ddb — interactive kernel debugger
In order to enable kernel debugging facilities include:
To prevent activation of the debugger on kernel panic(9):
In order to print a stack trace of the current thread on the console for a panic:
To print the numerical value of symbols in addition to the symbolic representation, define:
The ddb kernel debugger is an interactive debugger with a syntax inspired by gdb(1). If linked into the running kernel, it can be invoked locally with the ’debug’ keymap(5) action. The debugger is also invoked on kernel panic(9) if the debug.debugger_on_panic sysctl(8) MIB variable is set non-zero, which is the default unless the KDB_UNATTENDED option is specified.
The current location is called dot. The dot is displayed with a hexadecimal format at a prompt. The commands examine and write update dot to the address of the last line examined or the last location modified, and set next to the address of the next location to be examined or changed. Other commands do not change dot, and set next to be the same as dot.
The general command syntax is: command[/modifier] [
A blank line repeats the previous command from the address next with count 1 and no modifiers. Specifying addr sets dot to the address. Omitting addr uses dot. A missing count is taken to be 1 for printing commands or infinity for stack traces. A count of -1 is equivalent to a missing count. Options that are supplied but not supported by the given command are usually ignored.
The ddb debugger has a pager feature (like the more(1) command) for the output. If an output line exceeds the number set in the lines variable, it displays ’’--More--’’ and waits for a response. The valid responses for it are:
one more page
one more line
abort the current command, and return to the command input mode
Finally, ddb provides a small (currently 10 items) command history, and offers simple emacs-style command line editing capabilities. In addition to the emacs control keys, the usual ANSI arrow keys may be used to browse through the history buffer, and move the cursor within the current line.
examine[/AISabcdghilmorsuxz ...] [
x[/AISabcdghilmorsuxz ...] [
Display the addressed locations according to the formats in the modifier. Multiple modifier formats display multiple locations. If no format is specified, the last format specified for this command is used.
The format characters are:
look at by bytes (8 bits)
look at by half words (16 bits)
look at by long words (32 bits)
look at by quad words (64 bits)
print the location being displayed
print the location with a line number if possible
display in unsigned hex
display in signed hex
display in unsigned octal
display in signed decimal
display in unsigned decimal
display in current radix, signed
display low 8 bits as a character. Non-printing characters are displayed as an octal escape code (e.g., ’\000’).
display the null-terminated string at the location. Non-printing characters are displayed as octal escapes.
display in unsigned hex with character dump at the end of each line. The location is also displayed in hex at the beginning of each line.
display as an instruction
display as an instruction with possible alternate formats depending on the machine. On i386, this selects the alternate format for the instruction decoding (16 bits in a 32-bit code segment and vice versa).
display a symbol name for the pointer stored at the address
Examine forward: execute an examine command with the last specified parameters to it except that the next address displayed by it is used as the start address.
Examine backward: execute an examine command with the last specified parameters to it except that the last start address subtracted by the size displayed by it is used as the start address.
Print addrs according to the modifier character (as described above for examine). Valid formats are: a, x, z, o, d, u, r, and c. If no modifier is specified, the last one specified to it is used. The argument addr can be a string, in which case it is printed as it is. For example:
print/x "eax = " $eax "\necx = " $ecx "\n"
will print like:
eax = xxxxxx
ecx = yyyyyy
addr expr1 [expr2 ...]
w[/bhl] addr expr1 [expr2 ...]
Write the expressions specified after addr on the command line at succeeding locations starting with addr. The write unit size can be specified in the modifier with a letter b (byte), h (half word) or l (long word) respectively. If omitted, long word is assumed.
Warning: since there is no delimiter between expressions, strange things may happen. It is best to enclose each expression in parentheses.
set $variable [
Set the named variable or register with the value of expr. Valid variable names are described below.
Set a break point at addr. If count is supplied, the continue command will not stop at this break point on the first count − 1 times that it is hit. If the break point is set, a break point number is printed with ’#’. This number can be used in deleting the break point or adding conditions to it.
If the u modifier is specified, this command sets a break point in user address space. Without the u option, the address is considered to be in the kernel space, and a wrong space address is rejected with an error message. This modifier can be used only if it is supported by machine dependent routines.
Warning: If a user text is shadowed by a normal user space debugger, user space break points may not work correctly. Setting a break point at the low-level code paths may also cause strange behavior.
Delete the specified break point. The break point can be specified by a break point number with ’#’, or by using the same addr specified in the original break command, or by omitting addr to get the default address of dot.
