|
From: Tom H. <to...@co...> - 2009-07-10 13:57:09
|
On 10/07/09 15:17, Julian Seward wrote: >> The none/tests/amd64/bug127521-64 one is an unsupported instruction >> failure on cmpxchg16 by the looks of it: >> >> vex amd64->IR: unhandled instruction bytes: 0xF0 0x49 0xF 0xC7 0xA 0x49 > > Can you send me the /proc/cpuinfo for that machine? To support cmpxchg16b > properly, vex has to emit cmpxchg16b instruction into the instrumented code, > which is different from what it did before (some lame-ass fakery involving > normal 64-bit loads and stores). But some really old Opterons don't support > cmpxchg16b, which means vex now has to refuse to decode it if doesn't believe > the underlying machine can handle it. It's an Athon 64 3500+ and you're quite right - it doesn't have the cx16 feature bit set. > I guess what is comes to is, none/tests/amd64/bug127521-64 shouldn't be > run unless the machine supports cmpxchg16b. It would be interesting to > know if this test runs natively on the machine. No, it dies with SIGILL when run natively. So yes, that test needs to be excluded on machines with no cx16 support. Tom -- Tom Hughes (to...@co...) http://www.compton.nu/ |