On 27.07.2022 20:25, Andrew Cooper wrote: > On 26/07/2022 17:06, Jan Beulich wrote: >> While "type" can include PGT_pae_xen_l2, "x" can't, as the bit is >> cleared upon de-validation (see also the respective assertion earlier in >> the function). > > While this statement is true, it doesn't really explain why this is > relevant (or not) to TLB flushing. > > As far as the change goes, it's safe on 64bit builds of Xen (I think), > but would not be on 32bit builds when this logic was first written.
How that? The flush is needed if the type changes. If (for the purposes here) PGT_pae_xen_l2 was considered part of the type, then the bit being lost upon de-validation would mean we're flushing too little. Jan
