[PATCH v7 0/4] vfs: Use per-cpu list for SB's s_inodes list

From: Waiman Long
Date: Tue Apr 12 2016 - 18:55:24 EST


v6->v7:
- Fix the race condition in __pcpu_list_next_cpu() as reported by
Jan Kara.
- No changes in patches 2-4.

v5->v6:
- Remove patch 5 which can increase the kernel testing matrix.
- Disable preemption in pcpu_list_add() as it was complained by
the 0-day test even though it is not technically necessary.
- Add a PERCPU_LIST_WARN_ON() macro to simplify code.
- No changes in patches 2-4.

v4->v5:
- Fix the UP panic problem reported by 0day test by unifying the SMP
and UP code.
- Add patch 5 to add a new kernel config parameter to allow disabling
per-cpu list for small systems that won't benefit much from this
feature.

v3->v4:
- Fix some racing conditions in the code.
- Add another patch from Jan to replace list_for_each_entry_safe()
by list_for_each_entry().
- Add lockdep annotation.

v2->v3:
- Directly replace list_for_each_entry() and
list_for_each_entry_safe() by pcpu_list_iterate() and
pcpu_list_iterate_safe() respectively instead. Those 2 functions
provide a stateful per-cpu list iteration interface.
- Include Jan Kara's patch to clean up the fsnotify_unmount_inodes()
function.

v1->v2:
- Use separate structures for list head and nodes & provide a
cleaner interface.
- Use existing list_for_each_entry() or list_for_each_entry_safe()
macros for each of the sb's s_inodes iteration functions instead
of using list_for_each_entry_safe() for all of them which may not
be safe in some cases.
- Use an iterator interface to access all the nodes of a group of
per-cpu lists. This approach is cleaner than the previous double-for
macro which is kind of hacky. However, it does require more lines
of code changes.
- Add a preparatory patch 2 to extract out the per-inode codes from
the superblock s_inodes list iteration functions to minimize code
changes needed in the patch 3.

This patch is a replacement of my previous list batching patch -
https://lwn.net/Articles/674105/. Compared with the previous patch,
this one provides better performance and fairness. However, it also
requires a bit more changes in the VFS layer.

This patchset is a derivative of Andi Kleen's patch on "Initial per
cpu list for the per sb inode list"

https://git.kernel.org/cgit/linux/kernel/git/ak/linux-misc.git/commit/?h=hle315/
combined&id=f1cf9e715a40f44086662ae3b29f123cf059cbf4

Patch 1 introduces the per-cpu list.

Patch 2 cleans up the fsnotify_unmount_inodes() function by making
the code simpler and more standard.

Patch 3 replaces the use of list_for_each_entry_safe() in
evict_inodes() and invalidate_inodes() by list_for_each_entry().

Patch 4 modifies the superblock and inode structures to use the per-cpu
list. The corresponding functions that reference those structures
are modified.

Jan Kara (2):
fsnotify: Simplify inode iteration on umount
vfs: Remove unnecessary list_for_each_entry_safe() variants

Waiman Long (2):
lib/percpu-list: Per-cpu list with associated per-cpu locks
vfs: Use per-cpu list for superblock's inode list

fs/block_dev.c | 13 ++-
fs/drop_caches.c | 10 +-
fs/fs-writeback.c | 13 ++-
fs/inode.c | 40 +++-----
fs/notify/inode_mark.c | 53 +++--------
fs/quota/dquot.c | 16 ++--
fs/super.c | 7 +-
include/linux/fs.h | 8 +-
include/linux/percpu-list.h | 231 +++++++++++++++++++++++++++++++++++++++++++
lib/Makefile | 2 +-
lib/percpu-list.c | 100 +++++++++++++++++++
11 files changed, 397 insertions(+), 96 deletions(-)
create mode 100644 include/linux/percpu-list.h
create mode 100644 lib/percpu-list.c