Re: [PATCH 1/5] phylib: bus reset function

[Renaud Barbier <renaud.barbier@ge.com>]

On 21/09/12 09:30, Sascha Hauer wrote:
> On Fri, Sep 21, 2012 at 10:06:58AM +0200, Jean-Christophe PLAGNIOL-VILLARD wrote:
>> On 08:34 Fri 21 Sep     , Sascha Hauer wrote:
>>> On Thu, Sep 20, 2012 at 08:23:32PM +0200, Jean-Christophe PLAGNIOL-VILLARD wrote:
>>>> On 17:31 Thu 20 Sep     , Renaud Barbier wrote:
>>>>> Added an entry point for a bus reset function.
>>>>>
>>>>> Signed-off-by: Renaud Barbier <renaud.barbier@ge.com>
>>>>> ---
>>>>>  drivers/net/phy/phy.c |    3 +++
>>>>>  include/miidev.h      |    1 +
>>>>>  2 files changed, 4 insertions(+), 0 deletions(-)
>>>>>
>>>>> diff --git a/drivers/net/phy/phy.c b/drivers/net/phy/phy.c
>>>>> index 5406ea3..d9f5409 100644
>>>>> --- a/drivers/net/phy/phy.c
>>>>> +++ b/drivers/net/phy/phy.c
>>>>> @@ -129,6 +129,9 @@ int phy_device_connect(struct eth_device *edev, struct mii_device *bus, int addr
>>>>>  	unsigned int i;
>>>>>  	int ret = -EINVAL;
>>>>>  
>>>>> +	if (bus->reset)
>>>>> +		bus->reset(bus);
>>>>> +
>>>> I do not like this we should not touch the bus at phy connect
>>>
>>> Then please make a suggestion where it should be then. I also have some
>>> drivers where this function could be used.
>> I have too the need of a reset
>>
>> I get a big issue on reset the mii bus at eth open
>>
>> my only idea was to move the phy detect awqy from the eth_open to the mii bus
>> probe
>>
>> but this will increase the boot time as we do not need to probe the phy if we
>> do not use the network
> 
> Before we start adding quirks for do-the-reset-only-when necessary,
> access-the-phy-pnly-when necessary can we have a statement on which
> timescales we talk about here? A quick test on a i.MX board shows that
> detecting the phy takes 186us when its address is specified and 1.7ms
> when the whole bus is scanned for phys. This is really not worth adding
> quirks for.
> 
> BTW the reset function for the mii *bus* really should be done when the
> bus is registered. Doing it when a new phy is registered will break phys
> already present.
In summary, we agree that we need a bus reset function and that it
should be called at present by the function mii_register.
> 
> Sascha
> 


_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox