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: <be4e1cd8-a994-303d-9424-14439ce1f7d4@ideasonboard.com>
Date:   Thu, 17 Feb 2022 15:03:02 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.5.0
Subject: Re: [PATCH 0/3] drm: omapdrm: Fix excessive GEM buffers DMM/CMA usage
Content-Language: en-US
To:     Ivaylo Dimitrov <ivo.g.dimitrov.75@gmail.com>, tomba@kernel.org,
        airlied@linux.ie, daniel@ffwll.ch
Cc:     dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org,
        linux-omap@vger.kernel.org, merlijn@wizzup.org, tony@atomide.com
References: <1642587791-13222-1-git-send-email-ivo.g.dimitrov.75@gmail.com>
From:   Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
In-Reply-To: <1642587791-13222-1-git-send-email-ivo.g.dimitrov.75@gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi Ivaylo,

On 19/01/2022 12:23, Ivaylo Dimitrov wrote:
> This patch series fixes excessive DMM or CMA usage of GEM buffers leading to
> various runtime allocation failures. The series enables daily usage of devices
> without exausting limited resources like CMA or DMM space if GPU rendering is
> needed.
> 
> The first patch doesn't bring any functional changes, it just moves some
> TILER/DMM related code to a separate function, to simplify the review of the
> next two patches.
> 
> The second patch allows off-CPU rendering to non-scanout buffers. Without that
> patch, it is basically impossible to use the driver allocated GEM buffers on
> OMAP3 for anything else but a basic CPU rendered examples as if we want GPU
> rendering, we must allocate buffers as scanout buffers, which are CMA allocated.
> CMA soon gets fragmented and we start seeing allocation failures. Such failres
> in Xorg cannot be handeled gracefully, so the system is basically unusable.
> 
> Third patch fixes similar issue on OMAP4/5, where DMM/TILER spaces get
> fragmented with time, leading to allocation failures.

I think this is just hacking around the problem. The problem is that 
omapdrm is being used by some as a generic buffer allocator. Those users 
should be changed to use a their own allocator or a generic allocator. 
And we could then drop the OMAP_BO_SCANOUT flag, as all buffers would be 
scanout buffers.

Or do we have a regression in the driver? My understanding is that this 
has never really worked.

  Tomi