Embedded Development: Handling Exceptions and Interrupts in eCos

This chapter is from the book

Embedded Software Development with eCos

Learn More Buy

This chapter is from the book

This chapter is from the book 

Embedded Software Development with eCos

Learn More Buy

In this chapter, we begin with a look at exceptions and how they are handled at the HAL and application layers within the eCos system. Next, we cover the eCos interrupt model, including interrupt configuration, handling, and control. The exception and interrupt information prepares us for managing software and hardware events that occur.

3.1 Exceptions

An exception is a synchronous event that occurs during the execution of a thread that disrupts the normal flow of instructions. If exceptions are not properly processed during program execution, severe consequences, such as system failures, can occur. Exception handling is extremely important, especially in embedded systems, to avoid these failures, and improves the robustness of the software. Properly implemented exception handling can also aid in software execution recovery so that an application can proceed with its task after an exception has occurred.

Examples of exceptions that can occur in a system include those raised by hardware (such as a memory access error) and those raised by software (such as a divide by zero error). After the exception causes an interruption, the processor will jump to a defined address (or exception vector) and begin to run the instructions at that location. The address that the processor jumps to contains the exception handling code to process the error. Different processor architectures might locate the exception handlers at various places and implement this jump process in a variety of methods.

The method for implementing exception handling can vary. eCos does not provide an implementation similar to the throw and catch facilities provided in C++. For an embedded system, the simplest and most flexible method for handling exceptions is to call a function. This function needs a context or an area to do its work, and is typically passed in information, such as the exception number and possibly some optional parameters, to process the exception. After returning from this exception handler function, the thread can continue its execution.

There are two main methods for exception handling in eCos. The first is a combination of HAL and Kernel Exception Handling. The HAL provides the general hardware-level exception processing and then passes control on to the application for any extended exception support needed. This is the default configuration method.

The second exception handling method is Application Exception Handling. This allows the application to take full control over any or all of the exceptions and attaches a vector service routine directly to the hardware. The exception handler routine needs to be written in assembly language when using this configuration method.

The configuration option names described are found in the graphical configuration tool. In parentheses are the CDL names for the given option. For example, using the graphical configuration tool you could look up the configuration option name HAL Exception Support, which is located in the configuration window. This configuration option has a CDL macro associated with it—in this case, CYGPKG_HAL_EXCEPTIONS—which is located in the Properties window of the graphical configuration tool. We cover the graphical configuration tool in Chapter 11, The eCos Toolset.

The configuration options that control the exception handling are the HAL Exception Support (CYGPKG_HAL_EXCEPTIONS) option, located under the Platform-Independent HAL Options (CYGPKG_HAL_COMMON) component, and the Exception Handling (CYGPKG_KERNEL_EXCEPTIONS) component under the eCos Kernel package. Each of these configuration components requires the other to be enabled. Enabling these options allows HAL and Kernel Exception Handling. If both of these configuration options are disabled, it is up to you to provide exception handlers for Application Exception Handling.

Figure 3.1 shows the exception handling execution flow for the two main configuration options. The gray boxes illustrate, using a system call exception as an example, the HAL and Kernel Exception Handling execution flow. The white boxes show the execution flow for Application Exception Handling using an alignment exception as an example. We look at the execution flow shown in Figure 3.1 later in this chapter.

Figure 3.1 eCos exception handling execution flow.

As another option, all exception handling can be disabled in the system. If this configura-tion is selected and an exception occurs, the macro CYG_FAIL is called and an assertion is raised. If assertions are not enabled, the state of the processor is restored; however, the cause of the exception might still exist. This is not a very elegant method for exception handling and can lead to undefined system behavior when an exception occurs. However, if code space is extremely important, it might be necessary to eliminate any extended exception processing.

3.1.1 HAL and Kernel Exception Handling

It is important that an exception handler is installed for each exception supported by the processor. If a handler is not installed for a particular exception and it occurs during execution, the processor will jump to the exception address to begin execution only to find no code to execute. This can cause erratic behavior that can be very difficult to debug and might lead to system failures.

The eCos HAL uses a Vector Service Routine (VSR) table that is defined in each HAL package as hal_vsr_table. The VSR table is an array of pointers to the exception handler routines. The VSR table is the first place the processor looks to determine where to jump to execute the exception handler.¹ The size and base address of the hal_vsr_table is architecture specific. For example, the PowerPC architecture has the option of locating the VSR table at either address 0x0000_0000 or 0xFFF0_0000, based on the configuration option settings of certain processor registers. The MIPS architecture specifies that the exception vector location be at address 0xBFC0_0000. Linker script files are used to define the location of the exception vector table. Linker script files are located under the arch subdirectory for a given HAL architecture and have a .ld extension.

The VSR table is located at a fixed memory location. This allows an application running from RAM, which is a typical debug configuration, to take control over certain exception service routines while keeping the ROM monitor in charge of debug exception handlers.

The eCos HAL ensures that a default exception VSR is installed during the HAL startup process for each exception supported by a given architecture. Installation of the default exception VSR takes place in the routine hal_mon_init. There is one default exception VSR defined in each HAL package. Different architectures have different names for the default exception VSR, as shown in Table 3.1.

The ARM architecture is not listed in Table 3.1 because it does not use the hal_vsr_table. Instead, the ARM architecture defines separate handler routines for each exception it supports. These routines can be found in the file vectors.S.

Table 3.1 Default Exception Vector Service Routines

The job of the default exception VSR is to perform the common processing of all exceptions, which includes saving the processor's state, calling any kernel-level handler routine to perform additional processing, and restoring the state of the processor prior to returning to normal program execution.

After the HAL has completed the common exception processing, control is passed to the kernel level. The routine that is called to handle this HAL-to-kernel transition is cyg_hal_exception_handler. This routine is found in the file hal_misc.c under the HAL arch subdirectory. The cyg_hal_exception_handler routine has two different execution paths possible, based on configuration option settings. If we look at Figure 3.1, we can follow the execution path for a system call exception, shown in the gray boxes. At the kernel level, the two possible configuration options are labeled on the lines leading from the cyg_hal_exception_handler routine.

The first kernel-level configuration option we see in Figure 3.1 is Include GDB Stubs in HAL (CYGDBG_HAL_DEBUG_GDB_INCLUDE_STUBS), which is under the Source-Level Debug Support HAL (CYGPKG_HAL_DEBUG) common component. This option is typically used in a HAL package for a ROM monitor build to allow the debug stub code to process the exception. This exception processing occurs in the routine __handle_exception. The __handle_exception routine manages all debug exception processing such as breakpoints, single stepping, and debug packet protocol communication. Additional information about GDB and the GDB protocol can be found online at:

http://sources.redhat.com/gdb/onlinedocs/gdb_toc.html

The second kernel-level configuration option we see in Figure 3.1 is HAL Exception Support (CYGPKG_HAL_EXCEPTIONS), under the General Platform-Independent HAL common component. This HAL option requires the Kernel Exception Handling to be enabled in the eCos Kernel package. In this configuration, an application can install its own handler for exceptions to take care of any further processing, such as displaying or logging an error message.

Within the kernel Exception Handling (CYGPKG_KERNEL_EXCEPTIONS) option is the configuration suboption Use Global Exception Handlers (CYGSEM_KERNEL_EXCEPTIONS_ GLOBAL), which allows installation of handlers to be either global—one set of handlers for all exceptions in the entire system—or on a per-thread basis. The default configuration for this sub-option setting is to enable global exception handlers. In this configuration, the HAL exception handler calls the routine cyg_hal_deliver_exception, which is located in the file except.cxx under the kernel package subdirectory.

Item List 3.1

void
cyg_exception_set_handler(
 cyg_code_t exception_number,
 cyg_exception_handler_t *new_handler,
 cyg_addrword_t new_data,
 cyg_exception_handler_t **old_handler,
 void **old_data
);

void
cyg_exception_clear_handler(
 cyg_code_t exception_number,
); 

void
cyg_exception_call_handler(
 cyg_handle_t thread,
 cyg_code_t exception_number,
 cyg_addrword_t exception_info
 ); 

To control the exception service routines from the application, the eCos kernel defines functions within its API. Item List 3.1 shows the details of these functions.

