Re: [support-gang] the keyhandler.ppy mystery

2009-03-22 Thread quozl
I've reviewed the code in


and I cannot see a way to easily induce the symptom you report even with
mistaken changes done in an editor.

On Sun, Mar 22, 2009 at 09:58:19AM -0500, Aaron Konstam wrote:
 I was running build 801 and I opened in an editor. I did
 not mean to change it but maybe I mistakenly did.

Do you have a copy of the file after the change?

Which editor was used?

Were you doing this change in Terminal, or a text console, or via SSH?

Which user were you at the time of the edit ... olpc or root?

 Shortly after that the functioning of my XO keyboard slowly
 deteriorated. First after typing 2 or 3 characters it would freeze up.

Define freeze up?  At the time you observed the freeze, what actions
were you taking and what was the result, e.g. pressing keys and they
were not echoed, or moving finger on touchpad and the cursor did not
move, or the lights went out, or an SSH session via the wireless hung?

 Or it would type nonsense characters.

What made them nonsense?

 I installed the previous build 767
 by holding down the O game key at boot but things did not improve.

This would have used the alternate in the other /versions
pristine tree.  One that you had not edited.  Did you confirm after boot
that the build number was 767?  (Asking this to exclude the possibility
that the alternate build was no longer available or the firmware did not
capture the O game key from you).

Did you attempt a full power down, remove battery and external AC, wait
30 seconds, and restart?  (Asking this to exclude certain problems I
know of).

The slow deterioration you report is hard to relate to accidental edits.
You edits would have had to accidentally add some form of data storage
about the progression of the deterioration.

 The Home, Mesh, etc keys did not function.

This implies that the change prevented the module from working.  Those
keys are handled by this file.  Reference: _actions_table.

 But at this point I could
 synchronize the cursor by holding down the four corner keys.

This is handled by the embedded controller (EC) and has nothing to do
with the Sugar shell keyhandler.  One of the corner keys is the frame.
If the frame key did not work in Sugar, but the frame key worked for
touchpad recalibration, then this matches the possible cause well.

 erased the current line inn the terminal. does not intercept Control/U, so that's good.  Control/U
is intended to erase the current line in the Terminal.

 esc worked for a while.

keyhandler does intercept Alt/Esc for close_window.  It is as if the
change you mistakenly made hurt the key release function at the end of
the file, perhaps.

 Returning to 801 did not help and eventually all keys failed to work.

With olpc-update again?  Or by reboot?

 A usb keyboard worked without problem.

On plugging it in, this may have changed stored state.  Was it plugged
in prior to you finding that it worked without problem, or did you plug
it in at the time?

 A run of the self test showed
 that the keys were functional,

You mean the OpenFirmware self test?

 so it had to be a software problem.

If so, then I agree, but I'd also wonder if you tested in the text
console (Control/Alt/F2) whether the keyboard worked.  Doing so would
exclude Sugar and X11 leaving only the kernel and keyboard driver as
relevant, and would be a useful bisection.

 Finally a usb install that is described as erasing all XO data brought
 my XO back to life.


 Now this was a software problem. The only key related file I messed with
 was That seems to be the culprit and it seems that
 installing a new build under normal procedures does not change this

You mean with olpc-update? is unchanged from build 767 to build 801.

 Can anyone suggest another reason for my experiences?

Yes, a combination of other problems that you have managed to connect

(I had considered keyboard matrix lockup, which was a problem with B1
models that was fixed with B2.  But the matrix is vertical; if the F1
(zoom_mesh) key did not work, then 2 w s x keys would also not work.
You would have likely noticed that.  Cleared by full power removal for
30 seconds.  Might be induced by static discharge or electromagnetic

Devel mailing list

Re: [Olpc-france] extend the memory flash space with an SD card to build open80211s

2009-03-13 Thread quozl
On Fri, Mar 13, 2009 at 07:27:47PM +0100, Aime Vareille wrote:
 However I found not good
 because it puts the fedora kernel on the ubuntu which works with some
 discrepencies ; it takes also some time because of qemu and the boot is
 not optimal.

Fedora and Ubuntu use the same upstream source for the kernel.  What are
these discrepencies, and why aren't they fixed?

Devel mailing list

Re: debxo ohmd testers wanted

2009-03-09 Thread quozl
On Thu, Mar 05, 2009 at 04:49:37PM +0100, Holger Levsen wrote:
 Would you like to (co-)maintain the package in Debian? I'd be happy to
 sponsor uploads and/or co-maintain.

I guess so, but I don't know what that requires.

