Skip to contentSkip to conversation

spmi: prefix spmi bus device names with "spmi"

PATCH RESEND In review David Collins 3 participants 10 messages Sep 22, 2020 – Oct 14, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/1 spmi: prefix spmi bus device names with "spmi"

Change the format of spmi bus device names from:
<spmi_bus_number>-<spmi_device_sid>
Example: 0-01
to this:
spmi<spmi_bus_number>-<spmi_device_sid>
Example: spmi0-01

This helps to disambiguate SPMI device regmaps from I2C ones at /sys/kernel/debug/regmap since I2C devices use a very similar naming scheme: 0-0000.

---
 drivers/spmi/spmi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/spmi/spmi.c +2 -2
index c16b60f..ec94439 100644
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0-only
 /*
- * Copyright (c) 2012-2015, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2012-2015, 2020, The Linux Foundation. All rights reserved.
  */
 #include <linux/kernel.h>
 #include <linux/errno.h>
@@ -62,7 +62,7 @@ int spmi_device_add(struct spmi_device *sdev)
 	struct spmi_controller *ctrl = sdev->ctrl;
 	int err;
 
-	dev_set_name(&sdev->dev, "%d-%02x", ctrl->nr, sdev->usid);
+	dev_set_name(&sdev->dev, "spmi%d-%02x", ctrl->nr, sdev->usid);
 
 	err = device_add(&sdev->dev);
 	if (err < 0) {
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
maintainer
#1 lore raw

Quoting David Collins (2020-09-22 15:04:18)

Quoted 10 lines

Change the format of spmi bus device names from:
<spmi_bus_number>-<spmi_device_sid>
Example: 0-01
to this:
spmi<spmi_bus_number>-<spmi_device_sid>
Example: spmi0-01

This helps to disambiguate SPMI device regmaps from I2C ones at /sys/kernel/debug/regmap since I2C devices use a very similar naming scheme: 0-0000.

Can regmap debugfs prepend the bus name on the node made in debugfs? Does it do that already?

Quoted 27 lines
---
 drivers/spmi/spmi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/spmi/spmi.c +2 -2
index c16b60f..ec94439 100644
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0-only
 /*
- * Copyright (c) 2012-2015, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2012-2015, 2020, The Linux Foundation. All rights reserved.
  */
 #include <linux/kernel.h>
 #include <linux/errno.h>
@@ -62,7 +62,7 @@ int spmi_device_add(struct spmi_device *sdev)
        struct spmi_controller *ctrl = sdev->ctrl;
        int err;
 
-       dev_set_name(&sdev->dev, "%d-%02x", ctrl->nr, sdev->usid);
+       dev_set_name(&sdev->dev, "spmi%d-%02x", ctrl->nr, sdev->usid);
 
        err = device_add(&sdev->dev);
        if (err < 0) {
#2 lore raw
Quoted 8 lines · Stephen Boyd

Quoting David Collins (2020-09-22 15:04:18)

Quoted 3 lines

This helps to disambiguate SPMI device regmaps from I2C ones at /sys/kernel/debug/regmap since I2C devices use a very similar naming scheme: 0-0000.

Can regmap debugfs prepend the bus name on the node made in debugfs? Does it do that already?

It doesn't do that. I have to say that given the use of dev_name() in logging it does feel like it'd be useful to have distinct names for grepping if we're running into collisions, IIRC the reason I went with dev_name() was that it's a commonly used human readable handle for diagnostic infrastrucuture so it makes it easier to follow things around.

maintainer
#3 lore raw

Quoting Mark Brown (2020-10-01 10:43:26)

Quoted 7 lines · Stephen Boyd
Quoted 5 lines

Quoting David Collins (2020-09-22 15:04:18)

Quoted 3 lines

This helps to disambiguate SPMI device regmaps from I2C ones at /sys/kernel/debug/regmap since I2C devices use a very similar naming scheme: 0-0000.

Quoted 8 lines
Quoted 2 lines

Can regmap debugfs prepend the bus name on the node made in debugfs? Does it do that already?

It doesn't do that. I have to say that given the use of dev_name() in logging it does feel like it'd be useful to have distinct names for grepping if we're running into collisions, IIRC the reason I went with dev_name() was that it's a commonly used human readable handle for diagnostic infrastrucuture so it makes it easier to follow things around.

To me the dev_name() usage seems fine. Maybe David has some real reason to change this though?

In general I don't think userspace cares what the SPMI device name is, i.e. the device name isn't used for dev nodes because SPMI doesn't support ioctls or read/write APIs on the bus. That could be a nice feature addition though, to support something like i2c-dev.

Changing it so that regmap debugfs is less likely to collide looks weird. It doesn't actually collide anyway, so it seems like we're adding spmi prefix to make it easier to find in debugfs?

#4 lore raw
Quoted 20 lines · Stephen Boyd

Quoting Mark Brown (2020-10-01 10:43:26)

Quoted 7 lines · Stephen Boyd
Quoted 5 lines

Quoting David Collins (2020-09-22 15:04:18)

Quoted 3 lines

This helps to disambiguate SPMI device regmaps from I2C ones at /sys/kernel/debug/regmap since I2C devices use a very similar naming scheme: 0-0000.

Quoted 8 lines
Quoted 2 lines

Can regmap debugfs prepend the bus name on the node made in debugfs? Does it do that already?

