diff mbox

[dpdk-dev] eal: fix threads block on barrier

Message ID 1524847302-88110-1-git-send-email-jianfeng.tan@intel.com (mailing list archive)
State Accepted, archived
Headers

Checks

Context Check Description
ci/checkpatch warning coding style issues
ci/Intel-compilation fail apply patch file failure

Commit Message

Jianfeng Tan April 27, 2018, 4:41 p.m. UTC 
  Below commit introduced pthread barrier for synchronization.
But two IPC threads block on the barrier, and never wake up.

  (gdb) bt
  #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
      at ../sysdeps/unix/sysv/linux/futex-internal.h:61
  #1  futex_wait_simple (private=0, expected=0, futex_word=0x7fffffffcff4)
      at ../sysdeps/nptl/futex-internal.h:135
  #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at pthread_barrier_wait.c:184
  #3  rte_thread_init (arg=0x7fffffffcfe0)
      at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
  #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
  #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109

Through analysis, we find the barrier defined on the stack could be the
root cause. This patch will change to use heap memory as the barrier.

Fixes: d651ee4919cd ("eal: set affinity for control threads")

Cc: Olivier Matz <olivier.matz@6wind.com>
Cc: Anatoly Burakov <anatoly.burakov@intel.com>

Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
---
 lib/librte_eal/common/eal_common_thread.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)

Comments

Thomas Monjalon April 27, 2018, 4:42 p.m. UTC | #1 
27/04/2018 18:41, Jianfeng Tan:
> Below commit introduced pthread barrier for synchronization.
> But two IPC threads block on the barrier, and never wake up.
> 
>   (gdb) bt
>   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
>       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
>   #1  futex_wait_simple (private=0, expected=0, futex_word=0x7fffffffcff4)
>       at ../sysdeps/nptl/futex-internal.h:135
>   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at pthread_barrier_wait.c:184
>   #3  rte_thread_init (arg=0x7fffffffcfe0)
>       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
>   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
>   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> 
> Through analysis, we find the barrier defined on the stack could be the
> root cause. This patch will change to use heap memory as the barrier.
> 
> Fixes: d651ee4919cd ("eal: set affinity for control threads")

Shreyansh (Cc'ed) is seeing some bugs with this patch too.
Stephen Hemminger April 27, 2018, 5:03 p.m. UTC | #2 
On Fri, 27 Apr 2018 16:41:42 +0000
Jianfeng Tan <jianfeng.tan@intel.com> wrote:

> Below commit introduced pthread barrier for synchronization.
> But two IPC threads block on the barrier, and never wake up.
> 
>   (gdb) bt
>   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
>       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
>   #1  futex_wait_simple (private=0, expected=0, futex_word=0x7fffffffcff4)
>       at ../sysdeps/nptl/futex-internal.h:135
>   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at pthread_barrier_wait.c:184
>   #3  rte_thread_init (arg=0x7fffffffcfe0)
>       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
>   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
>   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> 
> Through analysis, we find the barrier defined on the stack could be the
> root cause. This patch will change to use heap memory as the barrier.
> 
> Fixes: d651ee4919cd ("eal: set affinity for control threads")
> 
> Cc: Olivier Matz <olivier.matz@6wind.com>
> Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> 
> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
> ---
>  lib/librte_eal/common/eal_common_thread.c | 20 +++++++++++++-------
>  1 file changed, 13 insertions(+), 7 deletions(-)
> 
> diff --git a/lib/librte_eal/common/eal_common_thread.c b/lib/librte_eal/common/eal_common_thread.c
> index 4e75cb8..da2b84f 100644
> --- a/lib/librte_eal/common/eal_common_thread.c
> +++ b/lib/librte_eal/common/eal_common_thread.c
> @@ -166,17 +166,21 @@ rte_ctrl_thread_create(pthread_t *thread, const char *name,
>  		const pthread_attr_t *attr,
>  		void *(*start_routine)(void *), void *arg)
>  {
> -	struct rte_thread_ctrl_params params = {
> -		.start_routine = start_routine,
> -		.arg = arg,
> -	};
> +	struct rte_thread_ctrl_params *params;
>  	unsigned int lcore_id;
>  	rte_cpuset_t cpuset;
>  	int cpu_found, ret;
>  
> -	pthread_barrier_init(&params.configured, NULL, 2);
> +	params = malloc(sizeof(*params));
> +	if (!params)
> +		return -1;
> +
> +	params->start_routine = start_routine;
> +	params->arg = arg;
>  
> -	ret = pthread_create(thread, attr, rte_thread_init, (void *)&params);
> +	pthread_barrier_init(&params->configured, NULL, 2);
> +
> +	ret = pthread_create(thread, attr, rte_thread_init, (void *)params);
>  	if (ret != 0)
>  		return ret;
>  
> @@ -203,12 +207,14 @@ rte_ctrl_thread_create(pthread_t *thread, const char *name,
>  	if (ret < 0)
>  		goto fail;
>  
> -	pthread_barrier_wait(&params.configured);
> +	pthread_barrier_wait(&params->configured);
> +	free(params);
>  
>  	return 0;
>  
>  fail:
>  	pthread_cancel(*thread);
>  	pthread_join(*thread, NULL);
> +	free(params);
>  	return ret;
>  }

