Re: [PATCh 0/3] x86,tlb: context switch optimizations

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Nov 14, 2024 at 12:36:17PM +0100, Peter Zijlstra wrote:
> On Wed, Nov 13, 2024 at 09:55:57AM -0500, Rik van Riel wrote:
> >  arch/x86/kernel/alternative.c | 11 +++--------
> >  1 file changed, 3 insertions(+), 8 deletions(-)
> > 
> > diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
> > index d17518ca19b8..f3caf5bc4df9 100644
> > --- a/arch/x86/kernel/alternative.c
> > +++ b/arch/x86/kernel/alternative.c
> > @@ -1830,6 +1830,9 @@ static inline void unuse_temporary_mm(temp_mm_state_t prev_state)
> >  	lockdep_assert_irqs_disabled();
> >  	switch_mm_irqs_off(NULL, prev_state.mm, current);
> >  
> > +	/* Force a TLB flush next time poking_mm is used. */
> > +	inc_mm_tlb_gen(poking_mm);
> > +
> >  	/*
> >  	 * Restore the breakpoints if they were disabled before the temporary mm
> >  	 * was loaded.
> > @@ -1940,14 +1943,6 @@ static void *__text_poke(text_poke_f func, void *addr, const void *src, size_t l
> >  	 */
> >  	unuse_temporary_mm(prev);
> >  
> > -	/*
> > -	 * Flushing the TLB might involve IPIs, which would require enabled
> > -	 * IRQs, but not if the mm is not used, as it is in this point.
> > -	 */
> > -	flush_tlb_mm_range(poking_mm, poking_addr, poking_addr +
> > -			   (cross_page_boundary ? 2 : 1) * PAGE_SIZE,
> > -			   PAGE_SHIFT, false);
> > -
> >  	if (func == text_poke_memcpy) {
> >  		/*
> >  		 * If the text does not match what we just wrote then something is
> 
> So I really don't like this.
> 
> Yes it avoids the immediate problem, but we're now sending IPIs where we
> shoulnd't be.
> 
> Fundamentally this whole text_poke thing is set up such that only a
> single CPU will have this magical mapping with the aliasses. Having it
> send IPIs is crazy.

I'm thinking the better fix is to make unuse_temporary_mm() explicitly
clear the bit if we don't want switch_mm() to do it.