Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'drm-fixes-2019-12-28' of git://anongit.freedesktop.org/drm/drm
Pull drm fixes from Dave Airlie:
"Post-xmas food coma recovery fixes. Only three fixes for i915 since I
expect most people are holidaying.
i915:
- power management rc6 fix
- framebuffer tracking fix
- display power management ratelimit fix"
* tag 'drm-fixes-2019-12-28' of git://anongit.freedesktop.org/drm/drm:
drm/i915: Hold reference to intel_frontbuffer as we track activity
drm/i915/gt: Ratelimit display power w/a
drm/i915/pmu: Ensure monotonic rc6
+139
-86
+1
-1
drivers/gpu/drm/i915/display/intel_display.c
+1
-1
drivers/gpu/drm/i915/display/intel_display.c
···
15112
15112
return ret;
15113
15113
15114
15114
fb_obj_bump_render_priority(obj);
15115
-
intel_frontbuffer_flush(obj->frontbuffer, ORIGIN_DIRTYFB);
15115
+
i915_gem_object_flush_frontbuffer(obj, ORIGIN_DIRTYFB);
15116
15116
15117
15117
if (!new_plane_state->base.fence) { /* implicit fencing */
15118
15118
struct dma_fence *fence;
+6
-10
drivers/gpu/drm/i915/display/intel_frontbuffer.c
+6
-10
drivers/gpu/drm/i915/display/intel_frontbuffer.c
···
229
229
vma->display_alignment = I915_GTT_MIN_ALIGNMENT;
230
230
spin_unlock(&obj->vma.lock);
231
231
232
-
obj->frontbuffer = NULL;
232
+
RCU_INIT_POINTER(obj->frontbuffer, NULL);
233
233
spin_unlock(&to_i915(obj->base.dev)->fb_tracking.lock);
234
234
235
235
i915_gem_object_put(obj);
236
-
kfree(front);
236
+
kfree_rcu(front, rcu);
237
237
}
238
238
239
239
struct intel_frontbuffer *
···
242
242
struct drm_i915_private *i915 = to_i915(obj->base.dev);
243
243
struct intel_frontbuffer *front;
244
244
245
-
spin_lock(&i915->fb_tracking.lock);
246
-
front = obj->frontbuffer;
247
-
if (front)
248
-
kref_get(&front->ref);
249
-
spin_unlock(&i915->fb_tracking.lock);
245
+
front = __intel_frontbuffer_get(obj);
250
246
if (front)
251
247
return front;
252
248
···
258
262
i915_active_may_sleep(frontbuffer_retire));
259
263
260
264
spin_lock(&i915->fb_tracking.lock);
261
-
if (obj->frontbuffer) {
265
+
if (rcu_access_pointer(obj->frontbuffer)) {
262
266
kfree(front);
263
-
front = obj->frontbuffer;
267
+
front = rcu_dereference_protected(obj->frontbuffer, true);
264
268
kref_get(&front->ref);
265
269
} else {
266
270
i915_gem_object_get(obj);
267
-
obj->frontbuffer = front;
271
+
rcu_assign_pointer(obj->frontbuffer, front);
268
272
}
269
273
spin_unlock(&i915->fb_tracking.lock);
270
274
+31
-3
drivers/gpu/drm/i915/display/intel_frontbuffer.h
+31
-3
drivers/gpu/drm/i915/display/intel_frontbuffer.h
···
27
27
#include <linux/atomic.h>
28
28
#include <linux/kref.h>
29
29
30
+
#include "gem/i915_gem_object_types.h"
30
31
#include "i915_active.h"
31
32
32
33
struct drm_i915_private;
33
-
struct drm_i915_gem_object;
34
34
35
35
enum fb_op_origin {
36
36
ORIGIN_GTT,
···
45
45
atomic_t bits;
46
46
struct i915_active write;
47
47
struct drm_i915_gem_object *obj;
48
+
struct rcu_head rcu;
48
49
};
49
50
50
51
void intel_frontbuffer_flip_prepare(struct drm_i915_private *i915,
···
54
53
unsigned frontbuffer_bits);
55
54
void intel_frontbuffer_flip(struct drm_i915_private *i915,
56
