* [PATCH V1] accel/amdxdna: Require carveout when PASID and force_iova are disabled
@ 2026-06-04 19:54 Lizhi Hou
2026-06-04 19:56 ` Mario Limonciello
2026-06-04 20:10 ` sashiko-bot
0 siblings, 2 replies; 4+ messages in thread
From: Lizhi Hou @ 2026-06-04 19:54 UTC (permalink / raw)
To: ogabbay, quic_jhugo, dri-devel, mario.limonciello,
karol.wachowski
Cc: Lizhi Hou, linux-kernel, max.zhen, sonal.santan
When both PASID and force_iova are disabled, carveout memory should be
used. Reject buffer allocations that cannot use carveout memory in this
configuration and return an error.
Fixes: 3cc5d7a59519 ("accel/amdxdna: Add carveout memory support for non-IOMMU systems")
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
---
drivers/accel/amdxdna/amdxdna_pci_drv.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c b/drivers/accel/amdxdna/amdxdna_pci_drv.c
index c677293c1ae7..1f066ed8d236 100644
--- a/drivers/accel/amdxdna/amdxdna_pci_drv.c
+++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c
@@ -120,8 +120,14 @@ static int amdxdna_drm_open(struct drm_device *ddev, struct drm_file *filp)
if (!amdxdna_iova_on(xdna)) {
/* No need to fail open since user may use pa + carveout later. */
- if (amdxdna_sva_init(client))
+ if (amdxdna_sva_init(client)) {
XDNA_WARN(xdna, "PASID not available for pid %d", client->pid);
+ if (!amdxdna_use_carveout(xdna)) {
+ XDNA_ERR(xdna, "PASID unavailable and carveout not configured");
+ kfree(client);
+ return -EINVAL;
+ }
+ }
}
mmgrab(client->mm);
init_srcu_struct(&client->hwctx_srcu);
--
2.34.1
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH V1] accel/amdxdna: Require carveout when PASID and force_iova are disabled
2026-06-04 19:54 [PATCH V1] accel/amdxdna: Require carveout when PASID and force_iova are disabled Lizhi Hou
@ 2026-06-04 19:56 ` Mario Limonciello
2026-06-05 16:22 ` Lizhi Hou
2026-06-04 20:10 ` sashiko-bot
1 sibling, 1 reply; 4+ messages in thread
From: Mario Limonciello @ 2026-06-04 19:56 UTC (permalink / raw)
To: Lizhi Hou, ogabbay, quic_jhugo, dri-devel, karol.wachowski
Cc: linux-kernel, max.zhen, sonal.santan
On 6/4/26 14:54, Lizhi Hou wrote:
> When both PASID and force_iova are disabled, carveout memory should be
> used. Reject buffer allocations that cannot use carveout memory in this
> configuration and return an error.
>
> Fixes: 3cc5d7a59519 ("accel/amdxdna: Add carveout memory support for non-IOMMU systems")
> Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
> ---
> drivers/accel/amdxdna/amdxdna_pci_drv.c | 8 +++++++-
> 1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c b/drivers/accel/amdxdna/amdxdna_pci_drv.c
> index c677293c1ae7..1f066ed8d236 100644
> --- a/drivers/accel/amdxdna/amdxdna_pci_drv.c
> +++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c
> @@ -120,8 +120,14 @@ static int amdxdna_drm_open(struct drm_device *ddev, struct drm_file *filp)
>
> if (!amdxdna_iova_on(xdna)) {
> /* No need to fail open since user may use pa + carveout later. */
> - if (amdxdna_sva_init(client))
> + if (amdxdna_sva_init(client)) {
> XDNA_WARN(xdna, "PASID not available for pid %d", client->pid);
> + if (!amdxdna_use_carveout(xdna)) {
> + XDNA_ERR(xdna, "PASID unavailable and carveout not configured");
> + kfree(client);
> + return -EINVAL;
> + }
> + }
> }
> mmgrab(client->mm);
> init_srcu_struct(&client->hwctx_srcu);
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH V1] accel/amdxdna: Require carveout when PASID and force_iova are disabled
2026-06-04 19:54 [PATCH V1] accel/amdxdna: Require carveout when PASID and force_iova are disabled Lizhi Hou
2026-06-04 19:56 ` Mario Limonciello
@ 2026-06-04 20:10 ` sashiko-bot
1 sibling, 0 replies; 4+ messages in thread
From: sashiko-bot @ 2026-06-04 20:10 UTC (permalink / raw)
To: Lizhi Hou; +Cc: dri-devel
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
Pre-existing issues:
- [High] Dangling pointer in `client->sva` leads to a Use-After-Free and double-free during DRM close if `iommu_sva_get_pasid()` fails.
--
commit a925d70753a15e6bc99a30d4fbc4ec99fe4b925e
Author: Lizhi Hou <lizhi.hou@amd.com>
accel/amdxdna: Require carveout when PASID and force_iova are disabled
When both PASID and force_iova are disabled, this patch ensures carveout
memory is configured, returning an error during DRM open if it is not.
> diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c b/drivers/accel/amdxdna/amdxdna_pci_drv.c
> index c677293c1ae7f..1f066ed8d2365 100644
> --- a/drivers/accel/amdxdna/amdxdna_pci_drv.c
> +++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c
> @@ -120,8 +120,14 @@ static int amdxdna_drm_open(struct drm_device *ddev, struct drm_file *filp)
>
> if (!amdxdna_iova_on(xdna)) {
> /* No need to fail open since user may use pa + carveout later. */
> - if (amdxdna_sva_init(client))
> + if (amdxdna_sva_init(client)) {
> XDNA_WARN(xdna, "PASID not available for pid %d", client->pid);
> + if (!amdxdna_use_carveout(xdna)) {
> + XDNA_ERR(xdna, "PASID unavailable and carveout not configured");
> + kfree(client);
> + return -EINVAL;
> + }
> + }
[Severity: High]
This is a pre-existing issue, but since this patch relies on the fallback
behavior when amdxdna_sva_init() fails, does this leave a dangling pointer in
client->sva?
Looking at amdxdna_sva_init(), if iommu_sva_bind_device() succeeds but
iommu_sva_get_pasid() fails, the sva handle is unbound but the pointer isn't
cleared:
amdxdna_sva_init() {
...
client->pasid = iommu_sva_get_pasid(client->sva);
if (client->pasid == IOMMU_PASID_INVALID) {
iommu_sva_unbind_device(client->sva);
XDNA_ERR(xdna, "SVA get pasid failed");
return -ENODEV;
}
...
}
If amdxdna_use_carveout() returns true, amdxdna_drm_open() continues
successfully, keeping the client object alive.
When the DRM file is later closed, amdxdna_sva_fini() is called:
amdxdna_sva_fini() {
if (IS_ERR_OR_NULL(client->sva))
return;
iommu_sva_unbind_device(client->sva);
}
Since client->sva still holds the old pointer instead of NULL or an error
pointer, the IS_ERR_OR_NULL check will pass, and iommu_sva_unbind_device()
will be called a second time. Can this cause a double-free or use-after-free
of the iommu_sva structure during cleanup?
> }
> mmgrab(client->mm);
> init_srcu_struct(&client->hwctx_srcu);
--
Sashiko AI review · https://sashiko.dev/#/patchset/20260604195459.2423279-1-lizhi.hou@amd.com?part=1
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH V1] accel/amdxdna: Require carveout when PASID and force_iova are disabled
2026-06-04 19:56 ` Mario Limonciello
@ 2026-06-05 16:22 ` Lizhi Hou
0 siblings, 0 replies; 4+ messages in thread
From: Lizhi Hou @ 2026-06-05 16:22 UTC (permalink / raw)
To: Mario Limonciello, ogabbay, quic_jhugo, dri-devel,
karol.wachowski
Cc: linux-kernel, max.zhen, sonal.santan
Applied to drm-misc-next-fixes
On 6/4/26 12:56, Mario Limonciello wrote:
>
>
> On 6/4/26 14:54, Lizhi Hou wrote:
>> When both PASID and force_iova are disabled, carveout memory should be
>> used. Reject buffer allocations that cannot use carveout memory in this
>> configuration and return an error.
>>
>> Fixes: 3cc5d7a59519 ("accel/amdxdna: Add carveout memory support for
>> non-IOMMU systems")
>> Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
> Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
>> ---
>> drivers/accel/amdxdna/amdxdna_pci_drv.c | 8 +++++++-
>> 1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c
>> b/drivers/accel/amdxdna/amdxdna_pci_drv.c
>> index c677293c1ae7..1f066ed8d236 100644
>> --- a/drivers/accel/amdxdna/amdxdna_pci_drv.c
>> +++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c
>> @@ -120,8 +120,14 @@ static int amdxdna_drm_open(struct drm_device
>> *ddev, struct drm_file *filp)
>> if (!amdxdna_iova_on(xdna)) {
>> /* No need to fail open since user may use pa + carveout
>> later. */
>> - if (amdxdna_sva_init(client))
>> + if (amdxdna_sva_init(client)) {
>> XDNA_WARN(xdna, "PASID not available for pid %d",
>> client->pid);
>> + if (!amdxdna_use_carveout(xdna)) {
>> + XDNA_ERR(xdna, "PASID unavailable and carveout not
>> configured");
>> + kfree(client);
>> + return -EINVAL;
>> + }
>> + }
>> }
>> mmgrab(client->mm);
>> init_srcu_struct(&client->hwctx_srcu);
>
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-06-05 16:22 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-06-04 19:54 [PATCH V1] accel/amdxdna: Require carveout when PASID and force_iova are disabled Lizhi Hou
2026-06-04 19:56 ` Mario Limonciello
2026-06-05 16:22 ` Lizhi Hou
2026-06-04 20:10 ` sashiko-bot
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.