Re: [PATCH v5] ASoC: tas2552: Support TI TAS2552 Amplifier
Mark Thanks for the review On 07/07/2014 03:08 AM, Mark Brown wrote: > On Thu, Jul 03, 2014 at 11:24:35AM -0500, Dan Murphy wrote: > >> +static int tas2552_power(struct tas2552_data *data, u8 power) >> +{ >> +int ret = 0; >> + >> +mutex_lock(>mutex); >> + >> +if (power) { >> +if (data->enable_gpio) >> +gpiod_set_value(data->enable_gpio, 1); >> + >> +data->power_state = 1; >> +} else { >> +if (data->enable_gpio) >> +gpiod_set_value(data->enable_gpio, 0); >> + >> +data->power_state = 0; >> +} >> + >> +mutex_unlock(>mutex); >> +return ret; >> +} > I don't understand this function. It appears to be the only place where > either power_state or mutex is used so it's just adding some wrapping > around setting the GPIO value which doesn't seem like it's doing much. > Why are we tracking power_state? This function and mutex are artifacts from the development and probably can be consolidated into the runtime PM calls. > >> +static void tas2552_sw_shutdown(struct tas2552_data *tas_data, int >> sw_shutdown) >> +{ >> +u8 cfg1_reg = 0x0; >> + >> +if (sw_shutdown) >> +cfg1_reg |= TAS2552_SWS_MASK; >> +else >> +cfg1_reg &= ~TAS2552_SWS_MASK; >> + >> +snd_soc_update_bits(tas_data->codec, TAS2552_CFG_1, >> + TAS2552_SWS_MASK, cfg1_reg); > Given that you're using _update_bits() clearing the bits in a register > that was just initialised to zero doesn't make a huge amount of sense. This was an artifact from RFC in which I was not using the snd_soc functions. I can remove the initialization of the variable. > >> +default: >> +dev_vdbg(codec->dev, "Substream sample rate is not found\n"); >> +return -EINVAL; >> +} > Better to print the rate. OK > >> +pm_runtime_get_sync(codec->dev); >> + >> +snd_soc_update_bits(codec, TAS2552_CFG_3, TAS2552_WCLK_MASK, wclk_reg); >> + >> +pm_runtime_put(codec->dev); > This seems really strange - why is the device being powered up to just > set a bit and then potentially powered down immediately? I'd expect to > just update the cache if the device is not active. You're also not > checking that the power up worked. I wanted to make sure that the device was on. I totally forgot that the device was using regmap and the values are cached when the device is not on. I can remove the get/put around the update calls. > >> + >> +static int tas2552_set_dai_fmt(struct snd_soc_dai *dai, unsigned int fmt) >> +{ >> +u8 serial_format; >> +u8 serial_control_mask = 0x00; >> +if (fmt & SND_SOC_DAIFMT_FORMAT_MASK) >> +serial_control_mask |= TAS2552_DATA_FORMAT_MASK; >> +if (serial_control_mask) { >> +pm_runtime_get_sync(codec->dev); >> + >> +snd_soc_update_bits(codec, TAS2552_SER_CTRL_1, >> serial_control_mask, >> +serial_format); >> + >> +pm_runtime_put(codec->dev); >> +} > This seems broken - if the format mask ever gets set then it won't be > cleared since we only do an update_bits() if the bit is being set. Why > isn't the driver just doing an _update_bits()? I do not understand what this statement means. Are you saying snd_soc_update_bits will not clear the bit if the bit mask is set appropriately? > > The comments about runtime PM also apply, they applies throughout the > driver. > >> +static int tas2552_set_dai_sysclk(struct snd_soc_dai *dai, int clk_id, >> + unsigned int freq, int dir) >> +{ >> +struct snd_soc_codec *codec = dai->codec; >> + >> +/* Fill in the PLL control registers for J & D >> + * PLL_CLK = (.5 * freq * J.D) / 2^p >> + * Need to fill in J and D here based on incoming freq >> + */ >> +pm_runtime_get_sync(codec->dev); >> + >> +snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, 0); >> + >> +snd_soc_write(codec, TAS2552_PLL_CTRL_1, 0x10); >> +snd_soc_write(codec, TAS2552_PLL_CTRL_2, 0x00); >> +snd_soc_write(codec, TAS2552_PLL_CTRL_3, 0x00); >> + >> +snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, >> +TAS2552_PLL_ENABLE); >> + >> +pm_runtime_put(codec->dev); > This makes no sense at all - please look at what other drivers are doing > with set_sysclk(). It should be used to get information about how the > device is clocked. > >> +static int tas2552_startup(struct snd_pcm_substream *substream, >> + struct snd_soc_dai *dai) >> +{ >> +struct snd_soc_codec *codec = dai->codec; >> + >> +pm_runtime_get_sync(codec->dev); >> + >> +/* Turn on Class D amplifier */ >> +snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_CLASSD_EN_MASK, >> +TAS2552_CLASSD_EN); > This should be done using DAPM. You
Re: [PATCH v5] ASoC: tas2552: Support TI TAS2552 Amplifier
On Thu, Jul 03, 2014 at 11:24:35AM -0500, Dan Murphy wrote: > +static int tas2552_power(struct tas2552_data *data, u8 power) > +{ > + int ret = 0; > + > + mutex_lock(>mutex); > + > + if (power) { > + if (data->enable_gpio) > + gpiod_set_value(data->enable_gpio, 1); > + > + data->power_state = 1; > + } else { > + if (data->enable_gpio) > + gpiod_set_value(data->enable_gpio, 0); > + > + data->power_state = 0; > + } > + > + mutex_unlock(>mutex); > + return ret; > +} I don't understand this function. It appears to be the only place where either power_state or mutex is used so it's just adding some wrapping around setting the GPIO value which doesn't seem like it's doing much. Why are we tracking power_state? > +static void tas2552_sw_shutdown(struct tas2552_data *tas_data, int > sw_shutdown) > +{ > + u8 cfg1_reg = 0x0; > + > + if (sw_shutdown) > + cfg1_reg |= TAS2552_SWS_MASK; > + else > + cfg1_reg &= ~TAS2552_SWS_MASK; > + > + snd_soc_update_bits(tas_data->codec, TAS2552_CFG_1, > + TAS2552_SWS_MASK, cfg1_reg); Given that you're using _update_bits() clearing the bits in a register that was just initialised to zero doesn't make a huge amount of sense. > + default: > + dev_vdbg(codec->dev, "Substream sample rate is not found\n"); > + return -EINVAL; > + } Better to print the rate. > + pm_runtime_get_sync(codec->dev); > + > + snd_soc_update_bits(codec, TAS2552_CFG_3, TAS2552_WCLK_MASK, wclk_reg); > + > + pm_runtime_put(codec->dev); This seems really strange - why is the device being powered up to just set a bit and then potentially powered down immediately? I'd expect to just update the cache if the device is not active. You're also not checking that the power up worked. > + > +static int tas2552_set_dai_fmt(struct snd_soc_dai *dai, unsigned int fmt) > +{ > + u8 serial_format; > + u8 serial_control_mask = 0x00; > + if (fmt & SND_SOC_DAIFMT_FORMAT_MASK) > + serial_control_mask |= TAS2552_DATA_FORMAT_MASK; > + if (serial_control_mask) { > + pm_runtime_get_sync(codec->dev); > + > + snd_soc_update_bits(codec, TAS2552_SER_CTRL_1, > serial_control_mask, > + serial_format); > + > + pm_runtime_put(codec->dev); > + } This seems broken - if the format mask ever gets set then it won't be cleared since we only do an update_bits() if the bit is being set. Why isn't the driver just doing an _update_bits()? The comments about runtime PM also apply, they applies throughout the driver. > +static int tas2552_set_dai_sysclk(struct snd_soc_dai *dai, int clk_id, > + unsigned int freq, int dir) > +{ > + struct snd_soc_codec *codec = dai->codec; > + > + /* Fill in the PLL control registers for J & D > + * PLL_CLK = (.5 * freq * J.D) / 2^p > + * Need to fill in J and D here based on incoming freq > + */ > + pm_runtime_get_sync(codec->dev); > + > + snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, 0); > + > + snd_soc_write(codec, TAS2552_PLL_CTRL_1, 0x10); > + snd_soc_write(codec, TAS2552_PLL_CTRL_2, 0x00); > + snd_soc_write(codec, TAS2552_PLL_CTRL_3, 0x00); > + > + snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, > + TAS2552_PLL_ENABLE); > + > + pm_runtime_put(codec->dev); This makes no sense at all - please look at what other drivers are doing with set_sysclk(). It should be used to get information about how the device is clocked. > +static int tas2552_startup(struct snd_pcm_substream *substream, > +struct snd_soc_dai *dai) > +{ > + struct snd_soc_codec *codec = dai->codec; > + > + pm_runtime_get_sync(codec->dev); > + > + /* Turn on Class D amplifier */ > + snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_CLASSD_EN_MASK, > + TAS2552_CLASSD_EN); This should be done using DAPM. > +static int tas2552_codec_probe(struct snd_soc_codec *codec) > +{ > + struct tas2552_data *tas2552 = snd_soc_codec_get_drvdata(codec); > + int ret; > + > + tas2552_power(tas2552, 1); > + tas2552_sw_shutdown(tas2552, 0); > + > + pm_runtime_set_active(codec->dev); > + pm_runtime_set_autosuspend_delay(codec->dev, 1000); > + pm_runtime_use_autosuspend(codec->dev); > + pm_runtime_enable(codec->dev); > + pm_runtime_mark_last_busy(codec->dev); > + pm_runtime_put_sync_autosuspend(codec->dev); This should all be done at the device level probe. > + /* 0dB gain */ > + snd_soc_write(codec, TAS2552_PGA_GAIN, 0x10); Use the hardware default, your default might not be sensible for some other user. > +
Re: [PATCH v5] ASoC: tas2552: Support TI TAS2552 Amplifier
On Thu, Jul 03, 2014 at 11:24:35AM -0500, Dan Murphy wrote: +static int tas2552_power(struct tas2552_data *data, u8 power) +{ + int ret = 0; + + mutex_lock(data-mutex); + + if (power) { + if (data-enable_gpio) + gpiod_set_value(data-enable_gpio, 1); + + data-power_state = 1; + } else { + if (data-enable_gpio) + gpiod_set_value(data-enable_gpio, 0); + + data-power_state = 0; + } + + mutex_unlock(data-mutex); + return ret; +} I don't understand this function. It appears to be the only place where either power_state or mutex is used so it's just adding some wrapping around setting the GPIO value which doesn't seem like it's doing much. Why are we tracking power_state? +static void tas2552_sw_shutdown(struct tas2552_data *tas_data, int sw_shutdown) +{ + u8 cfg1_reg = 0x0; + + if (sw_shutdown) + cfg1_reg |= TAS2552_SWS_MASK; + else + cfg1_reg = ~TAS2552_SWS_MASK; + + snd_soc_update_bits(tas_data-codec, TAS2552_CFG_1, + TAS2552_SWS_MASK, cfg1_reg); Given that you're using _update_bits() clearing the bits in a register that was just initialised to zero doesn't make a huge amount of sense. + default: + dev_vdbg(codec-dev, Substream sample rate is not found\n); + return -EINVAL; + } Better to print the rate. + pm_runtime_get_sync(codec-dev); + + snd_soc_update_bits(codec, TAS2552_CFG_3, TAS2552_WCLK_MASK, wclk_reg); + + pm_runtime_put(codec-dev); This seems really strange - why is the device being powered up to just set a bit and then potentially powered down immediately? I'd expect to just update the cache if the device is not active. You're also not checking that the power up worked. + +static int tas2552_set_dai_fmt(struct snd_soc_dai *dai, unsigned int fmt) +{ + u8 serial_format; + u8 serial_control_mask = 0x00; + if (fmt SND_SOC_DAIFMT_FORMAT_MASK) + serial_control_mask |= TAS2552_DATA_FORMAT_MASK; + if (serial_control_mask) { + pm_runtime_get_sync(codec-dev); + + snd_soc_update_bits(codec, TAS2552_SER_CTRL_1, serial_control_mask, + serial_format); + + pm_runtime_put(codec-dev); + } This seems broken - if the format mask ever gets set then it won't be cleared since we only do an update_bits() if the bit is being set. Why isn't the driver just doing an _update_bits()? The comments about runtime PM also apply, they applies throughout the driver. +static int tas2552_set_dai_sysclk(struct snd_soc_dai *dai, int clk_id, + unsigned int freq, int dir) +{ + struct snd_soc_codec *codec = dai-codec; + + /* Fill in the PLL control registers for J D + * PLL_CLK = (.5 * freq * J.D) / 2^p + * Need to fill in J and D here based on incoming freq + */ + pm_runtime_get_sync(codec-dev); + + snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, 0); + + snd_soc_write(codec, TAS2552_PLL_CTRL_1, 0x10); + snd_soc_write(codec, TAS2552_PLL_CTRL_2, 0x00); + snd_soc_write(codec, TAS2552_PLL_CTRL_3, 0x00); + + snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, + TAS2552_PLL_ENABLE); + + pm_runtime_put(codec-dev); This makes no sense at all - please look at what other drivers are doing with set_sysclk(). It should be used to get information about how the device is clocked. +static int tas2552_startup(struct snd_pcm_substream *substream, +struct snd_soc_dai *dai) +{ + struct snd_soc_codec *codec = dai-codec; + + pm_runtime_get_sync(codec-dev); + + /* Turn on Class D amplifier */ + snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_CLASSD_EN_MASK, + TAS2552_CLASSD_EN); This should be done using DAPM. +static int tas2552_codec_probe(struct snd_soc_codec *codec) +{ + struct tas2552_data *tas2552 = snd_soc_codec_get_drvdata(codec); + int ret; + + tas2552_power(tas2552, 1); + tas2552_sw_shutdown(tas2552, 0); + + pm_runtime_set_active(codec-dev); + pm_runtime_set_autosuspend_delay(codec-dev, 1000); + pm_runtime_use_autosuspend(codec-dev); + pm_runtime_enable(codec-dev); + pm_runtime_mark_last_busy(codec-dev); + pm_runtime_put_sync_autosuspend(codec-dev); This should all be done at the device level probe. + /* 0dB gain */ + snd_soc_write(codec, TAS2552_PGA_GAIN, 0x10); Use the hardware default, your default might not be sensible for some other user. + /** + * Data sheet indicates to write 0x0c to 0x0d during init but no + * additional information is given to what
Re: [PATCH v5] ASoC: tas2552: Support TI TAS2552 Amplifier
Mark Thanks for the review On 07/07/2014 03:08 AM, Mark Brown wrote: On Thu, Jul 03, 2014 at 11:24:35AM -0500, Dan Murphy wrote: +static int tas2552_power(struct tas2552_data *data, u8 power) +{ +int ret = 0; + +mutex_lock(data-mutex); + +if (power) { +if (data-enable_gpio) +gpiod_set_value(data-enable_gpio, 1); + +data-power_state = 1; +} else { +if (data-enable_gpio) +gpiod_set_value(data-enable_gpio, 0); + +data-power_state = 0; +} + +mutex_unlock(data-mutex); +return ret; +} I don't understand this function. It appears to be the only place where either power_state or mutex is used so it's just adding some wrapping around setting the GPIO value which doesn't seem like it's doing much. Why are we tracking power_state? This function and mutex are artifacts from the