From: Al V. <viro@ZenIV.linux.org.uk> - 2011-08-23 17:33:43
|
On Tue, Aug 23, 2011 at 09:20:12AM -0700, Linus Torvalds wrote: > It's EMULATING A SYSTEM CALL. That original "getregs" value is not > some "user space state". It's the *system call* state that you got > after the system call trapped. Setting it back is an insane operation, > but it would happen to work - if you make no changes. > > But UML *does* make changes. It takes that system call state, and then > EMULATES THE SYSTEM CALL INCORRECTLY. > > If you see it that way (which is the correct way), then it's clearly > an UML problem, and it's not at all "madness" that your > getregs/setregs pairing doesn't work. > > See? Buggy system call emulation. It's really that simple. Of course, > "simple" in this case is "really really subtle differences in how the > kernel treats syscall/sysenter/int80", so the *details* are certainly > not simple, but the concept is. It's a bit more than that (ptrace changes to syscall arguments *are* lost on syscall restart), but... as far as I'm concerned, the situation is simple now: * SYSCALL is not terminally broken wrt restarts. My apologies for misreading what was going on. * SYSENTER with Linus' patch does work just fine wrt restarts + ptrace * SYSCALL is losing ptrace-made changes to arguments when it restarts. Might or might not be a problem for somebody. * UML should not touch SYSCALL for 32bit. Not without serious changes in UML and I'm not convinced that it won't be worse than what we probably ought to do there: check if __kernel_vsyscall() does SYSCALL (recognizable by interaction with POKEUSER) and don't tell about vdso to guest processes. Anything well-behaving won't step on SYSCALL and the things that do deserve the subtle bugs they get. * asm glue is subtle, evil and doesn't have anywhere near enough documentation ;-/ |