cpu/esp32: support for light/deep sleep and pm_layered #13416

gschorcht · 2020-02-20T07:12:56Z

Contribution description

This PR adds support for EPS32's light and deep sleep power modes and make them available by pm_layered.

The power management of ESP32 supports now the following operating modes:

Modem-sleep: default operating mode when the WiFi interface is disabled.
Active mode: default operating mode when the WiFi interface is used
Light-sleep mode: the CPUs including peripherals are stalled, the SRAM is retained
Deep-sleep CPU and the SRAM are powered down, the RTC memory can be retained

Since the peripherals are not working during light-sleep and deep-sleep, the CPU cannot be woken up by internal interrupt sources such as timers. Therefore, RIOT's layered power management can't select them as idle mode. The application has to select them explicitly using the pm_set function.

To exit from light-sleep, the following wake-up sources can be used:

RTC timer (set the RTC timer alarm using rtc_set before calling pm_set)
GPIOs that are configured as input with enabled interrupt
RxD signal of UART0 and/or UART1 (configured with ESP_PM_WUP_UART0 and ESP_PM_WUP_UART1)

Please note Since only level interrupts are supported in light-sleep mode, defined edge interrupts of type GPIO_RISING and GPIO_FALLING are implicitly mapped to GPIO_HIGH and GPIO_LOW, respectively, when entering Light-sleep mode.

To exit from deep-sleep, only RTC resources can be used as wake-up sources:

RTC timer (set the RTC timer alarm using rtc_set before calling pm_set)
RTC GPIOs (configured by ESP_PM_WUP_PINS and ESP_PM_WUP_LEVEL)

In Deep-sleep mode the SRAM is powered down. However, the slow RTC memory can be retained. Therefore, data that must be retained during Deep-sleep and the subsequent system restart have to be stored in the slow RTC memory. For that purpose, use

__attribute__((section(".rtc.bss"))) to place uninitialized data in section .rtc.bss, and
__attribute__((section(".rtc.data"))) to place initialized data in section .rtc.data.

Testing procedure

Use a ESP32 board with a button at GPIO0 (usually default) and flash tests/periph_pm, for example:

CFLAGS='-DESP_PM_WUP_UART0=6 -DESP_PM_WUP_PINS=GPIO0 -DESP_PM_WUP_LEVEL=0' \
BOARD=esp32-wroom-32 make -C tests/periph_pm flash

This configures the RxD signal of UART0 as wake-up source from light-sleep and the button at GPIO0 as wake-up source from light and deep sleep.

Connect the ESP32 board via a USB multimeter and perform the following tests while observing the values on the USB multimeter. (Please note: Although the ESP32 consumes only about 800 uA in light sleep mode and only 10 uA in deep sleep mode, the USB multimeter can measure values between 10 mA and 34 mA. The reason for this is that there are a number of on-board components, such as the USB2UART bridge, that generate a base load. For the test it is sufficient that in light sleep mode approx. 21 mA less is consumed and in deep sleep approx. 22 mA less than in normal operating mode.)

Unblock the light-sleep (1) mode for 5 seconds:
```
unblock_rtc 1 5
```
ESP32 should go into light sleep mode (1) for 5 seconds and then continue working correctly.
Unblock the deep-sleep mode (0) for 15 seconds and afterwards unblock the light-sleep (1) mode:
```
unblock_rtc 0 15
unblock 1
```
Once the light-sleep mode (1) is unblocked by the second command, ESP32 should go into the deep-sleep mode for the remaining time. The system should restart with cause rst:0x5 (DEEPSLEEP_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT) after that.
Set light-sleep mode (1), wait a bit and press Enter at the console
```
set 1
```
The system should return from light-sleep once Enter was pressed.
Set deep-sleep mode (0), wait a bit and press the button at GPIO0.
```
set 0
```
The system should restart with cause rst:0x5 (DEEPSLEEP_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT) once the button was pressed.

Issues/PRs references

Fixes issue #13365
~~PR #13417 should be merged before this PR to ensure consistent documentation.~~
Depends on PR #13655

cpu/esp32/doc.txt

yegorich · 2020-02-20T10:02:22Z

cpu/esp32/include/gpio_arch.h

+#endif /* DOXYGEN */
+
+/**
+ * @brief   Current an output pin can drive in active and sleep modes


Could you somehow rewrite this sentence?

Hm, it says what it means. Any suggestion?

Maybe just "The current an output pin can drive in active and sleep modes"

cpu/esp32/periph/pm.c

cpu/esp32/vendor/esp-idf/include/driver/rtc_io.h

gschorcht · 2020-02-20T11:11:47Z

@yegorich Thanks for the first review. I pushed a fixup with a first round of fixes.

gschorcht · 2020-03-08T13:55:15Z

@leandrolanzieri As already told by email, I would like to get this PR into the next release because it seems to be a useful feature. Is it OK to set the Milestone 2020.04?

leandrolanzieri · 2020-03-09T08:25:27Z

@leandrolanzieri As already told by email, I would like to get this PR into the next release because it seems to be a useful feature. Is it OK to set the Milestone 2020.04?

Sure!

benpicco · 2020-03-17T18:25:32Z

~~Something broke during rebase:~~

make[4]: *** No rule to make target '/home/benpicco/dev/RIOT/cpu/esp32/esp-wifi/esp_wifi_netdev.c', needed by '/home/benpicco/dev/RIOT/examples/gnrc_networking/bin/esp32-wroom-32/esp_wifi/esp_wifi_netdev.o'.  Stop.

~~(When I build examples/gnrc_networking with esp_wifi)~~

nothing a make clean can't fix… sorry for the noise.

btw.: Looks like this PR came just in time for a new watch 😉

benpicco

When I run tests/periph_pm with CFLAGS='-DESP_PM_WUP_UART0=6 -DESP_PM_WUP_PINS=GPIO0 -DESP_PM_WUP_LEVEL=0' and do a unblock_rtc 1 5 I get

2020-03-17 22:04:28,495 # Unblocking power mode 1 for 5 seconds.
2020-03-17 22:04:28,496 #  BTN0 pressed.

That is, the GPIO interrupt is called when the RTC alarm rings.

benpicco · 2020-03-17T20:35:37Z

cpu/esp32/periph/pm.c

+    rtc_get_time(&tm_system);
+    rtc_get_alarm(&tm_alarm);
+
+    time_t t_system = mktime(&tm_system);
+    time_t t_alarm = mktime(&tm_alarm);
+    int _alarm_set = 0;
+
+    if (t_alarm > t_system) {
+        _alarm_set = 1;
+        esp_sleep_enable_timer_wakeup((uint64_t)(t_alarm - t_system) * US_PER_SEC);
    }


I'd suggest moving this to periph/rtc.
We started something similar for the sam0 platform where some hooks need to be executed before going to sleep / after leaving sleep.
Right now this is only implemented for periph/gpio, but there is more to come.

This way you can make use of internal driver structures and don't have to clutter periph/pm.

I will give a try.

benpicco · 2020-03-17T21:07:05Z

cpu/esp32/periph/pm.c

+    static const gpio_t wup_pins[] = { ESP_PM_WUP_PINS };
+
+    uint64_t wup_pin_mask = 0;
+    for (unsigned i = 0; i < ARRAY_SIZE(wup_pins); i++) {
+        wup_pin_mask |= 1ULL << wup_pins[i];
+    }
+#ifdef ESP_PM_WUP_LEVEL
+   esp_sleep_enable_ext1_wakeup(wup_pin_mask, ESP_PM_WUP_LEVEL);
+#else /* ESP_PM_WUP_LEVEL */
+   esp_sleep_enable_ext1_wakeup(wup_pin_mask, ESP_EXT1_WAKEUP_ANY_HIGH);
+#endif /* ESP_PM_WUP_LEVEL */
+#endif /* ESP_PM_WUP_PINS */


Couldn't this be done in gpio_init_int()?
The you could also drop the need of configuring the wake-up GPIOs at compile time.

Couldn't this be done in gpio_init_int()?

You mean for GPIOs (if they are RTC GPIOs) that are configured as GPIO_IN could also be used as wake-up source? That's not a good idea. It must be clear to the user that GPIOs in light/deep sleep mode are handled differently than usual. GPIOs cannot be controlled individually in sleep modes. A wake-up occurs when either all GPIOs become LOW (ESP_EXT1_WAKEUP_ALL_LOW) or any GPIO becomes HIGH (ESP_EXT1_WAKEUP_ANY_HIGH).

For example, if you want to wake up by a LOW active interrupt line, ESP_EXT1_WAKEUP_ALL_LOW would have to be selected as wake-up level. Then you would have to make sure that all other GPIOs configured as GPIO_IN are already LOW. This is hardly controllable. Using an explicit setting of the only one GPIO that is used as wake-up source is much easier in that case.

Unfortunately, the wake-up functions in ESP32 are not so sophisticated. It is something like all or nothing. Multiple GPIOs can only be used as a wake-up source if, for example, all connected interrupt lines were HIGH active, which is normally not the case.

