Re: [PATCH v2 10/12] HID: input: remove the need for HID_QUIRK_INVERT
From: Benjamin Tissoires
Date: Tue Feb 15 2022 - 03:52:14 EST
On Tue, Feb 15, 2022 at 3:04 AM Ping Cheng <pinglinux@xxxxxxxxx> wrote:
>
> On Mon, Feb 14, 2022 at 2:23 AM Benjamin Tissoires
> <benjamin.tissoires@xxxxxxxxxx> wrote:
> >
> > Hi Ping,
> >
> > On Thu, Feb 10, 2022 at 6:44 AM Ping Cheng <pinglinux@xxxxxxxxx> wrote:
> > >
> > > Sorry for the noise. Hit the wrong buttons...
> > >
> > > On Wed, Feb 9, 2022 at 9:21 PM Ping Cheng <pinglinux@xxxxxxxxx> wrote:
> > > >
> > > > On Thu, Feb 3, 2022 at 6:33 AM Benjamin Tissoires
> > > > <benjamin.tissoires@xxxxxxxxxx> wrote:
> > > > >
> > > > > HID_QUIRK_INVERT is kind of complex to deal with and was bogus.
> > > > >
> > > > > Furthermore, it didn't make sense to use a global per struct hid_device
> > > > > quirk for something dynamic as the current state.
> > > > >
> > > > > Store the current tool information in the report itself, and re-order
> > > > > the processing of the fields to enforce having all the tablet "state"
> > > > > fields before getting to In Range and other input fields.
> > > > >
> > > > > This way, we now have all the information whether a tool is present
> > > > > or not while processing In Range.
> > > > >
> > > > > This new behavior enforces that only one tool gets forwarded to userspace
> > > > > at the same time, and that if either eraser or invert is set, we enforce
> > > > > BTN_TOOL_RUBBER.
> > > > >
> > > > > Note that the release of the previous tool now happens in its own EV_SYN
> > > > > report so userspace doesn't get confused by having 2 tools.
> > > > >
> > > > > These changes are tested in the following hid-tools regression tests:
> > > > > https://gitlab.freedesktop.org/libevdev/hid-tools/-/merge_requests/127
> > > > >
> > > > > Signed-off-by: Benjamin Tissoires <benjamin.tissoires@xxxxxxxxxx>
> > > > >
> > > > > ---
> > > > >
> > > > > Changes in v2:
> > > > > - rework the entire tool switching, this makes the processing
> > > > > slightly more complex but is better for existing userspace.
> > > > > ---
> > > > > drivers/hid/hid-input.c | 98 +++++++++++++++++++++++++++++++++++++----
> > > > > include/linux/hid.h | 6 ++-
> > > > > 2 files changed, 95 insertions(+), 9 deletions(-)
> > > > >
> > > > > diff --git a/drivers/hid/hid-input.c b/drivers/hid/hid-input.c
> > > > > index 61d91117f4ae..9f8853640648 100644
> > > > > --- a/drivers/hid/hid-input.c
> > > > > +++ b/drivers/hid/hid-input.c
> > > > > @@ -63,8 +63,11 @@ static const struct {
> > > > > * This still leaves us 65535 individual priority values.
> > > > > */
> > > > > static const __u32 hidinput_usages_priorities[] = {
> > > > > + HID_DG_ERASER, /* Eraser (eraser touching) must always come before tipswitch */
> > > > > HID_DG_INVERT, /* Invert must always come before In Range */
> > > > > - HID_DG_INRANGE,
> > > > > + HID_DG_TIPSWITCH, /* Is the tip of the tool touching? */
> > > > > + HID_DG_TIPPRESSURE, /* Tip Pressure might emulate tip switch */
> > > > > + HID_DG_INRANGE, /* In Range needs to come after the other tool states */
> > > > > };
> > > > >
> > > > > #define map_abs(c) hid_map_usage(hidinput, usage, &bit, &max, EV_ABS, (c))
> > > > > @@ -1365,9 +1368,38 @@ static void hidinput_handle_scroll(struct hid_usage *usage,
> > > > > input_event(input, EV_REL, usage->code, hi_res);
> > > > > }
> > > > >
> > > > > +static void hid_report_release_tool(struct hid_report *report, struct input_dev *input,
> > > > > + unsigned int tool)
> > > > > +{
> > > > > + /* if the given tool is not currently reported, ignore */
> > > > > + if (!test_bit(tool, input->key))
> > > > > + return;
> > > > > +
> > > > > + /*
> > > > > + * if the given tool was previously set, release it,
> > > > > + * release any TOUCH and send an EV_SYN
> > > > > + */
> > > > > + input_event(input, EV_KEY, BTN_TOUCH, 0);
> > > > > + input_event(input, EV_KEY, tool, 0);
> > >
> > > Took a while for me to find the right device and setup the proper
> > > system for the testing. This block missing the serial number:
> > >
> > > input_event(input, EV_MSC, MSC_SERIAL, 0);
> > >
> > > Without this line, the next tool will not get its serial number.
