March 2019 - Devel - libvirt List Archives

[libvirt] [PATCH 00/21] qemu_hotplug: refactor device detach functions to fix one serious and other minor bugs
by Laine Stump 22 Mar '19

22 Mar '19

This all started with the observation that we were sending the DEVICE_REMOVED events for unplugged devices too soon (reported in https://bugzilla.redhat.com/1658198 ). While noticing that *all* of the device types send their DEVICE_REMOVED event too early, I also noticed how eerily similar all the different detach functions are (they have converged over time, as legacy code supporting old qemus with different methods of deleting different types of devices has been removed - pkrempa and mprivozn committed patches just in the last week that contributed to this convergence and removed some duplicated code). I also saw a few other definite bugs, as well as a couple that I think are bugs, but am not enough informed about zPCI to be certain (I've Cc'ed the author and reviewers of the original patch adding that code to the patches so that they can verify, or negate, my suspicions). Many/most of these patches are either very simple / easily verifiable code movement / removal / renaming of variables. Patch 20 is a bit more involved; viewing the new versions of the functions along with the patch would probably help it go much quicker. The first two patches in the series eliminate what I'm 99% certain is erroneous calling of the qemuDomainDetachExtensionDevice() when detaching a PCI device on S390. The commit log details my reasons for believing this code should be removed. The rest of the series eliminates small differences iteratively, arriving at a set of detach functions which are completely identical at the end (and only called from one place), and then replaces all those duplicates with a single block of common code in a higher level function that calls all of the individual functions (it was their only caller anyway). This eliminated the inconsistent application of checking for multifunction PCI devices and for the existence of a device alias. It also eliminated individual calls to qemuDomainRemove*Device() calls, and replaced them with a single call to qemuDomainRemoveDevice(). Eliminating all the calls (except one) to the individual qemuDomainRemove*Device() functions in turn made it possible to move the code that builds and queues the DEVICE_REMOVED event into a single place in the one remaining higher level caller - qemuDomainRemoveDevice() - thus eliminating the "premature event" for *all* device types at once. Laine Stump (21): qemu_hotplug: remove erroneous call to qemuDomainDetachExtensionDevice() qemu_hotplug: remove another erroneous qemuDomainDetachExtensionDevice() call qemu_hotplug: remove unnecessary check for valid PCI address qemu_hotplug: rename a virDomainDeviceInfoPtr to avoid confusion qemu_hotplug: eliminate multiple identical qemuDomainDetachHost*Device() functions qemu_hotplug: eliminate unnecessary call to qemuDomainDetachNetDevice() qemu_hotplug: refactor qemuDomainDetachDiskLive and qemuDomainDetachDiskDevice qemu_hotplug: don't call DetachThisHostDevice for hostdev network devices qemu_hotplug: merge qemuDomainDetachThisHostDevice into qemuDomainDetachHostDevice qemu_hotplug: move qemuDomainChangeGraphicsPasswords() qemu_hotplug: move (almost) all qemuDomainDetach*() functions together qemu_hotplug: move (Attach|Detach)Lease functions with others of same type qemu_hotplug: move qemuDomainDetachDeviceLive() to qemu_hotplug.c qemu_hotplug: remove extra function in middle of DetachController call chain qemu_hotplug: pull qemuDomainUpdateDeviceList out of qemuDomainDetachDeviceLive test: replace calls to individual detach functions with one call to main detach qemu_hotplug: rename dev to match in qemuDomainDetachDeviceLive qemu_hotplug: standardize the names/args/calling of qemuDomainDetach*() qemu_hotplug: new function qemuDomainRemoveAuditDevice() qemu_hotplug: consolidate all common detach code in qemuDomainDetachDeviceLive qemu_hotplug: delay sending DEVICE_REMOVED event until after *all* teardown src/qemu/qemu_driver.c | 102 +-- src/qemu/qemu_hotplug.c | 1407 ++++++++++++++++++--------------------- src/qemu/qemu_hotplug.h | 56 +- tests/qemuhotplugtest.c | 8 +- 4 files changed, 655 insertions(+), 918 deletions(-) Yay! jtomko would be proud - more deletions than insertions!!! -- 2.20.1

4 63

[libvirt] Bite sized task: #pragma once
by Daniel P. Berrangé 22 Mar '19

22 Mar '19

A heads up that I added a trivial bite sized task to convert header files to use #pragma once https://wiki.libvirt.org/page/BiteSizedTasks#Switch_headers_to_use_.23pragm… Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|

1 0

[libvirt] [PATCH] virDomainMomentAssignDef: Don't dereference a NULL pointer
by Michal Privoznik 22 Mar '19

22 Mar '19

This functions tries to add a domain moment (love the name!) onto a list of domain moments. Firstly, it checks if another moment with the same name already exists. Then, it creates an empty moment (without initializing its definition) and tries to add the moment onto the list dereferencing moment definition in that process. If it succeeds (which it never can), only after that it sets moment->def. Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com> --- src/conf/virdomainmomentobjlist.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/conf/virdomainmomentobjlist.c b/src/conf/virdomainmomentobjlist.c index b2122e7292..01e50ae1d4 100644 --- a/src/conf/virdomainmomentobjlist.c +++ b/src/conf/virdomainmomentobjlist.c @@ -227,7 +227,7 @@ virDomainMomentAssignDef(virDomainMomentObjListPtr moments, if (!(moment = virDomainMomentObjNew())) return NULL; - if (virHashAddEntry(moments->objs, moment->def->name, moment) < 0) { + if (virHashAddEntry(moments->objs, def->name, moment) < 0) { VIR_FREE(moment); return NULL; } -- 2.19.2

3 2

[libvirt] [PATCH] nwfilter: fix adding std MAC and IP values to filter binding
by Nikolay Shirokovskiy 22 Mar '19

22 Mar '19

Commit d1a7c08eb changed filter instantiation code to ignore MAC and IP variables explicitly specified for filter binding. It just replaces explicit values with values associated with the binding. Before the commit virNWFilterCreateVarsFrom was used so that explicit value take precedence. Let's bring old behavior back. This is useful. For example if domain has two interfaces it makes sense to list both mac adresses in MAC var of every interface filterref. So that if guest make a bond of these interfaces and start sending frames with one of the mac adresses from both interfaces we can pass outgress traffic from both interfaces too. Signed-off-by: Nikolay Shirokovskiy <nshirokovskiy(a)virtuozzo.com> --- src/nwfilter/nwfilter_gentech_driver.c | 92 ++++++++++++---------------------- 1 file changed, 32 insertions(+), 60 deletions(-) diff --git a/src/nwfilter/nwfilter_gentech_driver.c b/src/nwfilter/nwfilter_gentech_driver.c index 655f088..6d68189 100644 --- a/src/nwfilter/nwfilter_gentech_driver.c +++ b/src/nwfilter/nwfilter_gentech_driver.c @@ -127,60 +127,6 @@ virNWFilterRuleInstFree(virNWFilterRuleInstPtr inst) /** - * virNWFilterVarHashmapAddStdValues: - * @tables: pointer to hash tabel to add values to - * @macaddr: The string of the MAC address to add to the hash table, - * may be NULL - * @ipaddr: The string of the IP address to add to the hash table; - * may be NULL - * - * Returns 0 in case of success, -1 in case an error happened with - * error having been reported. - * - * Adds a couple of standard keys (MAC, IP) to the hash table. - */ -static int -virNWFilterVarHashmapAddStdValues(virHashTablePtr table, - const char *macaddr, - const virNWFilterVarValue *ipaddr) -{ - virNWFilterVarValue *val; - - if (macaddr) { - val = virNWFilterVarValueCreateSimpleCopyValue(macaddr); - if (!val) - return -1; - - if (virHashUpdateEntry(table, - NWFILTER_STD_VAR_MAC, - val) < 0) { - virNWFilterVarValueFree(val); - virReportError(VIR_ERR_INTERNAL_ERROR, - "%s", _("Could not add variable 'MAC' to hashmap")); - return -1; - } - } - - if (ipaddr) { - val = virNWFilterVarValueCopy(ipaddr); - if (!val) - return -1; - - if (virHashUpdateEntry(table, - NWFILTER_STD_VAR_IP, - val) < 0) { - virNWFilterVarValueFree(val); - virReportError(VIR_ERR_INTERNAL_ERROR, - "%s", _("Could not add variable 'IP' to hashmap")); - return -1; - } - } - - return 0; -} - - -/** * Convert a virHashTable into a string of comma-separated * variable names. */ @@ -705,6 +651,28 @@ virNWFilterDoInstantiate(virNWFilterTechDriverPtr techdriver, } +static int +virNWFilterVarHashmapAddStdValue(virHashTablePtr table, + const char *var, + const char *value) +{ + virNWFilterVarValue *val; + + if (virHashLookup(table, var)) + return 0; + + if (!(val = virNWFilterVarValueCreateSimpleCopyValue(value))) + return -1; + + if (virHashAddEntry(table, var, val) < 0) { + virNWFilterVarValueFree(val); + return -1; + } + + return 0; +} + + /* * Call this function while holding the NWFilter filter update lock */ @@ -717,7 +685,7 @@ virNWFilterInstantiateFilterUpdate(virNWFilterDriverStatePtr driver, bool forceWithPendingReq, bool *foundNewFilter) { - int rc; + int rc = -1; const char *drvname = EBIPTABLES_DRIVER_ID; virNWFilterTechDriverPtr techdriver; virNWFilterObjPtr obj; @@ -743,14 +711,18 @@ virNWFilterInstantiateFilterUpdate(virNWFilterDriverStatePtr driver, return -1; virMacAddrFormat(&binding->mac, vmmacaddr); + if (virNWFilterVarHashmapAddStdValue(binding->filterparams, + NWFILTER_STD_VAR_MAC, + vmmacaddr) < 0) + goto err_exit; ipaddr = virNWFilterIPAddrMapGetIPAddr(binding->portdevname); - - if (virNWFilterVarHashmapAddStdValues(binding->filterparams, - vmmacaddr, ipaddr) < 0) { - rc = -1; + if (ipaddr && + virNWFilterVarHashmapAddStdValue(binding->filterparams, + NWFILTER_STD_VAR_IP, + virNWFilterVarValueGetSimple(ipaddr)) < 0) goto err_exit; - } + filter = virNWFilterObjGetDef(obj); -- 1.8.3.1

2 4

[libvirt] [PATCH] xml: nodedev: make pci capability class element optional
by Nikolay Shirokovskiy 22 Mar '19

22 Mar '19

Commit 3bd4ed46 introduced this element as required which breaks backcompat for test driver. Let's make the element optional. Signed-off-by: Nikolay Shirokovskiy <nshirokovskiy(a)virtuozzo.com> --- docs/formatnode.html.in | 2 +- docs/schemas/nodedev.rng | 12 ++++++----- src/conf/node_device_conf.c | 23 +++++++++++----------- src/conf/node_device_conf.h | 2 +- src/node_device/node_device_udev.c | 2 +- .../pci_0000_02_10_7_sriov_pf_vfs_all.xml | 1 - 6 files changed, 22 insertions(+), 20 deletions(-) diff --git a/docs/formatnode.html.in b/docs/formatnode.html.in index 5095b97..c2a8f8f 100644 --- a/docs/formatnode.html.in +++ b/docs/formatnode.html.in @@ -71,7 +71,7 @@ include: <dl> <dt><code>class</code></dt> - <dd>Combined class, subclass and + <dd>Optional element for combined class, subclass and programming interface codes as 6-digit hexadecimal number. <span class="since">Since 5.2.0</span></dd> <dt><code>domain</code></dt> diff --git a/docs/schemas/nodedev.rng b/docs/schemas/nodedev.rng index 0f45d79..fe6ffa0 100644 --- a/docs/schemas/nodedev.rng +++ b/docs/schemas/nodedev.rng @@ -133,11 +133,13 @@ <value>pci</value> </attribute> - <element name='class'> - <data type="string"> - <param name="pattern">0x[0-9a-fA-F]{6}</param> - </data> - </element> + <optional> + <element name='class'> + <data type="string"> + <param name="pattern">0x[0-9a-fA-F]{6}</param> + </data> + </element> + </optional> <element name='domain'> <ref name='unsignedLong'/> </element> diff --git a/src/conf/node_device_conf.c b/src/conf/node_device_conf.c index 19c601a..b96d10c 100644 --- a/src/conf/node_device_conf.c +++ b/src/conf/node_device_conf.c @@ -208,7 +208,8 @@ virNodeDeviceCapPCIDefFormat(virBufferPtr buf, { size_t i; - virBufferAsprintf(buf, "<class>0x%.6x</class>\n", data->pci_dev.klass); + if (data->pci_dev.klass >= 0) + virBufferAsprintf(buf, "<class>0x%.6x</class>\n", data->pci_dev.klass); virBufferAsprintf(buf, "<domain>%d</domain>\n", data->pci_dev.domain); virBufferAsprintf(buf, "<bus>%d</bus>\n", data->pci_dev.bus); @@ -1645,16 +1646,16 @@ virNodeDevCapPCIDevParseXML(xmlXPathContextPtr ctxt, orignode = ctxt->node; ctxt->node = node; - if (virNodeDevCapsDefParseHexId("string(./class[1])", ctxt, - &pci_dev->klass, def, - _("no PCI class supplied for '%s'"), - _("invalid PCI class supplied for '%s'")) < 0) - goto out; - - if (pci_dev->klass > 0xffffff) { - virReportError(VIR_ERR_INTERNAL_ERROR, - _("invalid PCI class supplied for '%s'"), def->name); - goto out; + if ((tmp = virXPathString("string(./class[1])", ctxt))) { + if (virStrToLong_i(tmp, NULL, 16, &pci_dev->klass) < 0 || + pci_dev->klass > 0xffffff) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("invalid PCI class supplied for '%s'"), def->name); + goto out; + } + VIR_FREE(tmp); + } else { + pci_dev->klass = -1; } if (virNodeDevCapsDefParseULong("number(./domain[1])", ctxt, diff --git a/src/conf/node_device_conf.h b/src/conf/node_device_conf.h index b13bc13..e8cb315 100644 --- a/src/conf/node_device_conf.h +++ b/src/conf/node_device_conf.h @@ -153,7 +153,7 @@ struct _virNodeDevCapPCIDev { unsigned int function; unsigned int product; unsigned int vendor; - unsigned int klass; + int klass; char *product_name; char *vendor_name; virPCIDeviceAddressPtr physical_function; diff --git a/src/node_device/node_device_udev.c b/src/node_device/node_device_udev.c index f0e61e4..5ed3463 100644 --- a/src/node_device/node_device_udev.c +++ b/src/node_device/node_device_udev.c @@ -402,7 +402,7 @@ udevProcessPCI(struct udev_device *device, privileged = driver->privileged; nodeDeviceUnlock(); - if (udevGetUintProperty(device, "PCI_CLASS", &pci_dev->klass, 16) < 0) + if (udevGetIntProperty(device, "PCI_CLASS", &pci_dev->klass, 16) < 0) goto cleanup; if ((p = strrchr(def->sysfs_path, '/')) == NULL || diff --git a/tests/nodedevschemadata/pci_0000_02_10_7_sriov_pf_vfs_all.xml b/tests/nodedevschemadata/pci_0000_02_10_7_sriov_pf_vfs_all.xml index 74036a6..9e8dace 100644 --- a/tests/nodedevschemadata/pci_0000_02_10_7_sriov_pf_vfs_all.xml +++ b/tests/nodedevschemadata/pci_0000_02_10_7_sriov_pf_vfs_all.xml @@ -2,7 +2,6 @@ <name>pci_0000_02_10_7</name> <parent>pci_0000_00_04_0</parent> <capability type='pci'> - <class>0xffffff</class> <domain>0</domain> <bus>2</bus> <slot>16</slot> -- 1.8.3.1

2 6

[libvirt] [PATCH v2 0/4] tests: qemuxml2argv: support optional arguments
by Cole Robinson 22 Mar '19

22 Mar '19

Small/completion follow up to v1 here, which is mostly pushed: https://www.redhat.com/archives/libvir-list/2019-March/msg00943.html Patch 1 and 3 are new additions based on Andrea's reviews. Rest of the review comments should be addressed Cole Robinson (4): tests: qemuxml2argv: Tweak TEST_CAPS_PATH tests: qemuxml2argv: move DO_CAPS_TEST* qemuCaps init tests: qemuxml2argv: report error on ARG_* collisions tests: qemuxml2argv: add DO_TEST_INTERNAL tests/qemuxml2argvtest.c | 102 +++++++++++++++++++++++++-------------- 1 file changed, 66 insertions(+), 36 deletions(-) -- 2.21.0

2 8

[libvirt] [PATCH 00/21] tests: qemuxml2argv: support optional arguments
by Cole Robinson 21 Mar '19

21 Mar '19

Right now in qemuxml2argv we have a proliferation of DO_*TEST* macros. They essentially fill in data in the testInfo struct and invoke the test function. There's several bits of data that we want to specify for a small subset of tests: flags, parseFlags, migrateFrom/migrateFd, gic. The base macros need to handle these in their argument lists, and we provide many convenience macros that fill in default values for the less common parameters. Any time we want to add a new bit of data though, the base macros are extended, and every caller of those needs to be adjusted, and we are faced with the question of whether to extend the combinatorial explosion of convenience macros which have only a subset of options exposed. This series adds a testInfoSetArgs which uses va_args to make these mandatory parameters optional. The general format is: DO_TEST_FULL(... ARG_FOO, foovalue, ARG_BAR, barvalue, ...) Specifically, one of the migrate tests went from: DO_TEST_FULL("restore-v2", "exec:cat", 7, 0, 0, GIC_NONE, NONE); to: DO_TEST_FULL("restore-v2", ARG_MIGRATE_FROM, "exec:cat", ARG_MIGRATE_FD, 7, ARG_QEMU_CAPS, NONE); Which may not seem directly compelling, but adding new testInfo fields will be much easier. This will also be a base for a later series to share the VIR_TEST_CAPS infrastructure with qemuxml2xmltest Cole Robinson (21): qemu: add virQEMUCapsSetVList tests: qemuxml2argv: add testInfoSetArgs tests: qemuxml2argv: add va_arg enum handling tests: qemuxml2argv: push ARG_QEMU_CAPS to callers tests: qemuxml2argv: break apart testInitQEMUCaps tests: qemuxml2argv: handle gic with vaargs tests: qemuxml2argv: handle migrate* with varargs tests: qemuxml2argv: handle flags with varargs tests: qemuxml2argv: handle parseFlags with varargs tests: qemuxml2argv: remove DO_TEST_PARSE_FLAGS_ERROR tests: qemuxml2argv: remove unused DO_TEST_CAPS* macros tests: qemuxml2argv: add a comment separating DO_TEST* macros tests: qemuxml2argv: remove unused CAPS migrateFrom tests: qemuxml2argv: use varargs for CAPS flags tests: qemuxml2argv: remove full testInfo initialization tests: qemuxml2argv: centralize CAPS suffix building tests: qemuxml2argv: build capsfile in DO_TEST_CAPS_INTERNAL tests: qemuxml2argv: add testInfoClear tests: qemuxml2argv: move DO_TEST qemuCaps init tests: qemuxml2argv: move DO_CAPS_TEST* qemuCaps init tests: qemuxml2argv: add TEST_INTERNAL src/qemu/qemu_capabilities.c | 14 +- src/qemu/qemu_capabilities.h | 2 + tests/qemuxml2argvtest.c | 273 +++++++++++++++++++++++------------ 3 files changed, 193 insertions(+), 96 deletions(-) -- 2.20.1

4 56

Re: [libvirt] [Qemu-devel] [PULL v3 00/54] Kconfig conversion, excluding ARM and MIPS
by Laszlo Ersek 21 Mar '19

21 Mar '19

Hi Paolo, (+libvir-list) I'd like to report a regression introduced by this patch set: On 03/07/19 21:58, Paolo Bonzini wrote: > The following changes since commit > 6cb4f6db4f4367faa33da85b15f75bbbd2bed2a6: > > Merge remote-tracking branch > 'remotes/cleber/tags/python-next-pull-request' into staging > (2019-03-07 16:16:02 +0000) > > are available in the Git repository at: > > git://github.com/bonzini/qemu.git tags/for-upstream-kconfig > > for you to fetch changes up to > 576c3f2f16e7392e28cc9fe10d9a920d67d3645b: > > kconfig: add documentation (2019-03-07 21:54:22 +0100) > > ---------------------------------------------------------------- > Initial Kconfig work, excluding ARM and MIPS > > ---------------------------------------------------------------- I use "libvirt-daemon-4.5.0-10.el7.x86_64" at the moment. Here are the symptoms: (1) the output of the following command changes across this series: virsh domcapabilities qemu /opt/qemu-installed-optimized/bin/qemu-system-aarch64 aarch64 virt-4.0 as shown by the diff > @@ -92,6 +92,8 @@ > <video supported='yes'> > <enum name='modelType'> > <value>vga</value> > + <value>cirrus</value> > + <value>vmvga</value> > <value>virtio</value> > </enum> > </video> > @@ -111,10 +113,7 @@ > <value>scsi</value> > </enum> > <enum name='capsType'/> > - <enum name='pciBackend'> > - <value>default</value> > - <value>vfio</value> > - </enum> > + <enum name='pciBackend'/> > </hostdev> > </devices> > <features> (2) given the following domain XML snippet, in a qemu (not KVM) aarch64 guest that I have, > <video> > <model type='virtio' heads='1' primary='yes'/> > <address type='pci' domain='0x0000' bus='0x04' slot='0x00' function='0x0'/> > </video> the QEMU command line generated by libvirt changes: > - -device virtio-gpu-pci,id=video0,max_outputs=1,bus=pci.4,addr=0x0 \ > + -device virtio-vga,id=video0,max_outputs=1,bus=pci.4,addr=0x0 \ Perhaps this series causes (1), which in turn causes (2), or else this series causes both (1) and (2) independently of each other. I'm not sure about that. Either way, (2) is a problem, because virtio-vga is inappropriate for aarch64 guests (and the ArmVirtQemu platform firmware in edk2 won't even try to drive it -- a blank screen is the result). Now, I can't pinpoint an exact commit, because (after a *brutal* bisection) I get: > There are only 'skip'ped commits left to test. > The first bad commit could be any of: > a02c0edb35662de38d400f42b68845540669ca3d > 03b348bdcbd1eda4ce097b2b84527dec774d80c4 > d6e9c470fc91f75db1785f17a9d3567d5a27953d > a7e23159074c9d774fb1e88357b778994a0c9365 > bcb129b3154ba743f8e52c21c331a0dfcaee7c38 > 02017ee385ef574133c4a978d368640772978178 > 7c28b925b7e176b4e44ed05d23cf883561000546 > 1550b0e6bfe3ab6985e5ad75df1c528a0ca39468 > e9947d18df97e6c6584f020cf9cc995404cc8061 > 8f01b41e1098d8cb9491fa3ea7bd59cf187a5bd7 > 9533dcdd416530a0d72140c122bf90517b6c81eb > 32690c8bed5762518272876dcb6dd39a54f54fd1 > c3a98aa54c734dcb7a36d193c6330d8f04d4bf8e > ccf222a816d59af1318a7efb59c6b9c5578d1abf > f349474920d80838ecea3d421531fdb0660b8740 > 2ac041c2c3d89691cda1657981c41fe4bc20244b > e0e312f3525ad6ac18ba6633af29190dd9620cbc > b42075bb77672616127c9452c0f836e757e9ce1a > We cannot bisect more! Here's how I built each step of the bisection: - "git clean -ffdx" and "git submodule deinit --all --force" on the source directory - create a new, pristine build directory - configure QEMU for out-of-tree build, in that directory - select the i386, x86_64, arm, and aarch64 targets (all softmmu) - report "skip" to git-bisect iff the build breaks, otherwise test the aarch64 emulator through libvirt, and report good/bad dependent on the test result. The bisection log is attached -- "git bisect replay" and "git bisect visualize" should tell the story better than the above dump of commit hashes. Basically: - The tree builds up to and including commit 82f5181777eb ("kconfig: introduce kconfig files", 2019-03-07), and that commit works fine as well. - Then the build breaks (starting with commit e0e312f3525a, "build: switch to Kconfig", 2019-03-07). - The tree becomes buildable again at commit b42075bb7767 ("virtio: express virtio dependencies with Kconfig", 2019-03-07), but at that point, the functionality has been regressed. Thanks, Laszlo

4 8

[libvirt] [PATCH v2] numa: warn if numa 'mem' option or default RAM splitting between nodes is used.
by Igor Mammedov 20 Mar '19

20 Mar '19

Amend -numa option docs and print warnings if 'mem' option or default RAM splitting between nodes is used. It's intended to discourage users from using configuration that allows only to fake NUMA on guest side while leading to reduced performance of the guest due to inability to properly configure VM's RAM on the host. In NUMA case, it's recommended to always explicitly configure guest RAM using -numa node,memdev={backend-id} option. Signed-off-by: Igor Mammedov <imammedo(a)redhat.com> --- numa.c | 5 +++++ qemu-options.hx | 12 ++++++++---- 2 files changed, 13 insertions(+), 4 deletions(-) diff --git a/numa.c b/numa.c index 3875e1e..42838f9 100644 --- a/numa.c +++ b/numa.c @@ -121,6 +121,8 @@ static void parse_numa_node(MachineState *ms, NumaNodeOptions *node, if (node->has_mem) { numa_info[nodenr].node_mem = node->mem; + warn_report("Parameter -numa node,mem is obsolete," + " use -numa node,memdev instead"); } if (node->has_memdev) { Object *o; @@ -407,6 +409,9 @@ void numa_complete_configuration(MachineState *ms) if (i == nb_numa_nodes) { assert(mc->numa_auto_assign_ram); mc->numa_auto_assign_ram(mc, numa_info, nb_numa_nodes, ram_size); + warn_report("Default splitting of RAM between nodes is obsolete," + " Use '-numa node,memdev' to explicitly define RAM" + " allocation per node"); } numa_total = 0; diff --git a/qemu-options.hx b/qemu-options.hx index 1cf9aac..61035cb 100644 --- a/qemu-options.hx +++ b/qemu-options.hx @@ -206,10 +206,14 @@ For example: -numa cpu,node-id=0,socket-id=0 -numa cpu,node-id=1,socket-id=1 @end example -@samp{mem} assigns a given RAM amount to a node. @samp{memdev} -assigns RAM from a given memory backend device to a node. If -@samp{mem} and @samp{memdev} are omitted in all nodes, RAM is -split equally between them. +@samp{memdev} assigns RAM from a given memory backend device to a node. + +Legacy options/behaviour: @samp{mem} assigns a given RAM amount to a node. +If @samp{mem} and @samp{memdev} are omitted in all nodes, RAM is split equally +between them. Option @samp{mem} and default RAM splitting are obsolete as they +do not provide means to manage RAM on the host side and only allow QEMU to fake +NUMA support which in practice could degrade VM performance. +It's advised to always explicitly configure NUMA RAM by using the @samp{memdev} option. @samp{mem} and @samp{memdev} are mutually exclusive. Furthermore, if one node uses @samp{memdev}, all of them have to use it. -- 2.7.4

5 19

[libvirt] [PATCH 1/2] snapshots: allow create --redefine to force a new configuration
by Christian Ehrhardt 20 Mar '19

20 Mar '19

Currently there is no means to permanently modify the metadata stored within a snapshot if it does not pass the ABI compat checker. That is pretty much intentional but there are use cases where users verified that the change will work according to their needs and don't want to: snapshot-revert -> edit -> start, but instead store the modified domain definition within the metadata. To achive that this adds a --force argument to snapshot-create --redefine which will make it skip the check. Signed-off-by: Christian Ehrhardt <christian.ehrhardt(a)canonical.com> --- include/libvirt/libvirt-domain-snapshot.h | 3 +++ src/conf/snapshot_conf.c | 3 ++- src/qemu/qemu_driver.c | 7 ++++++- tools/virsh-snapshot.c | 6 ++++++ tools/virsh.pod | 8 +++++++- 5 files changed, 24 insertions(+), 3 deletions(-) diff --git a/include/libvirt/libvirt-domain-snapshot.h b/include/libvirt/libvirt-domain-snapshot.h index 602e5def59..dc8ba2b4a9 100644 --- a/include/libvirt/libvirt-domain-snapshot.h +++ b/include/libvirt/libvirt-domain-snapshot.h @@ -71,6 +71,9 @@ typedef enum { VIR_DOMAIN_SNAPSHOT_CREATE_LIVE = (1 << 8), /* create the snapshot while the guest is running */ + VIR_DOMAIN_SNAPSHOT_CREATE_NOCHECK = (1 << 9), /* take definition as + provided without + ABI compat check */ } virDomainSnapshotCreateFlags; /* Take a snapshot of the current VM state */ diff --git a/src/conf/snapshot_conf.c b/src/conf/snapshot_conf.c index ffb1313c89..f95b104ad5 100644 --- a/src/conf/snapshot_conf.c +++ b/src/conf/snapshot_conf.c @@ -474,7 +474,8 @@ virDomainSnapshotRedefineValidate(virDomainSnapshotDefPtr def, if (other->def->dom) { if (def->dom) { - if (!virDomainDefCheckABIStability(other->def->dom, + if (!(flags & VIR_DOMAIN_SNAPSHOT_CREATE_NOCHECK) && + !virDomainDefCheckABIStability(other->def->dom, def->dom, xmlopt)) return -1; } else { diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index a16eab5467..f3d032f633 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -15680,7 +15680,8 @@ qemuDomainSnapshotCreateXML(virDomainPtr domain, VIR_DOMAIN_SNAPSHOT_CREATE_REUSE_EXT | VIR_DOMAIN_SNAPSHOT_CREATE_QUIESCE | VIR_DOMAIN_SNAPSHOT_CREATE_ATOMIC | - VIR_DOMAIN_SNAPSHOT_CREATE_LIVE, NULL); + VIR_DOMAIN_SNAPSHOT_CREATE_LIVE | + VIR_DOMAIN_SNAPSHOT_CREATE_NOCHECK, NULL); VIR_REQUIRE_FLAG_RET(VIR_DOMAIN_SNAPSHOT_CREATE_QUIESCE, VIR_DOMAIN_SNAPSHOT_CREATE_DISK_ONLY, @@ -15689,6 +15690,10 @@ qemuDomainSnapshotCreateXML(virDomainPtr domain, VIR_DOMAIN_SNAPSHOT_CREATE_REDEFINE, NULL); + VIR_REQUIRE_FLAG_RET(VIR_DOMAIN_SNAPSHOT_CREATE_NOCHECK, + VIR_DOMAIN_SNAPSHOT_CREATE_REDEFINE, + NULL); + if ((redefine && !(flags & VIR_DOMAIN_SNAPSHOT_CREATE_CURRENT)) || (flags & VIR_DOMAIN_SNAPSHOT_CREATE_NO_METADATA)) update_current = false; diff --git a/tools/virsh-snapshot.c b/tools/virsh-snapshot.c index 6ea6e2744a..a680d2b410 100644 --- a/tools/virsh-snapshot.c +++ b/tools/virsh-snapshot.c @@ -146,6 +146,10 @@ static const vshCmdOptDef opts_snapshot_create[] = { .type = VSH_OT_BOOL, .help = N_("require atomic operation") }, + {.name = "force", + .type = VSH_OT_BOOL, + .help = N_("do not check XML for ABI compatibility") + }, VIRSH_COMMON_OPT_LIVE(N_("take a live snapshot")), {.name = NULL} }; @@ -177,6 +181,8 @@ cmdSnapshotCreate(vshControl *ctl, const vshCmd *cmd) flags |= VIR_DOMAIN_SNAPSHOT_CREATE_ATOMIC; if (vshCommandOptBool(cmd, "live")) flags |= VIR_DOMAIN_SNAPSHOT_CREATE_LIVE; + if (vshCommandOptBool(cmd, "force")) + flags |= VIR_DOMAIN_SNAPSHOT_CREATE_NOCHECK; if (!(dom = virshCommandOptDomain(ctl, cmd, NULL))) goto cleanup; diff --git a/tools/virsh.pod b/tools/virsh.pod index db72343159..a8cd15bad2 100644 --- a/tools/virsh.pod +++ b/tools/virsh.pod @@ -4570,7 +4570,7 @@ used to represent properties of snapshots. =item B<snapshot-create> I<domain> [I<xmlfile>] {[I<--redefine> [I<--current>]] | [I<--no-metadata>] [I<--halt>] [I<--disk-only>] [I<--reuse-external>] -[I<--quiesce>] [I<--atomic>] [I<--live>]} +[I<--quiesce>] [I<--atomic>] [I<--live>] [I<--force>]} Create a snapshot for domain I<domain> with the properties specified in I<xmlfile>. Normally, the only properties settable for a domain snapshot @@ -4629,6 +4629,12 @@ the guest is running. Both disk snapshot and domain memory snapshot are taken. This increases the size of the memory image of the external snapshot. This is currently supported only for full system external snapshots. +If I<--force> is specified, then the usual ABI compatibility check that +would be done on I<--redefine> will be skipped allowing to update the +domain definition stored in the snapshot metadata arbitrarily. +Note: this obviously involves extra risk, supplying I<--force> assures +libvirt that the snapshot will be compatible with the new configuration. + Existence of snapshot metadata will prevent attempts to B<undefine> a persistent domain. However, for transient domains, snapshot metadata is silently lost when the domain quits running (whether -- 2.17.1

3 7