Zephyr

From Wiki at Neela Nurseries
Revision as of 19:02, 4 January 2024 by Ted (talk | contribs) (→‎^ Zephyr Drivers: add section and first notes on SPI NOR FLASH and routine tracing)
Jump to: navigation, search

Zephyr RTOS :: Device Tree :: Kconfig :: cmake


Overview

This NN page a holding point for notes and links to external references regarding Zephyr RTOS, and Zephyr RTOS Project. Some key features of Zephyr RTOS include its designers' incorporation of Device Tree Source and Kconfig frameworks in this real time operating system.

As of 2023 April this page under reorganization, pulling multiple local pages into this page and removing older and defunct content - TMH

For device tree and Kconfig see also:


^ Zephyr Releases

Zephyr RTOS project itself, one of some one hundred twenty projects on Zephyr's Github page, has stable releases which become available once or twice a year. The Zephyr RTOS repository itself is generally at a cutting edge of development. It is `git tagged` or identified as a release candidate, until that point where the work has been finished to review all open issues and developments and create and tag a new stable release.

Nordic Semi forks a recent, but not necessarily most recent stable release of Zephyr RTOS Project as part of its NCS Software Development Kit.

A local NN wiki page with some early Zephyr release notes is Zephyr RTOS releases. There may not be much to save from this article, but contributor Ted to review it.

As of 2023 Q2 Zephyr RTOS stable release is v3.3.0.


^ Zephyr Toolchain Installation

Complete notes for Zephyr toolchain installation and config are provided at Zephyr Project's "Getting Started Guide".


^ Zephyr Build Process

Possible to use Kconfig to pass along compiler options, such as `CONFIG_COMPILER_OPT`:


----- ----- ----- ----- ----- ----- ----- ----- -----

^ Zephyr Internals

Zephyr RTOS run time starting point, for Zephyr v3.4.0 is function function z_cstart() in source file init.c.

Zephyr util.h header file:

    ${ZEPHYR_WORKSPACE}/zephyr/include/zephyr/sys/util.h:43:#define POINTER_TO_UINT(x) ((uintptr_t) (x))

Error codes enumerated in errno.h:

    zephyr/lib/libc/minimal/include/errno.h


----- ----- ----- ----- ----- ----- ----- ----- -----

^ Interrupts

Zephyr RTOS provides some API and code facilities to support hardware interrupts, but each processor architecture has its own ways of implementing interrupts. This local page section to cover issues and solutions to Zephyr application design with interrupts, and how to code for a given target processor.

First references:

A Zephyr documentation on architecture porting, with references to lower level interrupt configuration code:

Zephyr's API macro for connecting a routine to a hardware interrupt is defined in `irq.h`: 

 31 /**
 32  * @brief Initialize an interrupt handler.
 33  *
 34  * This routine initializes an interrupt handler for an IRQ. The IRQ must be
 35  * subsequently enabled before the interrupt handler begins servicing
 36  * interrupts.
 37  *
 38  * @warning
 39  * Although this routine is invoked at run-time, all of its arguments must be
 40  * computable by the compiler at build time.
 41  *
 42  * @param irq_p IRQ line number.
 43  * @param priority_p Interrupt priority.
 44  * @param isr_p Address of interrupt service routine.
 45  * @param isr_param_p Parameter passed to interrupt service routine.
 46  * @param flags_p Architecture-specific IRQ configuration flags..
 47  */
 48 #define IRQ_CONNECT(irq_p, priority_p, isr_p, isr_param_p, flags_p) \
 49         ARCH_IRQ_CONNECT(irq_p, priority_p, isr_p, isr_param_p, flags_p)


^ Zephyr Threads

An important executing element of a Zephyr based app is an application defined thread. Zephyr based applications written in C will normally have at least one app side thread , which Zephyr will default to giving the name `main`. This name reflects that application code begins executing from a function named `main()`, as is a long standing convention of C language.

Zephyr based applications can define threads of their own. This is a multi-step action in the code, both at time of development and at run time. Application defined threads are given numeric identifiers by Zephyr at run time; only "main line" code of the app gets a named thread by default. Developers however may optionally call a _Zephyr_API_ to give additional threads human meaningful names.

^ Anatomy of a Zephyr thread

To create a Zephyr thread in an app involves declarations of certain thread elements, a call to Zephyr's thread_create API function, and at minimum an entry point function for the given thread. The following pieces:

    • thread data structure
    • thread priority
    • thread stack size
    • call to Zephyr "create thread" API
    • thread entry point function

In English language, creation of a Zephyr application thread in simple terms involves the steps:

(1) create a thread's data structure of type `struct k_thread`
(2) pound define given thread's priority
(3) pound define given thread's static memory stack size in bytes
(4) call Zephyr's thread creation API function with its required ten parameters
(5) define an entry point function for the thread, which other application code most likely to call to start thread execution

A bit of repetition: Zephyr thread's data structure, priority and stack size are programmed in a declarative way. Stack size and thread data are typically declared as follows, the first of these involves a function like macro: 

K_THREAD_STACK_DEFINE(thread_led_stack_area, THREAD_LED_STACK_SIZE);

struct k_thread thread_led_data;

Interestingly, the stack area parameter first parameter to macro like function is defined in the above macro, and then referenced typically only one or two times, in the call to Zephyr's `k_thread_create()` API.

The call to `thread_create` <- REVIEW looks like this: 

int initialize_thread_led(void)
{
    int rstatus = 0;

    k_tid_t task_led_tid = k_thread_create(&thread_led_data,
                                           thread_led_stack_area,
                                           K_THREAD_STACK_SIZEOF(thread_led_stack_area),
                                           thread_led_entry_point,
                                           NULL, NULL, NULL,
                                           THREAD_LED_PRIORITY,
                                           0,
                                           K_MSEC(1000)); // K_NO_WAIT);

// REF https://docs.zephyrproject.org/2.6.0/reference/kernel/threads/index.html?highlight=k_thread_create#c.k_thread_name_set
// int k_thread_name_set(k_tid_t thread, const char *str)

    rstatus = k_thread_name_set(task_led_tid, MODULE_ID__THREAD_LED);
    if ( rstatus == 0 ) { } // avoid compiler warning about unused variable - TMH

    return (int)task_led_tid;
}

^ Thread API references

^ Zephyr Drivers

Zephyr RTOS has a notion of in-tree drivers and out-of-tree drivers. The following local page section begin to cover these topics with a few references to specific peripheral drivers.

    ^ In-tree drivers

    Early in contributor Ted's learning and use of Zephyr RTOS, "in tree" drivers and "out of tree" drivers concepts seemed distinct and referent to code bases which were completely located within Zephyr's source tree, and the latter located outside of the Zephyr RTOS source tree, respectively. This is not a correct view. Rather, Zephyr's "in tree" drivers have code in the Zephyr source tree which nearly always calls third party MCU library code bases.

    Out-out-tree drivers in contrast have their code bases entirely outside of Zephyr RTOS source tree.

    Zephyr RTOS code provides standardized structures to support peripherals and devices. These structures nearly always involve function pointers which ultimately point to a given target MCU or MCU family's driver library or Software Development Kit (SDK) code base.

    Zephyr RTOS code base makes heavy use of function pointers and of macros. These pointers and macros can make it a challenge to locate a function definiton during static code analysis. The `ctags` utility does not always have a clear chain of symbols to follow to get to the ultimate definition of a function.

    This NN article section on Zephyr in-tree drivers touches on this matter. The subsection here on SPI NOR FLASH touches on this challenge, and looks at one way of following list files from build artifacts, to get closer to driver function definitions.

    ^ Driver Code Factoring

    In-tree Zephyr drivers exist as part of Zephyr's source code tree, in other words, the code repo that entails Zephyr RTOS itself. Out-of-tree drivers written to cooperate with Zephyr exist outside of Zephyr's source code tree and are often their own independent code projects. See however this local page's In-tree drivers section for greater detail on driver code factoring in Zephyr RTOS.

    Example Zephyr app and supporting code base which entails an out-of-tree driver factored along side app code:

    ^ Out-of-tree drivers

    Stub section for Zephyr out-of-tree drivers.

^ Zephyr RTOS Logger and Backend Features

Zephyr has five log levels ranging from LOG_LEVEL_NONE to LOG_LEVEL_DBG. These are defined in log_core.h.

