Re: [PATCH v3 1/2] mm: migration: fix migration of huge PMD shared pages

[Mike Kravetz]

On 08/23/2018 05:48 AM, Michal Hocko wrote:
> On Tue 21-08-18 18:10:42, Mike Kravetz wrote:
> [...]
> 
> OK, after burning myself when trying to be clever here it seems like
> your proposed solution is indeed simpler.
> 
>> +bool huge_pmd_sharing_possible(struct vm_area_struct *vma,
>> +				unsigned long *start, unsigned long *end)
>> +{
>> +	unsigned long check_addr = *start;
>> +	bool ret = false;
>> +
>> +	if (!(vma->vm_flags & VM_MAYSHARE))
>> +		return ret;
>> +
>> +	for (check_addr = *start; check_addr < *end; check_addr += PUD_SIZE) {
>> +		unsigned long a_start = check_addr & PUD_MASK;
>> +		unsigned long a_end = a_start + PUD_SIZE;
> 
> I guess this should be rather in HPAGE_SIZE * PTRS_PER_PTE units as
> huge_pmd_unshare does.

Sure, I can do that.

However, I consider the statement making that calculation in huge_pmd_unshare
to be VERY ugly and confusing code.

	*addr = ALIGN(*addr, HPAGE_SIZE * PTRS_PER_PTE) - HPAGE_SIZE;

Note that it is adjusting the value of passed argument 'unsigned long *addr'.
This argument/value is part of a loop condition in all current callers of
huge_pmd_unshare.  For instance:

	for (; address < end; address += huge_page_size(h)) {

So, that calculation in huge_pmd_unshare gets the calling loop back to
the starting address of the unmapped range.  It even takes the loop increment
'huge_page_size(h)' into account.  That is why that ' - HPAGE_SIZE' is at
the end of the calculation.

ugly and confusing!  And on my list of things to clean up.
-- 
Mike Kravetz