cpu/esp: migrate to ztimer

[fjmolinas]

Contribution description

esp has a couple of uses of xtimer that can't be directly translated to ztimer, so it seems better to do it in its own PR. I also don't have hardware to tests so would appreciate if maybe @gschorcht could complement murdock's testing.

The most delicate change is https://github.com/fjmolinas/RIOT/blob/6a736e4cdfb57ee446050d8fd0c455f120dd381c/cpu/esp32/esp_ztimer.c#L153-L157 since this is not a direct translation, so depending on how sensitive this value is some adjustment might be needed, thougts @gschorcht?

Testing procedure

• All esp tests should pass
• Murdock tests should pass

Issues/PRs references

depends on #17385
part of #16903

[fjmolinas]

I'll run tests to see if somethin breaks....

[kaspar030]

needs rebase

[gschorcht]

@fjmolinas I will take a look later today.

[gschorcht]

Sorry, when I pushed you to change everything back to ztimer_usec, but I was a bit confused since you added it unconditionally here

RIOT/cpu/esp32/Makefile.dep

Line 9 in 8f5363d

USEMODULE += ztimer_usec

So I though that xtimer was enabled herd by default also before your changes and usec resolution is required. However, this wasn't the case.

The only place where it was enabled before and used usec resolution was here.

RIOT/cpu/esp_common/Makefile.dep

Line 71 in ab0e118

ifneq (,$(filter esp_freertos,$(USEMODULE)))

RIOT/cpu/esp_common/freertos/task.c

Lines 133 to 140 in e20f3e5

	void vTaskDelay( const TickType_t xTicksToDelay )
	{
	DEBUG("%s xTicksToDelay=%d\n", __func__, xTicksToDelay);
	#if defined(MODULE_ESP_WIFI_ANY)
	uint64_t us = xTicksToDelay * MHZ / xPortGetTickRateHz();
	xtimer_usleep(us);
	#endif
	}

But after looking into the code, I'm really sure that this could also be changed to msec. A tick in FreeRTOS is 10 or 20 ms.

RIOT/cpu/esp_common/freertos/portable.c

Lines 24 to 26 in e20f3e5

	uint32_t xPortGetTickRateHz(void) {
	return MSEC_PER_SEC / portTICK_PERIOD_MS;
	}

So we could change all timers to msec. Sorry for this inconvenience, I was just confused by the unconditional ztimer_usec added by this PR.

[gschorcht]

Since you touch this file, please correct

RIOT/cpu/esp_common/esp-now/esp_now_netdev.c

Line 120 in c706a6f

DEBUG("wifi_scan_get_ap_num ret=%d num=%d\n", ret ,ap_num);

-     DEBUG("wifi_scan_get_ap_num ret=%d num=%d\n", ret ,ap_num);
+     DEBUG("wifi_scan_get_ap_num ret=%d num=%d\n", ret, ap_num);

[gschorcht]

@fjmolinas It seems that we have a dependencies problem in Kconfig that wasn't catched by CI compilation when #17232 was merged. Let me try to fix it first. You can observe it when you compare the modules list with commands:

make BOARD=esp32-olimex-evb -C examples/gnrc_networking info-modules
TEST_KCONFIG=1 make BOARD=esp32-olimex-evb -C examples/gnrc_networking info-modules

Without TEST_KCONFIG=1 the following modules are enabled because the board has esp_eth as netdev default.

esp_eth
esp_idf_eth
esp_idf_eth_phy

where esp_eth then enables xtimer automatically.

With TEST_KCONFIG=1 these modules including xtimer are not enabled.

[gschorcht]

It seems that we have a dependencies problem in Kconfig

This is because esp_wifi*, esp_eth dependencies were not modeled at all til now.

[fjmolinas]

@gschorcht I conditionally included the ztimer file I hope you don't mind doing this instead, it's a bit more explicit regarding dependencies IMO.

[fjmolinas]

@gschorcht I conditionally included the ztimer file I hope you don't mind doing this instead, it's a bit more explicit regarding dependencies IMO.

But I can also revert if you prefer, I also realize this is turning into one big blob commit with many chances, I can split them up when squashing if you prefer.

[gschorcht]

You can squash your last changes directly to have a single commit.

[gschorcht]

Looks good, lets see what Murdock says.

[fjmolinas]

Yeiii! All green!

[kaspar030]

Nice one!

[fjmolinas]

Thanks for the review and suggestions @gschorcht!

[gschorcht]

@fjmolinas Great work. Thanks.

[gschorcht]

@fjmolinas Unfortunatly, it seems that this PR has side effects on esp_wifi for ESP8266. When I tested some of my changes in esp_wifi today I had to figure out, that the WiFi interface with WPA2 authentication doesn't work anymore since this PR. When we changed the timer from us to msec, I tested it only for ESP32 😟 I will try to figure out, what the problem is.

[gschorcht]

This would have to be:

-     *ztimer/*(.literal .text .literal.* .text.*)
+     *ztimer_core/*(.literal .text .literal.* .text.*)

When I was reviewing it, I didn't know that objects are not stored in ztimer but in ztimer_core although the module is called ztimer. But this alone doesn't solve the problem, unfortunatly 😟

[gschorcht]

@fjmolinas PR #17427 solves the problem with ztimer and ESP WiFi on ESP8266.