Re: [PATCH] drm/amdgpu/psp: keep TMR in visible vram region for SRIOV

["Yin, Tianci (Rico)" <Tianci.Yin@xxxxxxx>]

Great! That's good idea!

Thanks!

---

From: Koenig, Christian <Christian.Koenig@xxxxxxx>
Sent: Wednesday, August 28, 2019 18:25
To: Yin, Tianci (Rico) <Tianci.Yin@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
Cc: Xu, Feifei <Feifei.Xu@xxxxxxx>; Ma, Le <Le.Ma@xxxxxxx>; Xiao, Jack <Jack.Xiao@xxxxxxx>; Zhang, Hawking <Hawking.Zhang@xxxxxxx>
Subject: Re: [PATCH] drm/amdgpu/psp: keep TMR in visible vram region for SRIOV

 

It seems we have to add a new parameter for amdgpu_bo_create_kernel() to pass the prefer region.

You can also just pass a cpu pointer to the function allocating the stolen_vga_memory.

That's probably less work to change all callers and cleaner in general.

Christian.

Am 28.08.19 um 12:21 schrieb Yin, Tianci (Rico):

Hi Christian,

I have just realized the stolen_vga_memory problem,

I have ever thought that no one really use the stolen_vga_memory BO in KMD code, so it should be safe.

It seems we have to add a new parameter for amdgpu_bo_create_kernel() to pass the prefer region.

Thanks!

Rico

---

From: Christian König <ckoenig.leichtzumerken@xxxxxxxxx>
Sent: Wednesday, August 28, 2019 17:57
To: Yin, Tianci (Rico) <Tianci.Yin@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
Cc: Xu, Feifei <Feifei.Xu@xxxxxxx>; Ma, Le <Le.Ma@xxxxxxx>; Xiao, Jack <Jack.Xiao@xxxxxxx>; Zhang, Hawking <Hawking.Zhang@xxxxxxx>
Subject: Re: [PATCH] drm/amdgpu/psp: keep TMR in visible vram region for SRIOV

 

Am 28.08.19 um 11:25 schrieb Tianci Yin:
> From: "Tianci.Yin" <tianci.yin@xxxxxxx>
>
> Fix compute ring test failure in sriov scenario.
>
> Signed-off-by: Tianci.Yin <tianci.yin@xxxxxxx>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 14 ++++++++++----
>   drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h |  1 +
>   2 files changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
> index 9f7cc5b..92c68c9 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
> @@ -261,9 +261,15 @@ static int psp_tmr_init(struct psp_context *psp)
>                }
>        }
>  
> -     ret = amdgpu_bo_create_kernel(psp->adev, tmr_size, PSP_TMR_SIZE,
> -                                   AMDGPU_GEM_DOMAIN_VRAM,
> -                                   &psp->tmr_bo, &psp->tmr_mc_addr, NULL);
> +     psp->tmr_buf = NULL;
> +     if (!amdgpu_sriov_vf(psp->adev))
> +             ret = amdgpu_bo_create_kernel(psp->adev, tmr_size, PSP_TMR_SIZE,
> +                                           AMDGPU_GEM_DOMAIN_VRAM,
> +                                           &psp->tmr_bo, &psp->tmr_mc_addr, NULL);
> +     else
> +             ret = amdgpu_bo_create_kernel(psp->adev, tmr_size, PSP_TMR_SIZE,
> +                                           AMDGPU_GEM_DOMAIN_VRAM,
> +                                           &psp->tmr_bo, &psp->tmr_mc_addr, &psp->tmr_buf);

Might be better to use a local variable for the CPU pointer instead of
duplicating the calling parameters.

BTW: How did you solved the stolen_vga_memory problem I pointed out in
the original patch set?

Christian.

>  
>        return ret;
>   }
> @@ -1216,7 +1222,7 @@ static int psp_hw_fini(void *handle)
>  
>        psp_ring_destroy(psp, PSP_RING_TYPE__KM);
>  
> -     amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr, NULL);
> +     amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr, &psp->tmr_buf);
>        amdgpu_bo_free_kernel(&psp->fw_pri_bo,
>                              &psp->fw_pri_mc_addr, &psp->fw_pri_buf);
>        amdgpu_bo_free_kernel(&psp->fence_buf_bo,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
> index bc0947f..b73d4aa 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
> @@ -171,6 +171,7 @@ struct psp_context
>        /* tmr buffer */
>        struct amdgpu_bo                *tmr_bo;
>        uint64_t                        tmr_mc_addr;
> +     void                            *tmr_buf;
>  
>        /* asd firmware and buffer */
>        const struct firmware           *asd_fw;

_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx