OSDN Git Service
Andreas Gampe [Thu, 9 Jun 2016 01:18:47 +0000 (18:18 -0700)]
ART: Fix swap usage determination in dex2oat
The decision needs to be postponed until we have access to the
dex files.
Bug:
27878043
Change-Id: I55833c6ea827ec7a322711bd182fc723d1727f08
Anton Shamin [Mon, 16 May 2016 10:44:13 +0000 (16:44 +0600)]
ART: ArrayGet hoisting restriction added.
Currently if we hoist ArrayGet from loop there is no guarantee
that insn will be executed at runtime. Because of that we could
face issues like crashes in generated code.
This patch introduces restriction for ArrayGet hoisting. We say
that ArrayGet execution is guaranteed at least one time if its bb
dominates all exit blocks.
Signed-off-by: Anton Shamin <anton.shamin@intel.com>
(cherry picked from commit
f89381fed12faf96c45a83a989ae2fff82c05f3b)
BUG=
29145171
Change-Id: Ia5664dedb1543d78a7b4038801b8372572f069f6
David Sehr [Mon, 6 Jun 2016 20:16:10 +0000 (20:16 +0000)]
Merge "Enable profman pretty printing" into nyc-dev
Mathieu Chartier [Sat, 4 Jun 2016 00:47:32 +0000 (17:47 -0700)]
Revert "Revert "Hold dex caches live in class table""
Bug:
29083330
This reverts commit
f102faf1bcbdb2149e3e7bf27b1819f621b7894b.
Change-Id: Ibc6d260247b3113beec7d33552061512a36c9ce8
David Sehr [Thu, 2 Jun 2016 17:46:19 +0000 (10:46 -0700)]
Enable profman pretty printing
Bug:
28748264
Change-Id: I4dc366aa710c9ef05aa5d1b164d558232fe57a3d
Brian Carlstrom [Sat, 4 Jun 2016 00:22:32 +0000 (00:22 +0000)]
Revert "Hold dex caches live in class table"
This reverts commit
d6d49e56c2b7b11f474acb80cb02bb1fe9b7861e.
Bug:
29083330
Change-Id: Ie209b27897b8079f2d13fd0837fe5f83a7e61afc
Mathieu Chartier [Thu, 2 Jun 2016 18:48:30 +0000 (11:48 -0700)]
Hold dex caches live in class table
Prevents temporary dex caches being unloaded for the same dex file.
Usually this is OK, but if someone resolved a string in that dex
cache, it could leave stale pointers in BSS. Also it can use extra
memory in linear alloc if we allocate dex cache arrays multiple
times.
Bug:
29083330
(cherry picked from commit
f284d448e3edd428b6ade473d0993028638b2064)
Change-Id: Ie1b0b0cf835a998e19227cbb90014011a6cd40c4
Vladimir Marko [Wed, 1 Jun 2016 17:38:43 +0000 (18:38 +0100)]
Delay dex-to-dex compilation until Optimizing is done.
This fixes a race between inlining in the Optimizing
backend and dex-to-dex quickening where the Optimizing can
read the non-quickened opcode and then the quickened field
index or vtable index and look up the wrong field or method.
Even if we such tearing of the dex instruction does not
happen, the possible reordering of dex-to-dex and Optimizing
compilation makes the final oat file non-deterministic.
Also, remove VerificationResults::RemoveVerifiedMethod() as
we have only the Optimizing backend now and as such it was
dead code and would have interfered with this change.
Bug:
29043547
Bug:
29089975
(cherry picked from commit
492a7fa6df3b197a24099a50f5abf624164f3842)
Change-Id: I1337b772dc69318393845a790e5f6d38aa3de60f
Nicolas Geoffray [Thu, 2 Jun 2016 14:55:48 +0000 (15:55 +0100)]
Fix ArtMethod::GetInvokeType for static methods on interfaces.
And avoid calling ResolveMethod in the JIT, since it already
knows that method.
bug:
29089267
bug:
27521545
Change-Id: I5c5dae385804f1d2062defca459f69c26cdb4fef
Nicolas Geoffray [Wed, 1 Jun 2016 15:30:20 +0000 (16:30 +0100)]
Fix a bug in reference type propagation.
