arm32 has two special APIs to find exidx exception handling info,
dl_unwind_find_exidx and __gnu_Unwind_Find_exdix. The two functions have
identical behavior and function prototypes. libgcc's arm32 unwinder calls
__gnu_Unwind_Find_exdix, whereas LLVM's libunwind previously called
__gnu_Unwind_Find_exdix, but switched to dl_unwind_find_exidx as a result
of three patches (D30306, D30681, D39468).
In Bionic, for dynamic linking, __gnu_Unwind_Find_exdix in libc.so calls
dl_unwind_find_exidx in libdl.so.
For static executables, though, __gnu_Unwind_Find_exdix in libc.a used the
__exidx_* symbols, while dl_unwind_find_exidx in libdl.a(libdl_static.o)
was a return-0 no-op.
To fix the LLVM unwinder, replace the no-op dl_unwind_find_exidx in
libdl.a with a real function in libc.a(exidx_static.o), and have the GNU
function call the dl function for more consistency with dynamic linking.
dl_iterate_phdr follows a similar pattern, where the function exists in
libc.a and libdl.so (not libc.so or libdl.a).
This change makes unwinding work with an updated libunwind_llvm on arm32,
and it helps to allow unwinding in static executables without libdl.a.
Bug: https://github.com/android/ndk/issues/1094
Bug: http://b/141485154
Test: NDK tests, bionic unit tests
Change-Id: Ieeeb9b39a0e28544e21f9afe6fe51ef10d7c828c
Scudo creates a large map on 64 bit which can cause a slow down trying
to get the usage stats for all maps. Since the test only really cares
about a small subset of maps, only get the usage stats for those maps.
Test: Ran unit tests on scudo and jemalloc based systems.
Change-Id: Iba3cff1487ca304083aac323a3971b9f939f5c11
As the first pass, the malloc_info for scudo doesn't return anything.
This feature is hardly ever used, so this seems fine for the first
revision.
Bug: 137795072
Test: Ran unit tests on a taimen using scudo and a taimen using jemalloc.
Change-Id: I675386ae17ba5505c27bd9da12c60242d795464a
At heart a revert of "Don't #define __ANDROID_NDK__ for the platform build!",
which was commit eb61430374.
The original change was insufficiently motivated, and meant that the
NDK -- not just the platform -- no longer defines __ANDROID_NDK__. Which
then broke at least building toybox with NDK r19.
Change-Id: Ic616688e4d17d25714a9ef381269d7431deac9b0
* changes:
libc: remove now-unused FORTIFY functions
fortify: remove last uses of __bos_trivially_not*
fortify(string): emit diagnostics regardless of API level
fortify(fcntl): emit diagnostics regardless of API level
As of I2037548cc2061e46c379931588194c21dfe234b4, these are no longer
used. Since they're new in R, we can remove them instead of keeping
backwards compat 'forever'. Take that opportunity now.
Bug: 141267932
Test: TreeHugger
Change-Id: I13f94cdcff6e75ad19b964be76445f113f79559b
Since we're using the gt/ge ones a lot now, having `not` versions
probably just adds to confusion. Swap out their remaining uses and
delete them.
Bug: 141267932
Test: m checkbuild on internal-master
Change-Id: I2107ae65007a4995e4fa23371fefe4db7547f43b
HWASan-instrumented code needs TLS_SLOT_SANITIZER set up to run, and
that is not done until the new thread calls __hwasan_thread_enter. Block
all signals until that time to prevent hwasan-instrumented signal
handlers running (and crashing) on the new thread.
Bug: 141893397
Test: seq 0 10000000 | xargs -n 1 -P 200 adb shell am instrument \
-w -r -e command grant-all \
com.android.permissionutils/.PermissionInstrumentation
(cherry picked from commit d181585dd5)
Change-Id: Id65fae836edcacdf057327ccf16cf0b5e0f9474a
The bionic libs are now restricted to be in the runtime APEX and the
platform (for bootstrapping). It can still be referenced from other
APEXes but can't be included there.
Bug: 139870423
Test: m
Change-Id: I7f99eef27ccf75844ca5c9a7ea866496841b738f
The NDK doesn't support anything older than API level 16, so remove some
more clutter.
