There are a few things that you need to download and install before using MRI:
- Follow the instructions at https://github.com/adamgreen/gcc4mbed#quick-start to install the latest version of the GCC4MBED project which includes the MRI debug monitor.
- MRI doesn't work if the JTAG debug interface is actively being used on the device. Any such JTAG usage will need to be disabled before you can use MRI. What needs to be done will differ depending on your device:
- mbed 1768: Download http://mbed.org/media/uploads/simon/mbedmicrocontroller_21164.if. Copy this file to your mbed device and then reset it. This will update the firmware on the interface chip so that its JTAG usage can be disabled when MRI is used.
- LPCXpresso 1769: Cut away the debugger portion of the device along the white line which separates the main LPC1769 device from the JTAG hardware. You will now need to provide 3.3V of regulated power and a serial connection to UART0 on the LPC1769 portion of the board.
- Other: Remove JTAG debugger connection when using MRI and connect your PC to a UART on the device. Typically this will be done with something like a FTDI to USB to serial cable.
This section provides steps to setup MRI the first time and use it to debug one of the samples included with the GCC4MBED project.
You will first want to make sure that you can connect to your mbed device using the USB virtual serial port provided by the mbed interface chip. Follow the steps documented here in the mbed handbook to get your PC talking to the mbed device. Once you have this working, take note of the serial port identifier that worked for you (ie: COM3, /dev/ttyACM0, /dev/tty.usbmodem412, etc.) You will soon need to provide this identifier to the GNU debugger, GDB.
By default the GCC4MBED project builds binaries of a Release type which produces optimized code and doesn't include any debugging support. This should be changed to a Debug build type to disable optimizations (makes for easier debugging) and then enable the MRI debugger in the produced binaries. This can be done by following these steps:
- Run the BuildShellDebug script found in the root of the GCC4MBED project directory. This script will start a shell which is configured to create unoptimized binaries for your mbed.
- Run "cd samples" from the debug shell.
- Run "make clean all MRI_ENABLE=1" from the debug shell.
Using the BuildShell script will revert back to Release.
From within the debug shell, run "cd FileTest" to switch into the directory which contains the sample's sources and binaries. Make sure that your mbed device is connected to your PC and that you can access its file system from your PC. On macOS and Linux, you can probably now just run "make deploy" from the debug shell to have the FileTest.bin copied to your mbed device. On Windows, you will need to issue something like "copy FileTest.bin e:\" to copy the binary to your mbed device but you will need to switch the "e:\" drive designation to match whatever is appropriate for your particular machine.
You can read https://github.com/adamgreen/gcc4mbed/blob/master/notes/install.creole#deploy to learn more about how to configure the LPC_DEPLOY environment variable from your BuildShell* scripts to make deployment during build easier.
If you have a terminal application such as HyperTerminal or screen connected to your mbed's virtual serial port, then disconnect it now. This allows GDB to use the same serial port for remote debugging of the mbed device.
Press the reset button on the mbed device. The application will be restarted but not actually start execution as the MRI enabled builds halt at startup and wait for GDB to be connected. This allows you, the developer, to set breakpoints at locations of interest in your code (including global constructors) before starting your main() routine.
We are now going to connect GDB to your mbed device over the USB virtual serial port. The command you use to start GDB is similar across different operating systems and machines but the serial port identification portion at the end of the command line will depend on your particular machine. Just replace that portion of the command line examples shown next with what you found in the first step of this Getting Started guide:
- Windows: arm-none-eabi-gdb FileTest.elf --baud 460800 -ex "set target-charset ASCII" -ex "set print pretty on" -ex "set remotelogfile mri.log" -ex "target remote com3"
- macOS: arm-none-eabi-gdb FileTest.elf --baud 230400 -ex "set target-charset ASCII" -ex "set print pretty on" -ex "set remotelogfile mri.log" -ex "target remote /dev/tty.usbmodem412"
- Linux: arm-none-eabi-gdb FileTest.elf --baud 460800 -ex "set target-charset ASCII" -ex "set print pretty on" -ex "set remotelogfile mri.log" -ex "target remote /dev/ttyACM0"
A successful connection should result in text similar to this:
GNU gdb (GNU Tools for ARM Embedded Processors) 7.3.1.20120316-cvs Copyright (C) 2011 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html> This is free software: you are free to change and redistribute it. There is NO WARRANTY, to the extent permitted by law. Type "show copying" and "show warranty" for details. This GDB was configured as "--host=x86_64-apple-darwin11.4.0 --target=arm-none-eabi". For bug reporting instructions, please see: <http://www.gnu.org/software/gdb/bugs/>... Reading symbols from /depots/gcc4mbed/samples/FileTest/FileTest.elf...done. Remote debugging using /dev/tty.usbmodem412 _start () at ../../src/gcc4mbed.c:37 37 __debugbreak(); (gdb)
The last few lines of the GDB output indicate that our FileTest sample is currently halted at line 37 in gcc4mbed.c This is the halt that I mentioned earlier which lets us get in and set some breakpoints before we let things actually startup.
GDB is now waiting for us to give it some debugging commands so let's start out by looking at the code around the line we are currently halted at to get a feel for where we are in the sample code. We can do this by issuing the list command:
(gdb) list 32 33 if (MRI_ENABLE) 34 { 35 mriInit(MRI_INIT_PARAMETERS); 36 if (MRI_BREAK_ON_INIT) 37 __debugbreak(); 38 } 39 40 __libc_init_array(); 41 mainReturnValue = main();
So line 37 is actually a __debugbreak() instruction that was issued because the MRI_BREAK_ON_INIT variable was set in our debug build. We can also see that __libc_init_array() and main() will actually be called after we start up the application again.
So we haven't even gotten to main() yet. What function are we in then? We can issue the backtrace command to GDB and have it list the current call stack.
(gdb) backtrace #0 _start () at ../../src/gcc4mbed.c:37 #1 0x00002b90 in Reset_Handler ()
This shows us that we are currently in a function called _start() which itself was called from an assembly language routine called Reset_Handler().
We don't really care that much about this code so let's get into main where things make a bit more sense. We can use the break command to set breakpoints and then issue the continue to start the code executing at full speed until we hit this breakpoint:
(gdb) break main Breakpoint 1 at 0x248: file main.cpp, line 28. (gdb) continue Continuing. Note: automatically using hardware breakpoints for read-only addresses. Breakpoint 1, main () at main.cpp:28 28 for (i = 0 ; i < sizeof(TestBuffer) ; i++)
Ahh, now we are actually in the sample's main() C++ code itself. Let's list out some of the code in this main() function.
(gdb) list 23 unsigned char TestBuffer[256]; 24 unsigned char ReadBuffer[256]; 25 size_t i; 26 27 // Fill in test buffer with every byte value possible. 28 for (i = 0 ; i < sizeof(TestBuffer) ; i++) 29 { 30 TestBuffer[i] = i; 31 } 32 memset(ReadBuffer, 0, sizeof(ReadBuffer)); (gdb) list 33 34 // Create a file in the LocalFileSystem with this data. 35 FILE* fp = fopen(Filename, "w"); 36 if (!fp) 37 { 38 error("Failed to open %s for writing\n", Filename); 39 } 40 41 int BytesWritten = fwrite(TestBuffer, 1, sizeof(TestBuffer), fp); 42 if (BytesWritten != sizeof(TestBuffer))
Note that we can issue the list command multiple times and it just continues listing source from where the previous list stopped.
We see that the first bit of code initializes some buffers and then it starts using the LocalFileSystem to create files and write data to them. Let's start stepping through the code and see how it progresses. We can do this through the use of the next command.
