Re: [RFC V2 01/10] mm/mmap: Dynamically initialize protection_map[]

[Anshuman Khandual]

On 8/5/21 10:33 PM, Catalin Marinas wrote:
> On Mon, Jul 26, 2021 at 12:07:16PM +0530, Anshuman Khandual wrote:
>> The protection_map[] elements (__PXXX and __SXXX) might sometimes contain
>> runtime variables in certain platforms like arm64 preventing a successful
>> build because of the current static initialization. So it just defers the
>> initialization until mmmap_init() via a new helper init_protection_map().
>>
>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
>> ---
>>  mm/mmap.c | 26 ++++++++++++++++++++++----
>>  1 file changed, 22 insertions(+), 4 deletions(-)
>>
>> diff --git a/mm/mmap.c b/mm/mmap.c
>> index ca54d36..a95b078 100644
>> --- a/mm/mmap.c
>> +++ b/mm/mmap.c
>> @@ -100,10 +100,7 @@ static void unmap_region(struct mm_struct *mm,
>>   *								w: (no) no
>>   *								x: (yes) yes
>>   */
>> -pgprot_t protection_map[16] __ro_after_init = {
>> -	__P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
>> -	__S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
>> -};
>> +pgprot_t protection_map[16] __ro_after_init;
> 
> Mips, x86, sparc, arm32, m68k all adjust protection_map[] during boot.
> Could we do something similar here and avoid changing the generic code?

If __P[000..111] and __S[000..111] be made dummy values (e.g 0 or something
standard prot temporarily), hence the compilation problem could be avoided.
Later in the platform code, protection_map[] could be adjusted with actual
prot values which would involve variable.