[Skiboot] [PATCH v12 00/23] PCI Slot Managment
Gavin Shan
gwshan at linux.vnet.ibm.com
Mon Jun 20 21:58:54 AEST 2016
On Mon, Jun 20, 2016 at 04:02:11PM +1000, Stewart Smith wrote:
>Stewart Smith <stewart at linux.vnet.ibm.com> writes:
>> Gavin Shan <gwshan at linux.vnet.ibm.com> writes:
>>> The series of patches intend to support PCI slots for various tasks like
>>> retrieving PCI slot status, PCI slot reset, and PCI slot hotplug. The
>>> design is highlighted as below:
>>>
>>> * Each PHB is expected to have one associated "PHB" slot, which is
>>> created by platform.
>>> * Each PCI slot is created by platform as well, which might inherit
>>> some properties from PHB slot.
>>> * Currently, Firenze is the only one platform that populates PCI slots.
>>> * Each PCI slot has couple of callbacks implemented to support various
>>> functions it should support in pci-slot.c, which can be overrided by
>>> platforms to have their own particular behaviour or PCI slot properties.
>>> * The PCI slot functionality is implemented with state machine, which
>>> is inherited from Ben's original code to support PHBs. The PCI slots
>>> allowed to be overrided by platforms as well.
>>> * When removing or adding adapter to one particular PCI slot, the device
>>> nodes tied through the PCI slot will be removed or added (after rescanning).
>>> All device-tree changes is formated as message sent to OS, which needs
>>> those inforamtion to populate pci_dn and eeh_dev.
>>>
>>> Changelog
>>> =========
>>> v12:
>>> * Comments cleanup according to Russell's input
>>> * Code refactoring according to Andrew, Russell, Stewart's comments
>>> * Alistair's comments to merge the states will be done in future.
>>> v11:
>>> * Serialize the taks in cpu.c as Steward suggested.
>>> * opal_get_device_tree() is extended to return the memory length.
>>> * opal_pci_set_power_state() adds/removes PCI devices and device nodes
>>> that was done in platform backend. The backend function just issues
>>> I2C request to power off/on the slot.
>>> * opal_pci_set_power_state() is extended to slot's downstream adapters
>>> to offline or online without changing its power state as CAPI developers
>>> asked.
>>> * Documenting improvement.
>>> * Patch split and merge.
>>> v10:
>>> * Rebased to commit 0e47de7 ("mambo: Flatten device tree at the end")
>>> plus 3 additional patches: https://patchwork.ozlabs.org/patch/614760/
>>> * uint64_t argument for newly introduced APIs for extension in future.
>>> * Introduced opal_pci_poll2() that accepts two arguments.
>>> * Fix misc comments from Daniel.
>>> * Several patches that were merged to upstream and dropped from the
>>> series.
>>> v9:
>>> * Rebased to commit dfab2fb ("Fix endian flip in printf for FSP in
>>> hdata/fsp.c").
>>> * Patch split/merge for easier review.
>>> * Add retry mechanism required by fundamental reset on PHB3.
>>> * Created PHB slots for NPU PHBs.
>>> * Support PCI slots on astbmc platforms.
>>> * Fixed PCI slot power status and presence status with help of macros
>>> in opal-api.h.
>>> * More description added to doc/pci-slot.txt for the message sent at
>>> PCI hotplug/unplug time.
>>> v8:
>>> * Removed dt_counter. To transfer device sub-tree under the specified
>>> node with OPAL API opal_get_device_tree.
>>> v7:
>>> * Use dt_counter to track device-tree change.
>>> v6:
>>> * Rebased to commit 0d275bc ("Fix boot-tests for BMC machines due to FSP
>>> password in dotfile").
>>> * Remove SDC/SDL check when sending I2C data to slave.
>>> * Allow to change the default I2C request timeout, instead of disabling it.
>>> * Split PCI slot core to two files (pci-slot.c and pcie-slot.c).
>>> * PHB slot is created by default. It's not created by individual platforms
>>> any more.
>>> * Use flat device-tree blob to convey device sub-tree. The original one
>>> based on strings has been dropped.
>>> * Merged PATCH[v5 12/15] ("PCI: Reinitialize devices on slot reset") to
>>> PATCH[v5 7/15] ("core/pci: PCI slot management core").
>>> * Commit log cleanup.
>>> v5:
>>> * Fix I2C issues caused by wrong I2C master frequency.
>>> * Fix reserved PE# when unmapping RID on PHB3.
>>> * Add logic to support PCI hotplug: When one adapter is removed from the
>>> slot, the power supply and the device nodes tied through the slot will
>>> be cut off. When adding one adapter to PCI slot, the PCI bus leading
>>> from the slot will be rescanned. On any PCI topology changes, the device
>>> tree changes will be sent to OS for updating.
