OSDN Git Service

android-x86/dalvik.git
11 years agoUpdate houdini support ics-x86 android-x86-4.0-r1
Chih-Wei Huang [Thu, 6 Dec 2012 07:05:31 +0000 (15:05 +0800)]
Update houdini support

The patch requires new houdini for ICS.

Modified from JB patches of Daniel Fages <dfages@android-x86.org>.

11 years agoX86 trace JIT compiler support
Dong-Yuan Chen [Tue, 3 Jul 2012 20:13:07 +0000 (13:13 -0700)]
X86 trace JIT compiler support

This patch provides a fully functional x86 trace JIT compiler for Dalvik
VM. It is built on top of the existing x86 fast interpreter
with bug fixes and needed extension to support trace JIT interface. The
x86 trace JIT code generator was developed independent of the existing
template-based code generator and thus does not share exactly the same
infrastructure. Included in this patch are:

* Deprecated and removed the x86-atom fast interpreter that is no
  longer functional since ICS.
* Augmented x86 fast interpreter to provide interfaces for x86 trace JIT
  compiler.
* Added x86 trace JIT code generator with full JDWP debugging support.
* Method JIT and self-verification mode are not supported.

The x86 code generator uses the x86 instruction encoder/decoder library
from the Apache Harmony project. Additional wrapper extension and bug
fixes were added to support the x86 trace JIT code generator. The x86
instruction encoder/decoder is embedded inside the x86 code generator
under the libenc subdirectory.

Change-Id: I241113681963a16c13a3562390813cbaaa6eedf0
Signed-off-by: Dong-Yuan Chen <dong-yuan.chen@intel.com>
Signed-off-by: Yixin Shou <yixin.shou@intel.com>
Signed-off-by: Johnnie Birch <johnnie.l.birch.jr@intel.com>
Signed-off-by: Udayan <udayan.banerji@intel.com>
Signed-off-by: Sushma Kyasaralli Thimmappa <sushma.kyasaralli.thimmappa@intel.com>
Signed-off-by: Bijoy Jose <bijoy.a.jose@intel.com>
Signed-off-by: Razvan A Lupusoru <razvan.a.lupusoru@intel.com>
Signed-off-by: Tim Hartley <timothy.d.hartley@intel.com>
11 years agoFix a possible off-by-one in the verifier.
Elliott Hughes [Fri, 29 Jun 2012 23:45:06 +0000 (16:45 -0700)]
Fix a possible off-by-one in the verifier.

Bug: http://code.google.com/p/android/issues/detail?id=10863
Change-Id: I604f9a5ea08b6a6934b3ba522e1fda0bb738f81c

11 years agoExport jniRegisterSystemMethods() by C linkage
Chih-Wei Huang [Tue, 26 Jun 2012 09:58:36 +0000 (17:58 +0800)]
Export jniRegisterSystemMethods() by C linkage

It allows libdvm_houdini.so to be loaded correctly.

11 years agoEnable libhoudini support
Daniel Fages [Tue, 26 Jun 2012 09:55:51 +0000 (17:55 +0800)]
Enable libhoudini support

11 years agoMake dex merging stateless.
Xavier Ducrohet [Mon, 11 Jun 2012 22:17:31 +0000 (15:17 -0700)]
Make dex merging stateless.

This is required to run dx in Eclipse which doesn't unload the dx
library after each run.

(cherry-pick of 7aa5ce7e990dc3766eba97cd0932b62e4de21503.)

Change-Id: I7a69a3015d448ddd5558c307cd01346156cbc739

11 years agoSupport debug info in dexmerge.
Elliott Hughes [Thu, 7 Jun 2012 23:47:38 +0000 (16:47 -0700)]
Support debug info in dexmerge.

Bug: 4090053

(cherry-pick of bc23c4f3ebaefebb3f1be7732767631f91e165ea.)

Change-Id: I1108933fc03330ff91be3a2edef8b4966977dcd7
Signed-off-by: Jesse Wilson <jesse@swank.ca>
11 years agoFixing a crash when doing lock profiling
Mattias Petersson [Tue, 5 Jun 2012 08:43:01 +0000 (10:43 +0200)]
Fixing a crash when doing lock profiling

This is a fix for a crash that can happen when logging Contention
events. This logging is performed when lock profiling is enabled.
This is by default enabled on userdebug builds. The crash happened
when a thread was being destroyed. When a thread is being
destroyed it is normal that the frame depth is zero, and thus the
current frame is null. logContentionEvent() requires that the
current frame is not null, or it will crash.

The fix is to check if the current frame is null.

Change-Id: I4c2b9ad94b663398645497fdffa1ec6f7ea86a51

11 years agoRunning rebuild.sh found a few changes to be updated to vm/mterp/out/InterpAsm-mips...
Raghu Gandham [Mon, 21 May 2012 19:31:11 +0000 (12:31 -0700)]
Running rebuild.sh found a few changes to be updated to vm/mterp/out/InterpAsm-mips.S and vm/mterp/out/InterpC-mips.cpp

11 years agoFix: Issue 26627: Dalvik Heap Trim timer value after GC
SangWook Han [Fri, 11 May 2012 11:45:33 +0000 (20:45 +0900)]
Fix: Issue 26627: Dalvik Heap Trim timer value after GC

dvmRelativeCondWait need millisecods not seconds.

With this fix, trim will begins 5 seconds after concurrent GC

Bug: http://code.google.com/p/android/issues/detail?id=26627
Change-Id: I3802b48e345725fb78e4ef2c999dd243ef30a138

11 years agoPad the 8 and 16 bit Jvalue members
Chris Dearman [Tue, 31 Jan 2012 02:10:35 +0000 (18:10 -0800)]
Pad the 8 and 16 bit Jvalue members

This allows them to be consistently accessed as either 8-bit,
16-bit or 32-bit values in bigendian and little-endian mode.

