Commits f83c7c88 and 6eb1f2b9 broke the build on FreeBSD and OSX because
of symbols being undefined for those platforms.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
This is a replacement for the existing udevPCIGetMdevTypesCap which is
static to the udev backend. This simple helper constructs the sysfs path
from the device's base path for each mdev type and queries the
corresponding attributes of that type.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Right-aligning backslashes when defining macros or using complex
commands in Makefiles looks cute, but as soon as any changes is
required to the code you end up with either distractingly broken
alignment or unnecessarily big diffs where most of the changes
are just pushing all backslashes a few characters to one side.
Generated using
$ git grep -El '[[:blank:]][[:blank:]]\\$' | \
grep -E '*\.([chx]|am|mk)$$' | \
while read f; do \
sed -Ei 's/[[:blank:]]*[[:blank:]]\\$/ \\/g' "$f"; \
done
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
The mlx4 (Mellanox) netdev driver implements the sysfs phys_port_id
file for both VFs and PFs, so you can find the VF netdev plugged into
the same physical port as any given PF netdev by comparing the
contents of phys_port_id of the respective netdevs. That's what
libvirt does when attempting to find the PF netdev for a given VF
netdev (or vice versa).
Most other netdev's drivers don't implement phys_port_id, so the file
is visible in sysfs directory listing, but attempts to read it result
in ENOTSUPP. In these cases, libvirt is unable to read phys_port_id of
either the PF or the VF, so it just returns the first entry in the
PF/VF's list of netdevs.
But we've found that the i40e driver is in between those two
situations - it implements phys_port_id for PF netdevs, but doesn't
implement it for VF netdevs. So libvirt would successfully read the
phys_port_id of the PF netdev, then try to find a VF netdev with
matching phys_port_id, but would fail because phys_port_id is NULL for
all VFs. This would result in a message like the following:
Could not find network device with phys_port_id '3cfdfe9edc39'
under PCI device at /sys/class/net/ens4f1/device/virtfn0
To solve this problem in a way that won't break functionality for
anyone else, this patch saves the first netdev name we find for the
device, and returns that if we fail to find a netdev with the desired
phys_port_id.
Mellanox ConnectX-3 dual port SRIOV NICs present a bit of a challenge
when assigning one of their VFs to a guest using VFIO device
assignment.
These NICs have only a single PCI PF device, and that single PF has
two netdevs sharing the single PCI address - one for port 1 and one
for port 2. When a VF is created it can also have 2 netdevs, or it can
be setup in "single port" mode, where the VF has only a single netdev,
and that netdev is connected either to port 1 or to port 2.
When the VF is created in dual port mode, you get/set the MAC
address/vlan tag for the port 1 VF by sending a netlink message to the
PF's port1 netdev, and you get/set the MAC address/vlan tag for the
port 2 VF by sending a netlink message to the PF's port 2 netdev. (Of
course libvirt doesn't have any way to describe MAC/vlan info for 2
ports in a single hostdev interface, so that's a bit of a moot point)
When the VF is created in single port mode, you can *set* the MAC/vlan
info by sending a netlink message to *either* PF netdev - the driver
is smart enough to understand that there's only a single netdev, and
set the MAC/vlan for that netdev. When you want to *get* it, however,
the driver is more accurate - it will return 00:00:00:00:00:00 for the
MAC if you request it from the port 1 PF netdev when the VF was
configured to be single port on port 2, or if you request if from the
port 2 PF netdev when the VF was configured to be single port on port
1.
Based on this information, when *getting* the MAC/vlan info (to save
the original setting prior to assignment), we determine the correct PF
netdev by matching phys_port_id between VF and PF.
(IMPORTANT NOTE: this implies that to do PCI device assignment of the
VFs on dual port Mellanox cards using <interface type='hostdev'>
(i.e. if you want the MAC address/vlan tag to be set), not only must
the VFs be configured in single port mode, but also the VFs *must* be
bound to the host VF net driver, and libvirt must use managed='yes')
By the time libvirt is ready to set the new MAC/vlan tag, the VF has
already been unbound from the host net driver and bound to
vfio-pci. This isn't problematic though because, as stated earlier,
when a VF is created in single port mode, commands to configure it can
be sent to either the port 1 PF netdev or the port 2 PF netdev.
When it is time to restore the original MAC/vlan tag, again the VF
will *not* be bound to a host net driver, so it won't be possible to
learn from sysfs whether to use the port 1 or port 2 PF netdev for the
netlink commands. And again, it doesn't matter which netdev you
use. However, we must keep in mind that we saved the original settings
to a file called "${PF}_${VFNUM}". To solve this problem, we just
check for the existence of ${PF1}_${VFNUM} and ${PF2}_${VFNUM}, and
use whichever one we find (since we know that only one can be there)
A single PCI device may have multiple netdevs associated with it. Each
of those netdevs will have a different phys_port_id entry in
sysfs. This patch modifies virPCIGetNetName() to allow selecting one
of the potential many netdevs in two different ways:
1) by setting the "idx" argument, the caller can select the 1st (0),
2nd (1), etc. netdev from the PCI device's net subdirectory.