> >
> > I am tempted to simply disregard this request. It has been known for
> > ages that the evdev values are cached, so if you do not have the value
> > you need that means that the value has been sent previously (or you
> > can request it upon start with an ioctl). So in this particular case,
> > I don't really see the logic in forcing the SERIAL into the TOOL type
> > when the tool is clearly unrelated to the serial.
> >
> > My fear here is that by linking together too many axes, we may enter
> > in a state where we can not untangle them when needed.
>
> I see your point. We indeed added those ioctl's in the X driver to
> avoid missing the initial states of some of the key events.
>
> > >
> > > > > + input_event(input, EV_SYN, SYN_REPORT, 0);
> > > > > +
> > > > > + report->tool = 0;
> > > > > +}
> > > > > +
> > > > > +static void hid_report_set_tool(struct hid_report *report, struct input_dev *input,
> > > > > + unsigned int new_tool)
> > > > > +{
> > > > > + if (report->tool != new_tool)
> > > > > + hid_report_release_tool(report, input, report->tool);
> > > > > +
> > > > > + input_event(input, EV_KEY, new_tool, 1);
> > > > > + report->tool = new_tool;
> > > > > +}
> > > > > +
> > > > > void hidinput_hid_event(struct hid_device *hid, struct hid_field *field, struct hid_usage *usage, __s32 value)
> > > > > {
> > > > > struct input_dev *input;
> > > > > + struct hid_report *report = field->report;
> > > > > unsigned *quirks = &hid->quirks;
> > > > >
> > > > > if (!usage->type)
> > > > > @@ -1418,25 +1450,75 @@ void hidinput_hid_event(struct hid_device *hid, struct hid_field *field, struct
> > > > > }
> > > > >
> > > > > switch (usage->hid) {
> > > > > + case HID_DG_ERASER:
> > > > > + report->tool_active |= !!value;
> > > > > +
> > > > > + /*
> > > > > + * if eraser is set, we must enforce BTN_TOOL_RUBBER
> > > > > + * to accommodate for devices not following the spec.
> > > > > + */
> > > > > + if (value)
> > > > > + hid_report_set_tool(report, input, BTN_TOOL_RUBBER);
> > > > > + else if (report->tool != BTN_TOOL_RUBBER)
> > > > > + /* value is off, tool is not rubber, ignore */
> > > > > + return;
> > > > > +
> > > > > + /* let hid-input set BTN_TOUCH */
> > > > > + break;
> > > > > +
> > > > > case HID_DG_INVERT:
> > > > > - *quirks = value ? (*quirks | HID_QUIRK_INVERT) : (*quirks & ~HID_QUIRK_INVERT);
> > > > > + report->tool_active |= !!value;
> > > > > +
> > > > > + /*
> > > > > + * If invert is set, we store BTN_TOOL_RUBBER.
> > > > > + */
> > > > > + if (value)
> > > > > + hid_report_set_tool(report, input, BTN_TOOL_RUBBER);
> > > > > + else if (!report->tool_active)
> > > > > + /* tool_active not set means Invert and Eraser are not set */
> > > > > + hid_report_release_tool(report, input, BTN_TOOL_RUBBER);
> > > > > +
> > > > > + /* no further processing */
> > > > > return;
> > > > >
> > > > > case HID_DG_INRANGE:
> > > > > - if (value) {
> > > > > - input_event(input, usage->type, (*quirks & HID_QUIRK_INVERT) ? BTN_TOOL_RUBBER : usage->code, 1);
> > > > > - return;
> > > > > + report->tool_active |= !!value;
> > > > > +
> > > > > + if (report->tool_active) {
> > > > > + /*
> > > > > + * if tool is not set but is marked as active,
> > > > > + * assume ours
> > > > > + */
> > > > > + if (!report->tool)
> > > > > + hid_report_set_tool(report, input, usage->code);
> > > > > + } else {
> > > > > + hid_report_release_tool(report, input, usage->code);
> > > > > }
> > > > > - input_event(input, usage->type, usage->code, 0);
> > > > > - input_event(input, usage->type, BTN_TOOL_RUBBER, 0);
> > > > > +
> > > > > + /* reset tool_active for the next event */
> > > > > + report->tool_active = false;
> > > > > +
> > > > > + /* no further processing */
> > > > > return;
> > > > >
> > > > > + case HID_DG_TIPSWITCH:
> > > > > + report->tool_active |= !!value;
> > > > > +
> > > > > + /* if tool is set to RUBBER we should ignore the current value */
> > > > > + if (report->tool == BTN_TOOL_RUBBER)
> > > > > + return;
> > >
> > > This case should return before HID_DG_INVERT is checked since we may
> > > get TIPSWITCH bit before INVERT. In fact, the device that I tested
> > > sends the tipswatch bit before the invert bit. So, I alway get a PEN
> > > in and out before I get ERASER events, when I bring the pen in with
> > > the side-switch/invert bit pressed.
