On Fri, 23 Feb 2024 17:15:50 +1300 Barry Song <21cnbao@xxxxxxxxx> wrote:
> From: Barry Song <v-songbaohua@xxxxxxxx>
>
> While doing MADV_PAGEOUT, the current code will clear PTE young
> so that vmscan won't read young flags to allow the reclamation
> of madvised folios to go ahead.
> It seems we can do it by directly ignoring references, thus we
> can remove tlb flush in madvise and rmap overhead in vmscan.
>
> Regarding the side effect, in the original code, if a parallel
> thread runs side by side to access the madvised memory with the
> thread doing madvise, folios will get a chance to be re-activated
> by vmscan. But with the patch, they will still be reclaimed. But
> this behaviour doing PAGEOUT and doing access at the same time is
> quite silly like DoS. So probably, we don't need to care.
I think we might need to take care of the case, since users may use just a
best-effort estimation like DAMON for the target pages. In such cases, the
page granularity re-check of the access could be helpful. So I concern if this
could be a visible behavioral change for some valid use cases.
>
> A microbench as below has shown 6% decrement on the latency of
> MADV_PAGEOUT,
I assume some of the users may use MADV_PAGEOUT for proactive reclamation of
the memory. In the use case, I think latency of MADV_PAGEOUT might be not that
important.
Hence I think the cons of the behavioral change might outweigh the pros of the
latench improvement, for such best-effort proactive reclamation use case. Hope
to hear and learn from others' opinions.
>
> #define PGSIZE 4096
> main()
> {
> int i;
> #define SIZE 512*1024*1024
> volatile long *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE,
> MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
>
> for (i = 0; i < SIZE/sizeof(long); i += PGSIZE / sizeof(long))
> p[i] = 0x11;
>
> madvise(p, SIZE, MADV_PAGEOUT);
> }
>
> w/o patch w/ patch
> root@10:~# time ./a.out root@10:~# time ./a.out
> real 0m49.634s real 0m46.334s
> user 0m0.637s user 0m0.648s
> sys 0m47.434s sys 0m44.265s
>
> Signed-off-by: Barry Song <v-songbaohua@xxxxxxxx>
Thanks,
SJ
[...]
