On 28/07/2019 14:41, Anshuman Khandual wrote:
>
>
> On 07/22/2019 09:12 PM, Steven Price wrote:
>> It is useful to be able to skip parts of the page table tree even when
>> walking without VMAs. Add test_p?d callbacks similar to test_walk but
>> which are called just before a table at that level is walked. If the
>> callback returns non-zero then the entire table is skipped.
>>
>> Signed-off-by: Steven Price <steven.price@xxxxxxx>
>> ---
>> include/linux/mm.h | 11 +++++++++++
>> mm/pagewalk.c | 24 ++++++++++++++++++++++++
>> 2 files changed, 35 insertions(+)
>>
>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>> index b22799129128..325a1ca6f820 100644
>> --- a/include/linux/mm.h
>> +++ b/include/linux/mm.h
>> @@ -1447,6 +1447,11 @@ void unmap_vmas(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>> * value means "do page table walk over the current vma,"
>> * and a negative one means "abort current page table walk
>> * right now." 1 means "skip the current vma."
>> + * @test_pmd: similar to test_walk(), but called for every pmd.
>> + * @test_pud: similar to test_walk(), but called for every pud.
>> + * @test_p4d: similar to test_walk(), but called for every p4d.
>> + * Returning 0 means walk this part of the page tables,
>> + * returning 1 means to skip this range.
>> * @mm: mm_struct representing the target process of page table walk
>> * @vma: vma currently walked (NULL if walking outside vmas)
>> * @private: private data for callbacks' usage
>> @@ -1471,6 +1476,12 @@ struct mm_walk {
>> struct mm_walk *walk);
>> int (*test_walk)(unsigned long addr, unsigned long next,
>> struct mm_walk *walk);
>> + int (*test_pmd)(unsigned long addr, unsigned long next,
>> + pmd_t *pmd_start, struct mm_walk *walk);
>> + int (*test_pud)(unsigned long addr, unsigned long next,
>> + pud_t *pud_start, struct mm_walk *walk);
>> + int (*test_p4d)(unsigned long addr, unsigned long next,
>> + p4d_t *p4d_start, struct mm_walk *walk);
>> struct mm_struct *mm;
>> struct vm_area_struct *vma;
>> void *private;
>> diff --git a/mm/pagewalk.c b/mm/pagewalk.c
>> index 1cbef99e9258..6bea79b95be3 100644
>> --- a/mm/pagewalk.c
>> +++ b/mm/pagewalk.c
>> @@ -32,6 +32,14 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
>> unsigned long next;
>> int err = 0;
>>
>> + if (walk->test_pmd) {
>> + err = walk->test_pmd(addr, end, pmd_offset(pud, 0UL), walk);
>> + if (err < 0)
>> + return err;
>> + if (err > 0)
>> + return 0;
>> + }
>
> Though this attempts to match semantics with test_walk() and be comprehensive
> just wondering what are the real world situations when page walking need to be
> aborted based on error condition at a given page table level.
I'm not aware of a situation yet where aborting early is necessary - but
as you say this matches the semantics of test_walk() and was easy to
implement.
Steve