The upper bound of a bound type is always exact, so we should
not use it for setting the type of the bound type.
bug:
28730986
(cherry picked from commit
a90d489ab101891bbac7276b8748125e6b4fa0c2)
Change-Id: Ia3e9f88a274730be39e5c9b147204eec2a145de2
Calin Juravle [Thu, 2 Jun 2016 10:10:35 +0000 (10:10 +0000)]
Merge "Don't do any I/O when starting the profile saver" into nyc-dev
Mathieu Chartier [Wed, 1 Jun 2016 22:38:10 +0000 (22:38 +0000)]
Merge "Fix race with host_dlopen_handles_" into nyc-dev
Mathieu Chartier [Wed, 1 Jun 2016 17:48:19 +0000 (10:48 -0700)]
Fix race with host_dlopen_handles_
Thread 1 opens an already opened oat file.
Thread 2 dlcloses the oat file and removes it from
host_dlopen_handles_.
Thread 1 checks that it is not already in host_dlopen_handles_ and
proceeds to return the oat file. The problem now is that the BSS is
not cleared since it is the same oat file that was opened earlier.
The fix is to just hold the lock for dlopen / dlclose. This only
affects contention on host.
Bug:
28992179
Bug:
28990799
Bug:
28826195
(cherry picked from commit
c7d3f4b8c294f36209ea2c129f9714a36ae1ec6b)
Change-Id: Ib462720fab26427d8ca0c6323080604878edb8b5
David Sehr [Tue, 24 May 2016 21:52:31 +0000 (14:52 -0700)]
Initial profman support for verbose dump
Bug:
28748264
Change-Id: Ibc59e7f31afa4646fcf048eececb87437dd83de9
TreeHugger Robot [Sat, 28 May 2016 01:15:50 +0000 (01:15 +0000)]
Merge "Relax annotation visibility so runtime includes build." into nyc-dev
Jeff Hao [Fri, 27 May 2016 01:39:17 +0000 (18:39 -0700)]
Relax annotation visibility so runtime includes build.
Previous behavior in M and earlier would allow annotations marked
VISIBILITY_BUILD to be visible to the runtime when they should not
have been. When targeting older sdks, revert to this behavior.
Bug:
28826610
Change-Id: I95e10da899ed11107ca78c62dd2e263808008310
Richard Uhler [Fri, 27 May 2016 20:50:15 +0000 (20:50 +0000)]
Merge "Don't use dlopen on host for already loaded oat files." into nyc-dev
Richard Uhler [Tue, 24 May 2016 22:04:22 +0000 (15:04 -0700)]
Don't use dlopen on host for already loaded oat files.
Because the behavior of dlopen on the host is different then the
target in that case, and it causes tests to fail.
Bug:
28826195
(cherry picked from commit
a206c745dbb64b14f05c87891d425475c2f6d63a)
Change-Id: I29b04be07b4d26dc1ac5e6f35550745eb15e6728
Narayan Kamath [Fri, 27 May 2016 11:14:56 +0000 (11:14 +0000)]
Merge "Remove duplicate libcore failures expectations." into nyc-dev
Nicolas Geoffray [Thu, 14 Apr 2016 12:32:01 +0000 (13:32 +0100)]
Don't use JACK alpha anymore.
(cherry picked from commit
99e6d2d9b2585f605e02041f351b494f82f82121)
Bug:
28977632
Change-Id: Id37d90446be5d63cb0cf0131e34b562147e88335
Roland Levillain [Thu, 26 May 2016 16:38:46 +0000 (17:38 +0100)]
Remove duplicate libcore failures expectations.
Some jsr166 entries were already listed in
libcore/expectations/knownfailures.txt.
Bug:
28977632
Change-Id: I35788008262bac6f10d96f53e8e8f21e2364f091
Calin Juravle [Wed, 25 May 2016 15:45:16 +0000 (16:45 +0100)]
Don't do any I/O when starting the profile saver
The profile saver is started when the application is launched. During
the start we used to prepare the paths and transform them to canonical
paths to make sure we do the foreign dex checks on the actual paths.
Postpone the canonical path transformation to a later point to make sure
the impact is minimal.
Bug:
28785119
Bug:
28625993
Change-Id: I1aa378cdefde6ed6dc708762392db0f3a3158812
Vladimir Marko [Thu, 26 May 2016 11:35:24 +0000 (11:35 +0000)]
Merge "Fix OatWriter to update the header checksum correctly." into nyc-dev
Mathieu Chartier [Thu, 26 May 2016 01:38:14 +0000 (01:38 +0000)]
Merge "Prune class path classes from profile" into nyc-dev
Mathieu Chartier [Wed, 25 May 2016 22:05:59 +0000 (15:05 -0700)]
Prune class path classes from profile
Fixes a theoretical case where classes not in the app, but in the in
the class path could have been included in the image. The dex caches
for these classes are not properly handled and need to be pruned.
Not including the classes in the image classes makes sure the that
class linker automatically prunes them and frees the dex cache with
the explicit garbage collection.
Bug:
28452385
Change-Id: Ia44813863b5f1c79367049838021a64cc3842fc9
TreeHugger Robot [Wed, 25 May 2016 21:35:45 +0000 (21:35 +0000)]
Merge "Fix profile saver reference point when calculating the sleep time" into nyc-dev
Richard Uhler [Wed, 25 May 2016 19:18:07 +0000 (19:18 +0000)]
Merge changes Idf64b96b,Ie717089d into nyc-dev
* changes:
Base isDexOptNeeded result on OatFileAssistant::IsUpToDate.
Only compile dex files if they are not up to date.
Calin Juravle [Wed, 25 May 2016 17:09:53 +0000 (18:09 +0100)]
Fix profile saver reference point when calculating the sleep time
It is wrong to use last_time_ns_saver_woke_up_ as a reference. We should
use sleep_start.
Bug:
28953776
Change-Id: I88ab36c18b3b7b6e310bcdc6c59cadedc4cc9260
Vladimir Marko [Wed, 25 May 2016 11:49:49 +0000 (12:49 +0100)]
Fix OatWriter to update the header checksum correctly.
Make sure we update the checksum with dex file data, type
lookup tables and thunks produced by the relative patcher.
Bug:
28874264
Change-Id: I5b83237132ccb784acea6c195d0ed909ee33b99a
TreeHugger Robot [Wed, 25 May 2016 16:51:24 +0000 (16:51 +0000)]
Merge "Remove bogus DCHECK." into nyc-dev
Nicolas Geoffray [Wed, 25 May 2016 13:52:37 +0000 (14:52 +0100)]
Remove bogus DCHECK.
bug:
28771056
(cherry picked from commit
0950abe82f86a9943feb7719876ec291c6581037)
Change-Id: I8c5b737c0882ecb8d67d8f799356f30afe3dd87b
Brian Carlstrom [Wed, 25 May 2016 02:45:30 +0000 (19:45 -0700)]
Avoid long sleep in ProfileSaver
This was due to subtracting a >20s number of nanos from the constant for 20s.
Bug:
28880120
Change-Id: Ia30b80661f869843484cbac029008530128bd300
Mathieu Chartier [Wed, 25 May 2016 00:41:21 +0000 (17:41 -0700)]
Fix broken DCHECK
Dex cache may now contain references into boot image. Only check
classes that are actually in the application image.
Bug:
28295348
Change-Id: I9d92137c99a53924b13182ded6afabdeb441437d
Mathieu Chartier [Tue, 24 May 2016 23:30:03 +0000 (23:30 +0000)]
Merge "Add methods with samples during launch to profile" into nyc-dev
Mingyao Yang [Tue, 24 May 2016 23:09:36 +0000 (23:09 +0000)]
Merge "Fix an assert during jdwp debugging." into nyc-dev
Mingyao Yang [Tue, 24 May 2016 23:09:23 +0000 (23:09 +0000)]
Merge "Revert "Revert "Move rewritten StringFactory call results into dex registers for deopt""" into nyc-dev
Richard Uhler [Tue, 24 May 2016 22:42:37 +0000 (15:42 -0700)]
Base isDexOptNeeded result on OatFileAssistant::IsUpToDate.
So that it is consistent with when ART will try to run dex2oat.
Bug:
28826195
Change-Id: Idf64b96bf9a6f4ee67dd1af78e6177e078a3c593
Richard Uhler [Tue, 17 May 2016 17:34:52 +0000 (10:34 -0700)]
Only compile dex files if they are not up to date.
Instead of trying to compile them all the time, and relying on file
permissions to keep us from overwriting up-to-date dex files with
different compiler filters.
Bug:
27641809
Bug:
28639246
Bug:
28826195
Change-Id: Ie717089d58517427b6f9a522b48146761c57a1a9
Mathieu Chartier [Wed, 18 May 2016 15:51:52 +0000 (08:51 -0700)]
Add methods with samples during launch to profile
For the snapshot taken after application launch we look at all of
the methods with one or more samples.
