Re: [PATCH] selftests/mm: reuse FORCE_READ to replace "asm volatile("" : "+r" (XXX));"

[David Laight]

On Thu, 17 Jul 2025 13:43:45 +0200
David Hildenbrand <david@xxxxxxxxxx> wrote:

> On 17.07.25 12:48, wang lian wrote:
> >> On Wed, 16 Jul 2025 20:31:26 +0800 wang lian <lianux.mm@xxxxxxxxx> wrote:  
> >   
> >>> Several mm selftests use the `asm volatile("" : "+r" (variable));`
> >>> construct to force a read of a variable, preventing the compiler from
> >>> optimizing away the memory access. This idiom is cryptic and duplicated
> >>> across multiple test files.
> >>>
> >>> Following a suggestion from David[1], this patch refactors this
> >>> common pattern into a FORCE_READ() macro
> >>>
> >>>   tools/testing/selftests/mm/cow.c              | 30 +++++++++----------
> >>>   tools/testing/selftests/mm/hugetlb-madvise.c  |  5 +---
> >>>   tools/testing/selftests/mm/migration.c        | 13 ++++----
> >>>   tools/testing/selftests/mm/pagemap_ioctl.c    |  4 +--
> >>>   .../selftests/mm/split_huge_page_test.c       |  4 +--
> >>>   5 files changed, 24 insertions(+), 32 deletions(-)  
> >   
> >> The patch forgot to move the FORCE_READ definition into a header?  
> > 
> > Hi Andrew,
> > You are absolutely right. My apologies for the inconvenience.
> > This patch was sent standalone based on a suggestion from David during
> > the discussion of a previous, larger patch series. In that original series,
> > I had already moved the FORCE_READ() macro definition into vm_util.h.
> > 
> > You can find the original patch series and discussion at this link:
> > https://lore.kernel.org/lkml/20250714130009.14581-1-lianux.mm@xxxxxxxxx/
> > It should also be in your mailing list archive.
> > 
> > To make this easier to review and apply, I can send a new, small patch series
> > that first introduces the FORCE_READ() macro in vm_util.h and then applies this refactoring.  
> 
> Please simply perform the move of FORCE_READ() in this very patch where 
> you also use it elswehere.

Why not use READ_ONCE() instead (or even just make all the variables 'volatile char *').
I had to look up the definition to find the hidden indirection in FORCE_READ().

It has to be said that now writes to variables that are READ_ONCE() have to be
WRITE_ONCE() why not just make the variables 'volatile'.
That will stop things bleating about missing READ/WRITE_ONCE() wrappers.
There was a rational for not using volatile, but it is getting to be moot.

	David


> 
> I missed that when skimming over this patch.
>