Loading...
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/property.h>
37#include <linux/io.h>
38#include <linux/of.h>
39#include <linux/of_irq.h>
40#include <linux/mailbox_controller.h>
41#include <linux/mailbox/brcm-message.h>
42#include <linux/scatterlist.h>
43#include <linux/dma-direction.h>
44#include <linux/dma-mapping.h>
45#include <linux/dmapool.h>
46
47#define PDC_SUCCESS 0
48
49#define RING_ENTRY_SIZE sizeof(struct dma64dd)
50
51/* # entries in PDC dma ring */
52#define PDC_RING_ENTRIES 512
53/*
54 * Minimum number of ring descriptor entries that must be free to tell mailbox
55 * framework that it can submit another request
56 */
57#define PDC_RING_SPACE_MIN 15
58
59#define PDC_RING_SIZE (PDC_RING_ENTRIES * RING_ENTRY_SIZE)
60/* Rings are 8k aligned */
61#define RING_ALIGN_ORDER 13
62#define RING_ALIGN BIT(RING_ALIGN_ORDER)
63
64#define RX_BUF_ALIGN_ORDER 5
65#define RX_BUF_ALIGN BIT(RX_BUF_ALIGN_ORDER)
66
67/* descriptor bumping macros */
68#define XXD(x, max_mask) ((x) & (max_mask))
69#define TXD(x, max_mask) XXD((x), (max_mask))
70#define RXD(x, max_mask) XXD((x), (max_mask))
71#define NEXTTXD(i, max_mask) TXD((i) + 1, (max_mask))
72#define PREVTXD(i, max_mask) TXD((i) - 1, (max_mask))
73#define NEXTRXD(i, max_mask) RXD((i) + 1, (max_mask))
74#define PREVRXD(i, max_mask) RXD((i) - 1, (max_mask))
75#define NTXDACTIVE(h, t, max_mask) TXD((t) - (h), (max_mask))
76#define NRXDACTIVE(h, t, max_mask) RXD((t) - (h), (max_mask))
77
78/* Length of BCM header at start of SPU msg, in bytes */
79#define BCM_HDR_LEN 8
80
81/*
82 * PDC driver reserves ringset 0 on each SPU for its own use. The driver does
83 * not currently support use of multiple ringsets on a single PDC engine.
84 */
85#define PDC_RINGSET 0
86
87/*
88 * Interrupt mask and status definitions. Enable interrupts for tx and rx on
89 * ring 0
90 */
91#define PDC_RCVINT_0 (16 + PDC_RINGSET)
92#define PDC_RCVINTEN_0 BIT(PDC_RCVINT_0)
93#define PDC_INTMASK (PDC_RCVINTEN_0)
94#define PDC_LAZY_FRAMECOUNT 1
95#define PDC_LAZY_TIMEOUT 10000
96#define PDC_LAZY_INT (PDC_LAZY_TIMEOUT | (PDC_LAZY_FRAMECOUNT << 24))
97#define PDC_INTMASK_OFFSET 0x24
98#define PDC_INTSTATUS_OFFSET 0x20
99#define PDC_RCVLAZY0_OFFSET (0x30 + 4 * PDC_RINGSET)
100#define FA_RCVLAZY0_OFFSET 0x100
101
102/*
103 * For SPU2, configure MDE_CKSUM_CONTROL to write 17 bytes of metadata
104 * before frame
105 */
106#define PDC_SPU2_RESP_HDR_LEN 17
107#define PDC_CKSUM_CTRL BIT(27)
108#define PDC_CKSUM_CTRL_OFFSET 0x400
109
110#define PDC_SPUM_RESP_HDR_LEN 32
111
112/*
113 * Sets the following bits for write to transmit control reg:
114 * 11 - PtyChkDisable - parity check is disabled
115 * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
116 */
117#define PDC_TX_CTL 0x000C0800
118
119/* Bit in tx control reg to enable tx channel */
120#define PDC_TX_ENABLE 0x1
121
122/*
123 * Sets the following bits for write to receive control reg:
124 * 7:1 - RcvOffset - size in bytes of status region at start of rx frame buf
125 * 9 - SepRxHdrDescEn - place start of new frames only in descriptors
126 * that have StartOfFrame set
127 * 10 - OflowContinue - on rx FIFO overflow, clear rx fifo, discard all
128 * remaining bytes in current frame, report error
129 * in rx frame status for current frame
130 * 11 - PtyChkDisable - parity check is disabled
131 * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
132 */
133#define PDC_RX_CTL 0x000C0E00
134
135/* Bit in rx control reg to enable rx channel */
136#define PDC_RX_ENABLE 0x1
137
138#define CRYPTO_D64_RS0_CD_MASK ((PDC_RING_ENTRIES * RING_ENTRY_SIZE) - 1)
139
140/* descriptor flags */
141#define D64_CTRL1_EOT BIT(28) /* end of descriptor table */
142#define D64_CTRL1_IOC BIT(29) /* interrupt on complete */
143#define D64_CTRL1_EOF BIT(30) /* end of frame */
144#define D64_CTRL1_SOF BIT(31) /* start of frame */
145
146#define RX_STATUS_OVERFLOW 0x00800000
147#define RX_STATUS_LEN 0x0000FFFF
148
149#define PDC_TXREGS_OFFSET 0x200
150#define PDC_RXREGS_OFFSET 0x220
151
152/* Maximum size buffer the DMA engine can handle */
153#define PDC_DMA_BUF_MAX 16384
154
155enum pdc_hw {
156 FA_HW, /* FA2/FA+ hardware (i.e. Northstar Plus) */
157 PDC_HW /* PDC/MDE hardware (i.e. Northstar 2, Pegasus) */
158};
159
160struct pdc_dma_map {
161 void *ctx; /* opaque context associated with frame */
162};
163
164/* dma descriptor */
165struct dma64dd {
166 u32 ctrl1; /* misc control bits */
167 u32 ctrl2; /* buffer count and address extension */
168 u32 addrlow; /* memory address of the date buffer, bits 31:0 */
169 u32 addrhigh; /* memory address of the date buffer, bits 63:32 */
170};
171
172/* dma registers per channel(xmt or rcv) */
173struct dma64_regs {
174 u32 control; /* enable, et al */
175 u32 ptr; /* last descriptor posted to chip */
176 u32 addrlow; /* descriptor ring base address low 32-bits */
177 u32 addrhigh; /* descriptor ring base address bits 63:32 */
178 u32 status0; /* last rx descriptor written by hw */
179 u32 status1; /* driver does not use */
180};
181
182/* cpp contortions to concatenate w/arg prescan */
183#ifndef PAD
184#define _PADLINE(line) pad ## line
185#define _XSTR(line) _PADLINE(line)
186#define PAD _XSTR(__LINE__)
187#endif /* PAD */
188
189/* dma registers. matches hw layout. */
190struct dma64 {
191 struct dma64_regs dmaxmt; /* dma tx */
192 u32 PAD[2];
193 struct dma64_regs dmarcv; /* dma rx */
194 u32 PAD[2];
195};
196
197/* PDC registers */
198struct pdc_regs {
199 u32 devcontrol; /* 0x000 */
200 u32 devstatus; /* 0x004 */
201 u32 PAD;
202 u32 biststatus; /* 0x00c */
203 u32 PAD[4];
204 u32 intstatus; /* 0x020 */
205 u32 intmask; /* 0x024 */
206 u32 gptimer; /* 0x028 */
207
208 u32 PAD;
209 u32 intrcvlazy_0; /* 0x030 (Only in PDC, not FA2) */
210 u32 intrcvlazy_1; /* 0x034 (Only in PDC, not FA2) */
211 u32 intrcvlazy_2; /* 0x038 (Only in PDC, not FA2) */
212 u32 intrcvlazy_3; /* 0x03c (Only in PDC, not FA2) */
213
214 u32 PAD[48];
215 u32 fa_intrecvlazy; /* 0x100 (Only in FA2, not PDC) */
216 u32 flowctlthresh; /* 0x104 */
217 u32 wrrthresh; /* 0x108 */
218 u32 gmac_idle_cnt_thresh; /* 0x10c */
219
220 u32 PAD[4];
221 u32 ifioaccessaddr; /* 0x120 */
222 u32 ifioaccessbyte; /* 0x124 */
223 u32 ifioaccessdata; /* 0x128 */
224
225 u32 PAD[21];
226 u32 phyaccess; /* 0x180 */
227 u32 PAD;
228 u32 phycontrol; /* 0x188 */
229 u32 txqctl; /* 0x18c */
230 u32 rxqctl; /* 0x190 */
231 u32 gpioselect; /* 0x194 */
232 u32 gpio_output_en; /* 0x198 */
233 u32 PAD; /* 0x19c */
234 u32 txq_rxq_mem_ctl; /* 0x1a0 */
235 u32 memory_ecc_status; /* 0x1a4 */
236 u32 serdes_ctl; /* 0x1a8 */
237 u32 serdes_status0; /* 0x1ac */
238 u32 serdes_status1; /* 0x1b0 */
239 u32 PAD[11]; /* 0x1b4-1dc */
240 u32 clk_ctl_st; /* 0x1e0 */
241 u32 hw_war; /* 0x1e4 (Only in PDC, not FA2) */
242 u32 pwrctl; /* 0x1e8 */
243 u32 PAD[5];
244
245#define PDC_NUM_DMA_RINGS 4
246 struct dma64 dmaregs[PDC_NUM_DMA_RINGS]; /* 0x0200 - 0x2fc */
247
248 /* more registers follow, but we don't use them */
249};
250
251/* structure for allocating/freeing DMA rings */
252struct pdc_ring_alloc {
253 dma_addr_t dmabase; /* DMA address of start of ring */
254 void *vbase; /* base kernel virtual address of ring */
255 u32 size; /* ring allocation size in bytes */
256};
257
258/*
259 * context associated with a receive descriptor.
260 * @rxp_ctx: opaque context associated with frame that starts at each
261 * rx ring index.
262 * @dst_sg: Scatterlist used to form reply frames beginning at a given ring
263 * index. Retained in order to unmap each sg after reply is processed.
264 * @rxin_numd: Number of rx descriptors associated with the message that starts
265 * at a descriptor index. Not set for every index. For example,
266 * if descriptor index i points to a scatterlist with 4 entries,
267 * then the next three descriptor indexes don't have a value set.
268 * @resp_hdr: Virtual address of buffer used to catch DMA rx status
269 * @resp_hdr_daddr: physical address of DMA rx status buffer
270 */
271struct pdc_rx_ctx {
272 void *rxp_ctx;
273 struct scatterlist *dst_sg;
274 u32 rxin_numd;
275 void *resp_hdr;
276 dma_addr_t resp_hdr_daddr;
277};
278
279/* PDC state structure */
280struct pdc_state {
281 /* Index of the PDC whose state is in this structure instance */
282 u8 pdc_idx;
283
284 /* Platform device for this PDC instance */
285 struct platform_device *pdev;
286
287 /*
288 * Each PDC instance has a mailbox controller. PDC receives request
289 * messages through mailboxes, and sends response messages through the
290 * mailbox framework.
291 */
292 struct mbox_controller mbc;
293
294 unsigned int pdc_irq;
295
296 /* tasklet for deferred processing after DMA rx interrupt */
297 struct tasklet_struct rx_tasklet;
298
299 /* Number of bytes of receive status prior to each rx frame */
300 u32 rx_status_len;
301 /* Whether a BCM header is prepended to each frame */
302 bool use_bcm_hdr;
303 /* Sum of length of BCM header and rx status header */
304 u32 pdc_resp_hdr_len;
305
306 /* The base virtual address of DMA hw registers */
307 void __iomem *pdc_reg_vbase;
308
309 /* Pool for allocation of DMA rings */
310 struct dma_pool *ring_pool;
311
312 /* Pool for allocation of metadata buffers for response messages */
313 struct dma_pool *rx_buf_pool;
314
315 /*
316 * The base virtual address of DMA tx/rx descriptor rings. Corresponding
317 * DMA address and size of ring allocation.
318 */
319 struct pdc_ring_alloc tx_ring_alloc;
320 struct pdc_ring_alloc rx_ring_alloc;
321
322 struct pdc_regs *regs; /* start of PDC registers */
323
324 struct dma64_regs *txregs_64; /* dma tx engine registers */
325 struct dma64_regs *rxregs_64; /* dma rx engine registers */
326
327 /*
328 * Arrays of PDC_RING_ENTRIES descriptors
329 * To use multiple ringsets, this needs to be extended
330 */
331 struct dma64dd *txd_64; /* tx descriptor ring */
332 struct dma64dd *rxd_64; /* rx descriptor ring */
333
334 /* descriptor ring sizes */
335 u32 ntxd; /* # tx descriptors */
336 u32 nrxd; /* # rx descriptors */
337 u32 nrxpost; /* # rx buffers to keep posted */
338 u32 ntxpost; /* max number of tx buffers that can be posted */
339
340 /*
341 * Index of next tx descriptor to reclaim. That is, the descriptor
342 * index of the oldest tx buffer for which the host has yet to process
343 * the corresponding response.
344 */
345 u32 txin;
346
347 /*
348 * Index of the first receive descriptor for the sequence of
349 * message fragments currently under construction. Used to build up
350 * the rxin_numd count for a message. Updated to rxout when the host
351 * starts a new sequence of rx buffers for a new message.
352 */
353 u32 tx_msg_start;
354
355 /* Index of next tx descriptor to post. */
356 u32 txout;
357
358 /*
359 * Number of tx descriptors associated with the message that starts
360 * at this tx descriptor index.
361 */
362 u32 txin_numd[PDC_RING_ENTRIES];
363
364 /*
365 * Index of next rx descriptor to reclaim. This is the index of
366 * the next descriptor whose data has yet to be processed by the host.