Launch time:
books: 1933ms -> 1457ms (33% speedup).
plus: 1361ms -> 1260ms (8% speedup)
music: 3085ms -> 2963ms (4% speedup)
maps: 1583ms -> 1390ms (14% speedup)
Oat sizes:
books:
10871424 ->
13365888 (+23%)
plus:
17715624 ->
19493288 (+13%)
music:
17076864 ->
18289280 (+7%)
maps:
32264832 ->
35787392 (+12%)
Bug:
28750506
Change-Id: I3a955dc74b41a96911254ac5087de77bea2267c7
Aart Bik [Tue, 24 May 2016 19:51:37 +0000 (19:51 +0000)]
Merge "Fix bug in verifier: soft fail was hiding hard fail." into nyc-dev
TreeHugger Robot [Tue, 24 May 2016 19:44:05 +0000 (19:44 +0000)]
Merge "Don't update checksum if data ptr is null." into nyc-dev
TreeHugger Robot [Tue, 24 May 2016 18:26:18 +0000 (18:26 +0000)]
Merge "Put boot class loader classes and strings in dex cache of app images." into nyc-dev
Jeff Hao [Tue, 24 May 2016 02:17:04 +0000 (19:17 -0700)]
Don't update checksum if data ptr is null.
Passing a nullptr to adler32 resets it to its initial value.
Bug:
28874264
Change-Id: I863fa4d489618c1c6fa579cc89fb050f4cd23760
Aart Bik [Mon, 23 May 2016 21:58:49 +0000 (14:58 -0700)]
Fix bug in verifier: soft fail was hiding hard fail.
Rationale:
Dexfuzzing found a situation where a soft fail (modifying
a final field) was hiding a hard fail (type error on that
field), which caused a crash in the compiler later on.
Also added a crash-before/pass-after regression test,
so we don't add the return by accident later.
BUG=
28908555
(cherry picked from commit
c2bc26513ea7c5e2dd9f3b102fb23fd207e4ce63)
Change-Id: Ie79c4afa28cddbca4dfb78e5c75da6644612c15c
Mingyao Yang [Thu, 19 May 2016 17:48:40 +0000 (10:48 -0700)]
Revert "Revert "Move rewritten StringFactory call results into dex registers for deopt""
Potential gc points can make the result value stale. We now set the result value
to null proactively once it's moved to shadow frame registers. IsStringInit()
is written in a way that does string comparison instead of requiring method
resolution so that it doesn't have a gc point. Also we don't cache the callee
method during frame unwinding since the method may be rewritten already.
(cherry picked from commit
ffedec510c07cf0f77d0cce51940838f2e630514)
Bug:
28555675
Change-Id: Ic51511a4a0fc84a852d8d907f91e7835f49ac478
Vladimir Marko [Tue, 24 May 2016 15:40:04 +0000 (15:40 +0000)]
Merge "Compile JNI stubs for verify-profile and interpret-only." into nyc-dev
Dimitry Ivanov [Tue, 24 May 2016 15:35:58 +0000 (15:35 +0000)]
Merge "Remove no-op call to android_update_LD_LIBRARY_PATH" into nyc-dev
Roland Levillain [Wed, 18 May 2016 10:41:33 +0000 (11:41 +0100)]
Enforce class definition ordering rules in v37 Dex files.
These new checks also reject classes extending or
implementing themselves directly or transitively within
a single Dex file.
(cherry picked from commit
621b5eab8eed7746925bca489114d706ad455572)
Bug:
28812524
Bug:
28797012
Bug:
28685551
Bug:
27682580
Change-Id: I51c43f8b4b6a7abaade63682411e61f21767f5d4
Nicolas Geoffray [Tue, 26 Apr 2016 17:30:31 +0000 (18:30 +0100)]
Put boot class loader classes and strings in dex cache of app images.
The reason for b/
28295348 was that an ArtMethod of the boot image was
in the app image dex cache, but the declaring class of that
boot image method was not.
Since objects of boot images don't need fixups, the comparisons for
FixupStrings and FixupResolvedTypes was always false for them.
bug:
28295348
Change-Id: Ida6d42c902692914c302b7fe3d003f73710de6e1
Dimitry Ivanov [Tue, 24 May 2016 00:16:55 +0000 (17:16 -0700)]
Remove no-op call to android_update_LD_LIBRARY_PATH
With linker-namespaces enabled for all target sdk versions
the call to android_update_LD_LIBRARY_PATH is no longer needed.
Bug: http://b/
26040253
Bug: http://b/
27702070
Change-Id: I77c27c10a435e223824e1b69f727d0bc37a26306
Jeff Hao [Mon, 23 May 2016 23:18:09 +0000 (23:18 +0000)]
Merge "Fix oat_writer to use WriteData which updates checksum." into nyc-dev
Jeff Hao [Mon, 23 May 2016 21:30:44 +0000 (14:30 -0700)]
Fix oat_writer to use WriteData which updates checksum.
