[PATCH v7 2/7] fpga: add fit_fpga_load function

[Michal Simek]

On 5/9/22 15:34, Adrian Fiergolski wrote:
> Michal,
> 
> On 09.05.2022 15:28, Oleksandr Suvorov wrote:
>> Hi Adrian,
>>
>> On Mon, May 9, 2022 at 2:35 PM Adrian Fiergolski
>> <adrian.fiergolski at fastree3d.com> wrote:
>>> Hi Oleksandr,
>>>
>>> On 09.05.2022 13:02, Oleksandr Suvorov wrote:
>>>> Hi Adrian,
>>>>
>>>> On Wed, May 4, 2022 at 5:28 PM Adrian Fiergolski
>>>> <adrian.fiergolski at fastree3d.com> wrote:
>>>>> Hi Oleksandr,
>>>>>
>>>>> On 03.05.2022 09:42, Michal Simek wrote:
>>>>>> On 4/11/22 20:00, Adrian Fiergolski wrote:
>>>>>>> From: Oleksandr Suvorov <oleksandr.suvorov at foundries.io>
>>>>>>>
>>>>>>> Introduce a function which passes an fpga compatible string from
>>>>>>> FIT images to FPGA drivers. This lets the different implementations
>>>>>>> decide how to handle it.
>>>>>>>
>>>>>>> Some code of Jorge Ramirez-Ortiz <jorge at foundries.io> is reused.
>>>>>>>
>>>>>>> Signed-off-by: Oleksandr Suvorov <oleksandr.suvorov at foundries.io>
>>>>>>> Tested-by: Ricardo Salveti <ricardo at foundries.io>
>>>>>>> Co-developed-by: Adrian Fiergolski <adrian.fiergolski at fastree3d.com>
>>>>>>> Signed-off-by: Adrian Fiergolski <adrian.fiergolski at fastree3d.com>
>>>>>>> ---
>>>>>>>     common/spl/spl_fit.c |  6 ++----
>>>>>>>     drivers/fpga/fpga.c  | 23 ++++++++++++++++++++++-
>>>>>>>     include/fpga.h       |  4 ++++
>>>>>>>     3 files changed, 28 insertions(+), 5 deletions(-)
>>>>>>>
>>>>>>> diff --git a/common/spl/spl_fit.c b/common/spl/spl_fit.c
>>>>>>> index 1bbf824684..0e3c2a94b6 100644
>>>>>>> --- a/common/spl/spl_fit.c
>>>>>>> +++ b/common/spl/spl_fit.c
>>>>>>> @@ -588,11 +588,9 @@ static int spl_fit_upload_fpga(struct
>>>>>>> spl_fit_info *ctx, int node,
>>>>>>>         compatible = fdt_getprop(ctx->fit, node, "compatible", NULL);
>>>>>>>         if (!compatible)
>>>>>>>             warn_deprecated("'fpga' image without 'compatible' property");
>>>>>>> -    else if (strcmp(compatible, "u-boot,fpga-legacy"))
>>>>>>> -        printf("Ignoring compatible = %s property\n", compatible);
>>>>>>>     -    ret = fpga_load(0, (void *)fpga_image->load_addr,
>>>>>>> fpga_image->size,
>>>>>>> -            BIT_FULL);
>>>>>>> +    ret = fit_fpga_load(0, (void *)fpga_image->load_addr,
>>>>>>> fpga_image->size,
>>>>>>> +                BIT_FULL, compatible);
>>>>>>>         if (ret) {
>>>>>>>             printf("%s: Cannot load the image to the FPGA\n", __func__);
>>>>>>>             return ret;
>>>>>>> diff --git a/drivers/fpga/fpga.c b/drivers/fpga/fpga.c
>>>>>>> index 3b0a44b242..a306dd81f9 100644
>>>>>>> --- a/drivers/fpga/fpga.c
>>>>>>> +++ b/drivers/fpga/fpga.c
>>>>>>> @@ -249,8 +249,23 @@ int fpga_loads(int devnum, const void *buf,
>>>>>>> size_t size,
>>>>>>>     }
>>>>>>>     #endif
>>>>>>>     +int fit_fpga_load(int devnum, const void *buf, size_t bsize,
>>>>>>> +          bitstream_type bstype, const char *compatible)
>>>>>>> +{
>>>>>>> +    fpga_desc *desc = fpga_get_desc(devnum);
>>>>>> this generates warning which you should fix.
>>>>>>
>>>>>> +  fpga_desc *desc = fpga_get_desc(devnum);
>>>>>> +                    ^~~~~~~~~~~~~
>>>>>> w+../drivers/fpga/fpga.c: In function ‘fit_fpga_load’:
>>>>>> w+../drivers/fpga/fpga.c:255:20: warning: initialization discards
>>>>>> ‘const’ qualifier from pointer target type [-Wdiscarded-qualifiers]
>>>>> As it's your patch, could you address it?
>>>>>
>>>>> The 'compatible' field can't be set here, as fpga_get_desc returns
>>>>> 'const fpga_desc * const'. The descriptors table is populated (fpga_add)
>>>>> in board_init. At that stage, I am afraid, we don't have access to the
>>>>> fit image information yet to set a proper 'compatible' (would need to
>>>>> check the code more carefully). Moreover, IMHO it's not the cleanest way
>>>>> to change the 'compatible' field after the driver's initialisation.
>>>> Obviously, the "compatible" attribute belongs to an image, not to an FPGA
>>>> driver. Unfortunately, I don't see an easy way to stick this attribute to an
>>>> image in the current architecture. Maybe I missed that way, so I'd appreciate
>>>> any help with this.
>>>>
>>>> Anyway, we could come back to the FPGA driver subsystem later and rework
>>>> it better and deeper.
>>>>
>>>> For now, I'd only fix the warning. Are you ok with this?
>>> I haven't seen straightforward implementation in the current
>>> architecture as well. However, it's this series of patches which
>>> introduces 'compatible', so IMHO, it should be done properly. Michal,
>>> any ideas/opinions?
>> I see 2 ways to keep all "compatible"-logic in spl_fit_upload_fpga():
>> - by extending each vendor-specific *_load() function with another
>> parameter, which
>>    will contain a unique type for any supported "compatible" value.
>> - making up a better-fit structure like "fpga_bitstream", more
>> specific for fpga bs only
>>    and smaller than spl_image_info, packing there a bs ptr, bs size, bs
>> type, and bs
>>    "compatible" members.
>> Both ways require changing the interface for all fpga load()-family functions.
> 
> Which option from Oleksandr's proposal do you think will be easier to push 
> upstream?

I think you should convert compatible string from fit image to flags and don't 
parse any compatible string in the driver but only work with this flag. Maybe 
make sense also to record in the driver structure all supported flags by driver 
and let core to check if driver provides that feature or not. With conversion to 
DM this can allow to enable multiple drivers that one handle only simple 
unencrypted not authenticated bitstreams and others different types or different 
transport mechanisms.

Thanks,
Michal