2.1. CPU Dependent#

This class of code includes the foundation routines for the executive proper such as the context switch and the interrupt subroutine implementations. Sources for the supported processor families can be found in cpukit/score/cpu. A good starting point for a new family of processors is the no_cpu directory, which holds both prototypes and descriptions of each needed CPU dependent function.

CPU dependent code is further subcategorized if the implementation is dependent on a particular CPU model. For example, the MC68000 and MC68020 processors are both members of the m68k CPU family but there are significant differences between these CPU models which RTEMS must take into account.

The source code found in the cpukit/score/cpu is required to only depend upon the CPU model variations that GCC distinguishes for the purposes of multilib’ing. Multilib is the term the GNU community uses to refer to building a single library source multiple times with different compiler options so the binary code generated is compatible. As an example, from GCC’s perspective, many PowerPC CPU models are just a PPC603e. Remember that GCC only cares about the CPU code itself and need not be aware of any peripherals. In the embedded community, we are exposed to thousands of CPU models which are all based upon only a relative small number of CPU cores.

Similarly for the SPARC/ERC32 BSP, the RTEMS_CPU is specified as erc32 which is the name of the CPU model and BSP for this SPARC V7 system on chip. But the multilib variant used is actually v7 which indicates the ERC32 CPU core is a SPARC V7.

2.2. Board Dependent#

This class of code provides the most specific glue between RTEMS and a particular board. This code is represented by the Board Support Packages and associated Device Drivers. Sources for the BSPs included in the RTEMS distribution are located in the directory bsps. The BSP source directory is further subdivided based on the CPU family and BSP.

Some BSPs may support multiple board models within a single board family. This is necessary when the board supports multiple variants on a single base board. For example, the Motorola MVME162 board family has a fairly large number of variations based upon the particular CPU model and the peripherals actually placed on the board.

2.3. Peripheral Dependent#

This class of code provides a reusable library of peripheral device drivers which can be tailored easily to a particular board. The libchip library is a collection of reusable software objects that correspond to standard controllers. Just as the hardware engineer chooses a standard controller when designing a board, the goal of this library is to let the software engineer do the same thing.

The source code for the reusable peripheral driver library may be found in the directory cpukit/dev or bsps/shared/dev. The source code is further divided based upon the class of hardware. Example classes include serial communications controllers, real-time clocks, non-volatile memory, and network controllers.

2.4. Questions to Ask#

When evaluating what is required to support RTEMS applications on a particular target board, the following questions should be asked:

• Does a BSP for this board exist?

• Does a BSP for a similar board exists?

• Is the board’s CPU supported?

If there is already a BSP for the board, then things may already be ready to start developing application software. All that remains is to verify that the existing BSP provides device drivers for all the peripherals on the board that the application will be using. For example, the application in question may require that the board’s Ethernet controller be used and the existing BSP may not support this.

If the BSP does not exist and the board’s CPU model is supported, then examine the reusable chip library and existing BSPs for a close match. Other BSPs and libchip provide starting points for the development of a new BSP. It is often possible to copy existing components in the reusable chip library or device drivers from BSPs from different CPU families as the starting point for a new device driver. This will help reduce the development effort required.

If the board’s CPU family is supported but the particular CPU model on that board is not, then the RTEMS port to that CPU family will have to be augmented. After this is done, development of the new BSP can proceed.

Otherwise both CPU dependent code and the BSP will have to be written.

This type of development often requires specialized skills and there are people in the community who provide those services. If you need help in making these modifications to RTEMS try a search in a search engine with something like “RTEMS support”. The RTEMS Project encourages users to use support services however we do not endorse any providers.

2.5. CPU Dependent Executive Files#

The CPU dependent files in the RTEMS executive source code are found in the cpukit/score/cpu/${RTEMS_CPU} directories. The ${RTEMS_CPU} is a particular architecture, e.g. arm, powerpc, riscv, sparc, etc.

Within each CPU dependent directory inside the executive proper is a file named cpu.h which contains information about each of the supported CPU models within that family.

2.6. Board Support Package Structure#

The BSPs are all under the bsps directory. The structure in this source subtree is:

• bsps/shared

• bsps/${RTEMS_CPU}/shared

• bsps/${RTEMS_CPU}/${RTEMS_BSP_FAMILY}

The ${RTEMS_CPU} is a particular architecture, e.g. arm, powerpc, riscv, sparc, etc. The shared directories contain code shared by all BSPs or BSPs of a particular architecture. The ${RTEMS_BSP_FAMILY} directories contain BSPs for a particular system on chip (SoC) or processor family.

Use the following structure under the bsps/${RTEMS_CPU}/${RTEMS_BSP_FAMILY}:

• ata - the legacy ATA/IDE driver

• btimer - the legacy benchmark timer driver

• cache - cache controller support

• clock - the clock driver

• config - build system configuration files

• console - the console driver

• contrib - imports of external sources

  • the layout of external sources should be used as is if possible

• i2c - the I2C driver

• include - public header files

• irq - the interrupt controller support

• mpci - support for heterogeneous multiprocessing (RTEMS_MULTIPROCESSING)

• net - legacy network stack drivers

• rtc - the RTC driver

• spi - the SPI driver

• start - everything required to run a minimal application without devices

  • start.S - lowest level startup code

  • bspstart.c - low level startup code

  • bspsmp.c - SMP support

  • linkcmds - a linker command file

3.1. What is a “linkcmds” file?#

The linkcmds file is a script which is passed to the linker at linking time. This file describes the memory configuration of the board as needed to link the program. Specifically it specifies where the code and data for the application will reside in memory.

The format of the linker script is defined by the GNU Loader ld which is included as a component of the GNU Binary Utilities. If you are using GNU/Linux, then you probably have the documentation installed already and are using these same tools configured for native use. Please visit the Binutils project http://sourceware.org/binutils/ if you need more information.

3.2. Program Sections#

An embedded systems programmer must be much more aware of the placement of their executable image in memory than the average applications programmer. A program destined to be embedded as well as the target system have some specific properties that must be taken into account. Embedded machines often mean average performances and small memory usage. It is the memory usage that concerns us when examining the linker command file.

Two types of memories have to be distinguished:

• RAM - volatile offering read and write access

• ROM - non-volatile but read only

Even though RAM and ROM can be found in every personal computer, one generally doesn’t care about them. In a personal computer, a program is nearly always stored on disk and executed in RAM. Things are a bit different for embedded targets: the target will execute the program each time it is rebooted or switched on. The application program is stored in non-volatile memory such as ROM, PROM, EEPROM, or Flash. On the other hand, data processing occurs in RAM.

This leads us to the structure of an embedded program. In rough terms, an embedded program is made of sections. It is the responsibility of the application programmer to place these sections in the appropriate place in target memory. To make this clearer, if using the COFF object file format on the Motorola m68k family of microprocessors, the following sections will be present:

• code (.text) section: is the program’s code and it should not be modified. This section may be placed in ROM.

• non-initialized data (.bss) section: holds uninitialized variables of the program. It can stay in RAM.

• initialized data (.data) section: holds the initialized program data which may be modified during the program’s life. This means they have to be in RAM. On the other hand, these variables must be set to predefined values, and those predefined values have to be stored in ROM.

That brings us up to the notion of the image of an executable: it consists of the set of the sections that together constitute the application.

3.3. Image of an Executable#

As a program executable has many sections (note that the user can define their own, and that compilers define theirs without any notice), one has to specify the placement of each section as well as the type of memory (RAM or ROM) the sections will be placed into. For instance, a program compiled for a Personal Computer will see all the sections to go to RAM, while a program destined to be embedded will see some of his sections going into the ROM.

The connection between a section and where that section is loaded into memory is made at link time. One has to let the linker know where the different sections are to be placed once they are in memory.

The following example shows a simple layout of program sections. With some object formats, there are many more sections but the basic layout is conceptually similar.

3.4. Example Linker Command Script#

The GNU linker has a command language to specify the image format. This command language can be quite complicated but most of what is required can be learned by careful examination of a well-documented example. The following is a heavily commented version of the linker script used with the the gen68340 BSP This file can be found at $BSP340_ROOT/startup/linkcmds.

/*
 *  Specify that the output is to be coff-m68k regardless of what the
 *  native object format is.
 */
OUTPUT_FORMAT(coff-m68k)
/*
 *  Set the amount of RAM on the target board.
 *
 *  NOTE: The default may be overridden by passing an argument to ld.
 */
RamSize = DEFINED(RamSize) ? RamSize : 4M;
/*
 *  Set the amount of RAM to be used for the application heap.  Objects
 *  allocated using malloc() come from this area.  Having a tight heap
 *  size is somewhat difficult and multiple attempts to squeeze it may
 *  be needed reducing memory usage is important.  If all objects are
 *  allocated from the heap at system initialization time, this eases
 *  the sizing of the application heap.
 *
 *  NOTE 1: The default may be overridden by passing an argument to ld.
 *
 *  NOTE 2: The TCP/IP stack requires additional memory in the Heap.
 *
 *  NOTE 3: The GNAT/RTEMS run-time requires additional memory in
 *  the Heap.
 */
HeapSize = DEFINED(HeapSize) ? HeapSize : 0x10000;
/*
 *  Set the size of the starting stack used during BSP initialization
 *  until first task switch.  After that point, task stacks allocated
 *  by RTEMS are used.
 *
 *  NOTE: The default may be overridden by passing an argument to ld.
 */
StackSize = DEFINED(StackSize) ? StackSize : 0x1000;
/*
 *  Starting addresses and length of RAM and ROM.
 *
 *  The addresses must be valid addresses on the board.  The
 *  Chip Selects should be initialized such that the code addresses
 *  are valid.
 */
MEMORY {
ram : ORIGIN = 0x10000000, LENGTH = 4M
rom : ORIGIN = 0x01000000, LENGTH = 4M
}
/*
 *  This is for the network driver.  See the Networking documentation
 *  for more details.
 */
ETHERNET_ADDRESS =
DEFINED(ETHERNET_ADDRESS) ? ETHERNET_ADDRESS : 0xDEAD12;
/*
 *  The following defines the order in which the sections should go.
 *  It also defines a number of variables which can be used by the
 *  application program.
 *
 *  NOTE: Each variable appears with 1 or 2 leading underscores to
 *        ensure that the variable is accessible from C code with a
 *        single underscore.  Some object formats automatically add
 *        a leading underscore to all C global symbols.
 */
SECTIONS {
/*
 *  Make the RomBase variable available to the application.
 */
_RamSize = RamSize;
__RamSize = RamSize;
/*
 *  Boot PROM  - Set the RomBase variable to the start of the ROM.
 */
rom : {
  _RomBase = .;
  __RomBase = .;
} >rom
/*
 * Dynamic RAM - set the RamBase variable to the start of the RAM.
 */
ram : {
  _RamBase = .;
  __RamBase = .;
} >ram
/*
 *  Text (code) goes into ROM
 */
.text : {
  /*
   *  Create a symbol for each object (.o).
   */
  CREATE_OBJECT_SYMBOLS
  /*
   *  Put all the object files code sections here.
   */
  *(.text)
  . = ALIGN (16);      /*  go to a 16-byte boundary */
  /*
   *  C++ constructors and destructors
   *
   *  NOTE:  See the CROSSGCC mailing-list FAQ for
   *         more details about the "\[......]".
   */
  __CTOR_LIST__ = .;
   [......]
  __DTOR_END__ = .;
  /*
   *  Declares where the .text section ends.
   */
  etext = .;
 _etext = .;
} >rom
/*
 *  Exception Handler Frame section
 */
.eh_fram : {
  . = ALIGN (16);
  *(.eh_fram)
} >ram
/*
 *  GCC Exception section
 */
.gcc_exc : {
  . = ALIGN (16);
  *(.gcc_exc)
} >ram
/*
 *  Special variable to let application get to the dual-ported
 *  memory.
 */
dpram : {
  m340 = .;
  _m340 = .;
  . += (8 * 1024);
} >ram
/*
 *  Initialized Data section goes in RAM
 */
.data : {
  copy_start = .;
  *(.data)
  . = ALIGN (16);
  _edata = .;
  copy_end = .;
} >ram
/*
 *  Uninitialized Data section goes in ROM
 */
.bss : {
  /*
  *  M68K specific: Reserve some room for the Vector Table
  *  (256 vectors of 4 bytes).
  */
  M68Kvec = .;
  _M68Kvec = .;
  . += (256 * 4);
  /*
  *  Start of memory to zero out at initialization time.
  */
  clear_start = .;
  /*
   *  Put all the object files uninitialized data sections
   *  here.
   */
  *(.bss)
  *(COMMON)
  . = ALIGN (16);
  _end = .;
  /*
   *  Start of the Application Heap
   */
  _HeapStart = .;
  __HeapStart = .;
  . += HeapSize;
  /*
  *  The Starting Stack goes after the Application Heap.
  *  M68K stack grows down so start at high address.
  */
  . += StackSize;
  . = ALIGN (16);
  stack_init = .;
  clear_end = .;
  /*
  *  The RTEMS Executive Workspace goes here.  RTEMS
  *  allocates tasks, stacks, semaphores, etc. from this
  *  memory.
  */
  _WorkspaceBase = .;
  __WorkspaceBase = .;
} >ram

3.5. Initialized Data#

Now there’s a problem with the initialized data: the .data section has to be in RAM as this data may be modified during the program execution. But how will the values be initialized at boot time?

One approach is to place the entire program image in RAM and reload the image in its entirety each time the program is run. This is fine for use in a debug environment where a high-speed connection is available between the development host computer and the target. But even in this environment, it is cumbersome.

The solution is to place a copy of the initialized data in a separate area of memory and copy it into the proper location each time the program is started. It is common practice to place a copy of the initialized .data section at the end of the code (.text) section in ROM when building a PROM image. The GNU tool objcopy can be used for this purpose.

The following figure illustrates the steps a linked program goes through to become a downloadable image.

In Step 1, the program is linked together using the BSP linker script.

In Step 2, a copy is made of the .data section and placed after the .text section so it can be placed in PROM. This step is done after the linking time. There is an example of doing this in the file $RTEMS_ROOT/make/custom/gen68340.cfg:

# make a PROM image using objcopy
m68k-rtems-objcopy --adjust-section-vma \
.data=`m68k-rtems-objdump --section-headers $(basename $@).exe | awk '[...]'` \
$(basename $@).exe

Step 3 shows the final executable image as it logically appears in the target’s non-volatile program memory. The board initialization code will copy the “”copy of .data section” (which are stored in ROM) to their reserved location in RAM.

4.1. README Files#

Most BSPs provide one or more README files. Generally, there is a README file at the top of the BSP source. This file describes the board and its hardware configuration, provides vendor information, local configuration information, information on downloading code to the board, debugging, etc.. The intent of this file is to help someone begin to use the BSP faster.

A README file in a BSP subdirectory typically explains something about the contents of that subdirectory in greater detail. For example, it may list the documentation available for a particular peripheral controller and how to obtain that documentation. It may also explain some particularly cryptic part of the software in that directory or provide rationale on the implementation.

4.2. Times#

This file contains the results of the RTEMS Timing Test Suite. It is in a standard format so that results from one BSP can be easily compared with those of another target board.

If a BSP supports multiple variants, then there may be multiple times files. Usually these are named times.VARIANTn.

4.3. Tools Subdirectory#

Some BSPs provide additional tools that aid in using the target board. These tools run on the development host and are built as part of building the BSP. Most common is a script to automate running the RTEMS Test Suites on the BSP. Examples of this include:

• powerpc/psim includes scripts to ease use of the simulator

• m68k/mvme162 includes a utility to download across the VMEbus into target memory if the host is a VMEbus board in the same chasis.

4.4. bsp.h Include File#

The file include/bsp.h contains prototypes and definitions specific to this board. Every BSP is required to provide a bsp.h. The best approach to writing a bsp.h is copying an existing one as a starting point.

Many bsp.h files provide prototypes of variables defined in the linker script (linkcmds).

4.5. tm27.h Include File#

The tm27 test from the RTEMS Timing Test Suite is designed to measure the length of time required to vector to and return from an interrupt handler. This test requires some help from the BSP to know how to cause and manipulate the interrupt source used for this measurement. The following is a list of these:

• MUST_WAIT_FOR_INTERRUPT - modifies behavior of tm27.

• Install_tm27_vector - installs the interrupt service routine for the Interrupt Benchmark Test (tm27).

• Cause_tm27_intr - generates the interrupt source used in the Interrupt Benchmark Test (tm27).

• Clear_tm27_intr - clears the interrupt source used in the Interrupt Benchmark Test (tm27).

• Lower_tm27_intr - lowers the interrupt mask so the interrupt source used in the Interrupt Benchmark Test (tm27) can generate a nested interrupt.

All members of the Timing Test Suite are designed to run WITHOUT the Clock Device Driver installed. This increases the predictability of the tests’ execution as well as avoids occassionally including the overhead of a clock tick interrupt in the time reported. Because of this it is sometimes possible to use the clock tick interrupt source as the source of this test interrupt. On other architectures, it is possible to directly force an interrupt to occur.

4.6. sbrk() Implementation#

Although nearly all BSPs give all possible memory to the C Program Heap at initialization, it is possible for a BSP to configure the initial size of the heap small and let it grow on demand. If the BSP wants to dynamically extend the heap used by the C Library memory allocation routines (i.e. malloc family), then the``sbrk`` routine must be functional. The following is the prototype for this routine:

void * sbrk(ptrdiff_t increment)

The increment amount is based upon the sbrk_amount parameter passed to the bsp_libc_init during system initialization.

If your BSP does not want to support dynamic heap extension, then you do not have to do anything special. However, if you want to support sbrk, you must provide an implementation of this method and define CONFIGURE_MALLOC_BSP_SUPPORTS_SBRK in bsp.h. This informs rtems/confdefs.h to configure the Malloc Family Extensions which support sbrk.

4.7. bsp_fatal_extension() - Cleanup the Hardware#

The bsp_fatal_extension() is an optional BSP specific initial extension invoked once a fatal system state is reached. Most of the BSPs use the same shared version of bsp_fatal_extension() that does nothing or performs a system reset. This implementation is located in the bsps/shared/start/bspfatal-default.c file.

The bsp_fatal_extension() routine can be used to return to a ROM monitor, insure that interrupt sources are disabled, etc.. This routine is the last place to ensure a clean shutdown of the hardware. The fatal source, internal error indicator, and the fatal code arguments are available to evaluate the fatal condition. All of the non-fatal shutdown sequences ultimately pass their exit status to rtems_shutdown_executive and this is what is passed to this routine in case the fatal source is RTEMS_FATAL_SOURCE_EXIT.

On some BSPs, it prints a message indicating that the application completed execution and waits for the user to press a key before resetting the board. The PowerPC/gen83xx and PowerPC/gen5200 BSPs do this when they are built to support the FreeScale evaluation boards. This is convenient when using the boards in a development environment and may be disabled for production use.

4.8. Configuration Macros#

Each BSP can define macros in bsp.h which alter some of the the default configuration parameters in rtems/confdefs.h. This section describes those macros:

• CONFIGURE_MALLOC_BSP_SUPPORTS_SBRK must be defined if the BSP has proper support for sbrk. This is discussed in more detail in the previous section.

• BSP_IDLE_TASK_BODY may be defined to the entry point of a BSP specific IDLE thread implementation. This may be overridden if the application provides its own IDLE task implementation.

• BSP_IDLE_TASK_STACK_SIZE may be defined to the desired default stack size for the IDLE task as recommended when using this BSP.

• BSP_INTERRUPT_STACK_SIZE may be defined to the desired default interrupt stack size as recommended when using this BSP. This is sometimes required when the BSP developer has knowledge of stack intensive interrupt handlers.

• BSP_DEFAULT_UNIFIED_WORK_AREAS is defined when the BSP recommends that the unified work areas configuration should always be used. This is desirable when the BSP is known to always have very little RAM and thus saving memory by any means is desirable.

4.9. set_vector() - Install an Interrupt Vector#

On targets with Simple Vectored Interrupts, the BSP must provide an implementation of the set_vector routine. This routine is responsible for installing an interrupt vector. It invokes the support routines necessary to install an interrupt handler as either a “raw” or an RTEMS interrupt handler. Raw handlers bypass the RTEMS interrupt structure and are responsible for saving and restoring all their own registers. Raw handlers are useful for handling traps, debug vectors, etc.

The set_vector routine is a central place to perform interrupt controller manipulation and encapsulate that information. It is usually implemented as follows:

rtems_isr_entry set_vector(                 /* returns old vector */
  rtems_isr_entry handler,                  /* isr routine        */
  rtems_vector_number vector,               /* vector number      */
  int                 type                  /* RTEMS or RAW intr  */
)
{
  if the type is RAW
    install the raw vector
  else
    use rtems_interrupt_catch to install the vector
  perform any interrupt controller necessary to unmask the interrupt source
  return the previous handler
}

4.10. Interrupt Delay Profiling#

The RTEMS profiling needs support by the BSP for the interrupt delay times. In case profiling is enabled via the RTEMS build configuration option --enable-profiling (in this case the pre-processor symbol RTEMS_PROFILING is defined) a BSP may provide data for the interrupt delay times. The BSP can feed interrupt delay times with the _Profiling_Update_max_interrupt_delay() function (#include <rtems/score/profiling.h>). For an example please have a look at bsps/sparc/leon3/clock/ckinit.c.

4.11. Programmable Interrupt Controller API#

A BSP can use the PIC API to install Interrupt Service Routines through a set of generic methods. In order to do so, the header files <bsp/irq-generic.h> and <bsp/irq-info.h> must be included by the bsp specific irq.h file present in the include/ directory. The irq.h acts as a BSP interrupt support configuration file which is used to define some important MACROS. It contains the declarations for any required global functions like bsp_interrupt_dispatch(). Thus later on, every call to the PIC interface requires including <bsp/irq.h>

The generic interrupt handler table is intitalized by invoking the bsp_interrupt_initialize() method from bsp_start() in the bspstart.c file which sets up this table to store the ISR addresses, whose size is based on the definition of macros, BSP_INTERRUPT_VECTOR_MIN and BSP_INTERRUPT_VECTOR_MAX in include/bsp.h

For the generic handler table to properly function, some bsp specific code is required, that should be present in irq/irq.c. The bsp-specific functions required to be writen by the BSP developer are :

• bsp_interrupt_facility_initialize() contains bsp specific interrupt initialization code(Clear Pending interrupts by modifying registers, etc.). This method is called from bsp_interrupt_initialize() internally while setting up the table.

• bsp_interrupt_handler_default() acts as a fallback handler when no ISR address has been provided corresponding to a vector in the table.

• bsp_interrupt_dispatch() services the ISR by handling any bsp specific code & calling the generic method bsp_interrupt_handler_dispatch() which in turn services the interrupt by running the ISR after looking it up in the table. It acts as an entry to the interrupt switchboard, since the bsp branches to this function at the time of occurrence of an interrupt.

• bsp_interrupt_vector_enable() enables interrupts and is called in irq-generic.c while setting up the table.

• bsp_interrupt_vector_disable() disables interrupts and is called in irq-generic.c while setting up the table & during other important parts.

An interrupt handler is installed or removed with the help of the following functions :

rtems_status_code rtems_interrupt_handler_install(   /* returns status code */
  rtems_vector_number     vector,                    /* interrupt vector */
  const char             *info,                      /* custom identification text */
  rtems_option            options,                   /* Type of Interrupt */
  rtems_interrupt_handler handler,                   /* interrupt handler */
  void                   *arg                        /* parameter to be passed
                                                        to handler at the time of
                                                        invocation */
)
rtems_status_code rtems_interrupt_handler_remove(   /* returns status code */
  rtems_vector_number     vector,                   /* interrupt vector */
  rtems_interrupt_handler handler,                  /* interrupt handler */
  void                   *arg                       /* parameter to be passed to handler */
)

5.1. Introduction#

The system initialization consists of a low-level initialization performed by the start code in the start file (start.o) and a high-level initialization carried out by boot_card(). The final step of a successful high-level initialization is to switch to the initialization task and change into the normal system mode with multi-threading enabled. Errors during system initialization are fatal and end up in a call to _Terminate().

5.2. Low-Level Initialization via Start Code in the Start File (start.o)#

The start code in the start file (start.o) must be provided by the BSP. It is the first file presented to the linker and starts the process to link an executable (application image). It should contain the entry symbol of the executable. It is the responsibility of the linker script in conjunction with the compiler specifications file or compiler options to put the start code in the correct location in the executable. The start code is typically written in assembly language since it will tinker with the stack pointer. The general rule of thumb is that the start code in assembly language should do the minimum necessary to allow C code to execute to complete the initialization sequence.

The low-level system initialization may depend on a platform initialization carried out by a boot loader. The low-level system initialization may perform the following steps:

• Initialize the initialization stack. The initialization stack should use the ISR stack area. The symbols _ISR_Stack_area_begin, _ISR_Stack_area_end, and _ISR_Stack_size should be used to do this.

• Initialize processor registers and modes.

• Initialize pins.

• Initialize clocks (PLLs).

• Initialize memory controllers.

• Initialize instruction, data, and unified caches.

• Initialize memory management or protection units (MMU).

• Initialize processor exceptions.

• Copy the data sections from a read-only section to the runtime location.

• Set the BSS (.bss) section to zero.

• Initialize the C runtime environment.

• Call boot_card() to hand over to the high-level initialization.

For examples of start file codes see:

• bsps/arm/shared/start/start.S

• bsps/riscv/shared/start/start.S

5.3. High-Level Initialization via boot_card()#

The high-level initialization is carried out by boot_card(). For the high-level initialization steps see the Initialization Manager chapter in the RTEMS Classic API Guide. There are several system initialization steps which must be implemented by the BSP.

5.4. Error Handling#

Errors during system initialization are fatal and end up in a call to _Terminate(). See also the Fatal Error Manager chapter in the RTEMS Classic API Guide.

The BSP may use BSP-specific fatal error codes, see <bsp/fatal.h>.

The BSP should provide an initial extension which implements a fatal error handler. It should use the default implementation provided by <bsp/default-initial-extension.h> and bspfatal-default.c. If the default implementation is used, the BSP must implement a bsp_reset() function which should reset the platform.

6.1. Introduction#

This chapter describes the operation of a console driver using the RTEMS POSIX Termios support. Traditionally, RTEMS has referred to all serial device drivers as console drivers. Termios is defined by IEEE Std 1003.1-2008 (POSIX.1-2008). It supports various modes of operations at application level. This chapter focuses on the low-level serial device driver. Additional Termios information can be found in the Linux TERMIOS(3) manpage or the FreeBSD TERMIOS(4) manpage.

There are the following software layers.

In the default application configuration RTEMS opens during system initialization a /dev/console device file to create the file descriptors 0, 1 and 2 used for standard input, output and error, respectively. The corresponding device driver is usually a Termios serial device driver described here. The standard file descriptors are used by standard C library calls such as printf() or scanf() or directly via the read() or write() system calls.

6.2. Build System and Files#

A new serial device driver should consist of three parts.

• A section in the BSPs Makefile.am:

[...]
libbsp_a_SOURCES += ../../shared/dev/serial/console-termios.c
libbsp_a_SOURCES += console/console.c
[...]

• A general serial device specific low-level driver providing the handler table and the device context specialization for the Termios rtems_termios_device_install() function. This low-level driver could be used for more than one BSP.

• A BSP-specific initialization routine console_initialize(), that calls rtems_termios_device_install() providing a low-level driver context for each installed device. This is usually defined in the file console/console.c relative to the BSP base directory.

The low-level driver should provide a specialization of the Termios device context. The initialization routine must provide a context for each installed device via rtems_termios_device_install(). Here is an example header file for a low-level serial device driver.

#ifndef MY_DRIVER_H
#define MY_DRIVER_H

#include <some-chip/serial.h>

#include <rtems/termiostypes.h>

/* My low-level driver specialization of Termios device context */
typedef struct {
  rtems_termios_device_context base;
  const char *device_name;
  volatile some_chip_registers *regs;
  /* More stuff */
} my_driver_context;

extern const rtems_termios_device_handler my_driver_handler_polled;

extern const rtems_termios_device_handler my_driver_handler_interrupt;

#endif /* MY_DRIVER_H */

6.3. Driver Functioning Modes#

There are four main functioning modes for a Termios serial device driver. The mode must be set during device creation and cannot be changed afterwards.

Polled Mode (TERMIOS_POLLED)

In polled mode, the processor blocks on sending/receiving characters. This mode is not the most efficient way to utilize the serial device. But polled mode is usually necessary when one wants to print an error message in the event of a fatal error such as a fatal error in the BSP. This is also the simplest mode to program. Polled mode is generally preferred if the serial device is to be used primarily as a debug console. In a simple polled driver, the software will continuously check the status of the serial device when it is reading or writing to the serial device. Termios improves on this by delaying the caller for one clock tick between successive checks of the serial device on a read operation.

Interrupt Driven Mode (TERMIOS_IRQ_DRIVEN)

In interrupt driven mode, the processor does not block on sending/receiving characters. Data is buffered between the interrupt service routine and application code. Two buffers are used to insulate the application from the relative slowness of the serial device. One of the buffers is used for incoming characters, while the other is used for outgoing characters.

An interrupt is raised when a character is received by the serial device. The interrupt routine places the incoming character at the end of the input buffer. When an application asks for input, the characters at the front of the buffer are returned.

When the application prints to the serial device, the outgoing characters are placed at the end of the output buffer. The driver will place one or more characters in the serial device (the exact number depends on the serial device) An interrupt will be raised when all the characters have been transmitted. The interrupt service routine has to send the characters remaining in the output buffer the same way. When the transmitting side of the serial device is idle, it is typically necessary to prime the transmitter before the first interrupt will occur.

Interrupt Server Driven Mode (TERMIOS_IRQ_SERVER_DRIVEN)

The interrupt server driven mode is identical to the interrupt driven mode, except that a mutex is used to protect the low-level device state instead of an interrupt lock (disabled interrupts). Use this mode in case the serial device is connected via I2C or SPI and the I2C or SPI framework is used.

Task Driven Mode (TERMIOS_TASK_DRIVEN)

The task driven mode is similar to interrupt driven mode, but the actual data processing is done in dedicated tasks instead of interrupt routines. This mode is not available in SMP configurations. It has some implementation flaws and it is not well tested.

6.4. Polled Mode#

The handler table for the polled mode should look like the following.

const rtems_termios_device_handler my_driver_handler_polled = {
  .first_open = my_driver_first_open,
  .last_close = my_driver_last_close,
  .poll_read = my_driver_poll_read,
  .write = my_driver_poll_write,
  .set_attributes = my_driver_set_attributes,
  .ioctl = my_driver_ioctl, /* optional, may be NULL */
  .mode = TERMIOS_POLLED
};

The my_driver_poll_write() routine is responsible for writing n characters from buf to the serial device specified by base.

static void my_driver_poll_write(
  rtems_termios_device_context *base,
  const char                   *buf,
  size_t                        n
)
{
  my_driver_context *ctx;
  size_t             i;

  ctx = (my_driver_context *) base;

  for ( i = 0 ; i < n ; ++i ) {
    my_driver_write_char( ctx, buf[ i ] );
  }
}

The my_driver_poll_read() routine is responsible for reading a single character from the serial device specified by base. If no character is available, then the routine should immediately return minus one.

static int my_driver_poll_read( rtems_termios_device_context *base )
{
  my_driver_context *ctx;

  ctx = (my_driver_context *) base;

  if ( my_driver_can_read_char( ctx ) ) {
    /* Return the character (must be unsigned) */
    return my_driver_read_char( ctx );
  } else {
    /* Return -1 to indicate that no character is available */
    return -1;
  }
}

6.5. Interrupt Driven Mode#

The handler table for the interrupt driven mode should look like the following.

const rtems_termios_device_handler my_driver_handler_interrupt = {
  .first_open = my_driver_first_open,
  .last_close = my_driver_last_close,
  .poll_read = NULL,
  .write = my_driver_interrupt_write,
  .set_attributes = my_driver_set_attributes,
  .ioctl = my_driver_ioctl, /* optional, may be NULL */
  .mode = TERMIOS_IRQ_DRIVEN
};

There is no device driver read handler to be passed to Termios. Indeed a read() call returns the contents of Termios input buffer. This buffer is filled in the driver interrupt routine.

A serial device generally generates interrupts when it is ready to accept or to emit a number of characters. In this mode, the interrupt routine is the core of the driver.

The my_driver_interrupt_handler() is responsible for processing asynchronous interrupts from the serial device. There may be multiple interrupt handlers for a single serial device. Some serial devices can generate a unique interrupt vector for each interrupt source such as a character has been received or the transmitter is ready for another character.

In the simplest case, the my_driver_interrupt_handler() will have to check the status of the serial device and determine what caused the interrupt. The following describes the operation of an my_driver_interrupt_handler() which has to do this:

static void my_driver_interrupt_handler( void *arg )
{
  rtems_termios_tty *tty;
  my_driver_context *ctx;
  char               buf[N];
  size_t             n;

  tty = arg;
  ctx = rtems_termios_get_device_context( tty );

  /*
   * Check if we have received something.  The function reads the
   * received characters from the device and stores them in the
   * buffer.  It returns the number of read characters.
   */
  n = my_driver_read_received_chars( ctx, buf, N );
  if ( n > 0 ) {
    /* Hand the data over to the Termios infrastructure */
    rtems_termios_enqueue_raw_characters( tty, buf, n );
  }

  /*
   * Check if we have something transmitted.  The functions returns
   * the number of transmitted characters since the last write to the
   * device.
   */
  n = my_driver_transmitted_chars( ctx );
  if ( n > 0 ) {
    /*
     * Notify Termios that we have transmitted some characters.  It
     * will call now the interrupt write function if more characters
     * are ready for transmission.
     */
    rtems_termios_dequeue_characters( tty, n );
  }
}

The my_driver_interrupt_write() handler is responsible for telling the device that the n characters at buf are to be transmitted. It the value n is zero to indicate that no more characters are to send. The driver can disable the transmit interrupts now. This routine is invoked either from task context with disabled interrupts to start a new transmission process with exactly one character in case of an idle output state or from the interrupt handler to refill the transmitter. If the routine is invoked to start the transmit process the output state will become busy and Termios starts to fill the output buffer. If the transmit interrupt arises before Termios was able to fill the transmit buffer you will end up with one interrupt per character.

static void my_driver_interrupt_write(
  rtems_termios_device_context  *base,
  const char                    *buf,
  size_t                         n
)
{
  my_driver_context *ctx;

  ctx = (my_driver_context *) base;

  if ( n > 0 ) {
    /*
     * Tell the device to transmit some characters from buf (less than
     * or equal to n).  When the device is finished it should raise an
     * interrupt.  The interrupt handler will notify Termios that these
     * characters have been transmitted and this may trigger this write
     * function again.  You may have to store the number of outstanding
     * characters in the device data structure.
     */
  } else {
    /*
     * Termios will set n to zero to indicate that the transmitter is
     * now inactive.  The output buffer is empty in this case.  The
     * driver may disable the transmit interrupts now.
     */
  }
}

6.6. First Open#

Upon first open of the device, the my_driver_first_open() handler is called by Termios. The device registered as /dev/console (or CONSOLE_DEVICE_NAME) is opened automatically during RTEMS initialization.

static bool my_driver_first_open(
  rtems_termios_tty             *tty,
  rtems_termios_device_context  *base,
  struct termios                *term,
  rtems_libio_open_close_args_t *args
)
{
  my_driver_context *ctx;
  rtems_status_code  sc;
  bool               ok;

  ctx = (my_driver_context *) base;

  /*
   * You may add some initialization code here.
   */

  /*
   * Sets the initial baud rate.  This should be set to the value of
   * the boot loader.  This function accepts only exact Termios baud
   * values.
   */
  sc = rtems_termios_set_initial_baud( tty, MY_DRIVER_BAUD_RATE );
  if ( sc != RTEMS_SUCCESSFUL ) {
    /* Not a valid Termios baud */
  }

  /*
   * Alternatively you can set the best baud.
   */
  rtems_termios_set_best_baud( term, MY_DRIVER_BAUD_RATE );

  /*
   * To propagate the initial Termios attributes to the device use
   * this.
  */
  ok = my_driver_set_attributes( base, term );
  if ( !ok ) {
    /* This is bad */
  }

  /*
   * Return true to indicate a successful set attributes, and false
   * otherwise.
   */
  return true;
}

6.7. Last Close#

Termios will call the my_driver_last_close() handler if the last close happens on the device.

static void my_driver_last_close(
  rtems_termios_tty             *tty,
  rtems_termios_device_context  *base,
  rtems_libio_open_close_args_t *args
)
{
  my_driver_context *ctx;

  ctx = (my_driver_context *) base;

  /*
   * The driver may do some cleanup here.
   */
}

6.8. Set Attributes#

Termios will call the my_driver_set_attributes() handler if a serial line configuration parameter changed, e.g. baud, character size, number of stop bits, parity, etc.

static bool my_driver_set_attributes(
  rtems_termios_device_context *base,
  const struct termios         *term
)
{
  my_driver_context *ctx;

  ctx = (my_driver_context *) base;

  /*
   * Inspect the termios data structure and configure the device
   * appropriately.  The driver should only be concerned with the
   * parts of the structure that specify hardware setting for the
   * communications channel such as baud, character size, etc.
   */

  /*
   * Return true to indicate a successful set attributes, and false
   * otherwise.
   */
  return true;
}

6.9. IO Control#

Optionally, the my_driver_ioctl() routine may be provided for arbitrary device-specific functions.

static int my_driver_ioctl(
  rtems_termios_device_context *base,
  ioctl_command_t               request,
  void                         *buffer
)
{
  my_driver_context *ctx;

  ctx = (my_driver_context *) base;

  switch ( request ) {
    case MY_DRIVER_DO_XYZ:
      my_driver_do_xyz(ctx, buffer);
      break;
    default:
      rtems_set_errno_and_return_minus_one( EINVAL );
  }

  return 0;
}

6.10. Flow Control#

You can also provide handler for remote transmission control. This is not covered in this manual.

6.11. General Initialization#

The BSP-specific driver initialization is called once during the RTEMS initialization process.

The console_initialize() function may look like this:

#include <my-driver.h>

#include <rtems/console.h>

#include <bsp.h>
#include <bsp/fatal.h>

static my_driver_context driver_context_table[] = {
  { /* Some values for device 0 */ },
  { /* Some values for device 1 */ }
};

rtems_device_driver console_initialize(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg
)
{
  const rtems_termios_device_handler *handler;
  rtems_status_code                   sc;
  size_t                              i;

  #ifdef SOME_BSP_USE_INTERRUPTS
    handler = &my_driver_handler_interrupt;
  #else
    handler = &my_driver_handler_polled;
  #endif

  /*
   * Initialize the Termios infrastructure.  If Termios has already
   * been initialized by another device driver, then this call will
   * have no effect.
   */
  rtems_termios_initialize();

  /* Initialize each device */
  for ( i = 0; i < RTEMS_ARRAY_SIZE( driver_context_table ) ; ++i ) {
    my_driver_context *ctx;

    ctx = &driver_context_table[ i ];

    /*
     * Install this device in the file system and Termios.  In order
     * to use the console (i.e. being able to do printf, scanf etc.
     * on stdin, stdout and stderr), one device must be registered as
     * "/dev/console" (CONSOLE_DEVICE_NAME).
     */
    sc = rtems_termios_device_install( ctx->device_name, handler, NULL, ctx );
    if ( sc != RTEMS_SUCCESSFUL ) {
      bsp_fatal( SOME_BSP_FATAL_CONSOLE_DEVICE_INSTALL );
    }
  }

  return RTEMS_SUCCESSFUL;
}

7.1. Introduction#

The purpose of the clock driver is to provide two services for the operating system.

• A steady time basis to the kernel, so that the RTEMS primitives that need a clock tick work properly. See the Clock Manager chapter of the RTEMS Application C User’s Guide for more details.

• An optional timecounter to provide timestamps of the uptime and wall clock time with higher resolution than the clock tick.

The clock driver is usually located in the clock directory of the BSP. Clock drivers must use the Clock Driver Shell available via the clockimpl.h include file. This include file is not a normal header file and instead defines the clock driver functions declared in #include <rtems/clockdrv.h> which are used by RTEMS configuration file #include <rtems/confdefs.h>. In case the application configuration defines #define CONFIGURE_APPLICATION_NEEDS_CLOCK_DRIVER, then the clock driver is registered and should provide its services to the operating system. The clock tick interval is determined by the application configuration via #define CONFIGURE_MICROSECONDS_PER_TICK and can be obtained via rtems_configuration_get_microseconds_per_tick().

A hardware-specific clock driver must provide some functions, defines and macros for the Clock Driver Shell which are explained here step by step. A clock driver file looks in general like this.

/*
 * A section with functions, defines and macros to provide hardware-specific
 * functions for the Clock Driver Shell.
 */

#include "../../../shared/dev/clock/clockimpl.h"

Depending on the hardware capabilities one out of three clock driver variants must be selected.

Timecounter

The variant which provides all features needs a free running hardware counter and a periodic clock tick interrupt. This variant is mandatory in SMP configurations.

Simple Timecounter

A simple timecounter can be used if the hardware provides no free running hardware counter and only a periodic hardware counter synchronous to the clock tick interrupt is available.

Clock Tick Only

The most basic clock driver provides only a periodic clock tick interrupt. The timestamp resolution is limited to the clock tick interval.

7.2. Initialization#

The clock driver is initialized by the _Clock_Initialize() system initialization handler if requested by the application configuration option CONFIGURE_APPLICATION_NEEDS_CLOCK_DRIVER. The clock driver does not use the legacy IO driver framework.

#include <rtems/timecounter.h>

static struct timecounter some_tc;

static uint32_t some_tc_get_timecount( struct timecounter *tc )
{
  some.free_running_counter;
}

static void some_support_initialize_hardware( void )
{
  uint64_t us_per_tick;
  uint32_t counter_frequency_in_hz;
  uint32_t counter_ticks_per_clock_tick;

  us_per_tick = rtems_configuration_get_microseconds_per_tick();
  counter_frequency_in_hz = some_tc_get_frequency();

  /*
   * The multiplication must be done in 64-bit arithmetic to avoid an integer
   * overflow on targets with a high enough counter frequency.
   */
  counter_ticks_per_clock_tick =
    (uint32_t) ( counter_frequency_in_hz * us_per_tick ) / 1000000;

  /*
   * Initialize hardware and set up a periodic interrupt for the configuration
   * based counter ticks per clock tick.
   */

  some_tc.tc_get_timecount = some_tc_get_timecount;
  some_tc.tc_counter_mask = 0xffffffff;
  some_tc.tc_frequency = frequency;
  some_tc.tc_quality = RTEMS_TIMECOUNTER_QUALITY_CLOCK_DRIVER;
  rtems_timecounter_install( &some_tc );
}

#define Clock_driver_support_initialize_hardware() \
  some_support_initialize_hardware()

#include "../../../shared/dev/clock/clockimpl.h"

#include <rtems/timecounter.h>

static rtems_timecounter_simple some_tc;

static uint32_t some_tc_get( rtems_timecounter_simple *tc )
{
  return some.counter;
}

static bool some_tc_is_pending( rtems_timecounter_simple *tc )
{
  return some.is_pending;
}

static uint32_t some_tc_get_timecount( struct timecounter *tc )
{
  return rtems_timecounter_simple_downcounter_get(
    tc,
    some_tc_get,
    some_tc_is_pending
  );
}

static void some_tc_tick( rtems_timecounter_simple *tc )
{
  rtems_timecounter_simple_downcounter_tick( tc, some_tc_get );
}

static void some_support_initialize_hardware( void )
{
  uint64_t us_per_tick;
  uint32_t counter_frequency_in_hz;
  uint32_t counter_ticks_per_clock_tick;

  us_per_tick = rtems_configuration_get_microseconds_per_tick();
  counter_frequency_in_hz = some_tc_get_frequency();
  counter_ticks_per_clock_tick =
    (uint32_t) ( counter_frequency_in_hz * us_per_tick ) / 1000000;

  /* Initialize hardware */

  rtems_timecounter_simple_install(
    &some_tc,
    counter_frequency_in_hz,
    counter_ticks_per_clock_tick,
    some_tc_get_timecount
  );
}

#define Clock_driver_support_initialize_hardware() \
  some_support_initialize_hardware()
#define Clock_driver_timecounter_tick( arg ) \
  some_tc_tick( arg )

#include "../../../shared/dev/clock/clockimpl.h"

static void some_support_initialize_hardware( void )
{
  /* Initialize hardware */
}

#define Clock_driver_support_initialize_hardware() \
  some_support_initialize_hardware()

/* Indicate that this clock driver lacks a proper timecounter in hardware */

#define CLOCK_DRIVER_USE_DUMMY_TIMECOUNTER

#include "../../../shared/dev/clock/clockimpl.h"

7.3. Install Clock Tick Interrupt Service Routine#

The clock driver may provide a function to install the clock tick interrupt service routine via Clock_driver_support_install_isr( isr ). The clock tick interrupt service routine is passed as the one and only parameter to this macro. The default implementation will do nothing. The argument parameter (in the code below &some_instance) for the installed interrupt handler is available in the Clock_driver_support_at_tick( arg ) and Clock_driver_support_initialize_hardware( arg ) customization macros.

#include <bsp/irq.h>
#include <bsp/fatal.h>

static some_control some_instance;

static void some_support_install_isr( rtems_interrupt_handler isr )
{
  rtems_status_code sc;
  sc = rtems_interrupt_handler_install(
    SOME_IRQ,
    "Clock",
    RTEMS_INTERRUPT_UNIQUE,
    isr,
    &some_instance
  );
  if ( sc != RTEMS_SUCCESSFUL ) {
    bsp_fatal( SOME_FATAL_IRQ_INSTALL );
  }
}

#define Clock_driver_support_install_isr( isr ) \
  some_support_install_isr( isr )

#include "../../../shared/dev/clock/clockimpl.h"

7.4. Support At Tick#

The hardware-specific support at tick is specified by Clock_driver_support_at_tick( arg ). The arg is the argument used to install the clock interrupt handler. Device drivers may use this argument to access their control state.

static void some_support_at_tick( some_control *arg )
{
  /* Clear interrupt */
}

#define Clock_driver_support_at_tick( arg ) \
  some_support_at_tick( arg )

#include "../../../shared/dev/clock/clockimpl.h"

7.5. System Shutdown Support#

The clock driver system shutdown support was removed in RTEMS 5.1.

