Previously, if the user sets their own stack boundary, we still compute
the thread stack boundary (which is expensive), then immediately discard
the result. This change makes the expensive call only if we need it for
sure.
This should guarantee that the various macros required by
wasm_proposal.c are defined even if the build system does not supply
them to the compiler command.
For boundary checking, WAMR calls `pthread_attr_np`, which is
unfortunately quite slow on Linux when not called on the main thread
(see https://github.com/bytecodealliance/wasm-micro-runtime/issues/3966
for discussion).
This change moves the cost of stack bounds checking earlier in the
wasm_exec_env creation process. The idea is that it's perhaps better to
pay the price when creating the execution environment rather than in the
first function call.
The original code is left in place inside
`call_wasm_with_hw_bound_check` in case the `wasm_exec_env` is created
via `wasm_runtime_spawn_exec_env`.
While resolving linked globals in multi-module mode, WAMR tries to copy
the linked global's initial value into the destination global in the
current module. However, a bug in the implementation causes the copy to
be done from the InitializerExpression struct, not from its WASMValue
field.
This did not come up in WAMR's spec test runner because those are built
with WASM_ENABLE_SPEC_TEST, which means these globals are resolved as
builtins, not linked globals, which goes through a different (presumably
not faulty) path.
Previously, the loader would check the name of a new export against all
existing exports, leading to a quadratic running time.
This change makes the loader parse the entire export section. The
exports are then sorted by name, then adjacent exports are checked for
uniqueness.
Currently, the open-source builds of wamrc set WASM_ENABLE_DUMP_CALL_STACK,
which causes these two fields to be emitted. They are required by aot_emit_exception.c.
Internally at Google, we don't enable call stack dumps, so we've been using the
attached patch to make sure the fields are emitted anyway.
- Only retry on EAGAIN, ENOMEM or EINTR.
- On EINTR, don't count it against the retry budget, just keep retrying.
EINTR can happen in bursts.
- Log the errno on failure, and don't conditionalize that logging on
BH_ENABLE_TRACE_MMAP. In other parts of the code, error logging is not
conditional on that define, while turning on that tracing define makes
things overly verbose.
