by Matthew Gerlach

[permalink] [raw]

Subject: Re: [PATCH v8 3/4] fpga: dfl: add basic support for DFHv1

On Fri, 30 Dec 2022, Xu Yilun wrote:

> On 2022-12-28 at 10:16:23 -0800, [email protected] wrote:
>> From: Matthew Gerlach <[email protected]>
>>
>> Version 1 of the Device Feature Header (DFH) definition adds
>> functionality to the DFL bus.
>>
>> A DFHv1 header may have one or more parameter blocks that
>> further describes the HW to SW. Add support to the DFL bus
>> to parse the MSI-X parameter.
>>
>> The location of a feature's register set is explicitly
>> described in DFHv1 and can be relative to the base of the DFHv1
>> or an absolute address. Parse the location and pass the information
>> to DFL driver.
>>
>> Signed-off-by: Matthew Gerlach <[email protected]>
>> Reviewed-by: Ilpo J?rvinen <[email protected]>
>> ---
>> v8: use struct_size() from overflow.h
>> add dfh_get_u64_param_vals()
>
> Could you help check my comments?
> https://lore.kernel.org/linux-fpga/alpine.DEB.2.22.394.2212211421210.570436@rhweight-WRK1/T/#md86e3836130ebacd3c088f5c512ba741aac8a4d1

Sorry I missed your earlier comments. I have since responded to them
specificially.

>
> [...]
>
>>
>> +static u64 *find_param(u64 *params, resource_size_t max, int param_id)
>> +{
>> + u64 *end = params + max / sizeof(u64);
>> + u64 v, next;
>> +
>> + while (params < end) {
>> + v = *params;
>> + if (param_id == FIELD_GET(DFHv1_PARAM_HDR_ID, v))
>> + return params;
>> +
>> + next = FIELD_GET(DFHv1_PARAM_HDR_NEXT_OFFSET, v);
>> + params += next;
>> + if (FIELD_GET(DFHv1_PARAM_HDR_NEXT_EOP, v))
>> + break;
>> + }
>> +
>> + return NULL;
>> +}
>> +
>> +/**
>> + * dfh_find_param() - find parameter block for the given parameter id
>> + * @dfl_dev: dfl device
>> + * @param: id of dfl parameter
>> + *
>> + * Return: pointer to start of parameter block, NULL otherwise.
>> + */
>> +u64 *dfh_find_param(struct dfl_device *dfl_dev, int param_id)
>> +{
>> + return find_param(dfl_dev->params, dfl_dev->param_size, param_id);
>> +}
>> +EXPORT_SYMBOL_GPL(dfh_find_param);
>
> Didn't find where to use it?

I understand. Don't export a function unless there is a comsumer for it.

>
>> +
>> +/**
>> + * dfh_get_u64_param_vals() - get array of u64 param values for given parameter id
>
> There is no rule to say one u64 for each property in the parameter block.
> So I don't see the reason for DFL core to provide u64 array for the API,
> And the size of the parameter block is decided by HW, why make the user
> input the value?
>
> As we discussed before, dfl core doesn't try to look into the parameter
> block. So please just provide the const void *data & data_size for drivers.
> This is the most common way to represent a data block.

I will move the parameter parsing helper function to the driver itself.

Thanks for the feedback,
Matthew Gerlach

>
> Thanks,
> Yilun
>
>> + * @dfl_dev: dfl device
>> + * @param: id of dfl parameter
>> + * @pval: location of parameter data destination
>> + * @nvals: number of u64 elements of parameter data
>> + *
>> + * Return: pointer to start of parameter block, PTR_ERR otherwise
>> + */
>> +u64 *dfh_get_u64_param_vals(struct dfl_device *dfl_dev, int param_id, u64 *pval, int nvals)
>> +{
>> + u64 *param = find_param(dfl_dev->params, dfl_dev->param_size, param_id);
>> + u64 next;
>> + int i;
>> +
>> + if (!param)
>> + return ERR_PTR(-ENOENT);
>> +
>> + next = FIELD_GET(DFHv1_PARAM_HDR_NEXT_OFFSET, *param);
>> +
>> + if (nvals >= next)
>> + return ERR_PTR(-ENOMEM);
>> +
>> + for (i = 0; i < nvals; i++)
>> + *pval++ = param[i + 1];
>> +
>> + return param;
>> +}
>> +EXPORT_SYMBOL_GPL(dfh_get_u64_param_vals);
>> +
>

2023-01-02 18:50:09

by Matthew Gerlach

[permalink] [raw]

Subject: Re: [PATCH v8 3/4] fpga: dfl: add basic support for DFHv1

On Mon, 2 Jan 2023, Andy Shevchenko wrote:

> On Mon, Jan 02, 2023 at 07:27:06PM +0200, Andy Shevchenko wrote:
>> On Mon, Jan 02, 2023 at 08:54:48AM -0800, [email protected] wrote:
>>> On Thu, 29 Dec 2022, Andy Shevchenko wrote:
>>>> On Thu, Dec 29, 2022 at 08:18:03AM -0800, Tom Rix wrote:
>>>>> On 12/28/22 10:16 AM, [email protected] wrote:
>
> ...
>
>>>>>> + u64 params[];
>>>>> u64 *params
>>
>>>> This will break the overflow.h macros, no?
>>>> Besides that it will break the code for sure as it's not an equivalent.
>>
>>> I don't understand how this will break the overflow.h macros. The definition
>>> of struct dfl_feature_info and all of its uses are in a single file, dfl.c.
>>
>> Hint: __must_be_array()
>>
>> As I said, the proposed change is not acceptable since it's not an equivalent.
>
> Ah, you meant that there is no use of macros from overflow in the dfl.c?
> IIRC we discussed that some of the code may make use of them, or am I
> mistaken?

There currently is one usage of struct_size() from overflow.h in dfl.c,
and my patch adds another usage of struct_size(). struct dfl_feature_info
ends with a trailing array of u64.

I think the confusion is with struct dfl_feature and/or struct struct
dfl_device. Those structs don't end with a trailing array, and those
structs are not used with macros from overview.h.

Thanks for the feedback,
Matthew Gerlach

>
> --
> With Best Regards,
> Andy Shevchenko
>
>
>