Set a watchpoint for a region. Execution stops when an attempt to modify the region occurs. The size argument defaults to 4. If you specify a wrong space address, the request is rejected with an error message.
Warning: Attempts to watch wired kernel memory may cause unrecoverable error in some systems such as i386. Watchpoints on user addresses work best.
Set a hardware watchpoint for a region if supported by the architecture. Execution stops when an attempt to modify the region occurs. The size argument defaults to 4.
Warning: The hardware debug facilities do not have a concept of separate address spaces like the watch command does. Use hwatch for setting watchpoints on kernel address locations only, and avoid its use on user mode address spaces.
Delete specified hardware watchpoint.
Single step count times. If the p modifier is specified, print each instruction at each step. Otherwise, only print the last instruction.
Warning: depending on machine type, it may not be possible to single-step through some low-level code paths or user space code. On machines with software-emulated single-stepping (e.g., pmax), stepping through code executed by interrupt handlers will probably do the wrong thing.
Continue execution until a breakpoint or watchpoint. If the c modifier is specified, count instructions while executing. Some machines (e.g., pmax) also count loads and stores.
Warning: when counting, the debugger is really silently single-stepping. This means that single-stepping on low-level code may cause strange behavior.
Stop at the next call or return instruction. If the p modifier is specified, print the call nesting depth and the cumulative instruction count at each call or return. Otherwise, only print when the matching return is hit.
Stop at the matching return instruction. If the p modifier is specified, print the call nesting depth and the cumulative instruction count at each call or return. Otherwise, only print when the matching return is hit.
[pid | tid][,count]
t[/u] [pid | tid][,count]
where[/u] [pid | tid][,count]
bt[/u] [pid | tid][,count]
Stack trace. The u option traces user space; if omitted, trace only traces kernel space. The optional argument count is the number of frames to be traced. If count is omitted, all frames are printed.
Warning: User space stack trace is valid only if the machine dependent code supports it.
search[/bhl] addr value [mask][,count]
Search memory for value. The optional count argument limits the search.
Prints the thread address for a thread kernel-mode stack of which contains the specified address. If the thread is not found, search the thread stack cache and prints the cached stack address. Otherwise, prints nothing.
show all procs[/a]
Display all process information. The process information may not be shown if it is not supported in the machine, or the bottom of the stack of the target process is not in the main memory at that time. The a modifier will print command line arguments for each process.
show all trace
Show a stack trace for every thread in the system.
show all ttys
Show all TTY’s within the system. Output is similar to pstat(8), but also includes the address of the TTY structure.
show all vnets
Show the same output as "show vnet" does, but lists all virtualized network stacks within the system.
Show the same information like "show lockchain" does, but for every thread in the system.
Show all locks that are currently held. This command is only available if witness(4) is included in the kernel.
The same as "show pcpu", but for every CPU present in the system.
Show information related with resource management, including interrupt request lines, DMA request lines, I/O ports, I/O memory addresses, and Resource IDs.
Dump data about APIC IDT vector mappings.
Show breakpoints set with the "break" command.
show bio addr
Show information about the bio structure struct bio present at addr. See the sys/bio.h header file and g_bio(9) for more details on the exact meaning of the structure fields.
show buffer addr
Show information about the buf structure struct buf present at addr. See the sys/buf.h header file for more details on the exact meaning of the structure fields.
show callout addr
Show information about the callout structure struct callout present at addr.
Show brief information about the TTY subsystem.
Without argument, show the list of all created cdev’s, consisting of devfs node name and struct cdev address. When address of cdev is supplied, show some internal devfs state of the cdev.
Lists hooks currently waiting for completion in run_interrupt_driven_config_hooks().
Print numbered root and assigned CPU affinity sets. See cpuset(2) for more details.
Show registers specific to the Cyrix processor.
Prints the contents of the static device mapping table. Currently only available on the ARM architecture.
show domain addr
Print protocol domain structure struct domain at address addr. See the sys/domain.h header file for more details on the exact meaning of the structure fields.
show ffs [addr]
Show brief information about ffs mount at the address addr, if argument is given. Otherwise, provides the summary about each ffs mount.
show file addr
Show information about the file structure struct file present at address addr.
Show information about every file structure in the system.
Show the number of physical pages in each of the free lists.
show geom [addr]
If the addr argument is not given, displays the entire GEOM topology. If addr is given, displays details about the given GEOM object (class, geom, provider or consumer).
