xref: /linux/drivers/mailbox/bcm-pdc-mailbox.c (revision 352bce2ee19f9d479832c060548974da2ba51153)
1 /*
2  * Copyright 2016 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License, version 2, as
6  * published by the Free Software Foundation (the "GPL").
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License version 2 (GPLv2) for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * version 2 (GPLv2) along with this source code.
15  */
16 
17 /*
18  * Broadcom PDC Mailbox Driver
19  * The PDC provides a ring based programming interface to one or more hardware
20  * offload engines. For example, the PDC driver works with both SPU-M and SPU2
21  * cryptographic offload hardware. In some chips the PDC is referred to as MDE,
22  * and in others the FA2/FA+ hardware is used with this PDC driver.
23  *
24  * The PDC driver registers with the Linux mailbox framework as a mailbox
25  * controller, once for each PDC instance. Ring 0 for each PDC is registered as
26  * a mailbox channel. The PDC driver uses interrupts to determine when data
27  * transfers to and from an offload engine are complete. The PDC driver uses
28  * threaded IRQs so that response messages are handled outside of interrupt
29  * context.
30  *
31  * The PDC driver allows multiple messages to be pending in the descriptor
32  * rings. The tx_msg_start descriptor index indicates where the last message
33  * starts. The txin_numd value at this index indicates how many descriptor
34  * indexes make up the message. Similar state is kept on the receive side. When
35  * an rx interrupt indicates a response is ready, the PDC driver processes numd
36  * descriptors from the tx and rx ring, thus processing one response at a time.
37  */
38 
39 #include <linux/errno.h>
40 #include <linux/module.h>
41 #include <linux/init.h>
42 #include <linux/slab.h>
43 #include <linux/debugfs.h>
44 #include <linux/interrupt.h>
45 #include <linux/wait.h>
46 #include <linux/platform_device.h>
47 #include <linux/io.h>
48 #include <linux/of.h>
49 #include <linux/of_device.h>
50 #include <linux/of_address.h>
51 #include <linux/of_irq.h>
52 #include <linux/mailbox_controller.h>
53 #include <linux/mailbox/brcm-message.h>
54 #include <linux/scatterlist.h>
55 #include <linux/dma-direction.h>
56 #include <linux/dma-mapping.h>
57 #include <linux/dmapool.h>
58 
59 #define PDC_SUCCESS  0
60 
61 #define RING_ENTRY_SIZE   sizeof(struct dma64dd)
62 
63 /* # entries in PDC dma ring */
64 #define PDC_RING_ENTRIES  512
65 /*
66  * Minimum number of ring descriptor entries that must be free to tell mailbox
67  * framework that it can submit another request
68  */
69 #define PDC_RING_SPACE_MIN  15
70 
71 #define PDC_RING_SIZE    (PDC_RING_ENTRIES * RING_ENTRY_SIZE)
72 /* Rings are 8k aligned */
73 #define RING_ALIGN_ORDER  13
74 #define RING_ALIGN        BIT(RING_ALIGN_ORDER)
75 
76 #define RX_BUF_ALIGN_ORDER  5
77 #define RX_BUF_ALIGN	    BIT(RX_BUF_ALIGN_ORDER)
78 
79 /* descriptor bumping macros */
80 #define XXD(x, max_mask)              ((x) & (max_mask))
81 #define TXD(x, max_mask)              XXD((x), (max_mask))
82 #define RXD(x, max_mask)              XXD((x), (max_mask))
83 #define NEXTTXD(i, max_mask)          TXD((i) + 1, (max_mask))
84 #define PREVTXD(i, max_mask)          TXD((i) - 1, (max_mask))
85 #define NEXTRXD(i, max_mask)          RXD((i) + 1, (max_mask))
86 #define PREVRXD(i, max_mask)          RXD((i) - 1, (max_mask))
87 #define NTXDACTIVE(h, t, max_mask)    TXD((t) - (h), (max_mask))
88 #define NRXDACTIVE(h, t, max_mask)    RXD((t) - (h), (max_mask))
89 
90 /* Length of BCM header at start of SPU msg, in bytes */
91 #define BCM_HDR_LEN  8
92 
93 /*
94  * PDC driver reserves ringset 0 on each SPU for its own use. The driver does
95  * not currently support use of multiple ringsets on a single PDC engine.
96  */
97 #define PDC_RINGSET  0
98 
99 /*
100  * Interrupt mask and status definitions. Enable interrupts for tx and rx on
101  * ring 0
102  */
103 #define PDC_RCVINT_0         (16 + PDC_RINGSET)
104 #define PDC_RCVINTEN_0       BIT(PDC_RCVINT_0)
105 #define PDC_INTMASK	     (PDC_RCVINTEN_0)
106 #define PDC_LAZY_FRAMECOUNT  1
107 #define PDC_LAZY_TIMEOUT     10000
108 #define PDC_LAZY_INT  (PDC_LAZY_TIMEOUT | (PDC_LAZY_FRAMECOUNT << 24))
109 #define PDC_INTMASK_OFFSET   0x24
110 #define PDC_INTSTATUS_OFFSET 0x20
111 #define PDC_RCVLAZY0_OFFSET  (0x30 + 4 * PDC_RINGSET)
112 #define FA_RCVLAZY0_OFFSET   0x100
113 
114 /*
115  * For SPU2, configure MDE_CKSUM_CONTROL to write 17 bytes of metadata
116  * before frame
117  */
118 #define PDC_SPU2_RESP_HDR_LEN  17
119 #define PDC_CKSUM_CTRL         BIT(27)
120 #define PDC_CKSUM_CTRL_OFFSET  0x400
121 
122 #define PDC_SPUM_RESP_HDR_LEN  32
123 
124 /*
125  * Sets the following bits for write to transmit control reg:
126  * 11    - PtyChkDisable - parity check is disabled
127  * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
128  */
129 #define PDC_TX_CTL		0x000C0800
130 
131 /* Bit in tx control reg to enable tx channel */
132 #define PDC_TX_ENABLE		0x1
133 
134 /*
135  * Sets the following bits for write to receive control reg:
136  * 7:1   - RcvOffset - size in bytes of status region at start of rx frame buf
137  * 9     - SepRxHdrDescEn - place start of new frames only in descriptors
138  *                          that have StartOfFrame set
139  * 10    - OflowContinue - on rx FIFO overflow, clear rx fifo, discard all
140  *                         remaining bytes in current frame, report error
141  *                         in rx frame status for current frame
142  * 11    - PtyChkDisable - parity check is disabled
143  * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
144  */
145 #define PDC_RX_CTL		0x000C0E00
146 
147 /* Bit in rx control reg to enable rx channel */
148 #define PDC_RX_ENABLE		0x1
149 
150 #define CRYPTO_D64_RS0_CD_MASK   ((PDC_RING_ENTRIES * RING_ENTRY_SIZE) - 1)
151 
152 /* descriptor flags */
153 #define D64_CTRL1_EOT   BIT(28)	/* end of descriptor table */
154 #define D64_CTRL1_IOC   BIT(29)	/* interrupt on complete */
155 #define D64_CTRL1_EOF   BIT(30)	/* end of frame */
156 #define D64_CTRL1_SOF   BIT(31)	/* start of frame */
157 
158 #define RX_STATUS_OVERFLOW       0x00800000
159 #define RX_STATUS_LEN            0x0000FFFF
160 
161 #define PDC_TXREGS_OFFSET  0x200
162 #define PDC_RXREGS_OFFSET  0x220
163 
164 /* Maximum size buffer the DMA engine can handle */
165 #define PDC_DMA_BUF_MAX 16384
166 
167 enum pdc_hw {
168 	FA_HW,		/* FA2/FA+ hardware (i.e. Northstar Plus) */
169 	PDC_HW		/* PDC/MDE hardware (i.e. Northstar 2, Pegasus) */
170 };
171 
172 struct pdc_dma_map {
173 	void *ctx;          /* opaque context associated with frame */
174 };
175 
176 /* dma descriptor */
177 struct dma64dd {
178 	u32 ctrl1;      /* misc control bits */
179 	u32 ctrl2;      /* buffer count and address extension */
180 	u32 addrlow;    /* memory address of the date buffer, bits 31:0 */
181 	u32 addrhigh;   /* memory address of the date buffer, bits 63:32 */
182 };
183 
184 /* dma registers per channel(xmt or rcv) */
185 struct dma64_regs {
186 	u32  control;   /* enable, et al */
187 	u32  ptr;       /* last descriptor posted to chip */
188 	u32  addrlow;   /* descriptor ring base address low 32-bits */
189 	u32  addrhigh;  /* descriptor ring base address bits 63:32 */
190 	u32  status0;   /* last rx descriptor written by hw */
191 	u32  status1;   /* driver does not use */
192 };
193 
194 /* cpp contortions to concatenate w/arg prescan */
195 #ifndef PAD
196 #define _PADLINE(line)  pad ## line
197 #define _XSTR(line)     _PADLINE(line)
198 #define PAD             _XSTR(__LINE__)
199 #endif  /* PAD */
200 
201 /* dma registers. matches hw layout. */
202 struct dma64 {
203 	struct dma64_regs dmaxmt;  /* dma tx */
204 	u32          PAD[2];
205 	struct dma64_regs dmarcv;  /* dma rx */
206 	u32          PAD[2];
207 };
208 
209 /* PDC registers */
210 struct pdc_regs {
211 	u32  devcontrol;             /* 0x000 */
212 	u32  devstatus;              /* 0x004 */
213 	u32  PAD;
214 	u32  biststatus;             /* 0x00c */
215 	u32  PAD[4];
216 	u32  intstatus;              /* 0x020 */
217 	u32  intmask;                /* 0x024 */
218 	u32  gptimer;                /* 0x028 */
219 
220 	u32  PAD;
221 	u32  intrcvlazy_0;           /* 0x030 (Only in PDC, not FA2) */
222 	u32  intrcvlazy_1;           /* 0x034 (Only in PDC, not FA2) */
223 	u32  intrcvlazy_2;           /* 0x038 (Only in PDC, not FA2) */
224 	u32  intrcvlazy_3;           /* 0x03c (Only in PDC, not FA2) */
225 
226 	u32  PAD[48];
227 	u32  fa_intrecvlazy;         /* 0x100 (Only in FA2, not PDC) */
228 	u32  flowctlthresh;          /* 0x104 */
229 	u32  wrrthresh;              /* 0x108 */
230 	u32  gmac_idle_cnt_thresh;   /* 0x10c */
231 
232 	u32  PAD[4];
233 	u32  ifioaccessaddr;         /* 0x120 */
234 	u32  ifioaccessbyte;         /* 0x124 */
235 	u32  ifioaccessdata;         /* 0x128 */
236 
237 	u32  PAD[21];
238 	u32  phyaccess;              /* 0x180 */
239 	u32  PAD;
240 	u32  phycontrol;             /* 0x188 */
241 	u32  txqctl;                 /* 0x18c */
242 	u32  rxqctl;                 /* 0x190 */
243 	u32  gpioselect;             /* 0x194 */
244 	u32  gpio_output_en;         /* 0x198 */
245 	u32  PAD;                    /* 0x19c */
246 	u32  txq_rxq_mem_ctl;        /* 0x1a0 */
247 	u32  memory_ecc_status;      /* 0x1a4 */
248 	u32  serdes_ctl;             /* 0x1a8 */
249 	u32  serdes_status0;         /* 0x1ac */
250 	u32  serdes_status1;         /* 0x1b0 */
251 	u32  PAD[11];                /* 0x1b4-1dc */
252 	u32  clk_ctl_st;             /* 0x1e0 */
253 	u32  hw_war;                 /* 0x1e4 (Only in PDC, not FA2) */
254 	u32  pwrctl;                 /* 0x1e8 */
255 	u32  PAD[5];
256 
257 #define PDC_NUM_DMA_RINGS   4
258 	struct dma64 dmaregs[PDC_NUM_DMA_RINGS];  /* 0x0200 - 0x2fc */
259 
260 	/* more registers follow, but we don't use them */
261 };
262 
263 /* structure for allocating/freeing DMA rings */
264 struct pdc_ring_alloc {
265 	dma_addr_t  dmabase; /* DMA address of start of ring */
266 	void	   *vbase;   /* base kernel virtual address of ring */
267 	u32	    size;    /* ring allocation size in bytes */
268 };
269 
270 /*
271  * context associated with a receive descriptor.
272  * @rxp_ctx: opaque context associated with frame that starts at each
273  *           rx ring index.
274  * @dst_sg:  Scatterlist used to form reply frames beginning at a given ring
275  *           index. Retained in order to unmap each sg after reply is processed.
276  * @rxin_numd: Number of rx descriptors associated with the message that starts
277  *             at a descriptor index. Not set for every index. For example,
278  *             if descriptor index i points to a scatterlist with 4 entries,
279  *             then the next three descriptor indexes don't have a value set.
280  * @resp_hdr: Virtual address of buffer used to catch DMA rx status
281  * @resp_hdr_daddr: physical address of DMA rx status buffer
282  */
283 struct pdc_rx_ctx {
284 	void *rxp_ctx;
285 	struct scatterlist *dst_sg;
286 	u32  rxin_numd;
287 	void *resp_hdr;
288 	dma_addr_t resp_hdr_daddr;
289 };
290 
291 /* PDC state structure */
292 struct pdc_state {
293 	/* Index of the PDC whose state is in this structure instance */
294 	u8 pdc_idx;
295 
296 	/* Platform device for this PDC instance */
297 	struct platform_device *pdev;
298 
299 	/*
300 	 * Each PDC instance has a mailbox controller. PDC receives request
301 	 * messages through mailboxes, and sends response messages through the
302 	 * mailbox framework.
303 	 */
304 	struct mbox_controller mbc;
305 
306 	unsigned int pdc_irq;
307 
308 	/* tasklet for deferred processing after DMA rx interrupt */
309 	struct tasklet_struct rx_tasklet;
310 
311 	/* Number of bytes of receive status prior to each rx frame */
312 	u32 rx_status_len;
313 	/* Whether a BCM header is prepended to each frame */
314 	bool use_bcm_hdr;
315 	/* Sum of length of BCM header and rx status header */
316 	u32 pdc_resp_hdr_len;
317 
318 	/* The base virtual address of DMA hw registers */
319 	void __iomem *pdc_reg_vbase;
320 
321 	/* Pool for allocation of DMA rings */
322 	struct dma_pool *ring_pool;
323 
324 	/* Pool for allocation of metadata buffers for response messages */
325 	struct dma_pool *rx_buf_pool;
326 
327 	/*
328 	 * The base virtual address of DMA tx/rx descriptor rings. Corresponding
329 	 * DMA address and size of ring allocation.
330 	 */
331 	struct pdc_ring_alloc tx_ring_alloc;
332 	struct pdc_ring_alloc rx_ring_alloc;
333 
334 	struct pdc_regs *regs;    /* start of PDC registers */
335 
336 	struct dma64_regs *txregs_64; /* dma tx engine registers */
337 	struct dma64_regs *rxregs_64; /* dma rx engine registers */
338 
339 	/*
340 	 * Arrays of PDC_RING_ENTRIES descriptors
341 	 * To use multiple ringsets, this needs to be extended
342 	 */
343 	struct dma64dd   *txd_64;  /* tx descriptor ring */
344 	struct dma64dd   *rxd_64;  /* rx descriptor ring */
345 
346 	/* descriptor ring sizes */
347 	u32      ntxd;       /* # tx descriptors */
348 	u32      nrxd;       /* # rx descriptors */
349 	u32      nrxpost;    /* # rx buffers to keep posted */
350 	u32      ntxpost;    /* max number of tx buffers that can be posted */
351 
352 	/*
353 	 * Index of next tx descriptor to reclaim. That is, the descriptor
354 	 * index of the oldest tx buffer for which the host has yet to process
355 	 * the corresponding response.
356 	 */
357 	u32  txin;
358 
359 	/*
360 	 * Index of the first receive descriptor for the sequence of
361 	 * message fragments currently under construction. Used to build up
362 	 * the rxin_numd count for a message. Updated to rxout when the host
363 	 * starts a new sequence of rx buffers for a new message.
364 	 */
365 	u32  tx_msg_start;
366 
367 	/* Index of next tx descriptor to post. */
368 	u32  txout;
369 
370 	/*
371 	 * Number of tx descriptors associated with the message that starts
372 	 * at this tx descriptor index.
373 	 */
374 	u32      txin_numd[PDC_RING_ENTRIES];
375 
376 	/*
377 	 * Index of next rx descriptor to reclaim. This is the index of
378 	 * the next descriptor whose data has yet to be processed by the host.
379 	 */
380 	u32  rxin;
381 
382 	/*
383 	 * Index of the first receive descriptor for the sequence of
384 	 * message fragments currently under construction. Used to build up
385 	 * the rxin_numd count for a message. Updated to rxout when the host
386 	 * starts a new sequence of rx buffers for a new message.
387 	 */
388 	u32  rx_msg_start;
389 
390 	/*
391 	 * Saved value of current hardware rx descriptor index.
