1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright 2016 Broadcom
4 */
5
6/*
7 * Broadcom PDC Mailbox Driver
8 * The PDC provides a ring based programming interface to one or more hardware
9 * offload engines. For example, the PDC driver works with both SPU-M and SPU2
10 * cryptographic offload hardware. In some chips the PDC is referred to as MDE,
11 * and in others the FA2/FA+ hardware is used with this PDC driver.
12 *
13 * The PDC driver registers with the Linux mailbox framework as a mailbox
14 * controller, once for each PDC instance. Ring 0 for each PDC is registered as
15 * a mailbox channel. The PDC driver uses interrupts to determine when data
16 * transfers to and from an offload engine are complete. The PDC driver uses
17 * threaded IRQs so that response messages are handled outside of interrupt
18 * context.
19 *
20 * The PDC driver allows multiple messages to be pending in the descriptor
21 * rings. The tx_msg_start descriptor index indicates where the last message
22 * starts. The txin_numd value at this index indicates how many descriptor
23 * indexes make up the message. Similar state is kept on the receive side. When
24 * an rx interrupt indicates a response is ready, the PDC driver processes numd
25 * descriptors from the tx and rx ring, thus processing one response at a time.
26 */
27
28#include <linux/errno.h>
29#include <linux/module.h>
30#include <linux/init.h>
31#include <linux/slab.h>
32#include <linux/debugfs.h>
33#include <linux/interrupt.h>
34#include <linux/wait.h>
35#include <linux/platform_device.h>
36#include <linux/io.h>
37#include <linux/of.h>
38#include <linux/of_device.h>
39#include <linux/of_address.h>
40#include <linux/of_irq.h>
41#include <linux/mailbox_controller.h>
42#include <linux/mailbox/brcm-message.h>
43#include <linux/scatterlist.h>
44#include <linux/dma-direction.h>
45#include <linux/dma-mapping.h>
46#include <linux/dmapool.h>
47
48#define PDC_SUCCESS  0
49
50#define RING_ENTRY_SIZE   sizeof(struct dma64dd)
51
52/* # entries in PDC dma ring */
53#define PDC_RING_ENTRIES  512
54/*
55 * Minimum number of ring descriptor entries that must be free to tell mailbox
56 * framework that it can submit another request
57 */
58#define PDC_RING_SPACE_MIN  15
59
60#define PDC_RING_SIZE    (PDC_RING_ENTRIES * RING_ENTRY_SIZE)
61/* Rings are 8k aligned */
62#define RING_ALIGN_ORDER  13
63#define RING_ALIGN        BIT(RING_ALIGN_ORDER)
64
65#define RX_BUF_ALIGN_ORDER  5
66#define RX_BUF_ALIGN	    BIT(RX_BUF_ALIGN_ORDER)
67
68/* descriptor bumping macros */
69#define XXD(x, max_mask)              ((x) & (max_mask))
70#define TXD(x, max_mask)              XXD((x), (max_mask))
71#define RXD(x, max_mask)              XXD((x), (max_mask))
72#define NEXTTXD(i, max_mask)          TXD((i) + 1, (max_mask))
73#define PREVTXD(i, max_mask)          TXD((i) - 1, (max_mask))
74#define NEXTRXD(i, max_mask)          RXD((i) + 1, (max_mask))
75#define PREVRXD(i, max_mask)          RXD((i) - 1, (max_mask))
76#define NTXDACTIVE(h, t, max_mask)    TXD((t) - (h), (max_mask))
77#define NRXDACTIVE(h, t, max_mask)    RXD((t) - (h), (max_mask))
78
79/* Length of BCM header at start of SPU msg, in bytes */
80#define BCM_HDR_LEN  8
81
82/*
83 * PDC driver reserves ringset 0 on each SPU for its own use. The driver does
84 * not currently support use of multiple ringsets on a single PDC engine.
85 */
86#define PDC_RINGSET  0
87
88/*
89 * Interrupt mask and status definitions. Enable interrupts for tx and rx on
90 * ring 0
91 */
92#define PDC_RCVINT_0         (16 + PDC_RINGSET)
93#define PDC_RCVINTEN_0       BIT(PDC_RCVINT_0)
94#define PDC_INTMASK	     (PDC_RCVINTEN_0)
95#define PDC_LAZY_FRAMECOUNT  1
96#define PDC_LAZY_TIMEOUT     10000
97#define PDC_LAZY_INT  (PDC_LAZY_TIMEOUT | (PDC_LAZY_FRAMECOUNT << 24))
98#define PDC_INTMASK_OFFSET   0x24
99#define PDC_INTSTATUS_OFFSET 0x20
100#define PDC_RCVLAZY0_OFFSET  (0x30 + 4 * PDC_RINGSET)
101#define FA_RCVLAZY0_OFFSET   0x100
102
103/*
104 * For SPU2, configure MDE_CKSUM_CONTROL to write 17 bytes of metadata
105 * before frame
106 */
107#define PDC_SPU2_RESP_HDR_LEN  17
108#define PDC_CKSUM_CTRL         BIT(27)
109#define PDC_CKSUM_CTRL_OFFSET  0x400
110
111#define PDC_SPUM_RESP_HDR_LEN  32
112
113/*
114 * Sets the following bits for write to transmit control reg:
115 * 11    - PtyChkDisable - parity check is disabled
116 * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
117 */
118#define PDC_TX_CTL		0x000C0800
119
120/* Bit in tx control reg to enable tx channel */
121#define PDC_TX_ENABLE		0x1
122
123/*
124 * Sets the following bits for write to receive control reg:
125 * 7:1   - RcvOffset - size in bytes of status region at start of rx frame buf
126 * 9     - SepRxHdrDescEn - place start of new frames only in descriptors
127 *                          that have StartOfFrame set
128 * 10    - OflowContinue - on rx FIFO overflow, clear rx fifo, discard all
129 *                         remaining bytes in current frame, report error
130 *                         in rx frame status for current frame
131 * 11    - PtyChkDisable - parity check is disabled
132 * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
133 */
134#define PDC_RX_CTL		0x000C0E00
135
136/* Bit in rx control reg to enable rx channel */
137#define PDC_RX_ENABLE		0x1
138
139#define CRYPTO_D64_RS0_CD_MASK   ((PDC_RING_ENTRIES * RING_ENTRY_SIZE) - 1)
140
141/* descriptor flags */
142#define D64_CTRL1_EOT   BIT(28)	/* end of descriptor table */
143#define D64_CTRL1_IOC   BIT(29)	/* interrupt on complete */
144#define D64_CTRL1_EOF   BIT(30)	/* end of frame */
145#define D64_CTRL1_SOF   BIT(31)	/* start of frame */
146
147#define RX_STATUS_OVERFLOW       0x00800000
148#define RX_STATUS_LEN            0x0000FFFF
149
150#define PDC_TXREGS_OFFSET  0x200
151#define PDC_RXREGS_OFFSET  0x220
152
153/* Maximum size buffer the DMA engine can handle */
154#define PDC_DMA_BUF_MAX 16384
155
156enum pdc_hw {
157	FA_HW,		/* FA2/FA+ hardware (i.e. Northstar Plus) */
158	PDC_HW		/* PDC/MDE hardware (i.e. Northstar 2, Pegasus) */
159};
160
161struct pdc_dma_map {
162	void *ctx;          /* opaque context associated with frame */
163};
164
165/* dma descriptor */
166struct dma64dd {
167	u32 ctrl1;      /* misc control bits */
168	u32 ctrl2;      /* buffer count and address extension */
169	u32 addrlow;    /* memory address of the date buffer, bits 31:0 */
170	u32 addrhigh;   /* memory address of the date buffer, bits 63:32 */
171};
172
173/* dma registers per channel(xmt or rcv) */
174struct dma64_regs {
175	u32  control;   /* enable, et al */
176	u32  ptr;       /* last descriptor posted to chip */
177	u32  addrlow;   /* descriptor ring base address low 32-bits */
178	u32  addrhigh;  /* descriptor ring base address bits 63:32 */
179	u32  status0;   /* last rx descriptor written by hw */
180	u32  status1;   /* driver does not use */
181};
182
183/* cpp contortions to concatenate w/arg prescan */
184#ifndef PAD
185#define _PADLINE(line)  pad ## line
