[PATCH] drm/amdgpu: fix regression on CIK (v2)

[PATCH] drm/amdgpu: fix regression on CIK (v2)

[Alex Deucher]

From: Grigori Goronzy <greg@chown.ath.cx>

This fix was written against drm-next, but when it was
backported to 4.5 as a stable fix, the driver internal
structure change was missed.  Fix that up here to avoid
a hang due to waiting for the wrong sequence number.

v2: agd: fix up commit message

Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---

Hi Greg,

The code is correct for 4.6, this fix should be applied to
4.5 and older stable kernels.

Thanks,

Alex

 drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index 06602df..9b1c430 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -3628,7 +3628,7 @@ static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					unsigned vm_id, uint64_t pd_addr)
 {
 	int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX);
-	uint32_t seq = ring->fence_drv.sync_seq;
+	uint32_t seq = ring->fence_drv.sync_seq[ring->idx];
 	uint64_t addr = ring->fence_drv.gpu_addr;
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
-- 
2.5.0

Re: [PATCH] drm/amdgpu: fix regression on CIK (v2)

[Greg KH]

On Tue, Mar 22, 2016 at 03:48:18PM -0400, Alex Deucher wrote:
> From: Grigori Goronzy <greg@chown.ath.cx>
> 
> This fix was written against drm-next, but when it was
> backported to 4.5 as a stable fix, the driver internal
> structure change was missed.  Fix that up here to avoid
> a hang due to waiting for the wrong sequence number.
> 
> v2: agd: fix up commit message
> 
> Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
> Cc: stable@vger.kernel.org
> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> ---
> 
> Hi Greg,
> 
> The code is correct for 4.6, this fix should be applied to
> 4.5 and older stable kernels.

What is the patch in 4.6 that made it "correct"?  I would much rather
take that than something else.

thanks,

greg k-h

Re: [PATCH] drm/amdgpu: fix regression on CIK (v2)

[Alex Deucher]

On Tue, Mar 22, 2016 at 4:11 PM, Greg KH <gregkh@linuxfoundation.org> wrote:
> On Tue, Mar 22, 2016 at 03:48:18PM -0400, Alex Deucher wrote:
>> From: Grigori Goronzy <greg@chown.ath.cx>
>>
>> This fix was written against drm-next, but when it was
>> backported to 4.5 as a stable fix, the driver internal
>> structure change was missed.  Fix that up here to avoid
>> a hang due to waiting for the wrong sequence number.
>>
>> v2: agd: fix up commit message
>>
>> Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
>> Cc: stable@vger.kernel.org
>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>> ---
>>
>> Hi Greg,
>>
>> The code is correct for 4.6, this fix should be applied to
>> 4.5 and older stable kernels.
>
> What is the patch in 4.6 that made it "correct"?  I would much rather
> take that than something else.

It's correct as is for 4.6.  I cherry-picked it out of my drm-next
tree into 4.5 since it was a bug fix, but missed the internal driver
structure change in drm-next that was not in 4.5.

Alex

Re: [PATCH] drm/amdgpu: fix regression on CIK (v2)

[Greg KH]

On Tue, Mar 22, 2016 at 04:14:30PM -0400, Alex Deucher wrote:
> On Tue, Mar 22, 2016 at 4:11 PM, Greg KH <gregkh@linuxfoundation.org> wrote:
> > On Tue, Mar 22, 2016 at 03:48:18PM -0400, Alex Deucher wrote:
> >> From: Grigori Goronzy <greg@chown.ath.cx>
> >>
> >> This fix was written against drm-next, but when it was
> >> backported to 4.5 as a stable fix, the driver internal
> >> structure change was missed.  Fix that up here to avoid
> >> a hang due to waiting for the wrong sequence number.
> >>
> >> v2: agd: fix up commit message
> >>
> >> Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
> >> Cc: stable@vger.kernel.org
> >> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> >> ---
> >>
> >> Hi Greg,
> >>
> >> The code is correct for 4.6, this fix should be applied to
> >> 4.5 and older stable kernels.
> >
> > What is the patch in 4.6 that made it "correct"?  I would much rather
> > take that than something else.
> 
> It's correct as is for 4.6.  I cherry-picked it out of my drm-next
> tree into 4.5 since it was a bug fix, but missed the internal driver
> structure change in drm-next that was not in 4.5.

