Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-14 Thread Felix Kuehling
On 2023-12-14 10:06, Alex Deucher wrote: On Thu, Dec 14, 2023 at 9:24 AM Liu, Shaoyun wrote: [AMD Official Use Only - General] The gmc flush tlb function is used on both baremetal and sriov. But the function amdgpu_virt_kiq_reg_write_reg_wait is defined in amdgpu_virt.c with name

Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-14 Thread Mario Limonciello
On 12/14/2023 03:21, Christian König wrote: Am 13.12.23 um 20:44 schrieb Alex Deucher: On Wed, Dec 13, 2023 at 2:32 PM Mario Limonciello wrote: On 12/13/2023 13:12, Mario Limonciello wrote: On 12/13/2023 13:07, Alex Deucher wrote: On Wed, Dec 13, 2023 at 1:00 PM Mario Limonciello wrote:

RE: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-14 Thread Liu, Shaoyun
[AMD Official Use Only - General] I remembered we try to use kiq directly send the invalidation packet to mec FW with pasid as parameter since it's FW control the VMID and PASID mapping , but during some test , it shows performance drop compare to driver directly get the vmid/pasid

Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-14 Thread Alex Deucher
On Thu, Dec 14, 2023 at 9:24 AM Liu, Shaoyun wrote: > > [AMD Official Use Only - General] > > The gmc flush tlb function is used on both baremetal and sriov. But the > function amdgpu_virt_kiq_reg_write_reg_wait is defined in amdgpu_virt.c with > name 'virt' make it appear as a SRIOV

RE: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-14 Thread Liu, Shaoyun
[AMD Official Use Only - General] The gmc flush tlb function is used on both baremetal and sriov. But the function amdgpu_virt_kiq_reg_write_reg_wait is defined in amdgpu_virt.c with name 'virt' make it appear as a SRIOV only function, this sounds confusion . Will it make more sense to

Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-14 Thread Christian König
Am 13.12.23 um 20:44 schrieb Alex Deucher: On Wed, Dec 13, 2023 at 2:32 PM Mario Limonciello wrote: On 12/13/2023 13:12, Mario Limonciello wrote: On 12/13/2023 13:07, Alex Deucher wrote: On Wed, Dec 13, 2023 at 1:00 PM Mario Limonciello wrote: Some systems with MP1 13.0.4 or 13.0.11 have a

Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-13 Thread Alex Deucher
On Wed, Dec 13, 2023 at 2:32 PM Mario Limonciello wrote: > > On 12/13/2023 13:12, Mario Limonciello wrote: > > On 12/13/2023 13:07, Alex Deucher wrote: > >> On Wed, Dec 13, 2023 at 1:00 PM Mario Limonciello > >> wrote: > >>> > >>> Some systems with MP1 13.0.4 or 13.0.11 have a firmware bug that

Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-13 Thread Mario Limonciello
On 12/13/2023 13:12, Mario Limonciello wrote: On 12/13/2023 13:07, Alex Deucher wrote: On Wed, Dec 13, 2023 at 1:00 PM Mario Limonciello wrote: Some systems with MP1 13.0.4 or 13.0.11 have a firmware bug that causes the first MES packet after resume to fail. This packet is used to flush the

Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-13 Thread Mario Limonciello
On 12/13/2023 13:07, Alex Deucher wrote: On Wed, Dec 13, 2023 at 1:00 PM Mario Limonciello wrote: Some systems with MP1 13.0.4 or 13.0.11 have a firmware bug that causes the first MES packet after resume to fail. This packet is used to flush the TLB when GART is enabled. This issue is fixed

Re: [PATCH] drm/amd: Add a workaround for GFX11 systems that fail to flush TLB

2023-12-13 Thread Alex Deucher
On Wed, Dec 13, 2023 at 1:00 PM Mario Limonciello wrote: > > Some systems with MP1 13.0.4 or 13.0.11 have a firmware bug that > causes the first MES packet after resume to fail. This packet is > used to flush the TLB when GART is enabled. > > This issue is fixed in newer firmware, but as OEMs may