Closed dennisafa closed 4 years ago
In response to PR creation
Your results will arrive shortly
Nope def not looking at this until next year, good work tho :+1:
@koolzz -- will you have time look at this?
@koolzz -- will you have time look at this?
yeah I'll take a look this weekend
Instead of allocating rings for up to
MAX_NF
during the manager startup, allocate when an NF is actually spawned.Summary:
This PR moves the ring allocation functionality out of
onvm_init.c
and intoonvm_nf.c
. Rings are now allocated whenonvm_nf_start
is called by the NF initialization routine.Usage:
Merging notes:
TODO before merging :
Test Plan:
Made sure that the order of ring allocation didn't cause unexpected race conditions/other bugs. My main concern was shared core mode, in which all of the NF's rings are checked before they might be allocated. But since we NULL them out before the loop begins, it is OK. Tested by creating NF chains using speed_tester and deallocating/reallocating NF's.
Review:
@koolzz @kevindweb