Re: [PATCH v4 10/15] riscv: pgtable: move pagetable_dtor() to __tlb_remove_table()

[Kevin Brodsky <kevin.brodsky@arm.com>]

On 07/01/2025 13:31, Qi Zheng wrote:
> On 2025/1/7 19:58, Kevin Brodsky wrote:
>> On 07/01/2025 11:51, Qi Zheng wrote:
>>> [...]
>>>
>>> Author: Qi Zheng <zhengqi.arch@bytedance.com>
>>> Date:   Fri Dec 13 17:13:48 2024 +0800
>>>
>>>      mm: pgtable: completely move pagetable_dtor() to generic
>>> tlb_remove_table()
>>>
>>>      For the generic tlb_remove_table(), it is implemented in the
>>> following two
>>>      forms:
>>>
>>>      1) CONFIG_MMU_GATHER_TABLE_FREE is enabled
>>>
>>>      tlb_remove_table
>>>      --> generic __tlb_remove_table()
>>>
>>>      2) CONFIG_MMU_GATHER_TABLE_FREE is disabled
>>>
>>>      tlb_remove_table
>>>      --> tlb_remove_page
>>>
>>>      For case 1), the pagetable_dtor() has already been moved to
>>> generic
>>>      __tlb_remove_table().
>>>
>>>      For case 2), now only arm will call
>>> tlb_remove_table()/tlb_remove_ptdesc()
>>>      when CONFIG_MMU_GATHER_TABLE_FREE is disabled. Let's move
>>> pagetable_dtor()
>>>      completely to generic tlb_remove_table(), so that the
>>> architectures can
>>>      follow more easily.
>>>
>>>      Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>
> I missed your Suggested-by, will add it in v5.

Ah yes thanks!

>
>>>
>>> diff --git a/arch/arm/include/asm/tlb.h b/arch/arm/include/asm/tlb.h
>>> index b8eebdb598631..ea4fbe7b17f6f 100644
>>> --- a/arch/arm/include/asm/tlb.h
>>> +++ b/arch/arm/include/asm/tlb.h
>>> @@ -34,10 +34,6 @@ __pte_free_tlb(struct mmu_gather *tlb, pgtable_t
>>> pte, unsigned long addr)
>>>   {
>>>          struct ptdesc *ptdesc = page_ptdesc(pte);
>>>
>>> -#ifndef CONFIG_MMU_GATHER_TABLE_FREE
>>> -       pagetable_dtor(ptdesc);
>>> -#endif
>>
>> I guess this hunk will disappear since this call isn't present to start
>> with.
>
> Yes, I plan to add this in the patch #8, and remove it in this patch.

Right I guess this is required to keep patch 8 self-contained, makes sense.

>
>>
>>> -
>>>   #ifndef CONFIG_ARM_LPAE
>>>          /*
>>>           * With the classic ARM MMU, a pte page has two
>>> corresponding pmd
>>> diff --git a/include/asm-generic/tlb.h b/include/asm-generic/tlb.h
>>> index 69de47c7ef3c5..53ae7748f555b 100644
>>> --- a/include/asm-generic/tlb.h
>>> +++ b/include/asm-generic/tlb.h
>>> @@ -220,14 +220,20 @@ static inline void __tlb_remove_table(void
>>> *table)
>>>
>>>   extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
>>>
>>> -#else /* !CONFIG_MMU_GATHER_HAVE_TABLE_FREE */
>>> +#else /* !CONFIG_MMU_GATHER_TABLE_FREE */
>>
>> Good catch!
>>
>>>
>>> +static inline void tlb_remove_page(struct mmu_gather *tlb, struct
>>> page *page);
>>
>> Nit: might be better to move the declaration up, e.g. above #ifdef
>> CONFIG_MMU_GATHER_TABLE_FREE.
>
> Now only the tlb_remove_table() below calls it, maybe it's better to
> keep the impact to minimum?

I feel it might be better to make the declaration unconditional, but
this is really a detail, I don't mind either way.

>
>>
>>>   /*
>>>    * Without MMU_GATHER_TABLE_FREE the architecture is assumed to have
>>> page based
>>>    * page directories and we can use the normal page batching to free
>>> them.
>>>    */
>>> -#define tlb_remove_table(tlb, page) tlb_remove_page((tlb), (page))
>>> +static inline void tlb_remove_table(struct mmu_gather *tlb, void
>>> *table)
>>> +{
>>> +       struct page *page = (struct page *)table;
>>>
>>> +       pagetable_dtor(page_ptdesc(page));
>>> +       tlb_remove_page(tlb, page);
>>> +}
>>>   #endif /* CONFIG_MMU_GATHER_TABLE_FREE */
>>>
>>>   #ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
>>
>> Looks good to me otherwise.
>
> I will add your Reviewed-by to all patches (except yours) in v5, can
> I also add it to this new added patch? (if we agree with the discussion
> above) ;)

Yes please do, thanks!

- Kevin