Isolation groups will eventually allow us to make sure certain
devices, eg. PCI hostdevs, are assigned to guest PCI buses in
a way that guarantees improved isolation, error detection and
recovery for machine types and hypervisors that support it,
eg. pSeries guest on QEMU.

This patch merely defines storage for the new information
we're going to need later on and makes sure it is passed from
the hypervisor driver (QEMU / bhyve) down to the generic PCI
address allocation code.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

Now that we have a bit more control, let's convert our object into
a lockable object and let that magic handle the create and lock/unlock.

This also involves creating a virNodeDeviceEndAPI in order to handle
the object cleanup for API's that use the Add or Find API's in order
to get a locked/reffed object. The EndAPI will unlock and unref the
object returning NULL to indicate to the caller to not use the obj.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Move the structures to withing virnodedeviceobj.c

Move the typedefs from node_device_conf to virnodedeviceobj.h
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

In an overall effort to privatize access to virNodeDeviceObj and
virNodeDeviceObjList into the virnodedeviceobj module, move the
object list parsing from node_device_driver and replace with a
call to a virnodedeviceobj helper. This follows other similar
APIs/helpers which peruse the object list looking for some specific
data in order to get/return an @device (virNodeDevice) object to
the caller.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

We're about to move the call to nodeDeviceSysfsGetSCSIHostCaps from
node_device_driver into virnodedeviceobj, so move the guts of the code
from the driver specific node_device_linux_sysfs into its own API
since virnodedeviceobj cannot callback into the driver.

Nothing in the code deals with sysfs anyway, as that's hidden by the
various virSCSIHost* and virVHBA* utility function calls.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Create local @obj and @def for the API's rather than referencing the
devs->objs[i][->def->].  It'll make future patches easier to read.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Ensure that any function that walks the node device object list is prefixed
by virNodeDeviceObjList.

Also, modify the @filter param name for virNodeDeviceObjListExport to
be @aclfilter.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

In preparation to make things private, make the ->devs be pointers to a
virNodeDeviceObjList and then manage everything inside virnodedeviceobj
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Create an allocator for the virNodeDeviceObjPtr - include setting up
the mutex, saving the virNodeDeviceDefPtr, and locking the return object.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

A virNodeDeviceObjPtr is an @obj

A virNodeDeviceObjListPtr is a @devs

A virNodeDevicePtr is a @device
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

A virNodeDeviceObjPtr is an @obj

A virNodeDeviceObjListPtr is an @devs
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

 - In testDestroyVport rather than use a cleanup label, just return -1
   immediately since nothing else is needed.

 - In testStoragePoolDestroy, if !privpool, then just return -1 since
   nothing else will happen anyway.

 - Rather than "goto cleanup;" on failure to virNodeDeviceObjFindByName
   an @obj, just return directly.  This then allows the cleanup: label code
   to not have to check "if (obj)" before calling virNodeDeviceObjUnlock.
   This also simplifies some exit logic...

 - In testNodeDeviceObjFindByName use an error: label to handle the failure
   and don't do the ncaps++ within the VIR_STRDUP() source target index.
   Only increment ncaps after success. Easier on eyes at error label too.

 - In testNodeDeviceDestroy use "cleanup" rather than "out" for the goto
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Rather than passing the object to be removed by reference, pass by value
and then let the caller decide whether or not the object should be free'd
and how to handle the logic afterwards. This includes free'ing the object
and/or setting the local variable to NULL to prevent subsequent unexpected
usage (via something like virNodeDeviceObjRemove in testNodeDeviceDestroy).

For now this function will just handle the remove of the object from the
list for which it was placed during virNodeDeviceObjAssignDef.

This essentially reverts logic from commit id '61148074' that free'd the
device entry on list, set *dev = NULL and returned. Thus fixing a bug in
node_device_hal.c/dev_refresh() which would never call dev_create(udi)
since @dev would have been set to NULL.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Recent refactors made it so that the function may use uninitialized
pointer, but it actually wanted to use a different variable and value
at all.
Signed-off-by: NMartin Kletzander <mkletzan@redhat.com>

This reverts commit b3e71a88.