367 */
368 u32 rxin;
369
370 /*
371 * Index of the first receive descriptor for the sequence of
372 * message fragments currently under construction. Used to build up
373 * the rxin_numd count for a message. Updated to rxout when the host
374 * starts a new sequence of rx buffers for a new message.
375 */
376 u32 rx_msg_start;
377
378 /*
379 * Saved value of current hardware rx descriptor index.
380 * The last rx buffer written by the hw is the index previous to
381 * this one.
382 */
383 u32 last_rx_curr;
384
385 /* Index of next rx descriptor to post. */
386 u32 rxout;
387
388 struct pdc_rx_ctx rx_ctx[PDC_RING_ENTRIES];
389
390 /*
391 * Scatterlists used to form request and reply frames beginning at a
392 * given ring index. Retained in order to unmap each sg after reply
393 * is processed
394 */
395 struct scatterlist *src_sg[PDC_RING_ENTRIES];
396
397 /* counters */
398 u32 pdc_requests; /* number of request messages submitted */
399 u32 pdc_replies; /* number of reply messages received */
400 u32 last_tx_not_done; /* too few tx descriptors to indicate done */
401 u32 tx_ring_full; /* unable to accept msg because tx ring full */
402 u32 rx_ring_full; /* unable to accept msg because rx ring full */
403 u32 txnobuf; /* unable to create tx descriptor */
404 u32 rxnobuf; /* unable to create rx descriptor */
405 u32 rx_oflow; /* count of rx overflows */
406
407 /* hardware type - FA2 or PDC/MDE */
408 enum pdc_hw hw_type;
409};
410
411/* Global variables */
412
413struct pdc_globals {
414 /* Actual number of SPUs in hardware, as reported by device tree */
415 u32 num_spu;
416};
417
418static struct pdc_globals pdcg;
419
420/* top level debug FS directory for PDC driver */
421static struct dentry *debugfs_dir;
422
423static ssize_t pdc_debugfs_read(struct file *filp, char __user *ubuf,
424 size_t count, loff_t *offp)
425{
426 struct pdc_state *pdcs;
427 char *buf;
428 ssize_t ret, out_offset, out_count;
429
430 out_count = 512;
431
432 buf = kmalloc(out_count, GFP_KERNEL);
433 if (!buf)
434 return -ENOMEM;
435
436 pdcs = filp->private_data;
437 out_offset = 0;
438 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
439 "SPU %u stats:\n", pdcs->pdc_idx);
440 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
441 "PDC requests....................%u\n",
442 pdcs->pdc_requests);
443 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
444 "PDC responses...................%u\n",
445 pdcs->pdc_replies);
446 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
447 "Tx not done.....................%u\n",
448 pdcs->last_tx_not_done);
449 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
450 "Tx ring full....................%u\n",
451 pdcs->tx_ring_full);
452 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
453 "Rx ring full....................%u\n",
454 pdcs->rx_ring_full);
455 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
456 "Tx desc write fail. Ring full...%u\n",
457 pdcs->txnobuf);
458 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
459 "Rx desc write fail. Ring full...%u\n",
460 pdcs->rxnobuf);
461 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
462 "Receive overflow................%u\n",
463 pdcs->rx_oflow);
464 out_offset += scnprintf(buf + out_offset, out_count - out_offset,
465 "Num frags in rx ring............%u\n",
466 NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr,
467 pdcs->nrxpost));
468
469 if (out_offset > out_count)
470 out_offset = out_count;
471
472 ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset);
473 kfree(buf);
474 return ret;
475}
476
477static const struct file_operations pdc_debugfs_stats = {
478 .owner = THIS_MODULE,
479 .open = simple_open,
480 .read = pdc_debugfs_read,
481};
482
483/**
484 * pdc_setup_debugfs() - Create the debug FS directories. If the top-level
485 * directory has not yet been created, create it now. Create a stats file in
486 * this directory for a SPU.
487 * @pdcs: PDC state structure
488 */
489static void pdc_setup_debugfs(struct pdc_state *pdcs)
490{
491 char spu_stats_name[16];
492
493 if (!debugfs_initialized())
494 return;
495
496 snprintf(spu_stats_name, 16, "pdc%d_stats", pdcs->pdc_idx);
497 if (!debugfs_dir)
498 debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
499
500 /* S_IRUSR == 0400 */
501 debugfs_create_file(spu_stats_name, 0400, debugfs_dir, pdcs,
502 &pdc_debugfs_stats);
503}
504
505static void pdc_free_debugfs(void)
506{
507 debugfs_remove_recursive(debugfs_dir);
508 debugfs_dir = NULL;
509}
510
511/**
512 * pdc_build_rxd() - Build DMA descriptor to receive SPU result.
513 * @pdcs: PDC state for SPU that will generate result
514 * @dma_addr: DMA address of buffer that descriptor is being built for
515 * @buf_len: Length of the receive buffer, in bytes
516 * @flags: Flags to be stored in descriptor
517 */
518static inline void
519pdc_build_rxd(struct pdc_state *pdcs, dma_addr_t dma_addr,
520 u32 buf_len, u32 flags)
521{
522 struct device *dev = &pdcs->pdev->dev;
523 struct dma64dd *rxd = &pdcs->rxd_64[pdcs->rxout];
524
525 dev_dbg(dev,
526 "Writing rx descriptor for PDC %u at index %u with length %u. flags %#x\n",
527 pdcs->pdc_idx, pdcs->rxout, buf_len, flags);
528
529 rxd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
530 rxd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
531 rxd->ctrl1 = cpu_to_le32(flags);
532 rxd->ctrl2 = cpu_to_le32(buf_len);
533
534 /* bump ring index and return */
535 pdcs->rxout = NEXTRXD(pdcs->rxout, pdcs->nrxpost);
536}
537
538/**
539 * pdc_build_txd() - Build a DMA descriptor to transmit a SPU request to
540 * hardware.
541 * @pdcs: PDC state for the SPU that will process this request
542 * @dma_addr: DMA address of packet to be transmitted
543 * @buf_len: Length of tx buffer, in bytes
544 * @flags: Flags to be stored in descriptor
545 */
546static inline void
547pdc_build_txd(struct pdc_state *pdcs, dma_addr_t dma_addr, u32 buf_len,
548 u32 flags)
549{
550 struct device *dev = &pdcs->pdev->dev;
551 struct dma64dd *txd = &pdcs->txd_64[pdcs->txout];
552
553 dev_dbg(dev,
554 "Writing tx descriptor for PDC %u at index %u with length %u, flags %#x\n",
555 pdcs->pdc_idx, pdcs->txout, buf_len, flags);
556
557 txd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
558 txd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
559 txd->ctrl1 = cpu_to_le32(flags);
560 txd->ctrl2 = cpu_to_le32(buf_len);
561
562 /* bump ring index and return */
563 pdcs->txout = NEXTTXD(pdcs->txout, pdcs->ntxpost);
564}
565
566/**
567 * pdc_receive_one() - Receive a response message from a given SPU.
568 * @pdcs: PDC state for the SPU to receive from
569 *
570 * When the return code indicates success, the response message is available in
571 * the receive buffers provided prior to submission of the request.
572 *
573 * Return: PDC_SUCCESS if one or more receive descriptors was processed
574 * -EAGAIN indicates that no response message is available
575 * -EIO an error occurred
576 */
577static int
578pdc_receive_one(struct pdc_state *pdcs)
579{
580 struct device *dev = &pdcs->pdev->dev;
581 struct mbox_controller *mbc;
582 struct mbox_chan *chan;
583 struct brcm_message mssg;
584 u32 len, rx_status;
585 u32 num_frags;
586 u8 *resp_hdr; /* virtual addr of start of resp message DMA header */
587 u32 frags_rdy; /* number of fragments ready to read */
588 u32 rx_idx; /* ring index of start of receive frame */
589 dma_addr_t resp_hdr_daddr;
590 struct pdc_rx_ctx *rx_ctx;
591
592 mbc = &pdcs->mbc;
593 chan = &mbc->chans[0];
594 mssg.type = BRCM_MESSAGE_SPU;
595
596 /*
597 * return if a complete response message is not yet ready.
598 * rxin_numd[rxin] is the number of fragments in the next msg
599 * to read.
600 */
601 frags_rdy = NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr, pdcs->nrxpost);
602 if ((frags_rdy == 0) ||
603 (frags_rdy < pdcs->rx_ctx[pdcs->rxin].rxin_numd))
604 /* No response ready */
605 return -EAGAIN;
606
607 num_frags = pdcs->txin_numd[pdcs->txin];
608 WARN_ON(num_frags == 0);
609
610 dma_unmap_sg(dev, pdcs->src_sg[pdcs->txin],
611 sg_nents(pdcs->src_sg[pdcs->txin]), DMA_TO_DEVICE);
612
613 pdcs->txin = (pdcs->txin + num_frags) & pdcs->ntxpost;
614
615 dev_dbg(dev, "PDC %u reclaimed %d tx descriptors",
616 pdcs->pdc_idx, num_frags);
617
618 rx_idx = pdcs->rxin;
619 rx_ctx = &pdcs->rx_ctx[rx_idx];
620 num_frags = rx_ctx->rxin_numd;
621 /* Return opaque context with result */
622 mssg.ctx = rx_ctx->rxp_ctx;
623 rx_ctx->rxp_ctx = NULL;
624 resp_hdr = rx_ctx->resp_hdr;
625 resp_hdr_daddr = rx_ctx->resp_hdr_daddr;
626 dma_unmap_sg(dev, rx_ctx->dst_sg, sg_nents(rx_ctx->dst_sg),
627 DMA_FROM_DEVICE);
628
629 pdcs->rxin = (pdcs->rxin + num_frags) & pdcs->nrxpost;
630
631 dev_dbg(dev, "PDC %u reclaimed %d rx descriptors",
632 pdcs->pdc_idx, num_frags);
633
634 dev_dbg(dev,
635 "PDC %u txin %u, txout %u, rxin %u, rxout %u, last_rx_curr %u\n",
636 pdcs->pdc_idx, pdcs->txin, pdcs->txout, pdcs->rxin,
637 pdcs->rxout, pdcs->last_rx_curr);
638
639 if (pdcs->pdc_resp_hdr_len == PDC_SPUM_RESP_HDR_LEN) {
640 /*
641 * For SPU-M, get length of response msg and rx overflow status.
642 */
643 rx_status = *((u32 *)resp_hdr);
644 len = rx_status & RX_STATUS_LEN;
645 dev_dbg(dev,
646 "SPU response length %u bytes", len);
647 if (unlikely(((rx_status & RX_STATUS_OVERFLOW) || (!len)))) {
648 if (rx_status & RX_STATUS_OVERFLOW) {
649 dev_err_ratelimited(dev,
650 "crypto receive overflow");
651 pdcs->rx_oflow++;
652 } else {
653 dev_info_ratelimited(dev, "crypto rx len = 0");
654 }
655 return -EIO;
656 }
657 }
658
659 dma_pool_free(pdcs->rx_buf_pool, resp_hdr, resp_hdr_daddr);
660
661 mbox_chan_received_data(chan, &mssg);
662
663 pdcs->pdc_replies++;
664 return PDC_SUCCESS;
665}
666
667/**
668 * pdc_receive() - Process as many responses as are available in the rx ring.
669 * @pdcs: PDC state
670 *
671 * Called within the hard IRQ.
672 * Return:
673 */
674static int
675pdc_receive(struct pdc_state *pdcs)
676{
677 int rx_status;
678
679 /* read last_rx_curr from register once */
680 pdcs->last_rx_curr =
681 (ioread32((const void __iomem *)&pdcs->rxregs_64->status0) &
682 CRYPTO_D64_RS0_CD_MASK) / RING_ENTRY_SIZE;
683
684 do {
685 /* Could be many frames ready */
686 rx_status = pdc_receive_one(pdcs);
687 } while (rx_status == PDC_SUCCESS);
688
689 return 0;
690}
691
692/**
693 * pdc_tx_list_sg_add() - Add the buffers in a scatterlist to the transmit
694 * descriptors for a given SPU. The scatterlist buffers contain the data for a
695 * SPU request message.
696 * @pdcs: PDC state for the SPU that will process this request
697 * @sg: Scatterlist whose buffers contain part of the SPU request
698 *
699 * If a scatterlist buffer is larger than PDC_DMA_BUF_MAX, multiple descriptors
700 * are written for that buffer, each <= PDC_DMA_BUF_MAX byte in length.
701 *
702 * Return: PDC_SUCCESS if successful
703 * < 0 otherwise
704 */
705static int pdc_tx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
706{
707 u32 flags = 0;
708 u32 eot;
709 u32 tx_avail;
710
711 /*
712 * Num descriptors needed. Conservatively assume we need a descriptor
713 * for every entry in sg.
714 */
715 u32 num_desc;
716 u32 desc_w = 0; /* Number of tx descriptors written */
717 u32 bufcnt; /* Number of bytes of buffer pointed to by descriptor */
718 dma_addr_t databufptr; /* DMA address to put in descriptor */
719
720 num_desc = (u32)sg_nents(sg);
721
722 /* check whether enough tx descriptors are available */
723 tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
724 pdcs->ntxpost);
725 if (unlikely(num_desc > tx_avail)) {
726 pdcs->txnobuf++;
727 return -ENOSPC;
728 }
729
730 /* build tx descriptors */
731 if (pdcs->tx_msg_start == pdcs->txout) {
732 /* Start of frame */
733 pdcs->txin_numd[pdcs->tx_msg_start] = 0;
734 pdcs->src_sg[pdcs->txout] = sg;
735 flags = D64_CTRL1_SOF;
736 }
737
738 while (sg) {
739 if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
740 eot = D64_CTRL1_EOT;
741 else
742 eot = 0;
743
744 /*
745 * If sg buffer larger than PDC limit, split across
746 * multiple descriptors
747 */
748 bufcnt = sg_dma_len(sg);
749 databufptr = sg_dma_address(sg);
750 while (bufcnt > PDC_DMA_BUF_MAX) {
751 pdc_build_txd(pdcs, databufptr, PDC_DMA_BUF_MAX,
752 flags | eot);
753 desc_w++;
754 bufcnt -= PDC_DMA_BUF_MAX;
755 databufptr += PDC_DMA_BUF_MAX;
756 if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
757 eot = D64_CTRL1_EOT;
758 else
759 eot = 0;
760 }
761 sg = sg_next(sg);
762 if (!sg)
763 /* Writing last descriptor for frame */
764 flags |= (D64_CTRL1_EOF | D64_CTRL1_IOC);
765 pdc_build_txd(pdcs, databufptr, bufcnt, flags | eot);
766 desc_w++;
767 /* Clear start of frame after first descriptor */
768 flags &= ~D64_CTRL1_SOF;
769 }
770 pdcs->txin_numd[pdcs->tx_msg_start] += desc_w;
771
772 return PDC_SUCCESS;
773}
774
775/**
776 * pdc_tx_list_final() - Initiate DMA transfer of last frame written to tx
777 * ring.
