Hello, I'm trying to launch a thread - lcore_mainloop( from examples/timer/main.c ) that runs rte_manage_timer() every 2s from testpmd to ensure the timers i've registered in my driver are checked for expiry ( i even tried putting this thread in my driver as well, no difference in results) and i see that while this thread is running, i somehow seem to stop getting interrupts ..infact i don't even see eal_intr_process_interrupts () being called. diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c index ca4e1a4..a8d71d6 100644 --- a/app/test-pmd/testpmd.c +++ b/app/test-pmd/testpmd.c @@ -71,6 +71,8 @@ #include <rte_pci.h> #include <rte_ether.h> #include <rte_ethdev.h> +#include <rte_cycles.h> +#include <rte_timer.h> #include <rte_dev.h> #include <rte_string_fns.h> #ifdef RTE_LIBRTE_IXGBE_PMD @@ -2524,6 +2526,30 @@ signal_handler(int signum) } } +static int +lcore_mainloop(__attribute__((unused)) void *arg) +{ + uint64_t prev_tsc = 0, cur_tsc, diff_tsc; + unsigned int lcore_id; + + lcore_id = rte_lcore_id(); + printf("Starting mainloop on core %u\n", lcore_id); + + while (f_quit == 0) { + cur_tsc = rte_rdtsc(); + diff_tsc = cur_tsc - prev_tsc; + /* Schedule every 2 seconds */ + if (diff_tsc > rte_get_timer_hz() * 2) { + rte_timer_manage(); + prev_tsc = cur_tsc; + } else + sleep(1); + } + return 0; +} + int main(int argc, char** argv) { @@ -2627,6 +2653,7 @@ main(int argc, char** argv) if (strlen(cmdline_filename) != 0) cmdline_read_from_file(cmdline_filename); + rte_eal_remote_launch(lcore_mainloop, NULL, 3); if (interactive == 1) { if (auto_start) { printf("Start automatic packet forwarding\n"); My testpmd cmdline is like so: testpmd -c 0xff -n 3 -- -i portmask=0x3 --nb-cores=3 --rxq=1 --txq=1 Any idea what could be the problem ? Is this something that is expected or am i doing something wrong ? Thanks Som
On Fri, Nov 2, 2018 at 9:30 AM Somnath Kotur <somnath.kotur@broadcom.com>
wrote:
> Hello,
> I'm trying to launch a thread - lcore_mainloop( from
> examples/timer/main.c ) that runs rte_manage_timer() every 2s from testpmd
> to ensure the timers i've registered in my driver are checked for expiry (
> i even tried putting this thread in my driver as well, no difference in
> results) and i see that while this thread is running, i somehow seem to
> stop getting interrupts ..infact i don't even
> see eal_intr_process_interrupts () being called.
>
>
> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> index ca4e1a4..a8d71d6 100644
> --- a/app/test-pmd/testpmd.c
> +++ b/app/test-pmd/testpmd.c
> @@ -71,6 +71,8 @@
> #include <rte_pci.h>
> #include <rte_ether.h>
> #include <rte_ethdev.h>
> +#include <rte_cycles.h>
> +#include <rte_timer.h>
> #include <rte_dev.h>
> #include <rte_string_fns.h>
> #ifdef RTE_LIBRTE_IXGBE_PMD
> @@ -2524,6 +2526,30 @@ signal_handler(int signum)
> }
> }
>
> +static int
> +lcore_mainloop(__attribute__((unused)) void *arg)
> +{
> + uint64_t prev_tsc = 0, cur_tsc, diff_tsc;
> + unsigned int lcore_id;
> +
> + lcore_id = rte_lcore_id();
> + printf("Starting mainloop on core %u\n", lcore_id);
> +
> + while (f_quit == 0) {
> + cur_tsc = rte_rdtsc();
> + diff_tsc = cur_tsc - prev_tsc;
> + /* Schedule every 2 seconds */
> + if (diff_tsc > rte_get_timer_hz() * 2) {
> + rte_timer_manage();
> + prev_tsc = cur_tsc;
> + } else
> + sleep(1);
> + }
> + return 0;
> +}
> +
> int
> main(int argc, char** argv)
> {
> @@ -2627,6 +2653,7 @@ main(int argc, char** argv)
> if (strlen(cmdline_filename) != 0)
> cmdline_read_from_file(cmdline_filename);
>
> + rte_eal_remote_launch(lcore_mainloop, NULL, 3);
> if (interactive == 1) {
> if (auto_start) {
> printf("Start automatic packet forwarding\n");
>
>
> My testpmd cmdline is like so:
>
> testpmd -c 0xff -n 3 -- -i portmask=0x3 --nb-cores=3 --rxq=1 --txq=1
>
> Any idea what could be the problem ? Is this something that is expected or
> am i doing something wrong ?
>
> Thanks
> Som
>
On 02-Nov-18 4:00 AM, Somnath Kotur wrote:
> Hello,
> I'm trying to launch a thread - lcore_mainloop( from
> examples/timer/main.c ) that runs rte_manage_timer() every 2s from testpmd
> to ensure the timers i've registered in my driver are checked for expiry (
> i even tried putting this thread in my driver as well, no difference in
> results) and i see that while this thread is running, i somehow seem to
> stop getting interrupts ..infact i don't even
> see eal_intr_process_interrupts () being called.
>
>
> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> index ca4e1a4..a8d71d6 100644
> --- a/app/test-pmd/testpmd.c
> +++ b/app/test-pmd/testpmd.c
> @@ -71,6 +71,8 @@
> #include <rte_pci.h>
> #include <rte_ether.h>
> #include <rte_ethdev.h>
> +#include <rte_cycles.h>
> +#include <rte_timer.h>
> #include <rte_dev.h>
> #include <rte_string_fns.h>
> #ifdef RTE_LIBRTE_IXGBE_PMD
> @@ -2524,6 +2526,30 @@ signal_handler(int signum)
> }
> }
>
> +static int
> +lcore_mainloop(__attribute__((unused)) void *arg)
> +{
> + uint64_t prev_tsc = 0, cur_tsc, diff_tsc;
> + unsigned int lcore_id;
> +
> + lcore_id = rte_lcore_id();
> + printf("Starting mainloop on core %u\n", lcore_id);
> +
> + while (f_quit == 0) {
> + cur_tsc = rte_rdtsc();
> + diff_tsc = cur_tsc - prev_tsc;
> + /* Schedule every 2 seconds */
> + if (diff_tsc > rte_get_timer_hz() * 2) {
> + rte_timer_manage();
> + prev_tsc = cur_tsc;
> + } else
> + sleep(1);
> + }
> + return 0;
> +}
> +
> int
> main(int argc, char** argv)
> {
> @@ -2627,6 +2653,7 @@ main(int argc, char** argv)
> if (strlen(cmdline_filename) != 0)
> cmdline_read_from_file(cmdline_filename);
>
> + rte_eal_remote_launch(lcore_mainloop, NULL, 3);
> if (interactive == 1) {
> if (auto_start) {
> printf("Start automatic packet forwarding\n");
>
>
> My testpmd cmdline is like so:
>
> testpmd -c 0xff -n 3 -- -i portmask=0x3 --nb-cores=3 --rxq=1 --txq=1
>
> Any idea what could be the problem ? Is this something that is expected or
> am i doing something wrong ?
>
> Thanks
> Som
>
I may be completely off mark here, but as far as i understand, the EAL
Alarm API uses the interrupt thread. The rte_timer API is a high
performance timer API and is meant to be managed manually, by
periodically[1] calling rte_timer_manage(). If you want something to be
called every two seconds, just set up an rte_alarm - there's no need to
use the timer API (unless you are on FreeBSD, where alarm API is not
officially supported).
[1] as in, more frequently than every two seconds if you want to have
any semblance of timer precision!