OatWriter was using WriteFully in some places instead.
Bug:
28874264
Change-Id: Ibaea4536a7838f49feb98efe9c00a976fb5b2453
Vladimir Marko [Mon, 23 May 2016 14:32:42 +0000 (15:32 +0100)]
Compile JNI stubs for verify-profile and interpret-only.
This is the intended behavior to have a good JNI transition
performance.
Bug:
28902384
Change-Id: Iedb2ca343d66cf422ef2dc0527639e81d6ba8960
Roland Levillain [Fri, 20 May 2016 17:16:54 +0000 (18:16 +0100)]
Do not set property jpda.settings.syncPort in JDWP tests.
Let the OS choose the port automatically.
(cherry picked from commit
7235dd15e47d6f4b5e92400269cc206e01c514da)
Bug:
28869570
Change-Id: I79775f9e72f7f8d0f9ec3d68c6f0bdb057ca13ff
Mathieu Chartier [Fri, 20 May 2016 23:20:00 +0000 (23:20 +0000)]
Merge "Only fill methods with 0xFE on debug builds" into nyc-dev
Mathieu Chartier [Thu, 19 May 2016 17:13:04 +0000 (10:13 -0700)]
Only fill methods with 0xFE on debug builds
The GC scans classes without holding any locks, we can not fill
the methods if it is running. Added a GC critical section to address
this. Fixes random crash when scanning classes' methods.
Only for debug builds to not hurt performance.
Bug:
28699001
Change-Id: If96155eaf3fc0e6df31f57dcf32fbd4063b09345
Andreas Gampe [Thu, 19 May 2016 22:35:45 +0000 (22:35 +0000)]
Merge "ART: Check whether an oat file exists before opening" into nyc-dev
Aart Bik [Thu, 19 May 2016 22:34:19 +0000 (22:34 +0000)]
Merge "Do not allow OSR jump while debugging is active." into nyc-dev
Aart Bik [Wed, 18 May 2016 22:44:07 +0000 (15:44 -0700)]
Do not allow OSR jump while debugging is active.
Rationale:
This prevents dangerous transitions between interpreter
and osr compiled code while e.g. single stepping through
a method that is becoming hot.
BUG=
28672961
(cherry picked from commit
29bdaeef925f35eed68f3e52d04416daa0a62317)
Change-Id: I7647710dbe794a1b3320a9666d3e3be85fc40ccd
TreeHugger Robot [Thu, 19 May 2016 18:52:34 +0000 (18:52 +0000)]
Merge "Add missing check for IsShuttingDown in profile saver" into nyc-dev
Calin Juravle [Wed, 18 May 2016 22:49:36 +0000 (15:49 -0700)]
Add missing check for IsShuttingDown in profile saver
Bug:
28814718
Change-Id: I709fc30147047c7a420cd6ff2f0c3b57c54c2021
Mathieu Chartier [Thu, 19 May 2016 17:04:35 +0000 (17:04 +0000)]
Merge "Fix profile saver initial wait" into nyc-dev
TreeHugger Robot [Thu, 19 May 2016 14:30:18 +0000 (14:30 +0000)]
Merge "Revert "Move rewritten StringFactory call results into dex registers for deopt"" into nyc-dev
Roland Levillain [Thu, 19 May 2016 13:00:55 +0000 (13:00 +0000)]
Revert "Move rewritten StringFactory call results into dex registers for deopt"
This reverts commit
0ea451b4d74e8ddc28ea5ce87ef6d21113a3cdd5.
597-deopt-new-string is failing.
Bug:
28846692
Bug:
28555675
Change-Id: I8274e5ae1d065dedc562132882b39e7f3fe5baf4
Andreas Gampe [Thu, 19 May 2016 04:27:00 +0000 (21:27 -0700)]
ART: Fix Mac build
Follow-up to
9821cceebf077cb747aed543582b3d740a03ea7f.
Bug:
28801010
Change-Id: I0bbdf83d64ea81e56824aa19bfae3baa341a2b73
Andreas Gampe [Thu, 19 May 2016 04:10:42 +0000 (21:10 -0700)]
ART: Check whether an oat file exists before opening
The two-stage process of dlopen and our own loader is expensive.
Considering that the OatFileAssistant needs to check two locations,
one of which is unlikely to have a file for the common case of apps
installed in data, do a file existence check before attempting to
load the file.
Remove an unused variable.
Saves about 0.25ms on a N6P.
Bug:
28801010
Change-Id: I0aff9c276c0f539723f8a92663f14ce72160c44d
TreeHugger Robot [Thu, 19 May 2016 03:51:04 +0000 (03:51 +0000)]
Merge "ART: Optimize PreSetup pass for oat-file open" into nyc-dev
Mathieu Chartier [Thu, 19 May 2016 02:51:23 +0000 (19:51 -0700)]
Fix profile saver initial wait
Was using TimedWait for the initial wait, but this can return
early if the process receives any interrupt (due to futex). The
solution is to use a wait loop. If the wait returned early, not
enough classes were including in the profile. This negatively
affected application launch times in some scenarios.
Sample wait times before the change:
Waited 2.002s
Waited 97.808ms
Waited 182.676ms
Waited 2.000s
Waited 1.678s
Bug:
27688727
Change-Id: Ia84878e1278d70b4cc239374a2107d0f111dbdc5
Andreas Gampe [Wed, 18 May 2016 20:09:54 +0000 (13:09 -0700)]
ART: Optimize PreSetup pass for oat-file open
The PreSetup pass for opening an oatfile with dlopen parses all
shared objects and tries to retrieve the corresponding mem map
(so as to have a fake MemMap to reserve the space for any scanning
operations and so on). Optimistically optimize this pass by counting
the number of shared objects before the dlopen, and skipping as
many afterwards.
This saves about half a millisecond on an N6P.
Bug:
28801010
Change-Id: I70156f9875e89c5af3c9e36e8f01c5662e3f736e
Mingyao Yang [Mon, 16 May 2016 18:01:49 +0000 (11:01 -0700)]
Move rewritten StringFactory call results into dex registers for deopt
Bug:
28555675
(cherry picked from commit
8ca33bf04060fadd5b35fa93fa56547c62fe52e7)
Change-Id: I9236df283f2e83ca5dcde01f73dc0522d745cd59
Roland Levillain [Wed, 18 May 2016 14:48:45 +0000 (14:48 +0000)]
Merge "Catch classes inheriting from themselves in the class linker." into nyc-dev
Roland Levillain [Wed, 18 May 2016 11:25:38 +0000 (12:25 +0100)]
Catch classes inheriting from themselves in the class linker.
Bug:
28685551
Bug:
27682580
Bug:
28830038
Change-Id: If568013bf3c82c1df9b282522712d9af5ca5945d
Mingyao Yang [Tue, 10 May 2016 21:30:41 +0000 (14:30 -0700)]
Fix an assert during jdwp debugging.
When debugger attaches, we patch entry points of methods in framework code.
During that process, it's possible that some method's declaring class isn't
resolved yet. We need to relax one assert for that case.
Bug:
28630805
(cherry picked from commit
3fd448a8b7903b00fceaed5853306e2d7eb31c72)
Change-Id: I59fd488ad75417a64b52321677ffcac6ccdc5ce1
Andreas Gampe [Tue, 17 May 2016 17:13:10 +0000 (10:13 -0700)]
ART: Fix systrace monitor logging
Thinlock unlocking was incorrectly unconditionally ending a block.
Bug:
28423466
Change-Id: Ifaebd9e959041e157e292d4cba05675a37e9c700
TreeHugger Robot [Tue, 17 May 2016 01:46:56 +0000 (01:46 +0000)]
Merge "Revert "Reject classes inheriting from themselves directly or transitively."" into nyc-dev
Mathieu Chartier [Tue, 17 May 2016 01:21:19 +0000 (01:21 +0000)]
Merge "Visit IMT of all classes during image writing" into nyc-dev
Nicolas Geoffray [Tue, 17 May 2016 00:12:11 +0000 (00:12 +0000)]
Revert "Reject classes inheriting from themselves directly or transitively."
Bug:
28685551
Bug:
27682580
Bug:
28797012
This reverts commit
9de98ef4ad4311e85d124099b0b36ab81171ea20.
Change-Id: Ic4a21194ad602f4897f088a66672f3dd9f30f607
Mathieu Chartier [Mon, 16 May 2016 21:58:01 +0000 (14:58 -0700)]
Visit IMT of all classes during image writing
We still need to visit the IMT for classes that does not have any
methods. It is required to visit the IMT of every class to assign
image offsets for the conflict table methods.
Added regression test.