  1  #include <cyg/kernel/kapi.h>
  2  #include <cyg/infra/diag.h>
  3  #include <cyg/hal/hal_intr.h>
  4
  5  //
  6  // New System Call exception handler.
  7  //
  8  void system_call_exception_handler(
  9     cyg_addrword_t data,
10      cyg_code_t number,
11      cyg_addrword_t info)
12  { 
13    diag_printf( "ERROR: System Call Exception\n" );
14  } 
15
16
17  // 
18  // Main starting point for the application.
19  // 
20  void cyg_user_start(
21         void)
22  { 
23    cyg_exception_handler_t *old_handler;
24    cyg_addrword_t old_data;
25
26    //
27    // Install the exception handler for error output.
28    //
29    cyg_exception_set_handler(
30       CYGNUM_HAL_VECTOR_SYSTEM_CALL,
31       &system_call_exception_handler,
32       0,
33       &old_handler,
34       &old_data);
35  }

Code Listing 3.1 Example using the eCos kernel API for installing an exception handler within an application.

In Code Listing 3.1, we see an example of an application setting up the routine system_call_exception_handler, shown on line 8, for the PowerPC system call exception, CYGNUM_HAL_VECTOR_SYSTEM_CALL. CYGNUM_HAL_VECTOR_SYSTEM_CALL is defined in the PowerPC HAL package in the file hal_intr.h, which is included on line 3.

In this example, the exception handler simply writes out a message to the diagnostic port when this particular exception occurs, as shown on line 13. The function used is diag_printf, which is defined in the file diag.h included on line 2. When the call is made to install the handler, line 29, it can apply to either the thread or globally, depending on the con-figuration of the Use Global Exception Handlers (CYGSEM_KERNEL_EXCEPTIONS_GLOBAL) exception handling configuration suboption.

The cyg_user_start function, on line 20, is the main application entry point. The kernel startup process describing the cyg_user_start function is detailed in Chapter 5. To use the kernel API, the application must include the file kapi.h. The parameters passed into system_call_exception_handler, lines 9, 10, and 11, are:

data—the parameter setup in the cyg_exception_set_handlercall (0 in this example). It is often useful to use this parameter to pass a pointer to a structure for the exception handler routine to have information needed.

number—the exception number that occurred.

info—the processor-specific saved machine state. Currently, a pointer to the HAL_SavedRegisters structure is passed in this parameter. If this structure is modified, the saved state of the processor is altered. This allows the exception handler to correct the condition in order to allow the processor to continue. See the HAL macro definitions in Chapter 2 for an explanation of this architecture-specific structure.

If we look at Figure 3.1, the gray boxes illustrate the execution flow after setting up the exception handler in Code Listing 3.1.

The final routine that is called from the HAL is restore_state, as shown in Figure 3.1. This routine restores the processor registers to the state they were in prior to entering the default exception VSR. The state of the processor is contained in the HAL_SavedRegisters structure, which is passed in as a pointer. After restore_state returns, the processor can continue with its normal execution of the program.

3.1.2 Application Exception Handling

eCos provides a means for an application to completely take over all or some of the exception handling. This eliminates the HAL and kernel processing and allows the processor to vector directly to an application's VSR when an exception occurs. The application exception handler is then responsible for saving and restoring the processor's state, the same functionality provided by the HAL default exception VSR. It is also important to note that the VSR must be written in assembly language.

As we see in Figure 3.1, the white boxes illustrate how the application can take over exception handling. After the alignment exception occurs the VSR installed by the application is called to handle the exception condition. After the VSR completes, the handler must restore the processor's state so that the program can continue running.

The HAL defines macros to give the application access to the VSR table directly. These macros are described in Item List 3.2.

Item List 3.2 HAL Exception Vector Service Routine Macros

HAL_VSR_GET(
 _vector_,
 _pvsr_
 ) 

HAL_VSR_SET(
 vector_,
 vsr_,
 poldvsr_
 ) 

When installing a vector service routine, it is not necessary to call the HAL set and get macros directly. The eCos kernel API defines functions for the application to use instead. Item List 3.3 shows the kernel API functions for accessing the VSR table. These functions call the respective get and set HAL macros defined in Item List 3.3. For example, cyg_interrupt_get_vsr calls the HAL macro HAL_VSR_GET.

Item List 3.3 Kernel Exception Vector Service Routine API Functions

void
cyg_interrupt_get_vsr(
 cyg_vector_t vector,
 cyg_VSR_t **vsr
 );

void
cyg_interrupt_set_vsr(
 cyg_vector_t vector,
 cyg_VSR_t *vsr
 );