aboutsummaryrefslogtreecommitdiff
path: root/drivers/gpu/arm/midgard/mali_kbase_softjobs.c
blob: 6c19155808eb5b717b93b983496aa4cc68c86d3d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
/*
 *
 * (C) COPYRIGHT ARM Limited. All rights reserved.
 *
 * This program is free software and is provided to you under the terms of the
 * GNU General Public License version 2 as published by the Free Software
 * Foundation, and any use by you of this program is subject to the terms
 * of such GNU licence.
 *
 * A copy of the licence is included with the program, and can also be obtained
 * from Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
 * Boston, MA  02110-1301, USA.
 *
 */





#include <mali_kbase.h>

#ifdef CONFIG_SYNC
#include <linux/sync.h>
#include <linux/syscalls.h>
#include "mali_kbase_sync.h"
#endif


/* Mask to check cache alignment of data structures */
#define KBASE_CACHE_ALIGNMENT_MASK		((1<<L1_CACHE_SHIFT)-1)

/**
 * @file mali_kbase_softjobs.c
 *
 * This file implements the logic behind software only jobs that are
 * executed within the driver rather than being handed over to the GPU.
 */

static int kbase_dump_cpu_gpu_time(kbase_jd_atom *katom)
{
	kbase_va_region *reg;
	phys_addr_t addr;
	u64 pfn;
	u32 offset;
	char *page;
	struct timespec ts;
	base_dump_cpu_gpu_counters data;
	u64 system_time;
	u64 cycle_counter;
	mali_addr64 jc = katom->jc;
	kbase_context *kctx = katom->kctx;
	int pm_active_err;

	u32 hi1, hi2;

	memset(&data, 0, sizeof(data));

	/* Take the PM active reference as late as possible - otherwise, it could
	 * delay suspend until we process the atom (which may be at the end of a
	 * long chain of dependencies */
	pm_active_err = kbase_pm_context_active_handle_suspend(kctx->kbdev, KBASE_PM_SUSPEND_HANDLER_DONT_REACTIVATE);
	if (pm_active_err) {
		kbasep_js_device_data *js_devdata = &kctx->kbdev->js_data;

		/* We're suspended - queue this on the list of suspended jobs
		 * Use dep_item[1], because dep_item[0] is in use for 'waiting_soft_jobs' */
		mutex_lock(&js_devdata->runpool_mutex);
		list_add_tail(&katom->dep_item[1], &js_devdata->suspended_soft_jobs_list);
		mutex_unlock(&js_devdata->runpool_mutex);

		return pm_active_err;
	}

	kbase_pm_request_gpu_cycle_counter(kctx->kbdev);

	/* Read hi, lo, hi to ensure that overflow from lo to hi is handled correctly */
	do {
		hi1 = kbase_reg_read(kctx->kbdev, GPU_CONTROL_REG(CYCLE_COUNT_HI), NULL);
		cycle_counter = kbase_reg_read(kctx->kbdev, GPU_CONTROL_REG(CYCLE_COUNT_LO), NULL);
		hi2 = kbase_reg_read(kctx->kbdev, GPU_CONTROL_REG(CYCLE_COUNT_HI), NULL);
		cycle_counter |= (((u64) hi1) << 32);
	} while (hi1 != hi2);

	/* Read hi, lo, hi to ensure that overflow from lo to hi is handled correctly */
	do {
		hi1 = kbase_reg_read(kctx->kbdev, GPU_CONTROL_REG(TIMESTAMP_HI), NULL);
		system_time = kbase_reg_read(kctx->kbdev, GPU_CONTROL_REG(TIMESTAMP_LO), NULL);
		hi2 = kbase_reg_read(kctx->kbdev, GPU_CONTROL_REG(TIMESTAMP_HI), NULL);
		system_time |= (((u64) hi1) << 32);
	} while (hi1 != hi2);

	/* Record the CPU's idea of current time */
	getnstimeofday(&ts);

	kbase_pm_release_gpu_cycle_counter(kctx->kbdev);

	kbase_pm_context_idle(kctx->kbdev);

	data.sec = ts.tv_sec;
	data.usec = ts.tv_nsec / 1000;
	data.system_time = system_time;
	data.cycle_counter = cycle_counter;

	pfn = jc >> PAGE_SHIFT;
	offset = jc & ~PAGE_MASK;