(gdb) next 30 TestBuffer[i] = i; (gdb) next 28 for (i = 0 ; i < sizeof(TestBuffer) ; i++) (gdb) next 30 TestBuffer[i] = i; (gdb) next 28 for (i = 0 ; i < sizeof(TestBuffer) ; i++) (gdb) next 30 TestBuffer[i] = i;
We are just jumping between lines 28 and 30 of the for loop. Th code is going to go through 256 iterations of this loop so we should just let it run until we get to line 32. We can issue the until command to accomplish this:
(gdb) until main.cpp:32 main () at main.cpp:32 32 memset(ReadBuffer, 0, sizeof(ReadBuffer));
Let's step over this memset() call and get to the next line:
(gdb) next 35 FILE* fp = fopen(Filename, "w"); (gdb)
Lines 28 - 32 should have initialized 256 entries in the TestBuffer and ReadBuffer arrays. Let's dump the contents of these 256 bytes of memory and see if they contain the expected data of all zeroes in ReadBuffer and ascending values in TestBuffer. We use the x command (short for eXamine) for this:
(gdb) x/256bx TestBuffer 0x10007dd8: 0x00 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x10007de0: 0x08 0x09 0x0a 0x0b 0x0c 0x0d 0x0e 0x0f 0x10007de8: 0x10 0x11 0x12 0x13 0x14 0x15 0x16 0x17 0x10007df0: 0x18 0x19 0x1a 0x1b 0x1c 0x1d 0x1e 0x1f 0x10007df8: 0x20 0x21 0x22 0x23 0x24 0x25 0x26 0x27 0x10007e00: 0x28 0x29 0x2a 0x2b 0x2c 0x2d 0x2e 0x2f 0x10007e08: 0x30 0x31 0x32 0x33 0x34 0x35 0x36 0x37 0x10007e10: 0x38 0x39 0x3a 0x3b 0x3c 0x3d 0x3e 0x3f 0x10007e18: 0x40 0x41 0x42 0x43 0x44 0x45 0x46 0x47 0x10007e20: 0x48 0x49 0x4a 0x4b 0x4c 0x4d 0x4e 0x4f 0x10007e28: 0x50 0x51 0x52 0x53 0x54 0x55 0x56 0x57 0x10007e30: 0x58 0x59 0x5a 0x5b 0x5c 0x5d 0x5e 0x5f 0x10007e38: 0x60 0x61 0x62 0x63 0x64 0x65 0x66 0x67 0x10007e40: 0x68 0x69 0x6a 0x6b 0x6c 0x6d 0x6e 0x6f 0x10007e48: 0x70 0x71 0x72 0x73 0x74 0x75 0x76 0x77 0x10007e50: 0x78 0x79 0x7a 0x7b 0x7c 0x7d 0x7e 0x7f 0x10007e58: 0x80 0x81 0x82 0x83 0x84 0x85 0x86 0x87 0x10007e60: 0x88 0x89 0x8a 0x8b 0x8c 0x8d 0x8e 0x8f 0x10007e68: 0x90 0x91 0x92 0x93 0x94 0x95 0x96 0x97 0x10007e70: 0x98 0x99 0x9a 0x9b 0x9c 0x9d 0x9e 0x9f 0x10007e78: 0xa0 0xa1 0xa2 0xa3 0xa4 0xa5 0xa6 0xa7 0x10007e80: 0xa8 0xa9 0xaa 0xab 0xac 0xad 0xae 0xaf 0x10007e88: 0xb0 0xb1 0xb2 0xb3 0xb4 0xb5 0xb6 0xb7 0x10007e90: 0xb8 0xb9 0xba 0xbb 0xbc 0xbd 0xbe 0xbf 0x10007e98: 0xc0 0xc1 0xc2 0xc3 0xc4 0xc5 0xc6 0xc7 0x10007ea0: 0xc8 0xc9 0xca 0xcb 0xcc 0xcd 0xce 0xcf 0x10007ea8: 0xd0 0xd1 0xd2 0xd3 0xd4 0xd5 0xd6 0xd7 0x10007eb0: 0xd8 0xd9 0xda 0xdb 0xdc 0xdd 0xde 0xdf 0x10007eb8: 0xe0 0xe1 0xe2 0xe3 0xe4 0xe5 0xe6 0xe7 0x10007ec0: 0xe8 0xe9 0xea 0xeb 0xec 0xed 0xee 0xef 0x10007ec8: 0xf0 0xf1 0xf2 0xf3 0xf4 0xf5 0xf6 0xf7 0x10007ed0: 0xf8 0xf9 0xfa 0xfb 0xfc 0xfd 0xfe 0xff (gdb) x/256bx ReadBuffer 0x10007ed8: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007ee0: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007ee8: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007ef0: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007ef8: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f00: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f08: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f10: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f18: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f20: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f28: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f30: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f38: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f40: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f48: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f50: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f58: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f60: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f68: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f70: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f78: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