Re: [Xen-devel] credit scheduler svc->flags access race?

["Liu.yi" <liu.yi24@xxxxxxxxxx>]

Hi, Keir and George, thanks for your concerns.

Access race to svc->flags seems inevitably when csched_acct calls
vcpu_pause_nosync:
    vm excutes PAUSE instruction.
    hypercalls call csched_vcpu_yield.

I had think about taking csched_private->lock in csched_vcpu_yield, but
csched_acct takes this lock for a long time so that csched_vcpu_yield may
been blocked for long time too, so I chooes set_bit and clear_bit.

Atomic operations using LOCK prefix (like set_bit and clear_bit) will block
all the physical cpus which excute memory access. Does this fact implies
that spin locks are more efficient when their's granularity is small?

Sorry for my poor english.

Liuyi

--
View this message in context: 
http://xen.1045712.n5.nabble.com/credit-scheduler-svc-flags-access-race-tp5111504p5121613.html
Sent from the Xen - Dev mailing list archive at Nabble.com.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel