[PATCH v2 2/2] accel/rocket: keep core slots stable across unbind and rebind

[Jiaxing Hu]

Hi Igor,

All three are addressed, and in the right places. The .dev = NULL lands
in the rocket_core_init() failure path before num_cores is decremented,
which is what keeps the live slot count and the scheds allocation in
step. The .dev check in sched_to_core() and the note about synchronous
probe are both what I had in mind. Thanks for exercising the forced
init failure path on hardware too.

Reviewed-by: Jiaxing Hu <gahing at gahingwoo.com>

And thanks for the Tested-by on my v3. RK3588 is the part I cannot
check here, so having the soc_data conversion confirmed on all three
cores is worth a lot.

Cheers,
Jiaxing