When spilling a value to the stack, make sure to fill it as the same
type later. Otherwise, we might load garbage from the stack and violate
the assumption that the upper 32 bits of a 64 bit register are zero if
it currently holds a 32 bit value.
R=titzer@chromium.org
Bug: v8:7353, v8:6600
Change-Id: I7f2b1b31b7f3c13aa152c682cb59400fb5a3ebf0
Reviewed-on: https://chromium-review.googlesource.com/880682
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50797}
This removes the field in question to make it simpler to serialize and
deserialize modules without having to worry about the state of lazy
compilation. It is always possible to clone a non-anonymous builtin,
even without having this module-wide field.
R=clemensh@chromium.org
TEST=mjsunit/regress/wasm/regress-803427
BUG=chromium:803427
Change-Id: I72041e314eb6ee92859d45f1db0ed8500003edc4
Reviewed-on: https://chromium-review.googlesource.com/878581
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50771}
Tag RelocInfo which belongs to native wasm code, and fix printing to
not try to access the Code object for CODE_TARGET, but rather just
print "(wasm trampoline)".
Bug: chromium:801785
R=mstarzinger@chromium.org
Change-Id: I84a37f0c48ed7397cccf677b4d0f0352e5aceb9d
Reviewed-on: https://chromium-review.googlesource.com/875271
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50758}
This fixes a corner-case with lazy compilation in WebAssembly where
native-heap code did not expect to see WASM-to-JS wrappers in tables.
R=clemensh@chromium.org
TEST=mjsunit/regress/wasm/regress-803788
BUG=chromium:803788
Change-Id: Ie44b5c9efe2b171e1915295bb95d6cb61dfab3dc
Reviewed-on: https://chromium-review.googlesource.com/878262
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50755}
The method {SpillAllRegisters} should really just spill registers, and
not also constants. Also, since more code is inlined into that method
now, we can optimize it to update register use counters only once in
the end, since no used register should be left after executing this
method.
R=titzer@chromium.org
Bug: v8:6600, chromium:802244
Change-Id: I737a1a87f8e912151062224952c4f5dffd43f802
Reviewed-on: https://chromium-review.googlesource.com/868022
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50644}
This fixes a corner-case in the {NativeModuleSerializer} with modules
that do not contain any functions in the code table.
R=ahaas@chromium.org
TEST=mjsunit/regress/wasm/regress-801850
BUG=chromium:801850
Change-Id: I30cc3a26f30d8653fba2d7b99715830d12300ac2
Reviewed-on: https://chromium-review.googlesource.com/866773
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50581}
Ensure that for setcc, we only use a byte register as destination
register.
R=titzer@chromium.org
Bug: v8:6600, chromium:800756
Change-Id: Ie33f3faf602e7eda845205ba0ed2d9966460fd54
Reviewed-on: https://chromium-review.googlesource.com/860640
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50508}
In d8 a script is executed in a RealmScope. However, all micro task
which are created by the script are not executed within the RealmScope
at the moment. With this CL I move the execution of the micro task into
the RealmScope.
I thought about creating a new RealmScope for the micro tasks, but
(1) It did not fix the crashing repro;
(2) It seems wrong that the micro tasks are executed in a different
realm than the script;
Therefore I just moved the execution of the micro tasks into the
RealmScope of the script.
Thereby I moved the execution tasks also into the Context::Scope of the
script. The problem is that the Context::Scope surrounds the RealmScope,
and when I to open the RealmScope before the Context::Scope, not even
the execution of the script works anymore.
R=yangguo@chromium.org
Bug: chromium:797846
Change-Id: If152af282beec8f0b0564dcc9682fee8588e142c
Reviewed-on: https://chromium-review.googlesource.com/856497
Reviewed-by: Yang Guo <yangguo@chromium.org>
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#50442}
Moving a register to itself is not only unnecessary overhead, it also
breaks invariants in the StackTransferRecipe.
R=ahaas@chromium.org
Bug: v8:6600, chromium:793551
Change-Id: I659fd66b4f2d4564c437ed9fb048322af4299d97
Reviewed-on: https://chromium-review.googlesource.com/819231
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49992}
Ensure that the type is always stored correctly.
R=titzer@chromium.org
Bug: v8:6600, chromium:791810
Change-Id: Id3a3c20b14f8730b9550c548dec49ac47121e691
Reviewed-on: https://chromium-review.googlesource.com/811188
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49924}
When initializing the stack state at a merge point, don't force all
stack slots into registers. Allow constants to stay constants as long
as they are not part of the merge. Otherwise we might break assumptions
of outer blocks which then try to merge a register into a constant and
fail.
