Reviewed-by: Ken Wang
发件人: amd-gfx 代表 Rex Zhu
发送时间: 2016年8月3日 14:42:01
收件人: amd-gfx@lists.freedesktop.org
抄送: Zhu, Rex
主题: [PATCH] drm/amd/powrplay: delete code set/unset
From: Rex Zhu
This is a workaround to let VCE soft reset work.
RB1_BUSY bit is always set, so remove its checking now, and we
will depend on RB0_BUSY currently.
After we find the root cause of RB1_BUSY, we can add it back.
Change-Id: I0a4b622d8bc50f75d1a4187324cd2e9079c95013
Change-Id: I83a1cd8b19cbffc09c4e9b679b88cd65147446d5
Signed-off-by: Chunming Zhou
---
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 3 ---
1 file changed, 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
index
From: Michel Dänzer
At least with older kernels, the flip may never complete otherwise,
which can result in us hanging in drmmode_set_mode_major.
Fixes:
https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-ati/+bug/1577170
Signed-off-by: Michel Dänzer
Change-Id: Ia3f60107f7257f834be961128793af9b8a0389da
Signed-off-by: Chunming Zhou
---
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 -
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 3 ---
2 files changed, 4 deletions(-)
diff --git
VCE is always busy is since RB1_BUSY is always busy, now remove it.
And enable VCE soft reset.
Chunming Zhou (2):
drm/amdgpu: fix harvest config checking in vce3 check_soft_reset
drm/amdgpu: enable VCE soft reset
Rex Zhu (1):
drm/amdgpu: remove RB1_BUSY bit checking
Am 02.08.2016 um 03:58 schrieb zhoucm1:
On 2016年08月02日 02:30, Alex Deucher wrote:
This got leftover somehow when I cleaned this up.
Signed-off-by: Alex Deucher
Reviewed-by: Chunming Zhou
Reviewed-by: Christian König
Hi,
I think it would be helpful for at least one liner explaining actually
what is even getting "fixed" here? Obviously we assume, given that you
work for AMD, these magics are correct however it does help in case of
issue.
Just for a motivating example, suppose a magic is technically correct as
CG was being enabled in reverse sense from dpm/powerplay.
Also fix the default CLK_EN signal to enable all of the blocks.
Signed-off-by: Tom St Denis
---
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git
Only issue bypass if MGCG enabled
Signed-off-by: Tom St Denis
---
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
index
Signed-off-by: Tom St Denis
---
drivers/gpu/drm/amd/amdgpu/vi.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c
index 5449a3c8c9eb..07ea150f85f4 100644
---
It's handled by DPM/PP properly.
Signed-off-by: Tom St Denis
---
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 4
1 file changed, 4 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
index 93604bb78257..9d1924e4e2bc
From: Michel Dänzer
Now we can program a flip during a vertical blank period, if it's the
one targeted by the flip (or a later one). This allows simplifying
amdgpu_flip_work_func considerably.
Signed-off-by: Michel Dänzer
---
From: Michel Dänzer
With the previous change, it's safe to let page flips take effect
anytime during a vertical blank period.
This can avoid delaying a flip by a frame in some cases where we get to
radeon_flip_work_func -> adev->mode_info.funcs->page_flip during a
For virtual display feature, if user set the option "amdgpu.virtual_display=1"
when load amdgpu.ko. Then need to set the ip_blocks with virtual display ip
blocks. And when enable virtual display, the amdgpu_dal need to be set to zero.
Signed-off-by: Emily Deng
---
For virtual display feature, initialize dce_virtual_crtc_funcs.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/dce_virtual.c | 30 ++
1 file changed, 26 insertions(+), 4 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/dce_virtual.c
On 2016年08月03日 21:43, Christian König wrote:
Well that is a clear NAK to this whole approach.
Submitting the recovery jobs to the scheduler is reentrant because the
scheduler is the one who originally signaled us of a timeout.
we have reset all recovery jobs, right? Could we think those jobs
For virtual display feature, when the GPU has DCE engine, need to disable
the VGA render and CRTC, or it will hang when initialize GMC.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c | 27
For virtual display feature, initialize dce_virtual_display_funcs,
which will be used in function dce_virtual_set_display_funcs.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/dce_virtual.c | 222 +--
1 file changed, 208 insertions(+),
For virtual display feature, add virtual connector and encoder macros.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/ObjectID.h | 7 +++
1 file changed, 7 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/ObjectID.h
For virtual display feature, as there is no dce engine, so no pageflip irq
generated. So directly call pageflip irq funtion when received vysn interrupt.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/dce_virtual.c | 10 +++---
1 file changed, 7 insertions(+),
For virtual display feature, initialize dce_virtual_crtc_helper_funcs.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/dce_virtual.c | 137 +--
1 file changed, 128 insertions(+), 9 deletions(-)
diff --git
For virtual display feature, add one connector type in amdgpu_connector_add.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c | 95 ++
1 file changed, 95 insertions(+)
diff --git
On 2016年08月03日 20:16, Edward O'Callaghan wrote:
Hi,
I didn't see any of my previous review remarks on your previous
submission addresses in this submission of this series?
Hi Edward,
Sorry for that. Since these patch sets are big feature change and we
need to pin down the approach of the
From: Michel Dänzer
These flags allow userspace to explicitly specify the target vertical
blank period when a flip should take effect.
Signed-off-by: Michel Dänzer
---
Note that the previous patches in this series can avoid delaying page
flips
From: Michel Dänzer
With the previous change, it's safe to let page flips take effect
anytime during a vertical blank period.
This can avoid delaying a flip by a frame in some cases where we get to
amdgpu_flip_work_func -> adev->mode_info.funcs->page_flip during a
The purpose of this series is to allow drivers to avoid unnecessarily
delaying page flips, by explicitly telling the driver which vblank seqno
a flip is supposed to take effect in.
Patch 1 sets the target to the vblank seqno after the current one when
the DRM_IOCTL_MODE_PAGE_FLIP ioctl is called,
From: Michel Dänzer
Mostly the same as the existing page_flip hook, but takes an additional
parameter specifying the target vertical blank period when the flip
should take effect.
Signed-off-by: Michel Dänzer
---
drivers/gpu/drm/drm_crtc.c | 23
For virtual display feature, initialize dce_virtual_crtc_irq_funcs,
dce_virtual_pageflip_irq_funcs. As it has no dce engine, so the
pageflip interrupt won't be generated, and the vsync interrupt will
be generated by smu's periodic timer or software timer which will
be implemented later.
For virtual display feature, first need to initialize dce_virtual_ip_funcs,
which will be used when set ip blocks.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/Makefile | 3 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_display.c | 5 +-
For virtual display feature, use the software timer to
simulate the vsync interrupt.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h | 4
drivers/gpu/drm/amd/amdgpu/dce_virtual.c | 33 ++--
For virtual display feature, define virtual display ip blocks, and set
dce_virtual_ip_funcs to DCE block.
Signed-off-by: Emily Deng
---
drivers/gpu/drm/amd/amdgpu/cik.c | 341 +++
drivers/gpu/drm/amd/amdgpu/vi.c | 282
The logic is a bit confusing at first. "gated" means that SW CG is enabled not
that the clocks are turned off. When I sampled the registers in the original
configuration the clocks were reading 100% when the core was active. It would
read 0% when idle but only because power was gated.
With
> -Original Message-
> From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf
> Of Tom St Denis
> Sent: Wednesday, August 03, 2016 11:52 AM
> To: amd-gfx@lists.freedesktop.org
> Cc: StDenis, Tom
> Subject: [PATCH 4/6] drm/amd/amdgpu: Move VCE bypass after MGCG test
>
>
Maybe add a comment to that effect. With that added, this patch is:
Reviewed-by: Alex Deucher
From: StDenis, Tom
Sent: Wednesday, August 03, 2016 12:31 PM
To: Deucher, Alexander; amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH 2/6] drm/amd/amdgpu: Fix VCE CG order
Am 03.08.2016 um 17:51 schrieb Tom St Denis:
Signed-off-by: Tom St Denis
This one is Reviewed-by: Christian König .
For the rest I'm not deep enough into the topic to judge.
Regards,
Christian.
