| Message ID | 1392808032-25756-1-git-send-email-zonque@gmail.com (mailing list archive) |
|---|---|
| State | Accepted |
| Delegated to: | Mark Brown |
| Headers | show |
On Wed, Feb 19, 2014 at 12:07:12PM +0100, Daniel Mack wrote: > + if (gpio_is_valid(cs4271->gpio_nreset)) { > /* Set codec to the reset state */ > gpio_set_value(cs4271->gpio_nreset, 0); > + devm_gpio_free(codec->dev, cs4271->gpio_nreset); > + } If it's being requested as a managed resource shouldn't it be being freed automatically?
On 02/19/2014 12:15 PM, Mark Brown wrote: > On Wed, Feb 19, 2014 at 12:07:12PM +0100, Daniel Mack wrote: > >> + if (gpio_is_valid(cs4271->gpio_nreset)) { >> /* Set codec to the reset state */ >> gpio_set_value(cs4271->gpio_nreset, 0); >> + devm_gpio_free(codec->dev, cs4271->gpio_nreset); >> + } > > If it's being requested as a managed resource shouldn't it be being > freed automatically? > Nope, as the module itself will not go away necessarily. I hit the bug when unloading and reloading the machine driver. cs4271_probe(codec) will fail to acquire and drive the reset line, and the codec stays in reset. Daniel
On Wed, Feb 19, 2014 at 12:24:23PM +0100, Daniel Mack wrote: > On 02/19/2014 12:15 PM, Mark Brown wrote: > > On Wed, Feb 19, 2014 at 12:07:12PM +0100, Daniel Mack wrote: > >> + if (gpio_is_valid(cs4271->gpio_nreset)) { > >> /* Set codec to the reset state */ > >> gpio_set_value(cs4271->gpio_nreset, 0); > >> + devm_gpio_free(codec->dev, cs4271->gpio_nreset); > >> + } > > If it's being requested as a managed resource shouldn't it be being > > freed automatically? > Nope, as the module itself will not go away necessarily. I hit the bug > when unloading and reloading the machine driver. cs4271_probe(codec) > will fail to acquire and drive the reset line, and the codec stays in reset. The fix here is to move the resource acquisition to the bus level probe instead of the ASoC card startup.
On 02/19/2014 12:24 PM, Daniel Mack wrote: > On 02/19/2014 12:15 PM, Mark Brown wrote: >> On Wed, Feb 19, 2014 at 12:07:12PM +0100, Daniel Mack wrote: >> >>> + if (gpio_is_valid(cs4271->gpio_nreset)) { >>> /* Set codec to the reset state */ >>> gpio_set_value(cs4271->gpio_nreset, 0); >>> + devm_gpio_free(codec->dev, cs4271->gpio_nreset); >>> + } >> >> If it's being requested as a managed resource shouldn't it be being >> freed automatically? >> > > Nope, as the module itself will not go away necessarily. I hit the bug > when unloading and reloading the machine driver. cs4271_probe(codec) > will fail to acquire and drive the reset line, and the codec stays in reset. > In my opinion a better solution is to move the requesting of the GPIO from the CODEC probe() function to the SPI/I2C probe() functions. Actually it looks as if most of the code (if not all) in the CODEC probe() function should be done when the driver is probed. - Lars
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 02/19/2014 12:43 PM, Mark Brown wrote: > On Wed, Feb 19, 2014 at 12:24:23PM +0100, Daniel Mack wrote: >> On 02/19/2014 12:15 PM, Mark Brown wrote: >>> On Wed, Feb 19, 2014 at 12:07:12PM +0100, Daniel Mack wrote: > >>>> + if (gpio_is_valid(cs4271->gpio_nreset)) { /* Set codec to >>>> the reset state */ gpio_set_value(cs4271->gpio_nreset, 0); + >>>> devm_gpio_free(codec->dev, cs4271->gpio_nreset); + } > >>> If it's being requested as a managed resource shouldn't it be >>> being freed automatically? > >> Nope, as the module itself will not go away necessarily. I hit >> the bug when unloading and reloading the machine driver. >> cs4271_probe(codec) will fail to acquire and drive the reset >> line, and the codec stays in reset. > > The fix here is to move the resource acquisition to the bus level > probe instead of the ASoC card startup. > Alright. I thought there was a verdict once that resource claiming should be done right before usage, but you're right, it really rather belongs into the bus probe functions. I'll send another version. Thanks! Daniel -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/ iQIcBAEBAgAGBQJTBJ72AAoJELphMRr8Y1QkXt8P/RErTL26/X3s/MnYfhuPDzsx T7oU/YSpYvcqBVBab3rZjbyELRPqcrWO0XaiVFLiOqwt2KslKK/k8zz8gZIsvOxM YzheQMzI45sosYCFf1OssC7ojwTTks/hwgKzOj0Hi0aWtSVYRgV2kvg8S4tiwEzl OpM9Z4r0lCw9kJidlSwPCXcg+s4BN6xMcqaTIFfU1QCQ84dxHwzhf7cYZliOtWEv FlnMVKYAtBVCACVqBxHS+2OdopcTuwQ0aqts7UMXFq2MGCIYf6Fyq/qmmCMuaZrN yDcm3fpJodIoZo20nIf2zpaPGgD7ce+LK/Mu46sZ4MFRBxHgGlQ0C2UNZgNlVILI mY6hj8s44kCptX1sDPTviYoRnxQ4PTAdRUjfVRuShjkEFi8WHKEbIHwLqn0BvjSB FaaESTg5NhBJsAroTK83nqqY7MxcLZV4YBhnoH4Mhc2IZHa5hOxcHy5crGyUUEGF 4QWMJoHW+J3s8malsvwcNj8Fv4a6SaeptNr+4EkGW/lwrVAjl/03Ei7MvvqApyPm TSRuBXQUbiJhxBpkAMhFQ7jIj3Gw+CcgiudcwLA1sVdN2uTvbe/YRas/ZlWQn/W9 TVihbfTM8apHwCacyJYX7yE16eJ162AznYqCZC5ZCJ1L2k8ep30bhwEJk2nq4GP1 6FikGJ1wGOLdvDbwOc8b =dthk -----END PGP SIGNATURE-----
On Wed, Feb 19, 2014 at 01:09:27PM +0100, Daniel Mack wrote: > Alright. I thought there was a verdict once that resource claiming > should be done right before usage, but you're right, it really rather > belongs into the bus probe functions. No, that's not ever been the case as far as I can recall. There were some things it used to not be possible to do due to the ASoC level I/O only working after the card comes up but that's a different issue.
diff --git a/sound/soc/codecs/cs4271.c b/sound/soc/codecs/cs4271.c index ce05fd9..aeaa5c6 100644 --- a/sound/soc/codecs/cs4271.c +++ b/sound/soc/codecs/cs4271.c @@ -604,9 +604,11 @@ static int cs4271_remove(struct snd_soc_codec *codec) { struct cs4271_private *cs4271 = snd_soc_codec_get_drvdata(codec); - if (gpio_is_valid(cs4271->gpio_nreset)) + if (gpio_is_valid(cs4271->gpio_nreset)) { /* Set codec to the reset state */ gpio_set_value(cs4271->gpio_nreset, 0); + devm_gpio_free(codec->dev, cs4271->gpio_nreset); + } return 0; };
The reset GPIO line is acquired in cs4271_probe(), so it should be freed from cs4271_remove() as well. This way, the driver can unloaded and reloaded more than once. Signed-off-by: Daniel Mack <zonque@gmail.com> --- sound/soc/codecs/cs4271.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)