This will allow us to update dimm state on OSPM-initiated eject operations e.g.
with "echo 1 > /sys/bus/acpi/devices/PNP0C80\:00/eject"

This implements batch monitor operations for hot-add and hot-remove. These are
probably better suited for a higher-level management layer, but are useful for
testing. Let me know if there is interest for such commands upstream.

syntax: mem_increase poolid num
will hotplug num dimms from pool poolid. This starts from lowest unpopulated
physical memory (dimm) and trying to cover any existing physical holes.

syntax: mem_decrease poolid num
will hot-unplug num dimms from pool poolid, This starts from highest populated
physical memory (dimm).

Respective qmp commands are "mem-increase", "mem-decrease".

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Implement batch dimm creation command line options. These could be useful for
not bloating the command line with a large number of dimms.

syntax: -dimms pfx=poolid,size=sz,num=n
Will create numdimms dimms with ids poolid0, ..., poolidn-1. Each dimm has a
size of sz.

Implement -dimmpop option to populate dimms at bootup
syntax: -dimmpop pfx=poolid,num=n
This will populate n dimms with ids poolid0, ..., poolidn-1.

(live-migration could break here without patch 12/21: -dimmspop
needs to be reworked to support populating of individual dimms with
same prefix, and not only a range of dimms starting from 0)

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Returns total memory of guest in bytes, including hotplugged memory.

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

This allows failed hot operations to be retried at anytime. This only
works for guests that use _OST notification. Other guests cannot retry failed
hot operations on same devices until after reboot.

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

in case of hot-remove or hot-add failure, the dimm bitmaps in qemu and Seabios
are inconsistent with the true state of the DIMM devices. The "populated" field
of the DimmState reflects the true state of the device. This inconsistency means
that a failed operation cannot be retried.

Ths patch updates the bit array to the true state of the dimms on VM reboot.
This allows retry of failed hot-add or hot-remove operations after a reboot.

Retrying a failed hot operation is not yet possible before reboot (the following
patch removes this limitation for guests with _OST acpi support)

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

This allows qemu to receive notifications from the guest OS on success or
failure of a memory hotplug request. The guest OS needs to implement the _OST
functionality for this to work (linux-next: http://lkml.org/lkml/2012/6/25/321)
Also add new _OST registers in docs/specs/acpi_hotplug.txt

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Guest can respond to ACPI hotplug events e.g. with _EJ or _OST method.
This patch implements a tail queue to store guest notifications for memory
hot-add and hot-remove requests.

Guest responses for memory hotplug command on a per-dimm basis can be detected
with the new hmp command "info memhp" or the new qmp command "query-memhp"
Examples:

(qemu) dimm_add dimm0
(qemu) info memhp
Dimm: dimm0 hot-add success
or
Dimm: dimm0 hot-add failure

(qemu) dimm_del dimm0
(qemu) info memhp
Dimm: dimm0 hot-remove success
or
Dimm: dimm0 hot-remove failure

Results are removed from the queue once read.

This patch only queues _EJ events that signal hot-remove success.
For  _OST event queuing, which cover the hot-remove failure and
hot-add success/failure cases, the next 2 patches are also needed.

These notification items should probably be part of migration state (not yet
implemented)

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Live migration works after memory hot-add events, as long as the
qemu command line "-dimm" arguments are changed on the destination host
to specify "populated=on" for the dimms that have been hot-added.

If a command-line change has not occured, the destination host does not yet
have the corresponding ramblock in its ram_list. Activate the memslot on the
destination during ram_load.

Perhaps several fields of the DimmState struct should be part of a
VMStateDescription to handle migration in a cleaner way.

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Hot-add hmp syntax: dimm_add dimmid
Hot-remove hmp syntax: dimm_del dimmid

Respective qmp commands are "dimm-add", "dimm-del".

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Syntax: "-dimm id=name,size=sz,node=pxm,populated=on|off"

The starting physical address for all dimms is calculated automatically from top
of memory, skipping the pci hole at [PCI_HOLE_START, 4G).
"populated=on" means the dimm is populated at machine startup. Default is off.
"node" is defining numa proximity for this dimm. Default is node zero.

Example:
"-dimm id=dimm0,size=512M,node=0,populated=off"
will define a 512M memory slot belonging to numa node 0.

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

The numa_fw_cfg paravirt interface is extended to include SRAT information for
all hotplug-able dimms. There are 3 words for each hotplug-able memory slot,
denoting start address, size and node proximity. The new info is appended after
existing numa info, so that the fw_cfg layout does not break.  This information
is used by Seabios to build hotplug memory device objects at runtime.
nb_numa_nodes is set to 1 by default (not 0), so that we always pass srat info
to SeaBIOS.

v1->v2:
Dimm SRAT info (#dimms) is appended at end of existing numa fw_cfg in order not
to break existing layout
Documentation of the new fwcfg layout is included in docs/specs/fwcfg.txt

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Dimm physical address offsets are calculated automatically and memory map is
adjusted accordingly. If a DIMM can fit before the PCI_HOLE_START (currently
0xe0000000), it will be added normally, otherwise its physical address will be
above 4GB.

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

A 32-byte register is used to present up to 256 hotplug-able memory devices
to BIOS and OSPM. Hot-add and hot-remove functions trigger an ACPI hotplug
event through these. Only reads are allowed from these registers.

An ACPI hot-remove event but needs to wait for OSPM to eject the device.
We use a single-byte register to know when OSPM has called the _EJ function
for a particular dimm. A write to this byte will depopulate the respective dimm.
Only writes are allowed to this byte.

v1->v2:
mems_sts address moved from 0xaf20 to 0xaf80 (to accomodate more space for
cpu-hotplugging in the future).
_EJ array is reduced to a single byte.
Add documentation in docs/specs/acpi_hotplug.txt

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

Each hotplug-able memory slot is a SysBusDevice. A hot-add operation for a
particular dimm creates a new MemoryRegion of the given physical address
offset, size and node proximity, and attaches it to main system memory as a
sub_region. A hot-remove operation detaches and frees the MemoryRegion from
system memory.

This prototype still lacks proper qdev integration: a separate
hotplug side-channel is used and main system bus hotplug capability is
ignored.

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>

…u-kvm into uq/master

Link in oslib objects also for BSD user, but avoid using the version of
qemu_vmalloc() defined in oslib-posix.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Rename what is remaining of op_helper.c to seg_helper.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move memory access helpers to mem_helper.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move various functions to misc_helper.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move SMM helpers to smm_helper.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move SVM helpers to svm_helper.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move integer and bit field helpers to int_helper.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move condition code helpers to cc_helper.c.

Move the shared inline functions lshift(), cpu_load_eflags() and
cpu_cc_compute_all() to cpu.h.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Adjust function names and add an explicit CPUX86State
parameter instead of relying on AREG0.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move shift templates from helper_template.h to
shift_helper_template.h and the condition code helpers
to cc_helper_template.h.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Fix coding style in helper_template.h before next commit.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

SSE function tables could easily be corrupted because of use
of void pointers.

Introduce function pointer types and helper variables in order
to improve type safety.

Split sse_op_table3 according to types used.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move FPU and MMX/SSE helpers to fpu_helpers.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Fix coding style in ops_sse.h before next commit.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

After the previous patch, we can use the proper
declaration in a common header file.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Move exception handlers from op_helper.c to excp_helper.c.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Add an explicit CPUX86State parameter instead of relying on AREG0.

Merge raise_exception_env() to raise_exception(), likewise with
raise_exception_err_env() and raise_exception_err().

Introduce cpu_svm_check_intercept_param() and cpu_vmexit()
as wrappers.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Fix coding style and a few typos.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

The emulated devices can run simultaneously with the guest, so
we need to be careful with ordering of load and stores done by
them to the guest system memory, which need to be observed in
the right order by the guest operating system.

This adds a barrier call to the basic DMA read/write ops which
is currently implemented as a smp_mb(), but could be later
improved for more fine grained control of barriers.

Additionally, a _relaxed() variant of the accessors is provided
to easily convert devices who would be performance sensitive
and negatively impacted by the change.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>