778 * @pdcs: PDC state for SPU to process the request
779 *
780 * Sets the index of the last descriptor written in both the rx and tx ring.
781 *
782 * Return: PDC_SUCCESS
783 */
784static int pdc_tx_list_final(struct pdc_state *pdcs)
785{
786 /*
787 * write barrier to ensure all register writes are complete
788 * before chip starts to process new request
789 */
790 wmb();
791 iowrite32(pdcs->rxout << 4, &pdcs->rxregs_64->ptr);
792 iowrite32(pdcs->txout << 4, &pdcs->txregs_64->ptr);
793 pdcs->pdc_requests++;
794
795 return PDC_SUCCESS;
796}
797
798/**
799 * pdc_rx_list_init() - Start a new receive descriptor list for a given PDC.
800 * @pdcs: PDC state for SPU handling request
801 * @dst_sg: scatterlist providing rx buffers for response to be returned to
802 * mailbox client
803 * @ctx: Opaque context for this request
804 *
805 * Posts a single receive descriptor to hold the metadata that precedes a
806 * response. For example, with SPU-M, the metadata is a 32-byte DMA header and
807 * an 8-byte BCM header. Moves the msg_start descriptor indexes for both tx and
808 * rx to indicate the start of a new message.
809 *
810 * Return: PDC_SUCCESS if successful
811 * < 0 if an error (e.g., rx ring is full)
812 */
813static int pdc_rx_list_init(struct pdc_state *pdcs, struct scatterlist *dst_sg,
814 void *ctx)
815{
816 u32 flags = 0;
817 u32 rx_avail;
818 u32 rx_pkt_cnt = 1; /* Adding a single rx buffer */
819 dma_addr_t daddr;
820 void *vaddr;
821 struct pdc_rx_ctx *rx_ctx;
822
823 rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
824 pdcs->nrxpost);
825 if (unlikely(rx_pkt_cnt > rx_avail)) {
826 pdcs->rxnobuf++;
827 return -ENOSPC;
828 }
829
830 /* allocate a buffer for the dma rx status */
831 vaddr = dma_pool_zalloc(pdcs->rx_buf_pool, GFP_ATOMIC, &daddr);
832 if (unlikely(!vaddr))
833 return -ENOMEM;
834
835 /*
836 * Update msg_start indexes for both tx and rx to indicate the start
837 * of a new sequence of descriptor indexes that contain the fragments
838 * of the same message.
839 */
840 pdcs->rx_msg_start = pdcs->rxout;
841 pdcs->tx_msg_start = pdcs->txout;
842
843 /* This is always the first descriptor in the receive sequence */
844 flags = D64_CTRL1_SOF;
845 pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd = 1;
846
847 if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
848 flags |= D64_CTRL1_EOT;
849
850 rx_ctx = &pdcs->rx_ctx[pdcs->rxout];
851 rx_ctx->rxp_ctx = ctx;
852 rx_ctx->dst_sg = dst_sg;
853 rx_ctx->resp_hdr = vaddr;
854 rx_ctx->resp_hdr_daddr = daddr;
855 pdc_build_rxd(pdcs, daddr, pdcs->pdc_resp_hdr_len, flags);
856 return PDC_SUCCESS;
857}
858
859/**
860 * pdc_rx_list_sg_add() - Add the buffers in a scatterlist to the receive
861 * descriptors for a given SPU. The caller must have already DMA mapped the
862 * scatterlist.
863 * @pdcs: PDC state for the SPU that will process this request
864 * @sg: Scatterlist whose buffers are added to the receive ring
865 *
866 * If a receive buffer in the scatterlist is larger than PDC_DMA_BUF_MAX,
867 * multiple receive descriptors are written, each with a buffer <=
868 * PDC_DMA_BUF_MAX.
869 *
870 * Return: PDC_SUCCESS if successful
871 * < 0 otherwise (e.g., receive ring is full)
872 */
873static int pdc_rx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
874{
875 u32 flags = 0;
876 u32 rx_avail;
877
878 /*
879 * Num descriptors needed. Conservatively assume we need a descriptor
880 * for every entry from our starting point in the scatterlist.
881 */
882 u32 num_desc;
883 u32 desc_w = 0; /* Number of tx descriptors written */
884 u32 bufcnt; /* Number of bytes of buffer pointed to by descriptor */
885 dma_addr_t databufptr; /* DMA address to put in descriptor */
886
887 num_desc = (u32)sg_nents(sg);
888
889 rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
890 pdcs->nrxpost);
891 if (unlikely(num_desc > rx_avail)) {
892 pdcs->rxnobuf++;
893 return -ENOSPC;
894 }
895
896 while (sg) {
897 if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
898 flags = D64_CTRL1_EOT;
899 else
900 flags = 0;
901
902 /*
903 * If sg buffer larger than PDC limit, split across
904 * multiple descriptors
905 */
906 bufcnt = sg_dma_len(sg);
907 databufptr = sg_dma_address(sg);
908 while (bufcnt > PDC_DMA_BUF_MAX) {
909 pdc_build_rxd(pdcs, databufptr, PDC_DMA_BUF_MAX, flags);
910 desc_w++;
911 bufcnt -= PDC_DMA_BUF_MAX;
912 databufptr += PDC_DMA_BUF_MAX;
913 if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
914 flags = D64_CTRL1_EOT;
915 else
916 flags = 0;
917 }
918 pdc_build_rxd(pdcs, databufptr, bufcnt, flags);
919 desc_w++;
920 sg = sg_next(sg);
921 }
922 pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd += desc_w;
923
924 return PDC_SUCCESS;
925}
926
927/**
928 * pdc_irq_handler() - Interrupt handler called in interrupt context.
929 * @irq: Interrupt number that has fired
930 * @data: device struct for DMA engine that generated the interrupt
931 *
932 * We have to clear the device interrupt status flags here. So cache the
933 * status for later use in the thread function. Other than that, just return
934 * WAKE_THREAD to invoke the thread function.
935 *
936 * Return: IRQ_WAKE_THREAD if interrupt is ours
937 * IRQ_NONE otherwise
938 */
939static irqreturn_t pdc_irq_handler(int irq, void *data)
940{
941 struct device *dev = (struct device *)data;
942 struct pdc_state *pdcs = dev_get_drvdata(dev);
943 u32 intstatus = ioread32(pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
944
945 if (unlikely(intstatus == 0))
946 return IRQ_NONE;
947
948 /* Disable interrupts until soft handler runs */
949 iowrite32(0, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
950
951 /* Clear interrupt flags in device */
952 iowrite32(intstatus, pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
953
954 /* Wakeup IRQ thread */
955 tasklet_schedule(&pdcs->rx_tasklet);
956 return IRQ_HANDLED;
957}
958
959/**
960 * pdc_tasklet_cb() - Tasklet callback that runs the deferred processing after
961 * a DMA receive interrupt. Reenables the receive interrupt.
962 * @t: Pointer to the Altera sSGDMA channel structure
963 */
964static void pdc_tasklet_cb(struct tasklet_struct *t)
965{
966 struct pdc_state *pdcs = from_tasklet(pdcs, t, rx_tasklet);
967
968 pdc_receive(pdcs);
969
970 /* reenable interrupts */
971 iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
972}
973
974/**
975 * pdc_ring_init() - Allocate DMA rings and initialize constant fields of
976 * descriptors in one ringset.
977 * @pdcs: PDC instance state
978 * @ringset: index of ringset being used
979 *
980 * Return: PDC_SUCCESS if ring initialized
981 * < 0 otherwise
982 */
983static int pdc_ring_init(struct pdc_state *pdcs, int ringset)
984{
985 int i;
986 int err = PDC_SUCCESS;
987 struct dma64 *dma_reg;
988 struct device *dev = &pdcs->pdev->dev;
989 struct pdc_ring_alloc tx;
990 struct pdc_ring_alloc rx;
991
992 /* Allocate tx ring */
993 tx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &tx.dmabase);
994 if (unlikely(!tx.vbase)) {
995 err = -ENOMEM;
996 goto done;
997 }
998
999 /* Allocate rx ring */
1000 rx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &rx.dmabase);
1001 if (unlikely(!rx.vbase)) {
1002 err = -ENOMEM;
1003 goto fail_dealloc;
1004 }
1005
1006 dev_dbg(dev, " - base DMA addr of tx ring %pad", &tx.dmabase);
1007 dev_dbg(dev, " - base virtual addr of tx ring %p", tx.vbase);
1008 dev_dbg(dev, " - base DMA addr of rx ring %pad", &rx.dmabase);
1009 dev_dbg(dev, " - base virtual addr of rx ring %p", rx.vbase);
1010
1011 memcpy(&pdcs->tx_ring_alloc, &tx, sizeof(tx));
1012 memcpy(&pdcs->rx_ring_alloc, &rx, sizeof(rx));
1013
1014 pdcs->rxin = 0;
1015 pdcs->rx_msg_start = 0;
1016 pdcs->last_rx_curr = 0;
1017 pdcs->rxout = 0;
1018 pdcs->txin = 0;
1019 pdcs->tx_msg_start = 0;
1020 pdcs->txout = 0;
1021
1022 /* Set descriptor array base addresses */
1023 pdcs->txd_64 = (struct dma64dd *)pdcs->tx_ring_alloc.vbase;
1024 pdcs->rxd_64 = (struct dma64dd *)pdcs->rx_ring_alloc.vbase;
1025
1026 /* Tell device the base DMA address of each ring */
1027 dma_reg = &pdcs->regs->dmaregs[ringset];
1028
1029 /* But first disable DMA and set curptr to 0 for both TX & RX */
1030 iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1031 iowrite32((PDC_RX_CTL + (pdcs->rx_status_len << 1)),
1032 &dma_reg->dmarcv.control);
1033 iowrite32(0, &dma_reg->dmaxmt.ptr);
1034 iowrite32(0, &dma_reg->dmarcv.ptr);
1035
1036 /* Set base DMA addresses */
1037 iowrite32(lower_32_bits(pdcs->tx_ring_alloc.dmabase),
1038 &dma_reg->dmaxmt.addrlow);
1039 iowrite32(upper_32_bits(pdcs->tx_ring_alloc.dmabase),
1040 &dma_reg->dmaxmt.addrhigh);
1041
1042 iowrite32(lower_32_bits(pdcs->rx_ring_alloc.dmabase),
1043 &dma_reg->dmarcv.addrlow);
1044 iowrite32(upper_32_bits(pdcs->rx_ring_alloc.dmabase),
1045 &dma_reg->dmarcv.addrhigh);
1046
1047 /* Re-enable DMA */
1048 iowrite32(PDC_TX_CTL | PDC_TX_ENABLE, &dma_reg->dmaxmt.control);
1049 iowrite32((PDC_RX_CTL | PDC_RX_ENABLE | (pdcs->rx_status_len << 1)),
1050 &dma_reg->dmarcv.control);
1051
1052 /* Initialize descriptors */
1053 for (i = 0; i < PDC_RING_ENTRIES; i++) {
1054 /* Every tx descriptor can be used for start of frame. */
1055 if (i != pdcs->ntxpost) {
1056 iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF,
1057 &pdcs->txd_64[i].ctrl1);
1058 } else {
1059 /* Last descriptor in ringset. Set End of Table. */
1060 iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF |
1061 D64_CTRL1_EOT, &pdcs->txd_64[i].ctrl1);
1062 }
1063
1064 /* Every rx descriptor can be used for start of frame */
1065 if (i != pdcs->nrxpost) {
1066 iowrite32(D64_CTRL1_SOF,
1067 &pdcs->rxd_64[i].ctrl1);
1068 } else {
1069 /* Last descriptor in ringset. Set End of Table. */
1070 iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOT,
1071 &pdcs->rxd_64[i].ctrl1);
1072 }
1073 }
1074 return PDC_SUCCESS;
1075
1076fail_dealloc:
1077 dma_pool_free(pdcs->ring_pool, tx.vbase, tx.dmabase);
1078done:
1079 return err;
1080}
1081
1082static void pdc_ring_free(struct pdc_state *pdcs)
1083{
1084 if (pdcs->tx_ring_alloc.vbase) {
1085 dma_pool_free(pdcs->ring_pool, pdcs->tx_ring_alloc.vbase,
1086 pdcs->tx_ring_alloc.dmabase);
1087 pdcs->tx_ring_alloc.vbase = NULL;
1088 }
1089
1090 if (pdcs->rx_ring_alloc.vbase) {
1091 dma_pool_free(pdcs->ring_pool, pdcs->rx_ring_alloc.vbase,
1092 pdcs->rx_ring_alloc.dmabase);
1093 pdcs->rx_ring_alloc.vbase = NULL;
1094 }
1095}
1096
1097/**
1098 * pdc_desc_count() - Count the number of DMA descriptors that will be required
1099 * for a given scatterlist. Account for the max length of a DMA buffer.
1100 * @sg: Scatterlist to be DMA'd
1101 * Return: Number of descriptors required
1102 */
1103static u32 pdc_desc_count(struct scatterlist *sg)
1104{
1105 u32 cnt = 0;
1106
1107 while (sg) {
1108 cnt += ((sg->length / PDC_DMA_BUF_MAX) + 1);
1109 sg = sg_next(sg);
1110 }
1111 return cnt;
1112}
1113
1114/**
1115 * pdc_rings_full() - Check whether the tx ring has room for tx_cnt descriptors
1116 * and the rx ring has room for rx_cnt descriptors.
1117 * @pdcs: PDC state
1118 * @tx_cnt: The number of descriptors required in the tx ring
1119 * @rx_cnt: The number of descriptors required i the rx ring
1120 *
1121 * Return: true if one of the rings does not have enough space
1122 * false if sufficient space is available in both rings
1123 */
1124static bool pdc_rings_full(struct pdc_state *pdcs, int tx_cnt, int rx_cnt)
1125{
1126 u32 rx_avail;
1127 u32 tx_avail;
1128 bool full = false;
1129
1130 /* Check if the tx and rx rings are likely to have enough space */
1131 rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
1132 pdcs->nrxpost);
1133 if (unlikely(rx_cnt > rx_avail)) {
1134 pdcs->rx_ring_full++;
1135 full = true;
1136 }
1137
1138 if (likely(!full)) {
1139 tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
1140 pdcs->ntxpost);
1141 if (unlikely(tx_cnt > tx_avail)) {
1142 pdcs->tx_ring_full++;
1143 full = true;
1144 }
1145 }
1146 return full;
1147}
1148
1149/**
1150 * pdc_last_tx_done() - If both the tx and rx rings have at least
1151 * PDC_RING_SPACE_MIN descriptors available, then indicate that the mailbox
1152 * framework can submit another message.
1153 * @chan: mailbox channel to check
1154 * Return: true if PDC can accept another message on this channel
1155 */
1156static bool pdc_last_tx_done(struct mbox_chan *chan)
1157{
1158 struct pdc_state *pdcs = chan->con_priv;
1159 bool ret;
1160
1161 if (unlikely(pdc_rings_full(pdcs, PDC_RING_SPACE_MIN,
1162 PDC_RING_SPACE_MIN))) {
1163 pdcs->last_tx_not_done++;
1164 ret = false;
1165 } else {
1166 ret = true;
1167 }
1168 return ret;
1169}
1170
1171/**
1172 * pdc_send_data() - mailbox send_data function
1173 * @chan: The mailbox channel on which the data is sent. The channel
1174 * corresponds to a DMA ringset.
1175 * @data: The mailbox message to be sent. The message must be a
1176 * brcm_message structure.
1177 *
1178 * This function is registered as the send_data function for the mailbox
1179 * controller. From the destination scatterlist in the mailbox message, it
1180 * creates a sequence of receive descriptors in the rx ring. From the source
1181 * scatterlist, it creates a sequence of transmit descriptors in the tx ring.
1182 * After creating the descriptors, it writes the rx ptr and tx ptr registers to
1183 * initiate the DMA transfer.
1184 *
1185 * This function does the DMA map and unmap of the src and dst scatterlists in
1186 * the mailbox message.
1187 *
1188 * Return: 0 if successful
1189 * -ENOTSUPP if the mailbox message is a type this driver does not
1190 * support
1191 * < 0 if an error
1192 */
1193static int pdc_send_data(struct mbox_chan *chan, void *data)
1194{
1195 struct pdc_state *pdcs = chan->con_priv;
1196 struct device *dev = &pdcs->pdev->dev;
1197 struct brcm_message *mssg = data;
1198 int err = PDC_SUCCESS;
1199 int src_nent;
1200 int dst_nent;
1201 int nent;
1202 u32 tx_desc_req;
1203 u32 rx_desc_req;
1204
1205 if (unlikely(mssg->type != BRCM_MESSAGE_SPU))
1206 return -ENOTSUPP;
1207
1208 src_nent = sg_nents(mssg->spu.src);
1209 if (likely(src_nent)) {
1210 nent = dma_map_sg(dev, mssg->spu.src, src_nent, DMA_TO_DEVICE);
1211 if (unlikely(nent == 0))
1212 return -EIO;
1213 }
1214
1215 dst_nent = sg_nents(mssg->spu.dst);
1216 if (likely(dst_nent)) {
1217 nent = dma_map_sg(dev, mssg->spu.dst, dst_nent,
1218 DMA_FROM_DEVICE);
1219 if (unlikely(nent == 0)) {
1220 dma_unmap_sg(dev, mssg->spu.src, src_nent,
1221 DMA_TO_DEVICE);
1222 return -EIO;
1223 }
1224 }
1225
1226 /*
1227 * Check if the tx and rx rings have enough space. Do this prior to
1228 * writing any tx or rx descriptors. Need to ensure that we do not write
1229 * a partial set of descriptors, or write just rx descriptors but
1230 * corresponding tx descriptors don't fit. Note that we want this check
1231 * and the entire sequence of descriptor to happen without another
1232 * thread getting in. The channel spin lock in the mailbox framework
1233 * ensures this.
1234 */
1235 tx_desc_req = pdc_desc_count(mssg->spu.src);
1236 rx_desc_req = pdc_desc_count(mssg->spu.dst);
1237 if (unlikely(pdc_rings_full(pdcs, tx_desc_req, rx_desc_req + 1)))
1238 return -ENOSPC;
1239
1240 /* Create rx descriptors to SPU catch response */
1241 err = pdc_rx_list_init(pdcs, mssg->spu.dst, mssg->ctx);
1242 err |= pdc_rx_list_sg_add(pdcs, mssg->spu.dst);
1243
1244 /* Create tx descriptors to submit SPU request */
1245 err |= pdc_tx_list_sg_add(pdcs, mssg->spu.src);
1246 err |= pdc_tx_list_final(pdcs); /* initiate transfer */
1247
1248 if (unlikely(err))
1249 dev_err(&pdcs->pdev->dev,
1250 "%s failed with error %d", __func__, err);
1251
1252 return err;
1253}
1254
1255static int pdc_startup(struct mbox_chan *chan)
1256{
1257 return pdc_ring_init(chan->con_priv, PDC_RINGSET);
1258}
1259
1260static void pdc_shutdown(struct mbox_chan *chan)
1261{
1262 struct pdc_state *pdcs = chan->con_priv;
1263
1264 if (!pdcs)
1265 return;
1266
1267 dev_dbg(&pdcs->pdev->dev,
1268 "Shutdown mailbox channel for PDC %u", pdcs->pdc_idx);
1269 pdc_ring_free(pdcs);
1270}
1271
1272/**
1273 * pdc_hw_init() - Use the given initialization parameters to initialize the
1274 * state for one of the PDCs.
1275 * @pdcs: state of the PDC
1276 */
1277static
1278void pdc_hw_init(struct pdc_state *pdcs)
1279{
1280 struct platform_device *pdev;
1281 struct device *dev;
1282 struct dma64 *dma_reg;
1283 int ringset = PDC_RINGSET;
1284
1285 pdev = pdcs->pdev;
1286 dev = &pdev->dev;
1287
1288 dev_dbg(dev, "PDC %u initial values:", pdcs->pdc_idx);
1289 dev_dbg(dev, "state structure: %p",
1290 pdcs);
1291 dev_dbg(dev, " - base virtual addr of hw regs %p",
1292 pdcs->pdc_reg_vbase);
1293
1294 /* initialize data structures */
1295 pdcs->regs = (struct pdc_regs *)pdcs->pdc_reg_vbase;
1296 pdcs->txregs_64 = (struct dma64_regs *)
1297 (((u8 *)pdcs->pdc_reg_vbase) +
1298 PDC_TXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1299 pdcs->rxregs_64 = (struct dma64_regs *)
1300 (((u8 *)pdcs->pdc_reg_vbase) +
1301 PDC_RXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1302
1303 pdcs->ntxd = PDC_RING_ENTRIES;
1304 pdcs->nrxd = PDC_RING_ENTRIES;
1305 pdcs->ntxpost = PDC_RING_ENTRIES - 1;
1306 pdcs->nrxpost = PDC_RING_ENTRIES - 1;
1307 iowrite32(0, &pdcs->regs->intmask);
1308
1309 dma_reg = &pdcs->regs->dmaregs[ringset];
1310
1311 /* Configure DMA but will enable later in pdc_ring_init() */
1312 iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1313
1314 iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1315 &dma_reg->dmarcv.control);
1316
1317 /* Reset current index pointers after making sure DMA is disabled */
1318 iowrite32(0, &dma_reg->dmaxmt.ptr);
1319 iowrite32(0, &dma_reg->dmarcv.ptr);
1320
1321 if (pdcs->pdc_resp_hdr_len == PDC_SPU2_RESP_HDR_LEN)
1322 iowrite32(PDC_CKSUM_CTRL,
1323 pdcs->pdc_reg_vbase + PDC_CKSUM_CTRL_OFFSET);
1324}
1325
1326/**
1327 * pdc_hw_disable() - Disable the tx and rx control in the hw.
1328 * @pdcs: PDC state structure
1329 *
1330 */
1331static void pdc_hw_disable(struct pdc_state *pdcs)
1332{
1333 struct dma64 *dma_reg;
1334
1335 dma_reg = &pdcs->regs->dmaregs[PDC_RINGSET];
1336 iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1337 iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1338 &dma_reg->dmarcv.control);
1339}
1340
1341/**
1342 * pdc_rx_buf_pool_create() - Pool of receive buffers used to catch the metadata
1343 * header returned with each response message.
1344 * @pdcs: PDC state structure
1345 *
1346 * The metadata is not returned to the mailbox client. So the PDC driver
1347 * manages these buffers.
1348 *
1349 * Return: PDC_SUCCESS
1350 * -ENOMEM if pool creation fails
1351 */
1352static int pdc_rx_buf_pool_create(struct pdc_state *pdcs)
1353{
1354 struct platform_device *pdev;
1355 struct device *dev;
1356
1357 pdev = pdcs->pdev;
1358 dev = &pdev->dev;
1359
1360 pdcs->pdc_resp_hdr_len = pdcs->rx_status_len;
1361 if (pdcs->use_bcm_hdr)
1362 pdcs->pdc_resp_hdr_len += BCM_HDR_LEN;
1363
1364 pdcs->rx_buf_pool = dma_pool_create("pdc rx bufs", dev,
1365 pdcs->pdc_resp_hdr_len,
1366 RX_BUF_ALIGN, 0);
1367 if (!pdcs->rx_buf_pool)
1368 return -ENOMEM;
1369
1370 return PDC_SUCCESS;
1371}
1372
1373/**
1374 * pdc_interrupts_init() - Initialize the interrupt configuration for a PDC and
1375 * specify a threaded IRQ handler for deferred handling of interrupts outside of
1376 * interrupt context.
1377 * @pdcs: PDC state
1378 *
1379 * Set the interrupt mask for transmit and receive done.
1380 * Set the lazy interrupt frame count to generate an interrupt for just one pkt.
1381 *
1382 * Return: PDC_SUCCESS
1383 * <0 if threaded irq request fails
1384 */
1385static int pdc_interrupts_init(struct pdc_state *pdcs)
1386{
1387 struct platform_device *pdev = pdcs->pdev;
1388 struct device *dev = &pdev->dev;
1389 struct device_node *dn = pdev->dev.of_node;
1390 int err;
1391
1392 /* interrupt configuration */
1393 iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
1394
1395 if (pdcs->hw_type == FA_HW)
1396 iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1397 FA_RCVLAZY0_OFFSET);
1398 else
1399 iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1400 PDC_RCVLAZY0_OFFSET);
1401
1402 /* read irq from device tree */
1403 pdcs->pdc_irq = irq_of_parse_and_map(dn, 0);
1404 dev_dbg(dev, "pdc device %s irq %u for pdcs %p",
1405 dev_name(dev), pdcs->pdc_irq, pdcs);
1406
1407 err = devm_request_irq(dev, pdcs->pdc_irq, pdc_irq_handler, 0,
1408 dev_name(dev), dev);
1409 if (err) {
1410 dev_err(dev, "IRQ %u request failed with err %d\n",
1411 pdcs->pdc_irq, err);
1412 return err;
1413 }
1414 return PDC_SUCCESS;
1415}
1416
1417static const struct mbox_chan_ops pdc_mbox_chan_ops = {
1418 .send_data = pdc_send_data,
1419 .last_tx_done = pdc_last_tx_done,
1420 .startup = pdc_startup,
1421 .shutdown = pdc_shutdown
1422};
1423
1424/**
1425 * pdc_mb_init() - Initialize the mailbox controller.
1426 * @pdcs: PDC state
1427 *
1428 * Each PDC is a mailbox controller. Each ringset is a mailbox channel. Kernel
1429 * driver only uses one ringset and thus one mb channel. PDC uses the transmit
1430 * complete interrupt to determine when a mailbox message has successfully been
1431 * transmitted.
