Android mesa-10.6 branch is ready for testing

[Rick Carroll]

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org
Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org
Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org

Is building Haswell Graphics drivers into this possible? I am getting some crashes in the latest release with HW accelerated video.

[rbg]

need a hint here....getting the following build error..

Gen ES: libmesa_st_mesa <= dispatch.h
Traceback (most recent call last):
File "external/mesa/src/mesa/main/format_pack.py", line 3, in <module>
from mako.template import Template
ImportError: No module named mako.template
make: ***
[out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_pack.c]
Error 1
make: *** Deleting file
`out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_pack.c'
make: *** Waiting for unfinished jobs....
Traceback (most recent call last):
File "external/mesa/src/mesa/main/format_unpack.py", line 3, in <module>
from mako.template import Template
ImportError: No module named mako.template
Gen ES: libmesa_st_mesa <= remap_helper.h
make: ***
[out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_unpack.c]
Error 1
make: *** Deleting file
`out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_unpack.c'

#### make failed to build some targets (02:54 (mm:ss)) ####

did a complete repo sync yesterday, have done the

To test it (on top of android-x86 tree):

cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..

then tried just pulling the snapshot from git and got same issue....

then built with mrossi's 10.4 multiarec which worked fine.....

so before I go digging into it figured I was overlooking something....thanks

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org

[rbg]

Aha....was too focused on the format_pack.c that doesn't exist in 10.6 but
does exist in 10.4 / 9.2....which both compile fine.....

So this is a new environment dependency...I have added it to my AOSP env
build script which has been
fine for 3-4 years....thank you

not familiar with Python but you would think maybe a warning, conditional
would be present when the environment
it is running in doesn't have the dependencies it needs....such as what you
did for java 6 / java 7...!!!

thanks again

Post by rbg

Post by rbg
need a hint here....getting the following build error..
Gen ES: libmesa_st_mesa <= dispatch.h
File "external/mesa/src/mesa/main/format_pack.py", line 3, in <module>
from mako.template import Template
ImportError: No module named mako.template
make: ***

[out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_pack.c]

Post by rbg
Error 1
make: *** Deleting file

`out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_pack.c'

Post by rbg
make: *** Waiting for unfinished jobs....
File "external/mesa/src/mesa/main/format_unpack.py", line 3, in

<module>

Post by rbg
from mako.template import Template
ImportError: No module named mako.template

It said you need mako.template
In Ubuntu,
apt-get install python-mako

Post by rbg
Gen ES: libmesa_st_mesa <= remap_helper.h
make: ***

[out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_unpack.c]

Post by rbg
Error 1
make: *** Deleting file

`out/target/product/x86/gen/STATIC_LIBRARIES/libmesa_st_mesa_intermediates/main/format_unpack.c'

Post by rbg
#### make failed to build some targets (02:54 (mm:ss)) ####
did a complete repo sync yesterday, have done the
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
then tried just pulling the snapshot from git and got same issue....
then built with mrossi's 10.4 multiarec which worked fine.....
so before I go digging into it figured I was overlooking

something....thanks

[rbg]

The error is clear enough to me.

Well it may have been to me 30 years ago when I did it for a living...but
now
it takes me all day what I use to do in an hour..................LOL

Post by rbg
Aha....was too focused on the format_pack.c that doesn't exist in 10.6

but

Post by rbg
does exist in 10.4 / 9.2....which both compile fine.....
So this is a new environment dependency...I have added it to my AOSP env

Right... But that's a dependency required by
the new mesa 10.6 (or 10.5?), not by AOSP.

Post by rbg
build script which has been
fine for 3-4 years....thank you
not familiar with Python but you would think maybe a warning,

conditional

Post by rbg
would be present when the environment
it is running in doesn't have the dependencies it needs....such as what

you

Post by rbg
did for java 6 / java 7...!!!

The error is clear enough to me.
I think I got the error before but
have it fixed silently and then forgot it.

[Luigi Capriotti]

Post by rbg
Well it may have been to me 30 years ago when I did it for a living...but
now
it takes me all day what I use to do in an hour..................LOL

I hear you loud and clear! LOOL

[rbg]

forgot...am building on 5.1.0_r3

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org

[rbg]

(is there a way to switch between classical and gallium
drivers at runtime as before?)

this rang a bell but not sure what you meant by as before.....BUT

I remember somewhere, think it was when testing all the broadcom wifi
issues, that I could
do a blacklist on the command line when running "live usb" to eliminate
certain drivers from
being loaded....don't think that's what your looking for but will mention
it...

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org

[Mauro Rossi]

Hi Chih-Wei,

even if I could guess some parts Android.mk for dri state-tracker and
targets,
the loader/sym-linking changes in android specific code were beyond my
possibilities, thanks for solving the problem.
(for me also new lessons to learn)

I have tested kitkat-x86 with nouveau i.e. mesa EGL + static
gallium_dri.so (with nouveau gallium driver).

The results with nouveau at this point are the pretty much same for mesa
10.5.3 with egl_gallium and for mesa 10.6.0-devel, so "gallium_dri" mission
accomplished,
thank you very much.

I will provide futher info in the nouveau specif thread in this forum.

Mauro

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6

[Ravid Maduni]

Someone already build an iso on 5.1.0 with 4.0 kernel?

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org

[Rick Carroll]

Yes... but no Gapps. What you got?

Post by Ravid Maduni
Someone already build an iso on 5.1.0 with 4.0 kernel?

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org

--
You received this message because you are subscribed to a topic in the
Google Groups "Android-x86" group.
To unsubscribe from this topic, visit
https://groups.google.com/d/topic/android-x86/p1Qy_0t9efE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to
Visit this group at http://groups.google.com/group/android-x86.
For more options, visit https://groups.google.com/d/optout.

[Ravid Maduni]

My PC has 2.9 GH dual core cpu, gt 220 gpu and 4 GB RAM

[Rick Carroll]

No, i thought you were also compiling. The drivers for Hw Accel on your
chip aren't included in this mesa yet either. You've gotta recompile them
in. Straight OpenGL might work for you since nvidia has always had good
support for that, tho.

[The Black Death]

Ya RBG has a Build On http://www.android-x86.net/ in the Test Builds of the
download section, and as stated,you have to install Gapps yourself..

TBD

Post by Ravid Maduni
Someone already build an iso on 5.1.0 with 4.0 kernel?

