Enabling radeonsi on Android-x86 lollipop

[pstglia]

Hi everyone,

I've managed to compile radeonsi on Lollipop using Mesa 10.4.2 (which not
means that will work when testing on hardware :D )
At first I tried kitkat, but older llvm version was giving me problems.
That's why I tried lollipop, which uses a "mix" of version 3.4 / 3.5...

I had basically to change/create Android makefiles on "external/mesa" and
"external/llvm" (the the last one is a requirement for radeonsi).

I had also to create manually "format_srgb.c" file by executing these cmds
(not the first time I face this issue, but still can't figure out how to
solve it in order to generate it on compile time - Python doesn't like
me... ):
cd external/mesa/src/util/
python format_srgb.py > format_srgb.c

I've attached patches. If you want to build by your own, don't forget to
create format_srgb.c as described above (or by fixing Android.mk files,
your choice)

I had not the chance to test it (no hardware to test). If you have a CGN
AMD GPU (R7/R9 series) or an CGN APU (like A10 7850K, A4 1200, etc) *and
you are aware of the risks (damage/data loss)*, here's a testing ISO:
https://drive.google.com/file/d/0Bx12U5yGNcQCTHVsR0FHaVE2eVU/view?usp=sharing

Working or not, if possible collect this info to check gallium is being
used:

# Check gallium renderer and version:
logcat | grep "SurfaceFlinger" | grep -E "renderer|version|extensions"

# Check if radeondrmfb is being used
cat /proc/fb

Thanks,
Pstglia

[Weikun]

Hi Pstglia,

Here is ISO test log with radeon R7 240.

I can't entry launcher, just stop at bootanimation screen.

Hope it's helpful for you.

pstgliaæ–Œ 2015幎4月22日星期䞉 UTC+8䞊午10時34分49秒寫道

Post by pstglia
https://drive.google.com/file/d/0Bx12U5yGNcQCTHVsR0FHaVE2eVU/view?usp=sharing
logcat | grep "SurfaceFlinger" | grep -E "renderer|version|extensions"

https://drive.google.com/folderview?id=0Bx3B7X77tBUgfjY5VmF2TldsaXJiZlFkcTlaWk5DYXZzV01SQ2l1bHpzTmJQR29lT1EwQ28&usp=sharing

Post by pstglia
# Check if radeondrmfb is being used
cat /proc/fb

0 radeondrmfb

[Mauro Rossi]

Hi Paulo,

great!
Things are moving with consistent results

Did you had to change /hardware/drm_gralloc.git/Android.mk in order to
support radeonsi?

# Android.mk for drm_gralloc

DRM_GPU_DRIVERS := $(strip $(filter-out swrast, $(BOARD_GPU_DRIVERS)))

freedreno_drivers := freedreno
intel_drivers := i915 i965 i915g ilo
-radeon_drivers := r300g r600g
+radeon_drivers := r300g r600g radeonsi
nouveau_drivers := nouveau
vmwgfx_drivers := vmwgfx

Thanks for the info.

For me it's time to find a Southern Island videocard to support testing.
Mauro

[pstglia]

Hi Mauro,

Did you had to change /hardware/drm_gralloc.git/Android.mk in order to

Post by Mauro Rossi
support radeonsi?
# Android.mk for drm_gralloc
DRM_GPU_DRIVERS := $(strip $(filter-out swrast, $(BOARD_GPU_DRIVERS)))
freedreno_drivers := freedreno
intel_drivers := i915 i965 i915g ilo
-radeon_drivers := r300g r600g
+radeon_drivers := r300g r600g radeonsi

Yes. Also enabled on device/generic/common/BoardConfig.mk

BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g radeonsi

Post by Mauro Rossi
For me it's time to find a Southern Island videocard to support testing.

AMD could donate some for us... lol

[Blumi Milch]

Yay, great!

First time ever i see a speedy display with android-x86 on my desktop hardware. =)
There are still various crashes (Browser Setup, Sleep Mode, Antutu Install/Run)
but imho the most important step is done.

BTW, I had to add the -d parameter to logcat, the commandline did not work for me without it.


# logcat -d | grep "SurfaceFlinger" | grep -E "renderer|version|extensions"

I/SurfaceFlinger( 1777): Using composer version 1.0
I/SurfaceFlinger( 1777): version : 1.4 Android META-EGL
I/SurfaceFlinger( 1777): extensions: EGL_KHR_get_all_proc_addresses EGL_ANDROID_presentation_time EGL_KHR_image_base EGL_KHR_reusable_sync EGL_KHR_fence_sync EGL_ANDROID_image_native_buffer
I/SurfaceFlinger( 1777): renderer : Gallium 0.4 on AMD PITCAIRN
I/SurfaceFlinger( 1777): version : OpenGL ES 2.0 Mesa 10.4.2 (git-0d114a7)
I/SurfaceFlinger( 1777): extensions: GL_EXT_debug_marker GL_EXT_blend_minmax GL_EXT_multi_draw_arrays GL_EXT_texture_filter_anisotropic GL_EXT_texture_format_BGRA8888 GL_OES_compressed_ETC1_RGB8_texture GL_OES_depth24 GL_OES_element_index_uint GL_OES_fbo_render_mipmap GL_OES_mapbuffer GL_OES_rgb8_rgba8 GL_OES_standard_derivatives GL_OES_stencil8 GL_OES_texture_3D GL_OES_texture_npot GL_OES_EGL_image GL_OES_depth_texture GL_OES_packed_depth_stencil GL_EXT_texture_type_2_10_10_10_REV GL_OES_get_program_binary GL_APPLE_texture_max_level GL_EXT_discard_framebuffer GL_EXT_read_format_bgra GL_NV_fbo_color_attachments GL_OES_EGL_image_external GL_OES_vertex_array_object GL_EXT_texture_rg GL_EXT_unpack_subimage GL_NV_draw_buffers GL_NV_read_buffer GL_EXT_map_buffer_range GL_OES_depth_texture_cube_map GL_OES_surfaceless_context GL_EXT_separate_shader_objects GL_KHR_context_flush_control


# cat /proc/fb

0 radeondrmfb


Additional Hardware Info:

* AMD Radeon HD 7850, Catalyst Control Center:

Graphics Chipset AMD Radeon HD 7800 Series
Device ID 6819
Vendor ID 1002
Subsystem ID 3260
Subsystem Vendor ID 1682
Revision ID 00
Graphics Bus Capability PCI Express 3.0
Maximum Bus Setting PCI Express 2.0x16
BIOS Version 015.016.000.001
BIOS Part Number 113-785ACDF71_113-C4010400-X06
BIOS Date 2012/03/13
Memory Size 3455 MB
Memory Type HyperMemory
Core Clock in MHz 860 MHz
Memory Clock in MHz 1200 MHz


* Mainboard Asus P5E, CPU-Z Info:

CPU Intel Core 2 Duo E6750 @2.66GHz
Northbridge Intel X38 rev. 01
Southbridge Intel 82801IR (ICH9R) rev. 02
Graphic Interface PCI-Express
PCI-E Link Width x16
Memory Type DDR2
Memory Size 4096 MBytes

[pstglia]

Hi Chi Wei. Sure I can!

I'll try to start this evening

[pstglia]

Great work, Paulo!
Could you rebase the patches to the 10.6 branch?
It's the target mesa of lollipop-x86.

Hi again,

Here's the patches:

# to be applied under external/mesa
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.6.patch

# to be applied under external/llvm (attached on the initial post)
0001-Enable-R600-support-on-Android-x86-mesa-gallium-rade.patch

# to be applied under hardware/drm_gralloc (add radeonsi on radeon_drivers)
0001-Enable-radeonsi-on-Android-x86.patch

And also include it on device/generic/common/BoardConfig.mk
BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g radeonsi

I've compiled, but not tested.

Regards,
Pstglia

[Jarvis Chung]

Hi Pstglia,

Good job!!
Thanks for your share.

I couldn't run mesa 10.6 for HD6670 and R7 240.
but, I could run mesa 10.4.2 use your patch.

For mesa 10.6, I success boot Android, but always get error log as below:

<6>[ 48.332792] RenderThread[2805]: segfault at 8 ip ae403d52 sp a168c838

error 6 in Roboto-Regular.ttf[ae403000+3b000]
<3>[ 48.360088] binder: 2202: binder_alloc_buf, no vma
<6>[ 48.360093] binder: 1915:1915 transaction failed 29201, size 100-0
<3>[ 48.360739] binder: 2202: binder_alloc_buf, no vma
<6>[ 48.360742] binder: 1915:1915 transaction failed 29201, size 100-0

For mesa 10.4.2, I can run Angry Bird, but not run Aututu and watch video
from browser.

I attached log files.

Jarvis

pstgliaæ–Œ 2015幎4月23日星期四 UTC+8䞊午10時26分17秒寫道

Great work, Paulo!

Could you rebase the patches to the 10.6 branch?
It's the target mesa of lollipop-x86.

Hi again,
# to be applied under external/mesa
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.6.patch
# to be applied under external/llvm (attached on the initial post)
0001-Enable-R600-support-on-Android-x86-mesa-gallium-rade.patch
# to be applied under hardware/drm_gralloc (add radeonsi on radeon_drivers)
0001-Enable-radeonsi-on-Android-x86.patch
And also include it on device/generic/common/BoardConfig.mk
BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g radeonsi
I've compiled, but not tested.
Regards,
Pstglia

[pstglia]

Hi Jarvis,

Good job!!

Post by Jarvis Chung
Thanks for your share.
I couldn't run mesa 10.6 for HD6670 and R7 240.
but, I could run mesa 10.4.2 use your patch.

You used kernel 3.10.58 right? Maybe you could try a newer one ( 3.18 or
4.0 ). Many features for radeon are only implemented on newer kernels

Regards,
pstglia

[Jarvis Chung]

Hi Pstglia,

Thanks for your reply.

I tested 3.18 and 4.0 kernel for mesa 10.6. I got the same issue.
I attached log use kernel 4.0.

Thanks.
Jarvis

pstgliaæ–Œ 2015幎4月23日星期四 UTC+8䞋午7時31分50秒寫道

Post by pstglia
Hi Jarvis,
Good job!!

Post by Jarvis Chung
Thanks for your share.
I couldn't run mesa 10.6 for HD6670 and R7 240.
but, I could run mesa 10.4.2 use your patch.

You used kernel 3.10.58 right? Maybe you could try a newer one ( 3.18 or
4.0 ). Many features for radeon are only implemented on newer kernels
Regards,
pstglia

[Mauro Rossi]

Hi Jarvis,

logcat shows some errors to be further investigated:

E/OpenGLRenderer( 2968): GL error: GL_INVALID_OPERATION

here EGL trace could show some EGL primitive detail about the GL error code
(GL_INVALID_OPERATION i.e. 0x0502),
it can be done in settings or via [ALT]+[F1] command line:

setprop debug.egl.trace error


Regarding this error:

<3>[ 48.360739] binder: 2202: binder_alloc_buf, no vma

if I'm not wrong we had a workaround to set grub line at boot: vmalloc=
*256M*



We may also need to check why fonts rendering is causing segfault (SIGSEV
MAPERR) like this:

D/OpenGLRenderer( 3946): setting property extraRasterBucket to 0.95
D/OpenGLRenderer( 3946): failed
D/OpenGLRenderer( 3946): setting property extraRasterBucket to 0.0
D/OpenGLRenderer( 3946): failed
F/libc ( 3946): Fatal signal 11 (SIGSEGV), code 1, fault addr 0x8 in tid
4002 (RenderThread)
I/DEBUG ( 1636): *** *** *** *** *** *** *** *** *** *** *** *** *** ***
*** ***
I/DEBUG ( 1636): Build fingerprint:
'Android-x86/android_x86/x86:5.1/LMY47I/jarvis04231119:eng/test-keys'
I/DEBUG ( 1636): Revision: '0'
I/DEBUG ( 1636): ABI: 'x86'
I/DEBUG ( 1636): pid: 3946, tid: 4002, name: RenderThread >>>
com.android.systemui <<<
I/DEBUG ( 1636): signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x8
E/DEBUG ( 1636): AM write failure (32 / Broken pipe)
I/DEBUG ( 1636): eax 00000007 ebx b703353c ecx 00000001 edx
b7015861
I/DEBUG ( 1636): esi a168a924 edi 00000003
I/DEBUG ( 1636): xcs 00000073 xds 0000007b xes 0000007b xfs
000000ef xss 0000007b
I/DEBUG ( 1636): eip ae403d52 ebp 00000005 esp a168a838 flags
00010202
I/DEBUG ( 1636):
I/DEBUG ( 1636): backtrace:
I/DEBUG ( 1636): #00 pc 00000d52
/android/system/fonts/Roboto-Regular.ttf
W/libbacktrace( 1636): virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb3766a91
W/libbacktrace( 1636): virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb6a14823


M.

[pstglia]

I think the patch has to be refined a little.
I'll take it.

Thanks for this. Still trying to improve it (making better patches)

I have a little concern you changed
libLLVMExecutionEngine and libLLVMJIT
from the host lib to the target lib.

