[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZTbOUIAxuOGgLH+y@smile.fi.intel.com>
Date: Mon, 23 Oct 2023 22:49:36 +0300
From: Andy Shevchenko <andriy.shevchenko@...ux.intel.com>
To: Alexander Potapenko <glider@...gle.com>
Cc: catalin.marinas@....com, will@...nel.org, pcc@...gle.com,
andreyknvl@...il.com, aleksander.lobakin@...el.com,
linux@...musvillemoes.dk, yury.norov@...il.com,
alexandru.elisei@....com, linux-kernel@...r.kernel.org,
linux-arm-kernel@...ts.infradead.org, eugenis@...gle.com,
syednwaris@...il.com, william.gray@...aro.org
Subject: Re: [PATCH v8 2/2] lib/test_bitmap: add tests for
bitmap_{read,write}()
On Mon, Oct 23, 2023 at 03:50:42PM +0200, Alexander Potapenko wrote:
> On Mon, Oct 23, 2023 at 1:32 PM Andy Shevchenko
> <andriy.shevchenko@...ux.intel.com> wrote:
> > On Mon, Oct 23, 2023 at 12:23:27PM +0200, Alexander Potapenko wrote:
...
> > > + val = bitmap_read(bitmap, i, nbits);
> > > + (void)val;
> >
> > Is it marked with __must_check? Otherwise why do we need this?
>
> That was a weak attempt to prevent the compiler from completely
> optimizing away the bitmap_read() calls, but I haven't really looked
> at the result.
> The reality is that even with this check the calls are deleted, and
> the size of the function is only 68 bytes.
> Replacing the val assignment with a WRITE_ONCE() actually ensures the
> bitmap_read() calls are not deleted:
>
> diff --git a/lib/test_bitmap.c b/lib/test_bitmap.c
> index ba567f53feff1..ae57ae48ef3ad 100644
> --- a/lib/test_bitmap.c
> +++ b/lib/test_bitmap.c
> @@ -1360,8 +1360,7 @@ static void __init test_bitmap_read_perf(void)
> for (i = 0; i < TEST_BIT_LEN; i++) {
> if (i + nbits > TEST_BIT_LEN)
> break;
> - val = bitmap_read(bitmap, i, nbits);
> - (void)val;
> + WRITE_ONCE(val, bitmap_read(bitmap, i, nbits));
> }
> }
> }
Okay, whatever you choose, please add a comment explaining this.
--
With Best Regards,
Andy Shevchenko
Powered by blists - more mailing lists