1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Driver for sTec s1120 PCIe SSDs. sTec was acquired in 2013 by HGST and HGST
4 * was acquired by Western Digital in 2012.
5 *
6 * Copyright 2012 sTec, Inc.
7 * Copyright (c) 2017 Western Digital Corporation or its affiliates.
8 */
9
10#include <linux/kernel.h>
11#include <linux/module.h>
12#include <linux/init.h>
13#include <linux/pci.h>
14#include <linux/slab.h>
15#include <linux/spinlock.h>
16#include <linux/blkdev.h>
17#include <linux/blk-mq.h>
18#include <linux/sched.h>
19#include <linux/interrupt.h>
20#include <linux/compiler.h>
21#include <linux/workqueue.h>
22#include <linux/delay.h>
23#include <linux/time.h>
24#include <linux/hdreg.h>
25#include <linux/dma-mapping.h>
26#include <linux/completion.h>
27#include <linux/scatterlist.h>
28#include <linux/err.h>
29#include <linux/aer.h>
30#include <linux/wait.h>
31#include <linux/stringify.h>
32#include <scsi/scsi.h>
33#include <scsi/sg.h>
34#include <linux/io.h>
35#include <linux/uaccess.h>
36#include <asm/unaligned.h>
37
38#include "skd_s1120.h"
39
40static int skd_dbg_level;
41static int skd_isr_comp_limit = 4;
42
43#define SKD_ASSERT(expr) \
44	do { \
45		if (unlikely(!(expr))) { \
46			pr_err("Assertion failed! %s,%s,%s,line=%d\n",	\
47			       # expr, __FILE__, __func__, __LINE__); \
48		} \
49	} while (0)
50
51#define DRV_NAME "skd"
52#define PFX DRV_NAME ": "
53
54MODULE_LICENSE("GPL");
55
56MODULE_DESCRIPTION("STEC s1120 PCIe SSD block driver");
57
58#define PCI_VENDOR_ID_STEC      0x1B39
59#define PCI_DEVICE_ID_S1120     0x0001
60
61#define SKD_FUA_NV		(1 << 1)
62#define SKD_MINORS_PER_DEVICE   16
63
64#define SKD_MAX_QUEUE_DEPTH     200u
65
66#define SKD_PAUSE_TIMEOUT       (5 * 1000)
67
68#define SKD_N_FITMSG_BYTES      (512u)
69#define SKD_MAX_REQ_PER_MSG	14
70
71#define SKD_N_SPECIAL_FITMSG_BYTES      (128u)
72
73/* SG elements are 32 bytes, so we can make this 4096 and still be under the
74 * 128KB limit.  That allows 4096*4K = 16M xfer size
75 */
76#define SKD_N_SG_PER_REQ_DEFAULT 256u
77
78#define SKD_N_COMPLETION_ENTRY  256u
79#define SKD_N_READ_CAP_BYTES    (8u)
80
81#define SKD_N_INTERNAL_BYTES    (512u)
82
83#define SKD_SKCOMP_SIZE							\
84	((sizeof(struct fit_completion_entry_v1) +			\
85	  sizeof(struct fit_comp_error_info)) * SKD_N_COMPLETION_ENTRY)
86
87/* 5 bits of uniqifier, 0xF800 */
88#define SKD_ID_TABLE_MASK       (3u << 8u)
89#define  SKD_ID_RW_REQUEST      (0u << 8u)
90#define  SKD_ID_INTERNAL        (1u << 8u)
91#define  SKD_ID_FIT_MSG         (3u << 8u)
92#define SKD_ID_SLOT_MASK        0x00FFu
93#define SKD_ID_SLOT_AND_TABLE_MASK 0x03FFu
94
95#define SKD_N_MAX_SECTORS 2048u
96
97#define SKD_MAX_RETRIES 2u
98
99#define SKD_TIMER_SECONDS(seconds) (seconds)
100#define SKD_TIMER_MINUTES(minutes) ((minutes) * (60))
101
102#define INQ_STD_NBYTES 36
103
104enum skd_drvr_state {
105	SKD_DRVR_STATE_LOAD,
106	SKD_DRVR_STATE_IDLE,
107	SKD_DRVR_STATE_BUSY,
108	SKD_DRVR_STATE_STARTING,
109	SKD_DRVR_STATE_ONLINE,
110	SKD_DRVR_STATE_PAUSING,
111	SKD_DRVR_STATE_PAUSED,
112	SKD_DRVR_STATE_RESTARTING,
113	SKD_DRVR_STATE_RESUMING,
114	SKD_DRVR_STATE_STOPPING,
115	SKD_DRVR_STATE_FAULT,
116	SKD_DRVR_STATE_DISAPPEARED,
117	SKD_DRVR_STATE_PROTOCOL_MISMATCH,
118	SKD_DRVR_STATE_BUSY_ERASE,
119	SKD_DRVR_STATE_BUSY_SANITIZE,
120	SKD_DRVR_STATE_BUSY_IMMINENT,
121	SKD_DRVR_STATE_WAIT_BOOT,
122	SKD_DRVR_STATE_SYNCING,
123};
124
125#define SKD_WAIT_BOOT_TIMO      SKD_TIMER_SECONDS(90u)
126#define SKD_STARTING_TIMO       SKD_TIMER_SECONDS(8u)
127#define SKD_RESTARTING_TIMO     SKD_TIMER_MINUTES(4u)
128#define SKD_BUSY_TIMO           SKD_TIMER_MINUTES(20u)
129#define SKD_STARTED_BUSY_TIMO   SKD_TIMER_SECONDS(60u)
130#define SKD_START_WAIT_SECONDS  90u
131
132enum skd_req_state {
133	SKD_REQ_STATE_IDLE,
134	SKD_REQ_STATE_SETUP,
135	SKD_REQ_STATE_BUSY,
136	SKD_REQ_STATE_COMPLETED,
137	SKD_REQ_STATE_TIMEOUT,
138};
139
140enum skd_check_status_action {
141	SKD_CHECK_STATUS_REPORT_GOOD,
142	SKD_CHECK_STATUS_REPORT_SMART_ALERT,
143	SKD_CHECK_STATUS_REQUEUE_REQUEST,
144	SKD_CHECK_STATUS_REPORT_ERROR,
145	SKD_CHECK_STATUS_BUSY_IMMINENT,
146};
147
148struct skd_msg_buf {
149	struct fit_msg_hdr	fmh;
150	struct skd_scsi_request	scsi[SKD_MAX_REQ_PER_MSG];
151};
152
153struct skd_fitmsg_context {
154	u32 id;
155
156	u32 length;
157
158	struct skd_msg_buf *msg_buf;
159	dma_addr_t mb_dma_address;
160};
161
162struct skd_request_context {
163	enum skd_req_state state;
164
165	u16 id;
166	u32 fitmsg_id;
167
168	u8 flush_cmd;
169
170	enum dma_data_direction data_dir;
171	struct scatterlist *sg;
172	u32 n_sg;
173	u32 sg_byte_count;
174
175	struct fit_sg_descriptor *sksg_list;
176	dma_addr_t sksg_dma_address;
177
178	struct fit_completion_entry_v1 completion;
179
180	struct fit_comp_error_info err_info;
181	int retries;
182
183	blk_status_t status;
184};
185
186struct skd_special_context {
187	struct skd_request_context req;
188
189	void *data_buf;
190	dma_addr_t db_dma_address;
191
192	struct skd_msg_buf *msg_buf;
193	dma_addr_t mb_dma_address;
194};
195
196typedef enum skd_irq_type {
197	SKD_IRQ_LEGACY,
198	SKD_IRQ_MSI,
199	SKD_IRQ_MSIX
200} skd_irq_type_t;
201
202#define SKD_MAX_BARS                    2
203
204struct skd_device {
205	void __iomem *mem_map[SKD_MAX_BARS];
206	resource_size_t mem_phys[SKD_MAX_BARS];
207	u32 mem_size[SKD_MAX_BARS];
208
209	struct skd_msix_entry *msix_entries;
210
211	struct pci_dev *pdev;
212	int pcie_error_reporting_is_enabled;
213
214	spinlock_t lock;
215	struct gendisk *disk;
216	struct blk_mq_tag_set tag_set;
217	struct request_queue *queue;
218	struct skd_fitmsg_context *skmsg;
219	struct device *class_dev;
220	int gendisk_on;
221	int sync_done;
222
223	u32 devno;
224	u32 major;
225	char isr_name[30];
226
227	enum skd_drvr_state state;
228	u32 drive_state;
229
230	u32 cur_max_queue_depth;
231	u32 queue_low_water_mark;
232	u32 dev_max_queue_depth;
233
234	u32 num_fitmsg_context;
235	u32 num_req_context;
236
237	struct skd_fitmsg_context *skmsg_table;
238
239	struct skd_special_context internal_skspcl;
240	u32 read_cap_blocksize;
241	u32 read_cap_last_lba;
242	int read_cap_is_valid;
243	int inquiry_is_valid;
244	u8 inq_serial_num[13];  /*12 chars plus null term */
245
246	u8 skcomp_cycle;
247	u32 skcomp_ix;
248	struct kmem_cache *msgbuf_cache;
249	struct kmem_cache *sglist_cache;
250	struct kmem_cache *databuf_cache;
251	struct fit_completion_entry_v1 *skcomp_table;
252	struct fit_comp_error_info *skerr_table;
253	dma_addr_t cq_dma_address;
254
255	wait_queue_head_t waitq;
256
257	struct timer_list timer;
258	u32 timer_countdown;
259	u32 timer_substate;
260
261	int sgs_per_request;
262	u32 last_mtd;
263
264	u32 proto_ver;
265
266	int dbg_level;
267	u32 connect_time_stamp;
268	int connect_retries;
269#define SKD_MAX_CONNECT_RETRIES 16
270	u32 drive_jiffies;
271
272	u32 timo_slot;
273
274	struct work_struct start_queue;
275	struct work_struct completion_worker;
276};
277
278#define SKD_WRITEL(DEV, VAL, OFF) skd_reg_write32(DEV, VAL, OFF)
279#define SKD_READL(DEV, OFF)      skd_reg_read32(DEV, OFF)
280#define SKD_WRITEQ(DEV, VAL, OFF) skd_reg_write64(DEV, VAL, OFF)
281
282static inline u32 skd_reg_read32(struct skd_device *skdev, u32 offset)
283{
284	u32 val = readl(skdev->mem_map[1] + offset);
285
286	if (unlikely(skdev->dbg_level >= 2))
287		dev_dbg(&skdev->pdev->dev, "offset %x = %x\n", offset, val);
288	return val;
289}
290
291static inline void skd_reg_write32(struct skd_device *skdev, u32 val,
292				   u32 offset)
293{
294	writel(val, skdev->mem_map[1] + offset);
295	if (unlikely(skdev->dbg_level >= 2))
296		dev_dbg(&skdev->pdev->dev, "offset %x = %x\n", offset, val);
297}
298
299static inline void skd_reg_write64(struct skd_device *skdev, u64 val,
300				   u32 offset)
301{
302	writeq(val, skdev->mem_map[1] + offset);
303	if (unlikely(skdev->dbg_level >= 2))
304		dev_dbg(&skdev->pdev->dev, "offset %x = %016llx\n", offset,
305			val);
306}
307
308
309#define SKD_IRQ_DEFAULT SKD_IRQ_MSIX
310static int skd_isr_type = SKD_IRQ_DEFAULT;
311
312module_param(skd_isr_type, int, 0444);
313MODULE_PARM_DESC(skd_isr_type, "Interrupt type capability."
314		 " (0==legacy, 1==MSI, 2==MSI-X, default==1)");
315
316#define SKD_MAX_REQ_PER_MSG_DEFAULT 1
317static int skd_max_req_per_msg = SKD_MAX_REQ_PER_MSG_DEFAULT;
318
319module_param(skd_max_req_per_msg, int, 0444);
320MODULE_PARM_DESC(skd_max_req_per_msg,
321		 "Maximum SCSI requests packed in a single message."
322		 " (1-" __stringify(SKD_MAX_REQ_PER_MSG) ", default==1)");
323
324#define SKD_MAX_QUEUE_DEPTH_DEFAULT 64
325#define SKD_MAX_QUEUE_DEPTH_DEFAULT_STR "64"
326static int skd_max_queue_depth = SKD_MAX_QUEUE_DEPTH_DEFAULT;
327
328module_param(skd_max_queue_depth, int, 0444);
329MODULE_PARM_DESC(skd_max_queue_depth,
330		 "Maximum SCSI requests issued to s1120."
331		 " (1-200, default==" SKD_MAX_QUEUE_DEPTH_DEFAULT_STR ")");
332
333static int skd_sgs_per_request = SKD_N_SG_PER_REQ_DEFAULT;
334module_param(skd_sgs_per_request, int, 0444);
335MODULE_PARM_DESC(skd_sgs_per_request,
336		 "Maximum SG elements per block request."
337		 " (1-4096, default==256)");
338
339static int skd_max_pass_thru = 1;
340module_param(skd_max_pass_thru, int, 0444);
341MODULE_PARM_DESC(skd_max_pass_thru,
342		 "Maximum SCSI pass-thru at a time. IGNORED");
343
344module_param(skd_dbg_level, int, 0444);
345MODULE_PARM_DESC(skd_dbg_level, "s1120 debug level (0,1,2)");
346
347module_param(skd_isr_comp_limit, int, 0444);
348MODULE_PARM_DESC(skd_isr_comp_limit, "s1120 isr comp limit (0=none) default=4");
349
350/* Major device number dynamically assigned. */
351static u32 skd_major;
352
353static void skd_destruct(struct skd_device *skdev);
354static const struct block_device_operations skd_blockdev_ops;
355static void skd_send_fitmsg(struct skd_device *skdev,
356			    struct skd_fitmsg_context *skmsg);
357static void skd_send_special_fitmsg(struct skd_device *skdev,
358				    struct skd_special_context *skspcl);
359static bool skd_preop_sg_list(struct skd_device *skdev,
360			     struct skd_request_context *skreq);
361static void skd_postop_sg_list(struct skd_device *skdev,
362			       struct skd_request_context *skreq);
363
364static void skd_restart_device(struct skd_device *skdev);
365static int skd_quiesce_dev(struct skd_device *skdev);
366static int skd_unquiesce_dev(struct skd_device *skdev);
367static void skd_disable_interrupts(struct skd_device *skdev);
368static void skd_isr_fwstate(struct skd_device *skdev);
369static void skd_recover_requests(struct skd_device *skdev);
370static void skd_soft_reset(struct skd_device *skdev);
371
372const char *skd_drive_state_to_str(int state);
373const char *skd_skdev_state_to_str(enum skd_drvr_state state);
374static void skd_log_skdev(struct skd_device *skdev, const char *event);
375static void skd_log_skreq(struct skd_device *skdev,
376			  struct skd_request_context *skreq, const char *event);
377
378/*
379 *****************************************************************************
380 * READ/WRITE REQUESTS
381 *****************************************************************************
382 */
383static bool skd_inc_in_flight(struct request *rq, void *data, bool reserved)
384{
385	int *count = data;
386
387	count++;
388	return true;
389}
390
391static int skd_in_flight(struct skd_device *skdev)
392{
393	int count = 0;
394
395	blk_mq_tagset_busy_iter(&skdev->tag_set, skd_inc_in_flight, &count);
396
397	return count;
398}
399
400static void
401skd_prep_rw_cdb(struct skd_scsi_request *scsi_req,
402		int data_dir, unsigned lba,
403		unsigned count)
404{
405	if (data_dir == READ)
406		scsi_req->cdb[0] = READ_10;
407	else
408		scsi_req->cdb[0] = WRITE_10;
409
410	scsi_req->cdb[1] = 0;
411	scsi_req->cdb[2] = (lba & 0xff000000) >> 24;
412	scsi_req->cdb[3] = (lba & 0xff0000) >> 16;
413	scsi_req->cdb[4] = (lba & 0xff00) >> 8;
414	scsi_req->cdb[5] = (lba & 0xff);
415	scsi_req->cdb[6] = 0;
416	scsi_req->cdb[7] = (count & 0xff00) >> 8;
417	scsi_req->cdb[8] = count & 0xff;
418	scsi_req->cdb[9] = 0;
419}
420
421static void
422skd_prep_zerosize_flush_cdb(struct skd_scsi_request *scsi_req,
423			    struct skd_request_context *skreq)
424{
425	skreq->flush_cmd = 1;
426
427	scsi_req->cdb[0] = SYNCHRONIZE_CACHE;
428	scsi_req->cdb[1] = 0;
429	scsi_req->cdb[2] = 0;
430	scsi_req->cdb[3] = 0;
431	scsi_req->cdb[4] = 0;
432	scsi_req->cdb[5] = 0;
433	scsi_req->cdb[6] = 0;
434	scsi_req->cdb[7] = 0;
435	scsi_req->cdb[8] = 0;
436	scsi_req->cdb[9] = 0;
437}
438
439/*
440 * Return true if and only if all pending requests should be failed.
441 */
442static bool skd_fail_all(struct request_queue *q)
443{
444	struct skd_device *skdev = q->queuedata;
445
446	SKD_ASSERT(skdev->state != SKD_DRVR_STATE_ONLINE);
447
448	skd_log_skdev(skdev, "req_not_online");
449	switch (skdev->state) {
450	case SKD_DRVR_STATE_PAUSING:
451	case SKD_DRVR_STATE_PAUSED:
452	case SKD_DRVR_STATE_STARTING:
453	case SKD_DRVR_STATE_RESTARTING:
454	case SKD_DRVR_STATE_WAIT_BOOT:
455	/* In case of starting, we haven't started the queue,
456	 * so we can't get here... but requests are
457	 * possibly hanging out waiting for us because we
458	 * reported the dev/skd0 already.  They'll wait
459	 * forever if connect doesn't complete.
460	 * What to do??? delay dev/skd0 ??
461	 */
462	case SKD_DRVR_STATE_BUSY:
463	case SKD_DRVR_STATE_BUSY_IMMINENT:
464	case SKD_DRVR_STATE_BUSY_ERASE:
465		return false;
466
467	case SKD_DRVR_STATE_BUSY_SANITIZE:
468	case SKD_DRVR_STATE_STOPPING:
469	case SKD_DRVR_STATE_SYNCING:
470	case SKD_DRVR_STATE_FAULT:
471	case SKD_DRVR_STATE_DISAPPEARED:
472	default:
473		return true;
474	}
475}
476
477static blk_status_t skd_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
478				    const struct blk_mq_queue_data *mqd)
479{
480	struct request *const req = mqd->rq;
481	struct request_queue *const q = req->q;
482	struct skd_device *skdev = q->queuedata;
483	struct skd_fitmsg_context *skmsg;
484	struct fit_msg_hdr *fmh;
485	const u32 tag = blk_mq_unique_tag(req);
486	struct skd_request_context *const skreq = blk_mq_rq_to_pdu(req);
487	struct skd_scsi_request *scsi_req;
488	unsigned long flags = 0;
489	const u32 lba = blk_rq_pos(req);
490	const u32 count = blk_rq_sectors(req);
491	const int data_dir = rq_data_dir(req);
492
493	if (unlikely(skdev->state != SKD_DRVR_STATE_ONLINE))
494		return skd_fail_all(q) ? BLK_STS_IOERR : BLK_STS_RESOURCE;
495
496	if (!(req->rq_flags & RQF_DONTPREP)) {
497		skreq->retries = 0;
498		req->rq_flags |= RQF_DONTPREP;
499	}
500
501	blk_mq_start_request(req);
502
503	WARN_ONCE(tag >= skd_max_queue_depth, "%#x > %#x (nr_requests = %lu)\n",
504		  tag, skd_max_queue_depth, q->nr_requests);
505
506	SKD_ASSERT(skreq->state == SKD_REQ_STATE_IDLE);
507
508	dev_dbg(&skdev->pdev->dev,
509		"new req=%p lba=%u(0x%x) count=%u(0x%x) dir=%d\n", req, lba,
510		lba, count, count, data_dir);
511
512	skreq->id = tag + SKD_ID_RW_REQUEST;
513	skreq->flush_cmd = 0;
514	skreq->n_sg = 0;
515	skreq->sg_byte_count = 0;
516
517	skreq->fitmsg_id = 0;
518
519	skreq->data_dir = data_dir == READ ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
520
521	if (req->bio && !skd_preop_sg_list(skdev, skreq)) {
522		dev_dbg(&skdev->pdev->dev, "error Out\n");
523		skreq->status = BLK_STS_RESOURCE;
524		blk_mq_complete_request(req);
525		return BLK_STS_OK;
526	}
527
528	dma_sync_single_for_device(&skdev->pdev->dev, skreq->sksg_dma_address,
529				   skreq->n_sg *
530				   sizeof(struct fit_sg_descriptor),
531				   DMA_TO_DEVICE);
532
533	/* Either a FIT msg is in progress or we have to start one. */
534	if (skd_max_req_per_msg == 1) {
535		skmsg = NULL;
536	} else {
537		spin_lock_irqsave(&skdev->lock, flags);
538		skmsg = skdev->skmsg;
539	}
540	if (!skmsg) {
541		skmsg = &skdev->skmsg_table[tag];
542		skdev->skmsg = skmsg;
543
544		/* Initialize the FIT msg header */
545		fmh = &skmsg->msg_buf->fmh;
546		memset(fmh, 0, sizeof(*fmh));
547		fmh->protocol_id = FIT_PROTOCOL_ID_SOFIT;
548		skmsg->length = sizeof(*fmh);
549	} else {
550		fmh = &skmsg->msg_buf->fmh;
551	}
552
553	skreq->fitmsg_id = skmsg->id;
554
555	scsi_req = &skmsg->msg_buf->scsi[fmh->num_protocol_cmds_coalesced];
556	memset(scsi_req, 0, sizeof(*scsi_req));
557
558	scsi_req->hdr.tag = skreq->id;
559	scsi_req->hdr.sg_list_dma_address =
560		cpu_to_be64(skreq->sksg_dma_address);
561
562	if (req_op(req) == REQ_OP_FLUSH) {
563		skd_prep_zerosize_flush_cdb(scsi_req, skreq);
564		SKD_ASSERT(skreq->flush_cmd == 1);
565	} else {
566		skd_prep_rw_cdb(scsi_req, data_dir, lba, count);
567	}
568
569	if (req->cmd_flags & REQ_FUA)
570		scsi_req->cdb[1] |= SKD_FUA_NV;
571
572	scsi_req->hdr.sg_list_len_bytes = cpu_to_be32(skreq->sg_byte_count);
573
574	/* Complete resource allocations. */
575	skreq->state = SKD_REQ_STATE_BUSY;
576
577	skmsg->length += sizeof(struct skd_scsi_request);
578	fmh->num_protocol_cmds_coalesced++;
579
580	dev_dbg(&skdev->pdev->dev, "req=0x%x busy=%d\n", skreq->id,
581		skd_in_flight(skdev));
582
583	/*
584	 * If the FIT msg buffer is full send it.
585	 */
586	if (skd_max_req_per_msg == 1) {
587		skd_send_fitmsg(skdev, skmsg);
588	} else {
589		if (mqd->last ||
590		    fmh->num_protocol_cmds_coalesced >= skd_max_req_per_msg) {
591			skd_send_fitmsg(skdev, skmsg);
592			skdev->skmsg = NULL;
593		}
594		spin_unlock_irqrestore(&skdev->lock, flags);
595	}
596
597	return BLK_STS_OK;
598}
599
600static enum blk_eh_timer_return skd_timed_out(struct request *req,
601					      bool reserved)
602{
603	struct skd_device *skdev = req->q->queuedata;
604
605	dev_err(&skdev->pdev->dev, "request with tag %#x timed out\n",
606		blk_mq_unique_tag(req));
607
608	return BLK_EH_RESET_TIMER;
609}
610
611static void skd_complete_rq(struct request *req)
612{
613	struct skd_request_context *skreq = blk_mq_rq_to_pdu(req);
614
615	blk_mq_end_request(req, skreq->status);
616}
617
618static bool skd_preop_sg_list(struct skd_device *skdev,
619			     struct skd_request_context *skreq)
620{
621	struct request *req = blk_mq_rq_from_pdu(skreq);
622	struct scatterlist *sgl = &skreq->sg[0], *sg;
623	int n_sg;
624	int i;
625
626	skreq->sg_byte_count = 0;
627
628	WARN_ON_ONCE(skreq->data_dir != DMA_TO_DEVICE &&
629		     skreq->data_dir != DMA_FROM_DEVICE);
630
631	n_sg = blk_rq_map_sg(skdev->queue, req, sgl);
632	if (n_sg <= 0)
633		return false;
634
635	/*
636	 * Map scatterlist to PCI bus addresses.
637	 * Note PCI might change the number of entries.
638	 */
639	n_sg = dma_map_sg(&skdev->pdev->dev, sgl, n_sg, skreq->data_dir);
640	if (n_sg <= 0)
641		return false;
642
643	SKD_ASSERT(n_sg <= skdev->sgs_per_request);
644
645	skreq->n_sg = n_sg;
646
647	for_each_sg(sgl, sg, n_sg, i) {
648		struct fit_sg_descriptor *sgd = &skreq->sksg_list[i];
649		u32 cnt = sg_dma_len(sg);
650		uint64_t dma_addr = sg_dma_address(sg);
651
652		sgd->control = FIT_SGD_CONTROL_NOT_LAST;
653		sgd->byte_count = cnt;
654		skreq->sg_byte_count += cnt;
655		sgd->host_side_addr = dma_addr;
656		sgd->dev_side_addr = 0;
657	}
658
659	skreq->sksg_list[n_sg - 1].next_desc_ptr = 0LL;
660	skreq->sksg_list[n_sg - 1].control = FIT_SGD_CONTROL_LAST;
661
662	if (unlikely(skdev->dbg_level > 1)) {
663		dev_dbg(&skdev->pdev->dev,
664			"skreq=%x sksg_list=%p sksg_dma=%pad\n",
665			skreq->id, skreq->sksg_list, &skreq->sksg_dma_address);
666		for (i = 0; i < n_sg; i++) {
667			struct fit_sg_descriptor *sgd = &skreq->sksg_list[i];
668
669			dev_dbg(&skdev->pdev->dev,
670				"  sg[%d] count=%u ctrl=0x%x addr=0x%llx next=0x%llx\n",
671				i, sgd->byte_count, sgd->control,
672				sgd->host_side_addr, sgd->next_desc_ptr);
673		}
674	}
675
676	return true;
677}
678
679static void skd_postop_sg_list(struct skd_device *skdev,
680			       struct skd_request_context *skreq)
681{
682	/*
683	 * restore the next ptr for next IO request so we
684	 * don't have to set it every time.
685	 */
686	skreq->sksg_list[skreq->n_sg - 1].next_desc_ptr =
687		skreq->sksg_dma_address +
688		((skreq->n_sg) * sizeof(struct fit_sg_descriptor));
689	dma_unmap_sg(&skdev->pdev->dev, &skreq->sg[0], skreq->n_sg,
690		     skreq->data_dir);
691}
692
693/*
694 *****************************************************************************
695 * TIMER
696 *****************************************************************************
697 */
698
699static void skd_timer_tick_not_online(struct skd_device *skdev);
700
701static void skd_start_queue(struct work_struct *work)
702{
703	struct skd_device *skdev = container_of(work, typeof(*skdev),
704						start_queue);
705
706	/*
707	 * Although it is safe to call blk_start_queue() from interrupt
708	 * context, blk_mq_start_hw_queues() must not be called from
709	 * interrupt context.
710	 */
711	blk_mq_start_hw_queues(skdev->queue);
712}
713
714static void skd_timer_tick(struct timer_list *t)
715{
716	struct skd_device *skdev = from_timer(skdev, t, timer);
717	unsigned long reqflags;
718	u32 state;
719
720	if (skdev->state == SKD_DRVR_STATE_FAULT)
721		/* The driver has declared fault, and we want it to
722		 * stay that way until driver is reloaded.
723		 */
724		return;
725
726	spin_lock_irqsave(&skdev->lock, reqflags);
727
728	state = SKD_READL(skdev, FIT_STATUS);
729	state &= FIT_SR_DRIVE_STATE_MASK;
730	if (state != skdev->drive_state)
731		skd_isr_fwstate(skdev);
732
733	if (skdev->state != SKD_DRVR_STATE_ONLINE)
734		skd_timer_tick_not_online(skdev);
735
736	mod_timer(&skdev->timer, (jiffies + HZ));
737
738	spin_unlock_irqrestore(&skdev->lock, reqflags);
739}
740
741static void skd_timer_tick_not_online(struct skd_device *skdev)
742{
743	switch (skdev->state) {
744	case SKD_DRVR_STATE_IDLE:
745	case SKD_DRVR_STATE_LOAD:
746		break;
747	case SKD_DRVR_STATE_BUSY_SANITIZE:
748		dev_dbg(&skdev->pdev->dev,
749			"drive busy sanitize[%x], driver[%x]\n",
750			skdev->drive_state, skdev->state);
751		/* If we've been in sanitize for 3 seconds, we figure we're not
752		 * going to get anymore completions, so recover requests now
753		 */
754		if (skdev->timer_countdown > 0) {
755			skdev->timer_countdown--;
756			return;
757		}
758		skd_recover_requests(skdev);
759		break;
760
761	case SKD_DRVR_STATE_BUSY:
762	case SKD_DRVR_STATE_BUSY_IMMINENT:
763	case SKD_DRVR_STATE_BUSY_ERASE:
764		dev_dbg(&skdev->pdev->dev, "busy[%x], countdown=%d\n",
765			skdev->state, skdev->timer_countdown);
766		if (skdev->timer_countdown > 0) {
767			skdev->timer_countdown--;
768			return;
769		}
770		dev_dbg(&skdev->pdev->dev,
771			"busy[%x], timedout=%d, restarting device.",
772			skdev->state, skdev->timer_countdown);
773		skd_restart_device(skdev);
774		break;
775
776	case SKD_DRVR_STATE_WAIT_BOOT:
777	case SKD_DRVR_STATE_STARTING:
778		if (skdev->timer_countdown > 0) {
779			skdev->timer_countdown--;
780			return;
781		}
782		/* For now, we fault the drive.  Could attempt resets to
783		 * revcover at some point. */
784		skdev->state = SKD_DRVR_STATE_FAULT;
785
786		dev_err(&skdev->pdev->dev, "DriveFault Connect Timeout (%x)\n",
787			skdev->drive_state);
788
789		/*start the queue so we can respond with error to requests */
790		/* wakeup anyone waiting for startup complete */
791		schedule_work(&skdev->start_queue);
792		skdev->gendisk_on = -1;
793		wake_up_interruptible(&skdev->waitq);
794		break;
795
796	case SKD_DRVR_STATE_ONLINE:
797		/* shouldn't get here. */
798		break;
799
800	case SKD_DRVR_STATE_PAUSING:
801	case SKD_DRVR_STATE_PAUSED:
802		break;
803
804	case SKD_DRVR_STATE_RESTARTING:
805		if (skdev->timer_countdown > 0) {
806			skdev->timer_countdown--;
807			return;
808		}
809		/* For now, we fault the drive. Could attempt resets to
810		 * revcover at some point. */
811		skdev->state = SKD_DRVR_STATE_FAULT;
812		dev_err(&skdev->pdev->dev,
813			"DriveFault Reconnect Timeout (%x)\n",
814			skdev->drive_state);
815
816		/*
817		 * Recovering does two things:
818		 * 1. completes IO with error
819		 * 2. reclaims dma resources
820		 * When is it safe to recover requests?
821		 * - if the drive state is faulted
822		 * - if the state is still soft reset after out timeout
823		 * - if the drive registers are dead (state = FF)
824		 * If it is "unsafe", we still need to recover, so we will
825		 * disable pci bus mastering and disable our interrupts.
826		 */
827
828		if ((skdev->drive_state == FIT_SR_DRIVE_SOFT_RESET) ||
829		    (skdev->drive_state == FIT_SR_DRIVE_FAULT) ||
830		    (skdev->drive_state == FIT_SR_DRIVE_STATE_MASK))
831			/* It never came out of soft reset. Try to
832			 * recover the requests and then let them
833			 * fail. This is to mitigate hung processes. */
834			skd_recover_requests(skdev);
835		else {
836			dev_err(&skdev->pdev->dev, "Disable BusMaster (%x)\n",
837				skdev->drive_state);
838			pci_disable_device(skdev->pdev);
839			skd_disable_interrupts(skdev);
840			skd_recover_requests(skdev);
841		}
842
843		/*start the queue so we can respond with error to requests */
844		/* wakeup anyone waiting for startup complete */
845		schedule_work(&skdev->start_queue);
846		skdev->gendisk_on = -1;
847		wake_up_interruptible(&skdev->waitq);
848		break;
849
850	case SKD_DRVR_STATE_RESUMING:
851	case SKD_DRVR_STATE_STOPPING:
852	case SKD_DRVR_STATE_SYNCING:
853	case SKD_DRVR_STATE_FAULT:
854	case SKD_DRVR_STATE_DISAPPEARED:
855	default:
856		break;
857	}
858}
859
860static int skd_start_timer(struct skd_device *skdev)
861{
862	int rc;
863
864	timer_setup(&skdev->timer, skd_timer_tick, 0);
865
866	rc = mod_timer(&skdev->timer, (jiffies + HZ));
867	if (rc)
868		dev_err(&skdev->pdev->dev, "failed to start timer %d\n", rc);
869	return rc;
870}
871
872static void skd_kill_timer(struct skd_device *skdev)
873{
874	del_timer_sync(&skdev->timer);
875}
876
877/*
878 *****************************************************************************
879 * INTERNAL REQUESTS -- generated by driver itself
880 *****************************************************************************
881 */
882
883static int skd_format_internal_skspcl(struct skd_device *skdev)
884{
885	struct skd_special_context *skspcl = &skdev->internal_skspcl;
886	struct fit_sg_descriptor *sgd = &skspcl->req.sksg_list[0];
887	struct fit_msg_hdr *fmh;
888	uint64_t dma_address;
889	struct skd_scsi_request *scsi;
890
891	fmh = &skspcl->msg_buf->fmh;
892	fmh->protocol_id = FIT_PROTOCOL_ID_SOFIT;
893	fmh->num_protocol_cmds_coalesced = 1;
894
895	scsi = &skspcl->msg_buf->scsi[0];
896	memset(scsi, 0, sizeof(*scsi));
897	dma_address = skspcl->req.sksg_dma_address;
898	scsi->hdr.sg_list_dma_address = cpu_to_be64(dma_address);
899	skspcl->req.n_sg = 1;
900	sgd->control = FIT_SGD_CONTROL_LAST;
901	sgd->byte_count = 0;
902	sgd->host_side_addr = skspcl->db_dma_address;
903	sgd->dev_side_addr = 0;
904	sgd->next_desc_ptr = 0LL;
905
906	return 1;
907}
908
909#define WR_BUF_SIZE SKD_N_INTERNAL_BYTES
910
911static void skd_send_internal_skspcl(struct skd_device *skdev,
912				     struct skd_special_context *skspcl,
913				     u8 opcode)
914{
915	struct fit_sg_descriptor *sgd = &skspcl->req.sksg_list[0];
916	struct skd_scsi_request *scsi;
917	unsigned char *buf = skspcl->data_buf;
918	int i;
919
920	if (skspcl->req.state != SKD_REQ_STATE_IDLE)
921		/*
922		 * A refresh is already in progress.
923		 * Just wait for it to finish.
924		 */
925		return;
926
927	skspcl->req.state = SKD_REQ_STATE_BUSY;
928
929	scsi = &skspcl->msg_buf->scsi[0];
930	scsi->hdr.tag = skspcl->req.id;
931
932	memset(scsi->cdb, 0, sizeof(scsi->cdb));
933
934	switch (opcode) {
935	case TEST_UNIT_READY:
936		scsi->cdb[0] = TEST_UNIT_READY;
937		sgd->byte_count = 0;
938		scsi->hdr.sg_list_len_bytes = 0;
939		break;
940
941	case READ_CAPACITY:
942		scsi->cdb[0] = READ_CAPACITY;
943		sgd->byte_count = SKD_N_READ_CAP_BYTES;
944		scsi->hdr.sg_list_len_bytes = cpu_to_be32(sgd->byte_count);
945		break;
946
947	case INQUIRY:
948		scsi->cdb[0] = INQUIRY;
949		scsi->cdb[1] = 0x01;    /* evpd */
950		scsi->cdb[2] = 0x80;    /* serial number page */
951		scsi->cdb[4] = 0x10;
952		sgd->byte_count = 16;
953		scsi->hdr.sg_list_len_bytes = cpu_to_be32(sgd->byte_count);
954		break;
955
956	case SYNCHRONIZE_CACHE:
957		scsi->cdb[0] = SYNCHRONIZE_CACHE;
958		sgd->byte_count = 0;
959		scsi->hdr.sg_list_len_bytes = 0;
960		break;
961
962	case WRITE_BUFFER:
963		scsi->cdb[0] = WRITE_BUFFER;
964		scsi->cdb[1] = 0x02;
965		scsi->cdb[7] = (WR_BUF_SIZE & 0xFF00) >> 8;
966		scsi->cdb[8] = WR_BUF_SIZE & 0xFF;
967		sgd->byte_count = WR_BUF_SIZE;
968		scsi->hdr.sg_list_len_bytes = cpu_to_be32(sgd->byte_count);
969		/* fill incrementing byte pattern */
970		for (i = 0; i < sgd->byte_count; i++)
971			buf[i] = i & 0xFF;
972		break;
973
974	case READ_BUFFER:
975		scsi->cdb[0] = READ_BUFFER;
976		scsi->cdb[1] = 0x02;
977		scsi->cdb[7] = (WR_BUF_SIZE & 0xFF00) >> 8;
978		scsi->cdb[8] = WR_BUF_SIZE & 0xFF;
979		sgd->byte_count = WR_BUF_SIZE;
980		scsi->hdr.sg_list_len_bytes = cpu_to_be32(sgd->byte_count);
981		memset(skspcl->data_buf, 0, sgd->byte_count);
982		break;
983
984	default:
985		SKD_ASSERT("Don't know what to send");
986		return;
987
988	}
989	skd_send_special_fitmsg(skdev, skspcl);
990}
991
992static void skd_refresh_device_data(struct skd_device *skdev)
993{
994	struct skd_special_context *skspcl = &skdev->internal_skspcl;
995
996	skd_send_internal_skspcl(skdev, skspcl, TEST_UNIT_READY);
997}
998
999static int skd_chk_read_buf(struct skd_device *skdev,
1000			    struct skd_special_context *skspcl)
1001{
1002	unsigned char *buf = skspcl->data_buf;
1003	int i;
1004
1005	/* check for incrementing byte pattern */
1006	for (i = 0; i < WR_BUF_SIZE; i++)
1007		if (buf[i] != (i & 0xFF))
1008			return 1;
1009
1010	return 0;
1011}
1012
1013static void skd_log_check_status(struct skd_device *skdev, u8 status, u8 key,
1014				 u8 code, u8 qual, u8 fruc)
1015{
1016	/* If the check condition is of special interest, log a message */
1017	if ((status == SAM_STAT_CHECK_CONDITION) && (key == 0x02)
1018	    && (code == 0x04) && (qual == 0x06)) {
1019		dev_err(&skdev->pdev->dev,
1020			"*** LOST_WRITE_DATA ERROR *** key/asc/ascq/fruc %02x/%02x/%02x/%02x\n",
1021			key, code, qual, fruc);
1022	}
1023}
1024
1025static void skd_complete_internal(struct skd_device *skdev,
1026				  struct fit_completion_entry_v1 *skcomp,
1027				  struct fit_comp_error_info *skerr,
1028				  struct skd_special_context *skspcl)
1029{
1030	u8 *buf = skspcl->data_buf;
1031	u8 status;
1032	int i;
1033	struct skd_scsi_request *scsi = &skspcl->msg_buf->scsi[0];
1034
1035	lockdep_assert_held(&skdev->lock);
1036
1037	SKD_ASSERT(skspcl == &skdev->internal_skspcl);
1038
1039	dev_dbg(&skdev->pdev->dev, "complete internal %x\n", scsi->cdb[0]);
1040
1041	dma_sync_single_for_cpu(&skdev->pdev->dev,
1042				skspcl->db_dma_address,
1043				skspcl->req.sksg_list[0].byte_count,
1044				DMA_BIDIRECTIONAL);
1045
1046	skspcl->req.completion = *skcomp;
1047	skspcl->req.state = SKD_REQ_STATE_IDLE;
1048
1049	status = skspcl->req.completion.status;
1050
1051	skd_log_check_status(skdev, status, skerr->key, skerr->code,
1052			     skerr->qual, skerr->fruc);
1053
1054	switch (scsi->cdb[0]) {
1055	case TEST_UNIT_READY:
1056		if (status == SAM_STAT_GOOD)
1057			skd_send_internal_skspcl(skdev, skspcl, WRITE_BUFFER);
1058		else if ((status == SAM_STAT_CHECK_CONDITION) &&
1059			 (skerr->key == MEDIUM_ERROR))
1060			skd_send_internal_skspcl(skdev, skspcl, WRITE_BUFFER);
1061		else {
1062			if (skdev->state == SKD_DRVR_STATE_STOPPING) {
1063				dev_dbg(&skdev->pdev->dev,
1064					"TUR failed, don't send anymore state 0x%x\n",
1065					skdev->state);
1066				return;
1067			}
1068			dev_dbg(&skdev->pdev->dev,
1069				"**** TUR failed, retry skerr\n");
1070			skd_send_internal_skspcl(skdev, skspcl,
1071						 TEST_UNIT_READY);
1072		}
1073		break;
1074
1075	case WRITE_BUFFER:
1076		if (status == SAM_STAT_GOOD)
1077			skd_send_internal_skspcl(skdev, skspcl, READ_BUFFER);
1078		else {
1079			if (skdev->state == SKD_DRVR_STATE_STOPPING) {
1080				dev_dbg(&skdev->pdev->dev,
1081					"write buffer failed, don't send anymore state 0x%x\n",
1082					skdev->state);
1083				return;
1084			}
1085			dev_dbg(&skdev->pdev->dev,
1086				"**** write buffer failed, retry skerr\n");
1087			skd_send_internal_skspcl(skdev, skspcl,
1088						 TEST_UNIT_READY);
1089		}
1090		break;
1091
1092	case READ_BUFFER:
1093		if (status == SAM_STAT_GOOD) {
1094			if (skd_chk_read_buf(skdev, skspcl) == 0)
1095				skd_send_internal_skspcl(skdev, skspcl,
1096							 READ_CAPACITY);
1097			else {
1098				dev_err(&skdev->pdev->dev,
1099					"*** W/R Buffer mismatch %d ***\n",
1100					skdev->connect_retries);
1101				if (skdev->connect_retries <
1102				    SKD_MAX_CONNECT_RETRIES) {
1103					skdev->connect_retries++;
1104					skd_soft_reset(skdev);
1105				} else {
1106					dev_err(&skdev->pdev->dev,
1107						"W/R Buffer Connect Error\n");
1108					return;
1109				}
1110			}
1111
1112		} else {
1113			if (skdev->state == SKD_DRVR_STATE_STOPPING) {
1114				dev_dbg(&skdev->pdev->dev,
1115					"read buffer failed, don't send anymore state 0x%x\n",
1116					skdev->state);
1117				return;
1118			}
1119			dev_dbg(&skdev->pdev->dev,
1120				"**** read buffer failed, retry skerr\n");
1121			skd_send_internal_skspcl(skdev, skspcl,
1122						 TEST_UNIT_READY);
1123		}
1124		break;
1125
1126	case READ_CAPACITY:
1127		skdev->read_cap_is_valid = 0;
1128		if (status == SAM_STAT_GOOD) {
1129			skdev->read_cap_last_lba =
1130				(buf[0] << 24) | (buf[1] << 16) |
1131				(buf[2] << 8) | buf[3];
1132			skdev->read_cap_blocksize =
1133				(buf[4] << 24) | (buf[5] << 16) |
1134				(buf[6] << 8) | buf[7];
1135
1136			dev_dbg(&skdev->pdev->dev, "last lba %d, bs %d\n",
1137				skdev->read_cap_last_lba,
1138				skdev->read_cap_blocksize);
1139
1140			set_capacity(skdev->disk, skdev->read_cap_last_lba + 1);
1141
1142			skdev->read_cap_is_valid = 1;
1143
1144			skd_send_internal_skspcl(skdev, skspcl, INQUIRY);
1145		} else if ((status == SAM_STAT_CHECK_CONDITION) &&
1146			   (skerr->key == MEDIUM_ERROR)) {
1147			skdev->read_cap_last_lba = ~0;
1148			set_capacity(skdev->disk, skdev->read_cap_last_lba + 1);
1149			dev_dbg(&skdev->pdev->dev, "**** MEDIUM ERROR caused READCAP to fail, ignore failure and continue to inquiry\n");
1150			skd_send_internal_skspcl(skdev, skspcl, INQUIRY);
1151		} else {
1152			dev_dbg(&skdev->pdev->dev, "**** READCAP failed, retry TUR\n");
1153			skd_send_internal_skspcl(skdev, skspcl,
1154						 TEST_UNIT_READY);
1155		}
1156		break;
1157
1158	case INQUIRY:
1159		skdev->inquiry_is_valid = 0;
1160		if (status == SAM_STAT_GOOD) {
1161			skdev->inquiry_is_valid = 1;
1162
1163			for (i = 0; i < 12; i++)
1164				skdev->inq_serial_num[i] = buf[i + 4];
1165			skdev->inq_serial_num[12] = 0;
1166		}
1167
1168		if (skd_unquiesce_dev(skdev) < 0)
1169			dev_dbg(&skdev->pdev->dev, "**** failed, to ONLINE device\n");
1170		 /* connection is complete */
1171		skdev->connect_retries = 0;
1172		break;
1173
1174	case SYNCHRONIZE_CACHE:
1175		if (status == SAM_STAT_GOOD)
1176			skdev->sync_done = 1;
1177		else
1178			skdev->sync_done = -1;
1179		wake_up_interruptible(&skdev->waitq);
1180		break;
1181
1182	default:
1183		SKD_ASSERT("we didn't send this");
1184	}
1185}
1186
1187/*
1188 *****************************************************************************
1189 * FIT MESSAGES
1190 *****************************************************************************
1191 */
1192
1193static void skd_send_fitmsg(struct skd_device *skdev,
1194			    struct skd_fitmsg_context *skmsg)
1195{
1196	u64 qcmd;
1197
1198	dev_dbg(&skdev->pdev->dev, "dma address %pad, busy=%d\n",
1199		&skmsg->mb_dma_address, skd_in_flight(skdev));
1200	dev_dbg(&skdev->pdev->dev, "msg_buf %p\n", skmsg->msg_buf);
1201
1202	qcmd = skmsg->mb_dma_address;
1203	qcmd |= FIT_QCMD_QID_NORMAL;
1204
1205	if (unlikely(skdev->dbg_level > 1)) {
1206		u8 *bp = (u8 *)skmsg->msg_buf;
1207		int i;
1208		for (i = 0; i < skmsg->length; i += 8) {
1209			dev_dbg(&skdev->pdev->dev, "msg[%2d] %8ph\n", i,
1210				&bp[i]);
1211			if (i == 0)
1212				i = 64 - 8;
1213		}
1214	}
1215
1216	if (skmsg->length > 256)
1217		qcmd |= FIT_QCMD_MSGSIZE_512;
1218	else if (skmsg->length > 128)
1219		qcmd |= FIT_QCMD_MSGSIZE_256;
1220	else if (skmsg->length > 64)
1221		qcmd |= FIT_QCMD_MSGSIZE_128;
1222	else
1223		/*
1224		 * This makes no sense because the FIT msg header is
1225		 * 64 bytes. If the msg is only 64 bytes long it has
1226		 * no payload.
1227		 */
1228		qcmd |= FIT_QCMD_MSGSIZE_64;
1229
1230	dma_sync_single_for_device(&skdev->pdev->dev, skmsg->mb_dma_address,
1231				   skmsg->length, DMA_TO_DEVICE);
1232
1233	/* Make sure skd_msg_buf is written before the doorbell is triggered. */
1234	smp_wmb();
1235
1236	SKD_WRITEQ(skdev, qcmd, FIT_Q_COMMAND);
1237}
1238
1239static void skd_send_special_fitmsg(struct skd_device *skdev,
1240				    struct skd_special_context *skspcl)
1241{
1242	u64 qcmd;
1243
1244	WARN_ON_ONCE(skspcl->req.n_sg != 1);
1245
1246	if (unlikely(skdev->dbg_level > 1)) {
1247		u8 *bp = (u8 *)skspcl->msg_buf;
1248		int i;
1249
1250		for (i = 0; i < SKD_N_SPECIAL_FITMSG_BYTES; i += 8) {
1251			dev_dbg(&skdev->pdev->dev, " spcl[%2d] %8ph\n", i,
1252				&bp[i]);
1253			if (i == 0)
1254				i = 64 - 8;
1255		}
1256
1257		dev_dbg(&skdev->pdev->dev,
1258			"skspcl=%p id=%04x sksg_list=%p sksg_dma=%pad\n",
1259			skspcl, skspcl->req.id, skspcl->req.sksg_list,
1260			&skspcl->req.sksg_dma_address);
1261		for (i = 0; i < skspcl->req.n_sg; i++) {
1262			struct fit_sg_descriptor *sgd =
1263				&skspcl->req.sksg_list[i];
1264
1265			dev_dbg(&skdev->pdev->dev,
1266				"  sg[%d] count=%u ctrl=0x%x addr=0x%llx next=0x%llx\n",
1267				i, sgd->byte_count, sgd->control,
1268				sgd->host_side_addr, sgd->next_desc_ptr);
1269		}
1270	}
1271
1272	/*
1273	 * Special FIT msgs are always 128 bytes: a 64-byte FIT hdr
1274	 * and one 64-byte SSDI command.
1275	 */
1276	qcmd = skspcl->mb_dma_address;
1277	qcmd |= FIT_QCMD_QID_NORMAL + FIT_QCMD_MSGSIZE_128;
1278
1279	dma_sync_single_for_device(&skdev->pdev->dev, skspcl->mb_dma_address,
1280				   SKD_N_SPECIAL_FITMSG_BYTES, DMA_TO_DEVICE);
1281	dma_sync_single_for_device(&skdev->pdev->dev,
1282				   skspcl->req.sksg_dma_address,
1283				   1 * sizeof(struct fit_sg_descriptor),
1284				   DMA_TO_DEVICE);
1285	dma_sync_single_for_device(&skdev->pdev->dev,
1286				   skspcl->db_dma_address,
1287				   skspcl->req.sksg_list[0].byte_count,
1288				   DMA_BIDIRECTIONAL);
1289
1290	/* Make sure skd_msg_buf is written before the doorbell is triggered. */
1291	smp_wmb();
1292
1293	SKD_WRITEQ(skdev, qcmd, FIT_Q_COMMAND);
1294}
1295
1296/*
1297 *****************************************************************************
1298 * COMPLETION QUEUE
1299 *****************************************************************************
1300 */
1301
1302static void skd_complete_other(struct skd_device *skdev,
1303			       struct fit_completion_entry_v1 *skcomp,
1304			       struct fit_comp_error_info *skerr);
1305
1306struct sns_info {
1307	u8 type;
1308	u8 stat;
1309	u8 key;
1310	u8 asc;
1311	u8 ascq;
1312	u8 mask;
1313	enum skd_check_status_action action;
1314};
1315
1316static struct sns_info skd_chkstat_table[] = {
1317	/* Good */
1318	{ 0x70, 0x02, RECOVERED_ERROR, 0,    0,	   0x1c,
1319	  SKD_CHECK_STATUS_REPORT_GOOD },
1320
1321	/* Smart alerts */
1322	{ 0x70, 0x02, NO_SENSE,	       0x0B, 0x00, 0x1E,	/* warnings */
1323	  SKD_CHECK_STATUS_REPORT_SMART_ALERT },
1324	{ 0x70, 0x02, NO_SENSE,	       0x5D, 0x00, 0x1E,	/* thresholds */
1325	  SKD_CHECK_STATUS_REPORT_SMART_ALERT },
1326	{ 0x70, 0x02, RECOVERED_ERROR, 0x0B, 0x01, 0x1F,        /* temperature over trigger */
1327	  SKD_CHECK_STATUS_REPORT_SMART_ALERT },
1328
1329	/* Retry (with limits) */
1330	{ 0x70, 0x02, 0x0B,	       0,    0,	   0x1C,        /* This one is for DMA ERROR */
1331	  SKD_CHECK_STATUS_REQUEUE_REQUEST },
1332	{ 0x70, 0x02, 0x06,	       0x0B, 0x00, 0x1E,        /* warnings */
1333	  SKD_CHECK_STATUS_REQUEUE_REQUEST },
1334	{ 0x70, 0x02, 0x06,	       0x5D, 0x00, 0x1E,        /* thresholds */
1335	  SKD_CHECK_STATUS_REQUEUE_REQUEST },
1336	{ 0x70, 0x02, 0x06,	       0x80, 0x30, 0x1F,        /* backup power */
1337	  SKD_CHECK_STATUS_REQUEUE_REQUEST },
1338
1339	/* Busy (or about to be) */
1340	{ 0x70, 0x02, 0x06,	       0x3f, 0x01, 0x1F, /* fw changed */
1341	  SKD_CHECK_STATUS_BUSY_IMMINENT },
1342};
1343
1344/*
1345 * Look up status and sense data to decide how to handle the error
1346 * from the device.
1347 * mask says which fields must match e.g., mask=0x18 means check
1348 * type and stat, ignore key, asc, ascq.
1349 */
1350
1351static enum skd_check_status_action
1352skd_check_status(struct skd_device *skdev,
1353		 u8 cmp_status, struct fit_comp_error_info *skerr)
1354{
1355	int i;
1356
1357	dev_err(&skdev->pdev->dev, "key/asc/ascq/fruc %02x/%02x/%02x/%02x\n",
1358		skerr->key, skerr->code, skerr->qual, skerr->fruc);
1359
1360	dev_dbg(&skdev->pdev->dev,
1361		"stat: t=%02x stat=%02x k=%02x c=%02x q=%02x fruc=%02x\n",
1362		skerr->type, cmp_status, skerr->key, skerr->code, skerr->qual,
1363		skerr->fruc);
1364
1365	/* Does the info match an entry in the good category? */
1366	for (i = 0; i < ARRAY_SIZE(skd_chkstat_table); i++) {
1367		struct sns_info *sns = &skd_chkstat_table[i];
1368
1369		if (sns->mask & 0x10)
1370			if (skerr->type != sns->type)
1371				continue;
1372
1373		if (sns->mask & 0x08)
1374			if (cmp_status != sns->stat)
1375				continue;
1376
1377		if (sns->mask & 0x04)
1378			if (skerr->key != sns->key)
1379				continue;
1380
1381		if (sns->mask & 0x02)
1382			if (skerr->code != sns->asc)
1383				continue;
1384
1385		if (sns->mask & 0x01)
1386			if (skerr->qual != sns->ascq)
1387				continue;
1388
1389		if (sns->action == SKD_CHECK_STATUS_REPORT_SMART_ALERT) {
1390			dev_err(&skdev->pdev->dev,
1391				"SMART Alert: sense key/asc/ascq %02x/%02x/%02x\n",
1392				skerr->key, skerr->code, skerr->qual);
1393		}
1394		return sns->action;
1395	}
1396
1397	/* No other match, so nonzero status means error,
1398	 * zero status means good
1399	 */
1400	if (cmp_status) {
1401		dev_dbg(&skdev->pdev->dev, "status check: error\n");
1402		return SKD_CHECK_STATUS_REPORT_ERROR;
1403	}
1404
1405	dev_dbg(&skdev->pdev->dev, "status check good default\n");
1406	return SKD_CHECK_STATUS_REPORT_GOOD;
1407}
1408
1409static void skd_resolve_req_exception(struct skd_device *skdev,
1410				      struct skd_request_context *skreq,
1411				      struct request *req)
1412{
1413	u8 cmp_status = skreq->completion.status;
1414
1415	switch (skd_check_status(skdev, cmp_status, &skreq->err_info)) {
1416	case SKD_CHECK_STATUS_REPORT_GOOD:
1417	case SKD_CHECK_STATUS_REPORT_SMART_ALERT:
1418		skreq->status = BLK_STS_OK;
1419		if (likely(!blk_should_fake_timeout(req->q)))
1420			blk_mq_complete_request(req);
1421		break;
1422
1423	case SKD_CHECK_STATUS_BUSY_IMMINENT:
1424		skd_log_skreq(skdev, skreq, "retry(busy)");
1425		blk_mq_requeue_request(req, true);
1426		dev_info(&skdev->pdev->dev, "drive BUSY imminent\n");
1427		skdev->state = SKD_DRVR_STATE_BUSY_IMMINENT;
1428		skdev->timer_countdown = SKD_TIMER_MINUTES(20);
1429		skd_quiesce_dev(skdev);
1430		break;
1431
1432	case SKD_CHECK_STATUS_REQUEUE_REQUEST:
1433		if (++skreq->retries < SKD_MAX_RETRIES) {
1434			skd_log_skreq(skdev, skreq, "retry");
1435			blk_mq_requeue_request(req, true);
1436			break;
1437		}
1438		fallthrough;
1439
1440	case SKD_CHECK_STATUS_REPORT_ERROR:
1441	default:
1442		skreq->status = BLK_STS_IOERR;
1443		if (likely(!blk_should_fake_timeout(req->q)))
1444			blk_mq_complete_request(req);
1445		break;
1446	}
1447}
1448
1449static void skd_release_skreq(struct skd_device *skdev,
1450			      struct skd_request_context *skreq)
1451{
1452	/*
1453	 * Reclaim the skd_request_context
1454	 */
1455	skreq->state = SKD_REQ_STATE_IDLE;
1456}
1457
1458static int skd_isr_completion_posted(struct skd_device *skdev,
1459					int limit, int *enqueued)
1460{
1461	struct fit_completion_entry_v1 *skcmp;
1462	struct fit_comp_error_info *skerr;
1463	u16 req_id;
1464	u32 tag;
1465	u16 hwq = 0;
1466	struct request *rq;
1467	struct skd_request_context *skreq;
1468	u16 cmp_cntxt;
1469	u8 cmp_status;
1470	u8 cmp_cycle;
1471	u32 cmp_bytes;
1472	int rc = 0;
1473	int processed = 0;
1474
1475	lockdep_assert_held(&skdev->lock);
1476
1477	for (;; ) {
1478		SKD_ASSERT(skdev->skcomp_ix < SKD_N_COMPLETION_ENTRY);
1479
1480		skcmp = &skdev->skcomp_table[skdev->skcomp_ix];
1481		cmp_cycle = skcmp->cycle;
1482		cmp_cntxt = skcmp->tag;
1483		cmp_status = skcmp->status;
1484		cmp_bytes = be32_to_cpu(skcmp->num_returned_bytes);
1485
1486		skerr = &skdev->skerr_table[skdev->skcomp_ix];
1487
1488		dev_dbg(&skdev->pdev->dev,
1489			"cycle=%d ix=%d got cycle=%d cmdctxt=0x%x stat=%d busy=%d rbytes=0x%x proto=%d\n",
1490			skdev->skcomp_cycle, skdev->skcomp_ix, cmp_cycle,
1491			cmp_cntxt, cmp_status, skd_in_flight(skdev),
1492			cmp_bytes, skdev->proto_ver);
1493
1494		if (cmp_cycle != skdev->skcomp_cycle) {
1495			dev_dbg(&skdev->pdev->dev, "end of completions\n");
1496			break;
1497		}
1498		/*
1499		 * Update the completion queue head index and possibly
1500		 * the completion cycle count. 8-bit wrap-around.
1501		 */
1502		skdev->skcomp_ix++;
1503		if (skdev->skcomp_ix >= SKD_N_COMPLETION_ENTRY) {
1504			skdev->skcomp_ix = 0;
1505			skdev->skcomp_cycle++;
1506		}
1507
1508		/*
1509		 * The command context is a unique 32-bit ID. The low order
1510		 * bits help locate the request. The request is usually a
1511		 * r/w request (see skd_start() above) or a special request.
1512		 */
1513		req_id = cmp_cntxt;
1514		tag = req_id & SKD_ID_SLOT_AND_TABLE_MASK;
1515
1516		/* Is this other than a r/w request? */
1517		if (tag >= skdev->num_req_context) {
1518			/*
1519			 * This is not a completion for a r/w request.
1520			 */
1521			WARN_ON_ONCE(blk_mq_tag_to_rq(skdev->tag_set.tags[hwq],
1522						      tag));
1523			skd_complete_other(skdev, skcmp, skerr);
1524			continue;
1525		}
1526
1527		rq = blk_mq_tag_to_rq(skdev->tag_set.tags[hwq], tag);
1528		if (WARN(!rq, "No request for tag %#x -> %#x\n", cmp_cntxt,
1529			 tag))
1530			continue;
1531		skreq = blk_mq_rq_to_pdu(rq);
1532
1533		/*
1534		 * Make sure the request ID for the slot matches.
1535		 */
1536		if (skreq->id != req_id) {
1537			dev_err(&skdev->pdev->dev,
1538				"Completion mismatch comp_id=0x%04x skreq=0x%04x new=0x%04x\n",
1539				req_id, skreq->id, cmp_cntxt);
1540
1541			continue;
1542		}
1543
1544		SKD_ASSERT(skreq->state == SKD_REQ_STATE_BUSY);
1545
1546		skreq->completion = *skcmp;
1547		if (unlikely(cmp_status == SAM_STAT_CHECK_CONDITION)) {
1548			skreq->err_info = *skerr;
1549			skd_log_check_status(skdev, cmp_status, skerr->key,
1550					     skerr->code, skerr->qual,
1551					     skerr->fruc);
1552		}
1553		/* Release DMA resources for the request. */
1554		if (skreq->n_sg > 0)
1555			skd_postop_sg_list(skdev, skreq);
1556
1557		skd_release_skreq(skdev, skreq);
1558
1559		/*
1560		 * Capture the outcome and post it back to the native request.
1561		 */
1562		if (likely(cmp_status == SAM_STAT_GOOD)) {
1563			skreq->status = BLK_STS_OK;
1564			if (likely(!blk_should_fake_timeout(rq->q)))
1565				blk_mq_complete_request(rq);
1566		} else {
1567			skd_resolve_req_exception(skdev, skreq, rq);
1568		}
1569
1570		/* skd_isr_comp_limit equal zero means no limit */
1571		if (limit) {
1572			if (++processed >= limit) {
1573				rc = 1;
1574				break;
1575			}
1576		}
1577	}
1578
1579	if (skdev->state == SKD_DRVR_STATE_PAUSING &&
1580	    skd_in_flight(skdev) == 0) {
1581		skdev->state = SKD_DRVR_STATE_PAUSED;
1582		wake_up_interruptible(&skdev->waitq);
1583	}
1584
1585	return rc;
1586}
1587
1588static void skd_complete_other(struct skd_device *skdev,
1589			       struct fit_completion_entry_v1 *skcomp,
1590			       struct fit_comp_error_info *skerr)
1591{
1592	u32 req_id = 0;
1593	u32 req_table;
1594	u32 req_slot;
1595	struct skd_special_context *skspcl;
1596
1597	lockdep_assert_held(&skdev->lock);
1598
1599	req_id = skcomp->tag;
1600	req_table = req_id & SKD_ID_TABLE_MASK;
1601	req_slot = req_id & SKD_ID_SLOT_MASK;
1602
1603	dev_dbg(&skdev->pdev->dev, "table=0x%x id=0x%x slot=%d\n", req_table,
1604		req_id, req_slot);
1605
1606	/*
1607	 * Based on the request id, determine how to dispatch this completion.
1608	 * This swich/case is finding the good cases and forwarding the
1609	 * completion entry. Errors are reported below the switch.
1610	 */
1611	switch (req_table) {
1612	case SKD_ID_RW_REQUEST:
1613		/*
1614		 * The caller, skd_isr_completion_posted() above,
1615		 * handles r/w requests. The only way we get here
1616		 * is if the req_slot is out of bounds.
1617		 */
1618		break;
1619
1620	case SKD_ID_INTERNAL:
1621		if (req_slot == 0) {
1622			skspcl = &skdev->internal_skspcl;
1623			if (skspcl->req.id == req_id &&
1624			    skspcl->req.state == SKD_REQ_STATE_BUSY) {
1625				skd_complete_internal(skdev,
1626						      skcomp, skerr, skspcl);
1627				return;
1628			}
1629		}
1630		break;
1631
1632	case SKD_ID_FIT_MSG:
1633		/*
1634		 * These id's should never appear in a completion record.
1635		 */
1636		break;
1637
1638	default:
1639		/*
1640		 * These id's should never appear anywhere;
1641		 */
1642		break;
1643	}
1644
1645	/*
1646	 * If we get here it is a bad or stale id.
1647	 */
1648}
1649
1650static void skd_reset_skcomp(struct skd_device *skdev)
1651{
1652	memset(skdev->skcomp_table, 0, SKD_SKCOMP_SIZE);
1653
1654	skdev->skcomp_ix = 0;
1655	skdev->skcomp_cycle = 1;
1656}
1657
1658/*
1659 *****************************************************************************
1660 * INTERRUPTS
1661 *****************************************************************************
1662 */
1663static void skd_completion_worker(struct work_struct *work)
1664{
1665	struct skd_device *skdev =
1666		container_of(work, struct skd_device, completion_worker);
1667	unsigned long flags;
1668	int flush_enqueued = 0;
1669
1670	spin_lock_irqsave(&skdev->lock, flags);
1671
1672	/*
1673	 * pass in limit=0, which means no limit..
1674	 * process everything in compq
1675	 */
1676	skd_isr_completion_posted(skdev, 0, &flush_enqueued);
1677	schedule_work(&skdev->start_queue);
1678
1679	spin_unlock_irqrestore(&skdev->lock, flags);
1680}
1681
1682static void skd_isr_msg_from_dev(struct skd_device *skdev);
1683
1684static irqreturn_t
1685skd_isr(int irq, void *ptr)
1686{
1687	struct skd_device *skdev = ptr;
1688	u32 intstat;
1689	u32 ack;
1690	int rc = 0;
1691	int deferred = 0;
1692	int flush_enqueued = 0;
1693
1694	spin_lock(&skdev->lock);
1695
1696	for (;; ) {
1697		intstat = SKD_READL(skdev, FIT_INT_STATUS_HOST);
1698
1699		ack = FIT_INT_DEF_MASK;
1700		ack &= intstat;
1701
1702		dev_dbg(&skdev->pdev->dev, "intstat=0x%x ack=0x%x\n", intstat,
1703			ack);
1704
1705		/* As long as there is an int pending on device, keep
1706		 * running loop.  When none, get out, but if we've never
1707		 * done any processing, call completion handler?
1708		 */
1709		if (ack == 0) {
1710			/* No interrupts on device, but run the completion
1711			 * processor anyway?
1712			 */
1713			if (rc == 0)
1714				if (likely (skdev->state
1715					== SKD_DRVR_STATE_ONLINE))
1716					deferred = 1;
1717			break;
1718		}
1719
1720		rc = IRQ_HANDLED;
1721
1722		SKD_WRITEL(skdev, ack, FIT_INT_STATUS_HOST);
1723
1724		if (likely((skdev->state != SKD_DRVR_STATE_LOAD) &&
1725			   (skdev->state != SKD_DRVR_STATE_STOPPING))) {
1726			if (intstat & FIT_ISH_COMPLETION_POSTED) {
1727				/*
1728				 * If we have already deferred completion
1729				 * processing, don't bother running it again
1730				 */
1731				if (deferred == 0)
1732					deferred =
1733						skd_isr_completion_posted(skdev,
1734						skd_isr_comp_limit, &flush_enqueued);
1735			}
1736
1737			if (intstat & FIT_ISH_FW_STATE_CHANGE) {
1738				skd_isr_fwstate(skdev);
1739				if (skdev->state == SKD_DRVR_STATE_FAULT ||
1740				    skdev->state ==
1741				    SKD_DRVR_STATE_DISAPPEARED) {
1742					spin_unlock(&skdev->lock);
1743					return rc;
1744				}
1745			}
1746
1747			if (intstat & FIT_ISH_MSG_FROM_DEV)
1748				skd_isr_msg_from_dev(skdev);
1749		}
1750	}
1751
1752	if (unlikely(flush_enqueued))
1753		schedule_work(&skdev->start_queue);
1754
1755	if (deferred)
1756		schedule_work(&skdev->completion_worker);
1757	else if (!flush_enqueued)
1758		schedule_work(&skdev->start_queue);
1759
1760	spin_unlock(&skdev->lock);
1761
1762	return rc;
1763}
1764
1765static void skd_drive_fault(struct skd_device *skdev)
1766{
1767	skdev->state = SKD_DRVR_STATE_FAULT;
1768	dev_err(&skdev->pdev->dev, "Drive FAULT\n");
1769}
1770
1771static void skd_drive_disappeared(struct skd_device *skdev)
1772{
1773	skdev->state = SKD_DRVR_STATE_DISAPPEARED;
1774	dev_err(&skdev->pdev->dev, "Drive DISAPPEARED\n");
1775}
1776
1777static void skd_isr_fwstate(struct skd_device *skdev)
1778{
1779	u32 sense;
1780	u32 state;
1781	u32 mtd;
1782	int prev_driver_state = skdev->state;
1783
1784	sense = SKD_READL(skdev, FIT_STATUS);
1785	state = sense & FIT_SR_DRIVE_STATE_MASK;
1786
1787	dev_err(&skdev->pdev->dev, "s1120 state %s(%d)=>%s(%d)\n",
1788		skd_drive_state_to_str(skdev->drive_state), skdev->drive_state,
1789		skd_drive_state_to_str(state), state);
1790
1791	skdev->drive_state = state;
1792
1793	switch (skdev->drive_state) {
1794	case FIT_SR_DRIVE_INIT:
1795		if (skdev->state == SKD_DRVR_STATE_PROTOCOL_MISMATCH) {
1796			skd_disable_interrupts(skdev);
1797			break;
1798		}
1799		if (skdev->state == SKD_DRVR_STATE_RESTARTING)
1800			skd_recover_requests(skdev);
1801		if (skdev->state == SKD_DRVR_STATE_WAIT_BOOT) {
1802			skdev->timer_countdown = SKD_STARTING_TIMO;
1803			skdev->state = SKD_DRVR_STATE_STARTING;
1804			skd_soft_reset(skdev);
1805			break;
1806		}
1807		mtd = FIT_MXD_CONS(FIT_MTD_FITFW_INIT, 0, 0);
1808		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1809		skdev->last_mtd = mtd;
1810		break;
1811
1812	case FIT_SR_DRIVE_ONLINE:
1813		skdev->cur_max_queue_depth = skd_max_queue_depth;
1814		if (skdev->cur_max_queue_depth > skdev->dev_max_queue_depth)
1815			skdev->cur_max_queue_depth = skdev->dev_max_queue_depth;
1816
1817		skdev->queue_low_water_mark =
1818			skdev->cur_max_queue_depth * 2 / 3 + 1;
1819		if (skdev->queue_low_water_mark < 1)
1820			skdev->queue_low_water_mark = 1;
1821		dev_info(&skdev->pdev->dev,
1822			 "Queue depth limit=%d dev=%d lowat=%d\n",
1823			 skdev->cur_max_queue_depth,
1824			 skdev->dev_max_queue_depth,
1825			 skdev->queue_low_water_mark);
1826
1827		skd_refresh_device_data(skdev);
1828		break;
1829
1830	case FIT_SR_DRIVE_BUSY:
1831		skdev->state = SKD_DRVR_STATE_BUSY;
1832		skdev->timer_countdown = SKD_BUSY_TIMO;
1833		skd_quiesce_dev(skdev);
1834		break;
1835	case FIT_SR_DRIVE_BUSY_SANITIZE:
1836		/* set timer for 3 seconds, we'll abort any unfinished
1837		 * commands after that expires
1838		 */
1839		skdev->state = SKD_DRVR_STATE_BUSY_SANITIZE;
1840		skdev->timer_countdown = SKD_TIMER_SECONDS(3);
1841		schedule_work(&skdev->start_queue);
1842		break;
1843	case FIT_SR_DRIVE_BUSY_ERASE:
1844		skdev->state = SKD_DRVR_STATE_BUSY_ERASE;
1845		skdev->timer_countdown = SKD_BUSY_TIMO;
1846		break;
1847	case FIT_SR_DRIVE_OFFLINE:
1848		skdev->state = SKD_DRVR_STATE_IDLE;
1849		break;
1850	case FIT_SR_DRIVE_SOFT_RESET:
1851		switch (skdev->state) {
1852		case SKD_DRVR_STATE_STARTING:
1853		case SKD_DRVR_STATE_RESTARTING:
1854			/* Expected by a caller of skd_soft_reset() */
1855			break;
1856		default:
1857			skdev->state = SKD_DRVR_STATE_RESTARTING;
1858			break;
1859		}
1860		break;
1861	case FIT_SR_DRIVE_FW_BOOTING:
1862		dev_dbg(&skdev->pdev->dev, "ISR FIT_SR_DRIVE_FW_BOOTING\n");
1863		skdev->state = SKD_DRVR_STATE_WAIT_BOOT;
1864		skdev->timer_countdown = SKD_WAIT_BOOT_TIMO;
1865		break;
1866
1867	case FIT_SR_DRIVE_DEGRADED:
1868	case FIT_SR_PCIE_LINK_DOWN:
1869	case FIT_SR_DRIVE_NEED_FW_DOWNLOAD:
1870		break;
1871
1872	case FIT_SR_DRIVE_FAULT:
1873		skd_drive_fault(skdev);
1874		skd_recover_requests(skdev);
1875		schedule_work(&skdev->start_queue);
1876		break;
1877
1878	/* PCIe bus returned all Fs? */
1879	case 0xFF:
1880		dev_info(&skdev->pdev->dev, "state=0x%x sense=0x%x\n", state,
1881			 sense);
1882		skd_drive_disappeared(skdev);
1883		skd_recover_requests(skdev);
1884		schedule_work(&skdev->start_queue);
1885		break;
1886	default:
1887		/*
1888		 * Uknown FW State. Wait for a state we recognize.