Correct. I noticed gallivm requires some functions/methods declared under
LLVM's ExecutionEngine and JIT:

target SharedLib: gallium_dri
(out/target/product/x86/obj/SHARED_LIBRARIES/gallium_dri_intermediates/LINKED/gallium_dri.so)
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp:155: error:
undefined reference to 'llvm::JITMemoryManager::~JITMemoryManager()'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp:155: error:
undefined reference to 'llvm::JITMemoryManager::~JITMemoryManager()'
external/llvm/include/llvm/ExecutionEngine/ExecutionEngine.h:595: error:
undefined reference to 'llvm::EngineBuilder::InitEngine()'
external/llvm/include/llvm/ExecutionEngine/ExecutionEngine.h:723: error:
undefined reference to 'llvm::EngineBuilder::selectTarget()'
external/llvm/include/llvm/ExecutionEngine/ExecutionEngine.h:723: error:
undefined reference to 'llvm::EngineBuilder::create(llvm::TargetMachine*)'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp:155: error:
undefined reference to 'llvm::JITMemoryManager::~JITMemoryManager()'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp:550: error:
undefined reference to 'llvm::JITMemoryManager::CreateDefaultMemManager()'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_misc.cpp:155: error:
undefined reference to 'llvm::JITMemoryManager::~JITMemoryManager()'
out/target/product/x86/obj/STATIC_LIBRARIES/libmesa_gallium_intermediates/libmesa_gallium.a(lp_bld_misc.o):lp_bld_misc.cpp:vtable
for DelegatingJITMemoryManager: error: undefined reference to
'llvm::RTDyldMemoryManager::getSymbolAddress(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&)'
out/target/product/x86/obj/STATIC_LIBRARIES/libmesa_gallium_intermediates/libmesa_gallium.a(lp_bld_misc.o):lp_bld_misc.cpp:vtable
for ShaderMemoryManager: error: undefined reference to
'llvm::RTDyldMemoryManager::getSymbolAddress(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&)'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_init.c:168: error:
undefined reference to 'LLVMDisposeExecutionEngine'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_init.c:377: error:
undefined reference to 'LLVMLinkInJIT'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_init.c:240: error:
undefined reference to 'LLVMGetExecutionEngineTargetData'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_init.c:168: error:
undefined reference to 'LLVMDisposeExecutionEngine'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_init.c:377: error:
undefined reference to 'LLVMLinkInJIT'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_init.c:168: error:
undefined reference to 'LLVMDisposeExecutionEngine'
external/mesa/src/gallium/auxiliary/gallivm/lp_bld_init.c:609: error:
undefined reference to 'LLVMGetPointerToGlobal'

As host libraries, these functions are not included inside shared library
libLLVM (which I referenced on mesa). That's why I changed it to target
(the only way I could figure out at that moment)

I don't know if we can build these both as host and target libraries or
even link with the host libraries to solve these dependencies. Do you know
if this is possible? Any other option maybe?

Regards
Pstglia

[pstglia]

Post by pstglia

I have a little concern you changed
libLLVMExecutionEngine and libLLVMJIT
from the host lib to the target lib.

I don't know if we can build these both as host and target libraries or

even

Post by pstglia
link with the host libraries to solve these dependencies. Do you know if
this is possible? Any other option maybe?

Surely it can.
Many llvm libs are built in this way, e.g.,
lib/IR/Android.mk, lib/Linker/Android.mk.
So it's better to do the same way.

Thank you again for the explanation! I've updated the llvm patch (attached)
in order to fix this.

[Mauro Rossi]

Hi pstglia,

In configure.ac file mesa checks for llvm 3.4.2 to allow building radeonsi.

I'm trying to build llvm R600 targets with version that comes with
kitkat-x86,
which is llvm 3.4-0 from AOSP android-4.4.4_r2.0.1
<https://android.googlesource.com/platform/external/llvm/+/android-4.4.4_r2.0.1>

I wonder if 3.4.0 will suffice, doing the trick to feed mesa with
declaration that llvm 3.4.2 is avalable.


I have seen that you had to change llvm-gen-intrinsics.mk in order to
comply with tblgen option: -gen-intrinsic

tblgen --help

...
-gen-intrinsic - Generate intrinsic information
-gen-tgt-intrinsic - Generate target intrinsic information



[patched llvm-gen-intrinsics.mk]
ifeq ($(LOCAL_IS_HOST_MODULE),true)
$(call transform-host-td-to-out,intrinsic)
else
$(call transform-device-td-to-out,intrinsic)
endif

Do I need to replace the second intrinsic with tgt-intrinsic to properly
call transform-device-td-to-out for the device (target)?

If that is the case the same ifeq..else..end construct should be used in
llvm-tblgen-rules.mk,
in order to build generic "%GenIntrinsics" (? some weeks needed just to
understand what they are ...) differently for host and device.

Mauro

[pstglia]

Hi Mauro

Post by Mauro Rossi
Hi pstglia,
In configure.ac file mesa checks for llvm 3.4.2 to allow building radeonsi.
I'm trying to build llvm R600 targets with version that comes with
kitkat-x86,
which is llvm 3.4-0 from AOSP android-4.4.4_r2.0.1
<https://android.googlesource.com/platform/external/llvm/+/android-4.4.4_r2.0.1>
I wonder if 3.4.0 will suffice, doing the trick to feed mesa with
declaration that llvm 3.4.2 is avalable.

Not so sure forcing version 3.4.2... Some required extensions by mesa
(radeonsi/gallivm) seems to be missing on this llvm version.
In certains parts of code, Mesa checks the declared llvm version in use
(directives -DHAVE_LLVM and DLLVM_VERSION_PATCH) and decides which include
file or piece of code to be used. Also, if not matched with a minimum
version, it forces an error on compiling.

This is what I think, but I can be wrong.

Plus, by enabling the required llvm features needed by radeonsi
(ExecutionEngine, JIT, R600) You'll probably face these errors on llvm
compiling:

external/llvm/lib/ExecutionEngine/JIT/JITMemoryManager.cpp: In constructor
'{anonymous}::StatSymbols::StatSymbols()':
external/llvm/lib/ExecutionEngine/JIT/JITMemoryManager.cpp:810:66: error:
'open64' was not declared in this scope
...
target C++: libLLVMR600CodeGen <=
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp: In member function
'int
{anonymous}::AMDGPUCFGStructurizer::handleJumpintoIfImp(llvm::MachineBasicBlock*,
llvm::MachineBasicBlock*, llvm::MachineBasicBlock*)':
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp:1236:57: error:
lvalue required as increment operand
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp: In member function
'void
{anonymous}::AMDGPUCFGStructurizer::removeRedundantConditionalBranch(llvm::MachineBasicBlock*)':
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp:1716:49: error:
lvalue required as increment operand
make: ***
[out/target/product/x86/obj/STATIC_LIBRARIES/libLLVMR600CodeGen_intermediates/AMDILCFGStructurizer.o]
Error 1

This will require a lot of work and tricks.

Post by Mauro Rossi
I have seen that you had to change llvm-gen-intrinsics.mk in order to
comply with tblgen option: -gen-intrinsic
tblgen --help
...
-gen-intrinsic - Generate intrinsic information
-gen-tgt-intrinsic - Generate target intrinsic information
[patched llvm-gen-intrinsics.mk]
ifeq ($(LOCAL_IS_HOST_MODULE),true)
$(call transform-host-td-to-out,intrinsic)
else
$(call transform-device-td-to-out,intrinsic)
endif
Do I need to replace the second intrinsic with tgt-intrinsic to properly
call transform-device-td-to-out for the device (target)?

In fact I had changed only llvm-tblgen-rules.mk because it was configured
with this:

- $(call transform-td-to-out,tgt_intrinsics)

this resulted param "gen-tgt_intrinsics" producing an error. I replaced
with this:

+ $(call transform-td-to-out,tgt-intrinsic)

which results param "gen-tgt-intrinsic" matching what tblgen accepts. I'd
say it is a typo...




Regards
pstglia

[Mauro Rossi]

Post by pstglia
Plus, by enabling the required llvm features needed by radeonsi
(ExecutionEngine, JIT, R600) You'll probably face these errors on llvm
external/llvm/lib/ExecutionEngine/JIT/JITMemoryManager.cpp: In constructor
'open64' was not declared in this scope
...

This one I could solve by using part of this patch:

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20120409/140837.html

In the attachment the only two changes applicable.

Post by pstglia
target C++: libLLVMR600CodeGen <=
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp: In member function
'int
{anonymous}::AMDGPUCFGStructurizer::handleJumpintoIfImp(llvm::MachineBasicBlock*,
lvalue required as increment operand
external/llvm/lib/Target/R600/AMDILCFGStructurizer.cpp: In member function
'void
lvalue required as increment operand
make: ***
[out/target/product/x86/obj/STATIC_LIBRARIES/libLLVMR600CodeGen_intermediates/AMDILCFGStructurizer.o]
Error 1
This will require a lot of work and tricks.

Not solved yet :-(

I also encountered some errors due to missing LOCAL_C_INCLUDES, in InstPrinter/Android.mk,
MCTargetDesc/Android.mk and TargetInfo/Android.mk

LOCAL_C_INCLUDES := \
$(LOCAL_PATH)/.. \
$(intermediates)/../libLLVMR600CodeGen_intermediates \
+ $(intermediates)/../libLLVMCore_intermediates

Post by pstglia
- $(call transform-td-to-out,tgt_intrinsics)
this resulted param "gen-tgt_intrinsics" producing an error. I replaced
+ $(call transform-td-to-out,tgt-intrinsic)
which results param "gen-tgt-intrinsic" matching what tblgen accepts. I'd

say it is a typo...

I noticed when the generated file was not created, do the
scripts llvm-gen-intrinsics.mk and llvm-gen-intrinsics.mk need to check
LOCAL_MODULE_IS_HOST to select between intrinsic (for host librearies) and
tgt-intrinsic (for target ones)?
Thanks

I also tested your lollipop ISO with mesa radeonsi 10.4.2 it works with AMD
R7 250 and AMD HD7750.
I could not check OpenGL apps (besides the animated wallpapers) because I'm
having problems in setting up an HDD installation with gapps 5.1.

When I built ISO with mesa-10.6, I had problems at boot, do you have an ISO
with with mesa radeonsi 10.6 gallium_dri ? Thanks

I'm trying to test OpenGL using AndroidStudio run/debug, like I did for
kitkat-x86, by importing part of framework/native/opengl/tests (GLdual,
testPauseResume, testViewport) and the infamous harism_effects

Mauro

[Mauro Rossi]

Here is micropatch that solved the 'open64' problem.

M.

[Mauro Rossi]

It works with AMD R7 250 and AMD HD7750

Correction: tested with HD7750 (equal to R7 250) and HD7*9*50

Mauro

[pstglia]

Post by Mauro Rossi
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20120409/140837.html

Good catch!


I noticed when the generated file was not created, do the scripts

Post by Mauro Rossi
llvm-gen-intrinsics.mk and llvm-gen-intrinsics.mk need to check
LOCAL_MODULE_IS_HOST to select between intrinsic (for host librearies) and
tgt-intrinsic (for target ones)?
Thanks

All that I could figure out is that llvm-tblgen-rules.mk tries to find
files under generated dir with mask "*GenIntrinsics.inc". If found, it uses
llvm-tblgen to create some outputs.
Honestly, I have no idea the rule used to determine if it's a host or a
target.

I just changed it to fix this error I had during compiling:

make: Entering directory `/home/paulo/Downloads/android-x86-lollipop'
Host TableGen: libLLVMR600CodeGen (gen-tgt_intrinsics) <=
external/llvm/lib/Target/R600/AMDGPU.td
llvm-tblgen: Unknown command line argument '-gen-tgt_intrinsics'. Try:
'out/host/linux-x86/bin/llvm-tblgen -help'
llvm-tblgen: Did you mean '-gen-tgt-intrinsic'?
make: ***
[out/target/product/x86/gen/STATIC_LIBRARIES/libLLVMR600CodeGen_intermediates/AMDGPUGenIntrinsics.inc]
Error 1
make: Leaving directory `/home/paulo/Downloads/android-x86-lollipop'

Post by Mauro Rossi
I also tested your lollipop ISO with mesa radeonsi 10.4.2 it works with
AMD R7 250 and AMD HD7750.
I could not check OpenGL apps (besides the animated wallpapers) because
I'm having problems in setting up an HDD installation with gapps 5.1.

Something related to the fs type? Some other threads mention the need for
ext4...

Post by Mauro Rossi
When I built ISO with mesa-10.6, I had problems at boot, do you have an
ISO with with mesa radeonsi 10.6 gallium_dri ? Thanks

My ISO with 10.6 gives me an unusable flickering image and these errors:

# This crashes RenderThread
05-02 14:14:17.683 2129 2129 D OpenGLRenderer: setting property
extraRasterBucket to 0.95
05-02 14:14:17.683 2129 2129 D OpenGLRenderer: failed
05-02 14:14:17.683 2129 2129 D OpenGLRenderer: setting property
extraRasterBucket to 0.0
05-02 14:14:17.683 2129 2129 D OpenGLRenderer: failed

Seems related
to: https://android.googlesource.com/platform/frameworks/base/+/a736cd9%5E!/

# Also have lots of GL_INVALID_OPERATION
05-02 14:14:20.646 2587 2776 E OpenGLRenderer: GL error:
GL_INVALID_OPERATION

Don't know if these commits are related or if some include is missing on
Android Makefiles...:

commit 627991dbf74ce5aee9ce75155fc27a429bdd0548
Author: Brian Paul <***@vmware.com>
Date: Wed Mar 18 12:25:03 2015 -0600

dri: add _glapi_set_nop_handler(), _glapi_new_nop_table() to dri_test.c

commit 4bdbb588a9d385509f9168e38bfdb76952ba469c
Author: Brian Paul <***@vmware.com>
Date: Fri Mar 13 11:43:44 2015 -0600

mesa: reimplement dispatch table no-op function handling

commit 201e36e77d6ca616f75f14d5f1c31f0062ae4366
Author: Brian Paul <***@vmware.com>
Date: Fri Mar 13 10:20:29 2015 -0600

mapi: add new _glapi_new_nop_table() and _glapi_set_nop_handler()

[Mauro Rossi]

Thanks for the infos

Post by Mauro Rossi

make: *** [out/target/product/x86/obj/STATIC_LIBRARIES/libLLVMR600CodeGen_

intermediates/AMDILCFGStructurizer.o] Error 1

Post by Mauro Rossi

This will require a lot of work and tricks.

Not solved yet :-(

In order to solve this problem I defined the full list of sources and
replicated the makefile behavior for building all .inc generated files for
the three R600 sub targets (InstPrinter, MCTargetDesc and TargetInfo)

https://github.com/maurossi/llvm/commits/kitkat-x86_radeonsi_llvm34_WIP


Now I see a new, but old :-) problem, C++11 is required also to build llvm
R600 targets, as pointed out here: https://trac.macports.org/ticket/41548
STLport will not help according to people discussing this problem.

target C++: libLLVMR600CodeGen <=
external/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
external/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp: In member function
'virtual llvm::SDNode*
{anonymous}::AMDGPUDAGToDAGISel::Select(llvm::SDNode*)':
external/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp:254:42: error: 'class
*std::vector*<llvm::SDValue>' has no member named 'data'
...
external/llvm/lib/Target/R600/AMDGPUISelDAGToDAG.cpp:472:19: error: 'class
*std::vector*<llvm::SDValue>' has no member named 'data'
build/core/binary.mk:432: recipe for target
'out/target/product/x86/obj/STATIC_LIBRARIES/libLLVMR600CodeGen_intermediates/AMDGPUISelDAGToDAG.o'
failed
make: ***
[out/target/product/x86/obj/STATIC_LIBRARIES/libLLVMR600CodeGen_intermediates/AMDGPUISelDAGToDAG.o]
Error 1
make: Leaving directory '/home/utente/kitkat-x86'

Is lollipop's llvm3.5 using C++11 ?
Thanks

Mauro

[pstglia]

Good to know you are progressing on this.