	/* Assume this atom will be cancelled until we know otherwise */
	katom->event_code = BASE_JD_EVENT_JOB_CANCELLED;
	if (offset > 0x1000 - sizeof(data)) {
		/* Wouldn't fit in the page */
		return 0;
	}

	reg = kbase_region_tracker_find_region_enclosing_address(kctx, jc);
	if (!reg)
		return 0;

	if (!(reg->flags & KBASE_REG_GPU_WR)) {
		/* Region is not writable by GPU so we won't write to it either */
		return 0;
	}

	if (!reg->alloc->pages)
		return 0;

	addr = reg->alloc->pages[pfn - reg->start_pfn];
	if (!addr)
		return 0;

	page = kmap(pfn_to_page(PFN_DOWN(addr)));
	if (!page)
		return 0;

	memcpy(page + offset, &data, sizeof(data));
	kbase_sync_to_cpu(addr + offset, page + offset, sizeof(data));
	kunmap(pfn_to_page(PFN_DOWN(addr)));

	/* Atom was fine - mark it as done */
	katom->event_code = BASE_JD_EVENT_DONE;

	return 0;
}

#ifdef CONFIG_SYNC

/* Complete an atom that has returned '1' from kbase_process_soft_job (i.e. has waited)
 *
 * @param katom     The atom to complete
 */
static void complete_soft_job(kbase_jd_atom *katom)
{
	kbase_context *kctx = katom->kctx;

	mutex_lock(&kctx->jctx.lock);
	list_del(&katom->dep_item[0]);
	kbase_finish_soft_job(katom);
	if (jd_done_nolock(katom))
		kbasep_js_try_schedule_head_ctx(kctx->kbdev);
	mutex_unlock(&kctx->jctx.lock);
}

static base_jd_event_code kbase_fence_trigger(kbase_jd_atom *katom, int result)
{
	struct sync_pt *pt;
	struct sync_timeline *timeline;

	if (!list_is_singular(&katom->fence->pt_list_head)) {
		/* Not exactly one item in the list - so it didn't (directly) come from us */
		return BASE_JD_EVENT_JOB_CANCELLED;
	}

	pt = list_first_entry(&katom->fence->pt_list_head, struct sync_pt, pt_list);
	timeline = pt->parent;

	if (!kbase_sync_timeline_is_ours(timeline)) {
		/* Fence has a sync_pt which isn't ours! */
		return BASE_JD_EVENT_JOB_CANCELLED;
	}

	kbase_sync_signal_pt(pt, result);

	sync_timeline_signal(timeline);

	return (result < 0) ? BASE_JD_EVENT_JOB_CANCELLED : BASE_JD_EVENT_DONE;
}

static void kbase_fence_wait_worker(struct work_struct *data)
{
	kbase_jd_atom *katom;
	kbase_context *kctx;

	katom = container_of(data, kbase_jd_atom, work);
	kctx = katom->kctx;

	complete_soft_job(katom);
}

static void kbase_fence_wait_callback(struct sync_fence *fence, struct sync_fence_waiter *waiter)
{
	kbase_jd_atom *katom = container_of(waiter, kbase_jd_atom, sync_waiter);
	kbase_context *kctx;

	KBASE_DEBUG_ASSERT(NULL != katom);

	kctx = katom->kctx;

	KBASE_DEBUG_ASSERT(NULL != kctx);

	/* Propagate the fence status to the atom.
	 * If negative then cancel this atom and its dependencies.
	 */
	if (fence->status < 0)
	{
		katom->event_code = BASE_JD_EVENT_JOB_CANCELLED;
	}

	/* To prevent a potential deadlock we schedule the work onto the job_done_wq workqueue
	 *
	 * The issue is that we may signal the timeline while holding kctx->jctx.lock and
	 * the callbacks are run synchronously from sync_timeline_signal. So we simply defer the work.
	 */

	KBASE_DEBUG_ASSERT(0 == object_is_on_stack(&katom->work));
	INIT_WORK(&katom->work, kbase_fence_wait_worker);
	queue_work(kctx->jctx.job_done_wq, &katom->work);
}

static int kbase_fence_wait(kbase_jd_atom *katom)
{
	int ret;

