On Fri, Apr 24, 2015 at 5:00 AM, Brian Gerst <brge...@gmail.com> wrote:
>
> So actually this isn't a preemption issue, as the NULL SS is coming
> from an interrupt from userspace (timer tick, etc.).
It *is* a preemption issue, in the sense that the interrupt that
clears SS also then returns to user space using an "iret" that will
properly restore it.
So the only case we need to worry about is the preemption case, where
the interrupt has caused a task switch (typically because it woke
something up or it was the timer interrupt and the timeslice of the
previous task is up), and we switch to another context that returns to
user space using "sysret" instead.
> Another alternative to consider is setting SS=__KERNEL_DS on interrupt
> entry if it's NULL.
The interrupt path is likely more critical than the scheduler path.
Also, it's any exception, afaik, so it's a lot less targeted.
I like Andy's patch. It looks good and efficient. We need to keep this
issue in mind if we ever expand the "Use sysret to return to userspace
when possible" approach to other things than just system call returns,
but with the limitations of the contents of RCX/R11, that's unlikely
to ever be a useful thing anyway.
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
