UBI/UBIFS on NAND with PEB 4MB

UBI/UBIFS on NAND with PEB 4MB

[Matthias Auchmann]

Hi guys!

I'm trying to use UBI and UBIFS on a Micron MT29F32G08ABCDBJ4 SLC NAND with an erase block size 4MB.

First of all, mkfs.ubifs wouldn't support a LEB larger than UBIFS_MAX_LEB_SZ being (2*1024*1024).

When I use ubiattach though, it seems to be quite happy, giving me that on a 128MB partition:

UBI device number 0, total 32 LEBs (133169152 bytes, 127.0 MiB), available 8 LEBs (33292288 bytes, 31.8 MiB), LEB size 4161536 bytes (4.0 MiB)

Anyway, when I detach and reattach, it won't work. Details are in the bottom of that mail, but since I haven't thoroughly looked into it myself yet, this is only for those who care... I don't want to ask preliminary questions.

Anyway, my questions I guess then are:
-) Are there any patches/branches that enable UBIFS to be run with a LEB of 4161536?
-) Are there any known limitations with using a 4MiB PEB with UBI itself, or shall I assume for now that the problem is on my board/NAND driver regarding the above ubiattach errors upon the second attach?

Thanks!

BR,

Matthias



*** Please only read this if you care: ***
Details about attach/detach/attach problem.
Upon first attaching, dmesg says:
[  117.483503] ubi0: attaching mtd5
[  117.502347] ubi0: scanning is finished
[  117.502353] ubi0: empty MTD device detected
[  117.513280] ubi0: attached mtd5 (name "rootfs01", size 128 MiB)
[  117.513288] ubi0: PEB size: 4194304 bytes (4096 KiB), LEB size: 4161536 bytes
[  117.513293] ubi0: min./max. I/O unit sizes: 16384/16384, sub-page size 16384
[  117.513299] ubi0: VID header offset: 16384 (aligned 16384), data offset: 32768
[  117.513304] ubi0: good PEBs: 32, bad PEBs: 0, corrupted PEBs: 0
[  117.513309] ubi0: user volume: 0, internal volumes: 1, max. volumes count: 128
[  117.513314] ubi0: max/mean erase counter: 0/0, WL threshold: 4096, image sequence number: 34480849
[  117.513320] ubi0: available PEBs: 8, total reserved PEBs: 24, PEBs reserved for bad PEB handling: 20
[  117.513333] ubi0: background thread "ubi_bgt0d" started, PID 1637

Anyway, when I detach and reattach, I get this:
ubiattach: error!: cannot attach "/dev/mtd5"
           error 22 (Invalid argument)

dmesg says:
[  359.282384] ubi0: attaching mtd5
[  359.319433] ubi0: scanning is finished
[  359.321747] ubi0 error: vtbl_check: bad CRC at record 0: 0x2b4f6dd5, not 0x000000
[  359.321753] Volume table record 0 dump:
[  359.321757]  reserved_pebs   -16777216
[  359.321760]  alignment       0
[  359.321763]  data_pad        0
[  359.321766]  vol_type        0
[  359.321770]  upd_marker      0
[  359.321773]  name_len        0
[  359.321776]  name            NULL
[  359.321784] ubi0 error: vtbl_check: bad CRC at record 0: 0x2b4f6dd5, not 0x000000
[  359.321788] Volume table record 0 dump:
[  359.321791]  reserved_pebs   -16777216
[  359.321794]  alignment       0
[  359.321797]  data_pad        0
[  359.321800]  vol_type        0
[  359.321803]  upd_marker      0
[  359.321807]  name_len        0
[  359.321809]  name            NULL
[  359.321816] ubi0 error: ubi_read_volume_table: both volume tables are corrupted
[  359.321836] ubi0 error: ubi_attach_mtd_dev: failed to attach mtd5, error -22

RE: UBI/UBIFS on NAND with PEB 4MB

[Matthias Auchmann]

Oh please forgive me, I forgot to mention: I'm on Kernel 4.6.

-----Original Message-----
From: linux-mtd [mailto:linux-mtd-bounces@lists.infradead.org] On Behalf Of Matthias Auchmann
Sent: Tuesday, December 13, 2016 8:41 PM
To: linux-mtd@lists.infradead.org
Subject: UBI/UBIFS on NAND with PEB 4MB

Hi guys!

