Investigate SPI runtime PM count underflow

I noticed this when booting rk3588-test from 2024-03-22 on my RK3588 EVB1.

rockchip-spi feb20000.spi: Runtime PM usage count underflow!
rockchip-spi feb20000.spi: Runtime PM usage count underflow!

The underflow also happens on the Rock 5B in the CI. As this is an underflow, there is an unbalanced pm_runtime_put*(). The driver has a total of 3:

1. In rockchip_spi_setup(). That one has a matching pm_runtime_get_sync() a few lines before and should be fine
2. In rockchip_spi_remove(), directly after pm_runtime_get_sync(), so it should also be fine.
3. In rockchip_spi_set_cs. That one runs pm_runtime_put() when CS is deasserted and pm_runtime_get_sync() when it is asserted.

I'm pretty sure the last one is the root cause, since there is no need to have balanced calls to rockchip_spi_set_cs() and it kind of makes sense to explicitly de-assert CS at probe time.

It looks like this got fixed at some point. I no longer see the error after rebasing to latest master.

closed