[dpdk-dev] Segm fault to the update acl context

[dpdk-dev] Segm fault to the update acl context

[Victor Detoni]

hello,

I'm running l3fwd-acl example as a daemon and trying to update acl context
with as a "swap" acl context, but the program stops with a segment fault
statement. Even using variables to direct which acl context the main loop
has to use, the program breaks.

I have noticed that even when I update only the swap acl context, the acl
structures are updated and the segment fault also happens.

please, someone knows why this segment fault?

Thanks
Victor

Re: [dpdk-dev] Segm fault to the update acl context

[Ananyev, Konstantin]

Hi Victor,

> hello,
> 
> I'm running l3fwd-acl example as a daemon and trying to update acl context
> with as a "swap" acl context, but the program stops with a segment fault
> statement. Even using variables to direct which acl context the main loop
> has to use, the program breaks.
> 
> I have noticed that even when I update only the swap acl context, the acl
> structures are updated and the segment fault also happens.
> 
> please, someone knows why this segment fault?

As I remembers, l3fwd-acl doesn't support updating ACL ctx on the fly.
Could you explain a bit more what are you trying to do, or even better
provide some code example to reproduce the issue?

Konstantin

Re: [dpdk-dev] Segm fault to the update acl context

[Victor Detoni]

Hi Konstantin,

Thank you for your questions. Please find below a sample of my code:

After the program has started, I send sighup signals to update the acl
context, like this:

if (signup == SIGHUP)
  use_acl_ctx_swap = 1;
...
During the SIGHUP processes I create a new/swap acl ctx and data structures
using the app_acl_init, add_rules and setup_acl functions
...

main_loop()
...
if (use_acl_ctx_swap) {
   if (acl_search.num_ipv4) {
      rte_acl_classify(
         acl_config.swap_acx_ipv4[socketid], //use swap acl ctx
         acl_search.data_ipv4,
         acl_search.res_ipv4,
         acl_search.num_ipv4,
         DEFAULT_MAX_CATEGORIES);

         send_packets(acl_search.m_ipv4,
            acl_search.res_ipv4,
            acl_search.num_ipv4);

} else {
   if (acl_search.num_ipv4) {
      rte_acl_classify(
         acl_config.acx_ipv4[socketid],
         acl_search.data_ipv4,
         acl_search.res_ipv4,
         acl_search.num_ipv4,
         DEFAULT_MAX_CATEGORIES);

         send_packets(acl_search.m_ipv4,
            acl_search.res_ipv4,
            acl_search.num_ipv4);
   }
}

When I update the acl ctx, rarely the program doesn't crash, mainly when I
just increase de rules. Is there a another way to add new/remove rules in
the acl ctx instead of sighup or control variables?

thanks
Victor

On Mon, Mar 27, 2017 at 6:23 AM, Ananyev, Konstantin <
konstantin.ananyev@intel.com> wrote:

> Hi Victor,
>
> > hello,
> >
> > I'm running l3fwd-acl example as a daemon and trying to update acl
> context
> > with as a "swap" acl context, but the program stops with a segment fault
> > statement. Even using variables to direct which acl context the main loop
> > has to use, the program breaks.
> >
> > I have noticed that even when I update only the swap acl context, the acl
> > structures are updated and the segment fault also happens.
> >
> > please, someone knows why this segment fault?
>
> As I remembers, l3fwd-acl doesn't support updating ACL ctx on the fly.
> Could you explain a bit more what are you trying to do, or even better
> provide some code example to reproduce the issue?
>
> Konstantin
>
>

Re: [dpdk-dev] Segm fault to the update acl context

[Ananyev, Konstantin]

> 
> Hi Konstantin,
> 
> Thank you for your questions. Please find below a sample of my code:
> 
> After the program has started, I send sighup signals to update the acl context, like this:
> 
> if (signup == SIGHUP)
>   use_acl_ctx_swap = 1;
> ...
> During the SIGHUP processes I create a new/swap acl ctx and data structures using the app_acl_init, add_rules and setup_acl
> functions

Ok, so you do that inside your signal handler, correct?
Do you synchronize this context updates with your data-processing threads somehow?
Konstantin