392 	 * The last rx buffer written by the hw is the index previous to
393 	 * this one.
394 	 */
395 	u32  last_rx_curr;
396 
397 	/* Index of next rx descriptor to post. */
398 	u32  rxout;
399 
400 	struct pdc_rx_ctx rx_ctx[PDC_RING_ENTRIES];
401 
402 	/*
403 	 * Scatterlists used to form request and reply frames beginning at a
404 	 * given ring index. Retained in order to unmap each sg after reply
405 	 * is processed
406 	 */
407 	struct scatterlist *src_sg[PDC_RING_ENTRIES];
408 
409 	/* counters */
410 	u32  pdc_requests;     /* number of request messages submitted */
411 	u32  pdc_replies;      /* number of reply messages received */
412 	u32  last_tx_not_done; /* too few tx descriptors to indicate done */
413 	u32  tx_ring_full;     /* unable to accept msg because tx ring full */
414 	u32  rx_ring_full;     /* unable to accept msg because rx ring full */
415 	u32  txnobuf;          /* unable to create tx descriptor */
416 	u32  rxnobuf;          /* unable to create rx descriptor */
417 	u32  rx_oflow;         /* count of rx overflows */
418 
419 	/* hardware type - FA2 or PDC/MDE */
420 	enum pdc_hw hw_type;
421 };
422 
423 /* Global variables */
424 
425 struct pdc_globals {
426 	/* Actual number of SPUs in hardware, as reported by device tree */
427 	u32 num_spu;
428 };
429 
430 static struct pdc_globals pdcg;
431 
432 /* top level debug FS directory for PDC driver */
433 static struct dentry *debugfs_dir;
434 
435 static ssize_t pdc_debugfs_read(struct file *filp, char __user *ubuf,
436 				size_t count, loff_t *offp)
437 {
438 	struct pdc_state *pdcs;
439 	char *buf;
440 	ssize_t ret, out_offset, out_count;
441 
442 	out_count = 512;
443 
444 	buf = kmalloc(out_count, GFP_KERNEL);
445 	if (!buf)
446 		return -ENOMEM;
447 
448 	pdcs = filp->private_data;
449 	out_offset = 0;
450 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
451 			       "SPU %u stats:\n", pdcs->pdc_idx);
452 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
453 			       "PDC requests....................%u\n",
454 			       pdcs->pdc_requests);
455 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
456 			       "PDC responses...................%u\n",
457 			       pdcs->pdc_replies);
458 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
459 			       "Tx not done.....................%u\n",
460 			       pdcs->last_tx_not_done);
461 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
462 			       "Tx ring full....................%u\n",
463 			       pdcs->tx_ring_full);
464 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
465 			       "Rx ring full....................%u\n",
466 			       pdcs->rx_ring_full);
467 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
468 			       "Tx desc write fail. Ring full...%u\n",
469 			       pdcs->txnobuf);
470 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
471 			       "Rx desc write fail. Ring full...%u\n",
472 			       pdcs->rxnobuf);
473 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
474 			       "Receive overflow................%u\n",
475 			       pdcs->rx_oflow);
476 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
477 			       "Num frags in rx ring............%u\n",
478 			       NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr,
479 					  pdcs->nrxpost));
480 
481 	if (out_offset > out_count)
482 		out_offset = out_count;
483 
484 	ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset);
485 	kfree(buf);
486 	return ret;
487 }
488 
489 static const struct file_operations pdc_debugfs_stats = {
490 	.owner = THIS_MODULE,
491 	.open = simple_open,
492 	.read = pdc_debugfs_read,
493 };
494 
495 /**
496  * pdc_setup_debugfs() - Create the debug FS directories. If the top-level
497  * directory has not yet been created, create it now. Create a stats file in
498  * this directory for a SPU.
499  * @pdcs: PDC state structure
500  */
501 static void pdc_setup_debugfs(struct pdc_state *pdcs)
502 {
503 	char spu_stats_name[16];
504 
505 	if (!debugfs_initialized())
506 		return;
507 
508 	snprintf(spu_stats_name, 16, "pdc%d_stats", pdcs->pdc_idx);
509 	if (!debugfs_dir)
510 		debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
511 
512 	/* S_IRUSR == 0400 */
513 	debugfs_create_file(spu_stats_name, 0400, debugfs_dir, pdcs,
514 			    &pdc_debugfs_stats);
515 }
516 
517 static void pdc_free_debugfs(void)
518 {
519 	debugfs_remove_recursive(debugfs_dir);
520 	debugfs_dir = NULL;
521 }
522 
523 /**
524  * pdc_build_rxd() - Build DMA descriptor to receive SPU result.
525  * @pdcs:      PDC state for SPU that will generate result
526  * @dma_addr:  DMA address of buffer that descriptor is being built for
527  * @buf_len:   Length of the receive buffer, in bytes
528  * @flags:     Flags to be stored in descriptor
529  */
530 static inline void
531 pdc_build_rxd(struct pdc_state *pdcs, dma_addr_t dma_addr,
532 	      u32 buf_len, u32 flags)
533 {
534 	struct device *dev = &pdcs->pdev->dev;
535 	struct dma64dd *rxd = &pdcs->rxd_64[pdcs->rxout];
536 
537 	dev_dbg(dev,
538 		"Writing rx descriptor for PDC %u at index %u with length %u. flags %#x\n",
539 		pdcs->pdc_idx, pdcs->rxout, buf_len, flags);
540 
541 	rxd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
542 	rxd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
543 	rxd->ctrl1 = cpu_to_le32(flags);
544 	rxd->ctrl2 = cpu_to_le32(buf_len);
545 
546 	/* bump ring index and return */
547 	pdcs->rxout = NEXTRXD(pdcs->rxout, pdcs->nrxpost);
548 }
549 
550 /**
551  * pdc_build_txd() - Build a DMA descriptor to transmit a SPU request to
552  * hardware.
553  * @pdcs:        PDC state for the SPU that will process this request
554  * @dma_addr:    DMA address of packet to be transmitted
555  * @buf_len:     Length of tx buffer, in bytes
556  * @flags:       Flags to be stored in descriptor
557  */
558 static inline void
559 pdc_build_txd(struct pdc_state *pdcs, dma_addr_t dma_addr, u32 buf_len,
560 	      u32 flags)
561 {
562 	struct device *dev = &pdcs->pdev->dev;
563 	struct dma64dd *txd = &pdcs->txd_64[pdcs->txout];
564 
565 	dev_dbg(dev,
566 		"Writing tx descriptor for PDC %u at index %u with length %u, flags %#x\n",
567 		pdcs->pdc_idx, pdcs->txout, buf_len, flags);
568 
569 	txd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
570 	txd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
571 	txd->ctrl1 = cpu_to_le32(flags);
572 	txd->ctrl2 = cpu_to_le32(buf_len);
573 
574 	/* bump ring index and return */
575 	pdcs->txout = NEXTTXD(pdcs->txout, pdcs->ntxpost);
576 }
577 
578 /**
579  * pdc_receive_one() - Receive a response message from a given SPU.
580  * @pdcs:    PDC state for the SPU to receive from
581  *
582  * When the return code indicates success, the response message is available in
583  * the receive buffers provided prior to submission of the request.
584  *
585  * Return:  PDC_SUCCESS if one or more receive descriptors was processed
586  *          -EAGAIN indicates that no response message is available
587  *          -EIO an error occurred
588  */
589 static int
590 pdc_receive_one(struct pdc_state *pdcs)
591 {
592 	struct device *dev = &pdcs->pdev->dev;
593 	struct mbox_controller *mbc;
594 	struct mbox_chan *chan;
595 	struct brcm_message mssg;
596 	u32 len, rx_status;
597 	u32 num_frags;
598 	u8 *resp_hdr;    /* virtual addr of start of resp message DMA header */
599 	u32 frags_rdy;   /* number of fragments ready to read */
600 	u32 rx_idx;      /* ring index of start of receive frame */
601 	dma_addr_t resp_hdr_daddr;
602 	struct pdc_rx_ctx *rx_ctx;
603 
604 	mbc = &pdcs->mbc;
605 	chan = &mbc->chans[0];
606 	mssg.type = BRCM_MESSAGE_SPU;
607 
608 	/*
609 	 * return if a complete response message is not yet ready.
610 	 * rxin_numd[rxin] is the number of fragments in the next msg
611 	 * to read.
