summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/i915/gt/intel_gt_pm.c
blob: f553e2173bdadebc33dd7ee6786907e1ed407d05 (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
// SPDX-License-Identifier: MIT
/*
 * Copyright © 2019 Intel Corporation
 */

#include <linux/string_helpers.h>
#include <linux/suspend.h>

#include "i915_drv.h"
#include "i915_params.h"
#include "intel_context.h"
#include "intel_engine_pm.h"
#include "intel_gt.h"
#include "intel_gt_clock_utils.h"
#include "intel_gt_pm.h"
#include "intel_gt_requests.h"
#include "intel_llc.h"
#include "intel_pm.h"
#include "intel_rc6.h"
#include "intel_rps.h"
#include "intel_wakeref.h"
#include "pxp/intel_pxp_pm.h"

#define I915_GT_SUSPEND_IDLE_TIMEOUT (HZ / 2)

static void user_forcewake(struct intel_gt *gt, bool suspend)
{
	int count = atomic_read(&gt->user_wakeref);

	/* Inside suspend/resume so single threaded, no races to worry about. */
	if (likely(!count))
		return;

	intel_gt_pm_get(gt);
	if (suspend) {
		GEM_BUG_ON(count > atomic_read(&gt->wakeref.count));
		atomic_sub(count, &gt->wakeref.count);
	} else {
		atomic_add(count, &gt->wakeref.count);
	}
	intel_gt_pm_put(gt);
}

static void runtime_begin(struct intel_gt *gt)
{
	local_irq_disable();
	write_seqcount_begin(&gt->stats.lock);
	gt->stats.start = ktime_get();
	gt->stats.active = true;
	write_seqcount_end(&gt->stats.lock);
	local_irq_enable();
}

static void runtime_end(struct intel_gt *gt)
{
	local_irq_disable();
	write_seqcount_begin(&gt->stats.lock);
	gt->stats.active = false;
	gt->stats.total =
		ktime_add(gt->stats.total,
			  ktime_sub(ktime_get(), gt->stats.start));
	write_seqcount_end(&gt->stats.lock);
	local_irq_enable();
}

static int __gt_unpark(struct intel_wakeref *wf)
{
	struct intel_gt *gt = container_of(wf, typeof(*gt), wakeref);
	struct drm_i915_private *i915 = gt->i915;

	GT_TRACE(gt, "\n");

	/*
	 * It seems that the DMC likes to transition between the DC states a lot
	 * when there are no connected displays (no active power domains) during
	 * command submission.
	 *
	 * This activity has negative impact on the performance of the chip with
	 * huge latencies observed in the interrupt handler and elsewhere.
	 *
	 * Work around it by grabbing a GT IRQ power domain whilst there is any
	 * GT activity, preventing any DC state transitions.
	 */
	gt->awake = intel_display_power_get(i915, POWER_DOMAIN_GT_IRQ);
	GEM_BUG_ON(!gt->awake);

	intel_rc6_unpark(&gt->rc6);
	intel_rps_unpark(&gt->rps);
	i915_pmu_gt_unparked(i915);
	intel_guc_busyness_unpark(gt);

	intel_gt_unpark_requests(gt);
	runtime_begin(gt);

	return 0;
}

static int __gt_park(struct intel_wakeref *wf)
{
	struct intel_gt *gt = container_of(wf, typeof(*gt), wakeref);
	intel_wakeref_t wakeref = fetch_and_zero(&gt->awake);
	struct drm_i915_private *i915 = gt->i915;

	GT_TRACE(gt, "\n");

	runtime_end(gt);
	intel_gt_park_requests(gt);

	intel_guc_busyness_park(gt);
	i915_vma_parked(gt);
	i915_pmu_gt_parked(i915);
	intel_rps_park(&gt->rps);
	intel_rc6_park(&gt->rc6);

	/* Everything switched off, flush any residual interrupt just in case */
	intel_synchronize_irq(i915);

