linux-kernel - Re: [PATCH v11 2/3] rust: add dma coherent allocator abstraction.

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: <Z5fgYtIdOudE3ELF@tardis.local>
Date: Mon, 27 Jan 2025 11:37:06 -0800
From: Boqun Feng <boqun.feng@...il.com>
To: Alice Ryhl <aliceryhl@...gle.com>
Cc: Danilo Krummrich <dakr@...nel.org>,
	Abdiel Janulgue <abdiel.janulgue@...il.com>,
	rust-for-linux@...r.kernel.org, daniel.almeida@...labora.com,
	robin.murphy@....com, Miguel Ojeda <ojeda@...nel.org>,
	Alex Gaynor <alex.gaynor@...il.com>, Gary Guo <gary@...yguo.net>,
	Björn Roy Baron <bjorn3_gh@...tonmail.com>,
	Benno Lossin <benno.lossin@...ton.me>,
	Andreas Hindborg <a.hindborg@...nel.org>,
	Trevor Gross <tmgross@...ch.edu>,
	Valentin Obst <kernel@...entinobst.de>,
	open list <linux-kernel@...r.kernel.org>,
	Christoph Hellwig <hch@....de>,
	Marek Szyprowski <m.szyprowski@...sung.com>, airlied@...hat.com,
	"open list:DMA MAPPING HELPERS" <iommu@...ts.linux.dev>
Subject: Re: [PATCH v11 2/3] rust: add dma coherent allocator abstraction.

On Mon, Jan 27, 2025 at 11:21:52AM -0800, Boqun Feng wrote:
[...]
> > > Depending on the granularity that tearing can happen, if .foo is an enum
> > > (or any other type that not all bit combinations are valid) and tearing
> > > can happen at byte levels, then a racing dma_read() may read invalid
> > > data.
> > 
> > T: FromBytes + ToBytes is already required for these types. You can't
> > use these operations with such an enum.
> > 
> 
> I was talking about a wider problem, but fine ;-) So the assumption is
> the read_volatile() or copy_nonoverlapping() provide byte-level
> atomicity? Although unlikely, but if tearing happens at sub-byte level,
> then even if `T: FromBytes + ToBytes` you can still get invalid data.
> 

OK, `FromBytes + ToBytes` should tolerate sub-byte level tearing, so I
was wrong on this.

> > > I think it's fine to expect read_volatile() and write_volatile()
> > > themselves don't trigger UB, but we will need to be careful about the
> > > atomic granularity that we can expect on them. It would be more clear if
> > > we use the atomic API here (and implementation can be read_volatile()
> > > and write_volatile()), and it can avoid coding based on tribal knowledge
> > > such as "in kernel, read_volatile() and write_volatile() imply atomic".
> > 
> > Why should we use atomics for operations that don't need to be atomic?

And yes, given the `FromBytes + ToBytes`, it looks like you don't need
any atomicity for the basic DMA read/write operation.

Regards,
Boqun

> > Most of the time, dma memory is not *actually* changed while you read
> > it.
> > 
> 
> Because the requirement here actually needs atomic at byte level, it's
> similar to:
> 
> 	https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2022/p1478r8.html
> 
> also, notice that for byte level atomic, it's actually free on most of
> the architectures (i.e. no extra cost). Again, it's more of a "how do we
> express our assumption" question. If indeed we expect byte-level
> atomicity, then I see no harm to use atomic API here.
> 
> Regards,
> Boqun
> 
> > Alice