Message-Id: <1425300248-40277-1-git-send-email-mimu@linux.vnet.ibm.com>
Date:	Mon,  2 Mar 2015 13:43:52 +0100
From:	Michael Mueller <mimu@...ux.vnet.ibm.com>
To:	qemu-devel@...gnu.org, kvm@...r.kernel.org,
	linux-s390@...r.kernel.org, linux-kernel@...r.kernel.org
Cc:	Gleb Natapov <gleb@...nel.org>, Alexander Graf <agraf@...e.de>,
	Christian Borntraeger <borntraeger@...ibm.com>,
	"Jason J. Herne" <jjherne@...ux.vnet.ibm.com>,
	Cornelia Huck <cornelia.huck@...ibm.com>,
	Paolo Bonzini <pbonzini@...hat.com>,
	Andreas Faerber <afaerber@...e.de>,
	Richard Henderson <rth@...ddle.net>,
	Eduardo Habkost <ehabkost@...hat.com>,
	Michael Mueller <mimu@...ux.vnet.ibm.com>
Subject: [PATCH v3 00/16] s390x cpu model implementation

This patch set in combination with its kernel kvm patch set proposes an
implementation of S390 cpu models. The origin of this item is to provide
a means for management interfaces like libvirt to draw decisions if life
guest migration to a target hypervisor is reasonable.

A migration constraint is that a target hypervisor is capable to run a
guest with the same S390 cpu model as the source hypervisor does. To
verify this condition, the administration interface employes the existing
QMP command "query-cpu-definitions" which returns a list of all currently
supported S390 cpu models of a given host system. Together with the newly
defined QMP command "query-cpu-model", which returns the current active
S390 cpu model of a guest, a conclusion can be drawn if a migration is
possible.

A S390 cpu model is defined as a triple of machine type, cpu facility set
and IBC value. Each historic, current and future triple receives a name
composed of the machine type and its general availability counter. This name
forms the cpu model name (e.g.: "2817-ga2".)

With means of the Instruction Blocking Control feature (IBC), the instruction
set available to a given guest is limitable.

Details:
- The QMP command query-cpu-model returns the active cpu model and the
  accellerator it is using:

  {"name":"2066-ga1","accelerator":"kvm"}

Or just the empty model in case an accelerator does not implement cpu
models yet:

  {}

- A management instance like libvirt may probe by means of the QMP command
  query-cpu-definitions which models are defined and usable for all
  supporting accelerators. To implement this the cpu definition info type gets
  an optional field named 'accelerators' which holds a list defining
  which cpu model is 'runnable' and in addition which one the 'default'
  cpu model is (i.e. the model to be used in the 'host' case).

  [{"name":"2964-ga1",
    "accelerators":[{"name":"kvm","runnable":false,"default":false}]}

Or just 'host' in case an accelerator does not implement cpu models yet:

  [{"name":"host"}]

- For accel=kvm the cpu model initialization takes place in kvm_arch_init()

What's currently a little bit unclear to me is how to best initialize the
various accelerators for machine 'none'. I played around with different
options and finally came up with the following sugguestion:

Introduce a QEMU "probe mode" that gets entered in case the current machine
is "none" and no specific accelerator is requested on the cmd line. When
in that mode, loop trough a list of acellerators in configure_accelerator
and invoke all their init methods once. The last accelerator to init shall
be tcg.

In cpu model context that allows to initialize the S390 CPU classes for
each single accelertor which supports it. Whence the callback for
qemu-cpu-definitions allows to populate its answer string according to the
above sketched extended CpuDefinitionInfo type for multiplaccelerators. 

v2-v3:
- using GTK-Doc style format now for function descriptions
- typo fixed (2/16)
- gen-facilties now used to generate cpu model specific facility lists
  and the qemu side facility mask during build time (5/16)
- gen-facilities added to make magic (5/16)
- element of struct S390CPUMachineProps now statically in cpu class (6/16)
- element of struct S390CPUProcessorProps now statically in cpu class (6/16)
- facility list also static now (6/16)
- typo fixed (7/16)
- zBC12-ga1 model now active on zEC12-ga2 host (11/16)
- operations on facility lists use QEMU bitmap API now (11/16)
- routine s390_cpu_model_init() introduced, called during cpu object
  realization to prepare the current accelarator (12/16) if a cpu
  model was selected
- missing comment added in description of CpuModelInfo type (13/16)
- accelerator field now mandatory for "query-cpu-model" (13/16)
- sorted list related comment to "query-cpu-definitions" dropped in
  commit message (13/16)
- comment for AccelCpuInfo type updated (13/16)
- routine s390_facility_test() factored out (15/16)

v1-v2:
- QEMU-side facility list mask introduced: this allows to enable guest
  facilities that are handled by instruction interception handlers
  implemented on qemu side. Similar to the facilities enabled by means
  of the KVM side facility list mask which are handled by kvm/kernel.
- Concept of soft facilities has been dropped 
- Result type of QMP command query-cpu-definitions extended to hold
  additional information beside the cpu model name including which
  cpu model is runnable in current accelerator and machine context. 

Michael Mueller (16):
  Introduce probe mode for machine type none
  Introduce option --probe to switch into probe mode
  Introduce stub routine cpu_desc_avail
  target-s390x: Introduce cpu facilities
  target-s390x: Generate facility defines per cpu model
  target-s390x: Introduce cpu models
  target-s390x: Define cpu model specific facility lists
  target-s390x: Add cpu model alias definition routines
  target-s390x: Update linux-headers/asm-s390/kvm.h
  target-s390x: Add KVM VM attribute interface for cpu models
  target-s390x: Add cpu class initialization routines
  target-s390x: Prepare accelerator during cpu object realization
  target-s390x: New QMP command query-cpu-model
  target-s390x: Extend QMP command query-cpu-definitions
  target-s390x: Introduce facility test routine
  target-s390x: Enable cpu model usage

 Makefile.target               |   2 +-
 accel.c                       |  36 ++-
 hw/s390x/s390-virtio.c        |  12 +-
 include/hw/boards.h           |   1 +
 include/qemu-common.h         |   2 +
 include/sysemu/accel.h        |   2 +-
 include/sysemu/arch_init.h    |   1 +
 include/sysemu/kvm.h          |  10 +
 kvm-all.c                     |   3 +
 linux-headers/asm-s390/kvm.h  |  20 ++
 qapi-schema.json              |  57 +++-
 qemu-options.hx               |   8 +
 qmp-commands.hx               |   6 +
 qmp.c                         |   5 +
 rules.mak                     |   3 +
 stubs/Makefile.objs           |   2 +
 stubs/arch-query-cpu-mod.c    |   9 +
 stubs/cpu-desc-avail.c        |   6 +
 target-s390x/Makefile.objs    |  19 ++
 target-s390x/cpu-facilities.h |  76 +++++
 target-s390x/cpu-models.c     | 730 ++++++++++++++++++++++++++++++++++++++++++
 target-s390x/cpu-models.h     | 162 ++++++++++
 target-s390x/cpu-qom.h        |  25 ++
 target-s390x/cpu.c            | 149 ++++++++-
 target-s390x/gen-facilities.c | 401 +++++++++++++++++++++++
 target-s390x/helper.c         |   9 +-
 target-s390x/kvm.c            | 104 ++++++
 trace-events                  |   3 +
 vl.c                          |   9 +-
 29 files changed, 1845 insertions(+), 27 deletions(-)
 create mode 100644 stubs/arch-query-cpu-mod.c
 create mode 100644 stubs/cpu-desc-avail.c
 create mode 100644 target-s390x/cpu-facilities.h
 create mode 100644 target-s390x/cpu-models.c
 create mode 100644 target-s390x/cpu-models.h
 create mode 100644 target-s390x/gen-facilities.c

-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists