On the reader size, don't assume that the property size is PA_SIZE,
read it from the size of the file. Allows init to use a different
property size without recompiling statically linked executables.
(cherry picked from commit 285b42a04c)
Change-Id: I074204e9e6591b35faf7c1c58fb11ec162aff7bf
After download new version from upstream (OpenBSD 1.17) did the
following:
* changed all u_int* types to uint*
* add #include <sys/types.h>
All these changes are Android-specific and had been done before for
previous version (1.14).
Bug: http://code.google.com/p/android/issues/detail?id=54465
Change-Id: Ieb44e7fce4e794d997bb00ee0dd417fb61521720
Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
On the reader size, don't assume that the property size is PA_SIZE,
read it from the size of the file. Allows init to use a different
property size without recompiling statically linked executables.
Change-Id: I87fb0bf40c4724e3759a583fd9ea1f49492bc958
Names are immutable, so the fixed-sized arrays can be replaced with
variable-length ones to save memory (especially on internal tree nodes).
Change-Id: Iddf7856fba579b97f355e9ad4b3663a78767b96d
Signed-off-by: Greg Hackmann <ghackmann@google.com>
If a root process other than init calls __system_property_add, which
it should never do, it will break the design assumption that there is
only one mutator.
Pass O_EXCL to open() in map_prop_region_rw to ensure that only one
process ever has the property pages open for write.
Change-Id: I8233bfe0beaa40a5003ad53b98c661536b15f6b7
d329697 is too complicated. Change the multiple property pages back to
a single 128K property area that's mapped in entirely at initialization
(the memory will not get allocated until the pages are touched).
d329697 has other changes useful for testing (moving property area
initialization inside bionic and adding __system_property_set_filename)
so undo the change manually rather than with git revert.
Change-Id: Icd137669a4f8bc248e9dd2c1e8cc54e9193c9a6d
Signed-off-by: Greg Hackmann <ghackmann@google.com>
See the comments for an explanation of how properties are stored.
The trie structure is designed to scale better than the previous
array-based implementation. Searching an array with n properties
required average O(n) string compares of the entire key; searching the
trie requires average O(log n) string compares of each token (substrings
between '.' characters).
Change-Id: Ic28b3c5246004a3c502eb02d6e1c9b512884d872
Signed-off-by: Greg Hackmann <ghackmann@google.com>
Kernel provides virtual DSO for stack unwinding/exception handlind info for
signal usage case. Stack unwinding routines use 'dl_iterate_phdr' function
for additional DWARF info gathering from DSOs. Patch enables virtual DSO
enumeration via dl_iterate_phdr function.
Signed-off-by: Sergey Melnikov <sergey.melnikov@intel.com>
Change-Id: Ic2882b28f40b456a088bc1e63c50cbfda7e4a102
Fixes apps compiled statically against new libc but running with old
init
Bug: 9558625
Change-Id: I79e6f02575d278d4c7d8e8546d772ed0529bcaab
Signed-off-by: Greg Hackmann <ghackmann@google.com>
* commit '1642edb5208fe02ef64d5734fffaf7c4a724fd2f':
bionic: store property names as variable-length strings
bionic: prevent root processes from calling __system_property_add
bionic: revert to a single (larger) property area
bionic: reimplement property area as hybrid trie/binary tree
bionic: add missing memory barriers to system properties
bionic: make property area expandable
Names are immutable, so the fixed-sized arrays can be replaced with
variable-length ones to save memory (especially on internal tree nodes).
Signed-off-by: Greg Hackmann <ghackmann@google.com>
(cherry picked from commit 492ce95d9f)
Change-Id: Ib074192d1b71150233d78c58e9ffcf7ecf688b6b
If a root process other than init calls __system_property_add, which
it should never do, it will break the design assumption that there is
only one mutator.
Pass O_EXCL to open() in map_prop_region_rw to ensure that only one
process ever has the property pages open for write.
(cherry picked from commit fb9b7b436f)
Change-Id: I6df3afedbfb5d07891b095aa24b78278381a5aaf
d329697 is too complicated. Change the multiple property pages back to
a single 128K property area that's mapped in entirely at initialization
(the memory will not get allocated until the pages are touched).
d329697 has other changes useful for testing (moving property area
initialization inside bionic and adding __system_property_set_filename)
so undo the change manually rather than with git revert.
Signed-off-by: Greg Hackmann <ghackmann@google.com>
(cherry picked from commit 5f05348c18)
Change-Id: I690704552afc07a4dd410277893ca9c40bc13e5f
See the comments for an explanation of how properties are stored.
The trie structure is designed to scale better than the previous
array-based implementation. Searching an array with n properties
required average O(n) string compares of the entire key; searching the
trie requires average O(log n) string compares of each token (substrings
between '.' characters).
Signed-off-by: Greg Hackmann <ghackmann@google.com>
(cherry picked from commit 6ac8e6a46d)
Change-Id: Icbe31908572f33b4d9b85d5b62ac837cbd0f85e0
1) Reading the value must finish before checking whether it's intact
2) Setting the serial's dirty bit must visible before modifying the
value
3) The modified value must be visible before clearing the serial's dirty
bit
4) New properties and their TOC entries must be visible before updating
the property count
Signed-off-by: Greg Hackmann <ghackmann@google.com>
(cherry picked from commit 5bfa3ee8b3)
Change-Id: Id3fa45261fc2df2ae493ab5194bc2b6bff04e966
The property area is initially one 4K region, automatically expanding as
needed up to 64 regions.
To avoid duplicating code, __system_property_area_init() now allocates
and initializes the first region (previously it was allocated in init's
init_property_area() and initialized in bionic). For testing purposes,
__system_property_set_filename() may be used to override the file used
to map in regions.
Signed-off-by: Greg Hackmann <ghackmann@google.com>
(cherry picked from commit d32969701b)
Change-Id: I038d451fe8849b0c4863663eec6f57f6521bf4a7
Names are immutable, so the fixed-sized arrays can be replaced with
variable-length ones to save memory (especially on internal tree nodes).
Change-Id: Iee77874b4b55b0f9c5e531d1334be7f7f40086d2
Signed-off-by: Greg Hackmann <ghackmann@google.com>
If a root process other than init calls __system_property_add, which
it should never do, it will break the design assumption that there is
only one mutator.
