1495 Commits

Author	SHA1	Message	Date
liang.he@intel.com	09c664f974	Revert "fix a wamrc compile warning, and enable system include flag for third-party libraries to suppresses corresponding compile warnings (#4875)" ... This reverts commit 14d84eca0b.	2026-03-20 08:38:29 +08:00
TianlongLiang	14d84eca0b	fix a wamrc compile warning, and enable system include flag for third-party libraries to suppresses corresponding compile warnings (#4875)	2026-03-19 09:08:41 +08:00
TianlongLiang	c3a78cd159	fix sign-compare warnings (#4876)	2026-03-17 22:27:17 -07:00
TianlongLiang	c3dda4a7d6	Fix compile error when compiler don't support '__has_feature' (#4874)	2026-03-17 16:06:45 +08:00
James Ring	c9815ef2c7	fix msan errors (#3724) ... Mark argv as initialized so that msan doesn't complain	2026-03-16 19:39:08 -07:00
Vyacheslav Chigrin	26aa924acd	Move exception lock to ModuleInstance data. (#4772) ... * Move exception lock to ModuleInstance data. This lock acquired on each native function call. This cause performance impact on programs, containing many native function calls, and running in multithreaded environment. Update AOT_CURRENT_VERSION constant.	2026-03-16 16:22:07 +08:00
liang.he	9d1dc3e4d2	Fix: Fix null pointer dereference in wasm_c_api.c (#4863)	2026-03-09 19:10:35 -07:00
liang.he	f159cc56ac	Fix: initialize record_frame in aot_copy_callstack (#4861)	2026-03-09 18:45:24 -07:00
liang.he	9e4aa9c850	fuzzing: reject non-wasm files quickly and execute aot after compilation (#4780) ... * fix: disable unsigned integer overflow sanitization in build configurations FYI: from https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html `-fsanitize=unsigned-integer-overflow`: Unsigned integer overflow, where the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for lossy implicit conversions performed before such a computation. It brings a more common question: which is better, pre-additional-check or post-additional-check to fix a potential unsigned integer overflow? A pre-additional-check involves using a check to prevent integer overflow from the very beginning. A post-additional-check involves using a check after addition to see if there is an overflow. In this project, post-additional-checking is widely used. let's follow the routine. for performance sensitive logic, use __builtin_add_overflow etc. provide something like 9a5622791e/lib/platform.h (L176-L191) and encourage the use of them. ref. https://github.com/bytecodealliance/wasm-micro-runtime/pull/4549#issuecomment-3218687294 * fix: update AOT compiler configuration and enhance error handling in fuzz tests	2026-03-09 18:24:43 -07:00
Yi Liu	163a91b05a	Fix relocation addend sign extension on 32-bit platforms (#4846) ... * Fix relocation addend sign extension on 32-bit platforms When loading relocations on 32-bit platforms, the addend is read as uint32 and zero-extended to uint64, which corrupts negative addends. For example, -4 (0xFFFFFFFC) becomes 4294967292 instead of remaining -4. Use int32 with sign extension to int64, matching the Windows code path which already handles this correctly.	2026-03-09 10:09:28 +08:00
Yi Liu	539bebed9c	Fix missing IS_INVALID_TAGINDEX check in RETHROW handler (#4837) ... * Fix RETHROW handler missing IS_INVALID_TAGINDEX check Add validation for exception_tag_index in the RETHROW opcode handler to prevent out-of-bounds access to module->module->tags[] when the tag index is INVALID_TAGINDEX (0xFFFFFFFF). This matches the existing check in the THROW handler. When CATCH_ALL catches a cross-module exception with an unknown tag, it pushes INVALID_TAGINDEX onto the stack. Without this check, a subsequent RETHROW would access tags[0xFFFFFFFF]. * Fix incorrect code section byte counts in exception handling test The hand-crafted WASM binary in load_module_with_exception_handling had an off-by-one in the code section: body size was declared as 7 but the actual body (local count + try/catch_all/end/end) is only 6 bytes. This caused the WASM loader to fail with "unexpected end" when it tried to read past the available bytes. Fix the body size from 7 to 6 and the code section size from 9 to 8.	2026-03-09 10:04:18 +08:00
