tree: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
head: 4a0225c3d208cfa6e4550f2210ffd9114a952a81
commit: 8d4ba1be3d2257606e04aff412829d8972670750 ASoC: SOF: pci: split PCI into different drivers
date: 8 weeks ago
config: i386-randconfig-r024-20210426 (attached as .config)
compiler: gcc-9 (Debian 9.3.0-22) 9.3.0
reproduce (this is a W=1 build):
# https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=8d4ba1be3d2257606e04aff412829d8972670750
git remote add linus https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
git fetch --no-tags linus master
git checkout 8d4ba1be3d2257606e04aff412829d8972670750
# save the attached .config to linux build tree
make W=1 W=1 ARCH=i386
If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <[email protected]>
All errors (new ones prefixed by >>):
ld: sound/soc/sof/intel/byt.o: in function `sof_baytrail_probe':
>> sound/soc/sof/intel/byt.c:1057: undefined reference to `sof_acpi_probe'
>> ld: sound/soc/sof/intel/byt.o:(.data+0x44): undefined reference to `sof_acpi_remove'
>> ld: sound/soc/sof/intel/byt.o:(.data+0x94): undefined reference to `sof_acpi_pm'
vim +1057 sound/soc/sof/intel/byt.c
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1032
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1033 static int sof_baytrail_probe(struct platform_device *pdev)
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1034 {
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1035 struct device *dev = &pdev->dev;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1036 const struct sof_dev_desc *desc;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1037 const struct acpi_device_id *id;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1038 int ret;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1039
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1040 id = acpi_match_device(dev->driver->acpi_match_table, dev);
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1041 if (!id)
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1042 return -ENODEV;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1043
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1044 ret = snd_intel_acpi_dsp_driver_probe(dev, id->id);
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1045 if (ret != SND_INTEL_DSP_DRIVER_ANY && ret != SND_INTEL_DSP_DRIVER_SOF) {
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1046 dev_dbg(dev, "SOF ACPI driver not selected, aborting probe\n");
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1047 return -ENODEV;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1048 }
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1049
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1050 desc = device_get_match_data(&pdev->dev);
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1051 if (!desc)
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1052 return -ENODEV;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1053
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1054 if (desc == &sof_acpi_baytrail_desc && soc_intel_is_byt_cr(pdev))
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1055 desc = &sof_acpi_baytrailcr_desc;
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1056
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 @1057 return sof_acpi_probe(pdev, desc);
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1058 }
8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1059
:::::: The code at line 1057 was first introduced by commit
:::::: 8a49cd11e68ed0e6a687de04d25c06553bf96b0c ASoC: SOF: ACPI: avoid reverse module dependency
:::::: TO: Arnd Bergmann <[email protected]>
:::::: CC: Takashi Iwai <[email protected]>
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/[email protected]
On 4/27/21 6:10 AM, kernel test robot wrote:
> tree: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> head: 4a0225c3d208cfa6e4550f2210ffd9114a952a81
> commit: 8d4ba1be3d2257606e04aff412829d8972670750 ASoC: SOF: pci: split PCI into different drivers
> date: 8 weeks ago
> config: i386-randconfig-r024-20210426 (attached as .config)
> compiler: gcc-9 (Debian 9.3.0-22) 9.3.0
> reproduce (this is a W=1 build):
> # https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=8d4ba1be3d2257606e04aff412829d8972670750
> git remote add linus https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
> git fetch --no-tags linus master
> git checkout 8d4ba1be3d2257606e04aff412829d8972670750
> # save the attached .config to linux build tree
> make W=1 W=1 ARCH=i386
>
> If you fix the issue, kindly add following tag as appropriate
> Reported-by: kernel test robot <[email protected]>
>
> All errors (new ones prefixed by >>):
>
> ld: sound/soc/sof/intel/byt.o: in function `sof_baytrail_probe':
>>> sound/soc/sof/intel/byt.c:1057: undefined reference to `sof_acpi_probe'
>>> ld: sound/soc/sof/intel/byt.o:(.data+0x44): undefined reference to `sof_acpi_remove'
>>> ld: sound/soc/sof/intel/byt.o:(.data+0x94): undefined reference to `sof_acpi_pm'
Ack, this is again the partition between Baytrail and Merrifield that's
problematic in the corner case where ACPI is compiled as module and PCI
as built-in. It's a bit academic since the PCI probe wouldn't work, but
still it's not well partitioned.
I've been wanting to change this for a while, and split the common atom
parts from the ACPI/PCI parts. I guess it's time to fix this for good.
