Fix the Shift & Symbol key issue on GPIO-based keyboards

[tslmy]

Disclaimer: I used AI heavily in this fix. I at least ensured that I understood everything it wrote.

Background

For those who came across this PR in the future, here's the context.

Symptom

As noted by Nasubi on Discord (machine-translated):

On my Brain device:
When I try to enter a "." by pressing the "Symbol" key and "M" simultaneously, the "." repeats continuously and won't stop until I press another key.

Also, I just checked on my new Brain unit:
When entering ",", ".", or "/", a repeat input almost always occurs.
Regarding inputs other than ",", ".", or "/", the issue also occurred with "4" and "=".
Other keys seem fine based on my current testing.

(It seems to happen with those specific keys even when pressed simultaneously with Shift—specifically the E, F, N, M, and - keys.)

As written in this blog post:

I'm testing this on the PW-SH3, and sometimes the simultaneous key input for certain symbols (",", ".", "/") doesn't work right; even when the input registers, it often starts repeating endlessly for some reason.
I'm not sure if other models have the same issue.
In these cases, I deal with it by deleting the extra characters, but the only workaround I could think of to stop the repeating input was to immediately press another key.
(For example, to enter ".", I might press Symbol+M followed by M—resulting in ".m"—and then delete the "m".)
Since "." is frequently used for entering IP addresses and URLs, this is a bit frustrating.
Sometimes the input works fine, but other times it gets recognized as a repeating input, and I haven't been able to control it properly so far.

Context

Here's the actual Brain Row 4 (called "bank 0x10" in the code), from the device tree:

Row 4:  Col0=Shift   Col1=R   Col2=F   Col3=Symbol   Col4=N   Col5=M   Col6=Minus

Both modifiers -- Shift AND Symbol -- physically live on Row 4, together with N, M, and Minus.

Root cause

The original code drove columns like this (simplified):

for each column c:
    drive column c HIGH      ← "activate"
    (all other columns stay driven LOW)
    read the rows
    drive column c LOW again

And the rows were read as: pressed = HIGH. The key idea the original relied on: when you drive column c HIGH, a pressed key on column c pulls its row HIGH, so you see it.

The problem: the other columns were driven LOW the whole time. They are actively forced to ground, not disconnected.

The fix

Two coordinated changes.

Change A: idle columns go hi-Z instead of LOW

New scan:

for each column c:
    drive column c LOW          ← "activate" (note: LOW now, see below)
    set all other columns to INPUT mode  ← hi-Z = disconnected!
    read the rows
    set column c back to INPUT mode (hi-Z)

Change B: rows get pull-ups, and we flip the polarity

If idle columns are now disconnected, nothing holds a row wire at a defined value when none of its keys is pressed. It would float. On the device, it would appear as if all keys are dead.

We need to give those wires a "default value". We do this by adding pull-up resistors to the row wires. Now:

• Key not pressed → row floats up to HIGH via its pull-up → reads 1.
• Key pressed → connects row to the scanned column, which we drive LOW → strong path to ground beats the weak pull-up → reads 0.

So the logic flips to active-low: 0 now means pressed. (Before, with the drive-HIGH scheme, 1 meant pressed.)

That's why the active column is now driven LOW instead of HIGH: the pressed key needs to pull its pulled-up row down to be detected.

We had to flip the polarity because the i.MX28 has no pull-down resistors on these pins. Eviences:

1. The binding doc (fsl,mxs-pinctrl.txt) says, "The configuration on the pins includes drive strength, voltage and pull-up." No "pull-downs".

2. The constants in mxs-pinfunc.h only has two values, and notice they describe an on/off switch, not a direction:

#define MXS_PULL_DISABLE	0
#define MXS_PULL_ENABLE		1

So we can only weakly pull up, not weekly pull down. So the default value can only be "1", not "0". Thus we need to make "0" represent the "pressed" state.