612 	 */
613 	frags_rdy = NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr, pdcs->nrxpost);
614 	if ((frags_rdy == 0) ||
615 	    (frags_rdy < pdcs->rx_ctx[pdcs->rxin].rxin_numd))
616 		/* No response ready */
617 		return -EAGAIN;
618 
619 	num_frags = pdcs->txin_numd[pdcs->txin];
620 	WARN_ON(num_frags == 0);
621 
622 	dma_unmap_sg(dev, pdcs->src_sg[pdcs->txin],
623 		     sg_nents(pdcs->src_sg[pdcs->txin]), DMA_TO_DEVICE);
624 
625 	pdcs->txin = (pdcs->txin + num_frags) & pdcs->ntxpost;
626 
627 	dev_dbg(dev, "PDC %u reclaimed %d tx descriptors",
628 		pdcs->pdc_idx, num_frags);
629 
630 	rx_idx = pdcs->rxin;
631 	rx_ctx = &pdcs->rx_ctx[rx_idx];
632 	num_frags = rx_ctx->rxin_numd;
633 	/* Return opaque context with result */
634 	mssg.ctx = rx_ctx->rxp_ctx;
635 	rx_ctx->rxp_ctx = NULL;
636 	resp_hdr = rx_ctx->resp_hdr;
637 	resp_hdr_daddr = rx_ctx->resp_hdr_daddr;
638 	dma_unmap_sg(dev, rx_ctx->dst_sg, sg_nents(rx_ctx->dst_sg),
639 		     DMA_FROM_DEVICE);
640 
641 	pdcs->rxin = (pdcs->rxin + num_frags) & pdcs->nrxpost;
642 
643 	dev_dbg(dev, "PDC %u reclaimed %d rx descriptors",
644 		pdcs->pdc_idx, num_frags);
645 
646 	dev_dbg(dev,
647 		"PDC %u txin %u, txout %u, rxin %u, rxout %u, last_rx_curr %u\n",
648 		pdcs->pdc_idx, pdcs->txin, pdcs->txout, pdcs->rxin,
649 		pdcs->rxout, pdcs->last_rx_curr);
650 
651 	if (pdcs->pdc_resp_hdr_len == PDC_SPUM_RESP_HDR_LEN) {
652 		/*
653 		 * For SPU-M, get length of response msg and rx overflow status.
654 		 */
655 		rx_status = *((u32 *)resp_hdr);
656 		len = rx_status & RX_STATUS_LEN;
657 		dev_dbg(dev,
658 			"SPU response length %u bytes", len);
659 		if (unlikely(((rx_status & RX_STATUS_OVERFLOW) || (!len)))) {
660 			if (rx_status & RX_STATUS_OVERFLOW) {
661 				dev_err_ratelimited(dev,
662 						    "crypto receive overflow");
663 				pdcs->rx_oflow++;
664 			} else {
665 				dev_info_ratelimited(dev, "crypto rx len = 0");
666 			}
667 			return -EIO;
668 		}
669 	}
670 
671 	dma_pool_free(pdcs->rx_buf_pool, resp_hdr, resp_hdr_daddr);
672 
673 	mbox_chan_received_data(chan, &mssg);
674 
675 	pdcs->pdc_replies++;
676 	return PDC_SUCCESS;
677 }
678 
679 /**
680  * pdc_receive() - Process as many responses as are available in the rx ring.
681  * @pdcs:  PDC state
682  *
683  * Called within the hard IRQ.
684  * Return:
685  */
686 static int
687 pdc_receive(struct pdc_state *pdcs)
688 {
689 	int rx_status;
690 
691 	/* read last_rx_curr from register once */
692 	pdcs->last_rx_curr =
693 	    (ioread32(&pdcs->rxregs_64->status0) &
694 	     CRYPTO_D64_RS0_CD_MASK) / RING_ENTRY_SIZE;
695 
696 	do {
697 		/* Could be many frames ready */
698 		rx_status = pdc_receive_one(pdcs);
699 	} while (rx_status == PDC_SUCCESS);
700 
701 	return 0;
702 }
703 
704 /**
705  * pdc_tx_list_sg_add() - Add the buffers in a scatterlist to the transmit
706  * descriptors for a given SPU. The scatterlist buffers contain the data for a
707  * SPU request message.
708  * @spu_idx:   The index of the SPU to submit the request to, [0, max_spu)
709  * @sg:        Scatterlist whose buffers contain part of the SPU request
710  *
711  * If a scatterlist buffer is larger than PDC_DMA_BUF_MAX, multiple descriptors
712  * are written for that buffer, each <= PDC_DMA_BUF_MAX byte in length.
713  *
714  * Return: PDC_SUCCESS if successful
715  *         < 0 otherwise
716  */
717 static int pdc_tx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
718 {
719 	u32 flags = 0;
720 	u32 eot;
721 	u32 tx_avail;
722 
723 	/*
724 	 * Num descriptors needed. Conservatively assume we need a descriptor
725 	 * for every entry in sg.
726 	 */
727 	u32 num_desc;
728 	u32 desc_w = 0;	/* Number of tx descriptors written */
729 	u32 bufcnt;	/* Number of bytes of buffer pointed to by descriptor */
730 	dma_addr_t databufptr;	/* DMA address to put in descriptor */
731 
732 	num_desc = (u32)sg_nents(sg);
733 
734 	/* check whether enough tx descriptors are available */
735 	tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
736 					      pdcs->ntxpost);
737 	if (unlikely(num_desc > tx_avail)) {
738 		pdcs->txnobuf++;
739 		return -ENOSPC;
740 	}
741 
742 	/* build tx descriptors */
743 	if (pdcs->tx_msg_start == pdcs->txout) {
744 		/* Start of frame */
745 		pdcs->txin_numd[pdcs->tx_msg_start] = 0;
746 		pdcs->src_sg[pdcs->txout] = sg;
747 		flags = D64_CTRL1_SOF;
748 	}
749 
750 	while (sg) {
751 		if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
752 			eot = D64_CTRL1_EOT;
753 		else
754 			eot = 0;
755 
756 		/*
757 		 * If sg buffer larger than PDC limit, split across
758 		 * multiple descriptors
759 		 */
760 		bufcnt = sg_dma_len(sg);
761 		databufptr = sg_dma_address(sg);
762 		while (bufcnt > PDC_DMA_BUF_MAX) {
763 			pdc_build_txd(pdcs, databufptr, PDC_DMA_BUF_MAX,
764 				      flags | eot);
765 			desc_w++;
766 			bufcnt -= PDC_DMA_BUF_MAX;
767 			databufptr += PDC_DMA_BUF_MAX;
768 			if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
769 				eot = D64_CTRL1_EOT;
770 			else
771 				eot = 0;
772 		}
773 		sg = sg_next(sg);
774 		if (!sg)
775 			/* Writing last descriptor for frame */
776 			flags |= (D64_CTRL1_EOF | D64_CTRL1_IOC);
777 		pdc_build_txd(pdcs, databufptr, bufcnt, flags | eot);
778 		desc_w++;
779 		/* Clear start of frame after first descriptor */
780 		flags &= ~D64_CTRL1_SOF;
781 	}
782 	pdcs->txin_numd[pdcs->tx_msg_start] += desc_w;
783 
784 	return PDC_SUCCESS;
785 }
786 
787 /**
788  * pdc_tx_list_final() - Initiate DMA transfer of last frame written to tx
789  * ring.
790  * @pdcs:  PDC state for SPU to process the request
791  *
792  * Sets the index of the last descriptor written in both the rx and tx ring.
793  *
794  * Return: PDC_SUCCESS
795  */
796 static int pdc_tx_list_final(struct pdc_state *pdcs)
797 {
798 	/*
799 	 * write barrier to ensure all register writes are complete
800 	 * before chip starts to process new request
801 	 */
802 	wmb();
803 	iowrite32(pdcs->rxout << 4, &pdcs->rxregs_64->ptr);
804 	iowrite32(pdcs->txout << 4, &pdcs->txregs_64->ptr);
805 	pdcs->pdc_requests++;
806 
807 	return PDC_SUCCESS;
808 }
809 
810 /**
811  * pdc_rx_list_init() - Start a new receive descriptor list for a given PDC.
812  * @pdcs:   PDC state for SPU handling request
813  * @dst_sg: scatterlist providing rx buffers for response to be returned to
814  *	    mailbox client
815  * @ctx:    Opaque context for this request
816  *
817  * Posts a single receive descriptor to hold the metadata that precedes a
818  * response. For example, with SPU-M, the metadata is a 32-byte DMA header and
819  * an 8-byte BCM header. Moves the msg_start descriptor indexes for both tx and
820  * rx to indicate the start of a new message.
