tools: change cpumaps to uint8_t

Cpumap types in tools (libxc and libxl) are changed to be based on bytes like
in the interface to the hypervisor.
To make handling easier the size of used cpumaps is always based on the
number of physical cpus supported by the hypervisor. This eliminates the need
to keep track of the cpumap size in external interfaces.
In libxl a macro for cycling through a cpumap is added (libxl_for_each_cpu).

Interfaces changed:
libxl_set_vcpuaffinity()
libxl_cpumap_alloc()
xc_vcpu_setaffinity()
xc_vcpu_getaffinity()
xc_cpupool_freeinfo()

Signed-off-by: Juergen Gross <juergen.gross@ts.fujitsu.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

tools: gdbsx: Check return of write()

Not checking leads to warn_unused_result checks triggering in some
libraries and compilers.  Combined with -Werror this breaks the build.

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

VT-d: fix device assignment failure (regression from Xen c/s 19805:2f1fa2215e60)

If the device at <secbus>:00.0 is the device the mapping operation was
initiated for, trying to map it a second time will fail, and hence
this second mapping attempt must be prevented (as was done prior to
said c/s).

While at it, simplify the code a little, too.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Weidong Han <weidong.han@intel.com>

x86: xsave save/restore support for both PV and HVM guests.

Signed-off-by: Shan Haitao <haitao.shan@intel.com>
Signed-off-by: Han Weidong <weidong.han@intel.com>

x86 hvm: exposes AVX to guest.

Signed-off-by: Shan Haitao <haitao.shan@intel.com>
Signed-off-by: Han Weidong <weidong.han@intel.com>

x86: Xsave support for PV guests.

Signed-off-by: Shan Haitao <haitao.shan@intel.com>
Signed-off-by: Han Weidong <weidong.han@intel.com>

Fix ia64 tools build

22066:eccfdeb41b80 replaced tools/libxc/ia64/aclinux.h with
xen/include/acpi/platform/aclinux.h.
Because both are incompatible, tools cannot be built on ia64.

This patch fixes compilation errors by defining identifiers at
ac_ia64_tools.h.

Signed-off-by: KUWAMURA Shin'ya <kuwa@jp.fujitsu.com>

libxl: fix build since 22325:4ac03710fc32

"libxl: Fix migration of HVM guests" introduces a stat call without
including <sys/stat.h>

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

x86 hvm: Fixes for 22339:6fae4331d5aa

Primarily fix up register names in public header file for
compatibility with old unmodified versions of virtual firmware and
qemu-dm. Also fixes a bug in the read case of handle_evt_io().

Signed-off-by: Keir Fraser <keir@xen.org>

x86_emulate: Fix build with some versions of gcc targeting i386.

Signed-off-by: Keir Fraser <keir@xen.org>

i386: simplify pte_read_atomic()

There's no need for a loop here: if the value to be compared against
and the value to be stored are equal, cmpxchg{,8b} act just like a
load anyway.

Signed-off-by: Jan Beulich <jbeulich@novell.com>

x86 hvm: Introduce pmtimer_change_ioport and HVM_PARAM_ACPI_IOPORTS_LOCATION

By default, Xen will handle the old ACPI IO port. But it can switch to
the new one by setting the HVM_PARAM_ACPI_IOPORTS_LOCATION to 1.

Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Keir Fraser <keir@xen.org>

x86 hvm: Introduce unregister_io_handler

Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Keir Fraser <keir@xen.org>

VT-d: only scan secondary functions on multi-function devices

Signed-off-by: Jan Beulich <jbeulich@novell.com>

Xen: fix various checks of unsigned integers < 0

Some of these could be benignly discarded by the compiler but some are
actual bugs.

Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>

ACPI: fix typo

Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>

x86: Clean up existing XSAVE support

Signed-off-by: Han Weidong <weidong.han@intel.com>
Signed-off-by: Shan Haitao <haitao.shan@intel.com>

X86: Prefer TSC-deadline timer in Xen

The new TSC Deadline Timer offers system software a low overhead
per-logical-thread deadline timer in TSC units.

The timer is implemented via a new architectural 64-bit register,
IA32_TSC_DEADLINE_MSR. Reads and writes of this MSR occur in program
order, but are non-serializing.

The support for this feature is indicated by
CPUID.01H:ECX.TSC_Deadline[bit 24] =3D 1 as documented in the Intel
Architecture Software Developer's Manual.

The LOCAL APIC on new processors has a mode where its underlying
hardware timer can now be accessed via the non-serializing
IA32_TSC_DEADLINE_MSR in TSC tick units.

