[PATCH v3] lib/eal: fix segfaults due to thread exit order

[PATCH v3] lib/eal: fix segfaults due to thread exit order

[zhichaox.zeng]

From: Zhichao Zeng <zhichaox.zeng@intel.com>

The eal-intr-thread is not closed before memory cleanup in the
process of exiting. There is a small probability that when the
eal-intr-thread is about to use some pointers, the memory were
just cleaned, which cause the segment fault error caught by ASan.

This patch close the eal-intr-thread before memory cleanup when
exiting to avoid segment fault.

---
v2: add the same API for FreeBSD
---
v3: fix rte_eal_cleanup crash in debug_autotest

Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>
---
 lib/eal/common/eal_private.h     |  7 +++++++
 lib/eal/freebsd/eal.c            | 22 +++++++++++++++++++++-
 lib/eal/freebsd/eal_interrupts.c | 12 ++++++++++++
 lib/eal/linux/eal.c              | 21 ++++++++++++++++++++-
 lib/eal/linux/eal_interrupts.c   | 12 ++++++++++++
 5 files changed, 72 insertions(+), 2 deletions(-)

diff --git a/lib/eal/common/eal_private.h b/lib/eal/common/eal_private.h
index 44d14241f0..7adf41b7d7 100644
--- a/lib/eal/common/eal_private.h
+++ b/lib/eal/common/eal_private.h
@@ -152,6 +152,13 @@ int rte_eal_tailqs_init(void);
  */
 int rte_eal_intr_init(void);
 
+/**
+ * Destroy interrupt handling thread.
+ *
+ * This function is private to EAL.
+ */
+void rte_eal_intr_destroy(void);
+
 /**
  * Close the default log stream
  *
diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
index a6b20960f2..6875a9f6d2 100644
--- a/lib/eal/freebsd/eal.c
+++ b/lib/eal/freebsd/eal.c
@@ -72,6 +72,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge the running status of the eal */
+static uint32_t run_once;
 
 int
 eal_clean_runtime_dir(void)
@@ -574,12 +576,23 @@ static void rte_eal_init_alert(const char *msg)
 	RTE_LOG(ERR, EAL, "%s\n", msg);
 }
 
+static void warn_parent(void)
+{
+	RTE_LOG(WARNING, EAL, "fork() was called, DPDK won't work in the child "
+		"process unless it calls rte_eal_init()\n");
+}
+
+static void scratch_child(void)
+{
+	/* Scratch run_once so that a call to rte_eal_cleanup won't crash... */
+	__atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
 {
 	int i, fctret, ret;
-	static uint32_t run_once;
 	uint32_t has_run = 0;
 	char cpuset[RTE_CPU_AFFINITY_STR_LEN];
 	char thread_name[RTE_MAX_THREAD_NAME_LEN];
@@ -883,6 +896,8 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, warn_parent, scratch_child);
+
 	return fctret;
 }
 
@@ -891,8 +906,13 @@ rte_eal_cleanup(void)
 {
 	struct internal_config *internal_conf =
 		eal_get_internal_configuration();
+
+	if (__atomic_load_n(&run_once, __ATOMIC_RELAXED) == 0)
+		return 0;
+
 	rte_service_finalize();
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	/* after this point, any DPDK pointers will become dangling */
 	rte_eal_memory_detach();
 	rte_eal_alarm_cleanup();
diff --git a/lib/eal/freebsd/eal_interrupts.c b/lib/eal/freebsd/eal_interrupts.c
index 9f720bdc8f..cac3859b06 100644
--- a/lib/eal/freebsd/eal_interrupts.c
+++ b/lib/eal/freebsd/eal_interrupts.c
@@ -648,6 +648,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close kqueue */
+	close(kq);
+	kq = -1;
+}
+
 int
 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle,
 		int epfd, int op, unsigned int vec, void *data)
diff --git a/lib/eal/linux/eal.c b/lib/eal/linux/eal.c
index 1ef263434a..32c7adaa52 100644
--- a/lib/eal/linux/eal.c
+++ b/lib/eal/linux/eal.c
@@ -76,6 +76,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge the running status of the eal */
+static uint32_t run_once;
 
 int
 eal_clean_runtime_dir(void)
@@ -857,12 +859,23 @@ is_iommu_enabled(void)
 	return n > 2;
 }
 
+static void warn_parent(void)
+{
+	RTE_LOG(WARNING, EAL, "fork() was called, DPDK won't work in the child "
+		"process unless it calls rte_eal_init()\n");
+}
+
+static void scratch_child(void)
+{
+	/* Scratch run_once so that a call to rte_eal_cleanup won't crash... */
+	__atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
 {
 	int i, fctret, ret;
-	static uint32_t run_once;
 	uint32_t has_run = 0;
 	const char *p;
 	static char logid[PATH_MAX];
@@ -1228,6 +1241,8 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, warn_parent, scratch_child);
+
 	return fctret;
 }
 
@@ -1257,6 +1272,9 @@ rte_eal_cleanup(void)
 	struct internal_config *internal_conf =
 		eal_get_internal_configuration();
 
+	if (__atomic_load_n(&run_once, __ATOMIC_RELAXED) == 0)
+		return 0;
+
 	if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
 			internal_conf->hugepage_file.unlink_existing)
 		rte_memseg_walk(mark_freeable, NULL);
@@ -1266,6 +1284,7 @@ rte_eal_cleanup(void)
 	vfio_mp_sync_cleanup();
 #endif
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	/* after this point, any DPDK pointers will become dangling */
 	rte_eal_memory_detach();
 	eal_mp_dev_hotplug_cleanup();
diff --git a/lib/eal/linux/eal_interrupts.c b/lib/eal/linux/eal_interrupts.c
index d52ec8eb4c..7e9853e8e7 100644
--- a/lib/eal/linux/eal_interrupts.c
+++ b/lib/eal/linux/eal_interrupts.c
@@ -1199,6 +1199,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close the pipe used by epoll */
+	close(intr_pipe.writefd);
+	close(intr_pipe.readfd);
+}
+
 static void
 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
 {
-- 
2.25.1

Re: [PATCH v3] lib/eal: fix segfaults due to thread exit order

[Stephen Hemminger]

On Mon, 30 May 2022 13:47:38 +0000
zhichaox.zeng@intel.com wrote:

> +static void warn_parent(void)
> +{
> +	RTE_LOG(WARNING, EAL, "fork() was called, DPDK won't work in the child "
> +		"process unless it calls rte_eal_init()\n");
> +}
> +

Too wordy. Don't break messages across lines.

