Emmanuel Ackaouy [Wed, 13 Dec 2006 16:13:26 +0000 (16:13 +0000)]
[XEN] Improve scheduler cap mechanism
Somewhat unbastardize the scheduler cap mechanism. We now cleanly
pause and unpause running VCPUs of capped out domains instead of
using sub-idle priorities. This also improves the precision of
caps a bit. Signed-off-by: Emmanuel Ackaouy <ack@xensource.com>
Ewan Mellor [Wed, 13 Dec 2006 11:13:08 +0000 (11:13 +0000)]
Check return of new_domain inside dom0_init, avoiding segfault if there is
a second instance of xenstored running (and we're using different pid files).
Ewan Mellor [Wed, 13 Dec 2006 14:36:36 +0000 (14:36 +0000)]
When recreating the record for dom 0, read the current VCPU settings from Xen,
rather than from the store, allowing boot-time configuration to take precedence.
Hollis Blanchard [Tue, 12 Dec 2006 20:35:07 +0000 (14:35 -0600)]
[POWERPC][XEN] Add support for || as a xen/dom0 commandline divider. Signed-off-by: Jerone Young <jyoung5@us.ibm.com> Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>
--HG--
extra : transplant_source : %DB%C7M%B1JK9%D3Y6_%CF%82W%21m%96%8F%A2i
Emmanuel Ackaouy [Tue, 12 Dec 2006 19:37:21 +0000 (19:37 +0000)]
[XEN] Improve multi-core/thread scheduler support.
Relax granularity at which running VCPUs are migrated onto
cores or sockets with more idle execution vehicles and
simplify code a good deal in the process.
Also, some scheduler stat cleanup while in there. Signed-off-by: Emmanuel Ackaouy <ack@xensource.com>
Tim Deegan [Tue, 12 Dec 2006 16:52:13 +0000 (16:52 +0000)]
[XEN] Be sure to discard old PAE l2 shadows when guest l2e becomes null.
This fixes recent PAE live migrate failures. Signed-off-by: Tim Deegan <Tim.Deegan@xensource.com>
Jimi Xenidis [Tue, 12 Dec 2006 01:50:32 +0000 (20:50 -0500)]
[XEN][POWERPC] boot_of_allocator fixups
This patch:
- print devtree mod size correctly
- find next_bit rather then test all needed bits to be zero
- stop pre-allocating memory below image as used Signed-off-by: Jimi Xenidis <jimix@watson.ibm.com> Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>
--HG--
extra : transplant_source : %88%7E%1C%BA%C6%15M%A0%A3%A3%C2C%3F%BC%5B%0F%C2%A1%C9%B8
Jimi Xenidis [Tue, 12 Dec 2006 00:43:18 +0000 (19:43 -0500)]
[XEN][POWERPC] Fix systemsim-gpul failure to boot
This patch fixes the fencepost error and adds a check for the failure to
initialize the allocator, which would have make Mark's FTB much easier
to debug.
This works for me on the public systemsim-gpul release. Signed-off-by: Amos Waterland <apw@us.ibm.com>
Also fixes Maple, was just getting luck. Signed-off-by: Jimi Xenidis <jimix@watson.ibm.com> Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>
--HG--
extra : transplant_source : %87%8C%E1%F7%8A%D3%97oD%A6W%E58j%7Fma%9F%FA%C2
Hollis Blanchard [Tue, 12 Dec 2006 15:56:06 +0000 (09:56 -0600)]
[POWERPC] Merge in outstanding changes from xenppc-unstable.hg.
It's a long story, but basically a small divergence in xenppc-unstable meant a
large number of changesets couldn't be directly imported to xen-unstable, so
this changeset includes all of them. Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>
[XEN] The shadow FAST_FAULT_PATH optimisation assumes that pages never
transition between mmio and RAM-backed. This isn't true after a
an add_to_physmap memory op. Fix this by just blowing the shadow tables
after every such operation; they're rare enough that the performance
hit is not a concern.
Signed-off-by: Steven Smith <sos22@cam.ac.uk> Acked-by: Tim Deegan <Tim.Deegan@xensource.com>
[IA64] change XEN_DOMCTL_getmemlist to return gmfn instead of mfn
This patch tries to get 'xm dump-core' work on ia64 again.
With this patch, 'xm dump-core' results in creating core-dump file
as before with a slight format changes.
(machine address array -> pseudo physicall address array)
However xm dump-core has been broken on ia64 since
XEN_DOMCTL_getmemlist/ia64 is differenct from x86's.
This patch does NOT fix it.
[LIBXC] Make strerror() thread-safe by protecting it with a mutex.
Using strerror_r() is made difficult by the different GNU definition,
and different distros variously choose the POSIX or GNU prototype. Signed-off-by: Keir Fraser <keir@xensource.com>
Alastair Tse [Fri, 8 Dec 2006 13:31:21 +0000 (13:31 +0000)]
[XEND] Keep vif IP addresses as space separated string.
They were inconsistently represented as strings or lists making it
difficult to know the type of the IP parameter for vifs. Since both xm
and DevController represents these as space separated strings, that is
what it will be stored as.
[LIBXC] Clean up use of sterror(). Define a thread-safe
version which uses strerror_r(), and tries to cope with
the POSIX and GNU versions of that function. Signed-off-by: Keir Fraser <keir@xensource.com>
[IA64] workaround NaT consumption fault when hiding a PCI function from dom0
The free_irq() path seems to have trouble with devices hidden from
dom0 for driver domains. The result is a NaT consumption fault in
the reboot notifier_call_chain() when dom0 reboots. This workaround
prevents the problem until it can be further investigated.
Signed-off-by: Alex Williamson <alex.williamson@hp.com>
Alastair Tse [Thu, 7 Dec 2006 16:08:44 +0000 (16:08 +0000)]
[XEND] Sync Xen API with newer VDB and VIF specification.
Fix breakage with storing VIF and VBD refs, remove extraneous
debugging. Also fixing some type errors in default values for Xen API
if things are missing in XendConfig.
Tim Deegan [Thu, 7 Dec 2006 13:14:44 +0000 (13:14 +0000)]
[XEN] Init pae_l3_cache lock earlier
In particular, before hvm_vcpu_initialise causes a CR3 update. Signed-off-by: Tim Deegan <Tim.Deegan@xensource.com>
Ewan Mellor [Thu, 7 Dec 2006 12:23:35 +0000 (12:23 +0000)]
This patch checks the contents of XENDOMAINS_SAVE before the restore
process begins, and prevents xendomains from attempting to start any
domain that appears there, whether the domain started successfully or
not.
Ewan Mellor [Thu, 7 Dec 2006 12:15:30 +0000 (12:15 +0000)]
The new lifecycle patches in xen-unstable added 3 new operations which can
be performed on domains - new, start and delete. libvirt has had no-op
stub implementations of these APIs for a while & thus we'd enable their
use for Xen 3.0.4 builds. The new APIs are not currently exposed via the
existing SEXPR protocol, and since the new Xen-API protocol is unsupported
tech-preview for 3.0.4 we don't want to re-write against that just yet.
Thus the attached patch adds the 3 neccessary bindings for the SEXPR protocol
to allow the lifecycle functionality to be invoked. It also allows the 'state'
parameter to be specified when listing domains, so we can explicitly request
inactive domains if we know we're running against a 3.0.4 era XenD
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
projects / xen.git / log