Hi This is a continuation on top of my previous series:
tests: Make expliction defaults for tests Based-on: <20220902165126.1482-1-quint...@redhat.com> After the series is applied, the following combinations for x86_64-softmmu-devices.conf pass "make check" CONFIG_ISAPC=y CONFIG_I440FX=y CONFIG_Q35=y CONFIG_MICROVM=y CONFIG_ISAPC=n CONFIG_I440FX=y CONFIG_Q35=n CONFIG_MICROVM=n CONFIG_ISAPC=n CONFIG_I440FX=n CONFIG_Q35=y CONFIG_MICROVM=n ISAPC requires I440FX, so it can't be compiled alone. MICROVM is a mess, and you can't compile it right now without having CONFIG_PC defined, problem is at: /scratch/qemu/full/m64/../../../../mnt/code/qemu/full/hw/i386/acpi-common.c:39: multiple definition of `pc_madt_cpu_entry'; libcommon.fa.p/hw_acpi_acpi-x86-stub.c.o:/scratch/qemu/full/m64/../../../../mnt/code/qemu/full/hw/acpi/acpi-x86-stub.c:9: first defined here Basically the problem is that it expect CONFIG_PC to be defined. That happens if any other machine is defined, but if only MICROVM is defined it is not. So, now that I explained what I wanted to do, what it does: - make oem-fields consistent, only renames of functions - sort bios-tables-tests by machine type, so - it is easy to only run the things that are compiled in. Once this is done, we make tests depend on CONFIG_Q35 or CONFIG_I440FX in meson.build file. There are interesting cases, and are the tests are run with the default machine type (pc) but that work with either of them: qtests_i386_require_default = \ (config_all_devices.has_key('CONFIG_VGA') ? ['display-vga-test'] : []) + \ (config_host.has_key('CONFIG_POSIX') and config_all_devices.has_key('CONFIG_ACPI_ERST') ? ['erst-test'] : []) + \ (config_all_devices.has_key('CONFIG_SDHCI_PCI') ? ['fuzz-sdcard-test'] : []) + \ (config_host.has_key('CONFIG_LINUX') and config_all_devices.has_key('CONFIG_ISA_IPMI_BT') ? ['ipmi-bt-test'] : []) + \ (config_all_devices.has_key('CONFIG_ISA_IPMI_KCS') ? ['ipmi-kcs-test'] : []) + \ (config_all_devices.has_key('CONFIG_PVPANIC_ISA') ? ['pvpanic-test'] : []) + \ (config_all_devices.has_key('CONFIG_PVPANIC_PCI') ? ['pvpanic-pci-test'] : []) + \ (config_all_devices.has_key('CONFIG_RTL8139_PCI') ? ['rtl8139-test'] : []) + \ (config_all_devices.has_key('CONFIG_WDT_IB700') ? ['wdt_ib700-test'] : []) + \ (config_host.has_key('CONFIG_POSIX') ? ['test-filter-mirror'] : []) + \ (config_all_devices.has_key('CONFIG_USB_UHCI') ? ['usb-hcd-uhci-test'] : []) + \ (config_all_devices.has_key('CONFIG_USB_XHCI_NEC') ? ['usb-hcd-xhci-test'] : []) + \ ['boot-order-test', 'fw_cfg-test', 'migration-test', 'numa-test', 'rtc-test', 'test-filter-redirector', 'test-x86-cpuid-compat', 'vmgenid-test' ] This tests can be run with both machine types, but as we can't easily put -machine on them (several of them are supposed to run on other architectures), it is not "trivial to fix". I have a "hack" on my tree that gets the 1st machine available for this kind of tests and changed qtest_init() to qtest_init_first() that does exactly that. But I am not sure that is the way to go. Another way for me to fix it is just to change the Q35 machine to be the default for x86_64 when I400FX is not compiled in, but it has other kind of troubles. I started this wanting than all tests showed an explicit machine type, but there is nothing easier to be done for this multiarch tests. Notice that the problem already exist, and some tests do funny things to make sure that they get the "-machine pc". if (strcmp(arch, "i386") == 0 || strcmp(arch, "x86_64") == 0) { machine_addition = "-machine pc"; } QTestState *qtest = qtest_initf("%s -device virtio-mouse-pci,id=dev0", machine_addition); But this is not clear in my book. So, what I am doing this again. Number of files to compile and be able to run "make check" with minimal configurations (for x86_64-softmmu): - everything under the sun: 2757 files - everything that you can disable with configure: 1838 files - same than previous but only CONFIG_Q35 and this patches: 1759 files - on my tree with other multiple hacks: around 1500 files compiled. My goal would be that compiling with --without-default-devices, make check pass, but even after my changes, we get: Summary of Failures: 2/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/migration-test ERROR 0.16s killed by signal 6 SIGABRT 3/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/bios-tables-test ERROR 0.17s killed by signal 6 SIGABRT 4/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/test-hmp ERROR 0.19s killed by signal 6 SIGABRT 1/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/qom-test ERROR 0.33s killed by signal 6 SIGABRT 81/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/lpc-ich9-test ERROR 0.14s killed by signal 6 SIGABRT 83/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/q35-test ERROR 0.16s killed by signal 6 SIGABRT 85/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/tco-test ERROR 0.16s killed by signal 6 SIGABRT 86/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/device-plug-test ERROR 0.14s killed by signal 6 SIGABRT 88/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/drive_del-test ERROR 0.21s killed by signal 6 SIGABRT 89/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/fdc-test ERROR 0.21s killed by signal 6 SIGABRT 91/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/ide-test ERROR 0.13s killed by signal 6 SIGABRT 92/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/display-vga-test ERROR 0.14s killed by signal 6 SIGABRT 90/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/i440fx-test ERROR 0.33s killed by signal 6 SIGABRT 93/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/erst-test ERROR 0.15s killed by signal 6 SIGABRT 95/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/test-filter-mirror ERROR 0.14s killed by signal 6 SIGABRT 96/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/fw_cfg-test ERROR 0.13s killed by signal 6 SIGABRT 98/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/rtc-test ERROR 0.15s killed by signal 6 SIGABRT 99/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/test-filter-redirector ERROR 0.14s killed by signal 6 SIGABRT 100/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/vmgenid-test ERROR 0.18s killed by signal 6 SIGABRT 102/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/cpu-plug-test ERROR 0.18s killed by signal 6 SIGABRT 97/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/numa-test ERROR 0.46s killed by signal 6 SIGABRT 104/124 qemu:qtest+qtest-x86_64 / qtest-x86_64/test-x86-cpuid-compat ERROR 0.38s killed by signal 6 SIGABRT And we are not even starting to remove stuff that nobody is going to use in a modern guest (i.e. e1000e for instance is required). So, the question is, I am the only one that think this need to be improved, or should we left this as a hack on my devel tree. Please, comment. Thanks, Juan. Juan Quintela (8): bios-tables-test: Make oem-fields tests be consistent bios-tables-test: Sort all x86_64 tests by machine type bios-tables-test: Only run test for machine types compiled in tests: Only run intel-hda-tests if machine type is compiled in tests: sb16 has both pc and q35 tests tests: Make all tests that use q35 depend on it being compiled in tests: Unfold qtest_pci tests: Make all tests that depend on I440FX state that tests/qtest/bios-tables-test.c | 155 ++++++++++++++++++--------------- tests/qtest/fuzz-sb16-test.c | 10 ++- tests/qtest/intel-hda-test.c | 13 +-- tests/qtest/meson.build | 97 +++++++++++---------- 4 files changed, 154 insertions(+), 121 deletions(-) -- 2.37.2