55
unsigned frontbuffer_bits);
56
+
57
+
void intel_frontbuffer_put(struct intel_frontbuffer *front);
58
+
59
+
static inline struct intel_frontbuffer *
60
+
__intel_frontbuffer_get(const struct drm_i915_gem_object *obj)
61
+
{
62
+
struct intel_frontbuffer *front;
63
+
64
+
if (likely(!rcu_access_pointer(obj->frontbuffer)))
65
+
return NULL;
66
+
67
+
rcu_read_lock();
68
+
do {
69
+
front = rcu_dereference(obj->frontbuffer);
70
+
if (!front)
71
+
break;
72
+
73
+
if (unlikely(!kref_get_unless_zero(&front->ref)))
74
+
continue;
75
+
76
+
if (likely(front == rcu_access_pointer(obj->frontbuffer)))
77
+
break;
78
+
79
+
intel_frontbuffer_put(front);
80
+
} while (1);
81
+
rcu_read_unlock();
82
+
83
+
return front;
84
+
}
57
85
58
86
struct intel_frontbuffer *
59
87
intel_frontbuffer_get(struct drm_i915_gem_object *obj);
···
148
118
void intel_frontbuffer_track(struct intel_frontbuffer *old,
149
119
struct intel_frontbuffer *new,
150
120
unsigned int frontbuffer_bits);
151
-
152
-
void intel_frontbuffer_put(struct intel_frontbuffer *front);
153
121
154
122
#endif /* __INTEL_FRONTBUFFER_H__ */
+13
-4
drivers/gpu/drm/i915/display/intel_overlay.c
+13
-4
drivers/gpu/drm/i915/display/intel_overlay.c
···
279
279
struct i915_vma *vma)
280
280
{
281
281
enum pipe pipe = overlay->crtc->pipe;
282
+
struct intel_frontbuffer *from = NULL, *to = NULL;
282
283
283
284
WARN_ON(overlay->old_vma);
284
285
285
-
intel_frontbuffer_track(overlay->vma ? overlay->vma->obj->frontbuffer : NULL,
286
-
vma ? vma->obj->frontbuffer : NULL,
287
-
INTEL_FRONTBUFFER_OVERLAY(pipe));
286
+
if (overlay->vma)
287
+
from = intel_frontbuffer_get(overlay->vma->obj);
288
+
if (vma)
289
+
to = intel_frontbuffer_get(vma->obj);
290
+
291
+
intel_frontbuffer_track(from, to, INTEL_FRONTBUFFER_OVERLAY(pipe));
292
+
293
+
if (to)
294
+
intel_frontbuffer_put(to);
295
+
if (from)
296
+
intel_frontbuffer_put(from);
288
297
289
298
intel_frontbuffer_flip_prepare(overlay->i915,
290
299
INTEL_FRONTBUFFER_OVERLAY(pipe));
···
775
766
ret = PTR_ERR(vma);
776
767
goto out_pin_section;
777
768
}
778
-
intel_frontbuffer_flush(new_bo->frontbuffer, ORIGIN_DIRTYFB);
769
+
i915_gem_object_flush_frontbuffer(new_bo, ORIGIN_DIRTYFB);
779
770
780
771
if (!overlay->active) {
781
772
u32 oconfig;
+2
-1
drivers/gpu/drm/i915/gem/i915_gem_clflush.c
+2
-1
drivers/gpu/drm/i915/gem/i915_gem_clflush.c
+2
-2
drivers/gpu/drm/i915/gem/i915_gem_domain.c
+2
-2
drivers/gpu/drm/i915/gem/i915_gem_domain.c
···
664
664
i915_gem_object_unlock(obj);
665
665
666
666
if (write_domain)
667
-
intel_frontbuffer_invalidate(obj->frontbuffer, ORIGIN_CPU);
667
+
i915_gem_object_invalidate_frontbuffer(obj, ORIGIN_CPU);
668
668
669
669
out_unpin:
670
670
i915_gem_object_unpin_pages(obj);
···
784
784
}
785
785
786
786
out:
787
-
intel_frontbuffer_invalidate(obj->frontbuffer, ORIGIN_CPU);
787
+
i915_gem_object_invalidate_frontbuffer(obj, ORIGIN_CPU);
788
788
obj->mm.dirty = true;
789
789
/* return with the pages pinned */
790
790
return 0;
+25
-1
drivers/gpu/drm/i915/gem/i915_gem_object.c
+25
-1
drivers/gpu/drm/i915/gem/i915_gem_object.c
···
280
280
for_each_ggtt_vma(vma, obj)
281
281
intel_gt_flush_ggtt_writes(vma->vm->gt);
282
282
283
-
intel_frontbuffer_flush(obj->frontbuffer, ORIGIN_CPU);
283
+
i915_gem_object_flush_frontbuffer(obj, ORIGIN_CPU);
284
284
285
285
for_each_ggtt_vma(vma, obj) {
286
286
if (vma->iomap)
···
306
306
}
307
307
308
308
obj->write_domain = 0;
309
+
}
310
+
311
+
void __i915_gem_object_flush_frontbuffer(struct drm_i915_gem_object *obj,
312
+
enum fb_op_origin origin)
313
+
{
314
+
struct intel_frontbuffer *front;
315
+
316
+
front = __intel_frontbuffer_get(obj);
317
+
if (front) {
318
+
intel_frontbuffer_flush(front, origin);
319
+
intel_frontbuffer_put(front);
320
+
}
321
+
}
322
+
323
+
void __i915_gem_object_invalidate_frontbuffer(struct drm_i915_gem_object *obj,
324
+
enum fb_op_origin origin)
325
+
{
326
+
struct intel_frontbuffer *front;
327
+
328
+
front = __intel_frontbuffer_get(obj);
329
+
if (front) {
330
+
intel_frontbuffer_invalidate(front, origin);
331
+
intel_frontbuffer_put(front);
332
+
}
309
333
}
310
334
311
335
void i915_gem_init__objects(struct drm_i915_private *i915)
+22
-1
drivers/gpu/drm/i915/gem/i915_gem_object.h
+22
-1
drivers/gpu/drm/i915/gem/i915_gem_object.h
···
13
13
14
14
#include <drm/i915_drm.h>
15
15
16
+
#include "display/intel_frontbuffer.h"
16
17
#include "i915_gem_object_types.h"
17
-
18
18
#include "i915_gem_gtt.h"
19
19
20
20
void i915_gem_init__objects(struct drm_i915_private *i915);
···
462
462
int i915_gem_object_wait_priority(struct drm_i915_gem_object *obj,
463
463
unsigned int flags,
464
464
const struct i915_sched_attr *attr);
465
+
466
+
void __i915_gem_object_flush_frontbuffer(struct drm_i915_gem_object *obj,
467
+
enum fb_op_origin origin);
468
+
void __i915_gem_object_invalidate_frontbuffer(struct drm_i915_gem_object *obj,
469
+
enum fb_op_origin origin);
470
+
471
+
static inline void
472
+
i915_gem_object_flush_frontbuffer(struct drm_i915_gem_object *obj,
473
+
enum fb_op_origin origin)
474
+
{
475
+
if (unlikely(rcu_access_pointer(obj->frontbuffer)))
476
+
__i915_gem_object_flush_frontbuffer(obj, origin);
477
+
}
478
+
479
+
static inline void
480
+
i915_gem_object_invalidate_frontbuffer(struct drm_i915_gem_object *obj,
481
+
enum fb_op_origin origin)
482
+
{
483
+
if (unlikely(rcu_access_pointer(obj->frontbuffer)))
484
+
__i915_gem_object_invalidate_frontbuffer(obj, origin);
485
+
}
465
486
466
487
#endif
+1
-1
drivers/gpu/drm/i915/gem/i915_gem_object_types.h
+1
-1
drivers/gpu/drm/i915/gem/i915_gem_object_types.h
+2
-1
drivers/gpu/drm/i915/gt/intel_gt_pm.c
+2
-1
drivers/gpu/drm/i915/gt/intel_gt_pm.c
···
94
94
intel_uncore_forcewake_put(&i915->uncore, FORCEWAKE_ALL);
95
95
}
96
96
97
+
/* Defer dropping the display power well for 100ms, it's slow! */
97
98
GEM_BUG_ON(!wakeref);
98
-
intel_display_power_put(i915, POWER_DOMAIN_GT_IRQ, wakeref);
99
+
intel_display_power_put_async(i915, POWER_DOMAIN_GT_IRQ, wakeref);
99
100
100
101
i915_globals_park();
101
102
+5
-5
drivers/gpu/drm/i915/i915_gem.c
+5
-5
drivers/gpu/drm/i915/i915_gem.c
···
161
161
* We manually control the domain here and pretend that it
162
162
* remains coherent i.e. in the GTT domain, like shmem_pwrite.