7.6. SMP Support#

In SMP configurations, the clock tick service must be executed for each processor used by RTEMS. By default, the clock tick interrupt must be distributed to all processors used by RTEMS and each processor invokes the clock tick service individually. A clock driver may delegate all the work to the boot processor. It must define CLOCK_DRIVER_USE_ONLY_BOOT_PROCESSOR in this case.

Clock drivers must define Clock_driver_support_set_interrupt_affinity(online_processors) to set the interrupt affinity of the clock tick interrupt.

7.7. Multiple Clock Driver Ticks Per Clock Tick#

In case the hardware needs more than one clock driver tick per clock tick (e.g. due to a limited range of the hardware timer), then this can be specified with the optional #define CLOCK_DRIVER_ISRS_PER_TICK and #define CLOCK_DRIVER_ISRS_PER_TICK_VALUE defines. This is currently used only for x86 and it hopefully remains that way.

/* Enable multiple clock driver ticks per clock tick */
#define CLOCK_DRIVER_ISRS_PER_TICK 1

/* Specifiy the clock driver ticks per clock tick value */
#define CLOCK_DRIVER_ISRS_PER_TICK_VALUE 123

#include "../../../shared/dev/clock/clockimpl.h"

7.8. Clock Driver Ticks Counter#

The Clock Driver Shell provide a global variable that is simply a count of the number of clock driver interrupt service routines that have occurred. This information is valuable when debugging a system. This variable is declared as follows:

volatile uint32_t Clock_driver_ticks;

12.1. Introduction#

The Real-Time Clock (RTC) driver is responsible for providing an interface to an RTC device. The capabilities provided by this driver are:

• Set the RTC TOD to RTEMS TOD

• Set the RTEMS TOD to the RTC TOD

• Get the RTC TOD

• Set the RTC TOD to the Specified TOD

• Get the Difference Between the RTEMS and RTC TOD

The reference implementation for a real-time clock driver can be found in bsps/shared/dev/rtc/rtc-support.c. This driver is based on the libchip concept and can be easily configured to work with any of the RTC chips supported by the RTC chip drivers in the directory bsps/shared/dev/rtc. There is a README file in this directory for each supported RTC chip. Each of these README explains how to configure the shared libchip implementation of the RTC driver for that particular RTC chip.

The DY-4 DMV177 BSP used the shared libchip implementation of the RTC driver. There were no DMV177 specific configuration routines. A BSP could use configuration routines to dynamically determine what type of real-time clock is on a particular board. This would be useful for a BSP supporting multiple board models. The relevant ports of the DMV177’s RTC_Table configuration table is below:

#include <bsp.h>
#include <libchip/rtc.h>
#include <libchip/icm7170.h>

bool dmv177_icm7170_probe(int minor);

rtc_tbl RTC_Table[] = {
  { "/dev/rtc0",                 /* sDeviceName */
     RTC_ICM7170,                /* deviceType */
     &icm7170_fns,               /* pDeviceFns */
     dmv177_icm7170_probe,       /* deviceProbe */
     (void *) ICM7170_AT_1_MHZ,  /* pDeviceParams */
     DMV170_RTC_ADDRESS,         /* ulCtrlPort1 */
     0,                          /* ulDataPort */
     icm7170_get_register_8,     /* getRegister */
     icm7170_set_register_8,     /* setRegister */
  }
};
unsigned long RTC_Count = (sizeof(RTC_Table)/sizeof(rtc_tbl));
rtems_device_minor_number RTC_Minor;

bool dmv177_icm7170_probe(int minor)
{
  volatile uint16_t *card_resource_reg;
  card_resource_reg = (volatile uint16_t *) DMV170_CARD_RESORCE_REG;
  if ( (*card_resource_reg & DMV170_RTC_INST_MASK) == DMV170_RTC_INSTALLED )
    return TRUE;
  return FALSE;
}

12.2. Initialization#

The rtc_initialize routine is responsible for initializing the RTC chip so it can be used. The shared libchip implementation of this driver supports multiple RTCs and bases its initialization order on the order the chips are defined in the RTC_Table. Each chip defined in the table may or may not be present on this particular board. It is the responsibility of the deviceProbe to indicate the presence of a particular RTC chip. The first RTC found to be present is considered the preferred RTC.

In the shared libchip based implementation of the driver, the following actions are performed:

rtems_device_driver rtc_initialize(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor_arg,
  void                      *arg
)
{
  for each RTC configured in RTC_Table
    if the deviceProbe for this RTC indicates it is present
      set RTC_Minor to this device
      set RTC_Present to TRUE
      break out of this loop

    if RTC_Present is not TRUE
      return RTEMS_INVALID_NUMBER to indicate that no RTC is present

    register this minor number as the "/dev/rtc"

    perform the deviceInitialize routine for the preferred RTC chip

    for RTCs past this one in the RTC_Table
      if the deviceProbe for this RTC indicates it is present
        perform the deviceInitialize routine for this RTC chip
        register the configured name for this RTC
}

The deviceProbe routine returns TRUE if the device configured by this entry in the RTC_Table is present. This configuration scheme allows one to support multiple versions of the same board with a single BSP. For example, if the first generation of a board had Vendor A’s RTC chip and the second generation had Vendor B’s RTC chip, RTC_Table could contain information for both. The deviceProbe configured for Vendor A’s RTC chip would need to return TRUE if the board was a first generation one. The deviceProbe routines are very board dependent and must be provided by the BSP.

12.3. setRealTimeToRTEMS#

The setRealTimeToRTEMS routine sets the current RTEMS TOD to that of the preferred RTC.

void setRealTimeToRTEMS(void)
{
  if no RTCs are present
    return

  invoke the deviceGetTime routine for the preferred RTC
  set the RTEMS TOD using rtems_clock_set
}

12.4. setRealTimeFromRTEMS#

The setRealTimeFromRTEMS routine sets the preferred RTC TOD to the current RTEMS TOD.

void setRealTimeFromRTEMS(void)
{
  if no RTCs are present
    return

  obtain the RTEMS TOD using rtems_clock_get
  invoke the deviceSetTime routine for the preferred RTC
}

12.5. getRealTime#

The getRealTime returns the preferred RTC TOD to the caller.

void getRealTime( rtems_time_of_day *tod )
{
  if no RTCs are present
  return

  invoke the deviceGetTime routine for the preferred RTC
}

12.6. setRealTime#

The setRealTime routine sets the preferred RTC TOD to the TOD specified by the caller.

void setRealTime( rtems_time_of_day *tod )
{
  if no RTCs are present
    return

  invoke the deviceSetTime routine for the preferred RTC
}

12.7. checkRealTime#

The checkRealTime routine returns the number of seconds difference between the RTC TOD and the current RTEMS TOD.

int checkRealTime( void )
{
  if no RTCs are present
    return -1

  obtain the RTEMS TOD using rtems_clock_get
  get the TOD from the preferred RTC using the deviceGetTime routine
  convert the RTEMS TOD to seconds
  convert the RTC TOD to seconds

  return the RTEMS TOD in seconds - RTC TOD in seconds
}

13.1. Introduction#

This chapter is intended to provide an introduction to the procedure for writing RTEMS network device drivers. The example code is taken from the ‘Generic 68360’ network device driver. The source code for this driver is located in the bsps/m68k/gen68360/net directory in the RTEMS source code distribution. Having a copy of this driver at hand when reading the following notes will help significantly.

13.2. Learn about the network device#

Before starting to write the network driver become completely familiar with the programmer’s view of the device. The following points list some of the details of the device that must be understood before a driver can be written.

• Does the device use DMA to transfer packets to and from memory or does the processor have to copy packets to and from memory on the device?

• If the device uses DMA, is it capable of forming a single outgoing packet from multiple fragments scattered in separate memory buffers?

• If the device uses DMA, is it capable of chaining multiple outgoing packets, or does each outgoing packet require intervention by the driver?

• Does the device automatically pad short frames to the minimum 64 bytes or does the driver have to supply the padding?

• Does the device automatically retry a transmission on detection of a collision?

• If the device uses DMA, is it capable of buffering multiple packets to memory, or does the receiver have to be restarted after the arrival of each packet?

• How are packets that are too short, too long, or received with CRC errors handled? Does the device automatically continue reception or does the driver have to intervene?

• How is the device Ethernet address set? How is the device programmed to accept or reject broadcast and multicast packets?

• What interrupts does the device generate? Does it generate an interrupt for each incoming packet, or only for packets received without error? Does it generate an interrupt for each packet transmitted, or only when the transmit queue is empty? What happens when a transmit error is detected?

In addition, some controllers have specific questions regarding board specific configuration. For example, the SONIC Ethernet controller has a very configurable data bus interface. It can even be configured for sixteen and thirty-two bit data buses. This type of information should be obtained from the board vendor.

13.3. Understand the network scheduling conventions#

When writing code for the driver transmit and receive tasks, take care to follow the network scheduling conventions. All tasks which are associated with networking share various data structures and resources. To ensure the consistency of these structures the tasks execute only when they hold the network semaphore (rtems_bsdnet_semaphore). The transmit and receive tasks must abide by this protocol. Be very careful to avoid ‘deadly embraces’ with the other network tasks. A number of routines are provided to make it easier for the network driver code to conform to the network task scheduling conventions.

• void rtems_bsdnet_semaphore_release(void) This function releases the network semaphore. The network driver tasks must call this function immediately before making any blocking RTEMS request.

• void rtems_bsdnet_semaphore_obtain(void) This function obtains the network semaphore. If a network driver task has released the network semaphore to allow other network-related tasks to run while the task blocks, then this function must be called to reobtain the semaphore immediately after the return from the blocking RTEMS request.

• rtems_bsdnet_event_receive(rtems_event_set, rtems_option, rtems_interval, rtems_event_set *) The network driver task should call this function when it wishes to wait for an event. This function releases the network semaphore, calls rtems_event_receive to wait for the specified event or events and reobtains the semaphore. The value returned is the value returned by the rtems_event_receive.

13.4. Network Driver Makefile#

Network drivers are considered part of the BSD network package and as such are to be compiled with the appropriate flags. This can be accomplished by adding -D__INSIDE_RTEMS_BSD_TCPIP_STACK__ to the command line. If the driver is inside the RTEMS source tree or is built using the RTEMS application Makefiles, then adding the following line accomplishes this:

DEFINES += -D__INSIDE_RTEMS_BSD_TCPIP_STACK__

This is equivalent to the following list of definitions. Early versions of the RTEMS BSD network stack required that all of these be defined.

-D_COMPILING_BSD_KERNEL_ -DKERNEL -DINET -DNFS -DDIAGNOSTIC -DBOOTP_COMPAT

Defining these macros tells the network header files that the driver is to be compiled with extended visibility into the network stack. This is in sharp contrast to applications that simply use the network stack. Applications do not require this level of visibility and should stick to the portable application level API.

As a direct result of being logically internal to the network stack, network drivers use the BSD memory allocation routines This means, for example, that malloc takes three arguments. See the SONIC device driver (c/src/lib/libchip/network/sonic.c) for an example of this. Because of this, network drivers should not include <stdlib.h>. Doing so will result in conflicting definitions of malloc().

Application level code including network servers such as the FTP daemon are not part of the BSD kernel network code and should not be compiled with the BSD network flags. They should include <stdlib.h> and not define the network stack visibility macros.

13.5. Write the Driver Attach Function#

The driver attach function is responsible for configuring the driver and making the connection between the network stack and the driver.

Driver attach functions take a pointer to an rtems_bsdnet_ifconfig structure as their only argument. and set the driver parameters based on the values in this structure. If an entry in the configuration structure is zero the attach function chooses an appropriate default value for that parameter.

The driver should then set up several fields in the ifnet structure in the device-dependent data structure supplied and maintained by the driver:

ifp->if_softc

Pointer to the device-dependent data. The first entry in the device-dependent data structure must be an arpcom structure.

ifp->if_name

The name of the device. The network stack uses this string and the device number for device name lookups. The device name should be obtained from the name entry in the configuration structure.

ifp->if_unit

The device number. The network stack uses this number and the device name for device name lookups. For example, if ifp->if_name is scc and ifp->if_unit is 1, the full device name would be scc1. The unit number should be obtained from the name entry in the configuration structure.

ifp->if_mtu

The maximum transmission unit for the device. For Ethernet devices this value should almost always be 1500.

ifp->if_flags

The device flags. Ethernet devices should set the flags to IFF_BROADCAST|IFF_SIMPLEX, indicating that the device can broadcast packets to multiple destinations and does not receive and transmit at the same time.

ifp->if_snd.ifq_maxlen

The maximum length of the queue of packets waiting to be sent to the driver. This is normally set to ifqmaxlen.

ifp->if_init

The address of the driver initialization function.

ifp->if_start

The address of the driver start function.

ifp->if_ioctl

The address of the driver ioctl function.

ifp->if_output

The address of the output function. Ethernet devices should set this to ether_output.

RTEMS provides a function to parse the driver name in the configuration structure into a device name and unit number.

int rtems_bsdnet_parse_driver_name (
  const struct rtems_bsdnet_ifconfig  *config,
  char                               **namep
);

The function takes two arguments; a pointer to the configuration structure and a pointer to a pointer to a character. The function parses the configuration name entry, allocates memory for the driver name, places the driver name in this memory, sets the second argument to point to the name and returns the unit number. On error, a message is printed and -1 is returned.

Once the attach function has set up the above entries it must link the driver data structure onto the list of devices by calling if_attach. Ethernet devices should then call ether_ifattach. Both functions take a pointer to the device’s ifnet structure as their only argument.

The attach function should return a non-zero value to indicate that the driver has been successfully configured and attached.

13.6. Write the Driver Start Function.#

This function is called each time the network stack wants to start the transmitter. This occures whenever the network stack adds a packet to a device’s send queue and the IFF_OACTIVE bit in the device’s if_flags is not set.

For many devices this function need only set the IFF_OACTIVE bit in the if_flags and send an event to the transmit task indicating that a packet is in the driver transmit queue.

13.7. Write the Driver Initialization Function.#

This function should initialize the device, attach to interrupt handler, and start the driver transmit and receive tasks. The function:

rtems_id rtems_bsdnet_newproc(
  char *name,
  int   stacksize,
  void  (*entry)(void *),
  void *arg
);

should be used to start the driver tasks.

Note that the network stack may call the driver initialization function more than once. Make sure multiple versions of the receive and transmit tasks are not accidentally started.

13.8. Write the Driver Transmit Task#

This task is reponsible for removing packets from the driver send queue and sending them to the device. The task should block waiting for an event from the driver start function indicating that packets are waiting to be transmitted. When the transmit task has drained the driver send queue the task should clear the IFF_OACTIVE bit in if_flags and block until another outgoing packet is queued.

13.9. Write the Driver Receive Task#

This task should block until a packet arrives from the device. If the device is an Ethernet interface the function ether_input should be called to forward the packet to the network stack. The arguments to ether_input are a pointer to the interface data structure, a pointer to the ethernet header and a pointer to an mbuf containing the packet itself.

13.10. Write the Driver Interrupt Handler#

A typical interrupt handler will do nothing more than the hardware manipulation required to acknowledge the interrupt and send an RTEMS event to wake up the driver receive or transmit task waiting for the event. Network interface interrupt handlers must not make any calls to other network routines.

13.11. Write the Driver IOCTL Function#

This function handles ioctl requests directed at the device. The ioctl commands which must be handled are:

SIOCGIFADDR, SIOCSIFADDR

If the device is an Ethernet interface these commands should be passed on to ether_ioctl.

SIOCSIFFLAGS

This command should be used to start or stop the device, depending on the state of the interface IFF_UP and``IFF_RUNNING`` bits in if_flags:

IFF_RUNNING

Stop the device.

IFF_UP

Start the device.

IFF_UP|IFF_RUNNING

Stop then start the device.

0

Do nothing.

13.12. Write the Driver Statistic-Printing Function#

This function should print the values of any statistic/diagnostic counters the network driver may use. The driver ioctl function should call the statistic-printing function when the ioctl command is SIO_RTEMS_SHOW_STATS.

14.1. Introduction#

The purpose of the frame buffer driver is to provide an abstraction for graphics hardware. By using the frame buffer interface, an application can display graphics without knowing anything about the low-level details of interfacing to a particular graphics adapter. The parameters governing the mapping of memory to displayed pixels (planar or linear, bit depth, etc) is still implementation-specific, but device-independent methods are provided to determine and potentially modify these parameters.

The frame buffer driver is commonly located in the console directory of the BSP and registered by the name /dev/fb0. Additional frame buffers (if available) are named /dev/fb1*,*/dev/fb2, etc.

To work with the frame buffer, the following operation sequence is used:open(), ioctls() to get the frame buffer info, read() and/or write(), and close().

14.2. Driver Function Overview#

rtems_device_driver frame_buffer_initialize(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg)
{
  rtems_status_code status;

  printk( "frame buffer driver initializing..\n" );

  /*
   * Register the device
   */
  status = rtems_io_register_name("/dev/fb0", major, 0);
  if (status != RTEMS_SUCCESSFUL)
  {
    printk("Error registering frame buffer device!\n");
    rtems_fatal_error_occurred( status );
  }

  /*
   * graphics hardware initialization goes here for non-console
   * devices
   */

  return RTEMS_SUCCESSFUL;
}

rtems_device_driver frame_buffer_close(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg
)
{
  if (pthread_mutex_unlock(&mutex) == 0) {
    /* restore previous state.  for VGA this means return to text mode.
     * leave out if graphics hardware has been initialized in
     * frame_buffer_initialize()
     */
    ega_hwterm();
    printk( "FBVGA close called.\n" );
    return RTEMS_SUCCESSFUL;
  }
  return RTEMS_UNSATISFIED;
}

rtems_device_driver frame_buffer_close(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg)
{
  pthread_mutex_unlock(&mutex);

  /* TODO check mutex return value, RTEMS_UNSATISFIED if it failed.  we
   * don't want to unconditionally call ega_hwterm()... */
  /* restore previous state.  for VGA this means return to text mode.
   * leave out if graphics hardware has been initialized in
   * frame_buffer_initialize() */
  ega_hwterm();
  printk( "frame buffer close called.\n" );
  return RTEMS_SUCCESSFUL;
}

rtems_device_driver frame_buffer_read(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg
)
{
  rtems_libio_rw_args_t *rw_args = (rtems_libio_rw_args_t *)arg;
  rw_args->bytes_moved = ((rw_args->offset + rw_args->count) > fb_fix.smem_len ) ?
                           (fb_fix.smem_len - rw_args->offset) : rw_args->count;
  memcpy(rw_args->buffer,
         (const void *) (fb_fix.smem_start + rw_args->offset),
         rw_args->bytes_moved);
  return RTEMS_SUCCESSFUL;
}

rtems_device_driver frame_buffer_write(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg
)
{
  rtems_libio_rw_args_t *rw_args = (rtems_libio_rw_args_t *)arg;
  rw_args->bytes_moved = ((rw_args->offset + rw_args->count) > fb_fix.smem_len ) ?
                           (fb_fix.smem_len - rw_args->offset) : rw_args->count;
  memcpy((void *) (fb_fix.smem_start + rw_args->offset),
         rw_args->buffer,
         rw_args->bytes_moved);
  return RTEMS_SUCCESSFUL;
}

rtems_device_driver frame_buffer_control(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg
)
{
  rtems_libio_ioctl_args_t *args = arg;

  printk( "FBVGA ioctl called, cmd=%x\n", args->command  );

  switch( args->command ) {
    case FBIOGET_FSCREENINFO:
      args->ioctl_return =  get_fix_screen_info( ( struct fb_fix_screeninfo * ) args->buffer );
      break;
    case FBIOGET_VSCREENINFO:
      args->ioctl_return =  get_var_screen_info( ( struct fb_var_screeninfo * ) args->buffer );
      break;
    case FBIOPUT_VSCREENINFO:
      /* not implemented yet*/
      args->ioctl_return = -1;
      return RTEMS_UNSATISFIED;
    case FBIOGETCMAP:
      args->ioctl_return =  get_palette( ( struct fb_cmap * ) args->buffer );
      break;
    case FBIOPUTCMAP:
      args->ioctl_return =  set_palette( ( struct fb_cmap * ) args->buffer );
      break;
    default:
      args->ioctl_return = 0;
      break;
  }

  return RTEMS_SUCCESSFUL;
}

15.1. Introduction#

This chapter describes what is required to enable Ada interrupt and error exception handling when using GNAT over RTEMS.

The GNAT Ada95 interrupt support RTEMS was developed by Jiri Gaisler <jgais@ws.estec.esa.nl> who also wrote this chapter.

15.2. Mapping Interrupts to POSIX Signals#

In Ada95, interrupts can be attached with the interrupt_attach pragma. For most systems, the gnat run-time will use POSIX signal to implement the interrupt handling, mapping one signal per interrupt. For interrupts to be propagated to the attached Ada handler, the corresponding signal must be raised when the interrupt occurs.

The same mechanism is used to generate Ada error exceptions. Three error exceptions are defined: program, constraint and storage error. These are generated by raising the predefined signals: SIGILL, SIGFPE and SIGSEGV. These signals should be raised when a spurious or erroneous trap occurs.

To enable gnat interrupt and error exception support for a particular BSP, the following has to be done:

• Write an interrupt/trap handler that will raise the corresponding signal depending on the interrupt/trap number.

• Install the interrupt handler for all interrupts/traps that will be handled by gnat (including spurious).

• At startup, gnat calls __gnat_install_handler(). The BSP must provide this function which installs the interrupt/trap handlers.

Which CPU-interrupt will generate which signal is implementation defined. There are 32 POSIX signals (1 - 32), and all except the three error signals (SIGILL, SIGFPE and SIGSEGV) can be used. I would suggest to use the upper 16 (17 - 32) which do not have an assigned POSIX name.

Note that the pragma interrupt_attach will only bind a signal to a particular Ada handler - it will not unmask the interrupt or do any other things to enable it. This have to be done separately, typically by writing various device register.

15.3. Example Ada95 Interrupt Program#

An example program (irq_test) is included in the Ada examples package to show how interrupts can be handled in Ada95. Note that generation of the test interrupt (irqforce.c) is BSP specific and must be edited.

15.4. Version Requirements#

With RTEMS 4.0, a patch was required to psignal.c in RTEMS sources (to correct a bug associated to the default action of signals 15-32). The SPARC/ERC32 RTEMS BSP includes the``gnatsupp`` subdirectory that can be used as an example for other BSPs.

With GNAT 3.11p, a patch is required for a-init.c to invoke the BSP specific routine that installs the exception handlers.

16.1. Shared Memory Configuration Table#

The Shared Memory Configuration Table is defined in the following structure:

typedef volatile uint32_t vol_u32;

typedef struct {
  vol_u32 *address;        /* write here for interrupt    */
  vol_u32  value;          /* this value causes interrupt */
  vol_u32  length;         /* for this length (0,1,2,4)   */
} Shm_Interrupt_information;

struct shm_config_info {
  vol_u32           *base;                     /* base address of SHM         */
  vol_u32            length;                   /* length (in bytes) of SHM    */
  vol_u32            format;                   /* SHM is big or little endian */
  vol_u32          (*convert)();               /* neutral conversion routine  */
  vol_u32            poll_intr;                /* POLLED or INTR driven mode  */
  void             (*cause_intr)( uint32_t );
  Shm_Interrupt_information Intr;              /* cause intr information      */
};

typedef struct shm_config_info shm_config_table;

where the fields are defined as follows:

base

is the base address of the shared memory buffer used to pass messages between the nodes in the system.

length

is the length (in bytes) of the shared memory buffer used to pass messages between the nodes in the system.

format

is either SHM_BIG or SHM_LITTLE to indicate that the neutral format of the shared memory area is big or little endian. The format of the memory should be chosen to match most of the inter-node traffic.

convert

is the address of a routine which converts from native format to neutral format. Ideally, the neutral format is the same as the native format so this routine is quite simple.

poll_intr, cause_intr

is either INTR_MODE or POLLED_MODE to indicate how the node will be informed of incoming messages.

Intr

is the information required to cause an interrupt on a node. This structure contains the following fields:

address

is the address to write at to cause an interrupt on that node. For a polled node, this should be NULL.

value

is the value to write to cause an interrupt.

length

is the length of the entity to write on the node to cause an interrupt. This can be 0 to indicate polled operation, 1 to write a byte, 2 to write a sixteen-bit entity, and 4 to write a thirty-two bit entity.

16.2. Primitives#

void *Shm_Convert_address( void *address )
{
  return the local address version of this bus address
}

void Shm_Get_configuration(
  uint32_t           localnode,
  shm_config_table **shmcfg
)
{
  fill in the Shared Memory Configuration Table
}

typedef struct {
  vol_u32 lock;  /* lock field for this queue    */
  vol_u32 front; /* first envelope on queue      */
  vol_u32 rear;  /* last envelope on queue       */
  vol_u32 owner; /* receiving (i.e. owning) node */
} Shm_Locked_queue_Control;

void Shm_Initialize_lock(
  Shm_Locked_queue_Control *lq_cb
)
{
  lq_cb->lock = LQ_UNLOCKED;
}

void Shm_Lock(
  Shm_Locked_queue_Control *lq_cb
)
{
  disable processor interrupts
    set Shm_isrstat to previous interrupt disable level

  while ( TRUE ) {
    atomically attempt to acquire the lock
    if the lock was acquired
      return
    delay some small period of time
  }
}

void Shm_Unlock(
  Shm_Locked_queue_Control *lq_cb
)
{
  set the lock to the unlocked value
  reenable processor interrupts to their level prior
    to the lock being acquired.  This value was saved
    in the global variable Shm_isrstat
}

16.3. Installing the MPCI ISR#

The Shm_setvec is invoked by the portable portion of the shared memory to install the interrupt service routine that is invoked when an incoming message is announced. Some target boards support an interprocessor interrupt or mailbox scheme and this is where the ISR for that interrupt would be installed.

On an interrupt driven node, this routine would be implemented as follows:

void Shm_setvec( void )
{
  install the interprocessor communications ISR
}

On a polled node, this routine would be empty.

17.1. Benchmark Timer#

The RTEMS Timing Test Suite requires a benchmark timer. The RTEMS Timing Test Suite is very helpful for determining the performance of target hardware and comparing its performance to that of other RTEMS targets.

This section describes the routines which are assumed to exist by the RTEMS Timing Test Suite. The names used are EXACTLY what is used in the RTEMS Timing Test Suite so follow the naming convention.

void benchmark_timer_initialize(void)
{
  initialize the benchmark timer
}

benchmark_timer_t benchmark_timer_read(void)
{
  stop time = read the hardware timer
  if the subtract overhead feature is enabled
    subtract overhead from stop time
  return the stop time
}

void benchmark_timer_disable_subtracting_average_overhead(bool find_flag)
{
  disable the subtract overhead feature
}

17.2. gen68340 UART FIFO Full Mode#

The gen68340 BSP is an example of the use of the timer to support the UART input FIFO full mode (FIFO means First In First Out and roughly means buffer). This mode consists in the UART raising an interrupt when n characters have been received (n is the UART’s FIFO length). It results in a lower interrupt processing time, but the problem is that a scanf primitive will block on a receipt of less than n characters. The solution is to set a timer that will check whether there are some characters waiting in the UART’s input FIFO. The delay time has to be set carefully otherwise high rates will be broken:

• if no character was received last time the interrupt subroutine was entered, set a long delay,

• otherwise set the delay to the delay needed for n characters receipt.

18.1. Terms#

ATA device - physical device attached to an IDE controller

18.2. Introduction#

ATA driver provides generic interface to an ATA device. ATA driver is hardware independent implementation of ATA standard defined in working draft “AT Attachment Interface with Extensions (ATA-2)” X3T10/0948D revision 4c, March 18, 1996. ATA Driver based on IDE Controller Driver and may be used for computer systems with single IDE controller and with multiple as well. Although current implementation has several restrictions detailed below ATA driver architecture allows easily extend the driver. Current restrictions are:

• Only mandatory (see draft p.29) and two optional (READ/WRITE MULTIPLE) commands are implemented

• Only PIO mode is supported but both poll and interrupt driven

The reference implementation for ATA driver can be found in cpukit/libblock/src/ata.c.

18.3. Initialization#

The ata_initialize routine is responsible for ATA driver initialization. The main goal of the initialization is to detect and register in the system all ATA devices attached to IDE controllers successfully initialized by the IDE Controller driver.

In the implementation of the driver, the following actions are performed:

rtems_device_driver ata_initialize(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor,
  void                      *arg
)
{
  initialize internal ATA driver data structure

  for each IDE controller successfully initialized by the IDE Controller driver
    if the controller is interrupt driven
      set up interrupt handler

    obtain information about ATA devices attached to the controller
    with help of EXECUTE DEVICE DIAGNOSTIC command

    for each ATA device detected on the controller
      obtain device parameters with help of DEVICE IDENTIFY command

      register new ATA device as new block device in the system
}

Special processing of ATA commands is required because of absence of multitasking environment during the driver initialization.

Detected ATA devices are registered in the system as physical block devices (see libblock library description). Device names are formed based on IDE controller minor number device is attached to and device number on the controller (0 - Master, 1 - Slave). In current implementation 64 minor numbers are reserved for each ATA device which allows to support up to 63 logical partitions per device.

18.4. ATA Driver Architecture#

/* ATA request */
typedef struct ata_req_s {
  Chain_Node        link;   /* link in requests chain */
  char              type;   /* request type */
  ata_registers_t   regs;   /* ATA command */
  uint32_t          cnt;    /* Number of sectors to be exchanged */
  uint32_t          cbuf;   /* number of current buffer from breq in use */
  uint32_t          pos;    /* current position in 'cbuf' */
  blkdev_request   *breq;   /* blkdev_request which corresponds to the ata request */
  rtems_id          sema;   /* semaphore which is used if synchronous
                             * processing of the ata request is required */
  rtems_status_code status; /* status of ata request processing */
  int               error;  /* error code */
} ata_req_t;

/*
 * This structure describes controller state, devices configuration on the
 * controller and chain of ATA requests to the controller.
*/
typedef struct ata_ide_ctrl_s {
  bool          present;   /* controller state */
  ata_dev_t     device[2]; /* ata devices description */
  Chain_Control reqs;      /* requests chain */
} ata_ide_ctrl_t;

/*
 * Mapping of RTEMS ATA devices to the following pairs:
 * (IDE controller number served the device, device number on the controller)
*/
typedef struct ata_ide_dev_s {
  int ctrl_minor;/* minor number of IDE controller serves RTEMS ATA device */
  int device;    /* device number on IDE controller (0 or 1) */
} ata_ide_dev_t;

/* ATA driver events */
typedef enum ata_msg_type_s {
  ATA_MSG_GEN_EVT = 1,     /* general event */
  ATA_MSG_SUCCESS_EVT,     /* success event */
  ATA_MSG_ERROR_EVT,       /* error event */
  ATA_MSG_PROCESS_NEXT_EVT /* process next ata request event */
} ata_msg_type_t;

19.1. Introduction#

The IDE Controller driver is responsible for providing an interface to an IDE Controller. The capabilities provided by this driver are:

• Read IDE Controller register

• Write IDE Controller register

• Read data block through IDE Controller Data Register

• Write data block through IDE Controller Data Register

The reference implementation for an IDE Controller driver can be found in bsps/shared/dev/ide. This driver is based on the libchip concept and allows to work with any of the IDE Controller chips simply by appropriate configuration of BSP. Drivers for a particular IDE Controller chips locate in the following directories: drivers for well-known IDE Controller chips locate into bsps/shared/dev/ide and drivers for custom IDE Controller chips (for example, implemented on FPGA) locate into bsps/${RTEMS_CPU}/${RTEMS_BSP/ata. There is a README file in these directories for each supported IDE Controller chip. Each of these README explains how to configure a BSP for that particular IDE Controller chip.

19.2. Initialization#

IDE Controller chips used by a BSP are statically configured into IDE_Controller_Table. The ide_controller_initialize routine is responsible for initialization of all configured IDE controller chips. Initialization order of the chips based on the order the chips are defined in the IDE_Controller_Table.

The following actions are performed by the IDE Controller driver initialization routine:

rtems_device_driver ide_controller_initialize(
  rtems_device_major_number  major,
  rtems_device_minor_number  minor_arg,
  void                      *arg
)
{
  for each IDE Controller chip configured in IDE_Controller_Table
    if (BSP dependent probe(if exists) AND device probe for this IDE chip
        indicates it is present)
      perform initialization of the particular chip
      register device with configured name for this chip
}

19.3. Read IDE Controller Register#

The ide_controller_read_register routine reads the content of the IDE Controller chip register. IDE Controller chip is selected via the minor number. This routine is not allowed to be called from an application.

void ide_controller_read_register(
  rtems_device_minor_number  minor,
  unsigned32                 reg,
  unsigned32                *value
)
{
  get IDE Controller chip configuration information from
  IDE_Controller_Table by minor number

  invoke read register routine for the chip
}

19.4. Write IDE Controller Register#

The ide_controller_write_register routine writes IDE Controller chip register with specified value. IDE Controller chip is selected via the minor number. This routine is not allowed to be called from an application.

void ide_controller_write_register(
  rtems_device_minor_number minor,
  unsigned32                reg,
  unsigned32                value
)
{
  get IDE Controller chip configuration information from
  IDE_Controller_Table by minor number

  invoke write register routine for the chip
}

19.5. Read Data Block Through IDE Controller Data Register#

The ide_controller_read_data_block provides multiple consequent read of the IDE Controller Data Register. IDE Controller chip is selected via the minor number. The same functionality may be achieved via separate multiple calls of ide_controller_read_register routine but ide_controller_read_data_block allows to escape functions call overhead. This routine is not allowed to be called from an application.

void ide_controller_read_data_block(
  rtems_device_minor_number  minor,
  unsigned16                 block_size,
  blkdev_sg_buffer          *bufs,
  uint32_t                  *cbuf,
  uint32_t                  *pos
)
{
  get IDE Controller chip configuration information from
  IDE_Controller_Table by minor number

  invoke read data block routine for the chip
}

19.6. Write Data Block Through IDE Controller Data Register#

The ide_controller_write_data_block provides multiple consequent write into the IDE Controller Data Register. IDE Controller chip is selected via the minor number. The same functionality may be achieved via separate multiple calls of ide_controller_write_register routine but ide_controller_write_data_block allows to escape functions call overhead. This routine is not allowed to be called from an application.

void ide_controller_write_data_block(
  rtems_device_minor_number  minor,
  unsigned16                 block_size,
  blkdev_sg_buffer          *bufs,
  uint32_t                  *cbuf,
  uint32_t                  *pos
)
{
  get IDE Controller chip configuration information from
  IDE_Controller_Table by minor number

  invoke write data block routine for the chip
}

21.1. BSP Basics#

Embedded development is hard. Different CPUs have different instructions for doing the same thing, and different boards will have all sorts of different hardware that require unique drivers and interfaces. RTEMS handles this by having discrete packages, BSPs, to encapsulate code to accommodate for unique hardware. BSPs seek to implement the Hardware-Software interface. This, in a nutshell, is one of the core purposes. of RTEMS: To abstract (as much as is possible) away from the physical hardware and provide a standards compliant real-time environment for the embedded developer. If you think about it, the operating system on your normal computer serves a very similar purpose.

