[pull] amdgpu, amdkfd drm-fixes-5.10

Hi Dave, Daniel, Fixes for 5.10. The support for green sardine is not exactly a fix, but the changes are a minimal delta relative to renoir and should not affect any other asics. The following changes since commit 7babd126327b8b5a3904d2f8f01c95235801af2a: Merge tag 'drm-intel-fixes-2020-10-29

Re: [bugreport] [5.10-rc1] Oops: 0000 [#1] SMP NOPTI bug which always starts as page allocation failure

On Tue, Nov 3, 2020 at 4:05 PM Mikhail Gavrilov wrote: > > Hi folks. > I observed hard reproductible the set of bugs. > It always started as > 1) kworker/u64:2: page allocation failure: order:5, > mode:0x40dc0(GFP_KERNEL|__GFP_COMP|__GFP_ZERO), > nodemask=(null),cpuset=/,mems_allowed=0 > Continiou

Re: [PATCH] drm/amdkfd: replace idr_init() by idr_init_base()

On 2020-11-04 10:13 a.m., Deepak R Varma wrote: idr_init() uses base 0 which is an invalid identifier. The new function idr_init_base allows IDR to set the ID lookup from base 1. This avoids all lookups that otherwise starts from 0 since 0 is always unused. I disagree. We call idr_alloc with st

Re: [PATCH 0/2] amdgpu's drm_driver becomes const

On 2020-11-04 04:43, Daniel Vetter wrote: > On Tue, Nov 03, 2020 at 10:11:27PM -0500, Luben Tuikov wrote: >> Hi Daniel, >> >> These two patches follow up your latest >> DRM work to make definitions of struct drm_driver >> in DRM low-level drivers, constant, in amdgpu. >> >> This set doesn't descend

Re: [PATCH] drm/amdgpu: replace idr_init() by idr_init_base()

Am 04.11.20 um 16:16 schrieb Deepak R Varma: idr_init() uses base 0 which is an invalid identifier. The new function idr_init_base allows IDR to set the ID lookup from base 1. This avoids all lookups that otherwise starts from 0 since 0 is always unused. References: commit 6ce711f27500 ("idr: Ma

Re: [PATCH] drm/amdkfd: replace idr_init() by idr_init_base()

Am 04.11.20 um 16:13 schrieb Deepak R Varma: idr_init() uses base 0 which is an invalid identifier. The new function idr_init_base allows IDR to set the ID lookup from base 1. This avoids all lookups that otherwise starts from 0 since 0 is always unused. References: commit 6ce711f27500 ("idr: Ma

[PATCH] drm/amdgpu: replace idr_init() by idr_init_base()

idr_init() uses base 0 which is an invalid identifier. The new function idr_init_base allows IDR to set the ID lookup from base 1. This avoids all lookups that otherwise starts from 0 since 0 is always unused. References: commit 6ce711f27500 ("idr: Make 1-based IDRs more efficient") Signed-off-by

[PATCH] drm/amdkfd: replace idr_init() by idr_init_base()

idr_init() uses base 0 which is an invalid identifier. The new function idr_init_base allows IDR to set the ID lookup from base 1. This avoids all lookups that otherwise starts from 0 since 0 is always unused. References: commit 6ce711f27500 ("idr: Make 1-based IDRs more efficient") Signed-off-by

RE: [PATCH] drm/amdgpu: clip the ref divider max value at 100

[AMD Public Use] Thanks Alex, Christian, Let me quickly run this patch through someone in DAL team, and see how it looks. Regards Shashank From: Deucher, Alexander Sent: Wednesday, November 4, 2020 8:03 PM To: Koenig, Christian ; Sharma, Shashank ; amd-gfx@lists.freedesktop.org Cc: Qin, Eddy

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

Then feel free to stick an Acked-by: Christian König to the code and get it committed. Ideally somebody should have it on the TODO list to get rid of that code. Christian. Am 04.11.20 um 15:33 schrieb Deucher, Alexander: yeah, ideally.  Just need to get support for analog encoders. Alex --

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

yeah, ideally. Just need to get support for analog encoders. Alex From: Koenig, Christian Sent: Wednesday, November 4, 2020 9:31 AM To: Deucher, Alexander ; Sharma, Shashank ; amd-gfx@lists.freedesktop.org Cc: Qin, Eddy Subject: Re: [PATCH] drm/amdgpu: clip

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

In the long term we probably want to nuke this code anyway and switch to the DC code, don't we? Christian. Am 04.11.20 um 15:23 schrieb Deucher, Alexander: You might want to talk to the DAL team, they may have some advice.  In general, I would say test it as well as you can. It's probably safe

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

You might want to talk to the DAL team, they may have some advice. In general, I would say test it as well as you can. It's probably safe as radeon is still the default driver for SI parts and generally seems to be working well there. Alex From: Sharma, Shasha

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

Thanks Alex, Same question here, Should we go through some extensive test routine due to change in PLL values, or is it OK to go ahead based on our experience from Radeon values ? Regards Shashank On 04/11/20 7:36 pm, Deucher, Alexander wrote: > Acked-by: Alex Deucher >

