System Software Meeting Minutes - 2007-10-30

[Mitch Bradley]

Present:  Jim, Kim, Mitch, cjb, c_scott, dilinger, dwmw2, Jordan, 
Bernie, m_stone, ...

Topic - focused bug triage with particular emphasis on update.1 
delivery, then a discussion of vserver and containment issues.

1396 Update.2 Ebook reader should suspend on idleness
  Fixed - but needs to be turned on with the policy files
  (Touch a file in the filesystem to turn it on.)
  - CJB to turn it on, but only for >= C2 systems.

2679 Update.2 Suspend on idle
  - cjb to work on this after other ones that are lower-level
  - cjb to supply a list of those other ones
  
2765 Update.1 EC needs programmable delay to turn off DCON chip (XM-killjoy)
  - Reassigned to cjb to try doing it in OHM
  - We need to do tinderbox tests to determine the power impact
  - jg to check up on that tomorrow

1752 Update.2 USB wireless suspend/resume failure at setup phase
  - Closed

2353 Update.2 Hang (USB, DCON) going to suspend
  - Close as fixed  

2401 Update.2 rsmith Wakeup event is repeated continuously (EC and kernel)
  - The recipe given by MitchCharity has been reproduced by Mitch Bradley
  - Not a show-stopper because the system recovers if you press the 
button again
  - We don't want spin firmware for Update.1 unless a more serious 
problem surfaces

2699 Update.1  kernel usb storage does not write dirty blocks properly 
while unmounting
   - Important - reproducible, serious, probably fix-able
   - dilinger to work on with high priority

2833 Update.2 camera fails in os547
   - Works now because we reverted to an older gstreamer
   - Left open but deferred to Future to see what happens upstream with 
gstreamer and kernel

3470 rsmith No indication of low battery; laptop just shut down - No 
definite recipe for reproducing
3610 rsmith Mystery shutdown without red warning LED - No definite 
recipe for reproducing
4370 hardware No indication of low  battery, laptop just shut down
   - kim to collect all similar bugs

3579 Wireless hang after setconfig
   - Closed, fixed by hardware changes

3978 Journal getting blank
   - Might be fixed as of build 616
   - Expecting to land with datastore fixes

4032 Touchpad not that easy to use
   Longstanding bug - could be a kernel locking issue, could be 
resistive interference
   - Would be nice to be able to turn off the resistive area
   - jg to file new tickets for specific improvement opportunities

1407 Force recalibration when power situation changes
   - Move to update.1 and make it a blocker

4184 JFFS2 Dirent Anomaly
   - We have several workarounds for this, and a solid fix if we can 
remove vserver

4223 hardware Spinlock lockup on resume (LOOK) {This is the potential 
memory corruption thing}
   - Retest without vserver

4431 WLAN disappears after some number of suspend/resumes
   - Closed

4466 EC hang after large number of suspend/resumes (MP Start)
   - Low priority now

4217 Wireless/USB kernel panic on resume
   - Closed as duplicate

4401 rsmith Sugar battery capacity does not update (There is a patch for 
this; perhaps downgrade)
    - Reassign to kim for testing

4439 rsmith HAL doesn't understand our battery class anymore. (Patch 
available, also needs EC fix)
   Fix apparently available, needs testing
   - Can test latest stable kernel, but also awaiting new EC bits

4490 pdflush takes 80% cpu on 622
   - Dup of 4184  

Summary of Containment Issues:

    * With vserver, we have an existing system implementation, but the 
activities
    are only partially converted and have not been adequately tested.

    * The vserver kernel patch is unstable.  The systems team 
unanimously believes
    that it can't be fixed quickly.

    * The UID-based containment scheme has been designed but not yet 
implemented.
    The system-level implementation has no obvious roadblocks and thus 
might not
    take too long, but the activity conversion issue still remains.

    * We lack adequate resources for converting and testing activities.

    * The datastore needs modification to work across the isolation barrier.

    * c_scott and Ivan believe that, if we fail to ship filesystem 
isolation for
    Update.1, it won't be economical ever to do so (because it will break
    backwards compatibility).