Closed
Bug 898919
Opened 11 years ago
Closed 11 years ago
[A/V] GENLOCK_IOC_DEADLOCK failed occur while youtube streaming
Categories
(Firefox OS Graveyard :: General, defect, P1)
Tracking
(blocking-b2g:leo+, firefox26 unaffected, b2g18 fixed, b2g18-v1.0.0 wontfix, b2g18-v1.0.1 wontfix, b2g-v1.1hd fixed)
Tracking | Status | |
---|---|---|
firefox26 | --- | unaffected |
b2g18 | --- | fixed |
b2g18-v1.0.0 | --- | wontfix |
b2g18-v1.0.1 | --- | wontfix |
b2g-v1.1hd | --- | fixed |
People
(Reporter: leo.bugzilla.gecko, Assigned: sotaro)
References
Details
(Whiteboard: [TD-68993][SR 01255148])
Attachments
(2 files, 6 obsolete files)
1019.90 KB,
application/rar
|
Details | |
1.70 KB,
patch
|
sotaro
:
review+
|
Details | Diff | Splinter Review |
Precondition : Good wifi connection. (I think this problem is hard to be reproduced when connection is not good) STR Play any video without UI (Without title and progress bar). While playing, video frame is stuck for a while. Sometimes, it stopped until I touch the screen to show UI. And sometimes, it recovered after for a while, automatically. FYI Patch in bug 884188 is not applied on LG side because of regression. I backed out each of bug 891238, bug 885345, bug 888184. But I still see the symptom.
Reporter | ||
Updated•11 years ago
|
blocking-b2g: --- → leo+
Whiteboard: [TD-68993]
Comment 1•11 years ago
|
||
Strange. I think we saw something similar to this with bug 884188 in the build, but it was fixed when bug 884188 got backed out.
Assignee | ||
Comment 2•11 years ago
|
||
This could be different genlock error. I built a v1.1 leo ROM from today's source and confirmed that the following genlock. It happened at hw codec. It is different error compared to Bug 896286. ------------------------------------- 07-29 11:53:19.489 142 595 E libgenlock: perform_lock_unlock_operation: GENLOCK_IOC_DREADLOCK failed (lockType0x1, err=Connection timed out fd=37) 07-29 11:53:19.489 142 595 E omx_vdec: Failed to acquire genlock, ret = 1
Assignee | ||
Comment 3•11 years ago
|
||
The genlock error in comment #2 is fixed in Bug 879297 in the past.
Assignee | ||
Comment 4•11 years ago
|
||
Leo can you answer the question? - Is the status bar shown when doing "Play any video without UI"? - Does it happen when the app play a local video? - Can you attach logcat?
Flags: needinfo?(leo.bugzilla.gecko)
Assignee | ||
Comment 5•11 years ago
|
||
It needs to make clear which type of genlock happened at leo. Bug 896286 or Bug 879297.
Assignee | ||
Updated•11 years ago
|
Component: Gaia::Video → General
Assignee | ||
Comment 6•11 years ago
|
||
Need to confirm the problem happens at latest ROM on both hamachi(buri) and leo.
Keywords: qawanted
I was not able to reproduce this issue on both Leo and Buri devices.Was able to play the video without any obstructions and the video frame did not get stuck in the process. Tried playing three 20 min videos on both the devices. Environmental Variables Leo Moz RIL Build ID: 20130729070226 Gecko: http://hg.mozilla.org/releases/mozilla-b2g18/rev/8135299f3efd Gaia: 7aaffc8ccb6cf7ddd1e97943c108f1cb9eae5de0 Platform Version: 18.1 Firmware Version: D30008m Leaving qawanted for someone else to try and reproduce it.
QA Contact: dkumar
Reporter | ||
Comment 8•11 years ago
|
||
Flags: needinfo?(leo.bugzilla.gecko)
Reporter | ||
Comment 9•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #4) > Leo can you answer the question? > - Is the status bar shown when doing "Play any video without UI"? Yes, Status bar is shown normally. > - Does it happen when the app play a local video? I will check local file and inform you. > - Can you attach logcat? Please check attachment #782835 [details]. perform_lock_unlock_operation: GENLOCK_IOC_DREADLOCK failed (lockType0x1, err=Connection timed out fd=48)
Assignee | ||
Comment 10•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #5) > It needs to make clear which type of genlock happened at leo. Bug 896286 or > Bug 879297. From attachment 782835 [details], it becomes clear that the problem is similar to Bug 879297.
