OSDN Git Service
Vladimir Marko [Thu, 4 Aug 2016 14:44:10 +0000 (14:44 +0000)]
Merge "Change suspend entrypoint to save all registers."
Vladimir Marko [Thu, 28 Jul 2016 11:01:51 +0000 (12:01 +0100)]
Change suspend entrypoint to save all registers.
We avoid the need to save/restore registers in slow paths
and get significant code size savings. On Nexus 9, AOSP:
- 32-bit boot.oat: -1.4MiB (-1.9%)
- 64-bit boot.oat: -2.0MiB (-2.3%)
- other 32-bit oat files in dalvik-cache: -200KiB (-1.7%)
- other 64-bit oat files in dalvik-cache: -2.3MiB (-2.1%)
Test: Run ART test suite on host and Nexus 9 with gc stress.
Bug:
30212852
Change-Id: I7015afc1e7d30341618c9200a3dc9ae277afd134
Treehugger Robot [Thu, 4 Aug 2016 13:44:17 +0000 (13:44 +0000)]
Merge "Update known buildbot failures for ProcessBuilderTest"
Tobias Thierer [Wed, 3 Aug 2016 18:12:37 +0000 (19:12 +0100)]
Update known buildbot failures for ProcessBuilderTest
Also updated the comment string based on our latest understanding.
This is a native crash. Both of the affected tests as well as the
other tests in ProcessBuilderTest run 'sh' as a child process.
Only one of the tests that are affected to far accesses /proc/,
which had previously been suspected as related to the cause.
The statement that this is a native crash is based on there not
being a relevant stacktrace in logcat.
A 'sh' related native crash failure occurs in the logs but that
is not specific to these tests so it's not clear if it is related.
If not, it's not clear why the additional native crash doesn't
show up in the logs.
Test: Nothing
Bug:
30657148
Change-Id: I8347edab3cacd060baf1087a1939926f70fdd349
Tobias Thierer [Thu, 4 Aug 2016 10:51:13 +0000 (10:51 +0000)]
Merge "Note known failures for TCP_USER_TIMEOUT tests on host"
Narayan Kamath [Thu, 4 Aug 2016 09:18:08 +0000 (09:18 +0000)]
Merge "libcore_failures: Remove obsolete entries."
Narayan Kamath [Wed, 3 Aug 2016 09:13:22 +0000 (10:13 +0100)]
libcore_failures: Remove obsolete entries.
These tests were failing with the OpenJDK switch, but have since
been fixed.
bug:
26326992
test: art/tools/run-libcore-tests.sh --mode=host --variant=X32
Change-Id: I9210d3fd428e0e462d95c2a620fe014e26bb942c
Treehugger Robot [Thu, 4 Aug 2016 03:35:08 +0000 (03:35 +0000)]
Merge "Fix CC table lookup tests"
Mathieu Chartier [Thu, 4 Aug 2016 01:51:34 +0000 (18:51 -0700)]
Fix CC table lookup tests
Only set the mark bits for baker barrier, not sure why this
doesn't work for table lookup barrier yet.
Bug:
30162165
Test: test-art-host with CC table lookup
Change-Id: I4309ac22bfc0f8a5c9bf081365e3e38da43eee30
Treehugger Robot [Wed, 3 Aug 2016 23:40:55 +0000 (23:40 +0000)]
Merge "Change one read barrier bit to mark bit"
Mathieu Chartier [Fri, 29 Jul 2016 01:08:51 +0000 (18:08 -0700)]
Change one read barrier bit to mark bit
Optimization to help slow path performance. When the GC marks an
object through the read barrier slow path. The GC sets the mark bit
in the lock word of that reference. This bit is checked from the
assembly entrypoint the common case is that it is set. If the bit is
set, the read barrier knows the object is already marked and there is
no work to do.
To prevent dirty pages in zygote and image, the bit is set by the
image writer and zygote space creation.
