Re: [RFC PATCH v1 25/31] ARC: [plat-arcfpga] Hooking up platform to ARC UART

[Vineet Gupta <Vineet.Gupta1@synopsys.com>]

On Monday 07 January 2013 07:16 PM, Arnd Bergmann wrote:
> On Monday 07 January 2013, Vineet Gupta wrote:
>> On Wednesday 07 November 2012 07:46 PM, Arnd Bergmann wrote:
>>> On Wednesday 07 November 2012, Vineet Gupta wrote:
>> (1) Although I don't need the container "fpga" I'm forced to - because
>> of_platform_populate( ) -> of_match_node( ) expects the @match arg to be NOT NULL.
>> So we pass of_default_bus_match_table and have the compat string "simple-bus" in
>> the container. Per [1] it seemed it was possible to add the serial device directly
>> w/o the container.
> You could in theory make the serial port device itself be compatible to 
> somthing that is being probed by of_platform_populate, but putting it
> under a bus is the preferred way. Usually each system has at least one
> bus that devices are connected to and I would recommend to represent all
> buses in the device tree like they are in hardware.

OK.

>> (2) I need the following OF_DEV_AUXDATA to be able to "name" the device correctly
>> so that the registered driver [4] can bind with device. How do I match the driver
>> and devicetree node w/o this glue - it seems compatible="<manuf>,<model>" is not
>> enough. This also requires the uart base address to be specified (otherwise
>> of_dev_lookup() fails to identify the auxdata) which IMHO defeats the purpose of
>> devicetree in first place.
>> b
>> static struct of_dev_auxdata arcuart_auxdata_lookup[] __initdata = {
>> 	OF_DEV_AUXDATA("snps,arc-uart", UART0_BASE, "arc-uart", arc_uart_info),
>> 	{}
>> };
> It should be enough to fill the drv->of_match_table member of the
> platform_driver with the match table.

Not sure if I understand. My concern is the #define UART0_BASE (=0xc0fc1000) which
needs to be defined in code despite that value being present in the device tree.
And this is needed so that framework could match the device against the driver. I
would have thought that some device property (in device tree) could enable the
matching with Linux name (actually DRIVER_NAME defined in the uart driver).  Am I
missing something ?

>> (3) After above, driver's probe routine is getting called with platform_device->id
>> = -1 and it seems of_device_add() is doing that purposely. How do I handle that.
> What do you need the id for?

In case of multiple instances of UART, driver uses this value to index into struct
arc_uart_port [ ]

struct arc_uart_port {
    struct uart_port port;
    unsigned long baud;
    int is_emulated;
};

static struct arc_uart_port arc_uart_ports[CONFIG_SERIAL_ARC_NR_PORTS];

arc_serial_probe(struct platform_device *pdev)
    uart = &arc_uart_ports[pdev->id];
    arc_uart_init_one(pdev, ..)
          if (pdev->id < 0 ...)
                  return -ENOENT;
 
Although the current driver is flawed in that it checks for -ve value after
already indexing into the array :-(
I'll send a patch to serial list to fix that.

>> (4) Is above standalone "interrupts" string OK, or do I have to explicitly
>> instantiate the in-core intc as well. Since it is integral part of cpu, I really
>> don't need any support code to explicitly instantiate it. Also it is not accessed
>> via mem map - but special ARC instructions in aux address space of cpu.
> Interrupts are a little tricky. You need to create a bindind for your interrupt
> controller first and make the irqchip driver use irq domains in order for the
> irq description in the device tree to be mapped into a linux-internal irq number.
>
> In the simple case where you only have one irqchip in the system, you can use
> a "legacy" irq domain that simply translates the numbers 1:1. In some cases,
> it does make sense though (even with the legacy domain) to include additional
> flags in the device tree irq descriptor, e.g. the irq polarity and
> edge/level/message indication. Please read up on these and ask again if you have
> more questions.

Will do.

Thx,
-Vineet


>
> 	Arnd