Bug 213705 - RunLoop::dispatch should only call wakeUp when needed
Summary: RunLoop::dispatch should only call wakeUp when needed
Status: RESOLVED FIXED
Alias: None
Product: WebKit
Classification: Unclassified
Component: New Bugs (show other bugs)
Version: WebKit Nightly Build
Hardware: Unspecified Unspecified
: P2 Normal
Assignee: Geoffrey Garen
URL:
Keywords: InRadar
Depends on:
Blocks:
 
Reported: 2020-06-28 16:05 PDT by Geoffrey Garen
Modified: 2020-07-06 10:16 PDT (History)
9 users (show)

See Also:

Attachments
Patch (1.63 KB, patch)
2020-06-28 16:06 PDT, Geoffrey Garen 		no flags Details | Formatted Diff | Diff
Patch (4.94 KB, patch)
2020-06-28 16:33 PDT, Geoffrey Garen 		no flags Details | Formatted Diff | Diff
Patch (2.94 KB, patch)
2020-06-28 21:28 PDT, Geoffrey Garen 		no flags Details | Formatted Diff | Diff
Patch (2.98 KB, patch)
2020-06-30 21:54 PDT, Geoffrey Garen 		no flags Details | Formatted Diff | Diff
Patch (7.01 KB, patch)
2020-07-01 21:24 PDT, Geoffrey Garen 		no flags Details | Formatted Diff | Diff
Patch (8.88 KB, patch)
2020-07-02 15:34 PDT, Geoffrey Garen 		no flags Details | Formatted Diff | Diff
Show Obsolete (5) View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Geoffrey Garen 2020-06-28 16:05:09 PDT 
RunLoop::dispatch should only call wakeUp when needed
Comment 1 Geoffrey Garen 2020-06-28 16:06:25 PDT 
Created attachment 403013 [details]
Patch
Comment 2 Geoffrey Garen 2020-06-28 16:09:01 PDT 
Comment on attachment 403013 [details]
Patch

Oops! I had planned to remove the functionsHandled behavior separately from this patch -- but it's actually not correct until you do both, since functionsHandled is willing to exit without a wakeUp. :(
Comment 3 Geoffrey Garen 2020-06-28 16:33:11 PDT 
Created attachment 403014 [details]
Patch
Comment 4 Sam Weinig 2020-06-28 17:23:51 PDT 
Comment on attachment 403014 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=403014&action=review

> Source/WTF/ChangeLog:10
> +        RunLoop::wakeUp is super expensive (at least on Darwin platforms). Back
> +        when IndexedDB used RunLoop::dispatch, RunLoop::wakeUp accounted for 15%
> +        of all database lookup time.

Out of interest (and for the historical record) what makes it so expensive?
Comment 5 Geoffrey Garen 2020-06-28 20:03:25 PDT 
Most of the cost seemed to be its usage of pthread mutexes.
Comment 6 Sam Weinig 2020-06-28 20:22:36 PDT 
(In reply to Geoffrey Garen from comment #5)
> Most of the cost seemed to be its usage of pthread mutexes.

Is "its" here CFRunLoopWakeUp()? Or something we are doing in WTF?
Comment 7 Geoffrey Garen 2020-06-28 21:28:24 PDT 
Created attachment 403027 [details]
Patch
Comment 8 Geoffrey Garen 2020-06-28 21:29:09 PDT 
> Is "its" here CFRunLoopWakeUp()? Or something we are doing in WTF?

It's CFRunLoopSourceSignal and CFRunLoopWakeUp.
Comment 9 Geoffrey Garen 2020-06-30 21:54:01 PDT 
Created attachment 403264 [details]
Patch
Comment 10 Geoffrey Garen 2020-07-01 21:24:48 PDT 
Created attachment 403343 [details]
Patch
Comment 11 Geoffrey Garen 2020-07-02 15:34:39 PDT 
Created attachment 403408 [details]
Patch
Comment 12 Geoffrey Garen 2020-07-02 16:16:46 PDT 
Comment on attachment 403408 [details]
Patch

Looks like EWS is happy now.
Comment 13 EWS 2020-07-06 10:15:54 PDT 
Committed r263973: <https://trac.webkit.org/changeset/263973>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 403408 [details].
Comment 14 Radar WebKit Bug Importer 2020-07-06 10:16:18 PDT 
<rdar://problem/65137000>