EAAC score (lower is better):
N9: 777 -> 700 (average 31 of runs)
N6P (960000 mhz): 1737.48 -> 1442.31 (average of 25 runs)
Bug:
30162165
Bug:
12687968
Test: N9, N6P booting, test-art-host, test-art-target all with CC
Change-Id: Iae0cacfae221e33151d3c0ab65338d1c822ab63d
Treehugger Robot [Wed, 3 Aug 2016 22:21:08 +0000 (22:21 +0000)]
Merge "Use ArrayInstance.getLength method."
Roland Levillain [Wed, 3 Aug 2016 17:56:02 +0000 (17:56 +0000)]
Merge "ARM64: Use the zero register in the parallel-move resolver."
Tobias Thierer [Wed, 3 Aug 2016 17:33:33 +0000 (18:33 +0100)]
Note known failures for TCP_USER_TIMEOUT tests on host
TCP_USER_TIMEOUT is not currently defined on the prebuilt tcp.h (at least
not on x86-64), so its value at the non-native level remains 0, breaking
tests that rely on it.
Devices kernels and bionic support it, therefore these tests are only
expected to break on host.
Test: Nothing. Buildbot will let us know soon enough if the expected
failures do not match reality.
Bug:
30402085
Change-Id: Iacaf63ab341c6e4a41c7d21fb3064d58ada94800
Treehugger Robot [Wed, 3 Aug 2016 17:08:51 +0000 (17:08 +0000)]
Merge "Fix constant type"
Roland Levillain [Wed, 3 Aug 2016 17:03:25 +0000 (17:03 +0000)]
Merge "Fix an assertion in the non-Baker read barrier ARM64 slow path."
Richard Uhler [Wed, 3 Aug 2016 16:26:35 +0000 (09:26 -0700)]
Use ArrayInstance.getLength method.
Now that perflib provides it as a better way to get the length of an
array instance.
Change-Id: I6bf8bb5541229e16f36fa8ede0a07cd0eb7b931d
Test: m ahat-test
Bug:
30481345
David Sehr [Wed, 3 Aug 2016 16:05:20 +0000 (09:05 -0700)]
Fix constant type
Remove the unnecessary cast to long for DCHECK, a nit that was omitted
from the last CL, https://android-review.googlesource.com/#/c/248243/.
Bug: b/
28856653
Test: test-art-host-gtest-dex_file_test
Change-Id: I8ca18d19cbac61e40ddfbe32729aef5c20347174
Treehugger Robot [Wed, 3 Aug 2016 16:09:45 +0000 (16:09 +0000)]
Merge "ART: Extend run-test timeout logging time"
Nicolas Geoffray [Wed, 3 Aug 2016 15:21:40 +0000 (15:21 +0000)]
Merge "Scope the compilation and Create the ProfilingInfo in the loop."
Roland Levillain [Wed, 3 Aug 2016 14:02:10 +0000 (15:02 +0100)]
Fix an assertion in the non-Baker read barrier ARM64 slow path.
Test: m ART_USE_READ_BARRIER=true ART_READ_BARRIER_TYPE=TABLELOOKUP test-art-target (ARM64).
Bug:
30563701
Change-Id: Ib7969bd78e9c14cc2be9d5527d80385486b4f409
Alexandre Rames [Tue, 2 Aug 2016 12:45:28 +0000 (13:45 +0100)]
ARM64: Use the zero register in the parallel-move resolver.
When moving zero to a stack slot, this will change
mov temp, #0
str temp, [...]
into
str zr, [...]
Change-Id: I2211b00d70f3fa0a02e781c90198757290b2bf89
Treehugger Robot [Tue, 2 Aug 2016 22:41:39 +0000 (22:41 +0000)]
Merge "Return error directly when uncompressed length is 0"
ganxiaolin [Mon, 18 Jul 2016 03:21:44 +0000 (11:21 +0800)]
Return error directly when uncompressed length is 0
If uncompressed length is 0, in function MemMap::MapAnonymous, it will generate a MemMap instance whose base_begin is nullptr.
Then, when OpenMemory, it will create a DexFile instance.
At DexFile::DexFile constructor, header_ is asigned to 0, then header_->string_ids_off_ will cause crash.
