After checking all possible call chains to on26_test_port() here,
my tool finds that this function is never called in atomic context,
namely never in an interrupt handler or holding a spinlock.
And on26_test_port() is only called by pi_probe_unit() that calls
wait_event() through pi_claim(),
so it indicates that on26_test_port() can call functions that can sleep.
Thus mdelay can be replaced with msleep to avoid busy wait.
This is found by a static analysis tool named DCNS written by myself.
Signed-off-by: Jia-Ju Bai <[email protected]>
---
drivers/block/paride/on26.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/block/paride/on26.c b/drivers/block/paride/on26.c
index 95ba256..3240d6f 100644
--- a/drivers/block/paride/on26.c
+++ b/drivers/block/paride/on26.c
@@ -163,7 +163,7 @@ static int on26_test_port( PIA *pi) /* hard reset */
on26_write_regr(pi,0,6,0xb0);
y = on26_read_regr(pi,0,7);
if (!((x&0x80)||(y&0x80))) break;
- mdelay(100);
+ msleep(100);
}
if (i == RESET_WAIT)
--
1.7.9.5
On Fri, Jan 26, 2018 at 11:42:25PM +0800, Jia-Ju Bai wrote:
> After checking all possible call chains to on26_test_port() here,
> my tool finds that this function is never called in atomic context,
> namely never in an interrupt handler or holding a spinlock.
> And on26_test_port() is only called by pi_probe_unit() that calls
> wait_event() through pi_claim(),
> so it indicates that on26_test_port() can call functions that can sleep.
> Thus mdelay can be replaced with msleep to avoid busy wait.
Sigh... Here's how I would've written it:
"
on26_test_port() is never called from atomic contexts.
It has no direct callers and it is reachable only via ->test_port.
->test_port has only one user:
drivers/block/paride/paride.c:322: max = pi->proto->test_port(pi);
in pi_probe_unit(). That gets called only from pi_init(), called from
p{d,cd,f,t,g}_detect(), called from module_init stuff, all of the above
without entering atomic contexts along the way.
Despite never getting called from atomic contexts, on26_test_port() contains
mdelay(100), i.e. busy-loops for 0.1s; that's neither nice nor needed, since
msleep() would serve just as well.
Found by [reference to tool]"
On 2018/1/27 1:31, Al Viro wrote:
> On Fri, Jan 26, 2018 at 11:42:25PM +0800, Jia-Ju Bai wrote:
>> After checking all possible call chains to on26_test_port() here,
>> my tool finds that this function is never called in atomic context,
>> namely never in an interrupt handler or holding a spinlock.
>> And on26_test_port() is only called by pi_probe_unit() that calls
>> wait_event() through pi_claim(),
>> so it indicates that on26_test_port() can call functions that can sleep.
>> Thus mdelay can be replaced with msleep to avoid busy wait.
> Sigh... Here's how I would've written it:
>
> "
> on26_test_port() is never called from atomic contexts.
>
> It has no direct callers and it is reachable only via ->test_port.
> ->test_port has only one user:
> drivers/block/paride/paride.c:322: max = pi->proto->test_port(pi);
> in pi_probe_unit(). That gets called only from pi_init(), called from
> p{d,cd,f,t,g}_detect(), called from module_init stuff, all of the above
> without entering atomic contexts along the way.
>
> Despite never getting called from atomic contexts, on26_test_port() contains
> mdelay(100), i.e. busy-loops for 0.1s; that's neither nice nor needed, since
> msleep() would serve just as well.
>
> Found by [reference to tool]"
Okay, thanks for your patient guidance :)
I will revise the description and send v2 patch, and follow the advice
in my next patches.
Thanks,
Jia-Ju Bai