]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
power: supply: goldfish: Fix use-after-free in power_supply_changed()
authorWaqar Hameed <waqar.hameed@axis.com>
Sat, 20 Dec 2025 22:36:00 +0000 (23:36 +0100)
committerSebastian Reichel <sebastian.reichel@collabora.com>
Mon, 12 Jan 2026 01:52:30 +0000 (02:52 +0100)
commitb2ce982e2e0c888dc55c888ad0e20ea04daf2e6b
tree4242ce4eee57c71bc92dff0822aab51168112db9
parent642f33e34b969eedec334738fd5df95d2dc42742
power: supply: goldfish: Fix use-after-free in power_supply_changed()

Using the `devm_` variant for requesting IRQ _before_ the `devm_`
variant for allocating/registering the `power_supply` handle, means that
the `power_supply` handle will be deallocated/unregistered _before_ the
interrupt handler (since `devm_` naturally deallocates in reverse
allocation order). This means that during removal, there is a race
condition where an interrupt can fire just _after_ the `power_supply`
handle has been freed, *but* just _before_ the corresponding
unregistration of the IRQ handler has run.

This will lead to the IRQ handler calling `power_supply_changed()` with
a freed `power_supply` handle. Which usually crashes the system or
otherwise silently corrupts the memory...

Note that there is a similar situation which can also happen during
`probe()`; the possibility of an interrupt firing _before_ registering
the `power_supply` handle. This would then lead to the nasty situation
of using the `power_supply` handle *uninitialized* in
`power_supply_changed()`.

Fix this racy use-after-free by making sure the IRQ is requested _after_
the registration of the `power_supply` handle.

Fixes: 84d7b7687489 ("power: Add battery driver for goldfish emulator")
Signed-off-by: Waqar Hameed <waqar.hameed@axis.com>
Link: https://patch.msgid.link/500a606bb6fb6f2bb8d797e19a00cea9dd7b03c1.1766268280.git.waqar.hameed@axis.com
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
drivers/power/supply/goldfish_battery.c