Skip to content

Fix wait_info data race for deletion + fix atomic_wait logic #2016

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

Merged
wenyongh merged 15 commits into from
Mar 13, 2023
Merged

Fix wait_info data race for deletion + fix atomic_wait logic #2016

Show file tree
Hide file tree
Changes from 9 commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
c0f13de
Fix wait_info data race for deletion
g0djan Mar 8, 2023
8fdbc12
clang format
g0djan Mar 8, 2023
b30a14c
Wait_info->wait_list access always should be under its lock actually
g0djan Mar 8, 2023
962e46f
forgot to set initial count_acquisition to 1
g0djan Mar 8, 2023
327c1a3
Merge branch 'main' into godjan/fix-wait-info-data-race
g0djan Mar 10, 2023
2f69b0d
use wait_list->len, not counter
g0djan Mar 10, 2023
3e1a974
clang format
g0djan Mar 10, 2023
bc9fa8b
Fix wromg waiting on atomic when proc_exit was called before thread r…
g0djan Mar 11, 2023
388684d
clang-format
g0djan Mar 11, 2023
8bf4aab
remove timeouts to catch actual errors for atomic_wait with BLOCKING_…
g0djan Mar 11, 2023
1efcb64
clang-format-12
g0djan Mar 11, 2023
bcd639d
fix passed args for fast interpreter
g0djan Mar 11, 2023
6c21413
Address review comments
g0djan Mar 11, 2023
e81fa95
Added warning about using wait-info->wait_list
g0djan Mar 11, 2023
2970e7d
fix comments
g0djan Mar 11, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
93 changes: 55 additions & 38 deletions core/iwasm/common/wasm_shared_memory.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,9 @@

#include "bh_log.h"
#include "wasm_shared_memory.h"
#if WASM_ENABLE_THREAD_MGR != 0
#include "../libraries/thread-mgr/thread_manager.h"
#endif

static bh_list shared_memory_list_head;
static bh_list *const shared_memory_list = &shared_memory_list_head;
Expand Down Expand Up @@ -298,7 +301,7 @@ notify_wait_list(bh_list *wait_list, uint32 count)
}

static AtomicWaitInfo *
acquire_wait_info(void *address, bool create)
acquire_wait_info(void *address, bool create, AtomicWaitNode *wait_node)
Comment thread
wenyongh marked this conversation as resolved.
Outdated
{
AtomicWaitInfo *wait_info = NULL;
bh_list_status ret;
Expand All @@ -308,7 +311,7 @@ acquire_wait_info(void *address, bool create)
if (address)
wait_info = (AtomicWaitInfo *)bh_hash_map_find(wait_map, address);

if (!create) {
if (!create && !wait_info) {
os_mutex_unlock(&wait_map_lock);
return wait_info;
}
Expand All @@ -335,6 +338,13 @@ acquire_wait_info(void *address, bool create)
goto fail3;
}
}
if (wait_node) {
os_mutex_lock(&wait_info->wait_list_lock);
ret = bh_list_insert(wait_info->wait_list, wait_node);
os_mutex_unlock(&wait_info->wait_list_lock);
bh_assert(ret == BH_LIST_SUCCESS);
(void)ret;
}

os_mutex_unlock(&wait_map_lock);

Expand Down Expand Up @@ -380,17 +390,24 @@ static bool
map_remove_wait_info(HashMap *wait_map_, AtomicWaitInfo *wait_info,
void *address)
{
os_mutex_lock(&wait_map_lock);
Comment thread
g0djan marked this conversation as resolved.
os_mutex_lock(&wait_info->wait_list_lock);
if (wait_info->wait_list->len > 0) {
Comment thread
g0djan marked this conversation as resolved.
os_mutex_unlock(&wait_info->wait_list_lock);
os_mutex_unlock(&wait_map_lock);
return false;
}
os_mutex_unlock(&wait_info->wait_list_lock);
Comment thread
wenyongh marked this conversation as resolved.

bh_hash_map_remove(wait_map_, address, NULL, NULL);
os_mutex_unlock(&wait_map_lock);
return true;
}

uint32
wasm_runtime_atomic_wait(WASMModuleInstanceCommon *module, void *address,
uint64 expect, int64 timeout, bool wait64)
uint64 expect, int64 timeout, bool wait64,
WASMExecEnv *exec_env)
Comment thread
wenyongh marked this conversation as resolved.
Outdated
{
WASMModuleInstance *module_inst = (WASMModuleInstance *)module;
AtomicWaitInfo *wait_info;
Expand Down Expand Up @@ -418,14 +435,6 @@ wasm_runtime_atomic_wait(WASMModuleInstanceCommon *module, void *address,
return -1;
}