186#define _XSTR(line)     _PADLINE(line)
187#define PAD             _XSTR(__LINE__)
188#endif  /* PAD */
189
190/* dma registers. matches hw layout. */
191struct dma64 {
192	struct dma64_regs dmaxmt;  /* dma tx */
193	u32          PAD[2];
194	struct dma64_regs dmarcv;  /* dma rx */
195	u32          PAD[2];
196};
197
198/* PDC registers */
199struct pdc_regs {
200	u32  devcontrol;             /* 0x000 */
201	u32  devstatus;              /* 0x004 */
202	u32  PAD;
203	u32  biststatus;             /* 0x00c */
204	u32  PAD[4];
205	u32  intstatus;              /* 0x020 */
206	u32  intmask;                /* 0x024 */
207	u32  gptimer;                /* 0x028 */
208
209	u32  PAD;
210	u32  intrcvlazy_0;           /* 0x030 (Only in PDC, not FA2) */
211	u32  intrcvlazy_1;           /* 0x034 (Only in PDC, not FA2) */
212	u32  intrcvlazy_2;           /* 0x038 (Only in PDC, not FA2) */
213	u32  intrcvlazy_3;           /* 0x03c (Only in PDC, not FA2) */
214
215	u32  PAD[48];
216	u32  fa_intrecvlazy;         /* 0x100 (Only in FA2, not PDC) */
217	u32  flowctlthresh;          /* 0x104 */
218	u32  wrrthresh;              /* 0x108 */
219	u32  gmac_idle_cnt_thresh;   /* 0x10c */
220
221	u32  PAD[4];
222	u32  ifioaccessaddr;         /* 0x120 */
223	u32  ifioaccessbyte;         /* 0x124 */
224	u32  ifioaccessdata;         /* 0x128 */
225
226	u32  PAD[21];
227	u32  phyaccess;              /* 0x180 */
228	u32  PAD;
229	u32  phycontrol;             /* 0x188 */
230	u32  txqctl;                 /* 0x18c */
231	u32  rxqctl;                 /* 0x190 */
232	u32  gpioselect;             /* 0x194 */
233	u32  gpio_output_en;         /* 0x198 */
234	u32  PAD;                    /* 0x19c */
235	u32  txq_rxq_mem_ctl;        /* 0x1a0 */
236	u32  memory_ecc_status;      /* 0x1a4 */
237	u32  serdes_ctl;             /* 0x1a8 */
238	u32  serdes_status0;         /* 0x1ac */
239	u32  serdes_status1;         /* 0x1b0 */
240	u32  PAD[11];                /* 0x1b4-1dc */
241	u32  clk_ctl_st;             /* 0x1e0 */
242	u32  hw_war;                 /* 0x1e4 (Only in PDC, not FA2) */
243	u32  pwrctl;                 /* 0x1e8 */
244	u32  PAD[5];
245
246#define PDC_NUM_DMA_RINGS   4
247	struct dma64 dmaregs[PDC_NUM_DMA_RINGS];  /* 0x0200 - 0x2fc */
248
249	/* more registers follow, but we don't use them */
250};
251
252/* structure for allocating/freeing DMA rings */
253struct pdc_ring_alloc {
254	dma_addr_t  dmabase; /* DMA address of start of ring */
255	void	   *vbase;   /* base kernel virtual address of ring */
256	u32	    size;    /* ring allocation size in bytes */
257};
258
259/*
260 * context associated with a receive descriptor.
261 * @rxp_ctx: opaque context associated with frame that starts at each
262 *           rx ring index.
263 * @dst_sg:  Scatterlist used to form reply frames beginning at a given ring
264 *           index. Retained in order to unmap each sg after reply is processed.
265 * @rxin_numd: Number of rx descriptors associated with the message that starts
266 *             at a descriptor index. Not set for every index. For example,
267 *             if descriptor index i points to a scatterlist with 4 entries,
268 *             then the next three descriptor indexes don't have a value set.
269 * @resp_hdr: Virtual address of buffer used to catch DMA rx status
270 * @resp_hdr_daddr: physical address of DMA rx status buffer
271 */
272struct pdc_rx_ctx {
273	void *rxp_ctx;
274	struct scatterlist *dst_sg;
275	u32  rxin_numd;
276	void *resp_hdr;
277	dma_addr_t resp_hdr_daddr;
278};
279
280/* PDC state structure */
281struct pdc_state {
282	/* Index of the PDC whose state is in this structure instance */
283	u8 pdc_idx;
284
285	/* Platform device for this PDC instance */
286	struct platform_device *pdev;
287
288	/*
289	 * Each PDC instance has a mailbox controller. PDC receives request
290	 * messages through mailboxes, and sends response messages through the
291	 * mailbox framework.
292	 */
293	struct mbox_controller mbc;
294
295	unsigned int pdc_irq;
296
297	/* tasklet for deferred processing after DMA rx interrupt */
298	struct tasklet_struct rx_tasklet;
299
300	/* Number of bytes of receive status prior to each rx frame */
301	u32 rx_status_len;
302	/* Whether a BCM header is prepended to each frame */
303	bool use_bcm_hdr;
304	/* Sum of length of BCM header and rx status header */
305	u32 pdc_resp_hdr_len;
306
307	/* The base virtual address of DMA hw registers */
308	void __iomem *pdc_reg_vbase;
309
310	/* Pool for allocation of DMA rings */
311	struct dma_pool *ring_pool;
312
313	/* Pool for allocation of metadata buffers for response messages */
314	struct dma_pool *rx_buf_pool;
315
316	/*
317	 * The base virtual address of DMA tx/rx descriptor rings. Corresponding
318	 * DMA address and size of ring allocation.
319	 */
320	struct pdc_ring_alloc tx_ring_alloc;
321	struct pdc_ring_alloc rx_ring_alloc;
322
323	struct pdc_regs *regs;    /* start of PDC registers */
324
325	struct dma64_regs *txregs_64; /* dma tx engine registers */
326	struct dma64_regs *rxregs_64; /* dma rx engine registers */
327
328	/*
329	 * Arrays of PDC_RING_ENTRIES descriptors
330	 * To use multiple ringsets, this needs to be extended
331	 */
332	struct dma64dd   *txd_64;  /* tx descriptor ring */
333	struct dma64dd   *rxd_64;  /* rx descriptor ring */
334
335	/* descriptor ring sizes */
336	u32      ntxd;       /* # tx descriptors */
337	u32      nrxd;       /* # rx descriptors */
338	u32      nrxpost;    /* # rx buffers to keep posted */
339	u32      ntxpost;    /* max number of tx buffers that can be posted */
340
341	/*
342	 * Index of next tx descriptor to reclaim. That is, the descriptor
343	 * index of the oldest tx buffer for which the host has yet to process
344	 * the corresponding response.
345	 */
346	u32  txin;
347
348	/*
349	 * Index of the first receive descriptor for the sequence of
350	 * message fragments currently under construction. Used to build up
351	 * the rxin_numd count for a message. Updated to rxout when the host
352	 * starts a new sequence of rx buffers for a new message.
353	 */
354	u32  tx_msg_start;
355
356	/* Index of next tx descriptor to post. */
357	u32  txout;
358
359	/*
360	 * Number of tx descriptors associated with the message that starts
361	 * at this tx descriptor index.
362	 */
363	u32      txin_numd[PDC_RING_ENTRIES];
364
365	/*
366	 * Index of next rx descriptor to reclaim. This is the index of
367	 * the next descriptor whose data has yet to be processed by the host.
368	 */
369	u32  rxin;
370
371	/*
372	 * Index of the first receive descriptor for the sequence of
373	 * message fragments currently under construction. Used to build up
374	 * the rxin_numd count for a message. Updated to rxout when the host
375	 * starts a new sequence of rx buffers for a new message.
376	 */
377	u32  rx_msg_start;
378
379	/*
380	 * Saved value of current hardware rx descriptor index.
381	 * The last rx buffer written by the hw is the index previous to
382	 * this one.