Krisztian	fa7c2d37f4	clean(zephyr): reduce warnings on zephyr platform (#4860) ... * clean(warnings): fix [-Wsign-compare] in zephyr_file buf_len is a long unsigned int, while bytes_* can be negative due to error values. We don't need to check bytes_* for negative value, as it was done during read/write op. * clean(warnings): fix "MIN" redefined warning Some platforms, like Zephyr, already define MIN and definition in WAMR cause `warning: "MIN" redefined` warning. Check if it was defined before, and do not redefine it. Signed-off-by: Krisztian Szilvasi <34309983+kr-t@users.noreply.github.com>	2026-03-05 08:58:04 +08:00
Yi Liu	595dcd564f	Fix off-by-one in aot_alloc_tiny_frame overflow check (#4845) ... * Fix off-by-one in aot_alloc_tiny_frame overflow check The boundary check in aot_alloc_tiny_frame only verifies that new_frame itself doesn't exceed top_boundary, but doesn't account for the sizeof(AOTTinyFrame) bytes that are about to be written. When new_frame equals top_boundary exactly, the check passes but the subsequent write to new_frame->func_index goes past the boundary. This matches the correct pattern used in aot_alloc_frame (line 4086) which includes the frame size.	2026-03-03 13:48:42 +08:00
liang.he	01f4ff4cdb	workaround: prevent usage of anyref in struct fields and array elements (#4819) ... Disable corresponding cases in spec test	2026-03-03 13:05:44 +08:00
Yi Liu	3d707f9173	Add bounds checking for output tensor buffer in wasi-nn llama.cpp (#4847) ... * Add bounds checking for output tensor buffer in wasi-nn llama.cpp The get_output function copies LLM output into output_tensor->buf without checking against output_tensor->size, allowing writes past the buffer when the model generates output longer than the caller-provided buffer. Add size checks for both the metadata path and the token output loop. Instead of silently truncating output when the buffer is too small, return the too_large error with a diagnostic message. This makes the behavior consistent with the OpenVINO backend's get_output and allows callers to distinguish between successful completion and insufficient buffer size.	2026-03-02 14:52:16 -08:00
kogum4	dcf137e961	windows: fix vectored exception handler lifecycle for repeated thread env init/destroy (#4842) ... * Fix Windows VEH registration/removal lifecycle * Remove redundant OS_THREAD_MUTEX_INITIALIZER guards Since the code is already inside #ifdef BH_PLATFORM_WINDOWS, the macro is always defined. Use NULL directly for clarity. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-27 08:20:44 +08:00
Yi Liu	6a504e5ade	Fix sock_connect using wrong rights constant (#4849) ... wasi_ssp_sock_connect checks __WASI_RIGHT_SOCK_BIND instead of __WASI_RIGHT_SOCK_CONNECT when validating file descriptor rights. This was a copy-paste error from wasi_ssp_sock_bind. As a result, connect operations check the wrong capability.	2026-02-27 08:20:10 +08:00
Taylor C. Richberger	9143b1a2e3	fix: WASM_API_EXTERN should get visibility("default") on gcc and clang (#4851) ... This allows the wasm_c_api to work in shared libraries on non-Windows platforms.	2026-02-27 08:18:44 +08:00
liang.he	0c9fe614ae	Improve shared_heap test cases. (#4834) ... Aligned allocation size can be significantly greater than the original size, and page size varies across platforms. > macOS on M1 (Apple Silicon) uses a memory page size of 16,384 bytes (16 KB). > This differs from the traditional 4 KB page size used on Intel Macs and many > other ARM64 systems, and is designed to improve performance by reducing page > table overhead and allowing for better cache utilization.	2026-02-15 12:45:48 +08:00