/* acquire the wait info, create new one if not exists */
wait_info = acquire_wait_info(address, true);

if (!wait_info) {
wasm_runtime_set_exception(module, "failed to acquire wait_info");
return -1;
}

node = search_module((WASMModuleCommon *)module_inst->module);
os_mutex_lock(&node->shared_mem_lock);
no_wait = (!wait64 && *(uint32 *)address != (uint32)expect)
Expand All @@ -435,40 +444,48 @@ wasm_runtime_atomic_wait(WASMModuleInstanceCommon *module, void *address,
if (no_wait) {
return 1;
}
else {
bh_list_status ret;

if (!(wait_node = wasm_runtime_malloc(sizeof(AtomicWaitNode)))) {
wasm_runtime_set_exception(module, "failed to create wait node");
return -1;
}
memset(wait_node, 0, sizeof(AtomicWaitNode));
if (!(wait_node = wasm_runtime_malloc(sizeof(AtomicWaitNode)))) {
wasm_runtime_set_exception(module, "failed to create wait node");
return -1;
}
memset(wait_node, 0, sizeof(AtomicWaitNode));

if (0 != os_mutex_init(&wait_node->wait_lock)) {
wasm_runtime_free(wait_node);
return -1;
}
if (0 != os_mutex_init(&wait_node->wait_lock)) {
wasm_runtime_free(wait_node);
return -1;
}

if (0 != os_cond_init(&wait_node->wait_cond)) {
os_mutex_destroy(&wait_node->wait_lock);
wasm_runtime_free(wait_node);
return -1;
}
if (0 != os_cond_init(&wait_node->wait_cond)) {
os_mutex_destroy(&wait_node->wait_lock);
wasm_runtime_free(wait_node);
return -1;
}

wait_node->status = S_WAITING;
os_mutex_lock(&wait_info->wait_list_lock);
ret = bh_list_insert(wait_info->wait_list, wait_node);
os_mutex_unlock(&wait_info->wait_list_lock);
bh_assert(ret == BH_LIST_SUCCESS);
(void)ret;
wait_node->status = S_WAITING;

/* acquire the wait info, create new one if not exists */
wait_info = acquire_wait_info(address, true, wait_node);
Comment thread
wenyongh marked this conversation as resolved.
Outdated

if (!wait_info) {
os_mutex_destroy(&wait_node->wait_lock);
wasm_runtime_free(wait_node);
wasm_runtime_set_exception(module, "failed to acquire wait_info");
return -1;
}

/* condition wait start */
os_mutex_lock(&wait_node->wait_lock);

os_cond_reltimedwait(&wait_node->wait_cond, &wait_node->wait_lock,
timeout < 0 ? BHT_WAIT_FOREVER
: (uint64)timeout / 1000);
#if WASM_ENABLE_THREAD_MGR != 0
if (!wasm_cluster_is_thread_terminated(exec_env)) {
Comment thread
wenyongh marked this conversation as resolved.
Outdated

@g0djan g0djan Mar 11, 2023

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Concerned about if and not while here: spurious wakeup
@wenyongh @eloparco any thought on what ret should be considered as appropriate one to exit loop or what codes should be considered a false wakeup?

@eloparco eloparco Mar 11, 2023

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You mean the flag being set between the call to wasm_cluster_is_thread_terminated and os_cond_reltimedwait, right? That would cause the termination to be missed and the os_cond_reltimedwait to wait forever in case of BHT_WAIT_FOREVER

@g0djan g0djan Mar 11, 2023 •
edited
Loading

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, the problem you talking about I solved with your idea with condition variable(I just found a way to do it with the existing one):
thread 1

  • lock wait_node->wait_lock
  • check whether it's been already terminated
  • call wait on the condition variable that releases the lock

thread 2

  • lock wait_node->wait_lock
  • call notify on condition variable
  • unlock wait_node->wait_lock

To check I ran the tests and it never hangs any more, no tsan warnings as well(besides unrelated LOAD-STORE one)

@g0djan g0djan Mar 11, 2023 •
edited
Loading

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here I'm concerned about spurious wakeup, on some systems thread can actually wake up from wait on condition variable even there was no other thread in this program that notified it. Usually it's prevented by checking exit condition rather in a loop than just with an if.
https://en.wikipedia.org/wiki/Spurious_wakeup

@eloparco eloparco Mar 12, 2023

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Then I presume we want to add a while with a check on the wait_node->status

#endif
os_cond_reltimedwait(&wait_node->wait_cond, &wait_node->wait_lock,
timeout < 0 ? BHT_WAIT_FOREVER
: (uint64)timeout / 1000);
#if WASM_ENABLE_THREAD_MGR != 0
}
#endif

is_timeout = wait_node->status == S_WAITING ? true : false;
os_mutex_unlock(&wait_node->wait_lock);
Expand All @@ -486,8 +503,8 @@ wasm_runtime_atomic_wait(WASMModuleInstanceCommon *module, void *address,
wasm_runtime_free(wait_node);

/* Release wait info if no wait nodes attached */
removed_from_map = map_remove_wait_info(wait_map, wait_info, address);
os_mutex_unlock(&wait_info->wait_list_lock);
removed_from_map = map_remove_wait_info(wait_map, wait_info, address);
if (removed_from_map)
destroy_wait_info(wait_info);
Comment thread
wenyongh marked this conversation as resolved.
Outdated
os_mutex_unlock(&node->shared_mem_lock);
Expand Down Expand Up @@ -523,7 +540,7 @@ wasm_runtime_atomic_notify(WASMModuleInstanceCommon *module, void *address,
return -1;
}

wait_info = acquire_wait_info(address, false);
wait_info = acquire_wait_info(address, false, NULL);

/* Nobody wait on this address */
if (!wait_info) {
Expand Down
7 changes: 6 additions & 1 deletion core/iwasm/common/wasm_shared_memory.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,10 @@
#define _WASM_SHARED_MEMORY_H

#include "bh_common.h"
#include "wasm_exec_env.h"
#if WASM_ENABLE_THREAD_MGR != 0
#include "../libraries/thread-mgr/thread_manager.h"
#endif
#if WASM_ENABLE_INTERP != 0
#include "wasm_runtime.h"
#endif
Expand Down Expand Up @@ -60,7 +64,8 @@ shared_memory_set_memory_inst(WASMModuleCommon *module,

uint32
wasm_runtime_atomic_wait(WASMModuleInstanceCommon *module, void *address,
uint64 expect, int64 timeout, bool wait64);
uint64 expect, int64 timeout, bool wait64,
WASMExecEnv *exec_env);

uint32
wasm_runtime_atomic_notify(WASMModuleInstanceCommon *module, void *address,
Expand Down
14 changes: 10 additions & 4 deletions core/iwasm/interpreter/wasm_interp_classic.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1195,7 +1195,10 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
goto got_exception;
}

HANDLE_OP(WASM_OP_NOP) { HANDLE_OP_END(); }
HANDLE_OP(WASM_OP_NOP)
{
HANDLE_OP_END();
}

HANDLE_OP(EXT_OP_BLOCK)
{
Expand Down Expand Up @@ -3032,7 +3035,10 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,
HANDLE_OP(WASM_OP_I32_REINTERPRET_F32)
HANDLE_OP(WASM_OP_I64_REINTERPRET_F64)
HANDLE_OP(WASM_OP_F32_REINTERPRET_I32)
HANDLE_OP(WASM_OP_F64_REINTERPRET_I64) { HANDLE_OP_END(); }
HANDLE_OP(WASM_OP_F64_REINTERPRET_I64)
{
HANDLE_OP_END();
}

HANDLE_OP(WASM_OP_I32_EXTEND8_S)
{
Expand Down Expand Up @@ -3428,7 +3434,7 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,

ret = wasm_runtime_atomic_wait(
(WASMModuleInstanceCommon *)module, maddr,
(uint64)expect, timeout, false);
(uint64)expect, timeout, false, exec_env);
if (ret == (uint32)-1)
goto got_exception;

Expand All @@ -3452,7 +3458,7 @@ wasm_interp_call_func_bytecode(WASMModuleInstance *module,

ret = wasm_runtime_atomic_wait(
(WASMModuleInstanceCommon *)module, maddr, expect,
timeout, true);
timeout, true, exec_env);
if (ret == (uint32)-1)
goto got_exception;

Expand Down