On Wed, Apr 6, 2022 at 10:03 AM Yongqiang Sun <yongqiang.sun@xxxxxxx> wrote:
>
> driver loading failed on VEGA10 SRIOV VF with linux host due to a wide
> range of stolen reserved vram.
> Since VEGA10 SRIOV VF need to reserve vram for firmware with windows
> Hyper_V host specifically, check high part of FB address
> to only reserve memory for it, and the range of the reserved vram
> can be limited to between 5M-7M area.
I'm not super familiar with how we set up things on the hypervisor
side, but I'll take your word for it that this would only affect
hyperv. Please add:
Fixes: 8011a76ad370 ("drm/amdgpu: Add stolen reserved memory for MI25 SRIOV.")
With that:
Acked-by: Alex Deucher <alexander.deucher@xxxxxxx>
>
> Signed-off-by: Yongqiang Sun <yongqiang.sun@xxxxxxx>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c | 9 +++++----
> drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h | 3 +++
> 2 files changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> index a66a0881a934..324b43c1bc4c 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> @@ -647,11 +647,12 @@ void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev)
> case CHIP_VEGA10:
> adev->mman.keep_stolen_vga_memory = true;
> /*
> - * VEGA10 SRIOV VF needs some firmware reserved area.
> + * VEGA10 SRIOV VF with hyper-v host needs some firmware reserved area.
> */
> - if (amdgpu_sriov_vf(adev)) {
> - adev->mman.stolen_reserved_offset = 0x100000;
> - adev->mman.stolen_reserved_size = 0x600000;
> + if (amdgpu_sriov_vf(adev) &&
> + (adev->gmc.vram_start & AMDGPU_GMC_HYPERV_FB_BASE_MASK)) {
> + adev->mman.stolen_reserved_offset = 0x500000;
> + adev->mman.stolen_reserved_size = 0x200000;
> }
> break;
> case CHIP_RAVEN:
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> index 032b0313f277..906ec15e8dac 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> @@ -61,6 +61,9 @@
> */
> #define AMDGPU_GMC_FAULT_TIMEOUT 5000ULL
>
> +/* High part (33 - 47bit) FB address for VEGA10 HYPERV Host*/
> +#define AMDGPU_GMC_HYPERV_FB_BASE_MASK 0xF400000000ULL
> +
> struct firmware;
>
> /*
> --
> 2.25.1
>
