1336517 - Issues with bear71.nfb.ca

Status: RESOLVED FIXED

(Core :: Plug-ins, defect)

Description

[Stefan [:StefanG_QA]]

Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:54.0) Gecko/20100101 Firefox/54.0
Flash Version: 24.0.0.194
e10s = false

STR
1. Open Nightly and navigate to http://bear71.nfb.ca/#/bear71
2. Watch a couple of videos
3. Grant Flash plugin with access to camera and mic when asked
4. Observe video

AR: Flash plugin is crashing
ER: Flash plugin should continue playing the video

Comment 1

[Benjamin Smedberg]

Stefan, is this win64-specific? This is likely sandbox-related, but you should test to be sure, and also see whether this is a recent regression or dates back to turning the strict plugin sandbox on.

Comment 2

[Stefan [:StefanG_QA]]

Update: flipping dom.ipc.plugins.asyncdrawing.enable or e10s does not make any difference nor any combination of them.

Issues is not reproducible in Nightly 32 bits.

Comment 3

[Jim Mathies [:jimm]]

I'm experiencing a number of issues with 32-bit builds on a slower laptop that I don't see with async drawing turned off:

1) lots of buffering in the video
2) resizing the window doesn't update the dims of the flash applet
3) mouse input is ignored on the video player
4) tab spinner when switching tabs

Comment 4

[Jim Mathies [:jimm]]

On my main workstation I have no issues at all.

Comment 5

[Jim Mathies [:jimm]]

(In reply to Jim Mathies [:jimm] from comment #3)
> I'm experiencing a number of issues with 32-bit builds on a slower laptop
> that I don't see with async drawing turned off:
> 
> 1) lots of buffering in the video
> 2) resizing the window doesn't update the dims of the flash applet
> 3) mouse input is ignored on the video player
> 4) tab spinner when switching tabs

These are all side effects of a serious performance issue.

Comment 6

[Jim Mathies [:jimm]]

got the crash too - 

https://crash-stats.mozilla.com/report/index/c28b81e0-8a3a-4752-8ce5-b55c42170207

Comment 7

[Jim Mathies [:jimm]]

https://clptr.io/2ko7qi5

Comment 8

[Jim Mathies [:jimm]]

Attachment: stack.jpg

Comment 9

[Jim Mathies [:jimm]]

milan, can someone on the gfx team who's familiar with the async drawing code poke at this profile? Seems like we're spending a lot of time shifting bits around.

Comment 10

[Jim Mathies [:jimm]]

Doing a little debugging here, appears the changed information coming in @ NPN_SetCurrentAsyncSurface is ok. If you pause the intro video you get regular updates but those are restricted to the playback status bar. So generally I don't see incorrect data coming from flash.

Comment 11

[Milan Sreckovic [:milan] (needinfo for best results)]

David, does the profile from comment 7 / comment 8 look "expected"?

Comment 12

[Chris Peterson [:cpeterson]]

David, do you have time to look at this profile? This is a serious performance regression with Flash that is blocking our rollout of 64-bit Firefox as the default install.

Comment 13

[David Anderson [:dvander] - inactive, e-mail if emergency]

The performance regression here is to be expected, I think, given the circumstances. Previously this page was probably getting hardware-accelerated blitting directly onto a window. This is exactly the reason we previously did not force plugins to windowless mode: that involves like four bitmap copies per frame, which is insanely slow compared to windowing.

To solve this async drawing introduced two new drawing modes: a faster bitmap mode, and a fast-as-windowed mode using DXGI.

For some reason - maybe because the HTML doesn't specify a wmode parameter? This plugin is getting the "faster bitmap" mode. The async bitmap mode is faster than the old windowless mode, but it's still ultimately bitmap copying every frame. There's no way it can compete with acceleration.

When I force the site to use wmode="direct", the plugin is as fast as windowed mode again, since it's using the async DXGI path.

tl;dr - the plugin is being downgraded to software rendering. We probably have to either convince the site to use wmode="direct", or ask Adobe if it's okay to force this on, maybe on their end or on ours, or on a case-by-case basis.