[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAGWkznFxkm5DihSGDy9W-aSmtS2kU3_db9Ex-ZygMGDm_DEeeg@mail.gmail.com>
Date: Thu, 27 Jun 2024 08:23:18 +0800
From: Zhaoyang Huang <huangzhaoyang@...il.com>
To: "zhaoyang.huang" <zhaoyang.huang@...soc.com>, Yu Zhao <yuzhao@...gle.com>
Cc: Andrew Morton <akpm@...ux-foundation.org>, David Hildenbrand <david@...hat.com>, linux-mm@...ck.org,
linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
steve.kang@...soc.com
Subject: Re: [RFC PATCH] mm: introduce gen information in /proc/xxx/smaps
loop yu.zhao
On Wed, Jun 26, 2024 at 4:46 PM zhaoyang.huang
<zhaoyang.huang@...soc.com> wrote:
>
> From: Zhaoyang Huang <zhaoyang.huang@...soc.com>
>
> Madvise_cold_and_pageout could break the LRU's balance by
> colding or moving the folio without taking activity information into
> consideration. This commit would like to introduce the folios' gen
> information based on VMA block via which the userspace could query
> the VA's activity before madvise.
>
> eg. The VMA(56c00000-56e14000) which has big Rss/Gen value suggest that
> it possesses larger proportion active folios than VMA(70dd7000-71090000)
> does and is not good candidate for madvise.
>
> 56c00000-56e14000 rw-p 00000000 00:00 0 [anon:dalvik-/system/framework/oat/arm64/services.art]
> Size: 2128 kB
> KernelPageSize: 4 kB
> MMUPageSize: 4 kB
> Rss: 2128 kB
> Pss: 2128 kB
> Pss_Dirty: 2128 kB
> Shared_Clean: 0 kB
> Shared_Dirty: 0 kB
> Private_Clean: 0 kB
> Private_Dirty: 2128 kB
> Referenced: 2128 kB
> Anonymous: 2128 kB
> KSM: 0 kB
> LazyFree: 0 kB
> AnonHugePages: 0 kB
> ShmemPmdMapped: 0 kB
> FilePmdMapped: 0 kB
> Shared_Hugetlb: 0 kB
> Private_Hugetlb: 0 kB
> Swap: 0 kB
> SwapPss: 0 kB
> Locked: 0 kB
> Gen: 664
> THPeligible: 0
> VmFlags: rd wr mr mw me ac
> 70dd7000-71090000 rw-p 00000000 00:00 0 [anon:dalvik-/system/framework/boot.art]
> Size: 2788 kB
> KernelPageSize: 4 kB
> MMUPageSize: 4 kB
> Rss: 2788 kB
> Pss: 275 kB
> Pss_Dirty: 275 kB
> Shared_Clean: 0 kB
> Shared_Dirty: 2584 kB
> Private_Clean: 0 kB
> Private_Dirty: 204 kB
> Referenced: 2716 kB
> Anonymous: 2788 kB
> KSM: 0 kB
> LazyFree: 0 kB
> AnonHugePages: 0 kB
> ShmemPmdMapped: 0 kB
> FilePmdMapped: 0 kB
> Shared_Hugetlb: 0 kB
> Private_Hugetlb: 0 kB
> Swap: 0 kB
> SwapPss: 0 kB
> Locked: 0 kB
> Gen: 1394
> THPeligible: 0
>
> Signed-off-by: Zhaoyang Huang <zhaoyang.huang@...soc.com>
> ---
> fs/proc/task_mmu.c | 15 +++++++++++++++
> 1 file changed, 15 insertions(+)
>
> diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> index f8d35f993fe5..9731f43aa639 100644
> --- a/fs/proc/task_mmu.c
> +++ b/fs/proc/task_mmu.c
> @@ -408,12 +408,23 @@ struct mem_size_stats {
> u64 pss_dirty;
> u64 pss_locked;
> u64 swap_pss;
> +#ifdef CONFIG_LRU_GEN
> + u64 gen;
> +#endif
> };
>
> static void smaps_page_accumulate(struct mem_size_stats *mss,
> struct folio *folio, unsigned long size, unsigned long pss,
> bool dirty, bool locked, bool private)
> {
> +#ifdef CONFIG_LRU_GEN
> + int gen = folio_lru_gen(folio);
> + struct lru_gen_folio *lrugen = &folio_lruvec(folio)->lrugen;
> +
> + if (gen >= 0)
> + mss->gen += (lru_gen_from_seq(lrugen->max_seq) - gen + MAX_NR_GENS) % MAX_NR_GENS;
> +#endif
> +
> mss->pss += pss;
>
> if (folio_test_anon(folio))
> @@ -852,6 +863,10 @@ static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
> SEQ_PUT_DEC(" kB\nLocked: ",
> mss->pss_locked >> PSS_SHIFT);
> seq_puts(m, " kB\n");
> +#ifdef CONFIG_LRU_GEN
> + seq_put_decimal_ull_width(m, "Gen: ", mss->gen, 8);
> + seq_puts(m, "\n");
> +#endif
> }
>
> static int show_smap(struct seq_file *m, void *v)
> --
> 2.25.1
>
Powered by blists - more mailing lists