OSDN Git Service

android-x86/art.git
9 years agoART: Allow class-loading during deopt
Andreas Gampe [Tue, 10 Feb 2015 23:37:27 +0000 (15:37 -0800)]
ART: Allow class-loading during deopt

When deoptimizing, we might touch code that uses unloaded classes.

Bug: 19290147
Change-Id: I5776f08ba366e9742336caba0d6af85f00629afc

9 years agoFix false alarm on thread suspend timeout
tony.ys_liu [Wed, 14 Jan 2015 10:28:03 +0000 (18:28 +0800)]
Fix false alarm on thread suspend timeout

Root cause:
  CLOCK_REALTIME will jump backward/forward when system time-of-day clock is changed.
  It implies now_abs_ts will jump a lot across end_abs_ts.
  Then, it makes a false alarm (process crash) on thread suspend timeout when doing ComputeRelativeTimeSpec.

      if (ComputeRelativeTimeSpec(&rel_ts, end_abs_ts, now_abs_ts)) {
        return false;  // Timed out.
      }

Solution:
  Use CLOCK_MONOTONIC instead

Bug: 19296054

(cherry picked from commit 071e48ecfc95b1c67b07c975190d51f646ac4276)

Change-Id: I68fa5afda75e53373e836dd7779821aeb6f1bd46

9 years agoCheck for null class of class
Mathieu Chartier [Thu, 29 Jan 2015 18:47:10 +0000 (10:47 -0800)]
Check for null class of class

Experimental fix.

Bug: 19193521
Change-Id: I11571dd5a63852acdbad5d4291510542c2e4b18a

9 years agoForce set resolved method for static invokes
Mathieu Chartier [Wed, 28 Jan 2015 21:11:44 +0000 (13:11 -0800)]
Force set resolved method for static invokes

For static invokes, we may dispatch to the static method in the
superclass but resolve using the subclass. To prevent getting slow
paths on each invoke, we force set the resolved method for the
super class dex method index if we are in the same dex file.

Added test.

Bug: 19175856
(cherry picked from commit e4a91bbeaa118bc01a5c338c6a6d9ee094a6a2c6)
Change-Id: I26f8644a7f725f5c2dc2a94a8e9578f573792507

9 years agoQuick: Fix range check for intrinsic String.charAt() on x86.
Vladimir Marko [Mon, 26 Jan 2015 14:06:46 +0000 (14:06 +0000)]
Quick: Fix range check for intrinsic String.charAt() on x86.

Bug: 19125146

(cherry picked from commit 00ca84730a21578dcc6b47bd8e08b78ab9b2dded)

Change-Id: I67184371597fdcc9d9186172c1cff4efd3ca3093

9 years agoMerge "ART: On shutdown, only warn on mutex contention" into lmp-mr1-dev
Andreas Gampe [Fri, 23 Jan 2015 22:34:24 +0000 (22:34 +0000)]
Merge "ART: On shutdown, only warn on mutex contention" into lmp-mr1-dev

9 years agoMerge "Allow preopted apps to have sharpened calls for non-x86 architectures." into...
Jeff Hao [Fri, 23 Jan 2015 18:46:31 +0000 (18:46 +0000)]
Merge "Allow preopted apps to have sharpened calls for non-x86 architectures." into lmp-mr1-dev

9 years agoART: On shutdown, only warn on mutex contention
Andreas Gampe [Fri, 23 Jan 2015 03:48:51 +0000 (19:48 -0800)]
ART: On shutdown, only warn on mutex contention

Do not abort, as daemon threads may still be active.

Bug: 17894429
Change-Id: I7c1d50ff8d4a5e150279e703a69c8f2f1d423e6b

9 years agoFix exception handling during deoptimization
Sebastien Hertz [Fri, 16 Jan 2015 18:49:09 +0000 (19:49 +0100)]
Fix exception handling during deoptimization

When interpreting a deoptimized shadow frame, we may start with a
pending exception thrown by a previous deoptimized shadow frame (from
a previous invoke). Therefore, we need to handle it before executing
any instruction, otherwise we execute incorrect code.

Because we need the DEX pc of the throwing instruction to find a
matching catch handler, we initialize deoptimized shadow frames with
the current DEX pc at the time the stack is deoptimized.
When we are about to interpret a deoptimized shadow frame, we need to
update the shadow frame with the DEX pc of the next instruction to
interpret. There are three cases:
- if there is no pending exception, this is the instruction following
the current one.
- if there is a pending exception and we found a matching catch
handler, this is the first instruction of this handler.
- if there is a pending exception but there is no matching catch
handler, we do not execute the deoptimized shadow frame and continue
to its caller.

The verifier now fails when a method starts with a move-exception
instruction. Indeed we cannot start executing a method with a pending
exception.

Bug: 19057915
Bug: 19041195
Bug: 18607595

(cherry picked from commit 270a0e16c3b8e5b95cbfdbd8996ac137c7c6322b)

Change-Id: Ib4fdd0ad704b4f2369d229737c9cc77f559cea55

9 years agoAllow preopted apps to have sharpened calls for non-x86 architectures.
Jeff Hao [Fri, 23 Jan 2015 02:25:30 +0000 (18:25 -0800)]
Allow preopted apps to have sharpened calls for non-x86 architectures.

Bug: 19100762

Change-Id: Iefd4265cb7864cdaa12c1f9e44bfaa5aa4fb93ec

9 years agoUse correct handle scope offset from StackVisitor
Sebastien Hertz [Wed, 17 Dec 2014 19:02:50 +0000 (20:02 +0100)]
Use correct handle scope offset from StackVisitor

Incorrect offset resulted in bad GC roots due to invalid
Stack::GetThisObject for native methods.

Bug: 19070497
Bug: 18785293

(cherry picked from commit e4b7c892c4f40e76c172a77069afde3fe5ce87da)
Change-Id: I10e144cffac00978e3c84d43a30caccd61559b27

9 years agoMerge "ART: Curb dex2oat verbosity" into lmp-mr1-dev
Andreas Gampe [Thu, 22 Jan 2015 17:38:38 +0000 (17:38 +0000)]
Merge "ART: Curb dex2oat verbosity" into lmp-mr1-dev

9 years agoMerge "JDWP: fix breakpoint for method in the image" into lmp-mr1-dev
Sebastien Hertz [Wed, 21 Jan 2015 19:49:40 +0000 (19:49 +0000)]
Merge "JDWP: fix breakpoint for method in the image" into lmp-mr1-dev

9 years agoART: Curb dex2oat verbosity
Andreas Gampe [Tue, 20 Jan 2015 22:50:21 +0000 (14:50 -0800)]
ART: Curb dex2oat verbosity

Only print dedupe collisions in a debug build or on verbose logging.

Bug: 19082662

(cherry picked from commit 8d295f8f0e88fa5c6373962d545020a06033b3fc)

Change-Id: I08bd04a453d95b6dba6cf6955b5741cd97daf480

9 years agoJDWP: fix breakpoint for method in the image
Sebastien Hertz [Wed, 26 Nov 2014 21:11:27 +0000 (22:11 +0100)]
JDWP: fix breakpoint for method in the image

When we set a breakpoint in a compiled method, we deoptimize it by
changing its entrypoint so it is executed with the interpreter.
However, methods in the image can be called with their direct code
pointer, ignoring the updated entrypoint. In that case, the method
is not executed with the interpreter and we miss the breakpoint.

This CL avoids that situation by forcing a full deoptimization so
everything runs with the interpreter. However, if the image has been
compiled in PIC mode, we keep using selective deoptimization because
direct code pointer is not used in this mode.

Bug: 17965285

(cherry picked from commit 6963e44331258b131bcc0599b868ba15902d6d22)

Change-Id: I9bf738f89b9eb6d18733503216b376b8a1d181f5

9 years agoFix infinite loop in GenerateIdentityHashCode
tony.ys_liu [Fri, 16 Jan 2015 11:16:45 +0000 (19:16 +0800)]
Fix infinite loop in GenerateIdentityHashCode

Root Cause:
  If no one changes the seed, it will become infinite loop due to below condition
    (expected_value & LockWord::kHashMask) == 0

Solution:
  Changes the seed before entering the next loop

Added test.

Bug: 19046417

(cherry picked from commit 7380c3175b443cdc6f12a2a3a91df188eaaa5a61)

Change-Id: I7d1c377dd1bda780681514b24d61ebc776bc80ab

9 years agoMerge "Print more info in MarkSweep::VerifyRoot" into lmp-mr1-dev
Mathieu Chartier [Thu, 15 Jan 2015 19:59:36 +0000 (19:59 +0000)]
Merge "Print more info in MarkSweep::VerifyRoot" into lmp-mr1-dev

9 years agoPrint more info in MarkSweep::VerifyRoot
Mathieu Chartier [Wed, 14 Jan 2015 22:55:47 +0000 (14:55 -0800)]
Print more info in MarkSweep::VerifyRoot

Refactored old root callback to use a new class called RootInfo.
RootInfo contains all the relevant info related to the root
associated with the callback. The MarkSweep::VerifyRoot function
now uses this info to print the StackVisitor's described location
if the GC root is of the type kRootJavaFrame.

Some other cleanup.

Example output:
E/art     (12167): Tried to mark 0x123 not contained by any spaces
E/art     (12167): Attempting see if it's a bad root
E/art     (12167): Found invalid root: 0x123 with type RootJavaFrame
E/art     (12167): Location=Visiting method
'void java.lang.Runtime.gc()' at dex PC 0xffffffff (native PC 0x0)
vreg=0

Bug: 18588862
Change-Id: Ic5a2781f704e931265ffb3621c2eab4b2e25f60f

9 years agoMerge "Fix wide volatile IGET/IPUT on ARM without atomic ldrd/strd." into lmp-mr1-dev
Vladimir Marko [Thu, 15 Jan 2015 09:42:38 +0000 (09:42 +0000)]
Merge "Fix wide volatile IGET/IPUT on ARM without atomic ldrd/strd." into lmp-mr1-dev

9 years agoMerge "Add a test for our NPE checks." into lmp-mr1-dev
Vladimir Marko [Thu, 15 Jan 2015 09:41:35 +0000 (09:41 +0000)]
Merge "Add a test for our NPE checks." into lmp-mr1-dev

9 years agoMerge "ART: Simple structural class check" into lmp-mr1-dev
Andreas Gampe [Wed, 14 Jan 2015 22:11:19 +0000 (22:11 +0000)]
Merge "ART: Simple structural class check" into lmp-mr1-dev

9 years agoART: Simple structural class check
Andreas Gampe [Fri, 7 Nov 2014 00:52:58 +0000 (16:52 -0800)]
ART: Simple structural class check

Adds a simple check to class-loading when the embedded dex file in
an oat file and the dex file on the class path where we found the
class do not match.

We require that the number of methods and fields do not change, as
that will almost certainly mean that quickened and other compiled
offsets are wrong now. This is a reasonably lightweight change, but
we should investigate a full comparison including name and type of
members.

Bug: 17937814
Bug: 18708951
Change-Id: Icb9638bebd369ab23822817f4a97c8dd8625fea5

9 years agoFix wide volatile IGET/IPUT on ARM without atomic ldrd/strd.
Vladimir Marko [Tue, 13 Jan 2015 17:34:28 +0000 (17:34 +0000)]
Fix wide volatile IGET/IPUT on ARM without atomic ldrd/strd.

If ldrd/strd isn't atomic, IPUT_WIDE uses ldrexd+strexd and
we need to record the safepoint for the ldrexd rather than
strexd. IGET_WIDE was simply missing the memory barrier.

Bug: 18993519

(cherry picked from commit ee5e273e4d0dd91b480c8d5dbcccad15c1b7353c)

Change-Id: I4e9270b994f413c1a047c1c4bb9cce5f29e42cb4

9 years agoAdd a test for our NPE checks.
Nicolas Geoffray [Thu, 11 Sep 2014 17:46:01 +0000 (18:46 +0100)]
Add a test for our NPE checks.

(cherry picked from commit 5f16c8838c8e4524101f5897dac70035c5cc271f)

Change-Id: I0b216fb868ed56d30103fe7843a7eb0169b0b28b

9 years agoJDWP: allow VirtualMachine.Resume on partial suspension
Sebastien Hertz [Fri, 9 Jan 2015 13:51:41 +0000 (14:51 +0100)]
JDWP: allow VirtualMachine.Resume on partial suspension

Allows to resume multiple threads with a single VirtualMachine.Resume
command. When some threads are suspended for an event by the debugger
with a suspend count of 1 (other threads are running), a debugger can
resume them all this way.

Bug: 18924933

(cherry picked from commit f9d233dae8ca66ed5a2a474155d6bee0d74c355b)

Change-Id: I948c3b50d43693e86ae31e0928a33a467328583f

9 years agoMerge "Fix sigchainlib's implementation of sigaction" into lmp-mr1-dev
Dmitriy Ivanov [Tue, 6 Jan 2015 23:34:46 +0000 (23:34 +0000)]
Merge "Fix sigchainlib's implementation of sigaction" into lmp-mr1-dev

9 years agoFix sigchainlib's implementation of sigaction
Dmitriy Ivanov [Tue, 6 Jan 2015 22:55:26 +0000 (14:55 -0800)]
Fix sigchainlib's implementation of sigaction

  Correctly handles the case when old_action == new_action

Bug: 18740478
Change-Id: I97092318439e4f6f0a2513d4336496c72f8c5599

9 years agoMerge "Mute common case DexFile.isDexOptNeeded log spam" into lmp-mr1-dev
Brian Carlstrom [Tue, 6 Jan 2015 22:52:48 +0000 (22:52 +0000)]
Merge "Mute common case DexFile.isDexOptNeeded log spam" into lmp-mr1-dev

9 years agoMute common case DexFile.isDexOptNeeded log spam
Brian Carlstrom [Tue, 6 Jan 2015 20:05:34 +0000 (12:05 -0800)]
Mute common case DexFile.isDexOptNeeded log spam

Bug: 18914409

(cherry picked from commit 9851595177fec4d5e328f201f1992719fb40db2a)

Change-Id: If9b741ff8e09400a421e0a0ba1bc0d9ef13a9f99

9 years agoART: Fix divide-by-zero for ARM
Andreas Gampe [Tue, 6 Jan 2015 03:30:59 +0000 (19:30 -0800)]
ART: Fix divide-by-zero for ARM

There was an infinite loop in the code generation for a divide
by literal zero.

Bug: 18887754

(cherry picked from commit cfe71e59c667abb35bc2363c49af7f8b549c44d0)

Change-Id: Ibd481918d3c6d7bc62fdd1a6807042009f561d95

9 years agoMerge "ART: Do not JNI abort on nullptr GetObjectRefType" into lmp-mr1-dev
Andreas Gampe [Mon, 22 Dec 2014 18:27:42 +0000 (18:27 +0000)]
Merge "ART: Do not JNI abort on nullptr GetObjectRefType" into lmp-mr1-dev

9 years agoART: Do not JNI abort on nullptr GetObjectRefType
Andreas Gampe [Sat, 20 Dec 2014 08:08:35 +0000 (00:08 -0800)]
ART: Do not JNI abort on nullptr GetObjectRefType

A nullptr is a valid input, as it is different from a null reference.

Bug: 18820997

(cherry picked from commit a87630724ef4f8760684fa69c8ecc685735aff88)

Change-Id: Ibda8907ba13b20d2055049492a356ffdf4ddc714

9 years agoMerge "ART: Use interpret-only for vmSafeMode" into lmp-mr1-dev
Andreas Gampe [Sat, 20 Dec 2014 07:10:32 +0000 (07:10 +0000)]
Merge "ART: Use interpret-only for vmSafeMode" into lmp-mr1-dev

9 years agoART: Use interpret-only for vmSafeMode
Andreas Gampe [Fri, 19 Dec 2014 17:53:27 +0000 (09:53 -0800)]
ART: Use interpret-only for vmSafeMode

Set the compiler filter when the zygote tells us that the app is
in vmSafeMode. This ensures that secondary dex files will also be
compiled interpret-only.

Bug: 18812762
Change-Id: Idb61870beebe2eeb5b11554f95ea6a725c77f16a

9 years agoPass the real capacity to CreateRosAlloc.
Hiroshi Yamauchi [Sat, 20 Dec 2014 00:02:07 +0000 (16:02 -0800)]
Pass the real capacity to CreateRosAlloc.

(cherrypick commit 94c41dfcaf4bc131964ddd3013432841b07c3839)

Bug: 18808732
Change-Id: Id65469c3deb70c9f750f89cf0e7606849b801a21

9 years agoMerge "Fix running out of temps when storing invoke-interface result." into lmp-mr1-dev
Vladimir Marko [Fri, 19 Dec 2014 21:27:35 +0000 (21:27 +0000)]
Merge "Fix running out of temps when storing invoke-interface result." into lmp-mr1-dev

9 years agoART: Do not compile swap file functionality on Mac
Andreas Gampe [Fri, 19 Dec 2014 20:04:06 +0000 (12:04 -0800)]
ART: Do not compile swap file functionality on Mac

Bug: 18596910
Change-Id: Ic06429f41840ce318cfaa837efecc69c2076440e

9 years agoFix running out of temps when storing invoke-interface result.
Vladimir Marko [Fri, 19 Dec 2014 19:27:26 +0000 (19:27 +0000)]
Fix running out of temps when storing invoke-interface result.

On ARM, after emitting invoke-interface we didn't have any
free temps to use for storing the result, so we would crash
if the result was an unpromoted dalvik register with stack
location too far from SP.

