]> git.ipfire.org Git - thirdparty/linux.git/commit
device_cgroup: remove branch hint after code refactor
authorBreno Leitao <leitao@debian.org>
Wed, 7 Jan 2026 14:06:36 +0000 (06:06 -0800)
committerChristian Brauner <brauner@kernel.org>
Wed, 14 Jan 2026 15:50:36 +0000 (16:50 +0100)
commit6784f274722559c0cdaaa418bc8b7b1d61c314f9
treec5fa2999421dd6e5898a605c4fd11454c9c52d60
parentedecd1ae594abc8b14ed36c438d1a7db729403ba
device_cgroup: remove branch hint after code refactor

commit 4ef4ac360101 ("device_cgroup: avoid access to ->i_rdev in the
common case in devcgroup_inode_permission()") reordered the checks in
devcgroup_inode_permission() to check the inode mode before checking
i_rdev, for better cache behavior.

However, the likely() annotation on the i_rdev check was not updated
to reflect the new code flow. Originally, when i_rdev was checked
first, likely(!inode->i_rdev) made sense because most inodes were(?)
regular files/directories, thus i_rdev == 0.

After the reorder, by the time we reach the i_rdev check, we have
already confirmed the inode IS a block or character device. Block and
character special files are precisely defined by having a device number
(i_rdev), so !inode->i_rdev is now the rare edge case, not the common
case.

Branch profiling confirmed this is 100% mispredicted:

  correct incorrect  %    Function                      File              Line
  ------- ---------  -    --------                      ----              ----
        0   2631904 100   devcgroup_inode_permission    device_cgroup.h   24

Remove likely() to avoid giving the wrong hint to the CPU.

Fixes: 4ef4ac360101 ("device_cgroup: avoid access to ->i_rdev in the common case in devcgroup_inode_permission()")
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260107-likely_device-v1-1-0c55f83a7e47@debian.org
Reviewed-by: Mateusz Guzik <mjguzik@gmail.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
include/linux/device_cgroup.h