]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
phy: apple: atc: Fix typec switch/mux leak on unbind
authorDavid Carlier <devnexen@gmail.com>
Fri, 8 May 2026 20:19:58 +0000 (21:19 +0100)
committerVinod Koul <vkoul@kernel.org>
Thu, 14 May 2026 15:47:54 +0000 (21:17 +0530)
commit1854082fe0ddb81bc93d1f8e8a00554217fd09d1
tree951db65ff563943f87181e970eee89e972085111
parenta4058c09dd6e28ec33316fd6eb45ddae4cab1f31
phy: apple: atc: Fix typec switch/mux leak on unbind

atcphy_probe_switch() and atcphy_probe_mux() discard the pointers
returned by typec_switch_register() and typec_mux_register(). The
platform driver has no .remove callback, so when the driver unbinds
(e.g. via sysfs unbind) neither typec_switch_unregister() nor
typec_mux_unregister() is called. The framework reference taken in
typec_switch_register() (device_initialize() + device_add() in
drivers/usb/typec/mux.c) is therefore never dropped and the
typec_switch_dev / typec_mux_dev objects stay live forever, with
their sysfs entries under the typec_mux class also left behind. A
subsequent rebind cannot recreate them with the same fwnode-derived
name.

Save the registered handles and unregister them through
devm_add_action_or_reset() so framework registration is torn down
in step with the driver's other devm-managed state. While here,
drop struct apple_atcphy::sw and ::mux: they were declared with the
consumer-side types (typec_switch *, typec_mux *) instead of the
provider-side types and were never assigned.

Scope of the fix
================
This patch fixes the registration leak only. It does not close the
use-after-free window that arises when a consumer that obtained a
reference via fwnode_typec_switch_get() / fwnode_typec_mux_get()
outlives the provider unbind: such consumers keep the underlying
typec_switch_dev / typec_mux_dev alive past device_unregister(),
and a later typec_switch_set() / typec_mux_set() still invokes the
registered atcphy_sw_set() / atcphy_mux_set(), which dereferences
the freed apple_atcphy through typec_{switch,mux}_get_drvdata().

On Apple Silicon the relevant consumers are the typec port and the
cd321x controller registered by drivers/usb/typec/tipd/core.c.
Cable plug / orientation events and alt-mode transitions trigger
the .set callbacks via:

  tps6598x_interrupt()                 drivers/usb/typec/tipd/core.c
    tps6598x_handle_plug_event()
      tps6598x_connect()/_disconnect()
        typec_set_orientation()        drivers/usb/typec/class.c
          typec_switch_set(port->sw)   drivers/usb/typec/mux.c
            atcphy_sw_set()            drivers/phy/apple/atc.c

  cd321x_update_work()                 drivers/usb/typec/tipd/core.c
    cd321x_typec_update_mode()
      typec_mux_set(cd321x->mux)       drivers/usb/typec/mux.c
        atcphy_mux_set()               drivers/phy/apple/atc.c

Closing that window requires framework support for invalidating
consumer-held references on provider unbind. The same
consumer-survives-provider pattern has been discussed for the PHY
framework [1] and is out of scope here.

[1] https://lore.kernel.org/linux-phy/aZejMSJ9qqRWb2pX@google.com/

Fixes: 8e98ca1e74db ("phy: apple: Add Apple Type-C PHY")
Signed-off-by: David Carlier <devnexen@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Tested-by: Joshua Peisach <jpeisach@ubuntu.com>
Link: https://lkml.kernel.org/r/6ec1ed08328340db42655287afd5fa4067316b11.camel@perches.com
Link: https://patch.msgid.link/20260508201958.30060-1-devnexen@gmail.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
drivers/phy/apple/atc.c