	KBASE_DEBUG_ASSERT(NULL != katom);
	KBASE_DEBUG_ASSERT(NULL != katom->kctx);

	sync_fence_waiter_init(&katom->sync_waiter, kbase_fence_wait_callback);

	ret = sync_fence_wait_async(katom->fence, &katom->sync_waiter);

	if (ret == 1) {
		/* Already signalled */
		return 0;
	} else if (ret < 0) {
		goto cancel_atom;
	}
	return 1;

 cancel_atom:
	katom->event_code = BASE_JD_EVENT_JOB_CANCELLED;
	/* We should cause the dependant jobs in the bag to be failed,
	 * to do this we schedule the work queue to complete this job */
	KBASE_DEBUG_ASSERT(0 == object_is_on_stack(&katom->work));
	INIT_WORK(&katom->work, kbase_fence_wait_worker);
	queue_work(katom->kctx->jctx.job_done_wq, &katom->work);
	return 1;
}

static void kbase_fence_cancel_wait(kbase_jd_atom *katom)
{
	if (sync_fence_cancel_async(katom->fence, &katom->sync_waiter) != 0)
	{
		/* The wait wasn't cancelled - leave the cleanup for kbase_fence_wait_callback */
		return;
	}

	/* Wait was cancelled - zap the atoms */
	katom->event_code = BASE_JD_EVENT_JOB_CANCELLED;

	kbase_finish_soft_job(katom);

	if (jd_done_nolock(katom))
		kbasep_js_try_schedule_head_ctx(katom->kctx->kbdev);
}
#endif /* CONFIG_SYNC */

int kbase_process_soft_job(kbase_jd_atom *katom)
{
	int status;
	switch (katom->core_req & BASEP_JD_REQ_ATOM_TYPE) {
	case BASE_JD_REQ_SOFT_DUMP_CPU_GPU_TIME:
		return kbase_dump_cpu_gpu_time(katom);
#ifdef CONFIG_SYNC
	case BASE_JD_REQ_SOFT_FENCE_TRIGGER:
		KBASE_DEBUG_ASSERT(katom->fence != NULL);
		katom->event_code = kbase_fence_trigger(katom, katom->event_code == BASE_JD_EVENT_DONE ? 0 : -EFAULT);
		/* Release the reference as we don't need it any more */
		sync_fence_put(katom->fence);
		katom->fence = NULL;
		break;
	case BASE_JD_REQ_SOFT_FENCE_WAIT:
		return kbase_fence_wait(katom);
#endif				/* CONFIG_SYNC */
	case BASE_JD_REQ_SOFT_REPLAY:
		status = kbase_replay_process(katom);
		if (status & MALI_REPLAY_FLAG_JS_RESCHED)
			pr_err("replay called from kbase_process_soft_job - missing resched!\n");
		return status & MALI_REPLAY_STATUS_MASK;
	}

	/* Atom is complete */
	return 0;
}

void kbase_cancel_soft_job(kbase_jd_atom *katom)
{
	switch (katom->core_req & BASEP_JD_REQ_ATOM_TYPE) {
#ifdef CONFIG_SYNC
	case BASE_JD_REQ_SOFT_FENCE_WAIT:
		kbase_fence_cancel_wait(katom);
		break;
#endif
	default:
		/* This soft-job doesn't support cancellation! */
		KBASE_DEBUG_ASSERT(0);
	}
}

mali_error kbase_prepare_soft_job(kbase_jd_atom *katom)
{
	switch (katom->core_req & BASEP_JD_REQ_ATOM_TYPE) {
	case BASE_JD_REQ_SOFT_DUMP_CPU_GPU_TIME:
		{
			if(0 != (katom->jc & KBASE_CACHE_ALIGNMENT_MASK))
				return MALI_ERROR_FUNCTION_FAILED;
		}
		break;
#ifdef CONFIG_SYNC
	case BASE_JD_REQ_SOFT_FENCE_TRIGGER:
		{
			base_fence fence;
			int fd;
			if (0 != copy_from_user(&fence, (__user void *)(uintptr_t) katom->jc, sizeof(fence)))
				return MALI_ERROR_FUNCTION_FAILED;

			fd = kbase_stream_create_fence(fence.basep.stream_fd);
			if (fd < 0)
				return MALI_ERROR_FUNCTION_FAILED;

