linux-mtd: ubiattach taking long time

linux-mtd: ubiattach taking long time

[Pintu Agarwal]

Hi All,

We are using arm64, quad-core, with Linux Kernel 5.15 with 1GB NAND.
We have A/B partitioning and around 9 ubi volumes.

During boot-up we observed that the ubiattach alone is taking around
1.2 secs (either user space or kernel space).
The snapshot of the log is shown below.

I have tried using fastmap as well, but still no difference.
Are there any other techniques to improve the ubiattach timing ?

Logs:
----------
Doing from initramfs:
{{{
[    6.949143][  T214] ubi0: attaching mtd54
[    8.023766][  T214] ubi0: scanning is finished
[    8.035901][  T214] ubi0: attached mtd54 (name "system", size 867 MiB)
[    8.042635][  T214] ubi0: PEB size: 262144 bytes (256 KiB), LEB
size: 253952 bytes
[    8.050423][  T214] ubi0: min./max. I/O unit sizes: 4096/4096,
sub-page size 4096
[    8.058134][  T214] ubi0: VID header offset: 4096 (aligned 4096),
data offset: 8192
[    8.066011][  T214] ubi0: good PEBs: 3470, bad PEBs: 0, corrupted PEBs: 0
[    8.072995][  T214] ubi0: user volume: 10, internal volumes: 1,
max. volumes count: 128
[    8.081225][  T214] ubi0: max/mean erase counter: 3/0, WL
threshold: 4096, image sequence number: 302691963
[    8.091242][  T214] ubi0: available PEBs: 0, total reserved PEBs:
3470, PEBs reserved for bad PEB handling: 80
[    8.101544][  T215] ubi0: background thread "ubi_bgt0d" started, PID 215
UBI device number 0, total 3470 LEBs (881213440 bytes, 840.3 MiB),
available 0 LEBs (0 bytes), LEB size 253952 bytes (248.0 KiB)
[    8.136907][  T227] block ubiblock0_0: created from ubi0:0(rootfs_a)
}}}

Doing from Kernel using cmdline: ubi.mtd=54,0,30 :
{{{
[    6.702817][    T1] ubi0: default fastmap pool size: 170
[    6.702822][    T1] ubi0: default fastmap WL pool size: 85
[    6.702826][    T1] ubi0: attaching mtd54
[…]
[    7.784955][    T1] ubi0: scanning is finished
[    7.797135][    T1] ubi0: attached mtd54 (name "system", size 867 MiB)
}}}


Thanks,
Pintu

Re: linux-mtd: ubiattach taking long time

[Richard Weinberger]

Pintu,

----- Ursprüngliche Mail -----
> Von: "Pintu Agarwal" <pintu.ping@gmail.com>
> I have tried using fastmap as well, but still no difference.
> Are there any other techniques to improve the ubiattach timing ?
> 
> Logs:
> ----------
> Doing from initramfs:
> {{{
> [    6.949143][  T214] ubi0: attaching mtd54
> [    8.023766][  T214] ubi0: scanning is finished

No fastmap attach here.

Make sure to set fm_autoconvert:
http://www.linux-mtd.infradead.org/doc/ubi.html#L_fastmap
If set, UBI create after a few writes a fastmap.

Speaking of other techniques, you can improve scanning time also by
tuning UBI for your NAND controller/setup.
E.g. transferring only the amount of bytes needed for an header.
Or reading without ECC and re-reading with ECC if the header CRC-check fails.

Thanks,
//richard

Re: linux-mtd: ubiattach taking long time

[Pintu Agarwal]

Dear Richard, Thank you so much for your reply. Please find some
comments inline.

On Mon, 25 Mar 2024 at 23:50, Richard Weinberger <richard@nod.at> wrote:
>
> Pintu,
>
> ----- Ursprüngliche Mail -----
> > Von: "Pintu Agarwal" <pintu.ping@gmail.com>
> > I have tried using fastmap as well, but still no difference.
> > Are there any other techniques to improve the ubiattach timing ?
> >
> > Logs:
> > ----------
> > Doing from initramfs:
> > {{{
> > [    6.949143][  T214] ubi0: attaching mtd54
> > [    8.023766][  T214] ubi0: scanning is finished
>
> No fastmap attach here.
>
> Make sure to set fm_autoconvert:
> http://www.linux-mtd.infradead.org/doc/ubi.html#L_fastmap
> If set, UBI create after a few writes a fastmap.
>
Yes, I have set fm_autoconvert=1 on cmdline and enabled fastmap on the
other logs.
{{{
[    0.000000][    T0] Kernel command line: ... fm_autoconvert=1
ubi.mtd=54,0,30 ...
[...]
[    6.702817][    T1] ubi0: default fastmap pool size: 170
[    6.702822][    T1] ubi0: default fastmap WL pool size: 85
[    6.702826][    T1] ubi0: attaching mtd54
[…]
[    7.784955][    T1] ubi0: scanning is finished
[    7.797135][    T1] ubi0: attached mtd54 (name "system", size 867 MiB)
}}}
Is there anything missing here ?
Can we increase the pool size ? Will it help to improve the timing ?

> Speaking of other techniques, you can improve scanning time also by
> tuning UBI for your NAND controller/setup.
> E.g. transferring only the amount of bytes needed for an header.
> Or reading without ECC and re-reading with ECC if the header CRC-check fails.
>
Sorry, I could not get this fully.
Is it possible to elaborate more with some reference ?
Do we have some special commands/parameters to do it if we use initramfs ?

Re: linux-mtd: ubiattach taking long time

[Richard Weinberger]

Pintu,

----- Ursprüngliche Mail -----
> Von: "Pintu Agarwal" <pintu.ping@gmail.com>

> Is there anything missing here ?
> Can we increase the pool size ? Will it help to improve the timing ?

As long you see "scanning is finished", fastmap was not used.
Usually a fastmap is created after ubidetach or writing more than pool size.
 
>> Speaking of other techniques, you can improve scanning time also by
>> tuning UBI for your NAND controller/setup.
>> E.g. transferring only the amount of bytes needed for an header.
>> Or reading without ECC and re-reading with ECC if the header CRC-check fails.
>>
> Sorry, I could not get this fully.
> Is it possible to elaborate more with some reference ?
> Do we have some special commands/parameters to do it if we use initramfs ?

You need to touch code for that. These are highly specific adoptions.

Thanks,
//richard