 Comments on files I looked at so far:
   Standards-Version should be upgraded to 3.8.0 (after checking)

I don't know how to check.

   long description is too short (you wrote more in the mail I'm replying 
 to :)

Fixed.  See patch attached.

   misses homepage: and vcs-pseudoheaders

Fixed homepage, don't understand vcs-pseudoheaders.

   misses copyright owner
   misses download location
   misses short GPL blurbs
   mentions GPL but links to GPL-2


   unneeded with current content (see above)

Replaced with different content.

   version number should rather be something like 0.1.1.git20090304-1

Left as is, until you join in.

   we should file an ITP bug, so we can close it here ;) (rather to tell 
   others about out intentions to put this software into Debian)

Don't know how to do that.

   unconditionally ending with exit 0 looks wrong

Okay.  107 of 1051 packages on my system
/var/lib/dpkg/info/*.postinst ended with exit 0.  Should the package
fail to install if update-rc.d fails?

   are those files needed at all? I mean, shouldnt debhelper/cdbs create 
   if they contain autogenerated code only?

Don't know.

 rules: is unused, and if, I'd prefer

Don't know.

 Do you have a vcs where you maintain this?

Created just now.

git clone

diff --git a/debian/README.Debian b/debian/README.Debian
index aa31ee0..1614f1f 100644
--- a/debian/README.Debian
+++ b/debian/README.Debian
@@ -1,7 +1,16 @@
-README for Debian-style packaging of ohmd
+README for Debian packaging of ohmd
 by James Cameron
-This package was built from source downloaded using
-git clone git+ssh://
+- receives and generates D-Bus events,
- -- James Cameron, Wed,  4 Mar 2009 15:22:10 +1100
+- maintains a connection to the X server using an .Xauthority file in
+  /home/olpc or /root ... without this the detection of keyboard and
+  mouse idle is not performed,
+- directly uses /sys entries specific to the OLPC XO-1,
+- with gnome-power-manager installed the user will see a power dialog
+  in response to a power button press, often after the XO has been
+  woken from suspend.
+ -- James Cameron, Mon,  9 Mar 2009 20:11:51 +1100
diff --git a/debian/changelog b/debian/changelog
index e49f532..5cbfc50 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -1,8 +1,13 @@
 ohmd (0.1.1-6.21.20090304git.quozl-1) unstable; urgency=low
   * start ohmd on install
+  * start ohmd after hald (fixes a segfault)
+  * add homepage
+  * fix copyright
+  * fix build dependencies for Lenny, tested in a pbuilder
+  * add a brief system description to README.Debian
- -- James Cameron  Fri, 06 Mar 2009 11:44:22 +1100
+ -- James Cameron  Mon, 09 Mar 2009 20:43:51 +1100
 ohmd (0.1.1-6.21.20090304git.quozl) unstable; urgency=low
diff --git a/debian/control b/debian/control
index 690f129..ea32096 100644
--- a/debian/control
+++ b/debian/control
@@ -2,11 +2,14 @@ Source: ohmd
 Maintainer: James Cameron
 Section: unknown
 Priority: optional
-Build-Depends: debhelper (= 4.2.32), gtk-doc-tools, libhal-dev
+Build-Depends: debhelper (= 4.2.32), cdbs, gtk-doc-tools, libhal-dev, libglib2.0-dev, libdbus-glib-1-dev, libgtk2.0-dev
 Standards-Version: 3.7.3
 Package: ohmd
 Architecture: any
 Depends: ${shlibs:Depends}
 Description: Open Hardware Manager, for OLPC XO-1 Laptops
- Provides suspend on lid close, and suspend on idle.
+ For OLPC XO-1 laptops, provides suspend on power button press,
+ suspend on lid close, display dimming on idle, and suspend on
+ extended idle.
diff --git a/debian/copyright b/debian/copyright
index a9a6903..237b7a3 100644
--- a/debian/copyright
+++ b/debian/copyright
@@ -1,4 +1,35 @@
-Copyright: GPL
+This package was debianised by James Cameron on
-On Debian and Ubuntu GNU/Linux systems, the complete text of the GNU General
-Public License can be found in `/usr/share/common-licenses/GPL-2'.
+The current Debian maintainer is YOUR NAME y...@email.address
+It was downloaded from: 
+git clone git+ssh://
+ * Copyright (C) 2007 Richard Hughes
+ *
+ * Licensed under the GNU Lesser General Public License Version 2
+ *
+ * This program

Re: XO memory size

2009-02-25 Thread quozl
On Mon, Feb 23, 2009 at 04:33:59PM -0800, Derek Zhou wrote:
 Another thing is X drawing is very slow; however if I add:
 Option FBSize 8388608
 to xorg.conf, it becomes visibly faster.