This looks like a library bug. If there is a race on the configured barrier, then
putting on heap is just moving problem. It still has bug where other thread is referring to freed memory.
Shreyansh Jain April 27, 2018, 5:36 p.m. UTC | #3 
> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jianfeng Tan
> Sent: Friday, April 27, 2018 10:12 PM
> To: dev@dpdk.org
> Cc: thomas@monjalon.net; Jianfeng Tan <jianfeng.tan@intel.com>; Olivier
> Matz <olivier.matz@6wind.com>; Anatoly Burakov
> <anatoly.burakov@intel.com>
> Subject: [dpdk-dev] [PATCH] eal: fix threads block on barrier
> 
> Below commit introduced pthread barrier for synchronization.
> But two IPC threads block on the barrier, and never wake up.
> 
>   (gdb) bt
>   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
>       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
>   #1  futex_wait_simple (private=0, expected=0,
> futex_word=0x7fffffffcff4)
>       at ../sysdeps/nptl/futex-internal.h:135
>   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
> pthread_barrier_wait.c:184
>   #3  rte_thread_init (arg=0x7fffffffcfe0)
>       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
>   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
>   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> 
> Through analysis, we find the barrier defined on the stack could be the
> root cause. This patch will change to use heap memory as the barrier.
> 
> Fixes: d651ee4919cd ("eal: set affinity for control threads")
> 
> Cc: Olivier Matz <olivier.matz@6wind.com>
> Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> 
> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>

Though I have seen Stephen's comment on this (possibly a library bug), this at least fixes an issue which was dogging dpaa and dpaa2 - generating bus errors and futex errors with variation in core masks provided to applications.

Thanks a lot for this.

Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Stephen Hemminger April 27, 2018, 5:39 p.m. UTC | #4 
On Fri, 27 Apr 2018 17:36:56 +0000
Shreyansh Jain <shreyansh.jain@nxp.com> wrote:

> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jianfeng Tan
> > Sent: Friday, April 27, 2018 10:12 PM
> > To: dev@dpdk.org
> > Cc: thomas@monjalon.net; Jianfeng Tan <jianfeng.tan@intel.com>; Olivier
> > Matz <olivier.matz@6wind.com>; Anatoly Burakov
> > <anatoly.burakov@intel.com>
> > Subject: [dpdk-dev] [PATCH] eal: fix threads block on barrier
> > 
> > Below commit introduced pthread barrier for synchronization.
> > But two IPC threads block on the barrier, and never wake up.
> > 
> >   (gdb) bt
> >   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
> >       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
> >   #1  futex_wait_simple (private=0, expected=0,
> > futex_word=0x7fffffffcff4)
> >       at ../sysdeps/nptl/futex-internal.h:135
> >   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
> > pthread_barrier_wait.c:184
> >   #3  rte_thread_init (arg=0x7fffffffcfe0)
> >       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
> >   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
> >   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> > 
> > Through analysis, we find the barrier defined on the stack could be the
> > root cause. This patch will change to use heap memory as the barrier.
> > 
> > Fixes: d651ee4919cd ("eal: set affinity for control threads")
> > 
> > Cc: Olivier Matz <olivier.matz@6wind.com>
> > Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> > 
> > Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>  
> 
> Though I have seen Stephen's comment on this (possibly a library bug), this at least fixes an issue which was dogging dpaa and dpaa2 - generating bus errors and futex errors with variation in core masks provided to applications.
> 
> Thanks a lot for this.
> 
> Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

