Double-checked locking pattern for destruction was missing the acquire
barrier for the initial load.
TSAN complained with a data race where:
T1: ClearAllUsedNodes(), clearing out the node
T2: a. if(GetNodeSafe()) { Lock; ... }
T2: b. operator delete
Since GetNodeSafe() was a relaxed load, operator delete was allowed to
be reordered which raced with ClearAllUsedNodes().
Bug: chromium:1239081, chromium:1242795
Change-Id: I3906555b13cc51538a1a54b7ca481a96d81fd84e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3132264
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Cr-Commit-Position: refs/heads/main@{#76599}
This avoids a benign race in setting the raw pointer inside CTP
destructor by not emitting the write at all. The handle is destructed
which means that we only need to destroy any backing node but may
leave the handle untouched.
Drive-by:
- Add a few more docs.
- Make Clear() thread-safe.
- Make assignment of a sentinel pointer thread-safe.
- Make assignment of a nullptr thread-safe.
Depends on the Blink change: https://crrev.com/c/3116259
Bug: chromium:1242795, chromium:1243350
Change-Id: I8d76da30893c165e3946322b6d02f6ea2c8e529e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3114064
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Cr-Commit-Position: refs/heads/main@{#76493}
This reverts commit 5a6c7dee4e.
Reason for revert: Speculative: Lots of Chrome crashes:
https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Blink%20Linux/13353/overview
Original change's description:
> cppgc: Fix CTP for destruction
>
> This avoids a benign race in setting the raw pointer inside CTP
> destructor by not emitting the write at all. The handle is destructed
> which means that we only need to destroy any backing node but may
> leave the handle untouched.
>
> Drive-by:
> - Add a few more docs.
> - Make Clear() thread-safe.
> - Make assignment of a sentinel pointer thread-safe.
> - Make assignment of a nullptr thread-safe.
>
> Bug: chromium:1242795
> Change-Id: I0d9dafa31c298053e87ba1eb75f99fa6e33fa10b
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3114134
> Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
> Reviewed-by: Anton Bikineev <bikineev@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#76455}
Bug: chromium:1242795
Change-Id: Ia96d66f4908894091a4e498116d9568bd7b0e0a3
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3114058
Auto-Submit: Michael Achenbach <machenbach@chromium.org>
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Owners-Override: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/main@{#76482}
This avoids a benign race in setting the raw pointer inside CTP
destructor by not emitting the write at all. The handle is destructed
which means that we only need to destroy any backing node but may
leave the handle untouched.
Drive-by:
- Add a few more docs.
- Make Clear() thread-safe.
- Make assignment of a sentinel pointer thread-safe.
- Make assignment of a nullptr thread-safe.
Bug: chromium:1242795
Change-Id: I0d9dafa31c298053e87ba1eb75f99fa6e33fa10b
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3114134
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Cr-Commit-Position: refs/heads/main@{#76455}
Consider reading the internal node pointer instead of the actual pointer
when trying to figure out whether a node needs to be destroyed. This
preserves the non-atomiticity of the actual pointer which highlights
races using TSAN while fixing destruction.
Bug: chromium:1239081
Change-Id: I1d1fa29d40d86e4b156269abc90142ee71a8d8f4
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3110199
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Cr-Commit-Position: refs/heads/main@{#76415}
Allow CrossThreadPersistent and its weak form to access ASAN poisoned
memory from the GC entry points.
In general, payloads of to-be-finalized objects are poisoned until the
finalizer actually runs to avoid accidentally touching that payload.
In the case of cross-thread handles, these may need to be cleared by a
different thread before the finalizer actually runs. In order to clear
those references, the slot needs to be unpoisoned.
This is issue is ASAN-only and does not affect production or other
debug builds.
Bug: chromium:1230599, chromium:1056170
Change-Id: If4d0808953047319b02653821abbb5c638084dc5
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3040845
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#75846}
This is a reland of 7458e67c4e
Original change's description:
> cppgc: Implement basic Member and Persistent checks
>
> Adds check for
> - same heap on assignment
> - header and containment
>
> The verification state is eagerly created for on-heap Member
> references using caged heap and lazily created on first assignment for
> all others.
>
> Bug: chromium:1056170
> Change-Id: I38ee18eeb7ac489f69a46670cc5e5abe07f62dfa
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2878745
> Reviewed-by: Omer Katz <omerkatz@chromium.org>
> Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#74449}
Bug: chromium:1056170
Change-Id: I9cecfcf7ba2cb70650fd51f345fbf740b96ff6ba
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2882804
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#74468}
This reverts commit 7458e67c4e.
Reason for revert: Crash on windows https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Win64%20-%20debug/37698/overview
Original change's description:
> cppgc: Implement basic Member and Persistent checks
>
> Adds check for
> - same heap on assignment
> - header and containment
>
> The verification state is eagerly created for on-heap Member
> references using caged heap and lazily created on first assignment for
> all others.
>
> Bug: chromium:1056170
> Change-Id: I38ee18eeb7ac489f69a46670cc5e5abe07f62dfa
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2878745
> Reviewed-by: Omer Katz <omerkatz@chromium.org>
> Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#74449}
Bug: chromium:1056170
Change-Id: I466522a7d879560c99dabbd96c3b097894743a87
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2880149
Auto-Submit: Zhi An Ng <zhin@chromium.org>
Commit-Queue: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Cr-Commit-Position: refs/heads/master@{#74450}
Adds check for
- same heap on assignment
- header and containment
The verification state is eagerly created for on-heap Member
references using caged heap and lazily created on first assignment for
all others.
Bug: chromium:1056170
Change-Id: I38ee18eeb7ac489f69a46670cc5e5abe07f62dfa
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2878745
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#74449}
cpplint rules change over time, and we change the exact rules we enable
for v8. This CL removes NOLINT annotations which are not needed
according to the currently enabled rules.
R=mlippautz@chromium.org
Bug: v8:11717
Change-Id: I41c4c18dd3f70ec255e9d2769ffd25a38f6f2784
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2862764
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#74307}
This CL adds missing locks to the PersistentRegions for
(Weak)CrossThreadPersistents.
To make sure no locks are missed in the future, this CL also splits
PersistentRegion and introduces CrossThreadPersistentRegion that checks
whether a lock is taken whenever it is accessed.
Bug: chromium:1056170
Change-Id: Iaaef4a28af0f02bcb896706e9abf1ee5ad2ee1e1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2737299
Commit-Queue: Omer Katz <omerkatz@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73264}
Two threads might get the same PersistentNode because the
BasicCrossThreadPersistent ctor wasn't taking a lock. Then if one thread
frees the node and the other initalizes it or updates its owner, we get
some random object in our free list of PersistentNodes.
I debug a crash in Assign(Unsafe) and Clear where the PersistentNode
seemed to be allocated on stack. Empirically, adding this guard resolved
it. I can't confirm in the code that the scenario above is what was
happening.
Drive-by: adding a few DCHECKs.
Bug: chromium:1056170
Change-Id: I37d8ed5bb942a124c98d7524b7f04fe8ccb2aefd
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2718144
Commit-Queue: Omer Katz <omerkatz@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73023}
Assert that the lock is help whenever AssignUnsafe is called.
LazyMutex::AssertHeld resolves to a DCHECK so this should not
regress production performance (other than the call itself
that might not be inlined).
Bug: chromium:1056170
Change-Id: Ic2005d180e6960c24dff7743aa3e0d5e57a63d80
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2716286
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Omer Katz <omerkatz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73000}
This allows construction and destruction of empty Persistent and
friends, which simplifiest the use for embedders.
Bug: chromium:1056170
Change-Id: I4286639aa5d50f9f98654b859de10bb80cbada21
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2655505
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72396}
Adds support for Member<const T> by keeping the untyped storage in
MemberBase const, which is stronger than the required constness. All
accesses go through BasicMember which can re-add the appropriate
constness specified by the user.
The same concept is applied to all Member and Persistent handles.
Bug: chromium:1056170
Change-Id: I5a620258be3acb6a1b4b1437e69b8d7d1ec5ce6f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2625871
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72090}
Adds a cross-thread reference for strongly and weakly retaining
objects on a thread other than the thread that owns the object.
The intended use of the reference is by setting it up on the
originating thread, holding the object alive from another thread, and
ultimately accessing the object again on the originating thread.
The reference has known caveats:
- It's unsafe to use when the heap may terminate;
- It's unsafe to transitively reach through the graph because of
compaction;
Change-Id: I84fbdde69a099eb54af5b93c34e2169915b17e64
Bug: chromium:1056170
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2436449
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Omer Katz <omerkatz@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70428}
