[RFC PATCH] mm: Add large folio support for async readahead

[Yafang Shao <laoar.shao@xxxxxxxxx>]

When testing large folio support with XFS on our servers, we observed that
only a few large folios are mapped when reading large files via mmap. This
behavior occurs because large folio support is currently implemented only
for sync readahead, not for async readahead. Consequently, while the
first filemap fault may map to a large folio, subsequent filemap faults are
mapped to regular folios. This can be verified with a simple test case, as
shown below:

      #define LEN (1024 * 1024 * 1024) // 1GB file
      int main(int argc, char *argv[])
      {
          char *addr;
          int fd, i;

          fd = open("data", O_RDWR);
          if (fd < 0) {
              perror("open");
              exit(-1);
          }

          addr = mmap(NULL, LEN, PROT_READ|PROT_WRITE,
                      MAP_SHARED, fd, 0);
          if (addr == MAP_FAILED) {
              perror("mmap");
              exit(-1);
          }

          if (madvise(addr, LEN, MADV_HUGEPAGE)) {
              perror("madvise");
              exit(-1);
          }

          for (i = 0; i < LEN / 4096; i++)
                memset(addr + i * 4096, 1, 1);

          while (1) {} // Verifiable with /proc/meminfo

          munmap(addr, LEN);
          close(fd);
          exit(0);
      }

This patch adds large folio support to async readahead, aligning its
behavior with sync readahead.

Signed-off-by: Yafang Shao <laoar.shao@xxxxxxxxx>
---
 include/linux/pagemap.h | 4 +++-
 mm/filemap.c            | 4 ++++
 mm/readahead.c          | 9 +++++++++
 3 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index 68a5f1ff3301..b77eb5d3a3dd 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -1350,8 +1350,10 @@ struct readahead_control {
 	pgoff_t _index;
 	unsigned int _nr_pages;
 	unsigned int _batch_count;
-	bool _workingset;
 	unsigned long _pflags;
+	bool _workingset;
+	bool _large_folio;
+	bool _rand_read;
 };
 
 #define DEFINE_READAHEAD(ractl, f, r, m, i)				\
diff --git a/mm/filemap.c b/mm/filemap.c
index 36d22968be9a..c9695effdd88 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -3226,6 +3226,10 @@ static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
 		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
 
 	if (folio_test_readahead(folio)) {
+		if (vmf->vma->vm_flags & VM_HUGEPAGE)
+			ractl._large_folio = true;
+		if (vmf->vma->vm_flags & VM_RAND_READ)
+			ractl._rand_read = true;
 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
 		page_cache_async_ra(&ractl, folio, ra->ra_pages);
 	}
diff --git a/mm/readahead.c b/mm/readahead.c
index 3dc6c7a128dd..8c4f95f87b71 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -667,6 +667,15 @@ void page_cache_async_ra(struct readahead_control *ractl,
 	ra->async_size = ra->size;
 readit:
 	ractl->_index = ra->start;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if (ractl->_large_folio) {
+		ractl->_index &= ~((unsigned long)HPAGE_PMD_NR - 1);
+		if (!ractl->_rand_read)
+			ra->size = 2 * HPAGE_PMD_NR;
+		ra->async_size = HPAGE_PMD_NR;
+		order = HPAGE_PMD_ORDER;
+	}
+#endif
 	page_cache_ra_order(ractl, ra, order);
 }
 EXPORT_SYMBOL_GPL(page_cache_async_ra);
-- 
2.43.5