Could you verify there is not a use after free by using valgrind or some library that poisons memory on free.
Shreyansh Jain April 27, 2018, 5:45 p.m. UTC | #5 
> -----Original Message-----
> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> Sent: Friday, April 27, 2018 11:10 PM
> To: Shreyansh Jain <shreyansh.jain@nxp.com>
> Cc: Jianfeng Tan <jianfeng.tan@intel.com>; dev@dpdk.org;
> thomas@monjalon.net; Olivier Matz <olivier.matz@6wind.com>; Anatoly
> Burakov <anatoly.burakov@intel.com>
> Subject: Re: [dpdk-dev] [PATCH] eal: fix threads block on barrier
> 
> On Fri, 27 Apr 2018 17:36:56 +0000
> Shreyansh Jain <shreyansh.jain@nxp.com> wrote:
> 
> > > -----Original Message-----
> > > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jianfeng Tan
> > > Sent: Friday, April 27, 2018 10:12 PM
> > > To: dev@dpdk.org
> > > Cc: thomas@monjalon.net; Jianfeng Tan <jianfeng.tan@intel.com>;
> Olivier
> > > Matz <olivier.matz@6wind.com>; Anatoly Burakov
> > > <anatoly.burakov@intel.com>
> > > Subject: [dpdk-dev] [PATCH] eal: fix threads block on barrier
> > >
> > > Below commit introduced pthread barrier for synchronization.
> > > But two IPC threads block on the barrier, and never wake up.
> > >
> > >   (gdb) bt
> > >   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
> > >       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
> > >   #1  futex_wait_simple (private=0, expected=0,
> > > futex_word=0x7fffffffcff4)
> > >       at ../sysdeps/nptl/futex-internal.h:135
> > >   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
> > > pthread_barrier_wait.c:184
> > >   #3  rte_thread_init (arg=0x7fffffffcfe0)
> > >       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
> > >   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
> > >   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> > >
> > > Through analysis, we find the barrier defined on the stack could be
> the
> > > root cause. This patch will change to use heap memory as the
> barrier.
> > >
> > > Fixes: d651ee4919cd ("eal: set affinity for control threads")
> > >
> > > Cc: Olivier Matz <olivier.matz@6wind.com>
> > > Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> > >
> > > Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
> >
> > Though I have seen Stephen's comment on this (possibly a library
> bug), this at least fixes an issue which was dogging dpaa and dpaa2 -
> generating bus errors and futex errors with variation in core masks
> provided to applications.
> >
> > Thanks a lot for this.
> >
> > Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
> 
> Could you verify there is not a use after free by using valgrind or
> some library that poisons memory on free.

I will probably do that soon - but for the time being I don't want this issue to block the dpaa/dpaa2 for RC1 - these drivers were completely unusable without this patch.
Thomas Monjalon April 27, 2018, 7:52 p.m. UTC | #6 
27/04/2018 19:45, Shreyansh Jain:
> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> > Shreyansh Jain <shreyansh.jain@nxp.com> wrote:
> > > From: Jianfeng Tan
> > > > Below commit introduced pthread barrier for synchronization.
> > > > But two IPC threads block on the barrier, and never wake up.
> > > >
> > > >   (gdb) bt
> > > >   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
> > > >       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
> > > >   #1  futex_wait_simple (private=0, expected=0,
> > > > futex_word=0x7fffffffcff4)
> > > >       at ../sysdeps/nptl/futex-internal.h:135
> > > >   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
> > > > pthread_barrier_wait.c:184
> > > >   #3  rte_thread_init (arg=0x7fffffffcfe0)
> > > >       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
> > > >   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
> > > >   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> > > >
> > > > Through analysis, we find the barrier defined on the stack
> > > > could be the root cause. This patch will change to use heap
> > > > memory as the barrier.
> > > >
> > > > Fixes: d651ee4919cd ("eal: set affinity for control threads")
> > > >
> > > > Cc: Olivier Matz <olivier.matz@6wind.com>
> > > > Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> > > >
> > > > Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
> > >
> > > Though I have seen Stephen's comment on this (possibly a library
> > bug), this at least fixes an issue which was dogging dpaa and dpaa2 -
> > generating bus errors and futex errors with variation in core masks
> > provided to applications.
> > >
> > > Thanks a lot for this.
> > >
> > > Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