f80: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f88: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f90: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007f98: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007fa0: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007fa8: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007fb0: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007fb8: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007fc0: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007fc8: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x10007fd0: 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00
What is that /256bx text all about? Time to check out the help command in GDB:
(gdb) help x Examine memory: x/FMT ADDRESS. ADDRESS is an expression for the memory address to examine. FMT is a repeat count followed by a format letter and a size letter. Format letters are o(octal), x(hex), d(decimal), u(unsigned decimal), t(binary), f(float), a(address), i(instruction), c(char) and s(string). Size letters are b(byte), h(halfword), w(word), g(giant, 8 bytes). The specified number of objects of the specified size are printed according to the format. Defaults for format and size letters are those previously used. Default count is 1. Default address is following last thing printed with this command or "print".
The 256 is the count of items to dump. The b indicates that a byte at a time should be dumped and the x indicates that the values should be dumped in hexadecimal format.
Where are we again? Let's issue another list command:
(gdb) list 30 TestBuffer[i] = i; 31 } 32 memset(ReadBuffer, 0, sizeof(ReadBuffer)); 33 34 // Create a file in the LocalFileSystem with this data. 35 FILE* fp = fopen(Filename, "w"); 36 if (!fp) 37 { 38 error("Failed to open %s for writing\n", Filename); 39 }
So we are going to open a file whose name is located in the Filename variable. Let's use the print command to take a look and see what is in that string variable:
(gdb) print Filename $1 = "/local/foo.bar"
Ahh, we are going to create and write out data to a file called "/local/foo.bar".
We can continue stepping through this code for a bit more:
(gdb) next 36 if (!fp) (gdb) next 41 int BytesWritten = fwrite(TestBuffer, 1, sizeof(TestBuffer), fp); (gdb) next 42 if (BytesWritten != sizeof(TestBuffer)) (gdb) next 47 fclose(fp); (gdb) next 48 fp = NULL; (gdb) next 51 fp = fopen(Filename, "r");
Normally this foo.bar file would be created on the 2MB FLASH device built into the mbed device. There is a problem with that though. MRI needs to disable the mbed interface JTAG connection so that it can act as the debugger instead. Unfortunately the LocalFileSystem uses this JTAG interface to talk to this embedded FLASH. So where did the foo.bar file end up? If we actually look at the FileTest sample directory where we are currently running GDB, we will see the following (Note: Use shell dir on Windows instead of shell ls -l:
(gdb) shell ls -l total 1192 -rwxr-xr-x 1 adamgr admin 64632 Aug 9 22:49 FileTest.bin -rwxr-xr-x 1 adamgr admin 309396 Aug 9 22:49 FileTest.elf -rw-r--r-- 1 adamgr admin 181818 Aug 9 22:49 FileTest.hex drwxr-xr-x 8 adamgr admin 272 Aug 9 22:49 LPC176x -rwxr-xr-x 1 adamgr admin 225 May 17 14:44 debug -rw-r--r-- 1 adamgr admin 256 Aug 9 23:34 foo.bar -rw-r--r-- 1 adamgr admin 2334 Aug 8 13:49 main.cpp -rw-r--r-- 1 adamgr admin 723 Aug 8 13:49 makefile -rw-r--r-- 1 adamgr admin 30751 Aug 9 23:34 mri.log
We see that foo.bar is sitting in this location and it is the expected 256 bytes in length. What happened here? MRI actually intercepts these LocalFileSystem calls and instead sends them to GDB where the file I/O actually ends up taking place on your PC. This means that you can store files here and even read configuration files from this source directory while debugging.
