2014-11-11 22:55 GMT+03:00 Jeff Law <l...@redhat.com>: > On 11/11/14 11:42, Mike Stump wrote: >> >> On Nov 11, 2014, at 8:02 AM, Ilya Enkovich <enkovich....@gmail.com> wrote: >>> >>> This patch adds simple runtime MPX tests. >>> >>> Make check shows no fails for mpx.exp suite for >>> x86_64-unknown-linux-gnu{-m32,-mx32,}. OK for trunk? >> >> >> Ok with any changes necessary for: >> >> I’m assuming that this won’t break i386 (targets without any mpx) testing. > > It shouldn't. There's check in mpx.exp that's supposed to prevent the tests > from running. I haven't verified that though. > > Ilya, if you could verify the tests aren't going to cause failures on x86-64 > targets w/o MPX it'd be appreciated. > > jeff >
Hi, tests are compiled ignoring CPUID, mpx.exp only checks tests may be compiled (i.e. binutils support MPX, runtime wasn't disabled etc.). All MPX binaries are supposed to work correctly on hardware without MPX, we just shouldn't expect any bound violations detected. Thus all tests run even on hardware with no MPX. mpx-check.h is included into each (or almost each test). It checks CPUID and just exits with expected code in case MPX is not supported. I tried it on hardware with and without MPX and check log is same for both cases. But in more detailed gcc.log we may see that tests are SKIPPED when we have no MPX in hardware. Thanks, Ilya