I'm sorry, I don't understand at all what you mean here :(

What changed between 4.5.0 and 4.6-rc1 that made 4.5 "broken"?  What
fixed it in 4.6-rc1?

totally confused,

greg k-h

Re: [PATCH] drm/amdgpu: fix regression on CIK (v2)

[Alex Deucher]

On Tue, Mar 22, 2016 at 4:27 PM, Greg KH <gregkh@linuxfoundation.org> wrote:
> On Tue, Mar 22, 2016 at 04:14:30PM -0400, Alex Deucher wrote:
>> On Tue, Mar 22, 2016 at 4:11 PM, Greg KH <gregkh@linuxfoundation.org> wrote:
>> > On Tue, Mar 22, 2016 at 03:48:18PM -0400, Alex Deucher wrote:
>> >> From: Grigori Goronzy <greg@chown.ath.cx>
>> >>
>> >> This fix was written against drm-next, but when it was
>> >> backported to 4.5 as a stable fix, the driver internal
>> >> structure change was missed.  Fix that up here to avoid
>> >> a hang due to waiting for the wrong sequence number.
>> >>
>> >> v2: agd: fix up commit message
>> >>
>> >> Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
>> >> Cc: stable@vger.kernel.org
>> >> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>> >> ---
>> >>
>> >> Hi Greg,
>> >>
>> >> The code is correct for 4.6, this fix should be applied to
>> >> 4.5 and older stable kernels.
>> >
>> > What is the patch in 4.6 that made it "correct"?  I would much rather
>> > take that than something else.
>>
>> It's correct as is for 4.6.  I cherry-picked it out of my drm-next
>> tree into 4.5 since it was a bug fix, but missed the internal driver
>> structure change in drm-next that was not in 4.5.
>
> I'm sorry, I don't understand at all what you mean here :(
>
> What changed between 4.5.0 and 4.6-rc1 that made 4.5 "broken"?  What
> fixed it in 4.6-rc1?

The patch was written and tested against drm-next, but I cherry-picked
it to 4.5 as is since it was a bug fix.  However, I missed the
internal driver structure change that it depended on.  Now that
drm-next has been merged into 4.6 (including the driver structure
change), the code from 4.5 works fine.  However 4.5 and older kernels
are broken since they don't have the changes for 4.6.

Alex

Re: [PATCH] drm/amdgpu: fix regression on CIK (v2)

[Greg KH]

