Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <c3ac5d95-218f-4f36-fbec-0172af69d5a0@molgen.mpg.de>
Date:   Mon, 21 Feb 2022 11:46:34 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.6.1
Subject: Re: [PATCH] drm/amdgpu: check vm ready by evicting
Content-Language: en-US
To:     Qiang Yu <qiang.yu@amd.com>
Cc:     Alex Deucher <alexander.deucher@amd.com>,
        =?UTF-8?Q?Christian_K=c3=b6nig?= <christian.koenig@amd.com>,
        "Pan, Xinhui" <Xinhui.Pan@amd.com>,
        David Airlie <airlied@linux.ie>,
        Daniel Vetter <daniel@ffwll.ch>,
        dri-devel@lists.freedesktop.org, amd-gfx@lists.freedesktop.org,
        linux-kernel@vger.kernel.org
References: <20220221101239.2863-1-qiang.yu@amd.com>
From:   Paul Menzel <pmenzel@molgen.mpg.de>
In-Reply-To: <20220221101239.2863-1-qiang.yu@amd.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Dear Qiang Yu,


Am 21.02.22 um 11:12 schrieb Qiang Yu:


Thank you for your patch. Reading the commit message summary, I have no 
idea what “check vm ready by evicting” means. Can you please rephrase it?

> Workstation application ANSA/META get this error dmesg:

What version, and how can this be reproduced exactly? Just by starting 
the application?

> [drm:amdgpu_gem_va_ioctl [amdgpu]] *ERROR* Couldn't update BO_VA (-16)
> 
> This is caused by:
> 1. create a 256MB buffer in invisible VRAM
> 2. CPU map the buffer and access it causes vm_fault and try to move
>     it to visible VRAM
> 3. force visible VRAM space and traverse all VRAM bos to check if
>     evicting this bo is valuable
> 4. when checking a VM bo (in invisible VRAM), amdgpu_vm_evictable()
>     will set amdgpu_vm->evicting, but latter due to not in visible
>     VRAM, won't really evict it so not add it to amdgpu_vm->evicted
> 5. before next CS to clear the amdgpu_vm->evicting, user VM ops
>     ioctl will pass amdgpu_vm_ready() (check amdgpu_vm->evicted)
>     but fail in amdgpu_vm_bo_update_mapping() (check
>     amdgpu_vm->evicting) and get this error log
> 
> This error won't affect functionality as next CS will finish the
> waiting VM ops. But we'd better clear the error log by check the

s/check/checking/

> evicting flag which really stop VM ops latter.

stop*s*?

Can you please elaborate. Christian’s and your discussions was quite 
long, so adding a summary, why this approach works and what possible 
regressions there are going to be might be warranted.


Kind regards,

Paul


> Signed-off-by: Qiang Yu <qiang.yu@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 9 +++++++--
>   1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index 37acd8911168..2cd9f1a2e5fa 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -770,11 +770,16 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
>    * Check if all VM PDs/PTs are ready for updates
>    *
>    * Returns:
> - * True if eviction list is empty.
> + * True if VM is not evicting.
>    */
>   bool amdgpu_vm_ready(struct amdgpu_vm *vm)
>   {
> -	return list_empty(&vm->evicted);
> +	bool ret;
> +
> +	amdgpu_vm_eviction_lock(vm);
> +	ret = !vm->evicting;
> +	amdgpu_vm_eviction_unlock(vm);
> +	return ret;
>   }
>   
>   /**