Benjamin Herrenschmidt <b...@kernel.crashing.org> writes: > On Fri, 2016-09-09 at 09:53 +0530, Nikunj A Dadhania wrote: >> tlbie (and H_REMOVE for pseries) should have a global effect. This is >> achieved by iterating and setting tlb_need_flush in all the CPUs. >> >> Suggested-by: Benjamin Herrenschmidt <b...@kernel.crashing.org> >> Signed-off-by: Nikunj A Dadhania <nik...@linux.vnet.ibm.com> >> >> -- >> >> Note: Haven't changed following POWERPC_MMU_32B and POWERPC_MMU_601 >> yet. >> As I am not sure about it. > > 604 and 7400 can do SMP.
Sure, will add similar logic there. > That said, I think the approach in your patch is going to be a bit big > of a hammer. > We should have a separate flag indicating that we need to > broadcast a flush and only set it on tlbie (non-l) and tlbivax (on > BookE) so that we don't end up doing expensive broadcasts on things > like context switches. I had implemented that initially, and was checking that in check_tlb_flush, the logic was getting complicated, so thought about this way. Moreover, I was thinking about it, that needs to be a global tcg flag and not part the CPUPPCState structure, I was then worried about how to synchronise a global tcg variable. > We keep the existing logic to flush locally. We additionally replace > the one in ptesync (BookS) or tlbsync (BookE) to test for the broadcast > flag, and flush the "other" CPUs if set. > > That also means you have a nice spot to do the more complex MT-TCG > broadcast only when needed in the future. > Regards Nikunj
