* [PATCH 0/2] selftests/mm: fix deadlock after pthread_create
@ 2024-10-03 21:17 Edward Liaw
2024-10-03 21:17 ` [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t Edward Liaw
` (2 more replies)
0 siblings, 3 replies; 7+ messages in thread
From: Edward Liaw @ 2024-10-03 21:17 UTC (permalink / raw)
To: linux-kselftest, Andrew Morton, Shuah Khan, Lokesh Gidra,
Edward Liaw, Peter Xu
Cc: linux-kernel, kernel-team, linux-mm
On Android arm, pthread_create followed by a fork caused a deadlock in
the case where the fork required work to be completed by the created
thread.
Updated the synchronization primitive to use pthread_barrier instead of
atomic_bool.
Applied the same fix to the wp-fork-with-event test.
Edward Liaw (2):
selftests/mm: replace atomic_bool with pthread_barrier_t
selftests/mm: fix deadlock for fork after pthread_create on ARM
tools/testing/selftests/mm/uffd-common.c | 5 +++--
tools/testing/selftests/mm/uffd-common.h | 3 +--
tools/testing/selftests/mm/uffd-unit-tests.c | 21 ++++++++++++++------
3 files changed, 19 insertions(+), 10 deletions(-)
--
2.46.1.824.gd892dcdcdd-goog
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t
2024-10-03 21:17 [PATCH 0/2] selftests/mm: fix deadlock after pthread_create Edward Liaw
@ 2024-10-03 21:17 ` Edward Liaw
2024-10-06 17:53 ` Muhammad Usama Anjum
2024-10-18 14:37 ` Ryan Roberts
2024-10-03 21:17 ` [PATCH 2/2] selftests/mm: fix deadlock for fork after pthread_create on ARM Edward Liaw
2024-10-04 23:54 ` [PATCH 0/2] selftests/mm: fix deadlock after pthread_create Andrew Morton
2 siblings, 2 replies; 7+ messages in thread
From: Edward Liaw @ 2024-10-03 21:17 UTC (permalink / raw)
To: linux-kselftest, Andrew Morton, Shuah Khan, Lokesh Gidra,
Edward Liaw, Peter Xu
Cc: linux-kernel, kernel-team, linux-mm
Swaps synchronization primitive with pthread_barrier, so that
stdatomic.h does not need to be included.
The synchronization is needed on Android ARM64; we see a deadlock with
pthread_create when the parent thread races forward before the child has
a chance to start doing work.
Fixes: 8c864371b2a1 ("selftests/mm: fix ARM related issue with fork after
pthread_create")
Signed-off-by: Edward Liaw <edliaw@google.com>
---
tools/testing/selftests/mm/uffd-common.c | 5 +++--
tools/testing/selftests/mm/uffd-common.h | 3 +--
tools/testing/selftests/mm/uffd-unit-tests.c | 14 ++++++++------
3 files changed, 12 insertions(+), 10 deletions(-)
diff --git a/tools/testing/selftests/mm/uffd-common.c b/tools/testing/selftests/mm/uffd-common.c
index 717539eddf98..852e7281026e 100644
--- a/tools/testing/selftests/mm/uffd-common.c
+++ b/tools/testing/selftests/mm/uffd-common.c
@@ -18,7 +18,7 @@ bool test_uffdio_wp = true;
unsigned long long *count_verify;
uffd_test_ops_t *uffd_test_ops;
uffd_test_case_ops_t *uffd_test_case_ops;
-atomic_bool ready_for_fork;
+pthread_barrier_t ready_for_fork;
static int uffd_mem_fd_create(off_t mem_size, bool hugetlb)
{
@@ -519,7 +519,8 @@ void *uffd_poll_thread(void *arg)
pollfd[1].fd = pipefd[cpu*2];
pollfd[1].events = POLLIN;
- ready_for_fork = true;
+ /* Ready for parent thread to fork */
+ pthread_barrier_wait(&ready_for_fork);
for (;;) {
ret = poll(pollfd, 2, -1);
diff --git a/tools/testing/selftests/mm/uffd-common.h b/tools/testing/selftests/mm/uffd-common.h
index a70ae10b5f62..3e6228d8e0dc 100644
--- a/tools/testing/selftests/mm/uffd-common.h
+++ b/tools/testing/selftests/mm/uffd-common.h
@@ -33,7 +33,6 @@
#include <inttypes.h>
#include <stdint.h>
#include <sys/random.h>
-#include <stdatomic.h>
#include "../kselftest.h"
#include "vm_util.h"
@@ -105,7 +104,7 @@ extern bool map_shared;
extern bool test_uffdio_wp;
extern unsigned long long *count_verify;
extern volatile bool test_uffdio_copy_eexist;
-extern atomic_bool ready_for_fork;
+extern pthread_barrier_t ready_for_fork;
extern uffd_test_ops_t anon_uffd_test_ops;
extern uffd_test_ops_t shmem_uffd_test_ops;
diff --git a/tools/testing/selftests/mm/uffd-unit-tests.c b/tools/testing/selftests/mm/uffd-unit-tests.c
index b3d21eed203d..3db2296ac631 100644
--- a/tools/testing/selftests/mm/uffd-unit-tests.c
+++ b/tools/testing/selftests/mm/uffd-unit-tests.c
@@ -774,7 +774,7 @@ static void uffd_sigbus_test_common(bool wp)
char c;
struct uffd_args args = { 0 };
- ready_for_fork = false;
+ pthread_barrier_init(&ready_for_fork, NULL, 2);
fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
@@ -791,8 +791,9 @@ static void uffd_sigbus_test_common(bool wp)
if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
err("uffd_poll_thread create");
- while (!ready_for_fork)
- ; /* Wait for the poll_thread to start executing before forking */
+ /* Wait for child thread to start before forking */
+ pthread_barrier_wait(&ready_for_fork);
+ pthread_barrier_destroy(&ready_for_fork);
pid = fork();
if (pid < 0)
@@ -833,7 +834,7 @@ static void uffd_events_test_common(bool wp)
char c;
struct uffd_args args = { 0 };
- ready_for_fork = false;
+ pthread_barrier_init(&ready_for_fork, NULL, 2);
fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
if (uffd_register(uffd, area_dst, nr_pages * page_size,
@@ -844,8 +845,9 @@ static void uffd_events_test_common(bool wp)
if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
err("uffd_poll_thread create");
- while (!ready_for_fork)
- ; /* Wait for the poll_thread to start executing before forking */
+ /* Wait for child thread to start before forking */
+ pthread_barrier_wait(&ready_for_fork);
+ pthread_barrier_destroy(&ready_for_fork);
pid = fork();
if (pid < 0)
--
2.46.1.824.gd892dcdcdd-goog
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH 2/2] selftests/mm: fix deadlock for fork after pthread_create on ARM
2024-10-03 21:17 [PATCH 0/2] selftests/mm: fix deadlock after pthread_create Edward Liaw
2024-10-03 21:17 ` [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t Edward Liaw
@ 2024-10-03 21:17 ` Edward Liaw
2024-10-04 23:54 ` [PATCH 0/2] selftests/mm: fix deadlock after pthread_create Andrew Morton
2 siblings, 0 replies; 7+ messages in thread
From: Edward Liaw @ 2024-10-03 21:17 UTC (permalink / raw)
To: linux-kselftest, Andrew Morton, Shuah Khan, Lokesh Gidra,
Edward Liaw, Peter Xu
Cc: linux-kernel, kernel-team, linux-mm
On Android with arm, there is some synchronization needed to avoid a
deadlock when forking after pthread_create.
Fixes: cff294582798 ("selftests/mm: extend and rename uffd pagemap test")
Signed-off-by: Edward Liaw <edliaw@google.com>
---
tools/testing/selftests/mm/uffd-unit-tests.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/tools/testing/selftests/mm/uffd-unit-tests.c b/tools/testing/selftests/mm/uffd-unit-tests.c
index 3db2296ac631..c8a3b1c7edff 100644
--- a/tools/testing/selftests/mm/uffd-unit-tests.c
+++ b/tools/testing/selftests/mm/uffd-unit-tests.c
@@ -241,6 +241,9 @@ static void *fork_event_consumer(void *data)
fork_event_args *args = data;
struct uffd_msg msg = { 0 };
+ /* Ready for parent thread to fork */
+ pthread_barrier_wait(&ready_for_fork);
+
/* Read until a full msg received */
while (uffd_read_msg(args->parent_uffd, &msg));
@@ -308,8 +311,12 @@ static int pagemap_test_fork(int uffd, bool with_event, bool test_pin)
/* Prepare a thread to resolve EVENT_FORK */
if (with_event) {
+ pthread_barrier_init(&ready_for_fork, NULL, 2);
if (pthread_create(&thread, NULL, fork_event_consumer, &args))
err("pthread_create()");
+ /* Wait for child thread to start before forking */
+ pthread_barrier_wait(&ready_for_fork);
+ pthread_barrier_destroy(&ready_for_fork);
}
child = fork();
--
2.46.1.824.gd892dcdcdd-goog
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 0/2] selftests/mm: fix deadlock after pthread_create
2024-10-03 21:17 [PATCH 0/2] selftests/mm: fix deadlock after pthread_create Edward Liaw
2024-10-03 21:17 ` [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t Edward Liaw
2024-10-03 21:17 ` [PATCH 2/2] selftests/mm: fix deadlock for fork after pthread_create on ARM Edward Liaw
@ 2024-10-04 23:54 ` Andrew Morton
2 siblings, 0 replies; 7+ messages in thread
From: Andrew Morton @ 2024-10-04 23:54 UTC (permalink / raw)
To: Edward Liaw
Cc: linux-kselftest, Shuah Khan, Lokesh Gidra, Peter Xu,
linux-kernel, kernel-team, linux-mm
On Thu, 3 Oct 2024 21:17:09 +0000 Edward Liaw <edliaw@google.com> wrote:
> On Android arm, pthread_create followed by a fork caused a deadlock in
> the case where the fork required work to be completed by the created
> thread.