Re: [PATCH v3] lib/eal: fix segfaults due to thread exit order

[Stephen Hemminger]

On Mon, 30 May 2022 13:47:38 +0000
zhichaox.zeng@intel.com wrote:

> @@ -883,6 +896,8 @@ rte_eal_init(int argc, char **argv)
>  
>  	eal_mcfg_complete();
>  
> +	pthread_atfork(NULL, warn_parent, scratch_child);
> +
>  	return fctret;
>  }

There are lots of other cases where DPDK will die if you fork()
in a DPDK process then call DPDK functions in child. 

Not sure what the problem you are trying to solve is?

RE: [PATCH v3] lib/eal: fix segfaults due to thread exit order

[Zeng, ZhichaoX]

Hi Stephen:

> -----Original Message-----
> From: Stephen Hemminger <stephen@networkplumber.org> 
> Sent: Tuesday, May 31, 2022 12:29 AM
> To: Zeng, ZhichaoX <zhichaox.zeng@intel.com>
> Cc: dev@dpdk.org; Yang, Qiming <qiming.yang@intel.com>; Richardson, Bruce <bruce.richardson@intel.com>; hkalra@marvell.com; david.marchand@redhat.com; aconole@redhat.com; Matz, Olivier > > <olivier.matz@6wind.com>; thomas@monjalon.net; stable@dpdk.org
> Subject: Re: [PATCH v3] lib/eal: fix segfaults due to thread exit order
>
> On Mon, 30 May 2022 13:47:38 +0000
> zhichaox.zeng@intel.com wrote:

> > @@ -883,6 +896,8 @@ rte_eal_init(int argc, char **argv)
> >  
> >  	eal_mcfg_complete();
> >  
> > +	pthread_atfork(NULL, warn_parent, scratch_child);
> > +
> >  	return fctret;
> >  }

> There are lots of other cases where DPDK will die if you fork() in a DPDK process then call DPDK functions in child. 

> Not sure what the problem you are trying to solve is?

The original goal of this patch was to cancel eal-intr-thread before memory cleanup to avoid a small probability of segfaults. 

But in the debug_autotest test of dpdk-test, fork() is called and the exit process is tested in the child process which will call the rte_eal_cleanup function and mistakenly clean up non-existing threads. This will cause segmentation fault. 

Based on patch v2, atomic operation is added to determine whether it is a child process, so that the cleaning process will not execute in the child process to avoid the above problem.

Regards

RE: [PATCH v3] lib/eal: fix segfaults due to thread exit order

[Zeng, ZhichaoX]

Hi David, Harman

Please review this patch at your convenience. Thanks!

In addition, I tried to figure out the reason for the failure of the CI test. There is a saying that
this is a problem with Asan when cancelling a thread that is waiting on epoll. After adding the 
'-fstack-protector-all' parameter, there will be no exception. But I don't know the software 
environment for automated testing, so I can't verify this statement.

Regards

> Subject: [PATCH v3] lib/eal: fix segfaults due to thread exit order

> From: Zhichao Zeng <zhichaox.zeng@intel.com>

> The eal-intr-thread is not closed before memory cleanup in the process of exiting. There is a small probability that when the eal-intr-thread is about to use some pointers, the memory were just cleaned, which cause the segment fault error caught by ASan.

> This patch close the eal-intr-thread before memory cleanup when exiting to avoid segment fault.

> ---
> v2: add the same API for FreeBSD
> ---
> v3: fix rte_eal_cleanup crash in debug_autotest

> Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>

[PATCH v4] lib/eal: fix segfaults due to thread exit order

[zhichaox.zeng]

From: Zhichao Zeng <zhichaox.zeng@intel.com>

The eal-intr-thread is not closed before memory cleanup in the
process of exiting. There is a small probability that when the
eal-intr-thread is about to use some pointers, the memory were
just cleaned, which cause the segment fault error caught by ASan.

This patch close the eal-intr-thread before memory cleanup when
exiting to avoid segment fault. And add some atomic operations
to avoid executing rte_eal_cleanup in the child process spawned
by fork() in some test cases, e.g. debug_autotest of dpdk-test.

