Monday 14th November 2011

This month's meetings

WorkingGroups/ToolChain/Meetings
<< <  2011 / 11 >  >>
Mon Tue Wed Thu Fri Sat Sun
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28 29 30        

Attendees

Agenda

  • Review action items from last meeting
  • PR51068: trunk fails to bootstrap
  • trunk is in stage 3
    • Pre-approval?
    • Will anyone review them?
  • QEMU KVM for this quarter
  • Across validation Panda benchmark results

Action Items from this Meeting

  • None

Minutes

PR51068:

  • Ulrich sees similar things on SPU
  • Seems that there was a rush before stage 1
  • Things should improve...
  • Is a failure in stage 1, should occur on a cross compiler?
  • Ramana will have a look into it
  • Run one of the regression hunter scripts?
  • Latency is high so many builds is better, but a cross applies here

trunk in stage 3:

  • Last time was:
    • Do fix bugs towards 4.7
    • Send for pre-approval
    • Backport early to our 4.6
  • Don't want to go 'dark' for three months
  • Send for comments/RFC?
  • Will anyone respond?
  • Didn't get any feedback last time even on the ARM specific parts...
  • Some way of making RFCs easier to pick up?
  • Any way of making our job easier when stage 1 opens again?
    • Nothing to hand

Across validation Panda benchmark results:

  • All of the variations is in denbench
  • Kernels are a locked down 2.6.35
  • Probably not the random timing bug of 873453

  • Revital has also seen variations when running DENBench
  • Ones that are completly CPU bound have very low variance

QEMU KVM for this quarter:

  • Could skip TCG, but then it's harder to test outside Fast Model
  • Don't need TCG when running in KVM mode
  • A15 TRM is now out on infocentre
  • ARMARM that integrates the different sub manuals isn't out yet but isn't a blocker
  • Want Dave to finish what he's currently working on
  • ...so the work can't be on the critical path
  • Current state of TCG on ARM?
    • Three bugs that Peter knows of
    • makecontext(), overlap of code gen buffer and libc heap, and x86 reset before grub
    • Might be able to ignore the x86 bug but may bight in the future
  • Just implement ucontext.h? Check with mkedwards first
  • virtio?
    • Kernel support is going up, but there's no QEMU side
    • Needs a moderate amount of refactoring

WorkingGroups/ToolChain/Meetings/Archive/2011-11-14 (last modified 2013-08-30 11:47:58)