Also, add some documentation to {InitMergeStackSlot} to document the
intent of the implementation.
R=titzer@chromium.org
Bug: v8:784050, v8:6600
Change-Id: I3a4c83b446909027be075d3207cb7c748a6b1aad
Reviewed-on: https://chromium-review.googlesource.com/766353
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49423}
This is a reland of 6c6132880a
Original change's description:
> [Liftoff] Implement parallel register moves
>
> This was not implemented so far, leading to lots of clusterfuzz issues.
> Testing this feature would require complicating the interface a lot and
> exposing more implementation details in the header file, hence we just
> go with regression tests for now.
>
> R=ahaas@chromium.org, titzer@chromium.org
>
> Bug: v8:6600, chromium:782280
> Change-Id: I12863f3eb59a8dffdcc7d3bfb8e1f0ae0eec15ee
> Reviewed-on: https://chromium-review.googlesource.com/758772
> Reviewed-by: Andreas Haas <ahaas@chromium.org>
> Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#49286}
Bug: v8:6600, chromium:782280
Change-Id: I82a75bfeaf83dc63a2917da3ccdc4721c5d689e7
Reviewed-on: https://chromium-review.googlesource.com/763387
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49292}
This reverts commit 6c6132880a.
Reason for revert: breaks compilation on win64 bot: https://build.chromium.org/p/client.v8/builders/V8%20Win64%20-%20msvc/builds/69
Original change's description:
> [Liftoff] Implement parallel register moves
>
> This was not implemented so far, leading to lots of clusterfuzz issues.
> Testing this feature would require complicating the interface a lot and
> exposing more implementation details in the header file, hence we just
> go with regression tests for now.
>
> R=ahaas@chromium.org, titzer@chromium.org
>
> Bug: v8:6600, chromium:782280
> Change-Id: I12863f3eb59a8dffdcc7d3bfb8e1f0ae0eec15ee
> Reviewed-on: https://chromium-review.googlesource.com/758772
> Reviewed-by: Andreas Haas <ahaas@chromium.org>
> Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#49286}
TBR=titzer@chromium.org,ahaas@chromium.org,clemensh@chromium.org
Change-Id: I386983da7f9f03d54cef2190b01fd2cc5785966e
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: v8:6600, chromium:782280
Reviewed-on: https://chromium-review.googlesource.com/763367
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49289}
This was not implemented so far, leading to lots of clusterfuzz issues.
Testing this feature would require complicating the interface a lot and
exposing more implementation details in the header file, hence we just
go with regression tests for now.
R=ahaas@chromium.org, titzer@chromium.org
Bug: v8:6600, chromium:782280
Change-Id: I12863f3eb59a8dffdcc7d3bfb8e1f0ae0eec15ee
Reviewed-on: https://chromium-review.googlesource.com/758772
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49286}
The C_WASM_ENTRY only calls wasm code, so (outgoing) parameters of this
code are always untagged.
R=ahaas@chromium.org
Bug: v8:7049
Change-Id: I989908eea668bd56aa88f3f1744105d218584ef1
Reviewed-on: https://chromium-review.googlesource.com/758245
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49221}
When initializing the cache state for a merge, we should never use
registers multiple times. Other code paths leading to the same merge
point might provide different values for the different slots there.
R=ahaas@chromium.org
Bug: v8:7035, v8:6600
Change-Id: I8e409b494af0fdc1a5045ec04571611b97fcaf86
Reviewed-on: https://chromium-review.googlesource.com/754816
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49150}
If the destination register of a binop is the same register as the
right hand side, we would first move the left hand side into that
register (overwriting the value of the rhs), and then use the rhs.
This CL fixes this issue and adds a regression test.
R=ahaas@chromium.org
Bug: v8:6600, v8:7033
Change-Id: Ief90b5bcffc65823037bc57fb00741b2448e6375
Reviewed-on: https://chromium-review.googlesource.com/753462
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49137}
The existing stack check only checked the number of stack frames on the
stack, not the actual size of the stack frames. In the test case, each
stack frame is huge, and the interpreter runs out of memory before the
stack check stops the execution. With this change we take the size of
the value stack and the size of the control stack and compare their sum
to the stack limit of V8. Note that this stack limit is kind of
arbitrary, because the stack space of the interpreter is not on the
actual runtime stack but allocated in zone memory, and the stack check
exists to simulate stack overflows in compiled code, not to prevent
actual stack overflows.
R=clemensh@chromium.org
TEST=mjsunit/regress/wasm/regress-778917
Bug: chromium:778917
Change-Id: Ife47631fcb1a178a68facab1e42c0069b12c0155
Reviewed-on: https://chromium-review.googlesource.com/744003
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#49071}
We have an internal limit of 50000 local variables per wasm function.
