[dpdk-users] L-thread subsystem: rte_eal_remote_launch() does not work

ikuzar RABE ikuzar9295 at gmail.com
Wed Apr 17 16:12:41 CEST 2019


I found what was wrong.

lthread_create(&lt, -1, initial_lthread, (void *) NULL); is not run because
the number of launched schedulers (the number of times lthread_run( ) is
called) set in function lthread_num_schedulers_set(num_sched) is not
reached.
At first, I set *num_sched *to the number of enabled lcores which is
greater than 2. Now I call lthread_num_schedulers_set(2), I see hello from
2 lcores (lcore 5 and lcore 6) and the corresponding lthreads as expected.

To summarize, if I want to see hello from lcore 5 and lcore 6 and their
corresponding lthreads, using rte_eal_remote_launch(), I must write:

lthread_num_schedulers_set(2);
rte_eal_remote_launch(lthread_scheduler, (void *)NULL, 5);
rte_eal_remote_launch(lthread_scheduler, (void *)NULL, 6);

Regards,
ikuzar


Le mar. 16 avr. 2019 à 16:05, ikuzar RABE <ikuzar9295 at gmail.com> a écrit :

> Hi all,
>
> I am experimenting lightweigth thread (lthread) over lcore. I use lcores
> 0, 5 and 6.
> I run this example:
> https://doc.dpdk.org/api/examples_2performance-thread_2pthread_shim_2main_8c-example.html and
> the result is OK.
> The result is Ok => I saw Hello from every lthread / lcore
>
> Now, I want to replace rte_eal_mp_remote_launch(lthread_scheduler, (void
> *)NULL, CALL_MASTER); by rte_eal_remote_launch(lthread_scheduler, (void
> *)NULL, 5); and rte_eal_remote_launch(lthread_scheduler, (void *)NULL, 6);
> But I do not see any Hello from lthread / lcore.
> With gdb I enter into lthread_scheduler() but lthread_create(&lt, -1,
> initial_lthread, (void *) NULL); seems to be not run. I put a breakpoint
> inside initial_lthread() but the execution does not stop on the breakpoint.
>
> My configuration:
> - Linux Debian 8
> - dpdk v18.05.1
> - NUMA node0 CPU(s):     0-11,24-35
> - NUMA node1 CPU(s):     12-23,36-47
>
> Would someone know how to run ligthweight threads with
> rte_eal_remote_launch( ) ?
>
> Regards,
> ikuzar
>


More information about the users mailing list