development and probably can be consolidated into the runtime PM calls. +static void tas2552_sw_shutdown(struct tas2552_data *tas_data, int sw_shutdown) +{ +u8 cfg1_reg = 0x0; + +if (sw_shutdown) +cfg1_reg |= TAS2552_SWS_MASK; +else +cfg1_reg = ~TAS2552_SWS_MASK; + +snd_soc_update_bits(tas_data-codec, TAS2552_CFG_1, + TAS2552_SWS_MASK, cfg1_reg); Given that you're using _update_bits() clearing the bits in a register that was just initialised to zero doesn't make a huge amount of sense. This was an artifact from RFC in which I was not using the snd_soc functions. I can remove the initialization of the variable. +default: +dev_vdbg(codec-dev, Substream sample rate is not found\n); +return -EINVAL; +} Better to print the rate. OK +pm_runtime_get_sync(codec-dev); + +snd_soc_update_bits(codec, TAS2552_CFG_3, TAS2552_WCLK_MASK, wclk_reg); + +pm_runtime_put(codec-dev); This seems really strange - why is the device being powered up to just set a bit and then potentially powered down immediately? I'd expect to just update the cache if the device is not active. You're also not checking that the power up worked. I wanted to make sure that the device was on. I totally forgot that the device was using regmap and the values are cached when the device is not on. I can remove the get/put around the update calls. + +static int tas2552_set_dai_fmt(struct snd_soc_dai *dai, unsigned int fmt) +{ +u8 serial_format; +u8 serial_control_mask = 0x00; +if (fmt SND_SOC_DAIFMT_FORMAT_MASK) +serial_control_mask |= TAS2552_DATA_FORMAT_MASK; +if (serial_control_mask) { +pm_runtime_get_sync(codec-dev); + +snd_soc_update_bits(codec, TAS2552_SER_CTRL_1, serial_control_mask, +serial_format); + +pm_runtime_put(codec-dev); +} This seems broken - if the format mask ever gets set then it won't be cleared since we only do an update_bits() if the bit is being set. Why isn't the driver just doing an _update_bits()? I do not understand what this statement means. Are you saying snd_soc_update_bits will not clear the bit if the bit mask is set appropriately? The comments about runtime PM also apply, they applies throughout the driver. +static int tas2552_set_dai_sysclk(struct snd_soc_dai *dai, int clk_id, + unsigned int freq, int dir) +{ +struct snd_soc_codec *codec = dai-codec; + +/* Fill in the PLL control registers for J D + * PLL_CLK = (.5 * freq * J.D) / 2^p + * Need to fill in J and D here based on incoming freq + */ +pm_runtime_get_sync(codec-dev); + +snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, 0); + +snd_soc_write(codec, TAS2552_PLL_CTRL_1, 0x10); +snd_soc_write(codec, TAS2552_PLL_CTRL_2, 0x00); +snd_soc_write(codec, TAS2552_PLL_CTRL_3, 0x00); + +snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_PLL_ENABLE, +TAS2552_PLL_ENABLE); + +pm_runtime_put(codec-dev); This makes no sense at all - please look at what other drivers are doing with set_sysclk(). It should be used to get information about how the device is clocked. +static int tas2552_startup(struct snd_pcm_substream *substream, + struct snd_soc_dai *dai) +{ +struct snd_soc_codec *codec = dai-codec; + +pm_runtime_get_sync(codec-dev); + +/* Turn on Class D amplifier */ +snd_soc_update_bits(codec, TAS2552_CFG_2, TAS2552_CLASSD_EN_MASK, +TAS2552_CLASSD_EN); This should be done using DAPM. You mentioned this before. Isn't the startup part of the DAPM calls? +static int tas2552_codec_probe(struct snd_soc_codec *codec) +{ +struct tas2552_data *tas2552 = snd_soc_codec_get_drvdata(codec); +int ret; +