coccinelle: api/stream_open: treat all wait_.*() calls as blocking
authorKirill Smelkov <kirr@nexedi.com>
Sun, 23 Jun 2019 07:28:53 +0000 (07:28 +0000)
committerMasahiro Yamada <yamada.masahiro@socionext.com>
Sun, 7 Jul 2019 17:25:59 +0000 (02:25 +0900)
commit0c4ab18fc33bba70299f51fac83e32397cf9a04b
tree5be658bb4f5ec66d531fc68eb380c4601faa0abc
parentf58c17c202e7aec9623754a7ec0b0669e043b838
coccinelle: api/stream_open: treat all wait_.*() calls as blocking

Previously steam_open.cocci was treating only wait_event_.* - e.g.
wait_event_interruptible - as a blocking operation. However e.g.
wait_for_completion_interruptible is also blocking, and so from this
point of view it would be more logical to treat all wait_.* as a
blocking point.

The logic of this change actually came up for real when
drivers/pci/switch/switchtec.c changed from using
wait_event_interruptible to wait_for_completion_interruptible:

https://lore.kernel.org/linux-pci/20190413170056.GA11293@deco.navytux.spb.ru/
https://lore.kernel.org/linux-pci/20190415145456.GA15280@deco.navytux.spb.ru/
https://lore.kernel.org/linux-pci/20190415154102.GB17661@deco.navytux.spb.ru/

For a driver that uses nonseekable_open with read/write having stream
semantic and read also calling e.g. wait_for_completion_interruptible,
running stream_open.cocci before this patch would produce:

WARNING: <driver>_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open.

while after this patch it will report:

ERROR: <driver>_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix.

Signed-off-by: Kirill Smelkov <kirr@nexedi.com>
Acked-by: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
scripts/coccinelle/api/stream_open.cocci