Re: [PATCH 1/3] mm: Print head flags in dump_page

[John Hubbard <jhubbard@nvidia.com>]

On 2020-06-30 04:59, Matthew Wilcox wrote:
...
> How about ...
> [   38.049155] flags: 0x17ffe000000000e(anon|referenced|uptodate|dirty|compound)
> 
> That is, change pageflag_names[] to print 'head' as 'compound' and move the
> 'anon' or 'ksm' to look like a pageflag.  Also CMA.  Like this:
> 
> +++ b/include/trace/events/mmflags.h
> @@ -96,7 +96,7 @@
>          {1UL << PG_private,             "private"       },              \
>          {1UL << PG_private_2,           "private_2"     },              \
>          {1UL << PG_writeback,           "writeback"     },              \
> -       {1UL << PG_head,                "head"          },              \
> +       {1UL << PG_head,                "compound"      },              \
>          {1UL << PG_mappedtodisk,        "mappedtodisk"  },              \
>          {1UL << PG_reclaim,             "reclaim"       },              \
>          {1UL << PG_swapbacked,          "swapbacked"    },              \
> +++ b/mm/debug.c
> @@ -54,7 +54,7 @@ void __dump_page(struct page *page, const char *reason)
>           * state for debugging, it should be fine to accept a bit of
>           * inaccuracy here due to racing.
>           */
> -       bool page_cma = is_migrate_cma_page(page);
> +       char *cma = is_migrate_cma_page(page) ? "|cma" : "";
>          int mapcount;
>          char *type = "";
>   
> @@ -71,6 +71,10 @@ void __dump_page(struct page *page, const char *reason)
>          if (page < head || (page >= head + MAX_ORDER_NR_PAGES)) {
>                  /* Corrupt page, cannot call page_mapping */
>                  mapping = page->mapping;
> +               if ((unsigned long)mapping & PAGE_MAPPING_ANON)
> +                       mapping = NULL;
> +               mapping = (struct address_space *)
> +                               ((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
>                  head = page;
>                  compound = false;
>          } else {
> [...]
>          if (PageKsm(page))
> -               type = "ksm ";
> +               type = "ksm|";
>          else if (PageAnon(page))
> -               type = "anon ";
> -       else if (mapping) {
> +               type = "anon|";
> +
> +       BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS + 1);
> +       pr_warn("flags: %#lx(%s%pGp%s)\n", page->flags, type, &page->flags,
> +                       cma);
> +
> +       if (mapping) {
>                  const struct inode *host;
>                  const struct address_space_operations *a_ops;
> @@ -163,11 +167,6 @@ void __dump_page(struct page *page, const char *reason)
>                  }
>          }
>   out_mapping:
> -       BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS + 1);
> -
> -       pr_warn("%sflags: %#lx(%pGp)%s\n", type, head->flags, &head->flags,
> -               page_cma ? " CMA" : "");
> -
>   hex_only:
>          print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32,
>                          sizeof(unsigned long), page,
> 
> Can also delete the 'out_mapping' label this way.

OK, so after applying that on top of your original series, and your
line-break approach, here's what the output looks like for regular
and THP, head and tail pages (all involving FOLL_PIN):

THP FOLL_PIN page: head page:

[   42.360473] page:0000000025f35fdb refcount:513 mapcount:1 mapping:0000000000000000 index:0x0
[   42.368012] head:0000000025f35fdb order:9 compound_mapcount:1 compound_pincount:512
[   42.374761] flags: 0x17ffe000001000e(anon|referenced|uptodate|dirty|compound)
[   42.380994] raw: 017ffe000001000e ffffffff83649ca0 ffffea0020c80008 ffff888898091901
[   42.387822] raw: 0000000000000000 0000000000000000 00000201ffffffff 0000000000000000
[   42.394680] page dumped because: gup_benchmark: head page: dump_page test

THP FOLL_PIN page: tail page:

[   42.408222] page:00000000803d233b refcount:513 mapcount:1 mapping:0000000000000000 index:0x11
[   42.415850] head:0000000025f35fdb order:9 compound_mapcount:1 compound_pincount:512
[   42.422607] flags: 0x17ffe0000000000(anon|)
[   42.425772] raw: 017ffe0000000000 ffffea0020c98001 ffffea0020c98448 dead000000000400
[   42.432636] raw: 0000000000000000 0000000000000000 00000000ffffffff 0000000000000000
[   42.439490] head: 017ffe000001000e ffffffff83649ca0 ffffea0020c80008 ffff888898091901
[   42.446431] head: 0000000000000000 0000000000000000 00000201ffffffff 0000000000000000
[   42.453355] page dumped because: gup_benchmark: tail page: dump_page test

Non-THP FOLL_PIN page: head page:

[   41.513677] page:00000000190e28ba refcount:1025 mapcount:1 mapping:0000000000000000 index:0x0
[   41.521331] flags: 0x17ffe0000080034(anon|uptodate|lru|active|swapbacked)
[   41.527189] raw: 017ffe0000080034 ffffea002228db08 ffffea0022215108 ffff888898090191
[   41.534020] raw: 0000000000000000 0000000000000000 0000040100000000 0000000000000000
[   41.540863] page dumped because: gup_benchmark: head page: dump_page test

Non-THP FOLL_PIN page: tail page:

[   41.554472] page:00000000696a8210 refcount:1025 mapcount:1 mapping:0000000000000000 index:0x11
[   41.562230] flags: 0x17ffe0000080034(anon|uptodate|lru|active|swapbacked)
[   41.568073] raw: 017ffe0000080034 ffffea0022195688 ffffea0021a4e608 ffff888898090191
[   41.574940] raw: 0000000000000011 0000000000000000 0000040100000000 0000000000000000
[   41.581768] page dumped because: gup_benchmark: tail page: dump_page test

> 
> (I really like it that we're debating this ... it feels like we've been
> in a bit of a push-pull with the debug patches over the years)
> 

Yes, it's good that we're pausing a moment to polish this up. Lots of goodness
here and I like the above output.

thanks,
-- 
John Hubbard
NVIDIA