---
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 13
> -Original Message-
> From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf
> Of Tom St Denis
> Sent: Wednesday, August 03, 2016 12:47 PM
> To: amd-gfx@lists.freedesktop.org
> Cc: StDenis, Tom
> Subject: [PATCH] drm/amd/amdgpu: Partially revert change to UVD v3 CG
>
>
> -Original Message-
> From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf
> Of Tom St Denis
> Sent: Wednesday, August 03, 2016 11:52 AM
> To: amd-gfx@lists.freedesktop.org
> Cc: StDenis, Tom
> Subject: [PATCH 2/6] drm/amd/amdgpu: Fix VCE CG order and resume
> defaults
>
> -Original Message-
> From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf
> Of Tom St Denis
> Sent: Wednesday, August 03, 2016 11:52 AM
> To: amd-gfx@lists.freedesktop.org
> Subject: VCE3 MGCG for Stoney/Carrizo
>
> These patches fix the MGCG initialization as well
Hi Alex,
Ok, a similar change was introduced for UVD in
7a6526488b60283cec91852196a5423b1dfb14d8, should that be reverted in part too?
Tom
From: Deucher, Alexander
Sent: Wednesday, August 3, 2016 12:27
To: 'Tom St Denis'; amd-gfx@lists.freedesktop.org
Cc:
Partially undo changes made by commit:
drm/amd/amdgpu: don't track state in UVD clockgating
To keep bypass even if CG flags are not set.
Signed-off-by: Tom St Denis
---
drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff
fence_put was called on an uninitialized variable.
Signed-off-by: Jay Cornwall
---
drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
Am 03.08.2016 um 20:39 schrieb Jay Cornwall:
fence_put was called on an uninitialized variable.
Signed-off-by: Jay Cornwall
Good catch, patch is Reviewed-by: Christian König
Regards,
Christian.
---
drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c | 3
Patch #1 looks like a fix to me, please add my rb and push ASAP.
You should separate such fixes from bugger sets, send out individually
and maybe ping Alex to review and add them to his fixes branch.
Patch #2:
+ amdgpu_ttm_placement_init(adev, ,
+
Well that is a clear NAK to this whole approach.
Submitting the recovery jobs to the scheduler is reentrant because the
scheduler is the one who originally signaled us of a timeout.
Why not submit the recovery jobs to the hardware ring directly?
Regards,
Christian.
Am 28.07.2016 um 12:13
Hi,
I didn't see any of my previous review remarks on your previous
submission addresses in this submission of this series?
Kind Regards,
Edward.
On 08/02/2016 05:51 PM, Chunming Zhou wrote:
> every vm has itself recovery entity, which is used to reovery page table from
> their shadow.
> They
Well, usually they are just the recommended settings from the hardware
team which try to fix or improve things by changing the default values
for the registers.
I work for AMD and sometimes it isn't clear to me either what those
changes are good for. E.g. I would need to dig though a whole
Am 01.08.2016 um 15:59 schrieb Alex Deucher:
On Mon, Aug 1, 2016 at 1:27 AM, Flora Cui wrote:
I'll push if no objection
Reviewed-by: Alex Deucher
Sorry for the delay, been on vacation last week.
Patch is Reviewed-by: Christian König
Reviewed-by: Christian König for the whole
series.
Regards,
Christian.
Am 03.08.2016 um 09:51 schrieb Chunming Zhou:
VCE is always busy is since RB1_BUSY is always busy, now remove it.
And enable VCE soft reset.
Chunming Zhou (2):
drm/amdgpu: fix harvest config
I forgot to say, if this is clear to everyone else but me you can just
ignore my noise here :p
On 08/03/2016 10:06 PM, Edward O'Callaghan wrote:
> Hi,
>
> I think it would be helpful for at least one liner explaining actually
> what is even getting "fixed" here? Obviously we assume, given that
On 2016年08月03日 22:01, Christian König wrote:
Well patch #10 is incorrect. The SA BO will be set to NULL by
amdgpu_sa_bo_free(), so it can't be freed twice and so you can't
reference the fence twice.
I see.
But amdgpu_job_free_resources still shouldn't be called twice, right?
That's an
51 matches
Mail list logo