Bug: b/
28856653
Test: test-art-host-gtest-dex_file_test
Signed-off-by: ganxiaolin <ganxiaolin@xiaomi.com>
Change-Id: Id37f7629f4646cbc385ef054cb83b15be4c59b00
Nicolas Geoffray [Tue, 2 Aug 2016 17:49:25 +0000 (18:49 +0100)]
Scope the compilation and Create the ProfilingInfo in the loop.
We need to release the mutator lock to avoid being in a deadlock.
Also allocating the profiling just before compiling in the unlickely
event that it got collected after releasing the mutator lock.
bug:
30351473
Change-Id: I98dba646536d857152949b92f6c69ff9a03cf783
Nicolas Geoffray [Tue, 2 Aug 2016 20:18:27 +0000 (20:18 +0000)]
Merge "Improve the graph visualizer's output for constant locations."
Nicolas Geoffray [Tue, 2 Aug 2016 20:06:39 +0000 (20:06 +0000)]
Merge "ARM64: Make the VIXL macro assembler part of ART ARM64's assembler."
Andreas Gampe [Tue, 2 Aug 2016 04:53:57 +0000 (21:53 -0700)]
ART: Extend run-test timeout logging time
Double the time a test is allowed to attempt logging after it
timed out to two minutes.
Also change from using nested timeouts to using "-k" in general,
not just for host dex2oat.
Bug:
30351473
Test: m test-art-host
Change-Id: I134c7b220f69eaab7ea430a5c7ce45d77b091996
Dan Willemsen [Tue, 2 Aug 2016 17:41:59 +0000 (17:41 +0000)]
Merge "Split art-run-tests between building and packaging"
Vladimir Marko [Tue, 2 Aug 2016 16:59:35 +0000 (16:59 +0000)]
Merge "Align method code rather than method header in oat files."
Treehugger Robot [Tue, 2 Aug 2016 16:52:14 +0000 (16:52 +0000)]
Merge "ART: Fix Mips Thread offset checks"
Alexandre Rames [Tue, 2 Aug 2016 12:45:28 +0000 (13:45 +0100)]
ARM64: Make the VIXL macro assembler part of ART ARM64's assembler.
This avoids a dynamic allocation of the VIXL macro assembler.
Change-Id: I4cd62678d0978f1ad6f32ea0ce7279e09152be38
Andreas Gampe [Tue, 2 Aug 2016 15:44:20 +0000 (08:44 -0700)]
ART: Fix Mips Thread offset checks
Follow-up to commit
542451cc546779f5c67840e105c51205a1b0a8fd.
Bug:
30373134
Test: m test-art-host
Change-Id: I1d0202896e683cb20db30e8aac1c5ebcacc96467
Alexandre Rames [Tue, 2 Aug 2016 12:45:28 +0000 (13:45 +0100)]
Improve the graph visualizer's output for constant locations.
Change-Id: I423fb378ee61fb53c3b328fc74f4e95cdef0992a
Narayan Kamath [Tue, 2 Aug 2016 12:36:14 +0000 (12:36 +0000)]
Merge "DexFile: Add basic support for v38 dex files."
Vladimir Marko [Tue, 2 Aug 2016 12:31:10 +0000 (12:31 +0000)]
Merge "ARM: Embed 0.0 in VCMP."
Vladimir Marko [Mon, 1 Aug 2016 16:41:45 +0000 (17:41 +0100)]
ARM: Embed 0.0 in VCMP.
Test: Run ART test suite on Nexus 5.
Change-Id: I5cbbd98c4d64a4d9213e27adcae929ead5099a39
Vladimir Marko [Mon, 1 Aug 2016 15:33:16 +0000 (16:33 +0100)]
Align method code rather than method header in oat files.
This has always been the intent. For example, the function
name CompiledMethod::AlignCode() shows this intent.
Test: Run ART test suite on host and Nexus 9.
Change-Id: I11ae8963fc537fee853fe82e3aca22e77907eae7
Narayan Kamath [Mon, 1 Aug 2016 13:20:31 +0000 (14:20 +0100)]
DexFile: Add basic support for v38 dex files.
Makes sure they aren't rejected, essentially. Also adds positive
and negative test cases in dex_file_test.
bug:
30550796
Change-Id: I8c95055a22f8a1435868c8bf3497f9641b7bb900
test: make test-art-host
Narayan Kamath [Tue, 2 Aug 2016 08:09:35 +0000 (08:09 +0000)]
Merge "dex_instruction: remove unused parameter from instruction definitions."
Narayan Kamath [Mon, 1 Aug 2016 16:32:37 +0000 (17:32 +0100)]
dex_instruction: remove unused parameter from instruction definitions.
bug:
30550796
test: make test-art-host
Change-Id: Ib80a2d5c548578cefb0552b597e1d50b813ce8ef
Dan Willemsen [Tue, 2 Aug 2016 05:58:31 +0000 (22:58 -0700)]
Split art-run-tests between building and packaging
We remove the 'obj/PACKAGING' directory during an installclean. Instead
of forcing a rebuild of all of these tests after an installclean, build
them in a different obj directory, then copy them into the PACKAGING
directory to be gathered together by LOCAL_PICKUP_FILES.
The built directory isn't used directly for LOCAL_PICKUP_FILES, since we
don't clean up tests that are no longer defined, so an installclean can
fix that. This matches the behavior of modules installed into the system
image, etc.
This brings my build time for "m -j55 tests" from 3m30s to 35s after
"m installclean".
The total size of these binaries are small (~10MB), so the extra copy is
insignificant.
Change-Id: I20f1c09df25b70df8c7f86167480b3a39e99642a
Treehugger Robot [Tue, 2 Aug 2016 03:47:35 +0000 (03:47 +0000)]
Merge "ART: Convert pointer size to enum"
Nicolas Geoffray [Tue, 2 Aug 2016 02:07:19 +0000 (02:07 +0000)]
Merge "Revert "cpp-define-generator: "make art-update-cpp-defines" to regen asm_support""
Nicolas Geoffray [Tue, 2 Aug 2016 02:07:07 +0000 (02:07 +0000)]
Revert "cpp-define-generator: "make art-update-cpp-defines" to regen asm_support"
Reverting to unbreak golem builds.
This reverts commit
3c92b129047b12f0887a2068b3800d7a7dcb7d78.
Change-Id: I78ef481f57fbcd3b3a9f4802dd6c21dabb186856
Andreas Gampe [Tue, 26 Jul 2016 16:02:02 +0000 (09:02 -0700)]
ART: Convert pointer size to enum
Move away from size_t to dedicated enum (class).
Bug:
30373134
Bug:
30419309
Test: m test-art-host
Change-Id: Id453c330f1065012e7d4f9fc24ac477cc9bb9269
Treehugger Robot [Tue, 2 Aug 2016 00:32:16 +0000 (00:32 +0000)]
Merge "Fixed bug in disassembly of roundss/roundsd"
Aart Bik [Mon, 1 Aug 2016 22:55:36 +0000 (15:55 -0700)]
Fixed bug in disassembly of roundss/roundsd
Rationale:
These instructions should be marked as load, so that, using
Intel syntax, destination (xmm0) appears at left hand side, as in
roundss xmm0, xmm1
and not the other way around. First I suspected a bug in the
encoding (hence the test) and even the register allocator, but
since the code behaved correctly, only disassembly was really wrong.
Test: disassembler_x86_test (but nothing for actual disassembly)
BUG=
26327751
Change-Id: I060ef57f4d5a64cdc04b97ae8a799d1c0d22da05
Treehugger Robot [Mon, 1 Aug 2016 22:45:01 +0000 (22:45 +0000)]
Merge "Added direct memory operand support for comiss/sd on x86."
Treehugger Robot [Mon, 1 Aug 2016 22:39:15 +0000 (22:39 +0000)]
Merge "libsigchain: intercept bsd_signal on lp32"
Dimitry Ivanov [Mon, 1 Aug 2016 18:19:03 +0000 (11:19 -0700)]
libsigchain: intercept bsd_signal on lp32
bsd_signal is a deprecated synonym for the signal. Nevertheless
there are apps and libraries calling this function.
libsigchain needs to intercept these calls and handle them the same
way it handles signal() call.
Bug: http://b/
30562229
Test: readelf --dyn-sym app_process32 and check that bsd_signal is exported
readelf --dyn-sym app_process64 and check that bsd_signal is not exported
Change-Id: I3aa41632bb015568d8524a82f1c97e4443ec0d6c
Aart Bik [Mon, 1 Aug 2016 21:11:20 +0000 (14:11 -0700)]
Added direct memory operand support for comiss/sd on x86.
Rationale:
Memory operands allows for shorter instruction sequences.
Something that may be used e.g. the upcoming implementation
of the round intrinsic.
Bug=
26327751
Change-Id: Ifd976ed86917de1879fa036cbbbff5edf4fb741d
Test: assembler_x86_test
Nicolas Geoffray [Mon, 1 Aug 2016 19:45:05 +0000 (19:45 +0000)]
Merge "Put the deletion of profiling info under a GC critical section."
Hiroshi Yamauchi [Mon, 1 Aug 2016 19:05:43 +0000 (19:05 +0000)]
Merge "Revert "Revert "Scan immune spaces using mod union tables."""
Colin Cross [Mon, 1 Aug 2016 18:14:07 +0000 (18:14 +0000)]
Merge "ART: remove gcc cruft from the makefiles"
Treehugger Robot [Mon, 1 Aug 2016 17:46:49 +0000 (17:46 +0000)]
Merge "cpp-define-generator: "make art-update-cpp-defines" to regen asm_support"
Przemyslaw Szczepaniak [Mon, 1 Aug 2016 10:54:37 +0000 (10:54 +0000)]
Merge "Fix art::ArchTest::FinalizeSetup not being called."
Przemyslaw Szczepaniak [Mon, 25 Jul 2016 16:31:06 +0000 (17:31 +0100)]
Fix art::ArchTest::FinalizeSetup not being called.
CommonRuntimeTestImpl::SetUp() is calling FinalizeSetup. Since
CommonRuntimeTestImpl::FinalizeSetup is not virtual (it's virtual
one step higher in class hierarchy in CommonRuntimeTestBase), overridden
version in ArchTest was never used. This is causing ArchTest to
fail with openjdk8u60 version of java.lang.reflect (it's calling
Class#getDeclaredField in clinit, which fails on 32 bit
version because of different pointer sizes [arch_test forces
64-bit ISA])
Bug:
28666126
Test: make -j 32 test-art-host-gtest
Change-Id: Ief370f5b18ef787575ac0f88ecbe17ebbf037542
David Brazdil [Mon, 1 Aug 2016 08:39:04 +0000 (08:39 +0000)]
Merge "ART: Fix RegTypeCache for instance field declaring class"
Nicolas Geoffray [Sat, 30 Jul 2016 21:49:11 +0000 (22:49 +0100)]
Put the deletion of profiling info under a GC critical section.
Otherwise the GC could see dangling pointers.
bug:
30033802
Change-Id: I839c4f6e537aba98769e99e4d75ff869c9ff13eb
Aart Bik [Sat, 30 Jul 2016 02:53:25 +0000 (02:53 +0000)]
Merge "Fix for interpreter crash on new instance of class"
Treehugger Robot [Sat, 30 Jul 2016 02:10:00 +0000 (02:10 +0000)]
Merge "Revert "Revert "Use try lock to fix class resolution race"""
Mathieu Chartier [Fri, 29 Jul 2016 23:26:01 +0000 (16:26 -0700)]
Revert "Revert "Use try lock to fix class resolution race""
Fix possible deadlock in EnsureResolved caused by interaction with
GC. Since we were sleeping while holding the mutator lock, it could
block thread suspension. This would deadlock if the thread that
had locked h_class is already suspended since we would spin forever
and not make progress.
Bug:
27417671
Bug:
30500547
Test: test-art-host ART_TEST_GC_STRESS=true
This reverts commit
69bf969c055c31a75d17ea92aeee756042678114.
Change-Id: Id8750df065dc3b9ef7dc874f2eb2cc2c58e5d1eb
Richard Uhler [Sat, 30 Jul 2016 00:10:18 +0000 (00:10 +0000)]
Merge "Create OatFileAssistant::OatFileInfo inner class."
Aart Bik [Tue, 26 Jul 2016 00:52:22 +0000 (17:52 -0700)]
Fix for interpreter crash on new instance of class
Rationale:
Fuzz testing found divergences between the compiler and interpreter
which turned out to be caused by calling instance on java.lang.Class
(this worked for compiler but crashed interpeter). Since RI does not
allow this construct, solution is to force interpreter in this
unlikely case and throw run time exception like RI. This fixes
two cases found with fuzz testing. CL also includes
fail-before/pass-after test.
Test: 600-verifier-fails
BUG=
29758098
Change-Id: Ie80f7758def44e6655d28fec4c10c34ffa0fd60b
Hiroshi Yamauchi [Fri, 29 Jul 2016 22:07:05 +0000 (15:07 -0700)]
Revert "Revert "Scan immune spaces using mod union tables.""
This reverts commit
ccf7c8a68972f3b7bfa38af0c6b792f712f83765.
Change-Id: Iee618d9ae78e1dcdea4a7ecbe9253821d4d4d015
Test: N9 boot, art tests, Ritz EAAC.
Bug:
29516968
Bug:
29517059
Bug:
12687968
Treehugger Robot [Fri, 29 Jul 2016 21:23:49 +0000 (21:23 +0000)]
Merge "Revert "Use try lock to fix class resolution race""
Mathieu Chartier [Fri, 29 Jul 2016 20:06:45 +0000 (20:06 +0000)]
Revert "Use try lock to fix class resolution race"
This reverts commit
a704eda0078989a73cac111ed309aca50d2e289b.
Bug:
27417671
Bug:
30500547
Change-Id: Ieea05236b9e61c722660cd9497c9d55d13ccd010
Treehugger Robot [Fri, 29 Jul 2016 18:57:26 +0000 (18:57 +0000)]
Merge "Add arena stack and priority queue type aliases"
Treehugger Robot [Fri, 29 Jul 2016 18:52:16 +0000 (18:52 +0000)]
Merge "Revert "Scan immune spaces using mod union tables.""
Colin Cross [Fri, 29 Jul 2016 17:57:57 +0000 (17:57 +0000)]
Merge "ART: remove ART_JIT makefile variable"
Hiroshi Yamauchi [Fri, 29 Jul 2016 17:41:01 +0000 (17:41 +0000)]
Revert "Scan immune spaces using mod union tables."
This reverts commit
8256da35feac0ba7589fcd4b692a4b9a936679a4.
Test failing.
Change-Id: I391dcbdba19c6f6313e06075b0a69ea406c7c50e
Matthew Gharrity [Fri, 29 Jul 2016 16:13:33 +0000 (09:13 -0700)]
Add arena stack and priority queue type aliases
These will be used in the graph coloring register allocator.
Test: m test-art-host
Change-Id: I4d3d063ae5a02e697f07fc8212f391b5347270e5
Igor Murashkin [Fri, 29 Jul 2016 16:19:44 +0000 (09:19 -0700)]
cpp-define-generator: "make art-update-cpp-defines" to regen asm_support
Change-Id: I1a550eda6c77f38e63bb1c8fb943541cbd3f9cdf
Nicolas Geoffray [Fri, 29 Jul 2016 14:28:44 +0000 (14:28 +0000)]
Merge "Pass the right class loader when inlining."
Colin Cross [Thu, 28 Jul 2016 23:01:42 +0000 (16:01 -0700)]
ART: remove gcc cruft from the makefiles
ART only builds with clang now, delete all logic to determine gcc
vs. clang, and move all LOCAL_CLANG_CFLAGS into LOCAL_CFLAGS.