2) If the physPortID arg is set (to a null-terminated string) then
virPCIGetNetName() returns the netdev that has that phys_port_id in
the sysfs file of the same name in the netdev's directory.
Commit 8e09663 "pci: recognize/report GEN4 (PCIe 4.0) card 16GT/s Link
speed" introduced another speed into enum, but mistakenly also altered
field width, so one bit of link width was included there.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Without this added enum value, nodedev-dumpxml of a GEN4 (PCIe 4.0)
card will fail (due to the unrecognized link speed), and since
nodedev-detach and nodedev-reattach internally do a dumpxml+parse,
they will also fail. With this patch, all those operations succeed.
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
This function unbinds a device from its driver, then immediately
rebinds it to its driver again. The code for this new function is just
the 2nd half of virPCIDeviceBindWithDriverOverride(), so that
function's 2nd half is replaced with a call to virPCIDeviceRebind().
...and cleanup the callers to report it when it *is* an error.
In many cases It's useful for virPCIGetNetName() to not log an error
and simply return a NULL pointer when the given device isn't bound to
a net driver (e.g. we're looking at a VF that is permanently bound to
vfio-pci). The existing code would silently return an error in this
case, which could eventually lead to the dreaded "An error occurred
but the cause is unknown" log message.
This patch changes virPCIGetNetName() to still return success if the
device simply isn't bound to a net driver, and adjusts all the callers
that require a non-null netname to check for that condition and log an
error when it happens.
It is destructive to attempt reset on a pci- or cardbus-bridge, the
host can crash. The bridges won't contain any guest data and neither
they can be passed through using vfio/stub. So, no point in allowing a
reset on them.
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
The path to the config file for a PCI device is conventiently stored
in a virPCIDevice object, but that object's contents aren't directly
visible outside of virpci.c, so we need to have an accessor function
for it if anyone needs to look at it.
libvirt uses the new_id PCI sysfs interface to bind a PCI stub driver
to a PCI device. The new_id interface is known to be buggy and racey,
hence a more deterministic interface was introduced in the 3.12 kernel:
driver_override. For more details see
https://www.redhat.com/archives/libvir-list/2016-June/msg02124.html
For more details about the driver_override interface and examples of
its usage, see
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/drivers/pci/pci-driver.c?h=v3.12&id=782a985d7af26db39e86070d28f987cad21313c0
This patch adds support for the driver_override interface by
- adding new virPCIDevice{BindTo,UnbindFrom}StubWithOverride functions
that use the driver_override interface
- renames the existing virPCIDevice{BindTo,UnbindFrom}Stub functions
to virPCIDevice{BindTo,UnbindFrom}StubWithNewid to perserve existing
behavior on new_id interface
- changes virPCIDevice{BindTo,UnbindFrom}Stub function to call one of
the above depending on availability of driver_override
The patch includes a bit of duplicate code, but allows for easily
dropping the new_id code once support for older kernels is no
longer desired.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
The directories we iterate over are unlikely to contain any entries
starting with a dot, other than '.' and '..' which is already skipped
by virDirRead.
Commit c8b1a83605 changed the function, making it
impossible for callers to be able to tell whether a
non-negative return value means "physical function
address found and parsed correctly" or "couldn't find
corresponding physical function".
The important difference between the two being that,
in the latter case, the returned pointer is NULL and
should never, ever be dereferenced.
In order to cope with these changes, the callers
have to be updated.
Rather than return 0/-1 and/or a pointer to some memory, adjust the
helper to just return the allocated structure or NULL on failure.
Adjust the callers in order to handle that
Signed-off-by: John Ferlan <jferlan@redhat.com>
If we get to the error: label and clear out the *virtual_functions[]
pointers and then return w/ error to the caller - the caller has it's
own cleanup of the same array in the out: label which is keyed off the
value of num_virt_fns, which wasn't reset to 0 in the called function
leading to a possible problem.
Just clear the value (found by Coverity)
Signed-off-by: John Ferlan <jferlan@redhat.com>
We had both and the only difference was that the latter also included
information about multifunction setting. The problem with that was that
we couldn't use functions made for only one of the structs (e.g.
parsing). To consolidate those two structs, use the one in virpci.h,
include that in domain_conf.h and add the multifunction member in it.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
If we expose this information, which is one byte in every PCI config
file, we let all mgmt apps know whether the device itself is an endpoint
or not so it's easier for them to decide whether such device can be
passed through into a VM (endpoint) or not (*-bridge).
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1317531
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
The affected functions are:
virPCIDeviceGetManaged()
virPCIDeviceGetUnbindFromStub()
virPCIDeviceGetRemoveSlot()
virPCIDeviceGetReprobe()
Change their return type from unsigned int to bool: the corresponding
members in struct _virPCIDevice are defined as bool, and even the
corresponding virPCIDeviceSet*() functions take a bool value as input
so there's no point in these functions having unsigned int as return
type.
Suggested-by: John Ferlan <jferlan@redhat.com>
Unbinding a PCI device from the stub driver can require several steps,
and it can be useful for debugging to be able to trace which of these
steps are performed and which are skipped for each device.
The name is confusing, and there are just two uses: one is a test case,
and the other will be removed as part of an upcoming refactoring of
the hostdev code.
Most of the changes to the list of active and inactive PCI devices
happen in virHostdev, where they are properly logged.
virPCIDeviceDetach() and virPCIDeviceReattach(), however, change the
inactive list as well, so they should be logging similar messages.
Due to debug logs like this:
virPCIGetDeviceAddressFromSysfsLink:2432 : Attempting to resolve device path from device link '/sys/class/net/eth1/device/virtfn6'
logStrToLong_ui:2369 : Converted '0000:07:00.7' to unsigned int 0
logStrToLong_ui:2369 : Converted '07:00.7' to unsigned int 7
logStrToLong_ui:2369 : Converted '00.7' to unsigned int 0
logStrToLong_ui:2369 : Converted '7' to unsigned int 7
virPCIGetDeviceAddressFromSysfs:1947 : virPCIDeviceAddress 0000:07:00.7
virPCIGetVirtualFunctions:2554 : Found virtual function 7
printed *once for each SR-IOV Virtual Function* of a Physical Function
each time libvirt retrieved the list of VFs (so if the system has 128
VFs, there would be 900 lines of log for each call), the debug logs on
any system with a large number of VFs was dominated by "information"
that was possibly useful for debugging when the code was being
written, but is now useless for debugging of any problem on a running
system, and only serves to obscure the real useful information. This
overkill has no place in production code, so this patch removes it.
This replaces the virPCIKnownStubs string array that was used
internally for stub driver validation.
Advantages:
* possible values are well-defined
* typos in driver names will be detected at compile time
* avoids having several copies of the same string around
* no error checking required when setting / getting value
The names used mirror those in the
virDomainHostdevSubsysPCIBackendType enumeration.
This internal function supports, in theory, binding to a different
stub driver than the one the PCI device has been configured to use.
In practice, it is only ever called like
virPCIDeviceBindToStub(dev, dev->stubDriver);
which makes its second parameter redundant. Get rid of it, along
with the extra string copy required to support it.
Instead of replicating the information (domain, bus, slot, function)
inside the virPCIDevice structure, use the already-existing
virPCIDeviceAddress structure.
For users of the module, this means that the object returned by
virPCIDeviceGetAddress() can no longer be NULL and must no longer
be freed by the caller.
A PCI device may have the capability to setup virtual functions (VFs)
but have them currently all disabled. Prior to this patch, if that was
the case the the node device XML for the device wouldn't report any
virtual_functions capability.
With this patch, if a file called "sriov_totalvfs" is found in the
device's sysfs directory, its contents will be interpreted as a
decimal number, and that value will be reported as "maxCount" in a
capability element of the device's XML, e.g.:
<capability type='virtual_functions' maxCount='7'/>
This will be reported regardless of whether or not any VFs are
currently enabled for the device.
NB: sriov_numvfs (the number of VFs currently active) is also
available in sysfs, but that value is implied by the number of items
in the list that is inside the capability element, so there is no
reason to explicitly provide it as an attribute.
sriov_totalvfs and sriov_numvfs are available in kernels at least as far
back as the 2.6.32 that is in RHEL6.7, but in the case that they
simply aren't there, libvirt will behave as it did prior to this patch
- no maxCount will be displayed, and the virtual_functions capability
will be absent from the device's XML when 0 VFs are enabled.
Convert virPCIDriverDir to return the buffer allocated (or not) and make the
appropriate check in the caller.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Convert virPCIDriverFile to return the buffer allocated (or not) and make the
appropriate check in the caller.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Convert virPCIFile to return the buffer allocated (or not) and make the
appropriate check in the caller.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Basically a getter function which is implemented for accessing the
address fields in virPCIDevice.
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
I noticed this while working on qemuDomainGetBlockInfo. Assigning
a bool value to an int variable compiles fine, but raises red flags
on the maintenance front as it becomes too easy to assign -1 or 2
or any other non-bool value to the same variable.
* cfg.mk (sc_prohibit_int_assign_bool): New rule.
* src/conf/snapshot_conf.c (virDomainSnapshotRedefinePrep): Fix
offenders.
* src/qemu/qemu_driver.c (qemuDomainGetBlockInfo)
(qemuDomainSnapshotCreateXML): Likewise.
* src/test/test_driver.c (testDomainSnapshotAlignDisks):
Likewise.
* src/util/vircgroup.c (virCgroupSupportsCpuBW): Likewise.
* src/util/virpci.c (virPCIDeviceBindToStub): Likewise.
* src/util/virutil.c (virIsCapableVport): Likewise.
* tools/virsh-domain-monitor.c (cmdDomMemStat): Likewise.
* tools/virsh-domain.c (cmdBlockResize, cmdScreenshot)
(cmdInjectNMI, cmdSendKey, cmdSendProcessSignal)
(cmdDetachInterface): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
