The old thread priority must be stored as part of the EPOCH(9) tracker.

[• hselasky]

Summary

Else recursive use of EPOCH(9) may cause the wrong priority to be restored.

MFC after: 1 week
Sponsored by: Mellanox Technologies // NVIDIA Networking

[• hselasky]

Added patch context.

[markj]

I forgot that I wrote this, which does priority propagation properly: https://reviews.freebsd.org/D24622

I will try to revisit it shortly. I think this change is ok in the meantime.

sys/sys/param.h

79

Because of the removal of a field from struct thread? In this case the change should not be MFCed. In practice it doesn't matter because the layout will not change due to alignment constraints. There will be some pad bytes between td_base_user_pri and td_rb_list. So in fact __FreeBSD_version does not need to be bumped at all.

[• hselasky]

In D30375#682598, @markj wrote:

I forgot that I wrote this, which does priority propagation properly: https://reviews.freebsd.org/D24622

I will try to revisit it shortly. I think this change is ok in the meantime.

Looks good. I'll have a closer look over the weekend.

--HPS

sys/sys/param.h

79

What about the change in the epoch tracker?

[• hselasky]

This comment was removed by • hselasky.

[• hselasky]

Ignore my last comment. Patch must be AS-IS. Will someone approve it?

[markj]

sys/sys/param.h

79

I forgot about this. In that case the change should not be MFCed.

[• hselasky]

sys/sys/param.h

79

Yes, I'll handle that.