On 03/01/2025 04:48, Qi Zheng wrote:
> Hi Kevin,
>
> On 2025/1/3 00:53, Kevin Brodsky wrote:
>> On 30/12/2024 10:07, Qi Zheng wrote:
>>> static inline void riscv_tlb_remove_ptdesc(struct mmu_gather *tlb,
>>> void *pt)
>>> {
>>> - if (riscv_use_sbi_for_rfence())
>>> + if (riscv_use_sbi_for_rfence()) {
>>> tlb_remove_ptdesc(tlb, pt);
>>> - else
>>> + } else {
>>> + pagetable_dtor(pt);
>>> tlb_remove_page_ptdesc(tlb, pt);
>>
>> I find the imbalance pretty confusing: pagetable_dtor() is called
>> explicitly before using tlb_remove_page() but not tlb_remove_ptdesc().
>> Doesn't that assume that CONFIG_MMU_GATHER_HAVE_TABLE_FREE is selected?
>> Could we not call pagetable_dtor() from __tlb_batch_free_encoded_pages()
>> to ensure that the dtor is always called just before freeing, and remove
>
> In __tlb_batch_free_encoded_pages(), we can indeed detect PageTable()
> and call pagetable_dtor() to dtor the page table pages.
> But __tlb_batch_free_encoded_pages() is also used to free normal pages
> (not page table pages), so I don't want to add overhead there.
Interesting, can a tlb batch refer to pages than are not PTPs then?
>
> But now I think maybe we can do this in tlb_remove_page_ptdesc(), like
> this:
>
> diff --git a/arch/csky/include/asm/pgalloc.h
> b/arch/csky/include/asm/pgalloc.h
> index f1ce5b7b28f22..e45c7e91dcbf9 100644
> --- a/arch/csky/include/asm/pgalloc.h
> +++ b/arch/csky/include/asm/pgalloc.h
> @@ -63,7 +63,6 @@ static inline pgd_t *pgd_alloc(struct mm_struct *mm)
>
> #define __pte_free_tlb(tlb, pte, address) \
> do { \
> - pagetable_dtor(page_ptdesc(pte)); \
> tlb_remove_page_ptdesc(tlb, page_ptdesc(pte)); \
> } while (0)
>
> [...]
>
> diff --git a/include/asm-generic/tlb.h b/include/asm-generic/tlb.h
> index a96d4b440f3da..a59205863f431 100644
> --- a/include/asm-generic/tlb.h
> +++ b/include/asm-generic/tlb.h
> @@ -506,6 +506,7 @@ static inline void tlb_remove_ptdesc(struct
> mmu_gather *tlb, void *pt)
> /* Like tlb_remove_ptdesc, but for page-like page directories. */
> static inline void tlb_remove_page_ptdesc(struct mmu_gather *tlb,
> struct ptdesc *pt)
> {
> + pagetable_dtor(pt);
> tlb_remove_page(tlb, ptdesc_page(pt));
> }
I think this is an interesting idea, it does make arch code easier to
follow. OTOH it would have been more natural to me to call
pagetable_dtor() from tlb_remove_page(). I can however see that this
doesn't work, because tlb_remove_table() is defined as tlb_remove_page()
if CONFIG_MMU_GATHER_HAVE_TABLE_FREE isn't selected. Which brings me
back to my earlier question: in that case, aren't we missing a call to
pagetable_dtor() when using tlb_remove_table() (or tlb_remove_ptdesc())?
- Kevin
