870198 - Support basic display functions on gonk-JB

Status: RESOLVED FIXED

(Core Graveyard :: Widget: Gonk, defect)

Description

[Michael Wu [:mwu]]

Attachment: WIP: Abstract some display functions behind GonkDisplay, and implement for JB

JB introduces a number of changes to graphics system. This patch introduces the GonkDisplay interface to abstract away these changes so we can support both JB and ICS.

Comment 1

[Michael Wu [:mwu]]

Attachment: WIP: Abstract some display functions behind GonkDisplay, v2

Added ICS support. Need to add HWC support back before this is ready.

Comment 2

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v3

HWC support put back

Comment 3

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v4 (for review)

This patch introduces a class named GonkDisplay which abstracts away the differences between ICS and JB. The boot animation code is now also called from GonkDisplay. The GonkDisplay and boot animation code is bundled up and linked into the b2g binary so the boot animation can start as soon as possible.

On ICS, FramebufferNativeWindow is used to both create gralloc'd buffers for drawing and providing a ANativeBuffer which EGL uses to grab buffers.

On JB, the Nexus 4 requires things to be drawn via HWC. There, we use FramebufferSurface/SurfaceTextureClient/BufferQueue to replace the function of FramebufferNativeWindow. SurfaceTextureClient provides a ANativeWindow so EGL can draw to buffers, and when buffers are ready to be shown on screen, FramebufferSurface provides them to GonkDisplay. We used to get a EGLDisplay and EGLSurface before, but now we get the buffer_handle_t's directly. GonkDisplay then creates a barebones display list for HWC to render the buffer.

HWC 1.1 and later require all this magic. HWC 1.0 doesn't AFAICT. It still tries to extract buffer handles out of EGLSurfaces. I don't think this code supports HWC 1.0 / no HWC right now, though romaxa is trying to get something working on the Nexus 7.

Comment 4

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v4

Comment 5

[Michael Wu [:mwu]]

https://tbpl.mozilla.org/?tree=Try&rev=cfc267721633

Comment 6

[Michael Wu [:mwu]]

Comment on attachment 747698 [details] [diff] [review]
Abstract some display functions behind GonkDisplay, v4 (for review)

Review of attachment 747698 [details] [diff] [review]:
-----------------------------------------------------------------

::: widget/gonk/libdisplay/BootAnimation.cpp
@@ +416,5 @@
>          return nullptr;
>      }
>  
> +    int format = GetGonkDisplay()->surfaceformat;
> +    GonkDisplay *display = GetGonkDisplay();

Need to reverse the order of these two lines.

::: widget/gonk/libdisplay/BootAnimation.h
@@ +1,1 @@
> +#ifndef BOOTANIMATION_H

Need to add license boilerplate.

::: widget/gonk/libdisplay/GonkDisplay.h
@@ +27,5 @@
> +    virtual void SetEnabled(bool enabled) = 0;
> +
> +    virtual void* GetHWCDevice() = 0;
> +
> +    virtual bool SwapBuffers(void* dpy, void* sur) = 0;

Need to use EGLSurface/EGLDisplay..

::: widget/gonk/libdisplay/Makefile.in
@@ +35,5 @@
> +	$(NULL)
> +else ifeq (15,$(ANDROID_VERSION))
> +CPPSRCS += GonkDisplayICS.cpp
> +else
> +$(error Unsupported )

Forgot to finish this sentence.

Comment 7

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v4 -> v5

Comment 8

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v5 (for review)

Comment 9

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v5

Comment 11

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Comment on attachment 748170 [details] [diff] [review]
Abstract some display functions behind GonkDisplay, v5 (for review)

Review of attachment 748170 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/gl/GLContextProviderEGL.cpp
@@ +273,5 @@
>  
>  #ifdef DEBUG
>          printf_stderr("Initializing context %p surface %p on display %p\n", mContext, mSurface, EGL_DISPLAY());
>  #endif
> +#if defined(MOZ_WIDGET_GONK) && ANDROID_VERSION == 15

may want to prefer < or > the appropriate version here instead of explicit == 15, just in case.

@@ +1593,5 @@
>      LOCAL_EGL_GREEN_SIZE,      8,
>      LOCAL_EGL_BLUE_SIZE,       8,
>      LOCAL_EGL_ALPHA_SIZE,      8,
> +#if defined(MOZ_WIDGET_GONK) && ANDROID_VERSION >= 17
> +    /* EGL_FRAMEBUFFER_TARGET_ANDROID */ 0x3147, LOCAL_EGL_TRUE,

add EGL_FRAMEBUFFER_TARGET_ANDROID to GLDefs.h?

::: widget/gonk/libdisplay/GonkDisplayICS.cpp
@@ +42,5 @@
> +    // framebuffer device after we restart with the screen off.
> +    //
> +    // NB: this *must* run BEFORE allocating the
> +    // FramebufferNativeWindow.  Do not separate these two C++
> +    // statements.

Can we update this comment, because we did, in fact, separate the two C++ statements? :)

Comment 12

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v5 -> v6

Comment 13

[Michael Wu [:mwu]]

Attachment: Abstract some display functions behind GonkDisplay, v6 -> v7

A couple of fixes to make everything ok on ICS. I forgot to zero a new variable in the constructor, and ICS doesn't need to link against some of the new libraries JB uses.

Comment 14

[Michael Wu [:mwu]]

https://hg.mozilla.org/projects/birch/rev/875188dad4bf

Comment 15

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/875188dad4bf

Comment 16

[Vincent Lin[:vilin]]

(In reply to Michael Wu [:mwu] from comment #14)
> https://hg.mozilla.org/projects/birch/rev/875188dad4bf

Hi Michael,

It seams Gonk just supports older version of HWC(for ICS) so far ?
In the latest version of HWC(for JB), hwcomposer device is constructed by hwc_composer_device_1_t which is quite different from hwc_composer_device_t.

HwcComposer2D::Init(hwc_display_t dpy, hwc_surface_t sur)

...

if (int err = init()) {
   mHwc = (hwc_composer_device_t*)GetGonkDisplay()->GetHWCDevice();
   if (!mHwc) {
      LOGE("Failed to initialize hwc");
      return err;
      return -1;
   }

I'm just tracing the code and find this Bug ID by coincidence.
Not sure if it's proper to post here.
Sorry for bothering.