RE: Large gpio interrupt latency

["François Legal" <devel@thom.fr.eu.org>]

Le Mercredi, Juin 16, 2021 11:05 CEST, "Chen, Hongzhan" <hongzhan.chen@intel.com> a écrit:

>
>
> >-----Original Message-----
> >From: François Legal <devel@thom.fr.eu.org>
> >Sent: Wednesday, June 16, 2021 4:19 PM
> >To: Chen, Hongzhan <hongzhan.chen@intel.com>
> >Cc: xenomai@xenomai.org
> >Subject: RE: Large gpio interrupt latency
> >
> >Le Mercredi, Juin 16, 2021 10:10 CEST, "Chen, Hongzhan" <hongzhan.chen@intel.com> a écrit:
> >
> >>
> >> >
> >> >-----Original Message-----
> >> >From: Xenomai <xenomai-bounces@xenomai.org> On Behalf Of François Legal via Xenomai
> >> >Sent: Wednesday, June 16, 2021 3:16 PM
> >> >To: xenomai@xenomai.org
> >> >Subject: Large gpio interrupt latency
> >> >
> >> >Hello,
> >> >
> >> >working on a realtime data recorder (for which I submitted a patch to add timestamp retrieval for net packets), I experience a strange latency problem with taking GPIO interrupts.
> >> >
> >> >So basically my app monitors network packets (which as received by the DUT would trigger toggling of a GPIO on the DUT) and one GPIO.
> >> >We know (from some ohter reference instrument) that the DUT would toggle the GPIO 1-2ms after receiving a specific network packet.
> >> >
> >> >My recorder app relies on the timestamping of events done in interrupts service routines for GPIOs and network interface. By checking the timestamps returned by the application, I get an about 50ms delay between the network packet and the GPIO (so something between 51 and 52ms delay).
> >> >
> >>
> >> Are you toggling gpio and access gpio device  through your rtdm device like on path /dev/rtdm/your_gpiocontroller/gpio*?
> >>
> >
> >Maybe my setup was not very clear.
> >I've got a DUT that receives network data, and toggles one of its GPIOs depending one the network data.
> >
> >I've got another device running my realtime recording app, that receives the same network data as the DUT, and that has one of its GPIO connected to the DUT GPIO.
> >
> >On the recording app, I use the RTDM device to open, ioctl (enable interrupt + timestamp), then select & read the GPIO value.
>
> When issue happen, the recording app side have got same number of network data packages and gpio interrupts with that DUT have been toggling?
> I am asking this because I met gpio hardware issue that would cause gpio interrupt missing or invalid gpio interrupts when gpio is connecting on two different boards.
>

AFAICT, I'm not missing GPIO edges/interrupts in the app. Regarding network, the DUT and the recoding device are connected to the same switch with the same port config.

Moreover, I tried changing the interrupt polarity of the recording device (switched from rising edge to falling edge), and the result is 1-2ms + ~30ms (the width of the pulse of the DUT) + 50ms latency

> >
> >François
> >
> >
> >> Hongzhan Chen
> >>
> >> >I wonder where that might be coming from.
> >> >
> >> >Any idea ?
> >> >
> >> >thanks in advance
> >> >
> >> >François
> >
> >
> >