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: <68cd25de-e767-725e-2e7b-703217230bb0@redhat.com>
Date:   Thu, 15 Jun 2023 18:39:20 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.10.0
Subject: Re: [PATCH drm-next v4 00/14] [RFC] DRM GPUVA Manager & Nouveau
 VM_BIND UAPI
Content-Language: en-US
To:     Liam.Howlett@oracle.com
Cc:     airlied@gmail.com, daniel@ffwll.ch, tzimmermann@suse.de,
        mripard@kernel.org, corbet@lwn.net, christian.koenig@amd.com,
        bskeggs@redhat.com, matthew.brost@intel.com,
        boris.brezillon@collabora.com, alexdeucher@gmail.com,
        ogabbay@kernel.org, bagasdotme@gmail.com, willy@infradead.org,
        jason@jlekstrand.net, dri-devel@lists.freedesktop.org,
        nouveau@lists.freedesktop.org, linux-doc@vger.kernel.org,
        linux-mm@kvack.org, linux-kernel@vger.kernel.org
References: <20230606223130.6132-1-dakr@redhat.com>
From:   Danilo Krummrich <dakr@redhat.com>
Organization: RedHat
In-Reply-To: <20230606223130.6132-1-dakr@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 6/7/23 00:31, Danilo Krummrich wrote:

>    Maple Tree:
>      - Maple tree uses the 'unsinged long' type for node entries. While this
>        works for 64bit, it's incompatible with the DRM GPUVA Manager on 32bit,
>        since the DRM GPUVA Manager uses the u64 type and so do drivers using it.
>        While it's questionable whether a 32bit kernel and a > 32bit GPU address
>        space make any sense, it creates tons of compiler warnings when compiling
>        for 32bit. Maybe it makes sense to expand the maple tree API to let users
>        decide which size to pick - other ideas / proposals are welcome.

I remember you told me that the filesystem folks had some interest in a 
64-bit maple tree for a 32-bit kernel as well. Are there any news or 
plans for such a feature?

For the short term I'd probably add a feature flag to the GPUVA manager, 
where drivers explicitly need to promise not to pass in addresses 
exceeding 32-bit on a 32-bit kernel, and if they don't refuse to 
initialize the GPUVA manager on 32-bit kernels - or something similar...