unsigned int *ptr = (unsigned int *)pc;
unsigned long result = (unsigned long) *ptr;
*ptr = BREAKPOINT_INST;
- *(ptr+1)=trap_Breakpoint;
os_flush_icache((os_vm_address_t)ptr, sizeof(unsigned long));
int op = orig_inst >> 26;
int reg_a = (orig_inst >> 21) & 0x1f;
int reg_b = (orig_inst >> 16) & 0x1f;
- int fn = orig_inst & 0xffff;
int disp = (orig_inst&(1<<20)) ? orig_inst | (-1 << 21) : orig_inst&0x1fffff;
int next_pc = *os_context_pc_addr(context);
int branch = 0; /* was NULL; */
static sigset_t orig_sigmask;
+/* Perform the instruction that we overwrote with a breakpoint. As we
+ * don't have a single-step facility, this means we have to:
+ * - put the instruction back
+ * - put a second breakpoint at the following instruction,
+ * set after_breakpoint and continue execution.
+ *
+ * When the second breakpoint is hit (very shortly thereafter, we hope)
+ * sigtrap_handler gets called again, but follows the AfterBreakpoint
+ * arm, which
+ * - puts a bpt back in the first breakpoint place (running across a
+ * breakpoint shouldn't cause it to be uninstalled)
+ * - replaces the second bpt with the instruction it was meant to be
+ * - carries on
+ *
+ * Clear?
+ */
+
void arch_do_displaced_inst(os_context_t *context,unsigned int orig_inst)
{
+ /* Apparent off-by-one errors ahoy. If you consult the Alpha ARM,
+ * it will tell you that after a BPT, the saved PC is the address
+ * of the instruction _after_ the instruction that caused the trap.
+ *
+ * However, we decremented PC by 4 before calling the Lisp-level
+ * handler that calls this routine (see alpha-arch.c line 322 and
+ * friends) so when we get to this point PC is actually pointing
+ * at the BPT instruction itself. This is good, because this is
+ * where we want to restart execution when we do that */
+
unsigned int *pc=(unsigned int *)(*os_context_pc_addr(context));
unsigned int *next_pc;
- unsigned int next_inst;
int op = orig_inst >> 26;;
- fprintf(stderr,"arch_do_displaced_inst depends on sigreturn, which is not implemented and will\nalways fail\n");
+
orig_sigmask = *os_context_sigmask_addr(context);
sigaddset_blockable(os_context_sigmask_addr(context));
- /* Figure out where the displaced inst is going. */
- if(op == 0x1a || (op&0xf) == 0x30) /* branch...ugh */
- /* The cast to long is just to shut gcc up. */
- next_pc = (unsigned int *)((long)emulate_branch(context,orig_inst));
- else
- next_pc = pc+1;
-
/* Put the original instruction back. */
*pc = orig_inst;
os_flush_icache((os_vm_address_t)pc, sizeof(unsigned long));
skipped_break_addr = pc;
+ /* Figure out where we will end up after running the displaced
+ * instruction */
+ if(op == 0x1a || (op&0xf) == 0x30) /* a branch */
+ /* The cast to long is just to shut gcc up. */
+ next_pc = (unsigned int *)((long)emulate_branch(context,orig_inst));
+ else
+ next_pc = pc+1;
+
/* Set the after breakpoint. */
displaced_after_inst = *next_pc;
*next_pc = BREAKPOINT_INST;
after_breakpoint=1;
os_flush_icache((os_vm_address_t)next_pc, sizeof(unsigned long));
-
- monitor_or_something();
- sigreturn(context);
-}
-
-#define AfterBreakpoint 100
-
-static void
-sigill_handler(int signal, siginfo_t *siginfo, os_context_t *context) {
- fake_foreign_function_call(context);
- monitor_or_something();
}
static void
sigtrap_handler(int signal, siginfo_t *siginfo, os_context_t *context)
{
+ unsigned int code;
+
/* Don't disallow recursive breakpoint traps. Otherwise, we can't */
/* use debugger breakpoints anywhere in here. */
-
sigset_t *mask=(os_context_sigmask_addr(context));
- unsigned int code;
- fprintf(stderr,"sigtrap_handler:signal %d context=%p ",signal,context);
sigsetmask(mask);
- /* this is different from how CMUCL does it. CMUCL used
- * "call_pal PAL_gentrap", which doesn't do anything on Linux
- * so screwed up our offsets in odd ways. We use "bpt" instead
- */
-
- /* probably we should
- assert(*(unsigned int*)(*os_context_pc_addr(context)-4) == BREAKPOINT_INST)
- but I've not decided a good way to handle it if it turns out not to be
- */
+ /* this is different from how CMUCL does it. CMUCL used "call_pal
+ * PAL_gentrap", which doesn't do anything on Linux (unless NL0
+ * contains certain specific values). We use "bugchk" instead.
+ * It's (for our purposes) just the same as bpt but has a
+ * different opcode so we can test whether we're dealing with a
+ * breakpoint or a "system service" */
+
+ if((*(unsigned int*)(*os_context_pc_addr(context)-4))== BREAKPOINT_INST) {
+ if(after_breakpoint) {
+ /* see comments above arch_do_displaced_inst. This is where
+ * we reinsert the breakpoint that we removed earlier */
+
+ *os_context_pc_addr(context) -=4;
+ *skipped_break_addr = BREAKPOINT_INST;
+ os_flush_icache((os_vm_address_t)skipped_break_addr,
+ sizeof(unsigned long));
+ skipped_break_addr = NULL;
+ *(unsigned int *)*os_context_pc_addr(context) =
+ displaced_after_inst;
+ os_flush_icache((os_vm_address_t)*os_context_pc_addr(context), sizeof(unsigned long));
+ *os_context_sigmask_addr(context)= orig_sigmask;
+ after_breakpoint=0; /* false */
+ return;
+ } else
+ code = trap_Breakpoint;
+ } else
+ /* a "system service" */
code=*((u32 *)(*os_context_pc_addr(context)));
- fprintf(stderr,"pc=%lx code=%d, inst=%x\n",
- *os_context_pc_addr(context), code,
- *(unsigned int*)(*os_context_pc_addr(context)-4));
+
switch (code) {
case trap_PendingInterrupt:
arch_skip_instruction(context);
interrupt_internal_error(signal, siginfo, context, code==trap_Cerror);
break;
- case trap_Breakpoint:
+ case trap_Breakpoint: /* call lisp-level handler */
*os_context_pc_addr(context) -=4;
handle_breakpoint(signal, siginfo, context);
break;
case trap_FunctionEndBreakpoint:
*os_context_pc_addr(context) -=4;
- *os_context_pc_addr(context) = (int)handle_function_end_breakpoint(signal, siginfo, context);
- break;
-
- case AfterBreakpoint:
- *os_context_pc_addr(context) -=4;
- *skipped_break_addr = BREAKPOINT_INST;
- os_flush_icache((os_vm_address_t)skipped_break_addr,
- sizeof(unsigned long));
- skipped_break_addr = NULL;
- *(unsigned int *)*os_context_pc_addr(context) = displaced_after_inst;
- os_flush_icache((os_vm_address_t)*os_context_pc_addr(context), sizeof(unsigned long));
- *os_context_sigmask_addr(context)= orig_sigmask;
- after_breakpoint=0; /* NULL; */
+ *os_context_pc_addr(context) =
+ (int)handle_function_end_breakpoint(signal, siginfo, context);
break;
default:
+ fprintf(stderr, "unidetified breakpoint/trap %d\n",code);
interrupt_handle_now(signal, siginfo, context);
break;
}
}
-#define FIXNUM_VALUE(lispobj) (((int)lispobj)>>2)
-
static void sigfpe_handler(int signal, int code, os_context_t *context)
{
+ /* what should this contain? interesting question. If it really
+ * is empty, why don't we just ignore the signal? -dan 2001.08.10
+ */
}
void arch_install_interrupt_handlers()
{
- undoably_install_low_level_interrupt_handler(SIGILL, sigill_handler);
undoably_install_low_level_interrupt_handler(SIGTRAP, sigtrap_handler);
undoably_install_low_level_interrupt_handler(SIGFPE, sigfpe_handler);
}
return call_into_lisp(function, args, 3);
}
-
-/* This is apparently called by emulate_branch, but isn't defined. So */
-/* just do nothing and hope it works... */
-void cacheflush(void)
-{
- /* hoping probably isn't _actually_ enough. we should call_pal imb,
- according to the arch ref manual
- */
-}
projects / sbcl.git / commitdiff