383	 */
384	u32  last_rx_curr;
385
386	/* Index of next rx descriptor to post. */
387	u32  rxout;
388
389	struct pdc_rx_ctx rx_ctx[PDC_RING_ENTRIES];
390
391	/*
392	 * Scatterlists used to form request and reply frames beginning at a
393	 * given ring index. Retained in order to unmap each sg after reply
394	 * is processed
395	 */
396	struct scatterlist *src_sg[PDC_RING_ENTRIES];
397
398	/* counters */
399	u32  pdc_requests;     /* number of request messages submitted */
400	u32  pdc_replies;      /* number of reply messages received */
401	u32  last_tx_not_done; /* too few tx descriptors to indicate done */
402	u32  tx_ring_full;     /* unable to accept msg because tx ring full */
403	u32  rx_ring_full;     /* unable to accept msg because rx ring full */
404	u32  txnobuf;          /* unable to create tx descriptor */
405	u32  rxnobuf;          /* unable to create rx descriptor */
406	u32  rx_oflow;         /* count of rx overflows */
407
408	/* hardware type - FA2 or PDC/MDE */
409	enum pdc_hw hw_type;
410};
411
412/* Global variables */
413
414struct pdc_globals {
415	/* Actual number of SPUs in hardware, as reported by device tree */
416	u32 num_spu;
417};
418
419static struct pdc_globals pdcg;
420
421/* top level debug FS directory for PDC driver */
422static struct dentry *debugfs_dir;
423
424static ssize_t pdc_debugfs_read(struct file *filp, char __user *ubuf,
425				size_t count, loff_t *offp)
426{
427	struct pdc_state *pdcs;
428	char *buf;
429	ssize_t ret, out_offset, out_count;
430
431	out_count = 512;
432
433	buf = kmalloc(out_count, GFP_KERNEL);
434	if (!buf)
435		return -ENOMEM;
436
437	pdcs = filp->private_data;
438	out_offset = 0;
439	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
440			       "SPU %u stats:\n", pdcs->pdc_idx);
441	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
442			       "PDC requests....................%u\n",
443			       pdcs->pdc_requests);
444	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
445			       "PDC responses...................%u\n",
446			       pdcs->pdc_replies);
447	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
448			       "Tx not done.....................%u\n",
449			       pdcs->last_tx_not_done);
450	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
451			       "Tx ring full....................%u\n",
452			       pdcs->tx_ring_full);
453	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
454			       "Rx ring full....................%u\n",
455			       pdcs->rx_ring_full);
456	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
457			       "Tx desc write fail. Ring full...%u\n",
458			       pdcs->txnobuf);
459	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
460			       "Rx desc write fail. Ring full...%u\n",
461			       pdcs->rxnobuf);
462	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
463			       "Receive overflow................%u\n",
464			       pdcs->rx_oflow);
465	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
466			       "Num frags in rx ring............%u\n",
467			       NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr,
468					  pdcs->nrxpost));
469
470	if (out_offset > out_count)
471		out_offset = out_count;
472
473	ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset);
474	kfree(buf);
475	return ret;
476}
477
478static const struct file_operations pdc_debugfs_stats = {
479	.owner = THIS_MODULE,
480	.open = simple_open,
481	.read = pdc_debugfs_read,
482};
483
484/**
485 * pdc_setup_debugfs() - Create the debug FS directories. If the top-level
486 * directory has not yet been created, create it now. Create a stats file in
487 * this directory for a SPU.
488 * @pdcs: PDC state structure
489 */
490static void pdc_setup_debugfs(struct pdc_state *pdcs)
491{
492	char spu_stats_name[16];
493
494	if (!debugfs_initialized())
495		return;
496
497	snprintf(spu_stats_name, 16, "pdc%d_stats", pdcs->pdc_idx);
498	if (!debugfs_dir)
499		debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
500
501	/* S_IRUSR == 0400 */
502	debugfs_create_file(spu_stats_name, 0400, debugfs_dir, pdcs,
503			    &pdc_debugfs_stats);
504}
505
506static void pdc_free_debugfs(void)
507{
508	debugfs_remove_recursive(debugfs_dir);
509	debugfs_dir = NULL;
510}
511
512/**
513 * pdc_build_rxd() - Build DMA descriptor to receive SPU result.
514 * @pdcs:      PDC state for SPU that will generate result
515 * @dma_addr:  DMA address of buffer that descriptor is being built for
516 * @buf_len:   Length of the receive buffer, in bytes
517 * @flags:     Flags to be stored in descriptor
518 */
519static inline void
520pdc_build_rxd(struct pdc_state *pdcs, dma_addr_t dma_addr,
521	      u32 buf_len, u32 flags)
522{
523	struct device *dev = &pdcs->pdev->dev;
524	struct dma64dd *rxd = &pdcs->rxd_64[pdcs->rxout];
525
526	dev_dbg(dev,
527		"Writing rx descriptor for PDC %u at index %u with length %u. flags %#x\n",
528		pdcs->pdc_idx, pdcs->rxout, buf_len, flags);
529
530	rxd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
531	rxd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
532	rxd->ctrl1 = cpu_to_le32(flags);
533	rxd->ctrl2 = cpu_to_le32(buf_len);
534
535	/* bump ring index and return */
536	pdcs->rxout = NEXTRXD(pdcs->rxout, pdcs->nrxpost);
537}
538
539/**
540 * pdc_build_txd() - Build a DMA descriptor to transmit a SPU request to
541 * hardware.
542 * @pdcs:        PDC state for the SPU that will process this request
543 * @dma_addr:    DMA address of packet to be transmitted
544 * @buf_len:     Length of tx buffer, in bytes
545 * @flags:       Flags to be stored in descriptor
546 */
547static inline void
548pdc_build_txd(struct pdc_state *pdcs, dma_addr_t dma_addr, u32 buf_len,
549	      u32 flags)
550{
551	struct device *dev = &pdcs->pdev->dev;
552	struct dma64dd *txd = &pdcs->txd_64[pdcs->txout];
553
554	dev_dbg(dev,
555		"Writing tx descriptor for PDC %u at index %u with length %u, flags %#x\n",
556		pdcs->pdc_idx, pdcs->txout, buf_len, flags);
557
558	txd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
559	txd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
560	txd->ctrl1 = cpu_to_le32(flags);
561	txd->ctrl2 = cpu_to_le32(buf_len);
562
563	/* bump ring index and return */
564	pdcs->txout = NEXTTXD(pdcs->txout, pdcs->ntxpost);
565}
566
567/**
568 * pdc_receive_one() - Receive a response message from a given SPU.
569 * @pdcs:    PDC state for the SPU to receive from
570 *
571 * When the return code indicates success, the response message is available in
572 * the receive buffers provided prior to submission of the request.
573 *
574 * Return:  PDC_SUCCESS if one or more receive descriptors was processed
575 *          -EAGAIN indicates that no response message is available
576 *          -EIO an error occurred
577 */
578static int
579pdc_receive_one(struct pdc_state *pdcs)
580{
581	struct device *dev = &pdcs->pdev->dev;
582	struct mbox_controller *mbc;
583	struct mbox_chan *chan;
584	struct brcm_message mssg;
585	u32 len, rx_status;
586	u32 num_frags;
587	u8 *resp_hdr;    /* virtual addr of start of resp message DMA header */
588	u32 frags_rdy;   /* number of fragments ready to read */
589	u32 rx_idx;      /* ring index of start of receive frame */
590	dma_addr_t resp_hdr_daddr;
591	struct pdc_rx_ctx *rx_ctx;
592
593	mbc = &pdcs->mbc;
594	chan = &mbc->chans[0];
595	mssg.type = BRCM_MESSAGE_SPU;
596
597	/*
598	 * return if a complete response message is not yet ready.
599	 * rxin_numd[rxin] is the number of fragments in the next msg
600	 * to read.
