Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Tegra host1x Command DMA
4 *
5 * Copyright (c) 2010-2013, NVIDIA Corporation.
6 */
7
8
9#include <asm/cacheflush.h>
10#include <linux/device.h>
11#include <linux/dma-mapping.h>
12#include <linux/host1x.h>
13#include <linux/interrupt.h>
14#include <linux/kernel.h>
15#include <linux/kfifo.h>
16#include <linux/slab.h>
17#include <trace/events/host1x.h>
18
19#include "cdma.h"
20#include "channel.h"
21#include "dev.h"
22#include "debug.h"
23#include "job.h"
24
25/*
26 * push_buffer
27 *
28 * The push buffer is a circular array of words to be fetched by command DMA.
29 * Note that it works slightly differently to the sync queue; fence == pos
30 * means that the push buffer is full, not empty.
31 */
32
33/*
34 * Typically the commands written into the push buffer are a pair of words. We
35 * use slots to represent each of these pairs and to simplify things. Note the
36 * strange number of slots allocated here. 512 slots will fit exactly within a
37 * single memory page. We also need one additional word at the end of the push
38 * buffer for the RESTART opcode that will instruct the CDMA to jump back to
39 * the beginning of the push buffer. With 512 slots, this means that we'll use
40 * 2 memory pages and waste 4092 bytes of the second page that will never be
41 * used.
42 */
43#define HOST1X_PUSHBUFFER_SLOTS 511
44
45/*
46 * Clean up push buffer resources
47 */
48static void host1x_pushbuffer_destroy(struct push_buffer *pb)
49{
50 struct host1x_cdma *cdma = pb_to_cdma(pb);
51 struct host1x *host1x = cdma_to_host1x(cdma);
52
53 if (!pb->mapped)
54 return;
55
56 if (host1x->domain) {
57 iommu_unmap(host1x->domain, pb->dma, pb->alloc_size);
58 free_iova(&host1x->iova, iova_pfn(&host1x->iova, pb->dma));
59 }
60
61 dma_free_wc(host1x->dev, pb->alloc_size, pb->mapped, pb->phys);
62
63 pb->mapped = NULL;
64 pb->phys = 0;
65}
66
67/*
68 * Init push buffer resources
69 */
70static int host1x_pushbuffer_init(struct push_buffer *pb)
71{
72 struct host1x_cdma *cdma = pb_to_cdma(pb);
73 struct host1x *host1x = cdma_to_host1x(cdma);
74 struct iova *alloc;
75 u32 size;
76 int err;
77
78 pb->mapped = NULL;
79 pb->phys = 0;
80 pb->size = HOST1X_PUSHBUFFER_SLOTS * 8;
81
82 size = pb->size + 4;
83
84 /* initialize buffer pointers */
85 pb->fence = pb->size - 8;
86 pb->pos = 0;
87
88 if (host1x->domain) {
89 unsigned long shift;
90
91 size = iova_align(&host1x->iova, size);
92
93 pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
94 GFP_KERNEL);
95 if (!pb->mapped)
96 return -ENOMEM;
97
98 shift = iova_shift(&host1x->iova);
99 alloc = alloc_iova(&host1x->iova, size >> shift,
100 host1x->iova_end >> shift, true);
101 if (!alloc) {
102 err = -ENOMEM;
103 goto iommu_free_mem;
104 }
105
106 pb->dma = iova_dma_addr(&host1x->iova, alloc);
107 err = iommu_map(host1x->domain, pb->dma, pb->phys, size,
108 IOMMU_READ);
109 if (err)
110 goto iommu_free_iova;
111 } else {
112 pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
113 GFP_KERNEL);
114 if (!pb->mapped)
115 return -ENOMEM;
116
117 pb->dma = pb->phys;
118 }
119
120 pb->alloc_size = size;
121
122 host1x_hw_pushbuffer_init(host1x, pb);
123
124 return 0;
125
126iommu_free_iova:
127 __free_iova(&host1x->iova, alloc);
128iommu_free_mem:
129 dma_free_wc(host1x->dev, size, pb->mapped, pb->phys);
130
131 return err;
132}
133
134/*
135 * Push two words to the push buffer
136 * Caller must ensure push buffer is not full
137 */
138static void host1x_pushbuffer_push(struct push_buffer *pb, u32 op1, u32 op2)
139{
140 u32 *p = (u32 *)((void *)pb->mapped + pb->pos);
141
142 WARN_ON(pb->pos == pb->fence);
143 *(p++) = op1;
144 *(p++) = op2;
145 pb->pos += 8;
146
147 if (pb->pos >= pb->size)
148 pb->pos -= pb->size;
149}
150
151/*
152 * Pop a number of two word slots from the push buffer
153 * Caller must ensure push buffer is not empty
154 */
155static void host1x_pushbuffer_pop(struct push_buffer *pb, unsigned int slots)
156{
157 /* Advance the next write position */
158 pb->fence += slots * 8;
159
160 if (pb->fence >= pb->size)
161 pb->fence -= pb->size;
162}
163
164/*
165 * Return the number of two word slots free in the push buffer
166 */
167static u32 host1x_pushbuffer_space(struct push_buffer *pb)
168{
169 unsigned int fence = pb->fence;
170
171 if (pb->fence < pb->pos)
172 fence += pb->size;
173
174 return (fence - pb->pos) / 8;
175}
176
177/*
178 * Sleep (if necessary) until the requested event happens
179 * - CDMA_EVENT_SYNC_QUEUE_EMPTY : sync queue is completely empty.
