On 11.11.2019 00:07, Tod E. Kurt wrote: > Hi Heiner, > > Yes, that was what I was starting from. I was hoping there was now an in-situ way of replacing kernel modules without a full kernel recompile. > > I have tried this patch on Raspian 4.19.81 and Ubuntu 19 kernel 5.0.0-32 with no config changes. I believe it solves the main issue. However, I'm seeing another problem that I can't explain. > > Specifically: > - On reboot, device shows up in /sys/class/hidraw/ > - Opening with hidapi and (mis)using hidapi functions works, but, > - Device is then removed from /sys/class/hidraw/ > > Under what circumstances can a device be removed from /sys/class/hidraw? > > If you'd like to trigger this yourself, run "blink1-tool --list" and see the blink(1) disappear. And yet "blink1-tool" can subsequently still communicate with the blink(1). Which I don't quite understand yet either. > Once a dedicated device driver (like hid-led) is loaded, the raw device disappears. I compiled blink1-tool on latest linux-next with the hid quirk patch I sent, but the blink(1) never disappears. According to strace the tool doesn't use the /sys/class/hidraw interface but a low-level USB interface via libusb. > Cheers, > -=Tod > >> On Nov 10, 2019, at 9:25 a, Heiner Kallweit <hkallweit1@xxxxxxxxx> wrote: >> >> On 10.11.2019 17:59, Tod E. Kurt wrote: >>> Hi Heiner, (and others on the list) >>> >>> Thanks for your quick and detailed response. >>> >>> I am endeavoring to try the patch out now. But I've not built a desktop Linux kernel or kernel modules in about 15 years. Do you have a preferred recipe for applying this patch to an existing distro? Since 'hid-quirks' isn't a module, this means recompiling the entire HID driver, correct? I mostly test against various modern Ubuntu or Raspian flavors of Debian, if that matters. >>> >> See e.g. here: https://wiki.ubuntu.com/Kernel/BuildYourOwnKernel >> >>> Also, is there a mitigation for users running production distros that doesn't involve a recompile? >>> >> I don't think so (at least from kernel perspective). What you could do is adding an integration with >> the /sys/class/led interface in your library (if it should be suitable for your use cases) and >> auto-detect which interface to use. >> >>> Thanks, >>> -=Tod >>> >> Heiner >> >>>> On Nov 10, 2019, at 2:44 a, Heiner Kallweit <hkallweit1@xxxxxxxxx> wrote: >>>> >>>> On 10.11.2019 04:26, Tod E. Kurt wrote: >>>>> HiI >>>>> >>>> Hi Tod, >>>> >>>>> Since you are listed as author of "hid-led.c": >>>>> https://github.com/torvalds/linux/blob/master/drivers/hid/hid-led.c, >>>>> I was wondering if you could offer some insight on an issue I'm seeing. >>>>> I'm a maintainer of "hidapi" and "node-hid", cross-platform C and NodeJs libraries for accessing HID devices, and I'm the creator of the ThingM blink(1) USB LED that "hid-led" controls. >>>>> >>>>> On the distros I've seen (Ubuntu, Raspian) where "hid-led" is enabled, when inserting a blink(1) device, the "hid-led" driver appears to grab the device and remove it from available "hidraw" devices. This makes generic hidraw-based systems (like "hidapi" and the upcoming Chrome WebHID) unable to see the blink(1) >>>>> >>>>> I have tried blacklisting the "hid-led" module but the problem persist. Ffrom dmesg and lsmod the module does appear to not be loaded on reboot. >>>>> >>>>> Any insights on what's going on? Any tips on how to debug this or how to prevent this from occurring? >>>>> >>>> When hid-led was written it was needed to add the LED devices to hid_have_special_driver list. >>>> Else the driver can't take control over the device. Side effect is that even if hid-led >>>> isn't loaded hid-generic can't take control. >>>> I think since e04a0442d33b ("HID: core: remove the absolute need of hid_have_special_driver[]") >>>> it's no longer needed to have the LED devices in hid_have_special_driver. >>>> Could you please test the following patch? If hid-led is loaded is should control the >>>> device, if it's blacklisted hid-generic should have the control. >>>> >>>> >>>> diff --git a/drivers/hid/hid-quirks.c b/drivers/hid/hid-quirks.c >>>> index c50bcd967..bdaab79f7 100644 >>>> --- a/drivers/hid/hid-quirks.c >>>> +++ b/drivers/hid/hid-quirks.c >>>> @@ -419,13 +419,6 @@ static const struct hid_device_id hid_have_special_driver[] = { >>>> #if IS_ENABLED(CONFIG_HID_LCPOWER) >>>> { HID_USB_DEVICE(USB_VENDOR_ID_LCPOWER, USB_DEVICE_ID_LCPOWER_LC1000) }, >>>> #endif >>>> -#if IS_ENABLED(CONFIG_HID_LED) >>>> - { HID_USB_DEVICE(USB_VENDOR_ID_DREAM_CHEEKY, USB_DEVICE_ID_DREAM_CHEEKY_WN) }, >>>> - { HID_USB_DEVICE(USB_VENDOR_ID_DREAM_CHEEKY, USB_DEVICE_ID_DREAM_CHEEKY_FA) }, >>>> - { HID_USB_DEVICE(USB_VENDOR_ID_MICROCHIP, USB_DEVICE_ID_LUXAFOR) }, >>>> - { HID_USB_DEVICE(USB_VENDOR_ID_RISO_KAGAKU, USB_DEVICE_ID_RI_KA_WEBMAIL) }, >>>> - { HID_USB_DEVICE(USB_VENDOR_ID_THINGM, USB_DEVICE_ID_BLINK1) }, >>>> -#endif >>>> #if IS_ENABLED(CONFIG_HID_LENOVO) >>>> { HID_USB_DEVICE(USB_VENDOR_ID_LENOVO, USB_DEVICE_ID_LENOVO_TPKBD) }, >>>> { HID_USB_DEVICE(USB_VENDOR_ID_LENOVO, USB_DEVICE_ID_LENOVO_CUSBKBD) }, >>>> -- >>>> 2.24.0 >>>> >>>>> I can send you blink(1) devices and "hidapi" test programs if you'd like to try to replicate this. >>>>> >>>> Thanks for the offer, I've got a blink(1) already. Just tested the patch with this device >>>> and it worked properly like described above. >>>> >>>>> Thanks! >>>>> -=Tod >>>> >>>> Heiner >>> >>> >> > >