[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[qemu-s390x] [PATCH v3 00/33] Multi-phase reset mechanism

From: Damien Hedde
Subject: [qemu-s390x] [PATCH v3 00/33] Multi-phase reset mechanism
Date: Mon, 29 Jul 2019 16:56:21 +0200

Hi all,

Here's the third version of the multi-phase reset proposal patches. I've dropped
the RFC tag. Previous version can be found here:

The purpose of this series is to split the current reset procedure into 3
phases. We need 2 phases to separate what's local from what's doing external
side effects (IO for example). And we need a 3rd additional phase to handle
staying in reset state.

This serie has side-effects on ~20 files calling reset qdev/device api. In
consequence there is a lot of people in the CC list, most of them because of 2 
simple patches.

# Overview

The phases are:
INIT PHASE: Reset the object internal state, put a resetting flag and do the
    same for the reset subtree. No side effect on other devices to guarantee
    that, in a reset domain, everything get initialized first. This corresponds
    mostly to what is currently done in the device/bus reset method.

HOLD PHASE: This phase allows to control a reset with a IO. When a IO control
    reset procedure based on the IO level (not edge), we may need to assert
    the reset, wait some time, and finally de-assert the reset. The consequence
    is that such a device can stay in a "resetting state" and may need to show
    this state to other devices through its outputs. For example, a clock
    controller will typically shutdown its clocks when it is in resetting state.

EXIT PHASE: This phase sets outputs to state after reset. For a clock controller
     it starts the clocks. It also clears the "resetting" flag. A device should
     not react to inputs until this flag has been cleared. During this phase,
     outputs are propagated.

The reset use a new QOM interface: "Resettable". It is detailed in the added
documentation. Here's an overview, it contains the following methods:
  - 3 methods corresponding to the 3 phases, to be implemented by specialized
  - others methods do handle the reset mechanics, implemented by base classes:
    + reset_count getter/modifier methods to know when we enter or leave reset
      if multiple reset sources are enabled at the same time.
    + set_cold method to handle warm/cold reset
    + foreach_child to do something on reset children. This method allow to
      build some reset hirearchy.

The reset order of the phase is the following
  - reset assertation
    1. init phases in parent-to-child order
    2. hold phases in child-to-parent order
  - reset de-assertation
    3. exit phases in child-to-parent order

# Reset deprecation

The switch to multi-phase reset deprecates the current way of handling reset
(for users or developers of Devices). It used a reset method in both
DeviceClass and BusClass.

1. qdev/qbus_reset_all

Theses functions are deprecated by this series and are replaced by included
patches without change of behavior.

2. old's device_reset

There was a few call to this function, I renamed it *device_legacy_reset* to
handle the transition. This function allowed to reset only a given device 
(and not its eventual qbus subtree). This behavior is not possible with
the Resettable interface. I changed existing calls. Most of the time there is
no change of behavior because devices have no sub-buses.
Here the summary after checking each one (I've reproduced the list made by
Peter with some additional commentary of myself):

  -- used by the HDA device to reset the HDACodecDevice, which I think has
     no child buses
  -- resets the SynICState, which I think has no child buses
  -- resets the APIC, which has no child buses. (This reset is only
     done as a workaround for lack of reset phases: the whole machine
     is reset and then the APIC is re-reset last to undo any changes
     that other devices might have made to it. Probably making the APIC
     support phased reset would allow us to drop this hack.)
  -- called here to reset the MicroDriveState object. It does have
     a child bus, but it seems to me that potential children (ide-cd,
     ide-hd or ide-drive) have no implementation of the reset method.
  -- resets the SpaprXive device, which I think has no child buses
  -- resets a XiveSource, which I think has no child buses
  -- resets every QOM child of the SpaprPhbState. This one will
     require closer checking, but my guess is that if there's a child
     with a child bus then failure to reset the bus would be a bug.
  -- resets an SpaprTceTable, which I think has no child buses.
  -- resets the S390PCIBusDevice Needs S390 expertise, but probably
     either no child buses or failure to reset them is a bug.
  -- resets an individual SCSIDevice. Specialization of this class ("generic"
     or "disks") don't have child buses I think.
  -- resetting an SDState, which has no child bus
  -- resetting an SDState, which has no child bus

In my opinion only hw/ppc/spapr_pci.c and hw/s390x/s390-pci-inst.c may imply
behavior changes.

3. DeviceClass's reset and BusClass's methods

This is the major change. The method is deprecated because, methods are now
located in the interface class. In the series, I make the exit phase redirect
to the original reset method of DeviceClass (or BusClass). There a lot of use
of the method and transitioning to 3-phases only reset will need some work I
did not address in this series.

# Migration

For devices, I've added a patch to automate the addition of reset related
subsection. In consequence it is not necessary to explicetely add the
reset subsection in every device specialization requiring it. Right know
this is kind of a hack into qdev to dynamically modify the vmsd before the
registration. There is probably a much cleaner way to do this but I prefered
to demonstrate it by keeping modification local to qdev.
As far as I can tell it's ok to dynamically add subsections at the end. This
does not prevent from further adding subsections in the orignal vmsd: the
subsections order in the array is irrelevant from migration point-of-view.
The loading part just lookup each subsection in the array by name uppon

For buses, I handle them in the parent parent post_load. It has some limitation
but should work pretty well for most of cases (if we hold reset qdev/qbus
subtree starting from a device not a bus). If we consider holding part of a soc
under reset goes through gpio, it have to be a device.

# Change in qdev/qbus hiearchy

It is possible to change the parent bus of a device with the
qdev_set_parent_bus function. We could add some code to adapt the resetting
count of the device if old and new bus have different reset state. I did not do
it right now because it doesnt not fill the right place do this: as long as 
hiearchy follows the qdev hiearchy this would be fine place, but there is
nothing enforcing that with Resettable.

Given that, there is actually no bus help in reset. The only case when it can
happen is during the reset of a bus. Running the qom-test which test every
machine (with the initial main reset) gives one case where it happens: the raspi
because sd-card is moved during the reset. IMO, during reset, it's hazardous to
change qdev/qbus hierarchy since reset follows it so I'll try to address this on
a separate patch.

# Test

I tested this using the xilinx-zynq-a9 machine. Reset gpio and migration during
reset works.

# Changes since v2

  - don't call device implementations of reset phase functions multiple times.
  - init phase are executed parent-to-children order, legacy reset method is
    mapped on exit phase in order to keep the legacy order.
  - migration: poc of automatic addition of subsection.
  - migration: basic support for sub-buses.
  - qdev/qbus_reset_all & device_reset replacement

The series is organised as follow:
  - Patch   1 adds Resettable interface
  - Patches 2 and 3 rename device_reset function by device_legacy_reset to
    prepare the transition.
  - Patches 4 to 8 makes the changes in Device and Bus classes.
  - Patche 9 adds some doc
  - Patches 10 to 17 replace qdev/qbus_reset calls
  - Patches 18 to 27 replace device_reset calls
  - Patch 28 cleans remaining legacy reset API
  - Patches 29 to 33 modify the xilinx_zynq to add 3-phases reset support in the
    uart and the slcr (the reset controller of the soc).


Damien Hedde (33):
  Create Resettable QOM interface
  add temporary device_legacy_reset function to replace device_reset
  Replace all call to device_reset by call to device_legacy_reset
  make Device and Bus Resettable
  Switch to new api in qdev/bus
  add the vmstate description for device reset state
  automatically add vmstate for reset support in devices
  Add function to control reset with gpio inputs
  add doc about Resettable interface
  vl.c: remove qbus_reset_all registration
  hw/s390x/ipl.c: remove qbus_reset_all registration
  hw/pci/: remove qdev/qbus_reset_all call
  hw/scsi/: remove qdev/qbus_reset_all call
  hw/s390x/s390-virtio-ccw.c: remove qdev_reset_all call
  hw/ide/piix.c: remove qdev_reset_all call
  hw/input/adb.c: remove qdev_reset_all call
  hw/usb/dev-uas.c: remove qdev_reset_all call
  hw/audio/intel-hda.c: remove device_legacy_reset call
  hw/sd/pl181.c & omap_mmc.c: remove device_legacy_reset call
  hw/hyperv/hyperv.c: remove device_legacy_reset call
  hw/intc/spapr_xive.c: remove device_legacy_reset call
  hw/ppc/pnv_psi.c: remove device_legacy_reset call
  hw/scsi/vmw_pvscsi.c: remove device_legacy_reset call
  hw/ppc/spapr: remove device_legacy_reset call
  hw/i386/pc.c: remove device_legacy_reset call
  hw/s390x/s390-pci-inst.c: remove device_legacy_reset call
  hw/ide/microdrive.c: remove device_legacy_reset calls
  qdev: Remove unused deprecated reset functions
  hw/misc/zynq_slcr: use standard register definition
  convert cadence_uart to 3-phases reset
  Convert zynq's slcr to 3-phases reset
  Add uart reset support in zynq_slcr
  Connect the uart reset gpios in the zynq platform

 Makefile.objs                  |   1 +
 docs/devel/reset.txt           | 165 ++++++++++
 hw/arm/xilinx_zynq.c           |  14 +-
 hw/audio/intel-hda.c           |   2 +-
 hw/char/cadence_uart.c         |  77 ++++-
 hw/core/Makefile.objs          |   2 +
 hw/core/bus.c                  |  85 ++++++
 hw/core/qdev-vmstate.c         | 101 ++++++
 hw/core/qdev.c                 | 169 ++++++++--
 hw/core/resettable.c           | 220 +++++++++++++
 hw/core/trace-events           |  39 +++
 hw/hyperv/hyperv.c             |   2 +-
 hw/i386/pc.c                   |   2 +-
 hw/ide/microdrive.c            |   8 +-
 hw/ide/piix.c                  |   2 +-
 hw/input/adb.c                 |   2 +-
 hw/intc/spapr_xive.c           |   2 +-
 hw/misc/zynq_slcr.c            | 543 ++++++++++++++++++---------------
 hw/pci/pci.c                   |   6 +-
 hw/pci/pci_bridge.c            |   2 +-
 hw/ppc/pnv_psi.c               |   2 +-
 hw/ppc/spapr_pci.c             |   2 +-
 hw/ppc/spapr_vio.c             |   2 +-
 hw/s390x/ipl.c                 |   6 +-
 hw/s390x/s390-pci-inst.c       |   2 +-
 hw/s390x/s390-virtio-ccw.c     |   2 +-
 hw/scsi/lsi53c895a.c           |   4 +-
 hw/scsi/megasas.c              |   2 +-
 hw/scsi/mptsas.c               |   8 +-
 hw/scsi/spapr_vscsi.c          |   2 +-
 hw/scsi/virtio-scsi.c          |   6 +-
 hw/scsi/vmw_pvscsi.c           |   6 +-
 hw/sd/omap_mmc.c               |   2 +-
 hw/sd/pl181.c                  |   2 +-
 hw/usb/dev-uas.c               |   2 +-
 include/hw/char/cadence_uart.h |  10 +-
 include/hw/qdev-core.h         | 158 +++++++++-
 include/hw/resettable.h        | 126 ++++++++
 stubs/Makefile.objs            |   1 +
 stubs/device.c                 |   7 +
 tests/Makefile.include         |   1 +
 vl.c                           |   6 +-
 42 files changed, 1470 insertions(+), 333 deletions(-)
 create mode 100644 docs/devel/reset.txt
 create mode 100644 hw/core/qdev-vmstate.c
 create mode 100644 hw/core/resettable.c
 create mode 100644 hw/core/trace-events
 create mode 100644 include/hw/resettable.h
 create mode 100644 stubs/device.c


reply via email to

[Prev in Thread] Current Thread [Next in Thread]