Bug: 18769895
Change-Id: Ie6c131d68f1853a8317b305a22eab22faea80e90

9 years agoART: Fix swap space free list
Andreas Gampe [Fri, 19 Dec 2014 08:14:38 +0000 (00:14 -0800)]
ART: Fix swap space free list

You should look for the element's size, not the freed size, when
coalescing nodes.

Bug: 18809642
Change-Id: I0c4396d52ef68f392f1e8c76a57e3fb5587b77fb

9 years agoART: Swap-space in the compiler
Andreas Gampe [Tue, 9 Dec 2014 00:59:43 +0000 (16:59 -0800)]
ART: Swap-space in the compiler

Introduce a swap-space and corresponding allocator to transparently
switch native allocations to memory backed by a file.

Bug: 18596910
Change-Id: I131448f3907115054a592af73db86d2b9257ea33

9 years agoMerge "JDWP: fix setting multiple breakpoints in the same method" into lmp-mr1-dev
Sebastien Hertz [Thu, 18 Dec 2014 06:57:03 +0000 (06:57 +0000)]
Merge "JDWP: fix setting multiple breakpoints in the same method" into lmp-mr1-dev

9 years agoART: Punt to the interpreter for VerifiedMethod errors
Andreas Gampe [Wed, 17 Dec 2014 22:35:46 +0000 (14:35 -0800)]
ART: Punt to the interpreter for VerifiedMethod errors

In case that the GC map can't be created (because of size restrictions),
do not fail the class. Instead punt to the interpreter.

Bug: 17791183
Change-Id: I348bb306dbfc85c235fa93c0c527fba6627551fe

9 years agoJDWP: fix setting multiple breakpoints in the same method
Sebastien Hertz [Wed, 17 Dec 2014 15:35:50 +0000 (16:35 +0100)]
JDWP: fix setting multiple breakpoints in the same method

When setting multiple breakpoints in the same method, we were
incorrectly setting the deoptimization kind of all the breakpoints
set after a first breakpoint. This resulted in incorrect
deoptimization/undeoptimization and even an abort. This was caught
by running the debugger with sanity checks enabled with libartd.so.

We now set next breakpoints with the deoptimization kind of the first
existing breakpoint (if any) so we trigger right [un]deoptimization
when adding or removing a breakpoint.

Bug: 18782753
Bug: 18651686

(cherry picked from commit abe93e0098a2648fa286cfea01954737e32c7be9)

Change-Id: I95424e361dc04df0adb54b5afac5b4da2b8b2731

9 years agoMerge "Increase default allocation stack size" into lmp-mr1-dev
Mathieu Chartier [Thu, 11 Dec 2014 20:39:08 +0000 (20:39 +0000)]
Merge "Increase default allocation stack size" into lmp-mr1-dev

9 years agoIncrease default allocation stack size
Mathieu Chartier [Thu, 11 Dec 2014 02:45:30 +0000 (18:45 -0800)]
Increase default allocation stack size

Prevents some GC for alloc which can occur if the allocation stack
overflows.

Bug: 18707404
Change-Id: I00e8e518aa5105b69c70603a8395c81617b87d14

9 years agoMerge "ART: More ELF and oat file safety measures" into lmp-mr1-dev
Andreas Gampe [Thu, 11 Dec 2014 18:27:18 +0000 (18:27 +0000)]
Merge "ART: More ELF and oat file safety measures" into lmp-mr1-dev

9 years agoART: More ELF and oat file safety measures
Andreas Gampe [Thu, 11 Dec 2014 04:51:45 +0000 (20:51 -0800)]
ART: More ELF and oat file safety measures

In an ELF file, look for a shstrtab section when loading in
program-header-only mode. If the section is outside the file size,
it strongly indicates a broken compile.

When compiling oat files in the class linker, explicitly unlink
on failure. This should catch cases when dex2oat is killed or
crashes and doesn't have a chance to delete its (partial) output.

Bug: 15567083
Change-Id: Ia0c75f151d91c6f26a71696967255d6d409ca882

9 years agoMerge "Link libsigchain.a instead of sigchain.o" into lmp-mr1-dev
Dmitriy Ivanov [Wed, 10 Dec 2014 22:13:46 +0000 (22:13 +0000)]
Merge "Link libsigchain.a instead of sigchain.o" into lmp-mr1-dev

9 years agoMerge "Fix verifier bug caused by confusing ArtMethod::IsDirect vs ArtMethod::IsStati...
Brian Carlstrom [Wed, 10 Dec 2014 20:37:04 +0000 (20:37 +0000)]
Merge "Fix verifier bug caused by confusing ArtMethod::IsDirect vs ArtMethod::IsStatic semantics." into lmp-mr1-dev

9 years agoMerge "Ensure void type is resolved when running without image" into lmp-mr1-dev
Sebastien Hertz [Wed, 10 Dec 2014 20:16:04 +0000 (20:16 +0000)]
Merge "Ensure void type is resolved when running without image" into lmp-mr1-dev

9 years agoLink libsigchain.a instead of sigchain.o
Dmitriy Ivanov [Mon, 20 Oct 2014 21:11:00 +0000 (14:11 -0700)]
Link libsigchain.a instead of sigchain.o

  Relative paths may sometimes lead to failed builds
  if sources are built in particular order.

  Also it is possible to export symbols from static lib
  with LOCAL_WHOLE_STATIC_LIBRARIES variable.

(cherry picked from commit e49a4f6db74f4e66486a6aaf6564f997a94e230f)

Bug: 18701723
Change-Id: I6d62594c2e84020d67254747c1fd465e1144920e

9 years agoEnsure void type is resolved when running without image
Sebastien Hertz [Tue, 9 Dec 2014 18:08:15 +0000 (19:08 +0100)]
Ensure void type is resolved when running without image

To initialize java.lang.Void correctly, we need to ensure the 'void'
class is initialized in the dex cache of core-libart (because Java
code relies on it). However, if we run without image, the dex cache
may not contain the void class before we initialize java.lang.Void
class.