601	 */
602	frags_rdy = NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr, pdcs->nrxpost);
603	if ((frags_rdy == 0) ||
604	    (frags_rdy < pdcs->rx_ctx[pdcs->rxin].rxin_numd))
605		/* No response ready */
606		return -EAGAIN;
607
608	num_frags = pdcs->txin_numd[pdcs->txin];
609	WARN_ON(num_frags == 0);
610
611	dma_unmap_sg(dev, pdcs->src_sg[pdcs->txin],
612		     sg_nents(pdcs->src_sg[pdcs->txin]), DMA_TO_DEVICE);
613
614	pdcs->txin = (pdcs->txin + num_frags) & pdcs->ntxpost;
615
616	dev_dbg(dev, "PDC %u reclaimed %d tx descriptors",
617		pdcs->pdc_idx, num_frags);
618
619	rx_idx = pdcs->rxin;
620	rx_ctx = &pdcs->rx_ctx[rx_idx];
621	num_frags = rx_ctx->rxin_numd;
622	/* Return opaque context with result */
623	mssg.ctx = rx_ctx->rxp_ctx;
624	rx_ctx->rxp_ctx = NULL;
625	resp_hdr = rx_ctx->resp_hdr;
626	resp_hdr_daddr = rx_ctx->resp_hdr_daddr;
627	dma_unmap_sg(dev, rx_ctx->dst_sg, sg_nents(rx_ctx->dst_sg),
628		     DMA_FROM_DEVICE);
629
630	pdcs->rxin = (pdcs->rxin + num_frags) & pdcs->nrxpost;
631
632	dev_dbg(dev, "PDC %u reclaimed %d rx descriptors",
633		pdcs->pdc_idx, num_frags);
634
635	dev_dbg(dev,
636		"PDC %u txin %u, txout %u, rxin %u, rxout %u, last_rx_curr %u\n",
637		pdcs->pdc_idx, pdcs->txin, pdcs->txout, pdcs->rxin,
638		pdcs->rxout, pdcs->last_rx_curr);
639
640	if (pdcs->pdc_resp_hdr_len == PDC_SPUM_RESP_HDR_LEN) {
641		/*
642		 * For SPU-M, get length of response msg and rx overflow status.
643		 */
644		rx_status = *((u32 *)resp_hdr);
645		len = rx_status & RX_STATUS_LEN;
646		dev_dbg(dev,
647			"SPU response length %u bytes", len);
648		if (unlikely(((rx_status & RX_STATUS_OVERFLOW) || (!len)))) {
649			if (rx_status & RX_STATUS_OVERFLOW) {
650				dev_err_ratelimited(dev,
651						    "crypto receive overflow");
652				pdcs->rx_oflow++;
653			} else {
654				dev_info_ratelimited(dev, "crypto rx len = 0");
655			}
656			return -EIO;
657		}
658	}
659
660	dma_pool_free(pdcs->rx_buf_pool, resp_hdr, resp_hdr_daddr);
661
662	mbox_chan_received_data(chan, &mssg);
663
664	pdcs->pdc_replies++;
665	return PDC_SUCCESS;
666}
667
668/**
669 * pdc_receive() - Process as many responses as are available in the rx ring.
670 * @pdcs:  PDC state
671 *
672 * Called within the hard IRQ.
673 * Return:
674 */
675static int
676pdc_receive(struct pdc_state *pdcs)
677{
678	int rx_status;
679
680	/* read last_rx_curr from register once */
681	pdcs->last_rx_curr =
682	    (ioread32((const void __iomem *)&pdcs->rxregs_64->status0) &
683	     CRYPTO_D64_RS0_CD_MASK) / RING_ENTRY_SIZE;
684
685	do {
686		/* Could be many frames ready */
687		rx_status = pdc_receive_one(pdcs);
688	} while (rx_status == PDC_SUCCESS);
689
690	return 0;
691}
692
693/**
694 * pdc_tx_list_sg_add() - Add the buffers in a scatterlist to the transmit
695 * descriptors for a given SPU. The scatterlist buffers contain the data for a
696 * SPU request message.
697 * @spu_idx:   The index of the SPU to submit the request to, [0, max_spu)
698 * @sg:        Scatterlist whose buffers contain part of the SPU request
699 *
700 * If a scatterlist buffer is larger than PDC_DMA_BUF_MAX, multiple descriptors
701 * are written for that buffer, each <= PDC_DMA_BUF_MAX byte in length.
702 *
703 * Return: PDC_SUCCESS if successful
704 *         < 0 otherwise
705 */
706static int pdc_tx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
707{
708	u32 flags = 0;
709	u32 eot;
710	u32 tx_avail;
711
712	/*
713	 * Num descriptors needed. Conservatively assume we need a descriptor
714	 * for every entry in sg.
715	 */
716	u32 num_desc;
717	u32 desc_w = 0;	/* Number of tx descriptors written */
718	u32 bufcnt;	/* Number of bytes of buffer pointed to by descriptor */
719	dma_addr_t databufptr;	/* DMA address to put in descriptor */
720
721	num_desc = (u32)sg_nents(sg);
722
723	/* check whether enough tx descriptors are available */
724	tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
725					      pdcs->ntxpost);
726	if (unlikely(num_desc > tx_avail)) {
727		pdcs->txnobuf++;
728		return -ENOSPC;
729	}
730
731	/* build tx descriptors */
732	if (pdcs->tx_msg_start == pdcs->txout) {
733		/* Start of frame */
734		pdcs->txin_numd[pdcs->tx_msg_start] = 0;
735		pdcs->src_sg[pdcs->txout] = sg;
736		flags = D64_CTRL1_SOF;
737	}
738
739	while (sg) {
740		if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
741			eot = D64_CTRL1_EOT;
742		else
743			eot = 0;
744
745		/*
746		 * If sg buffer larger than PDC limit, split across
747		 * multiple descriptors
748		 */
749		bufcnt = sg_dma_len(sg);
750		databufptr = sg_dma_address(sg);
751		while (bufcnt > PDC_DMA_BUF_MAX) {
752			pdc_build_txd(pdcs, databufptr, PDC_DMA_BUF_MAX,
753				      flags | eot);
754			desc_w++;
755			bufcnt -= PDC_DMA_BUF_MAX;
756			databufptr += PDC_DMA_BUF_MAX;
757			if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
758				eot = D64_CTRL1_EOT;
759			else
760				eot = 0;
761		}
762		sg = sg_next(sg);
763		if (!sg)
764			/* Writing last descriptor for frame */
765			flags |= (D64_CTRL1_EOF | D64_CTRL1_IOC);
766		pdc_build_txd(pdcs, databufptr, bufcnt, flags | eot);
767		desc_w++;
768		/* Clear start of frame after first descriptor */
769		flags &= ~D64_CTRL1_SOF;
770	}
771	pdcs->txin_numd[pdcs->tx_msg_start] += desc_w;
772
773	return PDC_SUCCESS;
774}
775
776/**
777 * pdc_tx_list_final() - Initiate DMA transfer of last frame written to tx
778 * ring.
779 * @pdcs:  PDC state for SPU to process the request
780 *
781 * Sets the index of the last descriptor written in both the rx and tx ring.
782 *
783 * Return: PDC_SUCCESS
784 */
785static int pdc_tx_list_final(struct pdc_state *pdcs)
786{
787	/*
788	 * write barrier to ensure all register writes are complete
789	 * before chip starts to process new request
790	 */
791	wmb();
792	iowrite32(pdcs->rxout << 4, &pdcs->rxregs_64->ptr);
793	iowrite32(pdcs->txout << 4, &pdcs->txregs_64->ptr);
794	pdcs->pdc_requests++;
795
796	return PDC_SUCCESS;
797}
798
799/**
800 * pdc_rx_list_init() - Start a new receive descriptor list for a given PDC.
801 * @pdcs:   PDC state for SPU handling request
802 * @dst_sg: scatterlist providing rx buffers for response to be returned to
803 *	    mailbox client
804 * @ctx:    Opaque context for this request
805 *
806 * Posts a single receive descriptor to hold the metadata that precedes a
807 * response. For example, with SPU-M, the metadata is a 32-byte DMA header and
808 * an 8-byte BCM header. Moves the msg_start descriptor indexes for both tx and
809 * rx to indicate the start of a new message.
