[Lguest] CONFIG_PREEMPT causes corruption of application's FPU stack

[Simon Holm Thøgersen]

tir, 03 06 2008 kl. 12:43 -0700, skrev Suresh Siddha:
> On Tue, Jun 03, 2008 at 03:23:30PM +0200, Simon Holm Thøgersen wrote:
> > > [patch] x86: fix blocking call (math_state_restore()) condition in __switch_to
> > > 
> > > Add tsk_used_math() checks to prevent calling math_state_restore()
> > > which can sleep in the case of !tsk_used_math(). This prevents
> > > making a blocking call in __switch_to().
> > > 
> > > Apparently "fpu_counter > 5" check is not enough, as in some signal handling
> > > and fork/exec scenarios, fpu_counter > 5 and !tsk_used_math() is possible.
> > > 
> > > Signed-off-by: Suresh Siddha <suresh.b.siddha at intel.com>
> > > ---
> > Hi Suresh,
> > 
> > and thanks for looking into this. The patch did not fix the issue, but
> 
> Ok. You are probably running into different issue (please see below).
> Above patch fixes a real issue and I think it should fix the fpu
> corruption issue encountered by Jürgen. I will wait for Jürgen's test
> results before pushing the above patch.
> 
> > I'm wondering if it is lguest calling math_state_restore in
> > drivers/lguest/x86/core.c that could be the problem?
> 
> I def see a problem. In lguest_arch_run_guest(), MSR_IA32_SYSENTER_CS is not
> restored before making the math_state_restore() call. As the
> math_state_restore() can now block, this can cause issues. Appending
> patch should fix this issue and from your oops report, it is not very
> clear if the below patch should help fix your issue or not. Can you
> please try the below appended patch.
> 
> > 
> > Regardless of whether that is the issue, I think you (and everybody
> > else) will be able to reproduce the issue by running lguest on a 32-bit
> > system with CONFIG_PREEMPT=y and CONFIG_DEBUG_SPINLOCKS_SLEEP=y (I'm
> > also using CONFIG_DEBUG_PREEMPT=y but I don't think that matter). If you
> > download http://xm-test.xensource.com/ramdisks/initrd-1.1-i386.img and
> > run
> > 
> > Documentation/lguest/lguest 64 vmlinux --block=initrd-1.1-i386.img
> > 
> > it will very likely trigger the backtraces I'm getting.
> 
> If the below patch doesn't help fix your issue, then I will try to reproduce
> it locally here.
> 
It didn't, I'm afraid. I had both patches applied, and was able to
reproduce the trace fairly easily. The patches might have made the issue
slightly more difficult to provoke, but I'm not sure.


Simon