Checking by the send function that at least two maximum size Ethernet frames fit in the remaining heap before the LwIP packet buffer is allocated seems to increase stability. This can be caused by the fact that WLAN hardware interrupts allocate additional memory when receiving a frame during the send attempt.
Error case:
1. thread_yield_higher() stores the thread's ucontext
2. creates an "isr ucontext" for isr_thread_yield, switches to it
Case 1: no signals are pending, continues in isr_thread_yield()
3a. sched_run is called
4a. return to sched_active_thread ucontext
Case 2: signals pending (the crashing scenario), continues in native_irq_handler()
3b. handles signals
4b. if sched_context_switch_request is set, call sched_run
5b. return to sched_active_thread ucontext
4b misses the call to sched_run(), leading to a possible return into a
non-ready thread.
The situation where the firmware `lwIP` packet buffer is exhausted is an important indication that the traffic sent to and sent from the esp8266 is more than the esp8266 is able to handle. Therefore, it should be an error message.
It is not necessary to realize timeout handling in send function or to disconnect from AP if lwIP packet buffer is exhausted. Waiting that the frame allocated in lwIP packet buffer is freed by MAC layer led to the complete blockage of send function on heavy network load. Disconnecting from AP is counterproductive since reconnecting usually fails on heavy network load.
Disconnecting from the AP in the send function if the lwIP packet buffer is exhausted is counterproductive since reconnecting usually fails on heavy network load. A better strategy is to slow down the sending of MAC frames from netif a bit to wait for flushing the buffer in the MAC layer.
Defines a number of lwIP functions that are required as symbols by Espressif's SDK libraries. These functions are only dummies without real functionality. Using these functions instead of real lwIP functions provided with the SDK saves arround 4 kBytes of RAM.
Since _esp_wifi_recv_cb is not executed in interrupt context but in the context of the `ets` thread, it is not necessary to pass the`NETDEV_EVENT_ISR` event first. Instead, the receive function can be called directly which result in much faster handling, a less frame lost rate and more robustness.
Since _esp_wifi_recv_cb is not executed in interrupt context but in the context of the `ets` thread, the receive function can be called directly. There is no need for a mutex anymore to synchronize the access to the receive buffer between _esp_wifi_recv_cb and _recv function.
When the size of a received frame is checked, always the total length should be used instead of the length of the first lwIP pbuf in the pbuf chain. Otherwise, the check that the length does not exceed ETHERNET_MAX_LEN will always be true since the maximum size of one lwIP pbuf in a pbuf chain is 512 bytes.
Although only the station interface is needed, the WiFi interface has to be used in SoftAP + Station mode. Otherwise the send function blocks sporadically. Since the SoftAP interface is not used, it is configured with a hidden SSID and a long beacon interval. Connections from other stations are not allowed.
Some SDK interrupt service routines obviously use malloc/free in the interrupt context. Because the rmutex-based lock/unlock approach of the malloc/free function suite of the newlib does not work in the interrupt context, the SDK memory management functions MUST therefore be used. To use the same memory management functions in RIOT as in the SDK, the malloc/free function suite has to be replaced by wrapper functions.
Fixes sporadic blocking of the wifi thread in esp_wifi_recv_cb function under heavy network load conditions when frames are coming in faster than they can be processed. Since esp_wifi_recv_cb function is not executed in interrupt context, the msg_send function used for ISR event can block when the message queue is full. With this change esp_wifi can be flooded with icmpv6 packets of maximum size without any problems over hours.
This prevent evaluating `LINKFLAGS` when not needed and when building
in docker so does not produce errors if `avr-ld` is not installed.
```
BUILD_IN_DOCKER=1 BOARD=arduino-mega2560 make --no-print-directory -C examples/hello-world/ clean
/srv/ilab-builds/workspace/git/riot_master/makefiles/toolchain/gnu.inc.mk:18: objcopy not found. Hex file will not be created.
/bin/sh: 1: avr-ld: not found
```
It removes the `/bin/sh: 1: avr-ld: not found`
During flash write access, the IROM cache cannot be used and is disabled therefore. During that time, ets_post crashes if a functions is called which is not in IRAM. Therefore thread_flags_set must not be called if IROM cache is disabled.
Changes of ETS task handling require the context switch by software interrupt. The context switch based on interrupt is therefore enabled by default. Furthermore, the number of priority levels are increased due to the new additional thread.
