What version of the kernel are you using? Newer versions of the linux
kernel will pass VM_EXEC on mmap which will cause the binder to fail.
You might need to fix the binder so that it is no longer a
FORBIDDEN_MMAP_FLAGS.
lqoyzy wrote:
> Hi Ben,
>
> I'm trying to port android to a s3c2410 develop board..which's only
> 64m ram..
> now I've successfully built out armv4t images with your submitted
> patch and sean's patches under reviewing.
> but I'm getting servicemanager exiting problem while using nfs as
> rootfs.
> I'm doubting it's the binder who cannot be inited.. likely you've met
> the same problem, as in your patch that add some log for this.
> what's the problem it is? about mmap? or is there some option to
> enable the nfs support this? or 64m limits this?
> I've checked the /dev/binder & /dev/ashmen, both are 777.
> looking forward for your replay :-), thanks!
> I've no usb cable now..and don't know how to make adb to use a
> usb2serial cable..so some help on this is also very nice~
>
> On Nov 2, 7:33 pm, "Ben Leslie" <[EMAIL PROTECTED]> wrote:
>
>> I haven't had time to clean up my patches sufficently to submit them
>> for formal review, but I'm going away for two weeks tomorrow, and I
>> don't want people unnecessarily re-inventing the wheel (although Sean
>> and I seem to have gone and invented two wheels in parallel anyway).
>>
>> I've been using two separate trees, one to try things on and get them
>> up and running, and a separate integration tree for submitting things
>> on. These patches come straight from the former, and hence won't apply
>> 100% cleanly, but it should be relatively obvious what the changes
>> are (mostly adding ARCH_ARM_ to the front of HAVE_<x> variables).
>>
>> The diffs are per project and in most cases should be split into
>> multiple separate diffs before committing.
>>
>> The main other thing that patch addresses other than ARMv4 support is
>> using filesystems other than yaffs2, and support for frame-buffer
>> devices that don't support panning.
>>
>> The rest of this just describes the patches in detail. Note, although
>> getting this running on the Neo 1973 was the ultimate driver for
>> this, most of these patches are really aimed to get Android compiling for
>> plain ARMv4 (_not_ ARMv4T). This is _not_ the best approach of running
>> Android on an actual OpenMoko phone, but hey there are still devices
>> with StrongARM chips out there too.
>>
>> Note: I am off the air for the next two weeks, and don't plan on submitting
>> these for review until I get back from holiday. I'm hoping that someone
>> (looks
>> like Sean), is going to be pushing through with ARMv4T support in the
>> meantime,
>> so most of this will probably addressed one way or another by the time I'm
>> back.
>>
>> If any of it is still relevant once I'm back I'll clean it up for
>> submissions then.
>>
>> system/core:
>>
>> libcutils/{atomic-android-arm.S,memset32.S}:
>>
>> * Use BX<cc> macro rather than directly using bx<cc> instructions.
>> (see change 1663 for definition of BX<cc> macros).
>>
>> libpixelflinger/Android.mk:
>>
>> * Only compile t32cbblend.S when halfword multiplies are available.
>> * (Note: this actual check could be cleaned up).
>>
>> libpixelflinger/scanline.cpp:
>>
>> * Disable codegeneration on ARMv4
>>
>> rootdir/Android.mk:
>> rootdir/gen_initrc.py:
>> rootdir/init.rc:
>>
>> * Use a script to generate init.rc to allow different file systems
>> to be specified. (see other changes in the build project).
>>
>> frameworks/base
>>
>> servicemanager/{service_manager.c,binder.c}:
>>
>> * Provide better debugger in error cases.
>>
>> audioflinger/AudioMixer.cpp
>>
>> * Conditionalise use of halfword multiple instructions.
>>
>> libs/ui/EGLDisplaySurface.cpp
>>
>> * Provide swapBuffers() implementation for case when PAGE_FLIP is
>> not supported. This is probably not the right way to do this.