Yi Liu	6db91d7e24	Fix off-by-one in AOT func_index bounds checks (#4836) ... The AOT relocation loader validates func_index using: (func_index = (uint32)atoi(p)) > module->func_count Since func_ptrs is an array of func_count elements (indices 0 to func_count-1), func_index == func_count is out of bounds. The check must use >= instead of > to reject this boundary case. Fix all 4 affected locations in aot_loader.c.	2026-02-13 19:21:54 +08:00
Zhenwei Jin	e4bdddadd1	align quick_aot_entry invoke_* signatures with call sites (#4829) ... Signed-off-by: zhenweijin <zhenwei.jin@intel.com>	2026-02-09 13:47:32 +08:00
liang.he	753a348b1f	Fix a compilation error about calling os_atomic_cmpxchg (#4822) ... ``` /workspaces/wasm-micro-runtime/core/iwasm/aot/aot_runtime.c:4701:30: error: expected expression return os_atomic_cmpxchg(_Atomic(void *)ptr, &old_val, new_val); ^ /workspaces/wasm-micro-runtime/core/shared/utils/../platform/include/platform_api_extension.h:135:27: note: expanded from macro 'os_atomic_cmpxchg' ```	2026-02-03 13:56:59 +08:00
TianlongLiang	733ac3f19d	Collective fix (#4808) ... * fix a bug in zephyr platform file api * fix a bug in bh queue * fix a bug in shared heap malloc when it's memory64	2026-01-28 15:55:19 +08:00
Zhenwei Jin	04abbbcfbe	gc: complete common heap type coverage in wasm_is_refheaptype_common() (#4801) ... Signed-off-by: zhenweijin <zhenwei.jin@intel.com>	2026-01-28 13:03:20 +08:00
TianlongLiang	8d7d93d9b9	Correct some error handling path in libc_emcc (#4803)	2026-01-26 17:15:26 +08:00
TianlongLiang	d58586a927	fix code and comment typo (#4802)	2026-01-26 14:44:47 +08:00
Zhenwei Jin	432ba8b3e0	Fix memcpy overlap issue in RECOVER_BR_INFO for i64/v128 copy (#4797) ... When copying single i64 or V128 values in RECOVER_BR_INFO, source and destination memory regions may overlap, causing memcpy-param-overlap errors Use temporary variables to separate read and write operations, preventing the overlap issue. This fix references the approach used in the other path (when arity != 1), which calls copy_stack_values, that function explicitly handles memcpy overlap. Signed-off-by: zhenweijin <zhenwei.jin@intel.com>	2026-01-26 09:59:34 +08:00
Zhenwei Jin	9a23968fda	Fix potential memory leak when discarding FreeMachineFunction (#4796) ... Signed-off-by: zhenweijin <zhenwei.jin@intel.com>	2026-01-21 11:02:33 +08:00
Zhenwei Jin	7b750d1652	Fix alignment calculation for large offsets (#4791) ... Signed-off-by: zhenweijin <zhenwei.jin@intel.com>	2026-01-20 09:37:29 +08:00
linear0211	a7fd6539dc	Add bh_strtok_r function (#4790)	2026-01-20 09:13:22 +08:00
Zhenwei Jin	23df0d4e55	fix: correct boundary check in dynamic_offset check (#4788) ... correct boundary check in check_dynamic_offset_pop when dynamic_offset is 0. When dynamic_offset = 0, check_dynamic_offset_pop will always return true, which may wrongly update dynamic_offset. also include a typo fix in SET_OPERAND_REF Signed-off-by: zhenweijin <zhenwei.jin@intel.com>	2026-01-20 08:53:11 +08:00
Steven	29767f6b51	fix(aot): reserve x18 register on macOS ARM64 (#4775) ... * fix(aot): reserve x18 register on macOS ARM64 Apple reserves CPU register x18 for TLS on ARM64. When generating AOT code for aarch64 on macOS, LLVM may use x18, causing crashes when the AOT code runs on macOS ARM64 (M1/M2/M3). This patch: 1. Detects darwin/macho ABI and sets correct vendor string 2. Detects darwin/apple in default triple for platform detection 3. Adds +reserve-x18 to LLVM target features for aarch64 on macOS The fix only applies when compiling on macOS ARM64 hosts, ensuring generated AOT code is compatible