From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 6CA7CC3DA5D for ; Mon, 22 Jul 2024 08:20:55 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id D881189321; Mon, 22 Jul 2024 08:20:54 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=gmail.com header.i=@gmail.com header.b="R7ieokFH"; dkim-atps=neutral Received: from mail-wm1-f42.google.com (mail-wm1-f42.google.com [209.85.128.42]) by gabe.freedesktop.org (Postfix) with ESMTPS id E40FA89321 for ; Mon, 22 Jul 2024 08:20:49 +0000 (UTC) Received: by mail-wm1-f42.google.com with SMTP id 5b1f17b1804b1-426526d30aaso30148775e9.0 for ; Mon, 22 Jul 2024 01:20:49 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20230601; t=1721636448; x=1722241248; darn=lists.freedesktop.org; h=in-reply-to:from:content-language:references:to:subject:user-agent :mime-version:date:message-id:from:to:cc:subject:date:message-id :reply-to; bh=sfYvH02SOak2/ITQpmxFlX0FRXLy2rILFtiVgF4HiDc=; b=R7ieokFHM3kmR+yqga2OKR9kUYBBazMYHtvEsoAqVu3F16G+wz7B0p44aMeMFeIlXk sO6vp0nngRHCvrjoRPBMw32Jd0UXspld8GhBnbM7TO9+TngZScdBj7EFYzcoIL/9thi5 1svM6cLnOO+PFboSKJWsp6Ghfsi9bRJreXhcTCxzIqseop28ZyM1GGsXmIpy9tTVA3Q7 Lx+DR1BsR5lmnWENqgeD3QgF5XVhlqXxDWNXy7FUCjP8URBC7747FspaN7w8fgNXumov XX9pJ/hcg/eVpFsXO7JYdsxYwlzQHoH0CTm7uTEVFveA2sN/K9JqYo5sVO4Xt6S+T0a8 QMvw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1721636448; x=1722241248; h=in-reply-to:from:content-language:references:to:subject:user-agent :mime-version:date:message-id:x-gm-message-state:from:to:cc:subject :date:message-id:reply-to; bh=sfYvH02SOak2/ITQpmxFlX0FRXLy2rILFtiVgF4HiDc=; b=H4V6vrdP7H3i7P/4MluAzFag2247uRtPViJBvhPDfZhBUoyTrGkHf11O5kzc7aEOUz GsQcgc41vQz8rKsAgcrNFKkOAYzz/TiZwtxfLClhwx9W75Z24OOB1Ew6W+gPOzumyVW5 cDKgHz1Eg+IGEfYEdSpQzY7PHrq+rLwVgtYKsAgpVWuldbScMDz5bi/N+FZJSpSefx53 +h2ZCM0b1yA5w4SROzyvt1CtmgCSkjMCoTBDeqgq7KiuwVMQkAaASZTdYfPl1qzuFasg 2AJIxFmL8vqOHmGq4zPQA17pNyFORLHWVK5QOFHIJjqFfqo4JDROxQUfOl3ppvuwFi2B ni2Q== X-Forwarded-Encrypted: i=1; AJvYcCWAXD22uvOaPLLvX4Bmalbk+aaoJOyIxt+3pq3kMmKyqr8P06bDCH4Qkt3rjttNyY/+2SeZcyMhiso+zSZTPOmd2hKlFb8KZC9sGJ/GDg== X-Gm-Message-State: AOJu0YzNqmZ8fWRospBuSz6iKed05SFqxaaJb7BPRjK9Z+/mTx7Of4/F bVz4ttMCpY93aHfXYtdSKN1Nb5pvw8EgcdHHrSqwV7dZ1K+hOa7E X-Google-Smtp-Source: AGHT+IH5GbqvXN2QozO5vA/ACLfvV3r11OIQ/6PYLyiFYKUOjuQmlbf+MsT2ZAeL3k8wa2Btij7r3A== X-Received: by 2002:a05:600c:1550:b0:426:5cdf:2674 with SMTP id 5b1f17b1804b1-427dc51a073mr39452395e9.4.1721636447350; Mon, 22 Jul 2024 01:20:47 -0700 (PDT) Received: from [10.254.108.81] (munvpn.amd.com. [165.204.72.6]) by smtp.gmail.com with ESMTPSA id ffacd0b85a97d-368786944c8sm7786418f8f.54.2024.07.22.01.20.46 (version=TLS1_3 cipher=TLS_AES_128_GCM_SHA256 bits=128/128); Mon, 22 Jul 2024 01:20:46 -0700 (PDT) Content-Type: multipart/alternative; boundary="------------Q0amzg4GUnZMtc0JlovGnSCi" Message-ID: <92f00b1c-79c2-431b-b8dc-70070a65b14a@gmail.com> Date: Mon, 22 Jul 2024 10:20:45 +0200 MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Subject: Re: [PATCH] drm/amdgpu/mes: fix mes ring buffer overflow To: "Xiao, Jack" , "amd-gfx@lists.freedesktop.org" , "Deucher, Alexander" References: <20240719091615.1534436-1-Jack.Xiao@amd.com> Content-Language: en-US From: =?UTF-8?Q?Christian_K=C3=B6nig?= In-Reply-To: X-BeenThere: amd-gfx@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Discussion list for AMD gfx List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: amd-gfx-bounces@lists.freedesktop.org Sender: "amd-gfx" This is a multi-part message in MIME format. --------------Q0amzg4GUnZMtc0JlovGnSCi Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit Thx, but in that case this patch here then won't help either it just mitigates the problem. Can you try to reduce num_hw_submission for the MES ring? Thanks, Christian. Am 22.07.24 um 05:27 schrieb Xiao, Jack: > > [AMD Official Use Only - AMD Internal Distribution Only] > > > />> I think we rather need to increase the MES ring size instead./ > > Unfortunately, it doesn't work. I guess mes firmware has limitation. > > Regards, > Jack > > ------------------------------------------------------------------------ > *From:* Christian König > *Sent:* Friday, 19 July 2024 23:44 > *To:* Xiao, Jack ; amd-gfx@lists.freedesktop.org > ; Deucher, Alexander > > *Subject:* Re: [PATCH] drm/amdgpu/mes: fix mes ring buffer overflow > Am 19.07.24 um 11:16 schrieb Jack Xiao: > > wait memory room until enough before writing mes packets > > to avoid ring buffer overflow. > > > > Signed-off-by: Jack Xiao > > --- > >   drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 18 ++++++++++++++---- > >   drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 18 ++++++++++++++---- > >   2 files changed, 28 insertions(+), 8 deletions(-) > > > > diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c > b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c > > index 8ce51b9236c1..68c74adf79f1 100644 > > --- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c > > +++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c > > @@ -168,7 +168,7 @@ static int > mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        const char *op_str, *misc_op_str; > >        unsigned long flags; > >        u64 status_gpu_addr; > > -     u32 status_offset; > > +     u32 seq, status_offset; > >        u64 *status_ptr; > >        signed long r; > >        int ret; > > @@ -196,6 +196,13 @@ static int > mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        if (r) > >                goto error_unlock_free; > > > > +     seq = ++ring->fence_drv.sync_seq; > > +     r = amdgpu_fence_wait_polling(ring, > > +                                   seq - > ring->fence_drv.num_fences_mask, > > +                                   timeout); > > +     if (r < 1) > > +             goto error_undo; > > + > >        api_status = (struct MES_API_STATUS *)((char *)pkt + > api_status_off); > >        api_status->api_completion_fence_addr = status_gpu_addr; > >        api_status->api_completion_fence_value = 1; > > @@ -208,8 +215,7 @@ static int > mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        mes_status_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS; > > mes_status_pkt.api_status.api_completion_fence_addr = > >                ring->fence_drv.gpu_addr; > > - mes_status_pkt.api_status.api_completion_fence_value = > > -             ++ring->fence_drv.sync_seq; > > + mes_status_pkt.api_status.api_completion_fence_value = seq; > > > >        amdgpu_ring_write_multiple(ring, &mes_status_pkt, > > sizeof(mes_status_pkt) / 4); > > @@ -229,7 +235,7 @@ static int > mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >                dev_dbg(adev->dev, "MES msg=%d was emitted\n", > >                        x_pkt->header.opcode); > > > > -     r = amdgpu_fence_wait_polling(ring, ring->fence_drv.sync_seq, > timeout); > > +     r = amdgpu_fence_wait_polling(ring, seq, timeout); > >        if (r < 1 || !