1889		 */
1890		break;
1891	}
1892	dev_err(&skdev->pdev->dev, "Driver state %s(%d)=>%s(%d)\n",
1893		skd_skdev_state_to_str(prev_driver_state), prev_driver_state,
1894		skd_skdev_state_to_str(skdev->state), skdev->state);
1895}
1896
1897static bool skd_recover_request(struct request *req, void *data, bool reserved)
1898{
1899	struct skd_device *const skdev = data;
1900	struct skd_request_context *skreq = blk_mq_rq_to_pdu(req);
1901
1902	if (skreq->state != SKD_REQ_STATE_BUSY)
1903		return true;
1904
1905	skd_log_skreq(skdev, skreq, "recover");
1906
1907	/* Release DMA resources for the request. */
1908	if (skreq->n_sg > 0)
1909		skd_postop_sg_list(skdev, skreq);
1910
1911	skreq->state = SKD_REQ_STATE_IDLE;
1912	skreq->status = BLK_STS_IOERR;
1913	blk_mq_complete_request(req);
1914	return true;
1915}
1916
1917static void skd_recover_requests(struct skd_device *skdev)
1918{
1919	blk_mq_tagset_busy_iter(&skdev->tag_set, skd_recover_request, skdev);
1920}
1921
1922static void skd_isr_msg_from_dev(struct skd_device *skdev)
1923{
1924	u32 mfd;
1925	u32 mtd;
1926	u32 data;
1927
1928	mfd = SKD_READL(skdev, FIT_MSG_FROM_DEVICE);
1929
1930	dev_dbg(&skdev->pdev->dev, "mfd=0x%x last_mtd=0x%x\n", mfd,
1931		skdev->last_mtd);
1932
1933	/* ignore any mtd that is an ack for something we didn't send */
1934	if (FIT_MXD_TYPE(mfd) != FIT_MXD_TYPE(skdev->last_mtd))
1935		return;
1936
1937	switch (FIT_MXD_TYPE(mfd)) {
1938	case FIT_MTD_FITFW_INIT:
1939		skdev->proto_ver = FIT_PROTOCOL_MAJOR_VER(mfd);
1940
1941		if (skdev->proto_ver != FIT_PROTOCOL_VERSION_1) {
1942			dev_err(&skdev->pdev->dev, "protocol mismatch\n");
1943			dev_err(&skdev->pdev->dev, "  got=%d support=%d\n",
1944				skdev->proto_ver, FIT_PROTOCOL_VERSION_1);
1945			dev_err(&skdev->pdev->dev, "  please upgrade driver\n");
1946			skdev->state = SKD_DRVR_STATE_PROTOCOL_MISMATCH;
1947			skd_soft_reset(skdev);
1948			break;
1949		}
1950		mtd = FIT_MXD_CONS(FIT_MTD_GET_CMDQ_DEPTH, 0, 0);
1951		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1952		skdev->last_mtd = mtd;
1953		break;
1954
1955	case FIT_MTD_GET_CMDQ_DEPTH:
1956		skdev->dev_max_queue_depth = FIT_MXD_DATA(mfd);
1957		mtd = FIT_MXD_CONS(FIT_MTD_SET_COMPQ_DEPTH, 0,
1958				   SKD_N_COMPLETION_ENTRY);
1959		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1960		skdev->last_mtd = mtd;
1961		break;
1962
1963	case FIT_MTD_SET_COMPQ_DEPTH:
1964		SKD_WRITEQ(skdev, skdev->cq_dma_address, FIT_MSG_TO_DEVICE_ARG);
1965		mtd = FIT_MXD_CONS(FIT_MTD_SET_COMPQ_ADDR, 0, 0);
1966		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1967		skdev->last_mtd = mtd;
1968		break;
1969
1970	case FIT_MTD_SET_COMPQ_ADDR:
1971		skd_reset_skcomp(skdev);
1972		mtd = FIT_MXD_CONS(FIT_MTD_CMD_LOG_HOST_ID, 0, skdev->devno);
1973		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1974		skdev->last_mtd = mtd;
1975		break;
1976
1977	case FIT_MTD_CMD_LOG_HOST_ID:
1978		/* hardware interface overflows in y2106 */
1979		skdev->connect_time_stamp = (u32)ktime_get_real_seconds();
1980		data = skdev->connect_time_stamp & 0xFFFF;
1981		mtd = FIT_MXD_CONS(FIT_MTD_CMD_LOG_TIME_STAMP_LO, 0, data);
1982		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1983		skdev->last_mtd = mtd;
1984		break;
1985
1986	case FIT_MTD_CMD_LOG_TIME_STAMP_LO:
1987		skdev->drive_jiffies = FIT_MXD_DATA(mfd);
1988		data = (skdev->connect_time_stamp >> 16) & 0xFFFF;
1989		mtd = FIT_MXD_CONS(FIT_MTD_CMD_LOG_TIME_STAMP_HI, 0, data);
1990		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1991		skdev->last_mtd = mtd;
1992		break;
1993
1994	case FIT_MTD_CMD_LOG_TIME_STAMP_HI:
1995		skdev->drive_jiffies |= (FIT_MXD_DATA(mfd) << 16);
1996		mtd = FIT_MXD_CONS(FIT_MTD_ARM_QUEUE, 0, 0);
1997		SKD_WRITEL(skdev, mtd, FIT_MSG_TO_DEVICE);
1998		skdev->last_mtd = mtd;
1999
2000		dev_err(&skdev->pdev->dev, "Time sync driver=0x%x device=0x%x\n",
2001			skdev->connect_time_stamp, skdev->drive_jiffies);
2002		break;
2003
2004	case FIT_MTD_ARM_QUEUE:
2005		skdev->last_mtd = 0;
2006		/*
2007		 * State should be, or soon will be, FIT_SR_DRIVE_ONLINE.
2008		 */
2009		break;
2010
2011	default:
2012		break;
2013	}
2014}
2015
2016static void skd_disable_interrupts(struct skd_device *skdev)
2017{
2018	u32 sense;
2019
2020	sense = SKD_READL(skdev, FIT_CONTROL);
2021	sense &= ~FIT_CR_ENABLE_INTERRUPTS;
2022	SKD_WRITEL(skdev, sense, FIT_CONTROL);
2023	dev_dbg(&skdev->pdev->dev, "sense 0x%x\n", sense);
2024
2025	/* Note that the 1s is written. A 1-bit means
2026	 * disable, a 0 means enable.
2027	 */
2028	SKD_WRITEL(skdev, ~0, FIT_INT_MASK_HOST);
2029}
2030
2031static void skd_enable_interrupts(struct skd_device *skdev)
2032{
2033	u32 val;
2034
2035	/* unmask interrupts first */
2036	val = FIT_ISH_FW_STATE_CHANGE +
2037	      FIT_ISH_COMPLETION_POSTED + FIT_ISH_MSG_FROM_DEV;
2038
2039	/* Note that the compliment of mask is written. A 1-bit means
2040	 * disable, a 0 means enable. */
2041	SKD_WRITEL(skdev, ~val, FIT_INT_MASK_HOST);
2042	dev_dbg(&skdev->pdev->dev, "interrupt mask=0x%x\n", ~val);
2043
2044	val = SKD_READL(skdev, FIT_CONTROL);
2045	val |= FIT_CR_ENABLE_INTERRUPTS;
2046	dev_dbg(&skdev->pdev->dev, "control=0x%x\n", val);
2047	SKD_WRITEL(skdev, val, FIT_CONTROL);
2048}
2049
2050/*
2051 *****************************************************************************
2052 * START, STOP, RESTART, QUIESCE, UNQUIESCE
2053 *****************************************************************************
2054 */
2055
2056static void skd_soft_reset(struct skd_device *skdev)
2057{
2058	u32 val;
2059
2060	val = SKD_READL(skdev, FIT_CONTROL);
2061	val |= (FIT_CR_SOFT_RESET);
2062	dev_dbg(&skdev->pdev->dev, "control=0x%x\n", val);
2063	SKD_WRITEL(skdev, val, FIT_CONTROL);
2064}
2065
2066static void skd_start_device(struct skd_device *skdev)
2067{
2068	unsigned long flags;
2069	u32 sense;
2070	u32 state;
2071
2072	spin_lock_irqsave(&skdev->lock, flags);
2073
2074	/* ack all ghost interrupts */
2075	SKD_WRITEL(skdev, FIT_INT_DEF_MASK, FIT_INT_STATUS_HOST);
2076
2077	sense = SKD_READL(skdev, FIT_STATUS);
2078
2079	dev_dbg(&skdev->pdev->dev, "initial status=0x%x\n", sense);
2080
2081	state = sense & FIT_SR_DRIVE_STATE_MASK;
2082	skdev->drive_state = state;
2083	skdev->last_mtd = 0;
2084
2085	skdev->state = SKD_DRVR_STATE_STARTING;
2086	skdev->timer_countdown = SKD_STARTING_TIMO;
2087
2088	skd_enable_interrupts(skdev);
2089
2090	switch (skdev->drive_state) {
2091	case FIT_SR_DRIVE_OFFLINE:
2092		dev_err(&skdev->pdev->dev, "Drive offline...\n");
2093		break;
2094
2095	case FIT_SR_DRIVE_FW_BOOTING:
2096		dev_dbg(&skdev->pdev->dev, "FIT_SR_DRIVE_FW_BOOTING\n");
2097		skdev->state = SKD_DRVR_STATE_WAIT_BOOT;
2098		skdev->timer_countdown = SKD_WAIT_BOOT_TIMO;
2099		break;
2100
2101	case FIT_SR_DRIVE_BUSY_SANITIZE:
2102		dev_info(&skdev->pdev->dev, "Start: BUSY_SANITIZE\n");
2103		skdev->state = SKD_DRVR_STATE_BUSY_SANITIZE;
2104		skdev->timer_countdown = SKD_STARTED_BUSY_TIMO;
2105		break;
2106
2107	case FIT_SR_DRIVE_BUSY_ERASE:
2108		dev_info(&skdev->pdev->dev, "Start: BUSY_ERASE\n");
2109		skdev->state = SKD_DRVR_STATE_BUSY_ERASE;
2110		skdev->timer_countdown = SKD_STARTED_BUSY_TIMO;
2111		break;
2112
2113	case FIT_SR_DRIVE_INIT:
2114	case FIT_SR_DRIVE_ONLINE:
2115		skd_soft_reset(skdev);
2116		break;
2117
2118	case FIT_SR_DRIVE_BUSY:
2119		dev_err(&skdev->pdev->dev, "Drive Busy...\n");
2120		skdev->state = SKD_DRVR_STATE_BUSY;
2121		skdev->timer_countdown = SKD_STARTED_BUSY_TIMO;
2122		break;
2123
2124	case FIT_SR_DRIVE_SOFT_RESET:
2125		dev_err(&skdev->pdev->dev, "drive soft reset in prog\n");
2126		break;
2127
2128	case FIT_SR_DRIVE_FAULT:
2129		/* Fault state is bad...soft reset won't do it...
2130		 * Hard reset, maybe, but does it work on device?
2131		 * For now, just fault so the system doesn't hang.
2132		 */
2133		skd_drive_fault(skdev);
2134		/*start the queue so we can respond with error to requests */
2135		dev_dbg(&skdev->pdev->dev, "starting queue\n");
2136		schedule_work(&skdev->start_queue);
2137		skdev->gendisk_on = -1;
2138		wake_up_interruptible(&skdev->waitq);
2139		break;
2140
2141	case 0xFF:
2142		/* Most likely the device isn't there or isn't responding
2143		 * to the BAR1 addresses. */
2144		skd_drive_disappeared(skdev);
2145		/*start the queue so we can respond with error to requests */
2146		dev_dbg(&skdev->pdev->dev,
2147			"starting queue to error-out reqs\n");
2148		schedule_work(&skdev->start_queue);
2149		skdev->gendisk_on = -1;
2150		wake_up_interruptible(&skdev->waitq);
2151		break;
2152
2153	default:
2154		dev_err(&skdev->pdev->dev, "Start: unknown state %x\n",
2155			skdev->drive_state);
2156		break;
2157	}
2158
2159	state = SKD_READL(skdev, FIT_CONTROL);
2160	dev_dbg(&skdev->pdev->dev, "FIT Control Status=0x%x\n", state);
2161
2162	state = SKD_READL(skdev, FIT_INT_STATUS_HOST);
2163	dev_dbg(&skdev->pdev->dev, "Intr Status=0x%x\n", state);
2164
2165	state = SKD_READL(skdev, FIT_INT_MASK_HOST);
2166	dev_dbg(&skdev->pdev->dev, "Intr Mask=0x%x\n", state);
2167
2168	state = SKD_READL(skdev, FIT_MSG_FROM_DEVICE);
2169	dev_dbg(&skdev->pdev->dev, "Msg from Dev=0x%x\n", state);
2170
2171	state = SKD_READL(skdev, FIT_HW_VERSION);
2172	dev_dbg(&skdev->pdev->dev, "HW version=0x%x\n", state);
2173
2174	spin_unlock_irqrestore(&skdev->lock, flags);
2175}
2176
2177static void skd_stop_device(struct skd_device *skdev)
2178{
2179	unsigned long flags;
2180	struct skd_special_context *skspcl = &skdev->internal_skspcl;
2181	u32 dev_state;
2182	int i;
2183
2184	spin_lock_irqsave(&skdev->lock, flags);
2185
2186	if (skdev->state != SKD_DRVR_STATE_ONLINE) {
2187		dev_err(&skdev->pdev->dev, "%s not online no sync\n", __func__);
2188		goto stop_out;
2189	}
2190
2191	if (skspcl->req.state != SKD_REQ_STATE_IDLE) {
2192		dev_err(&skdev->pdev->dev, "%s no special\n", __func__);
2193		goto stop_out;
2194	}
2195
2196	skdev->state = SKD_DRVR_STATE_SYNCING;
2197	skdev->sync_done = 0;
2198
2199	skd_send_internal_skspcl(skdev, skspcl, SYNCHRONIZE_CACHE);
2200
2201	spin_unlock_irqrestore(&skdev->lock, flags);
2202
2203	wait_event_interruptible_timeout(skdev->waitq,
2204					 (skdev->sync_done), (10 * HZ));
2205
2206	spin_lock_irqsave(&skdev->lock, flags);
2207
2208	switch (skdev->sync_done) {
2209	case 0:
2210		dev_err(&skdev->pdev->dev, "%s no sync\n", __func__);
2211		break;
2212	case 1:
2213		dev_err(&skdev->pdev->dev, "%s sync done\n", __func__);
2214		break;
2215	default:
2216		dev_err(&skdev->pdev->dev, "%s sync error\n", __func__);
2217	}
2218
2219stop_out:
2220	skdev->state = SKD_DRVR_STATE_STOPPING;
2221	spin_unlock_irqrestore(&skdev->lock, flags);
2222
2223	skd_kill_timer(skdev);
2224
2225	spin_lock_irqsave(&skdev->lock, flags);
2226	skd_disable_interrupts(skdev);
2227
2228	/* ensure all ints on device are cleared */
2229	/* soft reset the device to unload with a clean slate */
2230	SKD_WRITEL(skdev, FIT_INT_DEF_MASK, FIT_INT_STATUS_HOST);
2231	SKD_WRITEL(skdev, FIT_CR_SOFT_RESET, FIT_CONTROL);
2232
2233	spin_unlock_irqrestore(&skdev->lock, flags);
2234
2235	/* poll every 100ms, 1 second timeout */
2236	for (i = 0; i < 10; i++) {
2237		dev_state =
2238			SKD_READL(skdev, FIT_STATUS) & FIT_SR_DRIVE_STATE_MASK;
2239		if (dev_state == FIT_SR_DRIVE_INIT)
2240			break;
2241		set_current_state(TASK_INTERRUPTIBLE);
2242		schedule_timeout(msecs_to_jiffies(100));
2243	}
2244
2245	if (dev_state != FIT_SR_DRIVE_INIT)
2246		dev_err(&skdev->pdev->dev, "%s state error 0x%02x\n", __func__,
2247			dev_state);
2248}
2249
2250/* assume spinlock is held */
2251static void skd_restart_device(struct skd_device *skdev)
2252{
2253	u32 state;
2254
2255	/* ack all ghost interrupts */
2256	SKD_WRITEL(skdev, FIT_INT_DEF_MASK, FIT_INT_STATUS_HOST);
2257
2258	state = SKD_READL(skdev, FIT_STATUS);
2259
2260	dev_dbg(&skdev->pdev->dev, "drive status=0x%x\n", state);
2261
2262	state &= FIT_SR_DRIVE_STATE_MASK;
2263	skdev->drive_state = state;
2264	skdev->last_mtd = 0;
2265
2266	skdev->state = SKD_DRVR_STATE_RESTARTING;
2267	skdev->timer_countdown = SKD_RESTARTING_TIMO;
2268
2269	skd_soft_reset(skdev);
2270}
2271
2272/* assume spinlock is held */
2273static int skd_quiesce_dev(struct skd_device *skdev)
2274{
2275	int rc = 0;
2276
2277	switch (skdev->state) {
2278	case SKD_DRVR_STATE_BUSY:
2279	case SKD_DRVR_STATE_BUSY_IMMINENT:
2280		dev_dbg(&skdev->pdev->dev, "stopping queue\n");
2281		blk_mq_stop_hw_queues(skdev->queue);
2282		break;
2283	case SKD_DRVR_STATE_ONLINE:
2284	case SKD_DRVR_STATE_STOPPING:
2285	case SKD_DRVR_STATE_SYNCING:
2286	case SKD_DRVR_STATE_PAUSING:
2287	case SKD_DRVR_STATE_PAUSED:
2288	case SKD_DRVR_STATE_STARTING:
2289	case SKD_DRVR_STATE_RESTARTING:
2290	case SKD_DRVR_STATE_RESUMING:
2291	default:
2292		rc = -EINVAL;
2293		dev_dbg(&skdev->pdev->dev, "state [%d] not implemented\n",
2294			skdev->state);
2295	}
2296	return rc;
2297}
2298
2299/* assume spinlock is held */
2300static int skd_unquiesce_dev(struct skd_device *skdev)
2301{
2302	int prev_driver_state = skdev->state;
2303
2304	skd_log_skdev(skdev, "unquiesce");
2305	if (skdev->state == SKD_DRVR_STATE_ONLINE) {
2306		dev_dbg(&skdev->pdev->dev, "**** device already ONLINE\n");
2307		return 0;
2308	}
2309	if (skdev->drive_state != FIT_SR_DRIVE_ONLINE) {
2310		/*
2311		 * If there has been an state change to other than
2312		 * ONLINE, we will rely on controller state change
2313		 * to come back online and restart the queue.