>
> Updated the synchronization primitive to use pthread_barrier instead of
> atomic_bool.
>
> Applied the same fix to the wp-fork-with-event test.
>
> Edward Liaw (2):
> selftests/mm: replace atomic_bool with pthread_barrier_t
> selftests/mm: fix deadlock for fork after pthread_create on ARM
These fixes have different Fixes: targets, which might cause
backporting issues - some kernels might end up with one patch and not
the other. Was this intended? Is it OK?
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t
2024-10-03 21:17 ` [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t Edward Liaw
@ 2024-10-06 17:53 ` Muhammad Usama Anjum
2024-10-18 14:37 ` Ryan Roberts
1 sibling, 0 replies; 7+ messages in thread
From: Muhammad Usama Anjum @ 2024-10-06 17:53 UTC (permalink / raw)
To: Edward Liaw, linux-kselftest, Andrew Morton, Shuah Khan,
Lokesh Gidra, Peter Xu
Cc: Usama.Anjum, linux-kernel, kernel-team, linux-mm
On 10/4/24 2:17 AM, Edward Liaw wrote:
> Swaps synchronization primitive with pthread_barrier, so that
> stdatomic.h does not need to be included.
>
> The synchronization is needed on Android ARM64; we see a deadlock with
> pthread_create when the parent thread races forward before the child has
> a chance to start doing work.
>
> Fixes: 8c864371b2a1 ("selftests/mm: fix ARM related issue with fork after
> pthread_create")
> Signed-off-by: Edward Liaw <edliaw@google.com>
Reviewed-by: Muhammad Usama Anjum <usama.anjum@collabora.com>
> ---
> tools/testing/selftests/mm/uffd-common.c | 5 +++--
> tools/testing/selftests/mm/uffd-common.h | 3 +--
> tools/testing/selftests/mm/uffd-unit-tests.c | 14 ++++++++------
> 3 files changed, 12 insertions(+), 10 deletions(-)
>
> diff --git a/tools/testing/selftests/mm/uffd-common.c b/tools/testing/selftests/mm/uffd-common.c
> index 717539eddf98..852e7281026e 100644
> --- a/tools/testing/selftests/mm/uffd-common.c
> +++ b/tools/testing/selftests/mm/uffd-common.c
> @@ -18,7 +18,7 @@ bool test_uffdio_wp = true;
> unsigned long long *count_verify;
> uffd_test_ops_t *uffd_test_ops;
> uffd_test_case_ops_t *uffd_test_case_ops;
> -atomic_bool ready_for_fork;
> +pthread_barrier_t ready_for_fork;
>
> static int uffd_mem_fd_create(off_t mem_size, bool hugetlb)
> {
> @@ -519,7 +519,8 @@ void *uffd_poll_thread(void *arg)
> pollfd[1].fd = pipefd[cpu*2];
> pollfd[1].events = POLLIN;
>
> - ready_for_fork = true;
> + /* Ready for parent thread to fork */
> + pthread_barrier_wait(&ready_for_fork);
>
> for (;;) {
> ret = poll(pollfd, 2, -1);
> diff --git a/tools/testing/selftests/mm/uffd-common.h b/tools/testing/selftests/mm/uffd-common.h
> index a70ae10b5f62..3e6228d8e0dc 100644
> --- a/tools/testing/selftests/mm/uffd-common.h
> +++ b/tools/testing/selftests/mm/uffd-common.h
> @@ -33,7 +33,6 @@
> #include <inttypes.h>
> #include <stdint.h>
> #include <sys/random.h>
> -#include <stdatomic.h>
>
> #include "../kselftest.h"
> #include "vm_util.h"
> @@ -105,7 +104,7 @@ extern bool map_shared;
> extern bool test_uffdio_wp;
> extern unsigned long long *count_verify;
> extern volatile bool test_uffdio_copy_eexist;
> -extern atomic_bool ready_for_fork;
> +extern pthread_barrier_t ready_for_fork;
>
> extern uffd_test_ops_t anon_uffd_test_ops;
> extern uffd_test_ops_t shmem_uffd_test_ops;
> diff --git a/tools/testing/selftests/mm/uffd-unit-tests.c b/tools/testing/selftests/mm/uffd-unit-tests.c
> index b3d21eed203d..3db2296ac631 100644
> --- a/tools/testing/selftests/mm/uffd-unit-tests.c
> +++ b/tools/testing/selftests/mm/uffd-unit-tests.c
> @@ -774,7 +774,7 @@ static void uffd_sigbus_test_common(bool wp)
> char c;
> struct uffd_args args = { 0 };
>
> - ready_for_fork = false;
