linux-kernel - Re: [PATCH net v2] net: bcmgenet: Clear ID_MODE_DIS in EXT_RGMII_OOB

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <5f3d64c8-5b02-d697-c214-fb14bcff99ac@gmail.com>
Date:   Fri, 8 May 2020 10:42:43 -0700
From:   Florian Fainelli <f.fainelli@...il.com>
To:     Marek Szyprowski <m.szyprowski@...sung.com>,
        Nicolas Saenz Julienne <nsaenzjulienne@...e.de>,
        Doug Berger <opendmb@...il.com>,
        "David S. Miller" <davem@...emloft.net>,
        Stefan Wahren <wahrenst@....net>
Cc:     bcm-kernel-feedback-list@...adcom.com, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH net v2] net: bcmgenet: Clear ID_MODE_DIS in
 EXT_RGMII_OOB_CTRL when not needed



On 5/7/2020 11:38 PM, Marek Szyprowski wrote:
> Hi Florian,
> 
> On 07.05.2020 17:54, Florian Fainelli wrote:
>> On 5/7/2020 3:03 AM, Marek Szyprowski wrote:
>>> On 07.05.2020 11:46, Marek Szyprowski wrote:
>>>> On 25.02.2020 14:11, Nicolas Saenz Julienne wrote:
>>>>> Outdated Raspberry Pi 4 firmware might configure the external PHY as
>>>>> rgmii although the kernel currently sets it as rgmii-rxid. This makes
>>>>> connections unreliable as ID_MODE_DIS is left enabled. To avoid this,
>>>>> explicitly clear that bit whenever we don't need it.
>>>>>
>>>>> Fixes: da38802211cc ("net: bcmgenet: Add RGMII_RXID support")
>>>>> Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@...e.de>
>>>> I've finally bisected the network issue I have on my RPi4 used for
>>>> testing mainline builds. The bisect pointed to this patch. Once it got
>>>> applied in v5.7-rc1, the networking is broken on my RPi4 in ARM32bit
>>>> mode and kernel compiled from bcm2835_defconfig. I'm using u-boot to
>>>> tftp zImage/dtb/initrd there. After reverting this patch network is
>>>> working fine again. The strange thing is that networking works fine if
>>>> kernel is compiled from multi_v7_defconfig but I don't see any obvious
>>>> difference there.
>>>>
>>>> I'm not sure if u-boot is responsible for this break, but kernel
>>>> definitely should be able to properly reset the hardware to the valid
>>>> state.
>>>>
>>>> ...
>>> Okay, I've played a bit more with this and found that enabling
>>> CONFIG_BROADCOM_PHY fixes this network issue. I wonder if Genet driver
>>> should simply select CONFIG_BROADCOM_PHY the same way as it selects
>>> CONFIG_BCM7XXX_PHY.
>> Historically GENET has been deployed with an internal PHY and this is
>> still 90% of the GENET users out there on classic Broadcom STB
>> platforms, not counting the 2711. For external PHYs, there is a variety
>> of options here, so selecting CONFIG_BROADCOM_PHY would be just one of
>> the possibilities, I would rather fix this with the bcm2835_defconfig
>> and multi_v7_defconfig update. Would that work for you?
> 
> Frankly I was surprised that the Genet driver successfully probed and 
> registered eth0 even when no proper PHY driver was available in the 
> system. It even reported the link status change, but then didn't 
> transfer any packets. I expected at least a runtime check and error or 
> warning if proper PHY is not available.


> If this is really not possible, I would still advise to select proper potential PHY drivers, so users 
> won't be confused.

It is possible to issue a warning if we find ourselves running on a
BCM2711 SoC and we end-up using the Generic PHY driver, much like what
r8169 does (for similar reasons):

https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=f32593773549
https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=0c2006b29e5f62784c70209e71da7876267e0e2d

> 
> The Genet driver already selects CONFIG_BCM7XXX_PHY. How common is it? 

GENET is a roughly 15 years old Ethernet controller that has evolved and
is still being integrated into new chips, so we have probably hundreds
of millions of devices out there.

> Would it really hurt do the same for CONFIG_BROADCOM_PHY? I expect that 
> 2711 will be quite popular SoC with it soon.

My problem with a select BROADCOM_PHY is that it will make it impossible
for me to deselect the Broadcom PHY driver. We have probably about a
hundred or so reference boards with a variety of external PHYs some
Broadcom and we have managed to get them all working out of the box with
the Generic PHY driver. If I cannot deselect the Broadcom PHY driver
there will be RGMII regressions (much like the one you reported) which I
really have no interest in solving when it can be avoided.

Does the following work for you:

diff --git a/drivers/net/ethernet/broadcom/Kconfig
b/drivers/net/ethernet/broadcom/Kconfig
index 53055ce5dfd6..8a70b9152f7c 100644
--- a/drivers/net/ethernet/broadcom/Kconfig
+++ b/drivers/net/ethernet/broadcom/Kconfig
@@ -69,6 +69,7 @@ config BCMGENET
        select BCM7XXX_PHY
        select MDIO_BCM_UNIMAC
        select DIMLIB
+       imply BROADCOM_PHY if ARCH_BCM2835
        help
          This driver supports the built-in Ethernet MACs found in the
          Broadcom BCM7xxx Set Top Box family chipset.

-- 
Florian