Revision b91e1302ad9b80c174a4855533f7e3aa2873355e authored by Linus Torvalds on 27 December 2016, 19:40:38 UTC, committed by Linus Torvalds on 29 December 2016, 19:03:15 UTC
In commit 62906027091f ("mm: add PageWaiters indicating tasks are
waiting for a page bit") Nick Piggin made our page locking no longer
unconditionally touch the hashed page waitqueue, which not only helps
performance in general, but is particularly helpful on NUMA machines
where the hashed wait queues can bounce around a lot.

However, the "clear lock bit atomically and then test the waiters bit"
sequence turns out to be much more expensive than it needs to be,
because you get a nasty stall when trying to access the same word that
just got updated atomically.

On architectures where locking is done with LL/SC, this would be trivial
to fix with a new primitive that clears one bit and tests another
atomically, but that ends up not working on x86, where the only atomic
operations that return the result end up being cmpxchg and xadd.  The
atomic bit operations return the old value of the same bit we changed,
not the value of an unrelated bit.

On x86, we could put the lock bit in the high bit of the byte, and use
"xadd" with that bit (where the overflow ends up not touching other
bits), and look at the other bits of the result.  However, an even
simpler model is to just use a regular atomic "and" to clear the lock
bit, and then the sign bit in eflags will indicate the resulting state
of the unrelated bit #7.

So by moving the PageWaiters bit up to bit #7, we can atomically clear
the lock bit and test the waiters bit on x86 too.  And architectures
with LL/SC (which is all the usual RISC suspects), the particular bit
doesn't matter, so they are fine with this approach too.

This avoids the extra access to the same atomic word, and thus avoids
the costly stall at page unlock time.

The only downside is that the interface ends up being a bit odd and
specialized: clear a bit in a byte, and test the sign bit.  Nick doesn't
love the resulting name of the new primitive, but I'd rather make the
name be descriptive and very clear about the limitation imposed by
trying to work across all relevant architectures than make it be some
generic thing that doesn't make the odd semantics explicit.

So this introduces the new architecture primitive

    clear_bit_unlock_is_negative_byte();

and adds the trivial implementation for x86.  We have a generic
non-optimized fallback (that just does a "clear_bit()"+"test_bit(7)"
combination) which can be overridden by any architecture that can do
better.  According to Nick, Power has the same hickup x86 has, for
example, but some other architectures may not even care.

All these optimizations mean that my page locking stress-test (which is
just executing a lot of small short-lived shell scripts: "make test" in
the git source tree) no longer makes our page locking look horribly bad.
Before all these optimizations, just the unlock_page() costs were just
over 3% of all CPU overhead on "make test".  After this, it's down to
0.66%, so just a quarter of the cost it used to be.

(The difference on NUMA is bigger, but there this micro-optimization is
likely less noticeable, since the big issue on NUMA was not the accesses
to 'struct page', but the waitqueue accesses that were already removed
by Nick's earlier commit).

Acked-by: Nick Piggin <npiggin@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Bob Peterson <rpeterso@redhat.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Andrew Lutomirski <luto@kernel.org>
Cc: Andreas Gruenbacher <agruenba@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1 parent 2d706e7
Raw File
Kconfig
config UFS_FS
	tristate "UFS file system support (read only)"
	depends on BLOCK
	help
	  BSD and derivate versions of Unix (such as SunOS, FreeBSD, NetBSD,
	  OpenBSD and NeXTstep) use a file system called UFS. Some System V
	  Unixes can create and mount hard disk partitions and diskettes using
	  this file system as well. Saying Y here will allow you to read from
	  these partitions; if you also want to write to them, say Y to the
	  experimental "UFS file system write support", below. Please read the
	  file <file:Documentation/filesystems/ufs.txt> for more information.

          The recently released UFS2 variant (used in FreeBSD 5.x) is
          READ-ONLY supported.

	  Note that this option is generally not needed for floppies, since a
	  good portable way to transport files and directories between unixes
	  (and even other operating systems) is given by the tar program ("man
	  tar" or preferably "info tar").

	  When accessing NeXTstep files, you may need to convert them from the
	  NeXT character set to the Latin1 character set; use the program
	  recode ("info recode") for this purpose.

	  To compile the UFS file system support as a module, choose M here: the
	  module will be called ufs.

	  If you haven't heard about all of this before, it's safe to say N.

config UFS_FS_WRITE
	bool "UFS file system write support (DANGEROUS)"
	depends on UFS_FS
	help
	  Say Y here if you want to try writing to UFS partitions. This is
	  experimental, so you should back up your UFS partitions beforehand.

config UFS_DEBUG
	bool "UFS debugging"
	depends on UFS_FS
	help
	  If you are experiencing any problems with the UFS filesystem, say
	  Y here.  This will result in _many_ additional debugging messages to be
	  written to the system log.
back to top