Skip to content

L
libvirt

openeuler / libvirt

通知 3
Star 0
Fork 0
L
libvirt

体验新版 GitCode，发现更多精彩内容 >>

切换分支/标签
  1. 09 5月, 2019 6 次提交
  3. 07 5月, 2019 1 次提交
    • E
      conf: Add parameter to virDomainDiskSourceFormat · 1ec3e397
      由 Eric Blake 提交于 
      Commits 4bc42986 and 218c81ea removed virDomainStorageSourceFormat on
the grounds that there were no external callers; however, the upcoming
backup code wants to output a <target> (push mode) or <scratch> (pull
mode) element that is in all other respects identical to a domain's
<source> element, where the previous virDomainStorageSourceFormat fit
the bill nicely. But rather than reverting the commits, it's easier to
just add an additional parameter for the element name to use, and
update all callers.
Signed-off-by: NEric Blake <eblake@redhat.com>
      1ec3e397
  5. 17 4月, 2019 3 次提交
  7. 15 4月, 2019 2 次提交
    • D
      PPC64 support for NVIDIA V100 GPU with NVLink2 passthrough · 1a922648
      由 Daniel Henrique Barboza 提交于 
      The NVIDIA V100 GPU has an onboard RAM that is mapped into the
host memory and accessible as normal RAM via an NVLink2 bridge. When
passed through in a guest, QEMU puts the NVIDIA RAM window in a
non-contiguous area, above the PCI MMIO area that starts at 32TiB.
This means that the NVIDIA RAM window starts at 64TiB and go all the
way to 128TiB.

This means that the guest might request a 64-bit window, for each PCI
Host Bridge, that goes all the way to 128TiB. However, the NVIDIA RAM
window isn't counted as regular RAM, thus this window is considered
only for the allocation of the Translation and Control Entry (TCE).
For more information about how NVLink2 support works in QEMU,
refer to the accepted implementation [1].

This memory layout differs from the existing VFIO case, requiring its
own formula. This patch changes the PPC64 code of
@qemuDomainGetMemLockLimitBytes to:

- detect if we have a NVLink2 bridge being passed through to the
guest. This is done by using the @ppc64VFIODeviceIsNV2Bridge function
added in the previous patch. The existence of the NVLink2 bridge in
the guest means that we are dealing with the NVLink2 memory layout;

- if an IBM NVLink2 bridge exists, passthroughLimit is calculated in a
different way to account for the extra memory the TCE table can alloc.
The 64TiB..128TiB window is more than enough to fit all possible
GPUs, thus the memLimit is the same regardless of passing through 1 or
multiple V100 GPUs.

Further reading explaining the background
[1] https://lists.gnu.org/archive/html/qemu-devel/2019-03/msg03700.html
[2] https://www.redhat.com/archives/libvir-list/2019-March/msg00660.html
[3] https://www.redhat.com/archives/libvir-list/2019-April/msg00527.htmlSigned-off-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: NErik Skultety <eskultet@redhat.com>
      1a922648
    • D
      qemu_domain: NVLink2 bridge detection function for PPC64 · cc9f0380
      由 Daniel Henrique Barboza 提交于 
      The NVLink2 support in QEMU implements the detection of NVLink2
capable devices by verifying the attributes of the VFIO mem region
QEMU allocates for the NVIDIA GPUs. To properly allocate an
adequate amount of memLock, Libvirt needs this information before
a QEMU instance is even created, thus querying QEMU is not
possible and opening a VFIO window is too much.

An alternative is presented in this patch. Making the following
assumptions:

- if we want GPU RAM to be available in the guest, an NVLink2 bridge
must be passed through;

- an unknown PCI device can be classified as a NVLink2 bridge
if its device tree node has 'ibm,gpu', 'ibm,nvlink',
'ibm,nvlink-speed' and 'memory-region'.

This patch introduces a helper called @ppc64VFIODeviceIsNV2Bridge
that checks the device tree node of a given PCI device and
check if it meets the criteria to be a NVLink2 bridge. This
new function will be used in a follow-up patch that, using the
first assumption, will set up the rlimits of the guest
accordingly.
Signed-off-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
      cc9f0380
  9. 12 4月, 2019 1 次提交
  11. 10 4月, 2019 2 次提交
  13. 04 4月, 2019 1 次提交
  15. 03 4月, 2019 11 次提交
  17. 29 3月, 2019 3 次提交
  19. 27 3月, 2019 1 次提交
  21. 22 3月, 2019 7 次提交
    • E
      snapshot: Rename virDomainSnapshotObjPtr · e055a816
      由 Eric Blake 提交于 
      Now that the core of SnapshotObj is agnostic to snapshots and can be
shared with upcoming checkpoint code, it is time to rename the struct
and the functions specific to list operations. A later patch will
shuffle which file holds the common code. This is a fairly mechanical
patch.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
      e055a816
    • E
      snapshot: Switch type of virDomainSnapshotObj.def · 1ab05da2
      由 Eric Blake 提交于 
      Another step towards making the object list reusable for both
snapshots and checkpoints: the list code only ever needs items that
are in the common virDomainMomentDef base type. This undoes a lot of
the churn in accessing common members added in the previous patch, and
the bulk of the patch is mechanical. But there was one spot where I
had to unroll a VIR_STEAL_PTR to work around changed types.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
      1ab05da2
    • E
      snapshot: Factor out virDomainMomentDef class · ffc0fbeb
      由 Eric Blake 提交于 
      Pull out the common parts of virDomainSnapshotDef that will be reused
for virDomainCheckpointDef into a new base class.  Adjust all callers
that use the direct fields (some of it is churn that disappears when
the next patch refactors virDomainSnapshotObj; oh well...).