180 * - Returns 1
181 * - CDMA_EVENT_PUSH_BUFFER_SPACE : there is space in the push buffer
182 * - Return the amount of space (> 0)
183 * Must be called with the cdma lock held.
184 */
185unsigned int host1x_cdma_wait_locked(struct host1x_cdma *cdma,
186 enum cdma_event event)
187{
188 for (;;) {
189 struct push_buffer *pb = &cdma->push_buffer;
190 unsigned int space;
191
192 switch (event) {
193 case CDMA_EVENT_SYNC_QUEUE_EMPTY:
194 space = list_empty(&cdma->sync_queue) ? 1 : 0;
195 break;
196
197 case CDMA_EVENT_PUSH_BUFFER_SPACE:
198 space = host1x_pushbuffer_space(pb);
199 break;
200
201 default:
202 WARN_ON(1);
203 return -EINVAL;
204 }
205
206 if (space)
207 return space;
208
209 trace_host1x_wait_cdma(dev_name(cdma_to_channel(cdma)->dev),
210 event);
211
212 /* If somebody has managed to already start waiting, yield */
213 if (cdma->event != CDMA_EVENT_NONE) {
214 mutex_unlock(&cdma->lock);
215 schedule();
216 mutex_lock(&cdma->lock);
217 continue;
218 }
219
220 cdma->event = event;
221
222 mutex_unlock(&cdma->lock);
223 wait_for_completion(&cdma->complete);
224 mutex_lock(&cdma->lock);
225 }
226
227 return 0;
228}
229
230/*
231 * Sleep (if necessary) until the push buffer has enough free space.
232 *
233 * Must be called with the cdma lock held.
234 */
235int host1x_cdma_wait_pushbuffer_space(struct host1x *host1x,
236 struct host1x_cdma *cdma,
237 unsigned int needed)
238{
239 while (true) {
240 struct push_buffer *pb = &cdma->push_buffer;
241 unsigned int space;
242
243 space = host1x_pushbuffer_space(pb);
244 if (space >= needed)
245 break;
246
247 trace_host1x_wait_cdma(dev_name(cdma_to_channel(cdma)->dev),
248 CDMA_EVENT_PUSH_BUFFER_SPACE);
249
250 host1x_hw_cdma_flush(host1x, cdma);
251
252 /* If somebody has managed to already start waiting, yield */
253 if (cdma->event != CDMA_EVENT_NONE) {
254 mutex_unlock(&cdma->lock);
255 schedule();
256 mutex_lock(&cdma->lock);
257 continue;
258 }
259
260 cdma->event = CDMA_EVENT_PUSH_BUFFER_SPACE;
261
262 mutex_unlock(&cdma->lock);
263 wait_for_completion(&cdma->complete);
264 mutex_lock(&cdma->lock);
265 }
266
267 return 0;
268}
269/*
270 * Start timer that tracks the time spent by the job.
271 * Must be called with the cdma lock held.
272 */
273static void cdma_start_timer_locked(struct host1x_cdma *cdma,
274 struct host1x_job *job)
275{
276 struct host1x *host = cdma_to_host1x(cdma);
277
278 if (cdma->timeout.client) {
279 /* timer already started */
280 return;
281 }
282
283 cdma->timeout.client = job->client;
284 cdma->timeout.syncpt = host1x_syncpt_get(host, job->syncpt_id);
285 cdma->timeout.syncpt_val = job->syncpt_end;
286 cdma->timeout.start_ktime = ktime_get();
287
288 schedule_delayed_work(&cdma->timeout.wq,
289 msecs_to_jiffies(job->timeout));
290}
291
292/*
293 * Stop timer when a buffer submission completes.
294 * Must be called with the cdma lock held.
295 */
296static void stop_cdma_timer_locked(struct host1x_cdma *cdma)
297{
298 cancel_delayed_work(&cdma->timeout.wq);
299 cdma->timeout.client = NULL;
300}
301
302/*
303 * For all sync queue entries that have already finished according to the
304 * current sync point registers:
305 * - unpin & unref their mems
306 * - pop their push buffer slots
307 * - remove them from the sync queue
308 * This is normally called from the host code's worker thread, but can be
309 * called manually if necessary.
