[PATCH] input: fix data race __ps2_command

[Dmitry Vyukov]

The data race happens on ps2dev->cmdcnt and ps2dev->cmdbuf contents.
__ps2_command reads that data concurrently with the interrupt handler.
As the result, for example, if a response arrives just after the
timeout, __ps2_command can copy out garbage from ps2dev->cmdbuf
but then see that ps2dev->cmdcnt is 0 and return success.

Stop the interrupt handler with serio_pause_rx() before
reading the results.

The data race was found with KernelThreadSanitizer (KTSAN).

Signed-off-by: Dmitry Vyukov <dvy...@google.com>
---
drivers/input/serio/libps2.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/input/serio/libps2.c b/drivers/input/serio/libps2.c
index 7551699..8f93336 100644
--- a/drivers/input/serio/libps2.c
+++ b/drivers/input/serio/libps2.c
@@ -234,17 +234,19 @@ int __ps2_command(struct ps2dev *ps2dev, unsigned char *param, int command)
!(ps2dev->flags & PS2_FLAG_CMD), timeout);
}

+ serio_pause_rx(ps2dev->serio);
if (param)
for (i = 0; i < receive; i++)
param[i] = ps2dev->cmdbuf[(receive - 1) - i];

if (ps2dev->cmdcnt && (command != PS2_CMD_RESET_BAT || ps2dev->cmdcnt != 1))
- goto out;
-
+ goto out_paused;
rc = 0;
+ goto out_paused;

out:
serio_pause_rx(ps2dev->serio);
+out_paused:
ps2dev->flags = 0;
serio_continue_rx(ps2dev->serio);

--
2.5.0.457.gab17608

[Pali Rohár]

Hi! Was this patch queued? Or is there any problem with it? Because I
did not see any response for more weeks on any mailing list.

--
Pali Rohár
pali....@gmail.com