1432 *
1433 * Return: 0 on success
1434 * < 0 if there is an allocation or registration failure
1435 */
1436static int pdc_mb_init(struct pdc_state *pdcs)
1437{
1438 struct device *dev = &pdcs->pdev->dev;
1439 struct mbox_controller *mbc;
1440 int chan_index;
1441 int err;
1442
1443 mbc = &pdcs->mbc;
1444 mbc->dev = dev;
1445 mbc->ops = &pdc_mbox_chan_ops;
1446 mbc->num_chans = 1;
1447 mbc->chans = devm_kcalloc(dev, mbc->num_chans, sizeof(*mbc->chans),
1448 GFP_KERNEL);
1449 if (!mbc->chans)
1450 return -ENOMEM;
1451
1452 mbc->txdone_irq = false;
1453 mbc->txdone_poll = true;
1454 mbc->txpoll_period = 1;
1455 for (chan_index = 0; chan_index < mbc->num_chans; chan_index++)
1456 mbc->chans[chan_index].con_priv = pdcs;
1457
1458 /* Register mailbox controller */
1459 err = devm_mbox_controller_register(dev, mbc);
1460 if (err) {
1461 dev_crit(dev,
1462 "Failed to register PDC mailbox controller. Error %d.",
1463 err);
1464 return err;
1465 }
1466 return 0;
1467}
1468
1469/* Device tree API */
1470static const int pdc_hw = PDC_HW;
1471static const int fa_hw = FA_HW;
1472
1473static const struct of_device_id pdc_mbox_of_match[] = {
1474 {.compatible = "brcm,iproc-pdc-mbox", .data = &pdc_hw},
1475 {.compatible = "brcm,iproc-fa2-mbox", .data = &fa_hw},
1476 { /* sentinel */ }
1477};
1478MODULE_DEVICE_TABLE(of, pdc_mbox_of_match);
1479
1480/**
1481 * pdc_dt_read() - Read application-specific data from device tree.
1482 * @pdev: Platform device
1483 * @pdcs: PDC state
1484 *
1485 * Reads the number of bytes of receive status that precede each received frame.
1486 * Reads whether transmit and received frames should be preceded by an 8-byte
1487 * BCM header.
1488 *
1489 * Return: 0 if successful
1490 * -ENODEV if device not available
1491 */
1492static int pdc_dt_read(struct platform_device *pdev, struct pdc_state *pdcs)
1493{
1494 struct device *dev = &pdev->dev;
1495 struct device_node *dn = pdev->dev.of_node;
1496 const int *hw_type;
1497 int err;
1498
1499 err = of_property_read_u32(dn, "brcm,rx-status-len",
1500 &pdcs->rx_status_len);
1501 if (err < 0)
1502 dev_err(dev,
1503 "%s failed to get DMA receive status length from device tree",
1504 __func__);
1505
1506 pdcs->use_bcm_hdr = of_property_read_bool(dn, "brcm,use-bcm-hdr");
1507
1508 pdcs->hw_type = PDC_HW;
1509
1510 hw_type = device_get_match_data(dev);
1511 if (hw_type)
1512 pdcs->hw_type = *hw_type;
1513
1514 return 0;
1515}
1516
1517/**
1518 * pdc_probe() - Probe function for PDC driver.
1519 * @pdev: PDC platform device
1520 *
1521 * Reserve and map register regions defined in device tree.
1522 * Allocate and initialize tx and rx DMA rings.
1523 * Initialize a mailbox controller for each PDC.
1524 *
1525 * Return: 0 if successful
1526 * < 0 if an error
1527 */
1528static int pdc_probe(struct platform_device *pdev)
1529{
1530 int err = 0;
1531 struct device *dev = &pdev->dev;
1532 struct resource *pdc_regs;
1533 struct pdc_state *pdcs;
1534
1535 /* PDC state for one SPU */
1536 pdcs = devm_kzalloc(dev, sizeof(*pdcs), GFP_KERNEL);
1537 if (!pdcs) {
1538 err = -ENOMEM;
1539 goto cleanup;
1540 }
1541
1542 pdcs->pdev = pdev;
1543 platform_set_drvdata(pdev, pdcs);
1544 pdcs->pdc_idx = pdcg.num_spu;
1545 pdcg.num_spu++;
1546
1547 err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(39));
1548 if (err) {
1549 dev_warn(dev, "PDC device cannot perform DMA. Error %d.", err);
1550 goto cleanup;
1551 }
1552
1553 /* Create DMA pool for tx ring */
1554 pdcs->ring_pool = dma_pool_create("pdc rings", dev, PDC_RING_SIZE,
1555 RING_ALIGN, 0);
1556 if (!pdcs->ring_pool) {
1557 err = -ENOMEM;
1558 goto cleanup;
1559 }
1560
1561 err = pdc_dt_read(pdev, pdcs);
1562 if (err)
1563 goto cleanup_ring_pool;
1564
1565 pdcs->pdc_reg_vbase = devm_platform_get_and_ioremap_resource(pdev, 0, &pdc_regs);
1566 if (IS_ERR(pdcs->pdc_reg_vbase)) {
1567 err = PTR_ERR(pdcs->pdc_reg_vbase);
1568 goto cleanup_ring_pool;
1569 }
1570 dev_dbg(dev, "PDC register region res.start = %pa, res.end = %pa",
1571 &pdc_regs->start, &pdc_regs->end);
1572
1573 /* create rx buffer pool after dt read to know how big buffers are */
1574 err = pdc_rx_buf_pool_create(pdcs);
1575 if (err)
1576 goto cleanup_ring_pool;
1577
1578 pdc_hw_init(pdcs);
1579
1580 /* Init tasklet for deferred DMA rx processing */
1581 tasklet_setup(&pdcs->rx_tasklet, pdc_tasklet_cb);
1582
1583 err = pdc_interrupts_init(pdcs);
1584 if (err)
1585 goto cleanup_buf_pool;
1586
1587 /* Initialize mailbox controller */
1588 err = pdc_mb_init(pdcs);
1589 if (err)
1590 goto cleanup_buf_pool;
1591
1592 pdc_setup_debugfs(pdcs);
1593
1594 dev_dbg(dev, "pdc_probe() successful");
1595 return PDC_SUCCESS;
1596
1597cleanup_buf_pool:
1598 tasklet_kill(&pdcs->rx_tasklet);
1599 dma_pool_destroy(pdcs->rx_buf_pool);
1600
1601cleanup_ring_pool:
1602 dma_pool_destroy(pdcs->ring_pool);
1603
1604cleanup:
1605 return err;
1606}
1607
1608static void pdc_remove(struct platform_device *pdev)
1609{
1610 struct pdc_state *pdcs = platform_get_drvdata(pdev);
1611
1612 pdc_free_debugfs();
1613
1614 tasklet_kill(&pdcs->rx_tasklet);
1615
1616 pdc_hw_disable(pdcs);
1617
1618 dma_pool_destroy(pdcs->rx_buf_pool);
1619 dma_pool_destroy(pdcs->ring_pool);
1620}
1621
1622static struct platform_driver pdc_mbox_driver = {
1623 .probe = pdc_probe,
1624 .remove_new = pdc_remove,
1625 .driver = {
1626 .name = "brcm-iproc-pdc-mbox",
1627 .of_match_table = pdc_mbox_of_match,
1628 },
1629};
1630module_platform_driver(pdc_mbox_driver);
1631
1632MODULE_AUTHOR("Rob Rice <rob.rice@broadcom.com>");
1633MODULE_DESCRIPTION("Broadcom PDC mailbox driver");
1634MODULE_LICENSE("GPL v2");
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
167enum 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
172struct pdc_dma_map {
173 void *ctx; /* opaque context associated with frame */
174};
175
176/* dma descriptor */
177struct 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) */
185struct 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. */
202struct 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 */
210struct 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 */
264struct 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 */
283struct 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 */
292struct 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 struct dentry *debugfs_stats; /* debug FS stats file for this PDC */
410
411 /* counters */
412 u32 pdc_requests; /* number of request messages submitted */
413 u32 pdc_replies; /* number of reply messages received */
414 u32 last_tx_not_done; /* too few tx descriptors to indicate done */
415 u32 tx_ring_full; /* unable to accept msg because tx ring full */
416 u32 rx_ring_full; /* unable to accept msg because rx ring full */
417 u32 txnobuf; /* unable to create tx descriptor */
418 u32 rxnobuf; /* unable to create rx descriptor */
419 u32 rx_oflow; /* count of rx overflows */
420
421 /* hardware type - FA2 or PDC/MDE */
422 enum pdc_hw hw_type;
423};
424
425/* Global variables */
426
427struct pdc_globals {
428 /* Actual number of SPUs in hardware, as reported by device tree */
429 u32 num_spu;
430};
431
432static struct pdc_globals pdcg;
433
434/* top level debug FS directory for PDC driver */
435static struct dentry *debugfs_dir;
436
437static ssize_t pdc_debugfs_read(struct file *filp, char __user *ubuf,
438 size_t count, loff_t *offp)
439{
440 struct pdc_state *pdcs;
441 char *buf;
442 ssize_t ret, out_offset, out_count;
443
444 out_count = 512;
445
446 buf = kmalloc(out_count, GFP_KERNEL);
447 if (!buf)
448 return -ENOMEM;
449
450 pdcs = filp->private_data;
451 out_offset = 0;
452 out_offset += snprintf(buf + out_offset, out_count - out_offset,
453 "SPU %u stats:\n", pdcs->pdc_idx);
454 out_offset += snprintf(buf + out_offset, out_count - out_offset,
455 "PDC requests....................%u\n",
456 pdcs->pdc_requests);
457 out_offset += snprintf(buf + out_offset, out_count - out_offset,
458 "PDC responses...................%u\n",
459 pdcs->pdc_replies);
460 out_offset += snprintf(buf + out_offset, out_count - out_offset,
461 "Tx not done.....................%u\n",
462 pdcs->last_tx_not_done);
463 out_offset += snprintf(buf + out_offset, out_count - out_offset,
464 "Tx ring full....................%u\n",
465 pdcs->tx_ring_full);
466 out_offset += snprintf(buf + out_offset, out_count - out_offset,
467 "Rx ring full....................%u\n",
468 pdcs->rx_ring_full);
469 out_offset += snprintf(buf + out_offset, out_count - out_offset,
470 "Tx desc write fail. Ring full...%u\n",
471 pdcs->txnobuf);
472 out_offset += snprintf(buf + out_offset, out_count - out_offset,
473 "Rx desc write fail. Ring full...%u\n",
474 pdcs->rxnobuf);
475 out_offset += snprintf(buf + out_offset, out_count - out_offset,
476 "Receive overflow................%u\n",
477 pdcs->rx_oflow);
478 out_offset += snprintf(buf + out_offset, out_count - out_offset,
479 "Num frags in rx ring............%u\n",
480 NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr,
481 pdcs->nrxpost));
482
483 if (out_offset > out_count)
484 out_offset = out_count;
485
486 ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset);
487 kfree(buf);
488 return ret;
489}
490
491static const struct file_operations pdc_debugfs_stats = {
492 .owner = THIS_MODULE,
493 .open = simple_open,
494 .read = pdc_debugfs_read,
495};
496
497/**
498 * pdc_setup_debugfs() - Create the debug FS directories. If the top-level
499 * directory has not yet been created, create it now. Create a stats file in
500 * this directory for a SPU.
501 * @pdcs: PDC state structure
502 */
503static void pdc_setup_debugfs(struct pdc_state *pdcs)
504{
505 char spu_stats_name[16];
506
507 if (!debugfs_initialized())
508 return;
509
510 snprintf(spu_stats_name, 16, "pdc%d_stats", pdcs->pdc_idx);
511 if (!debugfs_dir)
512 debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
513
514 /* S_IRUSR == 0400 */
515 pdcs->debugfs_stats = debugfs_create_file(spu_stats_name, 0400,
516 debugfs_dir, pdcs,
517 &pdc_debugfs_stats);
518}
519
520static void pdc_free_debugfs(void)
521{
522 debugfs_remove_recursive(debugfs_dir);
523 debugfs_dir = NULL;
524}
525
526/**
527 * pdc_build_rxd() - Build DMA descriptor to receive SPU result.
528 * @pdcs: PDC state for SPU that will generate result
529 * @dma_addr: DMA address of buffer that descriptor is being built for
530 * @buf_len: Length of the receive buffer, in bytes
531 * @flags: Flags to be stored in descriptor
532 */
533static inline void
534pdc_build_rxd(struct pdc_state *pdcs, dma_addr_t dma_addr,
535 u32 buf_len, u32 flags)
536{
537 struct device *dev = &pdcs->pdev->dev;
538 struct dma64dd *rxd = &pdcs->rxd_64[pdcs->rxout];
539
540 dev_dbg(dev,
541 "Writing rx descriptor for PDC %u at index %u with length %u. flags %#x\n",
542 pdcs->pdc_idx, pdcs->rxout, buf_len, flags);
543
544 rxd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
545 rxd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
546 rxd->ctrl1 = cpu_to_le32(flags);
547 rxd->ctrl2 = cpu_to_le32(buf_len);
548
549 /* bump ring index and return */
550 pdcs->rxout = NEXTRXD(pdcs->rxout, pdcs->nrxpost);
551}
552
553/**
554 * pdc_build_txd() - Build a DMA descriptor to transmit a SPU request to
555 * hardware.
556 * @pdcs: PDC state for the SPU that will process this request
557 * @dma_addr: DMA address of packet to be transmitted
558 * @buf_len: Length of tx buffer, in bytes
559 * @flags: Flags to be stored in descriptor
560 */
561static inline void
562pdc_build_txd(struct pdc_state *pdcs, dma_addr_t dma_addr, u32 buf_len,
563 u32 flags)
564{
565 struct device *dev = &pdcs->pdev->dev;
566 struct dma64dd *txd = &pdcs->txd_64[pdcs->txout];
567
568 dev_dbg(dev,
569 "Writing tx descriptor for PDC %u at index %u with length %u, flags %#x\n",
570 pdcs->pdc_idx, pdcs->txout, buf_len, flags);
571
572 txd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
573 txd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
574 txd->ctrl1 = cpu_to_le32(flags);
575 txd->ctrl2 = cpu_to_le32(buf_len);
576
577 /* bump ring index and return */
578 pdcs->txout = NEXTTXD(pdcs->txout, pdcs->ntxpost);
579}
580
581/**
582 * pdc_receive_one() - Receive a response message from a given SPU.
583 * @pdcs: PDC state for the SPU to receive from
584 *
585 * When the return code indicates success, the response message is available in
586 * the receive buffers provided prior to submission of the request.
