diff options
author | Eric Biggers <ebiggers@google.com> | 2017-03-31 19:31:48 +0200 |
---|---|---|
committer | Al Viro <viro@zeniv.linux.org.uk> | 2017-04-03 07:05:57 +0200 |
commit | 64bd72048a2ac07efed70debe606a1c6e5e03554 (patch) | |
tree | 1666d9e34b9a4a47ed54ab9eceabf4f56d6dc39a /drivers/devfreq | |
parent | statx: remove incorrect part of vfs_statx() comment (diff) | |
download | linux-64bd72048a2ac07efed70debe606a1c6e5e03554.tar.xz linux-64bd72048a2ac07efed70debe606a1c6e5e03554.zip |
statx: optimize copy of struct statx to userspace
I found that statx() was significantly slower than stat(). As a
microbenchmark, I compared 10,000,000 invocations of fstat() on a tmpfs
file to the same with statx() passed a NULL path:
$ time ./stat_benchmark
real 0m1.464s
user 0m0.275s
sys 0m1.187s
$ time ./statx_benchmark
real 0m5.530s
user 0m0.281s
sys 0m5.247s
statx is expected to be a little slower than stat because struct statx
is larger than struct stat, but not by *that* much. It turns out that
most of the overhead was in copying struct statx to userspace, mostly in
all the stac/clac instructions that got generated for each __put_user()
call. (This was on x86_64, but some other architectures, e.g. arm64,
have something similar now too.)
stat() instead initializes its struct on the stack and copies it to
userspace with a single call to copy_to_user(). This turns out to be
much faster, and changing statx to do this makes it almost as fast as
stat:
$ time ./statx_benchmark
real 0m1.624s
user 0m0.270s
sys 0m1.354s
For zeroing the reserved fields, start by zeroing the full struct with
memset. This makes it clear that every byte copied to userspace is
initialized, even implicit padding bytes (though there are none
currently). In the scenarios I tested, it also performed the same as a
designated initializer. Manually initializing each field was still
slightly faster, but would have been more error-prone and less
verifiable.
Also rename statx_set_result() to cp_statx() for consistency with
cp_old_stat() et al., and make it noinline so that struct statx doesn't
add to the stack usage during the main portion of the syscall execution.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Diffstat (limited to 'drivers/devfreq')
0 files changed, 0 insertions, 0 deletions