coresight: potential uninitialized variable in probe()
PATCH 1/1 coresight: potential uninitialized variable in probe()
The "drvdata->atclk" clock is optional, but if it gets set to an error pointer then we're accidentally return an uninitialized variable instead of success.
--- drivers/hwtracing/coresight/coresight-funnel.c | 1 + 1 file changed, 1 insertion(+)
drivers/hwtracing/coresight/coresight-funnel.c +1 -0
@@ -244,6 +244,7 @@ static int funnel_probe(struct device *dev, struct resource *res) }pm_runtime_put (dev); + ret =0 ; out_disable_clk:if (ret && !IS_ERR_OR_NULL (drvdata->atclk))
Signature · 1 line
2.20.1
Hi Dan,
Quoted 3 lines · Dan Carpenter
The "
drvdata->atclk" clock is optional, but if it gets set to an error pointer then we're accidentally return an uninitialized variable instead of success.
You are right, thanks a lot for pointing out.
I'd like to initialize 'ret = 0' at the head of function, so we can has the same fashion with other CoreSight drivers (e.g. replicator).
static int funnel_probe(struct device *dev, struct resource *res) {
- int ret;
+ int ret = 0;
If you agree, could you send a new patch for this?
Thanks,
Leo Yan
Quoted 21 lines
Fixes 78e6427b4e7b ("coresight: funnel: Support static funnel")Signed-off-by Dan Carpenter <[email protected]>--- drivers/hwtracing/coresight/coresight-funnel.c | 1 + 1 file changed, 1 insertion(+)drivers/hwtracing/coresight/coresight-funnel.c +1
-0@@ -244,6 +244,7 @@ static int funnel_probe(struct device *dev, struct resource *res) }pm_runtime_put (dev); + ret =0 ; out_disable_clk:if (ret && !IS_ERR_OR_NULL (drvdata->atclk))Signature · 1 line
2.20.1
Quoted 17 lines · Leo Yan
Hi Dan,
Quoted 3 lines · Dan Carpenter
The "
drvdata->atclk" clock is optional, but if it gets set to an error pointer then we're accidentally return an uninitialized variable instead of success.You are right, thanks a lot for pointing out.
I'd like to initialize 'ret = 0' at the head of function, so we can has the same fashion with other CoreSight drivers (e.g. replicator).
static int
funnel_probe(struct device *dev, struct resource *res){(quoted diff) +1
-1- int ret; + int ret = 0;
Quoted 1 line
If you agree, could you send a new patch for this?
Obviously that's an option I considered... The reason I didn't go with that is that a common bug that I see is:
int ret = 0; p = kmalloc(); if (!p) goto free_whatever;
In my experience it's better to initialize the return as late as possible so that you get static checker warnings when you forget to set the error code.
Also I think my way is more readable. I like to make the success path as explicit as possible. I hate when people do things like:
if (!ret) return ret;
About 10% of the time when you see this it is a bug, but it's hard to tell because it's not readable like it would be if people did:
if (!ret) return 0;
Or sometimes you see things like:
if (corner_case)
goto free; /* success path */
Without the "/* success path */ comment explaining why we're returning zero most readers will assume it's a mistake.
regards,
dan carpenter
Hi Dan,
Quoted 29 lines · Dan Carpenter
Quoted 17 lines · Leo Yan
Hi Dan,
Quoted 3 lines · Dan Carpenter
The "
drvdata->atclk" clock is optional, but if it gets set to an error pointer then we're accidentally return an uninitialized variable instead of success.You are right, thanks a lot for pointing out.
I'd like to initialize 'ret = 0' at the head of function, so we can has the same fashion with other CoreSight drivers (e.g. replicator).
static int
funnel_probe(struct device *dev, struct resource *res){(quoted diff) +1
-1- int ret; + int ret = 0;Quoted 1 line
If you agree, could you send a new patch for this?
Obviously that's an option I considered... The reason I didn't go with that is that a common bug that I see is:
int ret = 0; p = kmalloc(); if (!p) goto free_whatever;
Quoted 3 lines
In my experience it's better to initialize the return as late as possible so that you get static checker warnings when you forget to set the error code.
Just want to check one thing, which static checker you are using? I use sparse but it doesn't report this issue (I learned coccinelle also can do this but it outputs verbose logs).
To be honest, I didn't often run static checker when committed patches, but hope later can improve for this.
Quoted 12 lines
Also I think my way is more readable. I like to make the success path as explicit as possible. I hate when people do things like:
if (!ret) return ret;About 10% of the time when you see this it is a bug, but it's hard to tell because it's not readable like it would be if people did:
if (!ret) return 0;
Quoted 7 lines
Or sometimes you see things like:
if (
corner_case)
goto free; /* success path */Without the "/* success path */ comment explaining why we're returning zero most readers will assume it's a mistake.
Thanks for sharing much knowledge; your change is okay for me.
I think the point is the good habit can avoid pitfall and traps :) [1]
Thanks,
Leo Yan
[1] https://www.amazon.com/C-Traps-Pitfalls-Andrew-Koenig/dp/0201179288
On Thu, Jun 13, 2019 at 05:56:37PM +0800, Leo Yan wrote:
[...]
Quoted 5 lines
Quoted 3 lines
In my experience it's better to initialize the return as late as possible so that you get static checker warnings when you forget to set the error code.
Just want to check one thing, which static checker you are using?
Okay, I think it's smatch :)
Thanks,
Leo Yan
Quoted 16 lines · Dan Carpenter
The "
drvdata->atclk" clock is optional, but if it gets set to an error pointer then we're accidentally return an uninitialized variable instead of success.Fixes 78e6427b4e7b ("coresight: funnel: Support static funnel")Signed-off-by Dan Carpenter <[email protected]>--- drivers/hwtracing/coresight/coresight-funnel.c | 1 + 1 file changed, 1 insertion(+)drivers/hwtracing/coresight/coresight-funnel.c +0
-0
drivers/hwtracing/coresight/coresight-funnel.c static int funnel_probe(struct device *dev, struct resource *res) · line 247
Quoted 4 lines
drivers/hwtracing/coresight/coresight-funnel.c +1
-0@@ -244,6 +244,7 @@ static int funnel_probe(struct device *dev, struct resource *res) }pm_runtime_put (dev); + ret =0 ;
Applied - thanks.
Quoted 6 lines
drivers/hwtracing/coresight/coresight-funnel.c +0
-1out_disable_clk:if (ret && !IS_ERR_OR_NULL (drvdata->atclk)) --2.20.1