As it turns out this ends up very badly as the @def could be Free'd
even though it's owned by @obj as a result of the AssignDef.

driverpath, allocated by virAsprintfQuiet, was not freed and leaked.
Signed-off-by: NZhipeng Lu <lu.zhipeng@zte.com.cn>

Create a common API to handle the instantiation path filter lookup.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

New API will be virNWFilterInstantiateFilterInternal as it's called from
the virNWFilterInstantiateFilter and virNWFilterUpdateInstantiateFilter.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Rename to virNWFilterInstantiateFilterUpdate and alter the callers to not
have one parameter per line.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Rename to virNWFilterDoInstantiate to better describe the action.

Also fix the @vmuuid parameter to not have the ATTRIBUTE_UNUSED since it
is used in the call to virNWFilterDHCPSnoopReq.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Use @nwfilter always for the name
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

No need to goto cleanup and check "if (obj)" if we know (obj) isn't there,
so just return immediately.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Merge code into virNWFilterSaveConfig
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Should compare < 0 to be correct.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>

Change from

  'controls whether to discard ... requests are ignored'

to

  'controls whether discard requests ... are ignored'

Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

When looking for slots suitable for a PCI device, libvirt
might need to add an extra PCI controller: for pSeries guests,
we want that extra controller to be a PHB (pci-root) rather
than a PCI bridge.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

PCI bus has to be numbered sequentially, and no index can be
missing, so libvirt will fill in the blanks automatically for
the user.

Up until now, it has done so using either pci-bridge, for machine
types based on legacy PCI, or pcie-root-port, for machine types
based on PCI Express. Neither choice is good for pSeries guests,
where PHBs (pci-root) should be used instead.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

These tests demonstrate that, while it's now possible for the
user to create PHB explicitly and manually assign devices to
them, libvirt still defaults to extending the guest PCI
topology using PCI bridges and making suboptimal device
placement choices.

The next few commits will improve on these behaviors and the
tests outputs will automatically be updated to reflect this.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

Now that the multi-phb support series is in, work on the TODO at
qemuDomainGetMemLockLimitBytes() to arrive at the correct memlock limit
value.
Signed-off-by: NShivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
Reviewed-by: NAndrea Bolognani <abologna@redhat.com>

Additional PHBs (pci-root controllers) will be created for
the guest using the spapr-pci-host-bridge QEMU device, if
available; the implicit default PHB, while present in the
guest configuration, will be skipped.

Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1431193Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

Usually, a controller with alias 'x' will create a bus with the
same name; however, the bus created by a PHBs with alias 'x' will
be named 'x.0' instead, so we need to account for that.

As an exception to the exception, the implicit PHB that's added
automatically to every pSeries guest creates the 'pci.0' bus.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

This new capability can be used to detect whether a QEMU
binary supports the spapr-pci-host-bridge controller.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

pSeries guests will soon need the new information; luckily,
we can figure it out automatically most of the time, so
users won't have to worry about it.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

Adding it to the virDomainControllerPCIModelName enumeration
is enough for existing code to handle it, so parsing and
formatting will work without further tweaking.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

pSeries guests will soon be allowed to have multiple
PHBs (pci-root controllers), meaning the current check
on the controller index no longer applies to them.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

pSeries guests will soon be allowed to have multiple
PHBs (pci-root controllers), which of course means that
all but one of them will have a non-zero index; hence,
we'll need to relax the current check.

However, right now the check is performed in the conf
module, which is generic rather than tied to the QEMU
driver, and where we don't have information such as the
guest machine type available.

To make this change of behavior possible down the line,
we need to move the check from the XML parser to the
drivers. Luckily, only QEMU and bhyve are using PCI
controllers, so this doesn't result in much duplication.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>

Moving the check and rewriting it this way doesn't alter
the current behavior, but will allow us to special-case
pci-root down the line.
Signed-off-by: NAndrea Bolognani <abologna@redhat.com>
Reviewed-by: NLaine Stump <laine@laine.org>