Assignee | ||
Comment 11•11 years ago
|
||
Two possibility about this problem. - Bug 879297 did not solve the all problem - Regression happened by some reason.
Assignee | ||
Comment 12•11 years ago
|
||
When Bug 884188 is committed again, regenerate of this bug might be more difficult.
Assignee | ||
Comment 13•11 years ago
|
||
I am going to confirm if Bug 879297 still works correctly.
Assignee: nobody → sotaro.ikeda.g
Assignee | ||
Comment 14•11 years ago
|
||
By applying the patch, confirmed that Bug 879297 is still working correctly and check when "GENLOCK_IOC_DREADLOCK failed" happens.
Assignee | ||
Comment 15•11 years ago
|
||
By attachment 783380 [details] [diff] [review], confirmed that Bug 879297 is working correctly even when "GENLOCK_IOC_DREADLOCK failed" happened.
Assignee | ||
Comment 16•11 years ago
|
||
logcat of until genlock failure. Applied attachment 783380 [details] [diff] [review] to ROM.
Assignee | ||
Comment 17•11 years ago
|
||
logcat of until genlock failure. Applied attachment 783380 [details] [diff] [review] to ROM.
Assignee | ||
Comment 18•11 years ago
|
||
From the logs and investigations, the following become clear. - Patch of Bug 879297 works correctly. After video frame rendering, a Texture was bound to mDummyGrallocBuffer correctly. - genlock failure happened short time after rendering change from Open GL to Hw Composer. - genlock failure happened always during Hw Composer rendering. - Problem did not happen, when video rendering is restricted only to Hw Composer rendering(disabling video GL rendering) by modifying the ROM.
Updated•11 years ago
|
Whiteboard: [TD-68993] → [TD-68993] [SR 01255148]
Comment 19•11 years ago
|
||
I believe the intention of bug 884188 is to make it so all video playback frames (including youtube video) are always rendered with HW composer. If that's so, then this bug should be also solved.
Assignee | ||
Comment 20•11 years ago
|
||
Suplement to comment #18 - genlock failure continued until next OpenGL rendering.
Assignee | ||
Comment 21•11 years ago
|
||
(In reply to Diego Wilson [:diego] from comment #19) > I believe the intention of bug 884188 is to make it so all video playback > frames (including youtube video) are always rendered with HW composer. If > that's so, then this bug should be also solved. Even when bug 884188 is fixed, video is sometimes rendered by OpenGL, when user touches UI or UI changed by some timing on youtube web site.
Assignee | ||
Comment 22•11 years ago
|
||
There was alway a pattern in the logs before output genlock failure. - ALL last OpenGL rendered video frame buffers before genlock failure were also rendered by Hw Composer.
Comment 23•11 years ago
|
||
Ah yes, you are correct. I feel like this is a regressions. I'm pretty sure I tested switching back and forth between HWC and GPU back when we were fixing long youtube videos.
Assignee | ||
Comment 24•11 years ago
|
||
Yeah, I also feel this is a regression.
Flags: needinfo?(leo.bugzilla.gecko)
Assignee | ||
Updated•11 years ago
|
Flags: needinfo?(leo.bugzilla.gecko)
Reporter | ||
Comment 25•11 years ago
|
||
I also think that this is a regression. This problem happen in current QE. And in last QE (1th July ~ 12th July), I didn't report this kind of issue. (It might not be detected then.) So, I have checked changes from 1th July and backed out several suspicious commits. However, I didn't figure out, yet.
Assignee | ||
Comment 26•11 years ago
|
||
Diego, do you have an assumption about the cause of the problem?
Flags: needinfo?(dwilson)
Assignee | ||
Comment 27•11 years ago
|
||
(In reply to leo.bugzilla.gecko from comment #25) > I also think that this is a regression. > > This problem happen in current QE. > And in last QE (1th July ~ 12th July), I didn't report this kind of issue. > (It might not be detected then.) I checked 2 moz built ROMs(June/10, June/28), but still see the problem. And Bug 879297 is fixed on June/06. From them, I re-think that it is not a regression. The problem is just not recognized by us. Actually, until Bug 879297 was reported by Diego, no one care about it. We had a lot of Youtube playback problems at that time and we might not care about it so much.