Change-Id: I1be5519f3666a3553b404368075f5a7b5d9e6b32
Signed-off-by: Chris Dearman <chris@mips.com>
Signed-off-by: Raghu Gandham <raghu@mips.com>
11 years agoDalvik fast interpreter support and JIT implementation
Raghu Gandham [Wed, 2 May 2012 21:27:16 +0000 (14:27 -0700)]
Dalvik fast interpreter support and JIT implementation

Change-Id: I9bb4f6875b7061d3ffaee73f204026cb8ba3ed39
Signed-off-by: Raghu Gandham <raghu@mips.com>
Signed-off-by: Chris Dearman <chris@mips.com>
Signed-off-by: Douglas Leung <douglas@mips.com>
Signed-off-by: Don Padgett <don@mips.com>
11 years agoFix a bug in dvmQuasiAtomicsShutdown.
Elliott Hughes [Wed, 2 May 2012 23:07:36 +0000 (16:07 -0700)]
Fix a bug in dvmQuasiAtomicsShutdown.

Spotted visually when merging to an internal branch.

Change-Id: Ic130608bfa46036cec0684cd50706178cfefd2a0

11 years agoSupport architectures that require correctly aligned doubles
Raghu Gandham [Tue, 31 Jan 2012 02:25:05 +0000 (18:25 -0800)]
Support architectures that require correctly aligned doubles

CTS: tests.api.java.lang.reflect.MethodTest#test_invokeLjava_lang_Object$Ljava_lang_Object

Change-Id: I93c391884d802b8e69fe3e1ced55a2f2c7866d16
Signed-off-by: Nikola Kuzmanovic <nikola.kuzmanovic@rt-rk.com>
Signed-off-by: Chris Dearman <chris@mips.com>
11 years agoSimplify the quasi-atomic stuff and add MIPS support.
Elliott Hughes [Wed, 2 May 2012 00:24:51 +0000 (17:24 -0700)]
Simplify the quasi-atomic stuff and add MIPS support.

Change-Id: Iac896c03ef707c998cc714628adea0255f21c035

11 years agoUse old frame for dvmReportPreNativeInvoke/dvmReportPostNativeInvoke
Chris Dearman [Tue, 31 Jan 2012 01:44:28 +0000 (17:44 -0800)]
Use old frame for dvmReportPreNativeInvoke/dvmReportPostNativeInvoke

The portable and standard interpreters were using different values.
The portable interpeter is correct...

Change-Id: I8669dfbc979f161666e6b7f63aae16f48612a0c6
Signed-off-by: Chris Dearman <chris@mips.com>
11 years agoRestore fp after calling JNI method in allstubs interpreter
Chris Dearman [Tue, 31 Jan 2012 01:44:28 +0000 (17:44 -0800)]
Restore fp after calling JNI method in allstubs interpreter

The allstubs interpreter uses self->interpSave.curFrame as fp.
The invokeMethod native code sets self->interpSave.curFrame directly
effectively updating fp, so restore fp after the call has completed.

Change-Id: I0ffb15b71a48c2780251bbcd66e551bb0a52fe94
Signed-off-by: Chris Dearman <chris@mips.com>
11 years agoFix the warning emitted by codesourcery compiler 4.5.2
Raghu Gandham [Tue, 31 Jan 2012 01:44:28 +0000 (17:44 -0800)]
Fix the warning emitted by codesourcery compiler 4.5.2

Change-Id: Ib2b8e87afff7af56281d83961edf79c17cbcf940

11 years agoFix instruction logging messages
Chris Dearman [Tue, 31 Jan 2012 01:44:28 +0000 (17:44 -0800)]
Fix instruction logging messages

Change-Id: Ifc7de30ea1c1221f7838523a5ebd269eec622e03
Signed-off-by: Chris Dearman <chris@mips.com>
11 years agoFix the portable interpreter build
Chris Dearman [Wed, 1 Feb 2012 01:19:49 +0000 (17:19 -0800)]
Fix the portable interpreter build

Change-Id: Ibf08d50c7a01a4c30559dd2ee9511c28fb1365a5
Signed-off-by: Chris Dearman <chris@mips.com>
11 years agoFixed bug in x86-atom fast interpreter.
Johnnie Birch [Fri, 25 Mar 2011 20:24:03 +0000 (13:24 -0700)]
Fixed bug in x86-atom fast interpreter.

Adds guard for potential null pointer. Adds guard for potential null
pointer when checking for debugger active on return from method.

Change-Id: Iaef2b6815fe6c64450cc0cc36f8920e4c38f40e6

11 years agoFix a leak in the JIT.
Elliott Hughes [Fri, 20 Apr 2012 00:16:08 +0000 (17:16 -0700)]
Fix a leak in the JIT.

We should use the JIT's arena rather than explicit allocation/deallocation.

Change-Id: I12b78cb1586222b082b140e4a1db00969ce23e3c

11 years agoFix a small memory leak.
Elliott Hughes [Sat, 7 Jan 2012 01:25:14 +0000 (17:25 -0800)]
Fix a small memory leak.

Found by someone reading the code, rather than because we saw a crash. This is
only a small leak, and you'd have to be dumping threads (because of an ANR) or
creating a new thread to provoke it.

(cherry picked from commit 6d1a1dfd0ef006e19067b6ffd927160d0c6d9647)

Change-Id: Id857efca8d34b20d1acaa452c3fe5d2975e2572b

11 years agoSet SYSTEM_PAGE_SIZE from PAGE_SHIFT
Chris Dearman [Tue, 31 Jan 2012 02:22:27 +0000 (18:22 -0800)]
Set SYSTEM_PAGE_SIZE from PAGE_SHIFT

Hardwiring the expected page size is not portable

Signed-off-by: Chris Dearman <chris@mips.com>
11 years agoObvious mix of = and ==
Mark Gordon [Tue, 27 Mar 2012 19:16:32 +0000 (15:16 -0400)]
Obvious mix of = and ==

Change-Id: I52183924c287b80e9ca7c20df6edfe132b43e088

11 years agoCompact merged dexes on their byte aligned size
Charlie Groves [Sun, 11 Mar 2012 07:40:47 +0000 (23:40 -0800)]
Compact merged dexes on their byte aligned size

Previously, if a merged dex had too much free space, the compaction step
would subtract the unused space from the section sizes generated by
adding the merged dexes section sizes. The merged sizes weren't 4-byte
aligned, which meant if a section had been bumped in size for the
boundary, the compaction would fail when loading into a section eg
'Section limit 3665072 exceeded by string data'.

This gets the compacted section size from the actual used space instead
of from the non-aligned estimate, so everything fits.

Change-Id: I1aee1abd721ec21998bb3ea78f6f424e7421b614

12 years agomerge in ics-mr1-release history after reset to ics-mr1
The Android Automerger [Sat, 21 Jan 2012 00:11:41 +0000 (16:11 -0800)]
merge in ics-mr1-release history after reset to ics-mr1

12 years agoFix to dex verifier to handle float types that look like Boolean.
jeffhao [Wed, 18 Jan 2012 02:18:36 +0000 (18:18 -0800)]
Fix to dex verifier to handle float types that look like Boolean.

Addresses http://code.google.com/p/android/issues/detail?id=23465, where
a Float value is potentially 0x0 or Float.MIN_VALUE (0x1), which combines
into a Boolean type, which should be allowed to be promoted to Float.

(cherry picked from commit 2a249c50903d669d2a9296f0fe407b67646626c8)

Change-Id: I1b81fad5e1e166246535bdfa4b7a98db3b2531bf

12 years agomerge in ics-mr1-release history after reset to ics-mr1
The Android Automerger [Thu, 19 Jan 2012 05:16:51 +0000 (21:16 -0800)]
merge in ics-mr1-release history after reset to ics-mr1

12 years agoBug fix for control-flow construction involving nested loops.
Ben Cheng [Wed, 18 Jan 2012 21:45:57 +0000 (13:45 -0800)]
Bug fix for control-flow construction involving nested loops.

Bug: 5884080
Change-Id: Ic2f42abae3e71f9c0729984902396f92902184a3

12 years agomerge in ics-mr1-release history after reset to ics-mr1
The Android Automerger [Sat, 14 Jan 2012 00:17:11 +0000 (16:17 -0800)]
merge in ics-mr1-release history after reset to ics-mr1

12 years agofix personality call.
Nick Kralevich [Fri, 13 Jan 2012 22:04:38 +0000 (14:04 -0800)]
fix personality call.

Change-Id: I25845344494ac15b4a6876784d2b040779892459

12 years agoPrevent memory fragmentation.
Selim Gurun [Fri, 13 Jan 2012 18:47:15 +0000 (10:47 -0800)]
Prevent memory fragmentation.

Bug: 5817320

Prevent memory fragmentation and potential allocation failures. This
change is temporary.

Change-Id: Id1e8f9606687648235ea9e18861125a8c799d812

12 years agomerge in ics-mr1-release history after reset to ics-mr1
The Android Automerger [Fri, 2 Dec 2011 00:45:10 +0000 (16:45 -0800)]
merge in ics-mr1-release history after reset to ics-mr1

12 years agoInitialize the vC field of inlined getters/setters.
Ben Cheng [Thu, 1 Dec 2011 20:24:49 +0000 (12:24 -0800)]
Initialize the vC field of inlined getters/setters.

To eliminate a benign Valgrind warning.

Change-Id: I5f0ae53b8c86fe476f0b7ddfb6589191aea464f6

12 years agoInitialize the JNI indirect ref table to make Valgrind happy.
Ben Cheng [Wed, 30 Nov 2011 18:43:09 +0000 (10:43 -0800)]
Initialize the JNI indirect ref table to make Valgrind happy.

Specifically, this is to eliminate the benign warning about the
'serial' field being uninitialized.

BUG: 5542417
Change-Id: I8550972e70db94f24c47c4e5dea0b59cc5840b48

12 years agomerge in ics-mr1-release history after reset to ics-mr1
The Android Automerger [Thu, 10 Nov 2011 14:45:54 +0000 (06:45 -0800)]
merge in ics-mr1-release history after reset to ics-mr1

12 years agoDO NOT MERGE Further refinement of card table clear
Andy McFadden [Tue, 8 Nov 2011 00:57:54 +0000 (16:57 -0800)]
DO NOT MERGE Further refinement of card table clear

We're currently using memset() to clear cards out to the growth
limit, which on a device configured for 48MB/128MB process caps
means we're sitting on 384KB or 1MB of pages.  We can reduce that
substantially.

This change attempts to reduce the memset() length to the currently
active portion of the card table, based on the placement of objects
in the "live objects" bitmap.  This should avoid faulting in card
pages that we don't actually need.

This is suboptimal when parts of the card table are used briefly
and then ignored for a while.  An alternative implementation, which
uses madvise(DONTNEED) aggressively, is also included (#ifdefed out).

