Commit graph

205 commits

Author SHA1 Message Date
Andrew Hsieh
112a674798 Add -mno-movbe to disable "movbe" instruction from code-gen
Intruction "movbe" (move data after swapping bytes) is only available in
Intel atom.  Disable it for more general x86 devices and x86 hosts which
run emulator-x86.  Although emulator-x86 can support "movbe", most of
the time we want to run emulator-x86 with VT support for speed (with KVM
in Linux, or HAXM in Windows/MacOSX).  The presence of "movbe" can crash
emulator-x86 in VT because non-atom host doesn't have it natively.

Change-Id: I86e037900ddcb8d544f92b8d8464c8627347e560
2012-05-18 15:53:35 +08:00
Jean-Baptiste Queru
feec98b139 resolved conflicts for merge of 62855a83 to jb-dev-plus-aosp
Change-Id: Ic424ff756396463b98d137f8353d4a230e806a07
2012-05-16 13:18:39 -07:00
Jean-Baptiste Queru
2436ea8a8c Merge "Use i686-linux-android-4.6 for x86 build" 2012-05-16 06:42:55 -07:00
Nick Kralevich
ee08fe3196 stop using a custom linker script for static executables.
This is causing problems with making the linker relocatable.

Change-Id: I53e37fba52eabd3145130712d6e9bff074aed493
2012-05-15 10:40:35 -07:00
H.J. Lu
402d7f3654 Use i686-linux-android-4.6 for x86 build
Change-Id: Ibbb0a899f147013503d6830bb93050c5ae588539
Author: "H.J. Lu" <hongjiu.lu@intel.com>
2012-05-14 08:22:27 -07:00
Nick Kralevich
9d602a7005 am ecb1a565: x86: compile everything with relro / bind_now
* commit 'ecb1a565849b40ab7aa45b7fe71e24e29a059376':
  x86: compile everything with relro / bind_now
2012-05-10 09:24:10 -07:00
Nick Kralevich
ecb1a56584 x86: compile everything with relro / bind_now
Enable relro / bind_now when compiling Android applications.
This marks certain regions of memory as read-only after linking,
making memory corruption security vulnerabilities are harder
to exploit.

See:
 * http://www.akkadia.org/drepper/nonselsec.pdf (section 6)
 * http://tk-blog.blogspot.com/2009/02/relro-not-so-well-known-memory.html

relro support has been enabled for ARM since 233d460f21.
This change enables it for x86.

Change-Id: Ib90704f2fecffcfdc3587607112804faa0e16385
2012-05-09 14:49:04 -07:00
Jean-Baptiste Queru
269a40d140 am b031afab: am bf3d3287: Merge "Use the same compilation flags on OS X 10.8 as 10.7"
* commit 'b031afab543d8bf9bb236ff4c297f5ced70f8943':
  Use the same compilation flags on OS X 10.8 as 10.7
2012-05-07 13:59:56 -07:00
Al Sutton
35bb6422bc Use the same compilation flags on OS X 10.8 as 10.7
Added Mountain Lion to the list of versions which don't need ranlib
and don't need the pre-Lion linker flags

Change-Id: I0c785f0c66e324af9a209520c5a5b3c9bf7df0d5
Signed-off-by: Al Sutton <al@funkyandroid.com>
2012-05-07 09:44:28 +01:00
Jean-Baptiste Queru
76cc34ca0a am 6dfbf024: am 9e4e3b37: Merge "TARGET_linux-arm.mk: use single slash version of -print-whatever"
* commit '6dfbf0240b77acaf7d6b650b4c206deb29337e7b':
  TARGET_linux-arm.mk: use single slash version of -print-whatever
2012-05-02 17:05:04 -07:00
Jean-Baptiste Queru
9bdc290792 am 8571ee47: am f562adc8: Merge "Force use of compatibility compilers on Apple\'s command line tools"
* commit '8571ee478ee352827ec37460d23e6e0dbb1e96db':
  Force use of compatibility compilers on Apple's command line tools
2012-05-02 14:18:28 -07:00
Jean-Baptiste Queru
9e4e3b378f Merge "TARGET_linux-arm.mk: use single slash version of -print-whatever" 2012-05-02 13:26:50 -07:00
Al Sutton
0d07c732c7 Force use of compatibility compilers on Apple's command line tools
The cc and c++ compilers included in Apple's command line tools show some
unusual behaviour which causes the build to fail in several projects under
Xcode 4.3

This patch uses the gnu compatibility gcc and g++ compilers on OS X instead
to reduce problems, and generates an error when an llvm based compiler is
detected because they currently do not generate usable emulator executables
due to them not honouring global register variables.

Change-Id: I506c22dad3dcbd41df3c7672802a675d3655e262
Signed-off-by: Al Sutton <al@funkyandroid.com>
2012-05-02 07:11:27 -07:00
Nick Kralevich
233d460f21 ARM: compile everything with relro / bind_now.
Enable relro / bind_now when compiling Android applications.
This marks certain regions of memory as read-only after linking,
making memory corruption security vulnerabilities are harder
to exploit.

See:
 * http://www.akkadia.org/drepper/nonselsec.pdf (section 6)
 * http://tk-blog.blogspot.com/2009/02/relro-not-so-well-known-memory.html

Stop using the custom linker script, which inhibits
relro / bind_now support.

Change-Id: Ie97ccdd2845886bbc2ba2fdd47eed0ff4b29b60b
2012-04-26 09:09:24 -07:00
Jean-Baptiste Queru
4e82d1fa7f On linux, always use the prebuilt host toolchain
Change-Id: I8077c4780e1e3540e5e2133cf0ffa61d474d3acc
2012-04-20 19:04:56 -07:00
Jean-Baptiste Queru
a50ef1c124 Switch to new locations of toolchains
Change-Id: I15a8be11b3d0f1985ffd97dd51a1af18e1ce7b67
2012-04-20 16:15:42 -07:00
Jing Yu
c856e31523 Workaround for a gcc-4.6 bug on -fstrict-volatile-bitfields.
http://gcc.gnu.org/ml/gcc-patches/2011-03/msg01477.html
http://gcc.gnu.org/ml/gcc-patches/2011-11/msg01390.html

-fstrict-volatile-bitfields is set on by default on gcc-4.6
for ARM target. This causes an inline asm() error. While
upstream gcc community is working on the patch, we shut
off this option temporarily.

This patch does not affect gcc-4.4.3 build.

Change-Id: Id7016dc5856a516e10f2617232945034728a74ab
2012-04-20 07:53:16 -07:00
Doug Kwan
422a3650d3 Fix build problems with gcc-4.6.x.
Change-Id: If02f18efde4cec39c97bb0e7226b008cd6b85dd4
2012-04-20 07:52:47 -07:00
Nick Kralevich
b3789be766 Revert "ARM: compile everything with relro / bind_now."
This reverts commit 020e4919e8.
2012-04-10 13:13:29 -07:00
Nick Kralevich
020e4919e8 ARM: compile everything with relro / bind_now.
Enable relro / bind_now when compiling Android applications.
This marks certain regions of memory as read-only after linking,
making memory corruption security vulnerabilities are harder
to exploit.

See:
 * http://www.akkadia.org/drepper/nonselsec.pdf (section 6)
 * http://tk-blog.blogspot.com/2009/02/relro-not-so-well-known-memory.html

Stop using the custom linker script, which inhibits
relro / bind_now support.

Change-Id: Iaebfbbb492299004f9da92b649985b6cd67d8c97
2012-04-09 12:04:32 -07:00
Jing Yu
f5172c7564 Switch default gcc toolchain from 4.4.3 to 4.6.
Switch both arm-linux-androideabi toolchainand arm-eabi toolchain.

Change-Id: I7b2771fe684853ab37e3b5b7cfe4146f9f901e02
2012-03-29 20:45:50 -07:00
Andrew Hsieh
2245d17dc8 Point to new linux prebuilt toolchain for SDK build
Change-Id: I21849119b0f3f45f950e9e6d1da0833aff381fb7
2012-03-22 09:55:36 -07:00
Andrew Hsieh
0e63fd37dd Point to new linux prebuilt toolchain for SDK build
Change-Id: I21849119b0f3f45f950e9e6d1da0833aff381fb7
2012-03-13 02:17:35 -07:00
Andrew Hsieh
ed6b815505 BUILD_HOST_64bit=1 to build host tools in 64-bit
By default we build tool in 32-bit.  This flag allow individual tool and its
dependencies to be built in 64-bit (eg. Emulator).  Fixes to resolve 64-bit
porting issues (in other git) will be submitted seperately.

