rtc: rv8803: re-initialize all Epson RX8803 registers on voltage loss

The reference manuals of both the RX8803 and RV8803 dictate that
"[On V2F/VLF = ] all registers must be initialized".

The RV-8803 application manual (rev. 1.6) further specifies that crossing
V_LOW2 threshold enables flag V2F and triggers a Power-On reset.
According to table 3.11 in the document, all control registers are
defined to sensible values.

However, The Epson RX-8803 doesn't offer the same guarantees.
It explicitly states:

  During the initial power-up, the TEST bit is reset to "0" and the VLF
  bit is set to "1".
  ∗ At this point, all other register values are _undefined_, so be sure to
  perform a reset before using the module.

Commit d3700b6b ("rtc: rv8803: Stop the clock while setting the time")
also had this rationale:

  Indeed, all the registers must be initialized if the voltage has been
  lower than VLOW2 (triggering V2F), but not low enough to trigger a POR.

We should follow the advice and initialize all applicable registers.
We can group the registers into 3 groups:

A) Already correctly handled registers:
  * 0B-0Ch | Timer Counter  | unused and disabled by clearing TE in 0Dh
  * 0Dh    | Extension Reg  | already initialized in rv8803_regs_configure
  * 0Eh    | Flag Reg       | handled in IRQ handler, except for VLF, VDET
  * 0Eh    | VLF, VDET      | cleared in ->set_time
  * 10h    | 100th Seconds  | Already reset via RESET bit
  * 20-21h | Capture Buffer | holds timestamp unused by driver
  * 2Fh    | Event Control  | resets automatically

B) Registers that are hardware initialized on POR, but not on VLF:
  * 0Fh    | Control Reg
  * 2Ch    | OSC Offset

C) RAM that is undefined on voltage loss:
  * 00-06h | Date/Time
  * 07h    | RAM
  * 08-0Ah | Alarm

This means we should initialize after VLF the registers in group B
(RV8803_CTRL and RV8803_OSC_OFFSET).

Group C is all-zero after voltage loss on the RV-8803, but undefined on
the RX-8803. This is ok for Date/Time because ->get_time returns an
error code for as long as the voltage loss flag is active. It's cleared
on ->set_time however. Zeroing both RAM and alarm ensures a fixed value
is read afterwards.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Link: https://lore.kernel.org/r/20220426071056.1187235-4-s.hauer@pengutronix.de