Re: [PATCH 1/2] powerpc: Refactor __kernel_map_pages()

[Michael Ellerman]

Christophe Leroy <christophe.leroy@xxxxxxxxxx> writes:
> Le 22/02/2024 à 06:32, Michael Ellerman a écrit :
>> Christophe Leroy <christophe.leroy@xxxxxxxxxx> writes:
>>> __kernel_map_pages() is almost identical for PPC32 and RADIX.
>>>
>>> Refactor it.
>>>
>>> On PPC32 it is not needed for KFENCE, but to keep it simple
>>> just make it similar to PPC64.
>>>
>>> Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxxxxxx>
>>> ---
>>>   arch/powerpc/include/asm/book3s/64/pgtable.h | 10 ----------
>>>   arch/powerpc/include/asm/book3s/64/radix.h   |  2 --
>>>   arch/powerpc/mm/book3s64/radix_pgtable.c     | 14 --------------
>>>   arch/powerpc/mm/pageattr.c                   | 19 +++++++++++++++++++
>>>   arch/powerpc/mm/pgtable_32.c                 | 15 ---------------
>>>   5 files changed, 19 insertions(+), 41 deletions(-)
>>>
>>> diff --git a/arch/powerpc/mm/pageattr.c b/arch/powerpc/mm/pageattr.c
>>> index 421db7c4f2a4..16b8d20d6ca8 100644
>>> --- a/arch/powerpc/mm/pageattr.c
>>> +++ b/arch/powerpc/mm/pageattr.c
>>> @@ -101,3 +101,22 @@ int change_memory_attr(unsigned long addr, int numpages, long action)
>>>   	return apply_to_existing_page_range(&init_mm, start, size,
>>>   					    change_page_attr, (void *)action);
>>>   }
>>> +
>>> +#if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KFENCE)
>>> +#ifdef CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC
>>> +void __kernel_map_pages(struct page *page, int numpages, int enable)
>>> +{
>>> +	unsigned long addr = (unsigned long)page_address(page);
>>> +
>>> +	if (PageHighMem(page))
>>> +		return;
>>> +
>>> +	if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !radix_enabled())
>>> +		hash__kernel_map_pages(page, numpages, enable);
>>> +	else if (enable)
>>> +		set_memory_p(addr, numpages);
>>> +	else
>>> +		set_memory_np(addr, numpages);
>>> +}
>> 
>> This doesn't build on 32-bit, eg. ppc32_allmodconfig:
>> 
>> ../arch/powerpc/mm/pageattr.c: In function '__kernel_map_pages':
>> ../arch/powerpc/mm/pageattr.c:116:23: error: implicit declaration of function 'hash__kernel_map_pages' [-Werror=implicit-function-declaration]
>>    116 |                 err = hash__kernel_map_pages(page, numpages, enable);
>>        |                       ^~~~~~~~~~~~~~~~~~~~~~
>> 
>> I couldn't see a nice way to get around it, so ended up with:
>> 
>> void __kernel_map_pages(struct page *page, int numpages, int enable)
>> {
>> 	int err;
>> 	unsigned long addr = (unsigned long)page_address(page);
>> 
>> 	if (PageHighMem(page))
>> 		return;
>> 
>> #ifdef CONFIG_PPC_BOOK3S_64
>> 	if (!radix_enabled())
>> 		err = hash__kernel_map_pages(page, numpages, enable);
>> 	else
>> #endif
>> 	if (enable)
>> 		err = set_memory_p(addr, numpages);
>> 	else
>> 		err = set_memory_np(addr, numpages);
>> 
>
>
> I missed something it seems. Not good to leave something unterminated 
> when you leave for vacation and think it was finished when you come back.
>
> The best solution I see is to move hash__kernel_map_pages() prototype 
> somewhere else.

> $ git grep -e hash__ -e radix__ -- arch/powerpc/include/asm/*.h
> arch/powerpc/include/asm/bug.h:void hash__do_page_fault(struct pt_regs *);
> arch/powerpc/include/asm/mmu.h:extern void radix__mmu_cleanup_all(void);
> arch/powerpc/include/asm/mmu_context.h:extern void radix__switch_mmu_context(struct mm_struct *prev,
> arch/powerpc/include/asm/mmu_context.h:         return radix__switch_mmu_context(prev, next);
> arch/powerpc/include/asm/mmu_context.h:extern int hash__alloc_context_id(void);
> arch/powerpc/include/asm/mmu_context.h:void __init hash__reserve_context_id(int id);
> arch/powerpc/include/asm/mmu_context.h: context_id = hash__alloc_context_id();
> arch/powerpc/include/asm/mmu_context.h:  * radix__flush_all_mm() to determine the scope (local/global)
> arch/powerpc/include/asm/mmu_context.h:         radix__flush_all_mm(mm);

If anything I'd prefer to move those out of there into the book3s/64/
headers :)

> Maybe asm/mmu.h ?
>
> Or mm/mmu_decl.h ?

Yeah I'll do that. It's a bit of a dumping ground, but at least it's
internal to arch code, not exported to the rest of the kernel.

cheers