June 2015 - Devel - libvirt List Archives

[libvirt] [PATCH sandbox] Ensure libvirt-sandbox-init-common uses correct ld-linux
by Daniel P. Berrange 01 Jul '15

01 Jul '15

The libc.so library requires certainly functionality to be provided by the ld-linux.so library. The ld-linux.so is loaded by the kernel based on the PT_INTERP ELF section, and as such LD_LIBRARY_PATH has no effect. So, although libvirt-sandbox-init-{qemu,lxc} set LD_LIBRARY_PATH to force use of the libs from the host OS image, the common init program was stil using the ld-linux.so from the guest OS image. Sometimes this works, sometimes it breaks. When trying to run Fedora 19 as the root filesystem from a Fedora 22 host, it would break thus: /etc/libvirt-sandbox/scratch/.libs/libvirt-sandbox-init-common: relocation error: /etc/libvirt-sandbox/scratch/.libs/libc.so.6: symbol _dl_find_dso_for_object, version GLIBC_PRIVATE not defined in file ld-linux-x86-64.so.2 with link time reference To deal with this we must ensure that we always invoke the libvirt-sandbox-init-common program using the ld-linux that was provided by the host OS image. The sandbox builder is tweaked so that it always copies the host ld-linux.so into the libs scratch directory, and gives it a predictable name 'ld.so', since every architecture seems to have a different name. The libvirt-sandbox-init-{lxc,qemu} files are changed so that instead of exec'ing libvirt-sandbox-init-common directly, they will load it via the ld.so file. --- libvirt-sandbox/libvirt-sandbox-builder.c | 28 +++++++++++++++++++++++----- libvirt-sandbox/libvirt-sandbox-init-lxc.c | 1 + libvirt-sandbox/libvirt-sandbox-init-qemu.c | 1 + 3 files changed, 25 insertions(+), 5 deletions(-) diff --git a/libvirt-sandbox/libvirt-sandbox-builder.c b/libvirt-sandbox/libvirt-sandbox-builder.c index 2726868..e615606 100644 --- a/libvirt-sandbox/libvirt-sandbox-builder.c +++ b/libvirt-sandbox/libvirt-sandbox-builder.c @@ -252,10 +252,11 @@ GVirConnection *gvir_sandbox_builder_get_connection(GVirSandboxBuilder *builder) static gboolean gvir_sandbox_builder_copy_file(const char *path, const char *libsdir, + const char *newname, GError **error) { gchar *name = g_path_get_basename(path); - gchar *target = g_build_filename(libsdir, name, NULL); + gchar *target = g_build_filename(libsdir, newname ? newname : name, NULL); GFile *srcFile = g_file_new_for_path(path); GFile *tgtFile = g_file_new_for_path(target); gboolean result = FALSE; @@ -285,7 +286,7 @@ static gboolean gvir_sandbox_builder_copy_program(const char *program, const gchar *argv[] = {LDD_PATH, program, NULL}; gboolean result = FALSE; - if (!gvir_sandbox_builder_copy_file(program, dest, error)) + if (!gvir_sandbox_builder_copy_file(program, dest, NULL, error)) goto cleanup; @@ -301,14 +302,31 @@ static gboolean gvir_sandbox_builder_copy_program(const char *program, *tmp = '\0'; /* Search the line for the library path */ - start = strstr(line, " => "); + start = strstr(line, "/"); end = strstr(line, " ("); if (start && end) { - start = start + 4; + const gchar *newname = NULL; *end = '\0'; - if (!gvir_sandbox_builder_copy_file(start, dest, error)) + /* There are countless different naming schemes for + * the ld-linux.so library across architectures. Pretty + * much the only thing in common is they start with + * the two letters 'ld'. The LDD program prints it + * out differently too - it doesn't include " => " + * as this library is special - its actually a static + * linked executable not a library. + * + * To make life easier for libvirt-sandbox-init-{qemu,lxc} + * we just call the file 'ld.so' when we copy it into our + * scratch dir, no matter what it was called on the host. + */ + if (!strstr(line, " => ") && + strstr(start, "/ld")) { + newname = "ld.so"; + } + + if (!gvir_sandbox_builder_copy_file(start, dest, newname, error)) goto cleanup; } diff --git a/libvirt-sandbox/libvirt-sandbox-init-lxc.c b/libvirt-sandbox/libvirt-sandbox-init-lxc.c index e2fe7f0..d1e4a79 100644 --- a/libvirt-sandbox/libvirt-sandbox-init-lxc.c +++ b/libvirt-sandbox/libvirt-sandbox-init-lxc.c @@ -77,6 +77,7 @@ main(int argc, char **argv) args[narg++] = "1000"; } + args[narg++] = SANDBOXCONFIGDIR "/.libs/ld.so"; args[narg++] = SANDBOXCONFIGDIR "/.libs/libvirt-sandbox-init-common"; if (debug) args[narg++] = "-d"; diff --git a/libvirt-sandbox/libvirt-sandbox-init-qemu.c b/libvirt-sandbox/libvirt-sandbox-init-qemu.c index 09580da..cd6055a 100644 --- a/libvirt-sandbox/libvirt-sandbox-init-qemu.c +++ b/libvirt-sandbox/libvirt-sandbox-init-qemu.c @@ -422,6 +422,7 @@ main(int argc ATTR_UNUSED, char **argv ATTR_UNUSED) args[narg++] = "1000"; } + args[narg++] = SANDBOXCONFIGDIR "/.libs/ld.so"; args[narg++] = SANDBOXCONFIGDIR "/.libs/libvirt-sandbox-init-common"; if (debug) args[narg++] = "-d"; -- 2.4.3

