Fixed the uninitialized use of a signed integer variable ret in
soc_probe_component when all its definitions are not executed. This
caused -ftrivial-auto-var-init=pattern to initialize the variable to
repeated 0xAA (i.e. a negative value) and triggered the following code
unintentionally.
err_probe:
if (ret < 0)
soc_cleanup_component(component);
Signed-off-by: Jian Cai <[email protected]>
---
sound/soc/soc-core.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c
index 068d809c349a..bfb813ba34f3 100644
--- a/sound/soc/soc-core.c
+++ b/sound/soc/soc-core.c
@@ -1180,7 +1180,7 @@ static int soc_probe_component(struct snd_soc_card *card,
snd_soc_component_get_dapm(component);
struct snd_soc_dai *dai;
int probed = 0;
- int ret;
+ int ret = 0;
if (!strcmp(component->name, "snd-soc-dummy"))
return 0;
--
2.25.0.341.g760bfbb309-goog
> Fixed the uninitialized use of a signed integer variable ret in
> soc_probe_component when all its definitions are not executed. This
> caused -ftrivial-auto-var-init=pattern to initialize the variable to
> repeated 0xAA (i.e. a negative value) and triggered the following code
> unintentionally.
> Signed-off-by: Jian Cai <[email protected]>
Hi Jian,
I don't quite follow; it looks like `ret` is assigned to multiple times in
`soc_probe_component`. Are one of the return values of one of the functions
that are called then assigned to `ret` undefined? What control flow path leaves
`ret` unitialized?
On Fri, Feb 7, 2020 at 12:55 AM Jian Cai <[email protected]> wrote:
>
> Hi Nick,
>
> 'ret' is only defined in if branches and for loops (e.g. for_each_component_dais). If none of these branches or loops get executed, then eventually we end up having
https://elixir.bootlin.com/linux/latest/source/sound/soc/soc-core.c#L1276
and
https://elixir.bootlin.com/linux/latest/source/sound/soc/soc-core.c#L1287
both assign to `ret` before any `goto` is taken. Are you perhaps
looking at an older branch of the LTS tree, but not the master branch
of the mainline tree? (Or it's possible that it's 1am here in Zurich,
and I should go to bed).
>
> int ret;
>
> err_probe:
> if (ret < 0)
> soc_cleanup_component(component);
>
> With -ftrivial-auto-var-init=pattern, this code becomes
>
> int ret;
>
> err_probe:
> ret = 0xAAAAAAAA;
> if (ret < 0)
> soc_cleanup_component(component);
>
> So soc_cleanup_component gets called unintentionally this case, which causes the built kernel to miss some files.
>
>
>
> On Thu, Feb 6, 2020 at 3:28 PM Nick Desaulniers <[email protected]> wrote:
>>
>> > Fixed the uninitialized use of a signed integer variable ret in
>> > soc_probe_component when all its definitions are not executed. This
>> > caused -ftrivial-auto-var-init=pattern to initialize the variable to
>> > repeated 0xAA (i.e. a negative value) and triggered the following code
>> > unintentionally.
>>
>> > Signed-off-by: Jian Cai <[email protected]>
>>
>> Hi Jian,
>> I don't quite follow; it looks like `ret` is assigned to multiple times in
>> `soc_probe_component`. Are one of the return values of one of the functions
>> that are called then assigned to `ret` undefined? What control flow path leaves
>> `ret` unitialized?
--
Thanks,
~Nick Desaulniers