Dheeraj Sangamkar wrote: The crash utility and the embedded gdb module only know about data structure types that you "feed" them. So normally it's only aware of data structures found in the debuginfo data of the vmlinux file and from any kernel modules that you load with "mod". That being said, you can use the gdb "add-symbol-file" command from within a crash session to load the debuginfo data from a user executable that has been built with -g. For example, take the crash utility itself. It's built with -g by default, so from within a crash session I can load its debuginfo data to make crash/gdb aware of its user-space data structures. For example, the crash utility has this data structure defined in its "defs.h" file that's used to keep basic information for each task: struct task_context { /* context stored for each task */ ulong task; ulong thread_info; ulong pid; char comm[16+1]; int processor; ulong ptask; ulong mm_struct; struct task_context *tc_next; }; Running on a live system: # crash ... crash> The session obviously has no clue as to what a user-space data structure looks like: crash> struct task_context struct: invalid data structure reference: task_context crash> Again, since crash is built with -g, the debuginfo data is contained within its binary file. So if I use the gdb "add-symbol-file" to load its debuginfo data: crash> add-symbol-file /usr/bin/crash add symbol table from file "/usr/bin/crash" at Reading symbols from /usr/bin/crash...done. crash> Now it knows what the structure is: crash> struct task_context struct task_context { ulong task; ulong thread_info; ulong pid; char comm[17]; int processor; ulong ptask; ulong mm_struct; struct task_context *tc_next; } SIZE: 48 crash> Then to print an instance of a structure that exists in the user-space of a process: (1) you'll need to know where in user space memory it exists. (2) you'll need to be in the *context* of that task/process (3) you'll need to use the "-u" argument to the "struct" command. So suppose I know that one of these task_context structures exists at virtual address 0xac38950 in the crash utility's user virtual memory. First I make sure I'm in the proper context: crash> set PID: 13284 COMMAND: "crash" TASK: ddcb0550 [THREAD_INFO: c6bc7000] CPU: 1 STATE: TASK_RUNNING (ACTIVE) crash> And let the "struct" command know that the address is argument is a user-space address with the -u option: crash> struct task_context ac38950 -u struct task_context { task = 0xdbd14aa0, thread_info = 0xd9916000, pid = 0x419c, comm = "run-mozilla.sh\000\000", processor = 0x0, ptask = 0xc76cf000, mm_struct = 0xde38ee40, tc_next = 0x0 } crash> Dave

Thanks Dave, ioctl related data structures are common. I did not find the -u switch in the help for struct command in crash. Hopefully, my version of crash supports it. Dheeraj On Fri, Mar 7, 2008 at 8:32 PM, Dave Anderson <anderson(a)redhat.com&gt; wrote: