Re: [PATCH v6 10/18] sh/tlb: Convert SH to generic mmu_gather

From: Peter Zijlstra
Date: Wed Dec 04 2019 - 08:35:22 EST


On Wed, Dec 04, 2019 at 01:32:58PM +0100, Geert Uytterhoeven wrote:

> > Does the below help?
>
> Unfortunately not.
>
> > diff --git a/arch/sh/include/asm/pgalloc.h b/arch/sh/include/asm/pgalloc.h
> > index 22d968bfe9bb..73a2c00de6c5 100644
> > --- a/arch/sh/include/asm/pgalloc.h
> > +++ b/arch/sh/include/asm/pgalloc.h
> > @@ -36,9 +36,8 @@ do { \
> > #if CONFIG_PGTABLE_LEVELS > 2
> > #define __pmd_free_tlb(tlb, pmdp, addr) \
> > do { \
> > - struct page *page = virt_to_page(pmdp); \
> > - pgtable_pmd_page_dtor(page); \
> > - tlb_remove_page((tlb), page); \
> > + pgtable_pmd_page_dtor(pmdp); \
>
> expected âstruct page *â but argument is of type âpmd_t * {aka struct
> <anonymous> *}â
>
> > + tlb_remove_page((tlb), (pmdp)); \
>
> likewise

Duh.. clearly I misplaced my SH cross compiler. Let me go find it.

Also, looking at pgtable.c the pmd_t* actually comes from a kmemcach()
and should probably use pmd_free() (which is what the old code did too).

Also, since SH doesn't have ARCH_ENABLE_SPLIT_PMD_PTLOCK, it will never
need pgtable_pmd_page_dtor().

The below seems to build se7722_defconfig using sh4-linux-. That is, the
build fails, on 'node_reclaim_distance', not pgtable stuff.

Does this fare better?

---
arch/sh/include/asm/pgalloc.h | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/arch/sh/include/asm/pgalloc.h b/arch/sh/include/asm/pgalloc.h
index 22d968bfe9bb..c910e5bcde62 100644
--- a/arch/sh/include/asm/pgalloc.h
+++ b/arch/sh/include/asm/pgalloc.h
@@ -36,9 +36,7 @@ do { \
#if CONFIG_PGTABLE_LEVELS > 2
#define __pmd_free_tlb(tlb, pmdp, addr) \
do { \
- struct page *page = virt_to_page(pmdp); \
- pgtable_pmd_page_dtor(page); \
- tlb_remove_page((tlb), page); \
+ pmd_free((tlb)->mm, (pmdp)); \
} while (0);
#endif