I'm trying to use UBI and UBIFS on a Micron MT29F32G08ABCDBJ4 SLC NAND with an erase block size 4MB.

First of all, mkfs.ubifs wouldn't support a LEB larger than UBIFS_MAX_LEB_SZ being (2*1024*1024).

When I use ubiattach though, it seems to be quite happy, giving me that on a 128MB partition:

UBI device number 0, total 32 LEBs (133169152 bytes, 127.0 MiB), available 8 LEBs (33292288 bytes, 31.8 MiB), LEB size 4161536 bytes (4.0 MiB)

Anyway, when I detach and reattach, it won't work. Details are in the bottom of that mail, but since I haven't thoroughly looked into it myself yet, this is only for those who care... I don't want to ask preliminary questions.

Anyway, my questions I guess then are:
-) Are there any patches/branches that enable UBIFS to be run with a LEB of 4161536?
-) Are there any known limitations with using a 4MiB PEB with UBI itself, or shall I assume for now that the problem is on my board/NAND driver regarding the above ubiattach errors upon the second attach?

Thanks!

BR,

Matthias



*** Please only read this if you care: *** Details about attach/detach/attach problem.
Upon first attaching, dmesg says:
[  117.483503] ubi0: attaching mtd5
[  117.502347] ubi0: scanning is finished [  117.502353] ubi0: empty MTD device detected [  117.513280] ubi0: attached mtd5 (name "rootfs01", size 128 MiB) [  117.513288] ubi0: PEB size: 4194304 bytes (4096 KiB), LEB size: 4161536 bytes [  117.513293] ubi0: min./max. I/O unit sizes: 16384/16384, sub-page size 16384 [  117.513299] ubi0: VID header offset: 16384 (aligned 16384), data offset: 32768 [  117.513304] ubi0: good PEBs: 32, bad PEBs: 0, corrupted PEBs: 0 [  117.513309] ubi0: user volume: 0, internal volumes: 1, max. volumes count: 128 [  117.513314] ubi0: max/mean erase counter: 0/0, WL threshold: 4096, image sequence number: 34480849 [  117.513320] ubi0: available PEBs: 8, total reserved PEBs: 24, PEBs reserved for bad PEB handling: 20 [  117.513333] ubi0: background thread "ubi_bgt0d" started, PID 1637

Anyway, when I detach and reattach, I get this:
ubiattach: error!: cannot attach "/dev/mtd5"
           error 22 (Invalid argument)

dmesg says:
[  359.282384] ubi0: attaching mtd5
[  359.319433] ubi0: scanning is finished [  359.321747] ubi0 error: vtbl_check: bad CRC at record 0: 0x2b4f6dd5, not 0x000000 [  359.321753] Volume table record 0 dump:
[  359.321757]  reserved_pebs   -16777216
[  359.321760]  alignment       0
[  359.321763]  data_pad        0
[  359.321766]  vol_type        0
[  359.321770]  upd_marker      0
[  359.321773]  name_len        0
[  359.321776]  name            NULL
[  359.321784] ubi0 error: vtbl_check: bad CRC at record 0: 0x2b4f6dd5, not 0x000000 [  359.321788] Volume table record 0 dump:
[  359.321791]  reserved_pebs   -16777216
[  359.321794]  alignment       0
[  359.321797]  data_pad        0
[  359.321800]  vol_type        0
[  359.321803]  upd_marker      0
[  359.321807]  name_len        0
[  359.321809]  name            NULL
[  359.321816] ubi0 error: ubi_read_volume_table: both volume tables are corrupted [  359.321836] ubi0 error: ubi_attach_mtd_dev: failed to attach mtd5, error -22

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/

Re: UBI/UBIFS on NAND with PEB 4MB

[Boris Brezillon]

Hi Matthias,

On Tue, 13 Dec 2016 20:40:59 +0100
Matthias Auchmann <m.auchmann@artech.at> wrote:

> Hi guys!
> 
> I'm trying to use UBI and UBIFS on a Micron MT29F32G08ABCDBJ4 SLC NAND with an erase block size 4MB.

Woh! an SLC NAND with 4MB eraseblocks, that's surprizing.

> 
> First of all, mkfs.ubifs wouldn't support a LEB larger than UBIFS_MAX_LEB_SZ being (2*1024*1024).

This is the easy part ;). We already have patches to increase the limit
in different ubi utils (IIRC, it's also needed in ubinize).

