On Wed, Jan 13, 2021 at 03:51:42PM +0100, Andrey Konovalov wrote: > On Fri, Nov 20, 2020 at 9:34 PM Peter Collingbourne <pcc@xxxxxxxxxx> wrote: > > The kernel currently clears the tag bits (i.e. bits 56-63) in the fault > > address exposed via siginfo.si_addr and sigcontext.fault_address. However, > > the tag bits may be needed by tools in order to accurately diagnose > > memory errors, such as HWASan [1] or future tools based on the Memory > > Tagging Extension (MTE). > > > > Expose these bits via the arch_untagged_si_addr mechanism, so that > > they are only exposed to signal handlers with the SA_EXPOSE_TAGBITS > > flag set. > > > > [1] http://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html > > > > Signed-off-by: Peter Collingbourne <pcc@xxxxxxxxxx> > > Reviewed-by: Catalin Marinas <catalin.marinas@xxxxxxx> > > Link: https://linux-review.googlesource.com/id/Ia8876bad8c798e0a32df7c2ce1256c4771c81446 [...] > Not sure if it's expected, but this patch changes the way in-kernel > MTE faults are reported. The address of any in-kernel fault now gets > its top byte zeroed out, even though it was set to 0xf. However I > guess it's related to the fact that MTE always uses pointer tags in > 0x0_ format. Is this for _any_ faults or just MTE tag check faults? I tried the former and seems fine. Do you have a concrete example to make sure I understand the issue? -- Catalin
