[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <c9d83a4a-8942-49dd-aaf8-962770f820f7@gmail.com>
Date: Thu, 2 Jan 2025 15:52:06 +0000
From: Pavel Begunkov <asml.silence@...il.com>
To: Jakub Kicinski <kuba@...nel.org>
Cc: David Wei <dw@...idwei.uk>, io-uring@...r.kernel.org,
netdev@...r.kernel.org, Jens Axboe <axboe@...nel.dk>,
Paolo Abeni <pabeni@...hat.com>, "David S. Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>, Jesper Dangaard Brouer
<hawk@...nel.org>, David Ahern <dsahern@...nel.org>,
Mina Almasry <almasrymina@...gle.com>,
Stanislav Fomichev <stfomichev@...il.com>, Joe Damato <jdamato@...tly.com>,
Pedro Tammela <pctammela@...atatu.com>
Subject: Re: [PATCH net-next v9 03/20] net: generalise net_iov chunk owners
On 12/21/24 02:17, Jakub Kicinski wrote:
> On Sat, 21 Dec 2024 00:50:37 +0000 Pavel Begunkov wrote:
>>> Is there a good reason why dma addr is not part of net_iov_area?
>>> net_iov_area is one chunk of continuous address space.
>>> Instead of looping over pages in io_zcrx_map_area we could map
>>> the whole thing in one go.
>>
>> It's not physically contiguous. The registration API takes
>> contig user addresses, but that's not a hard requirement
>> either.
>
> Okay, I was thrown off by the base_virtual being in the common struct.
> But it appears you don't use that?
Right, but io_uring can make use of it, it just needs better types,
which is why I left it there to follow up later.
> AFAIR for devmem each area is physically contiguous if the region is
> not it gets split into areas.
Seems so, it's split into areas / chunk owners by following the
sg list layout.
--
Pavel Begunkov
Powered by blists - more mailing lists