Re: [Xen-ia64-devel] Some things to be considered for ptc.ga emulation

[Tristan Gingold <Tristan.Gingold@xxxxxxxx>]

Le Vendredi 07 Avril 2006 10:07, Xu, Anthony a écrit :
> > Tristan Gingold
> >Sent: 2006?4?6? 15:28
> >+void vhpt_flush_address_remote(int cpu,
> >+                           unsigned long vadr, unsigned long addr_range)
> >+{
> >+    while ((long)addr_range > 0) {
> >+            /* Get the VHPT entry.  */
> >+            unsigned int off = ia64_thash(vadr) - VHPT_ADDR;
> >+            volatile struct vhpt_lf_entry *v;
> >+            v =__va(per_cpu(vhpt_paddr, cpu) + off);
> >+            v->ti_tag = INVALID_TI_TAG;
> >             addr_range -= PAGE_SIZE;
> >             vadr += PAGE_SIZE;
> >     }
>
> I have a concern about this,
> This method seems be Ok if only modifying v->ti_tag, which is an atomic
> operation on bus.
> The possible optimization maybe check tag before setting it to invalid,
> (this is unreasonable to purge a irrelevant tlb mapping)
> There is a time between checking tag and setting it to invalid, this may
> cause race condition.
The race is we don't invalidate while we had to.  This means a ptc.ga is 
concurrent to an itc.
For me there is no race: the itc wins.  Of course, it may loose to do the cpu 
itc, but it doesn't matter.

Tristan.



_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel