Thread (14 messages) 14 messages, 3 authors, 2014-01-09

possible regression on 3.13 when calling flush_dcache_page

From: Joonsoo Kim <hidden>
Date: 2013-12-18 07:21:22
Also in: linux-mm, linux-mmc, lkml
Subsystem: memory management, the rest · Maintainers: Andrew Morton, Linus Torvalds

On Mon, Dec 16, 2013 at 03:43:43PM +0100, Ludovic Desroches wrote:
Hello,

On Fri, Dec 13, 2013 at 10:59:09AM +0900, Joonsoo Kim wrote:
quoted
On Thu, Dec 12, 2013 at 03:36:19PM +0100, Ludovic Desroches wrote:
quoted
fix mmc mailing list address error

On Thu, Dec 12, 2013 at 03:31:50PM +0100, Ludovic Desroches wrote:
quoted
Hi,

With v3.13-rc3 I have an error when the atmel-mci driver calls
flush_dcache_page (log at the end of the message).

Since I didn't have it before, I did a git bisect and the commit introducing
the error is the following one:

106a74e slab: replace free and inuse in struct slab with newly introduced active

I don't know if this commit has introduced a bug or if it has revealed a bug
in the atmel-mci driver.
Hello,

I think that this commit may not introduce a bug. This patch remove one
variable on slab management structure and replace variable name. So there
is no functional change.
If I have reverted this patch and other ones you did on top of it and
the issue disappear.
Hello,

Could you give me your '/proc/slabinfo' before/after this commit (106a74e)?

And how about testing with artificially increasing size of struct slab on
top of this commit (106a74e)?

I really wonder why the problem happens, because this doesn't cause any
functional change as far as I know. Only side-effect from this patch is
decreasing size of struct slab.

Thanks.
diff --git a/mm/slab.c b/mm/slab.c
index 2ec2336..d2240fd 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -174,6 +174,7 @@ struct slab {
        struct {
                struct list_head list;
                void *s_mem;            /* including colour offset */
+               unsigned int x;
                unsigned int active;    /* num of objs active in slab */
        };
 };
quoted
I doubt that side-effect of this patch reveals a bug in other place.
Side-effect is reduced memory usage for slab management structure. It would
makes some slabs have more objects with more density since slab management
structure is sometimes on the page for objects. So if it diminishes, more
objects can be in the page.

Anyway, I will look at it more. If you have any progress, please let me know.
No progress at the moment.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help