netdev - Re: [PATCH] net: phy: Ensure the state machine is called when phy is UP

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]

Date:	Tue, 17 May 2016 17:35:30 -0600
From:	David Mosberger <davidm@...uge.net>
To:	netdev@...r.kernel.org, Florian Fainelli <f.fainelli@...il.com>,
	Alexandre Belloni <alexandre.belloni@...e-electrons.com>,
	David Mosberger <davidm@...uge.net>
Subject: Re: [PATCH] net: phy: Ensure the state machine is called when phy is UP

[Resending with mailing list address corrected.  Sorry about that.]

I believe I'm seeing the same issue as in this thread:

  http://www.spinics.net/lists/netdev/msg373670.html

macb Ethernet driver and after upgrading from 4.4.0 to 4.60, the
Ethernet interface would not initialize unless I force autonegation
(e.g., with mii-diag -r).  It looks to me the code may always have
been broken, but since this check for PHY_POLL was introduced to
phy_state_machine():

/* Only re-schedule a PHY state machine change if we are polling the
* PHY, if PHY_IGNORE_INTERRUPT is set, then we will be moving
* between states from phy_mac_interrupt()
*/
if (phydev->irq == PHY_POLL)
        queue_delayed_work(system_power_efficient_wq, &phydev->state_queue,
  PHY_STATE_TIME * HZ);

The bug no longer gets papered over with.  The sequence of events I'm seeing:

[    0.810000] macb f8008000.ethernet: invalid hw address, using random
[    0.820000] libphy: MACB_mii_bus: probed
[    0.900000] kszphy_config_intr: mask 200 MII_KSZPHY_INTCS 0
[    0.900000] kszphy_ack_interrupt: rc 0
[    0.900000] kszphy_config_intr: mask 200 MII_KSZPHY_INTCS 500
[    0.910000] Micrel KSZ8081 or KSZ8091 f8008000.etherne:01: attached
PHY driver [Micrel KSZ8081 or KSZ8091]
(mii_bus:phy_addr=f8008000.etherne:01, irq=117)
[    0.930000] macb f8008000.ethernet eth0: Cadence GEM rev 0x00020203
at 0xf8008000 irq 27 (ca:83:b2:75:81:e2)
[    0.940000] kszphy_config_intr: mask 200 MII_KSZPHY_INTCS 0
[    0.940000] kszphy_ack_interrupt: rc 0
[    0.950000] kszphy_config_intr: mask 200 MII_KSZPHY_INTCS 500
[    0.950000] libphy: phy_state_machine: state 2
 :
[    3.060000] kszphy_config_intr: mask 200 MII_KSZPHY_INTCS 0
[    3.060000] kszphy_ack_interrupt: rc 1
[    3.070000] kszphy_config_intr: mask 200 MII_KSZPHY_INTCS 500
[    3.070000] libphy: phy_state_machine: state 2
 :
[    7.960000] macb: macb_open: calling phy_start()
[    7.970000] libphy: phy_start: old state 2
[    7.970000] libphy: phy_start: new state 4

So if I understand this right, at 3.06 seconds, a LINKUP interrupt
gets reported, but since the phy is in state PHY_READY, the interrupt
gets ignored.

At 7.96 seconds, phy_start() gets called, but even though this enters
state PHY_UP, it does not trigger phy_state_machine() and therefore
doesn't trigger autonegotiation.  I'm not quite sure how this is
intended to be handled properly.  I believe Alexandre Belloni's patch
would do the trick:

 http://www.spinics.net/lists/netdev/msg373324.html

Or has this been fixed in some other way already?

  --david
-- 
Love Linux kernel hacking?  Know Python?  We are hiring!  Contact us
at jobs@...uge.net