Re: [PATCH V3 1/2] mm/mmap: Restrict generic protection_map[] array visibility

[Christophe Leroy <christophe.leroy@csgroup.eu>]

Le 20/06/2022 à 07:16, Anshuman Khandual a écrit :
> 
> 
> On 6/16/22 11:05, Christophe Leroy wrote:
>>
>> Le 16/06/2022 à 06:09, Anshuman Khandual a écrit :
>>> Restrict generic protection_map[] array visibility only for platforms which
>>> do not enable ARCH_HAS_VM_GET_PAGE_PROT. For other platforms that do define
>>> their own vm_get_page_prot() enabling ARCH_HAS_VM_GET_PAGE_PROT, could have
>>> their private static protection_map[] still implementing an array look up.
>>> These private protection_map[] array could do without __PXXX/__SXXX macros,
>>> making them redundant and dropping them off as well.
>>>
>>> But platforms which do not define their custom vm_get_page_prot() enabling
>>> ARCH_HAS_VM_GET_PAGE_PROT, will still have to provide __PXXX/__SXXX macros.
>>>
>>> Cc: Andrew Morton <akpm@linux-foundation.org>
>>> Cc: linux-mm@kvack.org
>>> Cc: linux-kernel@vger.kernel.org
>>> Acked-by: Christoph Hellwig <hch@lst.de>
>>> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
>>> ---
>>>    arch/arm64/include/asm/pgtable-prot.h | 18 ------------------
>>>    arch/arm64/mm/mmap.c                  | 21 +++++++++++++++++++++
>>>    arch/powerpc/include/asm/pgtable.h    |  2 ++
>>>    arch/powerpc/mm/book3s64/pgtable.c    | 20 ++++++++++++++++++++
>>>    arch/sparc/include/asm/pgtable_64.h   | 19 -------------------
>>>    arch/sparc/mm/init_64.c               |  3 +++
>>>    arch/x86/include/asm/pgtable_types.h  | 19 -------------------
>>>    arch/x86/mm/pgprot.c                  | 19 +++++++++++++++++++
>>>    include/linux/mm.h                    |  2 ++
>>>    mm/mmap.c                             |  2 +-
>>>    10 files changed, 68 insertions(+), 57 deletions(-)
>>>
>>> diff --git a/arch/powerpc/include/asm/pgtable.h b/arch/powerpc/include/asm/pgtable.h
>>> index d564d0ecd4cd..8ed2a80c896e 100644
>>> --- a/arch/powerpc/include/asm/pgtable.h
>>> +++ b/arch/powerpc/include/asm/pgtable.h
>>> @@ -21,6 +21,7 @@ struct mm_struct;
>>>    #endif /* !CONFIG_PPC_BOOK3S */
>>>    
>>>    /* Note due to the way vm flags are laid out, the bits are XWR */
>>> +#ifndef CONFIG_ARCH_HAS_VM_GET_PAGE_PROT
>> This ifdef if not necessary for now, it doesn't matter if __P000 etc
>> still exist thought not used.
>>
>>>    #define __P000	PAGE_NONE
>>>    #define __P001	PAGE_READONLY
>>>    #define __P010	PAGE_COPY
>>> @@ -38,6 +39,7 @@ struct mm_struct;
>>>    #define __S101	PAGE_READONLY_X
>>>    #define __S110	PAGE_SHARED_X
>>>    #define __S111	PAGE_SHARED_X
>>> +#endif
>>>    
>>>    #ifndef __ASSEMBLY__
>>>    
>>> diff --git a/arch/powerpc/mm/book3s64/pgtable.c b/arch/powerpc/mm/book3s64/pgtable.c
>>> index 7b9966402b25..d3b019b95c1d 100644
>>> --- a/arch/powerpc/mm/book3s64/pgtable.c
>>> +++ b/arch/powerpc/mm/book3s64/pgtable.c
>>> @@ -551,6 +551,26 @@ unsigned long memremap_compat_align(void)
>>>    EXPORT_SYMBOL_GPL(memremap_compat_align);
>>>    #endif
>>>    
>>> +/* Note due to the way vm flags are laid out, the bits are XWR */
