rwsem: wake queued readers when writer blocks on active read lock
in [Kernel]
|
Prev: ARM: Fix gen_nand probe structures contents
Next: + drivers-acpi-apei-erst-dbgc-get_useru64-doesnt-work-on-i386.patch added to -mm tree
From: Tony Luck on 11 Aug 2010 21:30 On Mon, May 24, 2010 at 1:31 PM, Michel Lespinasse <walken(a)google.com> wrote: > This change addresses the following situation: > > - Thread A acquires the rwsem for read > - Thread B tries to acquire the rwsem for write, notices there is already > �an active owner for the rwsem. > - Thread C tries to acquire the rwsem for read, notices that thread B already > �tried to acquire it. > - Thread C grabs the spinlock and queues itself on the wait queue. > - Thread B grabs the spinlock and queues itself behind C. At this point A is > �the only remaining active owner on the rwsem. > > In this situation thread B could notice that it was the last active writer > on the rwsem, and decide to wake C to let it proceed in parallel with A > since they both only want the rwsem for read. > > Signed-off-by: Michel Lespinasse <walken(a)google.com> Linus tree this morning[1] was behaving badly on ia64 ... processes would wander off into some unkillable state ... and since this happened to processes starting from rc*.d I couln't get the system up to a login prompt. System is a 32-way (4 sockets * quad-core * hyperthread). git bisect pins the blame on this change (commit 424acaaeb...). Reverting it (and it's successor a8618a0e - because I assumed that it depended on 424...) gives me a kernel that works fine. Not sure what is wrong with this change. Maybe ia64 needs some more memory ordering bits than the changed code provides? I can dig into it a bit harder tomorrow, but I thought you'd like an early heads-up in case anyone else is seeing similar problems. -Tony [1] N.B. I've been on vacation - so this morning's tree was the first I'd tried since 2.6.35 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Michel Lespinasse on 12 Aug 2010 01:10
On Wed, Aug 11, 2010 at 10:02 PM, Michel Lespinasse <walken(a)google.com> wrote: > In arch/ia64/include/asm/rwsem.h I see RWSEM_WAITING_BIAS defined as > -__IA64_UL_CONST(0x0000000100000000) > > This makes it a large, positive unsigned value. This is probably > throwing off the rwsem_atomic_update(0, sem) < RWSEM_WAITING_BIAS > comparison in my patch (supposed to be long versus long, but actually > is long versus unsigned long on ia64). FYI, I just verified that RWSEM_WAITING_BIAS is defined as signed on all architectures except for ia64. So, this would be consistent with the issue being observed only on ia64. -- Michel "Walken" Lespinasse A program is never fully debugged until the last user dies. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |