gluebi problems on 2.6.28 backport

gluebi problems on 2.6.28 backport

[Holger Brunck]

Hi all, 
we are working on a powerpc (ppc 8xx) based system and we use ubi/ubifs
on a NOR flash device. Our kernel version is 2.6.28.

I have tried to integrate the newest UBI patches from the backport for
2.6.28 from:  
http://git.infradead.org/users/dedekind/ubifs-v2.6.28.git

Now ubi crashes if I try to write into a mtdblock device. Via "git
bisect" I have seen that the rework of the gluebi patch causes the
problem: 
	UBI: make gluebi a separate module
or
        UBI: remove built-in gluebi

Unfortunately the second patch did not compile standalone for that, both
patches can cause the problem. 

I have formated a ubi device, attached and then created three ubi
volumes. ubinfo seems to be ok afterwards. 

If I try to copy something in a volume: 
cp /usr/local/bin/testfile /dev/mtdblock5
the kernel crashes. 

The error message is: 
-bash-3.2# cp /usr/local/bin/testfile /dev/mtdblock5
Unable to handle kernel paging request for data at address 0x00000000
Faulting instruction address: 0xc0197cec
Oops: Kernel access of bad area, sig: 11 [#1]
PREEMPT MGSUVD
Modules linked in:
NIP: c0197cec LR: c019e154 CTR: c019e084
REGS: c38f3e60 TRAP: 0300   Not tainted  (2.6.28-b001-dev-3.1.3)
MSR: 00009032 <EE,ME,IR,DR>  CR: 33555899  XER: c0000000
DAR: 00000000, DSISR: c0000000
TASK = c3846800[126] 'mtdblockd' THREAD: c38f2000
GPR00: c019e154 c38f3f10 c3846800 00000000 00000001 c3a83800 0000a880
00000200 
GPR08: 00000000 00015700 00000000 0001ff80 0000a880 04110900 03e7ab00
00000000 
GPR16: 03e6fa28 03e747f0 00000000 00000000 00000000 c00409b8 c02c5ff0
c38f3f78 
GPR24: 00000200 c3a83800 c3954d90 00000001 c3a83800 c3954d90 00000200
00000200 
NIP [c0197cec] ubi_leb_read+0x18/0x138
LR [c019e154] gluebi_read+0xd0/0x120
Call Trace:
[c38f3f10] [00001032] 0x1032 (unreliable)
[c38f3f30] [c019e154] gluebi_read+0xd0/0x120
[c38f3f70] [c01869cc] mtdblock_readsect+0xdc/0x120
[c38f3fb0] [c0185d2c] mtd_blktrans_thread+0x198/0x2a4
[c38f3fe0] [c0040a04] kthread+0x4c/0x88
[c38f3ff0] [c000d530] kernel_thread+0x4c/0x68
Instruction dump:
3860ffe2 80010014 83e1000c 7c0803a6 38210010 4e800020 7c0802a6 9421ffe0 
7ccc3378 bf810010 90010024 7cbc2b78 <83e30000> 7cea3b78 83bf0170
7d094378 
---[ end trace 5690d91bc0c9bd48 ]---

Do I miss something or is there anybody who has a idea whats going
wrong? 

Regards 
Holger Brunck

Re: gluebi problems on 2.6.28 backport

[Artem Bityutskiy]

Hi,

On Thu, 2009-07-09 at 17:19 +0200, Holger Brunck wrote:
> Hi all, 
> we are working on a powerpc (ppc 8xx) based system and we use ubi/ubifs
> on a NOR flash device. Our kernel version is 2.6.28.
> 
> I have tried to integrate the newest UBI patches from the backport for
> 2.6.28 from:  
> http://git.infradead.org/users/dedekind/ubifs-v2.6.28.git

Note, I've just back-ported and pushed all latest updates to the
back-port trees.

> Unfortunately the second patch did not compile standalone for that, both
> patches can cause the problem. 

Yeah, for mainline this problem does not exist, though.

> Unable to handle kernel paging request for data at address 0x00000000
> Faulting instruction address: 0xc0197cec
> Oops: Kernel access of bad area, sig: 11 [#1]
> PREEMPT MGSUVD
> Modules linked in:
> NIP: c0197cec LR: c019e154 CTR: c019e084
> REGS: c38f3e60 TRAP: 0300   Not tainted  (2.6.28-b001-dev-3.1.3)
> MSR: 00009032 <EE,ME,IR,DR>  CR: 33555899  XER: c0000000
> DAR: 00000000, DSISR: c0000000
> TASK = c3846800[126] 'mtdblockd' THREAD: c38f2000
> GPR00: c019e154 c38f3f10 c3846800 00000000 00000001 c3a83800 0000a880
> 00000200 
> GPR08: 00000000 00015700 00000000 0001ff80 0000a880 04110900 03e7ab00
> 00000000 
> GPR16: 03e6fa28 03e747f0 00000000 00000000 00000000 c00409b8 c02c5ff0
> c38f3f78 
> GPR24: 00000200 c3a83800 c3954d90 00000001 c3a83800 c3954d90 00000200
> 00000200 
> NIP [c0197cec] ubi_leb_read+0x18/0x138
> LR [c019e154] gluebi_read+0xd0/0x120
> Call Trace:
> [c38f3f10] [00001032] 0x1032 (unreliable)
> [c38f3f30] [c019e154] gluebi_read+0xd0/0x120
> [c38f3f70] [c01869cc] mtdblock_readsect+0xdc/0x120
> [c38f3fb0] [c0185d2c] mtd_blktrans_thread+0x198/0x2a4
> [c38f3fe0] [c0040a04] kthread+0x4c/0x88
> [c38f3ff0] [c000d530] kernel_thread+0x4c/0x68
> Instruction dump:
> 3860ffe2 80010014 83e1000c 7c0803a6 38210010 4e800020 7c0802a6 9421ffe0 
> 7ccc3378 bf810010 90010024 7cbc2b78 <83e30000> 7cea3b78 83bf0170
> 7d094378 

The problem is that the mtd_blkdevs module does not open MTD devices
before working with them. Gluebi changes revealed this problem. Below
is the patch which should fix the issue. Please, try and let me know
if it helps.

======================================================================

From: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Subject: [PATCH] mtd: blkdevs: do not for get to get MTD devices

Nowadays MTD devices have to be "get" before they can be
used. This has to be done with 'put_mtd_device()'. The
'blktrans_open()' function did not do this and instead
used 'try_module_get()'. Fixe this.

Since 'put_mtd_device()' already gets the module, extra
'try_module_get()' is not needed.

This fixes oops when one tries to use mtdbloc on tope of
gluebi.

Reported-by: Holger Brunck <holger.brunck@keymile.com>
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
---
 drivers/mtd/mtd_blkdevs.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/mtd/mtd_blkdevs.c b/drivers/mtd/mtd_blkdevs.c
index 1409f01..2f55242 100644
--- a/drivers/mtd/mtd_blkdevs.c
+++ b/drivers/mtd/mtd_blkdevs.c
@@ -139,7 +139,7 @@ static int blktrans_open(struct block_device *bdev, fmode_t mode)
 	struct mtd_blktrans_ops *tr = dev->tr;
 	int ret = -ENODEV;
 
-	if (!try_module_get(dev->mtd->owner))
+	if (!get_mtd_device(NULL, dev->mtd->index))
 		goto out;
 
 	if (!try_module_get(tr->owner))
@@ -153,7 +153,7 @@ static int blktrans_open(struct block_device *bdev, fmode_t mode)
 	ret = 0;
 	if (tr->open && (ret = tr->open(dev))) {
 		dev->mtd->usecount--;
-		module_put(dev->mtd->owner);
+		put_mtd_device(dev->mtd);
 	out_tr:
 		module_put(tr->owner);
 	}
@@ -172,7 +172,7 @@ static int blktrans_release(struct gendisk *disk, fmode_t mode)
 
 	if (!ret) {
 		dev->mtd->usecount--;
-		module_put(dev->mtd->owner);
+		put_mtd_device(dev->mtd);
 		module_put(tr->owner);
 	}
 
-- 
1.6.0.6

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: gluebi problems on 2.6.28 backport

[Holger Brunck]

Hi, 

> The problem is that the mtd_blkdevs module does not open MTD devices
> before working with them. Gluebi changes revealed this problem. Below
> is the patch which should fix the issue. Please, try and let me know
> if it helps.
> 
> ======================================================================
> 
> From: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
> Subject: [PATCH] mtd: blkdevs: do not for get to get MTD devices
> 
> Nowadays MTD devices have to be "get" before they can be
> used. This has to be done with 'put_mtd_device()'. The
> 'blktrans_open()' function did not do this and instead
> used 'try_module_get()'. Fixe this.
> 
> Since 'put_mtd_device()' already gets the module, extra
> 'try_module_get()' is not needed.
> 
> This fixes oops when one tries to use mtdbloc on tope of
> gluebi.
> 
> Reported-by: Holger Brunck <holger.brunck@keymile.com>
> Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
> ---
>  drivers/mtd/mtd_blkdevs.c |    6 +++---
>  1 files changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/mtd/mtd_blkdevs.c b/drivers/mtd/mtd_blkdevs.c
> index 1409f01..2f55242 100644
> --- a/drivers/mtd/mtd_blkdevs.c
> +++ b/drivers/mtd/mtd_blkdevs.c
> @@ -139,7 +139,7 @@ static int blktrans_open(struct block_device *bdev, fmode_t mode)
>  	struct mtd_blktrans_ops *tr = dev->tr;
>  	int ret = -ENODEV;
>  
> -	if (!try_module_get(dev->mtd->owner))
> +	if (!get_mtd_device(NULL, dev->mtd->index))
>  		goto out;
>  
>  	if (!try_module_get(tr->owner))
> @@ -153,7 +153,7 @@ static int blktrans_open(struct block_device *bdev, fmode_t mode)
>  	ret = 0;
>  	if (tr->open && (ret = tr->open(dev))) {
>  		dev->mtd->usecount--;
> -		module_put(dev->mtd->owner);
> +		put_mtd_device(dev->mtd);
>  	out_tr:
>  		module_put(tr->owner);
>  	}
> @@ -172,7 +172,7 @@ static int blktrans_release(struct gendisk *disk, fmode_t mode)
>  
>  	if (!ret) {
>  		dev->mtd->usecount--;
> -		module_put(dev->mtd->owner);
> +		put_mtd_device(dev->mtd);
>  		module_put(tr->owner);
>  	}
>  
> -- 
> 1.6.0.6
> 
yes this patch solves my problem. 

Thanks a lot.

Regards 
Holger Brunck

Re: gluebi problems on 2.6.28 backport

[Artem Bityutskiy]

On Fri, 2009-07-10 at 14:50 +0200, Holger Brunck wrote:
> > 
> yes this patch solves my problem. 

OK, I'll push it to back-port trees then and try to persuade dwmw2
to send it upstresm for 2.6.31 inclusion.

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)