Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Native stack overflow issue when issuing wasm_runtime_init_thread_env() API #3909

Open
rpatwa701 opened this issue Nov 13, 2024 · 8 comments
Open

Native stack overflow issue when issuing wasm_runtime_init_thread_env() API #3909

rpatwa701 opened this issue Nov 13, 2024 · 8 comments

Comments

@rpatwa701
Copy link

rpatwa701 commented Nov 13, 2024 •
edited
Loading

Hi All,

I have developed a WASM containerized application. I run the application using iwasm runtime. I have a scenario where the WASM application registers a callback with the native library. I am using the below sequence of APIs to lookup and invoke the callback from the native library:

void register_callback_native(wasm_exec_env_t exec_env, int val, void *test, char *cb, int len)
{
    wasm_module_inst_t module_inst;
    wasm_function_inst_t func = NULL;
    struct example testapp = {10,'a'};
    uint32_t argv[1];

    argv[0] = 15;
    if (!wasm_runtime_init_thread_env()) {
        printf("Failed to initialize thread environment\n");
        return;
    }

    memcpy(test, &testapp, sizeof(struct example));
    module_inst = wasm_runtime_get_module_inst(exec_env);

    func = wasm_runtime_lookup_function(module_inst, cb);

    if (func == NULL) {
        printf("lookup failed\n");
    } else {
        if (!wasm_runtime_call_wasm(exec_env, func, 1, argv)) {
            printf("%s[%d]: %s", __func__, __LINE__,
                    wasm_runtime_get_exception(module_inst));
        }
    }
}

The above code works just fine on my x86 Ubuntu machine, however when I try running the same application on my ARM platform I see the below error:
Exception: native stack overflow
I narrowed it down and found wasm_runtime_init_thread_env() API is leading to this error.

I tried increassing the stack and heap size allocation when starting the application but that doesn't seem to help. Is there something missing here or am I using the wrong WASM runtime APIs?
@TianlongLiang
Copy link
Contributor

Is it still Ubuntu, or is it MacOS or Windows?

@rpatwa701
Copy link
Author

It is still Ubuntu on ARM platform

@rpatwa701
Copy link
Author

rpatwa701 commented Nov 14, 2024 •
edited
Loading

Running this on iwasm-1.3.2

After putting additional debug we observe the issue is in the below call flow when we invoke wasm_runtime_init_thread_env():
wam_runtime_init_thread_env
runtime_signal_init
os_thread_signal_init
init_stack_guard_pages
touch_pages

In touch_pages() seems like after a point os_alloca() leads to the native stack overflow error. I observed touch_pages called even during wasm_runtime_init() and for that function I see the touch_pages() API returns succesfully. Attaching the logs with debug prints leading upto the error:

Entered touch_pages function
Stack Min Address: 0x7fe5bae000
Page Size: 4096
uint8 sum = 0;
while (1)
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
if (touch_addr < stack_min_addr + page_size)
*touch_addr = 0;
sum += *touch_addr;
volatile uint8 *touch_addr = (volatile uint8 *)os_alloca(page_size / 2);
Exception: native stack overflow

Would appreciate if someone can help understand when would os_alloca() call cause stack overflow and how can this issue be resolved.

@rpatwa701
Copy link
Author

rpatwa701 commented Nov 14, 2024 •
edited
Loading

Building iwasm with the below flags, attaching trimmed patch:

Index: wasm-micro-runtime-WAMR-1.3.2/CMakeLists.txt
===================================================================
--- wasm-micro-runtime-WAMR-1.3.2.orig/CMakeLists.txt
+++ wasm-micro-runtime-WAMR-1.3.2/CMakeLists.txt
@@ -17,6 +17,10 @@ set (CMAKE_SHARED_LIBRARY_LINK_CXX_FLAGS
 
 set (CMAKE_C_STANDARD 99)
 
+if("${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "arm")
+  set(WAMR_BUILD_TARGET "ARMv7")
+endif()

@@ -82,7 +86,7 @@ endif ()
 
+  set (WAMR_BUILD_LIB_PTHREAD 1)
 endif ()
 
 if (NOT DEFINED WAMR_BUILD_LIB_WASI_THREADS)
@@ -154,6 +158,8 @@ if (MINGW)
   target_link_libraries (iwasm_shared -lWs2_32)
 endif ()
 
+ADD_SUBDIRECTORY(product-mini/platforms/linux)
+
 
 # HEADERS
Index: wasm-micro-runtime-WAMR-1.3.2/product-mini/platforms/linux/CMakeLists.txt
===================================================================
--- wasm-micro-runtime-WAMR-1.3.2.orig/product-mini/platforms/linux/CMakeLists.txt
+++ wasm-micro-runtime-WAMR-1.3.2/product-mini/platforms/linux/CMakeLists.txt
@@ -71,6 +71,8 @@ if (NOT DEFINED WAMR_BUILD_LIBC_WASI)
   set (WAMR_BUILD_LIBC_WASI 1)
 endif ()
 
+set(WAMR_BUILD_LIBC_EMCC 1)

+  set (WAMR_BUILD_LIB_PTHREAD 1)
 endif ()
 
+set (RUNTIME_SOURCE_ALL
+    ${WAMR_ROOT_DIR}/product-mini/platforms/posix/main.c
+    ${UNCOMMON_SHARED_SOURCE}
+)
+

@rpatwa701
Copy link
Author

I was able to go past the issue on my AARCH platform by building iwasm with cmake -DWAMR_DISABLE_STACK_HW_BOUND_CHECK=1 as suggested in this thread #2901. However I am still unable to understand what this check does and what are the implications of disabling this?

@TianlongLiang
Copy link
Contributor

TianlongLiang commented Nov 18, 2024 •
edited
Loading

Could you print the touch_addr too?

The alloca dynamically allocates memory on the stack rather than the heap. From your log, I think there could be something wrong with the stack_min_addr calculation on that AARCH platform(due to memory layout difference maybe?), so it simply keeps allocating until the native stack boundary is actually touched.

If stack_min_addr is calculated correctly, this if statement would simply break the outer for loop, and touch_addr wouldn't be actually touched(dereferenced)

        if (touch_addr < stack_min_addr + page_size) {
            sum += *(stack_min_addr + page_size - 1);
            break;
        }

@TianlongLiang
Copy link
Contributor

TianlongLiang commented Nov 18, 2024 •
edited
Loading

I was able to go past the issue on my AARCH platform by building iwasm with cmake -DWAMR_DISABLE_STACK_HW_BOUND_CHECK=1 as suggested in this thread #2901. However I am still unable to understand what this check does and what are the implications of disabling this?

Basically this flag does is that for each thread's stack, WAMR calculates the stack boundary for it and sets a guard page for it(touch it will raise a signal), to keep it from stack overflow. When you disable it, chances are you are fine, it's not that easy to actually cause a stack overflow in normal cases.

@schorg
Copy link

schorg commented Nov 18, 2024 •
edited
Loading

Just as a hint: I experience the same problem on Linux aarch64 (Virtual Machine on a Mac) with WAMR-1.3.3.

Did not upgrade for a long time. Everything worked fine, in the commit I used for a long time, maybe this helps locating the issue.

commit f0632edc37543aa1493efb9dbcd836eb58dbed23 (HEAD)
Author: liang.he <[email protected]>
Date:   Mon Aug 21 19:43:28 2023 +0800

    Clone the input binary during wasm_module_validate (#2483)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@schorg @rpatwa701 @TianlongLiang