Re: [PULL 11/14] tests/tcg: reduce the number of plugin tests combinations

[Gustavo Romero <gustavo.romero@linaro.org>]

Hi Stefan and Pierrick,

On 7/31/25 16:23, Pierrick Bouvier wrote:
> On 7/31/25 11:58 AM, Gustavo Romero wrote:
>> Hi,
>>
>> On 7/28/25 17:14, Gustavo Romero wrote:
>>> Hi,
>>>
>>> On 7/28/25 16:03, Pierrick Bouvier wrote:
>>>> Hi Alex,
>>>>
>>>> On 7/27/25 1:32 AM, Alex Bennée wrote:
>>>>> As our set of multiarch tests has grown the practice of running every
>>>>> plugin with every test is becoming unsustainable. If we switch to
>>>>> ensuring every test gets run with at least one plugin we can speed
>>>>> things up.
>>>>>
>>>>> Some plugins do need to be run with specific tests (for example the
>>>>> memory instrumentation test). We can handle this by manually adding
>>>>> them to EXTRA_RUNS. We also need to wrap rules in a CONFIG_PLUGIN test
>>>>> so we don't enable the runs when plugins are not enabled.
>>>>>
>>>>> Reviewed-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
>>>>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>>>>> Message-ID: <20250725154517.3523095-12-alex.bennee@linaro.org>
>>>>>
>>>>> diff --git a/tests/tcg/Makefile.target b/tests/tcg/Makefile.target
>>>>> index a12b15637ea..18afd5be194 100644
>>>>> --- a/tests/tcg/Makefile.target
>>>>> +++ b/tests/tcg/Makefile.target
>>>>> @@ -173,14 +173,25 @@ PLUGINS=$(filter-out $(DISABLE_PLUGINS), \
>>>>>    # We need to ensure expand the run-plugin-TEST-with-PLUGIN
>>>>>    # pre-requistes manually here as we can't use stems to handle it. We
>>>>>    # only expand MULTIARCH_TESTS which are common on most of our targets
>>>>> -# to avoid an exponential explosion as new tests are added. We also
>>>>> -# add some special helpers the run-plugin- rules can use below.
>>>>> +# and rotate the plugins so we don't grow too out of control as new
>>>>> +# tests are added. Plugins that need to run with a specific test
>>>>> +# should ensure they add their combination to EXTRA_RUNS.
>>>>>    ifneq ($(MULTIARCH_TESTS),)
>>>>> -$(foreach p,$(PLUGINS), \
>>>>> -    $(foreach t,$(MULTIARCH_TESTS),\
>>>>> -        $(eval run-plugin-$(t)-with-$(p): $t $p) \
>>>>> -        $(eval RUN_TESTS+=run-plugin-$(t)-with-$(p))))
>>>>> +
>>>>> +NUM_PLUGINS := $(words $(PLUGINS))
>>>>> +NUM_TESTS := $(words $(MULTIARCH_TESTS))
>>>>> +
>>>>> +define mod_plus_one
>>>>> +  $(shell $(PYTHON) -c "print( ($(1) % $(2)) + 1 )")
>>>>> +endef
>>>>> +
>>>>> +$(foreach _idx, $(shell seq 1 $(NUM_TESTS)), \
>>>>> +    $(eval _test := $(word $(_idx), $(MULTIARCH_TESTS))) \
>>>>> +    $(eval _plugin := $(word $(call mod_plus_one, $(_idx), $(NUM_PLUGINS)), $(PLUGINS))) \
>>>>> +    $(eval run-plugin-$(_test)-with-$(_plugin): $(_test) $(_plugin)) \
>>>>> +    $(eval RUN_TESTS+=run-plugin-$(_test)-with-$(_plugin)))
>>>>> +
>>>>>    endif # MULTIARCH_TESTS
>>>>>    endif # CONFIG_PLUGIN
>>>>> diff --git a/tests/tcg/multiarch/Makefile.target b/tests/tcg/multiarch/Makefile.target
>>>>> index bfdf7197a7b..38345ff8805 100644
>>>>> --- a/tests/tcg/multiarch/Makefile.target
>>>>> +++ b/tests/tcg/multiarch/Makefile.target
>>>>> @@ -189,6 +189,10 @@ run-plugin-semiconsole-with-%:
>>>>>    TESTS += semihosting semiconsole
>>>>>    endif
>>>>> +test-plugin-mem-access: CFLAGS+=-pthread -O0
>>>>> +test-plugin-mem-access: LDFLAGS+=-pthread -O0
>>>>> +
>>>>> +ifeq ($(CONFIG_PLUGIN),y)
>>>>>    # Test plugin memory access instrumentation
>>>>>    run-plugin-test-plugin-mem-access-with-libmem.so: \
>>>>>        PLUGIN_ARGS=$(COMMA)print-accesses=true
>>>>> @@ -197,8 +201,8 @@ run-plugin-test-plugin-mem-access-with-libmem.so: \
>>>>>        $(SRC_PATH)/tests/tcg/multiarch/check-plugin-output.sh \
>>>>>        $(QEMU) $<
>>>>> -test-plugin-mem-access: CFLAGS+=-pthread -O0
>>>>> -test-plugin-mem-access: LDFLAGS+=-pthread -O0