21.2. Common Features Found In BSPs#

Although the actual implementation code will differ between BSPs, all BSPs will share some degree of common functionality. This is because that no matter what exact hardware you have, you need some basic features implemented in order to have a real time system you can develop on. Some of the most common shared features across most boards include:

• console: is technically the serial driver for the BSP rather than

just a console driver, it deals with the board UART (i.e. serial devices) * clock: support for the clock tick - a regular time basis for the kernel * timer: support of timer devices, used for timing tests * rtc or tod: support for the hardware real time clock * network: the Ethernet driver * shmsupp: support of shared memory driver MPCI layer in a multiprocessor system * gnatsupp: BSP specific support for the GNU Ada run-time * irq: support for how the processor handles interrupts (probably the most common module shared by all boards) * tm27: specific routines for the tm27 timing test * start and startup: C and assembly used to initialize the board during startups/resets/reboots

These are just some of the things you should be looking for when adding Doxygen to a BSP.

Note that there is no guarantee a particular BSP will implement all of these features, or even some of them. These are just the most common ones to look for. RTEMS follows a standardized naming convention for the BSP sub directories, so you should be able to tell in most cases what has been implemented on the BSP level and what has not.

21.3. Shared Features#

Some of the RTEMS executive is hardware independent and can be abstracted so that the same piece of code can be shared across multiple CPU architectures, or across multiple boards on the same architecture. This is done so that chunks of software can be reused, as well as aiding in reducing the development and debugging time for implementing new BSPs. This greatly aids the developer, but as someone seeking to document this code, this can make your life a little bit harder. It is hard to tell by looking at the directory of a BSP which features have simply been left out and which features are being implemented by using shared code from either from the architecture (../shared) or the base bsps/ shared directory (../../shared). You may be looking at the BSP headers and notice that you have an irq.h, but no irq.c implementing it, or you might even be missing both. You know that the processor has interrupt support somehow, but where is it? The easiest way to figure this out is by looking at the Makefile.am for a BSP. We’ll detail this process more in a bit.

21.4. Rationale#

As someone adding documentation and not doing actual development work, you might think it is not necessary to know some of the in and outs of BSPs. In actuality, this information will prove to be very useful. Doxygen documentation works by grouping things and their components (i.e. functions and other definitions), and by having brief descriptions of what each group does. You can’t know what to look for or know how to group it or know how to describe it without some basic knowledge of what a BSP is. For more information on any of the above or BSPs in general, check out the BSP Development Guide. .

21.5. The Structure of the bsps/ directory#

All BSPs are found within the bsps/ directory, which is itself very well ordered. At the first level, we find a directory for each CPU architecture RTEMS supports, as well as a directory for code shared by all implementations.

$ cd bsps
$ ls
arm   bsp.am  lm32  m68k             mips   no_cpu         README  sparc
avr   h8300   m32c  Makefile.am      moxie  powerpc        sh      sparc64
bfin  i386    m32r  MERGE.PROCEDURE  nios2  preinstall.am  shared  v850

If we cd into a specific architecture, we see that a similar structure is employed. bsps/arm/ contains directories for each Board Support Package for boards with an ARM cpu, along with a folder for files and .h’s shared by all BSPs of that architecture.

$ cd arm
$ ls
acinclude.m4  edb7312    gumstix   Makefile.am    realview-pbx-a9  stm32f4
configure.ac  gba        lm3s69xx  nds            rtl22xx          xilinx-zynq
csb336        lpc24xx   preinstall.am  shared     csb337           gp32
lpc32xx   raspberrypi    smdk2410

Finally, if we cd into a specific BSP, we see the files and .h’s that compose the package for that particular board. You may recognize the directory names as some of the [common features] we outlined above, like ‘’’irq’’’, ‘’’clock’’’, ‘’’console’’’, and ‘’’startup’’’. These directories contain implementations of these features.

$ cd raspberrypi
$ ls
include       misc        README   clock          console          irq
start

Another way to get an idea of the structure of bsps/ is to navigate to a directory and execute the “tree -f” command. This outputs a nice graphic that conveys some of the hierarchical properties of a particular directory.

$ pwd
~/rtems/bsps/arm/raspberrypi
$ tree -f
.
|-- ./clock
|   `-- ./clock/clockdrv.c
|-- ./configure.ac
|-- ./console
|   |-- ./console/console-config.c
        |   `-- ./console/usart.c
|-- ./include
|   |-- ./include/bsp.h
|   |-- ./include/irq.h
|   |-- ./include/mmu.h
|   |-- ./include/raspberrypi.h
|   `-- ./include/usart.h
|-- ./irq
|   `-- ./irq/irq.c
|-- ./make
|   `-- ./make/custom
|       `-- ./make/custom/raspberrypi.cfg
|-- ./Makefile.am
|-- ./misc
|   `-- ./misc/timer.c
|-- ./preinstall.am
|-- ./README
`-- ./startup
    |-- ./startup/bspreset.c
    |-- ./startup/bspstart.c
    |-- ./startup/bspstarthooks.c
    |-- ./startup/linkcmds
    `-- ./startup/mm_config_table.c

In short, BSPs will use the following directories:

• bsps/shared <- code used that is shared by all BSPs

• bsps/CPU/shared <- code used shared by all BSPs of a particular CPU architecture

• bsps/CPU/BSP <- code unique to this BSP

As you can see, the bsps/ directory has a very logical and easy to understand structure to it. The documentation generated by Doxygen should attempt to match this structure as closely as possible. We want an overarching parent group to serve the same purpose as the bsps/ directory. In it, we want groups for each CPU architecture and a group for the shared files. We then want groups for each BSP. Breaking our documentation up into discrete groups like this will greatly simplify the process and make the documentation much easier to go through. By learning about the existing structure of the bsps/ directory, we get an idea of how we should structure the Doxygen groups we create. More on this in the next section.

21.6. Doxygen#

Now that we have covered some of the preliminaries, we can move on to what you are actually reading this wiki page for: adding Doxygen to the bsps/ directory. Let’s start with some Doxygen basics. Skip this if you are already comfortable with Doxygen.

In addition to this, check out the page on `Doxygen Recommendations <wiki:Developer/Coding/Doxygen >`_. , which also contains a fair amount of information that will not be covered here.

21.7. Doxygen Basics#

Doxygen is a documentation generator. It allows for documentation to be written right by the source code, greatly easing the pains of keeping documentation relevant and up to date. Doxygen has many commands, used for things like annotating functions with descriptions, parameter information, or return value information. You can reference other files or even other documentation.

The core component of Doxygen (that we care about right now at least) is what’s called a group, or module. These are used to add structure and associate groups of files that serve a similar purpose or implement the same thing.

21.8. Doxygen Headers#

Doxygen is always found in a special Doxygen comment block, known as a Doxygen header. In RTEMS, this block comes in the form of a multiline comment with some included Doxygen commands, which are preceded by the ‘@’ tag. Take a look at this Doxygen header that declares the arm_raspberrypi module, which houses the documentation in the BSP for the Raspberry Pi.

bsps/arm/raspberrypi/include/bsp.h:

/**
 * @defgroup arm_raspberrypi Raspberry Pi Support
 *
 * @ingroup bsp_arm
 *
 * @brief Raspberry Pi support package
 *
 */

You see a few commands here that we’ll cover in the following sections. Briefly, the @defgroup command declares a new group, the @ingroup command nests this group as a submodule of some other group (in this case bsp_arm), and the @brief command provides a brief description of what this group is.

21.9. The @defgroup Command#

The @defgroup command is used to declare new groups or modules. Think “define group”. The syntax of this command is as follows:

@defgroup <group name> <group description>

The group name is the name used by Doxygen elsewhere to reference this group. The group description is what is displayed when the end user navigates to this module in the resulting documentation. The group description is a couple words formatted as how it would be in a table of contents. This part is what actually shows up in the documentation, when the user navigates to this group’s module, this description will be the modules name.

Groups should only be declared (@defgroup) in .h files. This is because Doxygen is used primarily to document interfaces, which are only found in .h files. Placing @defgroups in .h files is the only real restriction. Which .h file you place the group declaration in surprisingly doesn’t matter. There is no information in the resulting documentation that indicates where the group was declared. You will see that we do have some rules for where you should place these declarations, but we also use this fact that it doesn’t matter to our advantage, in order to standardize things.

The @defgroup command is used only to define ‘’structure’’. No actual documentation is generated as a result of its use. We must @ingroup things to the group we declare in order to create documentation. Even though it does not generate visible documentation, the @defgroup command is still very important. We use it in a way that seeks to emulate the structure of the bsps/ directory itself. We do this by creating a hierarchy of groups for each CPU architecture and each BSP.

21.10. The @ingroup Command#

The @ingroup command is used to add ‘things’ to already declared groups or modules. These ‘things’ can either be other groups, or files themselves. The syntax of the @ingroup command is as follows:

@ingroup <group name>

The group name is the actual name, not description, of the group you want to add yourself to. Remember that group name was the second argument passed to the @defgroup command.

Using the @ingroup command is how we add ‘’meaning’’ to the ‘’structure’’ created by using @defgroup. @ingroup associates the file it is found in and all other Doxygen found within (function annotations, prototypes, etc) with the group we declared with the @defgroup command. We add related files and headers to the same groups to create a logical and cohesive body of documentation. If the end user wanted to read documentation about how the raspberry pi handles interrupts, all they would have to do would be to navigate to the raspberry pi’s interrupt support module (which we created with a @defgroup command), and read the documentation contained within (which we added with @ingroup commands).

@ingroup is found within all Doxygen headers, along with an @brief statement. There are two types of Doxygen headers, which we will go over after we see a description of the @brief command.

21.11. The @brief Command#

The @brief command is used to give either a) a brief description in the form of an entry as you would see it in a table of contents (i.e. Capitalized, only a couple of words) or b) a brief topic sentence giving a basic idea of what the group does. The reason you have two uses for the brief command is that it is used differently in the two types of Doxygen headers, as we will see shortly. The syntax of the brief command is self evident, but included for the sake of completion:

@brief <Table of Contents entry '''or''' Topic Sentence>

21.12. The Two Types of Doxygen Headers#

There are two types of Doxygen Headers. The first type is found at the beginning of a file, and contains an @file command. This type of header is used when @ingroup-ing the file into another doxygen group. The form of the @brief command in this case is a topic sentence, often very close to the file name or one of it’s major functions. An example of this type of header, found in bsps/arm/raspberrypi/include/bsp.h is as follows:

Header type 1: used to add files to groups, always found at the beginning of a file
/**
 * @file
 *
 * @ingroup raspberrypi
 *
 * @brief Global BSP definitions.
 */

/*
 *  Copyright (c) YYYY NAME
 *
 *   <LICENSE TERMS>
 */