>>
>> opengl/libGLES_CM/gl_wrapper.cpp
>>
>> * Handle return correctly for non-thumb targets.
>>
>> opengl/libagl/fixed_asm.S
>>
>> * Use macros for BX<cc>
>>
>> dalvik:
>>
>> Android.mk
>>
>> * Revert to using C version of interpreter if 64bit data instructions
>> are not available. Sean's more comprehensive patches to change the
>> asm are probably better here once they are cleaned up. But this
>> does work for now
>>
>> vm/alloc/clz.{ch}:
>>
>> * Avoid using builtin clz if clz instruction is not available. Yet to
>> determine
>> exactly why this should be necessary.
>>
>> vm/arch/arm/CallEABI.S:
>>
>> * This patch works but is hacky. The important thing for this file is that
>> if
>> HAVE_THUMB is not defined, or HAVE_FASTINTERWORKING is defined, then using
>> 'bx' is not needed and values can safely be loaded directly into
>> pc. Of course
>> on ARMv4T where you have THUMB, but no FASTINTERWORKING, more
>> extensive changes
>> are required, see Sean's patches.
>>
>> build:
>>
>> core/{Makefile,config.mk}:
>>
>> * Add support for using different filesystems for system and user, as
>> well as support
>> for yaffs (not yaffs2) and jffs2. Rationale is: jffs2 is better
>> compression and can
>> work well for ro /system. yaffs works on NAND with 512 byte pages (as
>> opposed to yaffs2
>> for 2k pages). For NAND constrained devices you end up needing both
>> filesystems.
>>
>> core/prelink-linux-arm.map:
>>
>> * Very small change to layout which seems to work for both ARMv4 and ARMv5.
>>
>> bionic:
>>
>> libc/Android.mk,libc/string/ffs.c:
>>
>> * Nasty hack for ffs() implementation. Note: the whole subject of gcc
>> builtins
>> should be better addressed. I think Sean has an assembler implementation
>> of
>> ffs().
>>
>> strlen,memcpy,memset:
>>
>> * Use PLD() macro to avoid unknown instruction on ARMv4.
>>
>> gensyscalls.py:
>>
>> * Updated to use BX<cc> macro
>>
>> everything else:
>>
>> * Regeneration of syscalls.
>>
>> (Due to regenerating all syscalls this patch is quite large, hence it
>> is gzipped before attaching).
>>
>> external/yaffs2:
>>
>> * Add support for generating yaffs as well as yaffs2 images.
>>
>> external/sonivox:
>>
>> * Only compile optimised ASM if half word multiply is available.
>>
>> external/skia:
>>
>> * Conditionalise use of __built_clz, and halfword multiple optimisations.
>>
>> external/opencore:
>>
>> * Conditionalise compilation based on having halfword multiply and
>> clz instructions as necessary.
>>
>> external/jpeg:
>>
>> * Conditionalise jpeg optimisation based on having halfword multiplies.
>>
>> * Only use prefetch-loop-arrays when PLD instruction exists. This is
>> not a great approach here, I think Marcello's is probably better.
>>
>> external/extras:
>>
>> * Use BX<cc> macros.
>>
>> build.diff
>> 5KViewDownload
>>
>> dalvik.diff
>> 3KViewDownload
>>
>> external-extras.diff
>> 1KViewDownload
>>
>> external-jpeg.diff
>> 1KViewDownload
>>
>> external-opencore.diff
>> 3KViewDownload
>>
>> external-skia.diff
>> 1KViewDownload
>>
>> external-sonivox.diff
>> < 1KViewDownload
>>
>> external-yaffs2.diff
>> 1KViewDownload
>>
>> frameworks-base.diff
>> 8KViewDownload
>>
>> system-core.diff
>> 8KViewDownload
>>
>> bionic.diff.gz
>> 16KViewDownload
>>
> >
>
--~--~---------~--~----~------------~-------~--~----~
unsubscribe: [EMAIL PROTECTED]
website: http://groups.google.com/group/android-porting
-~----------~----~----~----~------~----~------~--~---