587 *
588 * Return: PDC_SUCCESS if one or more receive descriptors was processed
589 * -EAGAIN indicates that no response message is available
590 * -EIO an error occurred
591 */
592static int
593pdc_receive_one(struct pdc_state *pdcs)
594{
595 struct device *dev = &pdcs->pdev->dev;
596 struct mbox_controller *mbc;
597 struct mbox_chan *chan;
598 struct brcm_message mssg;
599 u32 len, rx_status;
600 u32 num_frags;
601 u8 *resp_hdr; /* virtual addr of start of resp message DMA header */
602 u32 frags_rdy; /* number of fragments ready to read */
603 u32 rx_idx; /* ring index of start of receive frame */
604 dma_addr_t resp_hdr_daddr;
605 struct pdc_rx_ctx *rx_ctx;
606
607 mbc = &pdcs->mbc;
608 chan = &mbc->chans[0];
609 mssg.type = BRCM_MESSAGE_SPU;
610
611 /*
612 * return if a complete response message is not yet ready.
613 * rxin_numd[rxin] is the number of fragments in the next msg
614 * to read.
615 */
616 frags_rdy = NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr, pdcs->nrxpost);
617 if ((frags_rdy == 0) ||
618 (frags_rdy < pdcs->rx_ctx[pdcs->rxin].rxin_numd))
619 /* No response ready */
620 return -EAGAIN;
621
622 num_frags = pdcs->txin_numd[pdcs->txin];
623 WARN_ON(num_frags == 0);
624
625 dma_unmap_sg(dev, pdcs->src_sg[pdcs->txin],
626 sg_nents(pdcs->src_sg[pdcs->txin]), DMA_TO_DEVICE);
627
628 pdcs->txin = (pdcs->txin + num_frags) & pdcs->ntxpost;
629
630 dev_dbg(dev, "PDC %u reclaimed %d tx descriptors",
631 pdcs->pdc_idx, num_frags);
632
633 rx_idx = pdcs->rxin;
634 rx_ctx = &pdcs->rx_ctx[rx_idx];
635 num_frags = rx_ctx->rxin_numd;
636 /* Return opaque context with result */
637 mssg.ctx = rx_ctx->rxp_ctx;
638 rx_ctx->rxp_ctx = NULL;
639 resp_hdr = rx_ctx->resp_hdr;
640 resp_hdr_daddr = rx_ctx->resp_hdr_daddr;
641 dma_unmap_sg(dev, rx_ctx->dst_sg, sg_nents(rx_ctx->dst_sg),
642 DMA_FROM_DEVICE);
643
644 pdcs->rxin = (pdcs->rxin + num_frags) & pdcs->nrxpost;
645
646 dev_dbg(dev, "PDC %u reclaimed %d rx descriptors",
647 pdcs->pdc_idx, num_frags);
648
649 dev_dbg(dev,
650 "PDC %u txin %u, txout %u, rxin %u, rxout %u, last_rx_curr %u\n",
651 pdcs->pdc_idx, pdcs->txin, pdcs->txout, pdcs->rxin,
652 pdcs->rxout, pdcs->last_rx_curr);
653
654 if (pdcs->pdc_resp_hdr_len == PDC_SPUM_RESP_HDR_LEN) {
655 /*
656 * For SPU-M, get length of response msg and rx overflow status.
657 */
658 rx_status = *((u32 *)resp_hdr);
659 len = rx_status & RX_STATUS_LEN;
660 dev_dbg(dev,
661 "SPU response length %u bytes", len);
662 if (unlikely(((rx_status & RX_STATUS_OVERFLOW) || (!len)))) {
663 if (rx_status & RX_STATUS_OVERFLOW) {
664 dev_err_ratelimited(dev,
665 "crypto receive overflow");
666 pdcs->rx_oflow++;
667 } else {
668 dev_info_ratelimited(dev, "crypto rx len = 0");
669 }
670 return -EIO;
671 }
672 }
673
674 dma_pool_free(pdcs->rx_buf_pool, resp_hdr, resp_hdr_daddr);
675
676 mbox_chan_received_data(chan, &mssg);
677
678 pdcs->pdc_replies++;
679 return PDC_SUCCESS;
680}
681
682/**
683 * pdc_receive() - Process as many responses as are available in the rx ring.
684 * @pdcs: PDC state
685 *
686 * Called within the hard IRQ.
687 * Return:
688 */
689static int
690pdc_receive(struct pdc_state *pdcs)
691{
692 int rx_status;
693
694 /* read last_rx_curr from register once */
695 pdcs->last_rx_curr =
696 (ioread32(&pdcs->rxregs_64->status0) &
697 CRYPTO_D64_RS0_CD_MASK) / RING_ENTRY_SIZE;
698
699 do {
700 /* Could be many frames ready */
701 rx_status = pdc_receive_one(pdcs);
702 } while (rx_status == PDC_SUCCESS);
703
704 return 0;
705}
706
707/**
708 * pdc_tx_list_sg_add() - Add the buffers in a scatterlist to the transmit
709 * descriptors for a given SPU. The scatterlist buffers contain the data for a
710 * SPU request message.
711 * @spu_idx: The index of the SPU to submit the request to, [0, max_spu)
712 * @sg: Scatterlist whose buffers contain part of the SPU request
713 *
714 * If a scatterlist buffer is larger than PDC_DMA_BUF_MAX, multiple descriptors
715 * are written for that buffer, each <= PDC_DMA_BUF_MAX byte in length.
716 *
717 * Return: PDC_SUCCESS if successful
718 * < 0 otherwise
719 */
720static int pdc_tx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
721{
722 u32 flags = 0;
723 u32 eot;
724 u32 tx_avail;
725
726 /*
727 * Num descriptors needed. Conservatively assume we need a descriptor
728 * for every entry in sg.
729 */
730 u32 num_desc;
731 u32 desc_w = 0; /* Number of tx descriptors written */
732 u32 bufcnt; /* Number of bytes of buffer pointed to by descriptor */
733 dma_addr_t databufptr; /* DMA address to put in descriptor */
734
735 num_desc = (u32)sg_nents(sg);
736
737 /* check whether enough tx descriptors are available */
738 tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
739 pdcs->ntxpost);
740 if (unlikely(num_desc > tx_avail)) {
741 pdcs->txnobuf++;
742 return -ENOSPC;
743 }
744
745 /* build tx descriptors */
746 if (pdcs->tx_msg_start == pdcs->txout) {
747 /* Start of frame */
748 pdcs->txin_numd[pdcs->tx_msg_start] = 0;
749 pdcs->src_sg[pdcs->txout] = sg;
750 flags = D64_CTRL1_SOF;
751 }
752
753 while (sg) {
754 if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
755 eot = D64_CTRL1_EOT;
756 else
757 eot = 0;
758
759 /*
760 * If sg buffer larger than PDC limit, split across
761 * multiple descriptors
762 */
763 bufcnt = sg_dma_len(sg);
764 databufptr = sg_dma_address(sg);
765 while (bufcnt > PDC_DMA_BUF_MAX) {
766 pdc_build_txd(pdcs, databufptr, PDC_DMA_BUF_MAX,
767 flags | eot);
768 desc_w++;
769 bufcnt -= PDC_DMA_BUF_MAX;
770 databufptr += PDC_DMA_BUF_MAX;
771 if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
772 eot = D64_CTRL1_EOT;
773 else
774 eot = 0;
775 }
776 sg = sg_next(sg);
777 if (!sg)
778 /* Writing last descriptor for frame */
779 flags |= (D64_CTRL1_EOF | D64_CTRL1_IOC);
780 pdc_build_txd(pdcs, databufptr, bufcnt, flags | eot);
781 desc_w++;
782 /* Clear start of frame after first descriptor */
783 flags &= ~D64_CTRL1_SOF;
784 }
785 pdcs->txin_numd[pdcs->tx_msg_start] += desc_w;
786
787 return PDC_SUCCESS;
788}
789
790/**
791 * pdc_tx_list_final() - Initiate DMA transfer of last frame written to tx
792 * ring.
793 * @pdcs: PDC state for SPU to process the request
794 *
795 * Sets the index of the last descriptor written in both the rx and tx ring.
796 *
797 * Return: PDC_SUCCESS
798 */
799static int pdc_tx_list_final(struct pdc_state *pdcs)
800{
801 /*
802 * write barrier to ensure all register writes are complete
803 * before chip starts to process new request
804 */
805 wmb();
806 iowrite32(pdcs->rxout << 4, &pdcs->rxregs_64->ptr);
807 iowrite32(pdcs->txout << 4, &pdcs->txregs_64->ptr);
808 pdcs->pdc_requests++;
809
810 return PDC_SUCCESS;
811}
812
813/**
814 * pdc_rx_list_init() - Start a new receive descriptor list for a given PDC.
815 * @pdcs: PDC state for SPU handling request
816 * @dst_sg: scatterlist providing rx buffers for response to be returned to
817 * mailbox client
818 * @ctx: Opaque context for this request
819 *
820 * Posts a single receive descriptor to hold the metadata that precedes a
821 * response. For example, with SPU-M, the metadata is a 32-byte DMA header and
822 * an 8-byte BCM header. Moves the msg_start descriptor indexes for both tx and
823 * rx to indicate the start of a new message.
824 *
825 * Return: PDC_SUCCESS if successful
826 * < 0 if an error (e.g., rx ring is full)
827 */
828static int pdc_rx_list_init(struct pdc_state *pdcs, struct scatterlist *dst_sg,
829 void *ctx)
830{
831 u32 flags = 0;
832 u32 rx_avail;
833 u32 rx_pkt_cnt = 1; /* Adding a single rx buffer */
834 dma_addr_t daddr;
835 void *vaddr;
836 struct pdc_rx_ctx *rx_ctx;
837
838 rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
839 pdcs->nrxpost);
840 if (unlikely(rx_pkt_cnt > rx_avail)) {
841 pdcs->rxnobuf++;
842 return -ENOSPC;
843 }
844
845 /* allocate a buffer for the dma rx status */
846 vaddr = dma_pool_zalloc(pdcs->rx_buf_pool, GFP_ATOMIC, &daddr);
847 if (unlikely(!vaddr))
848 return -ENOMEM;
849
850 /*
851 * Update msg_start indexes for both tx and rx to indicate the start
852 * of a new sequence of descriptor indexes that contain the fragments
853 * of the same message.
854 */
855 pdcs->rx_msg_start = pdcs->rxout;
856 pdcs->tx_msg_start = pdcs->txout;
857
858 /* This is always the first descriptor in the receive sequence */
859 flags = D64_CTRL1_SOF;
860 pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd = 1;
861
862 if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
863 flags |= D64_CTRL1_EOT;
864
865 rx_ctx = &pdcs->rx_ctx[pdcs->rxout];
866 rx_ctx->rxp_ctx = ctx;
867 rx_ctx->dst_sg = dst_sg;
868 rx_ctx->resp_hdr = vaddr;
869 rx_ctx->resp_hdr_daddr = daddr;
870 pdc_build_rxd(pdcs, daddr, pdcs->pdc_resp_hdr_len, flags);
871 return PDC_SUCCESS;
872}
873
874/**
875 * pdc_rx_list_sg_add() - Add the buffers in a scatterlist to the receive
876 * descriptors for a given SPU. The caller must have already DMA mapped the
877 * scatterlist.
878 * @spu_idx: Indicates which SPU the buffers are for
879 * @sg: Scatterlist whose buffers are added to the receive ring
880 *
881 * If a receive buffer in the scatterlist is larger than PDC_DMA_BUF_MAX,
882 * multiple receive descriptors are written, each with a buffer <=
883 * PDC_DMA_BUF_MAX.
884 *
885 * Return: PDC_SUCCESS if successful
886 * < 0 otherwise (e.g., receive ring is full)
887 */
888static int pdc_rx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
889{
890 u32 flags = 0;
891 u32 rx_avail;
892
893 /*
894 * Num descriptors needed. Conservatively assume we need a descriptor
895 * for every entry from our starting point in the scatterlist.
896 */
897 u32 num_desc;
898 u32 desc_w = 0; /* Number of tx descriptors written */
899 u32 bufcnt; /* Number of bytes of buffer pointed to by descriptor */
900 dma_addr_t databufptr; /* DMA address to put in descriptor */
901
902 num_desc = (u32)sg_nents(sg);
903
904 rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
905 pdcs->nrxpost);
906 if (unlikely(num_desc > rx_avail)) {
907 pdcs->rxnobuf++;
908 return -ENOSPC;
909 }
910
911 while (sg) {
912 if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
913 flags = D64_CTRL1_EOT;
914 else
915 flags = 0;
916
917 /*
918 * If sg buffer larger than PDC limit, split across
919 * multiple descriptors
920 */
921 bufcnt = sg_dma_len(sg);
922 databufptr = sg_dma_address(sg);
923 while (bufcnt > PDC_DMA_BUF_MAX) {
924 pdc_build_rxd(pdcs, databufptr, PDC_DMA_BUF_MAX, flags);
925 desc_w++;
926 bufcnt -= PDC_DMA_BUF_MAX;
927 databufptr += PDC_DMA_BUF_MAX;
928 if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
929 flags = D64_CTRL1_EOT;
930 else
931 flags = 0;
932 }
933 pdc_build_rxd(pdcs, databufptr, bufcnt, flags);
934 desc_w++;
935 sg = sg_next(sg);
936 }
937 pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd += desc_w;
938
939 return PDC_SUCCESS;
940}
941
942/**
943 * pdc_irq_handler() - Interrupt handler called in interrupt context.
944 * @irq: Interrupt number that has fired
945 * @data: device struct for DMA engine that generated the interrupt
946 *
947 * We have to clear the device interrupt status flags here. So cache the
948 * status for later use in the thread function. Other than that, just return
949 * WAKE_THREAD to invoke the thread function.
