re: staging: dgnc: Fix frame size is larger than 1024B

re: staging: dgnc: Fix frame size is larger than 1024B

[Dan Carpenter]

Hello Konrad Zapalowicz,

The patch ea6e9dea2e72: "staging: dgnc: Fix frame size is larger than
1024B" from Aug 6, 2014, leads to the following static checker
warning:

	drivers/staging/dgnc/dgnc_tty.c:479 dgnc_sniff_nowait_nolock()
	error: scheduling with locks held: 'spin_lock:ch_lock'

drivers/staging/dgnc/dgnc_tty.c
   467  void dgnc_sniff_nowait_nolock(struct channel_t *ch, uchar *text, uchar *buf, int len)
   468  {
   469          struct timeval tv;
   470          int n;
   471          int r;
   472          int nbuf;
   473          int i;
   474          int tmpbuflen;
   475          char *tmpbuf;
   476          char *p;
   477          int too_much_data;
   478  
   479          tmpbuf = kzalloc(TMPBUFLEN, GFP_KERNEL);
                                            ^^^^^^^^^^

This is a sleeping under spinlock bug.

   480          if (!tmpbuf)
   481                  return;
   482          p = tmpbuf;

Normally the way to fix these is to just do a GFP_ATOMIC allocation but
sometimes you can shuffle the lock and the allocation around to avoid
the problem.

regards,
dan carpenter
_______________________________________________
devel mailing list
devel@linuxdriverproject.org
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel

Re: staging: dgnc: Fix frame size is larger than 1024B

[Konrad Zapalowicz]

On 08/25, Dan Carpenter wrote:
> Hello Konrad Zapalowicz,
> 
> The patch ea6e9dea2e72: "staging: dgnc: Fix frame size is larger than
> 1024B" from Aug 6, 2014, leads to the following static checker
> warning:
> 
> 	drivers/staging/dgnc/dgnc_tty.c:479 dgnc_sniff_nowait_nolock()
> 	error: scheduling with locks held: 'spin_lock:ch_lock'
> 
> drivers/staging/dgnc/dgnc_tty.c
>    467  void dgnc_sniff_nowait_nolock(struct channel_t *ch, uchar *text, uchar *buf, int len)
>    468  {
>    469          struct timeval tv;
>    470          int n;
>    471          int r;
>    472          int nbuf;
>    473          int i;
>    474          int tmpbuflen;
>    475          char *tmpbuf;
>    476          char *p;
>    477          int too_much_data;
>    478  
>    479          tmpbuf = kzalloc(TMPBUFLEN, GFP_KERNEL);
>                                             ^^^^^^^^^^
> 
> This is a sleeping under spinlock bug.
> 
>    480          if (!tmpbuf)
>    481                  return;
>    482          p = tmpbuf;
> 
> Normally the way to fix these is to just do a GFP_ATOMIC allocation but
> sometimes you can shuffle the lock and the allocation around to avoid
> the problem.

Thanks for the info, I was not aware about this kind of consequences.
I'll fix it in the upcoming series of patches for the dgnc driver.

By the way, now I  understand why someone has put there that local,
huge table which was fixed by my patch.

cheers,
konrad
 
> regards,
> dan carpenter
_______________________________________________
devel mailing list
devel@linuxdriverproject.org
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel

[PATCH] staging: dgnc: Fix sleeping under spinlock bug

[Konrad Zapalowicz]

From: Konrad Zapalowicz <bergo.torino+kernel@gmail.com>

This commit changes the memory allocation flags to ATOMIC in order to
avoid sleeping in the nowait/nolock code.

Signed-off-by: Konrad Zapalowicz <bergo.torino+kernel@gmail.com>
---
 drivers/staging/dgnc/dgnc_tty.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/staging/dgnc/dgnc_tty.c b/drivers/staging/dgnc/dgnc_tty.c
index afb20a3..a798476 100644
--- a/drivers/staging/dgnc/dgnc_tty.c
+++ b/drivers/staging/dgnc/dgnc_tty.c
@@ -462,7 +462,7 @@ void dgnc_sniff_nowait_nolock(struct channel_t *ch, uchar *text, uchar *buf, int
 	char *p;
 	int too_much_data;
 
-	tmpbuf = kzalloc(TMPBUFLEN, GFP_KERNEL);
+	tmpbuf = kzalloc(TMPBUFLEN, GFP_ATOMIC);
 	if (!tmpbuf)
 		return;
 	p = tmpbuf;
-- 
1.8.1.2