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

On 04/11/20 5:24 pm, Christian König wrote: > Am 04.11.20 um 11:40 schrieb Sharma, Shashank: >> [AMD Public Use] >> >> Hello Christian, >> Yes, that 100 is hardcoded in Radeon, and git blame says it was one of your >> patches which made it 100 from 128 😊. >> Would you mind having a look at commit

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

Acked-by: Alex Deucher From: Koenig, Christian Sent: Wednesday, November 4, 2020 6:54 AM To: Sharma, Shashank ; amd-gfx@lists.freedesktop.org Cc: Deucher, Alexander ; Qin, Eddy Subject: Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100 Am 04.11.20

Re: [PATCH] drm/amdgpu/virt: fix handling of the atomic flag

On Wed, Nov 4, 2020 at 4:39 AM Daniel Vetter wrote: > > On Tue, Nov 03, 2020 at 04:54:50PM -0500, Alex Deucher wrote: > > Use the per device drm driver feature flags rather than the > > global one. This way we can make the drm driver struct const. > > > > Signed-off-by: Alex Deucher > > Reviewed

RE: [PATCH] drm/amdgpu: add missing clock gating info in amdgpu_pm_info

[AMD Public Use] Reviewed-by: Hawking Zhang Regards, Hawking -Original Message- From: Wang, Kevin(Yang) Sent: Wednesday, November 4, 2020 15:36 To: amd-gfx@lists.freedesktop.org Cc: Zhang, Hawking ; Deucher, Alexander ; Wang, Kevin(Yang) Subject: [PATCH] drm/amdgpu: add missing clock

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

Am 04.11.20 um 11:40 schrieb Sharma, Shashank: [AMD Public Use] Hello Christian, Yes, that 100 is hardcoded in Radeon, and git blame says it was one of your patches which made it 100 from 128 😊. Would you mind having a look at commit id: 4b21ce1b4b5d262e7d4656b8ececc891fc3cb806 ? Ah, yes tha

RE: [PATCH] drm/amdgpu: clip the ref divider max value at 100

[AMD Public Use] Hello Christian, Yes, that 100 is hardcoded in Radeon, and git blame says it was one of your patches which made it 100 from 128 😊. Would you mind having a look at commit id: 4b21ce1b4b5d262e7d4656b8ececc891fc3cb806 ? Regards Shashank -Original Message- From: Koenig,

RE: [PATCH] drm/amdgpu: add missing clock gating info in amdgpu_pm_info

[AMD Official Use Only - Internal Distribution Only] Reviewed-by: Kenneth Feng Best Regards Kenneth -Original Message- From: amd-gfx On Behalf Of Kevin Wang Sent: Wednesday, November 4, 2020 3:36 PM To: amd-gfx@lists.freedesktop.org Cc: Deucher, Alexander ; Wang, Kevin(Yang) ; Zhang,

Re: [PATCH] drm/amdgpu: clip the ref divider max value at 100

Am 03.11.20 um 18:13 schrieb Shashank Sharma: This patch limits the ref_div_max value to 100, during the calculation of PLL feedback reference divider. With current value (128), the produced fb_ref_div value generates unstable output at particular frequencies. Radeon driver limits this value at 1

[PATCH 1/6] drm/radeon: Stop changing the drm_driver struct

With only the kms driver left, we can fold this in. This means we need to move the ioctl table, which means one additional ioctl must be defined in headers. Also there's a conflict between the radeon_init macro and the module init function, so rename the module functions to avoid that. Reviewed-b

Re: [PATCH 2/2] drm/amdgpu: Make struct drm_driver const

On Tue, Nov 03, 2020 at 10:11:29PM -0500, Luben Tuikov wrote: > Make the definition of struct drm_driver > a constant, to follow the latest developments > in the DRM layer. > > Signed-off-by: Luben Tuikov > --- > drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 32 + > drivers/g

Re: [PATCH] drm/amdgpu: fix NULL pointer crash on navi10 headless SKU

[AMD Official Use Only - Internal Distribution Only] Hi Alex, With DCN enabled, I take amdgpu_test/kfdtest/PhinoexMiner/S3 tests, they are all PASS. I ask @Long, Gang to take a large coverage test, if it's ok, I will enable DCN block. Thanks! Rico __

Re: [PATCH 0/2] amdgpu's drm_driver becomes const

On Tue, Nov 03, 2020 at 10:11:27PM -0500, Luben Tuikov wrote: > Hi Daniel, > > These two patches follow up your latest > DRM work to make definitions of struct drm_driver > in DRM low-level drivers, constant, in amdgpu. > > This set doesn't descend from my previous patch > "drm/amdgpu: Convert to

Re: [PATCH] drm/amdgpu/virt: fix handling of the atomic flag

On Tue, Nov 03, 2020 at 04:54:50PM -0500, Alex Deucher wrote: > Use the per device drm driver feature flags rather than the > global one. This way we can make the drm driver struct const. > > Signed-off-by: Alex Deucher Reviewed-by: Daniel Vetter Can we merge this through drm-misc-next so I c