Someday, I hope to switch this type to be a subclass of virObject, but
that requires a more thorough audit of cleanup paths, and besides
minimal incremental changes are easier to review.

As for the choice of naming:
I promised my teenage daughter Evelyn that I'd give her credit for her
contribution to this commit. I asked her "What would be a good name
for a base class for DomainSnapshot and DomainCheckpoint". After
explaining what a base class was (using the classic OOB Square and
Circle inherit from Shape), she came up with "DomainMoment", which is
way better than my initial thought of "DomainPointInTime" or
"DomainPIT".
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
      ffc0fbeb
    • E
      snapshot: Access snapshot def directly when needed · 55c2ab3e
      由 Eric Blake 提交于 
      An upcoming patch will rework virDomainSnapshotObjList to be generic
for both snapshots and checkpoints; reduce the churn by adding a new
accessor virDomainSnapshotObjGetDef() which returns the
snapshot-specific definition even when the list is rewritten to
operate only on a base class, then using it at sites that that are
specific to snapshots.  Use VIR_STEAL_PTR when appropriate in the
affected lines.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
      55c2ab3e
    • E
      snapshot: Add accessors for updating snapshot list relations · 02c4e24d
      由 Eric Blake 提交于 
      Rather than allowing a leaky abstraction where multiple drivers have
to open-code operations that update the relations in a
virDomainSnapshotObjList, it is better to add accessor functions so
that updates to relations are maintained closer to the internals.
This patch finishes the job started in the previous patch, by getting
rid of all direct access to nchildren, first_child, or sibling outside
of the lowest level functions, making it easier to refactor later on.

The lone new caller to virDomainSnapshotObjListSize() checks for a
return != 0, because it wants to handles errors (-1, only possible if
the hash table wasn't allocated) and existing snapshots (> 0) in the
same manner; we can drop the check for a current snapshot on the
grounds that there shouldn't be one if there are no snapshots.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
      02c4e24d
    • E
      snapshot: Track current snapshot in virDomainSnapshotObjList · 4819f54b
      由 Eric Blake 提交于 
      It is easier to track the current snapshot as part of the list of
snapshots. In particular, doing so lets us guarantee that the current
snapshot is cleared if that snapshot is removed from the list (rather
than depending on the caller to do so, and risking a use-after-free
problem, such as the one recently patched in 1db9d0ef).  This
requires the addition of several new accessor functions, as well as a
useful return type for virDomainSnapshotObjListRemove().  A few error
handling sites that were previously setting vm->current_snapshot =
NULL can now be dropped, because the previous function call has now
done it already.  Also, qemuDomainRevertToSnapshot() was setting the
current vm twice, so keep only the one used on the success path.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
      4819f54b
    • E
      snapshot: Drop virDomainSnapshotDef.current · f1056279
      由 Eric Blake 提交于 
      The only use for the 'current' member of virDomainSnapshotDef was with
the PARSE/FORMAT_INTERNAL flag for controlling an internal-use
<active> element marking whether a particular snapshot definition was
current, and even then, only by the qemu driver on output, and by qemu
and test driver on input. But this duplicates vm->snapshot_current,
and gets in the way of potential simplifications to have qemu store a
single file for all snapshots rather than one file per snapshot.  Get
rid of the member by adding a bool* parameter during parse (ignored if
the PARSE_INTERNAL flag is not set), and by adding a new flag during
format (if FORMAT_INTERNAL is set, the value printed in <active>
depends on the new FORMAT_CURRENT).

Then update the qemu driver accordingly, which involves hoisting
assignments to vm->current_snapshot to occur prior to any point where
a snapshot XML file is written (although qemu kept
vm->current_snapshot and snapshot->def_current in sync by the end of
the function, they were not always identical in the middle of
functions, so the shuffling gets a bit interesting). Later patches
will clean up some of that confusing churn to vm->current_snapshot.

Note: even if later patches refactor qemu to no longer use
FORMAT_INTERNAL for output (by storing bulk snapshot XML instead), we
will always need PARSE_INTERNAL for input (because on upgrade, a new
libvirt still has to parse XML left from a previous libvirt).
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
      f1056279
  23. 20 3月, 2019 2 次提交