> owner@xxxxxxxxxxxxxxx] On Behalf Of Kevin Hilman
> Sent: Wednesday, October 01, 2008 5:47 AM
> This patch does the accounting use 'dev->last_state' state instead of
> the governor-chosen state. If the target's enter_idle hook enters a
> different state than it was asked, it should update dev->last_state.
Isn't this the same as the patch Venki pushed on the pm list? I've been using a version of it for a while from him and haven't seen any issue.
I had also done a version which was much more evasive. In the version I was hacking at I tried to push the aborted state information back into the governor as it locally caches the last state tried.
Doing this simpler way like in patch does make sure the accounting is correct which is a plus, powertop doesn't 'lie' any more. It however, does not stop a following likely mis-prediction based on the local state kept in the governor.
Doing at least this much is necessary. However, if you don't want to feed the false state data into the predictor more linkage has to be done.
Regards,
Richard W.
--- Begin Message ---
As noted in PM minisummit notes here
http://lwn.net/Articles/292447/
Richard Woodruff <r-woodruff2@xxxxxx> reported a bug
"the CPUIDLE bug that if target is avoided due to BM activity the original
target state is still accounted the time."
Following provides a mechanism to fix the problem for all drivers and fixes
the problem for acpi cpuidle driver.
--
_______________________________________________
linux-pm mailing list
linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/linux-pm
--- End Message ---
_______________________________________________
linux-pm mailing list
linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/linux-pm