> 
> When I use ubiattach though, it seems to be quite happy, giving me that on a 128MB partition:
> 
> UBI device number 0, total 32 LEBs (133169152 bytes, 127.0 MiB), available 8 LEBs (33292288 bytes, 31.8 MiB), LEB size 4161536 bytes (4.0 MiB)
> 
> Anyway, when I detach and reattach, it won't work. Details are in the bottom of that mail, but since I haven't thoroughly looked into it myself yet, this is only for those who care... I don't want to ask preliminary questions.
> 
> Anyway, my questions I guess then are:
> -) Are there any patches/branches that enable UBIFS to be run with a LEB of 4161536?
> -) Are there any known limitations with using a 4MiB PEB with UBI itself, or shall I assume for now that the problem is on my board/NAND driver regarding the above ubiattach errors upon the second attach?

I don't think so. We did several tests on different MLC NANDs exposing
4MB blocks, and the UBI/UBIFS layers were not complaining about that
(we had other problems related to the MLC constraints).
What's your controller?

Regards,

Boris

RE: UBI/UBIFS on NAND with PEB 4MB

[Matthias Auchmann]

Hi Boris!

Thanks for your reply. Just now I found out that when I disable the DMA of my NAND driver, ubiattach/detach works. So it must be a problem with the DMA and/or cache coherency on this brand-new cache coherent interconnect or whatever. I will investigate it later! Sorry for having bothered you guys with that, but as I said yesterday night it was kind of preliminary ;-)

On the patches for the UBIFS_MAX_LEB_SZ, can you tell me where I can find those patches? That would be great ;-)

BR,

Matthias


-----Original Message-----
From: Boris Brezillon [mailto:boris.brezillon@free-electrons.com] 
Sent: Wednesday, December 14, 2016 8:43 AM
To: Matthias Auchmann
Cc: linux-mtd@lists.infradead.org
Subject: Re: UBI/UBIFS on NAND with PEB 4MB

Hi Matthias,

On Tue, 13 Dec 2016 20:40:59 +0100
Matthias Auchmann <m.auchmann@artech.at> wrote:

> Hi guys!
> 
> I'm trying to use UBI and UBIFS on a Micron MT29F32G08ABCDBJ4 SLC NAND with an erase block size 4MB.

Woh! an SLC NAND with 4MB eraseblocks, that's surprizing.

> 
> First of all, mkfs.ubifs wouldn't support a LEB larger than UBIFS_MAX_LEB_SZ being (2*1024*1024).

This is the easy part ;). We already have patches to increase the limit in different ubi utils (IIRC, it's also needed in ubinize).

> 
> When I use ubiattach though, it seems to be quite happy, giving me that on a 128MB partition:
> 
> UBI device number 0, total 32 LEBs (133169152 bytes, 127.0 MiB), 
> available 8 LEBs (33292288 bytes, 31.8 MiB), LEB size 4161536 bytes 
> (4.0 MiB)
> 
> Anyway, when I detach and reattach, it won't work. Details are in the bottom of that mail, but since I haven't thoroughly looked into it myself yet, this is only for those who care... I don't want to ask preliminary questions.
> 
> Anyway, my questions I guess then are:
> -) Are there any patches/branches that enable UBIFS to be run with a LEB of 4161536?
> -) Are there any known limitations with using a 4MiB PEB with UBI itself, or shall I assume for now that the problem is on my board/NAND driver regarding the above ubiattach errors upon the second attach?

I don't think so. We did several tests on different MLC NANDs exposing 4MB blocks, and the UBI/UBIFS layers were not complaining about that (we had other problems related to the MLC constraints).
What's your controller?

Regards,

Boris

RE: UBI/UBIFS on NAND with PEB 4MB

[Matthias Auchmann]

Hi guys!

Sorry that not much happened for such a long time. My colleague Markus has now picked up the investigation!

BR,

Matthias


-----Original Message-----
From: linux-mtd [mailto:linux-mtd-bounces@lists.infradead.org] On Behalf Of Matthias Auchmann
Sent: Wednesday, December 14, 2016 9:59 AM
To: Boris Brezillon
Cc: linux-mtd@lists.infradead.org
Subject: RE: UBI/UBIFS on NAND with PEB 4MB

Hi Boris!