821  *
822  * Return:  PDC_SUCCESS if successful
823  *          < 0 if an error (e.g., rx ring is full)
824  */
825 static int pdc_rx_list_init(struct pdc_state *pdcs, struct scatterlist *dst_sg,
826 			    void *ctx)
827 {
828 	u32 flags = 0;
829 	u32 rx_avail;
830 	u32 rx_pkt_cnt = 1;	/* Adding a single rx buffer */
831 	dma_addr_t daddr;
832 	void *vaddr;
833 	struct pdc_rx_ctx *rx_ctx;
834 
835 	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
836 					      pdcs->nrxpost);
837 	if (unlikely(rx_pkt_cnt > rx_avail)) {
838 		pdcs->rxnobuf++;
839 		return -ENOSPC;
840 	}
841 
842 	/* allocate a buffer for the dma rx status */
843 	vaddr = dma_pool_zalloc(pdcs->rx_buf_pool, GFP_ATOMIC, &daddr);
844 	if (unlikely(!vaddr))
845 		return -ENOMEM;
846 
847 	/*
848 	 * Update msg_start indexes for both tx and rx to indicate the start
849 	 * of a new sequence of descriptor indexes that contain the fragments
850 	 * of the same message.
851 	 */
852 	pdcs->rx_msg_start = pdcs->rxout;
853 	pdcs->tx_msg_start = pdcs->txout;
854 
855 	/* This is always the first descriptor in the receive sequence */
856 	flags = D64_CTRL1_SOF;
857 	pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd = 1;
858 
859 	if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
860 		flags |= D64_CTRL1_EOT;
861 
862 	rx_ctx = &pdcs->rx_ctx[pdcs->rxout];
863 	rx_ctx->rxp_ctx = ctx;
864 	rx_ctx->dst_sg = dst_sg;
865 	rx_ctx->resp_hdr = vaddr;
866 	rx_ctx->resp_hdr_daddr = daddr;
867 	pdc_build_rxd(pdcs, daddr, pdcs->pdc_resp_hdr_len, flags);
868 	return PDC_SUCCESS;
869 }
870 
871 /**
872  * pdc_rx_list_sg_add() - Add the buffers in a scatterlist to the receive
873  * descriptors for a given SPU. The caller must have already DMA mapped the
874  * scatterlist.
875  * @spu_idx:    Indicates which SPU the buffers are for
876  * @sg:         Scatterlist whose buffers are added to the receive ring
877  *
878  * If a receive buffer in the scatterlist is larger than PDC_DMA_BUF_MAX,
879  * multiple receive descriptors are written, each with a buffer <=
880  * PDC_DMA_BUF_MAX.
881  *
882  * Return: PDC_SUCCESS if successful
883  *         < 0 otherwise (e.g., receive ring is full)
884  */
885 static int pdc_rx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
886 {
887 	u32 flags = 0;
888 	u32 rx_avail;
889 
890 	/*
891 	 * Num descriptors needed. Conservatively assume we need a descriptor
892 	 * for every entry from our starting point in the scatterlist.
893 	 */
894 	u32 num_desc;
895 	u32 desc_w = 0;	/* Number of tx descriptors written */
896 	u32 bufcnt;	/* Number of bytes of buffer pointed to by descriptor */
897 	dma_addr_t databufptr;	/* DMA address to put in descriptor */
898 
899 	num_desc = (u32)sg_nents(sg);
900 
901 	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
902 					      pdcs->nrxpost);
903 	if (unlikely(num_desc > rx_avail)) {
904 		pdcs->rxnobuf++;
905 		return -ENOSPC;
906 	}
907 
908 	while (sg) {
909 		if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
910 			flags = D64_CTRL1_EOT;
911 		else
912 			flags = 0;
913 
914 		/*
915 		 * If sg buffer larger than PDC limit, split across
916 		 * multiple descriptors
917 		 */
918 		bufcnt = sg_dma_len(sg);
919 		databufptr = sg_dma_address(sg);
920 		while (bufcnt > PDC_DMA_BUF_MAX) {
921 			pdc_build_rxd(pdcs, databufptr, PDC_DMA_BUF_MAX, flags);
922 			desc_w++;
923 			bufcnt -= PDC_DMA_BUF_MAX;
924 			databufptr += PDC_DMA_BUF_MAX;
925 			if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
926 				flags = D64_CTRL1_EOT;
927 			else
928 				flags = 0;
929 		}
930 		pdc_build_rxd(pdcs, databufptr, bufcnt, flags);
931 		desc_w++;
932 		sg = sg_next(sg);
933 	}
934 	pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd += desc_w;
935 
936 	return PDC_SUCCESS;
937 }
938 
939 /**
940  * pdc_irq_handler() - Interrupt handler called in interrupt context.
941  * @irq:      Interrupt number that has fired
942  * @data:     device struct for DMA engine that generated the interrupt
943  *
944  * We have to clear the device interrupt status flags here. So cache the
945  * status for later use in the thread function. Other than that, just return
946  * WAKE_THREAD to invoke the thread function.
947  *
948  * Return: IRQ_WAKE_THREAD if interrupt is ours
949  *         IRQ_NONE otherwise
950  */
951 static irqreturn_t pdc_irq_handler(int irq, void *data)
952 {
953 	struct device *dev = (struct device *)data;
954 	struct pdc_state *pdcs = dev_get_drvdata(dev);
955 	u32 intstatus = ioread32(pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
956 
957 	if (unlikely(intstatus == 0))
958 		return IRQ_NONE;
959 
960 	/* Disable interrupts until soft handler runs */
961 	iowrite32(0, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
962 
963 	/* Clear interrupt flags in device */
964 	iowrite32(intstatus, pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
965 
966 	/* Wakeup IRQ thread */
967 	tasklet_schedule(&pdcs->rx_tasklet);
968 	return IRQ_HANDLED;
969 }
970 
971 /**
972  * pdc_tasklet_cb() - Tasklet callback that runs the deferred processing after
973  * a DMA receive interrupt. Reenables the receive interrupt.
974  * @data: PDC state structure
975  */
976 static void pdc_tasklet_cb(unsigned long data)
977 {
978 	struct pdc_state *pdcs = (struct pdc_state *)data;
979 
980 	pdc_receive(pdcs);
981 
982 	/* reenable interrupts */
983 	iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
984 }
985 
986 /**
987  * pdc_ring_init() - Allocate DMA rings and initialize constant fields of
988  * descriptors in one ringset.
