[PATCH] drm/amdgpu: hold the reference of finished fence
Tao, Yintian
Yintian.Tao at amd.com
Mon Mar 23 12:26:28 UTC 2020
Hi Christian
The error fence is finished fence not the hw ring fence. Please the call trace below.
[ 732.920906] RIP: 0010:dma_fence_signal_locked+0x3e/0x100
[ 732.939364] dma_fence_signal+0x29/0x50 ===>drm sched finished fence
[ 732.940036] drm_sched_fence_finished+0x12/0x20 [gpu_sched]
[ 732.940996] drm_sched_process_job+0x34/0xa0 [gpu_sched]
[ 732.941910] dma_fence_signal_locked+0x85/0x100
[ 732.942692] dma_fence_signal+0x29/0x50 ====> hw fence
[ 732.943457] amdgpu_fence_process+0x99/0x120 [amdgpu]
[ 732.944393] sdma_v4_0_process_trap_irq+0x81/0xa0 [amdgpu]
[ 732.945398] amdgpu_irq_dispatch+0xaf/0x1d0 [amdgpu]
[ 732.946317] amdgpu_ih_process+0x8c/0x110 [amdgpu]
[ 732.947206] amdgpu_irq_handler+0x24/0xa0 [amdgpu]
Best Regards
Yintian Tao
-----Original Message-----
From: Koenig, Christian <Christian.Koenig at amd.com>
Sent: 2020年3月23日 20:06
To: Tao, Yintian <Yintian.Tao at amd.com>; Deucher, Alexander <Alexander.Deucher at amd.com>
Cc: amd-gfx at lists.freedesktop.org
Subject: Re: [PATCH] drm/amdgpu: hold the reference of finished fence
I've just double checked and your analyses actually can't be correct.
When we call dma_fence_signal() in amdgpu_fence_process() we still have a reference to the fence.
See the code here:
> r = dma_fence_signal(fence);
> if (!r)
> DMA_FENCE_TRACE(fence, "signaled from irq
> context\n");
> else
> BUG();
>
> dma_fence_put(fence);
So I'm not sure how you ran into the crash in the first place, this is most likely something else.
Regards,
Christian.
Am 23.03.20 um 12:49 schrieb Yintian Tao:
> There is one one corner case at dma_fence_signal_locked which will
> raise the NULL pointer problem just like below.
> ->dma_fence_signal
> ->dma_fence_signal_locked
> ->test_and_set_bit
> here trigger dma_fence_release happen due to the zero of fence refcount.
>
> ->dma_fence_put
> ->dma_fence_release
> ->drm_sched_fence_release_scheduled
> ->call_rcu
> here make the union fled “cb_list” at finished fence to NULL because
> struct rcu_head contains two pointer which is same as struct list_head
> cb_list
>
> Therefore, to hold the reference of finished fence at amdgpu_job_run
> to prevent the null pointer during dma_fence_signal
>
> [ 732.912867] BUG: kernel NULL pointer dereference, address:
> 0000000000000008 [ 732.914815] #PF: supervisor write access in kernel
> mode [ 732.915731] #PF: error_code(0x0002) - not-present page [
> 732.916621] PGD 0 P4D 0 [ 732.917072] Oops: 0002 [#1] SMP PTI
> [ 732.917682] CPU: 7 PID: 0 Comm: swapper/7 Tainted: G OE 5.4.0-rc7 #1
> [ 732.918980] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
> BIOS rel-1.8.2-0-g33fbe13 by qemu-project.org 04/01/2014 [
> 732.920906] RIP: 0010:dma_fence_signal_locked+0x3e/0x100
> [ 732.938569] Call Trace:
> [ 732.939003] <IRQ>
> [ 732.939364] dma_fence_signal+0x29/0x50 [ 732.940036]
> drm_sched_fence_finished+0x12/0x20 [gpu_sched] [ 732.940996]
> drm_sched_process_job+0x34/0xa0 [gpu_sched] [ 732.941910]
> dma_fence_signal_locked+0x85/0x100
> [ 732.942692] dma_fence_signal+0x29/0x50 [ 732.943457]
> amdgpu_fence_process+0x99/0x120 [amdgpu] [ 732.944393]
> sdma_v4_0_process_trap_irq+0x81/0xa0 [amdgpu]
>
> Signed-off-by: Yintian Tao <yttao at amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 19 ++++++++++++++++++-
> drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 2 ++
> drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 3 +++
> 3 files changed, 23 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> index 7531527067df..03573eff660a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> @@ -52,7 +52,7 @@
>
> struct amdgpu_fence {
> struct dma_fence base;
> -
> + struct dma_fence *finished;
> /* RB, DMA, etc. */
> struct amdgpu_ring *ring;
> };
> @@ -149,6 +149,7 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring,
> struct dma_fence **f,
>
> seq = ++ring->fence_drv.sync_seq;
> fence->ring = ring;
> + fence->finished = NULL;
> dma_fence_init(&fence->base, &amdgpu_fence_ops,
> &ring->fence_drv.lock,
> adev->fence_context + ring->idx, @@ -182,6 +183,21 @@ int
> amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f,
> return 0;
> }
>
> +void amdgpu_fence_get_finished(struct dma_fence *base,
> + struct dma_fence *finished) {
> + struct amdgpu_fence *afence = to_amdgpu_fence(base);
> +
> + afence->finished = dma_fence_get(finished); }
> +
> +void amdgpu_fence_put_finished(struct dma_fence *base) {
> + struct amdgpu_fence *afence = to_amdgpu_fence(base);
> +
> + dma_fence_put(afence->finished);
> +}
> +
> /**
> * amdgpu_fence_emit_polling - emit a fence on the requeste ring
> *
> @@ -276,6 +292,7 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring)
> BUG();
>
> dma_fence_put(fence);
> + amdgpu_fence_put_finished(fence);
> pm_runtime_mark_last_busy(adev->ddev->dev);
> pm_runtime_put_autosuspend(adev->ddev->dev);
> } while (last_seq != seq);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> index 4981e443a884..deb2aeeadfb3 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> @@ -229,6 +229,8 @@ static struct dma_fence *amdgpu_job_run(struct drm_sched_job *sched_job)
> &fence);
> if (r)
> DRM_ERROR("Error scheduling IBs (%d)\n", r);
> + else
> + amdgpu_fence_get_finished(fence, finished);
> }
> /* if gpu reset, hw fence will be replaced here */
> dma_fence_put(job->fence);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> index 448c76cbf3ed..fd4da91859aa 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> @@ -96,6 +96,9 @@ void amdgpu_fence_driver_suspend(struct amdgpu_device *adev);
> void amdgpu_fence_driver_resume(struct amdgpu_device *adev);
> int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence,
> unsigned flags);
> +void amdgpu_fence_get_finished(struct dma_fence *base,
> + struct dma_fence *finished); void
> +amdgpu_fence_put_finished(struct dma_fence *base);
> int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s);
> bool amdgpu_fence_process(struct amdgpu_ring *ring);
> int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
More information about the amd-gfx
mailing list