On Tue, Nov 5, 2024 at 12:21 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
>
> On Tue, Nov 05, 2024 at 11:01:01AM +0800, Yafang Shao wrote:
> > Upon further analysis, it appears that the behavior depends on the
> > /sys/block/*/queue/read_ahead_kb setting. On my test server, this
> > parameter is set to 128KB. For the current approach to work without
> > modification, it would need to be 2MB-aligned.
> >
> > However, I believe MADV_HUGEPAGE behavior should not be dependent on
> > the value of read_ahead_kb. It would be more robust if the kernel
> > automatically aligned it to 2MB when MADV_HUGEPAGE is enabled. The
> > following changes ensure compatibility with non-2MB-aligned
> > read_ahead_kb values:
>
> or maybe:
>
> static unsigned long get_next_ra_size(struct file_ra_state *ra,
> unsigned long max)
> {
> unsigned long cur = ra->size;
>
> if (cur < max / 16)
> return 4 * cur;
> if (cur <= max / 2)
> return 2 * cur;
> + if (cur > max)
> + return cur;
> return max;
> }
>
It is better. Thanks for your suggestion.
--
Regards
Yafang
