On 7/3/25 09:48, Benjamin Tissoires wrote:
Hi Derek,
[I'll answer to this email with a very high level overview of it, as I'm
not sure I'll have time to dig much deeper in 6/6 today.]
I'll touch on my two patches at the front of the series and let Derek
get to the questions/comments on the later ones.
On Jul 02 2025, Derek J. Clark wrote:
This series adds initial support for the Legion Go S's built-in
controller HID configuration interface. In the first patch a new HID
uevent property is added, HID_FIRMWARE_VERSION, so as to permit fwupd
to read the firmware version of the HID interface without detaching the
kernel driver.
That immediately raise red flags on my side. HID_FIRMWARE_VERSION will
likely be used only for this new driver, and that means a special case
in each and every client.
Actually Richard and I had envisioned that all updatable HID devices
would start exporting their firmware version through this HID property.
lenovo-legos-hid was just the first.
The idea would then be that userspace software like fwupd would know to
parse this property to show the current version and never need to
interrogate the device directly unless it was actually being updated.
We had to deal with firmware versions in the past in the HID drivers,
and we ended up relying on the uniq field of the hid_device (because the
serial+firmware version uniquely identify the device).
I think this is a different case. We don't care so much about the
unique identification of the device as much as we care about the stream
of firmware applied to the device.
If HID_UNIQ is the right way to get the firmware version but some
drivers encode a serial+firmware and others encode firmware that's going
to make for a very messy "generic" property to read the firmware version
of a device.
The second patch adds the ability for an hid_driver to
assign new/arbitrary uevent properties for static data that doesn't
benefit from having a sysfs entry.
That, in my mind, is even worse (for the reasons above).
Do clients actually need to know about all the properties? My thought
was that if a client encounters a property it doesn't care about it can
just ignore it.
If that's misplaced; what would you prefer for all this static
information? A pile of sysfs files?
The third patch adds the VID and PID
for the Lenovo Legion Go S MCU.
Which shouldn't be in its own patch, but part of the driver initial
patch.
The fourth patch adds ABI documentation
for the config interface introduced in the final patch. The fifth patch
introduces the core lenovo-legos-hid driver which acts as a routing
interface for the different endpoints.
That "core" patch is IMO useless. All it does is:
- check for the USB endpoint (but in the wrong way, because if you
insert a device through uhid with the same PID/VID it will crash)
- replace the HID-core core functions with the same code
Really, this should be squashed into the next patch (with 3/6 then).
Also, why adding a new subdirectory? All the hid drivers are flat in the
drivers/hid/ directory, and the subdirs are for transport layers. There
is one exception for the surface driver but I don't see why you need
such an exception (yeah, the code is big, but what's the difference in
having a 1500 lines of code source in its own subdir vs at the root?)
The sixth path introduces the
config lenovo-legos-hid driver wich uses both the HID_FIRMWARE_VERSION
as well as arbitrary uevent properties. Additional interfaces and config
properties are planned to be added in a future series.
That one is too big for my liking. Generally speaking, a commit
descrition which says "this does this and that" can be split into 2
patches at least :)
What kind of future interfaces and config properties are you planning?
Patch 6 introduces a checkpatch WARNING that I'm unable to resolve:
WARNING: ENOSYS means 'invalid syscall nr' and nothing else
1292: FILE: drivers/hid/lenovo-legos-hid/lenovo-legos-hid-config.c:1085:
+ case -ENOSYS: /* during rmmod -ENOSYS is expected */
We can losely waive those while merging. We do it quite often actually.
But trying to minimize checkpatch warnings is a good thing, so thanks
for that.
This error handling case was added as it is experienced in the real world
when the driver is rmmod. The LED subsystem produces this error code in
its legacy code and this is not a new novel use of -ENOSYS, we are simply
catching the case to avoid spurious errors in dmesg when the driver is
removed. If there is a way to prevent this error from being triggered by
checkpatch in the first place, that would be an ideal remedy, but I'm not
aware how that can be done at this time.
Again, nothing to worry about.
Cheers,
Benjamin
Signed-off-by: Derek J. Clark <derekjohn.clark@xxxxxxxxx>
Derek J. Clark (4):
HID: Add Legion Go S ID's
HID: Add documentation for lenovo-legos-hid driver
HID: Add lenovo-legos-hid core
HID: Add lenovo-legos-hid configuration endpoint interface
Mario Limonciello (2):
HID: Include firmware version in the uevent
HID: Allow HID drivers to add more uevent variables
.../ABI/testing/sysfs-driver-lenovo-legos-hid | 269 +++
MAINTAINERS | 7 +
drivers/hid/Kconfig | 2 +
drivers/hid/Makefile | 2 +
drivers/hid/hid-core.c | 11 +
drivers/hid/hid-ids.h | 4 +
drivers/hid/lenovo-legos-hid/Kconfig | 11 +
drivers/hid/lenovo-legos-hid/Makefile | 6 +
drivers/hid/lenovo-legos-hid/config.c | 1518 +++++++++++++++++
drivers/hid/lenovo-legos-hid/config.h | 19 +
drivers/hid/lenovo-legos-hid/core.c | 122 ++
drivers/hid/lenovo-legos-hid/core.h | 25 +
include/linux/hid.h | 2 +
13 files changed, 1998 insertions(+)
create mode 100644 Documentation/ABI/testing/sysfs-driver-lenovo-legos-hid
create mode 100644 drivers/hid/lenovo-legos-hid/Kconfig
create mode 100644 drivers/hid/lenovo-legos-hid/Makefile
create mode 100644 drivers/hid/lenovo-legos-hid/config.c
create mode 100644 drivers/hid/lenovo-legos-hid/config.h
create mode 100644 drivers/hid/lenovo-legos-hid/core.c
create mode 100644 drivers/hid/lenovo-legos-hid/core.h
--
2.50.0
