Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix a problem I noticed a while back but had too many other things 

in the air to deal with it.
Basically, following a kernel fault (eg. dereferencing a NULL pointer
in kernel mode) a DDB 'trace' did not show the function where the
fault occurred. For example:

	db> trace
	_Debugger()
	_panic()
	_trap()
	faultstkadj()
	_pool_drain()
	_uvm_pageout()
	_start_pagedaemon()
	_proc_trampoline()
	db>

The 'faultstkadj()' line here is bogus. It is shown because the return
address to 'trap()' happens to point there, and since faultstkadj() has
no stack frame, DDB assumes it was the faulting function. In this example,
the _real_ function was pool_reclaim(), but you would have to look at
the program counter at the time of the fault to figure that one out.

This fix makes the trace command do the dirty work for you by grubbing
around in 'trap()'s argument list to find the *real* PC value at the
time of the fault, replacing the 'faultstkadj()' line with the real
function's name.
This commit is contained in:
scw 1999-04-04 11:33:02 +00:00
parent a6a272d18a
commit f7a019c61a
1 changed files with 35 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
sys/arch/m68k/m68k/db_trace.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: db_trace.c,v 1.23 1999/01/15 23:15:50 thorpej Exp $ */
/* $NetBSD: db_trace.c,v 1.24 1999/04/04 11:33:02 scw Exp $ */
/*
* Mach Operating System
@ -99,6 +99,8 @@ extern int curpcb;
#define get(addr, space) \
(db_get_value((db_addr_t)(addr), sizeof(int), FALSE))
#define offsetof(type, member) ((size_t)(&((type *)0)->member))
#define NREGISTERS 16
struct stackpos {
@ -475,6 +477,7 @@ db_stack_trace_cmd(addr, have_addr, count, modif)
char * name;
struct stackpos pos;
boolean_t kernel_only = TRUE;
int fault_pc = 0;
{
char *cp = modif;
@ -535,6 +538,37 @@ db_stack_trace_cmd(addr, have_addr, count, modif)
val = MAXINT;
}
}
/*
* Since faultstkadj doesn't set up a valid stack frame,
* we would assume it was the source of the fault. To
* get around this we peek at the fourth argument of
* "trap()" (the stack frame at the time of the fault)
* to determine the _real_ value of PC when things wen
* wrong.
*
* NOTE: If the argument list for 'trap()' ever changes,
* we lose.
*/
if ( strcmp("_trap", name) == 0 ) {
/* Point to 'trap()'s argument list */
regp = pos.k_fp + FR_SAVFP + 4;
/* Step over the arguments to the frame structure */
regp += (4 * 4) + offsetof(struct frame, f_pc);
/* Get the PC at the time of the fault */
fault_pc = get(regp, DSP);
} else
if ( fault_pc ) {
if ( strcmp("faultstkadj", name) == 0 ) {
db_find_sym_and_offset(fault_pc, &name, &val);
if (name == 0) {
name = "?";
val = MAXINT;
}
}
fault_pc = 0;
}
db_printf("%s", name);
if (pos.k_entry != MAXINT && name) {
char * entry_name;