810 *
811 * Return:  PDC_SUCCESS if successful
812 *          < 0 if an error (e.g., rx ring is full)
813 */
814static int pdc_rx_list_init(struct pdc_state *pdcs, struct scatterlist *dst_sg,
815			    void *ctx)
816{
817	u32 flags = 0;
818	u32 rx_avail;
819	u32 rx_pkt_cnt = 1;	/* Adding a single rx buffer */
820	dma_addr_t daddr;
821	void *vaddr;
822	struct pdc_rx_ctx *rx_ctx;
823
824	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
825					      pdcs->nrxpost);
826	if (unlikely(rx_pkt_cnt > rx_avail)) {
827		pdcs->rxnobuf++;
828		return -ENOSPC;
829	}
830
831	/* allocate a buffer for the dma rx status */
832	vaddr = dma_pool_zalloc(pdcs->rx_buf_pool, GFP_ATOMIC, &daddr);
833	if (unlikely(!vaddr))
834		return -ENOMEM;
835
836	/*
837	 * Update msg_start indexes for both tx and rx to indicate the start
838	 * of a new sequence of descriptor indexes that contain the fragments
839	 * of the same message.
840	 */
841	pdcs->rx_msg_start = pdcs->rxout;
842	pdcs->tx_msg_start = pdcs->txout;
843
844	/* This is always the first descriptor in the receive sequence */
845	flags = D64_CTRL1_SOF;
846	pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd = 1;
847
848	if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
849		flags |= D64_CTRL1_EOT;
850
851	rx_ctx = &pdcs->rx_ctx[pdcs->rxout];
852	rx_ctx->rxp_ctx = ctx;
853	rx_ctx->dst_sg = dst_sg;
854	rx_ctx->resp_hdr = vaddr;
855	rx_ctx->resp_hdr_daddr = daddr;
856	pdc_build_rxd(pdcs, daddr, pdcs->pdc_resp_hdr_len, flags);
857	return PDC_SUCCESS;
858}
859
860/**
861 * pdc_rx_list_sg_add() - Add the buffers in a scatterlist to the receive
862 * descriptors for a given SPU. The caller must have already DMA mapped the
863 * scatterlist.
864 * @spu_idx:    Indicates which SPU the buffers are for
865 * @sg:         Scatterlist whose buffers are added to the receive ring
866 *
867 * If a receive buffer in the scatterlist is larger than PDC_DMA_BUF_MAX,
868 * multiple receive descriptors are written, each with a buffer <=
869 * PDC_DMA_BUF_MAX.
870 *
871 * Return: PDC_SUCCESS if successful
872 *         < 0 otherwise (e.g., receive ring is full)
873 */
874static int pdc_rx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
875{
876	u32 flags = 0;
877	u32 rx_avail;
878
879	/*
880	 * Num descriptors needed. Conservatively assume we need a descriptor
881	 * for every entry from our starting point in the scatterlist.
882	 */
883	u32 num_desc;
884	u32 desc_w = 0;	/* Number of tx descriptors written */
885	u32 bufcnt;	/* Number of bytes of buffer pointed to by descriptor */
886	dma_addr_t databufptr;	/* DMA address to put in descriptor */
887
888	num_desc = (u32)sg_nents(sg);
889
890	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
891					      pdcs->nrxpost);
892	if (unlikely(num_desc > rx_avail)) {
893		pdcs->rxnobuf++;
894		return -ENOSPC;
895	}
896
897	while (sg) {
898		if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
899			flags = D64_CTRL1_EOT;
900		else
901			flags = 0;
902
903		/*
904		 * If sg buffer larger than PDC limit, split across
905		 * multiple descriptors
906		 */
907		bufcnt = sg_dma_len(sg);
908		databufptr = sg_dma_address(sg);
909		while (bufcnt > PDC_DMA_BUF_MAX) {
910			pdc_build_rxd(pdcs, databufptr, PDC_DMA_BUF_MAX, flags);
911			desc_w++;
912			bufcnt -= PDC_DMA_BUF_MAX;
913			databufptr += PDC_DMA_BUF_MAX;
914			if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
915				flags = D64_CTRL1_EOT;
916			else
917				flags = 0;
918		}
919		pdc_build_rxd(pdcs, databufptr, bufcnt, flags);
920		desc_w++;
921		sg = sg_next(sg);
922	}
923	pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd += desc_w;
924
925	return PDC_SUCCESS;
926}
927
928/**
929 * pdc_irq_handler() - Interrupt handler called in interrupt context.
930 * @irq:      Interrupt number that has fired
931 * @data:     device struct for DMA engine that generated the interrupt
932 *
933 * We have to clear the device interrupt status flags here. So cache the
934 * status for later use in the thread function. Other than that, just return
935 * WAKE_THREAD to invoke the thread function.
936 *
937 * Return: IRQ_WAKE_THREAD if interrupt is ours
938 *         IRQ_NONE otherwise
939 */
940static irqreturn_t pdc_irq_handler(int irq, void *data)
941{
942	struct device *dev = (struct device *)data;
943	struct pdc_state *pdcs = dev_get_drvdata(dev);
944	u32 intstatus = ioread32(pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
945
946	if (unlikely(intstatus == 0))
947		return IRQ_NONE;
948
949	/* Disable interrupts until soft handler runs */
950	iowrite32(0, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
951
952	/* Clear interrupt flags in device */
953	iowrite32(intstatus, pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
954
955	/* Wakeup IRQ thread */
956	tasklet_schedule(&pdcs->rx_tasklet);
957	return IRQ_HANDLED;
958}
959
960/**
961 * pdc_tasklet_cb() - Tasklet callback that runs the deferred processing after
962 * a DMA receive interrupt. Reenables the receive interrupt.
963 * @data: PDC state structure
964 */
965static void pdc_tasklet_cb(struct tasklet_struct *t)
966{
967	struct pdc_state *pdcs = from_tasklet(pdcs, t, rx_tasklet);
968
969	pdc_receive(pdcs);
970
971	/* reenable interrupts */
972	iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
973}
974
975/**
976 * pdc_ring_init() - Allocate DMA rings and initialize constant fields of
977 * descriptors in one ringset.