*status_ptr) { > > > >                if (misc_op_str) > > @@ -252,6 +258,10 @@ static int > mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        amdgpu_device_wb_free(adev, status_offset); > >        return 0; > > > > +error_undo: > > +     dev_err(adev->dev, "MES ring buffer is full.\n"); > > +     amdgpu_ring_undo(ring); > > + > >   error_unlock_free: > > spin_unlock_irqrestore(&mes->ring_lock, flags); > > > > diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c > b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c > > index c9f74231ad59..48e01206bcc4 100644 > > --- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c > > +++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c > > @@ -154,7 +154,7 @@ static int > mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        const char *op_str, *misc_op_str; > >        unsigned long flags; > >        u64 status_gpu_addr; > > -     u32 status_offset; > > +     u32 seq, status_offset; > >        u64 *status_ptr; > >        signed long r; > >        int ret; > > @@ -182,6 +182,13 @@ static int > mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        if (r) > >                goto error_unlock_free; > > > > +     seq = ++ring->fence_drv.sync_seq; > > +     r = amdgpu_fence_wait_polling(ring, > > +                                   seq - > ring->fence_drv.num_fences_mask, > > That's what's amdgpu_fence_emit_polling() does anyway. > > So this here just moves the polling a bit earlier. > > I think we rather need to increase the MES ring size instead. > > Regards, > Christian. > > > > +                                   timeout); > > +     if (r < 1) > > +             goto error_undo; > > + > >        api_status = (struct MES_API_STATUS *)((char *)pkt + > api_status_off); > >        api_status->api_completion_fence_addr = status_gpu_addr; > >        api_status->api_completion_fence_value = 1; > > @@ -194,8 +201,7 @@ static int > mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        mes_status_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS; > > mes_status_pkt.api_status.api_completion_fence_addr = > >                ring->fence_drv.gpu_addr; > > - mes_status_pkt.api_status.api_completion_fence_value = > > -             ++ring->fence_drv.sync_seq; > > + mes_status_pkt.api_status.api_completion_fence_value = seq; > > > >        amdgpu_ring_write_multiple(ring, &mes_status_pkt, > > sizeof(mes_status_pkt) / 4); > > @@ -215,7 +221,7 @@ static int > mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >                dev_dbg(adev->dev, "MES msg=%d was emitted\n", > >                        x_pkt->header.opcode); > > > > -     r = amdgpu_fence_wait_polling(ring, ring->fence_drv.sync_seq, > timeout); > > +     r = amdgpu_fence_wait_polling(ring, seq, timeout); > >        if (r < 1 || !*status_ptr) { > > > >                if (misc_op_str) > > @@ -238,6 +244,10 @@ static int > mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, > >        amdgpu_device_wb_free(adev, status_offset); > >        return 0; > > > > +error_undo: > > +     dev_err(adev->dev, "MES ring buffer is full.\n"); > > +     amdgpu_ring_undo(ring); > > + > >   error_unlock_free: > > spin_unlock_irqrestore(&mes->ring_lock, flags); > > > --------------Q0amzg4GUnZMtc0JlovGnSCi Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: 8bit Thx, but in that case this patch here then won't help either it just mitigates the problem.