Change-Id: I486cf7ddac727d3c374ed890857d497c3a69e598
2012-02-28 10:24:39 -08:00
Ying Wang
92ca609cd1 Fix bug with the HOST_AR.
Change-Id: I28ab86d3b6e7cbd4391e2640cf8e2be269b03e76
2012-02-23 16:27:43 -08:00
Kenny Root
87f4d5a926 Enable TLS register by default for ARMv7a arch
Everything that is based on ARMv7a should have a cp15 TLS register.
Enable it by default so it's not accidentally missed in newer board
configurations. In fact, this could be enabled for ARMv6 as well, but we
currently don't distinguish between ARMv5 and ARMv6 in the build system.

This can still be disabled by setting it to "false" in the board
configuration, but this shouldn't ever be needed.

Change-Id: Ic2918f32899c8bcfa482f92c98f5a192fa318470
2012-02-21 09:17:18 -08:00
Ying Wang
5dea692715 Use @rpath and @loader_path to avoid absolute path
With "-Wl,-rpath,@loader_path/../lib" the linker embeds
@loader_path/../lib as the target binary's rpath.
Here @loader_path means the path of the binary that initiates the
loading.
With "-install_name @rpath/libfoo.dylib " the linker runtime searches
libfoo.dylib in the binary's embeded rpath.

With this change you can call dlopen() without specifying the full path
of the shared library.

Change-Id: If9beb3f6f4642a377bff603bab5ba3fdb96211bf
2012-02-01 13:20:30 -08:00
Ying Wang
80469e4d6c Delete unused var.
Change-Id: Ib6c295287f064529a8b4d8ada15a9fbb73064f1e
2012-02-01 11:28:00 -08:00
Ying Wang
b4c2a64040 Set install name with absolute path for Mac host shared libraries.
Before this change, path of the install name is relative to the top dir.
That means you can execute dynamically-linked binaries only in the top dir.
With this change, you can execute dynamically-linked binaries anywhere.

Change-Id: I1c6441579ffb68505ea678296aceb2e66a6df1be
2012-01-31 21:04:25 -08:00
Jim Huang
5d84f1659f TARGET_linux-arm.mk: use single slash version of -print-whatever
The double slash variants are not documented in GCC manual.  Therefore,
it is better to use the single slash variants as usual.
2012-01-31 11:11:29 +08:00
Ying Wang
fcf246c989 Workaround for lack of "-Wl,--whole-archive" in MacOS's linker.
Change-Id: I2b1de0a253d341bcb7ee05fdc96bdc0d0967985e
2012-01-20 12:59:58 -08:00
Jean-Baptiste Queru
3b1dbe9302 Merge "Revert "Reference new location of SDK toolchain"" 2012-01-11 09:04:25 -08:00
Jean-Baptiste Queru
d75dbec617 Revert "Reference new location of SDK toolchain"
This reverts commit 1b834c6660.
2012-01-11 09:00:41 -08:00
Jean-Baptiste Queru
b6737c1366 Merge "Reference new location of SDK toolchain" 2012-01-11 06:36:55 -08:00
Jean-Baptiste Queru
cb8f66305c Revert "Point arm toolchain to the new gcc-4.6 one."
This reverts commit 1f87795eef.
2012-01-11 04:23:06 -08:00
Jean-Baptiste Queru
1b834c6660 Reference new location of SDK toolchain
Change-Id: If470d742a92061006d94b72f21df5c642348e179
2012-01-09 15:07:59 -08:00
Jean-Baptiste Queru
ea91f0a40e Reference x86 toolchain in its new location
Change-Id: I990b03107d1596a87a8d4cea6973735c60db4c64
2012-01-09 13:02:03 -08:00
Ying Wang
71b5cc1b8e Reference the new prebuilt ccache location.
Change-Id: I4450a8254fe82ed3a670ce653f49381d9c075c8e
2012-01-09 11:52:23 -08:00
Jing Yu
1f87795eef Point arm toolchain to the new gcc-4.6 one.
Tested build mysid-eng, sojus-eng, trygon-eng, yakju-eng.

Change-Id: I742cf9fcf2407b9098c99136d67161ecf062ac01

Conflicts:

	core/combo/TARGET_linux-arm.mk
2012-01-06 14:14:06 -08:00
Ying Wang
c7a10ddd31 Include toolchain makefile based on the current config.
So we can have the same set of module names in different host arch
/ toolchain version combinations.