On Tue, Mar 22, 2016 at 04:31:30PM -0400, Alex Deucher wrote:
> On Tue, Mar 22, 2016 at 4:27 PM, Greg KH <gregkh@linuxfoundation.org> wrote:
> > On Tue, Mar 22, 2016 at 04:14:30PM -0400, Alex Deucher wrote:
> >> On Tue, Mar 22, 2016 at 4:11 PM, Greg KH <gregkh@linuxfoundation.org> wrote:
> >> > On Tue, Mar 22, 2016 at 03:48:18PM -0400, Alex Deucher wrote:
> >> >> From: Grigori Goronzy <greg@chown.ath.cx>
> >> >>
> >> >> This fix was written against drm-next, but when it was
> >> >> backported to 4.5 as a stable fix, the driver internal
> >> >> structure change was missed.  Fix that up here to avoid
> >> >> a hang due to waiting for the wrong sequence number.
> >> >>
> >> >> v2: agd: fix up commit message
> >> >>
> >> >> Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
> >> >> Cc: stable@vger.kernel.org
> >> >> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> >> >> ---
> >> >>
> >> >> Hi Greg,
> >> >>
> >> >> The code is correct for 4.6, this fix should be applied to
> >> >> 4.5 and older stable kernels.
> >> >
> >> > What is the patch in 4.6 that made it "correct"?  I would much rather
> >> > take that than something else.
> >>
> >> It's correct as is for 4.6.  I cherry-picked it out of my drm-next
> >> tree into 4.5 since it was a bug fix, but missed the internal driver
> >> structure change in drm-next that was not in 4.5.
> >
> > I'm sorry, I don't understand at all what you mean here :(
> >
> > What changed between 4.5.0 and 4.6-rc1 that made 4.5 "broken"?  What
> > fixed it in 4.6-rc1?
> 
> The patch was written and tested against drm-next, but I cherry-picked
> it to 4.5 as is since it was a bug fix.  However, I missed the
> internal driver structure change that it depended on.  Now that
> drm-next has been merged into 4.6 (including the driver structure
> change), the code from 4.5 works fine.  However 4.5 and older kernels
> are broken since they don't have the changes for 4.6.

I still don't understand.  What is the commit in Linus's tree that
matches this change?  Why can't I just cherry-pick that patch instead of
this "custom" one?

thanks,

greg k-h

Re: [PATCH] drm/amdgpu: fix regression on CIK (v2)

[Alex Deucher]

On Mon, Apr 11, 2016 at 3:21 AM, Christian König
<christian.koenig@amd.com> wrote:
> Am 10.04.2016 um 18:22 schrieb Greg KH:
>>
>> On Tue, Mar 22, 2016 at 04:31:30PM -0400, Alex Deucher wrote:
>>>
>>> On Tue, Mar 22, 2016 at 4:27 PM, Greg KH <gregkh@linuxfoundation.org>
>>> wrote:
>>>>
>>>> On Tue, Mar 22, 2016 at 04:14:30PM -0400, Alex Deucher wrote:
>>>>>
>>>>> On Tue, Mar 22, 2016 at 4:11 PM, Greg KH <gregkh@linuxfoundation.org>
>>>>> wrote:
>>>>>>
>>>>>> On Tue, Mar 22, 2016 at 03:48:18PM -0400, Alex Deucher wrote:
>>>>>>>
>>>>>>> From: Grigori Goronzy <greg@chown.ath.cx>
>>>>>>>
>>>>>>> This fix was written against drm-next, but when it was
>>>>>>> backported to 4.5 as a stable fix, the driver internal
>>>>>>> structure change was missed.  Fix that up here to avoid
>>>>>>> a hang due to waiting for the wrong sequence number.
>>>>>>>
>>>>>>> v2: agd: fix up commit message
>>>>>>>
>>>>>>> Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
>>>>>>> Cc: stable@vger.kernel.org
>>>>>>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>>>>>>> ---
>>>>>>>
>>>>>>> Hi Greg,
>>>>>>>
>>>>>>> The code is correct for 4.6, this fix should be applied to
>>>>>>> 4.5 and older stable kernels.
>>>>>>
>>>>>> What is the patch in 4.6 that made it "correct"?  I would much rather
>>>>>> take that than something else.
>>>>>
>>>>> It's correct as is for 4.6.  I cherry-picked it out of my drm-next
>>>>> tree into 4.5 since it was a bug fix, but missed the internal driver
>>>>> structure change in drm-next that was not in 4.5.
>>>>
>>>> I'm sorry, I don't understand at all what you mean here :(
>>>>
>>>> What changed between 4.5.0 and 4.6-rc1 that made 4.5 "broken"?  What
>>>> fixed it in 4.6-rc1?
>>>
>>> The patch was written and tested against drm-next, but I cherry-picked
>>> it to 4.5 as is since it was a bug fix.  However, I missed the
>>> internal driver structure change that it depended on.  Now that
>>> drm-next has been merged into 4.6 (including the driver structure
>>> change), the code from 4.5 works fine.  However 4.5 and older kernels
>>> are broken since they don't have the changes for 4.6.
>>
>> I still don't understand.  What is the commit in Linus's tree that
>> matches this change?  Why can't I just cherry-pick that patch instead of
>> this "custom" one?
>
>
> The change in Linus tree this bug fix was based on is the improvement of the
> sequence handling and removal of hardware semaphores I did for 4.6. And
> those two new features in turn are based on the introduction of the GPU
> scheduler earlier in the 4.6 cycle.
>
> In total I think you would need to cherry pick something between 30 and 50
> patches to get to the same state and none of them are bug fixes which should
> go into 4.5 or older kernels.
>
> So Alex and Grigori's argumentation is valid that this should be fixed by
> this custom single line change instead of backporting everything it's based
> upon.
>
> Sorry for not being clear about this,
> Christian.

