on 9/1/2023 5:17 PM, Mel Gorman wrote:
> On Fri, Sep 01, 2023 at 11:51:38PM +0800, Kemeng Shi wrote:
>> In strict mode, we should return 0 if there is any hole in pageblock. If
>> we successfully isolated pages at beginning at pageblock and then have a
>> bogus compound_order outside pageblock in next page. We will abort search
>> loop with blockpfn > end_pfn. Although we will limit blockpfn to end_pfn,
>> we will treat it as a successful isolation in strict mode as blockpfn is
>> not < end_pfn and return partial isolated pages. Then
>> isolate_freepages_range may success unexpectly with hole in isolated
>> range.
>>
>> Fixes: 9fcd6d2e052e ("mm, compaction: skip compound pages by order in free scanner")
>> Signed-off-by: Kemeng Shi <[email protected]>
>> Reviewed-by: Baolin Wang <[email protected]>
>> ---
>> mm/compaction.c | 6 +++---
>> 1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index a40550a33aee..9ecbfbc695e5 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -626,11 +626,12 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
>> if (PageCompound(page)) {
>> const unsigned int order = compound_order(page);
>>
>> - if (likely(order <= MAX_ORDER)) {
>> + if (blockpfn + (1UL << order) <= end_pfn) {
>> blockpfn += (1UL << order) - 1;
>> page += (1UL << order) - 1;
>> nr_scanned += (1UL << order) - 1;
>> }
>> +
>> goto isolate_fail;
>> }
>>
>> @@ -678,8 +679,7 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
>> spin_unlock_irqrestore(&cc->zone->lock, flags);
>>
>> /*
>> - * There is a tiny chance that we have read bogus compound_order(),
>> - * so be careful to not go outside of the pageblock.
>> + * Be careful to not go outside of the pageblock.
>> */
>> if (unlikely(blockpfn > end_pfn))
>> blockpfn = end_pfn;
>
> Is this check still necessary after the first hunk?
>
Actually, I removed this check in the first version, but Baolin thought remove this check is not
cheap and not worth it. More discussion can be found in [1]. Thanks!
[1] https://lore.kernel.org/all/[email protected]/
On Fri, Sep 01, 2023 at 05:32:49PM +0800, Kemeng Shi wrote:
>
>
> on 9/1/2023 5:17 PM, Mel Gorman wrote:
> > On Fri, Sep 01, 2023 at 11:51:38PM +0800, Kemeng Shi wrote:
> >> In strict mode, we should return 0 if there is any hole in pageblock. If
> >> we successfully isolated pages at beginning at pageblock and then have a
> >> bogus compound_order outside pageblock in next page. We will abort search
> >> loop with blockpfn > end_pfn. Although we will limit blockpfn to end_pfn,
> >> we will treat it as a successful isolation in strict mode as blockpfn is
> >> not < end_pfn and return partial isolated pages. Then
> >> isolate_freepages_range may success unexpectly with hole in isolated
> >> range.
> >>
> >> Fixes: 9fcd6d2e052e ("mm, compaction: skip compound pages by order in free scanner")
> >> Signed-off-by: Kemeng Shi <[email protected]>
> >> Reviewed-by: Baolin Wang <[email protected]>
> >> ---
> >> mm/compaction.c | 6 +++---
> >> 1 file changed, 3 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/mm/compaction.c b/mm/compaction.c
> >> index a40550a33aee..9ecbfbc695e5 100644
> >> --- a/mm/compaction.c
> >> +++ b/mm/compaction.c
> >> @@ -626,11 +626,12 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
> >> if (PageCompound(page)) {
> >> const unsigned int order = compound_order(page);
> >>
> >> - if (likely(order <= MAX_ORDER)) {
> >> + if (blockpfn + (1UL << order) <= end_pfn) {
> >> blockpfn += (1UL << order) - 1;
> >> page += (1UL << order) - 1;
> >> nr_scanned += (1UL << order) - 1;
> >> }
> >> +
> >> goto isolate_fail;
> >> }
> >>
> >> @@ -678,8 +679,7 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
> >> spin_unlock_irqrestore(&cc->zone->lock, flags);
> >>
> >> /*
> >> - * There is a tiny chance that we have read bogus compound_order(),
> >> - * so be careful to not go outside of the pageblock.
> >> + * Be careful to not go outside of the pageblock.
> >> */
> >> if (unlikely(blockpfn > end_pfn))
> >> blockpfn = end_pfn;
> >
> > Is this check still necessary after the first hunk?
> >
> Actually, I removed this check in the first version, but Baolin thought remove this check is not
> cheap and not worth it. More discussion can be found in [1]. Thanks!
>
Ok, fair enough. While I think the check is redundant right now, it's a
reasonable defensive check and this is not a fast path so
Acked-by: Mel Gorman <[email protected]>
--
Mel Gorman
SUSE Labs