			katom->fence = sync_fence_fdget(fd);

			if (katom->fence == NULL) {
				/* The only way the fence can be NULL is if userspace closed it for us.
				 * So we don't need to clear it up */
				return MALI_ERROR_FUNCTION_FAILED;
			}
			fence.basep.fd = fd;
			if (0 != copy_to_user((__user void *)(uintptr_t) katom->jc, &fence, sizeof(fence))) {
				katom->fence = NULL;
				sys_close(fd);
				return MALI_ERROR_FUNCTION_FAILED;
			}
		}
		break;
	case BASE_JD_REQ_SOFT_FENCE_WAIT:
		{
			base_fence fence;
			if (0 != copy_from_user(&fence, (__user void *)(uintptr_t) katom->jc, sizeof(fence)))
				return MALI_ERROR_FUNCTION_FAILED;

			/* Get a reference to the fence object */
			katom->fence = sync_fence_fdget(fence.basep.fd);
			if (katom->fence == NULL)
				return MALI_ERROR_FUNCTION_FAILED;
		}
		break;
#endif				/* CONFIG_SYNC */
	case BASE_JD_REQ_SOFT_REPLAY:
		break;
	default:
		/* Unsupported soft-job */
		return MALI_ERROR_FUNCTION_FAILED;
	}
	return MALI_ERROR_NONE;
}

void kbase_finish_soft_job(kbase_jd_atom *katom)
{
	switch (katom->core_req & BASEP_JD_REQ_ATOM_TYPE) {
	case BASE_JD_REQ_SOFT_DUMP_CPU_GPU_TIME:
		/* Nothing to do */
		break;
#ifdef CONFIG_SYNC
	case BASE_JD_REQ_SOFT_FENCE_TRIGGER:
		if (katom->fence) {
			/* The fence has not yet been signalled, so we do it now */
			kbase_fence_trigger(katom, katom->event_code == BASE_JD_EVENT_DONE ? 0 : -EFAULT);
			sync_fence_put(katom->fence);
			katom->fence = NULL;
		}
		break;
	case BASE_JD_REQ_SOFT_FENCE_WAIT:
		/* Release the reference to the fence object */
		sync_fence_put(katom->fence);
		katom->fence = NULL;
		break;
#endif				/* CONFIG_SYNC */
	}
}

void kbase_resume_suspended_soft_jobs(kbase_device *kbdev)
{
	LIST_HEAD(local_suspended_soft_jobs);
	kbase_jd_atom *tmp_iter;
	kbase_jd_atom *katom_iter;
	kbasep_js_device_data *js_devdata;
	mali_bool resched = MALI_FALSE;
	KBASE_DEBUG_ASSERT(kbdev);

	js_devdata = &kbdev->js_data;

	/* Move out the entire list */
	mutex_lock(&js_devdata->runpool_mutex);
	list_splice_init(&js_devdata->suspended_soft_jobs_list, &local_suspended_soft_jobs);
	mutex_unlock(&js_devdata->runpool_mutex);

	/* Each atom must be detached from the list and ran separately - it could
	 * be re-added to the old list, but this is unlikely */
	list_for_each_entry_safe(katom_iter, tmp_iter, &local_suspended_soft_jobs, dep_item[1])
	{
		kbase_context *kctx = katom_iter->kctx;
		mutex_lock(&kctx->jctx.lock);

		/* Remove from the global list */
		list_del(&katom_iter->dep_item[1]);
		/* Remove from the context's list of waiting soft jobs */
		list_del(&katom_iter->dep_item[0]);

		if (kbase_process_soft_job(katom_iter) == 0) {
			kbase_finish_soft_job(katom_iter);
			resched |= jd_done_nolock(katom_iter);
		} else {
			/* The job has not completed */
			KBASE_DEBUG_ASSERT((katom_iter->core_req & BASEP_JD_REQ_ATOM_TYPE)
						  		!= BASE_JD_REQ_SOFT_REPLAY);
			list_add_tail(&katom_iter->dep_item[0], &kctx->waiting_soft_jobs);
		}

		mutex_unlock(&kctx->jctx.lock);
	}

	if (resched)
		kbasep_js_try_schedule_head_ctx(kbdev);
}