mfd: flipper-one-mcu: register the SW button interrupt

[munzzyy]

fomcu_irqs[] has no entry for FOMCU_INT_INPUT_SWBTN, so the SW button interrupt is allocated but can never fire.

REGMAP_IRQ_REG() is a designated initialiser — [_irq] = { .reg_offset = ..., .mask = ... } — so the enum value is the array index, not the position in the list. The enum runs BTN, TOUCH, HEADSET, SWBTN, UCSI_EVENT, and the array initialises 0, 1, 2 and 4. Index 3 is left as an implicit zero.

Because UCSI_EVENT is index 4, ARRAY_SIZE() is still 5, so num_irqs looks right and nothing complains. Building the array exactly as it is on flipper-devel:

ARRAY_SIZE(fomcu_irqs) = 5  (num_irqs reported to regmap-irq)

  [0] INPUT_BTN      reg_offset=0 mask=0x01
  [1] INPUT_TOUCH    reg_offset=0 mask=0x02
  [2] INPUT_HEADSET  reg_offset=0 mask=0x04
  [3] INPUT_SWBTN    reg_offset=0 mask=0x00   <-- status & 0 is always 0
  [4] UCSI_EVENT     reg_offset=2 mask=0x01

A .mask of 0 means regmap_irq_thread() ANDs the status word with 0 for that irq, so it never calls the handler, and masking and unmasking it write nothing.

Nothing fails loudly, which is what makes it awkward to spot. fo_input_irqs declares the resource, the domain hands out a virq for index 3, platform_get_irq_byname(pdev, "flipper-one-input-swbtn") succeeds, devm_request_threaded_irq() succeeds and the input device registers advertising KEY_POWER. Probe is clean. The button just does nothing.

FOMCU_INTSTS_INPUT_SWBTN is already defined as BIT(3) in the header, so this is only a missing line in the table. With it, index 3 gets mask=0x08 and the array size is unchanged.

Caveat on testing: there's no aarch64 cross-compiler on this machine and I don't have the hardware, so I haven't built the kernel or pressed the button. The reasoning is from the macro expansion and regmap-irq's mask handling, and the numbers above come from compiling the real table. Worth a quick confirmation from someone with a board that KEY_POWER starts arriving.

[alchark]

Thanks for spotting this! It wasn’t working indeed, and I haven’t yet gad chance to debug.

Could you please amend the commit with your full name in the sign-off line per kernel requirements? I’ll then merge at once, and fold it into a co-developed-by upon next rebase+squash.

[munzzyy]

Done — commit is now signed off as Cole Munz, same diff otherwise. Good to hear the button really was dead on your end. Thanks for the quick turnaround!

[alchark]

Thanks a lot for your contribution!