Re: [PATCH v9 3/4] arm64: Implement page table free interfaces

From: Will Deacon
Date: Wed May 23 2018 - 09:08:08 EST


Hi Chintan,

[as a side note: I'm confused on the status of this patch series, as part
of it was reposted separately by Toshi. Please can you work together?]

On Mon, Apr 30, 2018 at 01:11:33PM +0530, Chintan Pandya wrote:
> Implement pud_free_pmd_page() and pmd_free_pte_page().
>
> Implementation requires,
> 1) Clearing off the current pud/pmd entry
> 2) Invalidate TLB which could have previously
> valid but not stale entry
> 3) Freeing of the un-used next level page tables
>
> Signed-off-by: Chintan Pandya <cpandya@xxxxxxxxxxxxxx>
> ---
> arch/arm64/mm/mmu.c | 29 +++++++++++++++++++++++++----
> 1 file changed, 25 insertions(+), 4 deletions(-)
>
> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index da98828..0f651db 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
> @@ -45,6 +45,7 @@
> #include <asm/memblock.h>
> #include <asm/mmu_context.h>
> #include <asm/ptdump.h>
> +#include <asm/tlbflush.h>
>
> #define NO_BLOCK_MAPPINGS BIT(0)
> #define NO_CONT_MAPPINGS BIT(1)
> @@ -973,12 +974,32 @@ int pmd_clear_huge(pmd_t *pmdp)
> return 1;
> }
>
> -int pud_free_pmd_page(pud_t *pud, unsigned long addr)
> +int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
> {
> - return pud_none(*pud);
> + pmd_t *table;
> +
> + if (pmd_present(READ_ONCE(*pmdp))) {

Might also be worth checking pmd_table here, just in case. (same for pud)

> + table = __va(pmd_val(*pmdp));

Can you avoid dereferencing *pmdp twice, and instead READ_ONCE into a local
variable, please? (same for pud)

> + pmd_clear(pmdp);
> + __flush_tlb_kernel_pgtable(addr);
> + free_page((unsigned long) table);

Shouldn't this be pte_free_kernel, to pair with pte_alloc_kernel which
was used to allocate the page in the first place? (similarly for pud)

> + }
> + return 1;
> }
>
> -int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
> +int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
> {
> - return pmd_none(*pmd);
> + pmd_t *table;
> + int i;
> +
> + if (pud_present(READ_ONCE(*pudp))) {
> + table = __va(pud_val(*pudp));
> + for (i = 0; i < PTRS_PER_PMD; i++)
> + pmd_free_pte_page(&table[i], addr + (i * PMD_SIZE));

I think it would be cleaner to write this as a do { ... } while, for
consistency with the ioremap and vmalloc code.

Will