Brian Stecher
|
Re: CPU synchronization during TLB flush
|
Brian Stecher
02/27/2008 9:50 AM
post5317
|
Re: CPU synchronization during TLB flush
On Wed, Feb 27, 2008 at 08:40:51AM -0500, Varaprasad Lingutla wrote:
> I am using Qnx 6.2 on 2 CPU system. Once in a while one of the CPUs end up
> looping in smp_flush_tlb() for "tlb_flush_cnt" to go zero. The idle()
thread
> is running on other CPU. Has anyone seen this type of problem before or
has
> any idea whats causing this?
I can't really say - we don't have a "tlb_flush_cnt" symbol in any of our
versions, so it looks like something your company has modified.
> In recent versions, I don't see any synchronization between CPUs when
> IPI_TLB_FLUSH is issued. Can we assume that TLBs are flushed?
Since IPI's are interrupts, they'll get handled before any further user
code is executed on the target CPUs, unless interrupts are being left
off.
Brian
--
Brian Stecher (bstecher@qnx.com) QNX Software Systems
phone: +1 (613) 591-0931 (voice) 175 Terence Matthews Cr.
+1 (613) 591-3579 (fax) Kanata, Ontario, Canada K2M 1W8
|
|
|