Bug:
28707801
Change-Id: I1cd21f20e1e3c606ecfe8cbabe1045cadd86f268
Mingyao Yang [Mon, 16 May 2016 23:15:03 +0000 (23:15 +0000)]
Merge "Fix assert failure in interpreter after deoptimization." into nyc-dev
Calin Juravle [Fri, 29 Apr 2016 15:44:11 +0000 (16:44 +0100)]
Adjust ProfileSaver strategy
Change to a notification based strategy (instead of polling):
- the JIT notifies the ProfileSaver everytime it attempts a compilation
or generate (in a task) a ProfileInfo object for the method
- once a certain number of hot methods notification are received,
ProfileSaver wakes up the thread to do the analysis
- a wake up does not equal to I/O since the during processing we might
realized that's actually not enough new data.
- ProfileSaver throtles the request and ensures it doesn't do the
processing more often than kMinSavePeriodNs.
The new strategy avoids waking the thread unnecessary during extended
idle periods when we don't JIT.
Bug:
26815940
Change-Id: I24cb15daead5c5a5395ae6f5f79549e39e2f3011
Jeff Hao [Mon, 16 May 2016 17:15:20 +0000 (17:15 +0000)]
Merge "Make quick frames of proxy methods visited if they're constructors." into nyc-dev
Roland Levillain [Thu, 12 May 2016 15:12:47 +0000 (16:12 +0100)]
Reject classes inheriting from themselves directly or transitively.
Also enforce class definition ordering with respect to super
classes within a Dex file.
(cherry picked from commit
498bb81f8725e2cf8f5ce1aa6b49ab5a9c5bb4c9)
Bug:
28685551
Bug:
27682580
Change-Id: If3eba782538eb4328d4b8a542236632379e7c050
Jeff Hao [Fri, 13 May 2016 21:46:28 +0000 (14:46 -0700)]
Make quick frames of proxy methods visited if they're constructors.
Also adds a test case for getting annotations during moving GC.
Bug:
28510922
Change-Id: Ic9b849cd16321da3a3a609182dcdea08782496f9
Przemyslaw Szczepaniak [Thu, 12 May 2016 13:55:48 +0000 (14:55 +0100)]
Ensure that java.lang.reflect.Field is initialized
Recent change in libcore (a move of clinit code
to register function for java.net.InetAddress
class) was triggering a failure in 119-noimage-patchoat
test. Failure was occurring when the test was running
without the image. This was caused by the
java.lang.reflect.Field not being initialized when
running InetAddress.<clinit>, which is a side effect
of calling JNIEnv->GetMethodId inside the libcore JNI_OnLoad.
This changes makes sure that this class is initialized
when we're running without image.
Bug:
28153851
Bug:
27265238
Change-Id: I8aeb5eb312a163c452c40b851112544a8340e488
(cherry picked from commit
a2b147eaf72c5af23302a382620360b4cac28934)
Aart Bik [Wed, 11 May 2016 17:30:47 +0000 (10:30 -0700)]
Fix oatdump crash on arm64/arm code.
Also adds 16 bit literal information.
Rationale:
When "run-away" instructions are disassembled, the literal
addresses may go out of range, causing oatdump to crash.
This CL guards memory access against the full memory range
allocated to assembly instructions and data (it is possible
but not really necessary to refine this a bit). Out of range
arguments are now displayed as (?) to denote the issue, which
is a lot nicer than crashing.
BUG=
28670871
(cherry picked from commit
d3059e77818a058513ed92557160bdb6d5102b67)
Change-Id: I322bbdf83509e253430c40428a05c0ad16b81354
Richard Uhler [Thu, 12 May 2016 15:12:28 +0000 (15:12 +0000)]
Merge "Don't expect permission to open foreign dex use marks." into nyc-dev
TreeHugger Robot [Thu, 12 May 2016 07:57:02 +0000 (07:57 +0000)]
Merge "Fix another case of live_in at irreducible loop entry." into nyc-dev
Mathieu Chartier [Wed, 11 May 2016 22:15:39 +0000 (22:15 +0000)]
Merge "Prevent ArtMethod clearing race" into nyc-dev
Mathieu Chartier [Wed, 11 May 2016 01:13:20 +0000 (18:13 -0700)]
Prevent ArtMethod clearing race
There was race condition where the GC would be visiting the methods
of a class while the class linker overwrote the contents with 0xFE
when copying to a new array.
Since the GC is holding the class table lock at this time, we can
use this lock in the class linker to prevent the race.
