Re: [RFC PATCH v2 05/19] x86, mm: Use cache of page tables

["Edgecombe, Rick P" <rick.p.edgecombe@intel.com>]

On Tue, 2021-08-31 at 11:40 +0300, Mike Rapoport wrote:
> On Mon, Aug 30, 2021 at 04:59:13PM -0700, Rick Edgecombe wrote:
<trim>
> > 
> > diff --git a/arch/x86/include/asm/pgalloc.h
> > b/arch/x86/include/asm/pgalloc.h
> > index c7ec5bb88334..1ff308ea76cd 100644
> > --- a/arch/x86/include/asm/pgalloc.h
> > +++ b/arch/x86/include/asm/pgalloc.h
> > @@ -7,6 +7,10 @@
> >  #include <linux/pagemap.h>
> >  
> >  #define __HAVE_ARCH_PTE_ALLOC_ONE
> > +#ifdef CONFIG_PKS_PG_TABLES
> > +#define __HAVE_ARCH_FREE_TABLE
> > +#define __HAVE_ARCH_ALLOC_TABLE
> 
> I think one define would suffice. If we'd ever have an architecture
> that
> can implement only one of those, we update the ifdefery in
> asm-generic/pgalloc.h
> 
Good point, I'll change it.

> > +#endif
> >  #define __HAVE_ARCH_PGD_FREE
> >  #include <asm-generic/pgalloc.h>
> >  
> > @@ -162,7 +166,7 @@ static inline void p4d_free(struct mm_struct
> > *mm, p4d_t *p4d)
> >  		return;
> >  
> >  	BUG_ON((unsigned long)p4d & (PAGE_SIZE-1));
> > -	free_page((unsigned long)p4d);
> > +	free_table(virt_to_page(p4d));
> >  }
> >  
> >  extern void ___p4d_free_tlb(struct mmu_gather *tlb, p4d_t *p4d);
> 
> ...
> 
> > diff --git a/include/asm-generic/pgalloc.h b/include/asm-
> > generic/pgalloc.h
> > index 02932efad3ab..e576c19abc8c 100644
> > --- a/include/asm-generic/pgalloc.h
> > +++ b/include/asm-generic/pgalloc.h
> > @@ -2,11 +2,26 @@
> >  #ifndef __ASM_GENERIC_PGALLOC_H
> >  #define __ASM_GENERIC_PGALLOC_H
> >  
> > +#include <linux/mm.h>
> > +
> 
> Why is this required?
> 
> >  #ifdef CONFIG_MMU
> >  
> >  #define GFP_PGTABLE_KERNEL	(GFP_KERNEL | __GFP_ZERO)
> >  #define GFP_PGTABLE_USER	(GFP_PGTABLE_KERNEL | __GFP_ACCOUNT)
> >  
> > +#ifndef __HAVE_ARCH_ALLOC_TABLE
> > +static inline struct page *alloc_table(gfp_t gfp)
> > +{
> > +	return alloc_page(gfp);
> > +}
> > +#else /* __HAVE_ARCH_ALLOC_TABLE */
> > +extern struct page *alloc_table(gfp_t gfp);
> > +#endif /* __HAVE_ARCH_ALLOC_TABLE */
> > +
> > +#ifdef __HAVE_ARCH_FREE_TABLE
> > +extern void free_table(struct page *);
> > +#endif /* __HAVE_ARCH_FREE_TABLE */
> > +
> >  /**
> >   * __pte_alloc_one_kernel - allocate a page for PTE-level kernel
> > page table
> >   * @mm: the mm_struct of the current context
> 
> ...
> 
> > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > index c13c7af7cad3..ab63d5a201cb 100644
> > --- a/include/linux/mm.h
> > +++ b/include/linux/mm.h
> > @@ -2327,6 +2327,13 @@ static inline bool ptlock_init(struct page
> > *page) { return true; }
> >  static inline void ptlock_free(struct page *page) {}
> >  #endif /* USE_SPLIT_PTE_PTLOCKS */
> >  
> > +#ifndef CONFIG_PKS_PG_TABLES
> > +static inline void free_table(struct page *table_page)
> > +{
> > +	__free_pages(table_page, 0);
> > +}
> > +#endif /* CONFIG_PKS_PG_TABLES */
> > +
> 
> Can't this live in asm-generic/pgalloc.h?
> Then you won't need to include linux/mm.h there.
Some architectures couldn't find it in asm-generic/pgalloc.h. There is
some other page table stuff in linux/mm.h so didn't seem too strange.

> 
> >  static inline void pgtable_init(void)
> >  {
> >  	ptlock_cache_init();
> > @@ -2337,7 +2344,6 @@ static inline bool
> > pgtable_pte_page_ctor(struct page *page)
> >  {
> >  	if (!ptlock_init(page))
> >  		return false;
> > -	__SetPageTable(page);
> 
> This change is only valid when __HAVE_ARCH_ALLOC_TABLE is set.
Argh, thanks. Overall solution should still work I think. I'll rework
it to support both.

> 
> >  	inc_lruvec_page_state(page, NR_PAGETABLE);
> >  	return true;
> >  }
> > @@ -2345,7 +2351,6 @@ static inline bool
> > pgtable_pte_page_ctor(struct page *page)
> >  static inline void pgtable_pte_page_dtor(struct page *page)
> >  {
> >  	ptlock_free(page);
> > -	__ClearPageTable(page);
> >  	dec_lruvec_page_state(page, NR_PAGETABLE);
> >  }
> >  
> 
>