Cc: stable@dpdk.org

---
v2:
add the same API for FreeBSD
---
v3:
fix rte_eal_cleanup crash in debug_autotest
---
v4:
shorten the prompt message and optimize the commit log

Suggested-by: David Marchand <david.marchand@redhat.com>
Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>
---
 lib/eal/common/eal_private.h     |  7 +++++++
 lib/eal/freebsd/eal.c            | 21 ++++++++++++++++++++-
 lib/eal/freebsd/eal_interrupts.c | 12 ++++++++++++
 lib/eal/linux/eal.c              | 20 +++++++++++++++++++-
 lib/eal/linux/eal_interrupts.c   | 12 ++++++++++++
 5 files changed, 70 insertions(+), 2 deletions(-)

diff --git a/lib/eal/common/eal_private.h b/lib/eal/common/eal_private.h
index 44d14241f0..7adf41b7d7 100644
--- a/lib/eal/common/eal_private.h
+++ b/lib/eal/common/eal_private.h
@@ -152,6 +152,13 @@ int rte_eal_tailqs_init(void);
  */
 int rte_eal_intr_init(void);
 
+/**
+ * Destroy interrupt handling thread.
+ *
+ * This function is private to EAL.
+ */
+void rte_eal_intr_destroy(void);
+
 /**
  * Close the default log stream
  *
diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
index a6b20960f2..4882f27abd 100644
--- a/lib/eal/freebsd/eal.c
+++ b/lib/eal/freebsd/eal.c
@@ -72,6 +72,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge the running status of the eal */
+static uint32_t run_once;
 
 int
 eal_clean_runtime_dir(void)
@@ -574,12 +576,22 @@ static void rte_eal_init_alert(const char *msg)
 	RTE_LOG(ERR, EAL, "%s\n", msg);
 }
 
+static void warn_parent(void)
+{
+	RTE_LOG(WARNING, EAL, "DPDK won't work in the child process\n");
+}
+
+static void scratch_child(void)
+{
+	/* Scratch run_once so that a call to rte_eal_cleanup won't crash... */
+	__atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
 {
 	int i, fctret, ret;
-	static uint32_t run_once;
 	uint32_t has_run = 0;
 	char cpuset[RTE_CPU_AFFINITY_STR_LEN];
 	char thread_name[RTE_MAX_THREAD_NAME_LEN];
@@ -883,6 +895,8 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, warn_parent, scratch_child);
+
 	return fctret;
 }
 
@@ -891,8 +905,13 @@ rte_eal_cleanup(void)
 {
 	struct internal_config *internal_conf =
 		eal_get_internal_configuration();
+
+	if (__atomic_load_n(&run_once, __ATOMIC_RELAXED) == 0)
+		return 0;
+
 	rte_service_finalize();
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	/* after this point, any DPDK pointers will become dangling */
 	rte_eal_memory_detach();
 	rte_eal_alarm_cleanup();
diff --git a/lib/eal/freebsd/eal_interrupts.c b/lib/eal/freebsd/eal_interrupts.c
index 9f720bdc8f..cac3859b06 100644
--- a/lib/eal/freebsd/eal_interrupts.c
+++ b/lib/eal/freebsd/eal_interrupts.c
@@ -648,6 +648,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close kqueue */
+	close(kq);
+	kq = -1;
+}
+
 int
 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle,
 		int epfd, int op, unsigned int vec, void *data)
diff --git a/lib/eal/linux/eal.c b/lib/eal/linux/eal.c
index 1ef263434a..effebb33a6 100644
--- a/lib/eal/linux/eal.c
+++ b/lib/eal/linux/eal.c
@@ -76,6 +76,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge the running status of the eal */
+static uint32_t run_once;
 
 int
 eal_clean_runtime_dir(void)
@@ -857,12 +859,22 @@ is_iommu_enabled(void)
 	return n > 2;
 }
 
+static void warn_parent(void)
+{
+	RTE_LOG(WARNING, EAL, "DPDK won't work in the child process\n");
+}
+
+static void scratch_child(void)
+{
+	/* Scratch run_once so that a call to rte_eal_cleanup won't crash... */
+	__atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
 {
 	int i, fctret, ret;
-	static uint32_t run_once;
 	uint32_t has_run = 0;
 	const char *p;
 	static char logid[PATH_MAX];
@@ -1228,6 +1240,8 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, warn_parent, scratch_child);
+
 	return fctret;
 }
 
@@ -1257,6 +1271,9 @@ rte_eal_cleanup(void)
 	struct internal_config *internal_conf =
 		eal_get_internal_configuration();
 
+	if (__atomic_load_n(&run_once, __ATOMIC_RELAXED) == 0)
+		return 0;
+
 	if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
 			internal_conf->hugepage_file.unlink_existing)
 		rte_memseg_walk(mark_freeable, NULL);
@@ -1266,6 +1283,7 @@ rte_eal_cleanup(void)
 	vfio_mp_sync_cleanup();
 #endif
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	/* after this point, any DPDK pointers will become dangling */
 	rte_eal_memory_detach();
 	eal_mp_dev_hotplug_cleanup();
diff --git a/lib/eal/linux/eal_interrupts.c b/lib/eal/linux/eal_interrupts.c
index d52ec8eb4c..7e9853e8e7 100644
--- a/lib/eal/linux/eal_interrupts.c
+++ b/lib/eal/linux/eal_interrupts.c
@@ -1199,6 +1199,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close the pipe used by epoll */
+	close(intr_pipe.writefd);
+	close(intr_pipe.readfd);
+}
+
 static void
 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
 {
-- 
2.25.1

RE: [PATCH v4] lib/eal: fix segfaults due to thread exit order

[Zeng, ZhichaoX]

Hi David, Harman
    Please review this patch at your convenience, it's been about a month since the v1 version.
    Thanks!

