Re: [RFC PATCH v6 7/9] accel/rocket: add RK3576 NPU (RKNN) support

[Jiaxing Hu]

Hi Igor,

> One item outlives it, because it is not part of the poll machinery. In
> rocket_core_init(), the new multi-power-domain attach returns without
> unwinding rocket_job_init():

You are right, and the path immediately above it is the model, as you
say. RK3576 still needs the attach in v7, so the shape does survive the
rewrite, and v7 unwinds rocket_job_init() on that return.

> the commit message says nothing changes for RK3588, but struct
> rocket_core's clks[] grows from 4 to 6 there while the two extra names
> only arrive in 7/9

Moved rather than explained: the array grows in the patch that adds the
two names, so the RK3588 patch really does change nothing for RK3588.

The poll is gone in v7 entirely, not reduced. With the TASK_CON fix the
interrupt retires every job, so there is nothing for a fallback to fall
back to, and a fallback that never runs is a fallback nobody tests. The
job_lock move goes out on its own with a Fixes tag rather than riding in
a feature series, so it can be picked up on its own timeline.

> I would rather send you results than a tag that covers less than it
> looks like it does.

So would I, and thank you for putting it that way. Probe, multi-task
jobs, all three cores in parallel, a forced timeout and reset, and
runtime-PM cycling against a bit-exact oracle is more coverage than any
tag I could ask you for. I will send v7 to you directly when it goes
out.

Jiaxing