Towards the end of main, we will call remove(Filename) which should delete the file from our PC. Let's see if that happens by setting a breakpoint on the remove() function and then continuing execution until we hit this breakpoint:
(gdb) break remove Breakpoint 2 at 0x2f90 (gdb) c Continuing. Breakpoint 2, 0x00002f90 in remove ()
There are a few things to note with this GDB interaction:
- First is that when I issued the continue command, I just typed c instead. GDB allows such abbreviations for many of its commands. Usually the first couple of characters from a command name is enough. It should also be noted that just pressing Enter in many instances will simply repeat the last command you typed. That means that you don't need to keep typing next to step through your code. You can use next once and then just keep pressing the Enter key to issue subsequent steps.
- When breakpoint 2 was hit, it tells us that we are in remove() but it doesn't give a source filename and line number. That is because this function is part of the mbed SDK library which doesn't ship with symbols.
Abbreviations for some commands that we have already used:
b | break |
c | continue |
n | next |
p |
Let's take a look at the call stack now:
(gdb) bt #0 0x00002f90 in remove () #1 0x00000430 in main () at main.cpp:71
Let's just finish running the remove function and return back to our main() function. We can do this by issuing the finish command which allows the sample to run until is leaves the current function and returns to the caller:
(gdb) finish Run till exit from #0 0x00002f90 in remove () main () at main.cpp:73 73 printf("\r\nTest completed\r\n");
We are back in main after having called the remove() function. Let's take a look at the sample source directory again to see if foo.bar was deleted as expected:
(gdb) shell ls -l total 1192 -rwxr-xr-x 1 adamgr admin 64632 Aug 9 22:49 FileTest.bin -rwxr-xr-x 1 adamgr admin 309396 Aug 9 22:49 FileTest.elf -rw-r--r-- 1 adamgr admin 181818 Aug 9 22:49 FileTest.hex drwxr-xr-x 8 adamgr admin 272 Aug 9 22:49 LPC176x -rwxr-xr-x 1 adamgr admin 225 May 17 14:44 debug -rw-r--r-- 1 adamgr admin 2334 Aug 8 13:49 main.cpp -rw-r--r-- 1 adamgr admin 723 Aug 8 13:49 makefile -rw-r--r-- 1 adamgr admin 32929 Aug 9 23:45 mri.log
Gone as expected. Excellent!
Our debugging session is almost complete. Let's take a look at a few other interesting GDB commands:
(gdb) info breakpoints Num Type Disp Enb Address What 1 breakpoint keep y 0x00000248 in main() at main.cpp:28 breakpoint already hit 1 time 2 breakpoint keep y 0x00002f90 <remove+4> breakpoint already hit 1 time
info breakpoints shows us the currently set breakpoints. Issuing the help breakpoints command will give you an overview of other breakpoint related commands.
While I have shown you the next command here for stepping through your code, there is another stepping command in GDB called step. The difference is that next runs and advances to the next instruction in the current function while step will actually step into functions that are called (as long as symbols exist for those functions.)
Let's just let the program continue until it leaves main and ends up calling the exit() routine and then issue the quit command to leave this debugging session:
(gdb) c Continuing. Test completed Program received signal SIGTRAP, Trace/breakpoint trap. 0x00003870 in semihost_exit () (gdb) q A debugging session is active. Inferior 1 [Remote target] will be killed. Quit anyway? (y or n) y
One thing to note is that after the continue command is sent, there is some text, "Test completed", displayed in GDB before the program traps into the exit() call at the end of the program. This text actually comes from a printf() call at the end of the main() routine. MRI intercepts these printf() calls and sends them over to GDB. This means that you can combine printf() debugging with the more full featured GDB debugging.
Want more information about how to use GDB? If so checkout its documentation which will have been installed in your gcc4mbed project at gcc-arm-none-eabi/share/doc/gcc-arm-none-eabi/pdf/gdb.pdf