Opened 13 years ago
Last modified 13 years ago
#5 started defect
Using %cr8/TPR for spinlock_enter/exit
| Reported by: | john | Owned by: | john |
|---|---|---|---|
| Priority: | major | Component: | kernel |
| Version: | Keywords: | x86 | |
| Cc: | Blocked By: | ||
| Blocking: | Parent Tickets: | ||
| P4 Branch: | //depot/user/jhb/cr8/... | GIT Branch: | |
| FreeBSD PR: | Due Date: |
Description
I have a branch of this implemented. The trick so far is finding a workload where it helps.
Change History (3)
comment:1 Changed 13 years ago by john
- Owner set to john
- Status changed from new to accepted
comment:2 Changed 13 years ago by john
comment:3 Changed 13 years ago by john
- Status changed from accepted to started
Note: See
TracTickets for help on using
tickets.
I have tested doing 'make -j N' of kernel builds on machines with various numbers of cores and varying N and cr8 did not make a positive difference.
Similarly I tested running concurrent sha1 jobs of a really large file and also observed no meaningful difference.