@@ -623,6 +623,8 @@ int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
extern void delete_from_page_cache(struct page *page);
extern void __delete_from_page_cache(struct page *page, void *shadow);
int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
+void delete_from_page_cache_batch(struct address_space *mapping, int count,
+ struct page **pages);
/*
* Like add_to_page_cache_locked, but used to add newly allocated pages:
@@ -304,6 +304,90 @@ void delete_from_page_cache(struct page *page)
}
EXPORT_SYMBOL(delete_from_page_cache);
+/*
+ * page_cache_tree_delete_batch - delete several pages from page cache
+ * @mapping: the mapping to which pages belong
+ * @count: the number of pages to delete
+ * @pages: pages that should be deleted
+ *
+ * The function walks over mapping->page_tree and removes pages passed in
+ * @pages array from the radix tree. The function expects @pages array to
+ * sorted by page index. It tolerates holes in @pages array (radix tree
+ * entries at those indices are not modified). The function expects only THP
+ * head pages to be present in the @pages array and takes care to delete all
+ * corresponding tail pages from the radix tree as well.
+ *
+ * The function expects mapping->tree_lock to be held.
+ */
+static void
+page_cache_tree_delete_batch(struct address_space *mapping, int count,
+ struct page **pages)
+{
+ struct radix_tree_iter iter;
+ void **slot;
+ int total_pages = 0;
+ int i = 0, tail_pages = 0;
+ struct page *page;
+ pgoff_t start;
+
+ start = pages[0]->index;
+ radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
+ if (i >= count && !tail_pages)
+ break;
+ page = radix_tree_deref_slot_protected(slot,
+ &mapping->tree_lock);
+ if (radix_tree_exceptional_entry(page))
+ continue;
+ if (!tail_pages) {
+ /*
+ * Some page got inserted in our range? Skip it. We
+ * have our pages locked so they are protected from
+ * being removed.
+ */
+ if (page != pages[i])
+ continue;
+ WARN_ON_ONCE(!PageLocked(page));
+ if (PageTransHuge(page) && !PageHuge(page))
+ tail_pages = HPAGE_PMD_NR - 1;
+ page->mapping = NULL;
+ /*
+ * Leave page->index set: truncation lookup relies
+ * upon it
+ */
+ i++;
+ } else {
+ tail_pages--;
+ }
+ radix_tree_clear_tags(&mapping->page_tree, iter.node, slot);
+ __radix_tree_replace(&mapping->page_tree, iter.node, slot, NULL,
+ workingset_update_node, mapping);
+ total_pages++;
+ }
+ mapping->nrpages -= total_pages;
+}
+
+void delete_from_page_cache_batch(struct address_space *mapping, int count,
+ struct page **pages)
+{
+ int i;
+ unsigned long flags;
+
+ if (!count)
+ return;
+
+ spin_lock_irqsave(&mapping->tree_lock, flags);
+ for (i = 0; i < count; i++) {
+ trace_mm_filemap_delete_from_page_cache(pages[i]);
+
+ unaccount_page_cache_page(mapping, pages[i]);
+ }
+ page_cache_tree_delete_batch(mapping, count, pages);
+ spin_unlock_irqrestore(&mapping->tree_lock, flags);
+
+ for (i = 0; i < count; i++)
+ page_cache_free_page(mapping, pages[i]);
+}
+
int filemap_check_errors(struct address_space *mapping)
{
int ret = 0;
@@ -294,6 +294,14 @@ void truncate_inode_pages_range(struct address_space *mapping,
while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
min(end - index, (pgoff_t)PAGEVEC_SIZE),
indices)) {
+ /*
+ * Pagevec array has exceptional entries and we may also fail
+ * to lock some pages. So we store pages that can be deleted
+ * in an extra array.
+ */
+ struct page *pages[PAGEVEC_SIZE];
+ int batch_count = 0;
+
for (i = 0; i < pagevec_count(&pvec); i++) {
struct page *page = pvec.pages[i];
@@ -315,9 +323,17 @@ void truncate_inode_pages_range(struct address_space *mapping,
unlock_page(page);
continue;
}
- truncate_inode_page(mapping, page);
- unlock_page(page);
+ if (page->mapping != mapping) {
+ unlock_page(page);
+ continue;
+ }
+ pages[batch_count++] = page;
}
+ for (i = 0; i < batch_count; i++)
+ truncate_cleanup_page(mapping, pages[i]);
+ delete_from_page_cache_batch(mapping, batch_count, pages);
+ for (i = 0; i < batch_count; i++)
+ unlock_page(pages[i]);
pagevec_remove_exceptionals(&pvec);
pagevec_release(&pvec);
cond_resched();
Currently we remove pages from the radix tree one by one. To speed up page cache truncation, lock several pages at once and free them in one go. This allows us to batch radix tree operations in a more efficient way and also save round-trips on mapping->tree_lock. As a result we gain about 20% speed improvement in page cache truncation. Data from a simple benchmark timing 10000 truncates of 1024 pages (on ext4 on ramdisk but the filesystem is barely visible in the profiles). The range shows 1% and 95% percentiles of the measured times: 4.14-rc2 4.14-rc2 + batched truncation 248-256 209-219 249-258 209-217 248-255 211-239 248-255 209-217 247-256 210-218 Signed-off-by: Jan Kara <jack@suse.cz> --- include/linux/pagemap.h | 2 ++ mm/filemap.c | 84 +++++++++++++++++++++++++++++++++++++++++++++++++ mm/truncate.c | 20 ++++++++++-- 3 files changed, 104 insertions(+), 2 deletions(-)