[RFC PATCH] Unifiy ACPI & DT handling inside SOF #1315
Conversation
dbaluta
requested review from
RanderWang,
bardliao,
juimonen,
keyonjie,
lgirdwood,
libinyang,
lyakh,
plbossart and
ranj063
as code owners
dbaluta
force-pushed
the
topic/sof-dev
branch
from
8a46d15 to
5f639fe
Compare
SOF probes machine driver by creating its associated device. On Intel platforms the machine driver is probed by creating a platform device and the information describing the machine driver is syntethically generated via tabels of snd_soc_acpi_mach structures. On ARM platforms (specifically on i.MX) the machine driver can be described via a DT node. In both cases the operation of probing the machine driver is implemented via generic function snd_sof_machine_register. This function calls into device specific machine_register function. This patch does the following: (1) Introduce snd_sof_machine_register (2) Replace current creation of machine driver by snd_sof_machine_register (3) Introduce a generic function that creates a platform_device and use it for Intel platforms (4) Introduce imx8_machine_register which expects that the machine driver for i.MX platforms is described inside SOF DSP node. Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
snd_soc_fw_mach will be the generic machine driver descriptor. It contains the type of the machine driver (e.g ACPI/DT) and a pointer to the specific machine driver acpi/of. On i.MX platforms machine driver description will be read from Device Tree, so we don't actually need nothing more than a pointer to the corresponding device_node. We will use it to read any machine driver property. Because the nocodec implementation is generic enough we can keep the existing ACPI approach even for i.MX8. Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
dbaluta
force-pushed
the
topic/sof-dev
branch
from
5f639fe to
7e4eae1
Compare
plbossart
left a comment
plbossart
left a comment
There was a problem hiding this comment.
No real objections, but @ranj063 's work on MFD is going to change the way we deal with machine drivers, i.e. we may have more that one machine driver loaded for the same ACPI/PCI ID.
In short, we should probably pause a bit here and avoid going too fast since we will have conflicting changes going in at the same time
|
|
||
| /* machine driver probing */ | ||
| int (*machine_register)(struct snd_sof_dev *sof_dev); /* mandatory */ | ||
|
|
There was a problem hiding this comment.
I think yes, although I haven't seen any platform which doesn't have a "codec" scenario.
|
|
||
| static int imx8_machine_register(struct snd_sof_dev *sdev) | ||
| { | ||
| return devm_of_platform_populate(sdev->dev); |
There was a problem hiding this comment.
The SOF DSP DT will look like this:
dsp {
compatible = "fsl,imx8qxp-dsp";
reg = <0x596e8000 0x88000>;
clocks = <..>
mach_drv: mach_drv {
compatible = "simple-scu-audio-card";
simple-audio-card,widgets =<..>
simple-audio-card,routing =<..>
simple-audio-card,dai-link {
format = "i2s";
codec {
sound-dai = <&cs42888>;
};
};
};
devm_of_platform_populate - walks the device tree and creates devices from nodes.
Now, it starts from DSP root node and it will find mach_drv child node. It reads the node and creates a platform_device from it. It is similar with platform_device_register_data but it reads the information from DT instead of having them passed via pdata argument of platform_device_register_data.
|
@plbossart sure lets wait for @ranj063 PR to get in and then I will rebase my PR. |
2 participants
This is just to get your opinion on two matters regarding having a generic approach for ACPI / DT platforms in SOF.
There are two patches in the series each one tackling a different problem.
Currently SOF probes machine driver by manually creating a platform device. While this will just work fine for DT platforms also, it is not really necessary. We can describe the machine driver inside a DT node and then create a device platform from that node by calling of_platform_populate.
Because Intel uses platform_device_register / ARM uses of_platform_populate we need a function to abstract these two operations. Thus we introduce a new SOF operation: sof_machine_register which each platform will implement with specific code.
snd_sof_pdata should really be generic. For this reason we replace snd_soc_acpi_mach with snd_soc_fw_mach. The latter will be an union containing specific machine descriptors for ACPI / DT.
The code is far from being complete or proper tested but I need to understand if this is the correct direction.
Cc: @groncarolo @paulstelian97 @jlacla