exanic ptp issue on heavily loaded system

[AlexeySaff]

I believe that on a heavily loaded system where housekeeping kernel can't respond in a time this function provide an error

`static enum hrtimer_restart exanic_ptp_soft_hrtimer_callback(
struct hrtimer *timer)
{
struct exanic *exanic =
container_of(timer, struct exanic, ptp_clock_hrtimer);
uint32_t hw_time =
readl(exanic->regs_virt + REG_EXANIC_OFFSET(REG_EXANIC_HW_TIME));
uint64_t time_ticks = exanic_ptp_soft_extend_hw_time(exanic, hw_time);
int32_t error;

if ((exanic->tick_rollover_counter & 1) == 0)
    /* Timer should have fired near halfway point */
    error = hw_time - 0x80000000;
else
    /* Timer should have fired around rollover time */
    error = hw_time;

if (error < -MAX_ERROR_TICKS || error > MAX_ERROR_TICKS)
    dev_err(exanic_dev(exanic),
        "Rollover timer fired at an unexpected time: "
        "counter 0x%llx hwtime 0x%08x\n", exanic->tick_rollover_counter,
        hw_time);

exanic->tick_rollover_counter++;
exanic_ptp_update_info_page(exanic);

hrtimer_forward_now(&exanic->ptp_clock_hrtimer,
        next_rollover_update_time(exanic, time_ticks));

return HRTIMER_RESTART;

}`

cause hrtimer callback is executed in softirq context. Would it be possible somehow to give more priority to that check, as it's negatively affecting the whole PTP process and measurements when an error happens?

[2208433.063915] exanic 0000:01:00.0: Rollover timer fired at an unexpected time: counter 0xc584802 hwtime 0xffffffff
[2208433.064020] exanic 0000:01:00.0: Rollover timer fired at an unexpected time: counter 0xc584803 hwtime 0x809d611a

[AlexeySaff]

so basically the problem only when we read edge condition hwtime 0xffffffff - that needs to be checked

[bornlibra23]

Hi @AlexeySaff I am a noob with this aspect. Are you firing timers here? Who registers & who calls here? Perhaps you can point me to he relevant sections in the manual.

I am interested because I am sending orders at market open & would like to have a more accurate timing.