Implement the alwaysRunsAtBackgroundPriority WK2 setting using thread QoS

• Component: Web Template Framework
• Status: RESOLVED FIXED

[Andreas Kling]

Instead of manipulating the process assertion, let's implement the alwaysRunsAtBackgroundPriority setting by having a global thread QoS override.

<rdar://problem/29711409>

[Andreas Kling]

Created attachment 299634 [details]
WIP

[WebKit Commit Bot]

Attachment 299634 [details] did not pass style-queue:


ERROR: Source/WebKit2/UIProcess/ChildProcessProxy.cpp:56:  Should have a space between // and comment  [whitespace/comments] [4]
ERROR: Source/WTF/wtf/Threading.cpp:146:  Should have a space between // and comment  [whitespace/comments] [4]
ERROR: Source/WTF/wtf/Threading.cpp:149:  Should have a space between // and comment  [whitespace/comments] [4]
ERROR: Source/WTF/wtf/Threading.cpp:150:  Should have a space between // and comment  [whitespace/comments] [4]
Total errors found: 4 in 17 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Andreas Kling]

This hook point is not early enough, we're still creating some WorkQueues with too high QoS level.

[Andreas Kling]

Created attachment 299641 [details]
Patch for EWS

Okay, I think this will do the trick. Hooking in XPCServiceInitializer() means we set the QoS override before a ChildProcess object is constructed.

[Andreas Kling]

Created attachment 299643 [details]
Patch for review

[Build Bot]

Comment on attachment 299641 [details]
Patch for EWS

Attachment 299641 [details] did not pass mac-wk2-ews (mac-wk2):
Output: http://webkit-queues.webkit.org/results/2943433

New failing tests:
media/modern-media-controls/media-controller/media-controller-auto-hide-rewind-with-mouse-enter.html
media/modern-media-controls/media-controller/media-controller-auto-hide-mouse-leave-after-play.html
http/tests/misc/slow-loading-animated-image.html
accessibility/mac/aria-multiple-liveregions-notification.html

[Build Bot]

Created attachment 299655 [details]
Archive of layout-test-results from ews107 for mac-elcapitan-wk2

The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews.
Bot: ews107  Port: mac-elcapitan-wk2  Platform: Mac OS X 10.11.6

[Andreas Kling]

Created attachment 299668 [details]
Patch for EWS again

Let's try that patch for EWS again, apparently there were a lot of flaky tests getting skipped today.

[Build Bot]

Comment on attachment 299668 [details]
Patch for EWS again

Attachment 299668 [details] did not pass mac-wk2-ews (mac-wk2):
Output: http://webkit-queues.webkit.org/results/2944578

New failing tests:
media/modern-media-controls/media-controller/media-controller-auto-hide-rewind-with-mouse-enter.html
media/modern-media-controls/media-controller/media-controller-auto-hide-mouse-leave-after-play.html
http/tests/misc/slow-loading-animated-image.html

[Build Bot]

Created attachment 299680 [details]
Archive of layout-test-results from ews106 for mac-elcapitan-wk2

The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews.
Bot: ews106  Port: mac-elcapitan-wk2  Platform: Mac OS X 10.11.6

[Ryosuke Niwa]

Wait, there are two patches listed here. Which one is the good one?

[Andreas Kling]

(In reply to comment #11)
> Wait, there are two patches listed here. Which one is the good one?

Sorry about the confusion. The "Patch for EWS" has the alwaysRunsAtBackgroundPriority setting hardcoded to "on", because I wanted to exercise the full suite of layout tests in this mode.

The "Patch for review" is the one to actually read and review :)

@Gavin, could you take a look at my changes and let me know what you think?

[Antti Koivisto]

Comment on attachment 299643 [details]
Patch for review

View in context: https://bugs.webkit.org/attachment.cgi?id=299643&action=review

> Source/WTF/wtf/Threading.cpp:148
> +qos_class_t adjustedQOSClass(qos_class_t originalClass)
> +{
> +    if (globalQOSOverride != QOS_CLASS_UNSPECIFIED)
> +        return globalQOSOverride;
> +    return originalClass;
> +}

Since you are using setGlobalQOSOverride(QOS_CLASS_UTILITY) you are increasing the priority of the QOS_CLASS_BACKGROUND threads. Is that really the intention? Instead of an override should this rather be about setting a maximum QOS class?

> Source/WebKit2/UIProcess/ChildProcessProxy.cpp:34
> +ChildProcessProxy::ChildProcessProxy(bool alwaysRunsAtBackgroundPriority)

Enum would be more stylish.

> Source/bmalloc/bmalloc/Heap.h:73
> +    qos_class_t takeRequestedQOSClass() { return std::exchange(m_requestedQOSClass, QOS_CLASS_UNSPECIFIED); }
> +    void setQOSOverride(qos_class_t overrideClass) { m_requestedQOSClass = overrideClass; }

Should the names be more specific, like "ScavengerThreadQOSClass"?

[Andreas Kling]

Created attachment 300316 [details]
Patch for landing

[WebKit Commit Bot]

Comment on attachment 300316 [details]
Patch for landing

Clearing flags on attachment: 300316

Committed r211482: <http://trac.webkit.org/changeset/211482>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Chris Dumez]

FYI, we have a fairly major Speedometer regression on iOS and this commit is in the range. This is not the only suspicious commit though.