linux-kernel - Re: [PATCH 3/10] tmpfs: optimize clearing when writing

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Tue, 15 May 2012 18:51:16 +1000
From:	Nick Piggin <npiggin@...il.com>
To:	Hugh Dickins <hughd@...gle.com>
Cc:	Andrew Morton <akpm@...ux-foundation.org>,
	Christoph Hellwig <hch@...radead.org>, linux-mm@...ck.org,
	linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 3/10] tmpfs: optimize clearing when writing

On 12 May 2012 22:04, Hugh Dickins <hughd@...gle.com> wrote:
> Nick proposed years ago that tmpfs should avoid clearing its pages where
> write will overwrite them with new data, as ramfs has long done.  But I
> messed it up and just got bad data.  Tried again recently, it works fine.
>
> Here's time output for writing 4GiB 16 times on this Core i5 laptop:
>
> before: real    0m21.169s user  0m0.028s sys    0m21.057s
>        real    0m21.382s user  0m0.016s sys    0m21.289s
>        real    0m21.311s user  0m0.020s sys    0m21.217s
>
> after:  real    0m18.273s user  0m0.032s sys    0m18.165s
>        real    0m18.354s user  0m0.020s sys    0m18.265s
>        real    0m18.440s user  0m0.032s sys    0m18.337s
>
> ramfs:  real    0m16.860s user  0m0.028s sys    0m16.765s
>        real    0m17.382s user  0m0.040s sys    0m17.273s
>        real    0m17.133s user  0m0.044s sys    0m17.021s

Cool, thanks Hugh! Very big speedup.


>
> Yes, I have done perf reports, but they need more explanation than they
> deserve: in summary, clear_page vanishes, its cache loading shifts into
> copy_user_generic_unrolled; shmem_getpage_gfp goes down, and surprisingly
> mark_page_accessed goes way up - I think because they are respectively
> where the cache gets to be reloaded after being purged by clear or copy.
>
> Suggested-by: Nick Piggin <npiggin@...il.com>
> Signed-off-by: Hugh Dickins <hughd@...gle.com>
> ---
>  mm/shmem.c |   20 +++++++++++++++++---
>  1 file changed, 17 insertions(+), 3 deletions(-)
>
> --- 3045N.orig/mm/shmem.c       2012-05-05 10:46:05.732062006 -0700
> +++ 3045N/mm/shmem.c    2012-05-05 10:46:12.316062172 -0700
> @@ -1095,9 +1095,14 @@ repeat:
>                shmem_recalc_inode(inode);
>                spin_unlock(&info->lock);
>
> -               clear_highpage(page);
> -               flush_dcache_page(page);
> -               SetPageUptodate(page);
> +               /*
> +                * Let SGP_WRITE caller clear ends if write does not fill page
> +                */
> +               if (sgp != SGP_WRITE) {
> +                       clear_highpage(page);
> +                       flush_dcache_page(page);
> +                       SetPageUptodate(page);
> +               }
>                if (sgp == SGP_DIRTY)
>                        set_page_dirty(page);
>        }
> @@ -1307,6 +1312,14 @@ shmem_write_end(struct file *file, struc
>        if (pos + copied > inode->i_size)
>                i_size_write(inode, pos + copied);
>
> +       if (!PageUptodate(page)) {
> +               if (copied < PAGE_CACHE_SIZE) {
> +                       unsigned from = pos & (PAGE_CACHE_SIZE - 1);
> +                       zero_user_segments(page, 0, from,
> +                                       from + copied, PAGE_CACHE_SIZE);
> +               }
> +               SetPageUptodate(page);
> +       }
>        set_page_dirty(page);
>        unlock_page(page);
>        page_cache_release(page);
> @@ -1768,6 +1781,7 @@ static int shmem_symlink(struct inode *d
>                kaddr = kmap_atomic(page);
>                memcpy(kaddr, symname, len);
>                kunmap_atomic(kaddr);
> +               SetPageUptodate(page);
>                set_page_dirty(page);
>                unlock_page(page);
>                page_cache_release(page);