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@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
*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(-)