Is lollipop's llvm3.5 using C++11 ?
Yes it is. It uses external/libcxx, which is part of llvm package

Also, if you check llvm lollipop Makefiles (llvm-host-build.mk and
llvm-device-build.mk), they set "-std=c++11" flags

Regards,
Pstglia

[pstglia]

Post by pstglia
Thanks for this. Still trying to improve it (making better patches)

Sorry for reply late.
I've pushed the refined patch to the server.
Thank you!

Thanks Chih-Wei! I saw the changes. Much better that way, like enabling
llvm just when radeonsi is set.

[Mauro Rossi]

Hi,

in order to trigger the build of LLVM when radeonsi is declared in
BoardConfig.mk,
if I'm not wrong, one line should be changed in external/mesa/Android.mk,
because $(gallium_drivers) variable contains radeonsi by definition.

-MESA_ENABLE_LLVM := $(if $(filter radeonsi,$(gallium_drivers)),true,false)
+MESA_ENABLE_LLVM := $(if $(filter radeonsi,$(MESA_GPU_DRIVERS)),true,false)

M.

Post by pstglia

Post by pstglia
Thanks for this. Still trying to improve it (making better patches)
Sorry for reply late.
I've pushed the refined patch to the server.
Thank you!

Thanks Chih-Wei! I saw the changes. Much better that way, like enabling
llvm just when radeonsi is set.

[rbg]

http://git.android-x86.org/?p=platform/build.git;a=commit;h=c7c75c48ed2f6331332570cdecb32f8e62f0fe45

also needs to be pushed to lollipop...

Post by Mauro Rossi

Post by Mauro Rossi
Hi,
in order to trigger the build of LLVM when radeonsi is declared in
BoardConfig.mk,
if I'm not wrong, one line should be changed in

external/mesa/Android.mk,

Post by Mauro Rossi
because $(gallium_drivers) variable contains radeonsi by definition.
-MESA_ENABLE_LLVM := $(if $(filter

radeonsi,$(gallium_drivers)),true,false)

Post by Mauro Rossi
+MESA_ENABLE_LLVM := $(if $(filter

radeonsi,$(MESA_GPU_DRIVERS)),true,false)
You're right.
I'll correct it.
Thank you!

[pstglia]

If anyone is interested, I've uploaded a new build (lollipop, kernel 4.0,
mesa 10.5.4). Also there's a patch applied as an attempt to fix the video
playback issue on Lollipop. radeonsi is also enabled.

# ISO link
https://drive.google.com/file/d/0Bx12U5yGNcQCLVc4NzNkYzZjTms/view?usp=sharing

# Patch file for mesa 10.5.4
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.5.4.patch

# Patch for libstagefright (conversion color format forced
to HAL_PIXEL_FORMAT_RGB_565)
# apply on frameworks/av/media/libstagefright/colorconversion
0001-Use-HAL_PIXEL_FORMAT_RGB_565-as-default-conversion-f.patch

Did a very basic test on my machine (executed Antutu and some youtube
videos on embedded browser). Worked ok

Regards,
Pstglia

[rbg]

is this also needed?????

device/generic/common/BoardConfig.mk

#BOARD_GPU_DRIVERS ?= i915 i965 ilo r300g r600g nouveau vmwgfx
- BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g
+ BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g radeonsi

Post by pstglia
If anyone is interested, I've uploaded a new build (lollipop, kernel 4.0,
mesa 10.5.4). Also there's a patch applied as an attempt to fix the video
playback issue on Lollipop. radeonsi is also enabled.
# ISO link
https://drive.google.com/file/d/0Bx12U5yGNcQCLVc4NzNkYzZjTms/view?usp=sharing
# Patch file for mesa 10.5.4
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.5.4.patch
# Patch for libstagefright (conversion color format forced
to HAL_PIXEL_FORMAT_RGB_565)
# apply on frameworks/av/media/libstagefright/colorconversion
0001-Use-HAL_PIXEL_FORMAT_RGB_565-as-default-conversion-f.patch
Did a very basic test on my machine (executed Antutu and some youtube
videos on embedded browser). Worked ok
Regards,
Pstglia

[pstglia]

Hi Rbg,

device/generic/common/BoardConfig.mk

Post by rbg
#BOARD_GPU_DRIVERS ?= i915 i965 ilo r300g r600g nouveau vmwgfx
- BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g
+ BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g radeonsi

To enable radeonsi yes, it's required.

Let me make a more detailed list what is necessary in order to enable it on
a lollipop compiling:

# Mesa patches (external/mesa):

a) If you are going to build mesa 10.6, use this (posted April, 22th):
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.6.patch

b) If building with 10.5.4, use this (posted May, 5th):
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.5.4.patch

# LLVM patch (external/llvm)

Apply this patch (posted on May, 1st):
0001-Enable-R600-support-on-Android-x86-mesa-gallium-radeon_lollipop_20150501.patch

# drm gralloc (hardware/drm_gralloc)
Use this (posted April, 22th):
0001-Enable-radeonsi-on-Android-x86.patch

# BoardConfig.mk (device/generic/common)
Include radeonsi in BOARD_GPU_DRIVERS variable

* * *

Optionally (no relation with radeonsi), if you want a fix (hopefully - no
warranties yet :D ) for video playback on Lollipop, apply this patch
# Patch for libstagefright (conversion color format forced
to HAL_PIXEL_FORMAT_RGB_565)
# apply on frameworks/av/media/libstagefright/colorconversion
0001-Use-HAL_PIXEL_FORMAT_RGB_565-as-default-conversion-f.patch

Pstglia

[rbg]

Post by pstglia
Optionally (no relation with radeonsi), if you want a fix (hopefully - no

warranties yet :D ) for video playback on Lollipop, apply this patch

Post by pstglia
# Patch for libstagefright (conversion color format forced to

HAL_PIXEL_FORMAT_RGB_565)

Post by pstglia
# apply on frameworks/av/media/libstagefright/colorconversion
0001-Use-HAL_PIXEL_FORMAT_RGB_565-as-default-conversion-f.patch

confirmed on lollipop 5.1.1 / mesa 10.4.2, using browser
the only web site I had issues with playing videos was espn.....still only
get black screen instead of video

Yahoo / MSN / CNN and other sites showed video fine...

Youtube.com showed video fine...

tested on: asus PBZ77-M motherboard, Integrated Intel Graphics HD 4000
asus P5G41-M motherboard, Intel G41 Express ICH7 chipset,
Integrated Intel Graphics Media Accelerator(Intel® GMA X4500)

before this patch was unable to get any video anywhere............

Post by pstglia
Hi Rbg,
device/generic/common/BoardConfig.mk

Post by rbg
#BOARD_GPU_DRIVERS ?= i915 i965 ilo r300g r600g nouveau vmwgfx
- BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g
+ BOARD_GPU_DRIVERS ?= i915 i965 swrast r300g r600g radeonsi

To enable radeonsi yes, it's required.
Let me make a more detailed list what is necessary in order to enable it
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.6.patch
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.5.4.patch
# LLVM patch (external/llvm)
0001-Enable-R600-support-on-Android-x86-mesa-gallium-radeon_lollipop_20150501.patch
# drm gralloc (hardware/drm_gralloc)
0001-Enable-radeonsi-on-Android-x86.patch
# BoardConfig.mk (device/generic/common)
Include radeonsi in BOARD_GPU_DRIVERS variable
* * *
Optionally (no relation with radeonsi), if you want a fix (hopefully - no
warranties yet :D ) for video playback on Lollipop, apply this patch
# Patch for libstagefright (conversion color format forced
to HAL_PIXEL_FORMAT_RGB_565)
# apply on frameworks/av/media/libstagefright/colorconversion
0001-Use-HAL_PIXEL_FORMAT_RGB_565-as-default-conversion-f.patch
Pstglia

[pstglia]

So does that mean we have a working video Playback on 5.1/4.0 Intel??

I'd say yes. But it's better to test more hw just to make sure.

[pstglia]

Cool. OK. I'll do an install on GM45 chipset and report back.
Good work!
Mike

Tks... Just to make sure. Rbg tested a good amount of intel plataforms
already.

I've tested on my A10 5800K, working without problems I could notice.

Now let's try to conquer more issues :)

[hashirt]

Hi
I tested that iso
(android-x86-5.1_kernel_4.0_mesa_10.5.4_radeonsi_test_20150505.iso) on two
machines (pc and laptop).

Pc wouldn't pass through "android logo screen".
motherboard: biostar g31m
pentium 4
intel gma 3100
intel g33/g31 express chipset family
attaching: pc-log.txt, pc-dmesg.txt

On laptop, system boot was successful (as far as i can tell). I tested some
kernel 4.0 builds before, this build was the first one that boots. I
checked some video sites like youtube, dailymotion and had no issue playing
videos.
model: toshiba satellite a300
core 2 duo
ati mobility radeon hd 3470
attaching: laptop-log.txt, laptop-dmesg.txt

I used grub4dos for both cases.

I also have doubt that I am doing something wrong while acquiring dmesg and
log files. Just wanted to confirm. Here is what I did for this case.
-Boot into debug mode
-After boot is successful or boot fails-> #alt+f2
#mkdir /mnt/zzz
#mount /dev/sdb1 /mnt/zzz
#dmesg > /mnt/zzz/dmesg.txt
after dmesg.txt is on usb
-boot puppylinux, mount the partition android-x86 is installed
-copy android-2015-05-05/data/log.txt to usb

Regards

Post by pstglia
If anyone is interested, I've uploaded a new build (lollipop, kernel 4.0,
mesa 10.5.4). Also there's a patch applied as an attempt to fix the video
playback issue on Lollipop. radeonsi is also enabled.
# ISO link
https://drive.google.com/file/d/0Bx12U5yGNcQCLVc4NzNkYzZjTms/view?usp=sharing
# Patch file for mesa 10.5.4
0001-Enable-radeonsi-on-Android-x86-lollipop_mesa_10.5.4.patch
# Patch for libstagefright (conversion color format forced
to HAL_PIXEL_FORMAT_RGB_565)
# apply on frameworks/av/media/libstagefright/colorconversion
0001-Use-HAL_PIXEL_FORMAT_RGB_565-as-default-conversion-f.patch
Did a very basic test on my machine (executed Antutu and some youtube
videos on embedded browser). Worked ok
Regards,
Pstglia

[pstglia]

Hi hashirt

Post by hashirt
Pc wouldn't pass through "android logo screen".
motherboard: biostar g31m
pentium 4
intel gma 3100
intel g33/g31 express chipset family
attaching: pc-log.txt, pc-dmesg.txt

Take a look at the logcat. A core process (Zygote) is on an endless loop,
receiving a SIGILL signal.
This signal means there's a some instruction is invalid/not supported. As
this is an old processor (Pentium 4) which lacks support for newer
instructions used on this build (SSE 4.1 for mesa is one of them).
This is the most probably reason it's not working.

Post by hashirt
On laptop, system boot was successful (as far as i can tell). I tested
some kernel 4.0 builds before, this build was the first one that boots. I
checked some video sites like youtube, dailymotion and had no issue playing
videos.
model: toshiba satellite a300
core 2 duo
ati mobility radeon hd 3470
attaching: laptop-log.txt, laptop-dmesg.txt
I used grub4dos for both cases.

Great. Thanks for sharing your results

Post by hashirt
I also have doubt that I am doing something wrong while acquiring dmesg
and log files. Just wanted to confirm. Here is what I did for this case.
-Boot into debug mode
-After boot is successful or boot fails-> #alt+f2
#mkdir /mnt/zzz
#mount /dev/sdb1 /mnt/zzz
#dmesg > /mnt/zzz/dmesg.txt
after dmesg.txt is on usb
-boot puppylinux, mount the partition android-x86 is installed
-copy android-2015-05-05/data/log.txt to usb

It's ok. But you don't need to boot Linux in order to get log.txt. You can
get it from Android itself, from /data directory

[pstglia]

Post by pstglia
Take a look at the logcat. A core process (Zygote) is on an endless loop,
receiving a SIGILL signal.
This signal means there's a some instruction is invalid/not supported. As
this is an old processor (Pentium 4) which lacks support for newer
instructions used on this build (SSE 4.1 for mesa is one of them).
This is the most probably reason it's not working.

In fact, not really sure about SSE4.1 usage (although it's possible)... But
SSSE3 is really used. Pentium 4 just supports SSE3.

[Mauro Rossi]

Hi,
good news about radeonsi tests with mesa 10.5.4

Post by pstglia
# Patch file for mesa 10.5.4
....
Did a very basic test on my machine (executed Antutu and some youtube
videos on embedded browser). Worked ok
Regards,
Pstglia

I've tested lollipop with mesa 10.5.4 on HD7750 and HD7950, the result are
equivalent for the two videocards.

*Working perfectly OK:*

OpenGLES animated wallpapers
Graphics with OpenGL ES (by Thorsten Thormaehlen)
GLdual
testViewport
testPauseResume
testViewport
android_effects

*Little flaws (not much impact for usability):*

"green leaves" default wallpaper: gives a black wallpaper
OpenGL ES 1.0 Demo: 1st cube is rendered, following cube and demos are not
visible, showing a black surface
*gl_perfapp:* causes GUI restart (log attached)
*android_sharize:* it's running slow because of cyclically repeating 0x501
errors (log attached)


Mesa 10.5.4 radeonsi shows stability with most OpenGL tests/demos, it was
hard to find apps causing problems.

Mauro

PS: with mesa 10.6devel I get no GUI too, but I need to setup a stable test
environment to check, I'm still not able to properly reboot lollipop with
EXT4.
Since I am not able to install grub2 properly, I'm doing a "one-session
only" reformat EXT4 + install and selecting "Run Android" just after
installation.

[pstglia]

Post by Mauro Rossi
*Working perfectly OK:*
OpenGLES animated wallpapers
Graphics with OpenGL ES (by Thorsten Thormaehlen)
GLdual
testViewport
testPauseResume
testViewport
android_effects

Great. Have you tested some other x86 supported apps like Angry birds
series or Antutu?

*android_sharize:* it's running slow because of cyclically repeating 0x501

Post by Mauro Rossi
errors (log attached)

Do you have this error on other cards?

