[PATCH] drm/amdgpu: don't attach the tlb fence for SI

[PATCH] drm/amdgpu: don't attach the tlb fence for SI

[Hans de Goede]

From: Alex Deucher <alexander.deucher@amd.com>

commit eb296c09805ee37dd4ea520a7fb3ec157c31090f upstream.

SI hardware doesn't support pasids, user mode queues, or
KIQ/MES so there is no need for this.  Doing so results in
a segfault as these callbacks are non-existent for SI.

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4744
Fixes: f3854e04b708 ("drm/amdgpu: attach tlb fence to the PTs update")
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 676e24fb8864..cdcafde3c71a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1066,7 +1066,9 @@ amdgpu_vm_tlb_flush(struct amdgpu_vm_update_params *params,
 	}
 
 	/* Prepare a TLB flush fence to be attached to PTs */
-	if (!params->unlocked) {
+	if (!params->unlocked &&
+	    /* SI doesn't support pasid or KIQ/MES */
+	    params->adev->family > AMDGPU_FAMILY_SI) {
 		amdgpu_vm_tlb_fence_create(params->adev, vm, fence);
 
 		/* Makes sure no PD/PT is freed before the flush */
-- 
2.52.0

Re: [PATCH] drm/amdgpu: don't attach the tlb fence for SI

[Greg KH]

On Sun, Dec 14, 2025 at 10:53:36AM +0100, Hans de Goede wrote:
> From: Alex Deucher <alexander.deucher@amd.com>
> 
> commit eb296c09805ee37dd4ea520a7fb3ec157c31090f upstream.
> 
> SI hardware doesn't support pasids, user mode queues, or
> KIQ/MES so there is no need for this.  Doing so results in
> a segfault as these callbacks are non-existent for SI.
> 
> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4744
> Fixes: f3854e04b708 ("drm/amdgpu: attach tlb fence to the PTs update")
> Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)

What kernel tree(s) should this go to?

thanks,

greg k-jh

Re: [PATCH] drm/amdgpu: don't attach the tlb fence for SI

[Hans de Goede]

Hi greg,

On 15-Dec-25 9:12 AM, Greg KH wrote:
> On Sun, Dec 14, 2025 at 10:53:36AM +0100, Hans de Goede wrote:
>> From: Alex Deucher <alexander.deucher@amd.com>
>>
>> commit eb296c09805ee37dd4ea520a7fb3ec157c31090f upstream.
>>
>> SI hardware doesn't support pasids, user mode queues, or
>> KIQ/MES so there is no need for this.  Doing so results in
>> a segfault as these callbacks are non-existent for SI.
>>
>> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4744
>> Fixes: f3854e04b708 ("drm/amdgpu: attach tlb fence to the PTs update")
>> Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>> Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
>> ---
>>  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +++-
>>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> What kernel tree(s) should this go to?

This fixes a regression introduced in at least 6.17.y (6.17.11)
and 6.18.y (6.18.1). So it should at least go to those branches.

If any other branches also have gotten commit f3854e04b708
backported then those should get this too.

Regards,

Hans

Re: [PATCH] drm/amdgpu: don't attach the tlb fence for SI

[Greg KH]

On Mon, Dec 15, 2025 at 09:15:17AM +0100, Hans de Goede wrote:
> Hi greg,
> 
> On 15-Dec-25 9:12 AM, Greg KH wrote:
> > On Sun, Dec 14, 2025 at 10:53:36AM +0100, Hans de Goede wrote:
> >> From: Alex Deucher <alexander.deucher@amd.com>
> >>
> >> commit eb296c09805ee37dd4ea520a7fb3ec157c31090f upstream.
> >>
> >> SI hardware doesn't support pasids, user mode queues, or
> >> KIQ/MES so there is no need for this.  Doing so results in
> >> a segfault as these callbacks are non-existent for SI.
> >>
> >> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4744
> >> Fixes: f3854e04b708 ("drm/amdgpu: attach tlb fence to the PTs update")
> >> Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
> >> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> >> Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
> >> ---
> >>  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +++-
> >>  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > What kernel tree(s) should this go to?
> 
> This fixes a regression introduced in at least 6.17.y (6.17.11)
> and 6.18.y (6.18.1). So it should at least go to those branches.

But that commit is in 6.19-rc1, not anything older.

> If any other branches also have gotten commit f3854e04b708
> backported then those should get this too.

I don't see that commit in any stable tree, what am I missing?

confused,

greg k-h

Re: [PATCH] drm/amdgpu: don't attach the tlb fence for SI

[Hans de Goede]

Hi Greg,

On 15-Dec-25 10:05 AM, Greg KH wrote:
> On Mon, Dec 15, 2025 at 09:15:17AM +0100, Hans de Goede wrote:
>> Hi greg,
>>
>> On 15-Dec-25 9:12 AM, Greg KH wrote:
>>> On Sun, Dec 14, 2025 at 10:53:36AM +0100, Hans de Goede wrote:
>>>> From: Alex Deucher <alexander.deucher@amd.com>
>>>>
>>>> commit eb296c09805ee37dd4ea520a7fb3ec157c31090f upstream.
>>>>
>>>> SI hardware doesn't support pasids, user mode queues, or
>>>> KIQ/MES so there is no need for this.  Doing so results in
>>>> a segfault as these callbacks are non-existent for SI.
>>>>
>>>> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4744
>>>> Fixes: f3854e04b708 ("drm/amdgpu: attach tlb fence to the PTs update")
>>>> Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
>>>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>>>> Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
>>>> ---
>>>>  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +++-
>>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> What kernel tree(s) should this go to?
>>
>> This fixes a regression introduced in at least 6.17.y (6.17.11)
>> and 6.18.y (6.18.1). So it should at least go to those branches.
> 
> But that commit is in 6.19-rc1, not anything older.
> 
>> If any other branches also have gotten commit f3854e04b708
>> backported then those should get this too.
> 
> I don't see that commit in any stable tree, what am I missing?

Ah, I see now the Fixes tag in the original fix (which I cherry
picked) is wrong and does not point to the canonical commit id as
merged into Torvalds tree, sorry.

This is b4a7f4e7ad2b120a94f3111f92a11520052c762d  ("drm/amdgpu:
attach tlb fence to the PTs update") in Torvalds' tree:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b4a7f4e7ad2b120a94f3111f92a11520052c762d

and this actually made it into the v6.18 tag (vs being introduced
in 6.18.1 as I originally thought).

This is also in 6.17.11 as 23316ed02c228b52f871050f98a155f3d566c450

https://web.git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/gpu/drm/amd?h=linux-6.17.y&id=23316ed02c228b52f871050f98a155f3d566c450

FWIW I don't see it in 6.12.y (and I did not look further back).

Regards,

Hans

Re: [PATCH] drm/amdgpu: don't attach the tlb fence for SI

[Viktor Jägersküpper]

(Adding Sasha to this thread)

Hans de Goede wrote:
> Hi Greg,
> 
> On 15-Dec-25 10:05 AM, Greg KH wrote:
>> On Mon, Dec 15, 2025 at 09:15:17AM +0100, Hans de Goede wrote:
>>> Hi greg,
>>>
>>> On 15-Dec-25 9:12 AM, Greg KH wrote:
>>>> On Sun, Dec 14, 2025 at 10:53:36AM +0100, Hans de Goede wrote:
>>>>> From: Alex Deucher <alexander.deucher@amd.com>
>>>>>
>>>>> commit eb296c09805ee37dd4ea520a7fb3ec157c31090f upstream.
>>>>>
>>>>> SI hardware doesn't support pasids, user mode queues, or
>>>>> KIQ/MES so there is no need for this.  Doing so results in
>>>>> a segfault as these callbacks are non-existent for SI.
>>>>>
>>>>> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4744
>>>>> Fixes: f3854e04b708 ("drm/amdgpu: attach tlb fence to the PTs update")
>>>>> Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
>>>>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>>>>> Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
>>>>> ---
>>>>>  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +++-
>>>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> What kernel tree(s) should this go to?
>>>
>>> This fixes a regression introduced in at least 6.17.y (6.17.11)
>>> and 6.18.y (6.18.1). So it should at least go to those branches.
>>
>> But that commit is in 6.19-rc1, not anything older.
>>
>>> If any other branches also have gotten commit f3854e04b708
>>> backported then those should get this too.
>>
>> I don't see that commit in any stable tree, what am I missing?
> 
> Ah, I see now the Fixes tag in the original fix (which I cherry
> picked) is wrong and does not point to the canonical commit id as
> merged into Torvalds tree, sorry.
> 
> This is b4a7f4e7ad2b120a94f3111f92a11520052c762d  ("drm/amdgpu:
> attach tlb fence to the PTs update") in Torvalds' tree:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b4a7f4e7ad2b120a94f3111f92a11520052c762d
> 
> and this actually made it into the v6.18 tag (vs being introduced
> in 6.18.1 as I originally thought).
> 
> This is also in 6.17.11 as 23316ed02c228b52f871050f98a155f3d566c450
> 
> https://web.git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/gpu/drm/amd?h=linux-6.17.y&id=23316ed02c228b52f871050f98a155f3d566c450
> 
> FWIW I don't see it in 6.12.y (and I did not look further back).
> 
> Regards,
> 
> Hans
Can this fix be added to the 6.18 queue, please? All relevant
information can be found in the previous email from Hans.

