On Thu, 2018-01-25 at 11:26 +0100, Juergen Gross wrote: > On 25/01/18 11:22, Peter Zijlstra wrote: > > > > On Thu, Jan 25, 2018 at 10:02:05AM +0000, David Woodhouse wrote: > > > > > > On Tue, 2018-01-23 at 16:25 +0100, Peter Zijlstra wrote: > > > > > > > > Paravirt emits indirect calls which get flagged by objtool retpoline > > > > checks, annotate it away because all these indirect calls will be > > > > patched out before we start userspace. > > > I've seen this asserted repeatedly but I've never truly convinced > > > myself of it. Is this absolutely unconditionally true in every case, > > > even when we're running as a guest and there are *actual* calls to be > > > made? We turn them into direct calls, never leave them indirect? > > That is my understanding; and when I worked on the paravirt spinlock > > code and disassembled live guest code this seemed to have happend. > > > > But let me go read the paravirt code again to make a stronger argument > > in favour. > > > paravirt_patch_default() is the function you want to look at: it either > replaces the indirect call by some cutom code (which is never an > indirect call) or by a call of the target function. OK, my brain hurts a bit but I'm happy now. Thank you.