Thanks for your reports! Hope we have better luck with radeonsi than
nouveau :D

[Jarvis Chung]

Hi Pstglia,

I build a kernel 4.0 + mesa 10.5.4 to test asus R7 240 and rund Antutu , I
got error as below(Angry Birds Rio is ok):

4651 ActivityManager I START u0 {act=android.intent.action.MAIN
cat=[android.intent.category.LAUNCHER] flg=0x10200000
cmp=com.antutu.ABenchMark/.ABenchMarkStart} from uid 10014 on display 0
4651 AudioTrack W AUDIO_OUTPUT_FLAG_FAST denied by client
4420 AudioPolicyManager E getDeviceForStrategy() speaker device not
found for STRATEGY_SONIFICATION
4420 AudioPolicyManager E getDeviceForStrategy() speaker device not
found for STRATEGY_SONIFICATION
4420 AudioPolicyManager E getDeviceForStrategy() no device found
for STRATEGY_TRANSMITTED_THROUGH_SPEAKER
4651 WindowManager V addAppToken: AppWindowToken{16c1022f
token=Token{2995a80e ActivityRecord{3be5ce09 u0
com.antutu.ABenchMark/.ABenchMarkStart t8}}} to stack=1 task=8 at 0
4651 WindowManager V Adding window Window{6d45428 u0 Starting
com.antutu.ABenchMark} at 2 of 7 (after Window{20ab45be u0
com.cyanogenmod.trebuchet/com.android.launcher3.Launcher})
4420 audio_hw_primary_x86 D found out /dev/snd/pcmC0D3p
4651 ActivityManager I Start proc
5520:com.antutu.ABenchMark/u0a52 for activity
com.antutu.ABenchMark/.ABenchMarkStart
5520 ActivityThread I Switching default density from 160 to 240
5520 Settings W Setting android_id has moved from
android.provider.Settings.System to android.provider.Settings.Secure,
returning read-only value.
5520 OpenGLRenderer D Use EGL_SWAP_BEHAVIOR_PRESERVED: true
5520 GRALLOC-RADEON I detected chipset 0x6613 family 0x38 (vram
size 1948MiB, gart size 1019MiB)
5520 GRALLOC-DRM I create radeon for driver radeon
5520 Atlas D Validating map...
4651 WindowManager V Adding window Window{214b79 u0
com.antutu.ABenchMark/com.antutu.ABenchMark.ABenchMarkStart} at 2 of 8
(before Window{6d45428 u0 Starting com.antutu.ABenchMark})
4420 AudioPolicyManager E getDeviceForStrategy() speaker device not
found for STRATEGY_SONIFICATION
4420 AudioPolicyManager E getDeviceForStrategy() speaker device not
found for STRATEGY_SONIFICATION
4420 AudioPolicyManager E getDeviceForStrategy() no device found
for STRATEGY_TRANSMITTED_THROUGH_SPEAKER
5520 EGL-GALLIUM I using DRM screen
5520 EGL-GALLIUM I skip unsupported native format 0x3
5520 OpenGLRenderer I Initialized EGL, version 1.4
5520 OpenGLRenderer D Enabling debug mode 0
4651 WindowManager V Adding window Window{c10606c u0
SurfaceView} at 2 of 9 (before Window{214b79 u0
com.antutu.ABenchMark/com.antutu.ABenchMark.ABenchMarkStart})
5520 Settings W Setting android_id has moved from
android.provider.Settings.System to android.provider.Settings.Secure,
returning read-only value.
4651 ActivityManager I Start proc
5597:com.android.defcontainer/u0a3 for service
com.android.defcontainer/.DefaultContainerService
4651 ActivityManager I Displayed
com.antutu.ABenchMark/.ABenchMarkStart: +404ms
5597 ActivityThread I Switching default density from 160 to 240
4651 OpenGLRenderer D Use EGL_SWAP_BEHAVIOR_PRESERVED: true
4651 Atlas D Validating map...
4651 EGL-GALLIUM I using DRM screen
4651 EGL-GALLIUM I skip unsupported native format 0x3
4651 OpenGLRenderer I Initialized EGL, version 1.4
4651 OpenGLRenderer D Enabling debug mode 0
4651 EGL-GALLIUM E failed to import buffer 0xa0c7ad08
4651 OpenGLRenderer W Error creating image (0x3000)
4651 OpenGLRenderer W Could not create atlas image
4651 libc F Fatal signal 11 (SIGSEGV), code 1, fault
addr 0x4 in tid 5616 (RenderThread)
1393 DEBUG I *** *** *** *** *** *** *** *** *** ***
*** *** *** *** *** ***
1393 DEBUG I Build fingerprint:
'Android-x86/android_x86/x86:5.1.1/LMY47V/jarvis05111517:eng/test-keys'
1393 DEBUG I Revision: '0'
1393 DEBUG I ABI: 'x86'
1393 DEBUG I pid: 4651, tid: 5616, name: RenderThread

Post by Mauro Rossi

Post by Mauro Rossi

system_server <<<

1393 DEBUG I signal 11 (SIGSEGV), code 1
(SEGV_MAPERR), fault addr 0x4
1393 DEBUG I eax 00000000 ebx b6fd653c ecx ae35803c
edx 00000000
1393 DEBUG I esi ae18bac8 edi ae38e748
1393 DEBUG I xcs 00000073 xds 0000007b xes 0000007b
xfs 0000018f xss 0000007b
1393 DEBUG I eip b6f3c08c ebp 9c8ffac8 esp 9c8ffa60
flags 00210246
1393 DEBUG I backtrace:
1393 DEBUG I #00 pc 0001f08c /system/lib/libhwui.so
(android::uirenderer::AssetAtlas::init(android::sp<android::GraphicBuffer>,
long long*, int)+220)
1393 DEBUG I #01 pc 00092e1f /system/lib/libhwui.so
1393 DEBUG I #02 pc 0008cfe2 /system/lib/libhwui.so
1393 DEBUG I #03 pc 00093dd3 /system/lib/libhwui.so
1393 DEBUG I #04 pc 00093ba0 /system/lib/libhwui.so
1393 DEBUG I #05 pc 00095594 /system/lib/libhwui.so
1393 DEBUG I #06 pc 00096e62 /system/lib/libhwui.so
(android::uirenderer::renderthread::RenderThread::threadLoop()+162)
1393 DEBUG I #07 pc 00017de4 /system/lib/libutils.so
(android::Thread::_threadLoop(void*)+308)
1393 DEBUG I #08 pc 00070775
/system/lib/libandroid_runtime.so
(android::AndroidRuntime::javaThreadShell(void*)+117)
1393 DEBUG I #09 pc 000173f2 /system/lib/libutils.so
(thread_data_t::trampoline(thread_data_t const*)+130)
1393 DEBUG I #10 pc 00022168 /system/lib/libc.so
(__pthread_start(void*)+56)
1393 DEBUG I #11 pc 0001cc69 /system/lib/libc.so
(__start_thread+25)
1393 DEBUG I #12 pc 000137c6 /system/lib/libc.so
(__bionic_clone+70)
1393 libbacktrace W virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb750e82d
1393 libbacktrace W virtual bool
Backtrace::VerifyReadWordArgs(uintptr_t, word_t*): invalid pointer
0xb69b7873
1393 DEBUG I Tombstone written to:
/data/tombstones/tombstone_02
1387 ServiceManager I service 'wifiscanner' died
4893 WifiManager E Channel connection lost
4743 WifiManager E Channel connection lost
4420 AudioFlinger W power manager service died !!!
1387 ServiceManager I service 'clipboard' died
1387 ServiceManager I service 'statusbar' died

But, I success to run on AMD HD6670, include Antutu and 3D game.

Thanks.

pstgliaæ–Œ 2015幎5月12日星期二 UTC+8䞊午10時12分21秒寫道

Post by Mauro Rossi
*Working perfectly OK:*

Post by Mauro Rossi
OpenGLES animated wallpapers
Graphics with OpenGL ES (by Thorsten Thormaehlen)
GLdual
testViewport
testPauseResume
testViewport
android_effects

Great. Have you tested some other x86 supported apps like Angry birds
series or Antutu?
*android_sharize:* it's running slow because of cyclically repeating

Post by Mauro Rossi
0x501 errors (log attached)

Do you have this error on other cards?
Thanks for your reports! Hope we have better luck with radeonsi than
nouveau :D

[pstglia]

Hi Jarvis, thanks for testing:

4651 OpenGLRenderer D Use EGL_SWAP_BEHAVIOR_PRESERVED: true

Post by Jarvis Chung
4651 Atlas D Validating map...
4651 EGL-GALLIUM I using DRM screen
4651 EGL-GALLIUM I skip unsupported native format 0x3
4651 OpenGLRenderer I Initialized EGL, version 1.4
4651 OpenGLRenderer D Enabling debug mode 0
4651 EGL-GALLIUM E failed to import buffer 0xa0c7ad08
4651 OpenGLRenderer W Error creating image (0x3000)
4651 OpenGLRenderer W Could not create atlas image
4651 libc F Fatal signal 11 (SIGSEGV), code 1, fault
addr 0x4 in tid 5616 (RenderThread)

This error appears to be the same UCyborg reported. The first error message
"failed to import buffer" is returned
on external/mesa/src/gallium/state_trackers/egl/android/native_android.cpp
when there is no handle resource:

static struct pipe_resource *
import_buffer(struct android_display *adpy, const struct pipe_resource
*templ,
ANativeWindowBuffer *abuf)
{
...
if (!res)
ALOGE("failed to import buffer %p", abuf);

return res;


This "res" structure is populated by resource_from_handle. Handle type can
be "DRM_API_HANDLE_TYPE_SHARED" or "abuf->handle" depending on
"adpy->use_drm" value

I'd like to have more info on this (check if drm is being used or not for
wach hw). If possible, can you overwrite native_android.cpp by the one
attached here, create a new img and run again on both hardwares?
It contains some extra ALOGE print messages that should confirm this.

Thanks!

[Jarvis Chung]

Hi Pstglia,

Thanks for your help.

I tested my device again and attached log files.
*Asus R7 240 + debug patch*
* dmesg_pc_debug_kernel_4_mesa_10.5.4_20150514.log*
* logcat_pc_debug_kernel_4_mesa_10.5.4_20150514*

*Asus R7 240*
*dmesg_pc_kernel_4_mesa_10.5.4_20150514.log*
* logcat_pc_kernel_4_mesa_10.5.4_20150514.log*

HD6670 + debug patch
*dmesg_pc_hd6670_debug_kernel_4_mesa_10.5.4_20150514.log*
* logcat_pc_hd6670_debug_kernel_4_mesa_10.5.4_20150514.log*

The log had other debug log, ex: PCM, please skip it.

Thanks.
Jarvis


pstgliaæ–Œ 2015幎5月13日星期䞉 UTC+8䞋午9時17分03秒寫道

Post by Jarvis Chung
4651 OpenGLRenderer D Use EGL_SWAP_BEHAVIOR_PRESERVED: true

Post by Jarvis Chung
4651 Atlas D Validating map...
4651 EGL-GALLIUM I using DRM screen
4651 EGL-GALLIUM I skip unsupported native format 0x3
4651 OpenGLRenderer I Initialized EGL, version 1.4
4651 OpenGLRenderer D Enabling debug mode 0
4651 EGL-GALLIUM E failed to import buffer 0xa0c7ad08
4651 OpenGLRenderer W Error creating image (0x3000)
4651 OpenGLRenderer W Could not create atlas image
4651 libc F Fatal signal 11 (SIGSEGV), code 1,
fault addr 0x4 in tid 5616 (RenderThread)

This error appears to be the same UCyborg reported. The first error
message "failed to import buffer" is returned
on external/mesa/src/gallium/state_trackers/egl/android/native_android.cpp
static struct pipe_resource *
import_buffer(struct android_display *adpy, const struct pipe_resource
*templ,
ANativeWindowBuffer *abuf)
{
...
if (!res)
ALOGE("failed to import buffer %p", abuf);
return res;
This "res" structure is populated by resource_from_handle. Handle type can
be "DRM_API_HANDLE_TYPE_SHARED" or "abuf->handle" depending on
"adpy->use_drm" value
I'd like to have more info on this (check if drm is being used or not for
wach hw). If possible, can you overwrite native_android.cpp by the one
attached here, create a new img and run again on both hardwares?
It contains some extra ALOGE print messages that should confirm this.
Thanks!

[Mauro Rossi]

Post by Jarvis Chung
4651 EGL-GALLIUM E failed to import buffer 0xa0c7ad08
4651 OpenGLRenderer W Error creating image (0x3000)

Hi, could it be similar to this past problem, related to dri image formats
?

https://groups.google.com/forum/#!topic/android-x86/B903ihn37Ao

Mauro

[pstglia]

Hi Mauro,

Post by Mauro Rossi

Post by Jarvis Chung
4651 EGL-GALLIUM E failed to import buffer 0xa0c7ad08
4651 OpenGLRenderer W Error creating image (0x3000)

Hi, could it be similar to this past problem, related to dri image formats
?
https://groups.google.com/forum/#!topic/android-x86/B903ihn37Ao
Mauro

I'm not sure.... seems to be a different issue (lack of color format in
that case ) which resulted the same 0x3000 error. What do you suggest for
debugging?

What I found strange is KMS was not initialized for these cards Jarvis and
UCyborg tested. For your cards (which are quite similar to Blumi Milch one)
kms seems to be initialized (we have GRALLOC-KMS messages and Gallium
strings on logcat)

Maybe some unimplemented feature at kernel or libdrm side. Still debugging
to try to discover what's missing.

These cards lack info on Web also (Linux) which makes things a bit
difficult (remember: I'm just a Google searcher :D )

regards,
Pstglia

[Mauro Rossi]

Post by pstglia
I'm not sure.... seems to be a different issue (lack of color format in

that case ) which resulted the same 0x3000 error. What do you suggest for
debugging?

It's just a feeling, I would investigate the parameters (e.g. memtype) set
at dma-buf handle creation and check if they are supported.

Regarding debug logs (1 and 2 are well known):

1) enable drm tracing

drm.debug=7 as grub line, traces will be in dmesg log


2) enable EGL tracing in logcat (after GUI started ALT+F1)

setprop debug.egl.trace error


3) apitrace, probably the best if it works, to capture and replay the whole
EGL sequence

NOTE: part of the build procedure may be changed since the last time I
built and tested apitrace/qapitrace.

Build apitrace x86 on Ubuntu
--------------------------------------

git clone https://github.com/apitrace/apitrace

cd apitrace

export ANDROID_NDK=/home/utente/Downloads/android-ndk-r10d NOTE: your path
to NDK

cmake -DCMAKE_TOOLCHAIN_FILE=cmake/toolchain/android.toolchain.cmake
-DANDROID_ABI=x86 -DANDROID_TOOLCHAIN_NAME=x86-4.9 \
-DANDROID_API_LEVEL=9 -H. -Bbuild

make -C build


Upload the built egltrace.so library to Lollipop x86 device under test
-----------------------------------------------------------------------------------------

adb root
adb connect [your IP]
adb push libs/x86/egltrace.so data/
adb shell chmod 0777 /data/egltrace.so

Tracing with apitrace egltrace.so
-------------------------------------------

[example procedure for Settings - working]

adb root
PROCNAME=com.android.settings
adb shell setprop wrap.$PROCNAME LD_PRELOAD=/data/egltrace.so
adb shell setprop debug.apitrace.procname $PROCNAME
adb shell am start -S com.android.settings/.Settings

[procedure for Antutu - to be tried]

adb root
PROCNAME=com.antutu.ABenchMark
adb shell setprop wrap.$PROCNAME LD_PRELOAD=/data/egltrace.so
adb shell setprop debug.apitrace.procname $PROCNAME
adb shell am start -S com.antutu.ABenchMark/.ABenchMarkStart

Traces should be in folder data/data/com.antutu.ABenchMark/ and qapitrace
application can be used to examine the trace.

I would try myself, but as we speak I'm not able to reboot EXT4
installation.
How do you manage? Are you all using GRUB2 EFI boot?

Further instructions in the following links.

Mauro

*[how-to for building egltrace.so library with Android NDK]*

https://github.com/apitrace/apitrace/blob/master/INSTALL.markdown


*[how-to for tracing android apps]*

https://github.com/apitrace/apitrace/blob/master/Dalvik.markdown


*[how-to install latest version on Ubuntu to use qapitrace GUI and avoid
tracefiles version compatibility problems]*

sudo add-apt-repository ppa:gerboland/unity-mir
sudo apt-get update
sudo apt-get install apitrace-gl-frontend

[Mauro Rossi]

Post by Mauro Rossi
Build apitrace x86 on Ubuntu
-----------------------------------------

Fow those who wants to skip this part I have uploaded egltrace.so prebuilt
here:
https://drive.google.com/file/d/0B_OFHiIqgpSFUzVjcUF0NmdMX0U/view?usp=sharing


I could use some help with installing GRUB2 bios version to be able to
reboot lollipop-x86.

I have tried procedure described here: http://andex.exton.net/?p=165
and here: http://www.android-x86.net/index.php?topic=255.0

I see the new boot entry but I can boot Lollipop.
I'm using a PATA hard drive on ICH7 controller, do I need to load a module
at the beginning of grub 40_custom script?
Thanks

I'm having hard times, because I need to reinstall each time I need to
collect logs.
Mauro

[pstglia]

Post by Mauro Rossi
I could use some help with installing GRUB2 bios version to be able to
reboot lollipop-x86.
I have tried procedure described here: http://andex.exton.net/?p=165
and here: http://www.android-x86.net/index.php?topic=255.0
I see the new boot entry but I can boot Lollipop.
I'm using a PATA hard drive on ICH7 controller, do I need to load a module
at the beginning of grub 40_custom script?
Thanks
I'm having hard times, because I need to reinstall each time I need to
collect logs.
Mauro

Don't know if this helps. Here's my config entry for booting Lollipop (I
keep it on /boot/grub/custom.cfg):

menuentry "Android-x86 5.1" {
insmod ext2
set root='(hd0,8)'
linux /android-2015-05-14/kernel quiet root=/dev/ram0
androidboot.hardware=android_x86 video=-16 SRC=/android-2015-05-14
initrd /android-2015-05-14/initrd.img
}

My disk is SATA and Lollipop is installed in the 8th partition (extended).
I formatted it to ext4 inside Linux before installing Android:

Device Boot Start End Blocks Id System



/dev/sda1 XXXX XXXXXXX XXXXXXX 82 Linux swap
/dev/sda2 * XXXX XXXXXXX XXXXXXX 83 Linux
...
/dev/sda5 XXXX XXXXXXX XXXXXXX 83 Linux
...
/dev/sda8 XXXX XXXXXXX XXXXXXX 83 Linux


The date "2015-05-14" corresponds to the date my img was created as you can
see bellow (this is the content of my root directory):

drwxr-xr-x 4 root root 4096 May 14 17:34 *android-2015-05-14*

So, everytime I create a new ISO (so a new date on directory name), I have
to replace on custom.cfg and run "grub-install /dev/sda"

Note: Using Slackware distro, if that mathers.

[Mauro Rossi]

Thanks for your suggestions

I finally managed to understand the problem, grub has a command line with
essential tools.

ls [enter] command shows the list of hard drives

then I saw that /dev/sd*a*1 unexpectedly became (hd*1*,1), while I was
trying to load kernel from (hd0,1).

I was able to list the folders and files with:

ls (hd1,1)/ [enter]

ls (hd1,1)/android-2015-MM-DD [enter] MM and DD to be replaced with
actual month and day seen with previous command.

Now I could boot and I'm back in business with the debug / troubleshooting
!!!

Mauro

Post by Mauro Rossi
I could use some help with installing GRUB2 bios version to be able to

Post by Mauro Rossi
reboot lollipop-x86.
I have tried procedure described here: http://andex.exton.net/?p=165
and here: http://www.android-x86.net/index.php?topic=255.0
I see the new boot entry but I can boot Lollipop.
I'm using a PATA hard drive on ICH7 controller, do I need to load a
module at the beginning of grub 40_custom script?
Thanks
I'm having hard times, because I need to reinstall each time I need to
collect logs.
Mauro

Don't know if this helps. Here's my config entry for booting Lollipop (I
menuentry "Android-x86 5.1" {
insmod ext2
set root='(hd0,8)'
linux /android-2015-05-14/kernel quiet root=/dev/ram0
androidboot.hardware=android_x86 video=-16 SRC=/android-2015-05-14
initrd /android-2015-05-14/initrd.img
}
My disk is SATA and Lollipop is installed in the 8th partition (extended).
Device Boot Start End Blocks Id System
/dev/sda1 XXXX XXXXXXX XXXXXXX 82 Linux swap
/dev/sda2 * XXXX XXXXXXX XXXXXXX 83 Linux
...
/dev/sda5 XXXX XXXXXXX XXXXXXX 83 Linux
...
/dev/sda8 XXXX XXXXXXX XXXXXXX 83 Linux
The date "2015-05-14" corresponds to the date my img was created as you
drwxr-xr-x 4 root root 4096 May 14 17:34 *android-2015-05-14*
So, everytime I create a new ISO (so a new date on directory name), I have
to replace on custom.cfg and run "grub-install /dev/sda"
Note: Using Slackware distro, if that mathers.

[Luigi Capriotti]

What I do as I find very convenient is to have a linux install of any sort
on an ext4, then either install Android-x86 to the root partition without
formatting and with no grub.
BTW I can also copy Android-x86 files directly from the ISO/IMG as well.

With this arrangement I can have several Android-x86 builds on the same
partition and switch among them simply with a symbolic link (ln -s
Android-20150516 Android).

Ofc a grub custom entry needs to be added once in linux, but once done grub
stays untouched and allows to boot linux - ie the Android "recovery" ;) -
or any Andoid system previously linked.

[pstglia]

Hi Jarvis,

Post by Jarvis Chung
Thanks for your help.

Thank you and Mauro too for helping with debugging. I'll have to bother you
a little more :)

drm attributes are set. Now I'd like to know why KMS is not being
initialized for these cards.

Can you confirm if /proc/fb has "radeondrmfb"?

Also, can you replace these 2 files (gralloc.c and gralloc_drm_kms.c) under
hardware/drm_gralloc and create a new ISO?
It will print some info about drm_init and kms.

Another question: When do you get dmesg output? After GUI is showed or
before? It seems to be incomplete.

Regards,
pstglia

[Jarvis Chung]

Hi Pstglia,

Thanks for your help.

Sorry, I forgot to check my dmesg log.
I quickly to test and share my log as attach files.

**Antutu test:*
HD6670 is pass.
R7 240 is failed.

*com.aim.racingmgt(3D game) test:
HD6670 is pass.
* R7 240 is pass.*

The both card are use "radeondrmfb".
# cat /proc/fb
0 radeondrmfb


Thanks.
Jarvis

pstgliaæ–Œ 2015幎5月14日星期四 UTC+8䞋午10時36分49秒寫道

Post by pstglia
Hi Jarvis,

Post by Jarvis Chung
Thanks for your help.

Thank you and Mauro too for helping with debugging. I'll have to bother
you a little more :)
drm attributes are set. Now I'd like to know why KMS is not being
initialized for these cards.
Can you confirm if /proc/fb has "radeondrmfb"?
Also, can you replace these 2 files (gralloc.c and gralloc_drm_kms.c)
under hardware/drm_gralloc and create a new ISO?
It will print some info about drm_init and kms.
Another question: When do you get dmesg output? After GUI is showed or
before? It seems to be incomplete.
Regards,
pstglia

[pstglia]

Jarvis / Mauro,

**Antutu test:*

Post by Jarvis Chung
HD6670 is pass.
R7 240 is failed.
HD6670 is pass.
* R7 240 is pass.*

My debugs didn't return much help. But dmesg did. I bet this is the
responsible for this mess:

<3>[ 45.322662] radeon 0000:01:00.0: bo edb6fc00 va 0x0000000857 conflict
with (bo eb44a400 0x0000000868 0x0000000869)

This conflict seems to corrupt/crash graphic environment

By the end of April, Christian König submited some patches addressing
issues I believe are related to this. Check the link bellow:

http://www.spinics.net/lists/dri-devel/msg81506.html

We could apply these patches on our kernel and see if this solves our
problem. What do you think? If this not solves the problem we can use the
tips Mauro suggested for further investigation (drm.debug set to level 7,
using apitrace, etc)

Regards,
Pstglia

[Jarvis Chung]

Hi Pstglia,

Thanks for your suggestion.

I try to patch and fixed conflict issue, but I got the same error for
Antutu.

Maybe we could to try patch new kernel for drm:
http://permalink.gmane.org/gmane.comp.video.dri.devel/128426

Thanks.
Jarvis

pstgliaæ–Œ 2015幎5月15日星期五 UTC+8䞊午11時16分21秒寫道

Post by pstglia
Jarvis / Mauro,
**Antutu test:*

Post by Jarvis Chung
HD6670 is pass.
R7 240 is failed.
HD6670 is pass.
* R7 240 is pass.*

My debugs didn't return much help. But dmesg did. I bet this is the
<3>[ 45.322662] radeon 0000:01:00.0: bo edb6fc00 va 0x0000000857
conflict with (bo eb44a400 0x0000000868 0x0000000869)
This conflict seems to corrupt/crash graphic environment
By the end of April, Christian König submited some patches addressing
http://www.spinics.net/lists/dri-devel/msg81506.html
We could apply these patches on our kernel and see if this solves our
problem. What do you think? If this not solves the problem we can use the
tips Mauro suggested for further investigation (drm.debug set to level 7,
using apitrace, etc)
Regards,
Pstglia

[pstglia]

Post by Jarvis Chung
I try to patch and fixed conflict issue, but I got the same error for
Antutu.

Have you applied his 4 patches?

Post by Jarvis Chung
http://permalink.gmane.org/gmane.comp.video.dri.devel/128426

It's already commited on upstream (at least Christian König patches). We
can test it of course. You'll have to apply "
02d57d91aa55bc6e47f29c0622f2f77de0d86b3d" commit "
HACK: drm: disable GPU authentication" and use current config
( make -C kernel O=$OUT/obj/kernel ARCH=x86 oldconfig should be enough for
it).
Another option would be merge/rebase our kernel and upstream

Apart of this, try to collect some info with procedures Mauro suggested

Regards,
Pstglia

[pstglia]

Post by pstglia
It's already commited on upstream (at least Christian König patches). We
can test it of course. You'll have to apply "
02d57d91aa55bc6e47f29c0622f2f77de0d86b3d" commit "
HACK: drm: disable GPU authentication" and use current config
( make -C kernel O=$OUT/obj/kernel ARCH=x86 oldconfig should be enough
for it).
Another option would be merge/rebase our kernel and upstream

I've made a git rebase with Android-x86 4.0 kernel and 4.0.3 upstream. In
case you want it:

https://github.com/pstglia/linux/tree/v4.0.3_android

[Jarvis Chung]

Hi Pstglia,

Thanks for your help.

I try to rebuild this kernel, I also got the same issue.

About this issue, maybe we need to use Mauro suggestion to collecet more
log.


pstgliaæ–Œ 2015幎5月16日星期六 UTC+8䞊午3時35分15秒寫道

Post by pstglia
It's already commited on upstream (at least Christian König patches). We

Post by pstglia
can test it of course. You'll have to apply "
02d57d91aa55bc6e47f29c0622f2f77de0d86b3d" commit "
HACK: drm: disable GPU authentication" and use current config
( make -C kernel O=$OUT/obj/kernel ARCH=x86 oldconfig should be enough
for it).
Another option would be merge/rebase our kernel and upstream

I've made a git rebase with Android-x86 4.0 kernel and 4.0.3 upstream. In
https://github.com/pstglia/linux/tree/v4.0.3_android

[pstglia]

Hi Jarvis / Mauro / Everyone interested,

About this issue, maybe we need to use Mauro suggestion to collecet more

log.

I could do a little debug session on an Acer Aspire V5 (AMD E1 2100 -
Kanibi APU).

