Skip to contentSkip to conversation

drm/i915: Don't call synchronize_rcu_expedited under struct_mutex

PATCH In review Joonas Lahtinen 4 participants 7 messages Apr 7, 2017

Subsystems (2)
lore
Share & tools
lore

Joonas Lahtinen

author maintainer
#0 lore raw

PATCH 1/2 drm/i915: Don't call synchronize_rcu_expedited under struct_mutex

Only call synchronize_rcu_expedited after unlocking struct_mutex to avoid deadlock because the workqueues depend on struct_mutex.

From original patch by Andrea:

synchronize_rcu/synchronize_sched/synchronize_rcu_expedited() will hang until its own workqueues are run. The i915 gem workqueues will wait on the struct_mutex to be released. So we cannot wait for a quiescent state using those rcu primitives while holding the struct_mutex or it creates a circular lock dependency resulting in kernel hangs (which is reproducible but goes undetected by lockdep).

kswapd0 D 0 700 2 0x00000000

Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? _synchronize_rcu_expedited.constprop.65+0x2ef/0x300
? wake_up_bit+0x20/0x20
? rcu_stall_kick_kthreads.part.54+0xc0/0xc0
? rcu_exp_wait_wake+0x530/0x530
? i915_gem_shrink+0x34b/0x4b0
? i915_gem_shrinker_scan+0x7c/0x90
? i915_gem_shrinker_scan+0x7c/0x90
? shrink_slab.part.61.constprop.72+0x1c1/0x3a0
? shrink_zone+0x154/0x160
? kswapd+0x40a/0x720
? kthread+0xf4/0x130
? try_to_free_pages+0x450/0x450
? kthread_create_on_node+0x40/0x40
? ret_from_fork+0x23/0x30
plasmashell     D    0  4657   4614 0x00000000
Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? schedule_preempt_disabled+0xe/0x10
? __mutex_lock.isra.4+0x1c9/0x790
? i915_gem_close_object+0x26/0xc0
? i915_gem_close_object+0x26/0xc0
? drm_gem_object_release_handle+0x48/0x90
? drm_gem_handle_delete+0x50/0x80
? drm_ioctl+0x1fa/0x420
? drm_gem_handle_create+0x40/0x40
? pipe_write+0x391/0x410
? __vfs_write+0xc6/0x120
? do_vfs_ioctl+0x8b/0x5d0
? SyS_ioctl+0x3b/0x70
? entry_SYSCALL_64_fastpath+0x13/0x94
kworker/0:0     D    0 29186      2 0x00000000
Workqueue: events __i915_gem_free_work
Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? schedule_preempt_disabled+0xe/0x10
? __mutex_lock.isra.4+0x1c9/0x790
? del_timer_sync+0x44/0x50
? update_curr+0x57/0x110
? __i915_gem_free_objects+0x31/0x300
? __i915_gem_free_objects+0x31/0x300
? __i915_gem_free_work+0x2d/0x40
? process_one_work+0x13a/0x3b0
? worker_thread+0x4a/0x460
? kthread+0xf4/0x130
? process_one_work+0x3b0/0x3b0
? kthread_create_on_node+0x40/0x40
? ret_from_fork+0x23/0x30
Fixes 3d3d18f086cd ("drm/i915: Avoid rcu_barrier() from reclaim paths (shrinker)")
Cc Andrea Arcangeli <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Jani Nikula <[email protected]>
---
 drivers/gpu/drm/i915/i915_gem_shrinker.c | 26 ++++++++++++++++----------
 1 file changed, 16 insertions(+), 10 deletions(-)
drivers/gpu/drm/i915/i915_gem_shrinker.c +16 -10
index 2978acd..129ed30 100644
@@ -53,6 +53,17 @@ static bool i915_gem_shrinker_lock(struct drm_device *dev, bool *unlock)
 	BUG();
 }
 
+static void i915_gem_shrinker_unlock(struct drm_device *dev, bool unlock)
+{
+	if (!unlock)
+		return;
+
+	mutex_unlock(&dev->struct_mutex);
+
+	/* expedite the RCU grace period to free some request slabs */
+	synchronize_rcu_expedited();
+}
+
 static bool any_vma_pinned(struct drm_i915_gem_object *obj)
 {
 	struct i915_vma *vma;
@@ -232,11 +243,8 @@ i915_gem_shrink(struct drm_i915_private *dev_priv,
 		intel_runtime_pm_put(dev_priv);
 
 	i915_gem_retire_requests(dev_priv);
-	if (unlock)
-		mutex_unlock(&dev_priv->drm.struct_mutex);
 
-	/* expedite the RCU grace period to free some request slabs */
-	synchronize_rcu_expedited();
+	i915_gem_shrinker_unlock(&dev_priv->drm, unlock);
 
 	return count;
 }
@@ -296,8 +304,7 @@ i915_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)
 			count += obj->base.size >> PAGE_SHIFT;
 	}
 
-	if (unlock)
-		mutex_unlock(&dev->struct_mutex);
+	i915_gem_shrinker_unlock(dev, unlock);
 
 	return count;
 }
@@ -324,8 +331,8 @@ i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 					 sc->nr_to_scan - freed,
 					 I915_SHRINK_BOUND |
 					 I915_SHRINK_UNBOUND);
-	if (unlock)
-		mutex_unlock(&dev->struct_mutex);
+
+	i915_gem_shrinker_unlock(dev, unlock);
 
 	return freed;
 }
@@ -367,8 +374,7 @@ i915_gem_shrinker_unlock_uninterruptible(struct drm_i915_private *dev_priv,
 					 struct shrinker_lock_uninterruptible *slu)
 {
 	dev_priv->mm.interruptible = slu->was_interruptible;
-	if (slu->unlock)
-		mutex_unlock(&dev_priv->drm.struct_mutex);
+	i915_gem_shrinker_unlock(&dev_priv->drm, slu->unlock);
 }
 
 static int
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author maintainer
#1 lore raw

PATCH 2/2 drm/i915: Simplify shrinker locking

By using the same structure for both interruptible and uninterruptible locking in shrinker code, combined with the information that mm.interruptible is only being written to, the code can be greatly simplified.

Also removing the i915_gem_ prefix from the locking functions so that nobody in their wildest dreams considers exporting them.

---
 drivers/gpu/drm/i915/i915_drv.h          |  6 ----
 drivers/gpu/drm/i915/i915_gem.c          |  2 --
 drivers/gpu/drm/i915/i915_gem_shrinker.c | 56 +++++++++++---------------------
 drivers/gpu/drm/i915/intel_display.c     |  3 --
 4 files changed, 19 insertions(+), 48 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +0 -6
index c9b0949..f990f0e 100644
@@ -1511,12 +1511,6 @@ struct i915_gem_mm {
 	/** LRU list of objects with fence regs on them. */
 	struct list_head fence_list;
 
-	/**
-	 * Are we in a non-interruptible section of code like
-	 * modesetting?
-	 */
-	bool interruptible;
-
 	/* the indicator for dispatch video commands on two BSD rings */
 	atomic_t bsd_engine_dispatch_index;
 
drivers/gpu/drm/i915/i915_gem.c +0 -2
index 391aa69..c8139be 100644
@@ -4822,8 +4822,6 @@ i915_gem_load_init(struct drm_i915_private *dev_priv)
 
 	init_waitqueue_head(&dev_priv->pending_flip_queue);
 
-	dev_priv->mm.interruptible = true;
-
 	atomic_set(&dev_priv->mm.bsd_engine_dispatch_index, 0);
 
 	spin_lock_init(&dev_priv->fb_tracking.lock);
drivers/gpu/drm/i915/i915_gem_shrinker.c +19 -37

165 lines not on this page · raw message

drivers/gpu/drm/i915/intel_display.c +0 -3
index 3617927..c683508 100644
@@ -4861,12 +4861,9 @@ static void intel_crtc_dpms_overlay_disable(struct intel_crtc *intel_crtc)
 {
 	if (intel_crtc->overlay) {
 		struct drm_device *dev = intel_crtc->base.dev;
-		struct drm_i915_private *dev_priv = to_i915(dev);
 
 		mutex_lock(&dev->struct_mutex);
-		dev_priv->mm.interruptible = false;
 		(void) intel_overlay_switch_off(intel_crtc->overlay);
-		dev_priv->mm.interruptible = true;
 		mutex_unlock(&dev->struct_mutex);
 	}
 
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Reviewed-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2] drm/i915: Simplify shrinker locking

Quoted 10 lines · Joonas Lahtinen

By using the same structure for both interruptible and uninterruptible locking in shrinker code, combined with the information that mm.interruptible is only being written to, the code can be greatly simplified.

Also removing the i915_gem_ prefix from the locking functions so that nobody in their wildest dreams considers exporting them.

