Re: [PATCH v1 3/3] dt-bindings: google,cros-ec-keyb: add fn-key and f-keymap props

Re: [PATCH v1 3/3] dt-bindings: google,cros-ec-keyb: add fn-key and f-keymap props

[Fabio Baltieri]

Hey Rob, thanks for the review.

On Tue, Dec 09, 2025 at 01:22:43PM -0600, Rob Herring wrote:
> On Tue, Dec 09, 2025 at 03:47:06PM +0000, Fabio Baltieri wrote:
> > +  fn-key:
> > +    $ref: /schemas/types.yaml#/definitions/uint32
> > +    description: |
> > +      An u32 containing the coordinate of the Fn key, use the MATRIX_KEY(row,
> > +      col, code) macro, code is ignored.
> > +
> > +  fn-keymap:
> 
> If keymap is linux,keymap, then this should perhaps be linux,fn-keymap. 
> Depends if we still think linux,keymap is Linux specific?

I'm open for suggestions, trying to understand the pattern, these are
specific to this binding I think if anything they should be
google,fn-key and google,fn-keymap, similarly to the existing
google,needs-ghost-filter -- no idea why function-row-physmap was not
prefixed but I guess it slipped in and now it's not worth changing it.

Would it make sense?

Thanks,
Fabio

-- 
Fabio Baltieri

Re: [PATCH v1 3/3] dt-bindings: google,cros-ec-keyb: add fn-key and f-keymap props

[Dmitry Torokhov]

On Wed, Dec 10, 2025 at 06:00:29PM +0000, Fabio Baltieri wrote:
> Hey Rob, thanks for the review.
> 
> On Tue, Dec 09, 2025 at 01:22:43PM -0600, Rob Herring wrote:
> > On Tue, Dec 09, 2025 at 03:47:06PM +0000, Fabio Baltieri wrote:
> > > +  fn-key:
> > > +    $ref: /schemas/types.yaml#/definitions/uint32
> > > +    description: |
> > > +      An u32 containing the coordinate of the Fn key, use the MATRIX_KEY(row,
> > > +      col, code) macro, code is ignored.
> > > +
> > > +  fn-keymap:
> > 
> > If keymap is linux,keymap, then this should perhaps be linux,fn-keymap. 
> > Depends if we still think linux,keymap is Linux specific?
> 
> I'm open for suggestions, trying to understand the pattern, these are
> specific to this binding I think if anything they should be
> google,fn-key and google,fn-keymap, similarly to the existing
> google,needs-ghost-filter -- no idea why function-row-physmap was not
> prefixed but I guess it slipped in and now it's not worth changing it.

Just double the number of rows in the regular keymap to accommodate the
FN modifier, no need for separate keymap. Also no need to have fn-key
property, use whatever key that reports KEY_FN. See how it is done in
drivers/input/keyboard/tegra-kbc.c

Thanks.

-- 
Dmitry

Re: [PATCH v1 3/3] dt-bindings: google,cros-ec-keyb: add fn-key and f-keymap props

[Fabio Baltieri]

Hi Dmitry,

On Thu, Dec 11, 2025 at 08:44:02PM -0800, Dmitry Torokhov wrote:
> On Wed, Dec 10, 2025 at 06:00:29PM +0000, Fabio Baltieri wrote:
> > Hey Rob, thanks for the review.
> > 
> > On Tue, Dec 09, 2025 at 01:22:43PM -0600, Rob Herring wrote:
> > > On Tue, Dec 09, 2025 at 03:47:06PM +0000, Fabio Baltieri wrote:
> > > > +  fn-key:
> > > > +    $ref: /schemas/types.yaml#/definitions/uint32
> > > > +    description: |
> > > > +      An u32 containing the coordinate of the Fn key, use the MATRIX_KEY(row,
> > > > +      col, code) macro, code is ignored.
> > > > +
> > > > +  fn-keymap:
> > > 
> > > If keymap is linux,keymap, then this should perhaps be linux,fn-keymap. 
> > > Depends if we still think linux,keymap is Linux specific?
> > 
> > I'm open for suggestions, trying to understand the pattern, these are
> > specific to this binding I think if anything they should be
> > google,fn-key and google,fn-keymap, similarly to the existing
> > google,needs-ghost-filter -- no idea why function-row-physmap was not
> > prefixed but I guess it slipped in and now it's not worth changing it.
> 
> Just double the number of rows in the regular keymap to accommodate the
> FN modifier, no need for separate keymap. Also no need to have fn-key
> property, use whatever key that reports KEY_FN. See how it is done in
> drivers/input/keyboard/tegra-kbc.c

Had a look at the tegra-kbc driver as you suggested, first thing it
seems like the fn-key functionality there is dead code since 2013,
`use_fn_map` could only be enabled with platform data, not OF, and that
has been removed in 3a495aeada2b, as it stands kbc->use_fn_map can only
be false. I could send a patch to rip off that code if you want me to,
clearly it hasn't been used in a while (unless I'm missing something).