I've tried to reproduce this, and failed.  Please give me a copy of your
xorg.conf file.

What I did was install debxo 0.4 gnome.img, then investigate /etc/X11,
only to find that xorg.conf is no longer there in that version.  So I
don't know how you have one, and I don't know what you have in it.

I tried placing just the Option line you specified in an empty
xorg.conf, but X would not start, complaining of syntax error in the

Then I copied an xorg.conf from rsync://, and
looked through it.

Then I installed the Debian x11-apps package, and did some performance
timings, using time x11perf -time 1 -repeat 1 -all, with different
configurations, to try to reproduce your observation;

1.  debxo 0.4 standard gnome configuration, without xorg.conf file, the
test took 29m 43s,

2.  debxo 0.4 standard gnome configuration, with xorg.conf file as is
from build-ubuntu, the test took 29m 45s,

3.  debxo 0.4 standard gnome configuration, with the above xorg.conf
file, with your Option line added to the Driver section, the test took
29m 47s.

 Why is limiting the video ram to half the size make it faster? 

It doesn't.

Devel mailing list

Re: OS/X11 support for XO-1 hardware?

2009-02-25 Thread quozl
On Wed, Feb 25, 2009 at 10:47:07PM -0800, Hal Murray wrote:
 Is there anything I can do on my XO running terminal so that when I ssh to 
 another system and run a text based X program the fonts will come out useable 
 without a magnifying glass?

Not that I know of.  The X server dimensions and resolution are properly
conveyed through the X connection (test: ssh -X u...@host xdpyinfo), so
font selection by toolkits such as GTK+ and KDE appears to work fine
from a debxo 0.5 system ... konqueror and galeon used comfortably large
fonts in my tests just now, to a Debian Lenny host.  xterm without flags
yields bitmap fonts that are very tiny, as they should be.  xterm -fa
DejaVu LGC Sans Mono yields a 52 by 17 character cell area.

 Is there anything like an environment variable that says scale all
 fonts by N?

No.  The best I know of is to lie about the dimensions and resolution,
on the X server as a whole.

If the target system is in your control, ensure you have set toolkit
font sizes after verifying the dimensions and resolution reported by
xdpyinfo.  An error there can cause this symptom, because the toolkits
are adopting your erroneous choices.

Devel mailing list

Re: XO memory size

2009-02-23 Thread quozl
Sounds interesting.  Which version of debxo?

Show us the output of /proc/meminfo.

 Another thing is X drawing is very slow; however if I add:
 Option FBSize 8388608
 to xorg.conf, it becomes visibly faster. Why is limiting the video ram to 
 half the size make it faster? 

Good question, I'll try that too on debxo.

Devel mailing list

Re: On optimizing Theora

2009-02-19 Thread quozl
On Fri, Feb 20, 2009 at 12:28:42AM -0500, Benjamin M. Schwartz wrote:
 GCC 4.3 evidently does not do a very good job of optimizing for geode.

What percentage of CPU time was spent in libtheora?

Devel mailing list

Re: new mailing list for other distros?

2009-02-18 Thread quozl
On Wed, Feb 18, 2009 at 04:14:05PM +0800, Carlos Nazareno wrote:
  I will have my two XO's there, I will try to have them running different
  flavors of DebXO (from USB sticks if nothing else, but quite possibly from
  the NAND), since the future direction is to have them run relativly
  standard distros, having examples of the different distros would be nice.
 Would it be good to create a new separate mailman developer list for
 non XO OS (Fedora + Sugar) other linux distro ports? (debxo, ubuntu
 xo, fedora xo etc) And then keep devel for main XO OS to avoid

I disagree.  There is no clutter now, and concentrating all the XO
hardware related discussions here is very valuable.  Splitting by
distribution would halt collaboration.

 Off-topic, $250 a pop would be a sweet spot if that provides enough
 profit to fund OLPC considering that the $250-$300 niche is already
 filled with models that provide way more horsepower than the XO-1
 (yes, we can say that's not XO's objective niche!, but consumers
 will still look at cpu,ramstorage whatever anyone says).

In summary, you'd like to see a lower price.  I agree, but I don't see
it happening soon, because although an XO is a fantastic hacker tool,
we're really doing it for the kids who aren't yet hackers.  The other
products out there that have multiple other purposes, and ship in
quantities far greater than the XO, will have a lower price until the XO
ships in similar quantities.

Devel mailing list

Re: Price point plus sales to individuals

2009-02-18 Thread quozl
On Wed, Feb 18, 2009 at 04:37:34PM -1000, Mitch Bradley wrote:
 Okay, so those of you who are keen on there being a way for individuals 
 to buy XOs at $2xx dollars should place a volume order, set up a web 
 site, and start raking in the dough.


(I'm not volunteering to do that, but the point is well made, speaking
about something and not doing it is wasteful.)

Devel mailing list

Re: power consumption after shutdown

2009-02-15 Thread quozl
On Sun, Feb 15, 2009 at 12:36:36AM -0500, Mikus Grinbergs wrote:
 My apologies for not being clear.  I'll try again:
 I want to feel that the battery that I just put into the XO I'm 
 walking out the door with is as charged up as it normally can be.
 1)  If that battery came from an XO that was plugged into the AC
  24/7 for the past week -- I *think* it is fully charged up.

Yes, 95% likely.

 2)  If that battery came from an XO that had been shut down for 48
  hours; but then that XO had been plugged into the AC until the
  'power' light went green -- I *think* it is well charged up.

Yes, 95% likely.

 3)  If that battery had been sitting on the shelf (not in an XO) for
  a month -- what did I need to do to make it topped off ?

Discharge it to 90% capacity, then charge it.

(You can do this manually in OFW using the watch-battery command ... or
you can periodically attach the AC adaptor until you get an orange LED
instead of green.)

 Sorry - I did not mean that I cared exactly how charged up it was 
 -- what I want to know is how to ensure to myself, without going to 
 extremes, that the battery was well charged up.   [Is 94% reported 
 by the pop-up for an on-the-shelf battery believable, or is the 
 'true' value half that?  And if do I see 94%, do I need to worry 
 that this is a battery that requires some more charging?]

Given a battery with an unknown history, it is not practical to predict
the state of charge.

 I was under the impression that when the battery charge went 
 somewhere below about 97%, the XO would charge it some more.
 So I was surprised to have it stay at 94%.   [If 94% being shown on 
 the pop-up is not low enough for charging some more, then how can 
 I initiate charging being performed by the XO ?]

Discharge it some more.  (There is no facility for boost charging ...
one reason is that this would shorten the life of the battery.)

  But when the software pop-up is between 90-96% (for a previously
  out-of-case battery), and despite the AC adapter being plugged in
  that value does not increase in an hour -- what ought I to do to
  find the state of charge ?
  No possible way except a full discharge while measuring the power
 Apologies for being unclear.  I'm not particularly interested in the 
 precision of the state of charge.

My explanation of this precision was so that you would understand why
you cannot get what you want.  I had tried explaining that you cannot
get what you want but you persisted, so I thought you were interested.

 What I am interested in is is 
 the battery as 'well charged up' as I can make it be by using the 
 XO's 'built-in' charger - I will want to get the most minutes of use 
 of my XO when carried to a place that does not have electricity ?

Discharge to 90% and then charge to completion.  Power off, then remove
the battery.  Travel as fast as possible to the place that does not have
electricity, do not delay by a month.  Insert the battery and power on.
You will have the most minutes of use.

 And if it is not 'well charged up', what do I need to do to make it 
   so?   In particular -- if it *were* 90% charged up when I took 
 it off the shelf, would I have to fully discharge it in order to 
 then charge it up (using the XO as the 'charger') closer to 97-99% ?

No.  Assuming a battery that has only been used in the XO, you need only
complete one charge cycle, from orange LED to green LED.  Since you
cannot cause this cycle unless the state of charge value was low enough,
you need to manipulate the state of charge value down, and the only way
to do this in the design is to discharge for a short time.  Usually
about five minutes, but it depends on how busy the XO is.

Devel mailing list

Re: [Server-devel] mkusbinstall fails

2009-02-15 Thread quozl
You got Input/Output error during cp read of /media/cdtmp.Rf6276, which
I guess is the loopback mounted ISO 9660 file system image.  The most
common cause of this is truncation of the image, especially if the files
on which it occurs are near the end of the image.

checkisomd5 is missing.  Install it from package isomd5sum on Ubuntu.

There may be other tools missing.  Use script command to capture all the
output as text, so we can have another look.

Server-devel mailing list

Re: power consumption after shutdown

2009-02-14 Thread quozl
On Sat, Feb 14, 2009 at 01:56:22PM -0500, Mikus Grinbergs wrote:
 I have found the XO-1 batteries (mfg by BYD Company Ltd) to be very 
 stingy with energy loss.  When I put one back in after a month out 
 of the case, the XO software told me it was still 94% charged.

