On XO-1.5 with 11.3.0/11.3.1 -- hang during shutdown?

[Anish Mangal]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Thursday 14 June 2012 01:16 PM, James Cameron wrote:
> On Thu, Jun 14, 2012 at 12:03:39PM +0530, Anish Mangal wrote:
>> FYI, The testing efforts (with old v new SD cards are being
>> recorded at http://dev.laptop.org.au/issues/1323)
> 
> I've found microSD card performance can change slightly as a result
> of a reflash.  On #1323 it seems an fs-update was done prior to the
> test.
> 
> If you wish to keep analysing it to look at the differences
> between microSD cards, then:
> 
> - make the same number of shutdown tests for both the original
> microSD card and the different microSD card, so that the difference
> can be established statistically.
> 
> - restrict the testing to microSD cards from OLPC that we have 
> qualified.
> 

See http://dev.laptop.org.au/issues/1323#note-3 for the above two
points. We're testing with microSD cards that came with the OLPC
laptops (i.e.  OLPC approved/validated)

> - widen the testing to microSD cards from OLPC that have had little
> use.
> 
> - look for difference in behaviour with the microSD card written to
> in one laptop and used in another ... 'cause I'd hate to find that
> this was due to fs-update.
> 

Perhaps this could be one next step. Right now, we're doing (we have
one xo-1.5 and two microSD cards, one probably 'good', and the other
one 'bad', both OLPC approved)

* Insert bad microSD. Flash the new build (using fs-update)
* Test

* Insert good microSD. Flash the new build (using fs-update)
* Test

> I predict that this is a race condition during shutdown, which may 
> yield better to analysis with serial port attached.  The 11.3.x
> builds maintain a getty and shell on the serial port, if I recall
> correctly, and this may still be responsive at the time of the
> hang.  Using that shell it may be possible to find what processes
> are happening.  If that shell isn't available, try adding it.
> 

That's what we seem to have established in our initial debug (I think
it should be present somewhere in the thread history or the ticket).
Let me know if you can't find it.

> As to what is causing the different timing between different
> laptops, I predict that this is dependent on microSD card
> performance variation.  These cards contain a FLASH translation
> layer that processes SD commands and manages the remapping from
> virtual blocks to physical cells.  Their performance can vary.
> 
> I did consider the possibility of power cycling timing during 
> fs-update, but in the XO-1.5 units you have the microSD power is 
> managed by the embedded controller.  The SD card is power cycled
> by Open Firmware, but not the microSD card.
> 

The first thing we're trying to establish is that the problem happens
primarily due to a microSD card. Once we verify that it is the correct
direction, we'll go deeper into debug.

> Another thing you might try is run a microSD card in an SD card 
> adapter, see if there is a difference.
> 
> #1323 needs a pointer to your earlier work on #1033.  I found it
> in mail, but I shouldn't have had to.  ;-}
> 

Its already present in the 'Related Tickets' section on the same page.

- -- 
Anish

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJP2eV2AAoJEBoxUdDHDZVpx08H/jY46mT9vRI5Zn530BoeU4GJ
0LD2Ycf/0XWLjVPx7oS6cuySiOV2qWnNFUbh0iRVXOSHjafnm2Xsx2tMZX9t3CRe
JR3Yuzz1ymBPaYTK405+Kf5BadIHp6i0cJAG1jMtEvO7VvgQ8AxQCfiHOyjgxYe8
skI7w9xjSpIjB0nT76ePNaGG5FtLKQXMIixhcEbJt8pRoiBOLKYo2N6mEXgfniEU
k/UyvIbuMShdQzFJAIcQm8uEw8kZCHp4bQzjh+XMOL/H3eaL6JQT8K1bep9Ps+V6
YnC1D4ggFIm2uFUXy/aag0yaFepDoBvT3g0e66awPT/ZcT3p3+TwF8AlilwVOkw=
=GF4u
-----END PGP SIGNATURE-----