Re: [PATCH v2] mm/compaction: remove unnecessary detection code.

[Vlastimil Babka <vbabka@suse.cz>]

On 11/14/24 08:44, Vlastimil Babka wrote:
> On 11/14/24 07:57, Qiang Liu wrote:
>> It is impossible for the situation where blockpfn > end_pfn to arise,
>> The if statement here is not only unnecessary, but may also lead to
>> a misunderstanding that blockpfn > end_pfn could potentially happen.
>> so these unnecessary checking code should be removed.
>> 
>> Signed-off-by: Qiang Liu <liuq131@chinatelecom.cn>
> 
> I see that's since 3da0272a4c7d ("mm/compaction: correctly return failure
> with bogus compound_order in strict mode")

Hm but we still have:

for (; blockpfn < end_pfn; blockpfn += stride, page += stride) {

and this advance by stride can mix up with advance by isolated, initial pfn
might not be aligned... I don't see any guarantee that the for loop will
exit with exactly blockpfn == end_pfn, it may easily advance beyond end_pfn
so we shouldn't remove the check?

> I think that commit introduced a risk of overflow due to a bogus order
> (which we read in a racy way), and once blockpfn overflows it will satisfy
> <= end_pfn and might e.g. end up scanning a completely different zone?
> 
>                         if (blockpfn + (1UL << order) <= end_pfn) {
> 
>                                 blockpfn += (1UL << order) - 1;
>                                 page += (1UL << order) - 1;
>                                 nr_scanned += (1UL << order) - 1;
>                         }
> 
> We should better add back the MAX_ORDER sanity check?
> 
>> ---
>>  mm/compaction.c | 6 ------
>>  1 file changed, 6 deletions(-)
>> 
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index a2b16b08cbbf..baeda7132252 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -682,12 +682,6 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
>>  	if (locked)
>>  		spin_unlock_irqrestore(&cc->zone->lock, flags);
>>  
>> -	/*
>> -	 * Be careful to not go outside of the pageblock.
>> -	 */
>> -	if (unlikely(blockpfn > end_pfn))
>> -		blockpfn = end_pfn;
>> -
>>  	trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn,
>>  					nr_scanned, total_isolated);
>>  
> 
>