On 2022/10/20 20:10, Cezary Rojewski wrote:
On 2022-10-20 12:59 PM, Yang Yingliang wrote:
If snd_hdac_device_register() fails, 'codec' and name allocated in dev_set_name() called in snd_hdac_device_init() are leaked. Fix this by calling put_device(), so they can be freed in snd_hda_codec_dev_release() and kobject_cleanup().
Fixes: e4746d94d00c ("ASoC: Intel: Skylake: Introduce HDA codec init and exit routines") Fixes: dfe66a18780d ("ALSA: hdac_ext: add extended HDA bus")
I do not believe the second Fixes-tag is required as it's not connected directly with the fix you're providing.
When the reference of device is leaked, it leads to two memory leak: 'codec' and 'dev->kobj.name' which is allocated in dev_set_name(). The name leak is introduce by the second fix tag. This patch calling put_device() to free both of them.
Tag: Suggested-by: Cezary Rojewski cezary.rojewski@intel.com
would be most welcome though.
Also, if there would happen to be v3, please bundle Skylake and SOF patches together into a single patchset (still not a single patch!).
OK, It's my pleasure.
Signed-off-by: Yang Yingliang yangyingliang@huawei.com
...
sound/soc/intel/skylake/skl.c | 8 +------- 1 file changed, 1 insertion(+), 7 deletions(-)
diff --git a/sound/soc/intel/skylake/skl.c b/sound/soc/intel/skylake/skl.c index bbba2df33aaf..3312b57e3c0c 100644 --- a/sound/soc/intel/skylake/skl.c +++ b/sound/soc/intel/skylake/skl.c @@ -689,11 +689,6 @@ static void load_codec_module(struct hda_codec *codec) #endif /* CONFIG_SND_SOC_INTEL_SKYLAKE_HDAUDIO_CODEC */ -static void skl_codec_device_exit(struct device *dev) -{ - snd_hdac_device_exit(dev_to_hdac_dev(dev)); -}
static struct hda_codec *skl_codec_device_init(struct hdac_bus *bus, int addr) { struct hda_codec *codec; @@ -706,12 +701,11 @@ static struct hda_codec *skl_codec_device_init(struct hdac_bus *bus, int addr) } codec->core.type = HDA_DEV_ASOC; - codec->core.dev.release = skl_codec_device_exit; ret = snd_hdac_device_register(&codec->core); if (ret) { dev_err(bus->dev, "failed to register hdac device\n"); - snd_hdac_device_exit(&codec->core); + put_device(&codec->core.dev); return ERR_PTR(ret); }
Shy question: why my suggestion seems reasonable, I did not test it yet, proposed it based on static analysis of the code. Did you test it?
I did it by static detailed analysis, there is a common driver core mechanism to make sure snd_hda_codec_dev_release() is called in device_release() when refcount hit 0, and it's ok to call snd_hda_codec_dev_release(), because the member of codec that need be freed is still null, it won't cause any problem. Could you test this patch if you wish ?
Thanks, Yang
Regards, Czarek
.