Test: mma, make test-art-host, make test-art-target
Change-Id: I578615233ad3e71af18afb47bb11f3be8417216c
Colin Cross [Thu, 28 Jul 2016 23:03:13 +0000 (16:03 -0700)]
ART: remove ART_JIT makefile variable
The ART_JIT variable isn't used (and was never used?).
Test: mma, make test-art-host, make test-art-target
Change-Id: Ie6463c3c13fab43504f3c60fd5a15f696e2cf481
Richard Uhler [Tue, 19 Apr 2016 22:39:37 +0000 (15:39 -0700)]
Create OatFileAssistant::OatFileInfo inner class.
To share code reused for getting info about both the odex and oat files.
Test: oat file assistant tests.
Change-Id: Ifcc48e0ac481899285e39e3b02f3ade180e38728
Mathieu Chartier [Thu, 28 Jul 2016 22:27:34 +0000 (22:27 +0000)]
Merge "Avoid read barrier for IntArray::GetArrayClass"
Mathieu Chartier [Thu, 28 Jul 2016 18:02:38 +0000 (11:02 -0700)]
Avoid read barrier for IntArray::GetArrayClass
Changed the code use Mark instead of read barrier, this showed an
existing lock violation and possible deadlock which was fixed.
Prevent DCHECK failure from the assert.
Bug:
30469265
Test: test-art-host with CC
Change-Id: I275f953f06f6d13262043fc62eb88dca0356465a
Mathieu Chartier [Thu, 28 Jul 2016 18:12:56 +0000 (18:12 +0000)]
Merge "Use try lock to fix class resolution race"
Hiroshi Yamauchi [Thu, 28 Jul 2016 17:24:43 +0000 (17:24 +0000)]
Merge "Scan immune spaces using mod union tables."
Roland Levillain [Thu, 28 Jul 2016 16:54:49 +0000 (16:54 +0000)]
Merge "Disable CheckSwapUsage test with read barriers on x86."
Roland Levillain [Thu, 28 Jul 2016 16:20:19 +0000 (16:20 +0000)]
Merge "Remove two ReadBarrierMarkRegX entrypoints."
Nicolas Geoffray [Thu, 28 Jul 2016 02:49:14 +0000 (03:49 +0100)]
Pass the right class loader when inlining.
Otherwise, method and type resolution can resolve to the wrong
things and as a side effect update the dex cache with wrong data.
bug:
30403437
test:./art/test/run-test --host --jit --dev --no-prebuild 613
(cherry picked from commit
0a210d9b108c87c0e7c1d430a92ce6fc89790c95)
Change-Id: I8fdca96fccae079c8e99b8e86e7b6935acfce89d
David Brazdil [Wed, 27 Jul 2016 14:29:25 +0000 (15:29 +0100)]
ART: Fix RegTypeCache for instance field declaring class
For bytecode instruction:
iget(-*) ClassB->field:type
where the resolved field is actually declared in ClassA (ClassB
extends ClassA), MethodVerifier would create a RegType representing
ClassA but cache it under the descriptor "ClassB".
This is a bug but does not have any implications on correctness
because earlier resolution of ClassB (part of IGET handling) creates
another cache entry with the "ClassB" descriptor. Because the latter,
wrong entry is always imprecise (ClassA cannot be final because ClassB
extends it), either the earlier entry will be discovered or neither
of them will be a match.
Descriptor was replaced with the correct one and a DCHECK added when
creating a RegType to make sure the descriptor matches.
Bug:
30458218
Change-Id: I19e1bdd5dd79e5eac558122a06b9128d0a5c021f
Test: m test-art-host-run-test
Test: art/test/run-test --host 800-smali
Roland Levillain [Thu, 28 Jul 2016 15:37:28 +0000 (16:37 +0100)]
Disable CheckSwapUsage test with read barriers on x86.
This test does not produce the expected results on
some x86 systems when read barriers are enabled.
Also remove unused TEST_DISABLED_FOR_READ_BARRIER macro
definition.
Test: m test-art-host-gtest
Bug:
29259363
Bug:
12687968
Change-Id: If7c124226b32c10425725e62bb684150aba108ab
Vladimir Marko [Thu, 28 Jul 2016 14:10:54 +0000 (14:10 +0000)]
Merge "ARM64: Add comment to mterp int-to-long."
