As the work to get X86CPU subclasses static properties defined is taking very long to get reviewed and we risk not getting it finished in 1.5, I am sending an alternative mechanism to allow libvirt to probe for CPU model information. It's interesting to note that this alternative solution is actually _easier_ to use for libvirt, because libvirt logic is already based on CPUID bit values, not "feature names", so it should be very easy to convert the current libvirt code that (incorrectly) query the host CPU directly for available features to instead start QEMU with "-cpu host" and use the "feature-words" array information to find out which CPU features are supported by the host. The property will have two main use cases: - Checking host capabilities, by checking the features of the "host" CPU model; - Checking which features are enabled on each CPU model. This series has two alternative implementations, so both approaches can get feedback: patch 3 has a non-qapi implementation, that simply builds the list and sets the struct fields manually. Patch 4 converts that implementation to a qapi- based implementation. (BTW, I really missed some documentation on the list-related visitor functions, when writing this code) Example output: $ ./QMP/qmp --path=/tmp/m qom-get --path=/machine/unattached/device[1] --property=feature-words item[0].cpuid-register: EDX item[0].cpuid-input-eax: 1 item[0].features: 126614521 item[1].cpuid-register: ECX item[1].cpuid-input-eax: 1 item[1].features: 2155880449 item[2].cpuid-register: EBX item[2].cpuid-input-eax: 7 item[2].features: 0 item[2].cpuid-input-ecx: 0 item[3].cpuid-register: EDX item[3].cpuid-input-eax: 2147483649 item[3].features: 563346425 item[4].cpuid-register: ECX item[4].cpuid-input-eax: 2147483649 item[4].features: 101 item[5].cpuid-register: EDX item[5].cpuid-input-eax: 3221225473 item[5].features: 0 item[6].cpuid-register: EAX item[6].cpuid-input-eax: 1073741825 item[6].features: 0 item[7].cpuid-register: EDX item[7].cpuid-input-eax: 2147483658 item[7].features: 0 I plan to send another RFC as followup soon, that will add: * "-cpu force" (or "nofilter", or something else; I haven't decided about the name, yet) flag, to disable feature filtering based on host capabilities. * "filtered-features" property, listing which features were filtered out because of missing host capabilities (to allow libvirt to implement "enforce" mode in a more reliable way). Eduardo Habkost (4): target-i386: Add ECX information to FeatureWordInfo target-i386: Move feature word array outside get_register_name_32() target-i386: Add "feature-words" property target-i386: Use qapi for x86_cpu_get_feature_words() .gitignore | 2 + Makefile | 11 +++++ Makefile.objs | 5 +++ cpu-qapi-schema.json | 31 ++++++++++++++ target-i386/Makefile.objs | 1 + target-i386/cpu-qapi-schema.json | 31 ++++++++++++++ target-i386/cpu.c | 92 +++++++++++++++++++++++++++++++++------- 7 files changed, 158 insertions(+), 15 deletions(-) create mode 100644 cpu-qapi-schema.json create mode 100644 target-i386/cpu-qapi-schema.json -- 1.8.1.4 -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list