Thanks for your reply. Just now I found out that when I disable the DMA of my NAND driver, ubiattach/detach works. So it must be a problem with the DMA and/or cache coherency on this brand-new cache coherent interconnect or whatever. I will investigate it later! Sorry for having bothered you guys with that, but as I said yesterday night it was kind of preliminary ;-)

On the patches for the UBIFS_MAX_LEB_SZ, can you tell me where I can find those patches? That would be great ;-)

BR,

Matthias


-----Original Message-----
From: Boris Brezillon [mailto:boris.brezillon@free-electrons.com]
Sent: Wednesday, December 14, 2016 8:43 AM
To: Matthias Auchmann
Cc: linux-mtd@lists.infradead.org
Subject: Re: UBI/UBIFS on NAND with PEB 4MB

Hi Matthias,

On Tue, 13 Dec 2016 20:40:59 +0100
Matthias Auchmann <m.auchmann@artech.at> wrote:

> Hi guys!
> 
> I'm trying to use UBI and UBIFS on a Micron MT29F32G08ABCDBJ4 SLC NAND with an erase block size 4MB.

Woh! an SLC NAND with 4MB eraseblocks, that's surprizing.

> 
> First of all, mkfs.ubifs wouldn't support a LEB larger than UBIFS_MAX_LEB_SZ being (2*1024*1024).

This is the easy part ;). We already have patches to increase the limit in different ubi utils (IIRC, it's also needed in ubinize).

> 
> When I use ubiattach though, it seems to be quite happy, giving me that on a 128MB partition:
> 
> UBI device number 0, total 32 LEBs (133169152 bytes, 127.0 MiB), 
> available 8 LEBs (33292288 bytes, 31.8 MiB), LEB size 4161536 bytes
> (4.0 MiB)
> 
> Anyway, when I detach and reattach, it won't work. Details are in the bottom of that mail, but since I haven't thoroughly looked into it myself yet, this is only for those who care... I don't want to ask preliminary questions.
> 
> Anyway, my questions I guess then are:
> -) Are there any patches/branches that enable UBIFS to be run with a LEB of 4161536?
> -) Are there any known limitations with using a 4MiB PEB with UBI itself, or shall I assume for now that the problem is on my board/NAND driver regarding the above ubiattach errors upon the second attach?

I don't think so. We did several tests on different MLC NANDs exposing 4MB blocks, and the UBI/UBIFS layers were not complaining about that (we had other problems related to the MLC constraints).
What's your controller?

Regards,

Boris

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/

AW: UBI/UBIFS on NAND with PEB 4MB

[Markus Macsek]

[-- Attachment #1: Type: text/plain, Size: 4137 bytes --]

Hi,

we updated now to Linux kernel 4.9 and use mtd-utils 2.0. Unfortunately the
UBIFS problems remain the same.

Thus, I did comprehensive tests concerning our NAND and UBIFS (cp. attached
log files).

After all, I conclude that there is most probable an issue with the arasan nand
driver ([linux-xlnx.git]/drivers/mtd/nand/arasan_nand.c) regarding ECC. When
mounting the filesystem a corruption in the empty space at LEB 1:16384 is found
(cp. mount.log). I guess the problem is due to ECC changing a couple of "ff"
into "ef" (cp. nanddump2.log vs. nanddump3.log):

http://linux-mtd.infradead.narkive.com/Uvx5vHZc/ubifs-and-hardware-ecc-of-all-ff-pages-of-mlc-nand

Do you agree or are there other reasons for this behavior?

Note: In order to make mkfs.ubifs support the NANDs erase block size of 4M we
simply increased UBIFS_MAX_PEB_SZ and UBIFS_MAX_LEB_SZ from (2*1024*1024) to
(4*1024*1024), respectively (cp. increase-LEB-PEB-max-size-to-4M).

Best regards,
Markus

-----Ursprüngliche Nachricht-----
Von: Matthias Auchmann 
Gesendet: Dienstag, 11. April 2017 11:27
An: linux-mtd@lists.infradead.org
Cc: Markus Macsek
Betreff: RE: UBI/UBIFS on NAND with PEB 4MB

Hi guys!

Sorry that not much happened for such a long time. My colleague Markus has now picked up the investigation!