2314		 * The BUSY state means that driver is ready to
2315		 * continue normal processing but waiting for controller
2316		 * to become available.
2317		 */
2318		skdev->state = SKD_DRVR_STATE_BUSY;
2319		dev_dbg(&skdev->pdev->dev, "drive BUSY state\n");
2320		return 0;
2321	}
2322
2323	/*
2324	 * Drive has just come online, driver is either in startup,
2325	 * paused performing a task, or bust waiting for hardware.
2326	 */
2327	switch (skdev->state) {
2328	case SKD_DRVR_STATE_PAUSED:
2329	case SKD_DRVR_STATE_BUSY:
2330	case SKD_DRVR_STATE_BUSY_IMMINENT:
2331	case SKD_DRVR_STATE_BUSY_ERASE:
2332	case SKD_DRVR_STATE_STARTING:
2333	case SKD_DRVR_STATE_RESTARTING:
2334	case SKD_DRVR_STATE_FAULT:
2335	case SKD_DRVR_STATE_IDLE:
2336	case SKD_DRVR_STATE_LOAD:
2337		skdev->state = SKD_DRVR_STATE_ONLINE;
2338		dev_err(&skdev->pdev->dev, "Driver state %s(%d)=>%s(%d)\n",
2339			skd_skdev_state_to_str(prev_driver_state),
2340			prev_driver_state, skd_skdev_state_to_str(skdev->state),
2341			skdev->state);
2342		dev_dbg(&skdev->pdev->dev,
2343			"**** device ONLINE...starting block queue\n");
2344		dev_dbg(&skdev->pdev->dev, "starting queue\n");
2345		dev_info(&skdev->pdev->dev, "STEC s1120 ONLINE\n");
2346		schedule_work(&skdev->start_queue);
2347		skdev->gendisk_on = 1;
2348		wake_up_interruptible(&skdev->waitq);
2349		break;
2350
2351	case SKD_DRVR_STATE_DISAPPEARED:
2352	default:
2353		dev_dbg(&skdev->pdev->dev,
2354			"**** driver state %d, not implemented\n",
2355			skdev->state);
2356		return -EBUSY;
2357	}
2358	return 0;
2359}
2360
2361/*
2362 *****************************************************************************
2363 * PCIe MSI/MSI-X INTERRUPT HANDLERS
2364 *****************************************************************************
2365 */
2366
2367static irqreturn_t skd_reserved_isr(int irq, void *skd_host_data)
2368{
2369	struct skd_device *skdev = skd_host_data;
2370	unsigned long flags;
2371
2372	spin_lock_irqsave(&skdev->lock, flags);
2373	dev_dbg(&skdev->pdev->dev, "MSIX = 0x%x\n",
2374		SKD_READL(skdev, FIT_INT_STATUS_HOST));
2375	dev_err(&skdev->pdev->dev, "MSIX reserved irq %d = 0x%x\n", irq,
2376		SKD_READL(skdev, FIT_INT_STATUS_HOST));
2377	SKD_WRITEL(skdev, FIT_INT_RESERVED_MASK, FIT_INT_STATUS_HOST);
2378	spin_unlock_irqrestore(&skdev->lock, flags);
2379	return IRQ_HANDLED;
2380}
2381
2382static irqreturn_t skd_statec_isr(int irq, void *skd_host_data)
2383{
2384	struct skd_device *skdev = skd_host_data;
2385	unsigned long flags;
2386
2387	spin_lock_irqsave(&skdev->lock, flags);
2388	dev_dbg(&skdev->pdev->dev, "MSIX = 0x%x\n",
2389		SKD_READL(skdev, FIT_INT_STATUS_HOST));
2390	SKD_WRITEL(skdev, FIT_ISH_FW_STATE_CHANGE, FIT_INT_STATUS_HOST);
2391	skd_isr_fwstate(skdev);
2392	spin_unlock_irqrestore(&skdev->lock, flags);
2393	return IRQ_HANDLED;
2394}
2395
2396static irqreturn_t skd_comp_q(int irq, void *skd_host_data)
2397{
2398	struct skd_device *skdev = skd_host_data;
2399	unsigned long flags;
2400	int flush_enqueued = 0;
2401	int deferred;
2402
2403	spin_lock_irqsave(&skdev->lock, flags);
2404	dev_dbg(&skdev->pdev->dev, "MSIX = 0x%x\n",
2405		SKD_READL(skdev, FIT_INT_STATUS_HOST));
2406	SKD_WRITEL(skdev, FIT_ISH_COMPLETION_POSTED, FIT_INT_STATUS_HOST);
2407	deferred = skd_isr_completion_posted(skdev, skd_isr_comp_limit,
2408						&flush_enqueued);
2409	if (flush_enqueued)
2410		schedule_work(&skdev->start_queue);
2411
2412	if (deferred)
2413		schedule_work(&skdev->completion_worker);
2414	else if (!flush_enqueued)
2415		schedule_work(&skdev->start_queue);
2416
2417	spin_unlock_irqrestore(&skdev->lock, flags);
2418
2419	return IRQ_HANDLED;
2420}
2421
2422static irqreturn_t skd_msg_isr(int irq, void *skd_host_data)
2423{
2424	struct skd_device *skdev = skd_host_data;
2425	unsigned long flags;
2426
2427	spin_lock_irqsave(&skdev->lock, flags);
2428	dev_dbg(&skdev->pdev->dev, "MSIX = 0x%x\n",
2429		SKD_READL(skdev, FIT_INT_STATUS_HOST));
2430	SKD_WRITEL(skdev, FIT_ISH_MSG_FROM_DEV, FIT_INT_STATUS_HOST);
2431	skd_isr_msg_from_dev(skdev);
2432	spin_unlock_irqrestore(&skdev->lock, flags);
2433	return IRQ_HANDLED;
2434}
2435
2436static irqreturn_t skd_qfull_isr(int irq, void *skd_host_data)
2437{
2438	struct skd_device *skdev = skd_host_data;
2439	unsigned long flags;
2440
2441	spin_lock_irqsave(&skdev->lock, flags);
2442	dev_dbg(&skdev->pdev->dev, "MSIX = 0x%x\n",
2443		SKD_READL(skdev, FIT_INT_STATUS_HOST));
2444	SKD_WRITEL(skdev, FIT_INT_QUEUE_FULL, FIT_INT_STATUS_HOST);
2445	spin_unlock_irqrestore(&skdev->lock, flags);
2446	return IRQ_HANDLED;
2447}
2448
2449/*
2450 *****************************************************************************
2451 * PCIe MSI/MSI-X SETUP
2452 *****************************************************************************
2453 */
2454
2455struct skd_msix_entry {
2456	char isr_name[30];
2457};
2458
2459struct skd_init_msix_entry {
2460	const char *name;
2461	irq_handler_t handler;
2462};
2463
2464#define SKD_MAX_MSIX_COUNT              13
2465#define SKD_MIN_MSIX_COUNT              7
2466#define SKD_BASE_MSIX_IRQ               4
2467
2468static struct skd_init_msix_entry msix_entries[SKD_MAX_MSIX_COUNT] = {
2469	{ "(DMA 0)",	    skd_reserved_isr },
2470	{ "(DMA 1)",	    skd_reserved_isr },
2471	{ "(DMA 2)",	    skd_reserved_isr },
2472	{ "(DMA 3)",	    skd_reserved_isr },
2473	{ "(State Change)", skd_statec_isr   },
2474	{ "(COMPL_Q)",	    skd_comp_q	     },
2475	{ "(MSG)",	    skd_msg_isr	     },
2476	{ "(Reserved)",	    skd_reserved_isr },
2477	{ "(Reserved)",	    skd_reserved_isr },
2478	{ "(Queue Full 0)", skd_qfull_isr    },
2479	{ "(Queue Full 1)", skd_qfull_isr    },
2480	{ "(Queue Full 2)", skd_qfull_isr    },
2481	{ "(Queue Full 3)", skd_qfull_isr    },
2482};
2483
2484static int skd_acquire_msix(struct skd_device *skdev)
2485{
2486	int i, rc;
2487	struct pci_dev *pdev = skdev->pdev;
2488
2489	rc = pci_alloc_irq_vectors(pdev, SKD_MAX_MSIX_COUNT, SKD_MAX_MSIX_COUNT,
2490			PCI_IRQ_MSIX);
2491	if (rc < 0) {
2492		dev_err(&skdev->pdev->dev, "failed to enable MSI-X %d\n", rc);
2493		goto out;
2494	}
2495
2496	skdev->msix_entries = kcalloc(SKD_MAX_MSIX_COUNT,
2497			sizeof(struct skd_msix_entry), GFP_KERNEL);
2498	if (!skdev->msix_entries) {
2499		rc = -ENOMEM;
2500		dev_err(&skdev->pdev->dev, "msix table allocation error\n");
2501		goto out;
2502	}
2503
2504	/* Enable MSI-X vectors for the base queue */
2505	for (i = 0; i < SKD_MAX_MSIX_COUNT; i++) {
2506		struct skd_msix_entry *qentry = &skdev->msix_entries[i];
2507
2508		snprintf(qentry->isr_name, sizeof(qentry->isr_name),
2509			 "%s%d-msix %s", DRV_NAME, skdev->devno,
2510			 msix_entries[i].name);
2511
2512		rc = devm_request_irq(&skdev->pdev->dev,
2513				pci_irq_vector(skdev->pdev, i),
2514				msix_entries[i].handler, 0,
2515				qentry->isr_name, skdev);
2516		if (rc) {
2517			dev_err(&skdev->pdev->dev,
2518				"Unable to register(%d) MSI-X handler %d: %s\n",
2519				rc, i, qentry->isr_name);
2520			goto msix_out;
2521		}
2522	}
2523
2524	dev_dbg(&skdev->pdev->dev, "%d msix irq(s) enabled\n",
2525		SKD_MAX_MSIX_COUNT);
2526	return 0;
2527
2528msix_out:
2529	while (--i >= 0)
2530		devm_free_irq(&pdev->dev, pci_irq_vector(pdev, i), skdev);
2531out:
2532	kfree(skdev->msix_entries);
2533	skdev->msix_entries = NULL;
2534	return rc;
2535}
2536
2537static int skd_acquire_irq(struct skd_device *skdev)
2538{
2539	struct pci_dev *pdev = skdev->pdev;
2540	unsigned int irq_flag = PCI_IRQ_LEGACY;
2541	int rc;
2542
2543	if (skd_isr_type == SKD_IRQ_MSIX) {
2544		rc = skd_acquire_msix(skdev);
2545		if (!rc)
2546			return 0;
2547
2548		dev_err(&skdev->pdev->dev,
2549			"failed to enable MSI-X, re-trying with MSI %d\n", rc);
2550	}
2551
2552	snprintf(skdev->isr_name, sizeof(skdev->isr_name), "%s%d", DRV_NAME,
2553			skdev->devno);
2554
2555	if (skd_isr_type != SKD_IRQ_LEGACY)
2556		irq_flag |= PCI_IRQ_MSI;
2557	rc = pci_alloc_irq_vectors(pdev, 1, 1, irq_flag);
2558	if (rc < 0) {
2559		dev_err(&skdev->pdev->dev,
2560			"failed to allocate the MSI interrupt %d\n", rc);
2561		return rc;
2562	}
2563
2564	rc = devm_request_irq(&pdev->dev, pdev->irq, skd_isr,
2565			pdev->msi_enabled ? 0 : IRQF_SHARED,
2566			skdev->isr_name, skdev);
2567	if (rc) {
2568		pci_free_irq_vectors(pdev);
2569		dev_err(&skdev->pdev->dev, "failed to allocate interrupt %d\n",
2570			rc);
2571		return rc;
2572	}
2573
2574	return 0;
2575}
2576
2577static void skd_release_irq(struct skd_device *skdev)
2578{
2579	struct pci_dev *pdev = skdev->pdev;
2580
2581	if (skdev->msix_entries) {
2582		int i;
2583
2584		for (i = 0; i < SKD_MAX_MSIX_COUNT; i++) {
2585			devm_free_irq(&pdev->dev, pci_irq_vector(pdev, i),
2586					skdev);
2587		}
2588
2589		kfree(skdev->msix_entries);
2590		skdev->msix_entries = NULL;
2591	} else {
2592		devm_free_irq(&pdev->dev, pdev->irq, skdev);
2593	}
2594
2595	pci_free_irq_vectors(pdev);
2596}
2597
2598/*
2599 *****************************************************************************
2600 * CONSTRUCT
2601 *****************************************************************************
2602 */
2603
2604static void *skd_alloc_dma(struct skd_device *skdev, struct kmem_cache *s,
2605			   dma_addr_t *dma_handle, gfp_t gfp,
2606			   enum dma_data_direction dir)
2607{
2608	struct device *dev = &skdev->pdev->dev;
2609	void *buf;
2610
2611	buf = kmem_cache_alloc(s, gfp);
2612	if (!buf)
2613		return NULL;
2614	*dma_handle = dma_map_single(dev, buf,
2615				     kmem_cache_size(s), dir);
2616	if (dma_mapping_error(dev, *dma_handle)) {
2617		kmem_cache_free(s, buf);
2618		buf = NULL;
2619	}
2620	return buf;
2621}
2622
2623static void skd_free_dma(struct skd_device *skdev, struct kmem_cache *s,
2624			 void *vaddr, dma_addr_t dma_handle,
2625			 enum dma_data_direction dir)
2626{
2627	if (!vaddr)
2628		return;
2629
2630	dma_unmap_single(&skdev->pdev->dev, dma_handle,
2631			 kmem_cache_size(s), dir);
2632	kmem_cache_free(s, vaddr);
2633}
2634
2635static int skd_cons_skcomp(struct skd_device *skdev)
2636{
2637	int rc = 0;
2638	struct fit_completion_entry_v1 *skcomp;
2639
2640	dev_dbg(&skdev->pdev->dev,
2641		"comp pci_alloc, total bytes %zd entries %d\n",
2642		SKD_SKCOMP_SIZE, SKD_N_COMPLETION_ENTRY);
2643
2644	skcomp = dma_alloc_coherent(&skdev->pdev->dev, SKD_SKCOMP_SIZE,
2645				    &skdev->cq_dma_address, GFP_KERNEL);
2646
2647	if (skcomp == NULL) {
2648		rc = -ENOMEM;
2649		goto err_out;
2650	}
2651
2652	skdev->skcomp_table = skcomp;
2653	skdev->skerr_table = (struct fit_comp_error_info *)((char *)skcomp +
2654							   sizeof(*skcomp) *
2655							   SKD_N_COMPLETION_ENTRY);
2656
2657err_out:
2658	return rc;
2659}
2660
2661static int skd_cons_skmsg(struct skd_device *skdev)
2662{
2663	int rc = 0;
2664	u32 i;
2665
2666	dev_dbg(&skdev->pdev->dev,
2667		"skmsg_table kcalloc, struct %lu, count %u total %lu\n",
2668		sizeof(struct skd_fitmsg_context), skdev->num_fitmsg_context,
2669		sizeof(struct skd_fitmsg_context) * skdev->num_fitmsg_context);
2670
2671	skdev->skmsg_table = kcalloc(skdev->num_fitmsg_context,
2672				     sizeof(struct skd_fitmsg_context),
2673				     GFP_KERNEL);
2674	if (skdev->skmsg_table == NULL) {
2675		rc = -ENOMEM;
2676		goto err_out;
2677	}
2678
2679	for (i = 0; i < skdev->num_fitmsg_context; i++) {
2680		struct skd_fitmsg_context *skmsg;
2681
2682		skmsg = &skdev->skmsg_table[i];
2683
2684		skmsg->id = i + SKD_ID_FIT_MSG;
2685
2686		skmsg->msg_buf = dma_alloc_coherent(&skdev->pdev->dev,
2687						    SKD_N_FITMSG_BYTES,
2688						    &skmsg->mb_dma_address,
2689						    GFP_KERNEL);
2690		if (skmsg->msg_buf == NULL) {
2691			rc = -ENOMEM;
2692			goto err_out;
2693		}
2694
2695		WARN(((uintptr_t)skmsg->msg_buf | skmsg->mb_dma_address) &
2696		     (FIT_QCMD_ALIGN - 1),
2697		     "not aligned: msg_buf %p mb_dma_address %pad\n",
2698		     skmsg->msg_buf, &skmsg->mb_dma_address);
2699	}
2700
2701err_out:
2702	return rc;
2703}
2704
2705static struct fit_sg_descriptor *skd_cons_sg_list(struct skd_device *skdev,
2706						  u32 n_sg,
2707						  dma_addr_t *ret_dma_addr)
2708{
2709	struct fit_sg_descriptor *sg_list;
2710
2711	sg_list = skd_alloc_dma(skdev, skdev->sglist_cache, ret_dma_addr,
2712				GFP_DMA | __GFP_ZERO, DMA_TO_DEVICE);
2713
2714	if (sg_list != NULL) {
2715		uint64_t dma_address = *ret_dma_addr;
2716		u32 i;
2717
2718		for (i = 0; i < n_sg - 1; i++) {
2719			uint64_t ndp_off;
2720			ndp_off = (i + 1) * sizeof(struct fit_sg_descriptor);
2721
2722			sg_list[i].next_desc_ptr = dma_address + ndp_off;
2723		}
2724		sg_list[i].next_desc_ptr = 0LL;
2725	}
2726
2727	return sg_list;
2728}
2729
2730static void skd_free_sg_list(struct skd_device *skdev,
2731			     struct fit_sg_descriptor *sg_list,
2732			     dma_addr_t dma_addr)
2733{
2734	if (WARN_ON_ONCE(!sg_list))
2735		return;
2736
2737	skd_free_dma(skdev, skdev->sglist_cache, sg_list, dma_addr,
2738		     DMA_TO_DEVICE);
2739}
2740
2741static int skd_init_request(struct blk_mq_tag_set *set, struct request *rq,
2742			    unsigned int hctx_idx, unsigned int numa_node)
2743{
2744	struct skd_device *skdev = set->driver_data;
2745	struct skd_request_context *skreq = blk_mq_rq_to_pdu(rq);
2746
2747	skreq->state = SKD_REQ_STATE_IDLE;
2748	skreq->sg = (void *)(skreq + 1);
2749	sg_init_table(skreq->sg, skd_sgs_per_request);
2750	skreq->sksg_list = skd_cons_sg_list(skdev, skd_sgs_per_request,
2751					    &skreq->sksg_dma_address);
2752
2753	return skreq->sksg_list ? 0 : -ENOMEM;
2754}
2755
2756static void skd_exit_request(struct blk_mq_tag_set *set, struct request *rq,
2757			     unsigned int hctx_idx)
2758{
2759	struct skd_device *skdev = set->driver_data;
2760	struct skd_request_context *skreq = blk_mq_rq_to_pdu(rq);
2761
2762	skd_free_sg_list(skdev, skreq->sksg_list, skreq->sksg_dma_address);
2763}
2764
2765static int skd_cons_sksb(struct skd_device *skdev)
2766{
2767	int rc = 0;
2768	struct skd_special_context *skspcl;
2769
2770	skspcl = &skdev->internal_skspcl;
2771
2772	skspcl->req.id = 0 + SKD_ID_INTERNAL;
2773	skspcl->req.state = SKD_REQ_STATE_IDLE;
2774
2775	skspcl->data_buf = skd_alloc_dma(skdev, skdev->databuf_cache,
2776					 &skspcl->db_dma_address,
2777					 GFP_DMA | __GFP_ZERO,
2778					 DMA_BIDIRECTIONAL);
2779	if (skspcl->data_buf == NULL) {
2780		rc = -ENOMEM;
2781		goto err_out;
2782	}
2783
2784	skspcl->msg_buf = skd_alloc_dma(skdev, skdev->msgbuf_cache,
2785					&skspcl->mb_dma_address,
2786					GFP_DMA | __GFP_ZERO, DMA_TO_DEVICE);
2787	if (skspcl->msg_buf == NULL) {
2788		rc = -ENOMEM;
2789		goto err_out;
2790	}
2791
2792	skspcl->req.sksg_list = skd_cons_sg_list(skdev, 1,
2793						 &skspcl->req.sksg_dma_address);
2794	if (skspcl->req.sksg_list == NULL) {
2795		rc = -ENOMEM;
2796		goto err_out;
2797	}
2798
2799	if (!skd_format_internal_skspcl(skdev)) {
2800		rc = -EINVAL;
2801		goto err_out;
2802	}
2803
2804err_out:
2805	return rc;
2806}
2807
2808static const struct blk_mq_ops skd_mq_ops = {
2809	.queue_rq	= skd_mq_queue_rq,
2810	.complete	= skd_complete_rq,
2811	.timeout	= skd_timed_out,
2812	.init_request	= skd_init_request,
2813	.exit_request	= skd_exit_request,
2814};
2815
2816static int skd_cons_disk(struct skd_device *skdev)
2817{
2818	int rc = 0;
2819	struct gendisk *disk;
2820	struct request_queue *q;
2821	unsigned long flags;
2822
2823	disk = alloc_disk(SKD_MINORS_PER_DEVICE);
2824	if (!disk) {
2825		rc = -ENOMEM;
2826		goto err_out;
2827	}
2828
2829	skdev->disk = disk;
2830	sprintf(disk->disk_name, DRV_NAME "%u", skdev->devno);
2831
2832	disk->major = skdev->major;
2833	disk->first_minor = skdev->devno * SKD_MINORS_PER_DEVICE;
2834	disk->fops = &skd_blockdev_ops;
2835	disk->private_data = skdev;
2836
2837	memset(&skdev->tag_set, 0, sizeof(skdev->tag_set));
2838	skdev->tag_set.ops = &skd_mq_ops;
2839	skdev->tag_set.nr_hw_queues = 1;
2840	skdev->tag_set.queue_depth = skd_max_queue_depth;
2841	skdev->tag_set.cmd_size = sizeof(struct skd_request_context) +
2842		skdev->sgs_per_request * sizeof(struct scatterlist);
2843	skdev->tag_set.numa_node = NUMA_NO_NODE;
2844	skdev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
2845		BLK_ALLOC_POLICY_TO_MQ_FLAG(BLK_TAG_ALLOC_FIFO);
2846	skdev->tag_set.driver_data = skdev;
2847	rc = blk_mq_alloc_tag_set(&skdev->tag_set);
2848	if (rc)
2849		goto err_out;
2850	q = blk_mq_init_queue(&skdev->tag_set);
2851	if (IS_ERR(q)) {
2852		blk_mq_free_tag_set(&skdev->tag_set);
2853		rc = PTR_ERR(q);
2854		goto err_out;
2855	}
2856	q->queuedata = skdev;
2857
2858	skdev->queue = q;
2859	disk->queue = q;
2860
2861	blk_queue_write_cache(q, true, true);
2862	blk_queue_max_segments(q, skdev->sgs_per_request);
2863	blk_queue_max_hw_sectors(q, SKD_N_MAX_SECTORS);
2864
2865	/* set optimal I/O size to 8KB */
2866	blk_queue_io_opt(q, 8192);
2867
2868	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2869	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2870
2871	blk_queue_rq_timeout(q, 8 * HZ);
2872
2873	spin_lock_irqsave(&skdev->lock, flags);
2874	dev_dbg(&skdev->pdev->dev, "stopping queue\n");
2875	blk_mq_stop_hw_queues(skdev->queue);
2876	spin_unlock_irqrestore(&skdev->lock, flags);
2877
2878err_out:
2879	return rc;
2880}
2881
2882#define SKD_N_DEV_TABLE         16u
2883static u32 skd_next_devno;
2884
2885static struct skd_device *skd_construct(struct pci_dev *pdev)
2886{
2887	struct skd_device *skdev;
2888	int blk_major = skd_major;
2889	size_t size;
2890	int rc;
2891
2892	skdev = kzalloc(sizeof(*skdev), GFP_KERNEL);
2893
2894	if (!skdev) {
2895		dev_err(&pdev->dev, "memory alloc failure\n");
2896		return NULL;
2897	}
2898
2899	skdev->state = SKD_DRVR_STATE_LOAD;
2900	skdev->pdev = pdev;
2901	skdev->devno = skd_next_devno++;
2902	skdev->major = blk_major;
2903	skdev->dev_max_queue_depth = 0;
2904
2905	skdev->num_req_context = skd_max_queue_depth;
2906	skdev->num_fitmsg_context = skd_max_queue_depth;
2907	skdev->cur_max_queue_depth = 1;
2908	skdev->queue_low_water_mark = 1;
2909	skdev->proto_ver = 99;
2910	skdev->sgs_per_request = skd_sgs_per_request;
2911	skdev->dbg_level = skd_dbg_level;
2912
2913	spin_lock_init(&skdev->lock);
2914
2915	INIT_WORK(&skdev->start_queue, skd_start_queue);
2916	INIT_WORK(&skdev->completion_worker, skd_completion_worker);
2917
2918	size = max(SKD_N_FITMSG_BYTES, SKD_N_SPECIAL_FITMSG_BYTES);
2919	skdev->msgbuf_cache = kmem_cache_create("skd-msgbuf", size, 0,
2920						SLAB_HWCACHE_ALIGN, NULL);
2921	if (!skdev->msgbuf_cache)
2922		goto err_out;
2923	WARN_ONCE(kmem_cache_size(skdev->msgbuf_cache) < size,
2924		  "skd-msgbuf: %d < %zd\n",
2925		  kmem_cache_size(skdev->msgbuf_cache), size);
2926	size = skd_sgs_per_request * sizeof(struct fit_sg_descriptor);
2927	skdev->sglist_cache = kmem_cache_create("skd-sglist", size, 0,
2928						SLAB_HWCACHE_ALIGN, NULL);
2929	if (!skdev->sglist_cache)
2930		goto err_out;
2931	WARN_ONCE(kmem_cache_size(skdev->sglist_cache) < size,
2932		  "skd-sglist: %d < %zd\n",
2933		  kmem_cache_size(skdev->sglist_cache), size);
2934	size = SKD_N_INTERNAL_BYTES;
2935	skdev->databuf_cache = kmem_cache_create("skd-databuf", size, 0,
2936						 SLAB_HWCACHE_ALIGN, NULL);
2937	if (!skdev->databuf_cache)
2938		goto err_out;
2939	WARN_ONCE(kmem_cache_size(skdev->databuf_cache) < size,
2940		  "skd-databuf: %d < %zd\n",
2941		  kmem_cache_size(skdev->databuf_cache), size);
2942
2943	dev_dbg(&skdev->pdev->dev, "skcomp\n");
2944	rc = skd_cons_skcomp(skdev);
2945	if (rc < 0)
2946		goto err_out;
2947
2948	dev_dbg(&skdev->pdev->dev, "skmsg\n");
2949	rc = skd_cons_skmsg(skdev);
2950	if (rc < 0)
2951		goto err_out;
2952
2953	dev_dbg(&skdev->pdev->dev, "sksb\n");
2954	rc = skd_cons_sksb(skdev);
2955	if (rc < 0)
2956		goto err_out;
2957
2958	dev_dbg(&skdev->pdev->dev, "disk\n");
2959	rc = skd_cons_disk(skdev);
2960	if (rc < 0)
2961		goto err_out;
2962
2963	dev_dbg(&skdev->pdev->dev, "VICTORY\n");
2964	return skdev;
2965
2966err_out:
2967	dev_dbg(&skdev->pdev->dev, "construct failed\n");
2968	skd_destruct(skdev);
2969	return NULL;
2970}
2971
2972/*
2973 *****************************************************************************
2974 * DESTRUCT (FREE)
2975 *****************************************************************************
2976 */
2977
2978static void skd_free_skcomp(struct skd_device *skdev)
2979{
2980	if (skdev->skcomp_table)
2981		dma_free_coherent(&skdev->pdev->dev, SKD_SKCOMP_SIZE,
2982				  skdev->skcomp_table, skdev->cq_dma_address);
2983
2984	skdev->skcomp_table = NULL;
2985	skdev->cq_dma_address = 0;
2986}
2987
2988static void skd_free_skmsg(struct skd_device *skdev)
2989{
2990	u32 i;
2991
2992	if (skdev->skmsg_table == NULL)
2993		return;
2994
2995	for (i = 0; i < skdev->num_fitmsg_context; i++) {
2996		struct skd_fitmsg_context *skmsg;
2997
2998		skmsg = &skdev->skmsg_table[i];
2999
3000		if (skmsg->msg_buf != NULL) {
3001			dma_free_coherent(&skdev->pdev->dev, SKD_N_FITMSG_BYTES,
3002					  skmsg->msg_buf,
3003					    skmsg->mb_dma_address);
3004		}
3005		skmsg->msg_buf = NULL;
3006		skmsg->mb_dma_address = 0;
3007	}
3008
3009	kfree(skdev->skmsg_table);
3010	skdev->skmsg_table = NULL;
3011}
3012
3013static void skd_free_sksb(struct skd_device *skdev)
3014{
3015	struct skd_special_context *skspcl = &skdev->internal_skspcl;
3016
3017	skd_free_dma(skdev, skdev->databuf_cache, skspcl->data_buf,
3018		     skspcl->db_dma_address, DMA_BIDIRECTIONAL);
3019
3020	skspcl->data_buf = NULL;
3021	skspcl->db_dma_address = 0;
3022
3023	skd_free_dma(skdev, skdev->msgbuf_cache, skspcl->msg_buf,
3024		     skspcl->mb_dma_address, DMA_TO_DEVICE);
3025
3026	skspcl->msg_buf = NULL;
3027	skspcl->mb_dma_address = 0;
3028
3029	skd_free_sg_list(skdev, skspcl->req.sksg_list,
3030			 skspcl->req.sksg_dma_address);
3031
3032	skspcl->req.sksg_list = NULL;
3033	skspcl->req.sksg_dma_address = 0;
3034}
3035
3036static void skd_free_disk(struct skd_device *skdev)
3037{
3038	struct gendisk *disk = skdev->disk;
3039
3040	if (disk && (disk->flags & GENHD_FL_UP))
3041		del_gendisk(disk);
3042
3043	if (skdev->queue) {
3044		blk_cleanup_queue(skdev->queue);
3045		skdev->queue = NULL;
3046		if (disk)
3047			disk->queue = NULL;
3048	}
3049
3050	if (skdev->tag_set.tags)
3051		blk_mq_free_tag_set(&skdev->tag_set);
3052
3053	put_disk(disk);
3054	skdev->disk = NULL;
3055}
3056
3057static void skd_destruct(struct skd_device *skdev)
3058{
3059	if (skdev == NULL)
3060		return;
3061
3062	cancel_work_sync(&skdev->start_queue);
3063
3064	dev_dbg(&skdev->pdev->dev, "disk\n");
3065	skd_free_disk(skdev);
3066
3067	dev_dbg(&skdev->pdev->dev, "sksb\n");
3068	skd_free_sksb(skdev);
3069
3070	dev_dbg(&skdev->pdev->dev, "skmsg\n");
3071	skd_free_skmsg(skdev);
3072
3073	dev_dbg(&skdev->pdev->dev, "skcomp\n");
3074	skd_free_skcomp(skdev);
3075
3076	kmem_cache_destroy(skdev->databuf_cache);
3077	kmem_cache_destroy(skdev->sglist_cache);
3078	kmem_cache_destroy(skdev->msgbuf_cache);
3079
3080	dev_dbg(&skdev->pdev->dev, "skdev\n");
3081	kfree(skdev);
3082}
3083
3084/*
3085 *****************************************************************************
3086 * BLOCK DEVICE (BDEV) GLUE
3087 *****************************************************************************
3088 */
3089
3090static int skd_bdev_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3091{
3092	struct skd_device *skdev;
3093	u64 capacity;
3094
3095	skdev = bdev->bd_disk->private_data;
3096
3097	dev_dbg(&skdev->pdev->dev, "%s: CMD[%s] getgeo device\n",
3098		bdev->bd_disk->disk_name, current->comm);
3099
3100	if (skdev->read_cap_is_valid) {
3101		capacity = get_capacity(skdev->disk);
3102		geo->heads = 64;
3103		geo->sectors = 255;
3104		geo->cylinders = (capacity) / (255 * 64);
3105
3106		return 0;
3107	}
3108	return -EIO;
3109}
3110
3111static int skd_bdev_attach(struct device *parent, struct skd_device *skdev)
3112{
3113	dev_dbg(&skdev->pdev->dev, "add_disk\n");
3114	device_add_disk(parent, skdev->disk, NULL);
3115	return 0;
3116}
3117
3118static const struct block_device_operations skd_blockdev_ops = {
3119	.owner		= THIS_MODULE,
3120	.getgeo		= skd_bdev_getgeo,
3121};
3122
3123/*
3124 *****************************************************************************
3125 * PCIe DRIVER GLUE
3126 *****************************************************************************
3127 */
3128
3129static const struct pci_device_id skd_pci_tbl[] = {
3130	{ PCI_VENDOR_ID_STEC, PCI_DEVICE_ID_S1120,
3131	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, },
3132	{ 0 }                     /* terminate list */
3133};
3134
3135MODULE_DEVICE_TABLE(pci, skd_pci_tbl);
3136
3137static char *skd_pci_info(struct skd_device *skdev, char *str)
3138{
3139	int pcie_reg;
3140
3141	strcpy(str, "PCIe (");
3142	pcie_reg = pci_find_capability(skdev->pdev, PCI_CAP_ID_EXP);
3143
3144	if (pcie_reg) {
3145
3146		char lwstr[6];
3147		uint16_t pcie_lstat, lspeed, lwidth;
3148
3149		pcie_reg += 0x12;
3150		pci_read_config_word(skdev->pdev, pcie_reg, &pcie_lstat);
3151		lspeed = pcie_lstat & (0xF);
3152		lwidth = (pcie_lstat & 0x3F0) >> 4;
3153
3154		if (lspeed == 1)
3155			strcat(str, "2.5GT/s ");
3156		else if (lspeed == 2)
3157			strcat(str, "5.0GT/s ");
3158		else
3159			strcat(str, "<unknown> ");
3160		snprintf(lwstr, sizeof(lwstr), "%dX)", lwidth);
3161		strcat(str, lwstr);
3162	}
3163	return str;
3164}
3165
3166static int skd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3167{
3168	int i;
3169	int rc = 0;
3170	char pci_str[32];
3171	struct skd_device *skdev;
3172
3173	dev_dbg(&pdev->dev, "vendor=%04X device=%04x\n", pdev->vendor,
3174		pdev->device);
3175
3176	rc = pci_enable_device(pdev);
3177	if (rc)
3178		return rc;
3179	rc = pci_request_regions(pdev, DRV_NAME);
3180	if (rc)
3181		goto err_out;
3182	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3183	if (rc)
3184		rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3185	if (rc) {
3186		dev_err(&pdev->dev, "DMA mask error %d\n", rc);
3187		goto err_out_regions;
3188	}
3189
3190	if (!skd_major) {
3191		rc = register_blkdev(0, DRV_NAME);
3192		if (rc < 0)
3193			goto err_out_regions;
3194		BUG_ON(!rc);
3195		skd_major = rc;
3196	}
3197
3198	skdev = skd_construct(pdev);
3199	if (skdev == NULL) {
3200		rc = -ENOMEM;
3201		goto err_out_regions;
3202	}
3203
3204	skd_pci_info(skdev, pci_str);
3205	dev_info(&pdev->dev, "%s 64bit\n", pci_str);
3206
3207	pci_set_master(pdev);
3208	rc = pci_enable_pcie_error_reporting(pdev);
3209	if (rc) {
3210		dev_err(&pdev->dev,
3211			"bad enable of PCIe error reporting rc=%d\n", rc);
3212		skdev->pcie_error_reporting_is_enabled = 0;
