[PATCH] drm/amdgpu: Apply noretry setting for gfx10 and mmhub9.4

[Zeng, Oak]

I didn't have new information about the retry_mode setting for gfx10. I just saw the noretry kernel parameter wasn't applied to gfx10 but I didn't know gfx10 is excluded purposely. Let me drop the gfx10 part for now.

Oak

-----Original Message-----
From: Kuehling, Felix <Felix.Kuehling at amd.com> 
Sent: Friday, November 22, 2019 3:51 PM
To: Zeng, Oak <Oak.Zeng at amd.com>; amd-gfx at lists.freedesktop.org
Cc: Cornwall, Jay <Jay.Cornwall at amd.com>
Subject: Re: [PATCH] drm/amdgpu: Apply noretry setting for gfx10 and mmhub9.4


On 2019-11-22 3:23 p.m., Oak Zeng wrote:
> Config the translation retry behavior according to noretry kernel 
> parameter
>
> Change-Id: I5b91ea77715137cf8cb84e258ccdfbb19c7a4ed1
> Signed-off-by: Oak Zeng <Oak.Zeng at amd.com>
> Suggested-by: Jay Cornwall <Jay.Cornwall at amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 4 +++-
>   drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c | 5 +++--
>   2 files changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c 
> b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> index 4d6df35..7435487 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> @@ -1751,7 +1751,9 @@ static void gfx_v10_0_constants_init(struct amdgpu_device *adev)
>   	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
>   		nv_grbm_select(adev, 0, 0, 0, i);
>   		/* CP and shaders */
> -		WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
> +		tmp = REG_SET_FIELD(DEFAULT_SH_MEM_CONFIG, SH_MEM_CONFIG,
> +				RETRY_MODE, amdgpu_noretry ? 2 : 0);
> +		WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, tmp);
I didn't fix gfx_v10_0 on purpose last time I worked on this
("75ee64875e75 drm/amdkfd: Consistently apply noretry setting"), because the retry works differently on GFXv10. Do you have new information about how the RETRY_MODE setting works on GFXv10? Does it affect performance?


>   		if (i != 0) {
>   			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
>   				(adev->gmc.private_aperture_start >> 48)); diff --git 
> a/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c 
> b/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c
> index 753eea2..8599bfd 100644
> --- a/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c
> +++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c
> @@ -314,7 +314,8 @@ static void mmhub_v9_4_setup_vmid_config(struct amdgpu_device *adev, int hubid)
>   				    adev->vm_manager.block_size - 9);
>   		/* Send no-retry XNACK on fault to suppress VM fault storm. */
>   		tmp = REG_SET_FIELD(tmp, VML2VC0_VM_CONTEXT1_CNTL,
> -				    RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
> +				    RETRY_PERMISSION_OR_INVALID_PAGE_FAULT,
> +				    !amdgpu_noretry);

Right, this was missed in my previous commit because mmhub_v9_4 was added later.

Regards,
   Felix


>   		WREG32_SOC15_OFFSET(MMHUB, 0, mmVML2VC0_VM_CONTEXT1_CNTL,
>   				    hubid * MMHUB_INSTANCE_REGISTER_OFFSET + i,
>   				    tmp);
> @@ -905,4 +906,4 @@ static void mmhub_v9_4_query_ras_error_count(struct amdgpu_device *adev,
>   const struct amdgpu_mmhub_funcs mmhub_v9_4_funcs = {
>   	.ras_late_init = amdgpu_mmhub_ras_late_init,
>   	.query_ras_error_count = mmhub_v9_4_query_ras_error_count, -}; \ No 
> newline at end of file
> +};