[PATCH v6] um: Enable preemption in UML

[Johannes Berg]

> 
> 4. UML TLB flush is also invoked during a fork. This happens
> with interrupts and preempt disabled which disagrees with the
> standard mm locking via rwsem. The mm lock for this code path
> had to be replaced with an rcu.

For the record, even if I figured out this gets rid of the complaints,
I'm not entirely happy with this - yeah it's safe now, but it still
feels entirely wrong.

But I guess we can also do this and then remove it entirely like the
patch I just posted. Order probably doesn't matter much.

(Note my patch removes the locking completely since it's now invoked by
the kernel even under write mmap lock.)

> diff --git a/arch/um/kernel/fpu.c b/arch/um/kernel/fpu.c
> new file mode 100644
> index 000000000000..4817276b2a26
> --- /dev/null
> +++ b/arch/um/kernel/fpu.c

> +#ifdef CONFIG_64BIT
> +	if (likely(cpu_has(&boot_cpu_data, X86_FEATURE_XSAVEOPT)))
> +		__builtin_ia32_xsaveopt64(&current->thread.fpu, KNOWN_387_FEATURES);
> +	else {
> +		if (likely(cpu_has(&boot_cpu_data, X86_FEATURE_XSAVE)))
> +			__builtin_ia32_xsave64(&current->thread.fpu, KNOWN_387_FEATURES);
> +		else
> +			__builtin_ia32_fxsave64(&current->thread.fpu);

:)

OK, I'll stop mentioning it ;-)

> +EXPORT_SYMBOL_GPL(kernel_fpu_end);
> +

git am complained here about blank line at EOF

> @@ -597,8 +609,13 @@ void force_flush_all(void)
>  	struct vm_area_struct *vma;
>  	VMA_ITERATOR(vmi, mm, 0);
>  
> -	mmap_read_lock(mm);
> +	/* We use a RCU lock instead of a mm lock, because
> +	 * this can be invoked out of critical/atomic sections
> +	 * and that does not agree with the sleepable semantics
> +	 * of the standard semaphore based mm lock.
> +	 */
> +	rcu_read_lock();

Yeah I guess ... Seems like a very mechanical description of what's
going on, rather than a description of why this is correct (which
assumes no preempt and no SMP)?

I'd have preferred that, but with the patch I just posted we'll just
kill this entirely so it doesn't matter in the end.

johannes