> + pthread_barrier_init(&ready_for_fork, NULL, 2);
>
> fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
>
> @@ -791,8 +791,9 @@ static void uffd_sigbus_test_common(bool wp)
> if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
> err("uffd_poll_thread create");
>
> - while (!ready_for_fork)
> - ; /* Wait for the poll_thread to start executing before forking */
> + /* Wait for child thread to start before forking */
> + pthread_barrier_wait(&ready_for_fork);
> + pthread_barrier_destroy(&ready_for_fork);
>
> pid = fork();
> if (pid < 0)
> @@ -833,7 +834,7 @@ static void uffd_events_test_common(bool wp)
> char c;
> struct uffd_args args = { 0 };
>
> - ready_for_fork = false;
> + pthread_barrier_init(&ready_for_fork, NULL, 2);
>
> fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
> if (uffd_register(uffd, area_dst, nr_pages * page_size,
> @@ -844,8 +845,9 @@ static void uffd_events_test_common(bool wp)
> if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
> err("uffd_poll_thread create");
>
> - while (!ready_for_fork)
> - ; /* Wait for the poll_thread to start executing before forking */
> + /* Wait for child thread to start before forking */
> + pthread_barrier_wait(&ready_for_fork);
> + pthread_barrier_destroy(&ready_for_fork);
>
> pid = fork();
> if (pid < 0)
--
BR,
Muhammad Usama Anjum
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t
2024-10-03 21:17 ` [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t Edward Liaw
2024-10-06 17:53 ` Muhammad Usama Anjum
@ 2024-10-18 14:37 ` Ryan Roberts
2024-10-18 17:17 ` Edward Liaw
1 sibling, 1 reply; 7+ messages in thread
From: Ryan Roberts @ 2024-10-18 14:37 UTC (permalink / raw)
To: Edward Liaw, linux-kselftest, Andrew Morton, Shuah Khan,
Lokesh Gidra, Peter Xu
Cc: linux-kernel, kernel-team, linux-mm, Aishwarya TCV
On 03/10/2024 22:17, Edward Liaw wrote:
> Swaps synchronization primitive with pthread_barrier, so that
> stdatomic.h does not need to be included.
>
> The synchronization is needed on Android ARM64; we see a deadlock with
> pthread_create when the parent thread races forward before the child has
> a chance to start doing work.
>
> Fixes: 8c864371b2a1 ("selftests/mm: fix ARM related issue with fork after
> pthread_create")
> Signed-off-by: Edward Liaw <edliaw@google.com>
> ---
> tools/testing/selftests/mm/uffd-common.c | 5 +++--
> tools/testing/selftests/mm/uffd-common.h | 3 +--
> tools/testing/selftests/mm/uffd-unit-tests.c | 14 ++++++++------
> 3 files changed, 12 insertions(+), 10 deletions(-)
>
> diff --git a/tools/testing/selftests/mm/uffd-common.c b/tools/testing/selftests/mm/uffd-common.c
> index 717539eddf98..852e7281026e 100644
> --- a/tools/testing/selftests/mm/uffd-common.c
> +++ b/tools/testing/selftests/mm/uffd-common.c
> @@ -18,7 +18,7 @@ bool test_uffdio_wp = true;
> unsigned long long *count_verify;
> uffd_test_ops_t *uffd_test_ops;
> uffd_test_case_ops_t *uffd_test_case_ops;
> -atomic_bool ready_for_fork;
> +pthread_barrier_t ready_for_fork;
>
> static int uffd_mem_fd_create(off_t mem_size, bool hugetlb)
> {
> @@ -519,7 +519,8 @@ void *uffd_poll_thread(void *arg)
> pollfd[1].fd = pipefd[cpu*2];
> pollfd[1].events = POLLIN;
>
> - ready_for_fork = true;
> + /* Ready for parent thread to fork */
> + pthread_barrier_wait(&ready_for_fork);
Hi Edward,
This change is causing uffd-unit-tests to hang on arm64 when running the "move on anon" test. It's happening because this wait is never returning for this case. And that happens because ready_for_fork was never initialized for this test. It looks like there are other places where a thread is created for uffd_poll_thread() where ready_for_fork is not initialized too.