Assignee | ||
Comment 28•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #27) > (In reply to leo.bugzilla.gecko from comment #25) > > I also think that this is a regression. > > > > This problem happen in current QE. > > And in last QE (1th July ~ 12th July), I didn't report this kind of issue. > > (It might not be detected then.) > > I checked 2 moz built ROMs(June/10, June/28), Sorry, I failed to write ROM(June/10). So I confirmed only June/28. I am going to check other ROMs.
Assignee | ||
Comment 29•11 years ago
|
||
The genlock faileure happened on ROMs of June/11, June/14, June/28. v1.1 leo ROMs until June/10 have a problem(can not use data communication of WLAN/RIL), therefore I can not check the problem until June 10.
Assignee | ||
Comment 30•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #29) > The genlock faileure happened on ROMs of June/11, June/14, June/28. v1.1 leo > ROMs until June/10 have a problem Correction: v1.1 leo ROMs -> v1.1 leo ROMs stored at mozilla's repository.
Assignee | ||
Comment 31•11 years ago
|
||
Although I can not confirm the problem from June/06-June/10 because of the mozilla repository's v1.1 leo ROM's problem. From comment #29, I think it is not a regression. The problem was present and hidden by other youtube streaming problems.
Assignee | ||
Comment 32•11 years ago
|
||
From Comment 18, gecko works correctly to unbind the gralloc buffers from OpenGL texture. But the genlock failure happened at the situation witten in Comment 22. And from Comment 29, the problem also happens at ROMS of June/11, June/14 June/28. Therefore I feel that it is not a gecko's defect.
Assignee | ||
Comment 33•11 years ago
|
||
Leo, how do you think about it?
Flags: needinfo?(leo.bugzilla.gecko)
Assignee | ||
Comment 34•11 years ago
|
||
I feel following STR is easier to re-generate the genlock failure. -[1] Visit youtube web site by browser app -[2] Select relatively long video and start playback -[3] Sometimes change the playback position by seeking. repeat [3] until genlock failure happens.
Reporter | ||
Comment 35•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #33) > Leo, how do you think about it? I cannot understand your comment. So if this problem is not caused by Gecko.. Does it come from BSP or other side?
Flags: needinfo?(leo.bugzilla.gecko)
Reporter | ||
Comment 36•11 years ago
|
||
Patch from bug 887454 is not landed on LG side, yet. Sotaro, Did you see the same problem after using it?
Flags: needinfo?(sotaro.ikeda.g)
Assignee | ||
Comment 37•11 years ago
|
||
(In reply to leo.bugzilla.gecko from comment #35) > (In reply to Sotaro Ikeda [:sotaro] from comment #33) > > Leo, how do you think about it? > > I cannot understand your comment. > > So if this problem is not caused by Gecko.. > Does it come from BSP or other side? Yes. From comment #32, I became clear that it is not a regression and gecko side correctly try to unbind the gralloc buffer. Somehow there is a case that qcom's side code does not unbind the buffer in the following case even when gecko side try to unbind the gralloc buffer. - ALL last OpenGL rendered video frame buffers before genlock failure were also rendered by Hw Composer.
Flags: needinfo?(sotaro.ikeda.g)
Assignee | ||
Comment 38•11 years ago
|
||
We do not have a qcom's side code. We can not investigate about why qcom side does not unbind the gralloc buffer in the situation even when gecko side try to unbind the buffer.
Assignee | ||
Comment 39•11 years ago
|
||
(In reply to leo.bugzilla.gecko from comment #36) > Patch from bug 887454 is not landed on LG side, yet. > > Sotaro, > Did you see the same problem after using it? Yes.
Assignee | ||
Comment 40•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #39) > (In reply to leo.bugzilla.gecko from comment #36) > > Patch from bug 887454 is not landed on LG side, yet. > > > > Sotaro, > > Did you see the same problem after using it? > > Yes. I saw the problem both before and after applying the patch of bug 887454.
Assignee | ||
Comment 41•11 years ago
|
||
There is a difference the way of binding a gralloc buffer to a GL Texture. In android, SurfaceFlinger bind gralloc buffer very drawing update. It does not care that a Layer(Surface) will be rendered by hw composer or OpenGL. On the other hand, Firefox OS bind a graloc buffer to GL Texture only when a Layer is drawn by OpenGL and then bind to a dummy gralloc buffer soon after the Open GL drawing. This difference might affect to the problem. http://androidxref.com/4.0.4/xref/frameworks/base/services/surfaceflinger/Layer.cpp#403
Assignee | ||
Comment 43•11 years ago
|
||
(In reply to Jason Smith [:jsmith] from comment #42) > comment 34 has STR, so I'm pulling qawanted. For the comfirmation of ROMs(June/11, June/14), I flashed only gecko and gaia, but not flash ril. Because there days of ROM's ril did not work correctly.
Assignee | ||
Comment 44•11 years ago
|
||
Summarize the analysis of the problem until now. -[1] When gen lock failed happens, there were always a same pattern. + ALL last OpenGL rendered video frame buffers before genlock failure were also rendered by Hw Composer. -[2] From gecko, only thing to unlock genlock of gralloc buffers they are bounded to OpenGL Texture is following. + Bind another gralloc buffer to the texture. -[3] A similar genlock problem was fixed in Bug 879297. + It still works correctly. + It does [2] by using a dummy gralloc buffer. -[4] It is not a regression. + Bug 879297 is fixed on June/06. + A similar genlock failure happens also on the ROMs of June/11, June/14 June/28. -[5] The problem happens even when [3] works correctly.
Assignee | ||
Comment 45•11 years ago
|
||
I created the patch just to check if more agressive bind to dummy buffer could solve the problem. Because, the problem happen durin hw compsitor rendeirng and during the rendering no binding is done. The patch try to bind dummy buffer even during hw compositor rendeirng. The Result: The patch does not fix the problem.
Assignee | ||
Comment 46•11 years ago
|
||
bjacob, do you have any ideas what we can do from gecko side about this problem?
Flags: needinfo?(bjacob)
Comment 47•11 years ago
|
||
(Note: I am in France, so it's late today; I'm off tomorrow and on Tuesday; so if you need a quick reply, you might want to ask someone else, such as: Jeff Muizelaar, Peter Chang, or Nicolas Silva. Sorry!)
Comment 48•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #41) > There is a difference the way of binding a gralloc buffer to a GL Texture. > In android, SurfaceFlinger bind gralloc buffer very drawing update. It does > not care that a Layer(Surface) will be rendered by hw composer or OpenGL. > > On the other hand, Firefox OS bind a graloc buffer to GL Texture only when a > Layer is drawn by OpenGL and then bind to a dummy gralloc buffer soon after > the Open GL drawing. Nicolas Silva has recently landed on mozilla-central a patch making us not use a dummy gralloc buffer anymore; so maybe he'll have ideas about that here too. Flagging 'needinfo'.
Flags: needinfo?(bjacob) → needinfo?(nical.bugzilla)
Assignee | ||
Comment 49•11 years ago
|
||
Addition to the summary of comment #44 - [6] The genlock problem do not happen when hw composer is disabled. - [7] The genlock problem do not happen when vidoe frame is rendered only by HwComposer.
Comment 50•11 years ago
|
||
(In reply to Benoit Jacob [:bjacob] from comment #48) > (In reply to Sotaro Ikeda [:sotaro] from comment #41) > > There is a difference the way of binding a gralloc buffer to a GL Texture. > > In android, SurfaceFlinger bind gralloc buffer very drawing update. It does > > not care that a Layer(Surface) will be rendered by hw composer or OpenGL. > > > > On the other hand, Firefox OS bind a graloc buffer to GL Texture only when a > > Layer is drawn by OpenGL and then bind to a dummy gralloc buffer soon after > > the Open GL drawing. > > Nicolas Silva has recently landed on mozilla-central a patch making us not > use a dummy gralloc buffer anymore; so maybe he'll have ideas about that > here too. Flagging 'needinfo'. IIRC the idea was to let the driver unbind the previous gralloc buffer when binding the next one to the same GL texture (which is what surface flinger does). Now (m-c) we have one GL texture per texture unit globally (post-layers-refactoring it is stored in CompositorOGL, pre-refactoring the GL layer manager would be the best candidate) rather than having a GL texture per texture unit and per layer. The changeset is: http://hg.mozilla.org/mozilla-central/rev/3c7c31aaea78 The motivation for this was that this behavior yields better performances than what we were doing before (and still do in b2g18). It was not meant to fix a particular bug so I don't know for sure that it would help in this case. Although, reading the comments it looks like the problem is that we are doing things in a way that the driver has not been designed for, so maybe doing a similar change could help (since it brings us closer to what SurfaceFlinger does)
Flags: needinfo?(nical.bugzilla)
Comment 51•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #44) > -[1] When gen lock failed happens, there were always a same pattern. > + ALL last OpenGL rendered video frame buffers > before genlock failure were also rendered by Hw Composer. > -[3] A similar genlock problem was fixed in Bug 879297. > + It still works correctly. > + It does [2] by using a dummy gralloc buffer. In bug 879297 the issue was that the GPU held a buffer while switching to HWC rendering. If [1] is correct, this means the new issue is that HWC is holding a buffer while switching to GPU rendering.