Can you try to reduce num_hw_submission for the MES ring?

Thanks,
Christian.

Am 22.07.24 um 05:27 schrieb Xiao, Jack:

[AMD Official Use Only - AMD Internal Distribution Only]


>> I think we rather need to increase the MES ring size instead.

Unfortunately, it doesn't work. I guess mes firmware has limitation.

Regards,
Jack

From: Christian König <ckoenig.leichtzumerken@gmail.com>
Sent: Friday, 19 July 2024 23:44
To: Xiao, Jack <Jack.Xiao@amd.com>; amd-gfx@lists.freedesktop.org <amd-gfx@lists.freedesktop.org>; Deucher, Alexander <Alexander.Deucher@amd.com>
Subject: Re: [PATCH] drm/amdgpu/mes: fix mes ring buffer overflow
 
Am 19.07.24 um 11:16 schrieb Jack Xiao:
> wait memory room until enough before writing mes packets
> to avoid ring buffer overflow.
>
> Signed-off-by: Jack Xiao <Jack.Xiao@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 18 ++++++++++++++----
>   drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 18 ++++++++++++++----
>   2 files changed, 28 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
> index 8ce51b9236c1..68c74adf79f1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
> @@ -168,7 +168,7 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        const char *op_str, *misc_op_str;
>        unsigned long flags;
>        u64 status_gpu_addr;
> -     u32 status_offset;
> +     u32 seq, status_offset;
>        u64 *status_ptr;
>        signed long r;
>        int ret;
> @@ -196,6 +196,13 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        if (r)
>                goto error_unlock_free;
>  
> +     seq = ++ring->fence_drv.sync_seq;
> +     r = amdgpu_fence_wait_polling(ring,
> +                                   seq - ring->fence_drv.num_fences_mask,
> +                                   timeout);
> +     if (r < 1)
> +             goto error_undo;
> +
>        api_status = (struct MES_API_STATUS *)((char *)pkt + api_status_off);
>        api_status->api_completion_fence_addr = status_gpu_addr;
>        api_status->api_completion_fence_value = 1;
> @@ -208,8 +215,7 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        mes_status_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;
>        mes_status_pkt.api_status.api_completion_fence_addr =
>                ring->fence_drv.gpu_addr;
> -     mes_status_pkt.api_status.api_completion_fence_value =
> -             ++ring->fence_drv.sync_seq;
> +     mes_status_pkt.api_status.api_completion_fence_value = seq;
>  
>        amdgpu_ring_write_multiple(ring, &mes_status_pkt,
>                                   sizeof(mes_status_pkt) / 4);
> @@ -229,7 +235,7 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>                dev_dbg(adev->dev, "MES msg=%d was emitted\n",
>                        x_pkt->header.opcode);
>  
> -     r = amdgpu_fence_wait_polling(ring, ring->fence_drv.sync_seq, timeout);
> +     r = amdgpu_fence_wait_polling(ring, seq, timeout);
>        if (r < 1 || !*status_ptr) {
>  
>                if (misc_op_str)
> @@ -252,6 +258,10 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        amdgpu_device_wb_free(adev, status_offset);
>        return 0;
>  
> +error_undo:
> +     dev_err(adev->dev, "MES ring buffer is full.\n");
> +     amdgpu_ring_undo(ring);
> +
>   error_unlock_free:
>        spin_unlock_irqrestore(&mes->ring_lock, flags);
>  
> diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
> index c9f74231ad59..48e01206bcc4 100644
> --- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
> @@ -154,7 +154,7 @@ static int mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        const char *op_str, *misc_op_str;
>        unsigned long flags;
>        u64 status_gpu_addr;
> -     u32 status_offset;
> +     u32 seq, status_offset;
>        u64 *status_ptr;
>        signed long r;
>        int ret;
> @@ -182,6 +182,13 @@ static int mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        if (r)
>                goto error_unlock_free;
>  
> +     seq = ++ring->fence_drv.sync_seq;
> +     r = amdgpu_fence_wait_polling(ring,
> +                                   seq - ring->fence_drv.num_fences_mask,

That's what's amdgpu_fence_emit_polling() does anyway.

So this here just moves the polling a bit earlier.

I think we rather need to increase the MES ring size instead.

Regards,
Christian.


> +                                   timeout);
> +     if (r < 1)
> +             goto error_undo;
> +
>        api_status = (struct MES_API_STATUS *)((char *)pkt + api_status_off);
>        api_status->api_completion_fence_addr = status_gpu_addr;
>        api_status->api_completion_fence_value = 1;
> @@ -194,8 +201,7 @@ static int mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        mes_status_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;
>        mes_status_pkt.api_status.api_completion_fence_addr =
>                ring->fence_drv.gpu_addr;
> -     mes_status_pkt.api_status.api_completion_fence_value =
> -             ++ring->fence_drv.sync_seq;
> +     mes_status_pkt.api_status.api_completion_fence_value = seq;
>  
>        amdgpu_ring_write_multiple(ring, &mes_status_pkt,
>                                   sizeof(mes_status_pkt) / 4);
> @@ -215,7 +221,7 @@ static int mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>                dev_dbg(adev->dev, "MES msg=%d was emitted\n",
>                        x_pkt->header.opcode);
>  
> -     r = amdgpu_fence_wait_polling(ring, ring->fence_drv.sync_seq, timeout);
> +     r = amdgpu_fence_wait_polling(ring, seq, timeout);
>        if (r < 1 || !*status_ptr) {
>  
>                if (misc_op_str)
> @@ -238,6 +244,10 @@ static int mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
>        amdgpu_device_wb_free(adev, status_offset);
>        return 0;
>  
> +error_undo:
> +     dev_err(adev->dev, "MES ring buffer is full.\n");
> +     amdgpu_ring_undo(ring);
> +
>   error_unlock_free:
>        spin_unlock_irqrestore(&mes->ring_lock, flags);
>  


--------------Q0amzg4GUnZMtc0JlovGnSCi--