3213	} else
3214		skdev->pcie_error_reporting_is_enabled = 1;
3215
3216	pci_set_drvdata(pdev, skdev);
3217
3218	for (i = 0; i < SKD_MAX_BARS; i++) {
3219		skdev->mem_phys[i] = pci_resource_start(pdev, i);
3220		skdev->mem_size[i] = (u32)pci_resource_len(pdev, i);
3221		skdev->mem_map[i] = ioremap(skdev->mem_phys[i],
3222					    skdev->mem_size[i]);
3223		if (!skdev->mem_map[i]) {
3224			dev_err(&pdev->dev,
3225				"Unable to map adapter memory!\n");
3226			rc = -ENODEV;
3227			goto err_out_iounmap;
3228		}
3229		dev_dbg(&pdev->dev, "mem_map=%p, phyd=%016llx, size=%d\n",
3230			skdev->mem_map[i], (uint64_t)skdev->mem_phys[i],
3231			skdev->mem_size[i]);
3232	}
3233
3234	rc = skd_acquire_irq(skdev);
3235	if (rc) {
3236		dev_err(&pdev->dev, "interrupt resource error %d\n", rc);
3237		goto err_out_iounmap;
3238	}
3239
3240	rc = skd_start_timer(skdev);
3241	if (rc)
3242		goto err_out_timer;
3243
3244	init_waitqueue_head(&skdev->waitq);
3245
3246	skd_start_device(skdev);
3247
3248	rc = wait_event_interruptible_timeout(skdev->waitq,
3249					      (skdev->gendisk_on),
3250					      (SKD_START_WAIT_SECONDS * HZ));
3251	if (skdev->gendisk_on > 0) {
3252		/* device came on-line after reset */
3253		skd_bdev_attach(&pdev->dev, skdev);
3254		rc = 0;
3255	} else {
3256		/* we timed out, something is wrong with the device,
3257		   don't add the disk structure */
3258		dev_err(&pdev->dev, "error: waiting for s1120 timed out %d!\n",
3259			rc);
3260		/* in case of no error; we timeout with ENXIO */
3261		if (!rc)
3262			rc = -ENXIO;
3263		goto err_out_timer;
3264	}
3265
3266	return rc;
3267
3268err_out_timer:
3269	skd_stop_device(skdev);
3270	skd_release_irq(skdev);
3271
3272err_out_iounmap:
3273	for (i = 0; i < SKD_MAX_BARS; i++)
3274		if (skdev->mem_map[i])
3275			iounmap(skdev->mem_map[i]);
3276
3277	if (skdev->pcie_error_reporting_is_enabled)
3278		pci_disable_pcie_error_reporting(pdev);
3279
3280	skd_destruct(skdev);
3281
3282err_out_regions:
3283	pci_release_regions(pdev);
3284
3285err_out:
3286	pci_disable_device(pdev);
3287	pci_set_drvdata(pdev, NULL);
3288	return rc;
3289}
3290
3291static void skd_pci_remove(struct pci_dev *pdev)
3292{
3293	int i;
3294	struct skd_device *skdev;
3295
3296	skdev = pci_get_drvdata(pdev);
3297	if (!skdev) {
3298		dev_err(&pdev->dev, "no device data for PCI\n");
3299		return;
3300	}
3301	skd_stop_device(skdev);
3302	skd_release_irq(skdev);
3303
3304	for (i = 0; i < SKD_MAX_BARS; i++)
3305		if (skdev->mem_map[i])
3306			iounmap(skdev->mem_map[i]);
3307
3308	if (skdev->pcie_error_reporting_is_enabled)
3309		pci_disable_pcie_error_reporting(pdev);
3310
3311	skd_destruct(skdev);
3312
3313	pci_release_regions(pdev);
3314	pci_disable_device(pdev);
3315	pci_set_drvdata(pdev, NULL);
3316
3317	return;
3318}
3319
3320static int skd_pci_suspend(struct pci_dev *pdev, pm_message_t state)
3321{
3322	int i;
3323	struct skd_device *skdev;
3324
3325	skdev = pci_get_drvdata(pdev);
3326	if (!skdev) {
3327		dev_err(&pdev->dev, "no device data for PCI\n");
3328		return -EIO;
3329	}
3330
3331	skd_stop_device(skdev);
3332
3333	skd_release_irq(skdev);
3334
3335	for (i = 0; i < SKD_MAX_BARS; i++)
3336		if (skdev->mem_map[i])
3337			iounmap(skdev->mem_map[i]);
3338
3339	if (skdev->pcie_error_reporting_is_enabled)
3340		pci_disable_pcie_error_reporting(pdev);
3341
3342	pci_release_regions(pdev);
3343	pci_save_state(pdev);
3344	pci_disable_device(pdev);
3345	pci_set_power_state(pdev, pci_choose_state(pdev, state));
3346	return 0;
3347}
3348
3349static int skd_pci_resume(struct pci_dev *pdev)
3350{
3351	int i;
3352	int rc = 0;
3353	struct skd_device *skdev;
3354
3355	skdev = pci_get_drvdata(pdev);
3356	if (!skdev) {
3357		dev_err(&pdev->dev, "no device data for PCI\n");
3358		return -1;
3359	}
3360
3361	pci_set_power_state(pdev, PCI_D0);
3362	pci_enable_wake(pdev, PCI_D0, 0);
3363	pci_restore_state(pdev);
3364
3365	rc = pci_enable_device(pdev);
3366	if (rc)
3367		return rc;
3368	rc = pci_request_regions(pdev, DRV_NAME);
3369	if (rc)
3370		goto err_out;
3371	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3372	if (rc)
3373		rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3374	if (rc) {
3375		dev_err(&pdev->dev, "DMA mask error %d\n", rc);
3376		goto err_out_regions;
3377	}
3378
3379	pci_set_master(pdev);
3380	rc = pci_enable_pcie_error_reporting(pdev);
3381	if (rc) {
3382		dev_err(&pdev->dev,
3383			"bad enable of PCIe error reporting rc=%d\n", rc);
3384		skdev->pcie_error_reporting_is_enabled = 0;
3385	} else
3386		skdev->pcie_error_reporting_is_enabled = 1;
3387
3388	for (i = 0; i < SKD_MAX_BARS; i++) {
3389
3390		skdev->mem_phys[i] = pci_resource_start(pdev, i);
3391		skdev->mem_size[i] = (u32)pci_resource_len(pdev, i);
3392		skdev->mem_map[i] = ioremap(skdev->mem_phys[i],
3393					    skdev->mem_size[i]);
3394		if (!skdev->mem_map[i]) {
3395			dev_err(&pdev->dev, "Unable to map adapter memory!\n");
3396			rc = -ENODEV;
3397			goto err_out_iounmap;
3398		}
3399		dev_dbg(&pdev->dev, "mem_map=%p, phyd=%016llx, size=%d\n",
3400			skdev->mem_map[i], (uint64_t)skdev->mem_phys[i],
3401			skdev->mem_size[i]);
3402	}
3403	rc = skd_acquire_irq(skdev);
3404	if (rc) {
3405		dev_err(&pdev->dev, "interrupt resource error %d\n", rc);
3406		goto err_out_iounmap;
3407	}
3408
3409	rc = skd_start_timer(skdev);
3410	if (rc)
3411		goto err_out_timer;
3412
3413	init_waitqueue_head(&skdev->waitq);
3414
3415	skd_start_device(skdev);
3416
3417	return rc;
3418
3419err_out_timer:
3420	skd_stop_device(skdev);
3421	skd_release_irq(skdev);
3422
3423err_out_iounmap:
3424	for (i = 0; i < SKD_MAX_BARS; i++)
3425		if (skdev->mem_map[i])
3426			iounmap(skdev->mem_map[i]);
3427
3428	if (skdev->pcie_error_reporting_is_enabled)
3429		pci_disable_pcie_error_reporting(pdev);
3430
3431err_out_regions:
3432	pci_release_regions(pdev);
3433
3434err_out:
3435	pci_disable_device(pdev);
3436	return rc;
3437}
3438
3439static void skd_pci_shutdown(struct pci_dev *pdev)
3440{
3441	struct skd_device *skdev;
3442
3443	dev_err(&pdev->dev, "%s called\n", __func__);
3444
3445	skdev = pci_get_drvdata(pdev);
3446	if (!skdev) {
3447		dev_err(&pdev->dev, "no device data for PCI\n");
3448		return;
3449	}
3450
3451	dev_err(&pdev->dev, "calling stop\n");
3452	skd_stop_device(skdev);
3453}
3454
3455static struct pci_driver skd_driver = {
3456	.name		= DRV_NAME,
3457	.id_table	= skd_pci_tbl,
3458	.probe		= skd_pci_probe,
3459	.remove		= skd_pci_remove,
3460	.suspend	= skd_pci_suspend,
3461	.resume		= skd_pci_resume,
3462	.shutdown	= skd_pci_shutdown,
3463};
3464
3465/*
3466 *****************************************************************************
3467 * LOGGING SUPPORT
3468 *****************************************************************************
3469 */
3470
3471const char *skd_drive_state_to_str(int state)
3472{
3473	switch (state) {
3474	case FIT_SR_DRIVE_OFFLINE:
3475		return "OFFLINE";
3476	case FIT_SR_DRIVE_INIT:
3477		return "INIT";
3478	case FIT_SR_DRIVE_ONLINE:
3479		return "ONLINE";
3480	case FIT_SR_DRIVE_BUSY:
3481		return "BUSY";
3482	case FIT_SR_DRIVE_FAULT:
3483		return "FAULT";
3484	case FIT_SR_DRIVE_DEGRADED:
3485		return "DEGRADED";
3486	case FIT_SR_PCIE_LINK_DOWN:
3487		return "INK_DOWN";
3488	case FIT_SR_DRIVE_SOFT_RESET:
3489		return "SOFT_RESET";
3490	case FIT_SR_DRIVE_NEED_FW_DOWNLOAD:
3491		return "NEED_FW";
3492	case FIT_SR_DRIVE_INIT_FAULT:
3493		return "INIT_FAULT";
3494	case FIT_SR_DRIVE_BUSY_SANITIZE:
3495		return "BUSY_SANITIZE";
3496	case FIT_SR_DRIVE_BUSY_ERASE:
3497		return "BUSY_ERASE";
3498	case FIT_SR_DRIVE_FW_BOOTING:
3499		return "FW_BOOTING";
3500	default:
3501		return "???";
3502	}
3503}
3504
3505const char *skd_skdev_state_to_str(enum skd_drvr_state state)
3506{
3507	switch (state) {
3508	case SKD_DRVR_STATE_LOAD:
3509		return "LOAD";
3510	case SKD_DRVR_STATE_IDLE:
3511		return "IDLE";
3512	case SKD_DRVR_STATE_BUSY:
3513		return "BUSY";
3514	case SKD_DRVR_STATE_STARTING:
3515		return "STARTING";
3516	case SKD_DRVR_STATE_ONLINE:
3517		return "ONLINE";
3518	case SKD_DRVR_STATE_PAUSING:
3519		return "PAUSING";
3520	case SKD_DRVR_STATE_PAUSED:
3521		return "PAUSED";
3522	case SKD_DRVR_STATE_RESTARTING:
3523		return "RESTARTING";
3524	case SKD_DRVR_STATE_RESUMING:
3525		return "RESUMING";
3526	case SKD_DRVR_STATE_STOPPING:
3527		return "STOPPING";
3528	case SKD_DRVR_STATE_SYNCING:
3529		return "SYNCING";
3530	case SKD_DRVR_STATE_FAULT:
3531		return "FAULT";
3532	case SKD_DRVR_STATE_DISAPPEARED:
3533		return "DISAPPEARED";
3534	case SKD_DRVR_STATE_BUSY_ERASE:
3535		return "BUSY_ERASE";
3536	case SKD_DRVR_STATE_BUSY_SANITIZE:
3537		return "BUSY_SANITIZE";
3538	case SKD_DRVR_STATE_BUSY_IMMINENT:
3539		return "BUSY_IMMINENT";
3540	case SKD_DRVR_STATE_WAIT_BOOT:
3541		return "WAIT_BOOT";
3542
3543	default:
3544		return "???";
3545	}
3546}
3547
3548static const char *skd_skreq_state_to_str(enum skd_req_state state)
3549{
3550	switch (state) {
3551	case SKD_REQ_STATE_IDLE:
3552		return "IDLE";
3553	case SKD_REQ_STATE_SETUP:
3554		return "SETUP";
3555	case SKD_REQ_STATE_BUSY:
3556		return "BUSY";
3557	case SKD_REQ_STATE_COMPLETED:
3558		return "COMPLETED";
3559	case SKD_REQ_STATE_TIMEOUT:
3560		return "TIMEOUT";
3561	default:
3562		return "???";
3563	}
3564}
3565
3566static void skd_log_skdev(struct skd_device *skdev, const char *event)
3567{
3568	dev_dbg(&skdev->pdev->dev, "skdev=%p event='%s'\n", skdev, event);
3569	dev_dbg(&skdev->pdev->dev, "  drive_state=%s(%d) driver_state=%s(%d)\n",
3570		skd_drive_state_to_str(skdev->drive_state), skdev->drive_state,
3571		skd_skdev_state_to_str(skdev->state), skdev->state);
3572	dev_dbg(&skdev->pdev->dev, "  busy=%d limit=%d dev=%d lowat=%d\n",
3573		skd_in_flight(skdev), skdev->cur_max_queue_depth,
3574		skdev->dev_max_queue_depth, skdev->queue_low_water_mark);
3575	dev_dbg(&skdev->pdev->dev, "  cycle=%d cycle_ix=%d\n",
3576		skdev->skcomp_cycle, skdev->skcomp_ix);
3577}
3578
3579static void skd_log_skreq(struct skd_device *skdev,
3580			  struct skd_request_context *skreq, const char *event)
3581{
3582	struct request *req = blk_mq_rq_from_pdu(skreq);
3583	u32 lba = blk_rq_pos(req);
3584	u32 count = blk_rq_sectors(req);
3585
3586	dev_dbg(&skdev->pdev->dev, "skreq=%p event='%s'\n", skreq, event);
3587	dev_dbg(&skdev->pdev->dev, "  state=%s(%d) id=0x%04x fitmsg=0x%04x\n",
3588		skd_skreq_state_to_str(skreq->state), skreq->state, skreq->id,
3589		skreq->fitmsg_id);
3590	dev_dbg(&skdev->pdev->dev, "  sg_dir=%d n_sg=%d\n",
3591		skreq->data_dir, skreq->n_sg);
3592
3593	dev_dbg(&skdev->pdev->dev,
3594		"req=%p lba=%u(0x%x) count=%u(0x%x) dir=%d\n", req, lba, lba,
3595		count, count, (int)rq_data_dir(req));
3596}
3597
3598/*
3599 *****************************************************************************
3600 * MODULE GLUE
3601 *****************************************************************************
3602 */
3603
3604static int __init skd_init(void)
3605{
3606	BUILD_BUG_ON(sizeof(struct fit_completion_entry_v1) != 8);
3607	BUILD_BUG_ON(sizeof(struct fit_comp_error_info) != 32);
3608	BUILD_BUG_ON(sizeof(struct skd_command_header) != 16);
3609	BUILD_BUG_ON(sizeof(struct skd_scsi_request) != 32);
3610	BUILD_BUG_ON(sizeof(struct driver_inquiry_data) != 44);
3611	BUILD_BUG_ON(offsetof(struct skd_msg_buf, fmh) != 0);
3612	BUILD_BUG_ON(offsetof(struct skd_msg_buf, scsi) != 64);
3613	BUILD_BUG_ON(sizeof(struct skd_msg_buf) != SKD_N_FITMSG_BYTES);
3614
3615	switch (skd_isr_type) {
3616	case SKD_IRQ_LEGACY:
3617	case SKD_IRQ_MSI:
3618	case SKD_IRQ_MSIX:
3619		break;
3620	default:
3621		pr_err(PFX "skd_isr_type %d invalid, re-set to %d\n",
3622		       skd_isr_type, SKD_IRQ_DEFAULT);
3623		skd_isr_type = SKD_IRQ_DEFAULT;
3624	}
3625
3626	if (skd_max_queue_depth < 1 ||
3627	    skd_max_queue_depth > SKD_MAX_QUEUE_DEPTH) {
3628		pr_err(PFX "skd_max_queue_depth %d invalid, re-set to %d\n",
3629		       skd_max_queue_depth, SKD_MAX_QUEUE_DEPTH_DEFAULT);
3630		skd_max_queue_depth = SKD_MAX_QUEUE_DEPTH_DEFAULT;
3631	}
3632
3633	if (skd_max_req_per_msg < 1 ||
3634	    skd_max_req_per_msg > SKD_MAX_REQ_PER_MSG) {
3635		pr_err(PFX "skd_max_req_per_msg %d invalid, re-set to %d\n",
3636		       skd_max_req_per_msg, SKD_MAX_REQ_PER_MSG_DEFAULT);
3637		skd_max_req_per_msg = SKD_MAX_REQ_PER_MSG_DEFAULT;
3638	}
3639
3640	if (skd_sgs_per_request < 1 || skd_sgs_per_request > 4096) {
3641		pr_err(PFX "skd_sg_per_request %d invalid, re-set to %d\n",
3642		       skd_sgs_per_request, SKD_N_SG_PER_REQ_DEFAULT);
3643		skd_sgs_per_request = SKD_N_SG_PER_REQ_DEFAULT;
3644	}
3645
3646	if (skd_dbg_level < 0 || skd_dbg_level > 2) {
3647		pr_err(PFX "skd_dbg_level %d invalid, re-set to %d\n",
3648		       skd_dbg_level, 0);
3649		skd_dbg_level = 0;
3650	}
3651
3652	if (skd_isr_comp_limit < 0) {
3653		pr_err(PFX "skd_isr_comp_limit %d invalid, set to %d\n",
3654		       skd_isr_comp_limit, 0);
3655		skd_isr_comp_limit = 0;
3656	}
3657
3658	return pci_register_driver(&skd_driver);
3659}
3660
3661static void __exit skd_exit(void)
3662{
3663	pci_unregister_driver(&skd_driver);
3664
3665	if (skd_major)
3666		unregister_blkdev(skd_major, DRV_NAME);
3667}
3668
3669module_init(skd_init);
3670module_exit(skd_exit);
3671