Unstable bits CEWG project

Unstable bits CEWG project

[Ezequiel Garcia]

Hello all,

I'm considering to propose myself as a candidate for the CEWG funded
"Unstable bits" project [1].

I'd like to know if anyone has considered the posibility of adding
power-cut emulation to nandsim driver;
or if this is just a stupid idea.

Thanks,
Ezequiel.

[1] http://elinux.org/Start_working_on_the_%22unstable_bits%22_issue_to_make_UBIFS_more_robust

Re: Unstable bits CEWG project

[Ricard Wanderlof]

On Mon, 15 Oct 2012, Ezequiel Garcia wrote:

> I'm considering to propose myself as a candidate for the CEWG funded
> "Unstable bits" project [1].
>
> I'd like to know if anyone has considered the posibility of adding 
> power-cut emulation to nandsim driver; or if this is just a stupid idea.

I think the idea is good. My only doubts would be how well-defined the 
'unstable bits' behavior in real chips is, in order to be able to model it 
to the required accuracy (i.e. so that the resulting work in the software 
will be viable in a real-life situation ) in nandsim. Although most likely 
even with a fairly basic model quite an improvement could be made in the 
way it is handled.

/Ricard
-- 
Ricard Wolf Wanderlöf                           ricardw(at)axis.com
Axis Communications AB, Lund, Sweden            www.axis.com
Phone +46 46 272 2016                           Fax +46 46 13 61 30

Re: Unstable bits CEWG project

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 1263 bytes --]

On Mon, 2012-10-15 at 18:28 -0300, Ezequiel Garcia wrote:
> Hello all,
> 
> I'm considering to propose myself as a candidate for the CEWG funded
> "Unstable bits" project [1].
> 
> I'd like to know if anyone has considered the posibility of adding
> power-cut emulation to nandsim driver;
> or if this is just a stupid idea.

I'd suggest to add this in UBI instead.

1. nandsim is quite unreadable already and difficult to deal with
2. UBI is cleaner and all the I/O goes through functions in the io.c
   file.
3. By having it in UBI it is possible to force p-c emulation in
   interesting places, like when we are updating the volume table,
   or recovering from a failed write, or torturing a PEB, or writing
   EC/VID headers, or doing something fastmap-related.

   If you do this in nandsim, you will get 99% of power cuts when
   writing data, so you will mostly test this, and very rarely cover
   other places, unless you run for days.

   We passes through this when developing UBIFS, and this is why we
   put the power-cut emulation to UBIFS. I guess nowadays you can move
   it down to UBI.

Also, FYI, we have the integck test which supports emulated power-cut
testing for UBIFS.

-- 
Best Regards,
Artem Bityutskiy

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]