with Apple's platform requirements. * test(aot): add x18 register reservation test for macOS ARM64 Add a standalone test to verify that the +reserve-x18 LLVM flag is correctly applied when compiling AOT for macOS ARM64. On macOS ARM64, x18 is reserved by Apple for TLS (Thread Local Storage). Without the +reserve-x18 flag, LLVM may generate code that uses x18, causing random SIGSEGV crashes (~80% crash rate in testing). The test: - Creates a WASM module with 24 local variables to stress register allocation - Compiles to AOT with -O3 optimization (which would use x18 without the fix) - Runs 1000 iterations to verify no crashes occur - Only runs on macOS ARM64 (skipped on other platforms) Test results: - Without fix: 82/100 crash rate - With fix: 0/100 crash rate (1000 iterations verified)	2026-01-20 07:30:58 +08:00
Steven	5664589daa	fix: clear exec_env_tls when destroying exec_env (#4774) ... * fix: clear exec_env_tls when destroying exec_env When an exec_env is destroyed, check if it matches the current thread's exec_env_tls and clear it to avoid dangling pointer issues. Without this fix, in daemon-style execution where the same thread runs multiple WASM modules sequentially (like Cloudflare Workers), the exec_env_tls can point to freed memory after an exec_env is destroyed, causing crashes on subsequent executions when the signal handler tries to access it. This is critical for AOT mode with hardware bounds checking enabled, where signal handlers rely on exec_env_tls to handle SIGSEGV properly. * test(exec_env): add reproducer for exec_env_tls dangling pointer bug Add test case that reproduces the bug where exec_env_tls is not cleared on early return paths in invoke_native_with_hw_bound_check. The test triggers native stack overflow check failure, which causes wasm_runtime_call_wasm to return early after setting exec_env_tls but without clearing it. This leaves exec_env_tls pointing to a destroyed exec_env, causing subsequent calls to fail with "invalid exec env". Test confirms the fix in wasm_exec_env_destroy correctly clears exec_env_tls when destroying the exec_env it points to. * fix(runtime): clear exec_env_tls on early return from stack overflow check Move the fix to clear exec_env_tls at the source - in the early return path of invoke_native_with_hw_bound_check when native stack overflow check fails.	2026-01-20 07:09:26 +08:00
石博文	3c337dc3ab	fix: Wrong compiler parameter on MSVC (#4778) ... * fix: Wrong compiler parameter on MSVC * refactor: Move gcc flags to common build script. * refactor: Add both C/CXX flags	2026-01-08 09:10:07 +08:00
Huang Qi	b9159101cf	aot: avoid forcing mmap low addresses on NuttX (#4776) ... Signed-off-by: Huang Qi <huangqi3@xiaomi.com>	2026-01-05 10:57:18 +08:00
linear0211	07466407ba	Ensure --addr-pool mask accepts numbers only (#4619) ... * Ensure --addr-pool mask accepts numbers only * Add mask validation * Replace mask assignment position * Use a thread-safe function and free allocated memory Co-authored-by: liang.he <liang.he@intel.com>	2025-12-22 07:55:42 +08:00
TianlongLiang	f8448b6d79	Add a runtime API for reset shared heap(chain) (#4740) ... * add new runtime api for reset shared heap(chain) * add new unit test case	2025-12-15 09:56:55 +08:00
Krisztian	69446f4abb	Add missing functions from thumb arch (#4718) ... Signed-off-by: Krisztian Szilvasi <34309983+kr-t@users.noreply.github.com>	2025-11-28 17:08:24 +08:00
TianlongLiang	3c01e2ae0c	fix double promotion warning (#4731)	2025-11-28 11:06:43 +08:00