Based on my tests, I suppose the problem is really related to this message:

<3>[ 244.849191] radeon 0000:00:01.0: bo ccf78800 va 0x0000000858 conflict
with (bo cec3a000 0x0000000869 0x000000086a)

"ALOGE" debug msgs showed the error occurs
at external/mesa/src/gallium/drivers/radeon/r600_texture.c,
function r600_texture_from_handle (which is called by
"screen->resource_from_handle" on
external/mesa/src/gallium/state_trackers/egl/android/native_android.cpp).
Calling "buf = rscreen->ws->buffer_from_handle" is returning NULL (probably
a consequence of kernel message above):

05-19 18:21:45.486 1943 3452 I MESA-DEBUG: Inside r600_texture_from_handle
05-19 18:21:45.486 1943 3452 I MESA-DEBUG: templ->target = 2 ;
templ->depth0 = 1 ; templ->last_level = 0
05-19 18:21:45.486 1943 3452 E MESA-DEBUG:
rscreen->ws->buffer_from_handle returned NULL
05-19 18:21:45.486 1943 3452 E EGL-GALLIUM: failed to import buffer
0xa0c56f88

I tried to apply all latest available radeon kernel patches:
drm/radeon: fix lockup when BOs aren't part of the VM on release
drm/radeon: reset BOs address after clearing it.
drm/radeon: check new address before removing old one
drm/radeon: fix userptr return value checking (v2)

but no luck with them

Attached dmesg (kernel params: DEBUG=2; drm.debug=7 ; log_buf_len=262144)
and logcat with my debugs. Could not test with egltrace Mauro proposed (had
to return the machine to the owner :D)

Regards,
Pstglia

[pstglia]

I filled this bug on Bugzilla (Mesa/drm) in order to get more help:

https://bugs.freedesktop.org/show_bug.cgi?id=90537

[pstglia]

Hi everybody

Post by pstglia
https://bugs.freedesktop.org/show_bug.cgi?id=90537

Michel DÀnzer (Mesa developer) answered the bug I filled on freedesktop and
proposed a patch. I tested with this and it worked for the hw I tested (AMD
E1 2100 - KANIBI)

Anyone with radeon GCN hardwares (Rx GPU families, Radeon 78xx/79xx GPU
series, AMD Radeon E1, R2, and others) can test the iso bellow with Antutu
and/or any other apps that were producing messages like these ones (note:
uppercase letters are addresses, so the change every time)

dmesg: radeon 0000:00:01.0: bo XXXXX va YYYYYY conflict with (bo ZZZZZZZ
SSSSSSS DDDDDDDD)
logcat: EGL-GALLIUM: failed to import buffer 0xAAAAAAA

# ISO Link
https://drive.google.com/file/d/0Bx12U5yGNcQCRWdoWWtNdC0weW8/view?usp=sharing

Note: As pointed by MIchel, this patch requires kernel 4.0.3. I have a
rebased one (our kernel X mainline 4.0.3) on github:
https://github.com/pstglia/linux/tree/v4.0.3_android

Regards,
pstglia

[André Vitor]

Post by pstglia
Hi everybody
https://bugs.freedesktop.org/show_bug.cgi?id=90537
Michel DÀnzer (Mesa developer) answered the bug I filled on freedesktop and proposed a patch. I tested with this and it worked for the hw I tested (AMD E1 2100 - KANIBI)
Anyone with radeon GCN hardwares (Rx GPU families, Radeon 78xx/79xx GPU series, AMD Radeon E1, R2, and others) can test the iso bellow with Antutu and/or any other apps that were producing messages like these ones (note: uppercase letters are addresses, so the change every time)
dmesg: radeon 0000:00:01.0: bo XXXXX va YYYYYY conflict with (bo ZZZZZZZ SSSSSSS DDDDDDDD)
logcat: EGL-GALLIUM: failed to import buffer 0xAAAAAAA
# ISO Link
https://drive.google.com/file/d/0Bx12U5yGNcQCRWdoWWtNdC0weW8/view?usp=sharing
https://github.com/pstglia/linux/tree/v4.0.3_android
Regards,
pstglia

Hey pstglia,

Tested out here with my Radeon HD 7750 and everything works ok. 0 Errors.

[pstglia]

Hi André,

Hey pstglia,

Post by André Vitor
Tested out here with my Radeon HD 7750 and everything works ok. 0 Errors.

Thanks for the feedback. Could you run a complete Antutu test? If
possible, can you confirm if you had the same bug others were having with
this previous ISO from 2015-05-10?
https://drive.google.com/file/d/0Bx12U5yGNcQCS2k0eDdlQ0lIVWs/view?usp=sharing


Thanks!

[André Vitor]

Yes, I had the same problems before, I couldn't run Antutu. Now I can run a
complete test.

Post by pstglia
Hi André,
Hey pstglia,

Post by André Vitor
Tested out here with my Radeon HD 7750 and everything works ok. 0 Errors.

Thanks for the feedback. Could you run a complete Antutu test? If
possible, can you confirm if you had the same bug others were having with
this previous ISO from 2015-05-10?
https://drive.google.com/file/d/0Bx12U5yGNcQCS2k0eDdlQ0lIVWs/view?usp=sharing
Thanks!

[pstglia]

Post by André Vitor
Yes, I had the same problems before, I couldn't run Antutu. Now I can run
a complete test.

Thanks again André.

Anyone with different GCN/radeonsi cards could give a try on this testing
ISO please? We need to confirm with the proposed patch (bug 90537) is
working ok.

Thanks,
Pstglia

[Jarvis Chung]

Hi Pstglia,

I success to run Antutu.
Thanks your help.

Jarvis

pstgliaæ–Œ 2015幎5月23日星期六 UTC+8䞊午10時57分09秒寫道

Post by pstglia

Post by André Vitor
Yes, I had the same problems before, I couldn't run Antutu. Now I can run
a complete test.

Thanks again André.
Anyone with different GCN/radeonsi cards could give a try on this testing
ISO please? We need to confirm with the proposed patch (bug 90537) is
working ok.
Thanks,
Pstglia

[Jarvis Chung]

Test graphic card: *ASUS Radeon R7 240.*


Jarvis Chungæ–Œ 2015幎5月23日星期六 UTC+8䞊午11時45分33秒寫道

Post by Weikun
Hi Pstglia,
I success to run Antutu.
Thanks your help.
Jarvis
pstgliaæ–Œ 2015幎5月23日星期六 UTC+8䞊午10時57分09秒寫道

Post by pstglia

Post by André Vitor
Yes, I had the same problems before, I couldn't run Antutu. Now I can
run a complete test.

Thanks again André.
Anyone with different GCN/radeonsi cards could give a try on this testing
ISO please? We need to confirm with the proposed patch (bug 90537) is
working ok.
Thanks,
Pstglia

[Mauro Rossi]

Hi,

reporting test results on HD7950.

I could launch Antutu this time but I could not complete the 3D rendering
session.
RADEON_VA_UNMAP patch is producing a regression in the stability of the
GUI.

Strange thing that the green leaves default wallpaper is not showing up.

There are still cases where bo conflict appears in dmesg and I noticed
crashes that produces a deadlock, the GUI restarts sometimes OK, but at
some point it stays perpetually on the android logo.
In between I've seen very strange things happen, simple OpenGL apps working
fine but producing restarts when pressing home button or closing them

At some point I see what we saw in nouveau lockup scenarios, but the system
is not completely stucked and it was possible to collect logs.

On the other hand comparing mesa 10.5.4 and mesa 10.6rc1 (both patched with
RADEON_VA_UNMAP and with kernel 4.0.3) I see the same behavior on HD7950,
Attaching also full logs for 10.5.4 and 10.6rc1 test sessions.

M.

[pstglia]

Post by Mauro Rossi
Hi,
reporting test results on HD7950.

Thanks for the report Mauro. I've updated bug 90537.

So far we have 3 working devices with this patch (no reported bugs so far):
- E1 2100 (KANIBI)
- Radeon HD 7750 (BONAIRE)
- R7 240 (OALAND)

And problems with the one you tested:
- Radeon HD 7950 (TAHITI)

Hello UCyborg / Blumi Milch,
Can you test the latest iso I provided (2015-05-21 post) with your AMD
Radeon R2 / Radeon HD 7850 if possible? Check Antutu and other stuff (some
OpenGL Games, Video playback, etc)

Thank you very much!
Pstglia

[André Vitor]

Hi pstglia,

I just tried this ISO again and now I'm getting this after android's logo:

radeon 0000:01:00.0: bo e0b8a000 va 0x0000000857 conflict with (bo e094a000

Post by pstglia
0x0000000868 0x0000000869)

Sometimes the green background appears, other times do not.
Nothing more appears after this.

Tested on Radeon HD 7750.

dmesg and logcat attached.

Post by pstglia

Post by André Vitor
Yes, I had the same problems before, I couldn't run Antutu. Now I can run
a complete test.

Thanks again André.
Anyone with different GCN/radeonsi cards could give a try on this testing
ISO please? We need to confirm with the proposed patch (bug 90537) is
working ok.
Thanks,
Pstglia

[André Vitor]

Uhh, that's weird. Now it works.

dmesg and logcat with drm.debug=7 attached.

Post by Mauro Rossi
Hi pstglia,
radeon 0000:01:00.0: bo e0b8a000 va 0x0000000857 conflict with (bo

e094a000 0x0000000868 0x0000000869)

Sometimes the green background appears, other times do not.
Nothing more appears after this.
Tested on Radeon HD 7750.
dmesg and logcat attached.

Post by André Vitor
Yes, I had the same problems before, I couldn't run Antutu. Now I can
run a complete test.

Thanks again André.
Anyone with different GCN/radeonsi cards could give a try on this testing
ISO please? We need to confirm with the proposed patch (bug 90537) is
working ok.
Thanks,
Pstglia

[André Vitor]

It's booting now but I also can't finish antutu test anymore. It crashes in
the end of 3D test. I also noticed some hangs during the 3D test, it drop
frames to ~21 from 60 sometimes.

dmesg file drm.debug=7

https://drive.google.com/file/d/0B78EUF6fAoIFYUxhSEhGY0lnNFU/view?usp=sharing

I got it using "cat /proc/kmsg" right after android's logo until antutu 3D
test end. So it's a very very very long log.

Post by André Vitor
Uhh, that's weird. Now it works.
dmesg and logcat with drm.debug=7 attached.

Post by Mauro Rossi
Hi pstglia,
radeon 0000:01:00.0: bo e0b8a000 va 0x0000000857 conflict with (bo

e094a000 0x0000000868 0x0000000869)

Sometimes the green background appears, other times do not.
Nothing more appears after this.
Tested on Radeon HD 7750.
dmesg and logcat attached.

Post by André Vitor
Yes, I had the same problems before, I couldn't run Antutu. Now I can
run a complete test.

Thanks again André.
Anyone with different GCN/radeonsi cards could give a try on this
testing ISO please? We need to confirm with the proposed patch (bug 90537)
is working ok.
Thanks,
Pstglia

[pstglia]

Post by André Vitor
It's booting now but I also can't finish antutu test anymore. It crashes
in the end of 3D test. I also noticed some hangs during the 3D test, it
drop frames to ~21 from 60 sometimes.
dmesg file drm.debug=7
https://drive.google.com/file/d/0B78EUF6fAoIFYUxhSEhGY0lnNFU/view?usp=sharing
I got it using "cat /proc/kmsg" right after android's logo until antutu 3D
test end. So it's a very very very long log.

According to radeon devs, these conflicts are a known issue related to the
kernel part. They still don't have a definitive solution for this (see
https://bugs.freedesktop.org/show_bug.cgi?id=90537)

Mauro and me are trying to provide info on this bug in order to help them
solving this issue.

Let's keep testing new kernel releases when they mentioned some fix about
lockup

Regards,
pstglia

[Jarvis Chung]

Hi Pstglia,

# ISO Link

Post by pstglia
https://drive.google.com/file/d/0Bx12U5yGNcQCRWdoWWtNdC0weW8/view?usp=sharing

I try to rebuild a new ISO use kernel 4.0.3, but I couldn't run Antutu(the
same issue).
The following is my build info:
Android-x86 5.x + mesa 10.5.4(apply radeonsi patch) + kernel 4.0.3

Your ISO file is pass to run Antutu. Could you provide your build info to
me?

Thanks.
Jarvis

[pstglia]

Hi Jarvis,

I try to rebuild a new ISO use kernel 4.0.3, but I couldn't run Antutu(the

Post by Jarvis Chung
same issue).
Android-x86 5.x + mesa 10.5.4(apply radeonsi patch) + kernel 4.0.3
Your ISO file is pass to run Antutu. Could you provide your build info to
me?

You have to apply the patch provided with bug 90537:
https://bugs.freedesktop.org/attachment.cgi?id=115936

cd external/mesa
patch -p1 < PATH_TO_PATCH_FILE

Regards,
Pstglia

[Jarvis Chung]

Hi Pstglia,

Thanks for your help, I could run Antutu for my build.

About Antutu testing, I try to run full testing and shown success screen,
then services will died.
I also test your ISO and got the same issue.

The attached files is my dump log.

Thanks.
Jarvis

pstgliaæ–Œ 2015幎5月26日星期二 UTC+8䞊午12時35分12秒寫道

Post by pstglia
Hi Jarvis,
I try to rebuild a new ISO use kernel 4.0.3, but I couldn't run Antutu(the

Post by Jarvis Chung
same issue).
Android-x86 5.x + mesa 10.5.4(apply radeonsi patch) + kernel 4.0.3
Your ISO file is pass to run Antutu. Could you provide your build info to
me?

https://bugs.freedesktop.org/attachment.cgi?id=115936
cd external/mesa
patch -p1 < PATH_TO_PATCH_FILE
Regards,
Pstglia

[Joan R. Serra]

I run live image on my Athlon 5350 with Radeon R3. On about tablet->
OpenGL: Null,null,null

Thanks for your work !!!

Post by pstglia
Hi everybody

Post by pstglia
https://bugs.freedesktop.org/show_bug.cgi?id=90537

Michel DÀnzer (Mesa developer) answered the bug I filled on freedesktop
and proposed a patch. I tested with this and it worked for the hw I tested
(AMD E1 2100 - KANIBI)
Anyone with radeon GCN hardwares (Rx GPU families, Radeon 78xx/79xx GPU
series, AMD Radeon E1, R2, and others) can test the iso bellow with Antutu
uppercase letters are addresses, so the change every time)
dmesg: radeon 0000:00:01.0: bo XXXXX va YYYYYY conflict with (bo ZZZZZZZ
SSSSSSS DDDDDDDD)
logcat: EGL-GALLIUM: failed to import buffer 0xAAAAAAA
# ISO Link
https://drive.google.com/file/d/0Bx12U5yGNcQCRWdoWWtNdC0weW8/view?usp=sharing
Note: As pointed by MIchel, this patch requires kernel 4.0.3. I have a
https://github.com/pstglia/linux/tree/v4.0.3_android
Regards,
pstglia

[pstglia]

Post by Joan R. Serra
I run live image on my Athlon 5350 with Radeon R3. On about tablet->
OpenGL: Null,null,null

Thanks for testing it. Could you run Antutu complete test, watch video, etc?

Currently, OpenGL info is not being shown. However, you can get this info
from logcat

[Joan R. Serra]

Sorry but I try to install to harddisk and after I don't see system and
data folders on partition !?

Post by Joan R. Serra
I run live image on my Athlon 5350 with Radeon R3. On about tablet->

Post by Joan R. Serra
OpenGL: Null,null,null

Thanks for testing it. Could you run Antutu complete test, watch video, etc?
Currently, OpenGL info is not being shown. However, you can get this info
from logcat

[pstglia]

Post by Mauro Rossi
Hi pstglia,
I didn't follow the long thread carefully.
Except the vanilla patches of 4.0.3,
could you specify what additional patches you applied?

The only missing on git is LLVM patch (posted on May, 1st):

0001-Enable-R600-support-on-Android-x86-mesa-gallium-radeon_lollipop_20150501.patch

Regards,
Pstglia

[pstglia]

Post by pstglia
https://github.com/pstglia/linux/tree/v4.0.3_android

Hi, this repo is just a rebase result between our kernel (4.0.0) and and
mainline (4.0.3)

I did this because a developer from freedesktop (Michel DÀnzer) proposed a
Mesa patch for a Bug I filled on freedesktop (90537 - radeonsi bo/va
conflict on RADEON_GEM_VA (rscreen->ws->buffer_from_handle returns NULL).
He stated that this patch required kernel 4.0.3 due some fixes synced to
upstream.


Regards,
Pstglia

[pstglia]

FIY,

The bug I filled on freedesktop (90537) was marked as fixed

The changes were commited to kernel 4.1 (see link bellow). Devs stated that backporting this fix requires backporting other patches (they enumerated these commits on the link)

https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-4.1.y&id=3bc980bf19bb62007e923691fa2869ba113be895

[Mauro Rossi]

Hi,
I've come to know that 90537 bug was fixed by seeing this commit in mesa
master:

http://cgit.freedesktop.org/mesa/mesa/commit/?id=7796e8889a9a2cc1b454dc32d8da3d756404339a

this correction in mesa source could be necessary to avoid the radeonsi bo
conflict.
M.

[pstglia]

Hi Chih-Wei,

OK, so no additional kernel patches? I see.

As you probably noticed, after this message, Mesa devs have pushed fixes to
upstream (Kernel and Mesa) in order to fix the issue we (and others)
reported.

They are using the initial proposed patch + a new drm query
(RADEON_INFO_VA_UNMAP_WORKING)

Kernel patch was pushed to stable Kernel 4.1. As I could check, 4.0.6 match
all requirements they pointed, so I'm trying to testing it with our version

Mesa patch was synced to Mesa dev git (10.7). It can be backported to
previous releases (requires the Kernel patches for proper work)

BTW, I just merged 4.0.6 into into the android-4.0 branch.

Thanks. Just a question. Are these patches supposed to be on git?

drivers/net/wireless/wl/linux-recent.patch
drivers/net/wireless/wl/wl.patch

They are causing some hunks during kernel compiling, which aborts the build

Regards,
Pstglia

[rbg]

FWIW:

had no aborted kernel builds with the wl patches with 4.06 kernel...

built KitKat & Lollipop last week from git and my WiFi tested fine with the
b43/wl intergrated

Post by pstglia
Hi Chih-Wei,

OK, so no additional kernel patches? I see.

As you probably noticed, after this message, Mesa devs have pushed fixes
to upstream (Kernel and Mesa) in order to fix the issue we (and others)
reported.
They are using the initial proposed patch + a new drm query
(RADEON_INFO_VA_UNMAP_WORKING)
Kernel patch was pushed to stable Kernel 4.1. As I could check, 4.0.6
match all requirements they pointed, so I'm trying to testing it with our
version
Mesa patch was synced to Mesa dev git (10.7). It can be backported to
previous releases (requires the Kernel patches for proper work)

BTW, I just merged 4.0.6 into into the android-4.0 branch.

Thanks. Just a question. Are these patches supposed to be on git?
drivers/net/wireless/wl/linux-recent.patch
drivers/net/wireless/wl/wl.patch
They are causing some hunks during kernel compiling, which aborts the build
Regards,
Pstglia

[Mauro Rossi]

Post by pstglia
drivers/net/wireless/wl/linux-recent.patch
drivers/net/wireless/wl/wl.patch
They are causing some hunks during kernel compiling, which aborts the build

As a temporary workaround, to be able to build kernel 4.0.6 from repo and
my 4.1 rebase, I had to comment the following line in kernel config file

# CONFIG_WL is not set


because I had some Hunk# step of patch failing for wl, may be due to
preexisting scripts in the source folder



Wrt to my kernel 4.1 rebase + mesa 10.6, GUI works on Intel, Nouveau -
checked both with kitkat-x86 and lollipop-x86 builds.

Here is the git:


https://github.com/maurossi/linux/commits/android-4.1_rebase


NOTE: for kitkat-x86 it may be necessary to patch logger.c related code:
https://github.com/maurossi/linux/commit/93fdd143e172aa1f1929ae519f5a4368f3841732


git cherry-pick 93fdd143e172aa1f1929ae519f5a4368f3841732



I tested Radeonsi with 10.7.0devel rebase, because I get no GUI with mesa
10.6 rebase.


https://github.com/maurossi/mesa/tree/10.7.0devel_lollipop_28-jun-2015



Test results on HD7950: OpenGL apps/test are working as before, but the
following apps are still causing restart of the Android GUI:

- OpenGLES 1.0 demo click at first rotating cube
- Rajawali can't click back after any demo
- running Antutu still causes GUI restart


I still have doubts about my rebase skills/procedures, so I would
appreciate if I could get an independent feedback about the aforementioned
apps with 4.0.6 kernel and mesa 10.6


Mauro

[rbg]

checking my notes shows that I did see that issue and checked kitkat
changes, lollipop still being under
development sometimes changes do not get merged to lollipop, and lollipop
default is NOT kernel 4.xx.....

Anyway I merged this change into my lollipop build using 4.x.x kernel and
it corrected the issue....

merge to Lollipop(if using kernel 4.x.x):


http://git.android-x86.org/?p=platform/build.git;a=commit;h=a3fc7d384332bcda0da49192dbea93f5c6ef23fc

Post by Mauro Rossi

Post by pstglia
drivers/net/wireless/wl/linux-recent.patch
drivers/net/wireless/wl/wl.patch
They are causing some hunks during kernel compiling, which aborts the

build
As a temporary workaround, to be able to build kernel 4.0.6 from repo and
my 4.1 rebase, I had to comment the following line in kernel config file
# CONFIG_WL is not set
because I had some Hunk# step of patch failing for wl, may be due to
preexisting scripts in the source folder
Wrt to my kernel 4.1 rebase + mesa 10.6, GUI works on Intel, Nouveau -
checked both with kitkat-x86 and lollipop-x86 builds.
https://github.com/maurossi/linux/commits/android-4.1_rebase
https://github.com/maurossi/linux/commit/93fdd143e172aa1f1929ae519f5a4368f3841732
git cherry-pick 93fdd143e172aa1f1929ae519f5a4368f3841732
I tested Radeonsi with 10.7.0devel rebase, because I get no GUI with mesa
10.6 rebase.
https://github.com/maurossi/mesa/tree/10.7.0devel_lollipop_28-jun-2015
Test results on HD7950: OpenGL apps/test are working as before, but the
- OpenGLES 1.0 demo click at first rotating cube
- Rajawali can't click back after any demo
- running Antutu still causes GUI restart
I still have doubts about my rebase skills/procedures, so I would
appreciate if I could get an independent feedback about the aforementioned
apps with 4.0.6 kernel and mesa 10.6
Mauro

[rbg]

following also needs to be merged to Lollipop, IMO

http://git.android-x86.org/?p=platform/manifest.git;a=commit;h=3e17feb28403a186d91fd1060ac0102abd993b9c

Post by rbg
checking my notes shows that I did see that issue and checked kitkat
changes, lollipop still being under
development sometimes changes do not get merged to lollipop, and lollipop
default is NOT kernel 4.xx.....
Anyway I merged this change into my lollipop build using 4.x.x kernel and
it corrected the issue....
http://git.android-x86.org/?p=platform/build.git;a=commit;h=a3fc7d384332bcda0da49192dbea93f5c6ef23fc

Post by Mauro Rossi

Post by pstglia
drivers/net/wireless/wl/linux-recent.patch
drivers/net/wireless/wl/wl.patch
They are causing some hunks during kernel compiling, which aborts the

build
As a temporary workaround, to be able to build kernel 4.0.6 from repo and
my 4.1 rebase, I had to comment the following line in kernel config file
# CONFIG_WL is not set
because I had some Hunk# step of patch failing for wl, may be due to
preexisting scripts in the source folder
Wrt to my kernel 4.1 rebase + mesa 10.6, GUI works on Intel, Nouveau -
checked both with kitkat-x86 and lollipop-x86 builds.
https://github.com/maurossi/linux/commits/android-4.1_rebase
https://github.com/maurossi/linux/commit/93fdd143e172aa1f1929ae519f5a4368f3841732
git cherry-pick 93fdd143e172aa1f1929ae519f5a4368f3841732
I tested Radeonsi with 10.7.0devel rebase, because I get no GUI with mesa
10.6 rebase.
https://github.com/maurossi/mesa/tree/10.7.0devel_lollipop_28-jun-2015
Test results on HD7950: OpenGL apps/test are working as before, but the
- OpenGLES 1.0 demo click at first rotating cube
- Rajawali can't click back after any demo
- running Antutu still causes GUI restart
I still have doubts about my rebase skills/procedures, so I would
appreciate if I could get an independent feedback about the aforementioned
apps with 4.0.6 kernel and mesa 10.6
Mauro

[pstglia]

Post by rbg
Anyway I merged this change into my lollipop build using 4.x.x kernel and
it corrected the issue....
http://git.android-x86.org/?p=platform/build.git;a=commit;h=a3fc7d384332bcda0da49192dbea93f5c6ef23fc

I reverted this commit as my goal was building a radeonsi enabled image
with patches for bug 90537.

In case someone wants to test it, here's the image link:
https://drive.google.com/file/d/0Bx12U5yGNcQCbE5NRUJpY1lyRFk/view?usp=sharing

It was build with:
- Kernel 4.0.6 + Kernel patch for fixing bug 90537 (commit
3bc980bf19bb62007e923691fa2869ba113be895);
(important: commits 26d4d129b6042197b4cbc8341c0618f99231af2f, 48afbd70ac7b6aa62e8d452091023941d8085f8a
and c29c0876ec05d51a93508a39b90b92c29ba6423d are pre-requisites for this;
4.0.6 already includes those)
- Mesa 10.5.8 + Mesa patch 7796e8889a9a2cc1b454dc32d8da3d756404339a
<http://cgit.freedesktop.org/mesa/mesa/commit/?id=7796e8889a9a2cc1b454dc32d8da3d756404339a> for
bug 90537
- radeonsi is enabled (GCN cards - Radeon 78xx/79xx, R7 and R9 series)

Additional Notes:

1) Built 64 bits kernel (32 bits environment). I did this because I
included "fugu houdini" support using the info Edgardo Gho provided (see
https://groups.google.com/forum/#!starred/android-x86/7O2qLZjzGbk).
I was able to run some arm apps (Candy crush saga, Candy crush Soda and
"Mussoumano")

Patch created to enable houdini as Edgardo described:
Patch file: 0001-build_Add-support-for-houdini-using-fugu-libraries.patch
Apply on dir: build

2) On radeon cards, I was receiving a error when querying a device id
through drm ioctl call. This was just happening when using 64 bits kernel.
I checked the problem was a cast with "long" type. It was changed to the
same cast type used in xorg-video-ati (uintptr_t)
Patch
file: 0001-hardware_drm_gralloc_Use-generic-type-for-radeon-drm-calls.patch
Apply on dir: hardware/drm_gralloc

Regards
Pstglia

[Mauro Rossi]

Hi,

here are dmesg, logcat, dumpsys for HD7750 with latest kernel 4.0.6 + mesa
10.5.8 ISO

There is a segfault on libdrm_radeon.so

surfaceflinger[15579]: segfault at 0 ip 00000000f705ce9a sp
00000000ffe72330 error 4 in libdrm_radeon.so[f705a000+10000]

Anyway good progress and soon radeonsi will be working
M.

Post by rbg
Anyway I merged this change into my lollipop build using 4.x.x kernel and

Post by rbg
it corrected the issue....
http://git.android-x86.org/?p=platform/build.git;a=commit;h=a3fc7d384332bcda0da49192dbea93f5c6ef23fc

I reverted this commit as my goal was building a radeonsi enabled image
with patches for bug 90537.
https://drive.google.com/file/d/0Bx12U5yGNcQCbE5NRUJpY1lyRFk/view?usp=sharing
- Kernel 4.0.6 + Kernel patch for fixing bug 90537 (commit
3bc980bf19bb62007e923691fa2869ba113be895);
(important: commits 26d4d129b6042197b4cbc8341c0618f99231af2f, 48afbd70ac7b6aa62e8d452091023941d8085f8a
and c29c0876ec05d51a93508a39b90b92c29ba6423d are pre-requisites for this;
4.0.6 already includes those)
- Mesa 10.5.8 + Mesa patch 7796e8889a9a2cc1b454dc32d8da3d756404339a
<http://cgit.freedesktop.org/mesa/mesa/commit/?id=7796e8889a9a2cc1b454dc32d8da3d756404339a> for
bug 90537
- radeonsi is enabled (GCN cards - Radeon 78xx/79xx, R7 and R9 series)
1) Built 64 bits kernel (32 bits environment). I did this because I
included "fugu houdini" support using the info Edgardo Gho provided (see
https://groups.google.com/forum/#!starred/android-x86/7O2qLZjzGbk).
I was able to run some arm apps (Candy crush saga, Candy crush Soda and
"Mussoumano")
Patch file: 0001-build_Add-support-for-houdini-using-fugu-libraries.patch
Apply on dir: build
2) On radeon cards, I was receiving a error when querying a device id
through drm ioctl call. This was just happening when using 64 bits kernel.
I checked the problem was a cast with "long" type. It was changed to the
same cast type used in xorg-video-ati (uintptr_t)
Patch
file: 0001-hardware_drm_gralloc_Use-generic-type-for-radeon-drm-calls.patch
Apply on dir: hardware/drm_gralloc
Regards
Pstglia

[rbg]

Hi Pstglia....

have you tried this also on 64bit kernel, 32bit user ......

curious as that's what most of my builds have been and I do remember
providing a build you checked
and had issues with....then we concluded that the stuff was only working
with 32bit kernel, 32bit user.....

It would be interesting to know as Lollipop seems to be heading in the
64bit kernel direction....

Post by pstglia
Hi Jarvis / Mauro / Everyone interested,
About this issue, maybe we need to use Mauro suggestion to collecet more

log.

I could do a little debug session on an Acer Aspire V5 (AMD E1 2100 -
Kanibi APU).
<3>[ 244.849191] radeon 0000:00:01.0: bo ccf78800 va 0x0000000858
conflict with (bo cec3a000 0x0000000869 0x000000086a)
"ALOGE" debug msgs showed the error occurs
at external/mesa/src/gallium/drivers/radeon/r600_texture.c,
function r600_texture_from_handle (which is called by
"screen->resource_from_handle" on
external/mesa/src/gallium/state_trackers/egl/android/native_android.cpp).
Calling "buf = rscreen->ws->buffer_from_handle" is returning NULL (probably
05-19 18:21:45.486 1943 3452 I MESA-DEBUG: Inside
r600_texture_from_handle
05-19 18:21:45.486 1943 3452 I MESA-DEBUG: templ->target = 2 ;
templ->depth0 = 1 ; templ->last_level = 0
rscreen->ws->buffer_from_handle returned NULL
05-19 18:21:45.486 1943 3452 E EGL-GALLIUM: failed to import buffer
0xa0c56f88
drm/radeon: fix lockup when BOs aren't part of the VM on release
drm/radeon: reset BOs address after clearing it.
drm/radeon: check new address before removing old one
drm/radeon: fix userptr return value checking (v2)
but no luck with them
Attached dmesg (kernel params: DEBUG=2; drm.debug=7 ; log_buf_len=262144)
and logcat with my debugs. Could not test with egltrace Mauro proposed (had
to return the machine to the owner :D)
Regards,
Pstglia

[pstglia]

Hi Rbg

Post by rbg
have you tried this also on 64bit kernel, 32bit user ......

Not yet. Just 32/32. I'll try it as soon as possible

It would be interesting to know as Lollipop seems to be heading in the

Post by rbg
64bit kernel direction....

I'll try first on my hardware (which is working ok with 32 bits kernel)

pstglia

[UCyborg]

I've tried the ISO with Mesa 10.5.4 and it's still not working on AMD Radeon R2. Kernel messages say radeon driver initialized successfully but logcat spits out the following:

I/SurfaceFlinger( 2005): SurfaceFlinger is starting
I/SurfaceFlinger( 2005): SurfaceFlinger's main thread ready to run. Initializing graphics H/W...
D/libEGL ( 2005): loaded /system/lib/egl/libGLES_mesa.so
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
W/EGL-GALLIUM( 2005): failed to create DRM screen
W/EGL-GALLIUM( 2005): will fall back to other EGL drivers if any
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
W/EGL-DRI2( 2005): fail to get drm fd
I/EGL-GALLIUM( 2005): using SW screen
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
E/hwcomposer( 2005): framebuffer_open failed (Invalid argument)
W/hwcomposer( 2005): getting VSYNC period from thin air: 16666666
D/hwcomposer( 2005): Intel hwcomposer module
E/SurfaceFlinger( 2005): ERROR: failed to open framebuffer (Invalid argument), aborting

[pstglia]

Post by UCyborg
I've tried the ISO with Mesa 10.5.4 and it's still not working on AMD
Radeon R2. Kernel messages say radeon driver initialized successfully but
I/SurfaceFlinger( 2005): SurfaceFlinger is starting
I/SurfaceFlinger( 2005): SurfaceFlinger's main thread ready to run.
Initializing graphics H/W...
D/libEGL ( 2005): loaded /system/lib/egl/libGLES_mesa.so
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
W/EGL-GALLIUM( 2005): failed to create DRM screen
W/EGL-GALLIUM( 2005): will fall back to other EGL drivers if any
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
W/EGL-DRI2( 2005): fail to get drm fd
I/EGL-GALLIUM( 2005): using SW screen
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
E/hwcomposer( 2005): framebuffer_open failed (Invalid argument)
W/hwcomposer( 2005): getting VSYNC period from thin air: 16666666
D/hwcomposer( 2005): Intel hwcomposer module
E/SurfaceFlinger( 2005): ERROR: failed to open framebuffer (Invalid argument), aborting

Thank you for your feedback.

This message (from drm gralloc) is telling your gpu chip id (0x9853 -
Mullins) was not found in the supported list (see
hardware/drm_gralloc/radeon/radeon.h and
hardware/drm_gralloc/radeon/radeon_chipinfo_gen.h)

I've added this (and others - HAWAII and KAVERY) to drm gralloc. Patch
attached:
# To be applied under hardware/drm_gralloc
0001-Include-newer-missing-chipset-families.patch

When possible, could you test again with this new ISO?
https://drive.google.com/file/d/0Bx12U5yGNcQCS2k0eDdlQ0lIVWs/view?usp=sharing

Thanks,
Pstglia

[UCyborg]

It boots now and the interface is smooth, but video playback is
problematic. On YouTube in default browser, there is just black screen.
Trying to play a local file either in VLC or MXPlayer makes the
RenderThread in libhwui.so segfault. Video plays for few seconds before it
happens, screen temporarily freezes, Android logo shows up for few seconds
and then I'm back at the home screen.

Post by pstglia

Post by UCyborg
I've tried the ISO with Mesa 10.5.4 and it's still not working on AMD
Radeon R2. Kernel messages say radeon driver initialized successfully but
I/SurfaceFlinger( 2005): SurfaceFlinger is starting
I/SurfaceFlinger( 2005): SurfaceFlinger's main thread ready to run.
Initializing graphics H/W...
D/libEGL ( 2005): loaded /system/lib/egl/libGLES_mesa.so
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
W/EGL-GALLIUM( 2005): failed to create DRM screen
W/EGL-GALLIUM( 2005): will fall back to other EGL drivers if any
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
W/EGL-DRI2( 2005): fail to get drm fd
I/EGL-GALLIUM( 2005): using SW screen
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
E/GRALLOC-RADEON( 2005): unknown device ID 0x9853
E/hwcomposer( 2005): framebuffer_open failed (Invalid argument)
W/hwcomposer( 2005): getting VSYNC period from thin air: 16666666
D/hwcomposer( 2005): Intel hwcomposer module
E/SurfaceFlinger( 2005): ERROR: failed to open framebuffer (Invalid argument), aborting

Thank you for your feedback.
This message (from drm gralloc) is telling your gpu chip id (0x9853 -
Mullins) was not found in the supported list (see
hardware/drm_gralloc/radeon/radeon.h and
hardware/drm_gralloc/radeon/radeon_chipinfo_gen.h)
I've added this (and others - HAWAII and KAVERY) to drm gralloc. Patch
# To be applied under hardware/drm_gralloc
0001-Include-newer-missing-chipset-families.patch
When possible, could you test again with this new ISO?
https://drive.google.com/file/d/0Bx12U5yGNcQCS2k0eDdlQ0lIVWs/view?usp=sharing
Thanks,
Pstglia

[pstglia]

Hi UCyborg

It boots now and the interface is smooth, but video playback is

Post by UCyborg
problematic. On YouTube in default browser, there is just black screen.
Trying to play a local file either in VLC or MXPlayer makes the
RenderThread in libhwui.so segfault. Video plays for few seconds before it
happens, screen temporarily freezes, Android logo shows up for few seconds
and then I'm back at the home screen.

I noticed from your logcat is the lack of GRALLOC-KMS outputs and Gallium
strings (renderer, version, etc). If possible, can you boot under debug
mode and collect logcat output again please?

Can you get the output from these cmds also?

getprop | grep "debug.mesa.software"
getprop | grep "ro.hardware.gralloc"
cat /proc/fb


How does this APU performs on a Linux distro? Everything works as expected
(OpenGL accel, video playback, etc)? If not ok, I'd say it's due some
features are still for this arch are under develoment on Mesa (same for
KAVERI and KANIBI). Here's one of them:

# src/gallium/drivers/radeonsi/si_state.c (function si_init_config)
case CHIP_KAVERI:
*/* XXX todo */*
case CHIP_KABINI:
*/* XXX todo */*
*case CHIP_MULLINS:*
* /* XXX todo */*


Regards,
Pstglia

[Luigi Capriotti]

I have been testing this new build on a desktop computer connected via LAN.
Upon starting, no IP is leased to the computer and I have to manually start
dhcpcd to receive IP configuration (switching to console with alt+F1).

Once IP and DNS settings are received, I can ping and traceroute, but DNS
resolution is not working. nslookup fails as well, even if I specify a
proper DNS server:

$nslookup www.google.it 8.8.8.8
Server: 0.0.0.0
Address 1: 0.0.0.0
nslookup: can't resolve 'www.google.it'

This is exactly what is described here
<http://android.stackexchange.com/questions/10657/android-terminal-not-able-to-nslookup>,
but being this issue not reported by other testers I wonder if it's really
the same problem.

[Luigi Capriotti]

I tried android-x86-4.4-RC2 on same configuration and I have the same
issue, it's not this specific build.

Opening a new thread to seek for advices.

[pstglia]

Hi Chih-Wei,

However, I'm considering whether if we should

maintain our own pci_id table in drm_gralloc.
Actually only radeon ids are maintained
in drm_gralloc. All the others (intel, nouveau, ...)
use id from mesa/drm directly.
Probably the radeon should be changed to
the similar way as well so we won't suffer
such a problem in the future.

Yes it would avoid this update process every release. The only thing is
how to deal with flags associated for each chipset (in special mobility and
igp):

typedef struct {
uint32_t pci_device_id;
RADEONChipFamily chip_family;
* int mobility;*
* int igp;*
* int nocrtc2;*
* int nointtvout;*
* int singledac;*
} RADEONCardInfo;

they are populated inside gralloc_drm_radeon.c, but apparently not directly
used.
xf86-video-ati appears just to declare these flags also

[UCyborg]

Hi everyone,

I see you're already on the right track solving this. I've been using
Ubuntu 15.04 for some time on my laptop when it was still in development.
3D acceleration and hardware video decoding worked. The biggest problem
when it came to graphics was that radeon driver would occasionally lock up
and bring down the entire X Server and when it recovered, graphics lagged
badly so a reboot was required. There was a also a glitch in Firefox, when
using VAAPI for video decoding, replaying a video on YouTube would result
in that video playing slowly and the picture was corrupted. Refresh of the
site fixed it. Playing video in VLC using VDPAU worked as expected though.

As for Android, 3D definitely works now, without accelerated 3D graphics,
OS is entirely useless, mouse takes seconds to move around. Gallium strings
are missing in my logcat because I cleared it before attempting video
playback. I also tried playing GL TRON game from F-Droid and it works.

[UCyborg]

I can confirm /proc/fb has "radeondrmfb". Strange thing in my case is that
logcat prints out renderer, version and extension strings, but in Settings
under About, it says "null" for each of those strings. Also, how do you
execute commands when booting in Debug mode? It doesn't respond to anything
I type in terminal.

And one more unrelated question. Is booting in UEFI mode supported? None of
the boot entries work for me, nothing happens so I have to switch to Legacy
BIOS mode to be able to play with it. Only the entry for Windows 8 works
and boots it.

[pstglia]

Post by UCyborg
I can confirm /proc/fb has "radeondrmfb".

Good. Thanks for this info

Post by UCyborg
Strange thing in my case is that logcat prints out renderer, version and
extension strings, but in Settings under About, it says "null" for each of
those strings.

It's a known issue on Lollipop. Maybe an EGL context must be initalized
prior to getting info

Post by UCyborg
Also, how do you execute commands when booting in Debug mode? It doesn't
respond to anything I type in terminal.

When on debugging mode, after typing "exit" twice (when requested) and
entering GUI, you can switch back console with alt+F1 and use the cmds.
Do you mean it's freezed?

Another option is using a terminal emulator. Should work on Lollipop

Post by UCyborg
And one more unrelated question. Is booting in UEFI mode supported? None
of the boot entries work for me, nothing happens so I have to switch to
Legacy BIOS mode to be able to play with it. Only the entry for Windows 8
works and boots it.

You can create images file both for tradicional booting method
(bios/legacy) or UEFI.

These imgs I made available were create with "make iso_img", so just
legacy. Compiling with "make efi_img" would create an efi (non-signed)

[UCyborg]

Post by pstglia
When on debugging mode, after typing "exit" twice (when requested) and
entering GUI, you can switch back console with alt+F1 and use the cmds.
Do you mean it's freezed?
Another option is using a terminal emulator. Should work on Lollipop

The problem was that I got into wrong terminal by pressing alt + left
arrow, just pressing this once accomplished the same thing as alt + F1 in
normal mode. All okay now, logcat works, but eg. "reboot" and "poweroff"
really doesn't do anything.

You can create images file both for tradicional booting method

Post by pstglia
(bios/legacy) or UEFI.
These imgs I made available were create with "make iso_img", so just
legacy. Compiling with "make efi_img" would create an efi (non-signed)

Oh, okay, thanks for this info. Guess I'll just setup Ubuntu to be able to
easily make my own image, preferably with Google Apps. I noticed someone
made a script for that. Never compiled whole Android from source though, so
that should be fun.

[UCyborg]

Post by pstglia
When on debugging mode, after typing "exit" twice (when requested) and
entering GUI, you can switch back console with alt+F1 and use the cmds.
Do you mean it's freezed?
Another option is using a terminal emulator. Should work on Lollipop

The problem was that I got into wrong terminal by pressing alt + left
arrow. Pressing this combo once in normal mode accomplishes the same as alt
+ F1. All okay now, logcat works, but eg. "poweroff" and "reboot" just
return without doing anything.

You can create images file both for tradicional booting method

Post by pstglia
(bios/legacy) or UEFI.
These imgs I made available were create with "make iso_img", so just
legacy. Compiling with "make efi_img" would create an efi (non-signed)

Oh okay, thanks for this info, guess I'll just setup Ubuntu to be able to
make my own image, preferably with Google Apps. I noticed someone made a
script for that. Never compiled whole Android from source though, so that
should be fun.

[pstglia]

Post by UCyborg
The problem was that I got into wrong terminal by pressing alt + left
arrow. Pressing this combo once in normal mode accomplishes the same as alt
+ F1. All okay now, logcat works, but eg. "poweroff" and "reboot" just
return without doing anything.

Some commands require you add "busybox" in front of it (busybox reboot, for
instance)

[Никита Донецкий]

Thank you! Work on Radeon R9 280.
But buggy. Not change language input for example)
Without GApps)