About the extended fn map, I've two problems with it:
- it seems very wasteful: the normal map is loaded in a linear array
  so it can be access directly, which make sense as that's typically
  very densely populated, but in the case of the fn keys that's going to
  be mostly empty, I'd expect ~20 keys top from a 18x8 matrix. So that
  would waste load of space, direct access is good but for ~20 keys I
  think it's fine to scan it, especially since it only happens when Fn
  is pressed.
- I'd end up with two values for cols kicking around the driver, the
  real one and the one used in the map, which I feel adds confusing in
  the code.
- more importantly, one would have to keep the offset in mind when
  setting the keys in dt, we rely on OEM doing this and I think having a
  separate property with a meaningful name and a map with the same
  row,col and different code is more intuitive and would make their life
  easier, especially since we ship with keyboard of different size
  and the offset would be different depending on the device.

As for the fn-key property, unfortunately based on past experience I'd
expect such OEM to want to change that code, I could specify the code
rather than the row,col but I would not plain hardcode. Even my
(thinkpad) laptop sends KEY_WAKEUP for Fn.

Cheers,
Fabio

-- 
Fabio Baltieri

Re: [PATCH v1 3/3] dt-bindings: google,cros-ec-keyb: add fn-key and f-keymap props

[Dmitry Torokhov]

Hi Fabio,

On Tue, Dec 16, 2025 at 12:23:06PM +0000, Fabio Baltieri wrote:
> Hi Dmitry,
> 
> On Thu, Dec 11, 2025 at 08:44:02PM -0800, Dmitry Torokhov wrote:
> > On Wed, Dec 10, 2025 at 06:00:29PM +0000, Fabio Baltieri wrote:
> > > Hey Rob, thanks for the review.
> > > 
> > > On Tue, Dec 09, 2025 at 01:22:43PM -0600, Rob Herring wrote:
> > > > On Tue, Dec 09, 2025 at 03:47:06PM +0000, Fabio Baltieri wrote:
> > > > > +  fn-key:
> > > > > +    $ref: /schemas/types.yaml#/definitions/uint32
> > > > > +    description: |
> > > > > +      An u32 containing the coordinate of the Fn key, use the MATRIX_KEY(row,
> > > > > +      col, code) macro, code is ignored.
> > > > > +
> > > > > +  fn-keymap:
> > > > 
> > > > If keymap is linux,keymap, then this should perhaps be linux,fn-keymap. 
> > > > Depends if we still think linux,keymap is Linux specific?
> > > 
> > > I'm open for suggestions, trying to understand the pattern, these are
> > > specific to this binding I think if anything they should be
> > > google,fn-key and google,fn-keymap, similarly to the existing
> > > google,needs-ghost-filter -- no idea why function-row-physmap was not
> > > prefixed but I guess it slipped in and now it's not worth changing it.
> > 
> > Just double the number of rows in the regular keymap to accommodate the
> > FN modifier, no need for separate keymap. Also no need to have fn-key
> > property, use whatever key that reports KEY_FN. See how it is done in
> > drivers/input/keyboard/tegra-kbc.c
> 
> Had a look at the tegra-kbc driver as you suggested, first thing it
> seems like the fn-key functionality there is dead code since 2013,
> `use_fn_map` could only be enabled with platform data, not OF, and that
> has been removed in 3a495aeada2b, as it stands kbc->use_fn_map can only
> be false. I could send a patch to rip off that code if you want me to,
> clearly it hasn't been used in a while (unless I'm missing something).