Flags: needinfo?(dwilson)
Assignee | ||
Comment 52•11 years ago
|
||
By using attachment 783380 [details] [diff] [review], I confirmed again that [1] happened on v1.1 leo. From comment #51, to analyze the problem, investigation of internal of HWC is necessary. I am not correct person for it. I un-assign myself from this bug.
Assignee: sotaro.ikeda.g → nobody
Comment 53•11 years ago
|
||
Jonas, CJ, who would be the the right person for "investigation of HWC is necessary"?
Flags: needinfo?(jonas)
Flags: needinfo?(cku)
Assignee | ||
Comment 54•11 years ago
|
||
I reminds that there is still a few thing need to check in gecko side. I am going to take the bug again a bit more.
Assignee: nobody → sotaro.ikeda.g
Comment 55•11 years ago
|
||
Loop in Morris, who is working on 4.3 HWC porting now
Assignee | ||
Comment 56•11 years ago
|
||
Current B2G do not call FramebufferNativeWindow::compositionComplete().
Assignee | ||
Comment 57•11 years ago
|
||
The patch adds a call to FramebufferNativeWindow::compositionComplete().
Assignee | ||
Comment 58•11 years ago
|
||
By applying attachment 789966 [details] [diff] [review] to v1.1 leo, I still do can not reproduce the genlock failure until now.
Assignee | ||
Comment 59•11 years ago
|
||
On ROM with attachment 789966 [details] [diff] [review], the genlock failure did not happen on my leo device by STR in comment #34. It seems that the patch fix the problem.
Assignee | ||
Comment 60•11 years ago
|
||
Add Comments and nit fix.
Attachment #787710 -
Attachment is obsolete: true
Attachment #789966 -
Attachment is obsolete: true
Assignee | ||
Comment 61•11 years ago
|
||
Comment on attachment 790230 [details] [diff] [review] patch v2 - Add calling compositionComplete() jrmuizel, can I have a feed back for the patch?
Attachment #790230 -
Flags: feedback?(jmuizelaar)
Assignee | ||
Comment 62•11 years ago
|
||
I tested the patch (In reply to Sotaro Ikeda [:sotaro] from comment #59) > On ROM with attachment 789966 [details] [diff] [review], the genlock failure > did not happen on my leo device by STR in comment #34. It seems that the > patch fix the problem. Today, I tested longer duration than yesterday. Still the genlock failure did not happen.
Comment 63•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #57) > Created attachment 789966 [details] [diff] [review] > patch - Add calling compositionComplete() > > The patch adds a call to FramebufferNativeWindow::compositionComplete(). That's quite the catch there Sotaro! How did you figure this out? Looking at surfaceflinger?
Assignee | ||
Comment 64•11 years ago
|
||
(In reply to Diego Wilson [:diego] from comment #63) > That's quite the catch there Sotaro! How did you figure this out? Looking at > surfaceflinger? Yes, I compared B2G to the ICS's SurfaceFlinger.
Updated•11 years ago
|
Attachment #790230 -
Flags: feedback?(jmuizelaar) → feedback+
Assignee | ||
Updated•11 years ago
|
Attachment #790230 -
Flags: review?(bgirard)
Comment 65•11 years ago
|
||
Comment on attachment 790230 [details] [diff] [review] patch v2 - Add calling compositionComplete() Review of attachment 790230 [details] [diff] [review]: ----------------------------------------------------------------- r+ with the following changed. ::: gfx/layers/opengl/LayerManagerOGL.cpp @@ +1043,5 @@ > > +#ifdef MOZ_WIDGET_GONK > + // Should be called when composition rendering is complete for a frame. > + // Required by certain older drivers for synchronization. > + // XXX Recent HWCs do not need the call What happens to HWCs that don't need this when you call it? It should be noted in the comment here because it leaves the reader wondering. This code should be moved to mWidget DrawWindowOverlay.