This CL fixes this by forcing the resolution of 'void' class when
initializing the runtime without image.

We also generalize that to other types by not skipping verification
when running with preopted oat files without image. By verifying
all classes, we ensure all resolved types are in the dex caches.

Bug: 18338511
Change-Id: I65e345da47be6eda1ee55897c01a9a4cc1444915

9 years agoFix verifier bug caused by confusing ArtMethod::IsDirect vs ArtMethod::IsStatic seman...
Brian Carlstrom [Wed, 10 Dec 2014 04:15:42 +0000 (20:15 -0800)]
Fix verifier bug caused by confusing ArtMethod::IsDirect vs ArtMethod::IsStatic semantics.

Bug: 18485243

(cherry picked from commit be6fa5eb4a571e14481cf43f4cb264629c069153)

Change-Id: I6e4d8e7587f7e03288ce039471154c1e6ebc7d8a

9 years agoMerge "Support proxy method in StackVisitor::GetThisObject" into lmp-mr1-dev
Sebastien Hertz [Tue, 9 Dec 2014 21:53:37 +0000 (21:53 +0000)]
Merge "Support proxy method in StackVisitor::GetThisObject" into lmp-mr1-dev

9 years agoMerge "ART: GenLongArith has to clobber same operands for 2-op instr" into lmp-mr1-dev
Bill Buzbee [Tue, 9 Dec 2014 20:06:41 +0000 (20:06 +0000)]
Merge "ART: GenLongArith has to clobber same operands for 2-op instr" into lmp-mr1-dev

9 years agoMerge "ART: GenNegLong incorrectly handled register overlap for x86" into lmp-mr1-dev
Bill Buzbee [Tue, 9 Dec 2014 20:06:25 +0000 (20:06 +0000)]
Merge "ART: GenNegLong incorrectly handled register overlap for x86" into lmp-mr1-dev

9 years agoART: GenLongArith has to clobber same operands for 2-op instr
nikolay serdjuk [Thu, 13 Nov 2014 12:15:23 +0000 (18:15 +0600)]
ART: GenLongArith has to clobber same operands for 2-op instr

When we generate code for 'add-long/2addr v4, v4' and v4 is in
memory, we load v4 in a physical register, do operation and
store result in memory. After that, we should clobber the
physical register because it is not valid anymore.

Signed-off-by: nikolay serdjuk <nikolay.y.serdjuk@intel.com>
(cherry picked from commit 6b9356cf0feb2a4100798481869fd4d18180c29f)

Bug: 18671047
Bug: https://code.google.com/p/android/issues/detail?id=81542
Change-Id: Ie11e29cd32358ddbdccf66eaad46aa07f1269133

9 years agoART: GenNegLong incorrectly handled register overlap for x86
nikolay serdjuk [Mon, 10 Nov 2014 09:53:27 +0000 (16:53 +0700)]
ART: GenNegLong incorrectly handled register overlap for x86

There is a bug in the GenNegLong: it should invoke
   OpRegCopy(temp_reg, rl_result.reg.GetHigh());
instead of
   OpRegCopy(temp_reg, rl_result.reg);
But, anyway there is no need to handle the overlap
anymore because it is already handled in OpRegCopyWide()
which is invoked from StoreValueWide().
Signed-off-by: nikolay serdjuk <nikolay.y.serdjuk@intel.com>
(cherry picked from commit d24c9346138b7b8723461d94d9866762a59929d9)

Bug: 18671045
Bug: https://code.google.com/p/android/issues/detail?id=81544
Change-Id: I6d67a2a1507043c15de4616d0df59459630d4104

9 years agoSupport proxy method in StackVisitor::GetThisObject
Sebastien Hertz [Tue, 25 Nov 2014 15:30:53 +0000 (16:30 +0100)]
Support proxy method in StackVisitor::GetThisObject

Adds function artQuickGetProxyThisObject which returns the 'this'
object of the proxy method using the QuickArgumentVisitor. Since
proxy methods have the same layout than the kRefsAndArgs runtime
method and 'this' is the 1st method argument, it is located in the
first GPR.

Bug: 17965861
(cherry picked from commit a836bc9760419af4a515f96c66100a39e865f3b9)

Change-Id: I09b5c1cdfc051d9395bba929d4650804eb3b1f7f

9 years agoMerge "JDWP: avoid crash on unsupported modifier" into lmp-mr1-dev
Sebastien Hertz [Tue, 9 Dec 2014 09:28:48 +0000 (09:28 +0000)]
Merge "JDWP: avoid crash on unsupported modifier" into lmp-mr1-dev

9 years agoFix reference table sorting
Brian Carlstrom [Tue, 9 Dec 2014 00:22:06 +0000 (16:22 -0800)]
Fix reference table sorting

Bug: 18638116
Change-Id: I30894d5c33a2702b0c0be7fc9789b42256eea8b1

9 years agoMerge "Make Barrier robust against spurious wakeups" into lmp-mr1-dev
Hans Boehm [Fri, 5 Dec 2014 22:22:14 +0000 (22:22 +0000)]
Merge "Make Barrier robust against spurious wakeups" into lmp-mr1-dev

9 years agoART: Fix variable formatting and CodeLayout's multiple visits
Jean Christophe Beyler [Thu, 4 Sep 2014 15:15:11 +0000 (08:15 -0700)]
ART: Fix variable formatting and CodeLayout's multiple visits

The CodeLayout pass visits multiple times the same BasicBlock. This patch
  fixes that issue to reduce its overhead.

The patch also fixes cUnit to c_unit in the bb_optimization files.

Signed-off-by: Jean Christophe Beyler <jean.christophe.beyler@intel.com>
Bug: 18507840

(cherry picked from commit 75bcc3780bc40dd7c265e150aff5b891135ff6e3)

Change-Id: I4675ba0b4886c35f0093ac54e171dd87548f60c8

9 years agoJDWP: avoid crash on unsupported modifier
Sebastien Hertz [Fri, 5 Dec 2014 09:52:15 +0000 (10:52 +0100)]
JDWP: avoid crash on unsupported modifier

Returns NOT_IMPLEMENTED error if we receive an event request with an
unsupported modifier.

Bug: https://code.google.com/p/android/issues/detail?id=81037
Bug: 18617787
Change-Id: I97729e2f98af3a75d24604926a89860255d4acae

9 years agoMake Barrier robust against spurious wakeups
Hans Boehm [Wed, 3 Dec 2014 02:31:31 +0000 (18:31 -0800)]
Make Barrier robust against spurious wakeups

Fix Barrier implementation so that Wait cannot return prematurely
due to a spurious condition variable wakeup or interrupted futex call.

Document the resulting semantics of barrier.h, which are a bit surprising,
but appear to be exactly what current clients need.

Fix the test so that it actually passes with the fixed barrier.h,
and no longer tests for properties that can't be correctly satisfied.

Improve comment for InitTimeSpec, which we almost used.

(Differs from AOSP patch in that Increment returns void here,
as does cv TimedWait.  In AOSP they return bool.)

Bug:18509123

Change-Id: I0b25f33bcd22322ba04e3951cd484843788c2bf5
(cherry picked from commit 5567c11b9157eec110c0631aa2bff5836631e868)

9 years agoam c1bfaa22: Remove dex2oat watchdog warning
Brian Carlstrom [Thu, 4 Dec 2014 23:06:44 +0000 (23:06 +0000)]
am c1bfaa22: Remove dex2oat watchdog warning

* commit 'c1bfaa22570b4f771ae1cd832201b01ec84e32c1':
  Remove dex2oat watchdog warning

9 years agoRemove dex2oat watchdog warning
Brian Carlstrom [Thu, 4 Dec 2014 06:29:37 +0000 (22:29 -0800)]
Remove dex2oat watchdog warning

The warning output is breaking vm-tests on some devices. It does not
seem to serve much of a purpose given that it does not go to logcat
and long running compiles can be identified by the time summary
printed at the end of dex2oat.

Bug: 18609574
Change-Id: Ic4f3d118d99e4b307cbb7c4c2b57f05a003e092e

9 years agoRemove method verification results right after compiling a method
Mathieu Chartier [Thu, 4 Dec 2014 01:38:22 +0000 (17:38 -0800)]
Remove method verification results right after compiling a method

This saves memory since it allows the code arrays from methods
compiled in future methods to use the ram we just freed from the
verification results.

GmsCore.apk:
Before: dex2oat took 77.383s (threads: 2) arena alloc=6MB java alloc=30MB native alloc=77MB free=13KB
After:  dex2oat took 72.180s (threads: 2) arena alloc=6MB java alloc=30MB native alloc=60MB free=13KB

Bug: 18596910
Change-Id: I5d6df380e4fe58751a2b304202083f4d30b33b7c

9 years agoART: Fix unclosed files in dex2oat
Andreas Gampe [Wed, 3 Dec 2014 22:28:02 +0000 (14:28 -0800)]
ART: Fix unclosed files in dex2oat

Under some error conditions files are not closed, and the close guard
will complain.

Bug: 18603475
Change-Id: I7155e3f4f6231844edd375b9302ecf1ba8079950

9 years agoFix mac build
Mathieu Chartier [Wed, 3 Dec 2014 02:23:21 +0000 (18:23 -0800)]
Fix mac build

Macs don't have malloc.h, hopefully this fixes the build.

Change-Id: Id7600ea79197f36d3dfb5e7d589ac2f6ee325f85

9 years agoPrint memory usage in dex2oat shutdown
Mathieu Chartier [Tue, 2 Dec 2014 23:43:48 +0000 (15:43 -0800)]
Print memory usage in dex2oat shutdown

Example on mako:
I/dex2oat (31071): dex2oat took 31.195s (threads: 2) arena alloc=1013KB java alloc=13MB native alloc=32MB free=1490KB

Bug: 18069309

Change-Id: I08eac00842be35d4e659bddc8513f2062be725c9

9 years agoMerge "Quick: Fix neg-long on ARM for overlapping regs." into lmp-mr1-dev
Vladimir Marko [Tue, 2 Dec 2014 16:50:54 +0000 (16:50 +0000)]
Merge "Quick: Fix neg-long on ARM for overlapping regs." into lmp-mr1-dev

9 years agoMerge "JDWP: only deoptimize when it is required" into lmp-mr1-dev
Sebastien Hertz [Tue, 2 Dec 2014 16:39:09 +0000 (16:39 +0000)]
Merge "JDWP: only deoptimize when it is required" into lmp-mr1-dev

9 years agoMerge "Try normal allocation if large object allocation fails" into lmp-mr1-dev
Mathieu Chartier [Tue, 2 Dec 2014 00:57:21 +0000 (00:57 +0000)]
Merge "Try normal allocation if large object allocation fails" into lmp-mr1-dev

9 years agoTry normal allocation if large object allocation fails
Mathieu Chartier [Mon, 1 Dec 2014 23:00:27 +0000 (15:00 -0800)]
Try normal allocation if large object allocation fails

If a large object allocation fails, we now try the normal allocators.

Bug: 18504942
Change-Id: I18b9759d6af885556941542c57fec584f18197f1

9 years agoFix growth limit / footprint limit problems
Mathieu Chartier [Mon, 1 Dec 2014 21:40:48 +0000 (13:40 -0800)]
Fix growth limit / footprint limit problems

We now properly clear the growth limit of all the spaces when
ClearGrowthLimit is called. Previously we didn't clear the growth
limit of the background space.

This wasn't caught by heap test since it only manifested itself
when we had a zygote space.