> ...
> 
> main_loop()
> ...
> if (use_acl_ctx_swap) {
>    if (acl_search.num_ipv4) {
>       rte_acl_classify(
>          acl_config.swap_acx_ipv4[socketid], //use swap acl ctx
>          acl_search.data_ipv4,
>          acl_search.res_ipv4,
>          acl_search.num_ipv4,
>          DEFAULT_MAX_CATEGORIES);
> 
>          send_packets(acl_search.m_ipv4,
>             acl_search.res_ipv4,
>             acl_search.num_ipv4);
> 
> } else {
>    if (acl_search.num_ipv4) {
>       rte_acl_classify(
>          acl_config.acx_ipv4[socketid],
>          acl_search.data_ipv4,
>          acl_search.res_ipv4,
>          acl_search.num_ipv4,
>          DEFAULT_MAX_CATEGORIES);
> 
>          send_packets(acl_search.m_ipv4,
>             acl_search.res_ipv4,
>             acl_search.num_ipv4);
>    }
> }
> 
> When I update the acl ctx, rarely the program doesn't crash, mainly when I just increase de rules. Is there a another way to add
> new/remove rules in the acl ctx instead of sighup or control variables?
> 
> thanks
> Victor
> 
> On Mon, Mar 27, 2017 at 6:23 AM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
> Hi Victor,
> 
> > hello,
> >
> > I'm running l3fwd-acl example as a daemon and trying to update acl context
> > with as a "swap" acl context, but the program stops with a segment fault
> > statement. Even using variables to direct which acl context the main loop
> > has to use, the program breaks.
> >
> > I have noticed that even when I update only the swap acl context, the acl
> > structures are updated and the segment fault also happens.
> >
> > please, someone knows why this segment fault?
> 
> As I remembers, l3fwd-acl doesn't support updating ACL ctx on the fly.
> Could you explain a bit more what are you trying to do, or even better
> provide some code example to reproduce the issue?
> 
> Konstantin

Re: [dpdk-dev] Segm fault to the update acl context

[Victor Detoni]

Hi,

I have tried to use rcu and rte_lock libs to synchronize but I put the
ctx_reload before rte_acl_classify function and it is working fine.

My code is working like this and I'm not having latency during reload
process.

while() {
...
if (acl_ctx_reload)
   do_acl_reload();

   if (acl_search.num_ipv4) {
      rte_acl_classify(
         acl_config.acx_ipv4[socketid],
         acl_search.data_ipv4,
         acl_search.res_ipv4,
         acl_search.num_ipv4,
         DEFAULT_MAX_CATEGORIES);

         send_packets(acl_search.m_ipv4,
            acl_search.res_ipv4,
            acl_search.num_ipv4);
   }
}

Thank you for your support.

On Mon, Mar 27, 2017 at 8:34 PM, Ananyev, Konstantin <
konstantin.ananyev@intel.com> wrote:

>
>
> >
> > Hi Konstantin,
> >
> > Thank you for your questions. Please find below a sample of my code:
> >
> > After the program has started, I send sighup signals to update the acl
> context, like this:
> >
> > if (signup == SIGHUP)
> >   use_acl_ctx_swap = 1;
> > ...
> > During the SIGHUP processes I create a new/swap acl ctx and data
> structures using the app_acl_init, add_rules and setup_acl
> > functions
>
> Ok, so you do that inside your signal handler, correct?
> Do you synchronize this context updates with your data-processing threads
> somehow?
> Konstantin
>
> > ...
> >
> > main_loop()
> > ...
> > if (use_acl_ctx_swap) {
> >    if (acl_search.num_ipv4) {
> >       rte_acl_classify(
> >          acl_config.swap_acx_ipv4[socketid], //use swap acl ctx
> >          acl_search.data_ipv4,
> >          acl_search.res_ipv4,
> >          acl_search.num_ipv4,
> >          DEFAULT_MAX_CATEGORIES);
> >
> >          send_packets(acl_search.m_ipv4,
> >             acl_search.res_ipv4,
> >             acl_search.num_ipv4);
> >
> > } else {
> >    if (acl_search.num_ipv4) {
> >       rte_acl_classify(
> >          acl_config.acx_ipv4[socketid],
> >          acl_search.data_ipv4,
> >          acl_search.res_ipv4,
> >          acl_search.num_ipv4,
> >          DEFAULT_MAX_CATEGORIES);
> >
> >          send_packets(acl_search.m_ipv4,
> >             acl_search.res_ipv4,
> >             acl_search.num_ipv4);
> >    }
> > }
> >
> > When I update the acl ctx, rarely the program doesn't crash, mainly when
> I just increase de rules. Is there a another way to add
> > new/remove rules in the acl ctx instead of sighup or control variables?
> >
> > thanks
> > Victor
> >
> > On Mon, Mar 27, 2017 at 6:23 AM, Ananyev, Konstantin <
> konstantin.ananyev@intel.com> wrote:
> > Hi Victor,
> >
> > > hello,
> > >
> > > I'm running l3fwd-acl example as a daemon and trying to update acl
> context
> > > with as a "swap" acl context, but the program stops with a segment
> fault
> > > statement. Even using variables to direct which acl context the main
> loop
> > > has to use, the program breaks.
> > >
> > > I have noticed that even when I update only the swap acl context, the
> acl
> > > structures are updated and the segment fault also happens.
> > >
> > > please, someone knows why this segment fault?
> >
> > As I remembers, l3fwd-acl doesn't support updating ACL ctx on the fly.
> > Could you explain a bit more what are you trying to do, or even better
> > provide some code example to reproduce the issue?
> >
> > Konstantin
>
>