Hi all,
I've pushed the mesa-10.6 branch to the git server.
Please help to review and test it.
http://git.android-x86.org/?p=platform/external/mesa.git;a=shortlog;h=refs/heads/mesa-10.6
It based on mesa 10.6-devel branch + Emil's
submit/android-fixes#1 patches (minor fixes) +
my patches to fix all building issues of
kitkat-x86 and lollipop-x86 for both 32- & 64-bit targets.
classic drivers: i915 i965
gallium drivers: i915g ilo nouveau r300g r600g swrast vmwgfx
The most significant change is the building rules of
gallium drivers. Since the mesa upstream has removed
the st/egl code (superseded by st/dri) used by android-x86,
I have created new Android.mk to build st/dri and the
related libraries. As a result, /system/lib/dri/gallium_dri.so
is created.
cd external/mesa
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
cd ../..
Then build the iso_img or efi_img as usual.
I've tested the gallium drivers i915g, ilo, r600g, vmwgfx
successfully on lollipop-x86 for both 32- & 64-bit targets.
It can pass the Antutu 5.7 tests.
The classical drivers was disabled to ensure
the gallium drivers are used in the tests.
The Antutu shows the OpenGL renderer
is gallium 0.4 as expected.
(is there a way to switch between classical and gallium
drivers at runtime as before?)
* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.
* Video playback is black screen.
Seems due to mesa lack of YV12/NV12 support.
I've seen peyo's ibox porting has tried to address
the issue, but not complete yet. Refer to
https://github.com/peyo-hd/external_mesa3d/commit/70c3231499b8fcb6b66be63cdfed96fac3997064
https://github.com/peyo-hd/external_mesa3d/commit/b35cda41e91626ff70c56ad837d25fd131be5669
By applying peyo's patches the video becomes
gray in the classical driver, but no help to
the gallium driver
* Enabling x86(-64) assembly causes
some apps (e.g., Gallery3D) crashing.
The issue is gone by disabling the x86-asm.
(To disable it, revert commit 1b7b6587)
* OpenGL info can't be shown in the Settings.
All info shown are null.
http://git.android-x86.org/?p=platform/packages/apps/Settings.git;a=commit;h=b2dd5e454215f673525bc6ab9ff96f3ab0ea94f3
(not mesa 10.6 specific, happens in all versions)
The same branch is also buildable with kitkat-x86.
I haven't spent much time to test it yet.
But the above issues don't exist in kitkat-x86.
I consider the first two major issues are
the gating bugs of lollipop-x86.
We need more experts of mesa and/or
android framework to conquer them.
Once the issues are resolved we may have
a stable lollipop-x86 release soon.
PS. Anybody knows why the gallium driver
is still version 0.4 as mesa 9.2?
--
Chih-Wei
Android-x86 project
http://www.android-x86.org

[pstglia]

* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.

As pointed, this crashes graphical. Appears something in mesa code crashes
PowerManagerService. This is tombstone output:

pid: 1783, tid: 1816, name: PowerManagerSer >>> system_server <<<
signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x10
eax 00000000 ebx 9b22ad0c ecx 9a598a14 edx 9a4043a8
esi a21765f0 edi 9a4043a8
xcs 00000073 xds 0000007b xes 0000007b xfs 000000af xss 0000007b
eip 9af4c21f ebp 9a598a14 esp a21764b0 flags 00210282

backtrace:
#00 pc 0042121f /android/system/lib/dri/gallium_dri.so
#01 pc 00197777 <unknown>
#02 pc 0040962f /android/system/lib/dri/gallium_dri.so

stack:
a2176470 9a529b64
a2176474 a21765f0 [stack:1816]
a2176478 b752a0d9 /system/lib/libc.so (je_calloc+9)
a217647c 9adbd2f6 /android/system/lib/dri/gallium_dri.so
a2176480 9a5974d4
a2176484 9a5974d4
a2176488 9a598954
a217648c 9adbd2f6 /android/system/lib/dri/gallium_dri.so

I used addr2line in order to obtain function names associated to pc
addresses. If this procedure is correct, the last call was (note: I
disabled strip on my mesa libs):

bash-4.3$
/home/paulo/Downloads/android-x86-lollipop/prebuilts/gcc/linux-x86/host/x86_64-linux-glibc2.11-4.8/bin/x86_64-linux-addr2line
-f -e $OUT/obj/lib/gallium_dri.so 0042121f
_ZNK21ir_function_signature13function_nameEv
*/home/paulo/Downloads/android-x86-lollipop/external/mesa/src/glsl/ir.h:1127*

Which correspont to this line:

1125 inline const char *ir_function_signature::function_name() const
1126 {
*1127 return this->_function->name;*
1128 }

I tried also to include lots of debug messages
on frameworks/base/services/core/java/com/android/server/power/PowerManagerService.java
to see which was the last function called before the crash. It resulted on
this:

*# Latest calls for thread 1816:*
06-07 13:15:31.083 1783 1816 D PowerManagerService:
updateDisplayPowerStateLocked: mDisplayReady=false, policy=0,
mWakefulness=0, mWakeLockSummary=0x0, mUserActivitySummary=0x0,
mBootCompleted=true, mScreenBrightnessBoostInProgress=false
06-07 13:15:31.083 1783 1816 I PowerManagerService: PST DEBUG - Inside
updatePowerStateLocked - Before updateDreamLocked
06-07 13:15:31.083 1783 1816 I PowerManagerService: PST DEBUG - BUG
HUNTER - Inside updateDreamLocked
06-07 13:15:31.083 1783 1816 I PowerManagerService: PST DEBUG - Inside
updatePowerStateLocked - Before updateSuspendBlockerLocked
06-07 13:15:31.083 1783 1816 I PowerManagerService: PST DEBUG - BUG
HUNTER - Inside updateSuspendBlockerLocked
06-07 13:15:31.083 1783 1816 I PowerManagerService: PST DEBUG - BUG
HUNTER - Inside needDisplaySuspendBlockerLocked
06-07 13:15:31.083 1783 1816 I PowerManagerService: PST DEBUG - Inside
updateSuspendBlockerLocked - Before if
(mDecoupleHalInteractiveModeFromDisplayConfig)
06-07 13:15:31.084 1783 1816 I PowerManagerService: PST DEBUG - Inside
updateSuspendBlockerLocked - After if
(mDecoupleHalInteractiveModeFromDisplayConfig)
06-07 13:15:31.084 1783 1816 I PowerManagerService: PST DEBUG - Inside
updatePowerStateLocked - After updateSuspendBlockerLocked
06-07 13:15:31.084 1783 1816 I PowerManagerService: PST DEBUG - Inside
handleSandman - After updatePowerStateLocked (after
reallyGoToSleepNoUpdateLocked)
06-07 13:15:31.084 1783 1816 I PowerManagerService: PST DEBUG - End of
handleSandman

