On 12/23/2015 07:57 AM, Joonsoo Kim wrote:
What are the cases where pageblock_pfn_to_page() is used for a subset of
the pageblock and the result would be problematic for compaction? I.e.,
do we actually care to use pageblocks that are not contiguous at all?
The problematic pageblocks are those that have pages from more than one zone in
them, so we just skip them. Supposedly that can only happen by switching once
between two zones somewhere in the middle of the pageblock, so it's sufficient
to check first and last pfn and compare their zones. So using
pageblock_pfn_to_page() on a subset from compaction would be wrong. Holes (==no
pages) within pageblock is a different thing checked by pfn_valid_within()
(#defined out on archs where such holes cannot happen) when scanning the block.
That's why I'm not entirely happy with how the patch conflates both the
first/last pfn's zone checks and pfn_valid_within() checks. Yes, a fully
contiguous zone does *imply* that pageblock_pfn_to_page() doesn't have to check
first/last pfn for a matching zone. But it's not *equality*. And any (now just
*potential*) user of pageblock_pfn_to_page() with pfn's different than
first/last pfn of a pageblock is likely wrong.
Now, I understand your concern. What makes me mislead is that
3 of 4 callers to pageblock_pfn_to_page() in compaction.c could call it with
non-pageblock boundary pfn.
Oh, I thought you were talking about potential new callers, now that the
function was exported. So let's see about the existing callers:
isolate_migratepages() - first pfn can be non-boundary when restarting
from a middle of pageblock, that's true. But it means the pageblock has
already passed the check in previous call where it was boundary, so it's
safe. Worst can happen that the restarting pfn will be in a
intra-pageblock hole so pageblock will be falsely skipped over.
isolate_freepages() - always boundary AFAICS?
isolate_migratepages_range() and isolate_freepages_range() - yeah the
CMA parts say it doesn't have to be aligned, I don't know about actual users
Maybe, they should be fixed first.
It would be probably best, even for isolate_migratepages() for
consistency and less-surprisibility.
Then, yes. I can
separate first/last pfn's zone checks and pfn_valid_within() checks.
If then, would you be entirely happy? :)
Maybe, if the patch also made me a coffee :P
Thanks.
Thanks!
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>