Best regards
Zhichao

> -----Original Message-----
> From: Zeng, ZhichaoX <zhichaox.zeng@intel.com> 
> Sent: Wednesday, June 15, 2022 2:02 PM
> To: dev@dpdk.org
> Cc: stable@dpdk.org; Yang, Qiming <qiming.yang@intel.com>; david.marchand@redhat.com; stephen@networkplumber.org; mb@smartsharesystems.com; Zeng, ZhichaoX <zhichaox.zeng@intel.com>; Richardson, Bruce > <bruce.richardson@intel.com>; Harman Kalra <hkalra@marvell.com>
> Subject: [PATCH v4] lib/eal: fix segfaults due to thread exit order
>
> From: Zhichao Zeng <zhichaox.zeng@intel.com>
>
> The eal-intr-thread is not closed before memory cleanup in the process of exiting. There is a small probability that when the eal-intr-thread is about to use some pointers, the memory were just cleaned, which cause the > segment fault error caught by ASan.
>
> This patch close the eal-intr-thread before memory cleanup when exiting to avoid segment fault. And add some atomic operations to avoid executing rte_eal_cleanup in the child process spawned by fork() in some test > cases, e.g. debug_autotest of dpdk-test.
>
> Cc: stable@dpdk.org
>
> ---
> v2:
> add the same API for FreeBSD
> ---
> v3:
> fix rte_eal_cleanup crash in debug_autotest
> ---
> v4:
> shorten the prompt message and optimize the commit log
>
> Suggested-by: David Marchand <david.marchand@redhat.com>
> Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>

Re: [PATCH v4] lib/eal: fix segfaults due to thread exit order

[David Marchand]

On Fri, Jun 24, 2022 at 3:42 AM Zeng, ZhichaoX <zhichaox.zeng@intel.com> wrote:
>
> Hi David, Harman
>     Please review this patch at your convenience, it's been about a month since the v1 version.
>     Thanks!

Yes, it's been a month.
Feel free to help on reviewing patches from others so that reviews can
be quicker for everyone.

Thanks.

-- 
David Marchand

RE: [PATCH v4] lib/eal: fix segfaults due to thread exit order

[Zeng, ZhichaoX]

Hi Harman, Bruce
    Could you please help to review this patch, thank you so much!

Best regards
Zhichao

> -----Original Message-----
> From: Zeng, ZhichaoX <zhichaox.zeng@intel.com> 
> Sent: Wednesday, June 15, 2022 2:02 PM
> To: dev@dpdk.org
> Cc: stable@dpdk.org; Yang, Qiming <qiming.yang@intel.com>; david.marchand@redhat.com; stephen@networkplumber.org; mb@smartsharesystems.com; Zeng, ZhichaoX <zhichaox.zeng@intel.com>; Richardson, Bruce > <bruce.richardson@intel.com>; Harman Kalra <hkalra@marvell.com>
> Subject: [PATCH v4] lib/eal: fix segfaults due to thread exit order
>
> From: Zhichao Zeng <zhichaox.zeng@intel.com>
>
> The eal-intr-thread is not closed before memory cleanup in the process of exiting. There is a small probability that when the eal-intr-thread is about to use some pointers, the memory were just cleaned, which cause the > segment fault error caught by ASan.
>
> This patch close the eal-intr-thread before memory cleanup when exiting to avoid segment fault. And add some atomic operations to avoid executing rte_eal_cleanup in the child process spawned by fork() in some test > cases, e.g. debug_autotest of dpdk-test.
>
> Cc: stable@dpdk.org
>
> ---
> v2:
> add the same API for FreeBSD
> ---
> v3:
> fix rte_eal_cleanup crash in debug_autotest
> ---
> v4:
> shorten the prompt message and optimize the commit log
>
> Suggested-by: David Marchand <david.marchand@redhat.com>
> Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>

Re: [PATCH v4] lib/eal: fix segfaults due to thread exit order

[Bruce Richardson]

On Wed, Jun 15, 2022 at 02:01:54PM +0800, zhichaox.zeng@intel.com wrote:
> From: Zhichao Zeng <zhichaox.zeng@intel.com>
> 
> The eal-intr-thread is not closed before memory cleanup in the
> process of exiting. There is a small probability that when the
> eal-intr-thread is about to use some pointers, the memory were
> just cleaned, which cause the segment fault error caught by ASan.
> 
> This patch close the eal-intr-thread before memory cleanup when
> exiting to avoid segment fault. And add some atomic operations
> to avoid executing rte_eal_cleanup in the child process spawned
> by fork() in some test cases, e.g. debug_autotest of dpdk-test.
> 
> Cc: stable@dpdk.org
> 

Hi,

some comments inline below.

/Bruce

> ---
> v2:
> add the same API for FreeBSD
> ---
> v3:
> fix rte_eal_cleanup crash in debug_autotest
> ---
> v4:
> shorten the prompt message and optimize the commit log
> 

Please put these updates below the cutline after the sign-offs, i.e.
immediately before the diffstat.

> Suggested-by: David Marchand <david.marchand@redhat.com>
> Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>
> ---
>  lib/eal/common/eal_private.h     |  7 +++++++
>  lib/eal/freebsd/eal.c            | 21 ++++++++++++++++++++-
>  lib/eal/freebsd/eal_interrupts.c | 12 ++++++++++++
>  lib/eal/linux/eal.c              | 20 +++++++++++++++++++-
>  lib/eal/linux/eal_interrupts.c   | 12 ++++++++++++
>  5 files changed, 70 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/eal/common/eal_private.h b/lib/eal/common/eal_private.h
> index 44d14241f0..7adf41b7d7 100644
> --- a/lib/eal/common/eal_private.h
> +++ b/lib/eal/common/eal_private.h
> @@ -152,6 +152,13 @@ int rte_eal_tailqs_init(void);
>   */
>  int rte_eal_intr_init(void);
>  
> +/**
> + * Destroy interrupt handling thread.
> + *
> + * This function is private to EAL.
> + */
> +void rte_eal_intr_destroy(void);
> +
>  /**
>   * Close the default log stream
>   *
> diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
> index a6b20960f2..4882f27abd 100644
> --- a/lib/eal/freebsd/eal.c
> +++ b/lib/eal/freebsd/eal.c
> @@ -72,6 +72,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
>  /* used by rte_rdtsc() */
>  int rte_cycles_vmware_tsc_map;
>  
> +/* used to judge the running status of the eal */
> +static uint32_t run_once;
>  

I don't like just moving this variable from the eal_init function. When in
eal_init the name "run_once" made sense as it tracked how often the EAL
init function was run. However, now as a global variable the name
"run_once" no longer makes sense.

Two suggestions:
1. Keep run_once in EAL init as-is, and use a different variable or value
   to indicate that DPDK is initialized for cleanup.
2. Move the variable as you have here, just rename it to a more meaningful
   name.


>  int
>  eal_clean_runtime_dir(void)
> @@ -574,12 +576,22 @@ static void rte_eal_init_alert(const char *msg)
>  	RTE_LOG(ERR, EAL, "%s\n", msg);
>  }
>  
> +static void warn_parent(void)
> +{
> +	RTE_LOG(WARNING, EAL, "DPDK won't work in the child process\n");
> +}

I wonder if this contains enough information. Can we identify briefly what
parts will or won't work, or if we just want to deny everything, can we
give a brief reason why?

> +
> +static void scratch_child(void)
> +{
> +	/* Scratch run_once so that a call to rte_eal_cleanup won't crash... */
> +	__atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
> +}
> +

I think the name of this function needs improvement. I'm not sure that
"scratch" is the best term to use. Something like "clear_eal_flag" is
probably better.

>  /* Launch threads, called at application init(). */
>  int
>  rte_eal_init(int argc, char **argv)
>  {
>  	int i, fctret, ret;
> -	static uint32_t run_once;
>  	uint32_t has_run = 0;
>  	char cpuset[RTE_CPU_AFFINITY_STR_LEN];
>  	char thread_name[RTE_MAX_THREAD_NAME_LEN];
> @@ -883,6 +895,8 @@ rte_eal_init(int argc, char **argv)
>  
>  	eal_mcfg_complete();
>  
> +	pthread_atfork(NULL, warn_parent, scratch_child);
> +
>  	return fctret;
>  }
>  
> @@ -891,8 +905,13 @@ rte_eal_cleanup(void)
>  {
>  	struct internal_config *internal_conf =
>  		eal_get_internal_configuration();
> +
> +	if (__atomic_load_n(&run_once, __ATOMIC_RELAXED) == 0)
> +		return 0;
> +
>  	rte_service_finalize();
>  	rte_mp_channel_cleanup();
> +	rte_eal_intr_destroy();
>  	/* after this point, any DPDK pointers will become dangling */
>  	rte_eal_memory_detach();
>  	rte_eal_alarm_cleanup();
<snip for brevity>

[PATCH v5] lib/eal: fix segfaults in exiting

[Zhichao Zeng]

The 'eal-intr-thread' is not closed before memory cleanup in the process of
exiting. There is a small chance that the 'eal-intr-thread' is about to use
some pointers, the memory was just cleaned, which causes segfaults
caught by ASan.

This patch closes the 'eal-intr-thread' before memory cleanup in
'rte_eal_cleanup' to avoid segfaults, and adds a flag to avoid executing
'rte_eal_cleanup' in the child process which is forked to execut some
test cases(e.g. debug_autotest of dpdk-test).

Bugzilla ID: 1006
Cc: stable@dpdk.org

Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>

---
v2: add same API for FreeBSD
---
v3: fix rte_eal_cleanup crash in debug_autotest
---
v4: shorten the prompt message and optimize the commit log
---
v5: simplify patch
---
 lib/eal/common/eal_private.h     |  7 +++++++
 lib/eal/freebsd/eal.c            | 13 +++++++++++++
 lib/eal/freebsd/eal_interrupts.c | 12 ++++++++++++
 lib/eal/linux/eal.c              | 13 +++++++++++++
 lib/eal/linux/eal_interrupts.c   | 12 ++++++++++++
 5 files changed, 57 insertions(+)

diff --git a/lib/eal/common/eal_private.h b/lib/eal/common/eal_private.h
index 44d14241f0..7adf41b7d7 100644
--- a/lib/eal/common/eal_private.h
+++ b/lib/eal/common/eal_private.h
@@ -152,6 +152,13 @@ int rte_eal_tailqs_init(void);
  */
 int rte_eal_intr_init(void);
 
+/**
+ * Destroy interrupt handling thread.
+ *
+ * This function is private to EAL.
+ */
+void rte_eal_intr_destroy(void);
+
 /**
  * Close the default log stream
  *
diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
index 26fbc91b26..f27c1d9f97 100644
--- a/lib/eal/freebsd/eal.c
+++ b/lib/eal/freebsd/eal.c
@@ -72,6 +72,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge if is forked */
+static int is_forked;
 
 int
 eal_clean_runtime_dir(void)
@@ -574,6 +576,11 @@ static void rte_eal_init_alert(const char *msg)
 	RTE_LOG(ERR, EAL, "%s\n", msg);
 }
 
+static void mark_forked(void)
+{
+	is_forked++;
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
@@ -883,16 +890,22 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, NULL, mark_forked);
+
 	return fctret;
 }
 
 int
 rte_eal_cleanup(void)
 {
+	if (is_forked)
+		return 0;
+
 	struct internal_config *internal_conf =
 		eal_get_internal_configuration();
 	rte_service_finalize();
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	rte_trace_save();
 	eal_trace_fini();
 	/* after this point, any DPDK pointers will become dangling */
diff --git a/lib/eal/freebsd/eal_interrupts.c b/lib/eal/freebsd/eal_interrupts.c
index 9f720bdc8f..cac3859b06 100644
--- a/lib/eal/freebsd/eal_interrupts.c
+++ b/lib/eal/freebsd/eal_interrupts.c
@@ -648,6 +648,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close kqueue */
+	close(kq);
+	kq = -1;
+}
+
 int
 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle,
 		int epfd, int op, unsigned int vec, void *data)
diff --git a/lib/eal/linux/eal.c b/lib/eal/linux/eal.c
index 37d29643a5..ea38ec183d 100644
--- a/lib/eal/linux/eal.c
+++ b/lib/eal/linux/eal.c
@@ -76,6 +76,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge if is forked */
+static int is_forked;
 
 int
 eal_clean_runtime_dir(void)
@@ -954,6 +956,11 @@ eal_worker_thread_create(unsigned int lcore_id)
 	return ret;
 }
 