>>> * Refactor the PCI slot states and remove unnecessary ones.
>>>
>>> Gavin Shan (23):
>>> hw/p8-i2c: Allow to set I2C request timeout
>>> core/fdt: Improve debugging functionality
>>> core/fdt: Introduce flatten_dt_properties()
>>> core/fdt: Allow to specify FDT blob
>>> core/fdt: Introduce __create_dtb()
>>> core/fdt: Allow to exclude root node
>>> core/fdt: Introduce opal_get_device_tree()
>>> core/cpu: Introduce DEBUG_SERIALIZE_CPU_JOBS
>>> core/pci: Fix wrong reserved PE# in enumeration
>>> core/pci: Extend pci_walk_dev() for PCI slot
>>> core/pci: Extend pci_restore_bridge_buses() for PCI slot
>>> core/pci: Return value for struct phb_ops::device_init
>>> core/pci: Support PCI slot
>>> hw/p7ioc: Support PHB slot
>>> hw/phb3: Remove useless comments
>>> hw/phb3: Support PHB slot
>>> hw/npu: Support PHB slot
>>> platforms/ibm-fsp: Support PCI slot
>>> platforms/astbmc: Support PCI slot
>>> core/pci: Use PCI slot in enumeration
>>> core/pci: Power off empty hotpluggable slot
>>> core/opal: Support PCI slot and new APIs
>>> doc: PCI slot
>>>
>>> core/Makefile.inc | 4 +-
>>> core/cpu.c | 5 +
>>> core/fdt.c | 171 ++--
>>> core/init.c | 2 +-
>>> core/pci-opal.c | 235 ++++-
>>> core/pci-slot.c | 206 ++++
>>> core/pci.c | 341 ++++---
>>> core/pcie-slot.c | 452 +++++++++
>>> doc/opal-api/opal-get-device-tree-118.txt | 24 +
>>> doc/opal-api/opal-pci-get-power-state-120.txt | 19 +
>>> doc/opal-api/opal-pci-get-presence-state-119.txt | 22 +
>>> doc/opal-api/opal-pci-set-power-state-121.txt | 36 +
>>> doc/pci-slot-properties.txt | 17 -
>>> doc/pci-slot.txt | 119 +++
>>> hw/npu.c | 62 +-
>>> hw/p7ioc-phb.c | 1152 +++++++++-------------
>>> hw/p8-i2c.c | 25 +-
>>> hw/phb3.c | 707 ++++++-------
>>> include/i2c.h | 9 +
>>> include/opal-api.h | 18 +-
>>> include/p7ioc.h | 48 +-
>>> include/pci-slot.h | 255 +++++
>>> include/pci.h | 162 +--
>>> include/phb3.h | 54 +-
>>> include/platform.h | 1 +
>>> include/skiboot.h | 2 +-
>>> platforms/astbmc/slots.c | 58 +-
>>> platforms/ibm-fsp/Makefile.inc | 3 +-
>>> platforms/ibm-fsp/apollo-pci.c | 94 ++
>>> platforms/ibm-fsp/apollo.c | 23 +-
>>> platforms/ibm-fsp/firenze-pci.c | 957 ++++++++++++++++++
>>> platforms/ibm-fsp/firenze.c | 345 +------
>>> platforms/ibm-fsp/ibm-fsp.h | 13 +
>>> platforms/ibm-fsp/lxvpd.c | 375 ++++---
>>> platforms/ibm-fsp/lxvpd.h | 35 +-
>>> 35 files changed, 3913 insertions(+), 2138 deletions(-)
>>> create mode 100644 core/pci-slot.c
>>> create mode 100644 core/pcie-slot.c
>>> create mode 100644 doc/opal-api/opal-get-device-tree-118.txt
>>> create mode 100644 doc/opal-api/opal-pci-get-power-state-120.txt
>>> create mode 100644 doc/opal-api/opal-pci-get-presence-state-119.txt
>>> create mode 100644 doc/opal-api/opal-pci-set-power-state-121.txt
>>> delete mode 100644 doc/pci-slot-properties.txt
>>> create mode 100644 doc/pci-slot.txt
>>> create mode 100644 include/pci-slot.h
>>> create mode 100644 platforms/ibm-fsp/apollo-pci.c
>>> create mode 100644 platforms/ibm-fsp/firenze-pci.c
>>>
>>
>> Thanks! After a few revisions and everything, I just merged the series
>> to master as of 8c2f3fe.
>
>As discussed on IRC, something here seems to break slot names on
>OpenPOWER platforms - and I eagerly await your patch :)
>
Here's the fix as we talked on IRC:
https://patchwork.ozlabs.org/patch/637828/
Thanks,
Gavin
>--
>Stewart Smith
>OPAL Architect, IBM.
More information about the Skiboot
mailing list