Thread (8 messages) flat view 8 messages, 3 authors, 2024-05-27

Re: [PATCH 2/2] mm: debug: print correct information for slab folios

From: Matthew Wilcox <willy@infradead.org>
Date: 2024-05-22 12:32:35
Also in: linux-doc, linux-mm, lkml

On Wed, May 22, 2024 at 04:46:29PM +0900, Sukrit Bhatnagar wrote:
The function dump_page() prints "anon" even for slab pages.
This is not correct, especially now that struct slab is separated from
struct page, and that the slab pages cannot be mapped to userspace.

[    7.071985] page: refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x102768
[    7.072602] head: order:3 entire_mapcount:0 nr_pages_mapped:0 pincount:0
[    7.073085] anon flags: 0x8000000000000840(slab|head|zone=2)
[    7.073777] raw: 8000000000000840 ffff8881000419c0 0000000000000000 dead000000000001

This debugging output may be misleading, and it is not easy to understand
unless we read the source code.

If the folio tests true for slab, do not print information that does not
apply to it. Instead, print the slab flags stored in the kmem_cache field.

[    7.248722] page: refcount:1 mapcount:0 mapping:0000000000000000 index:0xffff888103e6aa87>
[    7.249135] head: order:3 entire_mapcount:0 nr_pages_mapped:0 pincount:0
[    7.249429] slab flags: 0x8000000000000840(slab|head|zone=2)
[    7.249664] cache flags: 0x10310(HWCACHE_ALIGN|PANIC|TYPESAFE_BY_RCU|CMPXCHG_DOUBLE)
[    7.249999] raw: 8000000000000000 ffffea00040f9a01 ffffea00040f9bc8 dead000000000400
You haven't tested this against the current codebase ...
quoted hunk ↗ jump to hunk
@@ -98,6 +101,8 @@ static void __dump_folio(struct folio *folio, struct page *page,
 		is_migrate_cma_folio(folio, pfn) ? " CMA" : "");
 	if (page_has_type(&folio->page))
 		pr_warn("page_type: %pGt\n", &folio->page.page_type);
+	else if (folio_test_slab(folio))
+		pr_warn("cache flags: %pGs\n", &((struct slab *)&folio->page)->slab_cache->flags);
 
... because page_has_type() is now true for slab; there is no more
PG_slab.  I think you also want:

	folio_slab(folio)->slab_cache->flags

Anyway, we have print_slab_info() which is currently static in slub.c.
Maybe that needs to become non-static and dump_page() should call that
for slabs?
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help