I guess you are right, we shoudl clean that up. We have another newer
driver that uses the same approach:

drivers/input/keyboard/pinephone-keyboard.c


> About the extended fn map, I've two problems with it:
> - it seems very wasteful: the normal map is loaded in a linear array
>   so it can be access directly, which make sense as that's typically
>   very densely populated, but in the case of the fn keys that's going to
>   be mostly empty, I'd expect ~20 keys top from a 18x8 matrix. So that
>   would waste load of space, direct access is good but for ~20 keys I
>   think it's fine to scan it, especially since it only happens when Fn
>   is pressed.

I am not concerned with this, as this is a singleton device. You
probably "waste" as much space in the code segment by implementing the
custom scanning logic. Additionally with the consolidated keymap
approach you are not breaking ioctls dealing with setting and retrieving
key codes.

> - I'd end up with two values for cols kicking around the driver, the
>   real one and the one used in the map, which I feel adds confusing in
>   the code.

Not sure I follow. You still have the same row and col reported, just
when figuring out the final keycode you need to add an offset.

> - more importantly, one would have to keep the offset in mind when
>   setting the keys in dt, we rely on OEM doing this and I think having a

Do we now? I thought we retain greater control over this. Maybe we
should sync internally.

>   separate property with a meaningful name and a map with the same
>   row,col and different code is more intuitive and would make their life
>   easier, especially since we ship with keyboard of different size
>   and the offset would be different depending on the device.
> 
> As for the fn-key property, unfortunately based on past experience I'd
> expect such OEM to want to change that code, I could specify the code
> rather than the row,col but I would not plain hardcode. Even my
> (thinkpad) laptop sends KEY_WAKEUP for Fn.

Again, we need to make sure we control OEMs better. On Lenovo Fn sends
wakeup only if it is not combined with another key, so it really has
custom logic with events delivered either through the main AT keyboard
or through custom interface in thinkpad platform driver. We do not need
this in oiur designs.

Thanks.

-- 
Dmitry

Re: [PATCH v1 3/3] dt-bindings: google,cros-ec-keyb: add fn-key and f-keymap props

[Fabio Baltieri]

On Wed, Dec 17, 2025 at 10:05:45AM -0800, Dmitry Torokhov wrote:
> > Had a look at the tegra-kbc driver as you suggested, first thing it
> > seems like the fn-key functionality there is dead code since 2013,
> > `use_fn_map` could only be enabled with platform data, not OF, and that
> > has been removed in 3a495aeada2b, as it stands kbc->use_fn_map can only
> > be false. I could send a patch to rip off that code if you want me to,
> > clearly it hasn't been used in a while (unless I'm missing something).
> 
> I guess you are right, we shoudl clean that up. We have another newer
> driver that uses the same approach:
> 
> drivers/input/keyboard/pinephone-keyboard.c

Alright I'll look into it and rework a v2 with the extended map, checked
the code again and I see your arguments, sounds reasonable.

Cheers,
Fabio

-- 
Fabio Baltieri

Re: [PATCH v1 2/3] Input: cros_ec_keyb: add function key support

[Fabio Baltieri]

Hey Simon,

On Thu, Dec 11, 2025 at 06:29:01AM -0700, Simon Glass wrote:
> > @@ -44,6 +52,13 @@
> >   * @bs_idev: The input device for non-matrix buttons and switches (or NULL).
> >   * @notifier: interrupt event notifier for transport devices
> >   * @vdata: vivaldi function row data
> > + * @fn_key: coordinate of the function key
> > + * @fn_keymap: array of coordinate and codes for the function keys
> > + * @fn_keymap_len: number of entries in the fn_keymap array
> > + * @fn_key_status: active function keys bitmap
> > + * @normal_key_status: active normal keys bitmap
> > + * @fn_key_pressed: tracks the function key status
> > + * @fn_key_triggered: tracks where any function key fired
> >   */
> >  struct cros_ec_keyb {
> >         unsigned int rows;
> > @@ -61,6 +76,14 @@ struct cros_ec_keyb {
> >         struct notifier_block notifier;
> >
> >         struct vivaldi_data vdata;
> > +
> > +       uint32_t fn_key;
> 
> Normally we use u32/u8 these days

Okay, I did notice the file was a bit of a mix, I'll change them in v2.

> 
> > +       uint32_t *fn_keymap;
> > +       int fn_keymap_len;
> > +       uint32_t fn_key_status;
> > +       uint8_t normal_key_status[CROS_EC_KEYBOARD_COLS_MAX];
> > +       bool fn_key_pressed;
> > +       bool fn_key_triggered;
> >  };
> >
> >  /**
> > @@ -166,16 +189,108 @@ static bool cros_ec_keyb_has_ghosting(struct cros_ec_keyb *ckdev, uint8_t *buf)
> >         return false;
> >  }
> >
> > +static bool cros_ec_key_is(int row, int col, uint32_t key)
> > +{
> > +       if (row == KEY_ROW(key) && col == KEY_COL(key))
> > +               return true;
> > +
> > +       return false;
> > +}
> > +
> > +static void cros_ec_keyb_process_one(struct cros_ec_keyb *ckdev,
> > +                                    int row, int col, bool state)
> > +{
> > +       struct input_dev *idev = ckdev->idev;
> > +       const unsigned short *keycodes = idev->keycode;
> > +       int pos = MATRIX_SCAN_CODE(row, col, ckdev->row_shift);
> > +       unsigned int code = keycodes[pos];
> > +
> > +       dev_dbg(ckdev->dev, "changed: [r%d c%d]: byte %02x\n", row, col, state);
> > +
> > +       if (ckdev->fn_keymap) {
> > +               if (cros_ec_key_is(row, col, ckdev->fn_key)) {
> > +                       ckdev->fn_key_pressed = state;
> > +
> > +                       if (state) {
> > +                               ckdev->fn_key_triggered = false;
> > +                       } else if (!ckdev->fn_key_triggered) {
> > +                               /*
> > +                                * Send the original code if nothing else has
> > +                                * been pressed together with Fn.
> > +                                */
> > +                               input_event(idev, EV_MSC, MSC_SCAN, pos);
> > +                               input_report_key(idev, code, true);
> > +                               input_sync(ckdev->idev);
> 
> What is this function? I might be missing a patch?

input_sync? it sends an EV_SYN, been there from the start, though I
noticed I miss one two lines below, was relying on the rest of the
function to send it but I changed the logic at some point and broke that
path, will fix that.

> 
> > +
> > +                               input_event(idev, EV_MSC, MSC_SCAN, pos);
> > +                               input_report_key(idev, code, false);
> > +                       }
> > +
> > +                       return;
> > +               }
> > +
> > +               if (!state) {
> > +                       /* Key release, may need to release the Fn code */
> > +                       for (int i = 0; i < ckdev->fn_keymap_len; i++) {
> > +                               if (!cros_ec_key_is(row, col,
> > +                                                   ckdev->fn_keymap[i]))
> > +                                       continue;
> > +
> > +                               if ((ckdev->fn_key_status & BIT(i)) == 0)
> > +                                       continue;
> > +
> > +                               code = KEY_VAL(ckdev->fn_keymap[i]);
> > +                               ckdev->fn_key_status &= ~BIT(i);
> > +
> > +                               input_event(idev, EV_MSC, MSC_SCAN, pos);
> > +                               input_report_key(idev, code, state);
> > +
> > +                               return;
> > +                       }
> > +
> > +                       if ((ckdev->normal_key_status[col] & BIT(row)) == 0)
> > +                               /* Discard, key press code was not sent */
> > +                               return;
> > +               } else if (ckdev->fn_key_pressed) {
> > +                       /* Key press while holding Fn */
> > +                       ckdev->fn_key_triggered = true;
> > +
> > +                       for (int i = 0; i < ckdev->fn_keymap_len; i++) {
> > +                               if (!cros_ec_key_is(row, col,
> > +                                                   ckdev->fn_keymap[i]))
> > +                                       continue;
> > +
> > +                               code = KEY_VAL(ckdev->fn_keymap[i]);
> > +                               ckdev->fn_key_status |= BIT(i);
> > +
> > +                               input_event(idev, EV_MSC, MSC_SCAN, pos);
> > +                               input_report_key(idev, code, state);
> > +
> > +                               return;
> > +                       }
> > +
> > +                       /* Do not emit a code if the key is not mapped */
> > +                       return;
> > +               }
> > +       }
> 
> I think this function could do with splitting a bit

Yeah, I don't love it either but there's a lot of logic intertwined in
there, tried to split it myself and ended up breaking stuff, the logic
for the fn key itsel though can go that's a good 16 lines, I'll start
with that, send a v2 and then go from there.

> Can the sandbox driver support this too?

Not sure what you are referring to, can you give me a pointer?

Hey thanks for the review, good to hear from you. :-)

Cheers,
Fabio

-- 
Fabio Baltieri