I still feel uneasy that I'm not using mm.uninterruptible in some parts of modesetting, nevertheless the change is correct, simplies the current code and means I have to fix gen2 properly if anyone ever notices.

-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#3 lore raw

✓ Fi.CI.BAT: success for series starting with [1/2] drm/i915: Don't call synchronize_rcu_expedited under struct_mutex

In reply to Joonas Lahtinen
== Series Details ==

== Series Details ==

Series: series starting with [1/2] drm/i915: Don't call synchronize_rcu_expedited under struct_mutex
URL : https://patchwork.freedesktop.org/series/22652/
State : success

== Summary ==

Series 22652v1 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/22652/revisions/1/mbox/

Test gem_exec_flush:

        Subgroup basic-batch-kernel-default-uc:
                fail       -> PASS       (fi-snb-2600) fdo#100007

fdo#100007 https://bugs.freedesktop.org/show_bug.cgi?id=100007

fi-bdw-5557u     total:278  pass:266  dwarn:0   dfail:0   fail:1   skip:11  time:430s
fi-bdw-gvtdvm    total:278  pass:255  dwarn:8   dfail:0   fail:1   skip:14  time:424s
fi-bsw-n3050     total:278  pass:241  dwarn:0   dfail:0   fail:1   skip:36  time:577s
fi-bxt-j4205     total:278  pass:258  dwarn:0   dfail:0   fail:1   skip:19  time:506s
fi-byt-j1900     total:278  pass:253  dwarn:0   dfail:0   fail:1   skip:24  time:479s
fi-byt-n2820     total:278  pass:249  dwarn:0   dfail:0   fail:1   skip:28  time:488s
fi-hsw-4770      total:278  pass:261  dwarn:0   dfail:0   fail:1   skip:16  time:412s
fi-hsw-4770r     total:278  pass:261  dwarn:0   dfail:0   fail:1   skip:16  time:410s
fi-ilk-650       total:278  pass:228  dwarn:0   dfail:0   fail:1   skip:49  time:428s
fi-ivb-3520m     total:278  pass:259  dwarn:0   dfail:0   fail:1   skip:18  time:475s
fi-ivb-3770      total:278  pass:259  dwarn:0   dfail:0   fail:1   skip:18  time:487s
fi-kbl-7500u     total:278  pass:259  dwarn:0   dfail:0   fail:1   skip:18  time:456s
fi-kbl-7560u     total:278  pass:267  dwarn:0   dfail:0   fail:1   skip:10  time:565s
fi-skl-6260u     total:278  pass:267  dwarn:0   dfail:0   fail:1   skip:10  time:451s
fi-skl-6700hq    total:278  pass:260  dwarn:0   dfail:0   fail:1   skip:17  time:567s
fi-skl-6700k     total:278  pass:255  dwarn:4   dfail:0   fail:1   skip:18  time:467s
fi-skl-6770hq    total:278  pass:267  dwarn:0   dfail:0   fail:1   skip:10  time:486s
fi-skl-gvtdvm    total:278  pass:264  dwarn:0   dfail:0   fail:1   skip:13  time:436s
fi-snb-2520m     total:278  pass:249  dwarn:0   dfail:0   fail:1   skip:28  time:527s
fi-snb-2600      total:278  pass:248  dwarn:0   dfail:0   fail:1   skip:29  time:412s

fdc0fa0d6347cadec14396a172a98636e9dfebc0 drm-tip: 2017y-04m-07d-10h-32m-49s UTC integration manifest 08e143d drm/i915: Simplify shrinker locking
60dc19c drm/i915: Don't call synchronize_rcu_expedited under struct_mutex

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_4439/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

author maintainer
#4 lore raw

Re: ✓ Fi.CI.BAT: success for series starting with [1/2] drm/i915: Don't call synchronize_rcu_expedited under struct_mutex

Thanks for the review, pushing series.

Quoted 45 lines · Patchwork

== Series Details ==

Series: series starting with [1/2] drm/i915: Don't call synchronize_rcu_expedited under struct_mutex URL   : https://patchwork.freedesktop.org/series/22652/
State : success

== Summary ==

Series 22652v1 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/22652/revisions/1/mbox/

Test gem_exec_flush:
        Subgroup basic-batch-kernel-default-uc:
                fail       -> PASS       (fi-snb-2600) fdo#100007

fdo#100007 https://bugs.freedesktop.org/show_bug.cgi?id=100007

fi-bdw-5557u     total:278  pass:266  dwarn:0   dfail:0   fail:1   skip:11  time:430s fi-bdw-gvtdvm    total:278  pass:255  dwarn:8   dfail:0   fail:1   skip:14  time:424s fi-bsw-n3050     total:278  pass:241  dwarn:0   dfail:0   fail:1   skip:36  time:577s fi-bxt-j4205     total:278  pass:258  dwarn:0   dfail:0   fail:1   skip:19  time:506s fi-byt-j1900     total:278  pass:253  dwarn:0   dfail:0   fail:1   skip:24  time:479s fi-byt-n2820     total:278  pass:249  dwarn:0   dfail:0   fail:1   skip:28  time:488s fi-hsw-4770      total:278  pass:261  dwarn:0   dfail:0   fail:1   skip:16  time:412s fi-hsw-4770r     total:278  pass:261  dwarn:0   dfail:0   fail:1   skip:16  time:410s fi-ilk-650       total:278  pass:228  dwarn:0   dfail:0   fail:1   skip:49  time:428s fi-ivb-3520m     total:278  pass:259  dwarn:0   dfail:0   fail:1   skip:18  time:475s fi-ivb-3770      total:278  pass:259  dwarn:0   dfail:0   fail:1   skip:18  time:487s fi-kbl-7500u     total:278  pass:259  dwarn:0   dfail:0   fail:1   skip:18  time:456s fi-kbl-7560u     total:278  pass:267  dwarn:0   dfail:0   fail:1   skip:10  time:565s fi-skl-6260u     total:278  pass:267  dwarn:0   dfail:0   fail:1   skip:10  time:451s fi-skl-6700hq    total:278  pass:260  dwarn:0   dfail:0   fail:1   skip:17  time:567s fi-skl-6700k     total:278  pass:255  dwarn:4   dfail:0   fail:1   skip:18  time:467s fi-skl-6770hq    total:278  pass:267  dwarn:0   dfail:0   fail:1   skip:10  time:486s fi-skl-gvtdvm    total:278  pass:264  dwarn:0   dfail:0   fail:1   skip:13  time:436s fi-snb-2520m     total:278  pass:249  dwarn:0   dfail:0   fail:1   skip:28  time:527s fi-snb-2600      total:278  pass:248  dwarn:0   dfail:0   fail:1   skip:29  time:412s

fdc0fa0d6347cadec14396a172a98636e9dfebc0 drm-tip: 2017y-04m-07d-10h-32m-49s UTC integration manifest 08e143d drm/i915: Simplify shrinker locking
60dc19c drm/i915: Don't call synchronize_rcu_expedited under struct_mutex

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_4439/

Signature · 7 lines
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#5 lore raw
In reply to Joonas Lahtinen

Hello Joonas,

Quoted 34 lines · Joonas Lahtinen
drivers/gpu/drm/i915/i915_gem_shrinker.c +12 -4
index 2978acd..129ed30 100644
@@ -53,6 +53,17 @@ static bool i915_gem_shrinker_lock(struct drm_device *dev, bool *unlock)
 	BUG();
 }
 
+static void i915_gem_shrinker_unlock(struct drm_device *dev, bool unlock)
+{
+	if (!unlock)
+		return;
+
+	mutex_unlock(&dev->struct_mutex);
+
+	/* expedite the RCU grace period to free some request slabs */
+	synchronize_rcu_expedited();
+}
+
 static bool any_vma_pinned(struct drm_i915_gem_object *obj)
 {
 	struct i915_vma *vma;
@@ -232,11 +243,8 @@ i915_gem_shrink(struct drm_i915_private *dev_priv,
 		intel_runtime_pm_put(dev_priv);
 
 	i915_gem_retire_requests(dev_priv);
-	if (unlock)
-		mutex_unlock(&dev_priv->drm.struct_mutex);
 
-	/* expedite the RCU grace period to free some request slabs */
-	synchronize_rcu_expedited();
+	i915_gem_shrinker_unlock(&dev_priv->drm, unlock);
 }

drivers/gpu/drm/i915/i915_gem_shrinker.c i915_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc) · line 307

Quoted 6 lines
drivers/gpu/drm/i915/i915_gem_shrinker.c +1 -2
@@ -296,8 +304,7 @@ i915_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)
 			count += obj->base.size >> PAGE_SHIFT;
 	}
 
-	if (unlock)
-		mutex_unlock(&dev->struct_mutex);
+	i915_gem_shrinker_unlock(dev, unlock);

