Re: [PATCH] wait_on_bit: add an acquire memory barrier

[Will Deacon <will@kernel.org>]

On Thu, Aug 25, 2022 at 05:03:40PM -0400, Mikulas Patocka wrote:
> Here I reworked your patch, so that test_bit_acquire is defined just like 
> test_bit. There's some code duplication (in 
> include/asm-generic/bitops/generic-non-atomic.h and in 
> arch/x86/include/asm/bitops.h), but that duplication exists in the 
> test_bit function too.
> 
> I tested it on x86-64 and arm64. On x86-64 it generates the "bt" 
> instruction for variable-bit test and "shr; and $1" for constant bit test. 
> On arm64 it generates the "ldar" instruction for both constant and 
> variable bit test.
> 
> For me, the kernel 6.0-rc2 doesn't boot in an arm64 virtual machine at all 
> (with or without this patch), so I only compile-tested it on arm64. I have 
> to bisect it.

It's working fine for me and I haven't had any other reports that it's not
booting. Please could you share some more details about your setup so we
can try to reproduce the problem?

> From: Mikulas Patocka <mpatocka@redhat.com>
> 
> There are several places in the kernel where wait_on_bit is not followed
> by a memory barrier (for example, in drivers/md/dm-bufio.c:new_read). On
> architectures with weak memory ordering, it may happen that memory
> accesses that follow wait_on_bit are reordered before wait_on_bit and they
> may return invalid data.
> 
> Fix this class of bugs by introducing a new function "test_bit_acquire"
> that works like test_bit, but has acquire memory ordering semantics.
> 
> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
> Cc: stable@vger.kernel.org
> 
>  arch/x86/include/asm/bitops.h                            |   13 +++++++++++++
>  include/asm-generic/bitops/generic-non-atomic.h          |   14 ++++++++++++++
>  include/asm-generic/bitops/instrumented-non-atomic.h     |   12 ++++++++++++
>  include/asm-generic/bitops/non-atomic.h                  |    1 +
>  include/asm-generic/bitops/non-instrumented-non-atomic.h |    1 +
>  include/linux/bitops.h                                   |    1 +
>  include/linux/buffer_head.h                              |    2 +-
>  include/linux/wait_bit.h                                 |    8 ++++----
>  kernel/sched/wait_bit.c                                  |    2 +-
>  9 files changed, 48 insertions(+), 6 deletions(-)

This looks good to me, thanks for doing it! Just one thing that jumped out
at me:

> Index: linux-2.6/include/linux/buffer_head.h
> ===================================================================
> --- linux-2.6.orig/include/linux/buffer_head.h
> +++ linux-2.6/include/linux/buffer_head.h
> @@ -156,7 +156,7 @@ static __always_inline int buffer_uptoda
>  	 * make it consistent with folio_test_uptodate
>  	 * pairs with smp_mb__before_atomic in set_buffer_uptodate
>  	 */
> -	return (smp_load_acquire(&bh->b_state) & (1UL << BH_Uptodate)) != 0;
> +	return test_bit_acquire(BH_Uptodate, &bh->b_state);

Do you think it would be worth adding set_bit_release() and then relaxing
set_buffer_uptodate() to use that rather than the smp_mb__before_atomic()?

Will