Re: Debug: sleeping function called from invalid context

[Matt Mackall <mpm@selenic.com>]

On Mon, Aug 18, 2003 at 05:15:45PM -0700, Andrew Morton wrote:
> Dave Jones <davej@redhat.com> wrote:
> >
> > How spooky. now I got one too, (minus the noise).
> > 
> > Call Trace:
> >  [<c0120022>] __might_sleep+0x5b/0x5f
> 
> It would be useful to know whether this was triggered by in_atomic() or by
> irqs_disabled().  We're suspecting the latter.

Everything points to it being a fault handler.

Here's my current understanding:

 some part of X calls sys_vm86()
 sys_vm86 stashes pointer to userspace structure
 do_sys_vm86 fiddles with register structures to setup 16-bit transition
 do_sys_vm86 goes to 16-bit mode _through the userspace return path_
 fault occurs in 16-bit code
 handle_vm86_fault invoked through interrupt
 save_v86_state writes into stashed userspace structure (might_sleep)
 return_to_32bit swaps register sets around
 return_to_32bit returns to the original userspace context

Because we never return to the context of the sys_vm86 syscall, we're
never again in an appropriate place to copy the registers over. A
cleaner way to do this is to setup return_to_32bit to return to the
point just after where sys_vm86 returns to 16bit mode and copy the
registers to userspace in normal process context.

-- 
Matt Mackall : http://www.selenic.com : of or relating to the moon