Bug: 18504942
Change-Id: Ibb3d76cba54a45ef755d7e2ced8c2e16cd2996b0

9 years agoSet dex_cache_strings_ when we call Class::SetDexCache
Mathieu Chartier [Mon, 1 Dec 2014 18:31:15 +0000 (10:31 -0800)]
Set dex_cache_strings_ when we call Class::SetDexCache

Ensures that these two variables never get out of sync. The error
was presumably related to not doing this for proxy classes. This
caused java code which was looking at the dex_cache_strings_ field
to incorrectly access a null array.

Bug: 18548887
Change-Id: If53c6ade3588b82a480d6674dfbd5caa2e5069fd

9 years agoQuick: Fix neg-long on ARM for overlapping regs.
Vladimir Marko [Mon, 1 Dec 2014 16:48:48 +0000 (16:48 +0000)]
Quick: Fix neg-long on ARM for overlapping regs.

Bug: 18569347

(cherry picked from commit 2f340a843ea5b3413c901f8c2365243b68864468)

Change-Id: Icde3bfdd7c90d51548823ce1f81caf9484de2be5

9 years agoMerge "Disable stack mprotects when running on valgrind" into lmp-mr1-dev
Dmitriy Ivanov [Thu, 27 Nov 2014 18:16:27 +0000 (18:16 +0000)]
Merge "Disable stack mprotects when running on valgrind" into lmp-mr1-dev

9 years agoAdd missing thread state change to kWaitingForCheckPointsToRun
Mathieu Chartier [Wed, 26 Nov 2014 23:37:52 +0000 (15:37 -0800)]
Add missing thread state change to kWaitingForCheckPointsToRun

Fixes tests

Change-Id: I4907d83ee1e9bcd2f6a823e402950d8051d4556a

9 years agoTrim reference tables when we trim the heap
Mathieu Chartier [Wed, 26 Nov 2014 19:21:15 +0000 (11:21 -0800)]
Trim reference tables when we trim the heap

Before:
System server:
 virtual                     shared   shared  private  private
  size      RSS      PSS    clean    dirty    clean    dirty    # object
2200      300      229        0       80        0      220   77 /dev/ashmem/dalvik-indirect ref table (deleted)
Location:
1896      128      102        0       28        0      100   39 /dev/ashmem/dalvik-indirect ref table (deleted)

After:
 virtual                     shared   shared  private  private
  size      RSS      PSS    clean    dirty    clean    dirty    # object
System server:
2216       64       64        0        0        0       64   79 /dev/ashmem/dalvik-indirect ref table (deleted)
Location:
2120       48       48        0        0        0       48   67 /dev/ashmem/dalvik-indirect ref table (deleted)

No pause time regression measured in memalloc test.

Bug: 17643507

Change-Id: I32d3e64cdcf8dd2f7aea509c81631597bbb9b392

9 years agoAdd a way to change the IMT size
Mathieu Chartier [Wed, 26 Nov 2014 02:36:01 +0000 (18:36 -0800)]
Add a way to change the IMT size

Useful for having smaller imts on memory constrainted devices.
Setting ART_IMT_SIZE=x will change the size of the IMT.

Bug: 17643507
Change-Id: Ia6bcfcb067c7e21c7f3003cf6254e4b2666ef169

9 years agoDisable stack mprotects when running on valgrind
Dmitriy Ivanov [Wed, 26 Nov 2014 18:30:11 +0000 (10:30 -0800)]
Disable stack mprotects when running on valgrind

Bug: 18155388
Change-Id: I17d7e3c707e83b864b0394ebbfb2438021faf76f

9 years agoMerge "Move dex cache strings from ArtMethod -> Class" into lmp-mr1-dev
Mathieu Chartier [Tue, 25 Nov 2014 23:32:03 +0000 (23:32 +0000)]
Merge "Move dex cache strings from ArtMethod -> Class" into lmp-mr1-dev

9 years agoMerge "Quick compiler: handle embedded switch data." into lmp-mr1-dev
buzbee [Tue, 25 Nov 2014 23:04:39 +0000 (23:04 +0000)]
Merge "Quick compiler: handle embedded switch data." into lmp-mr1-dev

9 years agoQuick compiler: handle embedded switch data.
buzbee [Tue, 25 Nov 2014 22:39:38 +0000 (14:39 -0800)]
Quick compiler: handle embedded switch data.

Although switch data is generally placed at the end of a dex
file by dx, it can occur elsewhere (and does via obsfucators).
This CL fixes a parsing error related to embedded switch data by
ensuring valid dex instructions following the embedded data appear
in their own basic blocks.

b/18524584

Change-Id: I815bb5ec939803f706f98beb3afbefc8b527f730

9 years agoFix oatwriter code deduping
Mathieu Chartier [Tue, 25 Nov 2014 19:20:28 +0000 (11:20 -0800)]
Fix oatwriter code deduping

Now that the GC maps are part of the oat method header they need
to be checked in CodeOffsetsKeyComparator.

Bug: 18523556

Change-Id: I539a6e7216166342b22515c1e2cf831dad32e41e

9 years agoMove dex cache strings from ArtMethod -> Class
Mathieu Chartier [Tue, 25 Nov 2014 02:29:54 +0000 (18:29 -0800)]
Move dex cache strings from ArtMethod -> Class

Adds one load for const strings which are not direct.

Saves >= 60KB of memory avg per app.
Image size: -350KB.

Bug: 17643507
Change-Id: I2d1a3253d9de09682be9bc6b420a29513d592cc8

9 years agoJDWP: only deoptimize when it is required
Sebastien Hertz [Mon, 17 Nov 2014 18:00:37 +0000 (19:00 +0100)]
JDWP: only deoptimize when it is required

We don't need to deoptimize anything when we forced the use of the
interpreter (-Xint). In this case, no compiled code is executed
(except native methods which are not concerned by deoptimization).
Therefore we even don't need to enable/disable deoptimization support
in instrumentation.

We also don't need to deoptimize a method that hasn't been compiled.
Since it will run with interpreter, there is no point deoptimizing
it. However this method may be inlined in a compiled caller method
so we still need to deoptimize everything in this case.

This CL updates breakpoint support by storing the required kind of
deoptimization for a particular method. There are 3 cases:
- kNothing: the method does not require deoptimization.
- kSelectiveDeoptimization: the method needs to be deoptimized.
- kFullDeoptimization: we must deoptimize everythinig.
When uninstalling a breakpoint, we need to do the reverse operation.

Also fixes the SanityCheckExistingBreakpoints function to control
breakpoints related to the given method only and adds extra verbose
ilogs when choosing the appropriate deoptimization kind.

Includes a partial cherry-pick of commit
87553c9fa1298ffb40127b2bb6413859fd3f79df to use method
ClassLinker::GetOatMethodQuickCodeFor.

Bug: 18407046

(cherry picked from commit f3928794a10516e2ac0ffe2686a10891788d4b9c)

Change-Id: I50853cc5fc5c52650485785a1198d35ea0f7fb8e

9 years agoFix failing tests
Mathieu Chartier [Mon, 24 Nov 2014 18:31:19 +0000 (10:31 -0800)]
Fix failing tests

Forgot to fix these with other CL.

Change-Id: I5ac158ca0d312b601fbf1e12722fe271751c88f2

9 years agoDelete ArtMethod gc_map_ field
Mathieu Chartier [Sat, 22 Nov 2014 00:51:29 +0000 (16:51 -0800)]
Delete ArtMethod gc_map_ field

Moved the gc_map field from OatMethod to OatQuickMethodHeader.
Deleted the ArtMethod gc_map_ field.

Bug: 17643507

Change-Id: Ifa0470c3e4c2f8a319744464d94c6838b76b3d48

9 years agoAdd fast string sharpening
Mathieu Chartier [Thu, 20 Nov 2014 23:04:10 +0000 (15:04 -0800)]
Add fast string sharpening

String sharpening changes const strings to PC relative loads instead
of always going through the dex cache. This saves code size and
probably improves performance slightly.

Before: 49602992 system@framework@boot.oat
After: 49385904 system@framework@boot.oat

Pre-cursor to removing dex_cache_strings_ field from ArtMethod.

Bug: 17643507

Change-Id: I1787f48774631eee0accafeea257aa8d0e91e8d6

9 years agoFix oatdump to use OatHeader pointer size
Mathieu Chartier [Fri, 21 Nov 2014 01:08:58 +0000 (17:08 -0800)]
Fix oatdump to use OatHeader pointer size

Bug: 18473190
Change-Id: If505b4f62105899f4f1257d3bccda3e6eb0dcd7c

9 years agoTune rosalloc buffer sizes.
Hiroshi Yamauchi [Thu, 20 Nov 2014 18:08:31 +0000 (10:08 -0800)]
Tune rosalloc buffer sizes.

We now use one-page buffers for size brackets 4-7, instead of two-page
buffers, and the first 8 size brackets for thread-local allocations,
instead of 11.

No slowdown observed with MemAllocTest, EvaluateAndApplyChanges, and
BinaryTrees.

Bug: 18377775

Change-Id: Ie1bb46bcf5d3729197e48e26a27da4cc39dd807e

9 years agoMerge "dex2oat: Pack likely-dirty objects together when generating the boot image...
Igor Murashkin [Thu, 20 Nov 2014 00:06:33 +0000 (00:06 +0000)]
Merge "dex2oat: Pack likely-dirty objects together when generating the boot image" into lmp-mr1-dev

9 years agodex2oat: Pack likely-dirty objects together when generating the boot image
Igor Murashkin [Fri, 14 Nov 2014 23:01:59 +0000 (15:01 -0800)]
dex2oat: Pack likely-dirty objects together when generating the boot image

This introduces a new algorithm into image writer which "bins" objects
by how likely they are to be dirtied at runtime. Objects in the same bin
are placed contiguously in memory (i.e. into the same page). We try to
tune the bin selection based on how clean or how dirty the object will
likely be at runtime.

As-is, this saves about 150KB per-process (private-dirty pages) and 700KB in
zygote (shared-dirty).

There is still about 800KB of objects that are clean but located in
dirty pages, so with more analysis we can tune the bin selection and get
even more memory savings.

Bug: 17611661
Change-Id: Ia1455e4c56ffd0a36ae2a723d35b7e06502980f7

9 years agoFix zygote space and non moving space map names
Mathieu Chartier [Wed, 19 Nov 2014 18:51:42 +0000 (10:51 -0800)]
Fix zygote space and non moving space map names

Space names:
"non moving space" -> "zygote space"
"alloc space" -> "non moving space"

Bug: 18447855
Change-Id: Ia937b6d046ccf7f66bf1f6bbb9f17a5e0d00c016

9 years agoMerge "Fix transitive equality error in set." into lmp-mr1-dev
Mathieu Chartier [Wed, 19 Nov 2014 18:37:22 +0000 (18:37 +0000)]
Merge "Fix transitive equality error in set." into lmp-mr1-dev

9 years agoQuick compiler: fix x86 special identity
buzbee [Wed, 19 Nov 2014 13:26:05 +0000 (05:26 -0800)]
Quick compiler: fix x86 special identity

The Quick compiler recognizes and specially handles a set of
small methods - including those that simply return one of their
arguments.  For x86, special identity was broken if the returned
argument was a double that was passed half in register and half
in memory.

internal b/17325447

(cherry picked from commit e8f74e58b476520cfc53970304c67b800994e33d)

Change-Id: I9afea5ef34498306528e0804489b5144e9e9aec3

9 years agoFix transitive equality error in set.
Mathieu Chartier [Tue, 18 Nov 2014 22:13:58 +0000 (14:13 -0800)]
Fix transitive equality error in set.

SubstringComparator used to be broken since it didn't have transitive
equality (ex. ab a ac). Now we use normal string comparison with
set.lower_bound saves an additional 12 bytes.

Also up frame size to 2700 to fix build.

Change-Id: Ie8d41faef37591e7feed1f5d2ad6258430b8ffb6