coredump: ensure the fpu state is flushed for proper multi-threaded core dump
authorSuresh Siddha <suresh.b.siddha@intel.com>
Wed, 16 May 2012 22:03:52 +0000 (15:03 -0700)
committerH. Peter Anvin <hpa@linux.intel.com>
Wed, 16 May 2012 22:16:48 +0000 (15:16 -0700)
commit11aeca0b3a083a457f5c34fe8c677d5e86a0c6b3
tree8d71ab524aa13aa8fd2bf10d7bbf3ddfa89ac4f7
parent55ccf3fe3f9a3441731aa79cf42a628fc4ecace9
coredump: ensure the fpu state is flushed for proper multi-threaded core dump

Nalluru reported hitting the BUG_ON(__thread_has_fpu(tsk)) in
arch/x86/kernel/xsave.c:__sanitize_i387_state() during the coredump
of a multi-threaded application.

A look at the exit seqeuence shows that other threads can still be on the
runqueue potentially at the below shown exit_mm() code snippet:

if (atomic_dec_and_test(&core_state->nr_threads))
complete(&core_state->startup);

===> other threads can still be active here, but we notify the thread
===> dumping core to wakeup from the coredump_wait() after the last thread
===> joins this point. Core dumping thread will continue dumping
===> all the threads state to the core file.

for (;;) {
set_task_state(tsk, TASK_UNINTERRUPTIBLE);
if (!self.task) /* see coredump_finish() */
break;
schedule();
}

As some of those threads are on the runqueue and didn't call schedule() yet,
their fpu state is still active in the live registers and the thread
proceeding with the coredump will hit the above mentioned BUG_ON while
trying to dump other threads fpustate to the coredump file.

BUG_ON() in arch/x86/kernel/xsave.c:__sanitize_i387_state() is
in the code paths for processors supporting xsaveopt. With or without
xsaveopt, multi-threaded coredump is broken and maynot contain
the correct fpustate at the time of exit.

In coredump_wait(), wait for all the threads to be come inactive, so
that we are sure all the extended register state is flushed to
the memory, so that it can be reliably copied to the core file.

Reported-by: Suresh Nalluru <suresh@aristanetworks.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1336692811-30576-2-git-send-email-suresh.b.siddha@intel.com
Acked-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
fs/exec.c