978 * @pdcs:    PDC instance state
979 * @ringset: index of ringset being used
980 *
981 * Return: PDC_SUCCESS if ring initialized
982 *         < 0 otherwise
983 */
984static int pdc_ring_init(struct pdc_state *pdcs, int ringset)
985{
986	int i;
987	int err = PDC_SUCCESS;
988	struct dma64 *dma_reg;
989	struct device *dev = &pdcs->pdev->dev;
990	struct pdc_ring_alloc tx;
991	struct pdc_ring_alloc rx;
992
993	/* Allocate tx ring */
994	tx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &tx.dmabase);
995	if (unlikely(!tx.vbase)) {
996		err = -ENOMEM;
997		goto done;
998	}
999
1000	/* Allocate rx ring */
1001	rx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &rx.dmabase);
1002	if (unlikely(!rx.vbase)) {
1003		err = -ENOMEM;
1004		goto fail_dealloc;
1005	}
1006
1007	dev_dbg(dev, " - base DMA addr of tx ring      %pad", &tx.dmabase);
1008	dev_dbg(dev, " - base virtual addr of tx ring  %p", tx.vbase);
1009	dev_dbg(dev, " - base DMA addr of rx ring      %pad", &rx.dmabase);
1010	dev_dbg(dev, " - base virtual addr of rx ring  %p", rx.vbase);
1011
1012	memcpy(&pdcs->tx_ring_alloc, &tx, sizeof(tx));
1013	memcpy(&pdcs->rx_ring_alloc, &rx, sizeof(rx));
1014
1015	pdcs->rxin = 0;
1016	pdcs->rx_msg_start = 0;
1017	pdcs->last_rx_curr = 0;
1018	pdcs->rxout = 0;
1019	pdcs->txin = 0;
1020	pdcs->tx_msg_start = 0;
1021	pdcs->txout = 0;
1022
1023	/* Set descriptor array base addresses */
1024	pdcs->txd_64 = (struct dma64dd *)pdcs->tx_ring_alloc.vbase;
1025	pdcs->rxd_64 = (struct dma64dd *)pdcs->rx_ring_alloc.vbase;
1026
1027	/* Tell device the base DMA address of each ring */
1028	dma_reg = &pdcs->regs->dmaregs[ringset];
1029
1030	/* But first disable DMA and set curptr to 0 for both TX & RX */
1031	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1032	iowrite32((PDC_RX_CTL + (pdcs->rx_status_len << 1)),
1033		  &dma_reg->dmarcv.control);
1034	iowrite32(0, &dma_reg->dmaxmt.ptr);
1035	iowrite32(0, &dma_reg->dmarcv.ptr);
1036
1037	/* Set base DMA addresses */
1038	iowrite32(lower_32_bits(pdcs->tx_ring_alloc.dmabase),
1039		  &dma_reg->dmaxmt.addrlow);
1040	iowrite32(upper_32_bits(pdcs->tx_ring_alloc.dmabase),
1041		  &dma_reg->dmaxmt.addrhigh);
1042
1043	iowrite32(lower_32_bits(pdcs->rx_ring_alloc.dmabase),
1044		  &dma_reg->dmarcv.addrlow);
1045	iowrite32(upper_32_bits(pdcs->rx_ring_alloc.dmabase),
1046		  &dma_reg->dmarcv.addrhigh);
1047
1048	/* Re-enable DMA */
1049	iowrite32(PDC_TX_CTL | PDC_TX_ENABLE, &dma_reg->dmaxmt.control);
1050	iowrite32((PDC_RX_CTL | PDC_RX_ENABLE | (pdcs->rx_status_len << 1)),
1051		  &dma_reg->dmarcv.control);
1052
1053	/* Initialize descriptors */
1054	for (i = 0; i < PDC_RING_ENTRIES; i++) {
1055		/* Every tx descriptor can be used for start of frame. */
1056		if (i != pdcs->ntxpost) {
1057			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF,
1058				  &pdcs->txd_64[i].ctrl1);
1059		} else {
1060			/* Last descriptor in ringset. Set End of Table. */
1061			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF |
1062				  D64_CTRL1_EOT, &pdcs->txd_64[i].ctrl1);
1063		}
1064
1065		/* Every rx descriptor can be used for start of frame */
1066		if (i != pdcs->nrxpost) {
1067			iowrite32(D64_CTRL1_SOF,
1068				  &pdcs->rxd_64[i].ctrl1);
1069		} else {
1070			/* Last descriptor in ringset. Set End of Table. */
1071			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOT,
1072				  &pdcs->rxd_64[i].ctrl1);
1073		}
1074	}
1075	return PDC_SUCCESS;
1076
1077fail_dealloc:
1078	dma_pool_free(pdcs->ring_pool, tx.vbase, tx.dmabase);
1079done:
1080	return err;
1081}
1082
1083static void pdc_ring_free(struct pdc_state *pdcs)
1084{
1085	if (pdcs->tx_ring_alloc.vbase) {
1086		dma_pool_free(pdcs->ring_pool, pdcs->tx_ring_alloc.vbase,
1087			      pdcs->tx_ring_alloc.dmabase);
1088		pdcs->tx_ring_alloc.vbase = NULL;
1089	}
1090
1091	if (pdcs->rx_ring_alloc.vbase) {
1092		dma_pool_free(pdcs->ring_pool, pdcs->rx_ring_alloc.vbase,
1093			      pdcs->rx_ring_alloc.dmabase);
1094		pdcs->rx_ring_alloc.vbase = NULL;
1095	}
1096}
1097
1098/**
1099 * pdc_desc_count() - Count the number of DMA descriptors that will be required
1100 * for a given scatterlist. Account for the max length of a DMA buffer.
1101 * @sg:    Scatterlist to be DMA'd
1102 * Return: Number of descriptors required
1103 */
1104static u32 pdc_desc_count(struct scatterlist *sg)
1105{
1106	u32 cnt = 0;
1107
1108	while (sg) {
1109		cnt += ((sg->length / PDC_DMA_BUF_MAX) + 1);
1110		sg = sg_next(sg);
1111	}
1112	return cnt;
1113}
1114
1115/**
1116 * pdc_rings_full() - Check whether the tx ring has room for tx_cnt descriptors
1117 * and the rx ring has room for rx_cnt descriptors.
1118 * @pdcs:  PDC state
1119 * @tx_cnt: The number of descriptors required in the tx ring
1120 * @rx_cnt: The number of descriptors required i the rx ring
1121 *
1122 * Return: true if one of the rings does not have enough space
1123 *         false if sufficient space is available in both rings
1124 */
1125static bool pdc_rings_full(struct pdc_state *pdcs, int tx_cnt, int rx_cnt)
1126{
1127	u32 rx_avail;
1128	u32 tx_avail;
1129	bool full = false;
1130
1131	/* Check if the tx and rx rings are likely to have enough space */
1132	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
1133					      pdcs->nrxpost);
1134	if (unlikely(rx_cnt > rx_avail)) {
1135		pdcs->rx_ring_full++;
1136		full = true;
1137	}
1138
1139	if (likely(!full)) {
1140		tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
1141						      pdcs->ntxpost);
1142		if (unlikely(tx_cnt > tx_avail)) {
1143			pdcs->tx_ring_full++;
1144			full = true;
1145		}
1146	}
1147	return full;
1148}
1149
1150/**
1151 * pdc_last_tx_done() - If both the tx and rx rings have at least
1152 * PDC_RING_SPACE_MIN descriptors available, then indicate that the mailbox
1153 * framework can submit another message.
1154 * @chan:  mailbox channel to check
1155 * Return: true if PDC can accept another message on this channel
1156 */
1157static bool pdc_last_tx_done(struct mbox_chan *chan)
1158{
1159	struct pdc_state *pdcs = chan->con_priv;
1160	bool ret;
1161
1162	if (unlikely(pdc_rings_full(pdcs, PDC_RING_SPACE_MIN,
1163				    PDC_RING_SPACE_MIN))) {
1164		pdcs->last_tx_not_done++;
1165		ret = false;
1166	} else {
1167		ret = true;
1168	}
1169	return ret;
1170}
1171
1172/**
1173 * pdc_send_data() - mailbox send_data function
1174 * @chan:	The mailbox channel on which the data is sent. The channel
1175 *              corresponds to a DMA ringset.
1176 * @data:	The mailbox message to be sent. The message must be a
1177 *              brcm_message structure.
1178 *
1179 * This function is registered as the send_data function for the mailbox
1180 * controller. From the destination scatterlist in the mailbox message, it
1181 * creates a sequence of receive descriptors in the rx ring. From the source
1182 * scatterlist, it creates a sequence of transmit descriptors in the tx ring.
1183 * After creating the descriptors, it writes the rx ptr and tx ptr registers to
1184 * initiate the DMA transfer.
1185 *
1186 * This function does the DMA map and unmap of the src and dst scatterlists in
1187 * the mailbox message.
1188 *
1189 * Return: 0 if successful
1190 *	   -ENOTSUPP if the mailbox message is a type this driver does not
1191 *			support
1192 *         < 0 if an error
1193 */
1194static int pdc_send_data(struct mbox_chan *chan, void *data)
1195{
1196	struct pdc_state *pdcs = chan->con_priv;
1197	struct device *dev = &pdcs->pdev->dev;
1198	struct brcm_message *mssg = data;
1199	int err = PDC_SUCCESS;
1200	int src_nent;
1201	int dst_nent;
1202	int nent;
1203	u32 tx_desc_req;
1204	u32 rx_desc_req;
1205
1206	if (unlikely(mssg->type != BRCM_MESSAGE_SPU))
1207		return -ENOTSUPP;
1208
1209	src_nent = sg_nents(mssg->spu.src);
1210	if (likely(src_nent)) {
1211		nent = dma_map_sg(dev, mssg->spu.src, src_nent, DMA_TO_DEVICE);
1212		if (unlikely(nent == 0))
1213			return -EIO;
1214	}
1215
1216	dst_nent = sg_nents(mssg->spu.dst);
1217	if (likely(dst_nent)) {
1218		nent = dma_map_sg(dev, mssg->spu.dst, dst_nent,
1219				  DMA_FROM_DEVICE);
1220		if (unlikely(nent == 0)) {
1221			dma_unmap_sg(dev, mssg->spu.src, src_nent,
1222				     DMA_TO_DEVICE);
1223			return -EIO;
1224		}
1225	}
1226
1227	/*
1228	 * Check if the tx and rx rings have enough space. Do this prior to
1229	 * writing any tx or rx descriptors. Need to ensure that we do not write
1230	 * a partial set of descriptors, or write just rx descriptors but
1231	 * corresponding tx descriptors don't fit. Note that we want this check
1232	 * and the entire sequence of descriptor to happen without another
1233	 * thread getting in. The channel spin lock in the mailbox framework
1234	 * ensures this.
1235	 */
1236	tx_desc_req = pdc_desc_count(mssg->spu.src);
1237	rx_desc_req = pdc_desc_count(mssg->spu.dst);
1238	if (unlikely(pdc_rings_full(pdcs, tx_desc_req, rx_desc_req + 1)))
1239		return -ENOSPC;
1240
1241	/* Create rx descriptors to SPU catch response */
1242	err = pdc_rx_list_init(pdcs, mssg->spu.dst, mssg->ctx);
1243	err |= pdc_rx_list_sg_add(pdcs, mssg->spu.dst);
1244
1245	/* Create tx descriptors to submit SPU request */
1246	err |= pdc_tx_list_sg_add(pdcs, mssg->spu.src);
1247	err |= pdc_tx_list_final(pdcs);	/* initiate transfer */
1248
1249	if (unlikely(err))
1250		dev_err(&pdcs->pdev->dev,
1251			"%s failed with error %d", __func__, err);
1252
1253	return err;
1254}
1255
1256static int pdc_startup(struct mbox_chan *chan)
1257{
1258	return pdc_ring_init(chan->con_priv, PDC_RINGSET);
1259}
1260
1261static void pdc_shutdown(struct mbox_chan *chan)
1262{
1263	struct pdc_state *pdcs = chan->con_priv;
1264
1265	if (!pdcs)
1266		return;
1267
1268	dev_dbg(&pdcs->pdev->dev,
1269		"Shutdown mailbox channel for PDC %u", pdcs->pdc_idx);
1270	pdc_ring_free(pdcs);
1271}
1272
1273/**
1274 * pdc_hw_init() - Use the given initialization parameters to initialize the
1275 * state for one of the PDCs.
1276 * @pdcs:  state of the PDC
1277 */
1278static
1279void pdc_hw_init(struct pdc_state *pdcs)
1280{
1281	struct platform_device *pdev;
1282	struct device *dev;
1283	struct dma64 *dma_reg;
1284	int ringset = PDC_RINGSET;
1285
1286	pdev = pdcs->pdev;
1287	dev = &pdev->dev;
1288
1289	dev_dbg(dev, "PDC %u initial values:", pdcs->pdc_idx);
1290	dev_dbg(dev, "state structure:                   %p",
1291		pdcs);
1292	dev_dbg(dev, " - base virtual addr of hw regs    %p",
1293		pdcs->pdc_reg_vbase);
1294
1295	/* initialize data structures */
1296	pdcs->regs = (struct pdc_regs *)pdcs->pdc_reg_vbase;
1297	pdcs->txregs_64 = (struct dma64_regs *)
1298	    (((u8 *)pdcs->pdc_reg_vbase) +
1299		     PDC_TXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1300	pdcs->rxregs_64 = (struct dma64_regs *)
1301	    (((u8 *)pdcs->pdc_reg_vbase) +
1302		     PDC_RXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1303
1304	pdcs->ntxd = PDC_RING_ENTRIES;
1305	pdcs->nrxd = PDC_RING_ENTRIES;
1306	pdcs->ntxpost = PDC_RING_ENTRIES - 1;
1307	pdcs->nrxpost = PDC_RING_ENTRIES - 1;
1308	iowrite32(0, &pdcs->regs->intmask);
1309
1310	dma_reg = &pdcs->regs->dmaregs[ringset];
1311
1312	/* Configure DMA but will enable later in pdc_ring_init() */
1313	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1314
1315	iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1316		  &dma_reg->dmarcv.control);
1317
1318	/* Reset current index pointers after making sure DMA is disabled */
1319	iowrite32(0, &dma_reg->dmaxmt.ptr);
1320	iowrite32(0, &dma_reg->dmarcv.ptr);
1321
1322	if (pdcs->pdc_resp_hdr_len == PDC_SPU2_RESP_HDR_LEN)
1323		iowrite32(PDC_CKSUM_CTRL,
1324			  pdcs->pdc_reg_vbase + PDC_CKSUM_CTRL_OFFSET);
1325}
1326
1327/**
1328 * pdc_hw_disable() - Disable the tx and rx control in the hw.
1329 * @pdcs: PDC state structure
1330 *
1331 */
1332static void pdc_hw_disable(struct pdc_state *pdcs)
1333{
1334	struct dma64 *dma_reg;
1335
1336	dma_reg = &pdcs->regs->dmaregs[PDC_RINGSET];
1337	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1338	iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1339		  &dma_reg->dmarcv.control);
1340}
1341
1342/**
1343 * pdc_rx_buf_pool_create() - Pool of receive buffers used to catch the metadata
1344 * header returned with each response message.
1345 * @pdcs: PDC state structure
1346 *
1347 * The metadata is not returned to the mailbox client. So the PDC driver
1348 * manages these buffers.
1349 *
1350 * Return: PDC_SUCCESS
1351 *         -ENOMEM if pool creation fails
1352 */
1353static int pdc_rx_buf_pool_create(struct pdc_state *pdcs)
1354{
1355	struct platform_device *pdev;
1356	struct device *dev;
1357
1358	pdev = pdcs->pdev;
1359	dev = &pdev->dev;
1360
1361	pdcs->pdc_resp_hdr_len = pdcs->rx_status_len;
1362	if (pdcs->use_bcm_hdr)
1363		pdcs->pdc_resp_hdr_len += BCM_HDR_LEN;
1364
1365	pdcs->rx_buf_pool = dma_pool_create("pdc rx bufs", dev,
1366					    pdcs->pdc_resp_hdr_len,
1367					    RX_BUF_ALIGN, 0);
1368	if (!pdcs->rx_buf_pool)
1369		return -ENOMEM;
1370
1371	return PDC_SUCCESS;
1372}
1373
1374/**
1375 * pdc_interrupts_init() - Initialize the interrupt configuration for a PDC and
1376 * specify a threaded IRQ handler for deferred handling of interrupts outside of
1377 * interrupt context.
1378 * @pdcs:   PDC state
1379 *
1380 * Set the interrupt mask for transmit and receive done.
1381 * Set the lazy interrupt frame count to generate an interrupt for just one pkt.
1382 *
1383 * Return:  PDC_SUCCESS
1384 *          <0 if threaded irq request fails
1385 */
1386static int pdc_interrupts_init(struct pdc_state *pdcs)
1387{
1388	struct platform_device *pdev = pdcs->pdev;
1389	struct device *dev = &pdev->dev;
1390	struct device_node *dn = pdev->dev.of_node;
1391	int err;
1392
1393	/* interrupt configuration */
1394	iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
1395
1396	if (pdcs->hw_type == FA_HW)
1397		iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1398			  FA_RCVLAZY0_OFFSET);
1399	else
1400		iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1401			  PDC_RCVLAZY0_OFFSET);
1402
1403	/* read irq from device tree */
1404	pdcs->pdc_irq = irq_of_parse_and_map(dn, 0);
1405	dev_dbg(dev, "pdc device %s irq %u for pdcs %p",
1406		dev_name(dev), pdcs->pdc_irq, pdcs);
1407
1408	err = devm_request_irq(dev, pdcs->pdc_irq, pdc_irq_handler, 0,
1409			       dev_name(dev), dev);
1410	if (err) {
1411		dev_err(dev, "IRQ %u request failed with err %d\n",
1412			pdcs->pdc_irq, err);
1413		return err;
1414	}
1415	return PDC_SUCCESS;
1416}
1417
1418static const struct mbox_chan_ops pdc_mbox_chan_ops = {
1419	.send_data = pdc_send_data,
1420	.last_tx_done = pdc_last_tx_done,
1421	.startup = pdc_startup,
1422	.shutdown = pdc_shutdown
1423};
1424
1425/**
1426 * pdc_mb_init() - Initialize the mailbox controller.
1427 * @pdcs:  PDC state
1428 *
1429 * Each PDC is a mailbox controller. Each ringset is a mailbox channel. Kernel
1430 * driver only uses one ringset and thus one mb channel. PDC uses the transmit
1431 * complete interrupt to determine when a mailbox message has successfully been
1432 * transmitted.
1433 *
1434 * Return: 0 on success
1435 *         < 0 if there is an allocation or registration failure
1436 */
1437static int pdc_mb_init(struct pdc_state *pdcs)
1438{
1439	struct device *dev = &pdcs->pdev->dev;
1440	struct mbox_controller *mbc;
1441	int chan_index;
1442	int err;
1443
1444	mbc = &pdcs->mbc;
1445	mbc->dev = dev;
1446	mbc->ops = &pdc_mbox_chan_ops;
1447	mbc->num_chans = 1;
1448	mbc->chans = devm_kcalloc(dev, mbc->num_chans, sizeof(*mbc->chans),
1449				  GFP_KERNEL);
1450	if (!mbc->chans)
1451		return -ENOMEM;
1452
1453	mbc->txdone_irq = false;
1454	mbc->txdone_poll = true;
1455	mbc->txpoll_period = 1;
1456	for (chan_index = 0; chan_index < mbc->num_chans; chan_index++)
1457		mbc->chans[chan_index].con_priv = pdcs;
1458
1459	/* Register mailbox controller */
1460	err = devm_mbox_controller_register(dev, mbc);
1461	if (err) {
1462		dev_crit(dev,
1463			 "Failed to register PDC mailbox controller. Error %d.",
1464			 err);
1465		return err;
1466	}
1467	return 0;
1468}
1469
1470/* Device tree API */
1471static const int pdc_hw = PDC_HW;
1472static const int fa_hw = FA_HW;
1473
1474static const struct of_device_id pdc_mbox_of_match[] = {
1475	{.compatible = "brcm,iproc-pdc-mbox", .data = &pdc_hw},
1476	{.compatible = "brcm,iproc-fa2-mbox", .data = &fa_hw},
1477	{ /* sentinel */ }
1478};
1479MODULE_DEVICE_TABLE(of, pdc_mbox_of_match);
1480
1481/**
1482 * pdc_dt_read() - Read application-specific data from device tree.
1483 * @pdev:  Platform device
1484 * @pdcs:  PDC state
1485 *
1486 * Reads the number of bytes of receive status that precede each received frame.
1487 * Reads whether transmit and received frames should be preceded by an 8-byte
1488 * BCM header.
1489 *
1490 * Return: 0 if successful
1491 *         -ENODEV if device not available
1492 */
1493static int pdc_dt_read(struct platform_device *pdev, struct pdc_state *pdcs)
1494{
1495	struct device *dev = &pdev->dev;
1496	struct device_node *dn = pdev->dev.of_node;
1497	const struct of_device_id *match;
1498	const int *hw_type;
1499	int err;
1500
1501	err = of_property_read_u32(dn, "brcm,rx-status-len",
1502				   &pdcs->rx_status_len);
1503	if (err < 0)
1504		dev_err(dev,
1505			"%s failed to get DMA receive status length from device tree",
1506			__func__);
1507
1508	pdcs->use_bcm_hdr = of_property_read_bool(dn, "brcm,use-bcm-hdr");
1509
1510	pdcs->hw_type = PDC_HW;
1511
1512	match = of_match_device(of_match_ptr(pdc_mbox_of_match), dev);
1513	if (match != NULL) {
1514		hw_type = match->data;
1515		pdcs->hw_type = *hw_type;
1516	}
1517
1518	return 0;
1519}
1520
1521/**
1522 * pdc_probe() - Probe function for PDC driver.
1523 * @pdev:   PDC platform device
1524 *
1525 * Reserve and map register regions defined in device tree.
1526 * Allocate and initialize tx and rx DMA rings.
1527 * Initialize a mailbox controller for each PDC.
1528 *
1529 * Return: 0 if successful
1530 *         < 0 if an error
1531 */
1532static int pdc_probe(struct platform_device *pdev)
1533{
1534	int err = 0;
1535	struct device *dev = &pdev->dev;
1536	struct resource *pdc_regs;
1537	struct pdc_state *pdcs;
1538
1539	/* PDC state for one SPU */
1540	pdcs = devm_kzalloc(dev, sizeof(*pdcs), GFP_KERNEL);
1541	if (!pdcs) {
1542		err = -ENOMEM;
1543		goto cleanup;
1544	}
1545
1546	pdcs->pdev = pdev;
1547	platform_set_drvdata(pdev, pdcs);
1548	pdcs->pdc_idx = pdcg.num_spu;
1549	pdcg.num_spu++;
1550
1551	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(39));
1552	if (err) {
1553		dev_warn(dev, "PDC device cannot perform DMA. Error %d.", err);
1554		goto cleanup;
1555	}
1556
1557	/* Create DMA pool for tx ring */
1558	pdcs->ring_pool = dma_pool_create("pdc rings", dev, PDC_RING_SIZE,
1559					  RING_ALIGN, 0);
1560	if (!pdcs->ring_pool) {
1561		err = -ENOMEM;
1562		goto cleanup;
1563	}
1564
1565	err = pdc_dt_read(pdev, pdcs);
1566	if (err)
1567		goto cleanup_ring_pool;
1568
1569	pdc_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1570	if (!pdc_regs) {
1571		err = -ENODEV;
1572		goto cleanup_ring_pool;
1573	}
1574	dev_dbg(dev, "PDC register region res.start = %pa, res.end = %pa",
1575		&pdc_regs->start, &pdc_regs->end);
1576
1577	pdcs->pdc_reg_vbase = devm_ioremap_resource(&pdev->dev, pdc_regs);
1578	if (IS_ERR(pdcs->pdc_reg_vbase)) {
1579		err = PTR_ERR(pdcs->pdc_reg_vbase);
1580		dev_err(&pdev->dev, "Failed to map registers: %d\n", err);
1581		goto cleanup_ring_pool;
1582	}
1583
1584	/* create rx buffer pool after dt read to know how big buffers are */
1585	err = pdc_rx_buf_pool_create(pdcs);
1586	if (err)
1587		goto cleanup_ring_pool;
1588
1589	pdc_hw_init(pdcs);
1590
1591	/* Init tasklet for deferred DMA rx processing */
1592	tasklet_setup(&pdcs->rx_tasklet, pdc_tasklet_cb);
1593
1594	err = pdc_interrupts_init(pdcs);
1595	if (err)
1596		goto cleanup_buf_pool;
1597
1598	/* Initialize mailbox controller */
1599	err = pdc_mb_init(pdcs);
1600	if (err)
1601		goto cleanup_buf_pool;
1602
1603	pdc_setup_debugfs(pdcs);
1604
1605	dev_dbg(dev, "pdc_probe() successful");
1606	return PDC_SUCCESS;
1607
1608cleanup_buf_pool:
1609	tasklet_kill(&pdcs->rx_tasklet);
1610	dma_pool_destroy(pdcs->rx_buf_pool);
1611
1612cleanup_ring_pool:
1613	dma_pool_destroy(pdcs->ring_pool);
1614
1615cleanup:
1616	return err;
1617}
1618
1619static int pdc_remove(struct platform_device *pdev)
1620{
1621	struct pdc_state *pdcs = platform_get_drvdata(pdev);
1622
1623	pdc_free_debugfs();
1624
1625	tasklet_kill(&pdcs->rx_tasklet);
1626
1627	pdc_hw_disable(pdcs);
1628
1629	dma_pool_destroy(pdcs->rx_buf_pool);
1630	dma_pool_destroy(pdcs->ring_pool);
1631	return 0;
1632}
1633
1634static struct platform_driver pdc_mbox_driver = {
1635	.probe = pdc_probe,
1636	.remove = pdc_remove,
1637	.driver = {
1638		   .name = "brcm-iproc-pdc-mbox",
1639		   .of_match_table = of_match_ptr(pdc_mbox_of_match),
1640		   },
1641};
1642module_platform_driver(pdc_mbox_driver);
1643
1644MODULE_AUTHOR("Rob Rice <rob.rice@broadcom.com>");
1645MODULE_DESCRIPTION("Broadcom PDC mailbox driver");
1646MODULE_LICENSE("GPL v2");
1647