Re: [RFC PATCH 2/3] x86/mm: make sure LAM is up-to-date during context switching

[Kirill A. Shutemov]

On Fri, Mar 08, 2024 at 07:23:58AM -0800, Dave Hansen wrote:
> On 3/7/24 17:34, Andy Lutomirski wrote:
> >> Fix this by making sure we write a new CR3 if LAM is not
> >> up-to-date. No problems were observed in practice, this was found
> >> by code inspection.
> > I think it should be fixed with a much bigger hammer: explicit IPIs.
> > Just don't ever let it get out of date, like install_ldt().
> I guess it matters whether the thing that matters is having a persistent
> inconsistency or a temporary one.  IPIs will definitely turn a permanent
> one into a temporary one.
> 
> But this is all easier to reason about if we can get rid of even the
> temporary inconsistency.
> 
> Wouldn't this be even simpler than IPIs?
> 
> static inline unsigned long set_tlbstate_lam_mode(struct mm_struct *mm)
> {
> 	unsigned long lam = READ_ONCE(mm->context.lam_cr3_mask);
> 
> +	/* LAM is for userspace only.  Ignore it for kernel threads: */
> +	if (tsk->flags & PF_KTHREAD)
> +		return 0;

I like this approach. kthread_use_mm() WARNs if it called for
non-PF_KTHREAD task, so it should be okay.

I was worried that it would also exclude io_uring, but worker threads
don't have the flag set.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov