platform: ace: notifying about idle thread readiness #7174
Conversation
tmleman
changed the title
tmleman
requested review from
lgirdwood,
plbossart,
mmaka1,
lbetlej and
dbaluta
as code owners
tmleman
force-pushed
the
topic/upstream/pr/ace/pm/secondary_core_d3
branch
from
2d8cf4a to
305785a
Compare
|
Can one of the admins verify this patch? |
tmleman
force-pushed
the
topic/upstream/pr/ace/pm/secondary_core_d3
branch
from
305785a to
affa6aa
Compare
|
@tmleman any update ? |
@lgirdwood I'm still waiting for the merge on the zephyr side |
Idle thread should be initialize only when core it booting for the first time. Doing this again would overwrite the kernel structs and the idle thread stack. Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
Informing the primary core that the Idle thread on secondary core is ready. During the D3 exit flow thread is not initialize again, but restored from previously saved context. This patch includes also zephyr version update to aba3b12e31 (total 15 commits). Changes related to intel_adsp contain refactor and fixes for ACE secondary cores power flows. Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
tmleman
force-pushed
the
topic/upstream/pr/ace/pm/secondary_core_d3
branch
from
affa6aa to
69c559f
Compare
tmleman
changed the title
| * and the idle thread stack. | ||
| */ | ||
| if (pm_state_next_get(id)->state == PM_STATE_ACTIVE) | ||
| z_init_cpu(id); |
There was a problem hiding this comment.
Given this is an open-coded of z_smp_start_cpu() (see comment above), I don't fully understand why we have a different implementation here, and if this is correct, why this is not needed in the Zephyr kernel/smp.c version. Wouldn't the same problem be there? Or is this specific to a configuration where state is saved to persistant memory? If yes, this would need to be ifdef'ed out so this applies only to applicable platforms.
There was a problem hiding this comment.
I don't know how zephyr is handling it but right now it's not working properly. After secondary core exits D3 it will repeat the path that led him to this state. Power will not be shut off so eventually it will return to the Idle state but this is not correct behavior. Maybe zephyr doesn't support situation in which each core can be turned off and on multiple times.
|
Usual PM fail in https://sof-ci.01.org/sofpr/PR7174/build6662/devicetest/index.html (but passes on at least one device) and https://sof-ci.01.org/sofpr/PR7174/build6663/devicetest/index.html has one fail on system PM. Proceeding with the fix. Thanks @tmleman for filing the issue. |
Informing the primary core that the Idle thread on secondary core is ready. During the D3 exit flow thread is not initialize again, but restored from previously saved context.
Requires: