[LEDE-DEV] Data corruption on kernel 4.9 with MIPS

[Rosen Penev]

Just thought I'd give a heads up given that the stable release is upon us.

I've tried making some noise about an issue affecting ramips with
kernel 4.9 where any device under /dev/sdX (maybe even /dev/mtdblock)
will start returning bad data after a while (how long seems dependent
on RAM size). Issue is here:
https://bugs.lede-project.org/index.php?do=details&task_id=1242&opened=21&openedsm=userid

It turns out, ar71xx is affected by this as well. I had an mvebu
router (Turris Omnia) running 4.9 with a USB hard drive connected that
I replaced with an Archer C7v4 after bricking the former. Hard drive
connected as well but this time, the drive got corrupt(mounts as
read-only) after several days uptime(I didn't check how long, less
than a week).

As for why this happens, current theory is some kind of DMA mapping
error in the kernel. The first suspect is this change:
https://github.com/lede-project/source/commit/668eb70157be59b17bb6da4a6de5d5e71a7c832b

But if memory serves me correctly, this was occurring before as well.

On the LEDE forums, people with XiaoMi MIR 3G routers were running
into this issue as well with several rolling back to 4.4 and reporting
good results. Link:
https://forum.lede-project.org/t/xiaomi-wifi-router-3g/5377/508