Re: [PATCH net] net: broadcom: Imply BROADCOM_PHY for BCMGENET

From: Florian Fainelli <f.fainelli@gmail.com>
To: Marek Szyprowski <m.szyprowski@samsung.com>, netdev@vger.kernel.org
Cc: nsaenzjulienne@suse.de, wahrenst@gmx.net,
	"David S. Miller" <davem@davemloft.net>,
	Thomas Gleixner <tglx@linutronix.de>,
	Tal Gilboa <talgi@mellanox.com>,
	Michael Chan <michael.chan@broadcom.com>,
	Geert Uytterhoeven <geert+renesas@glider.be>,
	Andy Gospodarek <gospo@broadcom.com>,
	Randy Dunlap <rdunlap@infradead.org>,
	open list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH net] net: broadcom: Imply BROADCOM_PHY for BCMGENET
Date: Mon, 11 May 2020 11:19:41 -0700	[thread overview]
Message-ID: <51710a87-5a99-35ee-5bea-92a5801cec09@gmail.com> (raw)
In-Reply-To: <350c88a9-eeaf-7859-d425-0ee4ca355ed3@samsung.com>

On 5/11/2020 12:21 AM, Marek Szyprowski wrote:
> Hi Florian,
> 
> On 09.05.2020 00:32, Florian Fainelli wrote:
>> The GENET controller on the Raspberry Pi 4 (2711) is typically
>> interfaced with an external Broadcom PHY via a RGMII electrical
>> interface. To make sure that delays are properly configured at the PHY
>> side, ensure that we get a chance to have the dedicated Broadcom PHY
>> driver (CONFIG_BROADCOM_PHY) enabled for this to happen.
>>
>> Fixes: 402482a6a78e ("net: bcmgenet: Clear ID_MODE_DIS in EXT_RGMII_OOB_CTRL when not needed")
>> Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
>> Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
>> ---
>> David,
>>
>> I would like Marek to indicate whether he is okay or not with this
>> change. Thanks!
> 
> It is better. It fixes the default values for ARM 32bit 
> bcm2835_defconfig and ARM 64bit defconfig, so you can add:
> 
> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
> 
> There is still an issue there. In case of ARM 64bit, when Genet driver 
> is configured as a module, BROADCOM_PHY is also set to module. When I 
> changed Genet to be built-in, BROADCOM_PHY stayed selected as module. 

OK.

> This case doesn't work, as Genet driver is loaded much earlier than the 
> rootfs/initrd/etc is available, thus broadcom phy driver is not loaded 
> at all. It looks that some kind of deferred probe is missing there.

In the absence of a specific PHY driver the Generic PHY driver gets used
instead. This is a valid situation as I described in my other email
because the boot loader/firmware could have left the PHY properly
configured with the expected RGMII delays and configuration such that
Linux does not need to be aware of anything. I suppose we could change
the GENET driver when running on the 2711 platform to reject the PHY
driver being "Generic PHY" on the premise that a specialized driver
should be used instead, but that seems a bit too restrictive IMHO.

Do you prefer a "select BROADCOM_PHY if ARCH_BCM2835" then?
-- 
Florian