On 10/24/19 6:29 AM, Vinod Koul wrote:

On 22-10-19, 18:48, Pierre-Louis Bossart wrote:

...

There is no good reason why the unique_id needs to be stored as 4 bits. The code will work without changes with a u8 since all values

Well this was due to the fact the slave id defined by MIPI has unique id as 4 bits. In fact if you look closely there are other fields in sdw_slave_id doing this

it's not because we extract 4 bits that we need to store the information in 4 bits.

...

are already filtered while parsing the ACPI tables and Slave devID registers.

Use u8 representation. This will allow us to encode a "IGNORE_UNIQUE_ID" value to account for firmware/BIOS creativity.

Why are we shoving firmware/BIOS issues into the core?

The core uses a matching formula which is too strict and does not work on multiple platforms.

You can argue that the BIOS should be fixed, but the counter argument is that the practice of ignoring the unique ID is allowed by the MIPI standard.

...

Signed-off-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com

include/linux/soundwire/sdw.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/include/linux/soundwire/sdw.h b/include/linux/soundwire/sdw.h index 688b40e65c89..28745b9ba279 100644 --- a/include/linux/soundwire/sdw.h +++ b/include/linux/soundwire/sdw.h @@ -403,6 +403,8 @@ int sdw_slave_read_prop(struct sdw_slave *slave);

• SDW Slave Structures and APIs

*/

+#define SDW_IGNORED_UNIQUE_ID 0xFF

• /**
  • struct sdw_slave_id - Slave ID
  • @mfg_id: MIPI Manufacturer ID

@@ -418,7 +420,7 @@ struct sdw_slave_id { __u16 mfg_id; __u16 part_id; __u8 class_id;

• __u8 unique_id:4;

• __u8 unique_id; __u8 sdw_version:4; };

-- 2.20.1

On 10/24/19 6:39 AM, Vinod Koul wrote:

On 22-10-19, 18:48, Pierre-Louis Bossart wrote:

...

The uniqueID is useful when there are two or more devices of the same type (identical manufacturer ID, part ID) on the same link.

Right!

the key part is "two or more". When it's "one" then the uniqueID has no defined meaning.

...

When there is a single device of a given type on a link, its uniqueID is irrelevant. It's not uncommon on actual platforms to see variations of the uniqueID, or differences between devID registers and ACPI _ADR fields.

Ideally this should be fixed in firmware, I do not like the fact the we are poking in core for firmware issues!

I will be the first to complain about BIOS issues, and the need for workarounds in the kernel, but here the BIOS vendors rely on permissions defined in the standard, see lines 3320..31 in the MIPI SoundWire 1.2 document. You will see that there is no requirement to use the full set of devID registers to identify a Slave device. The only requirement is to read devID0 first to force a state machine transition and enable arbitration.

In other words, it's a nice case of BIOS folks telling the kernel folks we are too strict in our interpretation of the standard, and what they do is a feature and not a bug.

...

This patch suggests a filter on startup to identify 'single' devices and tag them accordingly.

So you try to see if the board has a single device and mark them so that you can skip the unique id, did I get that right?

What about the boards which have multiple devices? How doing solve these?

The uniqueID is used when multiple devices of the same type are detected in ACPI tables. No change, see [1] below

...

The uniqueID is then not used for the probe, and the device name omits the uniqueID as well.

Signed-off-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com

drivers/soundwire/bus.c | 7 +++--- drivers/soundwire/slave.c | 52 ++++++++++++++++++++++++++++++++++++--- 2 files changed, 52 insertions(+), 7 deletions(-)

diff --git a/drivers/soundwire/bus.c b/drivers/soundwire/bus.c index fc53dbe57f85..be5d437058ed 100644 --- a/drivers/soundwire/bus.c +++ b/drivers/soundwire/bus.c @@ -422,10 +422,11 @@ static struct sdw_slave *sdw_get_slave(struct sdw_bus *bus, int i)

static int sdw_compare_devid(struct sdw_slave *slave, struct sdw_slave_id id) {

• if (slave->id.unique_id != id.unique_id ||

•    slave->id.mfg_id != id.mfg_id ||
  

• if (slave->id.mfg_id != id.mfg_id || slave->id.part_id != id.part_id ||

•    slave->id.class_id != id.class_id)
  

