[PATCH 1/2] net: designware: eqos: reset phy

[Ahmad Fatoum]

Hello Sascha,

On 08.06.21 00:22, Sascha Hauer wrote:
> On Mon, Jun 07, 2021 at 05:59:02PM +0200, Ahmad Fatoum wrote:
>> Hello Sascha,
>>
>> On 07.06.21 16:10, Sascha Hauer wrote:
>>> The designware eqos DT binding has support for specifying reset GPIOs.
>>> Add support for them.
>>>
>>> Signed-off-by: Sascha Hauer <s.hauer at pengutronix.de>
>>> ---
>>>  drivers/net/designware_eqos.c | 33 +++++++++++++++++++++++++++++++++
>>>  drivers/of/of_gpio.c          |  7 +++++++
>>>  2 files changed, 40 insertions(+)
>>>
>>> diff --git a/drivers/net/designware_eqos.c b/drivers/net/designware_eqos.c
>>> index d2baaeaf63..0321024169 100644
>>> --- a/drivers/net/designware_eqos.c
>>> +++ b/drivers/net/designware_eqos.c
>>> @@ -8,9 +8,11 @@
>>>  
>>>  #include <common.h>
>>>  #include <init.h>
>>> +#include <gpio.h>
>>>  #include <dma.h>
>>>  #include <net.h>
>>>  #include <of_net.h>
>>> +#include <of_gpio.h>
>>>  #include <linux/iopoll.h>
>>>  #include <linux/time.h>
>>>  #include <linux/sizes.h>
>>> @@ -189,6 +191,33 @@ struct eqos_desc {
>>>  
>>>  #define MII_BUSY		(1 << 0)
>>>  
>>> +static int eqos_phy_reset(struct device_d *dev, struct eqos *eqos)
>>> +{
>>> +	int phy_reset;
>>> +	int ret;
>>> +	u32 delays[3] = { 0, 0, 0 };
>>> +
>>> +	phy_reset = of_get_named_gpio(dev->device_node, "snps,reset-gpio", 0);
>>> +
>>> +        if (!gpio_is_valid(phy_reset))
>>> +		return 0;
>>
>> Whitespace is wrong.
>>
>>> +
>>> +	ret = gpio_request(phy_reset, "phy-reset");
>>> +	if (ret)
>>> +		return ret;
>>
>> Can you use gpiod_get instead? This would reduce the boilerplate here.
> 
> Sure. I didn't realize I don't honour the active high/low flags the way
> I did it.
> 
>>
>>> +
>>> +	of_property_read_u32_array(dev->device_node,
>>> +				   "snps,reset-delays-us",
>>> +				   delays, ARRAY_SIZE(delays));
>>> +
>>
>> Looks strange to read out a delay and not act on it. I'd prefer
>> waiting delays[0] here.
> 
> Yes, it looks strange, but that's because the binding doesn't make much
> sense. Why should I insert a delay before doing anything?

                   .--------.

     POR --------->|R  flip |---- Regulator ----> PHY VDD

                .->|S  flop |

                |  `--------'

                |   

                |   

                |

RESET GPIO -----'`-------------------------------> PHY Reset

(active low)

It's stupid, but it works with Linux and wouldn't with barebox
(if PHY VDD takes too long to stabilize)... ^^'

> I can a delay here, it wouldn't make much difference as all users except
> one specify zero delay here anyway. For the one exception I would bet
> someone has inserted the first delay without a good reason, they are all
> 10000.

That's probably true. I still think mimicking Linux' interpretation
of bindings is a good general rule to follow.

>>
>>> +	gpio_direction_active(phy_reset, 0);
>>
>> This always sets logical zero, because gpio_request above doesn't
>> accept a flag telling it otherwise. You'll need of_get_named_gpio_flags
>> here, at which point, you'll have basically open-coded gpiod_get(), so
>> you could use that.
> 
> Right.
> 
>>
>>> +	udelay(delays[1]);
>>
>> Linux rounds up to 1 msec granularity here. We should do likewise.
> 
> I don't see a good reason for that. The Linux driver used udelay()
> initially, that was changed to msleep as the times were too long for
> busy waiting. I have no clue why the author didn't use usleep_range
> instead.

Same reason: Device trees are tested with Linux. They've a better chance
of just working when we round up wait times the same way.

> 
>>
>>> +	gpio_set_active(phy_reset, 1);
>>
>> Nitpick: true/false.
> 
> Ok.
> 
> Sascha
> 

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |