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  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date:   Sat,  2 Jan 2021 15:17:32 +0000
From:   Pavel Begunkov <asml.silence@...il.com>
To:     linux-block@...r.kernel.org
Cc:     Jens Axboe <axboe@...nel.dk>,
        Christoph Hellwig <hch@...radead.org>,
        Matthew Wilcox <willy@...radead.org>,
        Ming Lei <ming.lei@...hat.com>,
        Johannes Weiner <hannes@...xchg.org>,
        Alexander Viro <viro@...iv.linux.org.uk>,
        "Darrick J . Wong" <darrick.wong@...cle.com>,
        "Martin K . Petersen" <martin.petersen@...cle.com>,
        Jonathan Corbet <corbet@....net>, linux-xfs@...r.kernel.org,
        linux-fsdevel@...r.kernel.org, io-uring@...r.kernel.org,
        linux-kernel@...r.kernel.org, target-devel@...r.kernel.org,
        linux-scsi@...r.kernel.org, linux-doc@...r.kernel.org
Subject: [PATCH v2 0/7] no-copy bvec

Currently, when iomap and block direct IO gets a bvec based iterator
the bvec will be copied, with all other accounting that takes much
CPU time and causes additional allocation for larger bvecs. The
patchset makes it to reuse the passed in iter bvec.

[1,2] are forbidding zero-length bvec segments to not pile special
cases, [3] skip/fix PSI tracking to not iterate over bvecs extra
time.


nullblk completion_nsec=0 submit_queues=NR_CORES, no merges, no stats
fio/t/io_uring /dev/nullb0 -d 128 -s 32 -c 32 -p 0 -B 1 -F 1 -b BLOCK_SIZE

BLOCK_SIZE             512     4K      8K      16K     32K     64K
===================================================================
old (KIOPS)            1208    1208    1131    1039    863     699
new (KIOPS)            1222    1222    1170    1137    1083    982

Previously, Jens got before 10% difference for polling real HW and small
block sizes, but that was for an older version that had one
iov_iter_advance() less


since RFC:
- add target_core_file patch by Christoph
- make no-copy default behaviour, remove iter flag
- iter_advance() instead of hacks to revert to work
- add bvec iter_advance() optimisation patch
- remove PSI annotations from direct IO (iomap, block and fs/direct)
- note in d/f/porting

since v1:
- don't allow zero-length bvec segments (Ming)
- don't add a BIO_WORKINGSET-less version of bio_add_page(), just clear
  the flag at the end and leave it for further cleanups (Christoph)
- commit message and comments rewording (Dave)
- other nits by Christoph

Christoph Hellwig (1):
  target/file: allocate the bvec array as part of struct
    target_core_file_cmd

Pavel Begunkov (6):
  splice: don't generate zero-len segement bvecs
  bvec/iter: disallow zero-length segment bvecs
  block/psi: remove PSI annotations from direct IO
  iov_iter: optimise bvec iov_iter_advance()
  bio: add a helper calculating nr segments to alloc
  bio: don't copy bvec for direct IO

 Documentation/filesystems/porting.rst | 16 ++++++
 block/bio.c                           | 71 +++++++++++++--------------
 drivers/target/target_core_file.c     | 20 +++-----
 fs/block_dev.c                        |  7 +--
 fs/direct-io.c                        |  2 +
 fs/iomap/direct-io.c                  |  9 ++--
 fs/splice.c                           |  9 ++--
 include/linux/bio.h                   | 13 +++++
 lib/iov_iter.c                        | 21 +++++++-
 9 files changed, 103 insertions(+), 65 deletions(-)

-- 
2.24.0

Powered by blists - more mailing lists