[v3] raw/ifpga: fix pthread cannot join

  From: Tianfei Zhang <tianfei.zhang@intel.com>

When we want to close a thread, we should set a flag to notify
thread handler function.

Fixes: 9c006c45 ("raw/ifpga: scan PCIe BDF device tree")
Cc: stable@dpdk.org

Signed-off-by: Tianfei Zhang <tianfei.zhang@intel.com>
---
v2: update commit log
---
v3: set thread id to 0 after pthread_join
---
 drivers/raw/ifpga/ifpga_rawdev.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
  

On Thu, Jan 20, 2022 at 7:01 AM Wei Huang <wei.huang@intel.com> wrote:
>
> From: Tianfei Zhang <tianfei.zhang@intel.com>
>
> When we want to close a thread, we should set a flag to notify
> thread handler function.
>
> Fixes: 9c006c45 ("raw/ifpga: scan PCIe BDF device tree")
> Cc: stable@dpdk.org
>
> Signed-off-by: Tianfei Zhang <tianfei.zhang@intel.com>
> ---
> v2: update commit log
> ---
> v3: set thread id to 0 after pthread_join

Evaluating/setting a pthread_t variable is a bad idea: this type
should be considered opaque.
Plus here, the ifpga_monitor_start variable already indicates whether
ifpga_monitor_start_thread is valid.

Only checking for ifpga_monitor_start in ifpga_rawdev_gsd_handle loop
should be enough.



> ---
>  drivers/raw/ifpga/ifpga_rawdev.c | 11 +++++++----
>  1 file changed, 7 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/raw/ifpga/ifpga_rawdev.c b/drivers/raw/ifpga/ifpga_rawdev.c
> index 8d9db58..2bc569b 100644
> --- a/drivers/raw/ifpga/ifpga_rawdev.c
> +++ b/drivers/raw/ifpga/ifpga_rawdev.c
> @@ -497,7 +497,7 @@ static int set_surprise_link_check_aer(
>         int gsd_enable, ret;
>  #define MS 1000
>
> -       while (1) {
> +       while (ifpga_monitor_start) {
>                 gsd_enable = 0;
>                 for (i = 0; i < IFPGA_RAWDEV_NUM; i++) {
>                         ifpga_rdev = &ifpga_rawdevices[i];
> @@ -542,18 +542,21 @@ static int set_surprise_link_check_aer(
>  static int
>  ifpga_monitor_stop_func(void)
>  {
> +       void *status;

This status variable is unused, no need for it.


>         int ret;
>
> -       if (ifpga_monitor_start == 1) {
> +       if ((ifpga_monitor_start == 1) && ifpga_monitor_start_thread) {
> +               ifpga_monitor_start = 0;
> +
>                 ret = pthread_cancel(ifpga_monitor_start_thread);
>                 if (ret)
>                         IFPGA_RAWDEV_PMD_ERR("Can't cancel the thread");
>
> -               ret = pthread_join(ifpga_monitor_start_thread, NULL);
> +               ret = pthread_join(ifpga_monitor_start_thread, &status);
>                 if (ret)
>                         IFPGA_RAWDEV_PMD_ERR("Can't join the thread");
>
> -               ifpga_monitor_start = 0;
> +               ifpga_monitor_start_thread = 0;
>
>                 return ret;
>         }
> --
> 1.8.3.1
>
  

On Thu, 20 Jan 2022 00:58:14 -0500
Wei Huang <wei.huang@intel.com> wrote:

> diff --git a/drivers/raw/ifpga/ifpga_rawdev.c b/drivers/raw/ifpga/ifpga_rawdev.c
> index 8d9db58..2bc569b 100644
> --- a/drivers/raw/ifpga/ifpga_rawdev.c
> +++ b/drivers/raw/ifpga/ifpga_rawdev.c
> @@ -497,7 +497,7 @@ static int set_surprise_link_check_aer(
>  	int gsd_enable, ret;
>  #define MS 1000
>  
> -	while (1) {
> +	while (ifpga_monitor_start) {

This will work on x86 CPU but not on CPU's with weak memory
ordering like ARM.

You need to use atomic_builtin when accessing a flag shared between
threads that is not being protected by a lock.
  

> -----Original Message-----
> From: Stephen Hemminger <stephen@networkplumber.org>
> Sent: Thursday, January 20, 2022 11:47 PM
> To: Huang, Wei <wei.huang@intel.com>
> Cc: dev@dpdk.org; Xu, Rosen <rosen.xu@intel.com>; Zhang, Qi Z
> <qi.z.zhang@intel.com>; stable@dpdk.org; Zhang, Tianfei
> <tianfei.zhang@intel.com>; Yigit, Ferruh <ferruh.yigit@intel.com>
> Subject: Re: [PATCH v3] raw/ifpga: fix pthread cannot join
> 
> On Thu, 20 Jan 2022 00:58:14 -0500
> Wei Huang <wei.huang@intel.com> wrote:
> 
> > diff --git a/drivers/raw/ifpga/ifpga_rawdev.c
> > b/drivers/raw/ifpga/ifpga_rawdev.c
> > index 8d9db58..2bc569b 100644
> > --- a/drivers/raw/ifpga/ifpga_rawdev.c
> > +++ b/drivers/raw/ifpga/ifpga_rawdev.c
> > @@ -497,7 +497,7 @@ static int set_surprise_link_check_aer(
> >  	int gsd_enable, ret;
> >  #define MS 1000
> >
> > -	while (1) {
> > +	while (ifpga_monitor_start) {
> 
> This will work on x86 CPU but not on CPU's with weak memory ordering like
> ARM.
> 
> You need to use atomic_builtin when accessing a flag shared between threads
> that is not being protected by a lock.

We submitted a v5 patch for your suggestion, but got a " coding style issues" which said that: 
Warning in drivers/raw/ifpga/ifpga_rawdev.c:
Using rte_atomicNN_xxx

We look at the devtools/checkpatches.sh script, it mentioned that rte_atomic APIs are forbidden.

ifpga_monitor_start just a flags to enable and disable the threads not share with threads. When the driver init it will start the thread, 
when the driver remove it want to stop the thread. I don't think it will has a concurrency issue on this scenario.