*# Latest calls before SIGSEGV*
06-07 13:15:31.127 1783 2641 I PowerManagerService: PST DEBUG - BUG
HUNTER - Inside powerHintInternal
06-07 13:15:31.127 1783 2641 I PowerManagerService-JNI: PST DEBUG - BUG
HUNTER JNI - Before int data_param = data
06-07 13:15:31.127 1783 2641 I PowerManagerService-JNI: PST DEBUG - BUG
HUNTER JNI - After int data_param = data
06-07 13:15:31.127 1783 2641 I PowerManagerService-JNI: PST DEBUG - BUG
HUNTER JNI - Before if(data)
06-07 13:15:31.127 1783 2641 I PowerManagerService-JNI: PST DEBUG - BUG
HUNTER JNI - if(data) is false - Before
gPowerModule->powerHint(gPowerModule, (power_hint_t)hintId, NULL)
06-07 13:15:31.127 1783 2641 I PowerManagerService-JNI: PST DEBUG - BUG
HUNTER JNI - End of nativeSendPowerHint
06-07 13:15:31.127 1783 2641 I PowerManagerService: PST DEBUG - BUG
HUNTER - End of powerHintInternal
06-07 13:15:31.127 1783 2641 I PowerManagerService: PST DEBUG - BUG
HUNTER - End of BinderService/powerHint
--------- beginning of crash
06-07 13:15:31.136 1783 1816 F libc : Fatal signal 11 (SIGSEGV), code
1, fault addr 0x10 in tid 1816 (PowerManagerSer)
06-07 13:15:31.187 1504 1504 I DEBUG : *** *** *** *** *** *** *** ***
*** *** *** *** *** *** *** ***
06-07 13:15:31.187 1504 1504 I DEBUG : Build fingerprint:
'Android-x86/android_x86/x86:5.1.1/LMY47V/paulo05050202:eng/test-keys'
06-07 13:15:31.187 1504 1504 I DEBUG : Revision: '0'
06-07 13:15:31.187 1504 1504 I DEBUG : ABI: 'x86'
06-07 13:15:31.187 1504 1504 I DEBUG : pid: 1783, tid: 1816, name:
PowerManagerSer >>> system_server <<<
06-07 13:15:31.187 1504 1504 I DEBUG : signal 11 (SIGSEGV), code 1
(SEGV_MAPERR), fault addr 0x10
06-07 13:15:31.189 1504 1504 I DEBUG : eax 00000000 ebx 9b22ad0c
ecx 9a598a14 edx 9a4043a8
06-07 13:15:31.189 1504 1504 I DEBUG : esi a21765f0 edi 9a4043a8
06-07 13:15:31.189 1504 1504 I DEBUG : xcs 00000073 xds 0000007b
xes 0000007b xfs 000000af xss 0000007b
06-07 13:15:31.189 1504 1504 I DEBUG : eip 9af4c21f ebp 9a598a14
esp a21764b0 flags 00210282
06-07 13:15:31.189 1504 1504 I DEBUG :
06-07 13:15:31.189 1504 1504 I DEBUG : backtrace:
06-07 13:15:31.189 1504 1504 I DEBUG : #00 pc 0042121f
/android/system/lib/dri/gallium_dri.so
06-07 13:15:31.189 1504 1504 I DEBUG : #01 pc 00197777 <unknown>
06-07 13:15:31.189 1504 1504 I DEBUG : #02 pc 0040962f
/android/system/lib/dri/gallium_dri.so
06-07 13:15:31.287 1504 1504 W libbacktrace: virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb752c82d
06-07 13:15:31.369 1504 1504 W libbacktrace: virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb69d5873

Blocking call to reallyGoToSleepNoUpdateLocked avoids crashing, but machine
never sleeps...

This is all I could get so far. Anyone has some advice on how to debug it?
Attached tombstone, logcat and 2 src files I inserted extra debugging msgs:
-
frameworks/base/services/core/java/com/android/server/power/PowerManagerService.java
-
frameworks/base/services/core/jni/com_android_server_power_PowerManagerService.cpp

Thanks,
Pstglia

[pstglia]

* Mesa crashes on android suspend,
no matter in the classical or gallium drivers.
This is not a 10.6 specific issue. It happens
in all mesa versions I've tested.

I could get a better backtrace now by striping more libs, running without
DEBUG=2 (this was giving me an incomplete output, don't know the reason)
and adding some ALOGI calls on suspect points. Bellow what I got:

I/PST-DEBUG-LNK( 2050): Inside ir_visitor_status::visit_enter - before
const ir_function_signature *const callee = ir->callee - addr is
0x9A9D9828, function_name addr is 0x9C9D6FF0
I/PST-DEBUG-LNK( 2050): Inside ir_visitor_status::visit_enter - after const
ir_function_signature *const callee = ir->callee
I/PST-DEBUG-LNK( 2050): Inside ir_visitor_status::visit_enter - before
const char *const name = callee->function_name()
--------- beginning of crash
F/libc ( 2050): Fatal signal 11 (SIGSEGV), code 1, fault addr 0x10 in
tid 2079 (PowerManagerSer)
I/DEBUG ( 1784): *** *** *** *** *** *** *** *** *** *** *** *** *** ***
*** ***
I/DEBUG ( 1784): Build fingerprint:
'Android-x86/android_x86/x86:5.1.1/LMY47V/paulo05050202:eng/test-keys'
I/DEBUG ( 1784): Revision: '0'
I/DEBUG ( 1784): ABI: 'x86'
I/DEBUG ( 1784): pid: 2050, tid: 2079, name: PowerManagerSer >>>
system_server <<<
I/DEBUG ( 1784): signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr
0x10
I/DEBUG ( 1784): eax 00000000 ebx 9ccb6cf4 ecx 7fe301d9 edx
7fe301d9
I/DEBUG ( 1784): esi a21915f0 edi 9a9d9814
I/DEBUG ( 1784): xcs 00000073 xds 0000007b xes 0000007b xfs
000000af xss 0000007b
I/DEBUG ( 1784): eip 9c9d72ef ebp 9cb74486 esp a21914a0 flags
00210286
I/DEBUG ( 1784):
I/DEBUG ( 1784): backtrace:
I/DEBUG ( 1784): #00 pc 004212ef /system/lib/dri/gallium_dri.so
((anonymous namespace)::call_link_visitor::visit_enter(ir_call*)+159)
I/DEBUG ( 1784): #01 pc 00417836 /system/lib/dri/gallium_dri.so
(ir_call::accept(ir_hierarchical_visitor*)+38)
I/DEBUG ( 1784): #02 pc 004172a9 /system/lib/dri/gallium_dri.so
(ir_function_signature::accept(ir_hierarchical_visitor*)+201)
I/DEBUG ( 1784): #03 pc 00417374 /system/lib/dri/gallium_dri.so
(ir_function::accept(ir_hierarchical_visitor*)+100)
I/DEBUG ( 1784): #04 pc 0041707f /system/lib/dri/gallium_dri.so
(visit_list_elements(ir_hierarchical_visitor*, exec_list*, bool)+127)
I/DEBUG ( 1784): #05 pc 00416cdb /system/lib/dri/gallium_dri.so
(ir_hierarchical_visitor::run(exec_list*)+43)
I/DEBUG ( 1784): #06 pc 00421b8e /system/lib/dri/gallium_dri.so
(link_function_calls(gl_shader_program*, gl_shader*, gl_shader**, unsigned
int)+142)
I/DEBUG ( 1784): #07 pc 0041f82b /system/lib/dri/gallium_dri.so
(link_shaders(gl_context*, gl_shader_program*)+7947)
I/DEBUG ( 1784): #08 pc 0029135f /system/lib/dri/gallium_dri.so
(_mesa_glsl_link_shader+287)
I/DEBUG ( 1784): #09 pc 0020618c /system/lib/dri/gallium_dri.so
(_mesa_LinkProgram+124)
I/DEBUG ( 1784): #10 pc 00010f57 /system/lib/libglapi.so
(shared_dispatch_stub_509+39)
I/DEBUG ( 1784): #11 pc 0008e86a /system/lib/libandroid_runtime.so
(android_glLinkProgram__I(_JNIEnv*, _jobject*, int)+26)


With this info, I could determine the exact point where SIGSEGV /
SEGV_MAPERR is caused:

external/mesa/src/glsl/link_functions.cpp
Class: class call_link_visitor : public ir_hierarchical_visitor
Method: virtual ir_visitor_status visit_enter(ir_call *ir)
instruction causing crash: const char *const name = callee->function_name();

This "callee" object is type "ir_function_signature" (got from ir->callee ;
"ir" is the class object param). It's method "function_name" should return
the glsl function name that will be used (most of calls it returns
"texture2D", according to my logs - see
logcat_android_build_mesa_10.7_kernel.4.0_20150614_1640.txt). But when
machine is entering sleep mode, this function apparently doesn't have a
valid pointer assignment, causing the crash.

The backtrace also shows that "glLinkProgram" is being used. It's a common
call, but I found this in particullar interesting by it's preceding comment:

frameworks/base/services/core/java/com/android/server/display/ColorFade.java
...
/**
* <p>
* Animates a screen transition from on to off or off to on by applying
* some GL transformations to a screenshot.
* </p><p>
* This component must only be created or accessed by the {@link Looper}
thread
* that belongs to the {@link *DisplayPowerController*}.
* </p>
*/
final class ColorFade {
...
private boolean initGLShaders(Context context) {
...
GLES20.glLinkProgram(mProgram);



Our crash is related to PowerManagerServices, and this file just exists on
Lollipop...

That's what I have so far. Don't take everything as 100% accurate, because
I may be wrong on my assumptions/conclusions

I attached the most recent logcat and tombstone I have.

Any help is appreciated.

Regards,
pstglia

[pstglia]

Post by pstglia
frameworks/base/services/core/java/com/android/server/display/ColorFade.java
...
/**
* <p>
* Animates a screen transition from on to off or off to on by applying
* some GL transformations to a screenshot.
* </p><p>
thread
* </p>
*/
final class ColorFade {
...
private boolean initGLShaders(Context context) {
...
GLES20.glLinkProgram(mProgram);
Our crash is related to PowerManagerServices, and this file just exists on
Lollipop...
That's what I have so far. Don't take everything as 100% accurate, because
I may be wrong on my assumptions/conclusions

Yes... this is our bad guy. Inserted a print before and after glLinkProgram:
Slog.d(TAG, "PST DEBUG - Inside ColorFade.java (initGLShaders) -
Before GLES20.glLinkProgram");
GLES20.glLinkProgram(mProgram);
Slog.d(TAG, "PST DEBUG - Inside ColorFade.java (initGLShaders) -
After GLES20.glLinkProgram");

And "After" debug point is not print. Instead we have the crash:

*06-17 22:46:52.811 1800 1829 D ColorFade: PST DEBUG - Inside
ColorFade.java (initGLShaders) - Before GLES20.glLinkProgram*
06-17 22:46:52.812 1800 1829 I PST-DEBUG-LNK: Inside
ir_visitor_status::visit_enter - before const ir_function_signature *const
callee = ir->callee - addr is 0x9B1BA828, function_name addr is 0x9C65EFF0
06-17 22:46:52.812 1800 1829 I PST-DEBUG-LNK: Inside
ir_visitor_status::visit_enter - after const ir_function_signature *const
callee = ir->callee
06-17 22:46:52.812 1800 1829 I PST-DEBUG-LNK: Inside
ir_visitor_status::visit_enter - before const char *const name =
callee->function_name()
--------- beginning of crash
06-17 22:46:52.812 1800 1829 F libc : Fatal signal 11 (SIGSEGV), code
1, fault addr 0x10 in tid 1829 (PowerManagerSer)
06-17 22:46:52.862 1518 1518 I DEBUG : *** *** *** *** *** *** *** ***
*** *** *** *** *** *** *** ***
06-17 22:46:52.862 1518 1518 I DEBUG : Build fingerprint:
'Android-x86/android_x86/x86:5.1.1/LMY47V/paulo05050202:eng/test-keys'
06-17 22:46:52.862 1518 1518 I DEBUG : Revision: '0'
06-17 22:46:52.862 1518 1518 I DEBUG : ABI: 'x86'
06-17 22:46:52.862 1518 1518 I DEBUG : pid: 1800, tid: 1829, name:
PowerManagerSer >>> system_server <<<
06-17 22:46:52.862 1518 1518 I DEBUG : signal 11 (SIGSEGV), code 1
(SEGV_MAPERR), fault addr 0x10
06-17 22:46:52.864 1518 1518 I DEBUG : eax 00000000 ebx 9c93ecf4
ecx c5eb9402 edx c5eb9402
06-17 22:46:52.864 1518 1518 I DEBUG : esi a21d15f0 edi 9b1ba814
06-17 22:46:52.864 1518 1518 I DEBUG : xcs 00000073 xds 0000007b
xes 0000007b xfs 000000af xss 0000007b
06-17 22:46:52.864 1518 1518 I DEBUG : eip 9c65f2ef ebp 9c7fc486
esp a21d14a0 flags 00210286
06-17 22:46:52.864 1518 1518 I DEBUG :
06-17 22:46:52.864 1518 1518 I DEBUG : backtrace:
06-17 22:46:52.864 1518 1518 I DEBUG : #00 pc 004212ef
/android/system/lib/dri/gallium_dri.so
06-17 22:46:52.864 1518 1518 I DEBUG : #01 pc 55424543 <unknown>
06-17 22:46:52.878 1518 1518 W libbacktrace: virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0x9c7fc48e
06-17 22:46:52.976 1518 1518 W libbacktrace: virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb755082d
06-17 22:46:53.048 1518 1518 W libbacktrace: virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb69f9873


This ColorFade is somehow problematic (there's a recent history of memory
leaks).

Now we need to know how to fix that (maybe disabling this ColorFade
function)

[Mauro Rossi]

Hi,

after mesa 10.6.0 release I've done a rebase of current x86/mesa-10.6
branch on top of freedesktop/10.6,
to pick nouveau unordered_set changes.

cd [kitkat-x86 folder]

cd external/mesa
git remote add freedesktop http://anongit.freedesktop.org/git/mesa/mesa.git
git fetch x86 mesa-10.6
git checkout FETCH_HEAD
git fetch freedesktop 10.6
git rebase freedesktop/10.6

[rebase is successful]


When building ISO with nouveau, I found two building errors (independent
from unordered_set):

1) error related to operator ""PRIx64, fixed by just adding one space:

https://github.com/maurossi/mesa/commit/e44e6f9d95c1aace41e0d9dbc500faaa9dc497cd

2) building error due to assert already defined, solved as per former
similar build error in src/util

https://github.com/maurossi/mesa/commit/31a90feb14c621dadbebfdc335dcf542e8f645a9

Checked the created ISO and it's working (with the known nouveau problems
with TestEGL com.android.test app, gldual test app and with harism effects
app).
M.

[pstglia]

Hi,

Post by pstglia
Now we need to know how to fix that (maybe disabling this ColorFade
function)

It looks like just to produce some fancy effects.
I think we can live without it.