Fixes sporadic blocking of the wifi thread in esp_wifi_recv_cb function under heavy network load conditions when frames are coming in faster than they can be processed. Since esp_wifi_recv_cb function is not executed in interrupt context, the msg_send function used for ISR event can block when the message queue is full. With this change esp_wifi can be flooded with icmpv6 packets of maximum size without any problems over hours.
RX callback function should be register when WiFi has been connected to AP successfully and should be unregistered when WiFi disconnects from AP. Therefore, esp_wifi_internal_reg_rxcb is called now in event handler on event SYSTEM_EVENT_STA_CONNECTED. It is reset now on event SYSTEM_EVENT_STA_DISCONNECTED.
Before function wifi_connect is executed, starting the WiFi driver should have been finished. This is indicated by the WiFi driver by sending event SYSTEM_EVENT_STA_START. Function wifi_connect is moved therefore to the event handler for SYSTEM_EVENT_STA_START.
The size of received and transmitted frames was stored in an uint8_t, which did not allow to process frames larger than 255 octets. However, WiFi has an MTU of 1500 octets.
Adds a memset function `system_secure_memset` which cannot be optimized out by the compiler. It uses the libsodium approach of weak symbols. Function system_secure_memset calls the standard memset. Calling an empty function declared with weak attribute after the memset call, prevents the compiler to optimize it out. The overhead is only one function call.
The driver is claiming it is needed for IPv6 / 6LoWPAN support, which
is not true (the long address is never used for NRFMIN in fact) and
this assumption actually leads to an assertion when compiled as with
the `gnrc_border_router` example.
Although ESP32 has four SPI controllers, only two of them can be effectively used (HSP and VSPI). The third one (FSPI) is used for external memory such as flash and PSRAM and can not be used for peripherals. FSPI is therefore removed from the API. In addition, the SPI0_DEV and SPI1_DEV configuration parameters are renamed SPI0_CTRL and SPI1_CTRL to better describe what they define and to avoid confusion with SPI_DEV (0) and SPI_DEV (1).
Although ESP32 has four SPI controllers, only two of them can be effectively used (HSP and VSPI). The third one (FSPI) is used for external memory such as flash and PSRAM and can not be used for peripherals. FSPI is therefore removed from the API. In addition, the SPI0_DEV and SPI1_DEV configuration parameters are renamed SPI0_CTRL and SPI1_CTRL to better describe what they define and to avoid confusion with SPI_DEV (0) and SPI_DEV (1).
Timer restart was moved from esp_now_scan_peers_done to esp_now_scan_peers_start to avoid that the scan for peers isn't triggered anymore if the esp_now_scan_peers_done callback isn't called because of errors.
The brr calculation on the datasheet is different than what is implmented.
This is intentional since it provides better rounding due to truncation.
There was no comment explaining that so this comment should prevent confusion.
Since it is not possible to reenter the function `esp_now_recv_cb`, and the functions netif::_ recv and esp_now_netdev::_ recv are called directly in the same thread context we can read directly from the `buf` and don't need a receive buffer anymmore.
Since it is not possible to reenter the function `esp_now_recv_cb`, and the functions netif::_ recv and esp_now_netdev::_ recv are called directly in the same thread context we only need a buffer which contains one frame and its source mac address.
Since it is not possible to reenter the function `esp_now_recv_cb`, and the functions netif::_ recv and esp_now_netdev::_ recv are called directly in the same thread context, neither a mutual exclusion has to be realized nor have the interrupts to be deactivated.
Since the esp_now_recv_cb function is not called directly as an ISR through interrupts, it is not executed in the interrupt context. _recv can therefore be called directly. The treatment of the recv_event is no longer necessary.
Although it should not be possible to reenter the function esp_now_recv_cb, this is avoided by an additional boolean variable. It can not be realized by mutex because it would reenter from same thread context. If macro NDEBUG is not defined, an assertion is used.
A second slot is defined with a calculated size, from the
remaining flash after the bootloader and the first slot.
Both slots are defined as equal size, but it can be overriden.
Avoids parsing IPv6 packets to determine destination address.
Allows using 6Lo over ESP-NOW, which is required due to the low MTU of
ESP-NOW.
Co-authored-by: Gunar Schorcht <gunar@schorcht.net>