310 * Must be called with the cdma lock held.
311 */
312static void update_cdma_locked(struct host1x_cdma *cdma)
313{
314 bool signal = false;
315 struct host1x *host1x = cdma_to_host1x(cdma);
316 struct host1x_job *job, *n;
317
318 /* If CDMA is stopped, queue is cleared and we can return */
319 if (!cdma->running)
320 return;
321
322 /*
323 * Walk the sync queue, reading the sync point registers as necessary,
324 * to consume as many sync queue entries as possible without blocking
325 */
326 list_for_each_entry_safe(job, n, &cdma->sync_queue, list) {
327 struct host1x_syncpt *sp =
328 host1x_syncpt_get(host1x, job->syncpt_id);
329
330 /* Check whether this syncpt has completed, and bail if not */
331 if (!host1x_syncpt_is_expired(sp, job->syncpt_end)) {
332 /* Start timer on next pending syncpt */
333 if (job->timeout)
334 cdma_start_timer_locked(cdma, job);
335
336 break;
337 }
338
339 /* Cancel timeout, when a buffer completes */
340 if (cdma->timeout.client)
341 stop_cdma_timer_locked(cdma);
342
343 /* Unpin the memory */
344 host1x_job_unpin(job);
345
346 /* Pop push buffer slots */
347 if (job->num_slots) {
348 struct push_buffer *pb = &cdma->push_buffer;
349
350 host1x_pushbuffer_pop(pb, job->num_slots);
351
352 if (cdma->event == CDMA_EVENT_PUSH_BUFFER_SPACE)
353 signal = true;
354 }
355
356 list_del(&job->list);
357 host1x_job_put(job);
358 }
359
360 if (cdma->event == CDMA_EVENT_SYNC_QUEUE_EMPTY &&
361 list_empty(&cdma->sync_queue))
362 signal = true;
363
364 if (signal) {
365 cdma->event = CDMA_EVENT_NONE;
366 complete(&cdma->complete);
367 }
368}
369
370void host1x_cdma_update_sync_queue(struct host1x_cdma *cdma,
371 struct device *dev)
372{
373 struct host1x *host1x = cdma_to_host1x(cdma);
374 u32 restart_addr, syncpt_incrs, syncpt_val;
375 struct host1x_job *job, *next_job = NULL;
376
377 syncpt_val = host1x_syncpt_load(cdma->timeout.syncpt);
378
379 dev_dbg(dev, "%s: starting cleanup (thresh %d)\n",
380 __func__, syncpt_val);
381
382 /*
383 * Move the sync_queue read pointer to the first entry that hasn't
384 * completed based on the current HW syncpt value. It's likely there
385 * won't be any (i.e. we're still at the head), but covers the case
386 * where a syncpt incr happens just prior/during the teardown.
387 */
388
389 dev_dbg(dev, "%s: skip completed buffers still in sync_queue\n",
390 __func__);
391
392 list_for_each_entry(job, &cdma->sync_queue, list) {
393 if (syncpt_val < job->syncpt_end) {
394
395 if (!list_is_last(&job->list, &cdma->sync_queue))
396 next_job = list_next_entry(job, list);
397
398 goto syncpt_incr;
399 }
400
401 host1x_job_dump(dev, job);
402 }
403
404 /* all jobs have been completed */
405 job = NULL;
406
407syncpt_incr:
408
409 /*
410 * Increment with CPU the remaining syncpts of a partially executed job.
411 *
412 * CDMA will continue execution starting with the next job or will get
413 * into idle state.