To solve ColorFade problem (in fact a GLSL fragment shader which Mesa
doesn't like that much) I have commented most of cmds used, keeping
just gl_FragColor variable set in order to make it black.

This is the patch 0001-Avoid-crashes-with-Suspendon-Lollipop-ColorFade.patch

Applying this patch, no more crashes on Mesa. However, on my HW, system
was freezed for a long time until watchdog kill processes and
SurfaceFlinger... In other words, another crash.

By analyzing ANR logs, I checked a message handler
on PowerManagerService.java (onDisplayStateChange) was calling a jni method
(setHalInteractiveModeLocked) and never ended.

I commented out all the cmds called inside this handler and I had no more
crashes. The result of this is that system doesn't really suspend (I mean
entering low power mode); It just fade screen to black on Suspend and shows
img again on resume.
This patch is 0001-WORKAROUND-Avoid-lock-on-suspend.patch

I suggest you test first
"0001-Avoid-crashes-with-Suspendon-Lollipop-ColorFade.patch" and see what
happens on your HW. If it freezes and recovers after some seconds/minutes
with some watchdog messages on logcat, apply
"0001-WORKAROUND-Avoid-lock-on-suspend.patch"

Regards,
Pstglia

[pstglia]

Hi again,


By analyzing ANR logs, I checked a message handler on

Post by pstglia
PowerManagerService.java (onDisplayStateChange) was calling a jni method
(setHalInteractiveModeLocked) and never ended.
I commented out all the cmds called inside this handler and I had no more
crashes. The result of this is that system doesn't really suspend (I mean
entering low power mode); It just fade screen to black on Suspend and shows
img again on resume.
This patch is 0001-WORKAROUND-Avoid-lock-on-suspend.patch

I believe the reason of this issue was found. 2 libsuspend necessary
patches from kitkat should be cherry-picked and applied on lollipop :


-----------------------------------------------------

commit f872453e01396c969104f440380ef299e26787e1

Author: Chih-Wei Huang <***@linux.org.tw>
Date: Wed Aug 21 23:39:58 2013 +0800


libsuspend: add debug.autosleep property

Only use autosleep first if debug.autosleep=1.
-----------------------------------------------------
commit 09e4512b48fb309c4f9c8c896bb01dcda9a46695
Author: Kyle Evans <***@android-x86.org>
Date: Tue Mar 4 01:39:49 2014 -0500


libsuspend: do not fail when sending on signal fails
-----------------------------------------------------


I also reverted this commit:
-----------------------------------------------------
commit: 6f26891d67a55515b46c3da38d63c4d41c3cfefc
Author: Todd Poynor
<http://git.android-x86.org/?p=platform/system/core.git;a=search;h=6f26891d67a55515b46c3da38d63c4d41c3cfefc;s=Todd+Poynor;st=author>
<***@google.com>
<http://git.android-x86.org/?p=platform/system/core.git;a=search;h=6f26891d67a55515b46c3da38d63c4d41c3cfefc;s=***@google.com;st=author>
Date: Tue, 4 MAr 2014 21:11:21 +0000
libsuspend: use wakeup count interface for system suspend

Use the wakeup count interface for requesting suspend, instead of
autosleep, to allow userspace to collect stats on suspend/resume
activity.
-----------------------------------------------------

With the commits mentioned above and libsuspend and
"0002-Avoid-crashes-with-Suspendon-Lollipop-ColorFade.patch" on
framework/base , I was able to suspend/resume on my machine.

I think this solves the suspend/resume bug on Lollipop. Please test it on
different HWs.

Regards,
Pstglia

PS: "0001-WORKAROUND-Avoid-lock-on-suspend.patch" was just a workaround to
block jni HAL call and is not needed.

[André Vitor]

Hi pstglia,

Can you make a ISO for testing? I would love test it, because my Zmax
tablet only miss this feature to be fully working as android 5.x. :D

Definitely gonna re-download the source code if works.

Thanks.

Post by pstglia
Hi again,
By analyzing ANR logs, I checked a message handler on

Post by pstglia
PowerManagerService.java (onDisplayStateChange) was calling a jni method
(setHalInteractiveModeLocked) and never ended.
I commented out all the cmds called inside this handler and I had no more
crashes. The result of this is that system doesn't really suspend (I mean
entering low power mode); It just fade screen to black on Suspend and shows
img again on resume.
This patch is 0001-WORKAROUND-Avoid-lock-on-suspend.patch

I believe the reason of this issue was found. 2 libsuspend necessary
-----------------------------------------------------
commit f872453e01396c969104f440380ef299e26787e1
Date: Wed Aug 21 23:39:58 2013 +0800
libsuspend: add debug.autosleep property
Only use autosleep first if debug.autosleep=1.
-----------------------------------------------------
commit 09e4512b48fb309c4f9c8c896bb01dcda9a46695
Date: Tue Mar 4 01:39:49 2014 -0500
libsuspend: do not fail when sending on signal fails
-----------------------------------------------------
-----------------------------------------------------
commit: 6f26891d67a55515b46c3da38d63c4d41c3cfefc
Author: Todd Poynor
<http://git.android-x86.org/?p=platform/system/core.git;a=search;h=6f26891d67a55515b46c3da38d63c4d41c3cfefc;s=Todd+Poynor;st=author>
Date: Tue, 4 MAr 2014 21:11:21 +0000
libsuspend: use wakeup count interface for system suspend
Use the wakeup count interface for requesting suspend, instead of
autosleep, to allow userspace to collect stats on suspend/resume
activity.
-----------------------------------------------------
With the commits mentioned above and libsuspend and
"0002-Avoid-crashes-with-Suspendon-Lollipop-ColorFade.patch" on
framework/base , I was able to suspend/resume on my machine.
I think this solves the suspend/resume bug on Lollipop. Please test it on
different HWs.
Regards,
Pstglia
PS: "0001-WORKAROUND-Avoid-lock-on-suspend.patch" was just a workaround to
block jni HAL call and is not needed.

[pstglia]

Hi André,

Can you make a ISO for testing? I would love test it, because my Zmax

Post by André Vitor
tablet only miss this feature to be fully working as android 5.x. :D
Definitely gonna re-download the source code if works.
Thanks.

Yes of course. I'll post the link when I upload it.

[pstglia]

Post by pstglia
Hi André,
Can you make a ISO for testing? I would love test it, because my Zmax

Post by André Vitor
tablet only miss this feature to be fully working as android 5.x. :D
Definitely gonna re-download the source code if works.
Thanks.

Yes of course. I'll post the link when I upload it.

Here's the testing ISO.

https://drive.google.com/file/d/0Bx12U5yGNcQCYkpqMWQ5THF2bEk/view?usp=sharing


Note: In my case sometimes takes 5~12 seconds before system suspends

[André Vitor]

Tested it.
dmesg and logcat attached.

It's working. My device is suspending (*instantly*), but I need to do some
things to make it work.