>>>>> +EXTRA_RUNS += run-plugin-test-plugin-mem-access-with-libmem.so
>>>>> +endif
>>>>>    # Update TESTS
>>>>>    TESTS += $(MULTIARCH_TESTS)
>>>>> diff --git a/tests/tcg/multiarch/system/Makefile.softmmu-target b/tests/tcg/multiarch/system/Makefile.softmmu-target
>>>>> index 5acf2700812..4171b4e6aa0 100644
>>>>> --- a/tests/tcg/multiarch/system/Makefile.softmmu-target
>>>>> +++ b/tests/tcg/multiarch/system/Makefile.softmmu-target
>>>>> @@ -71,8 +71,11 @@ endif
>>>>>    MULTIARCH_RUNS += run-gdbstub-memory run-gdbstub-interrupt \
>>>>>        run-gdbstub-untimely-packet run-gdbstub-registers
>>>>> +ifeq ($(CONFIG_PLUGIN),y)
>>>>>    # Test plugin memory access instrumentation
>>>>> -run-plugin-memory-with-libmem.so:         \
>>>>> -    PLUGIN_ARGS=$(COMMA)region-summary=true
>>>>> -run-plugin-memory-with-libmem.so:         \
>>>>> -    CHECK_PLUGIN_OUTPUT_COMMAND=$(MULTIARCH_SYSTEM_SRC)/validate-memory-counts.py $@.out
>>>>> +run-plugin-memory-with-libmem.so: memory libmem.so
>>>>> +run-plugin-memory-with-libmem.so: PLUGIN_ARGS=$(COMMA)region-summary=true
>>>>> +run-plugin-memory-with-libmem.so: CHECK_PLUGIN_OUTPUT_COMMAND=$(MULTIARCH_SYSTEM_SRC)/validate-memory-counts.py $@.out
>>>>> +
>>>>> +EXTRA_RUNS += run-plugin-memory-with-libmem.so
>>>>> +endif
>>>>
>>>> I'm not sure how it's related, but check-tcg on aarch64 host now fails [1] since this series was merged, and I suspect it may be related to this patch. I didn't spend time to reproduce and investigate it.
>>>>
>>>> [1] https://github.com/pbo-linaro/qemu-ci/actions/runs/16575679153/job/46879690693
>>>
>>> I was not able to reproduce it locally on 22.04, but in the CI indeed the test command is missing the "test-plugin-mem-access" binary at the end, just before redirection to .so.out, it should be:
>>>
>>> [...] -D test-plugin-mem-access-with-libmem.so.pout test-plugin-mem-access >  run-plugin-test-plugin-mem-access-with-libmem.so.out
>>>
>>> not:
>>>
>>> [...] -D test-plugin-mem-access-with-libmem.so.pout  >  run-plugin-test-plugin-mem-access-with-libmem.so.out
>>
>> The problem is that the new rules generated by a shuffled combination of tests and plugin
>> (the one using eval) sometimes does not include the rule for the test+plugin combination
>> passed via EXTRA_RUNs.
>>
>> So EXTRA_RUNS tests like run-plugin-test-plugin-mem-access-with-libmem.so might end up
>> with a proper rule if the test test-plugin-mem-access is combined with other plugin
>> randomly, since now a plugin is picked up based on a tests index _idx modulo with the
>> number of plugins.
>>
>> A possible fix is to generate correctly the rules for the tests with plugins passed
>> via EXTRA_RUNS and ideally, following the patch's mood, remove the test from running
>> if any other plugin (no shuffle with any other plugin).
>>
>> Of course, as Pierrick said, this patch could be a premature optimization. So maybe
>> it doesn´t justify adding more complexity to our Makefile (Makefiles are hard to debug,
>> with 'eval' tricks, it's _reallly_ hard, so the more we avoid it the better).
>>
>> That said, if we want to keep this commit, I've kicked off a test to fix it here:
>>
>> https://gitlab.com/gusbromero/qemu/-/pipelines/1959953122
>>
>> and sent the fix to the ML too. HTH.
>>
> 
> Thanks for taking a look at this Gustavo.
> 
> In the pipeline you shared, the build-some-softmmu fails with:
> Makefile:210: *** multiple target patterns.  Stop.
> make: *** [/builds/gusbromero/qemu/tests/Makefile.include:50: build-tcg-tests-x86_64-softmmu] Error 2
> make: *** Waiting for unfinished jobs....
> 
> Is that a new bug related to current patch?

v2 was busted. Please consider this fix, which passed the test pipeline:

https://gitlab.com/gusbromero/qemu/-/pipelines/1960296034

https://mail.gnu.org/archive/html/qemu-devel/2025-07/msg06926.html


Cheers,
Gustavo