Pass O_EXCL to open() in map_prop_region_rw to ensure that only one
process ever has the property pages open for write.
Change-Id: I6b7c118e5e2fd2b92a2b168b8454fe9707325462
d329697 is too complicated. Change the multiple property pages back to
a single 128K property area that's mapped in entirely at initialization
(the memory will not get allocated until the pages are touched).
d329697 has other changes useful for testing (moving property area
initialization inside bionic and adding __system_property_set_filename)
so undo the change manually rather than with git revert.
Change-Id: I0ecb27843404f93af5489f15bfe657d65175e4f0
Signed-off-by: Greg Hackmann <ghackmann@google.com>
See the comments for an explanation of how properties are stored.
The trie structure is designed to scale better than the previous
array-based implementation. Searching an array with n properties
required average O(n) string compares of the entire key; searching the
trie requires average O(log n) string compares of each token (substrings
between '.' characters).
Change-Id: I491305bc7aca59609abcd871a5f33d97f89ce714
Signed-off-by: Greg Hackmann <ghackmann@google.com>
Avoid keeping unnecessary file descriptors around when they're not
needed. Libc doesn't log so much that opening / closing overhead
matters.
Change-Id: I590ec5c27562db9bac025f781c48ec9a7724ce77
1) Reading the value must finish before checking whether it's intact
2) Setting the serial's dirty bit must visible before modifying the
value
3) The modified value must be visible before clearing the serial's dirty
bit
4) New properties and their TOC entries must be visible before updating
the property count
Change-Id: I26c680ec025fdb72362d5f618ec0d2b93d381233
Signed-off-by: Greg Hackmann <ghackmann@google.com>
The property area is initially one 4K region, automatically expanding as
needed up to 64 regions.
To avoid duplicating code, __system_property_area_init() now allocates
and initializes the first region (previously it was allocated in init's
init_property_area() and initialized in bionic). For testing purposes,
__system_property_set_filename() may be used to override the file used
to map in regions.
Change-Id: Ibe00ef52464bfa590953c4699a6d98383b0142b1
Signed-off-by: Greg Hackmann <ghackmann@google.com>
find_nth() will be inefficient on a trie. Since find_nth() is only used
internally and only for enumerating properties, we can add a foreach()
function to do this directly.
Signed-off-by: Greg Hackmann <ghackmann@google.com>
(cherry picked from commit 577418403d)
Change-Id: Iaca97d1182ce2c28863ba85241cbb5cf6185eb2f
find_nth() will be inefficient on a trie. Since find_nth() is only used
internally and only for enumerating properties, we can add a foreach()
function to do this directly.
Change-Id: I66bde9926c193073d74b244cce9fffd52108fff8
Signed-off-by: Greg Hackmann <ghackmann@google.com>
In 829c089f83, we disabled all
FORTIFY_SOURCE support when compiling under clang. At the time,
we didn't have proper test cases, and couldn't easily create targeted
clang tests.
This change re-enables FORTIFY_SOURCE support under clang for a
limited set of functions, where we have explicit unittests available.
The functions are:
* memcpy
* memmove
* strcpy
* strncpy
* strcat
* strncat
* memset
* strlen (with modifications)
* strchr (with modifications)
* strrchr (with modifications)
It may be possible, in the future, to enable other functions. However,
I need to write unittests first.
For strlen, strchr, and strrchr, clang unconditionally calls the
fortified version of the relevant function. If it doesn't know the
size of the buffer it's dealing with, it passes in ((size_t) -1),
which is the largest possible size_t.
I added two new clang specific unittest files, primarily copied
from fortify?_test.cpp.
I've also rebuild the entire system with these changes, and didn't
observe any obvious problems.
Change-Id: If12a15089bb0ffe93824b485290d05b14355fcaa
Define __errordecl and replace __attribute__((__error__("foo")))
with __errordecl. Make sure __errordecl is a no-op on clang, as it
generates a compile time warning.
Change-Id: Ifa1a2d3afd6881de9d479fc2adac6737871a2949
Move the implementation of writing to the system property area
from init to bionic, next to the reader implementation. This
will allow full property testing to be added to bionic tests.
Add new accessor and waiting functions to hide the implementation
from watchprops and various bionic users.
Also hide some of the implementation details of the property area
from init by moving them into _system_properties.h, and other details
from everybody by moving them into system_properties.h.
(cherry picked from commit dc1038b790)
Change-Id: I192d3825ee276c5047bc751039fe6cfe226a7cca
Move the implementation of writing to the system property area
from init to bionic, next to the reader implementation. This
will allow full property testing to be added to bionic tests.
Add new accessor and waiting functions to hide the implementation
from watchprops and various bionic users.
Also hide some of the implementation details of the property area
from init by moving them into _system_properties.h, and other details
from everybody by moving them into system_properties.h.
Change-Id: I9026e604109e30546b2849b60cab2e7e5ff00ba5
We were missing SIG_ATOMIC_MAX, SIG_ATOMIC_MIN, SIZE_MAX,
WCHAR_MAX, WCHAR_MIN, WINT_MAX, and WINT_MIN.
Change-Id: I2535f36bc220fbaea009b483599b7af811c4cb5c
Removed 'join_count' from pthread_internal_t and switched to using the flag
PTHREAD_ATTR_FLAG_JOINED to indicate if a thread is being joined. Combined with
a switch to a while loop in pthread_join, this fixes spurious wake-ups but
prevents a thread from being joined multiple times. This is fine for
two reasons:
1) The pthread_join specification allows for undefined behavior when multiple
threads try to join a single thread.
2) There is no thread safe way to allow multiple threads to join a single
thread with the pthread interface. The second thread calling pthread_join
could be pre-empted until the thread is destroyed and its handle reused for
a different thread. Therefore multi-join is always an error.
Bug: https://code.google.com/p/android/issues/detail?id=52255
Change-Id: I8b6784d47620ffdcdbfb14524e7402e21d46c5f7
This reverts commit d8627af159 which caused build breakage:
In file included from bionic/libc/include/limits.h:86:0,
from bionic/libc/include/stdint.h:33,
from bionic/libc/arch-arm/bionic/crtbegin.c:31:
bionic/libc/include/sys/limits.h:30:26: fatal error: linux/limits.h: No such file or directory
compilation terminated.
make: *** [out/target/product/generic/obj/lib/crtbegin_dynamic1.o] Error 1
Change-Id: I128095ecb99df92626e1f57e34c61e08c98a4078
* A dlmalloc usage error shouldn't call abort(3) because we want to
cause a SIGSEGV by writing the address dlmalloc didn't like to an
address the kernel won't like, so that debuggerd will dump the
memory around the address that upset dlmalloc.