This limit is checked when decoding the function body. For asm.js, we
skip function body validation, since by construction the code we
generate is correct. This makes us fail unexpectedly when trying to
(lazily) compile an asm.js function with more than 50000 locals.
Hence, check this limit in the asm parser and bail out if it is
exceeded.
R=mstarzinger@chromium.org
Bug: chromium:775710
Change-Id: I89d2069e133fb0f84947d477ae1ac5eda85571aa
Reviewed-on: https://chromium-review.googlesource.com/732660
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48929}
If the buffer associated with WebAssembly.Memory is used as memory
for asm.js modules, throw a range error on Memory.Grow.
Bug: chromium:776677
Change-Id: Iebcd7797fa7724002dd8073d1dbaeb98f080d316
Reviewed-on: https://chromium-review.googlesource.com/731844
Commit-Queue: Deepti Gandluri <gdeepti@chromium.org>
Reviewed-by: Brad Nelson <bradnelson@chromium.org>
Reviewed-by: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48837}
[wasm] Fix signature canonicalization for error case.
The decoder should not attempt to insert null signatures into the SignatureMap.
R=ahaas@chromium.org
Bug: chromium:775366
Change-Id: I0fbc0547dbf00fd25d37271a03b6756481a4c6a1
Reviewed-on: https://chromium-review.googlesource.com/730752
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48793}
The bug reference has been fixed, probably due to the new WasmContext
changes. We should keep a regression test for this anyway though.
Bug: v8:6931
Change-Id: Ie9d94690e764498d2153691d96414d0d26258794
Reviewed-on: https://chromium-review.googlesource.com/727022
Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Eric Holk <eholk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48712}
This patch introduces assertPromiseFulfills and assertPromiseFulfills as
a replacement for assertPromiseResult because it’s more JavaScript-y.
BUG=v8:6921
R=ahaas@chromium.org
Also-By: ahaas@chromium.org
Change-Id: I2f865dba3992ddf3b58987bf0b376d143edb5c31
Reviewed-on: https://chromium-review.googlesource.com/718746
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48578}
This reverts commit 5e76ff5a4a.
Reason for revert: tsan failures - https://build.chromium.org/p/client.v8/builders/V8%20Linux64%20TSAN/builds/17574
Original change's description:
> Reland "[wasm] always allocate memory when guard regions are needed"
>
> This reverts commit 7cf29d8df3.
>
> Original change's description:
> > [wasm] always allocate memory when guard regions are needed
> >
> > When using trap handlers, memory references do not get any checks inserted. This
> > means there is no check for a null memory as happens when the memory size is
> > 0. Normally this would be correctly caught as an out of bounds access, since the
> > low memory addresses are not normally mapped. However, if they were mapped for
> > some reason, we would not catch the out of bounds access.
> >
> > The fix is to ensure WebAssembly instances always have a guard region even if
> > the memory is size 0.
> >
> > Bug: chromium:769637
>
> Change-Id: I09fdaea92b7ccb3a6cc9e28392171ec098538a00
> Reviewed-on: https://chromium-review.googlesource.com/695812
> Commit-Queue: Eric Holk <eholk@chromium.org>
> Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
> Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#48293}
TBR=gdeepti@chromium.org,mtrofin@chromium.org,mlippautz@chromium.org,eholk@chromium.org,eholk@google.com,clemensh@chromium.org
Change-Id: I52d5354126158a92602b08c48703d562ac95075b
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/699599
Reviewed-by: Eric Holk <eholk@chromium.org>
Commit-Queue: Eric Holk <eholk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48294}
This reverts commit 7cf29d8df3.
Original change's description:
> [wasm] always allocate memory when guard regions are needed
>
> When using trap handlers, memory references do not get any checks inserted. This
> means there is no check for a null memory as happens when the memory size is
> 0. Normally this would be correctly caught as an out of bounds access, since the
> low memory addresses are not normally mapped. However, if they were mapped for
> some reason, we would not catch the out of bounds access.
>
> The fix is to ensure WebAssembly instances always have a guard region even if
> the memory is size 0.
>
> Bug: chromium:769637
Change-Id: I09fdaea92b7ccb3a6cc9e28392171ec098538a00
Reviewed-on: https://chromium-review.googlesource.com/695812
Commit-Queue: Eric Holk <eholk@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48293}
In JS to Wasm wrappers, arguments have to be converted from JavaScript's
representation to Wasm's representation. Because of property accessors, this can
result in JavaScript or even asm.js/Wasm code being run. We were previously
setting this flag before doing the parameter conversions, and if these
conversions triggered a Wasm property getter then we would try to set the flag
twice.