Bug:
28699001
(cherry picked from commit
10c5f56423feaf3eadb3d4c09c61d2b998404162)
Change-Id: I5ddca93106cb8e48962cf44e5ce434c45e05cd0c
Mathieu Chartier [Wed, 11 May 2016 18:35:39 +0000 (11:35 -0700)]
Add histogram for native allocations
Shows up in traces.txt, sample output:
Histogram of native allocation 0:4315,131072:33,
1179648:3 bucket size 131072
Histogram of native free 0:995,131072:19,
1179648:2 bucket size 131072
Bug:
28680116
Change-Id: I584eb85709d3ae28f5142a01af11075a2a6436d7
Richard Uhler [Tue, 10 May 2016 21:01:18 +0000 (14:01 -0700)]
Don't expect permission to open foreign dex use marks.
sepolicy says we can create the mark files, but open requires
additional permissions we don't need. Don't print a warning when we
aren't given those additional permissions.
Bug:
28241500
Change-Id: I6764a9b0e165ff9781a10f682d7387b094d34da9
Mingyao Yang [Thu, 28 Apr 2016 23:23:01 +0000 (16:23 -0700)]
Fix assert failure in interpreter after deoptimization.
There is code that does
new String(chararr, 0, chararr_count);
which under debuggable turns into a call into pNewEmptyString() plus a
call into pNewStringFromChars_CII().
Even though we currently don't patch return pc in a runtime method,
calling into pNewEmptyString() is special since it's hacked such
that it's acting as if the caller calls into the java method
StringFactory.newEmptyString() directly. So deoptimization can now
happen at the NewEmptyString site and the assert is triggered since
it's a new instance instead of an invoke instruction.
The fix relaxes the assert to allow the special case.
Bug:
28555675
(cherry picked from commit
504a69081f63818ca332ddaf54e8198448554538)
Change-Id: Idbb159b5aa450df2344cd93ae74fef5f55bdc534
Nicolas Geoffray [Tue, 10 May 2016 13:35:34 +0000 (14:35 +0100)]
Fix another case of live_in at irreducible loop entry.
GVN was implicitly extending the liveness of an instruction across
an irreducible loop.
Fix this problem by clearing the value set at loop entries that contain
an irreducible loop.
bug:
28252896
Change-Id: I68823cb88dceb4c2b4545286ba54fd0c958a48b0
Mathieu Chartier [Wed, 11 May 2016 02:32:30 +0000 (02:32 +0000)]
Merge changes from topic 'native_unloads' into nyc-dev
* changes:
Fix libarttest to be non debug
Change test 136 to use dlopen
Revert "Revert "Remove double unloading hack""
Revert "Revert "Revert "Revert "(dl)Close native libraries on unload""""
Mathieu Chartier [Thu, 5 May 2016 22:31:12 +0000 (15:31 -0700)]
Fix libarttest to be non debug
Previously both libarttest and libarttestd had debug enabled. This
made it hard to know which one was loaded (couldn't look at
kIsDebugBuild).
Fixes test 136 with ART_TEST_RUN_TEST_NDEBUG configuration.
Bug:
28406866
(cherry picked from commit
48b2b3e57e0510f549e9dc5c5a61f50eec697cf5)
Change-Id: I92983f71374b211f96f5b346fec326ea8543f876
Aart Bik [Tue, 10 May 2016 22:43:27 +0000 (22:43 +0000)]
Merge "Break the debug cycle!" into nyc-dev
Mathieu Chartier [Thu, 5 May 2016 18:09:58 +0000 (11:09 -0700)]
Change test 136 to use dlopen
dlopen self to prevent the native libary being closed from
DestroyJavaVM. We need this since it is the library currently
executing code.
Bug:
28406866
(cherry picked from commit
06fd2cff238cbdd8f6ceb4a85273c5a7fd0db023)
Change-Id: Ib23f09a97184abb69a142057b0ccb9d476f69c95
Mathieu Chartier [Tue, 10 May 2016 17:11:48 +0000 (10:11 -0700)]
Revert "Revert "Remove double unloading hack""
This reverts commit
e9cae49599856fe737cee87783438d2c5dabccfe.
Bug:
28406866
(cherry picked from commit
c0f2e678e45ff191f90651117f4e238caa521b80)
Change-Id: I806cec1a6baab8722262082a6d3db1ac3734a8ab
Mathieu Chartier [Tue, 10 May 2016 17:12:00 +0000 (10:12 -0700)]
Revert "Revert "Revert "Revert "(dl)Close native libraries on unload""""
This reverts commit
624b68ecbe67dd91cb50f588fbfa2243def47f1e.
Bug:
28406866
(cherry picked from commit
33bae7d52debe898879fc034a5524b1b165dbd88)
Change-Id: I4d6d6791b286cfeecaeebc388684ee8d3c44a920
Dimitry Ivanov [Tue, 10 May 2016 20:09:37 +0000 (20:09 +0000)]
Merge "Initialize native loader before starting runtime" into nyc-dev