virHostdevReattachPCIDevice() is a static that simply does
a wait loop with virPCIDeviceWaitForCleanup() before
calling virPCIDeviceReattach().
This loop traces back to commit d1e5676c0d, aiming to
solve a race condition between Libvirt returning the
device back to the host and QEMU trying to access it in
the meantime, which resulted in QEMU exiting on error
and killing the guest. This happens because device_del
is asynchronous, returning OK even if the guest didn't
release the device. Commit 01abc8a1b8 moved this code
to qemu_hostdev.c, 82e8dd4cf8 added the pci-stub conditional
for the loop, 899b261127 moved the code to virhostdev.c
where it stood until now.
The intent of this wait loop is still valid: device_del
is still not bullet proof into preventing the conditions
that commit d1e5676c0d aimed to fix, especially when considering
all the architectures we must support. However, this loop
is executed only in virHostdevReattachPCIDevice(), leaving
every other virPCIDeviceReattach() call prone to that error.
Let's move the wait loop code to virPCIDeviceReattach(). This
will:
- make every reattach call safe from this race condition
with the pci-stub;
- allow for a bit of code cleanup (virHostdevReattachPCIDevice()
can be erased, and virHostdevReAttachPCIDevices() can use
virPCIDeviceReattach() directly);
- make it easier to understand the overall reattach mechanisms in
Libvirt, without the risk of a newcomer wondering why reattach
is done slightly different in some instances.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
This code that executes virPCIDeviceReset in all virPCIDevicePtr
objects of a given virPCIDeviceListPtr list is replicated twice
in the code. Putting it in a helper function helps with
readability.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
There is no restriction on maximum value of PCI domain. In fact,
Linux kernel uses plain atomic inc when assigning PCI domains:
drivers/pci/pci.c:static int pci_get_new_domain_nr(void)
drivers/pci/pci.c-{
drivers/pci/pci.c- return atomic_inc_return(&__domain_nr);
drivers/pci/pci.c-}
Of course, this function is called only if kernel was compiled
without PCI domain support or ACPI did not provide PCI domain.
However, QEMU still has the same restriction as us: in
set_pci_host_devaddr() QEMU checks if domain isn't greater than
0xffff. But one can argue that that's a QEMU limitation. We still
want to be able to cope with other hypervisors that don't have
this limitation (possibly).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Currently, the way we format PCI address is using printf-s
precision, e.g. "%.4x". This works if we don't want to print any
value outside of bounds (which is usually the case). However,
turns out, PCI domain can be 0x10000 which doesn't work well with
our format strings. However, if we change the format string to
"%04x" then we still pad small values with zeroes but also we are
able to print values that are larger than four digits. In fact,
this format string used by kernel to print a PCI address:
"%04x:%02x:%02x.%d"
The other three format strings (for bus, device and function) are
changed too, so that we use the same format string as kernel.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The format string for a PCI address is copied over and over
again, often with slight adjustments. Introduce global
VIR_PCI_DEVICE_ADDRESS_FMT macro that holds the formatting string
and use it wherever possible.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In near future, the length restriction of PCI domain is going to
be lifted. This means that our assumption that PCI address is 13
bytes long is no longer true. We can avoid this problem by making
@name dynamically allocated and thus not bother with actual
length of stringified PCI address.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
This function declares @ret variable and then uses
VIR_STEAL_PTR() to avoid freeing temporary variable @dev which is
constructed. Well, as of 267f1e6da5 we have VIR_RETURN_PTR()
macro so that we can avoid this pattern.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
While it's true that older QEMUs were not able to deal with PCI
domains, we don't support those versions anymore (see
4a42ece13a). Therefore it is safe to always format fully
expanded PCI address. Format PCI domain always as it will
simplify next commits.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
A new algorithm for detecting the vcpus and monitor type conflicts
between new monitor an existing allocation and monitor groups.
After refactoring, since we are verifying both @vcpus and monitor
type @tag at the same time, the validating function name has been
renamed from virDomainResctrlMonValidateVcpus to
virDomainResctrlValidateMonitor.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Export virResctrlMonitorGetStats and make
virResctrlMonitorGetCacheOccupancy obsoleted.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Refactor 'virResctrlMonitorStats' to track multiple statistical
records.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Refactor and rename 'virResctrlMonitorFreeStats' to
'virResctrlMonitorStatsFree' to free one
'virResctrlMonitorStatsPtr' object.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
'virResctrlAllocIsEmpty' checks if cache allocation or memory
bandwidth allocation settings are specified in configuration
file. It is not proper to be used in checking memory bandwidth
allocation is specified in XML settings because this function
could not distinguish memory bandwidth allocations from cache
allocations.
Here using the local variable @n, which indicates the cache
allocation groups or memory bandwidth groups depending on the
context it is in, to decide if append a new @resctrl object.
If @n is zero and no monitors groups specified in XML, then
we should not append a new @resctrl object to @def->resctrls.
This kind of replacement is also more efficient and avoiding
a long function calling path.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Let 'virDomainResctrlVcpuMatch' to retrieve a pointer of
virDomainResctrlDefPtr in its third parameter instead
of virResctrlAllocPtr, if @vcpus is matched with the vcpus
of some resctrl allocation in list of @def->resctrls.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Creating object and judging if it is successfully created in fewer
lines.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
code cleanup for 'virDomainCachetuneDefParse' and
'virDomainMemorytuneDefParse'.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
'default monitor of an allocation' is defined as the resctrl
monitor group that created along with an resctrl allocation,
which is created by resctrl file system. If the monitor group
specified in domain configuration file is happened to be a
default monitor group of an allocation, then it is not necessary
to create monitor group since it is already created. But if
an monitor group is not an allocation default group, you
should create the group under folder
'/sys/fs/resctrl/mon_groups' and fill the vcpu PIDs to 'tasks'
file.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Fortunately, the code that handles metadata getting or setting is
driver agnostic, so all that is needed from individual hypervisor
drivers is to call the right functions.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1732306
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
"virsh console" on macOS cannot attach to a domain and it doesn't matter if
it's local or remote domain:
$ ~ virsh console vm
Connected to domain vm
Escape character is ^]
error: internal error: unable to wait on console condition
The error comes from pthread_cond_wait that fails with EINVAL. The mutex
in the parent is not initialized with pthread_mutex_init and it results
in silent failure of pthead_mutex_lock and the attach failure.
Fixes: 98361cc3b9 ("tools: console: make console virLockableObject")
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
In v5.6.0-rc1~347 I've mistakenly messed up news.xml as the
change I wanted to promote was added into a comment (I blame git
rebase for that). Anyway, restore the original state of the
comment so it can be copied again.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
readline-devel is an optional build dependency; when it is not
present, the output of 'virsh <<EOF ... EOF' is different in that the
input provided by the user is not echoed, and prompts become
interleaved on the same line as actual output, which in turn causes
the sed doing prompt filtering to mess up:
| ./virsh-snapshot
| --- exp 2019-07-31 18:42:31.107399428 -0300
| +++ out.cooked 2019-07-31 18:42:31.108399437 -0300
| @@ -1,8 +1,3 @@
| -
| -
| -Domain snapshot s3 created from 's3.xml'
| -Domain snapshot s2 created from 's2.xml'
| -Name: s2
| Domain: test
| Current: yes
| State: running
Maybe we should fix virsh in interactive mode to echo regardless of
whether readline-devel was used, but the quicker fix is to make the
test use 'virsh "..."' rather than reading its input from stdin.
Reported-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Tested-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Checkpoints are definitely a news-worthy addition, even if the
virDomainBackup API is not going to make it until a later release.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
I messed up formatting during conflict resolution across rebasing
while preparing my checkpoint patches :)
Signed-off-by: Eric Blake <eblake@redhat.com>
hv-spinlocks is not a CPUID feature and should not be checked as such.
While starting a domain with hv-spinlocks enabled, we would report a
warning about unsupported hyperv spinlocks feature even though it was
set properly.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Introduced by: commit e9528f41c6
'msrs' is a feature unrelated to Hyper-V Enlightenments, the commit message
which added it and the test have it right:
<features>
...
<msrs unknown='ignore'>
...
</features>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
SynIC stands for 'Synthetic Interrupt Controller', it is not a NIC. Fix the
spelling in accordance with Hypervisor Top Level Functional Specification.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
The example XML we have contains all other Hyper-V Enlightenments but
'stimer' is missing.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
CI flagged a failing mingw build, due to:
In file included from ../../src/conf/checkpoint_conf.c:24:
../gnulib/lib/configmake.h:8:17: error: expected identifier or '(' before string constant
8 | #define DATADIR "/usr/i686-w64-mingw32/sys-root/mingw/share"
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
As previously learned in commits bd205a90 and 976abdf6, gnulib's
configmake.h header does #define DATADIR "string...", while mingw's
<winsock2.h> expects to declare a type named DATADIR. As long as the
mingw system header is included first before configmake.h, the two
uses do not conflict, but until gnulib is patched to make configmake.h
automatically work around the issue, our immediate fix is the
workaround of rearranging our include order to insure no conflict.
Copy the paradigm used in domain_conf.c of using <unistd.h> to trigger
the indirect inclusion of <winsock2.h> on mingw.
Fixes: 1a4df34a
Signed-off-by: Eric Blake <eblake@redhat.com>
Turning a NULL URI instead the empty string is very misleading when
reading the debug logs as the distinction between the two is
functionally important.
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The recently added test27 spawns commandhelper. This is fine,
except, one of the things that commandhelper does is it records
arguments it was spawn with into commandhelper.log. Other test
cases then use checkoutput() to compare the arguments against the
expected ones and also unlink() the log file. However, test27()
is not doing that and thus it leaves the file behind. This
breaks distcheck.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
These test cases will start failing once the test driver provides
implementation for the virDomainGetCPUStats API.
Signed-off-by: Ilias Stamatis <stamatis.iliass@gmail.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
The returned array of qemuMonitorJobInfo structs must be freed.
164 (16 direct, 148 indirect) bytes in 1 blocks are definitely lost in loss record 64 of 84
at 0x4A3568B: realloc (vg_replace_malloc.c:826)
by 0x4D888BD: virReallocN (viralloc.c:244)
by 0x4D889B3: virExpandN (viralloc.c:293)
by 0x4D88C87: virInsertElementsN (viralloc.c:435)
by 0x214004: qemuMonitorJSONGetJobInfo (qemu_monitor_json.c:9185)
by 0x148B3F: testQueryJobs (qemumonitorjsontest.c:2979)
by 0x14C192: virTestRun (testutils.c:174)
by 0x14BF36: mymain (qemumonitorjsontest.c:3286)
by 0x14E256: virTestMain (testutils.c:1096)
by 0x14BFD9: main (qemumonitorjsontest.c:3298)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
ACKed-by: Peter Krempa <pkrempa@redhat.com>
Originally the names of the KVM CPU features were only used internally
for looking up their CPUID bits. So we used "__kvm_" prefix for them to
make sure the names do not collide with normal CPU features stored in
our CPU map.
But with QEMU 4.1 we check which features were enabled or disabled by a
freshly started QEMU process using their names rather than their CPUID
bits (mostly because of MSR features). Thus we need to change our made
up internal names into the actual names used by QEMU.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Most of the internally defined KVM CPUID features are not actually used
by libvirt. The QEMU driver may enable or disable them on the command
line, but we don't check for the associated CPU properties or CPUID
bits. They would be useless with QEMU 4.1 anyway since their names were
only remotely similar to the actual feature names.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
All the features are hyperv features even though they are provided by
KVM with QEMU. The "KVM" part in the macro names does not make a lot of
sense.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Starting with QEMU 4.1, we're using the canonical feature names on the
command line and avoid aliases to prepare for possible deprecation of
all aliases in QEMU. But we do so only for features from our CPU map,
hyperv features defined in the code were unchanged and this patch fixes
it. Some features use "hv-" prefix unconditionally because they were
introduced recently enough to always support spelling with a dash.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Originally the names of the hyperv CPU features were only used
internally for looking up their CPUID bits. So we used "__kvm_hv_"
prefix for them to make sure the names do not collide with normal CPU
features stored in our CPU map.
But with QEMU 4.1 we check which features were enabled or disabled by a
freshly started QEMU process using their names rather than their CPUID
bits (mostly because of MSR features). Thus we need to change our made
up internal names into the actual names used by QEMU. Most of the names
are only used with QEMU 4.1 and newer and the reset was introduced with
QEMU recently enough to already support spelling with "-". Thus we don't
need to define them as "hv_*" with a translation to "hv-*" for new QEMU.
Without this patch libvirt would mistakenly report all hyperv features
as unavailable and refuse to start any domain using them with QEMU 4.1.
Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
