Revert "ravb: Fixed to be able to unload modules"
authorGeert Uytterhoeven <geert+renesas@glider.be>
Tue, 22 Sep 2020 07:29:31 +0000 (09:29 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 24 Sep 2020 00:38:36 +0000 (17:38 -0700)
commit77972b55fb9d35d4a6b0abca99abffaa4ec6a85b
tree0128599d2b38f71f08acc75cf0712468eaaf1b44
parentef59b1953c26130ef23e6986b2ceea1efaa472e9
Revert "ravb: Fixed to be able to unload modules"

This reverts commit 1838d6c62f57836639bd3d83e7855e0ee4f6defc.

This commit moved the ravb_mdio_init() call (and thus the
of_mdiobus_register() call) from the ravb_probe() to the ravb_open()
call.  This causes a regression during system resume (s2idle/s2ram), as
new PHY devices cannot be bound while suspended.

During boot, the Micrel PHY is detected like this:

    Micrel KSZ9031 Gigabit PHY e6800000.ethernet-ffffffff:00: attached PHY driver [Micrel KSZ9031 Gigabit PHY] (mii_bus:phy_addr=e6800000.ethernet-ffffffff:00, irq=228)
    ravb e6800000.ethernet eth0: Link is Up - 1Gbps/Full - flow control off

During system suspend, (A) defer_all_probes is set to true, and (B)
usermodehelper_disabled is set to UMH_DISABLED, to avoid drivers being
probed while suspended.

  A. If CONFIG_MODULES=n, phy_device_register() calling device_add()
     merely adds the device, but does not probe it yet, as
     really_probe() returns early due to defer_all_probes being set:

       dpm_resume+0x128/0x4f8
 device_resume+0xcc/0x1b0
   dpm_run_callback+0x74/0x340
     ravb_resume+0x190/0x1b8
       ravb_open+0x84/0x770
 of_mdiobus_register+0x1e0/0x468
   of_mdiobus_register_phy+0x1b8/0x250
     of_mdiobus_phy_device_register+0x178/0x1e8
       phy_device_register+0x114/0x1b8
 device_add+0x3d4/0x798
   bus_probe_device+0x98/0xa0
     device_initial_probe+0x10/0x18
       __device_attach+0xe4/0x140
 bus_for_each_drv+0x64/0xc8
   __device_attach_driver+0xb8/0xe0
     driver_probe_device.part.11+0xc4/0xd8
       really_probe+0x32c/0x3b8

     Later, phy_attach_direct() notices no PHY driver has been bound,
     and falls back to the Generic PHY, leading to degraded operation:

       Generic PHY e6800000.ethernet-ffffffff:00: attached PHY driver [Generic PHY] (mii_bus:phy_addr=e6800000.ethernet-ffffffff:00, irq=POLL)
       ravb e6800000.ethernet eth0: Link is Up - 1Gbps/Full - flow control off

  B. If CONFIG_MODULES=y, request_module() returns early with -EBUSY due
     to UMH_DISABLED, and MDIO initialization fails completely:

       mdio_bus e6800000.ethernet-ffffffff:00: error -16 loading PHY driver module for ID 0x00221622
       ravb e6800000.ethernet eth0: failed to initialize MDIO
       PM: dpm_run_callback(): ravb_resume+0x0/0x1b8 returns -16
       PM: Device e6800000.ethernet failed to resume: error -16

     Ignoring -EBUSY in phy_request_driver_module(), like was done for
     -ENOENT in commit 21e194425abd65b5 ("net: phy: fix issue with loading
     PHY driver w/o initramfs"), would makes it fall back to the Generic
     PHY, like in the CONFIG_MODULES=n case.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: stable@vger.kernel.org
Reviewed-by: Sergei Shtylyov <sergei.shtylyov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/renesas/ravb_main.c