Applied, thanks Jianfeng.

> > Could you verify there is not a use after free by using valgrind or
> > some library that poisons memory on free.
> 
> I will probably do that soon - but for the time being I don't want
> this issue to block the dpaa/dpaa2 for RC1 - these drivers were
> completely unusable without this patch.

Please Shreyansh, continue the analysis of this bug.
Thanks
Stephen Hemminger April 28, 2018, 1:21 a.m. UTC | #7 
On Fri, 27 Apr 2018 21:52:26 +0200
Thomas Monjalon <thomas@monjalon.net> wrote:

> 27/04/2018 19:45, Shreyansh Jain:
> > From: Stephen Hemminger [mailto:stephen@networkplumber.org]  
> > > Shreyansh Jain <shreyansh.jain@nxp.com> wrote:  
> > > > From: Jianfeng Tan  
> > > > > Below commit introduced pthread barrier for synchronization.
> > > > > But two IPC threads block on the barrier, and never wake up.
> > > > >
> > > > >   (gdb) bt
> > > > >   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
> > > > >       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
> > > > >   #1  futex_wait_simple (private=0, expected=0,
> > > > > futex_word=0x7fffffffcff4)
> > > > >       at ../sysdeps/nptl/futex-internal.h:135
> > > > >   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
> > > > > pthread_barrier_wait.c:184
> > > > >   #3  rte_thread_init (arg=0x7fffffffcfe0)
> > > > >       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
> > > > >   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
> > > > >   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> > > > >
> > > > > Through analysis, we find the barrier defined on the stack
> > > > > could be the root cause. This patch will change to use heap
> > > > > memory as the barrier.
> > > > >
> > > > > Fixes: d651ee4919cd ("eal: set affinity for control threads")
> > > > >
> > > > > Cc: Olivier Matz <olivier.matz@6wind.com>
> > > > > Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> > > > >
> > > > > Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>  
> > > >
> > > > Though I have seen Stephen's comment on this (possibly a library  
> > > bug), this at least fixes an issue which was dogging dpaa and dpaa2 -
> > > generating bus errors and futex errors with variation in core masks
> > > provided to applications.  
> > > >
> > > > Thanks a lot for this.
> > > >
> > > > Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>  
> 
> Applied, thanks Jianfeng.
> 
> > > Could you verify there is not a use after free by using valgrind or
> > > some library that poisons memory on free.  
> > 
> > I will probably do that soon - but for the time being I don't want
> > this issue to block the dpaa/dpaa2 for RC1 - these drivers were
> > completely unusable without this patch.  
> 
> Please Shreyansh, continue the analysis of this bug.
> Thanks
> 
> 

I think the patch needs to change.
The attributes need be either global (or leak and never free).

The glibc source for init keeps the pointer to the attributes.


static const struct pthread_barrierattr default_barrierattr =
  {
    .pshared = PTHREAD_PROCESS_PRIVATE
  };


int
__pthread_barrier_init (pthread_barrier_t *barrier,
			const pthread_barrierattr_t *attr, unsigned int count)
{
  struct pthread_barrier *ibarrier;

  /* XXX EINVAL is not specified by POSIX as a possible error code for COUNT
     being too large.  See pthread_barrier_wait for the reason for the
     comparison with BARRIER_IN_THRESHOLD.  */
  if (__glibc_unlikely (count == 0 || count >= BARRIER_IN_THRESHOLD))
    return EINVAL;

  const struct pthread_barrierattr *iattr
    = (attr != NULL
       ? (struct pthread_barrierattr *) attr
       : &default_barrierattr);

  ibarrier = (struct pthread_barrier *) barrier;

  /* Initialize the individual fields.  */
  ibarrier->in = 0;
  ibarrier->out = 0;
  ibarrier->count = count;
  ibarrier->current_round = 0;
  ibarrier->shared = (iattr->pshared == PTHREAD_PROCESS_PRIVATE
		      ? FUTEX_PRIVATE : FUTEX_SHARED);

  return 0;
}
weak_alias (__pthread_barrier_init, pthread_barrier_init)
Stephen Hemminger April 28, 2018, 1:24 a.m. UTC | #8 
On Fri, 27 Apr 2018 21:52:26 +0200
Thomas Monjalon <thomas@monjalon.net> wrote:

> 27/04/2018 19:45, Shreyansh Jain:
> > From: Stephen Hemminger [mailto:stephen@networkplumber.org]  
> > > Shreyansh Jain <shreyansh.jain@nxp.com> wrote:  
> > > > From: Jianfeng Tan  
> > > > > Below commit introduced pthread barrier for synchronization.
> > > > > But two IPC threads block on the barrier, and never wake up.
> > > > >
> > > > >   (gdb) bt
> > > > >   #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
> > > > >       at ../sysdeps/unix/sysv/linux/futex-internal.h:61
> > > > >   #1  futex_wait_simple (private=0, expected=0,
> > > > > futex_word=0x7fffffffcff4)
> > > > >       at ../sysdeps/nptl/futex-internal.h:135
> > > > >   #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
> > > > > pthread_barrier_wait.c:184
> > > > >   #3  rte_thread_init (arg=0x7fffffffcfe0)
> > > > >       at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
> > > > >   #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
> > > > >   #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
> > > > >
> > > > > Through analysis, we find the barrier defined on the stack
> > > > > could be the root cause. This patch will change to use heap
> > > > > memory as the barrier.
> > > > >
> > > > > Fixes: d651ee4919cd ("eal: set affinity for control threads")
> > > > >
> > > > > Cc: Olivier Matz <olivier.matz@6wind.com>
> > > > > Cc: Anatoly Burakov <anatoly.burakov@intel.com>
> > > > >
> > > > > Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>  
> > > >
> > > > Though I have seen Stephen's comment on this (possibly a library  
> > > bug), this at least fixes an issue which was dogging dpaa and dpaa2 -
> > > generating bus errors and futex errors with variation in core masks
> > > provided to applications.  
> > > >
> > > > Thanks a lot for this.
> > > >
> > > > Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>  
> 
> Applied, thanks Jianfeng.
> 
> > > Could you verify there is not a use after free by using valgrind or
> > > some library that poisons memory on free.  
> > 
> > I will probably do that soon - but for the time being I don't want
> > this issue to block the dpaa/dpaa2 for RC1 - these drivers were
> > completely unusable without this patch.  
> 
> Please Shreyansh, continue the analysis of this bug.
> Thanks
> 
> 

The pthread_barrier should also be destroyed when it is no longer needed.
Jianfeng Tan April 28, 2018, 4:15 a.m. UTC | #9 
On 4/28/2018 9:21 AM, Stephen Hemminger wrote:
> On Fri, 27 Apr 2018 21:52:26 +0200
> Thomas Monjalon <thomas@monjalon.net> wrote:
>
>> 27/04/2018 19:45, Shreyansh Jain:
>>> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
>>>> Shreyansh Jain <shreyansh.jain@nxp.com> wrote:
>>>>> From: Jianfeng Tan
>>>>>> Below commit introduced pthread barrier for synchronization.
>>>>>> But two IPC threads block on the barrier, and never wake up.
>>>>>>
>>>>>>    (gdb) bt
>>>>>>    #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
>>>>>>        at ../sysdeps/unix/sysv/linux/futex-internal.h:61
>>>>>>    #1  futex_wait_simple (private=0, expected=0,
>>>>>> futex_word=0x7fffffffcff4)
>>>>>>        at ../sysdeps/nptl/futex-internal.h:135
>>>>>>    #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
>>>>>> pthread_barrier_wait.c:184
>>>>>>    #3  rte_thread_init (arg=0x7fffffffcfe0)
>>>>>>        at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
>>>>>>    #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
>>>>>>    #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
>>>>>>
>>>>>> Through analysis, we find the barrier defined on the stack
>>>>>> could be the root cause. This patch will change to use heap
>>>>>> memory as the barrier.
>>>>>>
>>>>>> Fixes: d651ee4919cd ("eal: set affinity for control threads")
>>>>>>
>>>>>> Cc: Olivier Matz <olivier.matz@6wind.com>
>>>>>> Cc: Anatoly Burakov <anatoly.burakov@intel.com>
>>>>>>
>>>>>> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
>>>>> Though I have seen Stephen's comment on this (possibly a library
>>>> bug), this at least fixes an issue which was dogging dpaa and dpaa2 -
>>>> generating bus errors and futex errors with variation in core masks
>>>> provided to applications.
>>>>> Thanks a lot for this.
>>>>>
>>>>> Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
>> Applied, thanks Jianfeng.
>>
>>>> Could you verify there is not a use after free by using valgrind or
>>>> some library that poisons memory on free.
>>> I will probably do that soon - but for the time being I don't want
>>> this issue to block the dpaa/dpaa2 for RC1 - these drivers were
>>> completely unusable without this patch.
>> Please Shreyansh, continue the analysis of this bug.
>> Thanks
>>
>>
> I think the patch needs to change.
> The attributes need be either global (or leak and never free).
>
> The glibc source for init keeps the pointer to the attributes.

Did not follow why we need to add attr here. Besides, init only uses 
attr to decide futex type (private or shared); seems that it does not 
keep  the pointer.

So I cannot understand why we need to add a non-null attr parameter.

Thanks,
Jianfeng

>
>
> static const struct pthread_barrierattr default_barrierattr =
>    {
>      .pshared = PTHREAD_PROCESS_PRIVATE
>    };
>
>
> int
> __pthread_barrier_init (pthread_barrier_t *barrier,
> 			const pthread_barrierattr_t *attr, unsigned int count)
> {
>    struct pthread_barrier *ibarrier;
>
>    /* XXX EINVAL is not specified by POSIX as a possible error code for COUNT
>       being too large.  See pthread_barrier_wait for the reason for the
>       comparison with BARRIER_IN_THRESHOLD.  */
>    if (__glibc_unlikely (count == 0 || count >= BARRIER_IN_THRESHOLD))
>      return EINVAL;
>
>    const struct pthread_barrierattr *iattr
>      = (attr != NULL
>         ? (struct pthread_barrierattr *) attr
>         : &default_barrierattr);
>
>    ibarrier = (struct pthread_barrier *) barrier;
>
>    /* Initialize the individual fields.  */
>    ibarrier->in = 0;
>    ibarrier->out = 0;
>    ibarrier->count = count;
>    ibarrier->current_round = 0;
>    ibarrier->shared = (iattr->pshared == PTHREAD_PROCESS_PRIVATE
> 		      ? FUTEX_PRIVATE : FUTEX_SHARED);
>
>    return 0;
> }
> weak_alias (__pthread_barrier_init, pthread_barrier_init)
Jianfeng Tan April 28, 2018, 4:22 a.m. UTC | #10 
On 4/28/2018 9:24 AM, Stephen Hemminger wrote:
> On Fri, 27 Apr 2018 21:52:26 +0200
> Thomas Monjalon <thomas@monjalon.net> wrote:
>
>> 27/04/2018 19:45, Shreyansh Jain:
>>> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
>>>> Shreyansh Jain <shreyansh.jain@nxp.com> wrote:
>>>>> From: Jianfeng Tan
>>>>>> Below commit introduced pthread barrier for synchronization.
>>>>>> But two IPC threads block on the barrier, and never wake up.
>>>>>>
>>>>>>    (gdb) bt
>>>>>>    #0  futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
>>>>>>        at ../sysdeps/unix/sysv/linux/futex-internal.h:61
>>>>>>    #1  futex_wait_simple (private=0, expected=0,
>>>>>> futex_word=0x7fffffffcff4)
>>>>>>        at ../sysdeps/nptl/futex-internal.h:135
>>>>>>    #2  __pthread_barrier_wait (barrier=0x7fffffffcff0) at
>>>>>> pthread_barrier_wait.c:184
>>>>>>    #3  rte_thread_init (arg=0x7fffffffcfe0)
>>>>>>        at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
>>>>>>    #4  start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
>>>>>>    #5  clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
>>>>>>
>>>>>> Through analysis, we find the barrier defined on the stack
>>>>>> could be the root cause. This patch will change to use heap
>>>>>> memory as the barrier.
>>>>>>
>>>>>> Fixes: d651ee4919cd ("eal: set affinity for control threads")
>>>>>>
>>>>>> Cc: Olivier Matz <olivier.matz@6wind.com>
>>>>>> Cc: Anatoly Burakov <anatoly.burakov@intel.com>
>>>>>>
>>>>>> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
>>>>> Though I have seen Stephen's comment on this (possibly a library
>>>> bug), this at least fixes an issue which was dogging dpaa and dpaa2 -
>>>> generating bus errors and futex errors with variation in core masks
>>>> provided to applications.
>>>>> Thanks a lot for this.
>>>>>
>>>>> Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
>> Applied, thanks Jianfeng.
>>
>>>> Could you verify there is not a use after free by using valgrind or
>>>> some library that poisons memory on free.
>>> I will probably do that soon - but for the time being I don't want
>>> this issue to block the dpaa/dpaa2 for RC1 - these drivers were
>>> completely unusable without this patch.
>> Please Shreyansh, continue the analysis of this bug.
>> Thanks
>>
>>
> The pthread_barrier should also be destroyed when it is no longer needed.

I tried this could also kick the sleeping thread; but due to "The effect 
of subsequent use of the barrier is undefined", I did not use that way.

Anyway, I agree that destroy() shall be called for completeness.

Thanks,
Jianfeng
diff mbox

Patch

diff --git a/lib/librte_eal/common/eal_common_thread.c b/lib/librte_eal/common/eal_common_thread.c
index 4e75cb8..da2b84f 100644
--- a/lib/librte_eal/common/eal_common_thread.c
+++ b/lib/librte_eal/common/eal_common_thread.c
@@ -166,17 +166,21 @@  rte_ctrl_thread_create(pthread_t *thread, const char *name,
 		const pthread_attr_t *attr,
 		void *(*start_routine)(void *), void *arg)
 {
-	struct rte_thread_ctrl_params params = {
-		.start_routine = start_routine,
-		.arg = arg,
-	};
+	struct rte_thread_ctrl_params *params;
 	unsigned int lcore_id;
 	rte_cpuset_t cpuset;
 	int cpu_found, ret;
 
-	pthread_barrier_init(&params.configured, NULL, 2);
+	params = malloc(sizeof(*params));
+	if (!params)
+		return -1;
+
+	params->start_routine = start_routine;
+	params->arg = arg;
 
-	ret = pthread_create(thread, attr, rte_thread_init, (void *)&params);
+	pthread_barrier_init(&params->configured, NULL, 2);
+
+	ret = pthread_create(thread, attr, rte_thread_init, (void *)params);
 	if (ret != 0)
 		return ret;
 
@@ -203,12 +207,14 @@  rte_ctrl_thread_create(pthread_t *thread, const char *name,
 	if (ret < 0)
 		goto fail;
 
-	pthread_barrier_wait(&params.configured);
+	pthread_barrier_wait(&params->configured);
+	free(params);
 
 	return 0;
 
 fail:
 	pthread_cancel(*thread);
 	pthread_join(*thread, NULL);
+	free(params);
 	return ret;
 }