+static void mark_forked(void)
+{
+	is_forked++;
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
@@ -1324,6 +1331,8 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, NULL, mark_forked);
+
 	return fctret;
 }
 
@@ -1347,6 +1356,9 @@ mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
 int
 rte_eal_cleanup(void)
 {
+	if (is_forked)
+		return 0;
+
 	/* if we're in a primary process, we need to mark hugepages as freeable
 	 * so that finalization can release them back to the system.
 	 */
@@ -1362,6 +1374,7 @@ rte_eal_cleanup(void)
 	vfio_mp_sync_cleanup();
 #endif
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	rte_trace_save();
 	eal_trace_fini();
 	/* after this point, any DPDK pointers will become dangling */
diff --git a/lib/eal/linux/eal_interrupts.c b/lib/eal/linux/eal_interrupts.c
index d52ec8eb4c..7e9853e8e7 100644
--- a/lib/eal/linux/eal_interrupts.c
+++ b/lib/eal/linux/eal_interrupts.c
@@ -1199,6 +1199,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close the pipe used by epoll */
+	close(intr_pipe.writefd);
+	close(intr_pipe.readfd);
+}
+
 static void
 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
 {
-- 
2.25.1

Re: [PATCH v5] lib/eal: fix segfaults in exiting

[Stephen Hemminger]

On Tue,  6 Sep 2022 10:51:31 +0800
Zhichao Zeng <zhichaox.zeng@intel.com> wrote:

>  
> +static void mark_forked(void)
> +{
> +	is_forked++;
> +}
> +

This will end up counting application threads as well.

Also, it would need to be atomic.

>  /* Launch threads, called at application init(). */
>  int
>  rte_eal_init(int argc, char **argv)
> @@ -1324,6 +1331,8 @@ rte_eal_init(int argc, char **argv)
>  
>  	eal_mcfg_complete();
>  
> +	pthread_atfork(NULL, NULL, mark_forked);
> +
>  	return fctret;
>  }

>  int
>  rte_eal_cleanup(void)
>  {
> +	if (is_forked)
> +		return 0;
> +

rte_eal_cleanup is supposed to be called only once by application.

RE: [PATCH v5] lib/eal: fix segfaults in exiting

[Zeng, ZhichaoX]

Hi Stephen,

> >
> > +static void mark_forked(void)
> > +{
> > +	is_forked++;
> > +}
> > +
> 
> This will end up counting application threads as well.
> 

I think it would be counted in the child process when 'fork()' is called,
and in the parent process, it would be zero.

> Also, it would need to be atomic.
> 

Thanks for your advice.

> >  /* Launch threads, called at application init(). */  int
> > rte_eal_init(int argc, char **argv) @@ -1324,6 +1331,8 @@
> > rte_eal_init(int argc, char **argv)
> >
> >  	eal_mcfg_complete();
> >
> > +	pthread_atfork(NULL, NULL, mark_forked);
> > +
> >  	return fctret;
> >  }
> 
> >  int
> >  rte_eal_cleanup(void)
> >  {
> > +	if (is_forked)
> > +		return 0;
> > +
> 
> rte_eal_cleanup is supposed to be called only once by application.

Yes. But in some case(e.g. debug_autotest of dpdk-test), it would fork
a child process to test 'rte_exit()', then it would call 'rte_eal_cleanup()'.
So 'is_forked' is introduced to avoid this situation.

Regards
Zhichao

[PATCH v6] lib/eal: fix segfaults in exiting

[Zhichao Zeng]

The 'eal-intr-thread' is not closed before memory cleanup in the process of
exiting. There is a small chance when 'eal-intr-thread' use some pointers,
meanwhile the memory was just cleaned, which causes segfaults.

This patch closes the 'eal-intr-thread' before memory cleanup in
'rte_eal_cleanup' to avoid segfaults, and adds a flag to avoid executing
'rte_eal_cleanup' in the child process which is forked to execute some
test cases(e.g. debug_autotest of dpdk-test).

Bugzilla ID: 1006
Cc: stable@dpdk.org

Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>

---
v6: use atomic operation
---
v5: simplify patch
---
v4: shorten the prompt message and optimize the commit log
---
v3: fix rte_eal_cleanup crash in debug_autotest
---
v2: add same API for FreeBSD
---
 lib/eal/common/eal_private.h     |  7 +++++++
 lib/eal/freebsd/eal.c            | 20 ++++++++++++++++++++
 lib/eal/freebsd/eal_interrupts.c | 12 ++++++++++++
 lib/eal/linux/eal.c              | 20 ++++++++++++++++++++
 lib/eal/linux/eal_interrupts.c   | 12 ++++++++++++
 5 files changed, 71 insertions(+)

diff --git a/lib/eal/common/eal_private.h b/lib/eal/common/eal_private.h
index 0f4d75bb89..2e3342bd15 100644
--- a/lib/eal/common/eal_private.h
+++ b/lib/eal/common/eal_private.h
@@ -152,6 +152,13 @@ int rte_eal_tailqs_init(void);
  */
 int rte_eal_intr_init(void);
 
+/**
+ * Destroy interrupt handling thread.
+ *
+ * This function is private to EAL.
+ */
+void rte_eal_intr_destroy(void);
+
 /**
  * Close the default log stream
  *
diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
index 1b58cd3da6..0839aa211c 100644
--- a/lib/eal/freebsd/eal.c
+++ b/lib/eal/freebsd/eal.c
@@ -73,6 +73,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* mark process is forked */
+static uint32_t forked_flag;
 
 int
 eal_clean_runtime_dir(void)
@@ -575,6 +577,18 @@ static void rte_eal_init_alert(const char *msg)
 	RTE_LOG(ERR, EAL, "%s\n", msg);
 }
 
+static void
+mark_forked(void)
+{
+	__atomic_add_fetch(&forked_flag, 1, __ATOMIC_RELAXED);
+}
+
+static uint32_t
+is_forked(void)
+{
+	return __atomic_load_n(&forked_flag, __ATOMIC_RELAXED);
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
@@ -884,16 +898,22 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, NULL, mark_forked);
+
 	return fctret;
 }
 
 int
 rte_eal_cleanup(void)
 {
+	if (is_forked())
+		return 0;
+
 	struct internal_config *internal_conf =
 		eal_get_internal_configuration();
 	rte_service_finalize();
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	eal_bus_cleanup();
 	rte_trace_save();
 	eal_trace_fini();
diff --git a/lib/eal/freebsd/eal_interrupts.c b/lib/eal/freebsd/eal_interrupts.c
index 9f720bdc8f..cac3859b06 100644
--- a/lib/eal/freebsd/eal_interrupts.c
+++ b/lib/eal/freebsd/eal_interrupts.c
@@ -648,6 +648,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close kqueue */
+	close(kq);
+	kq = -1;
+}
+
 int
 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle,
 		int epfd, int op, unsigned int vec, void *data)
diff --git a/lib/eal/linux/eal.c b/lib/eal/linux/eal.c
index e74542fc71..ef15d7e7f0 100644
--- a/lib/eal/linux/eal.c
+++ b/lib/eal/linux/eal.c
@@ -77,6 +77,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* mark process is forked */
+static uint32_t forked_flag;
 
 int
 eal_clean_runtime_dir(void)
@@ -955,6 +957,18 @@ eal_worker_thread_create(unsigned int lcore_id)
 	return ret;
 }
 
+static void
+mark_forked(void)
+{
+	__atomic_add_fetch(&forked_flag, 1, __ATOMIC_RELAXED);
+}
+
+static uint32_t
+is_forked(void)
+{
+	return __atomic_load_n(&forked_flag, __ATOMIC_RELAXED);
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
@@ -1325,6 +1339,8 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, NULL, mark_forked);
+
 	return fctret;
 }
 
@@ -1348,6 +1364,9 @@ mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
 int
 rte_eal_cleanup(void)
 {
+	if (is_forked())
+		return 0;
+
 	/* if we're in a primary process, we need to mark hugepages as freeable
 	 * so that finalization can release them back to the system.
 	 */
@@ -1363,6 +1382,7 @@ rte_eal_cleanup(void)
 	vfio_mp_sync_cleanup();
 #endif
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	eal_bus_cleanup();
 	rte_trace_save();
 	eal_trace_fini();
diff --git a/lib/eal/linux/eal_interrupts.c b/lib/eal/linux/eal_interrupts.c
index d52ec8eb4c..7e9853e8e7 100644
--- a/lib/eal/linux/eal_interrupts.c
+++ b/lib/eal/linux/eal_interrupts.c
@@ -1199,6 +1199,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close the pipe used by epoll */
+	close(intr_pipe.writefd);
+	close(intr_pipe.readfd);
+}
+
 static void
 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
 {
-- 
2.25.1

Re: [PATCH v6] lib/eal: fix segfaults in exiting

[Stephen Hemminger]

On Tue, 11 Oct 2022 13:25:14 +0800
Zhichao Zeng <zhichaox.zeng@intel.com> wrote:

> This patch closes the 'eal-intr-thread' before memory cleanup in
> 'rte_eal_cleanup' to avoid segfaults, and adds a flag to avoid executing
> 'rte_eal_cleanup' in the child process which is forked to execute some
> test cases(e.g. debug_autotest of dpdk-test

This is a test bug, not an DPDK bug. I don't think DPDK should account
for misuse of API in this way.

RE: [PATCH v6] lib/eal: fix segfaults in exiting

[Zeng, ZhichaoX]

Hi Stephen, thanks for your comments. 

The original goal of this patch is to close the 'eal-intr-thread' to fix segfaults
caught by Asan. But it breaks the debug_autotest unit test.

So the flag is added to fix the unit test.

Hi @David Marchand, what's your suggestions?

BR
Zhichao

> -----Original Message-----
> From: Stephen Hemminger <stephen@networkplumber.org>
> Sent: Tuesday, October 11, 2022 10:04 PM
> To: Zeng, ZhichaoX <zhichaox.zeng@intel.com>
> Cc: dev@dpdk.org; stable@dpdk.org; Zhou, YidingX
> <yidingx.zhou@intel.com>; Richardson, Bruce
> <bruce.richardson@intel.com>; Harman Kalra <hkalra@marvell.com>
> Subject: Re: [PATCH v6] lib/eal: fix segfaults in exiting
> 
> On Tue, 11 Oct 2022 13:25:14 +0800
> Zhichao Zeng <zhichaox.zeng@intel.com> wrote:
> 
> > This patch closes the 'eal-intr-thread' before memory cleanup in
> > 'rte_eal_cleanup' to avoid segfaults, and adds a flag to avoid
> > executing 'rte_eal_cleanup' in the child process which is forked to
> > execute some test cases(e.g. debug_autotest of dpdk-test
> 
> This is a test bug, not an DPDK bug. I don't think DPDK should account for
> misuse of API in this way.