Thanks,
Viktor

Re: [PATCH] drm/amdgpu: don't attach the tlb fence for SI

[Greg KH]

On Fri, Dec 19, 2025 at 02:41:17PM +0100, Viktor Jägersküpper wrote:
> (Adding Sasha to this thread)
> 
> Hans de Goede wrote:
> > Hi Greg,
> > 
> > On 15-Dec-25 10:05 AM, Greg KH wrote:
> >> On Mon, Dec 15, 2025 at 09:15:17AM +0100, Hans de Goede wrote:
> >>> Hi greg,
> >>>
> >>> On 15-Dec-25 9:12 AM, Greg KH wrote:
> >>>> On Sun, Dec 14, 2025 at 10:53:36AM +0100, Hans de Goede wrote:
> >>>>> From: Alex Deucher <alexander.deucher@amd.com>
> >>>>>
> >>>>> commit eb296c09805ee37dd4ea520a7fb3ec157c31090f upstream.
> >>>>>
> >>>>> SI hardware doesn't support pasids, user mode queues, or
> >>>>> KIQ/MES so there is no need for this.  Doing so results in
> >>>>> a segfault as these callbacks are non-existent for SI.
> >>>>>
> >>>>> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4744
> >>>>> Fixes: f3854e04b708 ("drm/amdgpu: attach tlb fence to the PTs update")
> >>>>> Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
> >>>>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> >>>>> Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
> >>>>> ---
> >>>>>  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +++-
> >>>>>  1 file changed, 3 insertions(+), 1 deletion(-)
> >>>>
> >>>> What kernel tree(s) should this go to?
> >>>
> >>> This fixes a regression introduced in at least 6.17.y (6.17.11)
> >>> and 6.18.y (6.18.1). So it should at least go to those branches.
> >>
> >> But that commit is in 6.19-rc1, not anything older.
> >>
> >>> If any other branches also have gotten commit f3854e04b708
> >>> backported then those should get this too.
> >>
> >> I don't see that commit in any stable tree, what am I missing?
> > 
> > Ah, I see now the Fixes tag in the original fix (which I cherry
> > picked) is wrong and does not point to the canonical commit id as
> > merged into Torvalds tree, sorry.
> > 
> > This is b4a7f4e7ad2b120a94f3111f92a11520052c762d  ("drm/amdgpu:
> > attach tlb fence to the PTs update") in Torvalds' tree:
> > 
> > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b4a7f4e7ad2b120a94f3111f92a11520052c762d
> > 
> > and this actually made it into the v6.18 tag (vs being introduced
> > in 6.18.1 as I originally thought).
> > 
> > This is also in 6.17.11 as 23316ed02c228b52f871050f98a155f3d566c450
> > 
> > https://web.git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/gpu/drm/amd?h=linux-6.17.y&id=23316ed02c228b52f871050f98a155f3d566c450
> > 
> > FWIW I don't see it in 6.12.y (and I did not look further back).
> > 
> > Regards,
> > 
> > Hans
> Can this fix be added to the 6.18 queue, please? All relevant
> information can be found in the previous email from Hans.

This is now in the 6.18.4 release.

thanks,

greg k-h