•    slave->id.class_id != id.class_id ||
  

•    (slave->id.unique_id != SDW_IGNORED_UNIQUE_ID &&
  

•     slave->id.unique_id != id.unique_id))
  

[1] this is where the unique_id is ignored if it was tagged as irrelevant while scanning ACPI tables. If it is not ignored, then the same comparison applied

...

return -ENODEV;

return 0; diff --git a/drivers/soundwire/slave.c b/drivers/soundwire/slave.c index 5dbc76772d21..19919975bb6d 100644 --- a/drivers/soundwire/slave.c +++ b/drivers/soundwire/slave.c @@ -29,10 +29,17 @@ static int sdw_slave_add(struct sdw_bus *bus, slave->dev.parent = bus->dev; slave->dev.fwnode = fwnode;

• /* name shall be sdw:link:mfg:part:class:unique */
• dev_set_name(&slave->dev, "sdw:%x:%x:%x:%x:%x",

•      bus->link_id, id->mfg_id, id->part_id,
  

•      id->class_id, id->unique_id);
  

• if (id->unique_id == SDW_IGNORED_UNIQUE_ID) {

• /* name shall be sdw:link:mfg:part:class */
  

• dev_set_name(&slave->dev, "sdw:%x:%x:%x:%x",
  

• 	     bus->link_id, id->mfg_id, id->part_id,
  

• 	     id->class_id);
  

• } else {

• /* name shall be sdw:link:mfg:part:class:unique */
  

• dev_set_name(&slave->dev, "sdw:%x:%x:%x:%x:%x",
  

• 	     bus->link_id, id->mfg_id, id->part_id,
  

• 	     id->class_id, id->unique_id);
  

• }

  slave->dev.release = sdw_slave_release; slave->dev.bus = &sdw_bus_type;

@@ -103,6 +110,7 @@ static bool find_slave(struct sdw_bus *bus, int sdw_acpi_find_slaves(struct sdw_bus *bus) { struct acpi_device *adev, *parent;

• struct acpi_device *adev2, *parent2;

  parent = ACPI_COMPANION(bus->dev); if (!parent) {

@@ -112,10 +120,46 @@ int sdw_acpi_find_slaves(struct sdw_bus *bus)

list_for_each_entry(adev, &parent->children, node) { struct sdw_slave_id id;

• struct sdw_slave_id id2;
  

• bool ignore_unique_id = true;
  
  

  if (!find_slave(bus, adev, &id)) continue;

• /* brute-force O(N^2) search for duplicates */
  

• parent2 = parent;
  

• list_for_each_entry(adev2, &parent2->children, node) {
  

• 	if (adev == adev2)
  

• 		continue;
  

• 	if (!find_slave(bus, adev2, &id2))
  

• 		continue;
  

• 	if (id.sdw_version != id2.sdw_version ||
  

• 	    id.mfg_id != id2.mfg_id ||
  

• 	    id.part_id != id2.part_id ||
  

• 	    id.class_id != id2.class_id)
  

• 		continue;
  

• 	if (id.unique_id != id2.unique_id) {
  

• 		dev_dbg(bus->dev,
  

• 			"Valid unique IDs %x %x for Slave mfg %x part %d\n",
  

• 			id.unique_id, id2.unique_id,
  

• 			id.mfg_id, id.part_id);
  

• 		ignore_unique_id = false;
  

• 	} else {
  

• 		dev_err(bus->dev,
  

• 			"Invalid unique IDs %x %x for Slave mfg %x part %d\n",
  

• 			id.unique_id, id2.unique_id,
  

• 			id.mfg_id, id.part_id);
  

• 		return -ENODEV;
  

• 	}
  

• }
  

• if (ignore_unique_id)
  

• 	id.unique_id = SDW_IGNORED_UNIQUE_ID;
  

• /*

  • don't error check for sdw_slave_add as we want to continue
  • adding Slaves

-- 2.20.1