> > >
> > > However, we know not all devices support INVERT. We probably have to
> > > keep the invert quirk to decide if BTN_TOOL_PEN should be set here or
> > > wait until HID_DG_INVERT is reached.
> >
> > I am under the impression that you completely missed the point of the
> > patch series. This series re-orders the processing of the events, and
> > if you look at the very first hunk in this patch, you'll see that the
> > new processing ensures we treat INVERT and ERASER before TIPSWITCH, no
> > matter the order of the device.
>
> Yeah, I read that part of the code, more than once. But, my evtest
> gave me an extra set of TOOL_PEN events before TOOL_ERASER. I was
> confused too. Then I thought maybe the code still relies on the actual
> order of the usages. Anyway, we've done enough discussion and testing,
> let's move on. The patchset is:
>
> Reviewed-by: Ping Cheng <ping.cheng@xxxxxxxxx>
> Tested-by: Ping Cheng <ping.cheng@xxxxxxxxx>
Thanks for the review.
However, if the device is not working as expected, can you send me
the hid-recorder[0] output off list so I can diagnose what is
happening?
Cheers,
Benjamin
[0] clone https://gitlab.freedesktop.org/libevdev/hid-tools, then pip
install ., then `sudo hid-recorder`
>
> Thank you for your effort, Benjamin!
> Ping
>
> > So when we process TIPSWITCH here, we are sure that if BTN_TOOL_RUBBER
> > is set, either INVERT or ERASER is set to 1, whether or not they come
> > before or after in the report.
> >
> > >
> > > If you are interested in the evtest output, please let me know. I'll
> > > share it with you offline so we don't abuse this list...
> >
> > evtest is the output of the kernel, and it doesn't give me much more
> > than "it's broken". However, if you can give me the hid-recorder
> > outputs, I'll be able to reproduce locally and compare the output as I
> > fix the code. Of course the best would be to add test cases in the
> > hid-tools repo, but I'll need the hid-recorder anyway to understand
> > what is failing and to be able to write the tests.
> >
> > Cheers,
> > Benjamin
> >
> > >
> > > Cheers,
> > > Ping
> > >
> > > > > + break;
> > > > > +
> > > > > case HID_DG_TIPPRESSURE:
> > > > > if (*quirks & HID_QUIRK_NOTOUCH) {
> > > > > int a = field->logical_minimum;
> > > > > int b = field->logical_maximum;
> > > > >
> > > > > - input_event(input, EV_KEY, BTN_TOUCH, value > a + ((b - a) >> 3));
> > > > > + if (value > a + ((b - a) >> 3)) {
> > > > > + input_event(input, EV_KEY, BTN_TOUCH, 1);
> > > > > + report->tool_active = true;
> > > > > + }
> > > > > }
> > > > > break;
> > > > >
> > > > > diff --git a/include/linux/hid.h b/include/linux/hid.h
> > > > > index eaad0655b05c..feb8df61168f 100644
> > > > > --- a/include/linux/hid.h
> > > > > +++ b/include/linux/hid.h
> > > > > @@ -347,7 +347,7 @@ struct hid_item {
> > > > > */
> > > > > #define MAX_USBHID_BOOT_QUIRKS 4
> > > > >
> > > > > -#define HID_QUIRK_INVERT BIT(0)
> > > > > +/* BIT(0) reserved for backward compatibility, was HID_QUIRK_INVERT */
> > > > > #define HID_QUIRK_NOTOUCH BIT(1)
> > > > > #define HID_QUIRK_IGNORE BIT(2)
> > > > > #define HID_QUIRK_NOGET BIT(3)
> > > > > @@ -515,6 +515,10 @@ struct hid_report {
> > > > > unsigned maxfield; /* maximum valid field index */
> > > > > unsigned size; /* size of the report (bits) */
> > > > > struct hid_device *device; /* associated device */
> > > > > +
> > > > > + /* tool related state */
> > > > > + bool tool_active; /* whether the current tool is active */
> > > > > + unsigned int tool; /* BTN_TOOL_* */
> > > > > };
> > > > >
> > > > > #define HID_MAX_IDS 256
> > > > > --
> > > > > 2.33.1
> > > > >
> > >
> >
>