So the actual use case will be to use one GPIO for an external interrupt line to wake up the ESP32 (one example). This can then be LOW or HIGH active.

@benpicco Unfortunately I can neither find the reason nor a solution for the problem that the GPIO interrupt is triggered in addition to the RTC alarm when leaving the sleep mode the first time. (Since it happens only the first time, I didn't realize that during my tests).

However, in a newer version of the SDK I have found a way to use GPIOs configured as input with interrupts directly as wakeup source. After merging the changes from new SDK version in the SDK version used by RIOT, I can use GPIOs directly as interrupt source 😄

Unfortunately only level interrupts and no edge interrupts are supported. What a bummer. 😟 I'm stumped. Ok, we could try to save the interrupt type of all GPIOs and map the edge interrupts to level interrupts before entering sleep mode and change them back when returning from the sleep mode. However, a unique mapping is difficult. Normally you would map GPIO_RISING to GPIO_HIGH and GPIO_FALLING to GPIO_LOW if it's only for waking up. Unfortunately, this does not work with our button in the tests/periph_pm, because GPIO_RISING is used, but the ESP32 button is LOW active, so it generates an interrupt when it is released and the mapping above would lead to GPIO_HIGH as wakeup level.

gschorcht · 2020-03-18T08:06:20Z

btw.: Looks like this PR came just in time for a new watch

Interesting. It uses an external RTC DS3231 because the internal one is not an RTC but only a RTT. BTW, I am thinking about removing the ESP32 RTC implementation and adding an RTT implementation. The RTC could than be realized with your RTT based RTC implementation.

It is a good example how GPIOs are used in deep sleep here. All buttons are pulled down and drawn to HIGH when pressed. The deep sleep configuration would then be

CFLAGS='-DESP_PM_WUP_PINS=GPIO34,GPIO35,GPIO36,GPIO39 -DESP_PM_WUP_LEVEL=1'

Other GPIOs like GPIO4, GPIO16, ... are configured as inputs but not used as wake-up source when in deep sleep.

One further remark. Unfortunately, it will probably never be possible to use BT on the ESP32 in RIOT. The BT stack of the SDK is very complex. It uses the underlying FreeRTOS very intensively and interacts with the hardware at a very low level that we will never be able to realize. Our FreeRTOS adaptation layer will never be able to provide all the required FreeRTOS features, also because FreeRTOS is tick oriented, while RIOT is tickless.

gschorcht · 2020-03-20T12:32:28Z

@benpicco I have pushed some further small changes. The background is that I'm working on changing the ESP8266 port for an SDK version update from 3.1 to 3.3 which will also support WPA2 Enterprise and light/deep sleep. The last changes were mainly changes that will allow to use the same periph_pm code for both ESP SoCs.

Should I squash the last changes to make them better readable?

benpicco

Looks good, tests/periph_pm works as expected and WiFi is stable too.
Just a few comments.

Feel free to squash.

cpu/esp_common/freertos/task.c

cpu/esp32/doc.txt

benpicco · 2020-03-22T16:45:12Z

cpu/esp32/periph/gpio.c

+#ifdef ESP_PM_WUP_LEVEL
+        esp_sleep_enable_ext1_wakeup(wup_pin_mask, ESP_PM_WUP_LEVEL);
+#else /* ESP_PM_WUP_LEVEL */
+        esp_sleep_enable_ext1_wakeup(wup_pin_mask, ESP_EXT1_WAKEUP_ANY_HIGH);
+#endif /* ESP_PM_WUP_LEVEL */


A

#ifndef ESP_PM_WUP_LEVEL #define ESP_PM_WUP_LEVEL ESP_EXT1_WAKEUP_ANY_HIGH #endif

would make this less cluttered.

Should we also use these SDK defined macros in documentation?

I don't really mind, although there is an advantage in spelling it out:

ESP_EXT1_WAKEUP_ALL_LOW: wake up when all selected GPIOs are low

ESP_EXT1_WAKEUP_ANY_HIGH: wake up when any of the selected GPIOs is high

cpu/esp32/periph/gpio.c

The function were used to set the GPIO pads in sleep mode. This function isn't required for deep or light sleep.

`rtc_init` is used after light sleep to update the system time from RTC timer. The fix corrects a small difference of about 230 ms which would sum up with each wakeup from light sleep.

gschorcht · 2020-03-22T23:02:50Z

Commit 04bbca3 refines the documentation a bit.

Commit f744cb0 fixes a small problem in the handling of events

SYSTEM_EVENT_STA_STOP - don't try to reconnect
SYSTEM_EVENT_STA_DISCONNECTED - only try to reset the internal state and reconnect if it was not a disconnect by intention (ASSOC_LEAVE).

The problem occured with commit 8cf397a

> set 1
CPU will enter power mode 1.
WiFi disconnected from ssid BSHS1, reason 8 (ASSOC_LEAVE)
esp_wifi_disconnect failed with return value 12290
esp_wifi_connect failed with return value 12290
The alarm rang
> 
> WiFi connected to ssid BSHS1, channel 8

The multiple disconnects and the failing reconnect are wrong.

cpu/esp32/doc.txt

benpicco

I think we are good here.
All remaining comments are addressed.

The #ifdef in vTaskDelay() is unfortunate, but a #ifdef MODULE_XTIMER could lead to hidden bugs, so this solution is OK.

Please squash.

Now, where the vendor files for light/deep sleep mode are added, function `pm_off` does not need to implement this mode by itself. Instead the existing deep sleep with disabled wakeup sources is used for pm_off.

The WiFi interface should be stopped before reboot or sleep. But stopping the WiFi interface disconnects an existing connection. Usually, esp_wifi_netdev tries to reconnect on an disconnect event. However, trying reconnect with a stopped WiFi interface may lead to a crash. Therefore, the stop event has to be handled.

gschorcht · 2020-03-23T12:41:10Z

@benpicco Squashed. But Murdock shows number of compilation errors that are definitely unrelated to this PR.

compilation of examples/suit_update failed on

lsn50
nrf52840-mdk
nrf52dk:llvm
nucleo-f302r8:gnu
nucleo-f303k8:gnu
nucleo-f303re:gnu
nucleo-l152re:gnu
nucleo-l496zg:gnu
openlabs-kw41z-mini-256kib
particle-boron
phynode-kw41z
saml11-xpro:gnu
samr21-xpro:llvm
sltb001a:gnu
slwstk6000b-slwrb4150a
stk3600

benpicco · 2020-03-23T12:44:49Z

Yea Murdock failed to fetch a pkg from Github - let's hope it's successful this time.

gschorcht requested review from aabadie, fjmolinas, kaspar030 and roberthartung as code owners February 20, 2020 07:12

gschorcht force-pushed the cpu/esp32/pm_layered branch 2 times, most recently from 6581329 to 18644ed Compare February 20, 2020 07:30

gschorcht mentioned this pull request Feb 20, 2020

boards/esp32: remove MCU feature table in board documentation #13417

Merged

gschorcht changed the title ~~cpu/esp32: support for light/deep sleep and pm layered~~ cpu/esp32: support for light/deep sleep and pm_layered Feb 20, 2020

yegorich reviewed Feb 20, 2020

View reviewed changes

cpu/esp32/doc.txt Outdated Show resolved Hide resolved

yegorich reviewed Feb 20, 2020

View reviewed changes

cpu/esp32/doc.txt Outdated Show resolved Hide resolved

yegorich reviewed Feb 20, 2020

View reviewed changes

cpu/esp32/periph/pm.c Outdated Show resolved Hide resolved

yegorich reviewed Feb 20, 2020

View reviewed changes

cpu/esp32/vendor/esp-idf/include/driver/rtc_io.h Show resolved Hide resolved

gschorcht mentioned this pull request Feb 20, 2020

ESP32: deep sleep modes #13365

Closed

benpicco added CI: ready for build If set, CI server will compile all applications for all available boards for the labeled PR and removed CI: ready for build If set, CI server will compile all applications for all available boards for the labeled PR labels Feb 21, 2020

gschorcht force-pushed the cpu/esp32/pm_layered branch from 2bca3f9 to 02ad1e7 Compare February 22, 2020 15:50

gschorcht added the CI: needs squashing Commits in this PR need to be squashed; If set, CI systems will mark this PR as unmergable label Feb 22, 2020

leandrolanzieri added this to the Release 2020.04 milestone Mar 9, 2020

fjmolinas mentioned this pull request Mar 17, 2020

Makefile.dep: filter periph_init modules #13648

Merged

benpicco reviewed Mar 17, 2020

View reviewed changes

gschorcht added the State: WIP State: The PR is still work-in-progress and its code is not in its final presentable form yet label Mar 20, 2020

gschorcht force-pushed the cpu/esp32/pm_layered branch from 3b7c530 to 0ccf4d3 Compare March 20, 2020 12:25

gschorcht removed the State: WIP State: The PR is still work-in-progress and its code is not in its final presentable form yet label Mar 20, 2020

benpicco reviewed Mar 22, 2020

View reviewed changes

cpu/esp32/periph/gpio.c Show resolved Hide resolved

gschorcht added 2 commits March 22, 2020 23:53

cpu/esp32: remove gpio_config_sleep_mode func

825b01e

The function were used to set the GPIO pads in sleep mode. This function isn't required for deep or light sleep.

cpu/esp32: small fix of rtc_init for light sleep

2cbf90d

`rtc_init` is used after light sleep to update the system time from RTC timer. The fix corrects a small difference of about 230 ms which would sum up with each wakeup from light sleep.

benpicco reviewed Mar 22, 2020

View reviewed changes

cpu/esp32/doc.txt Show resolved Hide resolved

gschorcht force-pushed the cpu/esp32/pm_layered branch from f744cb0 to 3eb9062 Compare March 23, 2020 09:47

benpicco approved these changes Mar 23, 2020

View reviewed changes

gschorcht added 9 commits March 23, 2020 12:25

cpu/esp32: add vendor files for light/deep sleep

3faf99a

cpu/esp32: use hibernate mode for pm_off

15ea6fe

Now, where the vendor files for light/deep sleep mode are added, function `pm_off` does not need to implement this mode by itself. Instead the existing deep sleep with disabled wakeup sources is used for pm_off.

cpu/esp32: function to set gpio drive strength

a7652df

cpu/esp32: several small doc fixes

5ec4113

cpu/esp32: add pm_layered support

dff74a3

boards/esp32: add button default interrupt flank definition

b1ec695

cpu/esp32/periph_rtc: funcs for enter/exit sleep

f4aa253

cpu/esp32/periph_gpio: funcs for enter/exit sleep

09899e4

gschorcht force-pushed the cpu/esp32/pm_layered branch from 1d88a79 to 4f97731 Compare March 23, 2020 11:48

benpicco merged commit 45b635f into RIOT-OS:master Mar 23, 2020

gschorcht deleted the cpu/esp32/pm_layered branch March 23, 2020 13:30

benpicco mentioned this pull request Mar 24, 2020

unittests/tests-rtt_rtc: blacklist test on esp32 #13705

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

cpu/esp32: support for light/deep sleep and pm_layered #13416

cpu/esp32: support for light/deep sleep and pm_layered #13416

gschorcht commented Feb 20, 2020 •

edited

Loading

yegorich Feb 20, 2020

gschorcht Feb 20, 2020

MrKevinWeiss Mar 17, 2020

gschorcht commented Feb 20, 2020

gschorcht commented Mar 8, 2020

leandrolanzieri commented Mar 9, 2020

benpicco commented Mar 17, 2020 •

edited

Loading

benpicco left a comment

benpicco Mar 17, 2020

gschorcht Mar 18, 2020

benpicco Mar 17, 2020

gschorcht Mar 18, 2020

gschorcht Mar 18, 2020 •

edited

Loading

gschorcht commented Mar 18, 2020

gschorcht commented Mar 20, 2020 •

edited

Loading

benpicco left a comment •

edited

Loading

benpicco Mar 22, 2020 •

edited

Loading

gschorcht Mar 22, 2020

gschorcht Mar 22, 2020

benpicco Mar 22, 2020

gschorcht commented Mar 22, 2020

benpicco left a comment

gschorcht commented Mar 23, 2020 •

edited

Loading

benpicco commented Mar 23, 2020

cpu/esp32: support for light/deep sleep and pm_layered #13416

cpu/esp32: support for light/deep sleep and pm_layered #13416

Conversation

gschorcht commented Feb 20, 2020 • edited Loading

Contribution description

Testing procedure

Issues/PRs references

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gschorcht commented Feb 20, 2020

gschorcht commented Mar 8, 2020

leandrolanzieri commented Mar 9, 2020

benpicco commented Mar 17, 2020 • edited Loading

benpicco left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gschorcht Mar 18, 2020 • edited Loading

Choose a reason for hiding this comment

gschorcht commented Mar 18, 2020

gschorcht commented Mar 20, 2020 • edited Loading

benpicco left a comment • edited Loading

Choose a reason for hiding this comment

benpicco Mar 22, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gschorcht commented Mar 22, 2020

benpicco left a comment

Choose a reason for hiding this comment

gschorcht commented Mar 23, 2020 • edited Loading

benpicco commented Mar 23, 2020

gschorcht commented Feb 20, 2020 •

edited

Loading

benpicco commented Mar 17, 2020 •

edited

Loading

gschorcht Mar 18, 2020 •

edited

Loading

gschorcht commented Mar 20, 2020 •

edited

Loading

benpicco left a comment •

edited

Loading

benpicco Mar 22, 2020 •

edited

Loading

gschorcht commented Mar 23, 2020 •

edited

Loading