Richard Henderson <richard.hender...@linaro.org> writes:
> On 6/19/23 19:34, Alex Bennée wrote: >> Richard Henderson <richard.hender...@linaro.org> writes: >> >>> On 6/12/23 04:50, Richard Henderson wrote: >>>> On 6/11/23 02:14, Alex Bennée wrote: >>>>> >>>>> Richard Henderson <richard.hender...@linaro.org> writes: >>>>> >>>>>> This is a perfectly natural occurrence for x86 "rep movb", >>>>>> where the "rep" prefix forms a counted loop of the one insn. >>>>>> >>>>>> During the tests/tcg/multiarch/memory test, this logging is >>>>>> triggered over 350000 times. Within the context of cross-i386-tci >>>>>> build, which is already slow by nature, the logging is sufficient >>>>>> to push the test into timeout. >>>>> >>>>> How does this get triggered because I added these: >>>>> >>>>> # non-inline runs will trigger the duplicate instruction heuristics in >>>>> libinsn.so >>>>> run-plugin-%-with-libinsn.so: >>>>> $(call run-test, $@, \ >>>>> $(QEMU) -monitor none -display none \ >>>>> -chardev file$(COMMA)path=$@.out$(COMMA)id=output \ >>>>> -plugin ../../plugin/libinsn.so$(COMMA)inline=on \ >>>>> -d plugin -D $*-with-libinsn.so.pout \ >>>>> $(QEMU_OPTS) $*) >>>>> >>>>> to prevent the callback versions from being called for x86. The original >>>>> intent of the check was to detect failures due to cpu_io_recompile, see >>>>> e025d799af (tests/plugin: expand insn test to detect duplicate >>>>> instructions) >>>> I have no idea how, but it's happening. >>>> >>>>>> Signed-off-by: Richard Henderson <richard.hender...@linaro.org> >>>>>> --- >>>>>> Irritatingly, it doesn't timeout locally, so I used staging to >>>>>> double-check: >>>>>> >>>>>> Fail: https://gitlab.com/qemu-project/qemu/-/jobs/4450754282#L5062 >>>>>> Pass: https://gitlab.com/qemu-project/qemu/-/jobs/4450927108 >>>> Note that in the pass case, we don't even log that the test ran. >>> >>> Any further thoughts on this? Otherwise I'll merge it to get rid of >>> the cross-i386-tci failure... >>> >>> >>> r~ >> I'm happy to drop the feature from the plugin but the clean-up also >> needs to be applied to the run-plugin-%-with-libinsn.so: rules for i386 >> and x86_64. > > Pardon? I don't know what you mean wrt changing the makefile. There are a couple of places that do: # non-inline runs will trigger the duplicate instruction heuristics in libinsn.so run-plugin-%-with-libinsn.so: $(call run-test, $@, $(QEMU) $(QEMU_OPTS) \ -plugin ../../plugin/libinsn.so$(COMMA)inline=on \ -d plugin -D $*-with-libinsn.so.pout $*) to prevent triggering the assert for x86 -- Alex Bennée Virtualisation Tech Lead @ Linaro