I added this change and it solves the problem, although it's pretty hacky.
This is blocking our arm64 testing on linux-next so would appreciate either a quick fix or removing the change until a fix is ready.
Thanks,
Ryan
----8<-----
diff --git a/tools/testing/selftests/mm/uffd-common.c b/tools/testing/selftests/mm/uffd-common.c
index 852e7281026e..a05eb705be02 100644
--- a/tools/testing/selftests/mm/uffd-common.c
+++ b/tools/testing/selftests/mm/uffd-common.c
@@ -19,6 +19,7 @@ unsigned long long *count_verify;
uffd_test_ops_t *uffd_test_ops;
uffd_test_case_ops_t *uffd_test_case_ops;
pthread_barrier_t ready_for_fork;
+bool wait_ready_for_fork;
static int uffd_mem_fd_create(off_t mem_size, bool hugetlb)
{
@@ -520,7 +521,8 @@ void *uffd_poll_thread(void *arg)
pollfd[1].events = POLLIN;
/* Ready for parent thread to fork */
- pthread_barrier_wait(&ready_for_fork);
+ if (wait_ready_for_fork)
+ pthread_barrier_wait(&ready_for_fork);
for (;;) {
ret = poll(pollfd, 2, -1);
diff --git a/tools/testing/selftests/mm/uffd-common.h b/tools/testing/selftests/mm/uffd-common.h
index 3e6228d8e0dc..e94329a39b34 100644
--- a/tools/testing/selftests/mm/uffd-common.h
+++ b/tools/testing/selftests/mm/uffd-common.h
@@ -105,6 +105,7 @@ extern bool test_uffdio_wp;
extern unsigned long long *count_verify;
extern volatile bool test_uffdio_copy_eexist;
extern pthread_barrier_t ready_for_fork;
+extern bool wait_ready_for_fork;
extern uffd_test_ops_t anon_uffd_test_ops;
extern uffd_test_ops_t shmem_uffd_test_ops;
diff --git a/tools/testing/selftests/mm/uffd-unit-tests.c b/tools/testing/selftests/mm/uffd-unit-tests.c
index 3db2296ac631..d1fc4cd6a948 100644
--- a/tools/testing/selftests/mm/uffd-unit-tests.c
+++ b/tools/testing/selftests/mm/uffd-unit-tests.c
@@ -775,6 +775,7 @@ static void uffd_sigbus_test_common(bool wp)
struct uffd_args args = { 0 };
pthread_barrier_init(&ready_for_fork, NULL, 2);
+ wait_ready_for_fork = true;
fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
@@ -794,6 +795,7 @@ static void uffd_sigbus_test_common(bool wp)
/* Wait for child thread to start before forking */
pthread_barrier_wait(&ready_for_fork);
pthread_barrier_destroy(&ready_for_fork);
+ wait_ready_for_fork = false;
pid = fork();
if (pid < 0)
@@ -835,6 +837,7 @@ static void uffd_events_test_common(bool wp)
struct uffd_args args = { 0 };
pthread_barrier_init(&ready_for_fork, NULL, 2);
+ wait_ready_for_fork = true;
fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
if (uffd_register(uffd, area_dst, nr_pages * page_size,
@@ -848,6 +851,7 @@ static void uffd_events_test_common(bool wp)
/* Wait for child thread to start before forking */
pthread_barrier_wait(&ready_for_fork);
pthread_barrier_destroy(&ready_for_fork);
+ wait_ready_for_fork = false;
pid = fork();
if (pid < 0)
----8<-----
>
> for (;;) {
> ret = poll(pollfd, 2, -1);
> diff --git a/tools/testing/selftests/mm/uffd-common.h b/tools/testing/selftests/mm/uffd-common.h
> index a70ae10b5f62..3e6228d8e0dc 100644
> --- a/tools/testing/selftests/mm/uffd-common.h
