Re: [PATCH] x86 spinlock: Fix memory corruption on completing completions

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 02/10/2015 06:23 AM, Linus Torvalds wrote:
On Mon, Feb 9, 2015 at 4:02 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
On Mon, Feb 09, 2015 at 03:04:22PM +0530, Raghavendra K T wrote:
So we have 3 choices,
1. xadd
2. continue with current approach.
3. a read before unlock and also after that.

For the truly paranoid we have probe_kernel_address(), suppose the lock
was in module space and the module just got unloaded under us.

That's much too expensive.

The xadd shouldn't be noticeably more expensive than the current
"add_smp()". Yes, "lock xadd" used to be several cycles slower than
just "lock add" on some early cores, but I think these days it's down
to a single-cycle difference, which is not really different from doing
a separate load after the add.

The real problem with xadd used to be that we always had to do magic
special-casing for i386, but that's one of the reasons we dropped
support for original 80386.

So I think Raghavendra's last version (which hopefully fixes the
lockup problem that Sasha reported) together with changing that

V2 did pass the stress, but getting confirmation Sasha would help.


         add_smp(&lock->tickets.head, TICKET_LOCK_INC);
         if (READ_ONCE(lock->tickets.tail) & TICKET_SLOWPATH_FLAG) ..

into something like

         val = xadd((&lock->ticket.head_tail, TICKET_LOCK_INC << TICKET_SHIFT);
         if (unlikely(val & TICKET_SLOWPATH_FLAG)) ...

would be the right thing to do. Somebody should just check that I got
that shift right, and that the tail is in the high bytes (head really
needs to be high to work, if it's in the low byte(s) the xadd would
overflow from head into tail which would be wrong).

Unfortunately xadd could result in head overflow as tail is high.

The other option was repeated cmpxchg which is bad I believe.
Any suggestions?

( I have the V3 with Oleg's suggestion and performance numbers but
without this getting resolved, It will be one unnecessary iteration).

How about getting rid off SLOW_PATH_FLAG in spinlock (i.e. use it only
 as hint for paravirt), but do unlock_kick whenever we see that
(tail-head) > TICKET_LOCK_INC?. (but this also may need cmpxchg in loop
in unlock but we will be able to get rid of clear slowpath logic)

Only problem is we may do unnecessary kicks even in 1x load.








_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization




[Index of Archives]     [KVM Development]     [Libvirt Development]     [Libvirt Users]     [CentOS Virtualization]     [Netdev]     [Ethernet Bridging]     [Linux Wireless]     [Kernel Newbies]     [Security]     [Linux for Hams]     [Netfilter]     [Bugtraq]     [Yosemite Forum]     [MIPS Linux]     [ARM Linux]     [Linux RAID]     [Linux Admin]     [Samba]

  Powered by Linux