When CONFIG_SPI is disabled, the newly added code for the DSP
firmware loading fails to link:
ERROR: "rt5677_spi_hotword_detected" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
ERROR: "rt5677_spi_write" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
Add a dependency to prevent this configuration.
Note: the does not work with the DT probing, as there is no binding
for the SPI half of the driver, but nothing seems to be using that
with the mainline kernel anyway.
Fixes: 461c623270e4 ("ASoC: rt5677: Load firmware via SPI using delayed work")
Signed-off-by: Arnd Bergmann <[email protected]>
---
sound/soc/codecs/Kconfig | 1 +
sound/soc/intel/boards/Kconfig | 1 +
sound/soc/mediatek/Kconfig | 2 +-
sound/soc/tegra/Kconfig | 2 +-
4 files changed, 4 insertions(+), 2 deletions(-)
diff --git a/sound/soc/codecs/Kconfig b/sound/soc/codecs/Kconfig
index 71b7286d14f2..8e9c461a84f8 100644
--- a/sound/soc/codecs/Kconfig
+++ b/sound/soc/codecs/Kconfig
@@ -1047,6 +1047,7 @@ config SND_SOC_RT5670
config SND_SOC_RT5677
tristate
+ depends on SPI_MASTER
select REGMAP_I2C
select REGMAP_IRQ
diff --git a/sound/soc/intel/boards/Kconfig b/sound/soc/intel/boards/Kconfig
index ef20316e83d1..da7f9111f3d3 100644
--- a/sound/soc/intel/boards/Kconfig
+++ b/sound/soc/intel/boards/Kconfig
@@ -34,6 +34,7 @@ if SND_SOC_INTEL_HASWELL || SND_SOC_SOF_BROADWELL
config SND_SOC_INTEL_BDW_RT5677_MACH
tristate "Broadwell with RT5677 codec"
depends on I2C
+ depends on SPI_MASTER
depends on I2C_DESIGNWARE_PLATFORM || COMPILE_TEST
depends on GPIOLIB || COMPILE_TEST
depends on X86_INTEL_LPSS || COMPILE_TEST
diff --git a/sound/soc/mediatek/Kconfig b/sound/soc/mediatek/Kconfig
index a656d2014127..4699787c93ed 100644
--- a/sound/soc/mediatek/Kconfig
+++ b/sound/soc/mediatek/Kconfig
@@ -97,7 +97,7 @@ config SND_SOC_MT8173_RT5650_RT5514
config SND_SOC_MT8173_RT5650_RT5676
tristate "ASoC Audio driver for MT8173 with RT5650 RT5676 codecs"
- depends on SND_SOC_MT8173 && I2C
+ depends on SND_SOC_MT8173 && I2C && SPI_MASTER
select SND_SOC_RT5645
select SND_SOC_RT5677
select SND_SOC_HDMI_CODEC
diff --git a/sound/soc/tegra/Kconfig b/sound/soc/tegra/Kconfig
index addadc827b91..df36e84c0116 100644
--- a/sound/soc/tegra/Kconfig
+++ b/sound/soc/tegra/Kconfig
@@ -122,7 +122,7 @@ config SND_SOC_TEGRA_MAX98090
config SND_SOC_TEGRA_RT5677
tristate "SoC Audio support for Tegra boards using a RT5677 codec"
- depends on SND_SOC_TEGRA && I2C && GPIOLIB
+ depends on SND_SOC_TEGRA && I2C && GPIOLIB && SPI_MASTER
select SND_SOC_RT5677
help
Say Y or M here if you want to add support for SoC audio on Tegra
--
2.20.0
On 10/12/2019 19:52, Arnd Bergmann wrote:
> When CONFIG_SPI is disabled, the newly added code for the DSP
> firmware loading fails to link:
>
> ERROR: "rt5677_spi_hotword_detected" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
> ERROR: "rt5677_spi_write" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
Would it be better if the above functions or the functions that call
these are conditional on CONFIG_SND_SOC_RT5677_SPI?
> Add a dependency to prevent this configuration.
>
> Note: the does not work with the DT probing, as there is no binding
Are you missing 'SPI' or something here?
> for the SPI half of the driver, but nothing seems to be using that
> with the mainline kernel anyway.
From a Tegra perspective, given that we don't use SPI in conjunction
with the rt5677 codec, only I2C so far, I am not sure we should make the
tegra_rt5677 driver dependent upon it. We should be able to operate
without the SPI bits enabled.