* Switch to the simpler FreeBSD/NetBSD style of registering stdio
cleanup. Hopefully this will let us simplify more of the stdio
implementation.
* Clear the stdio cleanup handler before we abort because of a dlmalloc
corruption error. This fixes the reported bug, where we'd hang inside
dlmalloc because the stdio cleanup reentered dlmalloc.
Bug: 9301265
Change-Id: Ief31b389455d6876e5a68f0f5429567d37277dbc
- Implemented chk_memalign.
- Fixed a few bugs in leak_memalign.
- Implemented {leak,fill,check,qemu}_malloc_usable_size.
- Make malloc_usable_size update at run time.
- Add malloc_test.cpp as a small set of tests for the
malloc debug routines.
- Fix the qemu routines since it's been broken since it moved to C++.
- Add support for the %u format to the out_vformat in libc_logging.cpp.
This is used by the emulator code.
Tested using the bionic-unit-tests with setprop libc.debug.malloc
set to 1, 5, and 10.
I tested as much as possible on the emulator, but tracing doesn't appear
to be working properly.
Bug: 6143477
Merge change from internal master.
(cherry-picked from commit 3d594c2580)
Change-Id: I4ae00fffba82315a8c283f35893fd554460722fb
- Implemented chk_memalign.
- Fixed a few bugs in leak_memalign.
- Implemented {leak,fill,check,qemu}_malloc_usable_size.
- Make malloc_usable_size update at run time.
- Add malloc_test.cpp as a small set of tests for the
malloc debug routines.
- Fix the qemu routines since it's been broken since it moved to C++.
- Add support for the %u format to the out_vformat in libc_logging.cpp.
This is used by the emulator code.
Tested using the bionic-unit-tests with setprop libc.debug.malloc
set to 1, 5, and 10.
I tested as much as possible on the emulator, but tracing doesn't appear
to be working properly.
Bug: 6143477
Change-Id: Ieba99b58c2228c88d80afd264501004a4dadd212
Add methods to attach/detach UID ranges to a specific dns cache/interface.
This mirrors the already existing code for attaching specific processes to
specific interfaces but will be used to push all processes from a given
user to a specific cache/interface.
Change-Id: Ic24391e92d3ca46fcb46cc4fc53e13984dec40b3
sigismember, sigaddset, and sigdelset had mixed code and declarations
which are not allowed in C90 and before.
Change-Id: I662af944fc1489e34bed228ce592e41f50d00e17
Signed-off-by: Erik Gilling <konkers@android.com>
__strcat_chk and __strncat_chk are slightly inefficient,
because they end up traversing over the same memory region
two times.
This change optimizes __strcat_chk / __strncat_chk so they
only access the memory once. Although I haven't benchmarked these
changes, it should improve the performance of these functions.
__strlen_chk - expose this function, even if -D_FORTIFY_SOURCE
isn't defined. This is needed to compile libc itself without
-D_FORTIFY_SOURCE.
Change-Id: Id2c70dff55a276b47c59db27a03734d659f84b74
The function should take a 'const void*' parameter, instead of 'void*'.
Note that the implementation in upstream-dlmalloc/malloc.c already does
this.
For context, see http://b.android.com/55725
Change-Id: Iefd55cdb8996699189e0545f9195972490306227
Fix license clause numbering.
Convert libc/stdio from K&R to ANSI C.
And add '__restrict' where it appeared in the header prototypes.
Change-Id: I5fdb22f79d3effa2298d03f9aa8412b4b087da04
Upstream: http://svnweb.freebsd.org/base?view=revision&revision=249810
Add CLOCK_MONOTONIC_RAW, CLOCK_REALTIME_COARSE, and CLOCK_MONOTONIC_COARSE
as supported by recent linux kernels.
(cherry-pick of 60e5144ca312b210b54ac8e6966108da0c97ff80.)
Bug: 8895727
Change-Id: If79a4d05d1301108f49a37588f9416c4be19277a
Add CLOCK_MONOTONIC_RAW, CLOCK_REALTIME_COARSE, and CLOCK_MONOTONIC_COARSE
as supported by recent linux kernels.
Bug: 8895727
Change-Id: I3d415a2edbcf2928dd855e876337bf2239ac134a
Streamline the memcpy a bit removing some unnecessary instructions.
The biggest speed improvement comes from changing the size of
the preload. On krait, the sweet spot for the preload in the main
loop is twice the L1 cache line size.
In most cases, these small tweaks yield > 1000MB/s speed ups. As
the size of the memcpy approaches about 1MB, the speed improvement
disappears.
Change-Id: Ief79694d65324e2db41bee4707dae19b8c24be62
Add support for fortify source level 2 to strncpy.
This will enable detection of more areas where strncpy
is used inappropriately. For example, this would have detected
bug 8727221.
Move the fortify_source tests out of string_test.cpp, and
put it into fortify1_test.cpp.
Create a new fortify2_test.cpp file, which copies all
the tests in fortify1_test.cpp, and adds fortify_source level
2 specific tests.
Change-Id: Ica0fba531cc7d0609e4f23b8176739b13f7f7a83
Pick up Linux kernel patch 2748e5dec7ca8a3804852c7c4171f9156384d15c
from 2007
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=2748e5dec7ca8a3804852c7c4171f9156384d15c
[NETFILTER]: Fix iptables ABI breakage on (at least) CRIS
With the introduction of x_tables we accidentally broke compatibility
by defining IPT_TABLE_MAXNAMELEN to XT_FUNCTION_MAXNAMELEN instead of
XT_TABLE_MAXNAMELEN, which is two bytes larger.
On most architectures it doesn't really matter since we don't have
any tables with names that long in the kernel and the structure
layout didn't change because of alignment requirements of following
members. On CRIS however (and other architectures that don't align
data) this changed the structure layout and thus broke compatibility
with old iptables binaries.
Changing it back will break compatibility with binaries compiled
against recent kernels again, but since the breakage has only been
there for three releases this seems like the better choice.
