Re: [RESEND PATCH v10 0/3] fpga: Use standard class dev_release function

[Tom Rix <trix@redhat.com>]

On 7/27/21 2:07 PM, Russ Weight wrote:
>
> On 7/27/21 10:44 AM, Tom Rix wrote:
>> On 7/27/21 10:16 AM, Russ Weight wrote:
>>> On 7/27/21 9:55 AM, Tom Rix wrote:
>>>> On 7/27/21 9:38 AM, Russ Weight wrote:
>>>>> On 7/27/21 9:22 AM, Tom Rix wrote:
>>>>>> On 7/27/21 9:15 AM, Russ Weight wrote:
>>>>>>> On 7/27/21 6:24 AM, Tom Rix wrote:
>>>>>>>> On 7/26/21 2:33 PM, Russ Weight wrote:
>>>>>>>>> Hi Moritz,
>>>>>>>>>
>>>>>>>>> I sent this version of the patchset out on July 8th. The only discussion has
>>>>>>>>> been around the compat-id structure. I think there is agreement that this can
>>>>>>>>> be treated separately. What are your thoughts? Do you think this patchset is
>>>>>>>>> ready?
>>>>>>>> While I agree the compat_id can be treated separately, i still have a problem with the 2 register functions.
>>>>>>>>
>>>>>>>> In this discussion,
>>>>>>>>
>>>>>>>> https://lore.kernel.org/linux-fpga/DM6PR11MB38194D113950FA4B75C7F2C485E49@DM6PR11MB3819.namprd11.prod.outlook.com/
>>>>>>>>
>>>>>>>> *register_full(..., *info) becomes
>>>>>>>>
>>>>>>>> *register(..., *info)
>>>>>>>>
>>>>>>>> and
>>>>>>>>
>>>>>>>> *register_simple() is moved out of the public api and becomes private to the subsystem.
>>>>>>> Somehow I missed that part of the email. I'm not sure I understand the part about
>>>>>>> register_simple() "becomes private to the subsystem". Most users would call the simple
>>>>>>> version of the API. Is the proposal to replicate the register_simple() interface
>>>>>>> for each driver that calls it?
>>>>>> Still use *register_simple() internally, so most of your patchset is unchanged.
>>>>>>
>>>>>> Move it out the public api. include/linux/fpga-mgr.h into someplace appropriate in drivers/fpga/
>>>>> Ok - I'll do another spin of the patch set.
>>>>>
>>>>> Just to be clear, *register_full() gets renamed back to *register(). The current *register()
>>>>> goes back to *register_simple(). *register_simple() is an exported symbol, but the function
>>>>> prototype moves out of the public header file into private header file under drivers/fpga.
>>>> Mostly.
>>>>
>>>> If register_simple does not need to be exported, it shouldn't be.
>>> If *register_simple() is to reside in the fpga_mgr or fpga_region module and be
>>> called by other modules, then it has to be exported... unless it is defined as a
>>> static function in a header file. Shall I go that route instead? I could create
>>> a new drivers/fpga/fpga.h file and add the *register_simple() functions there.
>>> Or I could create separate drivers/fpga/fpga-mgr.h and drivers/fpga/fpga-region.h
>>> header files with one static function each.
>> good point, exporting ok
> Before I do this, can I get others to chime in? There was a previous patchset that
> used *register_simple() and *register(). It seemed that the consensus at that time
> was that it should be *register() and *register_full(). I'm starting to feel like
> I'm going in circles...
>
> I personally don't have a problem with two public flavors of the register function
> (the way it is now), what do other's think? If it really is a problem to have two
> versions of the register function, maybe I should just change all of the callers
> to declare a structure and call the same register function?

It has been about 4 weeks and no one has chimed in.

Changing a public api is painful and should be done infrequently.

There should be only 1 register() call.

The introduction of the register(..., *info) is driven by compat_id.

compat_id is a broken api. I have been around with several ways to fix it.

So it looks like compat_id is going to stay so we will need the 
register(..., *info) version.

using *info is generally better for api stablity-wise than what we have 
now, tacking on parameters as needed.

how you want to get to a single public register(..., *info) call is up 
to you.

Changing all callers of register to use *info instead of the wrapper is 
a good way to go.