If this mode is present, prefer it over the traditional LAPIC timer
mode.  KERN_DEBUG dmesg will print "TSC deadline timer enabled" when
TDT is used.

Bootparam "tdt=off" is available to revert to LAPIC timer mode.

This patch is based on original work by Len Brown for Linux kernel.

cc: Len Brown <len.brown@intel.com>
Signed-off-by: Wei Gang <gang.wei@intel.com>
Signed-off-by: Keir Fraser <keir@xen.org>

vtd: consolidate VT-d quirks into a single file quirks.c

Consolidate VT-d quirks into a single file - quirks.c.  This includes
quirks to workaround OEM BIOS issue with VT-d enabling in IGD, Cantiga
VT-d buffer flush issue, Cantiga IGD Vt-d low power related errata,
and a quirk to workaround issues related to wifi direct assignment.

Signed-off-by: Allen Kay <allen.m.kay@intel.com>
Reviewed-by: Jan Beulich <JBeulich@novell.com>

credit2: Trace and debug key tweaks

* Add traces for credit reset and scheduling a tasklet
* Remove tsc for traces which probably don't need them
* Print domain info in the debug dump

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>

credit2: Fix runq_tickle to use idle, tickled masks

The previous code had a bug where, if a second vcpu woke up and ran
runq_tickle before the first vcpu had actually started running on a
tickled processor, the code would choose the same cpu to tickle, and
the result would be a "missed tickle" -- only one of the vcpus would
run, despite having idle processors.

This change:
* Keeps a mask of idle cpus
* Keeps a mask of cpus which have been tickled, but not entered
schedule yet.

The new tickle code first looks at the set of idle-but-not-tickled
cpus; if it's not empty, it tickles one.

If that doesn't work, it looks at the set of not-idle-but-not-tickled
cpus, finds the one with the lowest credit; if that's lower than the
waking vcpu, it tickles that one.

If any cpu is tickled, its bit in the tickled mask is set, and cleared
when schedule() is called.

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>

credit2: Make fake runq map initilization explicit

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>

hvmloader: Configure gPXE to use serial console

Sometimes we would rather connect to an HVM's serial console than it's
VGA display. In this case we cannot see or interact with the ethernet
booting process. This is a bit annoying so let's just enable it.

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>

QEMU_TAG update

avoid numa placement of cpus with active cpupools

When using cpupools don't pin vcpus to numa nodes as this might
conflict with the cpupool definition.
numa placement should be handled by cpupool configuration instead.

Signed-off-by: Juergen Gross <juergen.gross@ts.fujitsu.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxl: Fix migration of HVM guests

In the default "QemuDeviceModelRecord"-style HVM tail of the migration
protocol the size of the qemu save record is unknown at the receiver
and therefore it must read until EOF. This is not compatible with the
xl migration protocol which contains a post-migration handshake and
therefore cannot close the socket on the sending end.

What is required is an explicit length field for the save record,
which the "RemusDeviceModelState"-style HVM tail includes.

Rather than overload the "RemusDeviceModelState" name for a non-Remus
use case (on off chance that they need to diverge for some reason in
the future) introduce a third style called "DeviceModelRecord0002"
which is identical to current "RemusDeviceModelState"-style.

Hopefully the inclusion of a number here will allow easier extension
in the future without needing to come up with increasingly less
helpful names!

Also propagate errors from xc_domain_save and
libxl__domain_suspend_common to callers.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

xl: support -w (wait) option in xl shutdown

improves compatibility with xm.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

tools: caml-stubdom: fix "red zone" bug

According to the AMD64 ABI, the 128-byte area below %rsp is reserved
for scratch space and should not be used by signal, interrupt, or
exception handlers. However, Mini-OS does not adhere to this
interface. As a result, the -mno-red-zone flag must be provided to gcc
in order to force code to be compiled without this assumption. In the
case of the c-stubdom or the ioemu stubdom, Makefile configuration and
flags are inherited from Xen, which includes this flag. However, in
the case of the ocaml stubdom, since an independent ocaml version is
pulled and compiled, with its own make system in place, this flag is
not present.

As a default optimisation (no optimisation flags specified), gcc
generates function header code that uses mov instead of push (to save
a few instructions). However, with the 128-byte scratch space
assumption in place, registers are moved onto the stack *before* %rsp
is updated. This results in cases where when a function is being
called, *after* the first mov <foo>, <offset>(%rsp) but *before* the
%rsp update, sub <bar>, %rsp, if an interrupt is fired, then the stack
will be clobbered (as Mini-OS uses the application stack for its
stack, pushing registers onto the stack on entry and popping them off
on exit).

This patch fixes this issue by passing the -mno-red-zone flag to the
ocaml configure script in the stubdom Makefile.