Bug 5567332

(cherry-pick from master)

Change-Id: I6b181de20a0c8b6307acaebbc5d6dac20422f25d

12 years agomerge in ics-mr1-release history after reset to ics-mr1
The Android Automerger [Sun, 6 Nov 2011 14:45:27 +0000 (06:45 -0800)]
merge in ics-mr1-release history after reset to ics-mr1

12 years agoMerge "Don't pay for filename/line number lookup unless you need to." into ics-mr1
Elliott Hughes [Sat, 5 Nov 2011 00:22:17 +0000 (17:22 -0700)]
Merge "Don't pay for filename/line number lookup unless you need to." into ics-mr1

12 years agoMerge "Knock ::self() out of the ParseXml profile." into ics-mr1
Elliott Hughes [Sat, 5 Nov 2011 00:22:01 +0000 (17:22 -0700)]
Merge "Knock ::self() out of the ParseXml profile." into ics-mr1

12 years agoMerge "Reduce Dalvik card table overhead" into ics-mr1
Andy McFadden [Fri, 4 Nov 2011 23:46:14 +0000 (16:46 -0700)]
Merge "Reduce Dalvik card table overhead" into ics-mr1

12 years agoDon't decr profiling count without incr
Andy McFadden [Fri, 4 Nov 2011 22:58:37 +0000 (15:58 -0700)]
Don't decr profiling count without incr

The failure cleanup code in dvmMethodTraceStart() was assuming
that profiling had already been started, which was true until my
previous change.  The code in the interpreter isn't too picky, but
updateActiveProfilers() is trying to keep an accurate acount, and
gets upset.  If you pass in a bad filename, we disable something that
was never enabled, and the VM gets confused and aborts out of spite.

Related to bug 5564440

Change-Id: I17fd8bf4654c29f49efb10142d222a91c190b6ed

12 years agoReduce Dalvik card table overhead
Andy McFadden [Fri, 4 Nov 2011 19:56:55 +0000 (12:56 -0700)]
Reduce Dalvik card table overhead

The VM has a "growth limit" (e.g. 48MB) that is used to cap normal
apps, and a "max heap" (e.g. 128MB) that is used for apps with unusual
high-memory requirements.  The Dalvik GC uses a 128:1 "card table"
that spans the entire managed heap, out to the "max heap" limit.

The table is erased during every concurrent GC.  The first time
we do that post-zygote, the copy-on-write behavior causes physical
pages to be allocated.  We're currently clearing out to the heap max,
not the growth limit, which means we have a bunch of physical memory
allocated for pages that will never be used.

This changes it so we only clear out to the growth limit.  If the
growth limit is removed by the app, we clear out to the heap max.
On devices with a 128MB max heap, this reduces the private/dirty
usage for most apps by 640KB.

Bug 5567332

Change-Id: Iac633017c128d63b284d48016e47d40cc46df1de

12 years agoDon't pay for filename/line number lookup unless you need to.
Elliott Hughes [Fri, 4 Nov 2011 01:17:19 +0000 (18:17 -0700)]
Don't pay for filename/line number lookup unless you need to.

This wasn't a regression; the code's always been like this. But this shows up
in profiles of anything doing a lot of synchronized stuff, even when there's
no contention.

There are two awkward cases. One is that the old code used to offer a variety
of special-case messages for failures to find the current frame, save area,
and Method*. I assume this was just to help in debugging and shouldn't happen
in practice, so I don't think we've lost anything there. The other case was
that on unlocking, we used to explicitly say "unlocked". But I'm not sure we
could get into a situation where we'd be reporting contention on a monitor
that wasn't locked when we tried to lock it. So I think that's okay too.

Change-Id: Ib4401c771f717e9c8cc9c4e5346ec7a5f46a1636

12 years agoKnock ::self() out of the ParseXml profile.
Elliott Hughes [Thu, 3 Nov 2011 21:07:24 +0000 (14:07 -0700)]
Knock ::self() out of the ParseXml profile.

New profile (>=0.6, which is where the original seems to have cut off):

47       11.0070  dvmCallJNIMethod(unsigned int const*, JValue*, Method const*, Thread*)
30        7.0258  dvmChangeStatus(Thread*, ThreadStatus)
29        6.7916  addLocalReference(Thread*, Object*)
26        6.0890  dexDecodeDebugInfo(DexFile const*, DexCode const*, char const*, unsigned int, unsigned int, int (*)(void*, unsigned int, unsigned int), void (*)(void*, unsigned short, unsigned int, unsigned int, char const*, char const*, char const*), void*)
22        5.1522  dalvik_inst
15        3.5129  lockMonitor(Thread*, Monitor*)
14        3.2787  dvmLineNumFromPC
13        3.0445  javaLangString_equals(unsigned int, unsigned int, unsigned int, unsigned int, JValue*)
13        3.0445  scanObject(Object const*, GcMarkContext*)
12        2.8103  ScopedJniThreadState::ScopedJniThreadState(_JNIEnv*)
12        2.8103  common_invokeMethodNoRange
12        2.8103  dvmDecodeIndirectRef(Thread*, _jobject*)
9         2.1077  IndirectRefTable::add(unsigned int, Object*)
9         2.1077  ReleasePrimitiveArrayCritical(_JNIEnv*, _jarray*, void*, int)
9         2.1077  markObjectNonNull(Object const*, GcMarkContext*, bool)
9         2.1077  unpinPrimitiveArray(ArrayObject*)
8         1.8735  getCodeAddrCommon(unsigned short const*, bool)
7         1.6393  dexStringByTypeIdx(DexFile const*, unsigned int)
7         1.6393  dvmHeapSourceAlloc(unsigned int)
6         1.4052  GetPrimitiveArrayCritical(_JNIEnv*, _jarray*, unsigned char*)
6         1.4052  dvmPlatformInvoke
6         1.4052  pinPrimitiveArray(ArrayObject*)
6         1.4052  readUnsignedLeb128(unsigned char const**)
6         1.4052  scanFields(Object const*, GcMarkContext*)
5         1.1710  IndirectRefTable::get(void*) const
5         1.1710  dvmFindInReferenceTable(ReferenceTable const*, Object**, Object*)
4         0.9368  common_returnFromMethod
4         0.9368  dvmAddToReferenceTable(ReferenceTable*, Object*)
4         0.9368  dvmHeapBitmapScanWalk(HeapBitmap*, void (*)(Object*, void*, void*), void*)
4         0.9368  dvmInterpret(Thread*, Method const*, JValue*)
4         0.9368  dvmLockObject
4         0.9368  dvmMalloc(unsigned int, int)
4         0.9368  findPackedSwitchIndex(unsigned short const*, int, int)
4         0.9368  readStringIdx(DexFile const*, unsigned char const**)
4         0.9368  unlockMonitor(Thread*, Monitor*)
3         0.7026  dvmSetFinalizable

Change-Id: Ic5c36859f6810413bd0b48aad1d99da7daa6e8ba

12 years agoStart profiling after state prep
Andy McFadden [Thu, 3 Nov 2011 21:13:24 +0000 (14:13 -0700)]
Start profiling after state prep

The updateActiveProfilers() call was happening before the profiler
state setup was completed, creating a race that could lead to a
native crash.

Bug 5564440

Change-Id: I8c030fc87f106aa8b2a67876d604a5bd79e05175

12 years agoMerge "DO NOT MERGE: Optimize IndirectRefTable." into ics-mr1
Elliott Hughes [Thu, 3 Nov 2011 17:59:19 +0000 (10:59 -0700)]
Merge "DO NOT MERGE: Optimize IndirectRefTable." into ics-mr1

12 years agoDon't malloc between fork and exec
Andy McFadden [Wed, 2 Nov 2011 22:26:39 +0000 (15:26 -0700)]
Don't malloc between fork and exec

This was causing hangs in DEX file loading.

Bug 5329833

Change-Id: I19e9e0624f3823cc771c71a4b4ac177d57343c58

12 years agomerge in ics-mr1-release history after reset to ics-mr1
The Android Automerger [Wed, 2 Nov 2011 13:45:25 +0000 (06:45 -0700)]
merge in ics-mr1-release history after reset to ics-mr1

12 years agoam 3931f798: Fix cleanup (cherry-picked from ics-mr1)
Andy McFadden [Tue, 1 Nov 2011 23:34:39 +0000 (23:34 +0000)]
am 3931f798: Fix cleanup (cherry-picked from ics-mr1)

* commit '3931f79801b67577bf8ac12d0b3fb5b4aeaaf766':
  Fix cleanup (cherry-picked from ics-mr1)

12 years agoFix cleanup
Andy McFadden [Tue, 1 Nov 2011 18:04:46 +0000 (11:04 -0700)]
Fix cleanup
(cherry-picked from ics-mr1)

The code was attempting to null out a struct member after freeing
the struct.

Also, changed the order of directory permission tests so that
writable comes first.  Somehow "dalvik-cache directory not
writable" seems more direct than "not readable", since the code
isn't generally interested in reading the directory.

Bug 5549907

Change-Id: If737ab822b356aae98e47292d21946e33a04342b

12 years agoam e3d9cde4: Disable JIT single-stepping resume mode.
Ben Cheng [Tue, 1 Nov 2011 22:05:11 +0000 (22:05 +0000)]
am e3d9cde4: Disable JIT single-stepping resume mode.

* commit 'e3d9cde40f96bf40dd4ad3d495949869bde5dd1e':
  Disable JIT single-stepping resume mode.

12 years agoDisable JIT single-stepping resume mode.
Ben Cheng [Tue, 1 Nov 2011 20:53:39 +0000 (13:53 -0700)]
Disable JIT single-stepping resume mode.

BUG: 5551114
Change-Id: Id8ee75be5e71ee258994235952871ad181024e56

12 years agoFix cleanup
Andy McFadden [Tue, 1 Nov 2011 18:04:46 +0000 (11:04 -0700)]
Fix cleanup

The code was attempting to null out a struct member after freeing
the struct.

Also, changed the order of directory permission tests so that
writable comes first.  Somehow "dalvik-cache directory not
writable" seems more direct than "not readable", since the code
isn't generally interested in reading the directory.

Bug 5549907

Change-Id: Ie412ff6c0870357b92e53606f70cb9407e04e7cf

12 years agoDO NOT MERGE: Optimize IndirectRefTable.
Jeff Brown [Thu, 27 Oct 2011 00:04:54 +0000 (17:04 -0700)]
DO NOT MERGE: Optimize IndirectRefTable.

Removed unused debugging code that maintained pointers to the
4 previous objects in each slot.

Promoted the serial number property of a slot to be first class.
This is a sufficiently cheap and useful check to perform that we
might as well always do it.

Changed get() and remove() so they check the validity of the
references at the same time as they retrieve them.  They're
also a bit smarter about checking the reference kind given
that they can assume that the table will only contain references
of one kind.  The checks are now much more consistent.
For example, remove() used to check whether an index was
stale only in the case where it was not removing the top entry
(now it always checks).

Made add() return NULL instead of dvmAbort()'ing in the case
where the table overflowed and ensure we don't blow away the
table in that cases.  This change resolves an issue with
TestIndirectRefTable.cpp which deliberately overflows the table
and expects add() to return NULL (as documented!).  As it
happens, the add() method is called in exactly 3 places in
Jni.cpp.  In each of those cases, the code was written to
handle a NULL result and only in the case of adding a weak
global reference did it not abort.  Fixed the weak global
reference case to be consistent with the others.

