Re: [PATCH v3] HID: multitouch: Add quirks for flipped axes

From: Allen Ballway
Date: Tue Dec 13 2022 - 11:46:37 EST

Next message: Christoph Hellwig: "[GIT PULL] configfs updates for Linux 6.2"
Previous message: Vignesh Raghavendra: "[PATCH v3 3/5] dmaengine: ti: k3-psil-am62a: Add AM62Ax PSIL and PDMA data"
In reply to: Benjamin Tissoires: "Re: [PATCH v3] HID: multitouch: Add quirks for flipped axes"
Next in thread: Allen Ballway: "Re: [PATCH v3] HID: multitouch: Add quirks for flipped axes"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Tue, Dec 13, 2022 at 1:19 AM Benjamin Tissoires
<benjamin.tissoires@xxxxxxxxxx> wrote:
>
> On Tue, Dec 13, 2022 at 2:01 AM Allen Ballway <ballway@xxxxxxxxxxxx> wrote:
> >
> > Certain touchscreen devices, such as the ELAN9034, are oriented
> > incorrectly and report touches on opposite points on the X and Y axes.
> > For example, a 100x200 screen touched at (10,20) would report (90, 180)
> > and vice versa.
> >
> > This changed fixes the issue by adding device quirks to transform
> > the touch points into the correct spaces, from X -> MAX(X) - X,
> > and Y -> MAX(Y) - Y. These quirks are added in hid-quirks checking
> > both DMI information and device vendor and product IDs. The quirk
> > is handled in hid-multitouch to do the actual transformation.
> >
> > Signed-off-by: Allen Ballway <ballway@xxxxxxxxxxxx>
> > ---
> > V2 -> V3: Use existing HID_QUIRK_*_INVERT and match the quirk in
> > hid-quirk, passing down to hid-multitouch through the hid device.
> >
> > V1 -> V2: Address review comments, change to use DMI match. Confirmed
> > MT_TOOL_X/Y require transformation and update orientation based on
> > flipped axes.
> >
> >
> > drivers/hid/hid-multitouch.c | 43 ++++++++++++++++++++++++++++++++----
> > drivers/hid/hid-quirks.c | 33 +++++++++++++++++++++++++++
> > 2 files changed, 72 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/hid/hid-multitouch.c b/drivers/hid/hid-multitouch.c
> > index 91a4d3fc30e08..1f4c2aa511359 100644
> > --- a/drivers/hid/hid-multitouch.c
> > +++ b/drivers/hid/hid-multitouch.c
> > @@ -71,6 +71,8 @@ MODULE_LICENSE("GPL");
> > #define MT_QUIRK_SEPARATE_APP_REPORT BIT(19)
> > #define MT_QUIRK_FORCE_MULTI_INPUT BIT(20)
> > #define MT_QUIRK_DISABLE_WAKEUP BIT(21)
> > +#define MT_QUIRK_X_INVERT BIT(22)
> > +#define MT_QUIRK_Y_INVERT BIT(23)
>
> Why duplicate the already available quirks in struct hid_device?

This no longer seems necessary, and will be removed.

>
> >
> > #define MT_INPUTMODE_TOUCHSCREEN 0x02
> > #define MT_INPUTMODE_TOUCHPAD 0x03
> > @@ -1086,6 +1088,10 @@ static int mt_process_slot(struct mt_device *td, struct input_dev *input,
> > int orientation = wide;
> > int max_azimuth;
> > int azimuth;
> > + int x;
> > + int y;
> > + int cx;
> > + int cy;
> >
> > if (slot->a != DEFAULT_ZERO) {
> > /*
> > @@ -1104,6 +1110,16 @@ static int mt_process_slot(struct mt_device *td, struct input_dev *input,
> > if (azimuth > max_azimuth * 2)
> > azimuth -= max_azimuth * 4;
> > orientation = -azimuth;
> > +
> > + /* Orientation is inverted if the X or Y axes are
> > + * flipped, but normalized if both are inverted.
> > + */
> > + if (quirks & (MT_QUIRK_X_INVERT | MT_QUIRK_Y_INVERT) &&
> > + !((quirks & MT_QUIRK_X_INVERT)
> > + && (quirks & MT_QUIRK_Y_INVERT))) {
>
> This is done in every finger slot in every report. We should probably
> cache that information somewhere.

I can do this once in mt_probe and keep it saved as its own quirk.

>
> > + orientation = -orientation;
> > + }
> > +
> > }
> >
> > if (quirks & MT_QUIRK_TOUCH_SIZE_SCALING) {
> > @@ -1115,10 +1131,23 @@ static int mt_process_slot(struct mt_device *td, struct input_dev *input,
> > minor = minor >> 1;
> > }
> >
> > - input_event(input, EV_ABS, ABS_MT_POSITION_X, *slot->x);
> > - input_event(input, EV_ABS, ABS_MT_POSITION_Y, *slot->y);
> > - input_event(input, EV_ABS, ABS_MT_TOOL_X, *slot->cx);
> > - input_event(input, EV_ABS, ABS_MT_TOOL_Y, *slot->cy);
> > + x = quirks & MT_QUIRK_X_INVERT ?
> > + input_abs_get_max(input, ABS_MT_POSITION_X) - *slot->x :
> > + *slot->x;
> > + y = quirks & MT_QUIRK_Y_INVERT ?
> > + input_abs_get_max(input, ABS_MT_POSITION_Y) - *slot->y :
> > + *slot->y;
> > + cx = quirks & MT_QUIRK_X_INVERT ?
> > + input_abs_get_max(input, ABS_MT_POSITION_X) - *slot->cx :
> > + *slot->cx;
> > + cy = quirks & MT_QUIRK_Y_INVERT ?
> > + input_abs_get_max(input, ABS_MT_POSITION_Y) - *slot->cy :
> > + *slot->cy;
>
> I can't help but think that there must be a better way of doing that.
> If I didn't postpone HID-BPF to 6.3 I would have asked you to do this
> as a BPF program.
>
> Still, the whole point of the slot pointers was to remove the data
> copy everywhere, and you are adding it back. Not to mention that the
> same tests and accesses to variables are called multiple times.

I can move the transformation to mt_touch_input_mapping which has easier access
to the quirks coming from hid_device rather than using duplicated quirks for
mt_application/class.

>
> > +
> > + input_event(input, EV_ABS, ABS_MT_POSITION_X, x);
> > + input_event(input, EV_ABS, ABS_MT_POSITION_Y, y);
> > + input_event(input, EV_ABS, ABS_MT_TOOL_X, cx);
> > + input_event(input, EV_ABS, ABS_MT_TOOL_Y, cy);
> > input_event(input, EV_ABS, ABS_MT_DISTANCE, !*slot->tip_state);
> > input_event(input, EV_ABS, ABS_MT_ORIENTATION, orientation);
> > input_event(input, EV_ABS, ABS_MT_PRESSURE, *slot->p);
> > @@ -1735,6 +1764,12 @@ static int mt_probe(struct hid_device *hdev, const struct hid_device_id *id)
> > if (id->vendor == HID_ANY_ID && id->product == HID_ANY_ID)
> > td->serial_maybe = true;
> >
> > + if (hdev->quirks & HID_QUIRK_X_INVERT)
> > + td->mtclass.quirks |= MT_QUIRK_X_INVERT;
> > +
> > + if (hdev->quirks & HID_QUIRK_Y_INVERT)
> > + td->mtclass.quirks |= MT_QUIRK_Y_INVERT;
>
> As mentioned above, I don't see the point in doing that duplication of quirks.

Will remove

>
> > +
> > /* This allows the driver to correctly support devices
> > * that emit events over several HID messages.
> > */
> > diff --git a/drivers/hid/hid-quirks.c b/drivers/hid/hid-quirks.c
>
> AFAICT, Dmitry told you twice to use i2c-hid, not hid-quirks.c.
>
> We already have i2c-hid-dmi-quirks.c that you could extend. The
> rationale is that this way, the dmi check will be done only for
> internal peripherals, leaving aside the USB ones.

I had added the quirks to hid_device to i2c-hid-dmi-quirks before
trying this but they were getting stomped by the quirks coming from
hid-quirks. I will move these back to i2c-hid-dmi-quirks and change
hid-device to add them to whatever quirks come from hid-quirks.

>
> Cheers,
> Benjamin
>
> > index 0e9702c7f7d6c..47c6cd62f019a 100644
> > --- a/drivers/hid/hid-quirks.c
> > +++ b/drivers/hid/hid-quirks.c
> > @@ -16,6 +16,7 @@
> > #include <linux/export.h>
> > #include <linux/slab.h>
> > #include <linux/mutex.h>
> > +#include <linux/dmi.h>
> > #include <linux/input/elan-i2c-ids.h>
> >
> > #include "hid-ids.h"
> > @@ -957,6 +958,29 @@ static const struct hid_device_id hid_mouse_ignore_list[] = {
> > { }
> > };
> >
> > +static const struct hid_device_id elan_flipped_quirks[] = {
> > + { HID_DEVICE(BUS_I2C, HID_GROUP_MULTITOUCH_WIN_8, USB_VENDOR_ID_ELAN, 0x2dcd),
> > + HID_QUIRK_X_INVERT | HID_QUIRK_Y_INVERT },
> > + { }
> > +};
> > +
> > +/*
> > + * This list contains devices which have specific issues based on the system
> > + * they're on and not just the device itself. The driver_data will have a
> > + * specific hid device to match against.
> > + */
> > +static const struct dmi_system_id dmi_override_table[] = {
> > + {
> > + .ident = "DynaBook K50/FR",
> > + .matches = {
> > + DMI_EXACT_MATCH(DMI_SYS_VENDOR, "Dynabook Inc."),
> > + DMI_EXACT_MATCH(DMI_PRODUCT_NAME, "dynabook K50/FR"),
> > + },
> > + .driver_data = (void *)elan_flipped_quirks,
> > + },
> > + { } /* Terminate list */
> > +};
> > +
> > bool hid_ignore(struct hid_device *hdev)
> > {
> > int i;
> > @@ -1238,6 +1262,7 @@ static unsigned long hid_gets_squirk(const struct hid_device *hdev)
> > {
> > const struct hid_device_id *bl_entry;
> > unsigned long quirks = 0;
> > + const struct dmi_system_id *system_id;
> >
> > if (hid_match_id(hdev, hid_ignore_list))
> > quirks |= HID_QUIRK_IGNORE;
> > @@ -1249,6 +1274,14 @@ static unsigned long hid_gets_squirk(const struct hid_device *hdev)
> > if (bl_entry != NULL)
> > quirks |= bl_entry->driver_data;
> >
> > + system_id = dmi_first_match(dmi_override_table);
> > + if (system_id != NULL) {
> > + bl_entry = hid_match_id(hdev, system_id->driver_data);
> > + if (bl_entry != NULL)
> > + quirks |= bl_entry->driver_data;
> > + }
> > +
> > +
> > if (quirks)
> > dbg_hid("Found squirk 0x%lx for HID device 0x%04x:0x%04x\n",
> > quirks, hdev->vendor, hdev->product);
> > --
> > 2.39.0.rc1.256.g54fd8350bd-goog
> >
>

Thanks for the comments,
Allen

Next message: Christoph Hellwig: "[GIT PULL] configfs updates for Linux 6.2"
Previous message: Vignesh Raghavendra: "[PATCH v3 3/5] dmaengine: ti: k3-psil-am62a: Add AM62Ax PSIL and PDMA data"
In reply to: Benjamin Tissoires: "Re: [PATCH v3] HID: multitouch: Add quirks for flipped axes"
Next in thread: Allen Ballway: "Re: [PATCH v3] HID: multitouch: Add quirks for flipped axes"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]