Timeout in denali.c on Micron nandflash (Altera SoC)

[Richard Weinberger]

Thorsten,

On Tue, Mar 7, 2017 at 2:32 PM, Thorsten Christiansson
<thorsten.christiansson at idquantique.com> wrote:
> Hi all,
>
> I'm using Linux on an Altera SoC (Arria V), on which I'm using UBIFS on a
> nandflash from Micron (MT29F8G08ADADAH4).  I have a 400Mb r/w partition on
> which I have a sqlite3-based database. We're running an application that
> reads/writes fairly small blocks. After running for about a week at moderate
> load, I get an error message, and the filesystem becomes read-only.
>
> The message I get is a timeout, originating in the denali.c driver.
> [11744.733748] timeout occurred, status = 0x0, mask = 0x4
> [11745.733685] timeout occurred, status = 0x0, mask = 0x120
>
> I can also reproduce the error much faster (in ~1h) using the GNU 'stress'
> command, writing/reading small files continuously.
>
> I'm using Linux 4.4, with some patches from Altera. I have compared the
> denali.c that I'm using with the current HEAD on github, and the differences
> appear to be only cosmetic.
>
> I have asked Altera for help, but their only response so far has been that
> they can reproduce the issue on their latest SoCs (it apparently appears on
> both Arria10 and CycloneV) with the same flash. (We have also tested with a
> Macronix MX66U51235FMI-10G, with the same results.)
>
> At first we used the FASTMAP feature of the UBIFS, but then we ran into this
> issue after only a couple of hours running at moderate load. When we
> disabled
> that, we thought the problem was gone, but it appears that it was only
> hiding,
> and now comes out to bite us after about a week.
>
>
> My questions are the following:
> - Are there any known issues with the denali driver that could cause this?

Well, 4.4. is not very fresh. Maybe it saw fixes in recent version.

> - Could it be an issue in the MTD/UBI/UBIFS layers?

Since the denali driver prints the timeouts I'd say the root of the
problem is there.

> - Are there any other parameters that can be tuned in order to alleviate the
>   problem?
>
> and of course
> - Have I missed something obvious? I'm pulling my hair here...

Can you please give MTD tests a try?

-- 
Thanks,
//richard