Greg is there any chance this could be applied?

Alex

>
>>
>> thanks,
>>
>> greg k-h
>
>

Re: [PATCH] drm/amdgpu: fix regression on CIK (v2)

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 2827 bytes --]

On Sun, 2016-04-10 at 09:22 -0700, Greg KH wrote:
> On Tue, Mar 22, 2016 at 04:31:30PM -0400, Alex Deucher wrote:
> > 
> > On Tue, Mar 22, 2016 at 4:27 PM, Greg KH  wrote:
> > > 
> > > On Tue, Mar 22, 2016 at 04:14:30PM -0400, Alex Deucher wrote:
> > > > 
> > > > On Tue, Mar 22, 2016 at 4:11 PM, Greg KH  wrote:
> > > > > 
> > > > > On Tue, Mar 22, 2016 at 03:48:18PM -0400, Alex Deucher wrote:
> > > > > > 
> > > > > > From: Grigori Goronzy <greg@chown.ath.cx>
> > > > > > 
> > > > > > This fix was written against drm-next, but when it was
> > > > > > backported to 4.5 as a stable fix, the driver internal
> > > > > > structure change was missed.  Fix that up here to avoid
> > > > > > a hang due to waiting for the wrong sequence number.
> > > > > > 
> > > > > > v2: agd: fix up commit message
> > > > > > 
> > > > > > Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
> > > > > > Cc: stable@vger.kernel.org
> > > > > > Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> > > > > > ---
> > > > > > 
> > > > > > Hi Greg,
> > > > > > 
> > > > > > The code is correct for 4.6, this fix should be applied to
> > > > > > 4.5 and older stable kernels.
> > > > > What is the patch in 4.6 that made it "correct"?  I would much rather
> > > > > take that than something else.
> > > > It's correct as is for 4.6.  I cherry-picked it out of my drm-next
> > > > tree into 4.5 since it was a bug fix, but missed the internal driver
> > > > structure change in drm-next that was not in 4.5.
> > > I'm sorry, I don't understand at all what you mean here :(
> > > 
> > > What changed between 4.5.0 and 4.6-rc1 that made 4.5 "broken"?  What
> > > fixed it in 4.6-rc1?
> > The patch was written and tested against drm-next, but I cherry-picked
> > it to 4.5 as is since it was a bug fix.  However, I missed the
> > internal driver structure change that it depended on.  Now that
> > drm-next has been merged into 4.6 (including the driver structure
> > change), the code from 4.5 works fine.  However 4.5 and older kernels
> > are broken since they don't have the changes for 4.6.
> I still don't understand.  What is the commit in Linus's tree that
> matches this change?

This commit changed amdgpu_fence_driver::sync_seq from an array to a
scalar:

commit 5907a0d8af71d17811be49f2c056b3a89660e188
Author: Christian König <christian.koenig@amd.com>
Date:   Mon Jan 18 15:16:53 2016 +0100

    drm/amdgpu: cleanup sync_seq handling

> Why can't I just cherry-pick that patch instead of
> this "custom" one?
[...]