Signed-off-by: Patrick Colp <pjcolp@cs.ubc.ca>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxl: Fix format string abuses / vulnerabilities

There are a few places where libxl__xs_write is passed a variable
value to write to xenstore, but the semantics are that the first char*
is a format string.  So use "%s".

This fixes the following errors reported by some newer compilers:
 libxl.c: In function "libxl_create_cpupool":
 libxl.c:3981: error: format not a string literal and no format arguments
 libxl.c:3983: error: format not a string literal and no format arguments
 libxl.c: In function "libxl_cpupool_movedomain":
 libxl.c:4095: error: format not a string literal and no format arguments

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Gianni Tedesco <gianni.tedesco@citrix.com>

libxl: fix dom0 minimum memory threshold check

libxl_set_memory_target tries to set the memory target for a given
domain to new_target_memkb. The function includes a check to make sure
that dom0's memory is not reduced below a minimal threshold ie.
LIBXL_MIN_DOM0_MEM. However, this check is performed before the
new_target_memkb variable is properly initialised - when the value is
always zero. This means that the check always fails. Fix this by moving
the test to happen after the proper initialisation of new_target_memkb.

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

gdbsx: Makefiles: correctly generate dependencies

gdbsx hacks CFLAGS to remove optimisations for easier debugging, however
this breaks dep generation due to an early evaluation of CFLAGS and
leaves stale files called "..d" after a make clean. We could possibly
fix this by separating dep generation from the rest of CFLAGS but it
seems a little invasive. The gdbsx hackers can do their own thing during
development and testing anyway...

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>

ocaml: Makefile: delete dependency files during make clean

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

tools: Makefile: delete qemu during make distclean

The top-level makefile claims this to be the expected behaviour so make
it so.

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxl: Makefile: delete temporary files during make clean

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

blktap2: Makefiles: delete symlinks and deps during make clean

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

Do not wait for memory teardown when live migrate

When trying to live migrate guest with great memory (e.g. 20 GB),
there's delay caused by destroying source copy of domain.  To speed up
this process we can close socket before starting guest destroy, as
source guest is not running anymore.

Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: fix xc_version by handling all known command types.

xend was crashing since 22289:ca4a781c8ae8 due to missing handling of
XENVER_commandline.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>

Merge

"kdd" Windows debugger stub.

kdd runs in dom0, attaches to a domain and speaks the Windows kd serial
protocol over a TCP connection (which should go to kd or windbg, e.g.
by having another VM with its virtual COM1 set up as a TCP listener).

It doesn't do breakpoints &c yet, and windbg can get quite confused
since the kernel debugger's not actually running, but it's good enough
to extract backtraces from wedged VMs.

Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>

libxc: correct handling of switch_qemu_logdirty callback return value

The callback return convention follows xc_domain_save and therefore
returns 0 on success and >0 on error, correct the check accordingly.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

hvmloader: reduce iasl verbosity

I don't think anyone really needs iasl's banner displayed during the
build process.

Signed-off-by: Jan Beulich <jbeulich@novell.com>

ia64: fix build in grant table and tmem code

Signed-off-by: Jan Beulich <jbeulich@novell.com>

x86: enable support for {rd,wr}{fs,gs}base instructions

... so that once in a while Xen knows of a new CPU feature before
Linux starts making use of it.

While (obviously) I wasn't able to test this, it seemed strait forward
enough to enable anyway.

Signed-off-by: Jan Beulich <jbeulich@novell.com>

x86: delete not really used ioapic_renumber_irq variable

Signed-off-by: Jan Beulich <jbeulich@novell.com>

x86/kexec: fix very old regression and make compatible with modern Linux

c/s 13829 lost the (32-bit only) cpu_has_pae argument passed to the
primary kernel's stub (in the 32-bit Xen case only), and Linux
2.6.27/.30 (32-/64-bit) introduced a new argument (for KEXEC_JUMP)
which for now simply gets passed a hardcoded value.

Signed-off-by: Jan Beulich <jbeulich@novell.com>

Allow max_pages to be set to less than tot_pages

The memory allocation code sometimes needs to enforce that a guest
that's been told to balloon down isn't going to expand further
(because it's still executing a previous balloon-up operation).  That
means being able to set the desired max_pages even before the balloon
driver has brought tot_pages down to the right level.

Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>

cpupools: Make interface more consistent

The current cpupools code interface is a bit inconsistent.  This
patch addresses this by making the interaction for each
vcpu in a pool look like this:

alloc_vdata() -- allocates and sets up vcpu data
insert_vcpu() -- the vcpu is ready to run in this pool
remove_vcpu() -- take the vcpu out of the pool
free_vdata()  -- delete allocated vcpu data

(Previously, remove_vcpu and free_vdata were combined into a "destroy
vcpu", and insert_vcpu was only called for idle vcpus.)

This also addresses a bug in credit2 which was caused by a
misunderstanding of the cpupools interface.

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Acked-by: Juergen Gross <juergen.gross@ts.fujitsu.com>

libxc: finalise transition to hypercall buffers.

Rename xc_set_xen_guest_handle to set_xen_guest_handle[0] and remove now
unused functions:
  - xc_memalign
  - lock_pages
  - unlock_pages
  - hcall_buf_prep
  - hcall_buf_release

[0] sed -i -e 's/xc_set_xen_guest_handle/set_xen_guest_handle/g' \
tools/libxc/*.[ch] \
tools/python/xen/lowlevel/xc/xc.c \
tools/python/xen/lowlevel/acm/acm.c \
tools/libxc/ia64/xc_ia64_stubs.c \
tools/security/secpol_tool.c \
tools/misc/xenpm.c

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: do not align/lock buffers which do not need it

On restore:

region_mfn is passed to xc_map_foreign_range and
xc_map_foreign_bulk. In both cases the buffer is accessed from the
ioctl handler in the kernel and not from any hypercall. Therefore
normal copy_{to,from}_user handling in the kernel will cope with any
faulting access.

p2m_batch is passed to xc_domain_memory_populate_physmap which takes
care of bouncing the buffer already.

On save:

pfn_type is passed to xc_map_foreign_bulk which does not need locking
as per region_mfn above.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

secpol: use hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

xenpm: use hypercall buffers.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

python xc: use hypercall buffer interface.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

python acm: use hypercall buffer interface.

I have a suspicion these routines should be using libxc rather than
reimplementing all the hypercalls again, but I don't have the
enthusiasm to fix it.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert ia64 dom0vp interface to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: switch page offlining interfaces to hypercall buffers

There is no need to lock/bounce minfo->pfn_type in init_mem_info since
xc_get_pfn_type_batch() will take care of that for us.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert mmuext op interface over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert memory op interface over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert gnttab interfaces over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert tmem interface over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert mca interface over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert hvmop interfaces over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert flask interfaces over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert physdevop interface over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert schedop interfaces over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert evtchn interfaces over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert acm interfaces over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert watchdog interface over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert sysctl interfaces over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert shadow domctl interfaces and save/restore over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert domctl interfaces over to hypercall buffers

(defer save/restore and shadow related interfaces til a later patch)

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: convert xc_version over to hypercall buffers

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson.citrix.com>

libxc: infrastructure for hypercall safe data buffers.

(Modified version of this patch, with updated #includes in
xc_hcall_buf.c to fix stubdomain build.)

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

minios: add parentheses to mlock/mulock arguments.

Fixes warning/build error with non-trivial arguments.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

QEMU_TAG update

libxl: add a detailed explanation of the libxl domain memory model

Originally-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Converted to US-ASCII by iwj.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

tools: cleanup domain save switch_qemu_logdirty callback

Move the function into struct save_callbacks with the others and add
the void *closure to the callback arguments.

Add and propagate an error return code from the callback.

Use this in libxl to pass the save context to
libxl__domain_suspend_common_switch_qemu_logdirty allowing us to reuse
the parent's xenstore handle, gc context etc.

Also add an apparently missing libxl__free_all to
libxl__domain_suspend_common.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

tools: cpupools: add example cpupool config file

Adds an example configuration file for xm/xl pool-create

Signed-off-by: juergen.gross@ts.fujitsu.com
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

tools: cpupools: xl: commands and library changes

Support of cpu pools in libxl and xl:
  library functions
  xl cpupool-create
  xl cpupool-list
  xl cpupool-destroy
  xl cpupool-cpu-add
  xl cpupool-cpu-remove
  xl cpupool-migrate
Renamed all cpu pool related names to *cpupool*

Signed-off-by: juergen.gross@ts.fujitsu.com
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

tools: cpupools: Rename cpu-pool commands in xm to cpupool-*

The xm commands pool-* are renamed to cpupool-*
The old pool-* commands are defined as aliases.

Signed-off-by: juergen.gross@ts.fujitsu.com
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
--HG--
rename : tools/python/xen/xm/pool-create.py => tools/python/xen/xm/cpupool-create.py
rename : tools/python/xen/xm/pool-new.py => tools/python/xen/xm/cpupool-new.py
rename : tools/python/xen/xm/pool.py => tools/python/xen/xm/cpupool.py

tools: cpupools: Support arbitrary numbers of physical cpus

To be able to support arbitrary numbers of physical cpus it was necessary to
include the size of cpumaps in the xc-interfaces for cpu pools.
These were:
  definition of xc_cpupoolinfo_t
  xc_cpupool_getinfo()
  xc_cpupool_freeinfo()
xc_cpupool_getinfo() and xc_cpupool_freeinfo() are changed to allocate the
needed buffer and return it.

Signed-off-by: juergen.gross@ts.fujitsu.com
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

xenstore: add xenstore-watch command line client

# xenstore-watch x | while read w ; do
> echo "watch fired on $w"
> echo "value" $(xenstore-read $w)
> echo
> done

# xenstore-write x/y/z 42

output from while loop:
        watch fired on x/t/z
        value 42

[ also add line to .hgignore - iwj ]

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

xl: fix spelling mistake in xl.conf

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

xenstored: return EINVAL for attempt to watch invalid path

errno is essentially random at this point.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

x86 hvm: When sending INIT-SIPI IPIs, pause the sending vcpu until the
asynchronous tasklet completes its work.

This is a little bit cleaner than busy-spinning in a retry loop.

Signed-off-by: Keir Fraser <keir@xen.org>

hvmloader: Properly block on xenstore daemon rather than merely yielding.

Signed-off-by: Keir Fraser <keir@xen.org>

x86 hvm: Improve test for wait-for-SIPI state of a target VLAPIC.

This fixes a livelock in hvmloader with credit2 scheduler, whereby an
AP can be brought online, do its work, and shut itself down, before
the BSP re-emulates the VLAPIC write that sent the SIPI. BSP then ends
up in an endless re-emulation work where it sees the target vcpu is
down, therefore schedules a tasklet, which does no work because the
vcpu is already initialised. The fix is to check v->is_initialised
rather than VPF_down, before scheduling the tasklet.

Signed-off-by: Keir Fraser <keir.fraser@citrix.com>

svm: Re-enable 2MB super paging

Bitwise OR operator has higher precedence than conditional
operator. This patch fixes the incorrect associativity and re-enables
2MB paging for AMD family 10h system.

Signed-off-by: Wei Wang <wei.wang2@amd.com>
Acked-by: Wei Huang <wei.huang2@amd.com>

msi: Mask out multi-function flag from PCI_HEADER_TYPE in read_pci_mem_bar()

This leads to an erroneous WARN_ON and possibly other side effects. It
seems to me that even multi-function devices ought to enjoy the
privilege of MSI-X capabilities.

Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>

libxc: use generic xc_get_pfn_list on ia64

The ia64 specific xc_get_pfn_list doesn't seem any different to the
generic xc_get_pfn_list once the call to xc_ia64_get_pfn_list is
expanded so remove and just use the generic one.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: do not lock VCPU context in xc_ia64_pv_recv_vcpu_context

xc_ia64_pv_recv_vcpu_context does not need to lock the ctxt buffer
since it calls xc_ia64_recv_vcpu_context which calls
xc_vcpu_setcontext which takes care of any necessary bouncing.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: drop xc_get_max_pages

The function isn't really ia64 specific but since the result isn't
actually used in the only caller and the same info is available via
xc_domain_getinfo simply drop the function.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: simplify lock profiling API

Current function has heavily overloaded semantics for the various
arguments. Separate out into more specific functions.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: simplify performance counters API

Current function has heavily overloaded semantics for the various
arguments. Separate out into more specific functions.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: make do_memory_op's callers responsible for locking indirect buffers

Push responsibility for locking buffers refered to by the memory_op
argument up into the callers (which are now all internal to libxc).

This removes the last of the introspecation from do_memory_op and
generally makes the transistion to hypercall buffers smoother.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: make xc_memory_op library private

Now that all XENMEM_* callers go via an op specific function make
xc_memory_op private to libxc (and rename to do_memory_op for
consistency with other private functions).

Also change the interface to take a size parameter so that
do_memory_op knows how much memory to lock for the top-level argument,
removing some of the introspection.

[ Compatibility note: this breaks qemu-xen before 706bc8c588cb ]

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: remove compatibility macro for xc_domain_memory_populate_physmap

[ Compatibility note: this breaks qemu-xen before d3c5e73eb3ae ]

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

QEMU_TAG update

libxc: add xc_maximum_ram_page to wrap XENMEM_maximum_ram_page

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: add xc_machphys_mfn_list to wrap XENMEM_machphys_mfn_list

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>

libxc: add xc_domain_maximum_gpfn to wrap XENMEM_maximum_gpfn

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>