[PATCH v5] lib/eal: fix segfaults in exiting

[Zhichao Zeng]

The 'eal-intr-thread' is not closed before memory cleanup in the process of
exiting. There is a small chance that the 'eal-intr-thread' is about to use
some pointers, the memory was just cleaned, which causes segfaults
caught by ASan.

This patch closes the 'eal-intr-thread' before memory cleanup in
'rte_eal_cleanup' to avoid segfaults, and adds a flag to avoid executing
'rte_eal_cleanup' in the child process which is forked to execut some
test cases(e.g. debug_autotest of dpdk-test).

Bugzilla ID: 1006
Cc: stable@dpdk.org

Signed-off-by: Zhichao Zeng <zhichaox.zeng@intel.com>

---
v2: add same API for FreeBSD
---
v3: fix rte_eal_cleanup crash in debug_autotest
---
v4: shorten the prompt message and optimize the commit log
---
v5: simplify patch
---
 lib/eal/common/eal_private.h     |  7 +++++++
 lib/eal/freebsd/eal.c            | 13 +++++++++++++
 lib/eal/freebsd/eal_interrupts.c | 12 ++++++++++++
 lib/eal/linux/eal.c              | 13 +++++++++++++
 lib/eal/linux/eal_interrupts.c   | 12 ++++++++++++
 5 files changed, 57 insertions(+)

diff --git a/lib/eal/common/eal_private.h b/lib/eal/common/eal_private.h
index 44d14241f0..7adf41b7d7 100644
--- a/lib/eal/common/eal_private.h
+++ b/lib/eal/common/eal_private.h
@@ -152,6 +152,13 @@ int rte_eal_tailqs_init(void);
  */
 int rte_eal_intr_init(void);
 
+/**
+ * Destroy interrupt handling thread.
+ *
+ * This function is private to EAL.
+ */
+void rte_eal_intr_destroy(void);
+
 /**
  * Close the default log stream
  *
diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
index 26fbc91b26..f27c1d9f97 100644
--- a/lib/eal/freebsd/eal.c
+++ b/lib/eal/freebsd/eal.c
@@ -72,6 +72,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge if is forked */
+static int is_forked;
 
 int
 eal_clean_runtime_dir(void)
@@ -574,6 +576,11 @@ static void rte_eal_init_alert(const char *msg)
 	RTE_LOG(ERR, EAL, "%s\n", msg);
 }
 
+static void mark_forked(void)
+{
+	is_forked++;
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
@@ -883,16 +890,22 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, NULL, mark_forked);
+
 	return fctret;
 }
 
 int
 rte_eal_cleanup(void)
 {
+	if (is_forked)
+		return 0;
+
 	struct internal_config *internal_conf =
 		eal_get_internal_configuration();
 	rte_service_finalize();
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	rte_trace_save();
 	eal_trace_fini();
 	/* after this point, any DPDK pointers will become dangling */
diff --git a/lib/eal/freebsd/eal_interrupts.c b/lib/eal/freebsd/eal_interrupts.c
index 9f720bdc8f..cac3859b06 100644
--- a/lib/eal/freebsd/eal_interrupts.c
+++ b/lib/eal/freebsd/eal_interrupts.c
@@ -648,6 +648,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close kqueue */
+	close(kq);
+	kq = -1;
+}
+
 int
 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle,
 		int epfd, int op, unsigned int vec, void *data)
diff --git a/lib/eal/linux/eal.c b/lib/eal/linux/eal.c
index 37d29643a5..ea38ec183d 100644
--- a/lib/eal/linux/eal.c
+++ b/lib/eal/linux/eal.c
@@ -76,6 +76,8 @@ struct lcore_config lcore_config[RTE_MAX_LCORE];
 /* used by rte_rdtsc() */
 int rte_cycles_vmware_tsc_map;
 
+/* used to judge if is forked */
+static int is_forked;
 
 int
 eal_clean_runtime_dir(void)
@@ -954,6 +956,11 @@ eal_worker_thread_create(unsigned int lcore_id)
 	return ret;
 }
 
+static void mark_forked(void)
+{
+	is_forked++;
+}
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
@@ -1324,6 +1331,8 @@ rte_eal_init(int argc, char **argv)
 
 	eal_mcfg_complete();
 
+	pthread_atfork(NULL, NULL, mark_forked);
+
 	return fctret;
 }
 
@@ -1347,6 +1356,9 @@ mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
 int
 rte_eal_cleanup(void)
 {
+	if (is_forked)
+		return 0;
+
 	/* if we're in a primary process, we need to mark hugepages as freeable
 	 * so that finalization can release them back to the system.
 	 */
@@ -1362,6 +1374,7 @@ rte_eal_cleanup(void)
 	vfio_mp_sync_cleanup();
 #endif
 	rte_mp_channel_cleanup();
+	rte_eal_intr_destroy();
 	rte_trace_save();
 	eal_trace_fini();
 	/* after this point, any DPDK pointers will become dangling */
diff --git a/lib/eal/linux/eal_interrupts.c b/lib/eal/linux/eal_interrupts.c
index d52ec8eb4c..7e9853e8e7 100644
--- a/lib/eal/linux/eal_interrupts.c
+++ b/lib/eal/linux/eal_interrupts.c
@@ -1199,6 +1199,18 @@ rte_eal_intr_init(void)
 	return ret;
 }
 
+void
+rte_eal_intr_destroy(void)
+{
+	/* cancel the host thread to wait/handle the interrupt */
+	pthread_cancel(intr_thread);
+	pthread_join(intr_thread, NULL);
+
+	/* close the pipe used by epoll */
+	close(intr_pipe.writefd);
+	close(intr_pipe.readfd);
+}
+
 static void
 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
 {
-- 
2.25.1