950 *
951 * Return: IRQ_WAKE_THREAD if interrupt is ours
952 * IRQ_NONE otherwise
953 */
954static irqreturn_t pdc_irq_handler(int irq, void *data)
955{
956 struct device *dev = (struct device *)data;
957 struct pdc_state *pdcs = dev_get_drvdata(dev);
958 u32 intstatus = ioread32(pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
959
960 if (unlikely(intstatus == 0))
961 return IRQ_NONE;
962
963 /* Disable interrupts until soft handler runs */
964 iowrite32(0, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
965
966 /* Clear interrupt flags in device */
967 iowrite32(intstatus, pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
968
969 /* Wakeup IRQ thread */
970 tasklet_schedule(&pdcs->rx_tasklet);
971 return IRQ_HANDLED;
972}
973
974/**
975 * pdc_tasklet_cb() - Tasklet callback that runs the deferred processing after
976 * a DMA receive interrupt. Reenables the receive interrupt.
977 * @data: PDC state structure
978 */
979static void pdc_tasklet_cb(unsigned long data)
980{
981 struct pdc_state *pdcs = (struct pdc_state *)data;
982
983 pdc_receive(pdcs);
984
985 /* reenable interrupts */
986 iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
987}
988
989/**
990 * pdc_ring_init() - Allocate DMA rings and initialize constant fields of
991 * descriptors in one ringset.
992 * @pdcs: PDC instance state
993 * @ringset: index of ringset being used
994 *
995 * Return: PDC_SUCCESS if ring initialized
996 * < 0 otherwise
997 */
998static int pdc_ring_init(struct pdc_state *pdcs, int ringset)
999{
1000 int i;
1001 int err = PDC_SUCCESS;
1002 struct dma64 *dma_reg;
1003 struct device *dev = &pdcs->pdev->dev;
1004 struct pdc_ring_alloc tx;
1005 struct pdc_ring_alloc rx;
1006
1007 /* Allocate tx ring */
1008 tx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &tx.dmabase);
1009 if (unlikely(!tx.vbase)) {
1010 err = -ENOMEM;
1011 goto done;
1012 }
1013
1014 /* Allocate rx ring */
1015 rx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &rx.dmabase);
1016 if (unlikely(!rx.vbase)) {
1017 err = -ENOMEM;
1018 goto fail_dealloc;
1019 }
1020
1021 dev_dbg(dev, " - base DMA addr of tx ring %pad", &tx.dmabase);
1022 dev_dbg(dev, " - base virtual addr of tx ring %p", tx.vbase);
1023 dev_dbg(dev, " - base DMA addr of rx ring %pad", &rx.dmabase);
1024 dev_dbg(dev, " - base virtual addr of rx ring %p", rx.vbase);
1025
1026 memcpy(&pdcs->tx_ring_alloc, &tx, sizeof(tx));
1027 memcpy(&pdcs->rx_ring_alloc, &rx, sizeof(rx));
1028
1029 pdcs->rxin = 0;
1030 pdcs->rx_msg_start = 0;
1031 pdcs->last_rx_curr = 0;
1032 pdcs->rxout = 0;
1033 pdcs->txin = 0;
1034 pdcs->tx_msg_start = 0;
1035 pdcs->txout = 0;
1036
1037 /* Set descriptor array base addresses */
1038 pdcs->txd_64 = (struct dma64dd *)pdcs->tx_ring_alloc.vbase;
1039 pdcs->rxd_64 = (struct dma64dd *)pdcs->rx_ring_alloc.vbase;
1040
1041 /* Tell device the base DMA address of each ring */
1042 dma_reg = &pdcs->regs->dmaregs[ringset];
1043
1044 /* But first disable DMA and set curptr to 0 for both TX & RX */
1045 iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1046 iowrite32((PDC_RX_CTL + (pdcs->rx_status_len << 1)),
1047 &dma_reg->dmarcv.control);
1048 iowrite32(0, &dma_reg->dmaxmt.ptr);
1049 iowrite32(0, &dma_reg->dmarcv.ptr);
1050
1051 /* Set base DMA addresses */
1052 iowrite32(lower_32_bits(pdcs->tx_ring_alloc.dmabase),
1053 &dma_reg->dmaxmt.addrlow);
1054 iowrite32(upper_32_bits(pdcs->tx_ring_alloc.dmabase),
1055 &dma_reg->dmaxmt.addrhigh);
1056
1057 iowrite32(lower_32_bits(pdcs->rx_ring_alloc.dmabase),
1058 &dma_reg->dmarcv.addrlow);
1059 iowrite32(upper_32_bits(pdcs->rx_ring_alloc.dmabase),
1060 &dma_reg->dmarcv.addrhigh);
1061
1062 /* Re-enable DMA */
1063 iowrite32(PDC_TX_CTL | PDC_TX_ENABLE, &dma_reg->dmaxmt.control);
1064 iowrite32((PDC_RX_CTL | PDC_RX_ENABLE | (pdcs->rx_status_len << 1)),
1065 &dma_reg->dmarcv.control);
1066
1067 /* Initialize descriptors */
1068 for (i = 0; i < PDC_RING_ENTRIES; i++) {
1069 /* Every tx descriptor can be used for start of frame. */
1070 if (i != pdcs->ntxpost) {
1071 iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF,
1072 &pdcs->txd_64[i].ctrl1);
1073 } else {
1074 /* Last descriptor in ringset. Set End of Table. */
1075 iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF |
1076 D64_CTRL1_EOT, &pdcs->txd_64[i].ctrl1);
1077 }
1078
1079 /* Every rx descriptor can be used for start of frame */
1080 if (i != pdcs->nrxpost) {
1081 iowrite32(D64_CTRL1_SOF,
1082 &pdcs->rxd_64[i].ctrl1);
1083 } else {
1084 /* Last descriptor in ringset. Set End of Table. */
1085 iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOT,
1086 &pdcs->rxd_64[i].ctrl1);
1087 }
1088 }
1089 return PDC_SUCCESS;
1090
1091fail_dealloc:
1092 dma_pool_free(pdcs->ring_pool, tx.vbase, tx.dmabase);
1093done:
1094 return err;
1095}
1096
1097static void pdc_ring_free(struct pdc_state *pdcs)
1098{
1099 if (pdcs->tx_ring_alloc.vbase) {
1100 dma_pool_free(pdcs->ring_pool, pdcs->tx_ring_alloc.vbase,
1101 pdcs->tx_ring_alloc.dmabase);
1102 pdcs->tx_ring_alloc.vbase = NULL;
1103 }
1104
1105 if (pdcs->rx_ring_alloc.vbase) {
1106 dma_pool_free(pdcs->ring_pool, pdcs->rx_ring_alloc.vbase,
1107 pdcs->rx_ring_alloc.dmabase);
1108 pdcs->rx_ring_alloc.vbase = NULL;
1109 }
1110}
1111
1112/**
1113 * pdc_desc_count() - Count the number of DMA descriptors that will be required
1114 * for a given scatterlist. Account for the max length of a DMA buffer.
1115 * @sg: Scatterlist to be DMA'd
1116 * Return: Number of descriptors required
1117 */
1118static u32 pdc_desc_count(struct scatterlist *sg)
1119{
1120 u32 cnt = 0;
1121
1122 while (sg) {
1123 cnt += ((sg->length / PDC_DMA_BUF_MAX) + 1);
1124 sg = sg_next(sg);
1125 }
1126 return cnt;
1127}
1128
1129/**
1130 * pdc_rings_full() - Check whether the tx ring has room for tx_cnt descriptors
1131 * and the rx ring has room for rx_cnt descriptors.
1132 * @pdcs: PDC state
1133 * @tx_cnt: The number of descriptors required in the tx ring
1134 * @rx_cnt: The number of descriptors required i the rx ring
1135 *
1136 * Return: true if one of the rings does not have enough space
1137 * false if sufficient space is available in both rings
1138 */
1139static bool pdc_rings_full(struct pdc_state *pdcs, int tx_cnt, int rx_cnt)
1140{
1141 u32 rx_avail;
1142 u32 tx_avail;
1143 bool full = false;
1144
1145 /* Check if the tx and rx rings are likely to have enough space */
1146 rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
1147 pdcs->nrxpost);
1148 if (unlikely(rx_cnt > rx_avail)) {
1149 pdcs->rx_ring_full++;
1150 full = true;
1151 }
1152
1153 if (likely(!full)) {
1154 tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
1155 pdcs->ntxpost);
1156 if (unlikely(tx_cnt > tx_avail)) {
1157 pdcs->tx_ring_full++;
1158 full = true;
1159 }
1160 }
1161 return full;
1162}
1163
1164/**
1165 * pdc_last_tx_done() - If both the tx and rx rings have at least
1166 * PDC_RING_SPACE_MIN descriptors available, then indicate that the mailbox
1167 * framework can submit another message.
1168 * @chan: mailbox channel to check
1169 * Return: true if PDC can accept another message on this channel
1170 */
1171static bool pdc_last_tx_done(struct mbox_chan *chan)
1172{
1173 struct pdc_state *pdcs = chan->con_priv;
1174 bool ret;
1175
1176 if (unlikely(pdc_rings_full(pdcs, PDC_RING_SPACE_MIN,
1177 PDC_RING_SPACE_MIN))) {
1178 pdcs->last_tx_not_done++;
1179 ret = false;
1180 } else {
1181 ret = true;
1182 }
1183 return ret;
1184}
1185
1186/**
1187 * pdc_send_data() - mailbox send_data function
1188 * @chan: The mailbox channel on which the data is sent. The channel
1189 * corresponds to a DMA ringset.
1190 * @data: The mailbox message to be sent. The message must be a
1191 * brcm_message structure.
1192 *
1193 * This function is registered as the send_data function for the mailbox
1194 * controller. From the destination scatterlist in the mailbox message, it
1195 * creates a sequence of receive descriptors in the rx ring. From the source
1196 * scatterlist, it creates a sequence of transmit descriptors in the tx ring.
1197 * After creating the descriptors, it writes the rx ptr and tx ptr registers to
1198 * initiate the DMA transfer.
1199 *
1200 * This function does the DMA map and unmap of the src and dst scatterlists in
1201 * the mailbox message.
1202 *
1203 * Return: 0 if successful
1204 * -ENOTSUPP if the mailbox message is a type this driver does not
1205 * support
1206 * < 0 if an error
1207 */
1208static int pdc_send_data(struct mbox_chan *chan, void *data)
1209{
1210 struct pdc_state *pdcs = chan->con_priv;
1211 struct device *dev = &pdcs->pdev->dev;
1212 struct brcm_message *mssg = data;
1213 int err = PDC_SUCCESS;
1214 int src_nent;
1215 int dst_nent;
1216 int nent;
1217 u32 tx_desc_req;
1218 u32 rx_desc_req;
1219
1220 if (unlikely(mssg->type != BRCM_MESSAGE_SPU))
1221 return -ENOTSUPP;
1222
1223 src_nent = sg_nents(mssg->spu.src);
1224 if (likely(src_nent)) {
1225 nent = dma_map_sg(dev, mssg->spu.src, src_nent, DMA_TO_DEVICE);
1226 if (unlikely(nent == 0))
1227 return -EIO;
1228 }
1229
1230 dst_nent = sg_nents(mssg->spu.dst);
1231 if (likely(dst_nent)) {
1232 nent = dma_map_sg(dev, mssg->spu.dst, dst_nent,
1233 DMA_FROM_DEVICE);
1234 if (unlikely(nent == 0)) {
1235 dma_unmap_sg(dev, mssg->spu.src, src_nent,
1236 DMA_TO_DEVICE);
1237 return -EIO;
1238 }
1239 }
1240
1241 /*
1242 * Check if the tx and rx rings have enough space. Do this prior to
1243 * writing any tx or rx descriptors. Need to ensure that we do not write
1244 * a partial set of descriptors, or write just rx descriptors but
1245 * corresponding tx descriptors don't fit. Note that we want this check
1246 * and the entire sequence of descriptor to happen without another
1247 * thread getting in. The channel spin lock in the mailbox framework
1248 * ensures this.
1249 */
1250 tx_desc_req = pdc_desc_count(mssg->spu.src);
1251 rx_desc_req = pdc_desc_count(mssg->spu.dst);
1252 if (unlikely(pdc_rings_full(pdcs, tx_desc_req, rx_desc_req + 1)))
1253 return -ENOSPC;
1254
1255 /* Create rx descriptors to SPU catch response */
1256 err = pdc_rx_list_init(pdcs, mssg->spu.dst, mssg->ctx);
1257 err |= pdc_rx_list_sg_add(pdcs, mssg->spu.dst);
1258
1259 /* Create tx descriptors to submit SPU request */
1260 err |= pdc_tx_list_sg_add(pdcs, mssg->spu.src);
1261 err |= pdc_tx_list_final(pdcs); /* initiate transfer */
1262
1263 if (unlikely(err))
1264 dev_err(&pdcs->pdev->dev,
1265 "%s failed with error %d", __func__, err);
1266
1267 return err;
1268}
1269
1270static int pdc_startup(struct mbox_chan *chan)
1271{
1272 return pdc_ring_init(chan->con_priv, PDC_RINGSET);
1273}
1274
1275static void pdc_shutdown(struct mbox_chan *chan)
1276{
1277 struct pdc_state *pdcs = chan->con_priv;
1278
1279 if (!pdcs)
1280 return;
1281
1282 dev_dbg(&pdcs->pdev->dev,
1283 "Shutdown mailbox channel for PDC %u", pdcs->pdc_idx);
1284 pdc_ring_free(pdcs);
1285}
1286
1287/**
1288 * pdc_hw_init() - Use the given initialization parameters to initialize the
1289 * state for one of the PDCs.