Tobias Thierer [Thu, 28 Jul 2016 13:49:20 +0000 (13:49 +0000)]
Merge "Add ProcessBuilderTest#testRedirectInherit() to known failures"
Vladimir Marko [Thu, 28 Jul 2016 13:04:18 +0000 (14:04 +0100)]
ARM64: Add comment to mterp int-to-long.
Test: Build for Nexus 9.
Change-Id: I649e71fc1cdc2a4a103159bf72ceeb005231e66f
Vladimir Marko [Thu, 28 Jul 2016 12:43:29 +0000 (12:43 +0000)]
Merge "MIPS32: Improve string and class loads"
Vladimir Marko [Thu, 28 Jul 2016 09:00:59 +0000 (09:00 +0000)]
Merge "ARM64: More mterp improvements."
Treehugger Robot [Thu, 28 Jul 2016 05:35:28 +0000 (05:35 +0000)]
Merge "ART: Use old operator<< for MemBarrierKind"
Andreas Gampe [Thu, 28 Jul 2016 04:25:41 +0000 (21:25 -0700)]
ART: Use old operator<< for MemBarrierKind
To fix assumptions in the tests.
Test: m test-art-host
Change-Id: Ie2e738524ee8f75a3c41730f6a1449dab490e535
Alexey Frunze [Tue, 19 Jul 2016 22:00:40 +0000 (15:00 -0700)]
MIPS32: Improve string and class loads
Tested:
- MIPS32 Android boots in QEMU
- test-art-host-gtest
- test-art-target-run-test-optimizing in QEMU, on CI20
- test-art-target-gtest on CI20
Change-Id: I70fd5d5267f8594c3b29d5a4ccf66b8ca8b09df3
Treehugger Robot [Thu, 28 Jul 2016 01:54:06 +0000 (01:54 +0000)]
Merge "ART: Delete old compiler_enums.h"
Mathieu Chartier [Thu, 28 Jul 2016 01:05:07 +0000 (01:05 +0000)]
Merge "Reduce unnecessary read barriers in GC"
Andreas Gampe [Thu, 28 Jul 2016 00:53:11 +0000 (17:53 -0700)]
ART: Delete old compiler_enums.h
Holdover from the Quick days. Move the two enums that are still
used closer to the actual users (and prune no longer used cases).
Test: m test-art-host
Change-Id: I88aa49961a54635788cafac570ddc3125aa38262
Hiroshi Yamauchi [Wed, 27 Jul 2016 21:08:19 +0000 (14:08 -0700)]
Scan immune spaces using mod union tables.
It's faster to scan the immune spaces using the mod union tables (gray
objects on dirty pages) than the live bitmaps (gray objects).
The immune space scan time goes from ~13ms down to ~10ms in the
testWidgetsContainerFling test on angler at 1 GHz.
Change-Id: I0914d7bdb50cbf8a49c859070a3501b48eb79f3e
Test: N9 boot, art tests, Ritz EAAC.
Bug:
29516968
Bug:
29517059
Bug:
12687968
Mathieu Chartier [Wed, 13 Jul 2016 16:53:35 +0000 (09:53 -0700)]
Use try lock to fix class resolution race
There was some possible deadlocks related to EnsureResolved caused by
acquiring an object lock.
Scenario:
Thread 1 acquires lock on obj1
Thread 1 begins to resolve / initialize class1
Thread 1 blocks since it sees that class1 is already being resolved and
gets preempted before it can acquire the object lock on class1
Thread 2 finishes resolving and initializing class1 and locks class1
Thread 2 blocks attempting to lock obj1
Thread 1 blocks attempting to lock class1
Deadlock
Fixed the deadlock by changing EnsureResolved to use a try lock for the
unresolved case.
Added a test.
Test: Device boot, test-art-host, monitor_test
Bug:
27417671
Change-Id: Ic6e1c3ca6f45490cf8a7bf8e137dee71ac83ff64