10. Debugging Programs That Use Overlays

If your program is too large to fit completely in your target system's memory, you can sometimes use overlays to work around this problem. GDB provides some support for debugging programs that use overlays.

10.1 How Overlays Work		A general explanation of overlays.
10.2 Overlay Commands		Managing overlays in GDB.
10.3 Automatic Overlay Debugging		GDB can find out which overlays are mapped by asking the inferior.
10.4 Overlay Sample Program		A sample program using overlays.

10.1 How Overlays Work

Suppose you have a computer whose instruction address space is only 64 kilobytes long, but which has much more memory which can be accessed by other means: special instructions, segment registers, or memory management hardware, for example. Suppose further that you want to adapt a program which is larger than 64 kilobytes to run on this system.

One solution is to identify modules of your program which are relatively independent, and need not call each other directly; call these modules overlays. Separate the overlays from the main program, and place their machine code in the larger memory. Place your main program in instruction memory, but leave at least enough space there to hold the largest overlay as well.

Now, to call a function located in an overlay, you must first copy that overlay's machine code from the large memory into the space set aside for it in the instruction memory, and then jump to its entry point there.

Data Instruction Larger Address Space Address Space Address Space +-----------+ +-----------+ +-----------+ | | | | | | +-----------+ +-----------+ +-----------+<-- overlay 1 | program | | main | .----| overlay 1 | load address | variables | | program | | +-----------+ | and heap | | | | | | +-----------+ | | | +-----------+<-- overlay 2 | | +-----------+ | | | load address +-----------+ | | | .-| overlay 2 | | | | | | | mapped --->+-----------+ | | +-----------+ address | | | | | | | overlay | <-' | | | | area | <---' +-----------+<-- overlay 3 | | <---. | | load address +-----------+ `--| overlay 3 | | | | | +-----------+ | | +-----------+ | | +-----------+ A code overlay

The diagram (see A code overlay) shows a system with separate data and instruction address spaces. To map an overlay, the program copies its code from the larger address space to the instruction address space. Since the overlays shown here all use the same mapped address, only one may be mapped at a time. For a system with a single address space for data and instructions, the diagram would be similar, except that the program variables and heap would share an address space with the main program and the overlay area.

An overlay loaded into instruction memory and ready for use is called a mapped overlay; its mapped address is its address in the instruction memory. An overlay not present (or only partially present) in instruction memory is called unmapped; its load address is its address in the larger memory. The mapped address is also called the virtual memory address, or VMA; the load address is also called the load memory address, or LMA.

Unfortunately, overlays are not a completely transparent way to adapt a program to limited instruction memory. They introduce a new set of global constraints you must keep in mind as you design your program:

• Before calling or returning to a function in an overlay, your program must make sure that overlay is actually mapped. Otherwise, the call or return will transfer control to the right address, but in the wrong overlay, and your program will probably crash.

• If the process of mapping an overlay is expensive on your system, you will need to choose your overlays carefully to minimize their effect on your program's performance.

• The executable file you load onto your system must contain each overlay's instructions, appearing at the overlay's load address, not its mapped address. However, each overlay's instructions must be relocated and its symbols defined as if the overlay were at its mapped address. You can use GNU linker scripts to specify different load and relocation addresses for pieces of your program; see section `Overlay Description' in Using ld: the GNU linker.

• The procedure for loading executable files onto your system must be able to load their contents into the larger address space as well as the instruction and data spaces.

The overlay system described above is rather simple, and could be improved in many ways:

• If your system has suitable bank switch registers or memory management hardware, you could use those facilities to make an overlay's load area contents simply appear at their mapped address in instruction space. This would probably be faster than copying the overlay to its mapped area in the usual way.

• If your overlays are small enough, you could set aside more than one overlay area, and have more than one overlay mapped at a time.

• You can use overlays to manage data, as well as instructions. In general, data overlays are even less transparent to your design than code overlays: whereas code overlays only require care when you call or return to functions, data overlays require care every time you access the data. Also, if you change the contents of a data overlay, you must copy its contents back out to its load address before you can copy a different data overlay into the same mapped area.

10.2 Overlay Commands

To use GDB's overlay support, each overlay in your program must correspond to a separate section of the executable file. The section's virtual memory address and load memory address must be the overlay's mapped and load addresses. Identifying overlays with sections allows GDB to determine the appropriate address of a function or variable, depending on whether the overlay is mapped or not.

GDB's overlay commands all start with the word overlay; you can abbreviate this as ov or ovly. The commands are:

overlay off

Disable GDB's overlay support. When overlay support is disabled, GDB assumes that all functions and variables are always present at their mapped addresses. By default, GDB's overlay support is disabled.

overlay manual

Enable manual overlay debugging. In this mode, GDB relies on you to tell it which overlays are mapped, and which are not, using the overlay map-overlay and overlay unmap-overlay commands described below.

overlay map-overlay overlay

overlay map overlay

Tell GDB that overlay is now mapped; overlay must be the name of the object file section containing the overlay. When an overlay is mapped, GDB assumes it can find the overlay's functions and variables at their mapped addresses. GDB assumes that any other overlays whose mapped ranges overlap that of overlay are now unmapped.

overlay unmap-overlay overlay

overlay unmap overlay

Tell GDB that overlay is no longer mapped; overlay must be the name of the object file section containing the overlay. When an overlay is unmapped, GDB assumes it can find the overlay's functions and variables at their load addresses.

overlay auto

Enable automatic overlay debugging. In this mode, GDB consults a data structure the overlay manager maintains in the inferior to see which overlays are mapped. For details, see 10.3 Automatic Overlay Debugging.

overlay load-target

overlay load

Re-read the overlay table from the inferior. Normally, GDB re-reads the table GDB automatically each time the inferior stops, so this command should only be necessary if you have changed the overlay mapping yourself using GDB. This command is only useful when using automatic overlay debugging.

overlay list-overlays

overlay list

Display a list of the overlays currently mapped, along with their mapped addresses, load addresses, and sizes.

Normally, when GDB prints a code address, it includes the name of the function the address falls in:

(gdb) print main $3 = {int ()} 0x11a0 <main>

When overlay debugging is enabled, GDB recognizes code in unmapped overlays, and prints the names of unmapped functions with asterisks around them. For example, if foo is a function in an unmapped overlay, GDB prints it this way:

(gdb) overlay list No sections are mapped. (gdb) print foo $5 = {int (int)} 0x100000 <*foo*>

When foo's overlay is mapped, GDB prints the function's name normally:

(gdb) overlay list Section .ov.foo.text, loaded at 0x100000 - 0x100034, mapped at 0x1016 - 0x104a (gdb) print foo $6 = {int (int)} 0x1016 <foo>

When overlay debugging is enabled, GDB can find the correct address for functions and variables in an overlay, whether or not the overlay is mapped. This allows most GDB commands, like break and disassemble, to work normally, even on unmapped code. However, GDB's breakpoint support has some limitations:

• You can set breakpoints in functions in unmapped overlays, as long as GDB can write to the overlay at its load address.
• GDB can not set hardware or simulator-based breakpoints in unmapped overlays. However, if you set a breakpoint at the end of your overlay manager (and tell GDB which overlays are now mapped, if you are using manual overlay management), GDB will re-set its breakpoints properly.

10.3 Automatic Overlay Debugging

GDB can automatically track which overlays are mapped and which are not, given some simple co-operation from the overlay manager in the inferior. If you enable automatic overlay debugging with the overlay auto command (see section 10.2 Overlay Commands), GDB looks in the inferior's memory for certain variables describing the current state of the overlays.

Here are the variables your overlay manager must define to support GDB's automatic overlay debugging:

_ovly_table:

This variable must be an array of the following structures:

struct { /* The overlay's mapped address. */ unsigned long vma; /* The size of the overlay, in bytes. */ unsigned long size; /* The overlay's load address. */ unsigned long lma; /* Non-zero if the overlay is currently mapped; zero otherwise. */ unsigned long mapped; }

_novlys:

This variable must be a four-byte signed integer, holding the total number of elements in _ovly_table.

To decide whether a particular overlay is mapped or not, GDB looks for an entry in _ovly_table whose vma and lma members equal the VMA and LMA of the overlay's section in the executable file. When GDB finds a matching entry, it consults the entry's mapped member to determine whether the overlay is currently mapped.

In addition, your overlay manager may define a function called _ovly_debug_event. If this function is defined, GDB will silently set a breakpoint there. If the overlay manager then calls this function whenever it has changed the overlay table, this will enable GDB to accurately keep track of which overlays are in program memory, and update any breakpoints that may be set in overlays. This will allow breakpoints to work even if the overlays are kept in ROM or other non-writable memory while they are not being executed.

10.4 Overlay Sample Program

When linking a program which uses overlays, you must place the overlays at their load addresses, while relocating them to run at their mapped addresses. To do this, you must write a linker script (see section `Overlay Description' in Using ld: the GNU linker). Unfortunately, since linker scripts are specific to a particular host system, target architecture, and target memory layout, this manual cannot provide portable sample code demonstrating GDB's overlay support.

However, the GDB source distribution does contain an overlaid program, with linker scripts for a few systems, as part of its test suite. The program consists of the following files from `gdb/testsuite/gdb.base':

`overlays.c'

The main program file.

`ovlymgr.c'

A simple overlay manager, used by `overlays.c'.

`foo.c'

`bar.c'

`baz.c'

`grbx.c'

Overlay modules, loaded and used by `overlays.c'.

`d10v.ld'

`m32r.ld'

Linker scripts for linking the test program on the d10v-elf and m32r-elf targets.

You can build the test program using the d10v-elf GCC cross-compiler like this:

$ d10v-elf-gcc -g -c overlays.c $ d10v-elf-gcc -g -c ovlymgr.c $ d10v-elf-gcc -g -c foo.c $ d10v-elf-gcc -g -c bar.c $ d10v-elf-gcc -g -c baz.c $ d10v-elf-gcc -g -c grbx.c $ d10v-elf-gcc -g overlays.o ovlymgr.o foo.o bar.o \ baz.o grbx.o -Wl,-Td10v.ld -o overlays

The build process is identical for any other architecture, except that you must substitute the appropriate compiler and linker script for the target system for d10v-elf-gcc and d10v.ld.