BR,

Matthias


-----Original Message-----
From: linux-mtd [mailto:linux-mtd-bounces@lists.infradead.org] On Behalf Of Matthias Auchmann
Sent: Wednesday, December 14, 2016 9:59 AM
To: Boris Brezillon
Cc: linux-mtd@lists.infradead.org
Subject: RE: UBI/UBIFS on NAND with PEB 4MB

Hi Boris!

Thanks for your reply. Just now I found out that when I disable the DMA of my NAND driver, ubiattach/detach works. So it must be a problem with the DMA and/or cache coherency on this brand-new cache coherent interconnect or whatever. I will investigate it later! Sorry for having bothered you guys with that, but as I said yesterday night it was kind of preliminary ;-)

On the patches for the UBIFS_MAX_LEB_SZ, can you tell me where I can find those patches? That would be great ;-)

BR,

Matthias


-----Original Message-----
From: Boris Brezillon [mailto:boris.brezillon@free-electrons.com]
Sent: Wednesday, December 14, 2016 8:43 AM
To: Matthias Auchmann
Cc: linux-mtd@lists.infradead.org
Subject: Re: UBI/UBIFS on NAND with PEB 4MB

Hi Matthias,

On Tue, 13 Dec 2016 20:40:59 +0100
Matthias Auchmann <m.auchmann@artech.at> wrote:

> Hi guys!
> 
> I'm trying to use UBI and UBIFS on a Micron MT29F32G08ABCDBJ4 SLC NAND with an erase block size 4MB.

Woh! an SLC NAND with 4MB eraseblocks, that's surprizing.

> 
> First of all, mkfs.ubifs wouldn't support a LEB larger than UBIFS_MAX_LEB_SZ being (2*1024*1024).

This is the easy part ;). We already have patches to increase the limit in different ubi utils (IIRC, it's also needed in ubinize).

> 
> When I use ubiattach though, it seems to be quite happy, giving me that on a 128MB partition:
> 
> UBI device number 0, total 32 LEBs (133169152 bytes, 127.0 MiB), 
> available 8 LEBs (33292288 bytes, 31.8 MiB), LEB size 4161536 bytes
> (4.0 MiB)
> 
> Anyway, when I detach and reattach, it won't work. Details are in the bottom of that mail, but since I haven't thoroughly looked into it myself yet, this is only for those who care... I don't want to ask preliminary questions.
> 
> Anyway, my questions I guess then are:
> -) Are there any patches/branches that enable UBIFS to be run with a LEB of 4161536?
> -) Are there any known limitations with using a 4MiB PEB with UBI itself, or shall I assume for now that the problem is on my board/NAND driver regarding the above ubiattach errors upon the second attach?

I don't think so. We did several tests on different MLC NANDs exposing 4MB blocks, and the UBI/UBIFS layers were not complaining about that (we had other problems related to the MLC constraints).
What's your controller?

Regards,

Boris

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/

[-- Attachment #2: nandubi.tar.gz --]
[-- Type: application/x-gzip, Size: 26110 bytes --]

[-- Attachment #3: increase-LEB-PEB-max-size-to-4M.PATCH --]
[-- Type: application/octet-stream, Size: 826 bytes --]

diff --git a/include/libubi.h b/include/libubi.h
index 4d6a7ee..201f4c2 100644
--- a/include/libubi.h
+++ b/include/libubi.h
@@ -36,7 +36,7 @@ extern "C" {
 #define LIBUBI_UBI_VERSION 1
 
 /* Maximum physical eraseblock size in bytes */
-#define UBI_MAX_PEB_SZ (2*1024*1024)
+#define UBI_MAX_PEB_SZ (4*1024*1024)
 
 /* UBI library descriptor */
 typedef void * libubi_t;
diff --git a/ubifs-utils/mkfs.ubifs/ubifs.h b/ubifs-utils/mkfs.ubifs/ubifs.h
index 2f080a8..33e0818 100644
--- a/ubifs-utils/mkfs.ubifs/ubifs.h
+++ b/ubifs-utils/mkfs.ubifs/ubifs.h
@@ -26,7 +26,7 @@
 #define __UBIFS_H__
 
 /* Maximum logical eraseblock size in bytes */
-#define UBIFS_MAX_LEB_SZ (2*1024*1024)
+#define UBIFS_MAX_LEB_SZ (4*1024*1024)
 
 /* Minimum amount of data UBIFS writes to the flash */
 #define MIN_WRITE_SZ (UBIFS_DATA_NODE_SZ + 8)