On Wed, Sep 14, 2016 at 02:37:03PM +1000, Benjamin Herrenschmidt wrote: > On Wed, 2016-09-14 at 09:23 +0530, Nikunj A Dadhania wrote: > > Hr... this is confusing, let me rephrase ;-) > > > Due to lazy tlb flushes, propagation of the tlb flush is delayed. > Moreover, certain operations need to do broadcast flush, this too can > be > > delayed until we hit the operation that warrant a broadcast. > > Instead: > > We flush the qemu TLB lazily. check_tlb_flush is called whenever we > hit a context synchronizing event or instruction that requires a pending > flush to be performed. > > However, we fail to handle broadcast TLB flush operations. In order > to fix that efficiently, we want to differenciate whether check_tlb_flush() > needs to only apply pending local flushes (isync instructions, > interrupts, ...) or also global pending flush operations. The latter > is only needed when executing instructions that are defined architecturally > as synchronizing global TLB flush operations. This in our case is ptesync > on BookS and tlbsync on BookE along with the paravirtualized hypervisor > calls.
Nice.
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
signature.asc
Description: PGP signature