> vim +1057 sound/soc/sof/intel/byt.c
>
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1032
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1033 static int sof_baytrail_probe(struct platform_device *pdev)
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1034 {
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1035 struct device *dev = &pdev->dev;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1036 const struct sof_dev_desc *desc;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1037 const struct acpi_device_id *id;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1038 int ret;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1039
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1040 id = acpi_match_device(dev->driver->acpi_match_table, dev);
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1041 if (!id)
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1042 return -ENODEV;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1043
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1044 ret = snd_intel_acpi_dsp_driver_probe(dev, id->id);
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1045 if (ret != SND_INTEL_DSP_DRIVER_ANY && ret != SND_INTEL_DSP_DRIVER_SOF) {
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1046 dev_dbg(dev, "SOF ACPI driver not selected, aborting probe\n");
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1047 return -ENODEV;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1048 }
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1049
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1050 desc = device_get_match_data(&pdev->dev);
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1051 if (!desc)
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1052 return -ENODEV;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1053
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1054 if (desc == &sof_acpi_baytrail_desc && soc_intel_is_byt_cr(pdev))
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1055 desc = &sof_acpi_baytrailcr_desc;
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1056
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 @1057 return sof_acpi_probe(pdev, desc);
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1058 }
> 8a49cd11e68ed0 Arnd Bergmann 2021-03-01 1059
>
> :::::: The code at line 1057 was first introduced by commit
> :::::: 8a49cd11e68ed0e6a687de04d25c06553bf96b0c ASoC: SOF: ACPI: avoid reverse module dependency
>
> :::::: TO: Arnd Bergmann <[email protected]>
> :::::: CC: Takashi Iwai <[email protected]>
>
> ---
> 0-DAY CI Kernel Test Service, Intel Corporation
> https://lists.01.org/hyperkitty/list/[email protected]
>
On Tue, Apr 27, 2021 at 6:07 PM Pierre-Louis Bossart
<[email protected]> wrote:
> On 4/27/21 6:10 AM, kernel test robot wrote:
> > tree: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> > head: 4a0225c3d208cfa6e4550f2210ffd9114a952a81
> > commit: 8d4ba1be3d2257606e04aff412829d8972670750 ASoC: SOF: pci: split PCI into different drivers
> > date: 8 weeks ago
> > config: i386-randconfig-r024-20210426 (attached as .config)
> > compiler: gcc-9 (Debian 9.3.0-22) 9.3.0
> > reproduce (this is a W=1 build):
> > # https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=8d4ba1be3d2257606e04aff412829d8972670750
> > git remote add linus https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
> > git fetch --no-tags linus master
> > git checkout 8d4ba1be3d2257606e04aff412829d8972670750
> > # save the attached .config to linux build tree
> > make W=1 W=1 ARCH=i386
> >
> > If you fix the issue, kindly add following tag as appropriate
> > Reported-by: kernel test robot <[email protected]>
> >
> > All errors (new ones prefixed by >>):
> >
> > ld: sound/soc/sof/intel/byt.o: in function `sof_baytrail_probe':
> >>> sound/soc/sof/intel/byt.c:1057: undefined reference to `sof_acpi_probe'
> >>> ld: sound/soc/sof/intel/byt.o:(.data+0x44): undefined reference to `sof_acpi_remove'
> >>> ld: sound/soc/sof/intel/byt.o:(.data+0x94): undefined reference to `sof_acpi_pm'
>
> Ack, this is again the partition between Baytrail and Merrifield that's
> problematic in the corner case where ACPI is compiled as module and PCI
> as built-in. It's a bit academic since the PCI probe wouldn't work, but
> still it's not well partitioned.
>
> I've been wanting to change this for a while, and split the common atom
> parts from the ACPI/PCI parts. I guess it's time to fix this for good.
I wonder why I never see these in randconfig builds here. I can reproduce it
with the .config from the link above though, and I see these relevant config
lines:
CONFIG_SND_SOC_SOF_INTEL_TOPLEVEL=y
CONFIG_SND_SOC_SOF_INTEL_ATOM_HIFI_EP=y
CONFIG_SND_SOC_SOF_MERRIFIELD=y
CONFIG_SND_SOC_SOF_ACPI=m
CONFIG_SND_SOC_SOF_ACPI_DEV=m
CONFIG_SND_SOC_SOF_BAYTRAIL=m
An easy workaround would be to move the first 500 or so lines of
sound/soc/sof/intel/byt.c into a separate file and #include that from
distinct baytrail and merrifield modules, at the cost of a bit of
object code duplication and a somewhat dirty hack.
