Re: [PATCH] mm: Introduce GFP_PGTABLE

[Anshuman Khandual]

On 01/12/2019 10:18 PM, Shakeel Butt wrote:
>> --- a/arch/x86/kernel/espfix_64.c
>> +++ b/arch/x86/kernel/espfix_64.c
>> @@ -57,8 +57,6 @@
>>  # error "Need more virtual address space for the ESPFIX hack"
>>  #endif
>>
>> -#define PGALLOC_GFP (GFP_KERNEL | __GFP_ZERO)
>> -
>>  /* This contains the *bottom* address of the espfix stack */
>>  DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack);
>>  DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr);
>> @@ -172,7 +170,7 @@ void init_espfix_ap(int cpu)
>>         pud_p = &espfix_pud_page[pud_index(addr)];
>>         pud = *pud_p;
>>         if (!pud_present(pud)) {
>> -               struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0);
>> +               struct page *page = alloc_pages_node(node, GFP_PGTABLE, 0);
>>
>>                 pmd_p = (pmd_t *)page_address(page);
>>                 pud = __pud(__pa(pmd_p) | (PGTABLE_PROT & ptemask));
>> @@ -184,7 +182,7 @@ void init_espfix_ap(int cpu)
>>         pmd_p = pmd_offset(&pud, addr);
>>         pmd = *pmd_p;
>>         if (!pmd_present(pmd)) {
>> -               struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0);
>> +               struct page *page = alloc_pages_node(node, GFP_PGTABLE, 0);
>>
>>                 pte_p = (pte_t *)page_address(page);
>>                 pmd = __pmd(__pa(pte_p) | (PGTABLE_PROT & ptemask));
>> diff --git a/arch/x86/mm/pgtable.c b/arch/x86/mm/pgtable.c
>> index 7bd0170..d608b03 100644
>> --- a/arch/x86/mm/pgtable.c
>> +++ b/arch/x86/mm/pgtable.c
>> @@ -13,19 +13,17 @@ phys_addr_t physical_mask __ro_after_init = (1ULL << __PHYSICAL_MASK_SHIFT) - 1;
>>  EXPORT_SYMBOL(physical_mask);
>>  #endif
>>
>> -#define PGALLOC_GFP (GFP_KERNEL_ACCOUNT | __GFP_ZERO)
>> -
> You have silently dropped __GFP_ACCOUNT from all the allocations in this file.

Right, they need to be added back explicitly after GFP_PGTABLE. Matthew had
pointed this earlier. Will fix it next time around.

> 
> BTW why other archs not using __GFP_ACCOUNT for the user page tables?
> 

Some archs do and some dont. User page tables pages should use __GFP_ACCOUNT
for allocation. I am working on fixing it for arm64.