LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Arnd Bergmann <arnd@arndb.de>
To: Boqun Feng <boqun.feng@gmail.com>
Cc: linux-arch <linux-arch@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>,
	Wedson Almeida Filho <wedsonaf@google.com>,
	Gary Guo <gary@garyguo.net>, Hector Martin <marcan@marcan.st>,
	Arnd Bergmann <arnd@arndb.de>,
	Linus Walleij <linus.walleij@linaro.org>
Subject: Re: [Question] Alignment requirement for readX() and writeX()
Date: Fri, 30 Jul 2021 18:58:30 +0200	[thread overview]
Message-ID: <CAK8P3a0w09Ga_OXAqhA0JcgR-LBc32a296dZhpTyPDwVSgaNkw@mail.gmail.com> (raw)
In-Reply-To: <YQQr+twAYHk2jXs6@boqun-archlinux>

On Fri, Jul 30, 2021 at 6:43 PM Boqun Feng <boqun.feng@gmail.com> wrote:
>
> Hi,
>
> The background is that I'm reviewing Wedson's PR on IoMem for
> Rust-for-Linux project:
>
>         https://github.com/Rust-for-Linux/linux/pull/462
>
> readX() and writeX() are used to provide Rust code to read/write IO
> memory. And I want to find whether we need to check the alignment of the
> pointer. I wonder whether the addresses passed to readX() and writeX()
> need to be aligned to the size of the accesses (e.g. the parameter of
> readl() has to be a 4-byte aligned pointer).
>
> The only related information I get so far is the following quote in
> Documentation/driver-io/device-io.rst:
>
>         On many platforms, I/O accesses must be aligned with respect to
>         the access size; failure to do so will result in an exception or
>         unpredictable results.
>
> Does it mean all readX() and writeX() need to use aligned addresses?
> Or the alignment requirement is arch-dependent, i.e. if the architecture
> supports and has enabled misalignment load and store, no alignment
> requirement on readX() and writeX(), otherwise still need to use aligned
> addresses.
>
> I know different archs have their own alignment requirement on memory
> accesses, just want to make sure the requirement of the readX() and
> writeX() APIs.

I am not aware of any driver that requires unaligned access on __iomem
pointers, and since it definitely doesn't work on most architectures, I think
having an unconditional alignment check makes sense.

What would the alignment check look like? Is there a way to annotate
a pointer that is 'void __iomem *' in C as having a minimum alignment
when it gets passed into a function that uses readl()/writel() on it?

       Arnd

  reply	other threads:[~2021-07-30 16:58 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-30 16:42 Boqun Feng
2021-07-30 16:58 ` Arnd Bergmann [this message]
2021-07-30 17:30   ` Boqun Feng
2021-07-30 20:24     ` Arnd Bergmann
2021-07-31  1:51       ` Boqun Feng
2021-08-02  8:37   ` David Laight

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=CAK8P3a0w09Ga_OXAqhA0JcgR-LBc32a296dZhpTyPDwVSgaNkw@mail.gmail.com \
    --to=arnd@arndb.de \
    --cc=boqun.feng@gmail.com \
    --cc=gary@garyguo.net \
    --cc=linus.walleij@linaro.org \
    --cc=linux-arch@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=marcan@marcan.st \
    --cc=miguel.ojeda.sandonis@gmail.com \
    --cc=wedsonaf@google.com \
    --subject='Re: [Question] Alignment requirement for readX() and writeX()' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).