Another option would be a Kconfig hack to prevent the broken
configuration, such as
diff --git a/sound/soc/sof/intel/Kconfig b/sound/soc/sof/intel/Kconfig
index eebe784ff652..b2039b4af9cf 100644
--- a/sound/soc/sof/intel/Kconfig
+++ b/sound/soc/sof/intel/Kconfig
@@ -85,6 +85,8 @@ config SND_SOC_SOF_MERRIFIELD
tristate "SOF support for Tangier/Merrifield"
default SND_SOC_SOF_PCI
select SND_SOC_SOF_INTEL_ATOM_HIFI_EP
+ # Merrifield can't be built-in if snd-sof-acpi-intel-byt.ko is a module
+ depends on SND_SOC_SOF_BAYTRAIL || !SND_SOC_SOF_BAYTRAIL
help
This adds support for Sound Open Firmware for Intel(R) platforms
using the Tangier/Merrifield processors.
Arnd
>>> ld: sound/soc/sof/intel/byt.o: in function `sof_baytrail_probe':
>>>>> sound/soc/sof/intel/byt.c:1057: undefined reference to `sof_acpi_probe'
>>>>> ld: sound/soc/sof/intel/byt.o:(.data+0x44): undefined reference to `sof_acpi_remove'
>>>>> ld: sound/soc/sof/intel/byt.o:(.data+0x94): undefined reference to `sof_acpi_pm'
>>
>> Ack, this is again the partition between Baytrail and Merrifield that's
>> problematic in the corner case where ACPI is compiled as module and PCI
>> as built-in. It's a bit academic since the PCI probe wouldn't work, but
>> still it's not well partitioned.
>>
>> I've been wanting to change this for a while, and split the common atom
>> parts from the ACPI/PCI parts. I guess it's time to fix this for good.
>
> I wonder why I never see these in randconfig builds here. I can reproduce it
> with the .config from the link above though, and I see these relevant config
> lines:
>
> CONFIG_SND_SOC_SOF_INTEL_TOPLEVEL=y
> CONFIG_SND_SOC_SOF_INTEL_ATOM_HIFI_EP=y
> CONFIG_SND_SOC_SOF_MERRIFIELD=y
> CONFIG_SND_SOC_SOF_ACPI=m
> CONFIG_SND_SOC_SOF_ACPI_DEV=m
> CONFIG_SND_SOC_SOF_BAYTRAIL=m
yes, we end-up with a bad configuration with BAYTRAIL as m and
Merrifield as y. That's broken since they share the same file.
> An easy workaround would be to move the first 500 or so lines of
> sound/soc/sof/intel/byt.c into a separate file and #include that from
> distinct baytrail and merrifield modules, at the cost of a bit of
> object code duplication and a somewhat dirty hack.
I have a fix being tested here:
https://github.com/thesofproject/linux/pull/2874
The idea is what you suggested, move the common parts to a new file but
I used an 'atom' module instead #include to avoid duplication. The
exported functions are namespaced to avoid polluting the global symbol map.
The only other change needed is in the Makefile, We already have
CONFIG_SND_SOC_SOF_INTEL_ATOM_HIFI_EP in the Kconfig and that can be
used for this module.
> Another option would be a Kconfig hack to prevent the broken
> configuration, such as
>
> diff --git a/sound/soc/sof/intel/Kconfig b/sound/soc/sof/intel/Kconfig
> index eebe784ff652..b2039b4af9cf 100644
> --- a/sound/soc/sof/intel/Kconfig
> +++ b/sound/soc/sof/intel/Kconfig
> @@ -85,6 +85,8 @@ config SND_SOC_SOF_MERRIFIELD
> tristate "SOF support for Tangier/Merrifield"
> default SND_SOC_SOF_PCI
> select SND_SOC_SOF_INTEL_ATOM_HIFI_EP
> + # Merrifield can't be built-in if snd-sof-acpi-intel-byt.ko is a module
> + depends on SND_SOC_SOF_BAYTRAIL || !SND_SOC_SOF_BAYTRAIL
> help
> This adds support for Sound Open Firmware for Intel(R) platforms
> using the Tangier/Merrifield processors.
It'd rather do this the right way with a proper code partitioning, I was
tempted to do it multiple times and I've had it with this code.
On Tue, Apr 27, 2021 at 9:13 PM Pierre-Louis Bossart
<[email protected]> wrote:
> > An easy workaround would be to move the first 500 or so lines of
> > sound/soc/sof/intel/byt.c into a separate file and #include that from
> > distinct baytrail and merrifield modules, at the cost of a bit of
> > object code duplication and a somewhat dirty hack.
>
> I have a fix being tested here:
> https://github.com/thesofproject/linux/pull/2874
>
> The idea is what you suggested, move the common parts to a new file but
> I used an 'atom' module instead #include to avoid duplication. The
> exported functions are namespaced to avoid polluting the global symbol map.
>
> The only other change needed is in the Makefile, We already have
> CONFIG_SND_SOC_SOF_INTEL_ATOM_HIFI_EP in the Kconfig and that can be
> used for this module.
Makes sense. I was slightly worried this would require exporting a ton
of symbols and possibly circular dependencies but looking at your patch,
it seems entirely fine.
Arnd