The displayed state of charge is stored in the chip in the battery, and
is maintained by the EC.  The battery does not change it's own state of
charge value.

I don't think that batteries outside the XO will have their state of
charge value changed by the chemical self-discharge process.  But once
the terminal voltage becomes significantly inconsistent with the state
of charge value, I imagine that the terminal voltage is trusted more.

(If one discharges an XO battery outside the XO using a home lighting
circuit, the displayed state of charge will be inconsistent.  Persisting
in this practice results in increasing inconsistency.  Ceasing the
practice results in decreasing inconsistency over several XO moderated
charge and discharge cycles.  The inconsistency results in forced
power-down before the state of charge would suggest, manifesting as my
XO stops too soon.)

 James Cameron mentioned measuring a current draw of 24mA.  I suspect 
 this was *not* on a thoroughly shut-down G1G1.

Wrong.  It was on a thoroughly shutdown mass production unit.  The unit
was shutdown using the Sugar shutdown option, then the DC cable and the
battery were removed, 30 seconds were allowed to elapse, and then the DC
cable was reinserted.  The measurement was after this reinsertion.

 In my experience, 
 after two days in a shut-down state the XO software shows the 
 battery charge % to be somewhere in the mid-90's.

You observe a difference between my measurements and yours.  I knew the
technique was inadequate, but I was providing it as a maximum.  I had
already said that I thought the external DC supply path to contain
additional losses.  Richard has explained that the difference was due to
the EC not stopped, because it knows it is on external DC supply.

Devel mailing list

Re: power consumption after shutdown

2009-02-14 Thread quozl
On Sat, Feb 14, 2009 at 08:26:13PM -0500, Mikus Grinbergs wrote:
 This gets more and more bizarre !!

No, it's just physics and chemistry, constrained by engineering.
Please, if you think it is bizarre, explain why you think so, and I'll
happily explain the physics that I know.

 What I am now interested in is understanding a correct state of 
 charge value after having my battery out of the XO for one month. 

Not possible without discharging the battery fully to determine the
state of charge it had.  The test is charge destructive; once it is
completed there is no usable charge in the battery.

 [When I insert that battery and plug in the AC adapter, the 'power' 
 light is green, and the software pop-up says 94%.  But I notice that 
 after an hour, the software pop-up *still* says 94%.]

I hope you aren't surprised.  No reason it should change.  Battery is
not being used.

 I presume that when I merely shut down the XO, and leave it 
 sitting for two days with the battery still in the case:  Then when 
 I again connect the XO to the AC adapter (and let it charge until 
 the 'power' light goes from yellow to green), I should be able to 
 trust that what the software pop-up says is accurate.

No.  It will be close to the true value most of the time, unless there
is damage.  By close I mean within 20% or so, 90% of the time, assuming
no external use of the battery.  I'm just giving you personal estimates
there, I don't have all the data.

 But when the software pop-up is between 90-96% (for a previously 
 out-of-case battery), and despite the AC adapter being plugged in 
 that value does not increase in an hour -- what ought I to do to 
 find the state of charge ?

No possible way except a full discharge while measuring the power

The state of charge value that you refer to is calculated.  It is an
estimate.  It often represents reality.  The calculation is done by the
EC based on a measurement of current flowing into or out of the battery,
since the battery was manufactured.  The measurement is over time, so it
is called accumulation.  Measurement is done by the EC using the battery
support chip in the battery pack.  The results are stored in the chip.
Measurement does not occur of either chemical self-discharge, or
external discharge by some other means.

Devel mailing list

Re: power consumption after shutdown

2009-02-13 Thread quozl
Long ago I did some early measurements of the EC and it was consuming
current while operating.

The operating current has varied slightly according to the firmware
version.  There have been improvements, but I've not measured it

The symptom you describe is quite normal ... the battery has discharged,
and the state of charge evaluated by the EC shows that charging would be
of benefit.  So it goes yellow.

If you wish to avoid this draining by the EC, and instead rely only on
draining by self-discharge of the battery pack, then remove the pack
from the XO.  Self-discharge rate has a dependency on storage
temperature as well.

If you wish to measure the EC current, a simple way to do it is to
remove the battery pack, and place a current measuring device in series
with the DC cable to the XO.  This gives you a maximum.  The actual
current is smaller, because the DC socket path to the EC has more losses
than the DC battery path.

