On 19/03/2025 9:05 am, Roger Pau Monné wrote: > On Tue, Mar 18, 2025 at 08:10:33PM +0000, Andrew Cooper wrote: >> The LRET is detached from the PUSHes which set it up, and this is about to >> get >> worse with the changes to trampoline relocation. For the sake of one >> variable >> read, the complexity is not worth it. >> >> Reorder the logic to copy the trampoline into place, then switch stack and >> enter the trampoline. >> >> No functional change. >> >> Signed-off-by: Andrew Cooper <andrew.coop...@citrix.com> > Reviewed-by: Roger Pau Monné <roger....@citrix.com> > >> --- >> CC: Jan Beulich <jbeul...@suse.com> >> CC: Roger Pau Monné <roger....@citrix.com> >> >> Roger: I'd like this to be a prerequisite to your "[PATCH 4/7] x86/boot: >> apply >> trampoline relocations at destination position" to avoid the movement of >> reloc_trampoline32() making things worse. > I think you could commit this now-ish, and I can rebase on top?
CI said no, and the bug is hiding in plain sight. The setup for the rep movs: lea sym_esi(trampoline_start), %esi mov sym_esi(trampoline_phys), %edi is buggy. I'll try and find a nicer way to do this. ~Andrew
