Progress:
* UM-2 [QEMU upstream maintainership]
-- identified another piece of the reset mechanics that can be
tractably cleaned up: I think we should be able to get
everything (effectively) moved over to the 3-phase reset system
-- sent out various for-8.0 patchsets that convert devices that
were using device_class_set_parent_reset(), since getting rid
of that function is step 1 of that cleanup
-- code review etc
-- PMM
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
- Confirmed that gdbserver doesn't send a new XML target description
over the wire when the vector length changes, so there's no impact of
these patches on the amount of remote protocol traffic, which was a
concern for GDB maintainers.
- Started doing some up code cleanup and regression tests to prepare for
submitting v2 of the patches.
# [GNU-796] Stabilise GDB testsuite results in the CI
- Committed Gerrit change request adding the new CI job¹ to run fast
subset of GDB tests, which also has stable results. The list of tests
comes from Sourceware's builder project.
--
Thiago
¹ https://ci.linaro.org/view/tcwg_gnu_native/job/tcwg_gnu_native_fast_check_g…
Progress:
* UM-2 [QEMU upstream maintainership]
-- investigated and thought a bit more about a race condition involving
display devices. I know think I understand the problem but the
best idea I could come up with for a solution is pretty painful...
(Still hoping for feedback on whether I've missed something.)
-- collated and sent out a pull request with various bug fixes for rc0
-- sent some more reset-cleanup patches (these will be for 8.0)
-- PMM
Hi guys,
I will attempt to join today, but we've been finding that connectivity is dependent on the
number of rain drops and/or hail in between here and the mast. So I'm not overly hopeful
for later this evening.
Status report:
- Fixed a failure in my x86 TARGET_TB_PCREL patches, which caused
non-booting of some images (in today's tcg PR).
- Revised the TCGv_i128 patch set (not yet posted).
- Lots of work on accel/tcg/{cputlb,user-exec}.c, to honor
atomicity requirements of FEAT_LSE2.
Next up: Adding qemu_ld/st helpers and patterns for i128.
r~
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
- Finished implementing Luis' idea of having a process-wide target
description in addition to the thread-specific one. Started checking
changes for regressions.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- Usual code review and patch handling work
- Cross-checked our boot.c code against the kernel's booting.rst,
and sent patch fixing a few EL3 config bits we weren't setting
- Sent out the minutes from QEMU Summit
* QEMU-471 [QEMU ARM v9.0 Baseline CPU for TCG]
- Wrote and sent out patches to implement the extra traps
required for FEAT_EVT
-- PMM
Hello,
# [GNU-796] Stabilise GDB testsuite results in the CI
- Reworked and committed Gerrit change request for jenkins-scripts
adding a new CI job definition in tcwg_gnu-build.sh.
- Also reworked and committed another one to the same script adding a
default manifest file if one isn't specified on the command line.
# [GNU-767] Support changing SVE vector length in remote debugging
- Implemented Luis' suggestion of assuming that if ptrace can't read the
SVE state, then SVE is disabled (which is what gdbserver does before
my patches). It turns out that the only situation where this happens,
gdbserver doesn't actually care about the target description.
- Started exploring/implementing Luis' idea of having a process-wide
target description in addition to the thread-specific one. As Luis
suggests, it may simplify the code.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- Usual code review and patch handling work
- Spent a couple of days investigating a long-standing intermittent
test case failure booting a kernel on the raspberry pi. This turned
out to be caused by two linked bugs:
- the test framework unhelpfully disconnects immediately from the
guest serial console when the test ends, rather than staying
connected until the QEMU process exits to capture all the output
- QEMU's socket chardev loses data when the remote end closes the
connection if the guest UART hasn't read all of it
So the test was failing because the guest sometimes only saw
the 'h' of the final 'halt' command the test sent...
- Investigated and fixed a regression caused by some of last week's
reset patches. (Turns out the patch to fix it had been sent back
in 2021 but we accidentally dropped it on the floor. Oops.)
thanks
-- PMM
Hello,
# [GNU-767] Support changing SVE vector length in remote debugging
- Continued incorporating suggestions from Luis' review of the patches.
Working on a few regressions I introduced in the process.
# [GNU-796] Stabilise GDB testsuite results in the CI
- Started addressing Maxim's comments to my change requests for
jenkins-scripts.
--
Thiago
