Opened 15 years ago
Closed 15 years ago
#149 closed enhancement (fixed)
Flush windows in context_save_arch() instead of context_restore_arch()
| Reported by: | Jakub Jermář | Owned by: | Jakub Jermář |
|---|---|---|---|
| Priority: | major | Milestone: | 0.4.2 |
| Component: | helenos/kernel/sparc64 | Version: | mainline |
| Keywords: | Cc: | ||
| Blocker for: | Depends on: | ||
| See also: |
Description
Gedare Bloom suggests:
Issue the flushw instruction in context_save_arch instead of context_restore_arch, overwriting the window at cwp with the restored thread state, and "pretending" a flushw occurred by setting CANRESTORE to 0 and CANSAVE to NWINDOWS - 2
The reason I think this would be an improvement is that (1) it makes more sense to me to flush the windows while saving state, and (2) yousave a register spill and register fill of scheduler()'s window.
Change History (3)
comment:1 by , 15 years ago
comment:2 by , 15 years ago
| Summary: | Flush windows in context_save_arch() instead of context_reostre_arch() → Flush windows in context_save_arch() instead of context_restore_arch() |
|---|
Note:
See TracTickets
for help on using tickets.
I already tried this optimization and it appears to be a correct one. The only question is whether it will have the performance benefit it promises.
Moreover, if the windows are not flushed in context_save_arch(), there may be a problem with thread migration: thread is restored on processor 2 while its windows are still on processor 1. This is fortunately not possible in the current code due to an unrelated context_restore() at the end of scheduler() which flushes the windows.