Messages in this thread | | | Date | Fri, 07 Dec 2012 14:58:57 -0800 | From | Dave Hansen <> | Subject | Re: [PATCH] Debugging: Keep track of page owners |
| |
On 12/07/2012 02:44 PM, Andrew Morton wrote: > AFACIT that difference was undescribed. I can see that the new version > uses the stack-tracing infrastructure, but the change to > pagetypeinfo_showmixedcount_print() is a mystery.
Ahhh, I assume you're talking about this hunk:
>> @@ -976,10 +976,7 @@ static void pagetypeinfo_showmixedcount_print(struct seq_file *m, >> >> pagetype = allocflags_to_migratetype(page->gfp_mask); >> if (pagetype != mtype) { >> - if (is_migrate_cma(pagetype)) >> - count[MIGRATE_MOVABLE]++; >> - else >> - count[mtype]++; >> + count[mtype]++; >> break; >> }
That was to fix the comment that Laura Abbott made about it miscounting MIGRATE_CMA pages.
My patch-sending scripts were choking a bit on the text description in your patch. I'm using a long-ago-forked copy of your patch-utils and the DESC/EDESC in the patch I imported is giving them fits when I send via email and stripping large parts of the description. I'm happy to resend via email, too, but here, the raw patch (will the full description):
https://www.sr71.net/~dave/linux/pageowner.patch
The important description that the scripts managed to strip out when emailed was this:
Updated 12/4/2012 - should apply to 3.7 kernels. I did a quick sniff-test to make sure that this boots and produces some sane output, but it's not been exhaustively tested.
* Moved file over to debugfs (no reason to keep polluting /proc) * Now using generic stack tracking infrastructure * Added check for MIGRATE_CMA pages to explicitly count them as movable.
The new snprint_stack_trace() probably belongs in its own patch if this were to get merged, but it won't kill anyone as it stands.
-----
| |