163
163
*/
164
-
intel_frontbuffer_invalidate(obj->frontbuffer, ORIGIN_CPU);
164
+
i915_gem_object_invalidate_frontbuffer(obj, ORIGIN_CPU);
165
165
166
166
if (copy_from_user(vaddr, user_data, args->size))
167
167
return -EFAULT;
···
169
169
drm_clflush_virt_range(vaddr, args->size);
170
170
intel_gt_chipset_flush(&to_i915(obj->base.dev)->gt);
171
171
172
-
intel_frontbuffer_flush(obj->frontbuffer, ORIGIN_CPU);
172
+
i915_gem_object_flush_frontbuffer(obj, ORIGIN_CPU);
173
173
return 0;
174
174
}
175
175
···
589
589
goto out_unpin;
590
590
}
591
591
592
-
intel_frontbuffer_invalidate(obj->frontbuffer, ORIGIN_CPU);
592
+
i915_gem_object_invalidate_frontbuffer(obj, ORIGIN_CPU);
593
593
594
594
user_data = u64_to_user_ptr(args->data_ptr);
595
595
offset = args->offset;
···
631
631
user_data += page_length;
632
632
offset += page_length;
633
633
}
634
-
intel_frontbuffer_flush(obj->frontbuffer, ORIGIN_CPU);
634
+
i915_gem_object_flush_frontbuffer(obj, ORIGIN_CPU);
635
635
636
636
i915_gem_object_unlock_fence(obj, fence);
637
637
out_unpin:
···
721
721
offset = 0;
722
722
}
723
723
724
-
intel_frontbuffer_flush(obj->frontbuffer, ORIGIN_CPU);
724
+
i915_gem_object_flush_frontbuffer(obj, ORIGIN_CPU);
725
725
i915_gem_object_unlock_fence(obj, fence);
726
726
727
727
return ret;
+20
-53
drivers/gpu/drm/i915/i915_pmu.c
+20
-53
drivers/gpu/drm/i915/i915_pmu.c
···
144
144
return ktime_to_ns(ktime_sub(ktime_get(), kt));
145
145
}
146
146
147
-
static u64 __pmu_estimate_rc6(struct i915_pmu *pmu)
148
-
{
149
-
u64 val;
150
-
151
-
/*
152
-
* We think we are runtime suspended.
153
-
*
154
-
* Report the delta from when the device was suspended to now,
155
-
* on top of the last known real value, as the approximated RC6
156
-
* counter value.
157
-
*/
158
-
val = ktime_since(pmu->sleep_last);
159
-
val += pmu->sample[__I915_SAMPLE_RC6].cur;
160
-
161
-
pmu->sample[__I915_SAMPLE_RC6_ESTIMATED].cur = val;
162
-
163
-
return val;
164
-
}
165
-
166
-
static u64 __pmu_update_rc6(struct i915_pmu *pmu, u64 val)
167
-
{
168
-
/*
169
-
* If we are coming back from being runtime suspended we must
170
-
* be careful not to report a larger value than returned
171
-
* previously.