YAMAMOTO Takashi	912c2a6e31	add a set of apis to configure wasi via InstantiationArgs2 (#4707) ... * track if WASIArguments is configured by user i plan to use this to decide which wasi arguments (the one from module or the one from InstantiationArgs2) to use. * add WASIArguments to InstantiationArgs2 * use wasi configuration from InstantiationArgs2 if any fallback to the via-module configuration for now. * add a few api to configure wasi via InstantiationArgs2 * configure wasi via InstantiationArgs2 for platforms using libc_wasi.c * rt-thread: migrate to libc_wasi_set_init_args * common/libc_wasi.c: retire libc_wasi_init * fix build without wasi	2025-11-24 19:20:21 +08:00
Huang Qi	1d255ce0b5	refactor: Improve debug experience with CFI directives and proper label naming in Thumb invokeNative (#4719) ... This change enhances debugging support and improves assembly code quality: 1. Add CFI (Call Frame Information) directives (.cfi_startproc, .cfi_def_cfa_offset, .cfi_offset, .cfi_endproc) to provide proper call frame information for debuggers 2. Improve label naming by using dot-prefixed conventions (.Lreturn, .Lcall_func, etc.) following assembly best practices These improvements enhance debugging capabilities and code maintainability while preserving functionality. Signed-off-by: Huang Qi <huangqi3@xiaomi.com>	2025-11-24 19:19:18 +08:00
liang.he	53ffe451d0	Fast-interp should report unsupported opcode (#4723) ... when meeting simd opcode on non-posix platforms	2025-11-24 15:48:29 +08:00
liang.he	071e2aaed5	Merge commit from fork ... * fix: remove SIMD prefix handling from interpreter and update opcode definitions * test: add case for classic interpreter handling of unsupported SIMD opcodes	2025-11-24 14:37:46 +08:00
Zhenwei Jin	33fdc4a671	Merge commit from fork	2025-11-24 14:23:04 +08:00
Huang Qi	e2da849f11	wamr/arm: Add CFI directives to Thumb VFP assembly for better debugging (#4697) ... Add Call Frame Information (CFI) directives to invokeNative_thumb_vfp.s to improve stack unwinding and debugging capabilities on ARM platforms. This enables better backtrace generation and crash analysis. Changes include: - Add .cfi_startproc/.cfi_endproc directives for proper frame tracking - Add .cfi_def_cfa_offset and .cfi_def_cfa to track stack pointer changes - Add .cfi_offset directives to track saved register locations - Convert local labels to use .L prefix following assembly conventions These CFI directives allow debuggers and exception handlers to properly unwind the stack through native function calls, improving the debugging experience for WAMR applications on ARM targets. Signed-off-by: Huang Qi <huangqi3@xiaomi.com>	2025-11-17 17:55:05 +08:00
TianlongLiang	7898af9f21	fix bug and add unit test case for runtime api when shared heap is enabled (#4695)	2025-11-11 16:22:45 +08:00
Joe Polny	4b42cfdbce	fix: remove stack size check from exec_env_check (#4688) ... As per [this blogpost](https://bytecodealliance.github.io/wamr.dev/blog/understand-the-wamr-stacks/) WAMR does not require a WASM stack in AoT/JIT mode, but `wasm_runtime_exec_env_check` still always expects the size to be > 0	2025-11-05 08:10:02 +08:00
Joe Polny	238f45dddc	docs: clarify that * in signatures does address conversion (#4681) ... docs: clarify that * and $ in signatures does address conversion It was non-obvious that `*` means the runtime does the app to native conversion automatically. With the current docstring, one might assume that the runtime only does boundary checks and not address conversion. Under this assumption, passing the already-native address to `wasm_runtime_addr_app_to_native` will lead to unexpected behavior	2025-10-31 15:41:35 +08:00
YAMAMOTO Takashi	3b8b265f1f	Pass InstantiationArgs2 down to aot_instantiate/wasm_instantiate (#4594) ... This is a preparation for https://github.com/bytecodealliance/wasm-micro-runtime/issues/4364 No functional changes are intended.	2025-10-23 15:57:50 +08:00
Xenia Lu	4c371e655a	fix: false OOB in array.fill for interp (#4645) ... * cherry-pick gc spec test case * Fix false OOB in array.fill	2025-10-23 13:15:42 +08:00