From: "Christian König" <christian.koenig@amd.com>
To: jbmoore <jbmoore61@gmail.com>, alexander.deucher@amd.com
Cc: stable@vger.kernel.org
Subject: Re: [PATCH v4 4/4] drm/amdgpu: do not pass AMDGPU_FENCE_FLAG_64BIT to media rings
Date: Mon, 27 Apr 2026 20:03:58 +0200 [thread overview]
Message-ID: <1ef2ea5e-9da0-404e-8cec-16d75ef1daa9@amd.com> (raw)
In-Reply-To: <20260427163024.13512-1-jbmoore@nooks.dev>
On 4/27/26 18:30, jbmoore wrote:
> From: "John B. Moore" <jbmoore61@gmail.com>
>
> amdgpu_ib_schedule() unconditionally ORs AMDGPU_FENCE_FLAG_64BIT into
> the flags when emitting the user fence for every ring type:
>
> amdgpu_ring_emit_fence(ring, job->uf_addr, job->uf_sequence,
> fence_flags | AMDGPU_FENCE_FLAG_64BIT);
>
> VCN, UVD, VCE, and JPEG encoder/decoder rings only support 32-bit
> fence values. Their emit_fence callbacks contain bare WARN_ON()
> assertions for this flag, but the flag should never reach them in
> the first place.
>
> The VCN_ENC_CMD_FENCE hardware packet writes a single 32-bit
> sequence value to a 64-bit GPU address. There is no 64-bit fence
> variant in the VCN/UVD/VCE/JPEG command sets.
>
> Filter AMDGPU_FENCE_FLAG_64BIT at the call site in
> amdgpu_ib_schedule(), only setting it for ring types whose hardware
> supports 64-bit fence writes: GFX, compute, SDMA, KIQ, MES, and VPE.
>
> Also convert the bare WARN_ON() guards in the five affected VCN
> callbacks to WARN_ON_ONCE() to prevent kernel log flooding if
> the condition is somehow triggered via another path.
>
> Found by a custom amdgpu DRM ioctl fuzzer.
Well again clear NAK.
First of all this shouldn't be handled by the IB scheduler code and second that is already fixed upstream.
Regards,
Christian.
>
> Fixes: c660f40b1ef3 ("drm/amdgpu: fix user fence write race condition")
> Signed-off-by: John B. Moore <jbmoore61@gmail.com>
> Cc: stable@vger.kernel.org
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 18 +++++++++++++++++-
> drivers/gpu/drm/amd/amdgpu/vcn_sw_ring.c | 2 +-
> drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 4 ++--
> drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 4 ++--
> 4 files changed, 22 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
> index f1ed4a436..3c32a6197 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
> @@ -297,8 +297,24 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned int num_ibs,
>
> /* wrap the last IB with fence */
> if (job && job->uf_addr) {
> + unsigned int uf_flags = fence_flags;
> +
> + /*
> + * Only request 64-bit fence writes on rings whose hardware
> + * supports them. VCN/UVD/VCE/JPEG rings only support 32-bit
> + * fence values; passing AMDGPU_FENCE_FLAG_64BIT causes their
> + * emit_fence callbacks to WARN and emit a truncated fence.
> + */
> + if (ring->funcs->type == AMDGPU_RING_TYPE_GFX ||
> + ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ||
> + ring->funcs->type == AMDGPU_RING_TYPE_SDMA ||
> + ring->funcs->type == AMDGPU_RING_TYPE_KIQ ||
> + ring->funcs->type == AMDGPU_RING_TYPE_MES ||
> + ring->funcs->type == AMDGPU_RING_TYPE_VPE)
> + uf_flags |= AMDGPU_FENCE_FLAG_64BIT;
> +
> amdgpu_ring_emit_fence(ring, job->uf_addr, job->uf_sequence,
> - fence_flags | AMDGPU_FENCE_FLAG_64BIT);
> + uf_flags);
> }
>
> if (ring->funcs->emit_gfx_shadow && ring->funcs->init_cond_exec &&
> diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_sw_ring.c b/drivers/gpu/drm/amd/amdgpu/vcn_sw_ring.c
> index 2b9ddb3d2..9adc7607c 100644
> --- a/drivers/gpu/drm/amd/amdgpu/vcn_sw_ring.c
> +++ b/drivers/gpu/drm/amd/amdgpu/vcn_sw_ring.c
> @@ -27,7 +27,7 @@
> void vcn_dec_sw_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
> u64 seq, uint32_t flags)
> {
> - WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
> + WARN_ON_ONCE(flags & AMDGPU_FENCE_FLAG_64BIT);
>
> amdgpu_ring_write(ring, VCN_DEC_SW_CMD_FENCE);
> amdgpu_ring_write(ring, addr);
> diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
> index e9d790914..729c1c378 100644
> --- a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
> @@ -1548,7 +1548,7 @@ static void vcn_v1_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64
> {
> struct amdgpu_device *adev = ring->adev;
>
> - WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
> + WARN_ON_ONCE(flags & AMDGPU_FENCE_FLAG_64BIT);
>
> amdgpu_ring_write(ring,
> PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0));
> @@ -1724,7 +1724,7 @@ static void vcn_v1_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
> static void vcn_v1_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
> u64 seq, unsigned flags)
> {
> - WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
> + WARN_ON_ONCE(flags & AMDGPU_FENCE_FLAG_64BIT);
>
> amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE);
> amdgpu_ring_write(ring, addr);
> diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c
> index e35fae9cd..a020140fb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c
> @@ -1537,7 +1537,7 @@ void vcn_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
> {
> struct amdgpu_device *adev = ring->adev;
>
> - WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
> + WARN_ON_ONCE(flags & AMDGPU_FENCE_FLAG_64BIT);
> amdgpu_ring_write(ring, PACKET0(adev->vcn.inst[ring->me].internal.context_id, 0));
> amdgpu_ring_write(ring, seq);
>
> @@ -1722,7 +1722,7 @@ static void vcn_v2_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
> void vcn_v2_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
> u64 seq, unsigned flags)
> {
> - WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
> + WARN_ON_ONCE(flags & AMDGPU_FENCE_FLAG_64BIT);
>
> amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE);
> amdgpu_ring_write(ring, addr);
next prev parent reply other threads:[~2026-04-27 18:04 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-04-27 16:30 [PATCH v4 4/4] drm/amdgpu: do not pass AMDGPU_FENCE_FLAG_64BIT to media rings jbmoore
2026-04-27 18:03 ` Christian König [this message]
-- strict thread matches above, loose matches on Subject: below --
2026-04-27 16:32 John B. Moore
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1ef2ea5e-9da0-404e-8cec-16d75ef1daa9@amd.com \
--to=christian.koenig@amd.com \
--cc=alexander.deucher@amd.com \
--cc=jbmoore61@gmail.com \
--cc=stable@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox