The only case where the hardware capabilities influence the result
is when no <gic/> element was provided.
The test programs now ensure both that the correct GIC version is
picked in that case, and that hardware capabilities are not taken
into account when the user has already picked a GIC version.
Now that we choose the GIC version based on hardware features when
no <gic/> element has been provided, we need a way to fake the GIC
capabilities of the host.
Update the qemuxml2argv and qemuxml2xml tests to allow this.
When the <gic/> element in not present in the domain XML, use the
domain capabilities to figure out what GIC version is usable and
choose that one automatically.
This allows guests to be created on hardware that only supports
GIC v3 without having to update virt-manager and similar tools.
Keep using the default GIC version if the <gic/> element has been
added to the domain XML but no version has been specified, as not
to break existing guests.
The only QEMU versions that don't have such capability are <0.12,
which we no longer support anyway.
Additionally, this solves the issue of some QEMU binaries being
reported as not having such capability just because they lacked
the {kvm-}pci-assign QMP object.
We don't need them any longer; moreover, the previous structure
made it very easy for bugs to slip in, by having the result of one
check influence the following one.
By placing the check for "$with_init_script" = check front and
center, hopefully this won't happen (as easily) again.
Most distributions, including RHEL, have switched to systemd,
so we should detect it and act accordingly. This also means
that 'systemd+redhat' should be preferred to legacy 'redhat'.
Our witness for the check is the availability of the systemctl
command on the host.
If we didn't find a match, either because we're cross compiling
or because we're not building on RHEL, we won't install any
init script.
Make sure this is reported correctly in the configure summary.
Since commit 9b77ce63f1 we create a .in file while building all
man pages, including those in the tools/ directory; update the
ignore patterns to take this change into account.
The new ignore patterns are generic enough that we can get rid of
a few existing ones as well.
All top-level functions have been moved to this class.
On top of that, the app.warning() method has been defined,
so that calls to it - already present in rebuild() - can
actually succeed.
Since commit d195cffa2e, both $(srcdir) and $(abs_builddir)
are passed to the apibuild.py script; however, since the
former is a relative path and the latter an absolute one, the
script might not be able to detect whether they point to the
same location.
Pass both as relative paths to avoid the issue.
No need to special-case this one: we can add it to EXTRA_DIST so
that it will be shipped in any case, and if WITH_LOGIN_SHELL
happens to be enabled we mark it for installation as well.
We can't use eg. @sysconfdir@ directly in the .pod file, because
pod2man(1) will interpret that as a variable name and format it
accordingly.
Instead, we use eg. SYSCONFDIR and use a subsequent sed(1) call
to turn it into the expected @sysconfdir@.
The generated man pages may contain information that depend on
the build configuration, so they should not be shipped but
rather always built on the user's machine.
After this commit, all man pages are generated using the same two
steps:
1. Process a source $command.pod file with pod2man(1) to obtain
a valid man page in $command.$section.in
2. Process $command.$section.in with sed(1) to obtain the final
man page in $command.$section
Instead of embedding the pod information inside the respective
source files, store them in separate files.
This allows us to reduce the number of custom build rules as
most of the information can be inferred for the file name;
moreover, text editors are more likely to use proper syntax
highlighting for standalone pod files.
When describing attributes and elements, we mostly stick to
a certain pattern; however, there are a few cases when the
information is not presented in the usual way.
Since there doesn't seem to be any reason not to follow the
tried and true formula, rework those bits to fit the rest of
the documentation.
Commit 61b070cf20 cleaned up a number of cases where the <dt>
element was used to document symbols, but the symbol itself was
not inside a <code> element.
To make sure we don't end up having to clean up again a few
months from now, introduce a syntax-check rule that can spot
such mistakes.
All existing exceptions are marked as such, with either file
or line granularity depending on the case.
FreeBSD's sed(1) doesn't support using "\n" to insert a newline,
so the installed default.xml file ends up containing a literal
"n" between tags; to work around this problem, add a tr(1)
invocation as suggested by the sed FAQ[1].
[1] http://sed.sourceforge.net/sedfaq4.html (4.1 c)
According to the autoconf manual, using '$(LN_S) -f' is not
portable; remove the target explicitly beforehand to work around
this limitation.
Adjust some slightly awkward indentation while at it.
The current rule fails if the target already exists:
cd /home/jenkins/build/libvirt/lib && \
ln -s libnss_libvirt.so.1 nss_libvirt.so.1
ln: nss_libvirt.so.1: File exists
Makefile:3357: recipe for target 'install-exec-hook' failed
However, all other rules concerned with installation are
idempotent and will happily overwrite an existing target,
so this one should as well.
Commit 3a773c43c8 introduced the testCompareNetXML2XMLResult
enumeration; however, in one instance the result variable was
assigned a value from the very similar testCompareDocXML2XMLResult
enumeration, leading to a build error.
networkxml2xmltest.c:33:42: error:
implicit conversion from enumeration type 'testCompareDomXML2XMLResult'
to different enumeration type 'testCompareNetXML2XMLResult'
[-Werror,-Wenum-conversion]
testCompareNetXML2XMLResult result = TEST_COMPARE_DOM_XML2XML_RESULT_SUCCESS;
~~~~~~ ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Use the proper value (TEST_COMPARE_NET_XML2XML_RESULT_SUCCESS) instead.
QEMU introduced the query-gic-capabilities QMP command
with commit 4468d4e0f383: use the command, if available,
to probe available GIC capabilities.
The information obtained is stored in a virQEMUCaps
instance, and will be later used to fill in a
virDomainCaps instance.
The struct contains a single boolean field, 'supported':
the meaning of this field is too generic to be limited to
devices only, and in fact it's already being used for
other things like loaders and OSs.
Instead of trying to come up with a more generic name just
get rid of the struct altogether.
To prevent the error messages in cfg.mk from triggering the very
same rules they're supposed to explain, we split the message in
the middle of a symbol name, ending up with stuff like
'I am a me'ssage
Instead of relying on these quotation tricks, simply exclude
cfg.mk from the relevant checks.
virt-host-validate, just like virt-login-shell, doesn't make sense
on Windows, so we should avoid building it.
Make the tool optional and build it by default on all platforms
except Windows, erroring out if the user attempts to build it
anyway.
Being consistent is nice, especially when it comes to defining our
regular expression, where using single quotes instead of double
quotes allows us to leave out a few backslashes.
Changing this required altering a few error messages.
The only remaining use of double quotes is one where they are
actually required for the check to work.
Simply checking whether the cgroup name appears somewhere inside
/proc/self/cgroup is enough most of the time, but there are some
corner cases that require a more mindful parsing.
The check is supposed to stop users from trying to compile
virt-login-shell on Windows by erroring out during the
configure phase; however, there are two flaws in it:
* the value of "x$with_win" is compared to "yes" instead
of "xyes" (note the "x" in the first string)
* "test" is not being used, so the script will actually
try to run a command called "x$with_win" instead of
performing string comparison
This patch fixes both issues.
The existing code is built on the assumption that no cgroup
name can appear as part of another cgroup name; moreover, cgroups
are expected to always be listed in a specific order.
If that's not the case, eg. 'cpuacct' is listed before 'cpu', the
algorithm fails to detect the cgroup mount point.
Rewrite it to get rid of such assumptions.
virSocketAddrFormat() wants a single pointer, not a double pointer.
Fixes the following compilation error on FreeBSD:
util/virnetdev.c:1448:72: error: incompatible pointer types passing
'virSocketAddr **' to parameter of type 'const virSocketAddr *';
remove & [-Werror,-Wincompatible-pointer-types]
if (VIR_SOCKET_ADDR_VALID(peer) && !(peerstr = virSocketAddrFormat(&peer)))
^~~~~
./util/virsocketaddr.h:92:48: note: passing argument to parameter 'addr' here
char *virSocketAddrFormat(const virSocketAddr *addr);
^
Checking for the availability of this header, just like checking
for the availability of gnutls_rnd(), requires CFLAGS and LIBS to
be set appropriately.
Fixes the following compilation errors on FreeBSD:
qemu/qemu_domain.c:640:16: error: implicit declaration of function
'gnutls_rnd' is invalid in C99 [-Werror,-Wimplicit-function-declaration]
if ((ret = gnutls_rnd(GNUTLS_RND_RANDOM, key, nbytes)) < 0) {
^
qemu/qemu_domain.c:640:27: error: use of undeclared identifier
'GNUTLS_RND_RANDOM'; did you mean 'GNUTLS_CRD_ANON'?
if ((ret = gnutls_rnd(GNUTLS_RND_RANDOM, key, nbytes)) < 0) {
^~~~~~~~~~~~~~~~~
GNUTLS_CRD_ANON
Our use of gnutls_rnd(), introduced with commit ad7520e8, is
conditional to the availability of the <gnutls/crypto.h> header
file.
Such check, however, turns out not to be strict enough, as there
are some versions of GnuTLS (eg. 2.8.5 from CentOS 6) that provide
the header file, but not the function itself, which was introduced
only in GnuTLS 2.12.0.
Introduce an explicit check for the function.
The variables used for storing CFLAGS and LIBS before temporarily
modifying them was consistent when it comes to the name, but not
when it comes to the case.
Make sure names are completely consistent.
The previous value of CFLAGS was saved as old_cflags but later
restored from old_CFLAGS, which is clearly not correct.
Restore CFLAGS from the right variable.
Instead of relying on substring search, tokenize the input
and process each CPU flag separately. This ensures CPU flag
detection will continue to work correctly even if we start
looking for CPU flags whose name might appear as part of
other CPU flags' names.
The result of processing is stored in a virBitmap, which
means we don't have to parse /proc/cpuinfo in its entirety
for each single CPU flag we want to check.
Moreover, use of the newly-introduced virHostValidateCPUFlag
enumeration ensures we don't go looking for random CPU flags
which might actually be simple typos.
This reverts commit ee4cfb5643.
Since we're still not persisting our bookkeeping lists across
daemon restarts, we might have lost some information
virPCIDeviceReattach() relies on, for example whether the
device needs to be unbound from the stub driver.
As a result, if the daemon has been restarted in the meantime,
the device might end up remaining bound to the stub driver even
after 'virsh nodedev-reattach' or similar has been called, with
no way of giving it back to the host short of messing with
sysfs behind libvirt's back.
Revert back to the previous behavior of always trying to bind
the device to the host driver, regardless of its status when it
was detached, until persistent bookkeeping lists have been
implemented.
According to MDN[1], 'margin-left' and similar CSS properties,
including 'margin-right', cannot be applied to the '::first-line'
pseudo-element, so this rule will never have any effect and can
be safely removed.
[1] https://developer.mozilla.org/en/docs/Web/CSS/::first-line
If the cpu cgroup is not found when validating an host for
LXC support, virt-host-validate will suggest to enable the
CONFIG_CGROUP_SCHED kconfig option.
The appropriate option is really CONFIG_CGROUP_CPU. The
QEMU checks already get that right, so no changes needed.
Ensure the code behaves properly even for situations that were not
being considered before, such as simply detaching devices from the
host without attaching them to a guest and attaching devices as
managed even though they had already been manually detached from
the host.
Instead of forcing the values for the unbind_from_stub, remove_slot
and reprobe properties, look up the actual device and use that when
calling virPCIDeviceReattach().
This ensures the device is restored to its original state after
reattach: for example, if it was not bound to any driver before
detach, it will not be bound forcefully during reattach.
We would be just fine looking up the information in pcidevs most
of the time; however, some corner cases would not be handled
properly, so look up the actual device instead.
After this patch, ownership of virPCIDevice instances is very easy
to keep track of: for each host PCI device, the only instance that
actually matters is the one inside one of the bookkeeping list.
Whenever some operation needs to be performed on a PCI device, the
actual device is looked up first; when this is not the case, a
comment explains the reason.
Unmanaged devices, as the name suggests, are not detached
automatically from the host by libvirt before being attached to a
guest: it's the user's responsability to detach them manually
beforehand. If that preliminary step has not been performed, the
attach operation can't complete successfully.
Instead of relying on the lower layers to error out with cryptic
messages such as
error: Failed to attach device from /tmp/hostdev.xml
error: Path '/dev/vfio/12' is not accessible: No such file or directory
prevent the situation altogether and provide the user with a more
useful error message.
Unmanaged devices are attached to guests in two steps: first,
the device is detached from the host and marked as inactive;
subsequently, it is marked as active and attached to the guest.
If the daemon is restarted between these two operations, we lose
track of the inactive device.
Steps 5 and 6 of virHostdevPreparePCIDevices() already subtly
take care of this situation, but some planned changes will make
it so that's no longer the case. Plus, explicit is always better
than implicit.
These comments explain the difference between a virPCIDevice
instance used for lookups and an actual device instance; some
information is also provided for specific uses.
virHostdevGetPCIHostDeviceList() is similar but does not filter out
devices that are not in the active list; that said, we are looking
up the device in the active list just a few lines after anyway, so
we might as well just keep a single function around.
This also helps stress the fact the objects contained in pcidevs are
only for looking up the actual devices, which is something later
commits will make even more explicit.
We're in the hostdev module, so mgr is not an ambiguous name, and
in fact it's already used in some cases. Switch all the code over.
Take the chance to shorten declaration of
virHostdevIsPCINodeDeviceUsedData structures.
When we want to look up a device in a device list and we already
have the IDs from another source, we can simply use
virPCIDeviceListFindByIDs() instead of creating a temporary device
object.
If 'last_processed_hostdev_vf != -1' is false then, since the
loop counter 'i' starts at 0, 'i <= last_processed_hostdev_vf'
can't possibly be true and the loop body will never be executed.
However, since 'i' is unsigned and 'last_processed_hostdev_vf'
is signed, we can't just get rid of the check completely; what
we can do is move it outside of the loop to avoid checking its
value on every iteration and cluttering the actual loop
condition.
The comment claimed that virPCIDeviceReattach() does not reattach
a device to the host driver; except it actually does, so the
comment is just confusing and we're better off removing it.
Replace the term "loop" with the more generic "step". This allows us
to be more flexible and eg. have a step that consists in a single
function call.
Don't include the number of steps in the first comment of the
function, so that we can add or remove steps without having to worry
about keeping that comment in sync.
For the same reason, remove the summary contained in that comment.
Clean up some weird vertical spacing while we're at it.
Andrea Bolognani