Re: [RFC PATCH v3 2/2] fpga: set owner of fpga_manager_ops for existing low-level modules

[Marco Pagani]

On 2023-12-19 16:10, Greg Kroah-Hartman wrote:
> On Tue, Dec 19, 2023 at 03:54:25PM +0100, Marco Pagani wrote:
>>
>>
>> On 2023-12-18 21:33, Greg Kroah-Hartman wrote:
>>> On Mon, Dec 18, 2023 at 09:28:09PM +0100, Marco Pagani wrote:
>>>> This patch tentatively set the owner field of fpga_manager_ops to
>>>> THIS_MODULE for existing fpga manager low-level control modules.
>>>>
>>>> Signed-off-by: Marco Pagani <marpagan@xxxxxxxxxx>
>>>> ---
>>>>  drivers/fpga/altera-cvp.c             | 1 +
>>>>  drivers/fpga/altera-pr-ip-core.c      | 1 +
>>>>  drivers/fpga/altera-ps-spi.c          | 1 +
>>>>  drivers/fpga/dfl-fme-mgr.c            | 1 +
>>>>  drivers/fpga/ice40-spi.c              | 1 +
>>>>  drivers/fpga/lattice-sysconfig.c      | 1 +
>>>>  drivers/fpga/machxo2-spi.c            | 1 +
>>>>  drivers/fpga/microchip-spi.c          | 1 +
>>>>  drivers/fpga/socfpga-a10.c            | 1 +
>>>>  drivers/fpga/socfpga.c                | 1 +
>>>>  drivers/fpga/stratix10-soc.c          | 1 +
>>>>  drivers/fpga/tests/fpga-mgr-test.c    | 1 +
>>>>  drivers/fpga/tests/fpga-region-test.c | 1 +
>>>>  drivers/fpga/ts73xx-fpga.c            | 1 +
>>>>  drivers/fpga/versal-fpga.c            | 1 +
>>>>  drivers/fpga/xilinx-spi.c             | 1 +
>>>>  drivers/fpga/zynq-fpga.c              | 1 +
>>>>  drivers/fpga/zynqmp-fpga.c            | 1 +
>>>>  18 files changed, 18 insertions(+)
>>>>
>>>> diff --git a/drivers/fpga/altera-cvp.c b/drivers/fpga/altera-cvp.c
>>>> index 4ffb9da537d8..aeb913547dd8 100644
>>>> --- a/drivers/fpga/altera-cvp.c
>>>> +++ b/drivers/fpga/altera-cvp.c
>>>> @@ -520,6 +520,7 @@ static const struct fpga_manager_ops altera_cvp_ops = {
>>>>  	.write_init	= altera_cvp_write_init,
>>>>  	.write		= altera_cvp_write,
>>>>  	.write_complete	= altera_cvp_write_complete,
>>>> +	.owner		= THIS_MODULE,
>>>
>>> Note, this is not how to do this, force the compiler to set this for you
>>> automatically, otherwise everyone will always forget to do it.  Look at
>>> how functions like usb_register_driver() works.
>>>
>>> Also, are you _sure_ that you need a module owner in this structure?  I
>>> still don't know why...
>>>
>>
>> Do you mean moving the module owner field to the manager context and setting
>> it during registration with a helper macro?
> 
> I mean set it during registration with a helper macro.
> 
>> Something like:
>>
>> struct fpga_manager {
>> 	...
>> 	struct module *owner;
>> };
>>
>> #define fpga_mgr_register(parent, ...) \
>> 	__fpga_mgr_register(parent,..., THIS_MODULE)
>>
>> struct fpga_manager *
>> __fpga_mgr_register(struct device *parent, ..., struct module *owner)
>> {
>> 	...
>> 	mgr->owner = owner;
>> }
> 
> Yes.
> 
> But again, is a module owner even needed?  I don't think you all have
> proven that yet...

Programming an FPGA involves a potentially lengthy sequence of interactions
with the reconfiguration engine. The manager conceptually organizes these
interactions as a sequence of ops. Low-level modules implement these ops/steps
for a specific device. If we don't protect the low-level module, someone might
unload it right when we are in the middle of a low-level op programming the
FPGA. As far as I know, the kernel would crash in that case.

Thanks,
Marco