Change-Id: Ie9552b25892109c7042b9752132dc8ebd3552dc3
From the release notes:
Changes affecting current and future time stamps:
Palestine observed DST starting March 29, 2013. (Thanks to
Steffen Thorsen.) From 2013 on, Gaza and Hebron both observe DST,
with the predicted rules being the last Thursday in March at 24:00
to the first Friday on or after September 21 at 01:00.
Assume that the recent change to Paraguay's DST rules is permanent,
by moving the end of DST to the 4th Sunday in March every year.
(Thanks to Carlos Raúl Perasso.)
Changes affecting past time stamps:
Fix some historical data for Palestine to agree with that of
timeanddate.com, as follows:
The spring 2008 change in Gaza and Hebron was on 00:00 Mar 28, not
00:00 Apr 1.
The fall 2009 change in Gaza and Hebron on Sep 4 was at 01:00, not
02:00.
The spring 2010 change in Hebron was 00:00 Mar 26, not 00:01 Mar 27.
The spring 2011 change in Gaza was 00:01 Apr 1, not 12:01 Apr 2.
The spring 2011 change in Hebron on Apr 1 was at 00:01, not 12:01.
The fall 2011 change in Hebron on Sep 30 was at 00:00, not 03:00.
Fix times of habitation for Macquarie to agree with the Tasmania
Parks & Wildlife Service history, which indicates that permanent
habitation was 1899-1919 and 1948 on.
Changing affecting metadata only:
Macquarie Island is politically part of Australia, not Antarctica.
(Thanks to Tobias Conradi.)
Sort Macquarie more-consistently with other parts of Australia.
(Thanks to Tim Parenti.)
Change-Id: I3df146d046eda26dbc6ce2a0a26ad9214ec4eaca
Added the following headers for ALSA compressed
user space API:
- sound/compress_offload.h
- sound/compress_params.h
Change-Id: I9584c12acdadf31546d2921393b93d491e54a8d3
This uses the new code original submitted as memcpy.a15.S as
the base. However, the old code handled unaligned src/dst better
so that was spliced in. I optimized the original unaligned code by
removing a few unnecessary instructions. I optimized the a15 code by
rewriting the pre and post code. I also modified the main loop to add
a pld so that larger copies would not stall waiting for memory.
Test cases for the new memcpy:
- Copy all sized values from 0 to 1024 bytes, using whatever alignment
is returned by malloc.
For each alignment case described below, the test copied from 0 to 128
bytes.
- Src and dst pointers are both aligned to the same value, starting
at one going through every power of two up to and including 128.
- Src aligned to double word boundary, dst aligned to word boundary.
- Src aligned to word boundary, dst aligned to double word boundary.
- Src aligned to 16 bit boundary, dst aligned to word boundary.
- Src aligned to word boundary, dst aligned to 16 byte boundary.
- Src aligned to word boundary, dst aligned to 1 byte from a word
boundary.
- Src aligned to word boundary, dst aligned to 2 bytes from a word
boundary.
- Src aligned to word boundary, dst aligned to 3 bytes from a word
boundary.
- Src aligned to 1 byte from a word boundary, dst aligned to a word
boundary.
- Src aligned to 2 bytes from a word boundary, dst aligned to a word
boundary.
- Src aligned to 3 bytes from a word boundary, dst aligned to a word
boundary.
Cases to verify the unaligned source code properly aligns to a 16 bit
boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
4 + 128 bit boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
8 + 128 bit boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
12 + 128 bit boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
16 + 128 bit boundary.
In all cases, a two byte fencepost was placed at the end of the
destination to verify that only the requested number of bytes were copied.
Bug: 8005082
Merge from internal master.
(cherry-picked from commit 21ede92d79)
Change-Id: Ief70c9e6dc8c6473ae245b6570b2c266fed9618c
This adds __libc_fatal, cleans up the internal logging code a bit more,
and switches suitable callers over to __libc_fatal. In addition to logging,
__libc_fatal stashes the message somewhere that the debuggerd signal handler
can find it before calling abort.
In the debuggerd signal handler, we pass this address to debuggerd so that
it can come back with ptrace to read the message and present it to the user.
Bug: 8531731
(cherry picked from commit 0d787c1fa1)
Change-Id: I5daeeaa36c1fc23f7f437d73a19808d9d558dd4d
This adds __libc_fatal, cleans up the internal logging code a bit more,
and switches suitable callers over to __libc_fatal. In addition to logging,
__libc_fatal stashes the message somewhere that the debuggerd signal handler
can find it before calling abort.
In the debuggerd signal handler, we pass this address to debuggerd so that
it can come back with ptrace to read the message and present it to the user.
Bug: 8531731
Change-Id: I416ec1da38a8a1b0d0a582ccd7c8aaa681ed4a29
- eventfd.cpp and eventfd.s will output to the same file when building libc.a
out/target/product/*/obj/STATIC_LIBRARIES/libc_intermediates/WHOLE/libc_common_objs/eventfd.o
- And then `eventfd` will undefined when statically linked to libc.
Also add a unit test.
(cherry-pick of 8baa929d5d3bcf63381cf78ba76168c80c303f5e.)
Change-Id: Icd0eb0f4ce0511fb9ec00a504d491afd47d744d3
We use the system call constants from the kernel header files now,
so there's no need to check that they've been correctly transcribed
into SYSCALLS.TXT.
This is a work in progress. I've added TODOs to SYSCALLS.TXT explaining
what's left to do.
(cherry-pick of a51916b58b2d211bcf8ffdbe9cf7faa58e57382f.)
Change-Id: I4484acd946b1f548ac3d95327e58add9f98246ab
- eventfd.cpp and eventfd.s will output to the same file when building libc.a
out/target/product/*/obj/STATIC_LIBRARIES/libc_intermediates/WHOLE/libc_common_objs/eventfd.o
- And then `eventfd` will undefined when statically linked to libc.
Also add a unit test.
Change-Id: Ib310ade3256712ca617a90539e8eb07459c98505
We use the system call constants from the kernel header files now,
so there's no need to check that they've been correctly transcribed
into SYSCALLS.TXT.
This is a work in progress. I've added TODOs to SYSCALLS.TXT explaining
what's left to do.