Attachment #790230 -
Flags: review?(bgirard) → review+
Assignee | ||
Comment 66•11 years ago
|
||
(In reply to Benoit Girard (:BenWa) from comment #65) > > What happens to HWCs that don't need this when you call it? It should be > noted in the comment here because it leaves the reader wondering. Yeah, the above is not clear. I will update it in next patch. And this bug only care about B2G v1.1. For master, I am going to create another bug for it. On ICS gonk it is necessary always to call CompositionComplete(), On JB4.3 gonk, it is decided by HWC's version. HWC could be 3 versions(v1.0, v1.1 v1.2). CompositionComplete() is called only v1.0 HWC. http://androidxref.com/4.3_r2.1/xref/frameworks/native/services/surfaceflinger/DisplayHardware/HWComposer.cpp#758 > > This code should be moved to mWidget DrawWindowOverlay. Will fix in next patch.
Assignee | ||
Comment 67•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #66) > > This code should be moved to mWidget DrawWindowOverlay. > > Will fix in next patch. In an offline talk, it is going to keep stay in same place.
Assignee | ||
Comment 68•11 years ago
|
||
Attachment #790230 -
Attachment is obsolete: true
Assignee | ||
Comment 69•11 years ago
|
||
Comment on attachment 790356 [details] [diff] [review] patch v3 for b2g18 - Add calling compositionComplete() Carry "bgirard review+".
Attachment #790356 -
Flags: review+
Assignee | ||
Updated•11 years ago
|
Attachment #783380 -
Attachment is obsolete: true
Assignee | ||
Updated•11 years ago
|
Attachment #783382 -
Attachment is obsolete: true
Assignee | ||
Updated•11 years ago
|
Attachment #783384 -
Attachment is obsolete: true
Assignee | ||
Comment 70•11 years ago
|
||
Created Bug 905304 for master.
Assignee | ||
Comment 72•11 years ago
|
||
Leo, can you check if attachment 790356 [details] [diff] [review] works? Anyway, the patch is necessary.
Flags: needinfo?(leo.bugzilla.gecko)
Comment 73•11 years ago
|
||
https://hg.mozilla.org/releases/mozilla-b2g18/rev/09a8c5b6b287
Status: NEW → RESOLVED
Closed: 11 years ago
status-b2g18:
--- → fixed
status-b2g18-v1.0.0:
--- → wontfix
status-b2g18-v1.0.1:
--- → wontfix
status-b2g-v1.1hd:
--- → affected
status-firefox26:
--- → unaffected
Keywords: checkin-needed
Resolution: --- → FIXED
Whiteboard: [TD-68993] [SR 01255148] → [TD-68993][SR 01255148]
Reporter | ||
Comment 75•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #72) > Leo, can you check if attachment 790356 [details] [diff] [review] works? > Anyway, the patch is necessary. I cannot see the symptom after patching it!!
Flags: needinfo?(leo.bugzilla.gecko)
Comment 76•11 years ago
|
||
(In reply to Sotaro Ikeda [:sotaro] from comment #56) > Current B2G do not call FramebufferNativeWindow::compositionComplete(). Please refer this bug. https://bugzilla.mozilla.org/show_bug.cgi?id=883006 FramebufferNativeWindow::compositionComplete calls glFinish finally. And we fixed this bug by adding glFinish too.
Assignee | ||
Comment 77•11 years ago
|
||
ying.xu, thank for the info. FramebufferNativeWindow::compositionComplete() actually end up to calling glFinish on qcom platform. We choose not to call glFinish() directly but call FramebufferNativeWindow::compositionComplete(), because the api is provided by android to abstract the hw's difference. And in JB, the compositionComplete() is not called when HWC's version is v1.1 or more. http://androidxref.com/4.3_r2.1/xref/frameworks/native/services/surfaceflinger/DisplayHardware/DisplaySurface.h#35 http://androidxref.com/4.3_r2.1/xref/frameworks/native/services/surfaceflinger/DisplayHardware/HWComposer.cpp#758
Assignee | ||
Comment 78•11 years ago
|
||
This bug is already fixed. needinfo is not necessary now.
Flags: needinfo?(jonas)
You need to log in
before you can comment on or make changes to this bug.
Description
•