software team meeting notes march 20 2007

[Christopher Blizzard]

Software Team Meeting 03/20/2007

AI: Marcelo to get updated firmware available for builds
AI: Richard is working with wakeup on lan issue, should be done by thu
morning (which gpio you need to deal with)
AI: Richard needs to get a release of the firmware done w/ suspend by
thursday
AI: Jim will poke mitch to get everything in and then get richard to
release
AI: Marcelo will work with dilinger to get us a new stable kernel with
driver fixes

o progress on wireless
- wireless driver + firmware
- 992 - kernel oops from rmmod, fixed with rc3 mainline
- 1005 - ping mesh failure, fixed in firmware according to javier (972) 
- 1014 - low throughput in mesh mode - being looked at, even though it's
pretty good right now
  - marvel will keep looking at it internally but they are going to
close that bug?
  - want to lower the priority of it - what is the metric?
  - want the same speed as ad-hoc mode
  - comment in the middle of the entry - javier has a script
  - new release on the wiki - can we do some iperf testing against that?
- 972 - laptop leave ad-hoc mode after a while
  - can't reproduce with recent firmware, believe it's fixed
- 1019 - wifi channel setting fails
  - from luis has not been seen on recent firmware, but needs more
testing
  - think it's fixed with the most recent changes, made some fixes
around that area and haven't been able to reproduce
  - they feel it's safe to close
- 1102 - mesh routing/forwarding seizes after a few hours
  - hard to reproduce, haven't seen it after many hours of testing, feel
it might be fixed by another problem
- 1026 SCP failures across multi-hop paths
  - definitely fixed
  - problem was that buffer was too small between 3-4 address format (?)
  - mesh only problem
- some discussion of making sure that the mesh works for brazil later
with 35 nodes
- need more machines for marvel to test with
- need more feedback from the olpc development team if there are any
more inputs from the host/hardware side?  (trac #1060)
- wakeup on wireless lan: richard looked at schematics from quanta, they
have some changes they want to make - have a review for it tomorrow
(added AI for this)
  - gpio discussion (some can be used, some can't: see trac)
- someone from marvell/cozybit want to autonomous mode to the driver and
need to deal with power management, need to integrate with the olpc
folks?
  - richard will take that mail
  - marcelo is also important for power management/driver bits


o suspend/resume
- stable on the board
- need to do kernel integration testing
  - chris + jordan are working on issues
- official firmware release is probably B31
- fast boot + vsa-less is the future, we can cut over
- move to a "C" series for the firmware
- [ lots of details, can't keep up about specific areas ]
- jordan: suspend/resume is not stable with the kernel
  - need to figure out that everybody is following the same path
  - cut out an eng release of the firmware asap and make sure it's
checked into svn
  - need to start hitting serial and get into deep debug
  - everyone needs to be able to suspend the kernel and then can resume
  - need to start making sure our production path is where everyone is
testing
  - get everyone on the kernel because mitch thinks everything is great,
but everyone working on the kernel knows everything is broken
- richard: can get a release by thursday?
- olpc.rom is missing some files from svn
- debugger can't break when the system returns
  - resumes, returns then resets - don't know why
  - kernel is doing DMA on the way down?  can't tell
  - can't tell, though - goes in, doesn't come out
  - need a rom to do some testing
- when mitch gets the stuff in svn, richard can get a release right
after that

- Priority: LX firmware vs. suspend/resume?
  - lots of things are running, but might need to divert some for power
management?
  - need to parallelize the power management effort
  - need to have things done by mid-april
- only one person who can do the firmware - need to work on LX or power
management
- power management needs to take priority over LX at the moment
- rsmith is getting better with ofw, which helps - might be able to help
with low level stuff
- [ technical discussion about debugging suspend/breakpoints I couldn't
follow ]
- [ seriously, what is up with all this resistor talk? ]

o battery
- corruption from bios shadowing?
- ended up writing wrong things to the battery
- can fix with some forth code that rsmith wrote this weekend, has
recovered some batteries
- some batteries can't be recovered?  have to have faster hardware to
fix them (!)
  - can see super-fast on the scope
- cycled some batteries with 81, and things seem ok, but can't be sure
because it was hard to reproduce in the first place
- battery checking is now done 60 seconds after startup, enough time for
shadow
- double check will also help
- bottom line: some batteries have been lost for all time in the wild
without dedicated one wire hardware
- if batteries quit working after 81, we want to know about it
- some batteries can be saved by 81
- jim and rsmith will work on putting together an announcement of a
stable build

o tinderbox for power management
- no, not yet
- cjb hasn't had time yet
- battery lasts 3:10 even with the new firmware
  - X running, backlight on, wireless associated
  - not under full load

o action item list
- missing everything checked in for firmware
- cjb has done battery life testing
- dilinger kernel resume working on vsaless firmware?
  - git master does work
  - stable build does work
- power rail ECO is done
- richard to update tickets after change?  not done yet
  - richard will wrap it up, reset vs. host problem, work around works
so he hasn't spent cycles
  - not urgent, really
- mesh/avahi not working
  - crashing reports?  haven't seen them recently
  - only on first boot out of the box
  - may have to retest them, but haven't seen reports
  - need to get mixhalis to run 303 on his mesh boxes, or whatever is
stable
- wireless wakeup info
  - richard is working on it during
- no dcon breakout board
- cafe PCI cards are coming from marvell
  - asic pci cards
- lx board ID?
  - going to go with a nine-levels of EC checking/analog method - should
cover for a while