Change-Id: I3b86acfe7f84b4da1c802ee5a4ef13a2e83e7939
Currently, our getaddrinfo implementation does not conform to
any IETF standard. It follows draft-ietf-6man-rfc3484-revise-01,
but that draft has expired. Update the policy table to RFC6724.
(cherry-pick of e919b116d35aa7deb24ddece69c491e24c3b0d6f.)
Bug: 8276725
Change-Id: I2d17122defd966ac6c2c13d04887fb110f2598a0
Currently, our getaddrinfo implementation does not conform to
any IETF standard. It follows draft-ietf-6man-rfc3484-revise-01,
but that draft has expired. Update the policy table to RFC6724.
Bug: 8276725
Change-Id: I03c63abfcad9b2f3a3bab2718bd2fc6440531843
This uses the new code original submitted as memcpy.a15.S as
the base. However, the old code handled unaligned src/dst better
so that was spliced in. I optimized the original unaligned code by
removing a few unnecessary instructions. I optimized the a15 code by
rewriting the pre and post code. I also modified the main loop to add
a pld so that larger copies would not stall waiting for memory.
Test cases for the new memcpy:
- Copy all sized values from 0 to 1024 bytes, using whatever alignment
is returned by malloc.
For each alignment case described below, the test copied from 0 to 128
bytes.
- Src and dst pointers are both aligned to the same value, starting
at one going through every power of two up to and including 128.
- Src aligned to double word boundary, dst aligned to word boundary.
- Src aligned to word boundary, dst aligned to double word boundary.
- Src aligned to 16 bit boundary, dst aligned to word boundary.
- Src aligned to word boundary, dst aligned to 16 byte boundary.
- Src aligned to word boundary, dst aligned to 1 byte from a word
boundary.
- Src aligned to word boundary, dst aligned to 2 bytes from a word
boundary.
- Src aligned to word boundary, dst aligned to 3 bytes from a word
boundary.
- Src aligned to 1 byte from a word boundary, dst aligned to a word
boundary.
- Src aligned to 2 bytes from a word boundary, dst aligned to a word
boundary.
- Src aligned to 3 bytes from a word boundary, dst aligned to a word
boundary.
Cases to verify the unaligned source code properly aligns to a 16 bit
boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
4 + 128 bit boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
8 + 128 bit boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
12 + 128 bit boundary.
- Src aligned to 1 byte from a 128 bit boundary, dst aligned to
16 + 128 bit boundary.
In all cases, a two byte fencepost was placed at the end of the
destination to verify that only the requested number of bytes were copied.
Bug: 8005082
Change-Id: I700b2fab81941959d301ab1934c18fbd8ee3eee4
pthread_create returns EAGAIN when it can't allocate a pthread_internal_t,
when it can't allocate a stack for the new thread, or when clone(2) fails
because there are too many threads. It's useful to be able to know why your
pthread_create just failed, so add some logging.
Bug: 8470684
(cherry picked from commit cfa089df23)
Change-Id: Ibfc98a84c1817a931f9ae4c2b88762f0edfb6b79
pthread_create returns EAGAIN when it can't allocate a pthread_internal_t,
when it can't allocate a stack for the new thread, or when clone(2) fails
because there are too many threads. It's useful to be able to know why your
pthread_create just failed, so add some logging.
Bug: 8470684
Change-Id: I1bb4497d4f7528eacce0db35c2014771cba64569
Need to get the defintion of the FITRIM ioctl(). Also need
to update the blk_types.h header file as fs.h includes it.
Change-Id: I617622b620925942dc5aead9e49f8e17d17e5d74
The <asm/unistd.h> files contain the canonical data, and
<sys/glibc-syscalls.h> contain new glibc-compatible names,
and if you #include the standard <sys/syscall.h> you get
both sets of names.
Change-Id: I9919c080931c0ba1660f5e37c6a6265ea716d603
This lets us move all the ARM syscall stubs over to the kernel <asm/unistd.h>.
Our generated <sys/linux-syscalls.h> is now unused, but I'll remove that in a
later change.
Change-Id: Ie5ff2cc4abce1938576af7cbaef615a79c7f310d
Also add a more intention-revealing guard so we don't have loads of
places checking whether our inlining macro is defined.
Change-Id: I168860cedcfc798b07a5145bc48a125700265e47
Need to get the defintion of the FITRIM ioctl(). Also need
to update the blk_types.h header file as fs.h includes it.
Change-Id: I617622b620925942dc5aead9e49f8e17d17e5d74
For some reason, socketcalls.c was only being compiled for ARM, where
it makes no sense. For x86 we generate stubs for the socket functions
that use __NR_socketcall directly.
Change-Id: I84181e6183fae2314ae3ed862276eba82ad21e8e
<sys/linux-syscalls.h> only contains constants for the syscalls
we're generating stubs for. We want all the syscalls available
on the architecture in question.
Keep using <sys/linux-syscalls.h> on ARM for now because the
__NR_ARM_set_tls and __NR_ARM_cacheflush values aren't in <asm/unistd.h>.
Change-Id: I66683950d87d9b18d6107d0acc0ed238a4496f44
Fixes the MIPS and x86 builds. strace tests whether syscalls
are supported using #ifdef of the appropriate SYS_ constant.
Change-Id: I90be118dc42abfdaf5b0f9b1e676e8601f55106e
This uses the new strcmp.a15.S code as the basis for new versions
of strcmp.S.
The cortex-a15 code is the performance optimized version of strcmp.a15.S
taken with only the addition of a few pld instructions.
The cortex-a9 code is the same as the cortex-a15 code except that the
unaligned strcmp code was taken from the original strcmp.S.
The krait code is the same as the cortex-a15 code except that one path
in the unaligned strcmp code was taken from the original strcmp.S code
(the 2 byte overlap case).
The generic code is the original unmodified strmp.S from the bionic
subdirectory.
All three new versions underwent these test cases:
Strings the same, all same size:
- Both pointers double word aligned.
- One pointer double word aligned, one pointer word aligned.
- Both pointers word aligned.
- One pointer double word aligned, one pointer 1 off a word alignment.
- One pointer double word aligned, one pointer 2 off a word alignment.
- One pointer double word aligned, one pointer 3 off a word alignment.
- One pointer word aligned, one pointer 1 off a word alignment.
- One pointer word aligned, one pointer 2 off a word alignment.
- One pointer word aligned, one pointer 3 off a word alignment.
For all cases where it made sense, the two pointers were also tested
swapped.
Different strings, all same size:
- Single difference at double word boundary.
- Single difference at word boudary.
- Single difference at 1 off a word alignment.
- Single difference at 2 off a word alignment.
- Single difference at 3 off a word alignment.
Different sized strings, strings the same until the end:
- Shorter string ends on a double word boundary.
- Shorter string ends on word boundary.
- Shorter string ends at 1 off a word boundary.
- Shorter string ends at 2 off a word boundary.
- Shorter string ends at 3 off a word boundary.
For all different cases, run them through the same pointer alignment
cases when the strings are the same size.
For all cases the two pointers were also tested swapped.
Bug: 8005082
Merge from internal master.
(cherry-picked from commit a9a5870d16)
Change-Id: I4c2b98f8a50804fb98ab67f75e9d660f1315a144
We only need one logging API, and I prefer the one that does no
allocation and is thus safe to use in any context.
Also use O_CLOEXEC when opening the /dev/log files.
Move everything logging-related into one header file.
Change-Id: Ic1e3ea8e9b910dc29df351bff6c0aa4db26fbb58
The defines HAVE_32_BYTE_CACHE_LINES and ARCH_ARM_USE_NON_NEON_MEMCPY
are not used by any code. The previous memcpy code that used these
has been split into different architecture versions to avoid the need
for them.
Bug: 8005082
Merge from internal master.
(cherry-picked from commit 6e1a5cf31b)
Change-Id: Ib18fc3f4131b21cdbd19b9dde7697ac25d066fcf
In the old code, the index was a file to itself, so it made sense to
read until you hit the end of the file. In the new code, the index is
followed by hundreds of KiB of data, so we need to just search the
index.
Bug: 8368791
Change-Id: Icf5f8b5516cf3a93679fa849c9f6cd1cb100e0f1
Move arch specific code for arm, mips, x86 into separate
makefiles.
In addition, add different arm cpu versions of memcpy/memset.
Bug: 8005082
Merge from internal master (acdde8c1cf).
Change-Id: I04f3d0715104fab618e1abf7cf8f7eec9bec79df
This uses the new strcmp.a15.S code as the basis for new versions
of strcmp.S.
The cortex-a15 code is the performance optimized version of strcmp.a15.S
taken with only the addition of a few pld instructions.
The cortex-a9 code is the same as the cortex-a15 code except that the
unaligned strcmp code was taken from the original strcmp.S.
The krait code is the same as the cortex-a15 code except that one path
in the unaligned strcmp code was taken from the original strcmp.S code
(the 2 byte overlap case).
The generic code is the original unmodified strmp.S from the bionic
subdirectory.
All three new versions underwent these test cases:
Strings the same, all same size:
- Both pointers double word aligned.
- One pointer double word aligned, one pointer word aligned.
- Both pointers word aligned.
- One pointer double word aligned, one pointer 1 off a word alignment.
- One pointer double word aligned, one pointer 2 off a word alignment.
- One pointer double word aligned, one pointer 3 off a word alignment.
- One pointer word aligned, one pointer 1 off a word alignment.
- One pointer word aligned, one pointer 2 off a word alignment.
- One pointer word aligned, one pointer 3 off a word alignment.
For all cases where it made sense, the two pointers were also tested
swapped.
Different strings, all same size:
- Single difference at double word boundary.
- Single difference at word boudary.
- Single difference at 1 off a word alignment.
- Single difference at 2 off a word alignment.
- Single difference at 3 off a word alignment.
Different sized strings, strings the same until the end:
- Shorter string ends on a double word boundary.
- Shorter string ends on word boundary.
- Shorter string ends at 1 off a word boundary.
- Shorter string ends at 2 off a word boundary.
- Shorter string ends at 3 off a word boundary.
For all different cases, run them through the same pointer alignment
cases when the strings are the same size.
For all cases the two pointers were also tested swapped.
Bug: 8005082
Change-Id: I5f3dc02b48afba2cb9c13332ab45c828ff171a1c
Normally, the C library implicitly caches your timezone by virtue
of the fact that the prehistoric API assumes a single timezone for
the entire process.
The unfortunate mktime_tz and localtime_tz extensions work around
this, but represent timezones as strings to their callers, so code
that makes heavy use of these needs a cache to be able to perform
acceptably until it can hopefully one day be rewritten to use
java.util.Calendar or icu4c.
Bug: 8270865
Change-Id: I92e3964e86dc33ceac925f819cc5e26ff4203f50
From the release notes:
Changes affecting current and future time stamps:
Haiti uses US daylight-saving rules this year, and presumably future years.
This changes time stamps starting today. (Thanks to Steffen Thorsen.)
Paraguay will end DST on March 24 this year.
(Thanks to Steffen Thorsen.) For now, assume it's just this year.
Morocco does not observe DST during Ramadan;
try to predict Ramadan in Morocco as best we can.
(Thanks to Erik Homoet for the heads-up.)
Change-Id: I98d5290ea5a1d9fb1eeddab1c9e72135dc9e4bd1
This gets us back to using vfork now our ARM vfork assembler stub is
fixed, and adds the missing thread safety for the 'pidlist'.
Bug: 5335385
Change-Id: Ib08bfa65b2cb9fa695717aae629ea14816bf988d
From the release notes:
Chile's 2013 rules, and we guess rules for 2014 and later, will be
the same as 2012, namely Apr Sun>=23 03:00 UTC to Sep Sun>=2 04:00 UTC.
(Thanks to Steffen Thorsen and Robert Elz.)
New Zones Asia/Khandyga, Asia/Ust-Nera, Europe/Busingen.
(Thanks to Tobias Conradi and Arthur David Olson.)
Change-Id: I351e04b3348420ad7df7c648963c235b534033d2
This is actually a slightly newer upstream version than the one I
originally pulled. Hopefully now it's in upstream-freebsd it will
be easier to track upstream, though I still need to sit down and
write the necessary scripts at some point.
Bug: 5110679
Change-Id: I87e563f0f95aa8e68b45578e2a8f448bbf827a33
The defines HAVE_32_BYTE_CACHE_LINES and ARCH_ARM_USE_NON_NEON_MEMCPY
are not used by any code. The previous memcpy code that used these
has been split into different architecture versions to avoid the need
for them.