> +++ b/tools/testing/selftests/mm/uffd-common.h
> @@ -33,7 +33,6 @@
> #include <inttypes.h>
> #include <stdint.h>
> #include <sys/random.h>
> -#include <stdatomic.h>
>
> #include "../kselftest.h"
> #include "vm_util.h"
> @@ -105,7 +104,7 @@ extern bool map_shared;
> extern bool test_uffdio_wp;
> extern unsigned long long *count_verify;
> extern volatile bool test_uffdio_copy_eexist;
> -extern atomic_bool ready_for_fork;
> +extern pthread_barrier_t ready_for_fork;
>
> extern uffd_test_ops_t anon_uffd_test_ops;
> extern uffd_test_ops_t shmem_uffd_test_ops;
> diff --git a/tools/testing/selftests/mm/uffd-unit-tests.c b/tools/testing/selftests/mm/uffd-unit-tests.c
> index b3d21eed203d..3db2296ac631 100644
> --- a/tools/testing/selftests/mm/uffd-unit-tests.c
> +++ b/tools/testing/selftests/mm/uffd-unit-tests.c
> @@ -774,7 +774,7 @@ static void uffd_sigbus_test_common(bool wp)
> char c;
> struct uffd_args args = { 0 };
>
> - ready_for_fork = false;
> + pthread_barrier_init(&ready_for_fork, NULL, 2);
>
> fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
>
> @@ -791,8 +791,9 @@ static void uffd_sigbus_test_common(bool wp)
> if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
> err("uffd_poll_thread create");
>
> - while (!ready_for_fork)
> - ; /* Wait for the poll_thread to start executing before forking */
> + /* Wait for child thread to start before forking */
> + pthread_barrier_wait(&ready_for_fork);
> + pthread_barrier_destroy(&ready_for_fork);
>
> pid = fork();
> if (pid < 0)
> @@ -833,7 +834,7 @@ static void uffd_events_test_common(bool wp)
> char c;
> struct uffd_args args = { 0 };
>
> - ready_for_fork = false;
> + pthread_barrier_init(&ready_for_fork, NULL, 2);
>
> fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
> if (uffd_register(uffd, area_dst, nr_pages * page_size,
> @@ -844,8 +845,9 @@ static void uffd_events_test_common(bool wp)
> if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
> err("uffd_poll_thread create");
>
> - while (!ready_for_fork)
> - ; /* Wait for the poll_thread to start executing before forking */
> + /* Wait for child thread to start before forking */
> + pthread_barrier_wait(&ready_for_fork);
> + pthread_barrier_destroy(&ready_for_fork);
>
> pid = fork();
> if (pid < 0)
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t
2024-10-18 14:37 ` Ryan Roberts
@ 2024-10-18 17:17 ` Edward Liaw
0 siblings, 0 replies; 7+ messages in thread
From: Edward Liaw @ 2024-10-18 17:17 UTC (permalink / raw)
To: Ryan Roberts
Cc: linux-kselftest, Andrew Morton, Shuah Khan, Lokesh Gidra,
Peter Xu, linux-kernel, kernel-team, linux-mm, Aishwarya TCV
On Fri, Oct 18, 2024 at 7:37 AM Ryan Roberts <ryan.roberts@arm.com> wrote:
>
> On 03/10/2024 22:17, Edward Liaw wrote:
> > Swaps synchronization primitive with pthread_barrier, so that
> > stdatomic.h does not need to be included.
> >
> > The synchronization is needed on Android ARM64; we see a deadlock with
> > pthread_create when the parent thread races forward before the child has
> > a chance to start doing work.
> >
> > Fixes: 8c864371b2a1 ("selftests/mm: fix ARM related issue with fork after
> > pthread_create")
> > Signed-off-by: Edward Liaw <edliaw@google.com>
> > ---
> > tools/testing/selftests/mm/uffd-common.c | 5 +++--
> > tools/testing/selftests/mm/uffd-common.h | 3 +--
> > tools/testing/selftests/mm/uffd-unit-tests.c | 14 ++++++++------
> > 3 files changed, 12 insertions(+), 10 deletions(-)
> >
> > diff --git a/tools/testing/selftests/mm/uffd-common.c b/tools/testing/selftests/mm/uffd-common.c
> > index 717539eddf98..852e7281026e 100644
> > --- a/tools/testing/selftests/mm/uffd-common.c
> > +++ b/tools/testing/selftests/mm/uffd-common.c
> > @@ -18,7 +18,7 @@ bool test_uffdio_wp = true;
> > unsigned long long *count_verify;
> > uffd_test_ops_t *uffd_test_ops;
> > uffd_test_case_ops_t *uffd_test_case_ops;
> > -atomic_bool ready_for_fork;
> > +pthread_barrier_t ready_for_fork;
> >
> > static int uffd_mem_fd_create(off_t mem_size, bool hugetlb)
> > {
> > @@ -519,7 +519,8 @@ void *uffd_poll_thread(void *arg)
> > pollfd[1].fd = pipefd[cpu*2];
> > pollfd[1].events = POLLIN;
> >
> > - ready_for_fork = true;
> > + /* Ready for parent thread to fork */
> > + pthread_barrier_wait(&ready_for_fork);
>
> Hi Edward,
>
> This change is causing uffd-unit-tests to hang on arm64 when running the "move on anon" test. It's happening because this wait is never returning for this case. And that happens because ready_for_fork was never initialized for this test. It looks like there are other places where a thread is created for uffd_poll_thread() where ready_for_fork is not initialized too.
>
> I added this change and it solves the problem, although it's pretty hacky.
>
> This is blocking our arm64 testing on linux-next so would appreciate either a quick fix or removing the change until a fix is ready.
Ah, I should not have changed it to a pthread_barrier, since that
depends on the parent creating the barrier; I'm sending a revert and
subsequent fix.
>
> Thanks,
> Ryan
>
> ----8<-----
> diff --git a/tools/testing/selftests/mm/uffd-common.c b/tools/testing/selftests/mm/uffd-common.c
> index 852e7281026e..a05eb705be02 100644
> --- a/tools/testing/selftests/mm/uffd-common.c
> +++ b/tools/testing/selftests/mm/uffd-common.c
> @@ -19,6 +19,7 @@ unsigned long long *count_verify;
> uffd_test_ops_t *uffd_test_ops;
> uffd_test_case_ops_t *uffd_test_case_ops;
> pthread_barrier_t ready_for_fork;
> +bool wait_ready_for_fork;
>
> static int uffd_mem_fd_create(off_t mem_size, bool hugetlb)
> {
> @@ -520,7 +521,8 @@ void *uffd_poll_thread(void *arg)
> pollfd[1].events = POLLIN;
>
> /* Ready for parent thread to fork */
> - pthread_barrier_wait(&ready_for_fork);
> + if (wait_ready_for_fork)
> + pthread_barrier_wait(&ready_for_fork);
>
> for (;;) {
> ret = poll(pollfd, 2, -1);
> diff --git a/tools/testing/selftests/mm/uffd-common.h b/tools/testing/selftests/mm/uffd-common.h
> index 3e6228d8e0dc..e94329a39b34 100644
> --- a/tools/testing/selftests/mm/uffd-common.h
> +++ b/tools/testing/selftests/mm/uffd-common.h
> @@ -105,6 +105,7 @@ extern bool test_uffdio_wp;
> extern unsigned long long *count_verify;
> extern volatile bool test_uffdio_copy_eexist;
> extern pthread_barrier_t ready_for_fork;
> +extern bool wait_ready_for_fork;
>
> extern uffd_test_ops_t anon_uffd_test_ops;
> extern uffd_test_ops_t shmem_uffd_test_ops;
> diff --git a/tools/testing/selftests/mm/uffd-unit-tests.c b/tools/testing/selftests/mm/uffd-unit-tests.c
> index 3db2296ac631..d1fc4cd6a948 100644
> --- a/tools/testing/selftests/mm/uffd-unit-tests.c
> +++ b/tools/testing/selftests/mm/uffd-unit-tests.c
> @@ -775,6 +775,7 @@ static void uffd_sigbus_test_common(bool wp)
> struct uffd_args args = { 0 };
>
> pthread_barrier_init(&ready_for_fork, NULL, 2);
> + wait_ready_for_fork = true;
>
> fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
>
> @@ -794,6 +795,7 @@ static void uffd_sigbus_test_common(bool wp)
> /* Wait for child thread to start before forking */
> pthread_barrier_wait(&ready_for_fork);