Change-Id: Iec66584bf3de92aedd71a59f9dbe74b6ed025b2e
2012-01-06 11:59:30 -08:00
Al Sutton
f79ce15891 am 1964c8ed: am d8765f0f: am 80cfc37d: Partial fix for emulator under Lion.
* commit '1964c8ed2d6b1981ee991b5ecd95ab711affdf62':
  Partial fix for emulator under Lion.
2012-01-05 12:08:29 -08:00
Al Sutton
80cfc37d0f Partial fix for emulator under Lion.
Don't use -Wl,-dynamic on Lion
2012-01-04 15:16:50 -08:00
Al Sutton
a4f5991486 am 657aeac8: am 9ce06f11: Don\'t run ranlib on MacOS 10.7
* commit '657aeac8423d9e3cd31c0dffcecaad78a53d7604':
  Don't run ranlib on MacOS 10.7
2011-12-21 16:46:37 -08:00
Al Sutton
9ce06f11d6 Don't run ranlib on MacOS 10.7
Change-Id: I74f4623e6044dd804669aaa70c6e410fa8988d75
2011-12-21 13:12:33 -08:00
David 'Digit' Turner
6371da10f8 x86: Refine the way we configure the x86 build.
This is the result of a recent conversation around the x86 ABI
and its relationship with how we build platform images. Briefly:

- We want to use feature-based macros like ARCH_X86_HAVE_$FEATURE
  instead of relying on explicit compiler macros like -DUSE_$FEATURE

- We want to allow for other x86-based arch variants, e.g. if
  someone wants to build Android for a CPU that doesn't support
  the NDK x86 ABI. However, we need to ensure these are not
  tagged compatible by mistake (see check at the end of TARGET_linux-x86.mk)

- There are several flags which use is dubious now that we have
  a dedicated toolchain to build all the x86 stuff. Comment them
  as 'to be considered for removal'. We'll do the proper checks
  later.

Change-Id: I7768d7c34d73e274cbf4c09ae831b55280d6bb58
2011-12-14 18:08:47 +01:00
Bruce Beare
29371f157a am 38158161: am 3343b27c: x86: The default ATOM BSP flags need to be in sync with the NDK ABI
* commit '381581613aab988acc6f9d5b459fd4f533d8c2a5':
  x86: The default ATOM BSP flags need to be in sync with the NDK ABI
2011-12-12 15:39:05 -08:00
Bruce Beare
3343b27c53 x86: The default ATOM BSP flags need to be in sync with the NDK ABI
Change-Id: Ic741e1795c655ace1e33ae6708219f2708358090
Signed-off-by: Bruce Beare <bruce.j.beare@intel.com>
2011-12-09 13:27:08 -08:00
Jing Yu
7bf82e0171 Workaround for a gcc-4.6 bug on -fstrict-volatile-bitfields.
http://gcc.gnu.org/ml/gcc-patches/2011-03/msg01477.html
http://gcc.gnu.org/ml/gcc-patches/2011-11/msg01390.html

-fstrict-volatile-bitfields is set on by default on gcc-4.6
for ARM target. This causes an inline asm() error. While
upstream gcc community is working on the patch, we shut
off this option temporarily.

This patch does not affect gcc-4.4.3 build.

Change-Id: Id7016dc5856a516e10f2617232945034728a74ab
2011-11-18 14:22:55 -08:00
Jim Huang
ba2c734f71 am e39ee4d5: am cb48497f: HOST_linux-x86: Avoid doubly define macro _FORTIFY_SOURCE
* commit 'e39ee4d5e043872b3ef8053678a9c6020d82d4b6':
  HOST_linux-x86: Avoid doubly define macro _FORTIFY_SOURCE
2011-11-10 21:35:56 +00:00
Jim Huang
cb48497ff2 HOST_linux-x86: Avoid doubly define macro _FORTIFY_SOURCE
While building Android with latest host gcc, it causes the following
build error:

Install: out/host/linux-x86/bin/mkbootimg
host C++: obbtool <= frameworks/base/tools/obbtool/Main.cpp
<command-line>:0:0: error: "_FORTIFY_SOURCE" redefined [-Werror]
<built-in>:0:0: note: this is the location of the previous definition
cc1plus: all warnings being treated as errors

In order to make build system happy, this patch does unset macro
_FORTIFY_SOURCE and then define it in host cflags.

Change-Id: Ic37a90f05081a2c3650b3335bc87e9e895900fb2
2011-11-10 13:28:13 -08:00