	/* Defer dropping the display power well for 100ms, it's slow! */
	GEM_BUG_ON(!wakeref);
	intel_display_power_put_async(i915, POWER_DOMAIN_GT_IRQ, wakeref);

	return 0;
}

static const struct intel_wakeref_ops wf_ops = {
	.get = __gt_unpark,
	.put = __gt_park,
};

void intel_gt_pm_init_early(struct intel_gt *gt)
{
	/*
	 * We access the runtime_pm structure via gt->i915 here rather than
	 * gt->uncore as we do elsewhere in the file because gt->uncore is not
	 * yet initialized for all tiles at this point in the driver startup.
	 * runtime_pm is per-device rather than per-tile, so this is still the
	 * correct structure.
	 */
	intel_wakeref_init(&gt->wakeref, &gt->i915->runtime_pm, &wf_ops);
	seqcount_mutex_init(&gt->stats.lock, &gt->wakeref.mutex);
}

void intel_gt_pm_init(struct intel_gt *gt)
{
	/*
	 * Enabling power-management should be "self-healing". If we cannot
	 * enable a feature, simply leave it disabled with a notice to the
	 * user.
	 */
	intel_rc6_init(&gt->rc6);
	intel_rps_init(&gt->rps);
}

static bool reset_engines(struct intel_gt *gt)
{
	if (INTEL_INFO(gt->i915)->gpu_reset_clobbers_display)
		return false;

	return __intel_gt_reset(gt, ALL_ENGINES) == 0;
}

static void gt_sanitize(struct intel_gt *gt, bool force)
{
	struct intel_engine_cs *engine;
	enum intel_engine_id id;
	intel_wakeref_t wakeref;

	GT_TRACE(gt, "force:%s", str_yes_no(force));

	/* Use a raw wakeref to avoid calling intel_display_power_get early */
	wakeref = intel_runtime_pm_get(gt->uncore->rpm);
	intel_uncore_forcewake_get(gt->uncore, FORCEWAKE_ALL);

	intel_gt_check_clock_frequency(gt);

	/*
	 * As we have just resumed the machine and woken the device up from
	 * deep PCI sleep (presumably D3_cold), assume the HW has been reset
	 * back to defaults, recovering from whatever wedged state we left it
	 * in and so worth trying to use the device once more.
	 */
	if (intel_gt_is_wedged(gt))
		intel_gt_unset_wedged(gt);

	/* For GuC mode, ensure submission is disabled before stopping ring */
	intel_uc_reset_prepare(&gt->uc);

	for_each_engine(engine, gt, id) {
		if (engine->reset.prepare)
			engine->reset.prepare(engine);

		if (engine->sanitize)
			engine->sanitize(engine);
	}

	if (reset_engines(gt) || force) {
		for_each_engine(engine, gt, id)
			__intel_engine_reset(engine, false);
	}

	intel_uc_reset(&gt->uc, false);

	for_each_engine(engine, gt, id)
		if (engine->reset.finish)
			engine->reset.finish(engine);

	intel_rps_sanitize(&gt->rps);

	intel_uncore_forcewake_put(gt->uncore, FORCEWAKE_ALL);
	intel_runtime_pm_put(gt->uncore->rpm, wakeref);
}

void intel_gt_pm_fini(struct intel_gt *gt)
{
	intel_rc6_fini(&gt->rc6);
}

int intel_gt_resume(struct intel_gt *gt)
{
	struct intel_engine_cs *engine;
	enum intel_engine_id id;
	int err;

	err = intel_gt_has_unrecoverable_error(gt);
	if (err)
		return err;

	GT_TRACE(gt, "\n");

	/*
	 * After resume, we may need to poke into the pinned kernel
	 * contexts to paper over any damage caused by the sudden suspend.
	 * Only the kernel contexts should remain pinned over suspend,
	 * allowing us to fixup the user contexts on their first pin.
	 */
	gt_sanitize(gt, true);

	intel_gt_pm_get(gt);

	intel_uncore_forcewake_get(gt->uncore, FORCEWAKE_ALL);
	intel_rc6_sanitize(&gt->rc6);
	if (intel_gt_is_wedged(gt)) {
		err = -EIO;
		goto out_fw;
	}

	/* Only when the HW is re-initialised, can we replay the requests */
	err = intel_gt_init_hw(gt);
	if (err) {
		i915_probe_error(gt->i915,
				 "Failed to initialize GPU, declaring it wedged!\n");
		goto err_wedged;
	}

	intel_uc_reset_finish(&gt->uc);

	intel_rps_enable(&gt->rps);
	intel_llc_enable(&gt->llc);

	for_each_engine(engine, gt, id) {
		intel_engine_pm_get(engine);

		engine->serial++; /* kernel context lost */
		err = intel_engine_resume(engine);

		intel_engine_pm_put(engine);
		if (err) {
			drm_err(&gt->i915->drm,
				"Failed to restart %s (%d)\n",
				engine->name, err);
			goto err_wedged;
		}
	}

	intel_rc6_enable(&gt->rc6);

	intel_uc_resume(&gt->uc);

	intel_pxp_resume(&gt->pxp);

	user_forcewake(gt, false);

out_fw:
	intel_uncore_forcewake_put(gt->uncore, FORCEWAKE_ALL);
	intel_gt_pm_put(gt);
	return err;

err_wedged:
	intel_gt_set_wedged(gt);
	goto out_fw;
}

static void wait_for_suspend(struct intel_gt *gt)
{
	if (!intel_gt_pm_is_awake(gt))
		return;

	if (intel_gt_wait_for_idle(gt, I915_GT_SUSPEND_IDLE_TIMEOUT) == -ETIME) {
		/*
		 * Forcibly cancel outstanding work and leave
		 * the gpu quiet.
		 */
		intel_gt_set_wedged(gt);
		intel_gt_retire_requests(gt);
	}

	intel_gt_pm_wait_for_idle(gt);
}

void intel_gt_suspend_prepare(struct intel_gt *gt)
{
	user_forcewake(gt, true);
	wait_for_suspend(gt);

	intel_pxp_suspend_prepare(&gt->pxp);
}

static suspend_state_t pm_suspend_target(void)
{
#if IS_ENABLED(CONFIG_SUSPEND) && IS_ENABLED(CONFIG_PM_SLEEP)
	return pm_suspend_target_state;
#else
	return PM_SUSPEND_TO_IDLE;
#endif
}

void intel_gt_suspend_late(struct intel_gt *gt)
{
	intel_wakeref_t wakeref;

	/* We expect to be idle already; but also want to be independent */
	wait_for_suspend(gt);

	if (is_mock_gt(gt))
		return;

	GEM_BUG_ON(gt->awake);

	intel_uc_suspend(&gt->uc);
	intel_pxp_suspend(&gt->pxp);

	/*
	 * On disabling the device, we want to turn off HW access to memory
	 * that we no longer own.
	 *
	 * However, not all suspend-states disable the device. S0 (s2idle)
	 * is effectively runtime-suspend, the device is left powered on
	 * but needs to be put into a low power state. We need to keep
	 * powermanagement enabled, but we also retain system state and so
	 * it remains safe to keep on using our allocated memory.
	 */
	if (pm_suspend_target() == PM_SUSPEND_TO_IDLE)
		return;

	with_intel_runtime_pm(gt->uncore->rpm, wakeref) {
		intel_rps_disable(&gt->rps);
		intel_rc6_disable(&gt->rc6);
		intel_llc_disable(&gt->llc);
	}

	gt_sanitize(gt, false);

	GT_TRACE(gt, "\n");
}

void intel_gt_runtime_suspend(struct intel_gt *gt)
{
	intel_pxp_runtime_suspend(&gt->pxp);
	intel_uc_runtime_suspend(&gt->uc);

	GT_TRACE(gt, "\n");
}

int intel_gt_runtime_resume(struct intel_gt *gt)
{
	int ret;

	GT_TRACE(gt, "\n");
	intel_gt_init_swizzling(gt);
	intel_ggtt_restore_fences(gt->ggtt);

	ret = intel_uc_runtime_resume(&gt->uc);
	if (ret)
		return ret;

	intel_pxp_runtime_resume(&gt->pxp);

	return 0;
}

static ktime_t __intel_gt_get_awake_time(const struct intel_gt *gt)
{
	ktime_t total = gt->stats.total;

	if (gt->stats.active)
		total = ktime_add(total,
				  ktime_sub(ktime_get(), gt->stats.start));

	return total;
}

ktime_t intel_gt_get_awake_time(const struct intel_gt *gt)
{
	unsigned int seq;
	ktime_t total;

	do {
		seq = read_seqcount_begin(&gt->stats.lock);
		total = __intel_gt_get_awake_time(gt);
	} while (read_seqcount_retry(&gt->stats.lock, seq));

	return total;
}

#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
#include "selftest_gt_pm.c"
#endif