Re: [PATCH 11/12] mm: pgtable: introduce generic __tlb_remove_table()

From: Peter Zijlstra
Date: Mon Dec 16 2024 - 07:05:53 EST


On Mon, Dec 16, 2024 at 01:03:13PM +0100, Peter Zijlstra wrote:
> On Mon, Dec 16, 2024 at 01:00:43PM +0100, Peter Zijlstra wrote:
> > On Sat, Dec 14, 2024 at 05:02:57PM +0800, Qi Zheng wrote:
> >
> > > diff --git a/arch/s390/mm/pgalloc.c b/arch/s390/mm/pgalloc.c
> > > index c73b89811a264..3e002dea6278f 100644
> > > --- a/arch/s390/mm/pgalloc.c
> > > +++ b/arch/s390/mm/pgalloc.c
> > > @@ -193,13 +193,6 @@ void page_table_free(struct mm_struct *mm, unsigned long *table)
> > > pagetable_dtor_free(ptdesc);
> > > }
> > >
> > > -void __tlb_remove_table(void *table)
> > > -{
> > > - struct ptdesc *ptdesc = virt_to_ptdesc(table);
> > > -
> > > - pagetable_dtor_free(ptdesc);
> > > -}
> > > -
> > > #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > > static void pte_free_now(struct rcu_head *head)
> > > {
> >
> > > diff --git a/include/asm-generic/tlb.h b/include/asm-generic/tlb.h
> > > index 709830274b756..939a813023d7e 100644
> > > --- a/include/asm-generic/tlb.h
> > > +++ b/include/asm-generic/tlb.h
> >
> > > #define MAX_TABLE_BATCH \
> > > ((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
> > >
> > > +#ifndef __HAVE_ARCH_TLB_REMOVE_TABLE
> > > +static inline void __tlb_remove_table(void *_table)
> > > +{
> > > + struct ptdesc *ptdesc = (struct ptdesc *)_table;
> > > +
> > > + pagetable_dtor(ptdesc);
> > > + pagetable_free(ptdesc);
> > > +}
> > > +#endif
> >
> >
> > Spot the fail...
> >
> > That said, all this ptdesc stuff is another giant trainwreck. Let me
> > clean that up for you.
> >
> > ---
>
> > -static inline void __tlb_remove_table(void *_table)
> > +static inline void __tlb_remove_table(void *table)
> > {
> > - struct ptdesc *ptdesc = (struct ptdesc *)_table;
> > + struct ptdesc *ptdesc = page_to_ptdesc(table);
> >
> > pagetable_dtor(ptdesc);
> > pagetable_free(ptdesc);
>
> And now observe that __tlb_remove_table() is identical to
> asm-generic/pgalloc.h pte_free(), pmd_free(), __pud_free() and
> __p4d_free().
>
> And I'm sure we don't need 5 copies of this :-), wdyt?

Argh, nearly, it has the whole page vs virt nonsense still going on.
Bah.