how to achieve more granular polling for TI-RTOS?

[keffkeff]

I noticed that when using mongoose with TI-RTOS, mg_mgr_poll() does not work with timeout values more granular than seconds. I later realized that in the case of TI-RTOS, mg_millis() becomes time(NULL), which returns the UNIX time in seconds and not in milliseconds.

So my question is if it's possible to get more precise timeout values?

[keffkeff]

OK so I was able to get a more precise timeout value by using Clock_getTicks() from <ti/sysbios/knl/Clock.h> instead of time(NULL) * 1000. I know that Clock_getTicks() does not return the UNIX time, so if some function depends on the UNIX time being returned from mg_millis(), then that's obviously a problem. And now I see that the SNTP and DNS implementations use it for their packets, so that's probably not good news for my workaround.

Anyway, any advice would be much appreciated.

[scaprile]

#1814

[scaprile]

I'm checking the example code, and as far as I can see that does not use TI RTOS.
The fastest approach is perhaps you defining MG_ENABLE_CUSTOM_MILLIS and providing your own uint64_t mg_millis(void) implementation returning up time in milliseconds; as I think you already did.
It should be fine if you just return up time, ports to some architectures do just that:

mongoose/src/util.c

Lines 90 to 128 in 01390fe

	#if MG_ENABLE_CUSTOM_MILLIS
	#else
	uint64_t mg_millis(void) {
	#if MG_ARCH == MG_ARCH_WIN32
	return GetTickCount();
	#elif MG_ARCH == MG_ARCH_RP2040
	return time_us_64() / 1000;
	#elif MG_ARCH == MG_ARCH_ESP32
	return esp_timer_get_time() / 1000;
	#elif MG_ARCH == MG_ARCH_ESP8266
	return xTaskGetTickCount() * portTICK_PERIOD_MS;
	#elif MG_ARCH == MG_ARCH_FREERTOS_TCP || MG_ARCH == MG_ARCH_FREERTOS_LWIP
	return xTaskGetTickCount() * portTICK_PERIOD_MS;
	#elif MG_ARCH == MG_ARCH_AZURERTOS
	return tx_time_get() * (1000 /* MS per SEC */ / TX_TIMER_TICKS_PER_SECOND);
	#elif MG_ARCH == MG_ARCH_UNIX && defined(__APPLE__)
	// Apple CLOCK_MONOTONIC_RAW is equivalent to CLOCK_BOOTTIME on linux
	// Apple CLOCK_UPTIME_RAW is equivalent to CLOCK_MONOTONIC_RAW on linux
	return clock_gettime_nsec_np(CLOCK_UPTIME_RAW) / 1000000;
	#elif MG_ARCH == MG_ARCH_UNIX
	struct timespec ts = {0, 0};
	// See #1615 - prefer monotonic clock
	#if defined(CLOCK_MONOTONIC_RAW)
	// Raw hardware-based time that is not subject to NTP adjustment
	clock_gettime(CLOCK_MONOTONIC_RAW, &ts);
	#elif defined(CLOCK_MONOTONIC)
	// Affected by the incremental adjustments performed by adjtime and NTP
	clock_gettime(CLOCK_MONOTONIC, &ts);
	#else
	// Affected by discontinuous jumps in the system time and by the incremental
	// adjustments performed by adjtime and NTP
	clock_gettime(CLOCK_REALTIME, &ts);
	#endif
	return ((uint64_t) ts.tv_sec * 1000 + (uint64_t) ts.tv_nsec / 1000000);
	#else
	return (uint64_t) (time(NULL) * 1000);
	#endif
	}
	#endif

[cpq]

I second Sergio's suggestion. Mongoose's software timer implementation uses mg_millis(), which returns a uint64_t value - a system uptime in milliseconds. For those environments where we can't guess a "real" monotonic clock, we just fall back to C time(). Implementing your own mg_millis() is the way to get timers with a millisecond granularity.