Re: [alsa-devel] [RFC PATCH 2/7] soundwire: add Slave sysfs support

[Pierre-Louis Bossart]

On 5/7/19 12:19 AM, Vinod Koul wrote:
> On 06-05-19, 11:46, Pierre-Louis Bossart wrote:
> > On 5/6/19 11:22 AM, Vinod Koul wrote:
> > > On 06-05-19, 17:19, Greg KH wrote:
> > > > On Mon, May 06, 2019 at 09:42:35AM -0500, Pierre-Louis Bossart wrote:
> > > > > > > +
> > > > > > > +int sdw_sysfs_slave_init(struct sdw_slave *slave)
> > > > > > > +{
> > > > > > > +	struct sdw_slave_sysfs *sysfs;
> > > > > > > +	unsigned int src_dpns, sink_dpns, i, j;
> > > > > > > +	int err;
> > > > > > > +
> > > > > > > +	if (slave->sysfs) {
> > > > > > > +		dev_err(&slave->dev, "SDW Slave sysfs is already initialized\n");
> > > > > > > +		err = -EIO;
> > > > > > > +		goto err_ret;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	sysfs = kzalloc(sizeof(*sysfs), GFP_KERNEL);
> > > > > >
> > > > > > Same question as patch 1, why a new device?
> > > > >
> > > > > yes it's the same open. In this case, the slave devices are defined at a
> > > > > different level so it's also confusing to create a device to represent the
> > > > > slave properties. The code works but I am not sure the initial directions
> > > > > are correct.
> > > >
> > > > You can just make a subdir for your attributes by using the attribute
> > > > group name, if a subdirectory is needed just to keep things a bit more
> > > > organized.
> > >
> > > The key here is 'a subdir' which is not the case here. We did discuss
> > > this in the initial patches for SoundWire which had sysfs :)
> > >
> > > The way MIPI disco spec organized properties, we have dp0 and dpN
> > > properties each of them requires to have a subdir of their own and that
> > > was the reason why I coded it to be creating a device.
> >
> > Vinod, the question was not for dp0 and dpN, it's fine to have
> > subdirectories there, but rather why we need separate devices for the master
> > and slave properties.
>
> Slave does not have a separate device. IIRC the properties for Slave are
> in /sys/bus/soundwire/device/<slave>/...

I am not sure this is correct

ACPI defines the slaves devices under
/sys/bus/acpi/PRP0001, e.g.

/sys/bus/acpi/devices/PRP00001:00/device:17# ls
adr                                 mipi-sdw-dp-5-sink-subproperties
intel-endpoint-descriptor-0         mipi-sdw-dp-6-source-subproperties
intel-endpoint-descriptor-1         mipi-sdw-dp-7-sink-subproperties
mipi-sdw-dp-0-subproperties         mipi-sdw-dp-8-source-subproperties
mipi-sdw-dp-1-sink-subproperties    path
mipi-sdw-dp-1-source-subproperties  physical_node
mipi-sdw-dp-2-sink-subproperties    power
mipi-sdw-dp-2-source-subproperties  subsystem
mipi-sdw-dp-3-sink-subproperties    uevent
mipi-sdw-dp-4-source-subproperties

but the sysfs for slaves is shown as
/sys/bus/acpi/devices/PRP00001:00/int-sdw.0/sdw:0:25d:700:0:0# ls
bank_delay_support  master_count             sink_ports
ch_prep_timeout     mipi_revision            source_ports
clk_stop_mode1      modalias                 src-dp2
clk_stop_timeout    p15_behave               src-dp4
dp0                 paging_support           subsystem
driver              power                    test_mode_capable
firmware_node       reset_behave             uevent
hda_reg             simple_clk_stop_capable  wake_capable
high_PHY_capable    sink-dp1
index_reg           sink-dp3

and in sys/bus/soundwire/devices/sdw:0:25d:700:0:0# ls
bank_delay_support  master_count             sink_ports
ch_prep_timeout     mipi_revision            source_ports
clk_stop_mode1      modalias                 src-dp2
clk_stop_timeout    p15_behave               src-dp4
dp0                 paging_support           subsystem
driver              power                    test_mode_capable
firmware_node       reset_behave             uevent
hda_reg             simple_clk_stop_capable  wake_capable
high_PHY_capable    sink-dp1
index_reg           sink-dp3

So I would think we *do* create a new device for each slave instead of 
using the one that's already exposed by ACPI.

> For master yes we can skip the device creation, it was done for
> consistency sake of having these properties ties into sys/bus/soundwire/
>
> I don't mind if they are shown up in respective device node (PCI/platform
> etc) /sys/bus/foo/device/<>
>
> But for creating subdirectories you would need the new dpX devices.

yes, that's agreed.