2 1

[libvirt] [PATCH] qemu: Avoid using ".(null)" in UNIX socket path
by Jiri Denemark 01 Jul '15

01 Jul '15

The code which generates paths for UNIX socket blindly used target name without checking if it was set. Thus for the following device XML <channel type='unix'> <source mode='bind'/> <target type='virtio'/> </channel> we would generate "/var/lib/libvirt/qemu/channel/target/NAME.(null)" path which works but is not really correct. Let's not use the ".target_name" suffix at all if target name is not set. https://bugzilla.redhat.com/show_bug.cgi?id=1226854 Signed-off-by: Jiri Denemark <jdenemar(a)redhat.com> --- src/qemu/qemu_domain.c | 20 +++++++++--- .../qemuxml2argv-channel-virtio-unix.args | 19 +++++++++++ .../qemuxml2argv-channel-virtio-unix.xml | 37 ++++++++++++++++++++++ tests/qemuxml2argvtest.c | 5 +++ 4 files changed, 77 insertions(+), 4 deletions(-) create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.args create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.xml diff --git a/src/qemu/qemu_domain.c b/src/qemu/qemu_domain.c index 6213fd9..404489c 100644 --- a/src/qemu/qemu_domain.c +++ b/src/qemu/qemu_domain.c @@ -1249,11 +1249,23 @@ qemuDomainDeviceDefPostParse(virDomainDeviceDefPtr dev, dev->data.chr->deviceType == VIR_DOMAIN_CHR_DEVICE_TYPE_CHANNEL && dev->data.chr->targetType == VIR_DOMAIN_CHR_CHANNEL_TARGET_TYPE_VIRTIO && dev->data.chr->source.type == VIR_DOMAIN_CHR_TYPE_UNIX && - !dev->data.chr->source.data.nix.path && cfg) { - if (virAsprintf(&dev->data.chr->source.data.nix.path, "%s/%s.%s", - cfg->channelTargetDir, - def->name, dev->data.chr->target.name) < 0) + !dev->data.chr->source.data.nix.path) { + if (!cfg) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("cannot generate UNIX socket path")); goto cleanup; + } + + if (dev->data.chr->target.name) { + if (virAsprintf(&dev->data.chr->source.data.nix.path, "%s/%s.%s", + cfg->channelTargetDir, + def->name, dev->data.chr->target.name) < 0) + goto cleanup; + } else { + if (virAsprintf(&dev->data.chr->source.data.nix.path, "%s/%s", + cfg->channelTargetDir, def->name) < 0) + goto cleanup; + } dev->data.chr->source.data.nix.listen = true; } diff --git a/tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.args b/tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.args new file mode 100644 index 0000000..43a34ce --- /dev/null +++ b/tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.args @@ -0,0 +1,19 @@ +LC_ALL=C PATH=/bin HOME=/home/test USER=test LOGNAME=test QEMU_AUDIO_DRV=none \ +/usr/bin/qemu -S -M pc -m 214 -smp 1 -nographic -nodefconfig -nodefaults \ +-chardev socket,id=charmonitor,path=/tmp/test-monitor,server,nowait \ +-mon chardev=charmonitor,id=monitor,mode=readline \ +-no-acpi \ +-boot c \ +-device virtio-serial-pci,id=virtio-serial1,bus=pci.0,addr=0xa \ +-device virtio-serial-pci,id=virtio-serial0,bus=pci.0,addr=0x3 \ +-usb \ +-hda /dev/HostVG/QEMUGuest1 \ +-chardev socket,id=charchannel0,path=\ +/tmp/QEMUGuest1.org.qemu.guest_agent.0,server,nowait \ +-device virtserialport,bus=virtio-serial0.0,nr=1,chardev=charchannel0,id=channel0,\ +name=org.qemu.guest_agent.0 \ +-chardev socket,id=charchannel1,path=/tmp/QEMUGuest1,server,nowait \ +-device virtserialport,bus=virtio-serial0.0,nr=2,chardev=charchannel1,id=channel1 \ +-chardev socket,id=charchannel2,path=/tmp/QEMUGuest1.ble,server,nowait \ +-device virtserialport,bus=virtio-serial0.0,nr=3,chardev=charchannel2,id=channel2,\ +name=ble diff --git a/tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.xml b/tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.xml new file mode 100644 index 0000000..7fac943 --- /dev/null +++ b/tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-unix.xml @@ -0,0 +1,37 @@ +<domain type='qemu'> + <name>QEMUGuest1</name> + <uuid>c7a5fdbd-edaf-9455-926a-d65c16db1809</uuid> + <memory unit='KiB'>219136</memory> + <vcpu placement='static' cpuset='1-4,8-20,525'>1</vcpu> + <os> + <type arch='i686' machine='pc'>hvm</type> + <boot dev='hd'/> + </os> + <clock offset='utc'/> + <on_poweroff>destroy</on_poweroff> + <on_reboot>restart</on_reboot> + <on_crash>destroy</on_crash> + <devices> + <emulator>/usr/bin/qemu</emulator> + <disk type='block' device='disk'> + <source dev='/dev/HostVG/QEMUGuest1'/> + <target dev='hda' bus='ide'/> + <address type='drive' controller='0' bus='0' target='0' unit='0'/> + </disk> + <controller type='usb' index='0'/> + <controller type='ide' index='0'/> + <controller type='virtio-serial' index='1'> + <address type='pci' domain='0x0000' bus='0x00' slot='0x0a' function='0x0'/> + </controller> + <channel type='unix'> + <target type='virtio' name='org.qemu.guest_agent.0'/> + </channel> + <channel type='unix'> + <target type='virtio'/> + </channel> + <channel type='unix'> + <target type='virtio' name='ble'/> + </channel> + <memballoon model='none'/> + </devices> +</domain> diff --git a/tests/qemuxml2argvtest.c b/tests/qemuxml2argvtest.c index 4154601..6f8332d 100644 --- a/tests/qemuxml2argvtest.c +++ b/tests/qemuxml2argvtest.c @@ -519,6 +519,9 @@ mymain(void) driver.config->spiceTLS = 1; if (VIR_STRDUP_QUIET(driver.config->spicePassword, "123456") < 0) return EXIT_FAILURE; + VIR_FREE(driver.config->channelTargetDir); + if (VIR_STRDUP_QUIET(driver.config->channelTargetDir, "/tmp") < 0) + return EXIT_FAILURE; # define DO_TEST_FULL(name, migrateFrom, migrateFd, flags, ...) \ do { \ @@ -1113,6 +1116,8 @@ mymain(void) DO_TEST("channel-virtio-default", QEMU_CAPS_DEVICE, QEMU_CAPS_CHARDEV, QEMU_CAPS_NODEFCONFIG, QEMU_CAPS_SPICE, QEMU_CAPS_CHARDEV_SPICEVMC); + DO_TEST("channel-virtio-unix", + QEMU_CAPS_DEVICE, QEMU_CAPS_CHARDEV, QEMU_CAPS_NODEFCONFIG); DO_TEST("smartcard-host", QEMU_CAPS_CHARDEV, QEMU_CAPS_DEVICE, -- 2.4.5

2 2

[libvirt] [PATCH] qemu: agent: Don't automatically disable CPU0 via guest agent
by Peter Krempa 01 Jul '15

01 Jul '15

While CPU0 was made unpluggable in Linux a while ago it's not desirable to unplug it since some parts of the kernel (suspend-to-ram) still depend on it. This patch fixes the vCPU selection code in libvirt so that it will not be disabled. --- src/qemu/qemu_agent.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/src/qemu/qemu_agent.c b/src/qemu/qemu_agent.c index 043695b..7021a7a 100644 --- a/src/qemu/qemu_agent.c +++ b/src/qemu/qemu_agent.c @@ -1601,9 +1601,13 @@ qemuAgentUpdateCPUInfo(unsigned int nvcpus, size_t i; int nonline = 0; int nofflinable = 0; + ssize_t cpu0 = -1; /* count the active and offlinable cpus */ for (i = 0; i < ncpuinfo; i++) { + if (cpuinfo[i].id == 0) + cpu0 = i; + if (cpuinfo[i].online) nonline++; @@ -1618,6 +1622,15 @@ qemuAgentUpdateCPUInfo(unsigned int nvcpus, } } + /* CPU0 was made offlinable in linux a while ago, but certain parts (suspend + * to ram) of the kernel still don't cope well with that. Make sure that if + * all remaining vCPUs are offlinable, vCPU0 will not be selected to be + * offlined automatically */ + if (nofflinable == nonline && cpu0 >= 0 && cpuinfo[cpu0].online) { + cpuinfo[cpu0].offlinable = false; + nofflinable--; + } + /* the guest agent reported less cpus than requested */ if (nvcpus > ncpuinfo) { virReportError(VIR_ERR_INTERNAL_ERROR, "%s", -- 2.4.1

2 2

[libvirt] [PATCH 0/4] Some small libxl bug fixes
by Jim Fehlig 30 Jun '15

30 Jun '15

While working on a patch to add dom0 support to the libxl driver, I noticed a few existing bugs which are fixed by this small series. IMO 1 and 2 are safe even during the freeze as they fix quite annoying bugs. 3 and 4 are probably safe bug fixes too, unless I'm confused about the use of domainObj->{def,newDef,persistentDef} Jim Fehlig (4): libxl: don't overwrite domain state from statedir config libxl: don't remove persistent domain on start failure libxl: honor domainGetXMLDesc() --inactive flag libxl: Set def->vcpus after successfully modifying live vcpu count src/libxl/libxl_driver.c | 33 ++++++++++++++++++++++++++------- 1 file changed, 26 insertions(+), 7 deletions(-) -- 2.3.7

3 7

[libvirt] [libvirt-glib] More coverage for networking API v2
by Zeeshan Ali (Khattak) 30 Jun '15

30 Jun '15

Fixed the compiler warnings introduced by one of the patches.

3 18

[libvirt] [sandbox v2 00/11] Actually use host-image mounts on /
by Cédric Bosdonnat 30 Jun '15

30 Jun '15

Hi all, In the virt-sandbox documentation we had examples with host-image mounts targetting /. However the / in the sandbox was still the host one. The main goal of this patch series is to fix that problem. This will also be needed to run docker container with libvirt-sandbox. I also added some configure flags to disable lzma or zlib support at build time. At least opensuse 13.2 doesn't have static lzma. Diff to v1: * Fixed Dan's comments * Fixed the container case: * Made init-lxc all-static * Moved the init copy code to libvirt-sandbox-builder.c * Use setenv to set the LD_LIBRARY_PATH rather than reset the whole environment using execve Cédric Bosdonnat (11): Allow disabling build with lzma. Allow disabling zlib support. Enable strcmp checks in libvirt-sandbox-init-qemu.c Make init-lxc all static Copy all needed init programs and all its deps to config subdir Remove init-common dependency on libvirt-sandbox.so init-qemu: extract the mounts.cfg ntry mounting code qemu: use mounts targeting / as root Add function to check if there is a mount with / target Don't add sandbox:root device if we have a mount targetting / container builder: don't expose host rootfs if unneeded cfg.mk | 2 - configure.ac | 37 ++++- libvirt-sandbox/Makefile.am | 45 +++--- .../libvirt-sandbox-builder-container.c | 37 +++-- libvirt-sandbox/libvirt-sandbox-builder-machine.c | 22 +-- libvirt-sandbox/libvirt-sandbox-builder.c | 159 ++++++++++++++++++- libvirt-sandbox/libvirt-sandbox-builder.h | 2 + libvirt-sandbox/libvirt-sandbox-config-all.h | 62 ++++++++ libvirt-sandbox/libvirt-sandbox-config-disk.c | 2 +- libvirt-sandbox/libvirt-sandbox-config-initrd.c | 2 +- .../libvirt-sandbox-config-interactive.c | 2 +- .../libvirt-sandbox-config-mount-file.c | 2 +- .../libvirt-sandbox-config-mount-guest-bind.c | 2 +- .../libvirt-sandbox-config-mount-host-bind.c | 2 +- .../libvirt-sandbox-config-mount-host-image.c | 2 +- libvirt-sandbox/libvirt-sandbox-config-mount-ram.c | 2 +- libvirt-sandbox/libvirt-sandbox-config-mount.c | 2 +- .../libvirt-sandbox-config-network-address.c | 2 +- ...rt-sandbox-config-network-filterref-parameter.c | 2 +- .../libvirt-sandbox-config-network-filterref.c | 2 +- .../libvirt-sandbox-config-network-route.c | 2 +- libvirt-sandbox/libvirt-sandbox-config-network.c | 2 +- .../libvirt-sandbox-config-service-generic.c | 2 +- .../libvirt-sandbox-config-service-systemd.c | 2 +- libvirt-sandbox/libvirt-sandbox-config-service.c | 2 +- libvirt-sandbox/libvirt-sandbox-config.c | 23 ++- libvirt-sandbox/libvirt-sandbox-config.h | 1 + libvirt-sandbox/libvirt-sandbox-init-common.c | 5 +- libvirt-sandbox/libvirt-sandbox-init-lxc.c | 16 +- libvirt-sandbox/libvirt-sandbox-init-qemu.c | 175 +++++++++++++++++---- libvirt-sandbox/libvirt-sandbox-util.c | 2 +- libvirt-sandbox/libvirt-sandbox.sym | 1 + 32 files changed, 519 insertions(+), 104 deletions(-) create mode 100644 libvirt-sandbox/libvirt-sandbox-config-all.h -- 2.1.4

2 15

[libvirt] [PATCH] qemu: process: Improve update of maximum baloon state at startup
by Peter Krempa 30 Jun '15

30 Jun '15

In commit 641a145d73fdc3dd9350fd57b3d3247abf101c05 I've added code that resets the ballon memory value to full size prior to resuming the vCPUs since the size certainly was not reduced at that point. Since qemuProcess start is used also in code paths with already booted up guests (migration, save/restore) the assumption is not entirely true since the guest might already been running before. This patch adds a function that queries the monitor rather than using the full size since a balloon event would not be reissued in case we are recovering a saved migration state. Additionally the new function is used also when reconnecting to a VM after libvirtd restart since we might have missed a few ballon events while libvirtd was not running. --- src/qemu/qemu_process.c | 35 ++++++++++++++++++++++++++++++++++- 1 file changed, 34 insertions(+), 1 deletion(-) diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c index ba84182..5698404 100644 --- a/src/qemu/qemu_process.c +++ b/src/qemu/qemu_process.c @@ -2065,6 +2065,35 @@ qemuProcessReconnectRefreshChannelVirtioState(virQEMUDriverPtr driver, static int +qemuProcessRefreshBallonState(virQEMUDriverPtr driver, + virDomainObjPtr vm) +{ + unsigned long long balloon; + int rc; + + /* if no balloning is available, the current size equals to the current + * full memory size */ + if (!vm->def->memballoon || + vm->def->memballoon->model == VIR_DOMAIN_MEMBALLOON_MODEL_NONE) { + vm->def->mem.cur_balloon = virDomainDefGetMemoryActual(vm->def); + return 0; + } + + qemuDomainObjEnterMonitor(driver, vm); + rc = qemuMonitorGetBalloonInfo(qemuDomainGetMonitor(vm), &balloon); + if (qemuDomainObjExitMonitor(driver, vm) < 0) + rc = -1; + + if (rc < 0) + return -1; + + vm->def->mem.cur_balloon = balloon; + + return 0; +} + + +static int qemuProcessWaitForMonitor(virQEMUDriverPtr driver, virDomainObjPtr vm, int asyncJob, @@ -3779,6 +3808,9 @@ qemuProcessReconnect(void *opaque) if (qemuProcessReconnectRefreshChannelVirtioState(driver, obj) < 0) goto error; + if (qemuProcessRefreshBallonState(driver, obj) < 0) + goto error; + if (qemuProcessRecoverJob(driver, obj, conn, &oldjob) < 0) goto error; @@ -4922,7 +4954,8 @@ int qemuProcessStart(virConnectPtr conn, /* Since CPUs were not started yet, the ballon could not return the memory * to the host and thus cur_balloon needs to be updated so that GetXMLdesc * and friends return the correct size in case they can't grab the job */ - vm->def->mem.cur_balloon = virDomainDefGetMemoryActual(vm->def); + if (qemuProcessRefreshBallonState(driver, vm) < 0) + goto cleanup; VIR_DEBUG("Detecting actual memory size for video device"); if (qemuProcessUpdateVideoRamSize(driver, vm, asyncJob) < 0) -- 2.4.1

2 1

[libvirt] [PATCH 0/3] Address a couple of mpath/Multipath storage pool issues
by John Ferlan 30 Jun '15

30 Jun '15

My thought was to combine patch 2 and 3 unless it was felt that disallowing more than one Multipath pool to defined at one time (patch 3) was unnecessary For patch 1, I did some searching and came across the following to describe why /dev/mpath is no longer valid: http://comments.gmane.org/gmane.linux.kernel.device-mapper.devel/13864 I kept /dev/mpath as a valid option only just in case there's some strange back-compat issue or some platform where /dev/mpath is preferred/used instead of /dev/mapper. John Ferlan (3): mpath: Update path in CheckPool function docs: Clarify the Multipath target path value usage mpath: Don't allow more than one mpath pool at a time docs/formatstorage.html.in | 2 ++ docs/storage.html.in | 3 ++- src/conf/storage_conf.c | 3 +++ src/storage/storage_backend_mpath.c | 3 ++- 4 files changed, 9 insertions(+), 2 deletions(-) -- 2.1.0

2 5

[libvirt] [PATCH 0/2] Domain lifecycle onCrash adjustments
by John Ferlan 30 Jun '15

