On buildbot duty
- Several buildbot failures to investigate along with one linux kernel
regression
- Didn't manage to get the libcxx buildbot failures resolved on time.
I think the community are still trying to fix it (Script needs to be
Python2/Python3
Started work on LLD support for Morello
- Getting familiar with the toolchain
- First target is sufficient support to statically link the Howdy
"hello world" equivalent.
- Made the first couple of local patches to implement the easiest set
of static relocations.
Some upstream LLD patch review.
Progress:
* VIRT-65 [QEMU upstream maintainership]
+ investigated the cause of a hang when using single-threaded TCG
+ sent RFC patchset proposing a rework of the ptimer API/implementation
which will fix LP:1777777. Unfortunately we'll need to update all
the devices using ptimers. I started in on that process of conversion,
which turns out to be a little more awkward than I had expected;
I ran into at least one corner case I hadn't thought about.
thanks
-- PMM
== Progress ==
* Out of office on Monday
* Catching up after Connect
* Minor buildbot fix
* IR SVE Reviews [LLVM-545]
- Another round on the size queries patch
* SVE IR fuzzer [LLVM-586]
- Waiting for the size queries patch to get merged so I can give it a spin
* Support Arran fat pointers in LLDB [LLVM-597]
- Built lldb
- Looked a bit through the docs
== Plan ==
* Play more with lldb, read more docs
* A bit of globalisel maintenance
o LLVM:
* Buildbots babysitting:
- Troubleshot ARMv8 selfhosted bots issue
* Machine Outliner:
- Experiment outlining with -mexecute-only
- Tried to find a testcase which exhibits issues with unwinding
without much success.
o Misc
* Various meetings and discussions.
Progress:
* VIRT-65 [QEMU upstream maintainership]
+ identified cause of LP:1844597 and sent patch fixing it (we broke
direct-booting kernels when the emulated system has an AArch32 EL3)
+ tracked down the cause of LP:1777777 to a race condition that's
inherent to the current design of QEMU's 'ptimer' countdown timers;
fixing this will require us to rethink when we call the callback
when a ptimer hits 0. Posted some ideas to mailing list, for feedback
on whether I missed a better way to rework them.
+ usual upstream maintainer stuff; managed one way or another to
get a few of the most obstinately sticky patchsets off my to-review list
thanks
-- PMM
== Progress ==
* GCC:
- FDPIC: Whole series now committed. Reverted one patch (aiming at
cortex-M, caused problems in ARM mode, investigating)
* GCC upstream validation:
- reported a couple of issues. Helped with testing.
- getting ready to add cortex-m33 validation with qemu
* Binutils:
- Non-contiguous memory regions support in the BFD linker: not started yet.
* misc:
- infra fixes / troubleshooting / reviews
== Next ==
Linaro Connect San Diego
== Progress ==
* Out of office on Friday
* Use newer perf for benchmarking [TCWG-1512]
- Switched benchmarking to use bionic containers, with perf 4.18.0;
seems to work
* Investigate running benchmarks in containers [TCWG-1513]
- Done
* Uploaded LLVM 9.0.0-rc5 and rc6
* SVE IR fuzzer [LLVM-586]
- Unblocked my work on this, need to give it a serious spin
== Plan ==
* Buildbot babysitting this week
* LLVM-586
* Try to build LLDB
[VIRT-327 # Richard's upstream QEMU work ]
Convert notdirty and rom handling to cputlb; lots of cleanup around
that area. Version 3 is the first version that worked; two previous
RFCs took quite a bit of effort to work out why they didn't work.
[GCC]
Committed the lse out-of-line patch set.
Posted patches for two follow-on bugs affecting aarch64-elf.
r~
Progress:
* VIRT-65 [QEMU upstream maintainership]
+ sent v2 of the semihosting v2 API implementation patchset
+ sent patch to fix a Makefile bug I introduced that broke in-tree builds
+ code review etc
thanks
-- PMM
[VIRT-327 # Richard's upstream QEMU work ]
Posted v2 of fixing tlb bswap.
Fixed a regression in arm SMLAL*
Reviewed v2 of risc-v vector extension (ugh).
Created pull for Sven's target/hppa fixes.
[GNU Cauldron]
* Project Ranger: New extensible representation for value ranges.
* Lightning talks, a new gdbserver.
- too short to have enough details to be useful.
- incomplete enough that no demo possible.
- maybe look back at this when it's more developed.
* Register allocation BOF
* ARM BOF
- MTE is alive! In that there are now RFC patches for glibc and gcc.
It's probably time to revive my MTE patch set, at least for system mode.
I spent quite a long time talking with Richard Earnshaw about the
current state of affairs. I'm hoping that we can get the kernel folk
to agree on some fundamentals of the userland ABI, even perhaps before
the final support is in the kernel, so that we can implement the
linux-user side, which would probably be most helpful for glibc+gcc work.
* GCC Steering Committee
* Rethinking GCC development process + continuous testing
* Truly interprocedural IPA-SRA
* Lightning talks,
- EmBench
- Combined elimination for -Os.
* RISC-V BOF
r~
