Re: [PATCH] fs: avoid locking sb_lock in grab_super_passive()
From: Andrew Morton
Date: Fri Feb 20 2015 - 18:07:37 EST
On Thu, 19 Feb 2015 20:19:35 +0300 Konstantin Khlebnikov <khlebnikov@xxxxxxxxxxxxxx> wrote:
Please cc Dave Chinner on this.
> I've noticed significant locking contention in memory reclaimer around
> sb_lock inside grab_super_passive(). Grab_super_passive() is called from
> two places: in icache/dcache shrinkers (function super_cache_scan) and
> from writeback (function __writeback_inodes_wb). Both are required for
> progress in memory reclaimer.
> Also this lock isn't irq-safe. And I've seen suspicious livelock under
> serious memory pressure where reclaimer was called from interrupt which
> have happened right in place where sb_lock is held in normal context,
> so all other cpus were stuck on that lock too.
You mean someone is calling grab_super_passive() (ie: fs writeback)
from interrupt context? What's the call path?
> Grab_super_passive() acquires sb_lock to increment sb->s_count and check
> sb->s_instances. It seems sb->s_umount locked for read is enough here:
> super-block deactivation always runs under sb->s_umount locked for write.
> Protecting super-block itself isn't a problem: in super_cache_scan() sb
> is protected by shrinker_rwsem: it cannot be freed if its slab shrinkers
> are still active. Inside writeback super-block comes from inode from bdi
> writeback list under wb->list_lock.
> This patch removes locking sb_lock and checks s_instances under s_umount:
> generic_shutdown_super() unlinks it under sb->s_umount locked for write.
> Now successful grab_super_passive() only locks semaphore, callers must
> call up_read(&sb->s_umount) instead of drop_super(sb) when they're done.
The patch looks reasonable to me, but the grab_super_passive()
documentation needs further updating, please.
- It no longer "acquires a reference". All it does is to acquire an rwsem.
- What the heck is a "passive reference" anyway? It appears to be
the situation where we increment s_count without incrementing s_active.
After your patch, this superblock state no longer exists(?), so
perhaps the entire "passive reference" concept and any references to
it can be expunged from the kernel.
And grab_super_passive() should be renamed anyway. It no longer
"grabs" anything - it attempts to acquire ->s_umount.
- While we're dicking with the grab_super_passive() documentation,
let's turn it into kerneldoc by adding the /**.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/