On Fri, Jan 28 2022 at 1:13P -0500,
Christoph Hellwig <hch@xxxxxx> wrote:
> On Thu, Jan 27, 2022 at 11:17:51PM -0500, Mike Snitzer wrote:
> > + __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
> > + bio_op(bio), start_time);
> > }
> > +EXPORT_SYMBOL_GPL(bio_start_io_acct_time);
> >
> > /**
> > * bio_start_io_acct - start I/O accounting for bio based drivers
> > @@ -1084,14 +1096,15 @@ static unsigned long __part_start_io_acct(struct block_device *part,
> > */
> > unsigned long bio_start_io_acct(struct bio *bio)
> > {
> > - return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), bio_op(bio));
> > + return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
> > + bio_op(bio), jiffies);
>
> Is droppingthe READ_ONCE safe here? This is a honest question as these
> helpers still confuse me.
I'm not sure either, commit 956d510ee78ca ("block: add disk/bio-based
accounting helpers") doesn't offer any insight on the need.
Very little kernel code uses READ_ONCE(jiffies).
git diff 24d69293d9a561645e0b4d78c2fb179827e35f53^..e722fff238bbfe6308d7778a8c2163c181bf998a
shows that at the time the outgoing generic_{start,end}_io_acct didn't
use READ_ONCE (nor did any of the drivers that were updated to use the
new helpers).
ACCESS_ONCE() was used prior to READ_ONCE() -- see commit
cbbce82209490 and then 316c1608d15c7.
Anyway, looks to be cargo cult at this point right?
(if so, implies the use of READ_ONCE() in patch 3 isn't needed either)
> The rest looks good:
>
> Reviewed-by: Christoph Hellwig <hch@xxxxxx>
Thanks.
