From mboxrd@z Thu Jan  1 00:00:00 1970
From: andre.przywara@arm.com (Andre Przywara)
Date: Sat, 19 Sep 2015 22:45:11 +0100
Subject: [PATCH v13 2/5] uart: pl011: Introduce register accessor
In-Reply-To: <CABymUCPKqevt9tBEqLnxT-fen2PLy48QMkCfO+nm7-16wHQ8Og@mail.gmail.com>
References: <1438328959-16177-1-git-send-email-jun.nie@linaro.org>
 <1438328959-16177-3-git-send-email-jun.nie@linaro.org>
 <55FBECCD.2080906@arm.com>
 <CABymUCPKqevt9tBEqLnxT-fen2PLy48QMkCfO+nm7-16wHQ8Og@mail.gmail.com>
Message-ID: <55FDD767.7070109@arm.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

Hi Jun,

thanks for your reply. Just a quick one below...

...
>>> @@ -203,6 +206,24 @@ struct uart_amba_port {
>>>  #endif
>>>  };
>>>
>>> +static unsigned int pl011_readw(struct uart_amba_port *uap, int index)
>>> +{
>>> +       WARN_ON(index > REG_NR);
>>> +       return readw_relaxed(uap->port.membase + (index << 2));
>>> +}
>>> +
>>> +static void pl011_writew(struct uart_amba_port *uap, int val, int index)
>>> +{
>>> +       WARN_ON(index > REG_NR);
>>> +       writew_relaxed(val, uap->port.membase + (index << 2));
>>> +}
>>
>> I wonder if you could rename those to pl011_{read,write}, respectively
>> (loosing the "w" suffix).
>> The SBSA UART spec reads as the registers are actually accessible via
>> 32-bit accesses and rumour has it that there are implementations which
>> rely on that and don't work with ldrh/strh.
>> I am still waiting for reports about actual hardware to fail, but we
>> might be forced to change the access width to 32-bit for the SBSA subset
>> in the future. So having wrapper functions would make that change much
>> easier, but having them without a suffix from the beginning would even
>> be better, as I wouldn't be bothered to rename them later on.
> 
> OK, will change to 32-bit access in future version.

Sorry, there seems to be a misunderstanding here. 16-bit accesses are
totally fine for every _PL011_ part, it is just the SBSA-UART (which
uses this very same driver) which _may_ require 32-bit accesses.

So I was asking just to name the functions pl011_write and pl011_read,
so that their "w" suffix does not delude people into them being forever
16-bit wide. If needed (as said I am still waiting for failure reports),
we can then change the access width inside these functions to the
appropriate size later without requiring a name change again.

So please keep the {write,read}w_relaxed calls in there, just change the
function name.

Thanks!
Andre.