It doesn't do that. I have to say that given the use of dev_name() in logging it does feel like it'd be useful to have distinct names for grepping if we're running into collisions, IIRC the reason I went with dev_name() was that it's a commonly used human readable handle for diagnostic infrastrucuture so it makes it easier to follow things around.

To me the dev_name() usage seems fine. Maybe David has some real reason to change this though?

Quoted 8 lines

In general I don't think userspace cares what the SPMI device name is, i.e. the device name isn't used for dev nodes because SPMI doesn't support ioctls or read/write APIs on the bus. That could be a nice feature addition though, to support something like i2c-dev.

Changing it so that regmap debugfs is less likely to collide looks weird. It doesn't actually collide anyway, so it seems like we're adding spmi prefix to make it easier to find in debugfs?

Yes, that is correct. There isn't a collision since I2C uses 0-0000 and SPMI uses 0-00 naming scheme. However, those names are very similar and it is hard for a user to tell which is which inside /sys/kernel/debug/regmap without a deep understanding of the I2C and SPMI code.

The SPMI regmap debugfs files are used extensively for testing and debug purposes internally at Qualcomm and by our customers. It would be helpful if the more verbose naming scheme were accepted upstream to avoid confusion and broken test scripts.

Thanks,
David

Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
#5 lore raw
Quoted 4 lines · David Collins

The SPMI regmap debugfs files are used extensively for testing and debug purposes internally at Qualcomm and by our customers. It would be helpful if the more verbose naming scheme were accepted upstream to avoid confusion and broken test scripts.

...and doing this in the dev_name() should help other diagnostic users (like dev_printk() for example).

maintainer
#6 lore raw

Quoting Mark Brown (2020-10-02 09:03:24)

Quoted 9 lines · David Collins
Quoted 5 lines

The SPMI regmap debugfs files are used extensively for testing and debug purposes internally at Qualcomm and by our customers. It would be helpful if the more verbose naming scheme were accepted upstream to avoid confusion and broken test scripts.

...and doing this in the dev_name() should help other diagnostic users (like dev_printk() for example).

Don't thinks like dev_printk() prefix the bus name? See dev_driver_string()? So I agree that having the bus name is useful, but confused why there are testing scripts and things on top of regmap debugfs

Put another way, why not introduce something similar to i2c-dev where userspace can read/write registers for devices on the SPMI bus? Otherwise I presume the test scripts inside Qualcomm are just reading registers out of regmap?

#7 lore raw
Quoted 9 lines · Stephen Boyd

Quoting Mark Brown (2020-10-02 09:03:24)

Quoted 2 lines

...and doing this in the dev_name() should help other diagnostic users (like dev_printk() for example).

Don't thinks like dev_printk() prefix the bus name? See dev_driver_string()? So I agree that having the bus name is useful, but confused why there are testing scripts and things on top of regmap debugfs

Not that I've ever noticed, eg on the console.

Quoted 4 lines

Put another way, why not introduce something similar to i2c-dev where userspace can read/write registers for devices on the SPMI bus? Otherwise I presume the test scripts inside Qualcomm are just reading registers out of regmap?

I know some other vendors use the regmap debugfs for their diagnostic tools (obviously not with SPMI). It's generally so they can get the benefit of the cache, it's a combination of allowing the state to be inspected while the driver has the device powered down and for devices on slower buses being much more performant.

maintainer
#8 lore raw

Quoting Mark Brown (2020-10-02 11:04:30)

Quoted 6 lines · Stephen Boyd
Quoted 4 lines

Quoting Mark Brown (2020-10-02 09:03:24)

Quoted 2 lines

...and doing this in the dev_name() should help other diagnostic users (like dev_printk() for example).

Quoted 6 lines
Quoted 4 lines

Don't thinks like dev_printk() prefix the bus name? See dev_driver_string()? So I agree that having the bus name is useful, but confused why there are testing scripts and things on top of regmap debugfs

Not that I've ever noticed, eg on the console.

I see things like this on my console:

[    1.684617] spmi spmi-0: PMIC arbiter version v5 (0x50000000)

and 'spmi' is the bus name I'm thinking about. But I think that's because there isn't a driver attached. Nothing prints for the 0-00 device by default, so I enabled the debug print for it and I see

[    1.693280] pmic-spmi 0-00: 28: unknown v2.0

Anyway, the device name was written to follow i2c as far as I can tell.

If scripts, i.e. computers, have a hard time figuring out the name of the device then fix the script?

#9 lore raw
Quoted 8 lines · Stephen Boyd

I see things like this on my console:

[    1.684617] spmi spmi-0: PMIC arbiter version v5 (0x50000000)

and 'spmi' is the bus name I'm thinking about. But I think that's because there isn't a driver attached. Nothing prints for the 0-00 device by default, so I enabled the debug print for it and I see

Quoted 6 lines
[    1.693280] pmic-spmi 0-00: 28: unknown v2.0

Anyway, the device name was written to follow i2c as far as I can tell.

If scripts, i.e. computers, have a hard time figuring out the name of the device then fix the script?

I agree that we can drop this patch. There is no technical requirement for the spmi device naming scheme to be changed. We will update our downstream test scripts to use the upstream naming scheme and also socialize the naming difference internally.

Take care,
David

Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project