Printing of log messages and other strings can be deferred, and there is a mechanism in place in Zephyr RTOS project called `cbprintf packages` to help in this:

It is also possible to reduce code size in a Zephyr app by avoiding libc library functions in the s*printf() family:


Following links need updating to Zephyr 3.4.0:

Zephyr's thread_analyzer when not configured to dump reports to printk() instead sends thread data to LOG_INF(). Excerpt from Zephyr file `./zephyr/subsys/debug/thread_analyzer.c`: 

 11 #include <kernel.h>
 12 #include <debug/thread_analyzer.h>
 13 #include <debug/stack.h>
 14 #include <kernel.h>
 15 #include <logging/log.h>
 16 #include <stdio.h>
 17 
 18 LOG_MODULE_REGISTER(thread_analyzer, CONFIG_THREAD_ANALYZER_LOG_LEVEL);
 19 
 20 #if IS_ENABLED(CONFIG_THREAD_ANALYZER_USE_PRINTK)
 21 #define THREAD_ANALYZER_PRINT(...) printk(__VA_ARGS__)
 22 #define THREAD_ANALYZER_FMT(str)   str "\n"
 23 #define THREAD_ANALYZER_VSTR(str)  (str)
 24 #else
 25 #define THREAD_ANALYZER_PRINT(...) LOG_INF(__VA_ARGS__)
 26 #define THREAD_ANALYZER_FMT(str)   str
 27 #define THREAD_ANALYZER_VSTR(str)  log_strdup(str)
 28 #endif

Zephyr macro `LOG_INF` is in turn defined: 

ted@localhost:~/projects/zephyr-based/z4-sandbox-kionix-work/zephyr$ grep -nr LOG_INF ./* | grep define

./include/logging/log.h:61:#define LOG_INF(...)   Z_LOG(LOG_LEVEL_INF, __VA_ARGS__)

Define of Z_LOG: 

ted@localhost:~/projects/zephyr-based/z4-sandbox-kionix-work/zephyr$ grep -nr 'Z_LOG[^_]' ./*

./include/logging/log_core.h:295:#define Z_LOG2(_level, _source, _dsource, ...) do { \
./include/logging/log_core.h:329:#define Z_LOG(_level, ...) \



$ grep -nr 'Z_LOG[^_]' ./*

$ vi ./include/logging/log_core.h

329 #define Z_LOG(_level, ...) \
330         Z_LOG2(_level, __log_current_const_data, __log_current_dynamic_data, __VA_ARGS__)


A Zephyr RTOS sample to check out:

`ted@localhost:~/projects/zephyr-based/z4-sandbox-kionix-work/zephyr/samples/subsys/logging/logger$`


- 2022-10-05 -

Zephyr v2.6.0 source file `log_core.h` may also define `log_strdup()`: 

./subsys/logging/log_core.c:1017:char *z_log_strdup(const char *str)

But it looks like the routine is statically in-lined in Zephyr header file `log.h`: 

./include/logging/log.h:290:static inline char *log_strdup(const char *str)


^ Zephyr thread_analyzer_cb And Related

Zephyr 2.6.0's thread_analyzer_cb() routine gathers and populates a structure with all pertinent, reported run-time thread statistics. This is however a static routine so we cannot call it directly. Nor does it return the summary thread resource use data to its direct calling routine. There is yet hope, and to understand a path forward here excerpted is the definition of this important thread reporting routine from file `./zephyr/subsys/debug/thread_analyzer.c`: 

 59 static void thread_analyze_cb(const struct k_thread *cthread, void *user_data)
 60 {
 61         struct k_thread *thread = (struct k_thread *)cthread;
 62 #ifdef CONFIG_THREAD_RUNTIME_STATS
 63         k_thread_runtime_stats_t rt_stats_all;
 64         k_thread_runtime_stats_t rt_stats_thread;
 65         int ret;
 66 #endif
 67         size_t size = thread->stack_info.size;
 68         thread_analyzer_cb cb = user_data;
 69         struct thread_analyzer_info info;
 70         char hexname[PTR_STR_MAXLEN + 1];
 71         const char *name;
 72         size_t unused;
 73         int err;
 74 
 75 
 76 
 77         name = k_thread_name_get((k_tid_t)thread);
 78         if (!name || name[0] == '\0') {
 79                 name = hexname;
 80                 snprintk(hexname, sizeof(hexname), "%p", (void *)thread);
 81         }
 82 
 83         err = k_thread_stack_space_get(thread, &unused);
 84         if (err) {
 85                 THREAD_ANALYZER_PRINT(
 86                         THREAD_ANALYZER_FMT(
 87                                 " %-20s: unable to get stack space (%d)"),
 88                         name, err);

 88                         name, err);
 89 
 90                 unused = 0;
 91         }
 92 
 93         info.name = name;
 94         info.stack_size = size;
 95         info.stack_used = size - unused;
 96 
 97 #ifdef CONFIG_THREAD_RUNTIME_STATS
 98         ret = 0;
 99 
100         if (k_thread_runtime_stats_get(thread, &rt_stats_thread) != 0) {
101                 ret++;
102         }
103 
104         if (k_thread_runtime_stats_all_get(&rt_stats_all) != 0) {
105                 ret++;
106         }
107         if (ret == 0) {
108                 info.utilization = (rt_stats_thread.execution_cycles * 100U) /
109                         rt_stats_all.execution_cycles;
110         }
111 #endif
112         cb(&info);
113 }

The routine which calls this is: 

115 void thread_analyzer_run(thread_analyzer_cb cb)
116 {
117         if (IS_ENABLED(CONFIG_THREAD_ANALYZER_RUN_UNLOCKED)) {
118                 k_thread_foreach_unlocked(thread_analyze_cb, cb);
119         } else {
120                 k_thread_foreach(thread_analyze_cb, cb);
121         }
122 }

We should be able to directly call thread_analyzer_run(thread_analyzer_cb cb). If yes, and if we can learn how &info is defined, we should be able to redirect thread_analyzer reports to an arbitrary UART . . .


^ struct thread_analyzer_info

ted@localhost:~/projects/zephyr-based/z4-sandbox-kionix-work/zephyr$ grep -nr 'struct thread_analyzer_info' ./*
./include/debug/thre[[#top|^]] ad_analyzer.h:23:struct thread_analyzer_info {
./include/debug/thread_analyzer.h:44:typedef void (*thread_analyzer_cb)(struct thread_analyzer_info *info);

Structure definition for &info passed to thread_analyzer_cb() function above: 

/** @defgroup thread_analyzer Thread analyzer
 *  @brief Module for analyzing threads
 *
 *  This module implements functions and the configuration that simplifies
 *  thread analysis.
 *  @{
 */

struct thread_analyzer_info {
        /** The name of the thread or stringified address of the thread handle
         * if name is not set.
         */
        const char *name;
        /** The total size of the stack*/
        size_t stack_size;
        /** Stack size in used */
        size_t stack_used;

#ifdef CONFIG_THREAD_RUNTIME_STATS
        unsigned int utilization;
#endif
};


^ Zephyr Shell

This section started 2023-06-12. Growth goal at this time is to learn how Zephyr shell supports multiple shell sessions, across one or many interfaces.

First reference noting here relates to develop needing to separate shell communications from Zephyr logging. Not quite multi-session / multi-context shell support but somewhere in the ballpark:

This link asks salient questions to topic of multiple shell session support in Zephyr shell facility:


----- ----- ----- ----- ----- ----- ----- ----- -----

Zephyr Memory Management


^ Zephyr Kernel Timing

System timing, timing of hardware and firmware running together is an important non-trivial feature of nearly all embedded systems. In RTOS situations the operating system must expend some resources to keep track of a system timing mechanism with a time granularity of n microseconds. The given system may use another unit of time but the concept remains the same. Local article contributor Ted believes this applies for operating systems (and bare metal projects employing a timing mechanism) where there's a "system tick" or "systick" implemented.

Zephyr provides an API to permit firmware at runtime to obtain systick time granularity:

// A Zephyr API to obtain MCU clock cycles per second: // sys_clock_hw_cycles_per_sec()


Program uptime or "running hours" may also be of import to a given project. Following link at Zephyr Project documentation pages has valuable info on various Zephyr timing and work queueing facilities:

Related:

Important to follow up on this bug report relating to above mentioned Zephyr timing features:

2023-12-06

In Zephyr a k_timeout_t is a structure.


^ Zephyr Application Code Relocation


Retrieved from "https://wiki.neelanurseries.com/index.php?title=Zephyr&oldid=7681"