On Sun, May 03, 2026 at 04:57:05PM +0800, qiwu.chen wrote:
> Currently, reclaim_flags always contains RECLAIM_WB_ASYNC in lru_shrink
> tracepoints since commit 41ac1999c3e35 ("mm: vmscan: do not stall on
> writeback during memory compaction"), which is useless for debugging
> memory pressure issues. Other RECLAIM_WB_* flags are not used anywhere
> else, so they can be directly removed.
> This patch reworks the lru_shrink and write_folio tracepoints for better
> correlation and analysis:
> - traces each folio lru type instead of reclaim_flags.
> - traces each lru_shrink with reason.
You also removed the printing of the folio's PFN. Was this deliberate?
If so, it merits a mention in the commit description.
Also if you are going to do this (and I suspect we should do this!)
we don't need to do the folio -> pfn -> page conversion. We can just
store the folio pointer and print out the folio pointer.