Cheers
Jon
--
nvpublic
On 12/11/19 4:59 AM, Jon Hunter wrote:
>
> On 10/12/2019 19:52, Arnd Bergmann wrote:
>> When CONFIG_SPI is disabled, the newly added code for the DSP
>> firmware loading fails to link:
>>
>> ERROR: "rt5677_spi_hotword_detected" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
>> ERROR: "rt5677_spi_write" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
>
> Would it be better if the above functions or the functions that call
> these are conditional on CONFIG_SND_SOC_RT5677_SPI?
they are already conditional, with a fallback provided:
#if IS_ENABLED(CONFIG_SND_SOC_RT5677_SPI)
int rt5677_spi_read(u32 addr, void *rxbuf, size_t len);
int rt5677_spi_write(u32 addr, const void *txbuf, size_t len);
int rt5677_spi_write_firmware(u32 addr, const struct firmware *fw);
void rt5677_spi_hotword_detected(void);
#else
static inline int rt5677_spi_read(u32 addr, void *rxbuf, size_t len)
{
return -EINVAL;
}
static inline int rt5677_spi_write(u32 addr, const void *txbuf, size_t len)
{
return -EINVAL;
}
static inline int rt5677_spi_write_firmware(u32 addr, const struct
firmware *fw)
{
return -EINVAL;
}
static inline void rt5677_spi_hotword_detected(void){}
#endif
and since we have the following definition
config SND_SOC_RT5677_SPI
tristate
default SND_SOC_RT5677 && SPI
in theory if SPI is not enabled the fallback static inlines would always
be selected?
Arnd, if you can share the .config that exposes this problem it'd be nice
FWIW, there are other missing dependencies, the SPI controller was not
explicitly enabled so depending on the Kconfigs used by a distro the
machine driver probe could fail with the spi-RT5677AA component never
registered. The patch below seems to work for me (more testing needed)
diff --git a/sound/soc/intel/boards/Kconfig b/sound/soc/intel/boards/Kconfig
index b149e28a2076..cf1a2fde4c47 100644
--- a/sound/soc/intel/boards/Kconfig
+++ b/sound/soc/intel/boards/Kconfig
@@ -50,6 +50,9 @@ config SND_SOC_INTEL_BDW_RT5677_MACH
depends on I2C_DESIGNWARE_PLATFORM || COMPILE_TEST
depends on GPIOLIB || COMPILE_TEST
depends on X86_INTEL_LPSS || COMPILE_TEST
+ depends on SPI_MASTER
+ select SPI_PXA2XX
+ select SND_SOC_RT5677_SPI
select SND_SOC_RT5677
help
This adds support for Intel Broadwell platform based boards with
On Wed, Dec 11, 2019 at 3:00 AM Jon Hunter <[email protected]> wrote:
>
>
> On 10/12/2019 19:52, Arnd Bergmann wrote:
> > When CONFIG_SPI is disabled, the newly added code for the DSP
> > firmware loading fails to link:
> >
> > ERROR: "rt5677_spi_hotword_detected" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
> > ERROR: "rt5677_spi_write" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
>
> Would it be better if the above functions or the functions that call
> these are conditional on CONFIG_SND_SOC_RT5677_SPI?
>
> > Add a dependency to prevent this configuration.
> >
> > Note: the does not work with the DT probing, as there is no binding
>
> Are you missing 'SPI' or something here?
>
> > for the SPI half of the driver, but nothing seems to be using that
> > with the mainline kernel anyway.
>
> From a Tegra perspective, given that we don't use SPI in conjunction
> with the rt5677 codec, only I2C so far, I am not sure we should make the
> tegra_rt5677 driver dependent upon it. We should be able to operate
> without the SPI bits enabled.
>
There should be no changes needed for tegra, this should be isolated
to the bdw machine driver. The only things added to the machine driver
were some dai links.
> Cheers
> Jon
>
> --
> nvpublic
On Mon, Jan 6, 2020 at 10:55 PM Pierre-Louis Bossart
<[email protected]> wrote:
> On 12/11/19 4:59 AM, Jon Hunter wrote:
> >
> > On 10/12/2019 19:52, Arnd Bergmann wrote:
> >> When CONFIG_SPI is disabled, the newly added code for the DSP
> >> firmware loading fails to link:
> >>
> >> ERROR: "rt5677_spi_hotword_detected" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
> >> ERROR: "rt5677_spi_write" [sound/soc/codecs/snd-soc-rt5677.ko] undefined!
> >
> > Would it be better if the above functions or the functions that call
> > these are conditional on CONFIG_SND_SOC_RT5677_SPI?
>
> they are already conditional, with a fallback provided:
>
> #if IS_ENABLED(CONFIG_SND_SOC_RT5677_SPI)
> int rt5677_spi_read(u32 addr, void *rxbuf, size_t len);
Right, this fixed the problem.
> in theory if SPI is not enabled the fallback static inlines would always
> be selected?
>
> Arnd, if you can share the .config that exposes this problem it'd be nice
I just tried it again and it seems that the issue was already fixed by
commit fb3194413d1e ("ASoC: rt5677: Fix build error without CONFIG_SPI"),
which had not been merged by the time I created my patch as an
alternative workaround.
Arnd