[BUG] ssb: Possible sleep-in-atomic bugs in ssb_pcmcia_read8

[BUG] ssb: Possible sleep-in-atomic bugs in ssb_pcmcia_read8

[Jia-Ju Bai]

According to pcmcia.c, the driver may sleep under a spinlock.
The function call paths are:
ssb_pcmcia_read8 (acquire the spinlock)
   select_core_and_segment
     ssb_pcmcia_switch_segment
       ssb_pcmcia_cfg_write
         pcmcia_write_config_byte
           pcmcia_access_config (drivers/pcmcia/pcmcia_resource.c)
             mutex_lock --> may sleep

ssb_pcmcia_read8 (acquire the spinlock)
   select_core_and_segment
     ssb_pcmcia_switch_segment
       sssb_pcmcia_cfg_read
         pcmcia_read_config_byte
           pcmcia_access_config (drivers/pcmcia/pcmcia_resource.c)
             mutex_lock --> may sleep

A possible fix is to use spinlock instead of mutex lock in 
pcmcia_access_config in drivers/pcmcia/pcmcia_resource.c.

These bugs are found by my static analysis tool and my code review.


Thanks,
Jia-Ju Bai

Re: [BUG] ssb: Possible sleep-in-atomic bugs in ssb_pcmcia_read8

[Michael Büsch]

[-- Attachment #1: Type: text/plain, Size: 1287 bytes --]

On Mon, 9 Oct 2017 09:29:17 +0800
Jia-Ju Bai <baijiaju1990@163.com> wrote:

> According to pcmcia.c, the driver may sleep under a spinlock.
> The function call paths are:
> ssb_pcmcia_read8 (acquire the spinlock)
>    select_core_and_segment
>      ssb_pcmcia_switch_segment
>        ssb_pcmcia_cfg_write
>          pcmcia_write_config_byte
>            pcmcia_access_config (drivers/pcmcia/pcmcia_resource.c)
>              mutex_lock --> may sleep
> 
> ssb_pcmcia_read8 (acquire the spinlock)
>    select_core_and_segment
>      ssb_pcmcia_switch_segment
>        sssb_pcmcia_cfg_read
>          pcmcia_read_config_byte
>            pcmcia_access_config (drivers/pcmcia/pcmcia_resource.c)
>              mutex_lock --> may sleep
> 
> A possible fix is to use spinlock instead of mutex lock in 
> pcmcia_access_config in drivers/pcmcia/pcmcia_resource.c.
> 
> These bugs are found by my static analysis tool and my code review.



Thanks for scanning and your resulting bug notification.
I currently don't have the hardware at hand to develop and test a
proper fix for this.
That said, I'm not so sure anymore why bar_lock is a spinlock instead
of a mutex. It might be possible to convert this to mutex.

I will try to look into this.

-- 
Michael

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]