Bug: 8005082
(cherry picked from commit 6e1a5cf31b)
Change-Id: I69654d47db1458136782b5504290f620e924ee75
Don't pull in unnecessary header files. AFAIK, I've fixed all
the code which didn't include the correct header files.
Change-Id: If0b7bba74e77cb24a0cf9ce8968aa07400855e58
The attached patch provides a new implementation of strcmp for ARM,
using LDRD instead of LDR whenever possible.
For older architectures that do not support LDRD, this implementation
uses the same algorithm as before.
Testing and benchmarking:
* Validation: successfully passes a test that compares different strings
of length 1-128 and offsets 0-8 from a word boundary. Checked on
qemu/A15/A9, ARM/Thumb mode, Big/Little Endian.
* Integration with gcc: no regression on qemu for arm-none-eabi --with-cpu
a15/a9 --with-mode arm/thumb.
Change-Id: I9e230e1b99dbdc9119b69ee858a89038c516a4ea
Signed-off-by: Vassilis Laganakos <vasileios.laganakos@arm.com>
The strategy for large block sizes is LDRD and STRD with offset addressing,
where the main loop copies 64 bytes in every iteration, (i.e., 8 calls to
LDRD and STRD pairs), interleaving load and stores (i.e., the pairs of LDRD
and STRD of the same data are consecutive instructions), and the writeback
of an updated address is a separate instruction, which allows us to write
back the accumulated update once per iteration.
This strategy is implemented in memcpy.S. In some configurations, a plain
version of memcpy (included from memcpy-stub.c) is used instead of the
optimized one.
Validation:
* Correctness: checked memcpy using a test harness for block sizes
ranging between 1 to 128, and source and destination buffers alignment
ranging in { 0,1,2,3,4,8,12 } bytes each.
* Performance: benchmarking on Cortex-A15 FPGA indicates that this strategy
is better for A15 than the strategy used by glibc and even slightly better
than using NEON. Benchmarking on Cortex-A9 bare metal and Linux shows
that the proposed strategy is reasonable: not as fast as the version of
memcpy from glibc (which is the best open source strategy for A9), but
comparable with csl and bionic.
* Integration with GCC: no regression for arm-none-eabi --with-cpu
cortex-a15 and cortex-a9.
Change-Id: Ied56354d8992c62ae3e02d582a2bd55585d814b9
Signed-off-by: Vassilis Laganakos <vasileios.laganakos@arm.com>
Move arch specific code for arm, mips, x86 into separate
makefiles.
In addition, add different arm cpu versions of memcpy/memset.
Bug: 8005082
(cherry picked from commit acdde8c1cf)
Change-Id: I0108d432af9f6283ae99adfc92a3399e5ab3e31d
Stricter input parameters help avoid ugly casting when passing
pointers to immutable protobuf data.
While at it: an int return was dropped from 2 functions whose users
never used the result; one of the return paths was returning an
uninitialized value.
Size_t for portablity and warning supression, misc warnings addressed.
Change-Id: I2d5cbdaf0c9b6c4621a7d397772da13da5dc0943
The defines HAVE_32_BYTE_CACHE_LINES and ARCH_ARM_USE_NON_NEON_MEMCPY
are not used by any code. The previous memcpy code that used these
has been split into different architecture versions to avoid the need
for them.
Bug: 8005082
Change-Id: I14e0368c5bb4c3a80e72520f7cfd97e359918cce
Move arch specific code for arm, mips, x86 into separate
makefiles.
In addition, add different arm cpu versions of memcpy/memset.
Bug: 8005082
Change-Id: I04f3d0715104fab618e1abf7cf8f7eec9bec79df
Per "man capset", sys/capability.h is the appropriate header file
for the capget / capset definition, not unistd.h. Fixed.
As a short term hack, continue to include sys/capability.h in
unistd.h, until we can fix all the code which uses capget / capset.
Change-Id: I6e7cf55955d761ca785a14c5e4b7a44125d8fc15
The old scandir implementation didn't take into account the varying
size of directory entries, and didn't correctly clean up on its
error exits.
Bug: 7339844
Change-Id: Ib40e3564709752241a3119a496cbb2192e3f9abe
dnsproxyd can already determine our pid by looking at our
socket connection. It's dangerous (and unneeded) to pass it
ourselves.
Change-Id: I2596d02e361b302259ddb084be2fb75be59889c5
When an app doesn't have the internet permission, android_open_proxy
returns NULL, causing a segfault when calling fprintf. Fixed.
Change-Id: I598855350ed0db3cc88e5ae3b400145418a3a615
AFAIK, bionic only ever provided an implementation of bcmp
for x86, and even then, the code was never actually compiled.
Remove the prototype.
bcmp() has been obsoleted and replaced by memcmp()
Change-Id: I549d02ab6a9241a9acbbbfade0d98a9a02c2eaee
We could special-case raise(3) in non-threaded programs, but the more
conservative course is to make pthread_kill(3) work in signal handlers
at the cost of a race shared by other C libraries.
Change-Id: I59fb23d03bdabf403435e731704b33acdf3e0234
after change 32822 was rejected, this is the more light-weight
version of the fix: libc/include/sys/types.h already - via
libc/kernel/common/linux/posix_types.h - includes a definition
of __kernel_ssize_t from libc/kernel/arch-*/asm/posix_types.h
which is architecture-specific, toolchain-agnostic and also
gets rid of the gcc -Wformat warning (which it issues correctly,
since this i̲s̲ indeed a bug in bionic)
Change-Id: Ie4503ab16628bc25815a836d07556f665e9795c7
imgtec pointed out that pthread_kill(3) was broken, but most of the
other functions that ought to return ESRCH for invalid/exited threads
were equally broken.
Change-Id: I96347f6195549aee0c72dc39063e6c5d06d2e01f
Fix the pthread_setname_np test to take into account that emulator kernels are
so old that they don't support setting the name of other threads.
The CLONE_DETACHED thread is obsolete since 2.5 kernels.
Rename kernel_id to tid.
Fix the signature of __pthread_clone.
Clean up the clone and pthread_setname_np implementations slightly.