For some reason the device can't suspend unless I turn on Wi-Fi and connect
it. In dmesg it says that:

3Freezing of tasks failed after 20.002 seconds (1 tasks refusing to
After I turn on the Wi-Fi *AND* connect it to a network, the suspend thing
start working as should be, even if I turn Wi-Fi off again later.
Probably some app is trying to do something and it's stopping the whole
system from suspend.

Also, it's probably known, I can't suspend from the power button, I needed
to change the timeout sleep in the settings to test this.

Another bug that happens to my device when I try to sleep (after doing the
thing with Wi-Fi):

<3>[ 296.609520] pci_pm_suspend(): hcd_pci_suspend+0x0/0x16 returns -16

<3>[ 296.609536] dpm_run_callback(): pci_pm_suspend+0x0/0x172 returns -16
<3>[ 296.609545] PM: Device 0000:00:14.0 failed to suspend async: error
-16
<3>[ 296.620285] PM: Some devices failed to suspend, or early wake event
detected

Sometimes the device don't sleep because of this error, probably related to
kernel but still reporting this here.
This also was happening in kitkat-x86, that's why I think it's kernel
related.

Post by pstglia
Hi André,
Can you make a ISO for testing? I would love test it, because my Zmax

Post by André Vitor
tablet only miss this feature to be fully working as android 5.x. :D
Definitely gonna re-download the source code if works.
Thanks.

Yes of course. I'll post the link when I upload it.

Here's the testing ISO.
https://drive.google.com/file/d/0Bx12U5yGNcQCYkpqMWQ5THF2bEk/view?usp=sharing
Note: In my case sometimes takes 5~12 seconds before system suspends

[pstglia]

Hi André

Post by André Vitor
It's working. My device is suspending (*instantly*), but I need to do
some things to make it work.
For some reason the device can't suspend unless I turn on Wi-Fi and
3Freezing of tasks failed after 20.002 seconds (1 tasks refusing to
After I turn on the Wi-Fi *AND* connect it to a network, the suspend
thing start working as should be, even if I turn Wi-Fi off again later.
Probably some app is trying to do something and it's stopping the whole
system from suspend.

That's strange. If possible, can you attach a logcat with DEBUG=2? Hoping
this gives us more clues

Post by André Vitor
Also, it's probably known, I can't suspend from the power button, I needed
to change the timeout sleep in the settings to test this.
Another bug that happens to my device when I try to sleep (after doing the
<3>[ 296.609520] pci_pm_suspend(): hcd_pci_suspend+0x0/0x16 returns -16

<3>[ 296.609536] dpm_run_callback(): pci_pm_suspend+0x0/0x172 returns -16
<3>[ 296.609545] PM: Device 0000:00:14.0 failed to suspend async: error
-16
<3>[ 296.620285] PM: Some devices failed to suspend, or early wake event
detected

Sometimes the device don't sleep because of this error, probably related
to kernel but still reporting this here.
This also was happening in kitkat-x86, that's why I think it's kernel
related.

Do you experience this a regular Linux distro?

[Tom Briden]

I can confirm the 0001-Avoid-crashes-with-Suspendon-Lollipop-ColorFade
patch works to turn the screen off on my Sony Vaio Fit 13 running lollipop
but it doesn't put the device to sleep. Applying those 2 cherry-picks for
libsuspend do make it suspend but it must crash on wake as it ends up fully
rebooting the device. I'm happy enough with that patch as it at least means
the screen will go off when not in use so many thanks for that one

Is there anything i can do to debug the libsuspend changes? there aren't
any tombstones made and a logcat only shows everything up to the suspend?

[Tom Briden]

Looks like i spoke too soon. While the screen does go off after the sleep
timeout, after 20-30secs it starts to flicker on and off. any ideas?

Post by Tom Briden
I can confirm the 0001-Avoid-crashes-with-Suspendon-Lollipop-ColorFade
patch works to turn the screen off on my Sony Vaio Fit 13 running lollipop
but it doesn't put the device to sleep. Applying those 2 cherry-picks for
libsuspend do make it suspend but it must crash on wake as it ends up fully
rebooting the device. I'm happy enough with that patch as it at least means
the screen will go off when not in use so many thanks for that one
Is there anything i can do to debug the libsuspend changes? there aren't
any tombstones made and a logcat only shows everything up to the suspend?

[pstglia]

Hi

Looks like i spoke too soon. While the screen does go off after the sleep

Post by Tom Briden
timeout, after 20-30secs it starts to flicker on and off. any ideas?

Can you provide logcat and dmesg outputs? If possible, use the iso I
provided on Jun, 22th, as it contains some extra debugging msgs and libs
are not being stripped.

Regards
Pstglia

[Tom Briden]

Hm, suspend appears to work perfectly with your iso. The main differences I
can think of between are that I'm not using the 4.0 kernel and I use efi
builds (so had to switch my bios to legacy to boot your iso).

I'm currently running a 4.0 kernel build to see if that makes any
difference. Are your cherry-picks on the current head of system/core? I had
conflicts with one but it was easy enough to resolve. Just in case though,
can you post a git diff between your system/core and x86/lollipop-x86?

Thanks

Post by Tom Briden
Hi
Looks like i spoke too soon. While the screen does go off after the sleep

Post by Tom Briden
timeout, after 20-30secs it starts to flicker on and off. any ideas?

Can you provide logcat and dmesg outputs? If possible, use the iso I
provided on Jun, 22th, as it contains some extra debugging msgs and libs
are not being stripped.
Regards
Pstglia

[Tom Briden]

I had to disable CONFIG_WL to get it to compile but it now won't boot due
to libbinder errors:

07-04 21:39:35.785 2496 2496 E ProcessState: Binder driver protocol does
not match user space protocol!
07-04 21:39:35.785 2496 2496 E ProcessState: Binder ioctl to set max
threads failed: Bad file number
07-04 21:39:35.785 2496 2496 F ProcessState: Binder driver could not be
opened. Terminating.

Post by Tom Briden
Hm, suspend appears to work perfectly with your iso. The main differences
I can think of between are that I'm not using the 4.0 kernel and I use efi
builds (so had to switch my bios to legacy to boot your iso).
I'm currently running a 4.0 kernel build to see if that makes any
difference. Are your cherry-picks on the current head of system/core? I had
conflicts with one but it was easy enough to resolve. Just in case though,
can you post a git diff between your system/core and x86/lollipop-x86?
Thanks

Post by Tom Briden
Hi
Looks like i spoke too soon. While the screen does go off after the sleep

Post by Tom Briden
timeout, after 20-30secs it starts to flicker on and off. any ideas?

Can you provide logcat and dmesg outputs? If possible, use the iso I
provided on Jun, 22th, as it contains some extra debugging msgs and libs
are not being stripped.
Regards
Pstglia

[pstglia]

Hi Tom,

I'm currently running a 4.0 kernel build to see if that makes any

Post by Tom Briden
difference. Are your cherry-picks on the current head of system/core? I had
conflicts with one but it was easy enough to resolve. Just in case though,
can you post a git diff between your system/core and x86/lollipop-x86?

Here's my "git diff" output:

diff --git a/libsuspend/autosuspend.c b/libsuspend/autosuspend.c
index edd1007..3c88a33 100644
--- a/libsuspend/autosuspend.c
+++ b/libsuspend/autosuspend.c
@@ -18,6 +18,7 @@

#define LOG_TAG "libsuspend"
#include <cutils/log.h>
+#include <cutils/properties.h>

#include <suspend/autosuspend.h>

@@ -29,11 +30,15 @@ static bool autosuspend_inited;

static int autosuspend_init(void)
{
+ int autosleep;
+ char propbuf[PROPERTY_VALUE_MAX];
if (autosuspend_inited) {
return 0;
}

- autosuspend_ops = autosuspend_earlysuspend_init();
+ property_get("debug.autosleep", propbuf, "0");
+ autosleep = (propbuf[0] == '1');
+ autosuspend_ops = autosleep ? autosuspend_autosleep_init() :
autosuspend_earlysuspend_init();
if (autosuspend_ops) {
goto out;
}
diff --git a/libsuspend/autosuspend_earlysuspend.c
b/libsuspend/autosuspend_earlysuspend.c
index 2bece4c..8b89933 100644
--- a/libsuspend/autosuspend_earlysuspend.c
+++ b/libsuspend/autosuspend_earlysuspend.c
@@ -108,7 +108,6 @@ static int autosuspend_earlysuspend_enable(void)
if (ret < 0) {
strerror_r(errno, buf, sizeof(buf));
ALOGE("Error writing to %s: %s\n", EARLYSUSPEND_SYS_POWER_STATE,
buf);
- goto err;
}

if (wait_for_earlysuspend) {
@@ -123,8 +122,6 @@ static int autosuspend_earlysuspend_enable(void)

return 0;

-err:
- return ret;
}

static int autosuspend_earlysuspend_disable(void)
@@ -203,20 +200,10 @@ struct autosuspend_ops
*autosuspend_earlysuspend_init(void)
return NULL;
}

- ret = write(sPowerStatefd, "on", 2);
- if (ret < 0) {
- strerror_r(errno, buf, sizeof(buf));
- ALOGW("Error writing 'on' to %s: %s\n",
EARLYSUSPEND_SYS_POWER_STATE, buf);
- goto err_write;
- }
-
ALOGI("Selected early suspend\n");

start_earlysuspend_thread();

return &autosuspend_earlysuspend_ops;

-err_write:
- close(sPowerStatefd);
- return NULL;
}

[Tom Briden]

Thanks pstglia, that's identical to mine so it's probably down to the 3.18
kernel

Post by pstglia
Hi Tom,
I'm currently running a 4.0 kernel build to see if that makes any

Post by Tom Briden
difference. Are your cherry-picks on the current head of system/core? I had
conflicts with one but it was easy enough to resolve. Just in case though,
can you post a git diff between your system/core and x86/lollipop-x86?

diff --git a/libsuspend/autosuspend.c b/libsuspend/autosuspend.c
index edd1007..3c88a33 100644
--- a/libsuspend/autosuspend.c
+++ b/libsuspend/autosuspend.c
@@ -18,6 +18,7 @@
#define LOG_TAG "libsuspend"
#include <cutils/log.h>
+#include <cutils/properties.h>
#include <suspend/autosuspend.h>
@@ -29,11 +30,15 @@ static bool autosuspend_inited;
static int autosuspend_init(void)
{
+ int autosleep;
+ char propbuf[PROPERTY_VALUE_MAX];
if (autosuspend_inited) {
return 0;
}
- autosuspend_ops = autosuspend_earlysuspend_init();
+ property_get("debug.autosleep", propbuf, "0");
+ autosleep = (propbuf[0] == '1');
autosuspend_earlysuspend_init();
if (autosuspend_ops) {
goto out;
}
diff --git a/libsuspend/autosuspend_earlysuspend.c
b/libsuspend/autosuspend_earlysuspend.c
index 2bece4c..8b89933 100644
--- a/libsuspend/autosuspend_earlysuspend.c
+++ b/libsuspend/autosuspend_earlysuspend.c
@@ -108,7 +108,6 @@ static int autosuspend_earlysuspend_enable(void)
if (ret < 0) {
strerror_r(errno, buf, sizeof(buf));
ALOGE("Error writing to %s: %s\n", EARLYSUSPEND_SYS_POWER_STATE,
buf);
- goto err;
}
if (wait_for_earlysuspend) {
@@ -123,8 +122,6 @@ static int autosuspend_earlysuspend_enable(void)
return 0;
- return ret;
}
static int autosuspend_earlysuspend_disable(void)
@@ -203,20 +200,10 @@ struct autosuspend_ops
*autosuspend_earlysuspend_init(void)
return NULL;
}
- ret = write(sPowerStatefd, "on", 2);
- if (ret < 0) {
- strerror_r(errno, buf, sizeof(buf));
- ALOGW("Error writing 'on' to %s: %s\n",
EARLYSUSPEND_SYS_POWER_STATE, buf);
- goto err_write;
- }
-
ALOGI("Selected early suspend\n");
start_earlysuspend_thread();
return &autosuspend_earlysuspend_ops;
- close(sPowerStatefd);
- return NULL;
}

[Tom Briden]

Now running with 4.0.6 and still can't resume from suspend. Could it be
related to being in EFI mode? I had to
add CONFIG_ANDROID_BINDER_IPC_32BIT=y to the android_x86_64_defconfig to
get it booting, maybe i need to change something else in the config?

Post by Tom Briden
Thanks pstglia, that's identical to mine so it's probably down to the 3.18
kernel

Post by pstglia
Hi Tom,
I'm currently running a 4.0 kernel build to see if that makes any

Post by Tom Briden
difference. Are your cherry-picks on the current head of system/core? I had
conflicts with one but it was easy enough to resolve. Just in case though,
can you post a git diff between your system/core and x86/lollipop-x86?

diff --git a/libsuspend/autosuspend.c b/libsuspend/autosuspend.c
index edd1007..3c88a33 100644
--- a/libsuspend/autosuspend.c
+++ b/libsuspend/autosuspend.c
@@ -18,6 +18,7 @@
#define LOG_TAG "libsuspend"
#include <cutils/log.h>
+#include <cutils/properties.h>
#include <suspend/autosuspend.h>
@@ -29,11 +30,15 @@ static bool autosuspend_inited;
static int autosuspend_init(void)
{
+ int autosleep;
+ char propbuf[PROPERTY_VALUE_MAX];
if (autosuspend_inited) {
return 0;
}
- autosuspend_ops = autosuspend_earlysuspend_init();
+ property_get("debug.autosleep", propbuf, "0");
+ autosleep = (propbuf[0] == '1');
autosuspend_earlysuspend_init();
if (autosuspend_ops) {
goto out;
}
diff --git a/libsuspend/autosuspend_earlysuspend.c
b/libsuspend/autosuspend_earlysuspend.c
index 2bece4c..8b89933 100644
--- a/libsuspend/autosuspend_earlysuspend.c
+++ b/libsuspend/autosuspend_earlysuspend.c
@@ -108,7 +108,6 @@ static int autosuspend_earlysuspend_enable(void)
if (ret < 0) {
strerror_r(errno, buf, sizeof(buf));
ALOGE("Error writing to %s: %s\n", EARLYSUSPEND_SYS_POWER_STATE,
buf);
- goto err;
}
if (wait_for_earlysuspend) {
@@ -123,8 +122,6 @@ static int autosuspend_earlysuspend_enable(void)
return 0;
- return ret;
}
static int autosuspend_earlysuspend_disable(void)
@@ -203,20 +200,10 @@ struct autosuspend_ops
*autosuspend_earlysuspend_init(void)
return NULL;
}
- ret = write(sPowerStatefd, "on", 2);
- if (ret < 0) {
- strerror_r(errno, buf, sizeof(buf));
- ALOGW("Error writing 'on' to %s: %s\n",
EARLYSUSPEND_SYS_POWER_STATE, buf);
- goto err_write;
- }
-
ALOGI("Selected early suspend\n");
start_earlysuspend_thread();
return &autosuspend_earlysuspend_ops;
- close(sPowerStatefd);
- return NULL;
}

[pstglia]

Post by Tom Briden
Now running with 4.0.6 and still can't resume from suspend. Could it be
related to being in EFI mode? I had to
add CONFIG_ANDROID_BINDER_IPC_32BIT=y to the android_x86_64_defconfig to
get it booting, maybe i need to change something else in the config?

You are using a 64 bits kernel in a 32 bits userspace? Just asking this
because I'm having the same issue with a recent built (using 64 bits kernel
for fugu houdini tests)

In any case, try reverting this commit and see if you still have problems
(I know I said it was not needed reverting this, but just in case... :D ):
6f26891d67a55515b46c3da38d63c4d41c3cfefc
"libsuspend: use wakeup count interface for system suspend"

Regards,
Pstglia

[Tom Briden]

Same issue with that revert :\ I am indeed using a 64bit kernel in a 32bit
userspace. I also have the fugu houdini stuff in and that's all working
great.

Post by Tom Briden
Now running with 4.0.6 and still can't resume from suspend. Could it be

Post by Tom Briden
related to being in EFI mode? I had to
add CONFIG_ANDROID_BINDER_IPC_32BIT=y to the android_x86_64_defconfig to
get it booting, maybe i need to change something else in the config?

You are using a 64 bits kernel in a 32 bits userspace? Just asking this
because I'm having the same issue with a recent built (using 64 bits kernel
for fugu houdini tests)
In any case, try reverting this commit and see if you still have problems
6f26891d67a55515b46c3da38d63c4d41c3cfefc
"libsuspend: use wakeup count interface for system suspend"
Regards,
Pstglia

[pstglia]

Post by Tom Briden
Same issue with that revert :\ I am indeed using a 64bit kernel in a 32bit
userspace. I also have the fugu houdini stuff in and that's all working
great.

Ok, at least this confirms ithis revert is not really needed.

I'll try build my lollipop ISO again with 32 bits kernel and check if it
solves the resume problem.
If this fix my problem, we'll have to investigate for a fix on 64 bits...

[fgdn17]

think you'll find the code all takes care of the correct binder settings
for what your building...look at the git, look at the code, also
see the discussions and Chih-Wei solutions for this what 6 or more months
ago.....

the issue you are experiencing is very simply solved....merge / debug the
appropriate changes that already exist in kitkat-x86 into
lollipop-x86.....

sometimes we cannot see the forest through the trees....but that's how we
learn...repeating what has already been done!!

Post by Tom Briden
Now running with 4.0.6 and still can't resume from suspend. Could it be
related to being in EFI mode? I had to
add CONFIG_ANDROID_BINDER_IPC_32BIT=y to the android_x86_64_defconfig to
get it booting, maybe i need to change something else in the config?

Post by Tom Briden
Thanks pstglia, that's identical to mine so it's probably down to the
3.18 kernel

Post by pstglia
Hi Tom,
I'm currently running a 4.0 kernel build to see if that makes any

Post by Tom Briden
difference. Are your cherry-picks on the current head of system/core? I had
conflicts with one but it was easy enough to resolve. Just in case though,
can you post a git diff between your system/core and x86/lollipop-x86?

diff --git a/libsuspend/autosuspend.c b/libsuspend/autosuspend.c
index edd1007..3c88a33 100644
--- a/libsuspend/autosuspend.c
+++ b/libsuspend/autosuspend.c
@@ -18,6 +18,7 @@
#define LOG_TAG "libsuspend"
#include <cutils/log.h>
+#include <cutils/properties.h>
#include <suspend/autosuspend.h>
@@ -29,11 +30,15 @@ static bool autosuspend_inited;
static int autosuspend_init(void)
{
+ int autosleep;
+ char propbuf[PROPERTY_VALUE_MAX];
if (autosuspend_inited) {
return 0;
}
- autosuspend_ops = autosuspend_earlysuspend_init();
+ property_get("debug.autosleep", propbuf, "0");
+ autosleep = (propbuf[0] == '1');
autosuspend_earlysuspend_init();
if (autosuspend_ops) {
goto out;
}
diff --git a/libsuspend/autosuspend_earlysuspend.c
b/libsuspend/autosuspend_earlysuspend.c
index 2bece4c..8b89933 100644
--- a/libsuspend/autosuspend_earlysuspend.c
+++ b/libsuspend/autosuspend_earlysuspend.c
@@ -108,7 +108,6 @@ static int autosuspend_earlysuspend_enable(void)
if (ret < 0) {
strerror_r(errno, buf, sizeof(buf));
ALOGE("Error writing to %s: %s\n",
EARLYSUSPEND_SYS_POWER_STATE, buf);
- goto err;
}
if (wait_for_earlysuspend) {
@@ -123,8 +122,6 @@ static int autosuspend_earlysuspend_enable(void)
return 0;
- return ret;
}
static int autosuspend_earlysuspend_disable(void)
@@ -203,20 +200,10 @@ struct autosuspend_ops
*autosuspend_earlysuspend_init(void)
return NULL;
}
- ret = write(sPowerStatefd, "on", 2);
- if (ret < 0) {
- strerror_r(errno, buf, sizeof(buf));
- ALOGW("Error writing 'on' to %s: %s\n",
EARLYSUSPEND_SYS_POWER_STATE, buf);
- goto err_write;
- }
-
ALOGI("Selected early suspend\n");
start_earlysuspend_thread();
return &autosuspend_earlysuspend_ops;
- close(sPowerStatefd);
- return NULL;
}

[pstglia]

Post by pstglia
-----------------------------------------------------
commit: 6f26891d67a55515b46c3da38d63c4d41c3cfefc
Author: Todd Poynor
<http://git.android-x86.org/?p=platform/system/core.git;a=search;h=6f26891d67a55515b46c3da38d63c4d41c3cfefc;s=Todd+Poynor;st=author>
Date: Tue, 4 MAr 2014 21:11:21 +0000
libsuspend: use wakeup count interface for system suspend
Use the wakeup count interface for requesting suspend, instead of
autosleep, to allow userspace to collect stats on suspend/resume
activity.
-----------------------------------------------------

Reverting this is not needed.