I am currently tracking a hotlock reported by a customer on a large, 512 cores,
system, I am currently running 3.6.0 rc1 but the issue looks like it has been
this way for a very long time.
The offending lock is proc_dir_entry->pde_unload_lock.
In proc_reg_release we are doing a kfree under the spinlock which is ok but it
means we are holding the lock longer then required. Scaling improved when I
moved kfree out.
Also shouldn't the comment on pde_unload_lock also note that pde_openers and
pde_unload_completion are both used under the lock?
Here is some data from quick test program which just reads from /proc/cpuinfo.
Lower is better, as you can see the worst case scenario is improved.
baseline moved kfree
tasks read-sec read-sec
1 0.0141 0.0141
2 0.0140 0.0140
4 0.0140 0.0141
8 0.0145 0.0145
16 0.0553 0.0548
32 0.1688 0.1622
64 0.5017 0.3856
128 1.7005 0.9710
256 5.2513 2.6519
512 8.0529 6.2976
If the patch looks agreeable I will resend it properly.
diff --git a/fs/proc/inode.c b/fs/proc/inode.c
index 7ac817b..46016c1 100644
--- a/fs/proc/inode.c
+++ b/fs/proc/inode.c
@@ -403,9 +403,11 @@ static int proc_reg_release(struct inode *inode, struct file *file)
release = pde->proc_fops->release;
if (pdeo) {
list_del(&pdeo->lh);
- kfree(pdeo);
}
spin_unlock(&pde->pde_unload_lock);
+ if (pdeo) {
+ kfree(pdeo);
+ }
if (release)
rv = release(inode, file);
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