414 */
415 if (next_job)
416 restart_addr = next_job->first_get;
417 else
418 restart_addr = cdma->last_pos;
419
420 /* do CPU increments for the remaining syncpts */
421 if (job) {
422 dev_dbg(dev, "%s: perform CPU incr on pending buffers\n",
423 __func__);
424
425 /* won't need a timeout when replayed */
426 job->timeout = 0;
427
428 syncpt_incrs = job->syncpt_end - syncpt_val;
429 dev_dbg(dev, "%s: CPU incr (%d)\n", __func__, syncpt_incrs);
430
431 host1x_job_dump(dev, job);
432
433 /* safe to use CPU to incr syncpts */
434 host1x_hw_cdma_timeout_cpu_incr(host1x, cdma, job->first_get,
435 syncpt_incrs, job->syncpt_end,
436 job->num_slots);
437
438 dev_dbg(dev, "%s: finished sync_queue modification\n",
439 __func__);
440 }
441
442 /* roll back DMAGET and start up channel again */
443 host1x_hw_cdma_resume(host1x, cdma, restart_addr);
444}
445
446/*
447 * Create a cdma
448 */
449int host1x_cdma_init(struct host1x_cdma *cdma)
450{
451 int err;
452
453 mutex_init(&cdma->lock);
454 init_completion(&cdma->complete);
455
456 INIT_LIST_HEAD(&cdma->sync_queue);
457
458 cdma->event = CDMA_EVENT_NONE;
459 cdma->running = false;
460 cdma->torndown = false;
461
462 err = host1x_pushbuffer_init(&cdma->push_buffer);
463 if (err)
464 return err;
465
466 return 0;
467}
468
469/*
470 * Destroy a cdma
471 */
472int host1x_cdma_deinit(struct host1x_cdma *cdma)
473{
474 struct push_buffer *pb = &cdma->push_buffer;
475 struct host1x *host1x = cdma_to_host1x(cdma);
476
477 if (cdma->running) {
478 pr_warn("%s: CDMA still running\n", __func__);
479 return -EBUSY;
480 }
481
482 host1x_pushbuffer_destroy(pb);
483 host1x_hw_cdma_timeout_destroy(host1x, cdma);
484
485 return 0;
486}
487
488/*
489 * Begin a cdma submit
490 */
491int host1x_cdma_begin(struct host1x_cdma *cdma, struct host1x_job *job)
492{
493 struct host1x *host1x = cdma_to_host1x(cdma);
494
495 mutex_lock(&cdma->lock);
496
497 if (job->timeout) {
498 /* init state on first submit with timeout value */
499 if (!cdma->timeout.initialized) {
500 int err;
501
502 err = host1x_hw_cdma_timeout_init(host1x, cdma,
503 job->syncpt_id);
504 if (err) {
505 mutex_unlock(&cdma->lock);
506 return err;
507 }
508 }
509 }
510
511 if (!cdma->running)
512 host1x_hw_cdma_start(host1x, cdma);
513
514 cdma->slots_free = 0;
515 cdma->slots_used = 0;
516 cdma->first_get = cdma->push_buffer.pos;
517
518 trace_host1x_cdma_begin(dev_name(job->channel->dev));
519 return 0;
520}
521
522/*
523 * Push two words into a push buffer slot
524 * Blocks as necessary if the push buffer is full.
525 */
526void host1x_cdma_push(struct host1x_cdma *cdma, u32 op1, u32 op2)
527{
528 struct host1x *host1x = cdma_to_host1x(cdma);
529 struct push_buffer *pb = &cdma->push_buffer;
530 u32 slots_free = cdma->slots_free;
531
532 if (host1x_debug_trace_cmdbuf)
533 trace_host1x_cdma_push(dev_name(cdma_to_channel(cdma)->dev),
534 op1, op2);
535
536 if (slots_free == 0) {
537 host1x_hw_cdma_flush(host1x, cdma);
538 slots_free = host1x_cdma_wait_locked(cdma,
539 CDMA_EVENT_PUSH_BUFFER_SPACE);
540 }
541
542 cdma->slots_free = slots_free - 1;
543 cdma->slots_used++;
544 host1x_pushbuffer_push(pb, op1, op2);
545}
546
547/*
548 * Push four words into two consecutive push buffer slots. Note that extra
549 * care needs to be taken not to split the two slots across the end of the
550 * push buffer. Otherwise the RESTART opcode at the end of the push buffer
551 * that ensures processing will restart at the beginning will break up the
552 * four words.
553 *
554 * Blocks as necessary if the push buffer is full.
555 */
556void host1x_cdma_push_wide(struct host1x_cdma *cdma, u32 op1, u32 op2,
557 u32 op3, u32 op4)
558{
559 struct host1x_channel *channel = cdma_to_channel(cdma);
560 struct host1x *host1x = cdma_to_host1x(cdma);
561 struct push_buffer *pb = &cdma->push_buffer;
562 unsigned int needed = 2, extra = 0, i;
563 unsigned int space = cdma->slots_free;
564
565 if (host1x_debug_trace_cmdbuf)
566 trace_host1x_cdma_push_wide(dev_name(channel->dev), op1, op2,
567 op3, op4);
568
569 /* compute number of extra slots needed for padding */
570 if (pb->pos + 16 > pb->size) {
571 extra = (pb->size - pb->pos) / 8;
572 needed += extra;
573 }
574
575 host1x_cdma_wait_pushbuffer_space(host1x, cdma, needed);
576 space = host1x_pushbuffer_space(pb);
577
578 cdma->slots_free = space - needed;
579 cdma->slots_used += needed;
580
581 /*
582 * Note that we rely on the fact that this is only used to submit wide
583 * gather opcodes, which consist of 3 words, and they are padded with
584 * a NOP to avoid having to deal with fractional slots (a slot always
585 * represents 2 words). The fourth opcode passed to this function will
586 * therefore always be a NOP.
587 *
588 * This works around a slight ambiguity when it comes to opcodes. For
589 * all current host1x incarnations the NOP opcode uses the exact same
590 * encoding (0x20000000), so we could hard-code the value here, but a
591 * new incarnation may change it and break that assumption.
592 */
593 for (i = 0; i < extra; i++)
594 host1x_pushbuffer_push(pb, op4, op4);
595
596 host1x_pushbuffer_push(pb, op1, op2);
597 host1x_pushbuffer_push(pb, op3, op4);
598}
599
600/*
601 * End a cdma submit
602 * Kick off DMA, add job to the sync queue, and a number of slots to be freed
603 * from the pushbuffer. The handles for a submit must all be pinned at the same
604 * time, but they can be unpinned in smaller chunks.
605 */
606void host1x_cdma_end(struct host1x_cdma *cdma,
607 struct host1x_job *job)
608{
609 struct host1x *host1x = cdma_to_host1x(cdma);
610 bool idle = list_empty(&cdma->sync_queue);
611
612 host1x_hw_cdma_flush(host1x, cdma);
613
614 job->first_get = cdma->first_get;
615 job->num_slots = cdma->slots_used;
616 host1x_job_get(job);
617 list_add_tail(&job->list, &cdma->sync_queue);
618
619 /* start timer on idle -> active transitions */
620 if (job->timeout && idle)
621 cdma_start_timer_locked(cdma, job);
622
623 trace_host1x_cdma_end(dev_name(job->channel->dev));
624 mutex_unlock(&cdma->lock);
625}
626
627/*
628 * Update cdma state according to current sync point values
629 */
630void host1x_cdma_update(struct host1x_cdma *cdma)
631{
632 mutex_lock(&cdma->lock);
633 update_cdma_locked(cdma);
634 mutex_unlock(&cdma->lock);
635}
1/*
2 * Tegra host1x Command DMA
3 *
4 * Copyright (c) 2010-2013, NVIDIA Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program. If not, see <http://www.gnu.org/licenses/>.
17 */
18
19
20#include <asm/cacheflush.h>
21#include <linux/device.h>
22#include <linux/dma-mapping.h>
23#include <linux/host1x.h>
24#include <linux/interrupt.h>
25#include <linux/kernel.h>
26#include <linux/kfifo.h>
27#include <linux/slab.h>
28#include <trace/events/host1x.h>
29
30#include "cdma.h"
31#include "channel.h"
32#include "dev.h"
33#include "debug.h"
34#include "job.h"
35
36/*
37 * push_buffer
38 *
39 * The push buffer is a circular array of words to be fetched by command DMA.
40 * Note that it works slightly differently to the sync queue; fence == pos
41 * means that the push buffer is full, not empty.
42 */
43
44#define HOST1X_PUSHBUFFER_SLOTS 512
45
46/*
47 * Clean up push buffer resources
48 */
49static void host1x_pushbuffer_destroy(struct push_buffer *pb)
50{
51 struct host1x_cdma *cdma = pb_to_cdma(pb);
52 struct host1x *host1x = cdma_to_host1x(cdma);
53
54 if (!pb->phys)
55 return;
56
57 if (host1x->domain) {
58 iommu_unmap(host1x->domain, pb->dma, pb->alloc_size);
59 free_iova(&host1x->iova, iova_pfn(&host1x->iova, pb->dma));
60 }
61
62 dma_free_wc(host1x->dev, pb->alloc_size, pb->mapped, pb->phys);
63
64 pb->mapped = NULL;
65 pb->phys = 0;
66}
67
68/*
69 * Init push buffer resources
70 */
71static int host1x_pushbuffer_init(struct push_buffer *pb)
72{
73 struct host1x_cdma *cdma = pb_to_cdma(pb);
74 struct host1x *host1x = cdma_to_host1x(cdma);
75 struct iova *alloc;
76 u32 size;
77 int err;
78
79 pb->mapped = NULL;
80 pb->phys = 0;
81 pb->size = HOST1X_PUSHBUFFER_SLOTS * 8;
82
83 size = pb->size + 4;
84
85 /* initialize buffer pointers */
86 pb->fence = pb->size - 8;
87 pb->pos = 0;
88
89 if (host1x->domain) {
90 unsigned long shift;
91
92 size = iova_align(&host1x->iova, size);
93
94 pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
95 GFP_KERNEL);
96 if (!pb->mapped)
97 return -ENOMEM;
98
99 shift = iova_shift(&host1x->iova);
100 alloc = alloc_iova(&host1x->iova, size >> shift,
101 host1x->iova_end >> shift, true);
102 if (!alloc) {
103 err = -ENOMEM;
104 goto iommu_free_mem;
105 }
106
107 pb->dma = iova_dma_addr(&host1x->iova, alloc);
108 err = iommu_map(host1x->domain, pb->dma, pb->phys, size,
109 IOMMU_READ);
110 if (err)
111 goto iommu_free_iova;
112 } else {
113 pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
114 GFP_KERNEL);
115 if (!pb->mapped)