30 Jun '15

First patch uses the right symbols, second patch adds an extra check to qemu backend for the coredump-destroy action John Ferlan (2): Use the correct symbol for 'onCrash' qemu: Add missing on_crash lifecycle type src/lxc/lxc_native.c | 2 +- src/qemu/qemu_command.c | 5 +++-- src/vmx/vmx.c | 2 +- src/xenconfig/xen_sxpr.c | 2 +- 4 files changed, 6 insertions(+), 5 deletions(-) -- 2.1.0

2 3

[libvirt] [RFC] qemu: Redesigning guest CPU configuration
by Jiri Denemark 30 Jun '15

30 Jun '15

Hi all (and sorry for the long email), The current way QEMU driver handles guest CPU configuration is not ideal. We detect host CPU capabilities only by querying the CPU and we don't check with QEMU what features it supports. We don't check QEMU's definitions of CPU models, which may be different from libvirt's definitions. All this results in several issues: - guest CPU may change druing migration, save/restore - libvirt may ask for a CPU which QEMU cannot provide; the guest will see a slightly different CPU but libvirt client won't know about it - libvirt may come up with a CPU that doesn't make sense and which won't work for a guest (the guest may even crash) Although usually everything just works, it is very fragile. Since we want to fix all these issues, we need to: - guarantee stable guest ABI (a single domain XML should always results in the same guest ABI). Once a domain is started, its CPU definition should never change (unless someone changes the XML, of course, similar to, e.g. PCI addresses). However, there are a few exceptions: - host-passthrough CPU mode will always result in "-cpu host" - host-model CPU mode should recompute the CPU model on every start, but the CPU must not change during migration - always make sure QEMU provides the CPU we asked for. Starting a domain should fail in case QEMU cannot provide exactly the CPU we asked for. - provide usable host-model mode and custom mode with minimum match. We need to generate CPU configurations that actually work, i.e., we need to ask QEMU what CPU it can provide on current host rather than requesting a bunch of features on top of a CPU model which does not always match the host CPU. QEMU already provides or will soon provide everything we need to meet these requirements: - we can cover every configurable part of a CPU in our cpu_map.xml and instead of asking QEMU for a specific CPU model we can use "-cpu custom" with a fully specified CPU - we can use the additional data about CPU models to choose the right one for a host CPU - when starting a domain we can check whether QEMU filtered out any of the features we asked for and refuse to start the domain - we can ask QEMU what would "-cpu host" look like and use that for host-model and minimum match CPUs (it won't work for TCG mode, though, but we can keep using the current CPUID detection code for TCG) Once we start maintaining CPU models with all the details, we will likely meet the same issues QEMU folks meet, i.e., we will need to fix bugs in existing CPU models. And it's not just about adding removing CPU features but also fixing other parameters, such as wrong level, etc. It's clear every change will require a new CPU model to be defined. But I think we should do it in a way that applications or users should not need (if they don't want to) to care about it. I'm thinking about doing something similar to machine types. Each CPU model could be defined in several versions and a CPU specs without a version would be an alias to the latest version. The problem is, we need to maintain backward compatibility and we should avoid breaking existing domains (shouldn't we?) which just work even though their guest CPUs do not exactly match the domain XML definitions. So either we need to define all existing CPU models in all their variants used for various machine types and have a mapping between (model without a version, machine type) to a specific version of the model (which may be quite hard) or we need to be able to distinguish between an existing domain and a new domain with no CPU model version. While host-model and host-passthrough CPU modes are easy because they are designed to change everytime a domain starts (which means we don't need to be able to distinguish between existing and new domains), custom CPU mode are tricky. Currently, the only at least a bit reasonable thing which came to my mind is to have a new CPU mode, but it still seems awkward so please share your ideas if you have any. BTW, I don't think we should try to expose every part of the CPU model definitions in domain XML, they should remain hidden behind the CPU model name. It would be hard to explain what each of the extra parameters mean, each model would have to include them anyway since we can't expect users to provide all the details correctly, and once visible in domain XML it could encourage users to play with the values. I'm looking forward to any comments or ideas. Jirka

4 14