Looks like it depends on other changes to lock usage.

Ben.

-- 
Ben Hutchings
The generation of random numbers is too important to be left to chance.
                                                            - Robert Coveyou

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Patch "drm/amdgpu: fix regression on CIK (v2)" has been added to the 4.4-stable tree

[gregkh]

This is a note to let you know that I've just added the patch titled

    drm/amdgpu: fix regression on CIK (v2)

to the 4.4-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-fix-regression-on-cik-v2.patch
and it can be found in the queue-4.4 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@vger.kernel.org> know about it.


>From alexdeucher@gmail.com  Sun May  1 15:58:01 2016
From: Grigori Goronzy <greg@chown.ath.cx>
Date: Tue, 22 Mar 2016 15:48:18 -0400
Subject: drm/amdgpu: fix regression on CIK (v2)
To: gregkh@linuxfoundation.org
Cc: christian.koenig@amd.com, greg@chown.ath.cx, stable@vger.kernel.org, Alex Deucher <alexander.deucher@amd.com>
Message-ID: <1458676098-9543-1-git-send-email-alexander.deucher@amd.com>

From: Grigori Goronzy <greg@chown.ath.cx>

This fix was written against drm-next, but when it was
backported to 4.5 as a stable fix, the driver internal
structure change was missed.  Fix that up here to avoid
a hang due to waiting for the wrong sequence number.

v2: agd: fix up commit message

Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

---
 drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -3628,7 +3628,7 @@ static void gfx_v7_0_ring_emit_vm_flush(
 					unsigned vm_id, uint64_t pd_addr)
 {
 	int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX);
-	uint32_t seq = ring->fence_drv.sync_seq;
+	uint32_t seq = ring->fence_drv.sync_seq[ring->idx];
 	uint64_t addr = ring->fence_drv.gpu_addr;
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));


Patches currently in stable-queue which might be from greg@chown.ath.cx are

queue-4.4/drm-amdgpu-fix-regression-on-cik-v2.patch

Patch "drm/amdgpu: fix regression on CIK (v2)" has been added to the 4.5-stable tree

[gregkh]

This is a note to let you know that I've just added the patch titled

    drm/amdgpu: fix regression on CIK (v2)

to the 4.5-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-fix-regression-on-cik-v2.patch
and it can be found in the queue-4.5 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@vger.kernel.org> know about it.


>From alexdeucher@gmail.com  Sun May  1 15:58:01 2016
From: Grigori Goronzy <greg@chown.ath.cx>
Date: Tue, 22 Mar 2016 15:48:18 -0400
Subject: drm/amdgpu: fix regression on CIK (v2)
To: gregkh@linuxfoundation.org
Cc: christian.koenig@amd.com, greg@chown.ath.cx, stable@vger.kernel.org, Alex Deucher <alexander.deucher@amd.com>
Message-ID: <1458676098-9543-1-git-send-email-alexander.deucher@amd.com>

From: Grigori Goronzy <greg@chown.ath.cx>

This fix was written against drm-next, but when it was
backported to 4.5 as a stable fix, the driver internal
structure change was missed.  Fix that up here to avoid
a hang due to waiting for the wrong sequence number.

v2: agd: fix up commit message

Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

---
 drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -3628,7 +3628,7 @@ static void gfx_v7_0_ring_emit_vm_flush(
 					unsigned vm_id, uint64_t pd_addr)
 {
 	int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX);
-	uint32_t seq = ring->fence_drv.sync_seq;
+	uint32_t seq = ring->fence_drv.sync_seq[ring->idx];
 	uint64_t addr = ring->fence_drv.gpu_addr;
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));


Patches currently in stable-queue which might be from greg@chown.ath.cx are

queue-4.5/drm-amdgpu-fix-regression-on-cik-v2.patch