--
Thanks,
Anatoly
> On Nov 2, 2018, at 6:31 AM, Burakov, Anatoly <anatoly.burakov@intel.com> wrote: > > On 02-Nov-18 4:00 AM, Somnath Kotur wrote: >> Hello, >> I'm trying to launch a thread - lcore_mainloop( from >> examples/timer/main.c ) that runs rte_manage_timer() every 2s from testpmd >> to ensure the timers i've registered in my driver are checked for expiry ( >> i even tried putting this thread in my driver as well, no difference in >> results) and i see that while this thread is running, i somehow seem to >> stop getting interrupts ..infact i don't even >> see eal_intr_process_interrupts () being called. >> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c >> index ca4e1a4..a8d71d6 100644 >> --- a/app/test-pmd/testpmd.c >> +++ b/app/test-pmd/testpmd.c >> @@ -71,6 +71,8 @@ >> #include <rte_pci.h> >> #include <rte_ether.h> >> #include <rte_ethdev.h> >> +#include <rte_cycles.h> >> +#include <rte_timer.h> >> #include <rte_dev.h> >> #include <rte_string_fns.h> >> #ifdef RTE_LIBRTE_IXGBE_PMD >> @@ -2524,6 +2526,30 @@ signal_handler(int signum) >> } >> } >> +static int >> +lcore_mainloop(__attribute__((unused)) void *arg) >> +{ >> + uint64_t prev_tsc = 0, cur_tsc, diff_tsc; >> + unsigned int lcore_id; >> + >> + lcore_id = rte_lcore_id(); >> + printf("Starting mainloop on core %u\n", lcore_id); >> + >> + while (f_quit == 0) { >> + cur_tsc = rte_rdtsc(); >> + diff_tsc = cur_tsc - prev_tsc; >> + /* Schedule every 2 seconds */ >> + if (diff_tsc > rte_get_timer_hz() * 2) { >> + rte_timer_manage(); >> + prev_tsc = cur_tsc; >> + } else >> + sleep(1); >> + } >> + return 0; >> +} >> + >> int >> main(int argc, char** argv) >> { >> @@ -2627,6 +2653,7 @@ main(int argc, char** argv) >> if (strlen(cmdline_filename) != 0) >> cmdline_read_from_file(cmdline_filename); >> + rte_eal_remote_launch(lcore_mainloop, NULL, 3); >> if (interactive == 1) { >> if (auto_start) { >> printf("Start automatic packet forwarding\n"); >> My testpmd cmdline is like so: >> testpmd -c 0xff -n 3 -- -i portmask=0x3 --nb-cores=3 --rxq=1 --txq=1 >> Any idea what could be the problem ? Is this something that is expected or >> am i doing something wrong ? >> Thanks >> Som > > I may be completely off mark here, but as far as i understand, the EAL Alarm API uses the interrupt thread. The rte_timer API is a high performance timer API and is meant to be managed manually, by periodically[1] calling rte_timer_manage(). If you want something to be called every two seconds, just set up an rte_alarm - there's no need to use the timer API (unless you are on FreeBSD, where alarm API is not officially supported). > > [1] as in, more frequently than every two seconds if you want to have any semblance of timer precision! > Unless we changed it I believe the > -- > Thanks, > Anatoly Regards, Keith
> On Nov 2, 2018, at 9:35 AM, Wiles, Keith <keith.wiles@intel.com> wrote:
>
>
Sorry, meant to hit cancel for my previous email, Anatoly answered it correctly.
Regards,
Keith
Thank you Anatoly and keith for your replies. What i still don't get is why
the two cannot seem to co-exist?
That is , whenever this while loop was running in the EAL thread that i
showed in the code, my driver would stop receiving interrupts/async events
...Do we understand that?
But you are right Anatoly, i need something to be called in my driver at a
cadence of ~2s, I was registering timers in my driver and running
rte_timer_manage() in this seperate thread to check and ensure the
callbacks are executed on expiry.
so if you are saying, i can achieve same functionality using rte_alarm()
while i can get async events in my driver at the same time, i will explore
this option
Thanks a lot
Som
On Fri, Nov 2, 2018 at 8:07 PM Wiles, Keith <keith.wiles@intel.com> wrote:
>
>
> > On Nov 2, 2018, at 9:35 AM, Wiles, Keith <keith.wiles@intel.com> wrote:
> >
> >
> Sorry, meant to hit cancel for my previous email, Anatoly answered it
> correctly.
>
> Regards,
> Keith
>
>
Hi Anatoly/keith,
I just tried using rte_eal_alarm_set() in the driver
like so:
rte_eal_alarm_set(US_PER_S * 2, timer0_cb, (void *)data);
The moment i start this off, i stop getting async events/interrupt
notifications in my driver ...I'm hitting the same problem as it was before
with rte_timer_manage() as well.
What is going wrong here ? Something we are missing ?
Thanks
Som
On Sat, Nov 3, 2018 at 8:06 AM Somnath Kotur <somnath.kotur@broadcom.com>
wrote:
> Thank you Anatoly and keith for your replies. What i still don't get is
> why the two cannot seem to co-exist?
> That is , whenever this while loop was running in the EAL thread that i
> showed in the code, my driver would stop receiving interrupts/async events
> ...Do we understand that?
> But you are right Anatoly, i need something to be called in my driver at a
> cadence of ~2s, I was registering timers in my driver and running
> rte_timer_manage() in this seperate thread to check and ensure the
> callbacks are executed on expiry.
> so if you are saying, i can achieve same functionality using rte_alarm()
> while i can get async events in my driver at the same time, i will explore
> this option
> Thanks a lot
>
> Som
>
> On Fri, Nov 2, 2018 at 8:07 PM Wiles, Keith <keith.wiles@intel.com> wrote:
>
>>
>>
>> > On Nov 2, 2018, at 9:35 AM, Wiles, Keith <keith.wiles@intel.com> wrote:
>> >
>> >
>> Sorry, meant to hit cancel for my previous email, Anatoly answered it
>> correctly.
>>
>> Regards,
>> Keith
>>
>>
Hello Keith/Anatoly,
Another strange thing, it appears that if i keep a
breakpoint (using gdb) on my async intr handler(bnxt_int_handler), then it
seems to hit it (after the alarm callback is invoked first) and then i can
see both this alarm going off every 2s and those async/interrupts arriving
on my driver....
Thanks
Som
On Sat, Nov 3, 2018 at 9:58 AM Somnath Kotur <somnath.kotur@broadcom.com>
wrote:
> Hi Anatoly/keith,
> I just tried using rte_eal_alarm_set() in the driver
> like so:
>
> rte_eal_alarm_set(US_PER_S * 2, timer0_cb, (void *)data);
>
> The moment i start this off, i stop getting async events/interrupt
> notifications in my driver ...I'm hitting the same problem as it was before
> with rte_timer_manage() as well.
> What is going wrong here ? Something we are missing ?
>
> Thanks
> Som
>
> On Sat, Nov 3, 2018 at 8:06 AM Somnath Kotur <somnath.kotur@broadcom.com>
> wrote:
>
>> Thank you Anatoly and keith for your replies. What i still don't get is
>> why the two cannot seem to co-exist?
>> That is , whenever this while loop was running in the EAL thread that i
>> showed in the code, my driver would stop receiving interrupts/async events
>> ...Do we understand that?
>> But you are right Anatoly, i need something to be called in my driver at
>> a cadence of ~2s, I was registering timers in my driver and running
>> rte_timer_manage() in this seperate thread to check and ensure the
>> callbacks are executed on expiry.
>> so if you are saying, i can achieve same functionality using rte_alarm()
>> while i can get async events in my driver at the same time, i will explore
>> this option
>> Thanks a lot
>>
>> Som
>>
>> On Fri, Nov 2, 2018 at 8:07 PM Wiles, Keith <keith.wiles@intel.com>
>> wrote:
>>
>>>
>>>
>>> > On Nov 2, 2018, at 9:35 AM, Wiles, Keith <keith.wiles@intel.com>
>>> wrote:
>>> >
>>> >
>>> Sorry, meant to hit cancel for my previous email, Anatoly answered it
>>> correctly.
>>>
>>> Regards,
>>> Keith
>>>
>>>