The normal read_seqcount_begin() function will wait for any current
writers to exit their critical region by looping until the sequence
count is even.
That "wait for sequence count to stabilize" is the right thing to do if
the read-locker will just retry the whole operation on contention: no
point in doing a potentially expensive reader sequence if we know at the
beginning that we'll just end up re-doing it all.
HOWEVER. Some users don't actually retry the operation, but instead
will abort and do the operation with proper locking. So the sequence
count case may be the optimistic quick case, but in the presense of
writers you may want to do full locking in order to guarantee forward
progress. The prime example of this would be the RCU name lookup.
And in that case, you may well be better off without the "retry early",
and are in a rush to instead get to the failure handling. Thus this
"raw" interface that just returns the sequence number without testing it
- it just forces the low bit to zero so that read_seqcount_retry() will
always fail such a "active concurrent writer" scenario.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We really need to use a ACCESS_ONCE() on the sequence value read in
__read_seqcount_begin(), because otherwise the compiler might end up
reloading the value in between the test and the return of it. As a
result, it might end up returning an odd value (which means that a write
is in progress).
If the reader is then fast enough that that odd value is still the
current one when the read_seqcount_retry() is done, we might end up with
a "successful" read sequence, even despite the concurrent write being
active.
In practice this probably never really happens - there just isn't
anything else going on around the read of the sequence count, and the
common case is that we end up having a read barrier immediately
afterwards.
So the code sequence in which gcc might decide to reaload from memory is
small, and there's no reason to believe it would ever actually do the
reload. But if the compiler ever were to decide to do so, it would be
incredibly annoying to debug. Let's just make sure.
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It uses cpu_relax(), and so needs <asm/processor.h>
Without this patch, I see:
CC arch/mn10300/kernel/asm-offsets.s
In file included from include/linux/time.h:8,
from include/linux/timex.h:56,
from include/linux/sched.h:57,
from arch/mn10300/kernel/asm-offsets.c:7:
include/linux/seqlock.h: In function 'read_seqbegin':
include/linux/seqlock.h:91: error: implicit declaration of function 'cpu_relax'
whilst building asb2364_defconfig on MN10300.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
All static seqlock should be initialized with the lockdep friendly
__SEQLOCK_UNLOCKED() macro.
Remove legacy SEQLOCK_UNLOCKED() macro.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: David Miller <davem@davemloft.net>
Link: http://lkml.kernel.org/r/%3C1306238888.3026.31.camel%40edumazet-laptop%3E
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Move the smp_rmb after cpu_relax loop in read_seqlock and add
ACCESS_ONCE to make sure the test and return are consistent.
A multi-threaded core in the lab didn't like the update
from 2.6.35 to 2.6.36, to the point it would hang during
boot when multiple threads were active. Bisection showed
af5ab277de (clockevents:
Remove the per cpu tick skew) as the culprit and it is
supported with stack traces showing xtime_lock waits including
tick_do_update_jiffies64 and/or update_vsyscall.
Experimentation showed the combination of cpu_relax and smp_rmb
was significantly slowing the progress of other threads sharing
the core, and this patch is effective in avoiding the hang.
A theory is the rmb is affecting the whole core while the
cpu_relax is causing a resource rebalance flush, together they
cause an interfernce cadance that is unbroken when the seqlock
reader has interrupts disabled.
At first I was confused why the refactor in
3c22cd5709 (kernel: optimise
seqlock) didn't affect this patch application, but after some
study that affected seqcount not seqlock. The new seqcount was
not factored back into the seqlock. I defer that the future.
While the removal of the timer interrupt offset created
contention for the xtime lock while a cpu does the
additonal work to update the system clock, the seqlock
implementation with the tight rmb spin loop goes back much
further, and is just waiting for the right trigger.
Cc: <stable@vger.kernel.org>
Signed-off-by: Milton Miller <miltonm@bga.com>
Cc: <linuxppc-dev@lists.ozlabs.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Link: http://lkml.kernel.org/r/%3Cseqlock-rmb%40mdm.bga.com%3E
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Add branch annotations for seqlock read fastpath, and introduce
__read_seqcount_begin and __read_seqcount_end functions, that can avoid the
smp_rmb() if used carefully. These will be used by store-free path walking
algorithm performance is critical and seqlocks are in use.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Thomas Gleixner debugged a particularly ugly seqlock related livelock:
do not process the seq-read section if we know it beforehand that the
test at the end of the section will fail ...
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Correct mis-spellings of "algorithm", "appear", "consistent" and
(shame, shame) "kernel".
Signed-off-by: Robert P. J. Day <rpjday@mindspring.com>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
seqlock_init() needs to use spin_lock_init() for dynamic locks, so that
lockdep is notified about the presence of a new lock.
(this is a fallout of the recent networking merge, which started using
the so-far unused seqlock_init() API.)
This fix solves the following lockdep-internal warning on current -git:
INFO: trying to register non-static key.
the code is fine but needs lockdep annotation.
turning off the locking correctness validator.
__lock_acquire+0x10c/0x9f9
lock_acquire+0x56/0x72
_spin_lock+0x35/0x42
neigh_destroy+0x9d/0x12e
neigh_periodic_timer+0x10a/0x15c
run_timer_softirq+0x126/0x18e
__do_softirq+0x6b/0xe6
do_softirq+0x64/0xd2
ksoftirqd+0x82/0x138
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Locking init improvement:
- introduce and use __SPIN_LOCK_UNLOCKED for array initializations,
to pass in the name string of locks, used by debugging
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
In vsyscall function do_vgettimeofday(), some functions are declared as
inlined, which is a hint for gcc to compile the function inlined but it
not forced. Sometimes compiler does not compile the function as
inlined, so here inline is replaced by __always_inline prefix.
It does not happen in gcc compiler actually, but it possibly happens.
Signed-off-by: bibo mao <bibo.mao@intel.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