Change-Id: I16c2ff8845b67530544bbda9aa6618058603066d
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '0a2cb815974ea96af664fa966079966a08916722':
Simplify __stack_chk_fail, and fix it so we get debuggerd stack traces.
libc_bionic.a is already compiled -Werror, but this one file gets
compiled into its own library because it needs to be compiled with
-fno-stack-protector.
Change-Id: I273c535ab5c73ccaccbcf793fda1f788a2589abe
bionic/libc/bionic/ssp.cpp:41:31: warning: converting to non-pointer type 'uintptr_t {aka unsigned int}' from NULL [-Wconversion-null]
Change-Id: Id154ed4a99520cca64ffd3dbe4d743db6e2da28a
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '95b1ea1bb3c93369d96045420b91b7617992d8bd':
Add a bunch more missing ENDs to assembler routines.
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '991ee7d89574e8d04c8863a2850613073a2f96b3':
Simplify pthread_create, using more public API.
If r0 == 0, we're the child. If r0 > 0, we're the parent.
Otherwise set errno.
The __bionic_clone code I copy & pasted was wrong. This patch
fixes both.
Bug: 3461078
Change-Id: Ibb7d6cc7e54e666841f2f0dc59a141a0b31982e4
This reverts commit 6f94de3ca4
(Doesn't try to increase the number of TLS slots; that leads to
an inability to boot. Adds more tests.)
Change-Id: Ia7d25ba3995219ed6e686463dbba80c95cc831ca
# By Andrew Boie
# Via Andrew Boie (1) and Gerrit Code Review (1)
* commit '3fa67465365baf87f751bea09847e32f01005788':
add factory property file definition
# Via Android Git Automerger (1) and others
* commit '6b73d13fa414afeecba6718bf724e8ac922bac39':
Revert "Revert "Pull the pthread_key_t functions out of pthread.c.""
# Via Gerrit Code Review (2) and Android Git Automerger (1)
* commit 'e4b08318c13fac774b233a5459427563d2983f79':
Revert "Pull the pthread_key_t functions out of pthread.c."
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit 'cae7b2cfb509e7d5d10a8085b1ec319daaef768f':
Fix __pthread_clone on ARM to set errno on failure.
MIPS and x86 appear to have been correct already.
(Also fix unit tests that ASSERT_EQ with errno so that the
arguments are in the retarded junit order.)
Bug: 3461078
Change-Id: I2418ea98927b56e15b4ba9cfec97f5e7094c6291
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '1fea0f258a45d918fe5ae8e9769f45c0348bd095':
Clean up ARM assembler files to use ENTRY/END.
POSIX says pthread_create returns EAGAIN, not ENOMEM.
Also pull pthread_attr_t functions into their own file.
Also pull pthread_setname_np into its own file.
Also remove unnecessary #includes from pthread_key.cpp.
Also account for those pthread keys used internally by bionic,
so they don't count against the number of keys available to user
code. (They do with glibc, but glibc's limit is the much more
generous 1024.)
Also factor out the common errno-restoring idiom to reduce gotos.
Bug: 6702535
Change-Id: I555e66efffcf2c1b5a2873569e91489156efca42
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '2d3e72336e76180fb00822386da4f14203d117ce':
Revert "Revert "Pull the pthread_key_t functions out of pthread.c.""
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '09e89c3ced51d846e13c2508fbb6812bb61475cd':
Pull the pthread_key_t functions out of pthread.c.
This was originally motivated by noticing that we were setting the
wrong bits for the well-known tls entries. That was a harmless bug
because none of the well-known tls entries has a destructor, but
it's best not to leave land mines lying around.
Also add some missing POSIX constants, a new test, and fix
pthread_key_create's return value when we hit the limit.
Change-Id: Ife26ea2f4b40865308e8410ec803b20bcc3e0ed1
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '9a9bb243b50be5e3910b8edad72327bc216e72d0':
Switch to using AT_RANDOM for the stack guards.
There's now only one place where we deal with this stuff, it only needs to
be parsed once by the dynamic linker (rather than by each recipient), and it's
now easier for us to get hold of auxv data early on.
Change-Id: I6314224257c736547aac2e2a650e66f2ea53bef5
name server addresses are read from the dns
cache associated wih the interface on which
the request shall be done.
processes which has requested to issue dns request
using specific interface are now proxied to netd.
added methods to attach/detach a process to a specific
dns cache/interface.
added getaddrinfoforinface method which takes an
interface as an argument.
bug:4815099
bug:5465296
Change-Id: I7a8fe1980cdf99d4d296ddc5c6411f0c72162263
# Via Android Git Automerger (1) and others
* commit '4e9d9e4df8309493e65c5ca15513644361fda314':
Switch x86 syscall stubs over to the ENTER/END style of the ARM stubs.
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit 'd4187efd7f9c30ffaff4738862e3d08be27a52e9':
Switch x86 syscall stubs over to the ENTER/END style of the ARM stubs.
Also update the x86 asm.h to support this; we need it for libm assembler
anyway.
Also clean up the _FBSDID hack in <sys/cdefs.h>.
Change-Id: Iababd977b8110ec022bf7c93f4d62ece47630e7c
# Via Android Git Automerger (1) and others
* commit '16444e0f85585508e74d6166b8bbb4acb19738a0':
Don't claim there were no leaks if we weren't even checking.
# Via Elliott Hughes (1) and Gerrit Code Review (1)
* commit '3dc6b57cf54b15a756551deeda33df5715e519bb':
Don't claim there were no leaks if we weren't even checking.
This brings us up to date with FreeBSD HEAD, fixes various bugs, unifies
the set of functions we support on ARM, MIPS, and x86, fixes "long double",
adds ISO C99 support, and adds basic unit tests.
It turns out that our "long double" functions have always been broken
for non-normal numbers. This patch fixes that by not using the upstream
implementations and just forwarding to the regular "double" implementation
instead (since "long double" on Android is just "double" anyway, which is
what BSD doesn't support).
All the tests pass on ARM, MIPS, and x86, plus glibc on x86-64.
Bug: 3169850
Bug: 8012787
Bug: https://code.google.com/p/android/issues/detail?id=6697
Change-Id: If0c343030959c24bfc50d4d21c9530052c581837
Otherwise people trying to use this are left wondering "did I not leak, or did
the leak checking code not get called when I exited?".
Change-Id: If79b225f8a2e24dd69aba1fb836bf9e81bb00efe
