mirror of
https://github.com/torvalds/linux
synced 2024-10-19 01:39:28 +00:00
a93d2afd3f
When devm runs function in the "remove" path for a device it runs them
in the reverse order. That means that if you have parts of your driver
that aren't using devm or are using "roll your own" devm w/
devm_add_action_or_reset() you need to keep that in mind.
The mt8186 audio driver didn't quite get this right. Specifically, in
mt8186_init_clock() it called mt8186_audsys_clk_register() and then
went on to call a bunch of other devm function. The caller of
mt8186_init_clock() used devm_add_action_or_reset() to call
mt8186_deinit_clock() but, because of the intervening devm functions,
the order was wrong.
Specifically at probe time, the order was:
1. mt8186_audsys_clk_register()
2. afe_priv->clk = devm_kcalloc(...)
3. afe_priv->clk[i] = devm_clk_get(...)
At remove time, the order (which should have been 3, 2, 1) was:
1. mt8186_audsys_clk_unregister()
3. Free all of afe_priv->clk[i]
2. Free afe_priv->clk
The above seemed to be causing a use-after-free. Luckily, it's easy to
fix this by simply using devm more correctly. Let's move the
devm_add_action_or_reset() to the right place. In addition to fixing
the use-after-free, code inspection shows that this fixes a leak
(missing call to mt8186_audsys_clk_unregister()) that would have
happened if any of the syscon_regmap_lookup_by_phandle() calls in
mt8186_init_clock() had failed.
Fixes:
|
||
|---|---|---|
| .. | ||
| ac97 | ||
| aoa | ||
| arm | ||
| atmel | ||
| core | ||
| drivers | ||
| firewire | ||
| hda | ||
| i2c | ||
| isa | ||
| mips | ||
| oss | ||
| parisc | ||
| pci | ||
| pcmcia | ||
| ppc | ||
| sh | ||
| soc | ||
| sparc | ||
| spi | ||
| synth | ||
| usb | ||
| virtio | ||
| x86 | ||
| xen | ||
| ac97_bus.c | ||
| Kconfig | ||
| last.c | ||
| Makefile | ||
| sound_core.c | ||