| Message ID | 2750370.rlMuFcoBUA@wuerfel |
|---|---|
| State | Superseded |
| Headers | show |
On 9/8/2014 7:26 PM, Arnd Bergmann wrote: > gcc-4.9 found a potential condition under which the 'pending' > variable may be used uninitialized: > drivers/rtc/rtc-pcf8563.c: In function 'pcf8563_irq': > drivers/rtc/rtc-pcf8563.c:173:5: warning: 'pending' may be used uninitialized in this function [-Wmaybe-uninitialized] > This is because in the pcf8563_get_alarm_mode() function, we > check any nonzero return of pcf8563_read_block_data, but > in the irq function we only check for negative values, so > a possible positive value does not get detected if the compiler > chooses not to inline the entire call chain. > Checking for any non-zero value in the interrupt handler as well > is just as correct and lets the compiler know what we are doing, > without needing a bogus initialization. > As pointed out by Sergei Shtylyov, the same code section contains > another bug: an interrupt handler is not supposed to return > an errno value. Let's fix this as well by returning IRQ_NONE > in case of a communication error. > Signed-off-by: Arnd Bergmann <arnd@arndb.de> If you're fixing both issues in one patch, it probably needs somewhat modified subject, like "rtc: pcf8356: fix error handling". WBR, Sergei
diff --git a/drivers/rtc/rtc-pcf8563.c b/drivers/rtc/rtc-pcf8563.c index 5a197d9dc7e7..c2ef0a22ee94 100644 --- a/drivers/rtc/rtc-pcf8563.c +++ b/drivers/rtc/rtc-pcf8563.c @@ -167,8 +167,8 @@ static irqreturn_t pcf8563_irq(int irq, void *dev_id) char pending; err = pcf8563_get_alarm_mode(pcf8563->client, NULL, &pending); - if (err < 0) - return err; + if (err) + return IRQ_NONE; if (pending) { rtc_update_irq(pcf8563->rtc, 1, RTC_IRQF | RTC_AF);
gcc-4.9 found a potential condition under which the 'pending' variable may be used uninitialized: drivers/rtc/rtc-pcf8563.c: In function 'pcf8563_irq': drivers/rtc/rtc-pcf8563.c:173:5: warning: 'pending' may be used uninitialized in this function [-Wmaybe-uninitialized] This is because in the pcf8563_get_alarm_mode() function, we check any nonzero return of pcf8563_read_block_data, but in the irq function we only check for negative values, so a possible positive value does not get detected if the compiler chooses not to inline the entire call chain. Checking for any non-zero value in the interrupt handler as well is just as correct and lets the compiler know what we are doing, without needing a bogus initialization. As pointed out by Sergei Shtylyov, the same code section contains another bug: an interrupt handler is not supposed to return an errno value. Let's fix this as well by returning IRQ_NONE in case of a communication error. Signed-off-by: Arnd Bergmann <arnd@arndb.de>