9751
9752	hpsa_sas_phy = hpsa_alloc_sas_phy(hpsa_sas_port);
9753	if (!hpsa_sas_phy) {
9754		rc = -ENODEV;
9755		goto free_sas_port;
9756	}
9757
9758	rc = hpsa_sas_port_add_phy(hpsa_sas_phy);
9759	if (rc)
9760		goto free_sas_phy;
9761
9762	h->sas_host = hpsa_sas_node;
9763
9764	return 0;
9765
9766free_sas_phy:
9767	sas_phy_free(hpsa_sas_phy->phy);
9768	kfree(hpsa_sas_phy);
9769free_sas_port:
9770	hpsa_free_sas_port(hpsa_sas_port);
9771free_sas_node:
9772	hpsa_free_sas_node(hpsa_sas_node);
9773
9774	return rc;
9775}
9776
9777static void hpsa_delete_sas_host(struct ctlr_info *h)
9778{
9779	hpsa_free_sas_node(h->sas_host);
9780}
9781
9782static int hpsa_add_sas_device(struct hpsa_sas_node *hpsa_sas_node,
9783				struct hpsa_scsi_dev_t *device)
9784{
9785	int rc;
9786	struct hpsa_sas_port *hpsa_sas_port;
9787	struct sas_rphy *rphy;
9788
9789	hpsa_sas_port = hpsa_alloc_sas_port(hpsa_sas_node, device->sas_address);
9790	if (!hpsa_sas_port)
9791		return -ENOMEM;
9792
9793	rphy = sas_end_device_alloc(hpsa_sas_port->port);
9794	if (!rphy) {
9795		rc = -ENODEV;
9796		goto free_sas_port;
9797	}
9798
9799	hpsa_sas_port->rphy = rphy;
9800	device->sas_port = hpsa_sas_port;
9801
9802	rc = hpsa_sas_port_add_rphy(hpsa_sas_port, rphy);
9803	if (rc)
9804		goto free_sas_rphy;
9805
9806	return 0;
9807
9808free_sas_rphy:
9809	sas_rphy_free(rphy);
9810free_sas_port:
9811	hpsa_free_sas_port(hpsa_sas_port);
9812	device->sas_port = NULL;
9813
9814	return rc;
9815}
9816
9817static void hpsa_remove_sas_device(struct hpsa_scsi_dev_t *device)
9818{
9819	if (device->sas_port) {
9820		hpsa_free_sas_port(device->sas_port);
9821		device->sas_port = NULL;
9822	}
9823}
9824
9825static int
9826hpsa_sas_get_linkerrors(struct sas_phy *phy)
9827{
9828	return 0;
9829}
9830
9831static int
9832hpsa_sas_get_enclosure_identifier(struct sas_rphy *rphy, u64 *identifier)
9833{
9834	struct Scsi_Host *shost = phy_to_shost(rphy);
9835	struct ctlr_info *h;
9836	struct hpsa_scsi_dev_t *sd;
9837
9838	if (!shost)
9839		return -ENXIO;
9840
9841	h = shost_to_hba(shost);
9842
9843	if (!h)
9844		return -ENXIO;
9845
9846	sd = hpsa_find_device_by_sas_rphy(h, rphy);
9847	if (!sd)
9848		return -ENXIO;
9849
9850	*identifier = sd->eli;
9851
9852	return 0;
9853}
9854
9855static int
9856hpsa_sas_get_bay_identifier(struct sas_rphy *rphy)
9857{
9858	return -ENXIO;
9859}
9860
9861static int
9862hpsa_sas_phy_reset(struct sas_phy *phy, int hard_reset)
9863{
9864	return 0;
9865}
9866
9867static int
9868hpsa_sas_phy_enable(struct sas_phy *phy, int enable)
9869{
9870	return 0;
9871}
9872
9873static int
9874hpsa_sas_phy_setup(struct sas_phy *phy)
9875{
9876	return 0;
9877}
9878
9879static void
9880hpsa_sas_phy_release(struct sas_phy *phy)
9881{
9882}
9883
9884static int
9885hpsa_sas_phy_speed(struct sas_phy *phy, struct sas_phy_linkrates *rates)
9886{
9887	return -EINVAL;
9888}
9889
9890static struct sas_function_template hpsa_sas_transport_functions = {
9891	.get_linkerrors = hpsa_sas_get_linkerrors,
9892	.get_enclosure_identifier = hpsa_sas_get_enclosure_identifier,
9893	.get_bay_identifier = hpsa_sas_get_bay_identifier,
9894	.phy_reset = hpsa_sas_phy_reset,
9895	.phy_enable = hpsa_sas_phy_enable,
9896	.phy_setup = hpsa_sas_phy_setup,
9897	.phy_release = hpsa_sas_phy_release,
9898	.set_phy_speed = hpsa_sas_phy_speed,
9899};
9900
9901/*
9902 *  This is it.  Register the PCI driver information for the cards we control
9903 *  the OS will call our registered routines when it finds one of our cards.
9904 */
9905static int __init hpsa_init(void)
9906{
9907	int rc;
9908
9909	hpsa_sas_transport_template =
9910		sas_attach_transport(&hpsa_sas_transport_functions);
9911	if (!hpsa_sas_transport_template)
9912		return -ENODEV;
9913
9914	rc = pci_register_driver(&hpsa_pci_driver);
9915
9916	if (rc)
9917		sas_release_transport(hpsa_sas_transport_template);
9918
9919	return rc;
9920}
9921
9922static void __exit hpsa_cleanup(void)
9923{
9924	pci_unregister_driver(&hpsa_pci_driver);
9925	sas_release_transport(hpsa_sas_transport_template);
9926}
9927
9928static void __attribute__((unused)) verify_offsets(void)
9929{
9930#define VERIFY_OFFSET(member, offset) \
9931	BUILD_BUG_ON(offsetof(struct raid_map_data, member) != offset)
9932
9933	VERIFY_OFFSET(structure_size, 0);
9934	VERIFY_OFFSET(volume_blk_size, 4);
9935	VERIFY_OFFSET(volume_blk_cnt, 8);
9936	VERIFY_OFFSET(phys_blk_shift, 16);
9937	VERIFY_OFFSET(parity_rotation_shift, 17);
9938	VERIFY_OFFSET(strip_size, 18);
9939	VERIFY_OFFSET(disk_starting_blk, 20);
9940	VERIFY_OFFSET(disk_blk_cnt, 28);
9941	VERIFY_OFFSET(data_disks_per_row, 36);
9942	VERIFY_OFFSET(metadata_disks_per_row, 38);
9943	VERIFY_OFFSET(row_cnt, 40);
9944	VERIFY_OFFSET(layout_map_count, 42);
9945	VERIFY_OFFSET(flags, 44);
9946	VERIFY_OFFSET(dekindex, 46);
9947	/* VERIFY_OFFSET(reserved, 48 */
9948	VERIFY_OFFSET(data, 64);
9949
9950#undef VERIFY_OFFSET
9951
9952#define VERIFY_OFFSET(member, offset) \
9953	BUILD_BUG_ON(offsetof(struct io_accel2_cmd, member) != offset)
9954
9955	VERIFY_OFFSET(IU_type, 0);
9956	VERIFY_OFFSET(direction, 1);
9957	VERIFY_OFFSET(reply_queue, 2);
9958	/* VERIFY_OFFSET(reserved1, 3);  */
9959	VERIFY_OFFSET(scsi_nexus, 4);
9960	VERIFY_OFFSET(Tag, 8);
9961	VERIFY_OFFSET(cdb, 16);
9962	VERIFY_OFFSET(cciss_lun, 32);
9963	VERIFY_OFFSET(data_len, 40);
9964	VERIFY_OFFSET(cmd_priority_task_attr, 44);
9965	VERIFY_OFFSET(sg_count, 45);
9966	/* VERIFY_OFFSET(reserved3 */
9967	VERIFY_OFFSET(err_ptr, 48);
9968	VERIFY_OFFSET(err_len, 56);
9969	/* VERIFY_OFFSET(reserved4  */
9970	VERIFY_OFFSET(sg, 64);
9971
9972#undef VERIFY_OFFSET
9973
9974#define VERIFY_OFFSET(member, offset) \
9975	BUILD_BUG_ON(offsetof(struct io_accel1_cmd, member) != offset)
9976
9977	VERIFY_OFFSET(dev_handle, 0x00);
9978	VERIFY_OFFSET(reserved1, 0x02);
9979	VERIFY_OFFSET(function, 0x03);
9980	VERIFY_OFFSET(reserved2, 0x04);
9981	VERIFY_OFFSET(err_info, 0x0C);
9982	VERIFY_OFFSET(reserved3, 0x10);
9983	VERIFY_OFFSET(err_info_len, 0x12);
9984	VERIFY_OFFSET(reserved4, 0x13);
9985	VERIFY_OFFSET(sgl_offset, 0x14);
9986	VERIFY_OFFSET(reserved5, 0x15);
9987	VERIFY_OFFSET(transfer_len, 0x1C);
9988	VERIFY_OFFSET(reserved6, 0x20);
9989	VERIFY_OFFSET(io_flags, 0x24);
9990	VERIFY_OFFSET(reserved7, 0x26);
9991	VERIFY_OFFSET(LUN, 0x34);
9992	VERIFY_OFFSET(control, 0x3C);
9993	VERIFY_OFFSET(CDB, 0x40);
9994	VERIFY_OFFSET(reserved8, 0x50);
9995	VERIFY_OFFSET(host_context_flags, 0x60);
9996	VERIFY_OFFSET(timeout_sec, 0x62);
9997	VERIFY_OFFSET(ReplyQueue, 0x64);
9998	VERIFY_OFFSET(reserved9, 0x65);
9999	VERIFY_OFFSET(tag, 0x68);
10000	VERIFY_OFFSET(host_addr, 0x70);
10001	VERIFY_OFFSET(CISS_LUN, 0x78);
10002	VERIFY_OFFSET(SG, 0x78 + 8);
10003#undef VERIFY_OFFSET
10004}
10005
10006module_init(hpsa_init);
10007module_exit(hpsa_cleanup);
10008