116 return -ENOMEM;
117
118 pb->dma = pb->phys;
119 }
120
121 pb->alloc_size = size;
122
123 host1x_hw_pushbuffer_init(host1x, pb);
124
125 return 0;
126
127iommu_free_iova:
128 __free_iova(&host1x->iova, alloc);
129iommu_free_mem:
130 dma_free_wc(host1x->dev, pb->alloc_size, pb->mapped, pb->phys);
131
132 return err;
133}
134
135/*
136 * Push two words to the push buffer
137 * Caller must ensure push buffer is not full
138 */
139static void host1x_pushbuffer_push(struct push_buffer *pb, u32 op1, u32 op2)
140{
141 u32 *p = (u32 *)((void *)pb->mapped + pb->pos);
142
143 WARN_ON(pb->pos == pb->fence);
144 *(p++) = op1;
145 *(p++) = op2;
146 pb->pos = (pb->pos + 8) & (pb->size - 1);
147}
148
149/*
150 * Pop a number of two word slots from the push buffer
151 * Caller must ensure push buffer is not empty
152 */
153static void host1x_pushbuffer_pop(struct push_buffer *pb, unsigned int slots)
154{
155 /* Advance the next write position */
156 pb->fence = (pb->fence + slots * 8) & (pb->size - 1);
157}
158
159/*
160 * Return the number of two word slots free in the push buffer
161 */
162static u32 host1x_pushbuffer_space(struct push_buffer *pb)
163{
164 return ((pb->fence - pb->pos) & (pb->size - 1)) / 8;
165}
166
167/*
168 * Sleep (if necessary) until the requested event happens
169 * - CDMA_EVENT_SYNC_QUEUE_EMPTY : sync queue is completely empty.
170 * - Returns 1
171 * - CDMA_EVENT_PUSH_BUFFER_SPACE : there is space in the push buffer
172 * - Return the amount of space (> 0)
173 * Must be called with the cdma lock held.
174 */
175unsigned int host1x_cdma_wait_locked(struct host1x_cdma *cdma,
176 enum cdma_event event)
177{
178 for (;;) {
179 struct push_buffer *pb = &cdma->push_buffer;
180 unsigned int space;
181
182 switch (event) {
183 case CDMA_EVENT_SYNC_QUEUE_EMPTY:
184 space = list_empty(&cdma->sync_queue) ? 1 : 0;
185 break;
186
187 case CDMA_EVENT_PUSH_BUFFER_SPACE:
188 space = host1x_pushbuffer_space(pb);
189 break;
190
191 default:
192 WARN_ON(1);
193 return -EINVAL;
194 }
195
196 if (space)
197 return space;
198
199 trace_host1x_wait_cdma(dev_name(cdma_to_channel(cdma)->dev),
200 event);
201
202 /* If somebody has managed to already start waiting, yield */
203 if (cdma->event != CDMA_EVENT_NONE) {
204 mutex_unlock(&cdma->lock);
205 schedule();
206 mutex_lock(&cdma->lock);
207 continue;
208 }
209
210 cdma->event = event;
211
212 mutex_unlock(&cdma->lock);
213 down(&cdma->sem);
214 mutex_lock(&cdma->lock);
215 }
216
217 return 0;
218}
219
220/*
221 * Start timer that tracks the time spent by the job.
222 * Must be called with the cdma lock held.
223 */
224static void cdma_start_timer_locked(struct host1x_cdma *cdma,
225 struct host1x_job *job)
226{
227 struct host1x *host = cdma_to_host1x(cdma);
228
229 if (cdma->timeout.client) {
230 /* timer already started */
231 return;
232 }
233
234 cdma->timeout.client = job->client;
235 cdma->timeout.syncpt = host1x_syncpt_get(host, job->syncpt_id);
236 cdma->timeout.syncpt_val = job->syncpt_end;
237 cdma->timeout.start_ktime = ktime_get();
238
239 schedule_delayed_work(&cdma->timeout.wq,
240 msecs_to_jiffies(job->timeout));
241}
242
243/*
244 * Stop timer when a buffer submission completes.
245 * Must be called with the cdma lock held.
246 */
247static void stop_cdma_timer_locked(struct host1x_cdma *cdma)
248{
249 cancel_delayed_work(&cdma->timeout.wq);
250 cdma->timeout.client = 0;
251}
252
253/*
254 * For all sync queue entries that have already finished according to the
255 * current sync point registers:
256 * - unpin & unref their mems
257 * - pop their push buffer slots
258 * - remove them from the sync queue
259 * This is normally called from the host code's worker thread, but can be
260 * called manually if necessary.
261 * Must be called with the cdma lock held.
262 */
263static void update_cdma_locked(struct host1x_cdma *cdma)
264{
265 bool signal = false;
266 struct host1x *host1x = cdma_to_host1x(cdma);
267 struct host1x_job *job, *n;
268
269 /* If CDMA is stopped, queue is cleared and we can return */
270 if (!cdma->running)
271 return;
272
273 /*
274 * Walk the sync queue, reading the sync point registers as necessary,
275 * to consume as many sync queue entries as possible without blocking
276 */
277 list_for_each_entry_safe(job, n, &cdma->sync_queue, list) {
278 struct host1x_syncpt *sp =
279 host1x_syncpt_get(host1x, job->syncpt_id);
280
281 /* Check whether this syncpt has completed, and bail if not */
282 if (!host1x_syncpt_is_expired(sp, job->syncpt_end)) {
283 /* Start timer on next pending syncpt */
284 if (job->timeout)
285 cdma_start_timer_locked(cdma, job);
286
287 break;
288 }
289
290 /* Cancel timeout, when a buffer completes */
291 if (cdma->timeout.client)
292 stop_cdma_timer_locked(cdma);
293
294 /* Unpin the memory */
295 host1x_job_unpin(job);
296
297 /* Pop push buffer slots */
298 if (job->num_slots) {
299 struct push_buffer *pb = &cdma->push_buffer;
300
301 host1x_pushbuffer_pop(pb, job->num_slots);
302
303 if (cdma->event == CDMA_EVENT_PUSH_BUFFER_SPACE)
304 signal = true;
305 }
306
307 list_del(&job->list);
308 host1x_job_put(job);
309 }
310
311 if (cdma->event == CDMA_EVENT_SYNC_QUEUE_EMPTY &&
312 list_empty(&cdma->sync_queue))
313 signal = true;
314
315 if (signal) {
316 cdma->event = CDMA_EVENT_NONE;
317 up(&cdma->sem);
318 }
319}
320
321void host1x_cdma_update_sync_queue(struct host1x_cdma *cdma,
322 struct device *dev)
323{
324 struct host1x *host1x = cdma_to_host1x(cdma);
325 u32 restart_addr, syncpt_incrs, syncpt_val;
326 struct host1x_job *job = NULL;
327
328 syncpt_val = host1x_syncpt_load(cdma->timeout.syncpt);
329
330 dev_dbg(dev, "%s: starting cleanup (thresh %d)\n",
331 __func__, syncpt_val);
332
333 /*
334 * Move the sync_queue read pointer to the first entry that hasn't
335 * completed based on the current HW syncpt value. It's likely there
336 * won't be any (i.e. we're still at the head), but covers the case
337 * where a syncpt incr happens just prior/during the teardown.
338 */
339
340 dev_dbg(dev, "%s: skip completed buffers still in sync_queue\n",
341 __func__);
342
343 list_for_each_entry(job, &cdma->sync_queue, list) {
344 if (syncpt_val < job->syncpt_end)
345 break;
346
347 host1x_job_dump(dev, job);
348 }
349
350 /*
351 * Walk the sync_queue, first incrementing with the CPU syncpts that
352 * are partially executed (the first buffer) or fully skipped while
353 * still in the current context (slots are also NOP-ed).
354 *
355 * At the point contexts are interleaved, syncpt increments must be
356 * done inline with the pushbuffer from a GATHER buffer to maintain
357 * the order (slots are modified to be a GATHER of syncpt incrs).
358 *
359 * Note: save in restart_addr the location where the timed out buffer
360 * started in the PB, so we can start the refetch from there (with the
361 * modified NOP-ed PB slots). This lets things appear to have completed
362 * properly for this buffer and resources are freed.
363 */
364
365 dev_dbg(dev, "%s: perform CPU incr on pending same ctx buffers\n",
366 __func__);
367
368 if (!list_empty(&cdma->sync_queue))
369 restart_addr = job->first_get;
370 else
371 restart_addr = cdma->last_pos;
372
373 /* do CPU increments as long as this context continues */
374 list_for_each_entry_from(job, &cdma->sync_queue, list) {
375 /* different context, gets us out of this loop */
376 if (job->client != cdma->timeout.client)
377 break;
378
379 /* won't need a timeout when replayed */
380 job->timeout = 0;
381
382 syncpt_incrs = job->syncpt_end - syncpt_val;
383 dev_dbg(dev, "%s: CPU incr (%d)\n", __func__, syncpt_incrs);
384
385 host1x_job_dump(dev, job);
386
387 /* safe to use CPU to incr syncpts */
388 host1x_hw_cdma_timeout_cpu_incr(host1x, cdma, job->first_get,
389 syncpt_incrs, job->syncpt_end,
390 job->num_slots);
391
392 syncpt_val += syncpt_incrs;
393 }
394
395 /*
396 * The following sumbits from the same client may be dependent on the
397 * failed submit and therefore they may fail. Force a small timeout
398 * to make the queue cleanup faster.
