PERFORCE change 132683 for review

[John Birrell]

http://perforce.freebsd.org/chv.cgi?CH=132683

Change 132683 by jb at jb_freebsd1 on 2008/01/07 07:33:25

	Add the amd64 code to the Function Boundary Trace (fbt) provider.
	This is just OpenSolaris code rearranged to suit FreeBSD's linker
	sets.

Affected files ...

.. //depot/projects/dtrace/src/sys/cddl/dev/fbt/fbt.c#6 edit

Differences ...

==== //depot/projects/dtrace/src/sys/cddl/dev/fbt/fbt.c#6 (text+ko) ====

@@ -93,7 +93,7 @@
 #define	FBT_RET_IMM16		0xc2
 #define	FBT_LEAVE		0xc9
 
-#ifdef __amd64
+#ifdef __amd64__
 #define	FBT_PATCHVAL		0xcc
 #else
 #define	FBT_PATCHVAL		0xf0
@@ -198,6 +198,19 @@
 
 				cpu->cpu_dtrace_caller = 0;
 			} else {
+#ifdef __amd64__
+				/*
+				 * On amd64, we instrument the ret, not the
+				 * leave.  We therefore need to set the caller
+				 * to assure that the top frame of a stack()
+				 * action is correct.
+				 */
+				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
+				cpu->cpu_dtrace_caller = stack[0];
+				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT |
+				    CPU_DTRACE_BADADDR);
+#endif
+
 				dtrace_probe(fbt->fbtp_id, fbt->fbtp_roffset,
 				    rval, 0, 0, 0);
 				cpu->cpu_dtrace_caller = 0;
@@ -231,46 +244,6 @@
 		return (0);
 	}
 
-#ifdef DOODAD
-	/*
-	 * Still trying to narrow down why some of these things are
-	 * being called from the probe context. At the moment I suspect
-	 * there is some call which executes witness code and that
-	 * is *certain* to end in tears.
-	 */
-	if (name[0] == '_' ||
-	    strcmp(name, "trap") == 0 ||
-	    strcmp(name, "userret") == 0 ||
-	    strncmp(name, "realtimer", 9) == 0 ||
-	    strncmp(name, "ithread", 7) == 0 ||
-	    strncmp(name, "itimer", 6) == 0 ||
-	    strncmp(name, "ioapic", 6) == 0 ||
-	    strncmp(name, "isitmy", 6) == 0 ||
-	    strncmp(name, "db_", 3) == 0 ||
-	    strncmp(name, "gdb_", 4) == 0 ||
-	    strncmp(name, "kdb_", 4) == 0 ||
-	    strncmp(name, "intr_", 5) == 0 ||
-
-	    strncmp(name, "X_", 2) == 0 ||
-	    strncmp(name, "vm", 2) == 0 ||
-	    strncmp(name, "th", 2) == 0 ||
-	    strncmp(name, "sh", 2) == 0 ||
-	    strncmp(name, "fi", 2) == 0 ||
-	    strncmp(name, "fd", 2) == 0 ||
-	    strncmp(name, "elf", 3) == 0 ||
-	    strncmp(name, "str", 3) == 0 ||
-	    strncmp(name, "nfs", 3) == 0 ||
-	    strncmp(name, "ata_", 4) == 0 ||
-	    strncmp(name, "acpi", 4) == 0 ||
-	    strncmp(name, "link_", 5) == 0 ||
-	    strncmp(name, "bufobj", 6) == 0 ||
-	    strncmp(name, "linker", 6) == 0 ||
-	    strncmp(name, "witness", 7) == 0 ||
-	    strncmp(name, "spinlock", 8) == 0 ||
-	    strcmp(name, "pcpu_find") == 0)	/* used in stack() impl */
-		return (0);
-#endif
-
 	/*
 	 * These filters are set by sysctl to help debugging the initial
 	 * port of this provider. Normally the wildcard would be 
@@ -309,6 +282,26 @@
 	instr = (u_int8_t *) symval->value;
 	limit = (u_int8_t *) symval->value + symval->size;
 
+#ifdef __amd64__
+	while (instr < limit) {
+		if (*instr == FBT_PUSHL_EBP)
+			break;
+
+		if ((size = dtrace_instr_size(instr)) <= 0)
+			break;
+
+		instr += size;
+	}
+
+	if (instr >= limit || *instr != FBT_PUSHL_EBP) {
+		/*
+		 * We either don't save the frame pointer in this
+		 * function, or we ran into some disassembly
+		 * screw-up.  Either way, we bail.
+		 */
+		return (0);
+	}
+#else
 	if (instr[0] != FBT_PUSHL_EBP)
 		return (0);
 
@@ -317,6 +310,7 @@
 	    !(instr[1] == FBT_MOVL_ESP_EBP0_V1 &&
 	    instr[2] == FBT_MOVL_ESP_EBP1_V1))
 		return (0);
+#endif
 
 	fbt = malloc(sizeof (fbt_probe_t), M_FBT, M_WAITOK | M_ZERO);
 	fbt->fbtp_name = name;
@@ -347,6 +341,17 @@
 	if ((size = dtrace_instr_size(instr)) <= 0)
 		return (0);
 
+#ifdef __amd64__
+	/*
+	 * We only instrument "ret" on amd64 -- we don't yet instrument
+	 * ret imm16, largely because the compiler doesn't seem to
+	 * (yet) emit them in the kernel...
+	 */
+	if (*instr != FBT_RET) {
+		instr += size;
+		goto again;
+	}
+#else
 	if (!(size == 1 &&
 	    (*instr == FBT_POPL_EBP || *instr == FBT_LEAVE) &&
 	    (*(instr + 1) == FBT_RET ||
@@ -354,6 +359,7 @@
 		instr += size;
 		goto again;
 	}
+#endif
 
 	/*
 	 * We (desperately) want to avoid erroneously instrumenting a
@@ -403,6 +409,7 @@
 	fbt->fbtp_ctl = lf;
 	fbt->fbtp_loadcnt = lf->loadcnt;
 
+#ifndef __amd64__
 	if (*instr == FBT_POPL_EBP) {
 		fbt->fbtp_rval = DTRACE_INVOP_POPL_EBP;
 	} else {
@@ -412,6 +419,13 @@
 	fbt->fbtp_roffset =
 	    (uintptr_t)(instr - (uint8_t *) symval->value) + 1;
 
+#else
+	ASSERT(*instr == FBT_RET);
+	fbt->fbtp_rval = DTRACE_INVOP_RET;
+	fbt->fbtp_roffset =
+	    (uintptr_t)(instr - (uint8_t *) symval->value);
+#endif
+
 	fbt->fbtp_savedval = *instr;
 	fbt->fbtp_patchval = FBT_PATCHVAL;
 	fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)];
@@ -471,7 +485,7 @@
 	/*
 	 * List the functions in the module and the symbol values.
 	 */
-	linker_file_function_listall(lf, fbt_provide_module_function, modname);
+	(void) linker_file_function_listall(lf, fbt_provide_module_function, modname);
 }
 
 /* ARGSUSED */