I did this just now, on a unit running Q2E27, and another running Q2E30,
the current is 24mA at 12.9V powered from a large sealed lead acid
battery with nothing else attached.  Two days of this would be 1.15 amp
hours (Ah).  Three days would be 1.73 Ah.

Assuming 3.1 Ah OLPC CL1 Li-Fe battery, two days should be enough to
bring the state of charge down to about 63%, certainly time for a

Devel mailing list

Re: [Server-devel] rsync on xs-dev

2009-02-12 Thread quozl
On Thu, Feb 12, 2009 at 06:27:03PM -0800, Sameer Verma wrote:
 I was trying to rsync ISOs last night (updating an older release) but
 it didn't work...might help those who have limited bandwidth.

You can rsync over SSH, and you can use --partial to ensure that the
data you have got so far is not wasted for when you have to restart.

Bare rsync is merely slightly more efficient, since the data is not
encrypted and encapsulated by the SSH connection.

Server-devel mailing list

Re: 8.2.1 wireless testing results #2

2009-02-06 Thread quozl
Comparing staging-26 against 8.2-767, using a NetComm NB600W
(purchased 2007-12-13) configured for either open or WPA TKIP. 

Open works fine.  Connection is automatically reestablished on reboot.

WPA(TKIP): 8.2.0 reconnects on boot, staging-26 always fails to
reconnect on boot, bringing up the password request dialog, repeatedly.
On the fourth or fifth dialog retry, the icon for the access point
disappears from view and reappears a minute or two later.  When that
happens, clicking on it successfully connects.

Workaround: after a reboot, cancelling the password request dialog, then
using the Control Panel, Network, Discard network history, allows
immediate connection.

Also, resuming from suspend on 8.2.0 requires connecting again, but
there is no password prompt.  Resuming on staging-26 produces the
password prompt, and generates the same symptom as rebooting ...
repeated prompts solved by Discard.

Conclusion: the regressions reported by Daniel Drake are reproduced.

Devel mailing list

Re: 8.2.1 Thoughts

2009-02-05 Thread quozl
On Thu, Feb 05, 2009 at 06:53:44PM -0500, Michael Stone wrote:
   1) Daniel Drake discovered some annoying wifi regressions (#9235).
  We need to find root cause here, e.g. by bisecting the kernel patches 
  since 8.2.0 and testing each resulting kernel with both the new and the 

Briefly, how is this done?  Where can I get the current kernel source as
a bisectable repository, and is it as simple as reverting a patch, make,
scp the kernel to a unit, then reboot and attempt to reproduce?

Devel mailing list

Re: Treatise on Formatting FLASH Storage Devices

2009-02-04 Thread quozl
On Wed, Feb 04, 2009 at 12:40:38AM -1000, Mitch Bradley wrote:
 Read it and weep.


Fixed a couple of typos in the last section.

Also, re:

Conversely, if the layout is bad, every cluster write might split two
pages, forcing the FTL to perform four internal I/O operations instead
of one.

Is it therefore four times slower?

Devel mailing list

Re: looked for, but did not find, control knobs for mesh

2009-02-02 Thread quozl
On Mon, Feb 02, 2009 at 02:09:31AM -0500, Mikus Grinbergs wrote:
 To me, two kids under a tree is a very important scenario. 
 Although mesh fails on current Joyrides, I'm experimenting with 
 manual intervention (e.g., ifconfig) to get it going anyway.

Said manual intervention could be added as a button to Sugar, I guess.

 Aside from ifconfig up/down, what other control knobs exist for 
 starting/stopping radio communication using a 169.254.x.x address ?

None.  However, don't forget to investigate iwpriv and iwlist.

I've just done some of my early tests again on build 767 using three
XOs.  The minimum to get a manual mesh to establish is, on each XO:

0.  stop NetworkManager, disable it, and reboot,

service NetworkManager stop
chkconfig NetworkManager off

1.  configure the radio (via the eth prefix),

iwconfig eth0 mode ad-hoc essid channel 6

2.  configure the network interface (via the msh prefix),

ifconfig msh0

3.  emit packets,


The result is that the forwarding table will begin to contain MAC
addresses.  This is one way to test if the mesh is operational, before
moving units apart.

iwpriv eth0 fwt_list

It isn't necessary to configure eth0 network interface with ifconfig if
you have no other need to do so.

Devel mailing list

Re: What keeps me going...

2009-01-12 Thread quozl
Devel mailing list

Re: administrative security

2009-01-11 Thread quozl
Physical access to the system gives full access, especially once the
developer key is obtained, to install applications that their teachers
or government had not considered.  The system considers the user to be
the authorisation authority.

If specific applications are not welcome in a deployment, they should be
checked for.

(At my primary school it became illegal to use green or red pens.  But
they could never stop us, we just bought them from shops or each other.)

Devel mailing list

Re: UDP broadcast from an XO

2009-01-01 Thread quozl
Visibility in Neighbourhood View is determined by access from the XO to
the Jabber server.  The Jabber server does not relay these UDP packets
for you.  Therefore visibility is not an indicator of ability to operate
over UDP.

A wireless router will relay the UDP packets.  The relay is being done
by the router itself, and so any XO not associated with the router may
not receive the UDP packets unless the packets are forwarded by the
router to whatever router the other XO is associated with.

Can you show me one of these UDP packets?

Devel mailing list

Re: XOs with no sound

2008-12-30 Thread quozl
How to fix no sound caused by operating system.

1.  obtain the root prompt, e.g. by starting the Terminal activity and
clicking on become root button,

2.  if you wish to find out in which way the settings have been
corrupted, copy the file /etc/asound.state before proceeding,

cp /etc/asound.state /home/olpc/asound.state.orig

3.  obtain a copy of /etc/asound.state from a working XO, or from a
reinstalled XO, and place it in /etc/asound.state on the failed XO,

4.  restore the settings from the file,

alsactl restore

5.  test that sound now works.

Note: there is a possibility that the build 767 ALSA saved state file is
in some other place.  I've not checked.  I found out where the file was
on Joyride 2612 using a command:

strace -e open alsactl restore

Devel mailing list

Re: Help runnning a script after Installing an activity from .xo

2008-12-17 Thread quozl
On Wed, Dec 17, 2008 at 12:31:31PM +0530, shivaprasad javali wrote:
 So I have to modify the /etc/sysconfig/modules/olpc-1.modules file so
 that I ask the XO to load the OSS module when it boots.

Why when it boots?  Why not when the activity starts?

import os
os.system(/bin/su -c 'modprobe snd-pcm-oss' root)

Don't remove the module on termination, in case another activity needs

Also, isn't there another way to cause the module to be loaded other
than explicit loading?

Devel mailing list

Re: No surprise on memory

2008-12-15 Thread quozl
On Mon, Dec 15, 2008 at 11:21:18PM -0500, Benjamin M. Schwartz wrote:
 I'm no expert, but making the system work well without overcommit would
 probably require extensive modifications to the python interpreter, the
 fd.o libraries (dbus, gstreamer, telepathy, etc.), gecko, and maybe even
 X.  All of these would need to allocate only as much memory as they need,
 and react appropriately when malloc returns NULL.  In other words, 'tain't
 gonna happen.

Couldn't we instrument malloc to report when it returns NULL (into an
area of memory we have helpfully set aside for the purpose) and then
report those events during testing, in order to find out and fix those
instances of overallocation?

Devel mailing list

Re: 2588 - Journal unusable

2008-12-11 Thread quozl
On Thu, Dec 11, 2008 at 04:24:32PM -0500, Eben Eliason wrote:
 1) Joyide is the development build stream.  It doesn't by any means
 promise stability, in general.
 2) It's early in the release cycle, so things are even more likely to
 break in fairly big ways.

It is as if joyride is being used as part of developers' edit, compile
and test sequence.  Wouldn't it be more appropriate to do this sequence
privately before a developer releases their changes to the public?

Here's how it would happen ... the developer would build a new RPM,
install it on a test unit, test that it works, *before* publishing the
RPM for the joyride to pick up.  If any failures occur due to
integration against other developer changes, then stronger dependencies
should be added.

I'm also waiting for the joyride builds to be usable before I consume
download data resources on them.

Devel mailing list

Re: Announcing the NANDblaster

2008-12-05 Thread quozl
On Fri, Dec 05, 2008 at 11:28:45AM +0100, Bert Freudenberg wrote:
 Wow, that's great! With that reflashing many laptops might even be  
 fun, just watching the blinkenlights ;)

The wireless LEDs are not enabled, you have to watch the screen instead.
(Not a complaint, merely an observation, as the blinkenlights don't).

James Cameronmailto:[EMAIL PROTECTED]
Devel mailing list

Re: Touch pads

2008-11-25 Thread quozl
On Tue, Nov 25, 2008 at 02:42:42PM +0545, Bryan Berry wrote:
 One important point, make sure you hit the Fn key last when you do the
 4-finger salute.


