Source: qemu Severity: wishlist Hi.
I find myself sometimes running qemu-system-x86_64 and only minutes later that something is kind of fast, but not super fast. This usually is when I try complex desktop environments, and I always attributed this to use of software rasterizer in Mesa, and lack of virtgl. But in fact I was simply assuming that qemu is using kvm by default these days, even if -enable-kvm is not provided on command line. And that assumption was wrong. I wish qemu-system (and qemu-user) used kvm automatically. Try to open /dev/kvm and if it successes just assume kvm to be used, if it exist but can't be used, i.e. due to permissions emit a small warning about callback CPU emulation. If runnig as superuser, emit an error and exit. If -disable-kvm is used do not attempt to do that. If explicit -enable-kvm is used and kvm fails (no kvm or kvm support not built into binary, or missing kvm support on specific architecture), emit an error and exit. If -cpu XYZ or -machine xyz is used, or smp beyond what can be done on a machine, and kvm can't be used, emit a warning to stderr approprietly about using qemu emulation. If -disable-kvm is used, do not emit warning. If -enable-kvm is used, and kvm fails, emit an error and exit. To extend the wishlist, one could think about stuff like global config. /etc/qemu.conf could be also used for defaults, with default ones being cpu=host, enable-kvm, memory=xyz, smp=$(some_sane_number bigger than 1). And if a user wants they can modify it to add proper sound=ac97, [x86] net=nic,model=virtio, etc, to simply invokation of qemu by default. Configuration could have global, native and per-forgeing-target arch sections. Similalry I wish the default for qemu (when using with -kernel and -initrd parameters), and use virtio-net-pci, virtio-blk, virtio-scsi-pci (when passing block devices), virtio-serial-pci, virtio-rng-pci by default. This is because when using -kernel most of the time we are booting Linux, and most of the time we will have a virtio support on a guest side (if new virtio device is introduced only enable it by default after reasonable time ~2 years). Or a flag like -virtio-all, to enable all virtio types on a host instead of specificing a page long list of commands. Surely this can be done, but require a bit more research and care to not make existing users of qemu break. This would make use of manual flags unecassary in 95% of cases. I personally never had a problem with kvm, that made me need to disable it. I understand this might be going beyond what Debian can provide (that is why the bug is about only kvm autodetection), but I hope I am not the only one wanting this in qemu. Especially new users of qemu is going to appreciate it. Regards, Witold -- System Information: Debian Release: buster/sid APT prefers testing-debug APT policy: (500, 'testing-debug'), (500, 'testing') Architecture: amd64 (x86_64) Foreign Architectures: i386 Kernel: Linux 4.18.0-2-amd64 (SMP w/2 CPU cores) Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8), LANGUAGE=en_US:en (charmap=UTF-8) Shell: /bin/sh linked to /bin/dash Init: systemd (via /run/systemd/system) LSM: AppArmor: enabled
