On 28/08/2019 15:10, Jan Beulich wrote: > On 12.08.2019 20:21, Andrew Cooper wrote: >> load_TR() is used exclusively in the resume path, but jumps through a lot of >> unnecessary hoops. >> >> As suspend/resume is strictly on CPU0 in idle context, the correct GDT to use >> is boot_gdt, which means it doesn't need saving on suspend. Similarly, the >> correct IDT to use can be derived, and the LDT is guaranteed to be NUL. >> >> The TR is still correct in the GDT, but needs the busy bit clearing before we >> can reload it. >> >> Signed-off-by: Andrew Cooper <andrew.coop...@citrix.com> >> --- >> CC: Jan Beulich <jbeul...@suse.com> >> CC: Wei Liu <w...@xen.org> >> CC: Roger Pau Monné <roger....@citrix.com> >> >> A slightly different option would be to call load_system_tables() rather than >> opencoding part of it in restore_rest_processor_state(). However, that is >> more setup than is necessary. Thoughts? > This might indeed be better (despite the parts not needed on this > path), as it would further centralize the handling. Of course only > as long as there's no (significant) change needed to the function > in order to be used here.
No change at all. The AP's already do go through load_system_tables() on S3 resume. It is slightly suboptimal that we get two identical lgdt's this close together, but it is the only way of ending up entering C with __HYPERVISOR_CS. At some future point, I'll see how easy it is to arrange the trampoline GDT to be usable with __HYPERVISOR_CS, which would simplify the boot and resume paths. ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