Tom

>
> - Russ
>
>
>> Tom
>>
>>> - Russ
>>>
>>>> Tom
>>>>
>>>>> - Russ
>>>>>
>>>>>> Tom
>>>>>>
>>>>>>>> Tom
>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>>>> - Russ
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 7/26/21 2:27 PM, Russ Weight wrote:
>>>>>>>>>> The FPGA framework has a convention of using managed resource functions
>>>>>>>>>> to allow parent drivers to manage the data structures allocated by the
>>>>>>>>>> class drivers. They use an empty *_dev_release() function to satisfy the
>>>>>>>>>> class driver.
>>>>>>>>>>
>>>>>>>>>> This is inconsistent with linux driver model.
>>>>>>>>>>
>>>>>>>>>> These changes remove the managed resource functions and populate the class
>>>>>>>>>> dev_release callback functions. They also merge the create() and register()
>>>>>>>>>> functions into a single register() or register_full() function for each of
>>>>>>>>>> the fpga-mgr, fpga-region, and fpga-bridge class drivers.
>>>>>>>>>>
>>>>>>>>>> The new *register_full() functions accept an info data structure to provide
>>>>>>>>>> flexibility in passing optional parameters. The *register() functions
>>>>>>>>>> support the legacy parameter list for users that don't require the use of
>>>>>>>>>> optional parameters.
>>>>>>>>>>
>>>>>>>>>> For more context, refer to this email thread:
>>>>>>>>>>
>>>>>>>>>> https://marc.info/?l=linux-fpga&m=162127412218557&w=2
>>>>>>>>>>
>>>>>>>>>> I turned on the configs assocated with each of the modified files, but I
>>>>>>>>>> must have been missing some dependencies, because not all of them compiled.
>>>>>>>>>> I did a run-time test specifically with the dfl-fme infrastructure. This
>>>>>>>>>> would have exercised the region, bridge, and fpga-mgr frameworks.
>>>>>>>>>>
>>>>>>>>>> Changelog v9 -> v10:
>>>>>>>>>>        - Fixed commit messages to reference register_full() instead of
>>>>>>>>>>          register_simple().
>>>>>>>>>>        - Removed the fpga_bridge_register_full() function, because there is
>>>>>>>>>>          not need for it yet. Updated the documentation and commit message
>>>>>>>>>>          accordingly.
>>>>>>>>>>        - Updated documentation to reference the fpga_manager_info and
>>>>>>>>>>          fpga_region_info structures.
>>>>>>>>>>
>>>>>>>>>> Changelog v8 -> v9:
>>>>>>>>>>        - Cleaned up documentation for the FPGA Manager, Bridge, and Region
>>>>>>>>>>          register functions
>>>>>>>>>>        - Renamed fpga_*_register() to fpga_*_register_full()
>>>>>>>>>>        - Renamed fpga_*_register_simple() to fpga_*_register()
>>>>>>>>>>        - Renamed devm_fpga_mgr_register() to devm_fpga_mgr_register_full()
>>>>>>>>>>        - Renamed devm_fpga_mgr_register_simple() to devm_fpga_mgr_register()
>>>>>>>>>>
>>>>>>>>>> Changelog v7 -> v8:
>>>>>>>>>>        - Added reviewed-by tags.
>>>>>>>>>>        - Updated Documentation/driver-api/fpga/ files: fpga-mgr.rst,
>>>>>>>>>>          fpga-bridge.rst, and fpga-region.rst.
>>>>>>>>>>
>>>>>>>>>> Changelog v6 -> v7:
>>>>>>>>>>        - Update the commit messages to describe the new parameters for the
>>>>>>>>>>          *register() functions and to mention the *register_simple() functions.
>>>>>>>>>>        - Fix function prototypes in header file to rename dev to parent.
>>>>>>>>>>        - Make use of the PTR_ERR_OR_ZERO() macro when possible.
>>>>>>>>>>        - Some cleanup of comments.
>>>>>>>>>>        - Update function definitions/prototypes to apply const to the new info
>>>>>>>>>>          parameter.
>>>>>>>>>>        - Verify that info->br_ops is non-null in the fpga_bridge_register()
>>>>>>>>>>          function.
>>>>>>>>>>        - Verify a non-null info pointer in the fpga_region_register() function.
>>>>>>>>>>
>>>>>>>>>> Changelog v5 -> v6:
>>>>>>>>>>        - Moved FPGA manager/bridge/region optional parameters out of the ops
>>>>>>>>>>          structure and back into the FPGA class driver structure.
>>>>>>>>>>        - Changed fpga_*_register() function parameters to accept an info data
>>>>>>>>>>          structure to provide flexibility in passing optional parameters.
>>>>>>>>>>        - Added fpga_*_register_simple() functions to support current parameters
>>>>>>>>>>          for users that don't require use of optional parameters.
>>>>>>>>>>
>>>>>>>>>> Changelog v4 -> v5:
>>>>>>>>>>        - Rebased on top of recently accepted patches.
>>>>>>>>>>        - Removed compat_id from the fpga_mgr_register() parameter list
>>>>>>>>>>          and added it to the fpga_manager_ops structure. This also required
>>>>>>>>>>          dynamically allocating the dfl-fme-ops structure in order to add
>>>>>>>>>>          the appropriate compat_id.
>>>>>>>>>>        - Created the fpga_region_ops data structure which is optionally passed
>>>>>>>>>>          to fpga_region_register(). compat_id, the get_bridges() pointer, and
>>>>>>>>>>          the priv pointer are included in the fpga_region_ops structure.
>>>>>>>>>>
>>>>>>>>>> Changelog v3 -> v4:
>>>>>>>>>>        - Added the compat_id parameter to fpga_mgr_register() and
>>>>>>>>>>          devm_fpga_mgr_register() to ensure that the compat_id is set before
>>>>>>>>>>          the device_register() call.
>>>>>>>>>>        - Added the compat_id parameter to fpga_region_register() to ensure
>>>>>>>>>>          that the compat_id is set before the device_register() call.
>>>>>>>>>>        - Modified the dfl_fpga_feature_devs_enumerate() function to restore
>>>>>>>>>>          the fpga_region_register() call to the correct location.
>>>>>>>>>>
>>>>>>>>>> Changelog v2 -> v3:
>>>>>>>>>>        - Cleaned up comment headers for fpga_mgr_register(), fpga_bridge_register(),
>>>>>>>>>>          and fpga_region_register().
>>>>>>>>>>        - Fixed error return on ida_simple_get() failure for fpga_mgr_register(),
>>>>>>>>>>          fpga_bridge_register(), and fpga_region_register().
>>>>>>>>>>        - Fixed error return value for fpga_bridge_register(): ERR_PTR(ret) instead
>>>>>>>>>>          of NULL.
>>>>>>>>>>
>>>>>>>>>> Changelog v1 -> v2:
>>>>>>>>>>        - Restored devm_fpga_mgr_register() functionality to the fpga-mgr
>>>>>>>>>>          class driver, adapted for the combined create/register functionality.
>>>>>>>>>>        - All previous callers of devm_fpga_mgr_register() will continue to call
>>>>>>>>>>          devm_fpga_mgr_register().
>>>>>>>>>>        - replaced unnecessary ternary operators in return statements with
>>>>>>>>>>          standard if conditions.
>>>>>>>>>>
>>>>>>>>>> Russ Weight (3):
>>>>>>>>>>        fpga: mgr: Use standard dev_release for class driver
>>>>>>>>>>        fpga: bridge: Use standard dev_release for class driver
>>>>>>>>>>        fpga: region: Use standard dev_release for class driver
>>>>>>>>>>
>>>>>>>>>>       Documentation/driver-api/fpga/fpga-bridge.rst |   6 +-
>>>>>>>>>>       Documentation/driver-api/fpga/fpga-mgr.rst    |  38 +++-
>>>>>>>>>>       Documentation/driver-api/fpga/fpga-region.rst |  12 +-
>>>>>>>>>>       drivers/fpga/altera-cvp.c                     |  12 +-
>>>>>>>>>>       drivers/fpga/altera-fpga2sdram.c              |  12 +-
>>>>>>>>>>       drivers/fpga/altera-freeze-bridge.c           |  10 +-
>>>>>>>>>>       drivers/fpga/altera-hps2fpga.c                |  12 +-
>>>>>>>>>>       drivers/fpga/altera-pr-ip-core.c              |   7 +-
>>>>>>>>>>       drivers/fpga/altera-ps-spi.c                  |   9 +-
>>>>>>>>>>       drivers/fpga/dfl-fme-br.c                     |  10 +-
>>>>>>>>>>       drivers/fpga/dfl-fme-mgr.c                    |  22 +-
>>>>>>>>>>       drivers/fpga/dfl-fme-region.c                 |  17 +-
>>>>>>>>>>       drivers/fpga/dfl.c                            |  12 +-
>>>>>>>>>>       drivers/fpga/fpga-bridge.c                    | 122 +++-------
>>>>>>>>>>       drivers/fpga/fpga-mgr.c                       | 215 ++++++++----------
>>>>>>>>>>       drivers/fpga/fpga-region.c                    | 119 ++++------
>>>>>>>>>>       drivers/fpga/ice40-spi.c                      |   9 +-
>>>>>>>>>>       drivers/fpga/machxo2-spi.c                    |   9 +-
>>>>>>>>>>       drivers/fpga/of-fpga-region.c                 |  10 +-
>>>>>>>>>>       drivers/fpga/socfpga-a10.c                    |  16 +-
>>>>>>>>>>       drivers/fpga/socfpga.c                        |   9 +-
>>>>>>>>>>       drivers/fpga/stratix10-soc.c                  |  16 +-
>>>>>>>>>>       drivers/fpga/ts73xx-fpga.c                    |   9 +-
>>>>>>>>>>       drivers/fpga/xilinx-pr-decoupler.c            |  17 +-
>>>>>>>>>>       drivers/fpga/xilinx-spi.c                     |  11 +-
>>>>>>>>>>       drivers/fpga/zynq-fpga.c                      |  16 +-
>>>>>>>>>>       drivers/fpga/zynqmp-fpga.c                    |   9 +-
>>>>>>>>>>       include/linux/fpga/fpga-bridge.h              |  30 ++-
>>>>>>>>>>       include/linux/fpga/fpga-mgr.h                 |  62 +++--
>>>>>>>>>>       include/linux/fpga/fpga-region.h              |  36 ++-
>>>>>>>>>>       30 files changed, 383 insertions(+), 511 deletions(-)
>>>>>>>>>>