With this change, we wait until after all argument conversions are done to set
the flag.
Bug: chromium:769846
R=bradnelson@chromium.org
Change-Id: Ia4b56df45619dcad69f3750bb33cacfedcaeb5b2
Reviewed-on: https://chromium-review.googlesource.com/693414
Commit-Queue: Brad Nelson <bradnelson@chromium.org>
Reviewed-by: Brad Nelson <bradnelson@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48244}
This reverts commit 1f99c66b56.
Reason for revert: Test timeouts on Win64 Debug: https://build.chromium.org/p/client.v8/builders/V8%20Win64%20-%20debug/builds/19226
Original change's description:
> [wasm] always allocate memory when guard regions are needed
>
> When using trap handlers, memory references do not get any checks inserted. This
> means there is no check for a null memory as happens when the memory size is
> 0. Normally this would be correctly caught as an out of bounds access, since the
> low memory addresses are not normally mapped. However, if they were mapped for
> some reason, we would not catch the out of bounds access.
>
> The fix is to ensure WebAssembly instances always have a guard region even if
> the memory is size 0.
>
> Bug: chromium:769637
> Change-Id: I2d0f8c107563236c3780eb7746c2f820e319c65f
> Reviewed-on: https://chromium-review.googlesource.com/693137
> Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
> Commit-Queue: Eric Holk <eholk@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#48240}
TBR=gdeepti@chromium.org,mtrofin@chromium.org,eholk@chromium.org
Change-Id: I4065b367c6cfffe8dd601b67cd53ad54759ae96a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: chromium:769637
Reviewed-on: https://chromium-review.googlesource.com/692918
Reviewed-by: Eric Holk <eholk@chromium.org>
Commit-Queue: Eric Holk <eholk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48242}
When using trap handlers, memory references do not get any checks inserted. This
means there is no check for a null memory as happens when the memory size is
0. Normally this would be correctly caught as an out of bounds access, since the
low memory addresses are not normally mapped. However, if they were mapped for
some reason, we would not catch the out of bounds access.
The fix is to ensure WebAssembly instances always have a guard region even if
the memory is size 0.
Bug: chromium:769637
Change-Id: I2d0f8c107563236c3780eb7746c2f820e319c65f
Reviewed-on: https://chromium-review.googlesource.com/693137
Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
Commit-Queue: Eric Holk <eholk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48240}
Memory instantiate on initialize should always patch memory
references. If memory references are not patched for no initial
memory, on subsequent calls to grow_memory in wasm functions for
instances that share a module, the references will be patched
without resetting cloned compiled values to their correct initial
values.
BUG=chromium:763439
Change-Id: I666439332379b02aa344e99d61ef3dc88ab86cc8
Reviewed-on: https://chromium-review.googlesource.com/674707
Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Deepti Gandluri <gdeepti@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48097}
The wasm valiation incorrectly allowed simd locals, even without the
experimental flag turned on. This was not noted in the generated code
because simd opcodes were forbidden, but the interpreter could not
handle these locals.
R=clemensh@chromium.org
Bug: chromium:763697
Change-Id: I11d924ac21e50bce81d0504c2c7b252105a89f80
Reviewed-on: https://chromium-review.googlesource.com/660117
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#47946}
When lazy-compiling, it is important we reconstitute the
ModuleEnv accurately. Besides addressing a bug, this change
also does away with the need to relocate memory and globals
parameters (in lazy compilation), by using "the right ones" upfront.
Bug: chromium:753496
Change-Id: I1412a499f05d02d49319fced1b3047698328f3b5
Reviewed-on: https://chromium-review.googlesource.com/609376
Reviewed-by: Brad Nelson <bradnelson@chromium.org>
Commit-Queue: Brad Nelson <bradnelson@chromium.org>
Cr-Commit-Position: refs/heads/master@{#47280}
This brings the wasm-constants.js file inline with that (forked copy) in
the WebAssembly spec repo, which should make it easier to export tests
from V8 to the spec in the future.
R=clemensh@chromium.org
Bug:
Change-Id: I7db23efc2d671f65b614f9dbc97ae2f355f91b04
Reviewed-on: https://chromium-review.googlesource.com/586248
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46894}
The code was already there, but there was a bug in it: Because of the
missing reference, we were only updating a *copy* of the signature map,
hence the update had no effect.
This intentially is a minimal CL, in order to allow for easy
backmerging.
More mitigations and tests are coming in a separate CL.
R=titzer@chromium.org
Change-Id: Ifb462093f4b8f4d5380b6774636537c67c2b676c
Reviewed-on: https://chromium-review.googlesource.com/570278
Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46664}
It's ok that the instance of the called code object is different from
the caller instance. This happens if one instance calls an exported
function of another instance.
R=ahaas@chromium.org
Bug: chromium:739768
Change-Id: I6afa8332a9b33fe32e9332cdca573053f058421d
Reviewed-on: https://chromium-review.googlesource.com/568494
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46624}
This reverts commit 1520a8518a.
Reason for revert: This CL does not do what it should. All tasks which access the isolate have to be cancelable to guarantee that the isolate still exists when the task is executed. Foreground compilation tasks access the isolate, so they cannot be just normal tasks.
Original change's description:
> [wasm] Run foreground compilation tasks as normal tasks
>
> This CL makes foreground compilation tasks normal (i.e. not cancelable)
> again, because otherwise a deadlock can happen. I think the reason why
> the foreground tasks were cancelable was to make sure that all tasks
> either finish correctly or get canceled. However, since the isolate can
> only shut down on the main thread, this means that the foreground task
> should have already finished when the isolate shuts down, or it should
> not have started at all. I reordered the deletion of the AsyncCompileJob
> though to make sure that an AsyncCompileJob is removed from
> CompilationManager before its promise is resolved.
>
> Here is the deadlock: The JS code which is executed after a promise is
> resolved is executed within the task which resolves the promise. In case
> of async compilation this means that some JS code is executed within a
> CompileTask. In JS, the shutdown of the isolate can be triggered. During
> the shutdown of the isolate, the CancelableTaskManager waits for all
> registered cancelable tasks to complete, including the CompileTask of
> async compilation. This means that the CancelableTaskManager waits for
> itself to finish, which is a deadlock.
>
> R=clemensh@chromium.org, mtrofin@chromium.org
>
> Change-Id: I9f8c7fb2cfc5b9bfc53c761010b1590293bb82c9
> Reviewed-on: https://chromium-review.googlesource.com/554733
> Commit-Queue: Andreas Haas <ahaas@chromium.org>
> Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
> Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#46343}
TBR=mtrofin@chromium.org,ahaas@chromium.org,clemensh@chromium.org
Change-Id: I60fab90b46d70c703d827816503e7e23b8c50251
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/558284
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46353}
This CL makes foreground compilation tasks normal (i.e. not cancelable)
again, because otherwise a deadlock can happen. I think the reason why
the foreground tasks were cancelable was to make sure that all tasks
either finish correctly or get canceled. However, since the isolate can
only shut down on the main thread, this means that the foreground task
should have already finished when the isolate shuts down, or it should
not have started at all. I reordered the deletion of the AsyncCompileJob
though to make sure that an AsyncCompileJob is removed from
CompilationManager before its promise is resolved.
Here is the deadlock: The JS code which is executed after a promise is
resolved is executed within the task which resolves the promise. In case
of async compilation this means that some JS code is executed within a
CompileTask. In JS, the shutdown of the isolate can be triggered. During
the shutdown of the isolate, the CancelableTaskManager waits for all
registered cancelable tasks to complete, including the CompileTask of
async compilation. This means that the CancelableTaskManager waits for
itself to finish, which is a deadlock.
R=clemensh@chromium.org, mtrofin@chromium.org
Change-Id: I9f8c7fb2cfc5b9bfc53c761010b1590293bb82c9
Reviewed-on: https://chromium-review.googlesource.com/554733
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46343}
The implication was actually in the wrong direction: If there is no
memory start address, then the size must be 0.
If the size is 0 though, we might allocate nevertheless to have guard
pages around the accessible memory.
R=ahaas@chromium.org
BUG=chromium:736584
Change-Id: I297dece658d5eaf69c58ecb109ff21d3ca0b8a8d
Reviewed-on: https://chromium-review.googlesource.com/548635
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46221}
If one wasm instance imports an exported function of another instance,
we unwrap the js-to-wasm wrapper of the export and use the underlying
code object directly. However, the code object does not keep the wasm
instance alive. It is only connected via a WeakCell.
With this CL, we explicitly store a FixedArray of all wasm instances
from which we imported functions to keep them alive at least as long as
the instance which imports the code.
R=mtrofin@chromium.org, ahaas@chromium.org
BUG=chromium:734345
Change-Id: I8dcfc9a4ea2d791a62d8cb7255039e481c50bdfd
Reviewed-on: https://chromium-review.googlesource.com/539738
Reviewed-by: Mircea Trofin <mtrofin@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46062}