>>> +static const pgprot_t protection_map[16] = {
>>> +	[VM_NONE]					= PAGE_NONE,
>>> +	[VM_READ]					= PAGE_READONLY,
>>> +	[VM_WRITE]					= PAGE_COPY,
>>> +	[VM_WRITE | VM_READ]				= PAGE_COPY,
>>> +	[VM_EXEC]					= PAGE_READONLY_X,
>>> +	[VM_EXEC | VM_READ]				= PAGE_READONLY_X,
>>> +	[VM_EXEC | VM_WRITE]				= PAGE_COPY_X,
>>> +	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_X,
>>> +	[VM_SHARED]					= PAGE_NONE,
>>> +	[VM_SHARED | VM_READ]				= PAGE_READONLY,
>>> +	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
>>> +	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
>>> +	[VM_SHARED | VM_EXEC]				= PAGE_READONLY_X,
>>> +	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READONLY_X,
>>> +	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_X,
>>> +	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_X
>>> +};
>>> +
>> There is not much point is first additing that here and then move it
>> elsewhere in the second patch.
>>
>> I think with my suggestion to use #ifdef __P000 as a guard, the powerpc
>> changes could go in a single patch.
>>
>>>    pgprot_t vm_get_page_prot(unsigned long vm_flags)
>>>    {
>>>    	unsigned long prot = pgprot_val(protection_map[vm_flags &
>>> diff --git a/mm/mmap.c b/mm/mmap.c
>>> index 61e6135c54ef..e66920414945 100644
>>> --- a/mm/mmap.c
>>> +++ b/mm/mmap.c
>>> @@ -101,6 +101,7 @@ static void unmap_region(struct mm_struct *mm,
>>>     *								w: (no) no
>>>     *								x: (yes) yes
>>>     */
>>> +#ifndef CONFIG_ARCH_HAS_VM_GET_PAGE_PROT
>> You should use #ifdef __P000 instead, that way you could migrate
>> architectures one by one.
> 
> If vm_get_page_prot() gets moved into all platforms, wondering what would be
> the preferred method to organize this patch series ?
> 
> 1. Move protection_map[] inside platforms with ARCH_HAS_VM_PAGE_PROT (current patch 1)
> 2. Convert remaining platforms to use ARCH_HAS_VM_PAGE_PROT one after the other
> 3. Drop ARCH_HAS_VM_PAGE_PROT completely
> 
> Using "#ifdef __P000" for wrapping protection_map[] will leave two different #ifdefs
> in flight i.e __P000, ARCH_HAS_VM_PAGE_PROT in the generic mmap code, until both gets
> dropped eventually. But using "#ifdef __P000" does enable splitting the first patch
> into multiple changes for each individual platforms.

 From previous discussions and based on Christoph's suggestion, I guess 
we now aim at getting vm_get_page_prot() moved into all platforms 
together with protection_map[]. Therefore the use of #ifdef __P000 could 
be very temporary at the begining of the series:
1. Guard generic protection_map[] with #ifdef ___P000
2. Move protection_map[] into architecture and drop __Pxxx/__Sxxx  for arm64
3. Same for sparc
4. Same for x86
5. Convert entire powerpc to ARCH_HAS_VM_PAGE_PROT and move 
protection_map[] into architecture and drop __Pxxx/__Sxxx
6. Replace #ifdef __P000 by #ifdef CONFIG_ARCH_HAS_VM_PAGE_PROT
7. Convert all remaining platforms to CONFIG_ARCH_HAS_VM_PAGE_PROT one 
by one (but keep a protection_map[] table, don't use switch/case)
8. Drop ARCH_HAS_VM_PAGE_PROT completely.

Eventually you can squash step 6 into step 8.