399 */
400
401 list_for_each_entry_from(job, &cdma->sync_queue, list)
402 if (job->client == cdma->timeout.client)
403 job->timeout = min_t(unsigned int, job->timeout, 500);
404
405 dev_dbg(dev, "%s: finished sync_queue modification\n", __func__);
406
407 /* roll back DMAGET and start up channel again */
408 host1x_hw_cdma_resume(host1x, cdma, restart_addr);
409}
410
411/*
412 * Create a cdma
413 */
414int host1x_cdma_init(struct host1x_cdma *cdma)
415{
416 int err;
417
418 mutex_init(&cdma->lock);
419 sema_init(&cdma->sem, 0);
420
421 INIT_LIST_HEAD(&cdma->sync_queue);
422
423 cdma->event = CDMA_EVENT_NONE;
424 cdma->running = false;
425 cdma->torndown = false;
426
427 err = host1x_pushbuffer_init(&cdma->push_buffer);
428 if (err)
429 return err;
430
431 return 0;
432}
433
434/*
435 * Destroy a cdma
436 */
437int host1x_cdma_deinit(struct host1x_cdma *cdma)
438{
439 struct push_buffer *pb = &cdma->push_buffer;
440 struct host1x *host1x = cdma_to_host1x(cdma);
441
442 if (cdma->running) {
443 pr_warn("%s: CDMA still running\n", __func__);
444 return -EBUSY;
445 }
446
447 host1x_pushbuffer_destroy(pb);
448 host1x_hw_cdma_timeout_destroy(host1x, cdma);
449
450 return 0;
451}
452
453/*
454 * Begin a cdma submit
455 */
456int host1x_cdma_begin(struct host1x_cdma *cdma, struct host1x_job *job)
457{
458 struct host1x *host1x = cdma_to_host1x(cdma);
459
460 mutex_lock(&cdma->lock);
461
462 if (job->timeout) {
463 /* init state on first submit with timeout value */
464 if (!cdma->timeout.initialized) {
465 int err;
466
467 err = host1x_hw_cdma_timeout_init(host1x, cdma,
468 job->syncpt_id);
469 if (err) {
470 mutex_unlock(&cdma->lock);
471 return err;
472 }
473 }
474 }
475
476 if (!cdma->running)
477 host1x_hw_cdma_start(host1x, cdma);
478
479 cdma->slots_free = 0;
480 cdma->slots_used = 0;
481 cdma->first_get = cdma->push_buffer.pos;
482
483 trace_host1x_cdma_begin(dev_name(job->channel->dev));
484 return 0;
485}
486
487/*
488 * Push two words into a push buffer slot
489 * Blocks as necessary if the push buffer is full.
490 */
491void host1x_cdma_push(struct host1x_cdma *cdma, u32 op1, u32 op2)
492{
493 struct host1x *host1x = cdma_to_host1x(cdma);
494 struct push_buffer *pb = &cdma->push_buffer;
495 u32 slots_free = cdma->slots_free;
496
497 if (host1x_debug_trace_cmdbuf)
498 trace_host1x_cdma_push(dev_name(cdma_to_channel(cdma)->dev),
499 op1, op2);
500
501 if (slots_free == 0) {
502 host1x_hw_cdma_flush(host1x, cdma);
503 slots_free = host1x_cdma_wait_locked(cdma,
504 CDMA_EVENT_PUSH_BUFFER_SPACE);
505 }
506
507 cdma->slots_free = slots_free - 1;
508 cdma->slots_used++;
509 host1x_pushbuffer_push(pb, op1, op2);
510}
511
512/*
513 * End a cdma submit
514 * Kick off DMA, add job to the sync queue, and a number of slots to be freed
515 * from the pushbuffer. The handles for a submit must all be pinned at the same
516 * time, but they can be unpinned in smaller chunks.
517 */
518void host1x_cdma_end(struct host1x_cdma *cdma,
519 struct host1x_job *job)
520{
521 struct host1x *host1x = cdma_to_host1x(cdma);
522 bool idle = list_empty(&cdma->sync_queue);
523
524 host1x_hw_cdma_flush(host1x, cdma);
525
526 job->first_get = cdma->first_get;
527 job->num_slots = cdma->slots_used;
528 host1x_job_get(job);
529 list_add_tail(&job->list, &cdma->sync_queue);
530
531 /* start timer on idle -> active transitions */
532 if (job->timeout && idle)
533 cdma_start_timer_locked(cdma, job);
534
535 trace_host1x_cdma_end(dev_name(job->channel->dev));
536 mutex_unlock(&cdma->lock);
537}
538
539/*
540 * Update cdma state according to current sync point values
541 */
542void host1x_cdma_update(struct host1x_cdma *cdma)
543{
544 mutex_lock(&cdma->lock);
545 update_cdma_locked(cdma);
546 mutex_unlock(&cdma->lock);
547}