On 30/05/2017 16:37, Mark Wielaard wrote: > On Tue, 2017-05-30 at 13:28 +0530, Naveen N. Rao wrote: >>> + if (! readfunc(sp + LR_OFFSET, &newLr, arg)) >> >> should this be newSp + LR_OFFSET, since the LR save area is in the >> caller's stack frame? > > Of course. You are entirely right. > > This also explains why the testcase didn't seem to work correctly. > It actually does with this change. I should have been more suspicious > about that then just assuming it was because the link register couldn't > always be retrieved correctly. It actually can if we pick it from the > right location! > > Attached the fixed patch plus a testcase (actual binaries removed from > patch, but testcase documents how they were generated - precisely like > for the other arches).
Great, I'll resubmit my kernel patch then! Paolo