James Cameronmailto:[EMAIL PROTECTED]
Devel mailing list

Re: Packet loss during wireless scans (Testing needed)

2008-11-25 Thread quozl
Reproduced on 767 with 5.110.22.p18

Also observed packet loss when switching between text virtual consoles
using Alt/F1 and Alt/F2 ... but it was difficult to reproduce.

James Cameronmailto:[EMAIL PROTECTED]
Devel mailing list

Re: Simulating a lower resolution on the OLPC XO Laptop

2008-11-25 Thread quozl
On Tue, Nov 25, 2008 at 11:57:17AM +0100, Strider wrote:
 The problem with this is that the laptop isn't powerful enugh
 to handle fullscreen applications at this resolution.

All those I have tried have worked fine at this resolution.  Which
particular applications are you referring to?

I've tried; konqueror, galeon, gpredict, xclock, firefox, emacs, xterm,
inkscape, gimp, abiword, openoffice, wesnoth, and xastir.  I'm probably
not using an application you're using.  Which is it?  Perhaps there is a
design problem in the application.  Perhaps the application requires a
display bandwidth that exceeds what is available.

James Cameronmailto:[EMAIL PROTECTED]
Devel mailing list

Re: Test release of multicast NAND updater

2008-11-07 Thread quozl
Test 3, a different place, B4 acting as sender, five C2 as receiver,
channel 1, four C2 updated fine in one or two passes.

The C2 that I mentioned before is continuing to give trouble.  Same
symptom occurs, usually between 20 and 200 blocks after starting.  Have
tried varying position, channel, temperature, battery presence, remove
all power, USB load, screen brightness, but no change to symptom.  I
suspect something wrong with the unit, but I can't pin it down.

(This same unit when it arrived last week with a prior G1G1 OS build on
it wasn't able to activate wireless.  After I had upgraded it to 757 it
was able to.  That's just additional correlation.)

The unit works fine now after I flashed it with 757 from a USB key.

James Cameronmailto:[EMAIL PROTECTED]
Devel mailing list

Re: Test release of multicast NAND updater

2008-11-07 Thread quozl
Worked great.

Test 1, C2 as sender, channel 1, B4 as receiver, C2 copy-nand'd from
os757.img, broadcast update worked fine, only three lost packets, and
that was as I was handling the receiving unit.  On the next run through
the sequence numbers the missing packets were picked up, the NAND was
updated, and the B4 booted normally, and did first-boot sugar name

Test 2, same C2 as sender, but restarted, four C2 units as receiver, all
receivers started first, then sender started.  Three units captured the
whole stream in two passes, wrote NAND.  One unit stalled with this on

ok enand1
Boot device: /dropin-fs:mcastnand  Arguments: ether: 1 /nandflash
Waiting for server
Expecting 2389 blocks, 236511 total packets (99 ppb)
 need 24
0002 need 12

(where _ represents the block cursor)

It did not budge despite (a) the sender cycling through block 3 again,
and (b) the unit being moved right next to the sender.

No keyboard input did anything.

I power cycled, and did the enand1 again, and it worked fine.

Can Q2E22A be used on B2?  ;-)

General comment ... starting an enand1 while a broadcast is already in
progress ... seems to generate two three need responses.  Tried this
five times.  Feels like a settling issue.  You did mention it.  It makes
a single-pass problematic, but since the sender loops, no biggie.

James Cameronmailto:[EMAIL PROTECTED]
Devel mailing list

Re: Acoustic distance measurement test results

2007-11-18 Thread quozl
On Sun, Nov 18, 2007 at 06:01:22PM -0500, Benjamin M. Schwartz wrote:
 I also encountered some difficulty when sharing the activity over the
 mesh at distances greater than 25 meters.  This might be because the
 default mesh frequency (Channel 1) is the same as MIT's pervasive
 wireless network.  After switching to mesh channel 11, I had no
 difficulty sharing over distances up to 40 m.

Your observations are consistent with mine, subject to variables you
didn't mention.

How high off the ground were the laptops?  At ground height, with ears
up, the node to node transmission distance is very small.  That's why
school server antennas will be mounted high.

A football field normally has very little slope, so the terrain
obstruction is easily understood.

It might also be an area with a large radio noise background.  By
placing the laptops on the ground you may also have reduced the noise
from nearby transmitters.

With two laptops in a paddock or dirt road away from any city, I can
easily get to 300m on with the laptops at 1.5m above ground.

Nearby, I can reproduce 1.6km on a tar road with about 5% packet loss.

James Cameronmailto:[EMAIL PROTECTED]
Devel mailing list