Changed the signature of contains() to take an Object* since
that's what we're actually looking for inside the table.

Added a couple of calls to dump() in TestIndirectRefTable.cpp
for visual inspection of its correctness.

Performance as measured by TestIndirectRefTable on same hardware.

Old implementation:
- Add/remove 100 objects FIFO order, 100000 iterations, 0.023ms / iteration
- Add/remove 100 objects LIFO order, 100000 iterations, 0.020ms / iteration
- Get 100 objects, 100000 iterations, 0.009ms / iteration

New implementation:
- Add/remove 100 objects FIFO order, 100000 iterations, 0.010ms / iteration
- Add/remove 100 objects LIFO order, 100000 iterations, 0.009ms / iteration
- Get 100 objects, 100000 iterations, 0.002ms / iteration

Cherry-pick from master.

Conflicts:

vm/IndirectRefTable.cpp

Change-Id: I157f3c1ba598137222878b8e6a5890efb744fe76

12 years agoFix single-stepping resume mode in the interpreter.
Ben Cheng [Thu, 27 Oct 2011 21:04:12 +0000 (14:04 -0700)]
Fix single-stepping resume mode in the interpreter.

(cherry-picked from master)

When a resume attempt is cancelled due to other pending request make sure
the old native resume PC is cleared. Otherwise the JIT code cache may be
re-entered with mismatching Dalvik PC.

Also fix a code bloat problem where single-step count is not set properly
after executing a return instruction.

BUG: 5208786

Change-Id: I39db1633dcda1f464311c878ded4b5695a9b7ffd

12 years agoam b1e2271b: Fix single-stepping resume mode in the interpreter.
Ben Cheng [Fri, 28 Oct 2011 20:12:05 +0000 (20:12 +0000)]
am b1e2271b: Fix single-stepping resume mode in the interpreter.

* commit 'b1e2271b51cfb1f264b8966f1151652767259751':
  Fix single-stepping resume mode in the interpreter.

12 years agoFix single-stepping resume mode in the interpreter.
Ben Cheng [Thu, 27 Oct 2011 21:04:12 +0000 (14:04 -0700)]
Fix single-stepping resume mode in the interpreter.

(cherry-picked from master)

When a resume attempt is cancelled due to other pending request make sure
the old native resume PC is cleared. Otherwise the JIT code cache may be
re-entered with mismatching Dalvik PC.

Also fix a code bloat problem where single-step count is not set properly
after executing a return instruction.

BUG: 5208786

Change-Id: I39db1633dcda1f464311c878ded4b5695a9b7ffd

12 years agoFix single-stepping resume mode in the interpreter.
Ben Cheng [Thu, 27 Oct 2011 21:04:12 +0000 (14:04 -0700)]
Fix single-stepping resume mode in the interpreter.

(cherry-picked from master)

When a resume attempt is cancelled due to other pending request make sure
the old native resume PC is cleared. Otherwise the JIT code cache may be
re-entered with mismatching Dalvik PC.

Also fix a code bloat problem where single-step count is not set properly
after executing a return instruction.

BUG: 5208786

Change-Id: I54775215b11eae29ccdb6111dc0fdfa99e41e08d

12 years agomerge in ics-release history after reset to master
The Android Automerger [Thu, 20 Oct 2011 13:59:36 +0000 (06:59 -0700)]
merge in ics-release history after reset to master

12 years agoMerge "Provide debugging info should extended opcode lookup fail."
Jesse Wilson [Wed, 19 Oct 2011 20:26:19 +0000 (13:26 -0700)]
Merge "Provide debugging info should extended opcode lookup fail."

12 years agoProvide debugging info should extended opcode lookup fail.
Jesse Wilson [Wed, 19 Oct 2011 15:02:51 +0000 (11:02 -0400)]
Provide debugging info should extended opcode lookup fail.

Bug: http://code.google.com/p/android/issues/detail?id=20814
Change-Id: Iedaf07721c29dd3d24827c3f7a953b7371c9735d

12 years agomerge in ics-release history after reset to master
The Android Automerger [Tue, 18 Oct 2011 13:59:26 +0000 (06:59 -0700)]
merge in ics-release history after reset to master

12 years agoAvoid deadlock.
Elliott Hughes [Fri, 14 Oct 2011 21:36:46 +0000 (14:36 -0700)]
Avoid deadlock.

dvmIsHeapAddress is the pointer validity checker for use outside the GC. If
it doesn't take the heap lock, it risks looking at data structures that are
changing under its feet. If it does take the heap lock, it risks deadlock
between a thread doing an explicit GC and the signal catcher trying to dump
threads (causing the GC thread to be suspended while holding the heap lock).

Calling back into managed code while holding the heap lock and with other
threads resumed sounds like an inherently bad idea to me, but that's a battle
for another day.

With this change, we can handle SIGQUIT while we're in ReferenceQueue.add
as a result of an explicit GC in a system doing concurrent collections, so
the only known problem is fixed.

Bug: 5425802
Change-Id: I42d434d5ea3ffbcb77a4c544b81b08a4c7364a16

12 years agomerge in ics-release history after reset to master
The Android Automerger [Tue, 11 Oct 2011 12:32:21 +0000 (05:32 -0700)]
merge in ics-release history after reset to master

12 years agoMerge "Leave up to 512 chars of gDvm.lastMessage on dvmAbort's stack frame."
Ben Cheng [Mon, 10 Oct 2011 23:55:23 +0000 (16:55 -0700)]
Merge "Leave up to 512 chars of gDvm.lastMessage on dvmAbort's stack frame."

12 years agoLeave up to 512 chars of gDvm.lastMessage on dvmAbort's stack frame.
Ben Cheng [Mon, 10 Oct 2011 23:26:36 +0000 (16:26 -0700)]
Leave up to 512 chars of gDvm.lastMessage on dvmAbort's stack frame.

Bug: 5372634
Change-Id: I019a059e5a989de3dfc9e2f5e78f7caa7168cf75

12 years agoMerge "Reordered some checks and added more to the verifier."
jeffhao [Mon, 10 Oct 2011 21:51:45 +0000 (14:51 -0700)]
Merge "Reordered some checks and added more to the verifier."

12 years agoMerge "Revert "Use the card table to scan the immune region of the heap.""
Ben Cheng [Mon, 10 Oct 2011 21:08:17 +0000 (14:08 -0700)]
Merge "Revert "Use the card table to scan the immune region of the heap.""

12 years agoRevert "Use the card table to scan the immune region of the heap."
Ben Cheng [Mon, 10 Oct 2011 20:17:58 +0000 (13:17 -0700)]
Revert "Use the card table to scan the immune region of the heap."

This reverts commit 6c355e53332502314c3d82a7afcf898d66118f27.

Conflicts are resolved in:

vm/alloc/CardTable.cpp
vm/alloc/HeapBitmap.cpp
vm/alloc/HeapInternal.h
vm/alloc/HeapSource.cpp
vm/alloc/MarkSweep.cpp

Tested with overnight monkey runs + Delaunay.

Bug: 5037417
Change-Id: I400eff8542d0bddd865ce570a479f7750c1ba484

12 years agoMerge "Don't say UTF-8 in Dalvik's CheckJNI when we mean Modified UTF-8."
Elliott Hughes [Mon, 10 Oct 2011 18:17:13 +0000 (11:17 -0700)]
Merge "Don't say UTF-8 in Dalvik's CheckJNI when we mean Modified UTF-8."

12 years agoDon't say UTF-8 in Dalvik's CheckJNI when we mean Modified UTF-8.
Elliott Hughes [Fri, 7 Oct 2011 22:42:23 +0000 (15:42 -0700)]
Don't say UTF-8 in Dalvik's CheckJNI when we mean Modified UTF-8.

Change-Id: I4130e7f92275275a6d6f782e1ad979fbf9197a8e

12 years agoRevert "GC experiment."
The Android Automerger [Fri, 7 Oct 2011 00:32:54 +0000 (17:32 -0700)]
Revert "GC experiment."

This reverts commit b3fb1d08137ead7b6be016e1bc147cb781e76111.

12 years agoReordered some checks and added more to the verifier.
jeffhao [Thu, 6 Oct 2011 17:33:22 +0000 (10:33 -0700)]
Reordered some checks and added more to the verifier.

The verifier checks for 2 errors in put instructions. First, if the
field can't be resolved, the instruction is overwritten to throw a
verifier error. Second, if the data types of the field and instruction
mismatch, the verifier rejects the whole class. This change swaps the
order of the checks, so that if both errors occur, the class is
rejected.

Also, checks were added to prevent float values from being used as array
indexes or arguments to if statements.

Change-Id: I7b122f4ebc4043266be4466e10679c6319a788ef

12 years agoGC experiment.
Ben Cheng [Thu, 6 Oct 2011 05:43:34 +0000 (22:43 -0700)]
GC experiment.

Change-Id: Ifbeec98a796fe8347da25a1be50288d756b40f02

12 years agomerge in ics-release history after reset to master
The Android Automerger [Wed, 5 Oct 2011 13:58:05 +0000 (06:58 -0700)]
merge in ics-release history after reset to master

12 years agoFixed dx verifier issue when an error is detected.
jeffhao [Tue, 4 Oct 2011 21:33:57 +0000 (14:33 -0700)]
Fixed dx verifier issue when an error is detected.

The code needed to break out properly when an error was detected to
prevent it from causing even more problems.

Change-Id: If497655a406689aeb1b4aa6523421cc22f2b7593

12 years agomerge in ics-release history after reset to master
The Android Automerger [Fri, 30 Sep 2011 13:44:24 +0000 (06:44 -0700)]
merge in ics-release history after reset to master

12 years agoFix a typo in unused code.
Elliott Hughes [Thu, 29 Sep 2011 17:47:22 +0000 (10:47 -0700)]
Fix a typo in unused code.

Change-Id: Ied11a79337a3cc795428dfde29e405ec0bbcb98c

12 years agomerge in ics-release history after reset to master
The Android Automerger [Thu, 29 Sep 2011 00:57:05 +0000 (17:57 -0700)]
merge in ics-release history after reset to master

12 years agoFix memory barriers (Issue 3338450)
buzbee [Tue, 27 Sep 2011 18:47:28 +0000 (11:47 -0700)]
Fix memory barriers (Issue 3338450)

Add extra memory barrier on volatile stores.

Change-Id: Id4a4750cdfc910eda2f0b44ead0af2a569b5735e

12 years agomerge in ics-release history after reset to master
The Android Automerger [Wed, 21 Sep 2011 13:59:19 +0000 (06:59 -0700)]
merge in ics-release history after reset to master

12 years agoDon't spam the logs with thread priority information.
Jesse Wilson [Tue, 20 Sep 2011 15:52:02 +0000 (11:52 -0400)]
Don't spam the logs with thread priority information.

Ten minutes of logs contained 108 lines of 'Elevating priority from 10 to 0'
and 108 lines of 'Reset priority to 10'. This provides little value.

