Hi,
On 2022/10/20 16:39, Cezary Rojewski wrote:
On 2022-10-20 5:45 AM, Yang Yingliang wrote:
If snd_hdac_device_register() fails, 'codec' is leaked, free it in snd_hdac_device_exit(). And device_initialize() is called in snd_hdac_device_init(), it should call put_device() to give up reference or the name allocated in dev_set_name() is leaked. The snd_hdac_device_exit() will be called in dev->release().
Fixes: e4746d94d00c ("ASoC: Intel: Skylake: Introduce HDA codec init and exit routines") Signed-off-by: Yang Yingliang yangyingliang@huawei.com
sound/soc/intel/skylake/skl.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/sound/soc/intel/skylake/skl.c b/sound/soc/intel/skylake/skl.c index bbba2df33aaf..f0048e3fa619 100644 --- a/sound/soc/intel/skylake/skl.c +++ b/sound/soc/intel/skylake/skl.c @@ -691,7 +691,11 @@ static void load_codec_module(struct hda_codec *codec) static void skl_codec_device_exit(struct device *dev) { - snd_hdac_device_exit(dev_to_hdac_dev(dev)); + struct hdac_device *hdac_dev = dev_to_hdac_dev(dev); + struct hda_codec *codec = container_of(hdac_dev, struct hda_codec, core);
+ snd_hdac_device_exit(hdac_dev); + kfree(codec); } static struct hda_codec *skl_codec_device_init(struct hdac_bus *bus, int addr) @@ -711,7 +715,7 @@ static struct hda_codec *skl_codec_device_init(struct hdac_bus *bus, int addr) 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); }
Hello,
That's a tricky one. And that's because the code added in commit [1] which you're fixing here, basically mimics the original code that has been removed later on in the series of mine [2].
The key part is: function snd_hda_codec_device_init() replaced snd_hdac_device_init(). I believe the former may generate codec-leak in situations you speak of.
Yes, after using snd_hda_codec_device_init(), 'codec' is leaked. And there is another leak, the name allocated in dev_set_name() called in snd_hdac_device_init() is always leaked if snd_hdac_device_register() fails in snd_hdac_ext_bus_device_init() which is introduced by comit dfe66a18780d ("ALSA: hdac_ext: add extended HDA bus").
Now, on to the fix. My suggestion is to start with removing the following line: codec->core.dev.release = skl_codec_device_exit;
and skl_codec_device_exit() function itself. Let's leave the default as is and see if the leak still occurs.
OK, I will send a v2.
Thanks, Yang
Regards, Czarek .