Test: builds
Change-Id: If257a27841396af001b089b7ae0fbd8c3e0128e4
The run-time of dl_iterate_phdr can be noticeable for exception-handling.
e.g. I tested extest on the GitHub issue by:
- disabling the dlpi_subs/dlpi_adds-based caching in libgcc
- adding a weaker "load base hint" caching in libgcc that doesn't need
the dlpi_subs/dlpi_adds fields, but still has to iterate over every
module to validate a cache hit
extest throws 10000 exceptions, and I saw a regression from ~1550ms
runtime on Q to about ~1950ms on master. This CL reduces the regression to
about ~1700ms.
Bug: https://github.com/android/ndk/issues/1062
Test: bionic unit tests
Change-Id: I099e97e1a20f5b2aa6737789e49d965170eb85a8
The NDK only supports >= 16, so remove anything older than that to avoid
giving the misleading impression that such old targets are still
supported.
(This change doesn't touch <unistd.h>. I'll follow up with that once the
outstanding FORTIFY changes to that file are in.)
Test: builds
Change-Id: I6cc6ecdb99fe228a4afa71f78e5fd45309ba9786
Refactor some of the definitions to separate out all of the native
allocator pieces into one place that can be changed easily. This
should fix a few static libraries that appear to have accidentally
included jemalloc wrapper functions. For example, I verified that
libc_nomalloc.a no longer has references to any je_XXX functions.
Modify the bionic_libc_platform_headers to not include any libraries.
If this isn't updated, soong thinks there are cycles when used by
libscudo.
To enable scudo, change the libc_native_allocator_defaults defaults
from libc_jemalloc5_defaults to libc_scudo_defaults and comment out
the defaults: ["libc_scudo_wrapper_defaults"], line for the shared
library libc_scudo.
To do a final switch to scudo, it will be necessary to clean up
some code in other parts of the tree, but this allows a single cl
to enable or disable.
Bug: 137795072
Test: Builds with jemalloc or scudo with a small change.
Test: Ran bionic unit tests.
Change-Id: I07bb5432a0d2b2f405f92412e8d04fb9c9e51b31
Merged-In: I07bb5432a0d2b2f405f92412e8d04fb9c9e51b31
(cherry picked from commit ccff1b19ef)
The tables in the BSD tolower/toupper are slower for ASCII than just
doing the bit twiddling.
We can't actually remove the tables on LP32, so move them into the
"cruft" we keep around for backwards compatibility (but remove them for
LP64 where they were never exposed).
I noticed that the new bit-twiddling tolower(3) was performing better
on arm64 than toupper(3). The 0xdf constant was requiring an extra MOV,
and there isn't a BIC that takes an immediate value. Since we've already
done the comparison to check that we're in the right range (where the
bit is always set), though, we can EOR 0x20 to get the same result as
the missing BIC 0x20 in just one instruction.
I've applied that same optimization to towupper(3) too.
Before:
BM_ctype_tolower_n 3.30 ns 3.30 ns 212353035
BM_ctype_tolower_y 3.31 ns 3.30 ns 211234204
BM_ctype_toupper_n 3.30 ns 3.29 ns 214161246
BM_ctype_toupper_y 3.29 ns 3.28 ns 207643473
BM_wctype_towupper_ascii_n 3.53 ns 3.53 ns 195944444
BM_wctype_towupper_ascii_y 3.48 ns 3.48 ns 199233248
After:
BM_ctype_tolower_n 2.93 ns 2.92 ns 242373703
BM_ctype_tolower_y 2.88 ns 2.87 ns 245365309
BM_ctype_toupper_n 2.93 ns 2.93 ns 243049353
BM_ctype_toupper_y 2.89 ns 2.89 ns 245072521
BM_wctype_towupper_ascii_n 3.34 ns 3.33 ns 212951912
BM_wctype_towupper_ascii_y 3.29 ns 3.29 ns 214651254
(Why do both the "y" and "n" variants speed up with the EOR
change? Because the compiler transforms the code so that we
unconditionally do the bit twiddling and then use CSEL to decide whether
or not to actually use the result.)
We also save 1028 bytes of data in the LP64 libc.so.
Test: ran the bionic benchmarks and tests
Change-Id: I7829339f8cb89a58efe539c2a01c51807413aa2d