Change-Id: Id96e8304238aa3344af4dd6342ace575534acc0f

12 years agoAttempt to fix dalvik heap corruption.
Jesse Wilson [Tue, 6 Sep 2011 16:55:25 +0000 (12:55 -0400)]
Attempt to fix dalvik heap corruption.

This is bccheng's find, as reported here:
http://b/5208786#ISSUE_HistoryHeader49

Bug: http://b/5208786
Change-Id: I1616bac1ef6c197f723656187a784fc6838975d4

12 years agoAttempt to fix dalvik heap corruption.
Jesse Wilson [Tue, 6 Sep 2011 16:55:25 +0000 (12:55 -0400)]
Attempt to fix dalvik heap corruption.

This is bccheng's find, as reported here:
http://b/5208786#ISSUE_HistoryHeader49

Bug: http://b/5208786
Change-Id: I1616bac1ef6c197f723656187a784fc6838975d4

12 years agomerge in ics-release history after reset to master
The Android Automerger [Tue, 6 Sep 2011 13:33:14 +0000 (06:33 -0700)]
merge in ics-release history after reset to master

12 years agomerge in ics-release history after reset to master
The Android Automerger [Mon, 5 Sep 2011 13:33:20 +0000 (06:33 -0700)]
merge in ics-release history after reset to master

12 years agoFix document formatting error (html tags)
Logan Chien [Mon, 5 Sep 2011 13:14:25 +0000 (21:14 +0800)]
Fix document formatting error (html tags)

Change-Id: I57d10b69f400ce3ffda1a26dc632a4122a4f3d70

12 years agomerge in ics-release history after reset to master
The Android Automerger [Fri, 2 Sep 2011 23:01:51 +0000 (16:01 -0700)]
merge in ics-release history after reset to master

12 years agoMerge "Dx reg allocator fix for accidental long/double clobbering."
jeffhao [Fri, 2 Sep 2011 19:00:05 +0000 (12:00 -0700)]
Merge "Dx reg allocator fix for accidental long/double clobbering."

12 years agoMerge "Rename register to conform to instruction format"
Shih-wei Liao [Fri, 2 Sep 2011 14:15:10 +0000 (07:15 -0700)]
Merge "Rename register to conform to instruction format"

12 years agomerge in ics-release history after reset to master
The Android Automerger [Thu, 1 Sep 2011 13:33:23 +0000 (06:33 -0700)]
merge in ics-release history after reset to master

12 years agoRename register to conform to instruction format
Logan Chien [Thu, 1 Sep 2011 07:19:31 +0000 (15:19 +0800)]
Rename register to conform to instruction format

For instruction format 35c, 41c, 52c, and 5rc, the register
name was different from the Dalvik Instruction Format document,
and InstrUtils.cpp of libdex.  This commit updates the name,
so that they will look consistent.

Change-Id: I857b1f9c1ac7178dc147d0019df589c9c6d102e1

12 years agoFix a corner-case dex merging bug.
Jesse Wilson [Wed, 31 Aug 2011 19:00:57 +0000 (15:00 -0400)]
Fix a corner-case dex merging bug.

When doing a null merge (merging a dex file with an empty file) word-alignment
means that the output may be two bytes longer than the input. Always include the
rounded-up length in the bytes allocated to a section.

com.android.dx.util.DexException: Section limit 669894 exceeded by type list
at com.android.dx.io.DexBuffer$Section.ensureCapacity(DexBuffer.java:541)
at com.android.dx.io.DexBuffer$Section.writeShort(DexBuffer.java:574)
at com.android.dx.io.DexBuffer$Section.writeTypeList(DexBuffer.java:636)
at com.android.dx.merge.DexMerger$3.write(DexMerger.java:439)
at com.android.dx.merge.DexMerger$3.write(DexMerger.java:424)
at com.android.dx.merge.DexMerger$IdMerger.mergeUnsorted(DexMerger.java:321)

Change-Id: I67b81ba8c837b0307e313f4012a051b022b25db2

12 years agomerge in ics-release history after reset to master
The Android Automerger [Wed, 31 Aug 2011 13:33:16 +0000 (06:33 -0700)]
merge in ics-release history after reset to master

12 years agoam 4130f726: am 1dd1d924: am e4e76feb: Merge "Set DX stack size to 1M by default...
Raphael Moll [Tue, 30 Aug 2011 16:50:59 +0000 (09:50 -0700)]
am 4130f726: am 1dd1d924: am e4e76feb: Merge "Set DX stack size to 1M by default on Windows."

* commit '4130f726d82493748bf7252da631e1f315f2538c':
  Set DX stack size to 1M by default on Windows.

12 years agoam 1dd1d924: am e4e76feb: Merge "Set DX stack size to 1M by default on Windows."
Raphael Moll [Tue, 30 Aug 2011 01:01:08 +0000 (18:01 -0700)]
am 1dd1d924: am e4e76feb: Merge "Set DX stack size to 1M by default on Windows."

* commit '1dd1d9246c0804d1b4fda955c56a6d89c989ec6a':
  Set DX stack size to 1M by default on Windows.

12 years agoam e4e76feb: Merge "Set DX stack size to 1M by default on Windows."
Raphael Moll [Tue, 30 Aug 2011 00:58:08 +0000 (17:58 -0700)]
am e4e76feb: Merge "Set DX stack size to 1M by default on Windows."

* commit 'e4e76feb6aad0d0161fabe12cb7d97e665f7b109':
  Set DX stack size to 1M by default on Windows.

12 years agoMerge "Set DX stack size to 1M by default on Windows."
Raphael Moll [Fri, 26 Aug 2011 23:33:03 +0000 (16:33 -0700)]
Merge "Set DX stack size to 1M by default on Windows."