This is a note to let you know that I've just added the patch titled
drm/amdgpu/vkms: relax timer deactivation by hrtimer_try_to_cancel
to the 6.4-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
drm-amdgpu-vkms-relax-timer-deactivation-by-hrtimer_try_to_cancel.patch
and it can be found in the queue-6.4 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
>From b42ae87a7b3878afaf4c3852ca66c025a5b996e0 Mon Sep 17 00:00:00 2001
From: Guchun Chen <guchun.chen@xxxxxxx>
Date: Thu, 6 Jul 2023 15:57:21 +0800
Subject: drm/amdgpu/vkms: relax timer deactivation by hrtimer_try_to_cancel
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
From: Guchun Chen <guchun.chen@xxxxxxx>
commit b42ae87a7b3878afaf4c3852ca66c025a5b996e0 upstream.
In below thousands of screen rotation loop tests with virtual display
enabled, a CPU hard lockup issue may happen, leading system to unresponsive
and crash.
do {
xrandr --output Virtual --rotate inverted
xrandr --output Virtual --rotate right
xrandr --output Virtual --rotate left
xrandr --output Virtual --rotate normal
} while (1);
NMI watchdog: Watchdog detected hard LOCKUP on cpu 1
? hrtimer_run_softirq+0x140/0x140
? store_vblank+0xe0/0xe0 [drm]
hrtimer_cancel+0x15/0x30
amdgpu_vkms_disable_vblank+0x15/0x30 [amdgpu]
drm_vblank_disable_and_save+0x185/0x1f0 [drm]
drm_crtc_vblank_off+0x159/0x4c0 [drm]
? record_print_text.cold+0x11/0x11
? wait_for_completion_timeout+0x232/0x280
? drm_crtc_wait_one_vblank+0x40/0x40 [drm]
? bit_wait_io_timeout+0xe0/0xe0
? wait_for_completion_interruptible+0x1d7/0x320
? mutex_unlock+0x81/0xd0
amdgpu_vkms_crtc_atomic_disable
It's caused by a stuck in lock dependency in such scenario on different
CPUs.
CPU1 CPU2
drm_crtc_vblank_off hrtimer_interrupt
grab event_lock (irq disabled) __hrtimer_run_queues
grab vbl_lock/vblank_time_block amdgpu_vkms_vblank_simulate
amdgpu_vkms_disable_vblank drm_handle_vblank
hrtimer_cancel grab dev->event_lock
So CPU1 stucks in hrtimer_cancel as timer callback is running endless on
current clock base, as that timer queue on CPU2 has no chance to finish it
because of failing to hold the lock. So NMI watchdog will throw the errors
after its threshold, and all later CPUs are impacted/blocked.
So use hrtimer_try_to_cancel to fix this, as disable_vblank callback
does not need to wait the handler to finish. And also it's not necessary
to check the return value of hrtimer_try_to_cancel, because even if it's
-1 which means current timer callback is running, it will be reprogrammed
in hrtimer_start with calling enable_vblank to make it works.
v2: only re-arm timer when vblank is enabled (Christian) and add a Fixes
tag as well
v3: drop warn printing (Christian)
v4: drop superfluous check of blank->enabled in timer function, as it's
guaranteed in drm_handle_vblank (Christian)
Fixes: 84ec374bd580 ("drm/amdgpu: create amdgpu_vkms (v4)")
Cc: stable@xxxxxxxxxxxxxxx
Suggested-by: Christian König <christian.koenig@xxxxxxx>
Signed-off-by: Guchun Chen <guchun.chen@xxxxxxx>
Reviewed-by: Christian König <christian.koenig@xxxxxxx>
Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vkms.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vkms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vkms.c
@@ -55,8 +55,9 @@ static enum hrtimer_restart amdgpu_vkms_
DRM_WARN("%s: vblank timer overrun\n", __func__);
ret = drm_crtc_handle_vblank(crtc);
+ /* Don't queue timer again when vblank is disabled. */
if (!ret)
- DRM_ERROR("amdgpu_vkms failure on handling vblank");
+ return HRTIMER_NORESTART;
return HRTIMER_RESTART;
}
@@ -81,7 +82,7 @@ static void amdgpu_vkms_disable_vblank(s
{
struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
- hrtimer_cancel(&amdgpu_crtc->vblank_timer);
+ hrtimer_try_to_cancel(&amdgpu_crtc->vblank_timer);
}
static bool amdgpu_vkms_get_vblank_timestamp(struct drm_crtc *crtc,
Patches currently in stable-queue which might be from guchun.chen@xxxxxxx are
queue-6.4/drm-amdgpu-vkms-relax-timer-deactivation-by-hrtimer_try_to_cancel.patch
