Progress: (short week, 3 days)
* UM-2 [QEMU upstream maintainership]
- More code review, as softfreeze is now quite close. I think we've
finally got there with the SME patchset (the remaining problems
with v5 were very minor)
* QEMU-422 [QEMU Arm Neoverse V1 vCPU for TCG]
- QEMU-315 OS Lock/DoubleLock work now upstream
-- PMM
Hello,
# Linaro CI's GDB testsuite results
* Implemented new version of change in the Abe shell script to correctly
set auto-load-safe-path so that GDB can load libthread_db.so.1. Since
GDB's configure script needs to be passed an argument containing
literal dollar signs, use '@@' instead of '$' to avoid the shell
misinterpreting it and substitute at the point of use. This version
worked and was merged to the Abe repo.
* Implemented and merged change in the Abe shell script to set variables
pointing to language-specific target tools only if they are present in
the path. This fixes Ada test failures in the GDB testsuite.
* Implemented and merged improvement to our tcwg_gnu-build.sh script to
allow specifying on the command line the Abe repo and branch to use.
# [GNU-767] Support changing SVE vector length in remote debugging
* Analysed and fixed a couple more regressions in my branch.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- Code review (misc, plus another round on the SME patchset)
- KVM Forum programme committee work
* QEMU-422 [QEMU Arm Neoverse V1 vCPU for TCG]
- Wrote and sent patches that deal with the last of the "small and
uninteresting" dependencies of the V1 support epic (this one covers
Feat_DoubleLock, which lets guest OSes suppress debug events as
part of their power-down-a-CPU sequence)
-- PMM
Hi guys. I just realized that today's meeting will be right in the middle of dinner
plans. I'll organize better next week. Anyway, here's my status:
- Fourth SME patch set -- parts merged, and some bugs fixed.
- Base/arm-compat portion of semihosting rewrite now upstream.
Target-specific bits for m68k, mips, nios2, xtensa outstanding.
- Almost done with FEAT_HAFDBS (hardware access/dirty updates).
Big changes to ptw.c to make that work: 50+ patches.
The reorg should make FEAT_RME easier though.
r~
Hello,
I just noticed that I didn't send a report for week #24. Sorry about
that. For that reason, this report covers two weeks.
# Linaro CI's GDB testsuite results
* Submitted upstream patch fixing gdb_compile regexp to account for our
GNATMAKE_FOR_TARGET value containing spaces. This should solve the Ada
test failures in our CI.
* Fixed the Abe shell script to correctly set auto-load-safe-path so
that GDB can load libthread_db.so.1. Unfortunately the CI detected a
problem with using the generated manifest to reproduce the toolchain
build, so I'm back to the drawing board on this one.
* Sent a Gerrit change request for the Abe script updating it to use
release 12's branch and tag for GDB and gdbserver, and another one
removing a workaround for a build problem fixed in GDB 9.
* Sent a Gerrit change request for tcwg-dev-build.sh fixing a couple of
issues I found when using it to build and test GDB.
# [GNU-767] Support changing SVE vector length in remote debugging
* Analysed and fixed a couple more regressions in my branch.
# Misc
* Attended a few online sessions from Linaro San Francisco Get Together.
* Sent Gerrit change request updating the TCWG script that generates our
SSH known_hosts file to include the Ed25519 host keys, which are now
preferred in Ubuntu 22.04.
* Was out one day due to “Corpus Christi” public holiday.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- First week back after holiday: lots of catchup
- Good progress with the code review queue, including getting
rid of a few things that had been lurking in it for too
long, and another pass through the latest SME series
- KVM Forum programme committee work
-- PMM
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
* Debugged and fixed a few issues in my code where gdbserver was having
trouble with setting the target description for new threads. Also
fixed regressions in a couple more testcases.
# Linaro CI's GDB testsuite results
* Luis commented on IRC that the GDB testsuite results from the Linaro CI
seemed a bit high with about 540 failures, so I set out to investigate.
The two biggest sources of failures are:
1. Ada tests were failing because they can't build their test
executables. This is expected because we don't have the Ada compiler
in the CI, but normally these testcases would just bail out as
unsupported. Found out that there was a regex in the GDB testsuite
that was (IMHO) too strict for our particular setup and fixed it.
2. All CTF tests are failing. The compiler in the CI seems to be
building test executables with CTF debug info, so we do seem to have a
problem there.
The CI's GDB is also not loading libthread_db.so because of a GDB
security setting, so I fixed that as well and expect some more tests
to pass now.
I still haven't submitted my fixes for 1. to upstream and for the
libthread_db.so.1 issue to our CI scripts because I'm still working
with jenkins-scripts and Abe to reproduce the CI environment to verify
the fixes.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- This week was almost entirely code review and similar upstream tasks
- managed to review all the preliminary parts of the SME patchset
- reviewed several other refactoring series from rth
- reviewed a big PS/2 keyboard emulation refactoring patchset
- prep for KVM Forum program committee session
* QEMU-422 [QEMU Arm Neoverse V1 vCPU for TCG]
- FEAT_RASv1p1 and FEAT_DoubleFault now upstream
-- PMM
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
* Debugged and fixed issues in my code where gdbserver was in some cases
trying to read inferior registers before aarch64_target was fully
initialised, and also while the inferior thread was running. This
improved reliability but there are still some testsuite regressions to
look into.
# Sourceware bug 28561 - [gdb/testsuite] Error due to not reading \r\n
at end of mi prompt
* Posted comment to the bug report summarising my current results.
* Tried to reproduce the bug using the upstream v5.3 kernel (since I was
able to see it with openSUSE Leap's v5.3 kernel), but failed. Still
trying to understand what that implies.
--
Thiago
Progress (covers two half-weeks):
* UM-2 [QEMU upstream maintainership]
- investigating a reported bug with semihosting syscalls over gdbstub
- reviewed a massive series from RTH refactoring our SVE code
in preparation for handling SME
* QEMU-422 [QEMU Arm Neoverse V1 vCPU for TCG]
- looking at exactly what FEAT_RASv1p1 and FEAT_DoubleFault require
of a minimal RAS implementation like QEMU's
- sent a patch implementing FEAT_RASv1p1 (a no-op for QEMU)
- sent a patch implementing FEAT_DoubleFault (turns out to be
simpler than I initially feared it might be)
-- PMM
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
* Found out that one issue I was having with gdbserver was caused by an
incompatibility it has with zsh (sourceware bug 26116).
* Fixed a gdbserver crash in my code where a register set was NULL in
the target description.
* Now working on an issue where the target description says SVE isn't
supported, but the regsets information says they are. This is caused
by my code sometimes moving the first ptrace call to get the inferior
registers to a time before it is properly stopped.
# Sourceware bug 28561 - [gdb/testsuite] Error due to not reading \r\n
at end of mi prompt
* Set up a few containers and KVM guests to attempt to reproduce this
bug on different environments. Reproduced it successfully on an
openSUSE Leap 15.3 guest, as well as Ubuntu 20.04 and 22.04 containers
in said guest. I wasn't able to reproduce on Ubuntu 20.04 nor 22.04
guests. Suspecting it's a kernel issue where the fix was backported to
the Ubuntu kernels.
--
Thiago
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
- Rebased my code on current master branch.
- Ran into issues with running the GDB testsuite against gdbserver, both
on my branch and on master. Currently investigating.
- Regarding the ptrace issues I'm running into, at Richard's suggestion
I checked the errno and it's ESRCH. This means that the inferior isn't
in a state where gdbserver can fetch its registers. So maybe there's
a race condition in gdbserver itself and it's trying to do that before
the inferior is stopped. Currently investigating.
# Sourceware bug 28561 - [gdb/testsuite] Error due to not reading \r\n
at end of mi prompt
* Luis mentioned this bug so I started investigating it. Tried
reproducing it on Ubuntu 20.04 and 22.04 but without success.
Also tried setting up an openSUSE Leap 15.3 KVM guest to try to
reproduce it there, but the distro's installer hangs.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- Joint Linaro/Arm meeting on the Realm Management Extension (RME).
Mostly this was "how the software stack is going to work"; some
useful discussion on how this intersects with QEMU, both as a
potential emulation platform for software development and also
as the VMM for a realm-using KVM virtual machine.
- Kicked off a discussion thread on getting rid of the last parts
of our "ad-hoc CI" this release cycle. The main thing that's
not yet handled by Gitlab CI is FreeBSD and NetBSD hosts.
We might be able to do this if we can find an x86 Linux machine
to act as a gitlab runner (it needs to be able to run guest
BSD VMs).
* QEMU-422 [QEMU Arm Neoverse V1 vCPU for TCG]
- Several small features have now made it upstream so those
subtasks have been closed out
- Looking at FEAT_RASv1p1 and FEAT_DoubleFault. The first of these
is easy because we only model the minimal "no error records"
RAS: I think we can simply bump the ID register fields to
indicate support. The second is a little bit more interesting
because the new SCR_EL3.EASE bit means we suddenly have a
source of physical SErrors (from synchronous external aborts)
when we previously could not have any...
-- PMM
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
* Continued working on the testsuite regressions introduced by my code.
- Found out that the crash I was investigating (which happens when
gdbserver thinks that the kernel doesn't support SVE and crashes)
only occurs with kernel v5.4 from Ubuntu 20.04 and not with
Ubuntu 22.04's v5.15. In the former version,
ptrace(PTRACE_GETREGSET, …, NT_ARM_SVE, …) often returns -1 (but not
always) and gdbserver interprets this as meaning that SVE isn't
supported. The newer kernel always returns successfully from this
call. gdbserver still acts weird so there's more digging to do, but
at least it doesn't crash.
# Misc
- Took one day of vacation.
--
Thiago