linux-kernel - Re: [RFC v2 1/2] dmapool: Move pool metadata into non-DMA memory

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAOZ5it1H_bac-W7vptu5pAi=O6KhRWPCwLP8NgjD55+AyBvfwQ@mail.gmail.com>
Date: Thu, 21 Nov 2024 10:48:17 -0700
From: Brian Johannesmeyer <bjohannesmeyer@...il.com>
To: Christoph Hellwig <hch@...radead.org>
Cc: Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org, 
	linux-kernel@...r.kernel.org, linux-hardening@...r.kernel.org, 
	Raphael Isemann <teemperor@...il.com>, Cristiano Giuffrida <giuffrida@...vu.nl>, Herbert Bos <h.j.bos@...nl>, 
	Greg KH <gregkh@...uxfoundation.org>, Keith Busch <kbusch@...nel.org>
Subject: Re: [RFC v2 1/2] dmapool: Move pool metadata into non-DMA memory

On Wed, Nov 20, 2024 at 10:03 PM Christoph Hellwig <hch@...radead.org> wrote:
> Indeed.  You'd probably need to split the linkage of the pages into
> a list of those that have free blocks and those that don't as a minimum.
>
> Can you share your current version?

Sure, I can share the current version, though fair warning---it’s
still quite messy.

FWIW, I wonder if the bitmap approach might be more suitable as a
separate RFC. AFAICT, the primary issue with the currently submitted
patches is their runtime overhead. I’ve proposed a way to address this
in my recent response to [RFC v2 0/2]. Unfortunately, as I noted,
improving the memory overhead without worsening the runtime
performance is challenging---for example, removing the `next_block`
pointers would require iterating over all pages to find a free
`block`, which significantly impacts the runtime.

That said, how would you prefer I share my bitmap approach? Should I
submit it as a separate patch series or provide the patch directly in
this thread?

Thanks,

Brian Johannesmeyer