1290 * @pdcs: state of the PDC
1291 */
1292static
1293void pdc_hw_init(struct pdc_state *pdcs)
1294{
1295 struct platform_device *pdev;
1296 struct device *dev;
1297 struct dma64 *dma_reg;
1298 int ringset = PDC_RINGSET;
1299
1300 pdev = pdcs->pdev;
1301 dev = &pdev->dev;
1302
1303 dev_dbg(dev, "PDC %u initial values:", pdcs->pdc_idx);
1304 dev_dbg(dev, "state structure: %p",
1305 pdcs);
1306 dev_dbg(dev, " - base virtual addr of hw regs %p",
1307 pdcs->pdc_reg_vbase);
1308
1309 /* initialize data structures */
1310 pdcs->regs = (struct pdc_regs *)pdcs->pdc_reg_vbase;
1311 pdcs->txregs_64 = (struct dma64_regs *)
1312 (((u8 *)pdcs->pdc_reg_vbase) +
1313 PDC_TXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1314 pdcs->rxregs_64 = (struct dma64_regs *)
1315 (((u8 *)pdcs->pdc_reg_vbase) +
1316 PDC_RXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1317
1318 pdcs->ntxd = PDC_RING_ENTRIES;
1319 pdcs->nrxd = PDC_RING_ENTRIES;
1320 pdcs->ntxpost = PDC_RING_ENTRIES - 1;
1321 pdcs->nrxpost = PDC_RING_ENTRIES - 1;
1322 iowrite32(0, &pdcs->regs->intmask);
1323
1324 dma_reg = &pdcs->regs->dmaregs[ringset];
1325
1326 /* Configure DMA but will enable later in pdc_ring_init() */
1327 iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1328
1329 iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1330 &dma_reg->dmarcv.control);
1331
1332 /* Reset current index pointers after making sure DMA is disabled */
1333 iowrite32(0, &dma_reg->dmaxmt.ptr);
1334 iowrite32(0, &dma_reg->dmarcv.ptr);
1335
1336 if (pdcs->pdc_resp_hdr_len == PDC_SPU2_RESP_HDR_LEN)
1337 iowrite32(PDC_CKSUM_CTRL,
1338 pdcs->pdc_reg_vbase + PDC_CKSUM_CTRL_OFFSET);
1339}
1340
1341/**
1342 * pdc_hw_disable() - Disable the tx and rx control in the hw.
1343 * @pdcs: PDC state structure
1344 *
1345 */
1346static void pdc_hw_disable(struct pdc_state *pdcs)
1347{
1348 struct dma64 *dma_reg;
1349
1350 dma_reg = &pdcs->regs->dmaregs[PDC_RINGSET];
1351 iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1352 iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1353 &dma_reg->dmarcv.control);
1354}
1355
1356/**
1357 * pdc_rx_buf_pool_create() - Pool of receive buffers used to catch the metadata
1358 * header returned with each response message.
1359 * @pdcs: PDC state structure
1360 *
1361 * The metadata is not returned to the mailbox client. So the PDC driver
1362 * manages these buffers.
1363 *
1364 * Return: PDC_SUCCESS
1365 * -ENOMEM if pool creation fails
1366 */
1367static int pdc_rx_buf_pool_create(struct pdc_state *pdcs)
1368{
1369 struct platform_device *pdev;
1370 struct device *dev;
1371
1372 pdev = pdcs->pdev;
1373 dev = &pdev->dev;
1374
1375 pdcs->pdc_resp_hdr_len = pdcs->rx_status_len;
1376 if (pdcs->use_bcm_hdr)
1377 pdcs->pdc_resp_hdr_len += BCM_HDR_LEN;
1378
1379 pdcs->rx_buf_pool = dma_pool_create("pdc rx bufs", dev,
1380 pdcs->pdc_resp_hdr_len,
1381 RX_BUF_ALIGN, 0);
1382 if (!pdcs->rx_buf_pool)
1383 return -ENOMEM;
1384
1385 return PDC_SUCCESS;
1386}
1387
1388/**
1389 * pdc_interrupts_init() - Initialize the interrupt configuration for a PDC and
1390 * specify a threaded IRQ handler for deferred handling of interrupts outside of
1391 * interrupt context.
1392 * @pdcs: PDC state
1393 *
1394 * Set the interrupt mask for transmit and receive done.
1395 * Set the lazy interrupt frame count to generate an interrupt for just one pkt.
1396 *
1397 * Return: PDC_SUCCESS
1398 * <0 if threaded irq request fails
1399 */
1400static int pdc_interrupts_init(struct pdc_state *pdcs)
1401{
1402 struct platform_device *pdev = pdcs->pdev;
1403 struct device *dev = &pdev->dev;
1404 struct device_node *dn = pdev->dev.of_node;
1405 int err;
1406
1407 /* interrupt configuration */
1408 iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
1409
1410 if (pdcs->hw_type == FA_HW)
1411 iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1412 FA_RCVLAZY0_OFFSET);
1413 else
1414 iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1415 PDC_RCVLAZY0_OFFSET);
1416
1417 /* read irq from device tree */
1418 pdcs->pdc_irq = irq_of_parse_and_map(dn, 0);
1419 dev_dbg(dev, "pdc device %s irq %u for pdcs %p",
1420 dev_name(dev), pdcs->pdc_irq, pdcs);
1421
1422 err = devm_request_irq(dev, pdcs->pdc_irq, pdc_irq_handler, 0,
1423 dev_name(dev), dev);
1424 if (err) {
1425 dev_err(dev, "IRQ %u request failed with err %d\n",
1426 pdcs->pdc_irq, err);
1427 return err;
1428 }
1429 return PDC_SUCCESS;
1430}
1431
1432static const struct mbox_chan_ops pdc_mbox_chan_ops = {
1433 .send_data = pdc_send_data,
1434 .last_tx_done = pdc_last_tx_done,
1435 .startup = pdc_startup,
1436 .shutdown = pdc_shutdown
1437};
1438
1439/**
1440 * pdc_mb_init() - Initialize the mailbox controller.
1441 * @pdcs: PDC state
1442 *
1443 * Each PDC is a mailbox controller. Each ringset is a mailbox channel. Kernel
1444 * driver only uses one ringset and thus one mb channel. PDC uses the transmit
1445 * complete interrupt to determine when a mailbox message has successfully been
1446 * transmitted.
1447 *
1448 * Return: 0 on success
1449 * < 0 if there is an allocation or registration failure
1450 */
1451static int pdc_mb_init(struct pdc_state *pdcs)
1452{
1453 struct device *dev = &pdcs->pdev->dev;
1454 struct mbox_controller *mbc;
1455 int chan_index;
1456 int err;
1457
1458 mbc = &pdcs->mbc;
1459 mbc->dev = dev;
1460 mbc->ops = &pdc_mbox_chan_ops;
1461 mbc->num_chans = 1;
1462 mbc->chans = devm_kcalloc(dev, mbc->num_chans, sizeof(*mbc->chans),
1463 GFP_KERNEL);
1464 if (!mbc->chans)
1465 return -ENOMEM;
1466
1467 mbc->txdone_irq = false;
1468 mbc->txdone_poll = true;
1469 mbc->txpoll_period = 1;
1470 for (chan_index = 0; chan_index < mbc->num_chans; chan_index++)
1471 mbc->chans[chan_index].con_priv = pdcs;
1472
1473 /* Register mailbox controller */
1474 err = mbox_controller_register(mbc);
1475 if (err) {
1476 dev_crit(dev,
1477 "Failed to register PDC mailbox controller. Error %d.",
1478 err);
1479 return err;
1480 }
1481 return 0;
1482}
1483
1484/* Device tree API */
1485static const int pdc_hw = PDC_HW;
1486static const int fa_hw = FA_HW;
1487
1488static const struct of_device_id pdc_mbox_of_match[] = {
1489 {.compatible = "brcm,iproc-pdc-mbox", .data = &pdc_hw},
1490 {.compatible = "brcm,iproc-fa2-mbox", .data = &fa_hw},
1491 { /* sentinel */ }
1492};
1493MODULE_DEVICE_TABLE(of, pdc_mbox_of_match);
1494
1495/**
1496 * pdc_dt_read() - Read application-specific data from device tree.
1497 * @pdev: Platform device
1498 * @pdcs: PDC state
1499 *
1500 * Reads the number of bytes of receive status that precede each received frame.
1501 * Reads whether transmit and received frames should be preceded by an 8-byte
1502 * BCM header.
1503 *
1504 * Return: 0 if successful
1505 * -ENODEV if device not available
1506 */
1507static int pdc_dt_read(struct platform_device *pdev, struct pdc_state *pdcs)
1508{
1509 struct device *dev = &pdev->dev;
1510 struct device_node *dn = pdev->dev.of_node;
1511 const struct of_device_id *match;
1512 const int *hw_type;
1513 int err;
1514
1515 err = of_property_read_u32(dn, "brcm,rx-status-len",
1516 &pdcs->rx_status_len);
1517 if (err < 0)
1518 dev_err(dev,
1519 "%s failed to get DMA receive status length from device tree",
1520 __func__);
1521
1522 pdcs->use_bcm_hdr = of_property_read_bool(dn, "brcm,use-bcm-hdr");
1523
1524 pdcs->hw_type = PDC_HW;
1525
1526 match = of_match_device(of_match_ptr(pdc_mbox_of_match), dev);
1527 if (match != NULL) {
1528 hw_type = match->data;
1529 pdcs->hw_type = *hw_type;
1530 }
1531
1532 return 0;
1533}
1534
1535/**
1536 * pdc_probe() - Probe function for PDC driver.
1537 * @pdev: PDC platform device
1538 *
1539 * Reserve and map register regions defined in device tree.
1540 * Allocate and initialize tx and rx DMA rings.
1541 * Initialize a mailbox controller for each PDC.
1542 *
1543 * Return: 0 if successful
1544 * < 0 if an error
1545 */
1546static int pdc_probe(struct platform_device *pdev)
1547{
1548 int err = 0;
1549 struct device *dev = &pdev->dev;
1550 struct resource *pdc_regs;
1551 struct pdc_state *pdcs;
1552
1553 /* PDC state for one SPU */
1554 pdcs = devm_kzalloc(dev, sizeof(*pdcs), GFP_KERNEL);
1555 if (!pdcs) {
1556 err = -ENOMEM;
1557 goto cleanup;
1558 }
1559
1560 pdcs->pdev = pdev;
1561 platform_set_drvdata(pdev, pdcs);
1562 pdcs->pdc_idx = pdcg.num_spu;
1563 pdcg.num_spu++;
1564
1565 err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(39));
1566 if (err) {
1567 dev_warn(dev, "PDC device cannot perform DMA. Error %d.", err);
1568 goto cleanup;
1569 }
1570
1571 /* Create DMA pool for tx ring */
1572 pdcs->ring_pool = dma_pool_create("pdc rings", dev, PDC_RING_SIZE,
1573 RING_ALIGN, 0);
1574 if (!pdcs->ring_pool) {
1575 err = -ENOMEM;
1576 goto cleanup;
1577 }
1578
1579 err = pdc_dt_read(pdev, pdcs);
1580 if (err)
1581 goto cleanup_ring_pool;
1582
1583 pdc_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1584 if (!pdc_regs) {
1585 err = -ENODEV;
1586 goto cleanup_ring_pool;
1587 }
1588 dev_dbg(dev, "PDC register region res.start = %pa, res.end = %pa",
1589 &pdc_regs->start, &pdc_regs->end);
1590
1591 pdcs->pdc_reg_vbase = devm_ioremap_resource(&pdev->dev, pdc_regs);
1592 if (IS_ERR(pdcs->pdc_reg_vbase)) {
1593 err = PTR_ERR(pdcs->pdc_reg_vbase);
1594 dev_err(&pdev->dev, "Failed to map registers: %d\n", err);
1595 goto cleanup_ring_pool;
1596 }
1597
1598 /* create rx buffer pool after dt read to know how big buffers are */
1599 err = pdc_rx_buf_pool_create(pdcs);
1600 if (err)
1601 goto cleanup_ring_pool;
1602
1603 pdc_hw_init(pdcs);
1604
1605 /* Init tasklet for deferred DMA rx processing */
1606 tasklet_init(&pdcs->rx_tasklet, pdc_tasklet_cb, (unsigned long)pdcs);
1607
1608 err = pdc_interrupts_init(pdcs);
1609 if (err)
1610 goto cleanup_buf_pool;
1611
1612 /* Initialize mailbox controller */
1613 err = pdc_mb_init(pdcs);
1614 if (err)
1615 goto cleanup_buf_pool;
1616
1617 pdcs->debugfs_stats = NULL;
1618 pdc_setup_debugfs(pdcs);
1619
1620 dev_dbg(dev, "pdc_probe() successful");
1621 return PDC_SUCCESS;
1622
1623cleanup_buf_pool:
1624 tasklet_kill(&pdcs->rx_tasklet);
1625 dma_pool_destroy(pdcs->rx_buf_pool);
1626
1627cleanup_ring_pool:
1628 dma_pool_destroy(pdcs->ring_pool);
1629
1630cleanup:
1631 return err;
1632}
1633
1634static int pdc_remove(struct platform_device *pdev)
1635{
1636 struct pdc_state *pdcs = platform_get_drvdata(pdev);
1637
1638 pdc_free_debugfs();
1639
1640 tasklet_kill(&pdcs->rx_tasklet);
1641
1642 pdc_hw_disable(pdcs);
1643
1644 mbox_controller_unregister(&pdcs->mbc);
1645
1646 dma_pool_destroy(pdcs->rx_buf_pool);
1647 dma_pool_destroy(pdcs->ring_pool);
1648 return 0;
1649}
1650
1651static struct platform_driver pdc_mbox_driver = {
1652 .probe = pdc_probe,
1653 .remove = pdc_remove,
1654 .driver = {
1655 .name = "brcm-iproc-pdc-mbox",
1656 .of_match_table = of_match_ptr(pdc_mbox_of_match),
1657 },
1658};
1659module_platform_driver(pdc_mbox_driver);
1660
1661MODULE_AUTHOR("Rob Rice <rob.rice@broadcom.com>");
1662MODULE_DESCRIPTION("Broadcom PDC mailbox driver");
1663MODULE_LICENSE("GPL v2");