lkml.org 
[lkml]   [2022]   [Jan]   [31]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] docs/memory-barriers.txt: volatile is not a barrier() substitute
On Mon, Jan 31, 2022 at 02:52:47PM -0800, Nick Desaulniers wrote:
> Add text to memory-barriers.txt and deprecated.rst to denote that
> volatile-qualifying an asm statement is not a substitute for either a
> compiler barrier (``barrier();``) or a clobber list.
>
> This way we can point to this in code that strengthens existing
> volatile-qualified asm statements to use a compiler barrier.
>
> Suggested-by: Kees Cook <keescook@chromium.org>
> Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
> ---
> Example: https://godbolt.org/z/8PW549zz9
>
> Documentation/memory-barriers.txt | 24 ++++++++++++++++++++++++
> Documentation/process/deprecated.rst | 17 +++++++++++++++++
> 2 files changed, 41 insertions(+)
>
> diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
> index b12df9137e1c..f3908c0812da 100644
> --- a/Documentation/memory-barriers.txt
> +++ b/Documentation/memory-barriers.txt
> @@ -1726,6 +1726,30 @@ of optimizations:
> respect the order in which the READ_ONCE()s and WRITE_ONCE()s occur,
> though the CPU of course need not do so.
>
> + (*) Similarly, the compiler is within its rights to reorder instructions
> + around an asm statement so long as clobbers are not violated. For example,
> +
> + asm volatile ("");
> + flag = true;
> +
> + May be modified by the compiler to:
> +
> + flag = true;
> + asm volatile ("");
> +
> + Marking an asm statement as volatile is not a substitute for barrier(),
> + and is implicit for asm goto statements and asm statements that do not
> + have outputs (like the above example). Prefer either:
> +
> + asm ("":::"memory");
> + flag = true;
> +
> + Or:
> +
> + asm ("");
> + barrier();
> + flag = true;
> +

I like this!

> (*) The compiler is within its rights to invent stores to a variable,
> as in the following example:
>
> diff --git a/Documentation/process/deprecated.rst b/Documentation/process/deprecated.rst
> index 388cb19f5dbb..432816e2f79e 100644
> --- a/Documentation/process/deprecated.rst
> +++ b/Documentation/process/deprecated.rst
> @@ -329,3 +329,20 @@ struct_size() and flex_array_size() helpers::
> instance->count = count;
>
> memcpy(instance->items, source, flex_array_size(instance, items, instance->count));
> +
> +Volatile Qualified asm Statements
> +=================================

I would open with an example, like:

Instead of::

volatile asm("...");

just use::

asm("...");


> +
> +According to `the GCC docs on inline asm
> +https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Volatile`_:
> +
> + asm statements that have no output operands and asm goto statements,
> + are implicitly volatile.

Does this mean "volatile" _is_ needed when there are operands, etc?

> +
> +For many uses of asm statements, that means adding a volatile qualifier won't
> +hurt (making the implicit explicit), but it will not strengthen the semantics
> +for such cases where it would have been implied. Care should be taken not to
> +confuse ``volatile`` with the kernel's ``barrier()`` macro or an explicit
> +clobber list. See [memory-barriers]_ for more info on ``barrier()``.
> +
> +.. [memory-barriers] Documentation/memory-barriers.txt
> --
> 2.35.0.rc2.247.g8bbb082509-goog
>

--
Kees Cook

\
 
 \ /
  Last update: 2022-02-01 00:54    [W:2.383 / U:0.164 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site