> pthread_barrier_destroy(&ready_for_fork);
> + wait_ready_for_fork = false;
>
> pid = fork();
> if (pid < 0)
> @@ -835,6 +837,7 @@ static void uffd_events_test_common(bool wp)
> struct uffd_args args = { 0 };
>
> pthread_barrier_init(&ready_for_fork, NULL, 2);
> + wait_ready_for_fork = true;
>
> fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
> if (uffd_register(uffd, area_dst, nr_pages * page_size,
> @@ -848,6 +851,7 @@ static void uffd_events_test_common(bool wp)
> /* Wait for child thread to start before forking */
> pthread_barrier_wait(&ready_for_fork);
> pthread_barrier_destroy(&ready_for_fork);
> + wait_ready_for_fork = false;
>
> pid = fork();
> if (pid < 0)
> ----8<-----
>
>
> >
> > for (;;) {
> > ret = poll(pollfd, 2, -1);
> > diff --git a/tools/testing/selftests/mm/uffd-common.h b/tools/testing/selftests/mm/uffd-common.h
> > index a70ae10b5f62..3e6228d8e0dc 100644
> > --- a/tools/testing/selftests/mm/uffd-common.h
> > +++ b/tools/testing/selftests/mm/uffd-common.h
> > @@ -33,7 +33,6 @@
> > #include <inttypes.h>
> > #include <stdint.h>
> > #include <sys/random.h>
> > -#include <stdatomic.h>
> >
> > #include "../kselftest.h"
> > #include "vm_util.h"
> > @@ -105,7 +104,7 @@ extern bool map_shared;
> > extern bool test_uffdio_wp;
> > extern unsigned long long *count_verify;
> > extern volatile bool test_uffdio_copy_eexist;
> > -extern atomic_bool ready_for_fork;
> > +extern pthread_barrier_t ready_for_fork;
> >
> > extern uffd_test_ops_t anon_uffd_test_ops;
> > extern uffd_test_ops_t shmem_uffd_test_ops;
> > diff --git a/tools/testing/selftests/mm/uffd-unit-tests.c b/tools/testing/selftests/mm/uffd-unit-tests.c
> > index b3d21eed203d..3db2296ac631 100644
> > --- a/tools/testing/selftests/mm/uffd-unit-tests.c
> > +++ b/tools/testing/selftests/mm/uffd-unit-tests.c
> > @@ -774,7 +774,7 @@ static void uffd_sigbus_test_common(bool wp)
> > char c;
> > struct uffd_args args = { 0 };
> >
> > - ready_for_fork = false;
> > + pthread_barrier_init(&ready_for_fork, NULL, 2);
> >
> > fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
> >
> > @@ -791,8 +791,9 @@ static void uffd_sigbus_test_common(bool wp)
> > if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
> > err("uffd_poll_thread create");
> >
> > - while (!ready_for_fork)
> > - ; /* Wait for the poll_thread to start executing before forking */
> > + /* Wait for child thread to start before forking */
> > + pthread_barrier_wait(&ready_for_fork);
> > + pthread_barrier_destroy(&ready_for_fork);
> >
> > pid = fork();
> > if (pid < 0)
> > @@ -833,7 +834,7 @@ static void uffd_events_test_common(bool wp)
> > char c;
> > struct uffd_args args = { 0 };
> >
> > - ready_for_fork = false;
> > + pthread_barrier_init(&ready_for_fork, NULL, 2);
> >
> > fcntl(uffd, F_SETFL, uffd_flags | O_NONBLOCK);
> > if (uffd_register(uffd, area_dst, nr_pages * page_size,
> > @@ -844,8 +845,9 @@ static void uffd_events_test_common(bool wp)
> > if (pthread_create(&uffd_mon, NULL, uffd_poll_thread, &args))
> > err("uffd_poll_thread create");
> >
> > - while (!ready_for_fork)
> > - ; /* Wait for the poll_thread to start executing before forking */
> > + /* Wait for child thread to start before forking */
> > + pthread_barrier_wait(&ready_for_fork);
> > + pthread_barrier_destroy(&ready_for_fork);
> >
> > pid = fork();
> > if (pid < 0)
>
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2024-10-18 17:18 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-10-03 21:17 [PATCH 0/2] selftests/mm: fix deadlock after pthread_create Edward Liaw
2024-10-03 21:17 ` [PATCH 1/2] selftests/mm: replace atomic_bool with pthread_barrier_t Edward Liaw
2024-10-06 17:53 ` Muhammad Usama Anjum
2024-10-18 14:37 ` Ryan Roberts
2024-10-18 17:17 ` Edward Liaw
2024-10-03 21:17 ` [PATCH 2/2] selftests/mm: fix deadlock for fork after pthread_create on ARM Edward Liaw
2024-10-04 23:54 ` [PATCH 0/2] selftests/mm: fix deadlock after pthread_create Andrew Morton
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox