Re: [PATCH V2 2/4] ARM64 LPC: LPC driver implementation on Hip06

From: zhichang
Date: Tue Sep 13 2016 - 02:30:54 EST

On 2016å09æ08æ 18:00, Arnd Bergmann wrote:
> On Thursday, September 8, 2016 4:06:01 PM CEST zhichang.yuan wrote:
>>>> +struct lpc_io_ops {
>>>> + unsigned int periosz;
>>>> + int (*lpc_iord)(struct hisilpc_dev *pdev, struct lpc_cycle_para *para,
>>>> + unsigned long ptaddr, unsigned char *buf,
>>>> + unsigned long dlen);
>>>> + int (*lpc_iowr)(struct hisilpc_dev *pdev, struct lpc_cycle_para *para,
>>>> + unsigned long ptaddr,
>>>> + const unsigned char *buf,
>>>> + unsigned long dlen);
>>>> +};
>>> The operations are not needed unless we also put the earlycon support
>>> in, so maybe leave them out from the first patch and only add them
>>> later (or drop the earlycon support if possible).
>> Do you want to remove the struct lpc_io_ops member from struct hisilpc_dev??
>> I think we can not do that.
>> These two functions are essential rd/wr operation for Hip06 LPC. They will be fallen into
>> when the upper layer drivers call their own IO in/out functions, such as serial_in/serial_out
>> for 8250 serial.
>> I can define lpc_iord/lpc_iowr directly in struct hisilpc_dev and cancel the definition of
>> struct lpc_io_ops. In my original idea, several LPC cycle types will be supported. Each cycle
>> type has its specific ops. Now, only one cycle type is needed, the struct lpc_io_ops is not
>> meaningful.
> My point was that the indirect function calls are not needed at
> until patch four, so you can call the functions directly here,
> and make the logic for indirect function calls part of that
> later patch.
O. I think I got your meaning now.
In patch V2, the lpc_io_ops is not needed even if earlycon is supported. The early_in/early_out operation is
defined in hisi_lpc.c too, we can directly call the hisilpc_target_in/hisilpc_target_out to finish the LPC IO
At this moment, all the IO functions specific to the child devices of hip06 LPC have their own indirect call
method. This lpc_io_ops will be removed in V3.

> What are the other LPC cycle types that could be supported?
O. memory and firmware operations are supported too. But at this moment, we only use IO cycle.


> Arnd