Show IDT layout. The first column specifies the IDT vector. The second one is the name of the interrupt/trap handler. Those functions are machine dependent.
show igi_list addr
Show information about the IGMP structure struct igmp_ifsoftc present at addr.
show inodedeps [addr]
Show brief information about each inodedep structure. If addr is given, only inodedeps belonging to the fs located at the supplied address are shown.
show inpcb addr
Show information on IP Control Block struct in_pcb present at addr.
Dump information about interrupt handlers.
Dump the interrupt statistics.
Show interrupt lines and their respective kernel threads.
Show information from the local APIC registers for this CPU.
show lock addr
Show lock structure. The output format is as follows:
Name of the lock.
Flags passed to the lock initialization function. flags values are lock class specific.
Current state of a lock. state values are lock class specific.
show lockchain addr
Show all threads a particular thread at address addr is waiting on based on non-spin locks.
Show the same information as "show buf", but for every locked struct buf object.
List all locked vnodes in the system.
Prints all locks that are currently acquired. This command is only available if witness(4) is included in the kernel.
Prints malloc(9) memory allocator statistics. The output format is as follows:
Specifies a type of memory. It is the same as a description string used while defining the given memory type with MALLOC_DECLARE(9).
Number of memory allocations of the given type, for which free(9) has not been called yet.
Total memory consumed by the given allocation type.
Number of memory allocation requests for the given memory type.
The same information can be gathered in userspace with ’’vmstat −m’’.
Prints the VM map at addr. If the f modifier is specified the complete map is printed.
Print the system’s message buffer. It is the same output as in the ’’dmesg’’ case. It is useful if you got a kernel panic, attached a serial cable to the machine and want to get the boot messages from before the system hang.
Displays short info about all currently mounted file systems.
show mount addr
Displays details about the given mount point.
Prints the VM object at addr. If the f option is specified the complete object is printed.
Print the panic message if set.
Show statistics on VM pages.
Show statistics on VM page queues.
Print PCI bus registers. The same information can be gathered in userspace by running ’’pciconf −lv’’.
Print current processor state. The output format is as follows:
Thread pointer, process identifier and the name of the process.
Control block pointer.
FPU thread pointer.
Idle thread pointer.
CPU identifier coming from APIC.
spin locks held
Names of spin locks held.
Dump process groups present within the system.
show proc [addr]
If no [addr] is specified, print information about the current process. Otherwise, show information about the process at address addr.
Show process virtual memory layout.
show protosw addr
Print protocol switch structure struct protosw at address addr.
Display the register set. If the u modifier is specified, it displays user registers instead of kernel registers or the currently saved one.
Warning: The support of the u modifier depends on the machine. If not supported, incorrect information will be displayed.
show rman addr
Show resource manager object struct rman at address addr. Addresses of particular pointers can be gathered with "show allrman" command.
Show real time clock value. Useful for long debugging sessions.
Deprecated. Now an alias for show lockchain.
Both commands provide the same functionality. They show sleepqueue struct sleepqueue structure. Sleepqueues are used within the FreeBSD kernel to implement sleepable synchronization primitives (thread holding a lock might sleep or be context switched), which at the time of writing are: condvar(9), sx(9) and standard msleep(9) interface.
show sockbuf addr
show socket addr
Those commands print struct sockbuf and struct socket objects placed at addr. Output consists of all values present in structures mentioned. For exact interpretation and more details, visit sys/socket.h header file.
Show system registers (e.g., cr0-4 on i386.) Not present on some platforms.
show tcpcb addr
Print TCP control block struct tcpcb lying at address addr. For exact interpretation of output, visit netinet/tcp.h header file.
show thread [addr]
If no addr is specified, show detailed information about current thread. Otherwise, information about thread at addr is printed.
Show all threads within the system. Output format is as follows:
Thread identifier (TID)
Thread structure address
show tty addr
Display the contents of a TTY structure in a readable form.
show turnstile addr
Show turnstile struct turnstile structure at address addr. Turnstiles are structures used within the FreeBSD kernel to implement synchronization primitives which, while holding a specific type of lock, cannot sleep or context switch to another thread. Currently, those are: mutex(9), rwlock(9), rmlock(9).
Show UMA allocator statistics. Output consists five columns:
Name of the UMA zone. The same string that was passed to uma_zcreate(9) as a first argument.
Size of a given memory object (slab).
