[Intel-gfx] [CI 1/2] drm/i915: Avoiding recursing on ww_mutex inside shrinker

[Chris Wilson]

We have to avoid taking ww_mutex inside the shrinker as we use it as a
plain mutex type and so need to avoid recursive deadlocks:

[  602.771969] =================================
[  602.771970] [ INFO: inconsistent lock state ]
[  602.771973] 4.10.0gpudebug+ #122 Not tainted
[  602.771974] ---------------------------------
[  602.771975] inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-W} usage.
[  602.771978] kswapd0/40 [HC0[0]:SC0[0]:HE1:SE1] takes:
[  602.771979]  (reservation_ww_class_mutex){+.+.?.}, at: [<ffffffffa054680a>] i915_gem_object_wait+0x39a/0x410 [i915]
[  602.772020] {RECLAIM_FS-ON-W} state was registered at:
[  602.772024]   mark_held_locks+0x76/0x90
[  602.772026]   lockdep_trace_alloc+0xb8/0xc0
[  602.772028]   __kmalloc_track_caller+0x5d/0x130
[  602.772031]   krealloc+0x89/0xb0
[  602.772033]   reservation_object_reserve_shared+0xaf/0xd0
[  602.772055]   i915_gem_do_execbuffer.isra.35+0x1413/0x18b0 [i915]
[  602.772075]   i915_gem_execbuffer2+0x10e/0x1d0 [i915]
[  602.772078]   drm_ioctl+0x291/0x480
[  602.772079]   do_vfs_ioctl+0x695/0x6f0
[  602.772081]   SyS_ioctl+0x3c/0x70
[  602.772084]   entry_SYSCALL_64_fastpath+0x18/0xad
[  602.772085] irq event stamp: 5197423
[  602.772088] hardirqs last  enabled at (5197423): [<ffffffff8116751d>] kfree+0xdd/0x170
[  602.772091] hardirqs last disabled at (5197422): [<ffffffff811674f9>] kfree+0xb9/0x170
[  602.772095] softirqs last  enabled at (5190992): [<ffffffff8107bfe1>] __do_softirq+0x221/0x280
[  602.772097] softirqs last disabled at (5190575): [<ffffffff8107c294>] irq_exit+0x64/0xc0
[  602.772099]
               other info that might help us debug this:
[  602.772100]  Possible unsafe locking scenario:

[  602.772101]        CPU0
[  602.772101]        ----
[  602.772102]   lock(reservation_ww_class_mutex);
[  602.772104]   <Interrupt>
[  602.772105]     lock(reservation_ww_class_mutex);
[  602.772107]
                *** DEADLOCK ***

[  602.772109] 2 locks held by kswapd0/40:
[  602.772110]  #0:  (shrinker_rwsem){++++..}, at: [<ffffffff811337b5>] shrink_slab.constprop.62+0x35/0x280
[  602.772116]  #1:  (&dev->struct_mutex){+.+.+.}, at: [<ffffffffa0553957>] i915_gem_shrinker_lock+0x27/0x60 [i915]
[  602.772141]
               stack backtrace:
[  602.772144] CPU: 2 PID: 40 Comm: kswapd0 Not tainted 4.10.0gpudebug+ #122
[  602.772145] Hardware name: LENOVO 42433ZG/42433ZG, BIOS 8AET64WW (1.44 ) 07/26/2013
[  602.772147] Call Trace:
[  602.772151]  dump_stack+0x68/0xa1
[  602.772153]  print_usage_bug+0x1d4/0x1f0
[  602.772155]  mark_lock+0x390/0x530
[  602.772157]  ? print_irq_inversion_bug+0x200/0x200
[  602.772159]  __lock_acquire+0x405/0x1260
[  602.772181]  ? i915_gem_object_wait+0x39a/0x410 [i915]
[  602.772183]  lock_acquire+0x60/0x80
[  602.772205]  ? i915_gem_object_wait+0x39a/0x410 [i915]
[  602.772207]  mutex_lock_nested+0x69/0x760
[  602.772229]  ? i915_gem_object_wait+0x39a/0x410 [i915]
[  602.772231]  ? kfree+0xdd/0x170
[  602.772253]  ? i915_gem_object_wait+0x163/0x410 [i915]
[  602.772255]  ? trace_hardirqs_on_caller+0x18d/0x1c0
[  602.772256]  ? trace_hardirqs_on+0xd/0x10
[  602.772278]  i915_gem_object_wait+0x39a/0x410 [i915]
[  602.772300]  i915_gem_object_unbind+0x5e/0x130 [i915]
[  602.772323]  i915_gem_shrink+0x22d/0x3d0 [i915]
[  602.772347]  i915_gem_shrinker_scan+0x3f/0x80 [i915]
[  602.772349]  shrink_slab.constprop.62+0x1ad/0x280
[  602.772352]  shrink_node+0x52/0x80
[  602.772355]  kswapd+0x427/0x5c0
[  602.772358]  kthread+0x122/0x130
[  602.772360]  ? try_to_free_pages+0x270/0x270
[  602.772362]  ? kthread_stop+0x70/0x70
[  602.772365]  ret_from_fork+0x2e/0x40

v2: Add commentary about the pruning being opportunistic

Reported-by: Jan Nordholz <jckn at gmx.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99977#c10
Fixes: e54ca9774777 ("drm/i915: Remove completed fences after a wait")
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Cc: Matthew Auld <matthew.auld at intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index e29f9400c9d1..aca1eaddafb4 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -462,11 +462,16 @@ i915_gem_object_wait_reservation(struct reservation_object *resv,
 
 	dma_fence_put(excl);
 
+	/* Oportunistically prune the fences iff we know they have *all* been
+	 * signaled and that the reservation object has not been changed (i.e.
+	 * no new fences have been added).
+	 */
 	if (prune_fences && !__read_seqcount_retry(&resv->seq, seq)) {
-		reservation_object_lock(resv, NULL);
-		if (!__read_seqcount_retry(&resv->seq, seq))
-			reservation_object_add_excl_fence(resv, NULL);
-		reservation_object_unlock(resv);
+		if (reservation_object_trylock(resv)) {
+			if (!__read_seqcount_retry(&resv->seq, seq))
+				reservation_object_add_excl_fence(resv, NULL);
+			reservation_object_unlock(resv);
+		}
 	}
 
 	return timeout;
-- 
2.11.0