172
-
*/
173
-
if (val >= pmu->sample[__I915_SAMPLE_RC6_ESTIMATED].cur) {
174
-
pmu->sample[__I915_SAMPLE_RC6_ESTIMATED].cur = 0;
175
-
pmu->sample[__I915_SAMPLE_RC6].cur = val;
176
-
} else {
177
-
val = pmu->sample[__I915_SAMPLE_RC6_ESTIMATED].cur;
178
-
}
179
-
180
-
return val;
181
-
}
182
-
183
147
static u64 get_rc6(struct intel_gt *gt)
184
148
{
185
149
struct drm_i915_private *i915 = gt->i915;
186
150
struct i915_pmu *pmu = &i915->pmu;
187
151
unsigned long flags;
152
+
bool awake = false;
188
153
u64 val;
189
154
190
-
val = 0;
191
155
if (intel_gt_pm_get_if_awake(gt)) {
192
156
val = __get_rc6(gt);
193
157
intel_gt_pm_put_async(gt);
158
+
awake = true;
194
159
}
195
160
196
161
spin_lock_irqsave(&pmu->lock, flags);
197
162
198
-
if (val)
199
-
val = __pmu_update_rc6(pmu, val);
163
+
if (awake) {
164
+
pmu->sample[__I915_SAMPLE_RC6].cur = val;
165
+
} else {
166
+
/*
167
+
* We think we are runtime suspended.
168
+
*
169
+
* Report the delta from when the device was suspended to now,
170
+
* on top of the last known real value, as the approximated RC6
171
+
* counter value.
172
+
*/
173
+
val = ktime_since(pmu->sleep_last);
174
+
val += pmu->sample[__I915_SAMPLE_RC6].cur;
175
+
}
176
+
177
+
if (val < pmu->sample[__I915_SAMPLE_RC6_LAST_REPORTED].cur)
178
+
val = pmu->sample[__I915_SAMPLE_RC6_LAST_REPORTED].cur;
200
179
else
201
-
val = __pmu_estimate_rc6(pmu);
180
+
pmu->sample[__I915_SAMPLE_RC6_LAST_REPORTED].cur = val;
202
181
203
182
spin_unlock_irqrestore(&pmu->lock, flags);
204
183
···
189
210
struct i915_pmu *pmu = &i915->pmu;
190
211
191
212
if (pmu->enable & config_enabled_mask(I915_PMU_RC6_RESIDENCY))
192
-
__pmu_update_rc6(pmu, __get_rc6(&i915->gt));
213
+
pmu->sample[__I915_SAMPLE_RC6].cur = __get_rc6(&i915->gt);
193
214
194
215
pmu->sleep_last = ktime_get();
195
-
}
196
-
197
-
static void unpark_rc6(struct drm_i915_private *i915)
198
-
{
199
-
struct i915_pmu *pmu = &i915->pmu;
200
-
201
-
/* Estimate how long we slept and accumulate that into rc6 counters */
202
-
if (pmu->enable & config_enabled_mask(I915_PMU_RC6_RESIDENCY))
203
-
__pmu_estimate_rc6(pmu);
204
216
}
205
217
206
218
#else
···
202
232
}
203
233
204
234
static void park_rc6(struct drm_i915_private *i915) {}
205
-
static void unpark_rc6(struct drm_i915_private *i915) {}
206
235
207
236
#endif
208
237
···
249
280
* Re-enable sampling timer when GPU goes active.
250
281
*/
251
282
__i915_pmu_maybe_start_timer(pmu);
252
-
253
-
unpark_rc6(i915);
254
283
255
284
spin_unlock_irq(&pmu->lock);
256
285
}
+1
-1
drivers/gpu/drm/i915/i915_pmu.h
+1
-1
drivers/gpu/drm/i915/i915_pmu.h
+8
-2
drivers/gpu/drm/i915/i915_vma.c
+8
-2
drivers/gpu/drm/i915/i915_vma.c
···
1104
1104
return err;
1105
1105
1106
1106
if (flags & EXEC_OBJECT_WRITE) {
1107
-
if (intel_frontbuffer_invalidate(obj->frontbuffer, ORIGIN_CS))
1108
-
i915_active_add_request(&obj->frontbuffer->write, rq);
1107
+
struct intel_frontbuffer *front;
1108
+
1109
+
front = __intel_frontbuffer_get(obj);
1110
+
if (unlikely(front)) {
1111
+
if (intel_frontbuffer_invalidate(front, ORIGIN_CS))
1112
+
i915_active_add_request(&front->write, rq);
1113
+
intel_frontbuffer_put(front);
1114
+
}
1109
1115
1110
1116
dma_resv_add_excl_fence(vma->resv, &rq->fence);
1111
1117
obj->write_domain = I915_GEM_DOMAIN_RENDER;