Number of slabs being currently used.
Number of free slabs within the UMA zone.
Number of allocations requests to the given zone.
The very same information might be gathered in the userspace with the help of ’’vmstat −z’’.
show unpcb addr
Shows UNIX domain socket private control block struct unpcb present at the address addr.
Prints, whether the internal VM objects are in a map somewhere and none have zero ref counts.
This is supposed to show physical addresses consumed by a VM object. Currently, it is not possible to use this command when witness(4) is compiled in the kernel.
show vnet addr
Prints virtualized network stack struct vnet structure present at the address addr.
show vnode [addr]
Prints vnode struct vnode structure lying at [addr]. For the exact interpretation of the output, look at the sys/vnode.h header file.
show vnodebufs addr
Shows clean/dirty buffer lists of the vnode located at addr.
show vpath addr
Walk the namecache to lookup the pathname of the vnode located at addr.
Displays all watchpoints. Shows watchpoints set with "watch" command.
Shows information about lock acquisition coming from the witness(4) subsystem.
Toggles between remote GDB and DDB mode. In remote GDB mode, another machine is required that runs gdb(1) using the remote debug feature, with a connection to the serial console port on the target machine. Currently only available on the i386 architecture.
Halt the system.
kill sig pid
Send signal sig to process pid. The signal is acted on upon returning from the debugger. This command can be used to kill a process causing resource contention in the case of a hung system. See signal(3) for a list of signals. Note that the arguments are reversed relative to kill(2).
Hard reset the system. If the optional argument seconds is given, the debugger will wait for this long, at most a week, before rebooting.
Print a short summary of the available commands and command abbreviations.
ddb supports a basic output capture facility, which can be used to retrieve the results of debugging commands from userspace using sysctl(3). capture on enables output capture; capture off disables capture. capture reset will clear the capture buffer and disable capture. capture status will report current buffer use, buffer size, and disposition of output capture.
Userspace processes may inspect and manage ddb capture state using sysctl(8):
debug.ddb.capture.bufsize may be used to query or set the current capture buffer size.
debug.ddb.capture.maxbufsize may be used to query the compile-time limit on the capture buffer size.
debug.ddb.capture.bytes may be used to query the number of bytes of output currently in the capture buffer.
debug.ddb.capture.data returns the contents of the buffer as a string to an appropriately privileged process.
This facility is particularly useful in concert with the scripting and textdump(4) facilities, allowing scripted debugging output to be captured and committed to disk as part of a textdump for later analysis. The contents of the capture buffer may also be inspected in a kernel core dump using kgdb(1).
Run, define, list, and delete scripts. See the SCRIPTING section for more information on the scripting facility.
Use the textdump dump command to immediately perform a textdump. More information may be found in textdump(4). The textdump set command may be used to force the next kernel core dump to be a textdump rather than a traditional memory dump or minidump. textdump status reports whether a textdump has been scheduled. textdump unset cancels a request to perform a textdump as the next kernel core dump.
The debugger accesses registers and variables as $name. Register names are as in the ’’show registers’’ command. Some variables are suffixed with numbers, and may have some modifier following a colon immediately after the variable name. For example, register variables can have a u modifier to indicate user register (e.g., ’’$eax:u’’).
Built-in variables currently supported are:
Input and output radix.
Addresses are printed as ’’symbol+offset’’ unless offset is greater than maxoff.
The width of the displayed line.
The number of lines. It is used by the built-in pager. Setting it to 0 disables paging.
Tab stop width.
Work variable; xx can take values from 0 to 31.
Most expression operators in C are supported except ’~’, ’^’, and unary ’&’. Special rules in ddb are:
The name of a symbol is translated to the value of the symbol, which is the address of the corresponding object. ’.’ and ’:’ can be used in the identifier. If supported by an object format dependent routine, [
filename: ]func:lineno, [
filename: ]variable, and [
filename: ]lineno can be accepted as a symbol.
Radix is determined by the first two letters: ’0x’: hex, ’0o’: octal, ’0t’: decimal; otherwise, follow current radix.
address of the start of the last line examined. Unlike dot or next, this is only changed by examine or write command.
last address explicitly specified.
Translated to the value of the specified variable. It may be followed by a ’:’ and modifiers as described above.
A binary operator which rounds up the left hand side to the next multiple of right hand side.
Indirection. It may be followed by a ’:’ and modifiers as described above.
ddb supports a basic scripting facility to allow automating tasks or responses to specific events. Each script consists of a list of DDB commands to be executed sequentially, and is assigned a unique name. Certain script names have special meaning, and will be automatically run on various ddb events if scripts by those names have been defined.
The script command may be used to define a script by name. Scripts consist of a series of ddb commands separated with the ’;’ character. For example:
kdb.enter.panic=bt; show pcpu
script lockinfo=show alllocks; show lockedvnods
The scripts command lists currently defined scripts.
The run command execute a script by name. For example:
The unscript command may be used to delete a script by name. For example:
These functions may also be performed from userspace using the ddb(8) command.
Certain scripts are run automatically, if defined, for specific ddb events. The follow scripts are run when various events occur:
The kernel debugger was entered as a result of an acpi(4) event.
The kernel debugger was entered at boot as a result of the debugger boot flag being set.
The kernel debugger was entered as a result of a serial or console break.
The kernel debugger was entered as a result of a CAM(4) event.
The kernel debugger was entered as a result of an assertion failure in the mac_test(4) module of the TrustedBSD MAC Framework.
The kernel debugger was entered as a result of an ndis(4) breakpoint event.
The kernel debugger was entered as a result of a netgraph(4) event.
panic(9) was called.
The kernel debugger was entered as a result of a powerfail NMI on the sparc64 platform.
The kernel debugger was entered as a result of an unimplemented interrupt type on the powerpc platform.
The kernel debugger was entered as a result of the debug.kdb.enter sysctl being set.
The kernel debugger was entered as a result of a trapsig event on the sparc64 platform.
The kernel debugger was entered as a result of an assertion failure in the union file system.
The kernel debugger was entered, but no reason has been set.
The kernel debugger was entered as a result of a VFS lock violation.
The kernel debugger was entered as a result of a watchdog firing.
The kernel debugger was entered as a result of a witness(4) violation.
In the event that none of these scripts is found, ddb will attempt to execute a default script:
The kernel debugger was entered, but a script exactly matching the reason for entering was not defined. This can be used as a catch-all to handle cases not specifically of interest; for example, kdb.enter.witness might be defined to have special handling, and kdb.enter.default might be defined to simply panic and reboot.
On machines with an ISA expansion bus, a simple NMI generation card can be constructed by connecting a push button between the A01 and B01 (CHCHK# and GND) card fingers. Momentarily shorting these two fingers together may cause the bridge chipset to generate an NMI, which causes the kernel to pass control to ddb. Some bridge chipsets do not generate a NMI on CHCHK#, so your mileage may vary. The NMI allows one to break into the debugger on a wedged machine to diagnose problems. Other bus’ bridge chipsets may be able to generate NMI using bus specific methods. There are many PCI and PCIe add-in cards which can generate NMI for debugging. Modern server systems typically use IPMI to generate signals to enter the debugger. The devel/ipmitool port can be used to send the chassis power diag command which delivers an NMI to the processor. Embedded systems often use JTAG for debugging, but rarely use it in combination with ddb.
For serial consoles, you can enter the debugger by sending a BREAK condition on the serial line if options BREAK_TO_DEBUGGER is specified in the kernel. Most terminal emulation programs can send a break sequence with a special key sequence or via a menu item. However, in some setups, sending the break can be difficult to arrange or happens spuriously, so if the kernel contains options ALT_BREAK_TO_DEBUGGER then the sequence of CR TILDE CTRL-B enters the debugger; CR TILDE CTRL-P causes a panic instead of entering the debugger; and CR TILDE CTRL-R causes an immediate reboot. In all the above sequences, CR is a Carriage Return and is usually sent by hitting the Enter or Return key. TILDE is the ASCII tilde character (~). CTRL-x is Control x created by hitting the control key and then x and then releasing both.
The break to enter the debugger behavior may be enabled at run-time by setting the sysctl(8) debug.kdb.break_to_debugger to 1. The alternate sequence to enter the debugger behavior may be enabled at run-time by setting the sysctl(8) debug.kdb.alt_break_to_debugger to 1. The debugger may be entered by setting the sysctl(8) debug.kdb.enter to 1.
Header files mentioned in this manual page can be found below /usr/include directory.
The ddb debugger was developed for Mach, and ported to 386BSD 0.1. This manual page translated from man(7) macros by Garrett Wollman.
Robert N. M. Watson added support for ddb output capture, textdump(4) and scripting in FreeBSD 7.1.
BSD September 7, 2018 BSD