989  * @pdcs:    PDC instance state
990  * @ringset: index of ringset being used
991  *
992  * Return: PDC_SUCCESS if ring initialized
993  *         < 0 otherwise
994  */
995 static int pdc_ring_init(struct pdc_state *pdcs, int ringset)
996 {
997 	int i;
998 	int err = PDC_SUCCESS;
999 	struct dma64 *dma_reg;
1000 	struct device *dev = &pdcs->pdev->dev;
1001 	struct pdc_ring_alloc tx;
1002 	struct pdc_ring_alloc rx;
1003 
1004 	/* Allocate tx ring */
1005 	tx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &tx.dmabase);
1006 	if (unlikely(!tx.vbase)) {
1007 		err = -ENOMEM;
1008 		goto done;
1009 	}
1010 
1011 	/* Allocate rx ring */
1012 	rx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &rx.dmabase);
1013 	if (unlikely(!rx.vbase)) {
1014 		err = -ENOMEM;
1015 		goto fail_dealloc;
1016 	}
1017 
1018 	dev_dbg(dev, " - base DMA addr of tx ring      %pad", &tx.dmabase);
1019 	dev_dbg(dev, " - base virtual addr of tx ring  %p", tx.vbase);
1020 	dev_dbg(dev, " - base DMA addr of rx ring      %pad", &rx.dmabase);
1021 	dev_dbg(dev, " - base virtual addr of rx ring  %p", rx.vbase);
1022 
1023 	memcpy(&pdcs->tx_ring_alloc, &tx, sizeof(tx));
1024 	memcpy(&pdcs->rx_ring_alloc, &rx, sizeof(rx));
1025 
1026 	pdcs->rxin = 0;
1027 	pdcs->rx_msg_start = 0;
1028 	pdcs->last_rx_curr = 0;
1029 	pdcs->rxout = 0;
1030 	pdcs->txin = 0;
1031 	pdcs->tx_msg_start = 0;
1032 	pdcs->txout = 0;
1033 
1034 	/* Set descriptor array base addresses */
1035 	pdcs->txd_64 = (struct dma64dd *)pdcs->tx_ring_alloc.vbase;
1036 	pdcs->rxd_64 = (struct dma64dd *)pdcs->rx_ring_alloc.vbase;
1037 
1038 	/* Tell device the base DMA address of each ring */
1039 	dma_reg = &pdcs->regs->dmaregs[ringset];
1040 
1041 	/* But first disable DMA and set curptr to 0 for both TX & RX */
1042 	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1043 	iowrite32((PDC_RX_CTL + (pdcs->rx_status_len << 1)),
1044 		  &dma_reg->dmarcv.control);
1045 	iowrite32(0, &dma_reg->dmaxmt.ptr);
1046 	iowrite32(0, &dma_reg->dmarcv.ptr);
1047 
1048 	/* Set base DMA addresses */
1049 	iowrite32(lower_32_bits(pdcs->tx_ring_alloc.dmabase),
1050 		  &dma_reg->dmaxmt.addrlow);
1051 	iowrite32(upper_32_bits(pdcs->tx_ring_alloc.dmabase),
1052 		  &dma_reg->dmaxmt.addrhigh);
1053 
1054 	iowrite32(lower_32_bits(pdcs->rx_ring_alloc.dmabase),
1055 		  &dma_reg->dmarcv.addrlow);
1056 	iowrite32(upper_32_bits(pdcs->rx_ring_alloc.dmabase),
1057 		  &dma_reg->dmarcv.addrhigh);
1058 
1059 	/* Re-enable DMA */
1060 	iowrite32(PDC_TX_CTL | PDC_TX_ENABLE, &dma_reg->dmaxmt.control);
1061 	iowrite32((PDC_RX_CTL | PDC_RX_ENABLE | (pdcs->rx_status_len << 1)),
1062 		  &dma_reg->dmarcv.control);
1063 
1064 	/* Initialize descriptors */
1065 	for (i = 0; i < PDC_RING_ENTRIES; i++) {
1066 		/* Every tx descriptor can be used for start of frame. */
1067 		if (i != pdcs->ntxpost) {
1068 			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF,
1069 				  &pdcs->txd_64[i].ctrl1);
1070 		} else {
1071 			/* Last descriptor in ringset. Set End of Table. */
1072 			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF |
1073 				  D64_CTRL1_EOT, &pdcs->txd_64[i].ctrl1);
1074 		}
1075 
1076 		/* Every rx descriptor can be used for start of frame */
1077 		if (i != pdcs->nrxpost) {
1078 			iowrite32(D64_CTRL1_SOF,
1079 				  &pdcs->rxd_64[i].ctrl1);
1080 		} else {
1081 			/* Last descriptor in ringset. Set End of Table. */
1082 			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOT,
1083 				  &pdcs->rxd_64[i].ctrl1);
1084 		}
1085 	}
1086 	return PDC_SUCCESS;
1087 
1088 fail_dealloc:
1089 	dma_pool_free(pdcs->ring_pool, tx.vbase, tx.dmabase);
1090 done:
1091 	return err;
1092 }
1093 
1094 static void pdc_ring_free(struct pdc_state *pdcs)
1095 {
1096 	if (pdcs->tx_ring_alloc.vbase) {
1097 		dma_pool_free(pdcs->ring_pool, pdcs->tx_ring_alloc.vbase,
1098 			      pdcs->tx_ring_alloc.dmabase);
1099 		pdcs->tx_ring_alloc.vbase = NULL;
1100 	}
1101 
1102 	if (pdcs->rx_ring_alloc.vbase) {
1103 		dma_pool_free(pdcs->ring_pool, pdcs->rx_ring_alloc.vbase,
1104 			      pdcs->rx_ring_alloc.dmabase);
1105 		pdcs->rx_ring_alloc.vbase = NULL;
1106 	}
1107 }
1108 
1109 /**
1110  * pdc_desc_count() - Count the number of DMA descriptors that will be required
1111  * for a given scatterlist. Account for the max length of a DMA buffer.
1112  * @sg:    Scatterlist to be DMA'd
1113  * Return: Number of descriptors required
1114  */
1115 static u32 pdc_desc_count(struct scatterlist *sg)
1116 {
1117 	u32 cnt = 0;
1118 
1119 	while (sg) {
1120 		cnt += ((sg->length / PDC_DMA_BUF_MAX) + 1);
1121 		sg = sg_next(sg);
1122 	}
1123 	return cnt;
1124 }
1125 
1126 /**
1127  * pdc_rings_full() - Check whether the tx ring has room for tx_cnt descriptors
1128  * and the rx ring has room for rx_cnt descriptors.
1129  * @pdcs:  PDC state
1130  * @tx_cnt: The number of descriptors required in the tx ring
1131  * @rx_cnt: The number of descriptors required i the rx ring
1132  *
1133  * Return: true if one of the rings does not have enough space
1134  *         false if sufficient space is available in both rings
1135  */
1136 static bool pdc_rings_full(struct pdc_state *pdcs, int tx_cnt, int rx_cnt)
1137 {
1138 	u32 rx_avail;
1139 	u32 tx_avail;
1140 	bool full = false;
1141 
1142 	/* Check if the tx and rx rings are likely to have enough space */
1143 	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
1144 					      pdcs->nrxpost);
1145 	if (unlikely(rx_cnt > rx_avail)) {
1146 		pdcs->rx_ring_full++;
1147 		full = true;
1148 	}
1149 
1150 	if (likely(!full)) {
1151 		tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
1152 						      pdcs->ntxpost);
1153 		if (unlikely(tx_cnt > tx_avail)) {
1154 			pdcs->tx_ring_full++;
1155 			full = true;
1156 		}
1157 	}
1158 	return full;
1159 }
1160 
1161 /**
1162  * pdc_last_tx_done() - If both the tx and rx rings have at least
1163  * PDC_RING_SPACE_MIN descriptors available, then indicate that the mailbox
1164  * framework can submit another message.
1165  * @chan:  mailbox channel to check
1166  * Return: true if PDC can accept another message on this channel
1167  */
1168 static bool pdc_last_tx_done(struct mbox_chan *chan)
1169 {
1170 	struct pdc_state *pdcs = chan->con_priv;
1171 	bool ret;
1172 
1173 	if (unlikely(pdc_rings_full(pdcs, PDC_RING_SPACE_MIN,
1174 				    PDC_RING_SPACE_MIN))) {
1175 		pdcs->last_tx_not_done++;
1176 		ret = false;
1177 	} else {
1178 		ret = true;
1179 	}
1180 	return ret;
1181 }
1182 
1183 /**
1184  * pdc_send_data() - mailbox send_data function
1185  * @chan:	The mailbox channel on which the data is sent. The channel
1186  *              corresponds to a DMA ringset.
1187  * @data:	The mailbox message to be sent. The message must be a
1188  *              brcm_message structure.
1189  *
1190  * This function is registered as the send_data function for the mailbox
1191  * controller. From the destination scatterlist in the mailbox message, it
1192  * creates a sequence of receive descriptors in the rx ring. From the source
1193  * scatterlist, it creates a sequence of transmit descriptors in the tx ring.
1194  * After creating the descriptors, it writes the rx ptr and tx ptr registers to
1195  * initiate the DMA transfer.
1196  *
1197  * This function does the DMA map and unmap of the src and dst scatterlists in
1198  * the mailbox message.
1199  *
1200  * Return: 0 if successful
1201  *	   -ENOTSUPP if the mailbox message is a type this driver does not
1202  *			support
1203  *         < 0 if an error
1204  */
1205 static int pdc_send_data(struct mbox_chan *chan, void *data)
1206 {
1207 	struct pdc_state *pdcs = chan->con_priv;
1208 	struct device *dev = &pdcs->pdev->dev;
1209 	struct brcm_message *mssg = data;
1210 	int err = PDC_SUCCESS;
1211 	int src_nent;
1212 	int dst_nent;
1213 	int nent;
1214 	u32 tx_desc_req;
1215 	u32 rx_desc_req;
1216 
1217 	if (unlikely(mssg->type != BRCM_MESSAGE_SPU))
1218 		return -ENOTSUPP;
1219 
1220 	src_nent = sg_nents(mssg->spu.src);
1221 	if (likely(src_nent)) {
1222 		nent = dma_map_sg(dev, mssg->spu.src, src_nent, DMA_TO_DEVICE);
1223 		if (unlikely(nent == 0))
1224 			return -EIO;
1225 	}
1226 
1227 	dst_nent = sg_nents(mssg->spu.dst);
1228 	if (likely(dst_nent)) {
1229 		nent = dma_map_sg(dev, mssg->spu.dst, dst_nent,
1230 				  DMA_FROM_DEVICE);
1231 		if (unlikely(nent == 0)) {
1232 			dma_unmap_sg(dev, mssg->spu.src, src_nent,
1233 				     DMA_TO_DEVICE);
1234 			return -EIO;
1235 		}
1236 	}
1237 
1238 	/*
1239 	 * Check if the tx and rx rings have enough space. Do this prior to
1240 	 * writing any tx or rx descriptors. Need to ensure that we do not write
1241 	 * a partial set of descriptors, or write just rx descriptors but
1242 	 * corresponding tx descriptors don't fit. Note that we want this check
1243 	 * and the entire sequence of descriptor to happen without another
1244 	 * thread getting in. The channel spin lock in the mailbox framework
1245 	 * ensures this.
1246 	 */
1247 	tx_desc_req = pdc_desc_count(mssg->spu.src);
1248 	rx_desc_req = pdc_desc_count(mssg->spu.dst);
1249 	if (unlikely(pdc_rings_full(pdcs, tx_desc_req, rx_desc_req + 1)))
1250 		return -ENOSPC;
1251 
1252 	/* Create rx descriptors to SPU catch response */
1253 	err = pdc_rx_list_init(pdcs, mssg->spu.dst, mssg->ctx);
1254 	err |= pdc_rx_list_sg_add(pdcs, mssg->spu.dst);
1255 
1256 	/* Create tx descriptors to submit SPU request */
1257 	err |= pdc_tx_list_sg_add(pdcs, mssg->spu.src);
1258 	err |= pdc_tx_list_final(pdcs);	/* initiate transfer */
1259 
1260 	if (unlikely(err))
1261 		dev_err(&pdcs->pdev->dev,
1262 			"%s failed with error %d", __func__, err);
1263 
1264 	return err;
1265 }
1266 
1267 static int pdc_startup(struct mbox_chan *chan)
1268 {
1269 	return pdc_ring_init(chan->con_priv, PDC_RINGSET);
1270 }
1271 
1272 static void pdc_shutdown(struct mbox_chan *chan)
1273 {
1274 	struct pdc_state *pdcs = chan->con_priv;
1275 
1276 	if (!pdcs)
1277 		return;
1278 
1279 	dev_dbg(&pdcs->pdev->dev,
1280 		"Shutdown mailbox channel for PDC %u", pdcs->pdc_idx);
1281 	pdc_ring_free(pdcs);
1282 }
1283 
1284 /**
1285  * pdc_hw_init() - Use the given initialization parameters to initialize the
1286  * state for one of the PDCs.
1287  * @pdcs:  state of the PDC
1288  */
1289 static
1290 void pdc_hw_init(struct pdc_state *pdcs)
1291 {
1292 	struct platform_device *pdev;
1293 	struct device *dev;
1294 	struct dma64 *dma_reg;
1295 	int ringset = PDC_RINGSET;
1296 
1297 	pdev = pdcs->pdev;
1298 	dev = &pdev->dev;
1299 
1300 	dev_dbg(dev, "PDC %u initial values:", pdcs->pdc_idx);
1301 	dev_dbg(dev, "state structure:                   %p",
1302 		pdcs);
1303 	dev_dbg(dev, " - base virtual addr of hw regs    %p",
1304 		pdcs->pdc_reg_vbase);
1305 
1306 	/* initialize data structures */
1307 	pdcs->regs = (struct pdc_regs *)pdcs->pdc_reg_vbase;
1308 	pdcs->txregs_64 = (struct dma64_regs *)
1309 	    (((u8 *)pdcs->pdc_reg_vbase) +
1310 		     PDC_TXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1311 	pdcs->rxregs_64 = (struct dma64_regs *)
1312 	    (((u8 *)pdcs->pdc_reg_vbase) +
1313 		     PDC_RXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1314 
1315 	pdcs->ntxd = PDC_RING_ENTRIES;
1316 	pdcs->nrxd = PDC_RING_ENTRIES;
1317 	pdcs->ntxpost = PDC_RING_ENTRIES - 1;
1318 	pdcs->nrxpost = PDC_RING_ENTRIES - 1;
1319 	iowrite32(0, &pdcs->regs->intmask);
1320 
1321 	dma_reg = &pdcs->regs->dmaregs[ringset];
1322 
1323 	/* Configure DMA but will enable later in pdc_ring_init() */
1324 	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1325 
1326 	iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1327 		  &dma_reg->dmarcv.control);
1328 
1329 	/* Reset current index pointers after making sure DMA is disabled */
1330 	iowrite32(0, &dma_reg->dmaxmt.ptr);
1331 	iowrite32(0, &dma_reg->dmarcv.ptr);
1332 
1333 	if (pdcs->pdc_resp_hdr_len == PDC_SPU2_RESP_HDR_LEN)
1334 		iowrite32(PDC_CKSUM_CTRL,
1335 			  pdcs->pdc_reg_vbase + PDC_CKSUM_CTRL_OFFSET);
1336 }
1337 
1338 /**
1339  * pdc_hw_disable() - Disable the tx and rx control in the hw.
1340  * @pdcs: PDC state structure
1341  *
1342  */
1343 static void pdc_hw_disable(struct pdc_state *pdcs)
1344 {
1345 	struct dma64 *dma_reg;
1346 
1347 	dma_reg = &pdcs->regs->dmaregs[PDC_RINGSET];
1348 	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1349 	iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1350 		  &dma_reg->dmarcv.control);
1351 }
1352 
1353 /**
1354  * pdc_rx_buf_pool_create() - Pool of receive buffers used to catch the metadata
1355  * header returned with each response message.
1356  * @pdcs: PDC state structure
1357  *
1358  * The metadata is not returned to the mailbox client. So the PDC driver
1359  * manages these buffers.
1360  *
1361  * Return: PDC_SUCCESS
1362  *         -ENOMEM if pool creation fails
1363  */
1364 static int pdc_rx_buf_pool_create(struct pdc_state *pdcs)
1365 {
1366 	struct platform_device *pdev;
1367 	struct device *dev;
1368 
1369 	pdev = pdcs->pdev;
1370 	dev = &pdev->dev;
1371 
1372 	pdcs->pdc_resp_hdr_len = pdcs->rx_status_len;
1373 	if (pdcs->use_bcm_hdr)
1374 		pdcs->pdc_resp_hdr_len += BCM_HDR_LEN;
1375 
1376 	pdcs->rx_buf_pool = dma_pool_create("pdc rx bufs", dev,
1377 					    pdcs->pdc_resp_hdr_len,
1378 					    RX_BUF_ALIGN, 0);
1379 	if (!pdcs->rx_buf_pool)
1380 		return -ENOMEM;
1381 
1382 	return PDC_SUCCESS;
1383 }
1384 
1385 /**
1386  * pdc_interrupts_init() - Initialize the interrupt configuration for a PDC and
1387  * specify a threaded IRQ handler for deferred handling of interrupts outside of
1388  * interrupt context.
1389  * @pdcs:   PDC state
1390  *
1391  * Set the interrupt mask for transmit and receive done.
1392  * Set the lazy interrupt frame count to generate an interrupt for just one pkt.
1393  *
1394  * Return:  PDC_SUCCESS
1395  *          <0 if threaded irq request fails
1396  */
1397 static int pdc_interrupts_init(struct pdc_state *pdcs)
1398 {
1399 	struct platform_device *pdev = pdcs->pdev;
1400 	struct device *dev = &pdev->dev;
1401 	struct device_node *dn = pdev->dev.of_node;
1402 	int err;
1403 
1404 	/* interrupt configuration */
1405 	iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
1406 
1407 	if (pdcs->hw_type == FA_HW)
1408 		iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1409 			  FA_RCVLAZY0_OFFSET);
1410 	else
1411 		iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1412 			  PDC_RCVLAZY0_OFFSET);
1413 
1414 	/* read irq from device tree */
1415 	pdcs->pdc_irq = irq_of_parse_and_map(dn, 0);
1416 	dev_dbg(dev, "pdc device %s irq %u for pdcs %p",
1417 		dev_name(dev), pdcs->pdc_irq, pdcs);
1418 
1419 	err = devm_request_irq(dev, pdcs->pdc_irq, pdc_irq_handler, 0,
1420 			       dev_name(dev), dev);
1421 	if (err) {
1422 		dev_err(dev, "IRQ %u request failed with err %d\n",
1423 			pdcs->pdc_irq, err);
1424 		return err;
1425 	}
1426 	return PDC_SUCCESS;
1427 }
1428 
1429 static const struct mbox_chan_ops pdc_mbox_chan_ops = {
1430 	.send_data = pdc_send_data,
1431 	.last_tx_done = pdc_last_tx_done,
1432 	.startup = pdc_startup,
1433 	.shutdown = pdc_shutdown
1434 };
1435 
1436 /**
1437  * pdc_mb_init() - Initialize the mailbox controller.
1438  * @pdcs:  PDC state
1439  *
1440  * Each PDC is a mailbox controller. Each ringset is a mailbox channel. Kernel
1441  * driver only uses one ringset and thus one mb channel. PDC uses the transmit
1442  * complete interrupt to determine when a mailbox message has successfully been
1443  * transmitted.
1444  *
1445  * Return: 0 on success
1446  *         < 0 if there is an allocation or registration failure
1447  */
1448 static int pdc_mb_init(struct pdc_state *pdcs)
1449 {
1450 	struct device *dev = &pdcs->pdev->dev;
1451 	struct mbox_controller *mbc;
1452 	int chan_index;
1453 	int err;
1454 
1455 	mbc = &pdcs->mbc;
1456 	mbc->dev = dev;
1457 	mbc->ops = &pdc_mbox_chan_ops;
1458 	mbc->num_chans = 1;
1459 	mbc->chans = devm_kcalloc(dev, mbc->num_chans, sizeof(*mbc->chans),
1460 				  GFP_KERNEL);
1461 	if (!mbc->chans)
1462 		return -ENOMEM;
1463 
1464 	mbc->txdone_irq = false;
1465 	mbc->txdone_poll = true;
1466 	mbc->txpoll_period = 1;
1467 	for (chan_index = 0; chan_index < mbc->num_chans; chan_index++)
1468 		mbc->chans[chan_index].con_priv = pdcs;
1469 
1470 	/* Register mailbox controller */
1471 	err = devm_mbox_controller_register(dev, mbc);
1472 	if (err) {
1473 		dev_crit(dev,
1474 			 "Failed to register PDC mailbox controller. Error %d.",
1475 			 err);
1476 		return err;
1477 	}
1478 	return 0;
1479 }
1480 
1481 /* Device tree API */
1482 static const int pdc_hw = PDC_HW;
1483 static const int fa_hw = FA_HW;
1484 
1485 static const struct of_device_id pdc_mbox_of_match[] = {
1486 	{.compatible = "brcm,iproc-pdc-mbox", .data = &pdc_hw},
1487 	{.compatible = "brcm,iproc-fa2-mbox", .data = &fa_hw},
1488 	{ /* sentinel */ }
1489 };
1490 MODULE_DEVICE_TABLE(of, pdc_mbox_of_match);
1491 
1492 /**
1493  * pdc_dt_read() - Read application-specific data from device tree.
1494  * @pdev:  Platform device
1495  * @pdcs:  PDC state
1496  *
1497  * Reads the number of bytes of receive status that precede each received frame.
1498  * Reads whether transmit and received frames should be preceded by an 8-byte
1499  * BCM header.
1500  *
1501  * Return: 0 if successful
1502  *         -ENODEV if device not available
1503  */
1504 static int pdc_dt_read(struct platform_device *pdev, struct pdc_state *pdcs)
1505 {
1506 	struct device *dev = &pdev->dev;
1507 	struct device_node *dn = pdev->dev.of_node;
1508 	const struct of_device_id *match;
1509 	const int *hw_type;
1510 	int err;
1511 
1512 	err = of_property_read_u32(dn, "brcm,rx-status-len",
1513 				   &pdcs->rx_status_len);
1514 	if (err < 0)
1515 		dev_err(dev,
1516 			"%s failed to get DMA receive status length from device tree",
1517 			__func__);
1518 
1519 	pdcs->use_bcm_hdr = of_property_read_bool(dn, "brcm,use-bcm-hdr");
1520 
1521 	pdcs->hw_type = PDC_HW;
1522 
1523 	match = of_match_device(of_match_ptr(pdc_mbox_of_match), dev);
1524 	if (match != NULL) {
1525 		hw_type = match->data;
1526 		pdcs->hw_type = *hw_type;
1527 	}
1528 
1529 	return 0;
1530 }
1531 
1532 /**
1533  * pdc_probe() - Probe function for PDC driver.
1534  * @pdev:   PDC platform device
1535  *
1536  * Reserve and map register regions defined in device tree.
1537  * Allocate and initialize tx and rx DMA rings.
1538  * Initialize a mailbox controller for each PDC.
1539  *
1540  * Return: 0 if successful
1541  *         < 0 if an error
1542  */
1543 static int pdc_probe(struct platform_device *pdev)
1544 {
1545 	int err = 0;
1546 	struct device *dev = &pdev->dev;
1547 	struct resource *pdc_regs;
1548 	struct pdc_state *pdcs;
1549 
1550 	/* PDC state for one SPU */
1551 	pdcs = devm_kzalloc(dev, sizeof(*pdcs), GFP_KERNEL);
1552 	if (!pdcs) {
1553 		err = -ENOMEM;
1554 		goto cleanup;
1555 	}
1556 
1557 	pdcs->pdev = pdev;
1558 	platform_set_drvdata(pdev, pdcs);
1559 	pdcs->pdc_idx = pdcg.num_spu;
1560 	pdcg.num_spu++;
1561 
1562 	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(39));
1563 	if (err) {
1564 		dev_warn(dev, "PDC device cannot perform DMA. Error %d.", err);
1565 		goto cleanup;
1566 	}
1567 
1568 	/* Create DMA pool for tx ring */
1569 	pdcs->ring_pool = dma_pool_create("pdc rings", dev, PDC_RING_SIZE,
1570 					  RING_ALIGN, 0);
1571 	if (!pdcs->ring_pool) {
1572 		err = -ENOMEM;
1573 		goto cleanup;
1574 	}
1575 
1576 	err = pdc_dt_read(pdev, pdcs);
1577 	if (err)
1578 		goto cleanup_ring_pool;
1579 
1580 	pdc_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1581 	if (!pdc_regs) {
1582 		err = -ENODEV;
1583 		goto cleanup_ring_pool;
1584 	}
1585 	dev_dbg(dev, "PDC register region res.start = %pa, res.end = %pa",
1586 		&pdc_regs->start, &pdc_regs->end);
1587 
1588 	pdcs->pdc_reg_vbase = devm_ioremap_resource(&pdev->dev, pdc_regs);
1589 	if (IS_ERR(pdcs->pdc_reg_vbase)) {
1590 		err = PTR_ERR(pdcs->pdc_reg_vbase);
1591 		dev_err(&pdev->dev, "Failed to map registers: %d\n", err);
1592 		goto cleanup_ring_pool;
1593 	}
1594 
1595 	/* create rx buffer pool after dt read to know how big buffers are */
1596 	err = pdc_rx_buf_pool_create(pdcs);
1597 	if (err)
1598 		goto cleanup_ring_pool;
1599 
1600 	pdc_hw_init(pdcs);
1601 
1602 	/* Init tasklet for deferred DMA rx processing */
1603 	tasklet_init(&pdcs->rx_tasklet, pdc_tasklet_cb, (unsigned long)pdcs);
1604 
1605 	err = pdc_interrupts_init(pdcs);
1606 	if (err)
1607 		goto cleanup_buf_pool;
1608 
1609 	/* Initialize mailbox controller */
1610 	err = pdc_mb_init(pdcs);
1611 	if (err)
1612 		goto cleanup_buf_pool;
1613 
1614 	pdc_setup_debugfs(pdcs);
1615 
1616 	dev_dbg(dev, "pdc_probe() successful");
1617 	return PDC_SUCCESS;
1618 
1619 cleanup_buf_pool:
1620 	tasklet_kill(&pdcs->rx_tasklet);
1621 	dma_pool_destroy(pdcs->rx_buf_pool);
1622 
1623 cleanup_ring_pool:
1624 	dma_pool_destroy(pdcs->ring_pool);
1625 
1626 cleanup:
1627 	return err;
1628 }
1629 
1630 static int pdc_remove(struct platform_device *pdev)
1631 {
1632 	struct pdc_state *pdcs = platform_get_drvdata(pdev);
1633 
1634 	pdc_free_debugfs();
1635 
1636 	tasklet_kill(&pdcs->rx_tasklet);
1637 
1638 	pdc_hw_disable(pdcs);
1639 
1640 	dma_pool_destroy(pdcs->rx_buf_pool);
1641 	dma_pool_destroy(pdcs->ring_pool);
1642 	return 0;
1643 }
1644 
1645 static struct platform_driver pdc_mbox_driver = {
1646 	.probe = pdc_probe,
1647 	.remove = pdc_remove,
1648 	.driver = {
1649 		   .name = "brcm-iproc-pdc-mbox",
1650 		   .of_match_table = of_match_ptr(pdc_mbox_of_match),
1651 		   },
1652 };
1653 module_platform_driver(pdc_mbox_driver);
1654 
1655 MODULE_AUTHOR("Rob Rice <rob.rice@broadcom.com>");
1656 MODULE_DESCRIPTION("Broadcom PDC mailbox driver");
1657 MODULE_LICENSE("GPL v2");
1658