| Message ID | 20230130125504.2509710-1-fengwei.yin@intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | folio based filemap_map_pages() | expand |
On 1/30/2023 8:54 PM, Yin Fengwei wrote: > Current filemap_map_pages() still use page granularity > even when the underneath folio is large folio. Making it > use folio based granularity allows batched refcount, > rmap and mm counter opertion. Which brings some level > performance gain. This series tries to update > filemap_map_pages() > do_set_pte() > page_add_file_rmap() > to bring batched refcount, rmap and mm counter in. > > Patch 1 enabled the fault around for shared file page > write fault. As David suggested here: > https://lore.kernel.org/linux-mm/e14b4e9a-612d-fc02-edc0-8f3b6bcf4148@redhat.com/ > > Patch 2 update filemap_map_pages() to do map based on > folio granularity and batched refcount update > > Patch 3,4,5 Enable batched rmap and mm counter > > Function testing: > - two days daily usage: boot to xfs GUI, kernel build, editor and browser > > Performance testing: > - kernel build: no obvious performance impact > - micro benchmark: will-it-scale:page_fault3 (shared file write fault) > Patch 1 brings 3.75X performance gain > > - micro benchmark: will-it-scale:page_fault3 like but with file read fault > - Patch 2 brings about 2% performance gain (batched refcount) > - Patch 3,4,5 brings about 15% performance gain (batched rmap and mm counter) > Forgot to mention: the base of the series is next-20230119. > > Yin Fengwei (5): > mm: Enable fault around for shared file page fault > filemap: do file page mapping with folio granularity > rmap: add page_add_file_rmap_range() > mm: add do_set_pte_entry() > filemap: batched update mm counter,rmap when map file folio > > include/linux/mm.h | 2 + > include/linux/rmap.h | 2 + > mm/filemap.c | 106 +++++++++++++++++++++++++++++-------------- > mm/memory.c | 25 +++++++++- > mm/rmap.c | 70 +++++++++++++++++----------- > 5 files changed, 143 insertions(+), 62 deletions(-) >
Current filemap_map_pages() still use page granularity even when the underneath folio is large folio. Making it use folio based granularity allows batched refcount, rmap and mm counter opertion. Which brings some level performance gain. This series tries to update filemap_map_pages() do_set_pte() page_add_file_rmap() to bring batched refcount, rmap and mm counter in. Patch 1 enabled the fault around for shared file page write fault. As David suggested here: https://lore.kernel.org/linux-mm/e14b4e9a-612d-fc02-edc0-8f3b6bcf4148@redhat.com/ Patch 2 update filemap_map_pages() to do map based on folio granularity and batched refcount update Patch 3,4,5 Enable batched rmap and mm counter Function testing: - two days daily usage: boot to xfs GUI, kernel build, editor and browser Performance testing: - kernel build: no obvious performance impact - micro benchmark: will-it-scale:page_fault3 (shared file write fault) Patch 1 brings 3.75X performance gain - micro benchmark: will-it-scale:page_fault3 like but with file read fault - Patch 2 brings about 2% performance gain (batched refcount) - Patch 3,4,5 brings about 15% performance gain (batched rmap and mm counter) Yin Fengwei (5): mm: Enable fault around for shared file page fault filemap: do file page mapping with folio granularity rmap: add page_add_file_rmap_range() mm: add do_set_pte_entry() filemap: batched update mm counter,rmap when map file folio include/linux/mm.h | 2 + include/linux/rmap.h | 2 + mm/filemap.c | 106 +++++++++++++++++++++++++++++-------------- mm/memory.c | 25 +++++++++- mm/rmap.c | 70 +++++++++++++++++----------- 5 files changed, 143 insertions(+), 62 deletions(-)