netdev - Xilinx axienet + DP83620 in fiber mode won't set netif_carrier

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-next>] [day] [month] [year] [list]

Message-ID: <20180516091630.GA17469@salem.gmr.ssr.upm.es>
Date:   Wed, 16 May 2018 11:16:31 +0200
From:   "Alvaro G. M." <alvaro.gamez@...ent.com>
To:     netdev@...r.kernel.org
Subject: Xilinx axienet + DP83620 in fiber mode won't set netif_carrier_on

Hi,

I have a custom board with a Xilinx FPGA running Microblaze and fitting a
Xilinx Axi Ethernet IP core.  This core communicates through MII mode with a
DP83620 PHY from Texas that supports both cabled and fiber interfaces, of
which I'm using the latter.

Under these circumstances, I've noticed that the interface is pretty much
dead except for receiving broadcast packages, so I tried to dig on the
driver to find the cause. Please, beware that I'm not very familiar with the
netdev subsystem, so I may be mistaken on lots of things.

It seems that of_phy_connect ends up calling netif_carrier_off:

phy_device.c:1036
	/* Initial carrier state is off as the phy is about to be
	 * (re)initialized.
	 */
	netif_carrier_off(phydev->attached_dev);

	/* Do initial configuration here, now that
	 * we have certain key parameters
	 * (dev_flags and interface)
	 */
	err = phy_init_hw(phydev);
	if (err)
		goto error;

	phy_resume(phydev);

However, neither xilinx_axienet_main.c nor dp83848.c ever runs
netif_carrier_on. As a simple test, I tried this patch, and that was enough
to make the interface work.

diff --git a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
index e74e1e897864..d8bbe4c51b8a 100644
--- a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
+++ b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
@@ -957,6 +957,8 @@ static int axienet_open(struct net_device *ndev)
 	if (ret)
 		goto err_rx_irq;

+	netif_carrier_on(ndev);
+
 	return 0;

 err_rx_irq:

I understand, however, that this is just a proof of concept that shows the
underlying issue. I'd like to contribute to making this a proper patch, or
maybe anyone who is familiar with the netdev subsystem knows at first sight
what is the solution for this.

My understanding is that this code works fine with other PHY chips, as
pretty much the same code has been in the kernel for a long time, but that
probably before ee06b1728b95643668e40fc58ae118aeb7c1753e (which I
instigated) this Xilinx core and driver had never been tested with any
interface other than GMII and RGMII, which were back then written
explicitly, with an unknown PHY chip.

I should also note that axienet_adjust_link is never called in this
configuration, which is the place where I think the call to netif_carrier_on
should be (based on what I've read on other ethernet drivers), but it
seems that the dp83620 doesn't notify of any autonegotiation (at least while
on fiber mode).

I'm open to reading and testing whatever is needed, and please, feel free to
correct me if I've said anything incorrect, which most probably I've done.

Best regards

-- 
Alvaro G. M.