[ARM] Issue of memory compaction on kernel 2.6.35.9

[ARM] Issue of memory compaction on kernel 2.6.35.9

[naveen yadav]

Dear all,

we want to varify compaction on ARM and we are  using 2.6.25.9 kernel
on cortex a9.

Since ARM does not have HUGETLB_PAGE support and compaction is HUGE
PAGE independent so I removed from config file

******************************************************************************************************************************
# support for memory compaction
config COMPACTION
        bool "Allow for memory compaction"
        select MIGRATION
        #depends on EXPERIMENTAL && HUGETLB_PAGE && MMU
        depends on EXPERIMENTAL && MMU
   help
          Allows the compaction of memory for the allocation of huge pages.	
******************************************************************************************************************************
after triggering Memory Compaction by writing any value to
/proc/sys/vm/compact_memory i am getting the SVC mode crash
******************************************************************************************************************************
#echo 1 > /proc/sys/vm/compact_memory
Unable to handle kernel paging request at virtual address ee420be4
pgd = d9c6c000
[ee420be4] *pgd=00000000
Internal error: Oops: 805 [#1] PREEMPT
last sysfs file:
Modules linked in:
CPU: 0    Not tainted  (2.6.35.9 #16)
PC is at compact_zone+0x178/0x610
LR is at compact_zone+0x138/0x610
pc : [<c009f30c>]    lr : [<c009f2cc>]    psr: 40000093
sp : d9d75e40  ip : c0380978  fp : d9d75e94
r10: d9d74000  r9 : c03806c8  r8 : 00069704
r7 : 00069800  r6 : 00d2e080  r5 : c04ea080  r4 : d9d75e9c
r3 : 60000093  r2 : 00000002  r1 : ee420be4  r0 : ee430b82
Flags: nZcv  IRQs off  FIQs on  Mode SVC_32  ISA ARM  Segment us	
******************************************************************************************************************************


We  tried to narrow down the prob... I found crash is form
“del_page_from_lru_list(zone, page, page_lru(page)); ” function
isolate_migratepages

{
-----------
/* Successfully isolated */
		del_page_from_lru_list(zone, page, page_lru(page));

	--------------------------------------------------
}

In my ARM board have only one zone (Node 0, zone   Normal)

******************************************************************************************************************************
VDLinux#> cat /proc/buddyinfo
Node 0, zone   Normal      5      4      4      2      1      3      0
     2       3      4      2      1      7
******************************************************************************************************************************

Can some one guide why I am geeting the crash and how to debugg this problem


Thanks

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [ARM] Issue of memory compaction on kernel 2.6.35.9

[Mel Gorman]

On Wed, Apr 13, 2011 at 05:05:33PM +0530, naveen yadav wrote:
> Dear all,
> 
> we want to varify compaction on ARM and we are  using 2.6.25.9 kernel
> on cortex a9.
> 
> Since ARM does not have HUGETLB_PAGE support and compaction is HUGE
> PAGE independent so I removed from config file
> 

Bear in mind that if you intend to depend on compaction to allow
devices to use high-order allocations, you could be in trouble in
the future. Compaction gives no guarantees that high-order pages will
be available so a device must still be able to cope with allocation
failure. In the case of transparent hugepage support and hugetlbfs,
allocation failure is not a serious problem.

> ******************************************************************************************************************************
> # support for memory compaction
> config COMPACTION
>         bool "Allow for memory compaction"
>         select MIGRATION
>         #depends on EXPERIMENTAL && HUGETLB_PAGE && MMU
>         depends on EXPERIMENTAL && MMU
>    help
>           Allows the compaction of memory for the allocation of huge pages.	
> ******************************************************************************************************************************
> after triggering Memory Compaction by writing any value to
> /proc/sys/vm/compact_memory i am getting the SVC mode crash
> ******************************************************************************************************************************
> #echo 1 > /proc/sys/vm/compact_memory
> Unable to handle kernel paging request at virtual address ee420be4
> pgd = d9c6c000
> [ee420be4] *pgd=00000000
> Internal error: Oops: 805 [#1] PREEMPT
> last sysfs file:
> Modules linked in:
> CPU: 0    Not tainted  (2.6.35.9 #16)
> PC is at compact_zone+0x178/0x610
> LR is at compact_zone+0x138/0x610
> pc : [<c009f30c>]    lr : [<c009f2cc>]    psr: 40000093
> sp : d9d75e40  ip : c0380978  fp : d9d75e94
> r10: d9d74000  r9 : c03806c8  r8 : 00069704
> r7 : 00069800  r6 : 00d2e080  r5 : c04ea080  r4 : d9d75e9c
> r3 : 60000093  r2 : 00000002  r1 : ee420be4  r0 : ee430b82
> Flags: nZcv  IRQs off  FIQs on  Mode SVC_32  ISA ARM  Segment us	
> ******************************************************************************************************************************
> 
> 
> We  tried to narrow down the prob... I found crash is form
> ?del_page_from_lru_list(zone, page, page_lru(page)); ? function
> isolate_migratepages
> 

ARM punches holes in the mem_map structure to save memory and
memory compaction is not aware of them because it couldn't have been
configured.  Debug what PFN is failing and see if it is near a memory
hole. If it's within a memory hole, that's your problem.

-- 
Mel Gorman
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>