Why here? This doesn't make sense, all it matters is the throttling to happen when scan_objects is run, if count_objects is run it's not worth it to wait a quiescent point and to call synchronize_rcu_expedited(), it is *very* expensive. I recommend to reverse this hunk, I think it's worsening the runtime with zero benefits to increase the reliability of reclaim.

drivers/gpu/drm/i915/i915_gem_shrinker.c i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc) · line 338

Quoted 13 lines
drivers/gpu/drm/i915/i915_gem_shrinker.c +2 -2
 	return count;
 }
@@ -324,8 +331,8 @@ i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 					 sc->nr_to_scan - freed,
 					 I915_SHRINK_BOUND |
 					 I915_SHRINK_UNBOUND);
-	if (unlock)
-		mutex_unlock(&dev->struct_mutex);
+
+	i915_gem_shrinker_unlock(dev, unlock);
 
 	return freed;
 }

Perfect here, faster than re-reading the mutex too, already thought of checking unlock instead. Although if that's the only place it can as well be explicit.

drivers/gpu/drm/i915/i915_gem_shrinker.c i915_gem_shrinker_unlock_uninterruptible(struct drm_i915_private *dev_priv, · line 378

Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_shrinker.c +1 -2
@@ -367,8 +374,7 @@ i915_gem_shrinker_unlock_uninterruptible(struct drm_i915_private *dev_priv,
 					 struct shrinker_lock_uninterruptible *slu)
 {
 	dev_priv->mm.interruptible = slu->was_interruptible;
-	if (slu->unlock)
-		mutex_unlock(&dev_priv->drm.struct_mutex);
+	i915_gem_shrinker_unlock(&dev_priv->drm, slu->unlock);
 }

i915_gem_shrinker_unlock_uninterruptible is more of a helper function but it doesn't always make sense to wait a rcu grace period. I think it'd be better to be selective in when to explicitly run synchronize_rcu_expedited(); in fact in some case synchronize_sched() may be prefereable and it will cause less wasted CPU cycles at the only downside of higher latency, it's all about tradeoffs which one should be called as they're equivalent as far as i915 is concerned. I don't think calling synchronize_rcu_expedited(); unconditionally in a unlock helper is ok and it'd be better to decided if to call (and which one) on a case by case basis.

I kind I prefer my patch, just cleaned up with the synchronize_rcu_expedited under if (unlock) { mutex_unlock; synchronize_rcu... }.

I'd also like to see all those mutex_trylock_recursive dropped, the only place where it makes sense to use it really is i915_gem_shrinker_scan and i915_gem_shrinker_count, unless in fact we want to replace it there too with a mutex_trylock and stop the recursive behavior of reclaim into DRM code. The other cases where the code uses lock recursion don't make much sense to me, I think the code would be simpler if the information on the struct_mutex being hold would be passed as parameter in all other cases. It'd be likely faster as well for the same reason why checking "unlock" is better above than checking the mutex.
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#6 lore raw
Quoted 4 lines · Andrea Arcangeli

Hello Joonas,
I kind I prefer my patch, just cleaned up with the synchronize_rcu_expedited under if (unlock) { mutex_unlock; synchronize_rcu... }.

Here a cleaned up version below using "unlock" instead of checking the mutex again (otherwise equivalent).

From 930f1779c1430062852f7796e9c555db8e795319 Mon Sep 17 00:00:00 2001
From: Andrea Arcangeli <[email protected]>
Date: Fri, 7 Apr 2017 16:15:55 +0200
Subject: [PATCH 1/1] drm/i915: Don't call synchronize_rcu_expedited under
struct_mutex

synchronize_rcu/synchronize_sched/synchronize_rcu_expedited() will hang until its own workqueues are run. The i915 gem workqueues will wait on the struct_mutex to be released. So we cannot wait for a quiescent state using those rcu primitives while holding the struct_mutex or it creates a circular lock dependency resulting in kernel hangs (which is reproducible but goes undetected by lockdep).

This started in commit 3d3d18f086cdda72ee18a454db70ca72c6e3246c and lockdep didn't detect it apparently.

kswapd0 D 0 700 2 0x00000000

Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? _synchronize_rcu_expedited.constprop.65+0x2ef/0x300
? wake_up_bit+0x20/0x20
? rcu_stall_kick_kthreads.part.54+0xc0/0xc0
? rcu_exp_wait_wake+0x530/0x530
? i915_gem_shrink+0x34b/0x4b0
? i915_gem_shrinker_scan+0x7c/0x90
? i915_gem_shrinker_scan+0x7c/0x90
? shrink_slab.part.61.constprop.72+0x1c1/0x3a0
? shrink_zone+0x154/0x160
? kswapd+0x40a/0x720
? kthread+0xf4/0x130
? try_to_free_pages+0x450/0x450
? kthread_create_on_node+0x40/0x40
? ret_from_fork+0x23/0x30
plasmashell     D    0  4657   4614 0x00000000
Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? schedule_preempt_disabled+0xe/0x10
? __mutex_lock.isra.4+0x1c9/0x790
? i915_gem_close_object+0x26/0xc0
? i915_gem_close_object+0x26/0xc0
? drm_gem_object_release_handle+0x48/0x90
? drm_gem_handle_delete+0x50/0x80
? drm_ioctl+0x1fa/0x420
? drm_gem_handle_create+0x40/0x40
? pipe_write+0x391/0x410
? __vfs_write+0xc6/0x120
? do_vfs_ioctl+0x8b/0x5d0
? SyS_ioctl+0x3b/0x70
? entry_SYSCALL_64_fastpath+0x13/0x94
kworker/0:0     D    0 29186      2 0x00000000
Workqueue: events __i915_gem_free_work
Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? schedule_preempt_disabled+0xe/0x10
? __mutex_lock.isra.4+0x1c9/0x790
? del_timer_sync+0x44/0x50
? update_curr+0x57/0x110
? __i915_gem_free_objects+0x31/0x300
? __i915_gem_free_objects+0x31/0x300
? __i915_gem_free_work+0x2d/0x40
? process_one_work+0x13a/0x3b0
? worker_thread+0x4a/0x460
? kthread+0xf4/0x130
? process_one_work+0x3b0/0x3b0
? kthread_create_on_node+0x40/0x40
? ret_from_fork+0x23/0x30
---
 drivers/gpu/drm/i915/i915_gem.c          |  9 +++++++++
 drivers/gpu/drm/i915/i915_gem_shrinker.c | 15 ++++++++++-----
 2 files changed, 19 insertions(+), 5 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +9 -0
index 67b1fc5..3982489 100644
@@ -4742,6 +4742,13 @@ int i915_gem_freeze(struct drm_i915_private *dev_priv)
 	i915_gem_shrink_all(dev_priv);
 	mutex_unlock(&dev_priv->drm.struct_mutex);
 
+	/*
+	 * Cannot call synchronize_rcu() inside the struct_mutex
+	 * because it may block until workqueues complete, and the
+	 * running workqueue may wait on the struct_mutex.
+	 */
+	synchronize_rcu(); /* wait for our earlier RCU delayed slab frees */
+
 	intel_runtime_pm_put(dev_priv);
 
 	return 0;
@@ -4781,6 +4788,8 @@ int i915_gem_freeze_late(struct drm_i915_private *dev_priv)
 	}
 	mutex_unlock(&dev_priv->drm.struct_mutex);
 
+	synchronize_rcu_expedited();
+
 	return 0;
 }
 
drivers/gpu/drm/i915/i915_gem_shrinker.c +10 -5
index d5d2b4c..4636e52 100644
@@ -235,9 +235,6 @@ i915_gem_shrink(struct drm_i915_private *dev_priv,
 	if (unlock)
 		mutex_unlock(&dev_priv->drm.struct_mutex);
 
-	/* expedite the RCU grace period to free some request slabs */
-	synchronize_rcu_expedited();
-
 	return count;
 }
 
@@ -263,7 +260,6 @@ unsigned long i915_gem_shrink_all(struct drm_i915_private *dev_priv)
 				I915_SHRINK_BOUND |
 				I915_SHRINK_UNBOUND |
 				I915_SHRINK_ACTIVE);
-	synchronize_rcu(); /* wait for our earlier RCU delayed slab frees */
 
 	return freed;
 }
@@ -321,8 +317,17 @@ i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 					 sc->nr_to_scan - freed,
 					 I915_SHRINK_BOUND |
 					 I915_SHRINK_UNBOUND);
-	if (unlock)
+	if (unlock) {
 		mutex_unlock(&dev->struct_mutex);
+		/*
+		 * If reclaim was invoked by an allocation done while
+		 * holding the struct mutex, we cannot call
+		 * synchronize_rcu_expedited() as it depends on
+		 * workqueues to run but the running workqueue may be
+		 * blocked waiting on us to release struct_mutex.
+		 */
+		synchronize_rcu_expedited();
+	}
 
 	return freed;
 }

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx