Fixed profiling of trap, syscall and interrupt handlers and some

ordinary functions, essentially by backing out half of rev.1.115 of
amd64/exception.S.  The handlers must be between certain labels for
the purposes of profiling, and this was broken by scattering them in
separately compiled .S files, especially for ordinary functions that
ended up between the labels.  Merge the files by #including them as
before, except with different pathnames and better comments and
organization.  Changes to the scattered files are minimal -- just
move the labels to the file that does the #includes.

This also partly fixes profiling of IPIs -- all IPI handlers are now
correctly classified as interrupt handlers, but many are still missing
mcount calls.

sys/amd64/amd64/apic_vector.S

@@ -129,7 +129,6 @@ IDTVEC(spuriousint)
 
 	iretq
 
-MCOUNT_LABEL(bintr2)
 	ISR_VEC(1, apic_isr1)
 	ISR_VEC(2, apic_isr2)
 	ISR_VEC(3, apic_isr3)
@@ -137,7 +136,6 @@ MCOUNT_LABEL(bintr2)
 	ISR_VEC(5, apic_isr5)
 	ISR_VEC(6, apic_isr6)
 	ISR_VEC(7, apic_isr7)
-MCOUNT_LABEL(eintr2)
 
 #ifdef SMP
 /*

sys/amd64/amd64/exception.S

@@ -30,6 +30,9 @@
  * $FreeBSD$
  */
 
+#include "opt_atpic.h"
+#include "opt_ia32.h"
+
 #include <machine/asmacros.h>
 #include <machine/psl.h>
 #include <machine/trap.h>
@@ -294,8 +297,44 @@ ENTRY(fork_trampoline)
 	MEXITCOUNT
 	jmp	doreti			/* Handle any ASTs */
 
+/*
+ * To efficiently implement classification of trap and interrupt handlers
+ * for profiling, there must be only trap handlers between the labels btrap
+ * and bintr, and only interrupt handlers between the labels bintr and
+ * eintr.  This is implemented (partly) by including files that contain
+ * some of the handlers.  Before including the files, set up a normal asm
+ * environment so that the included files doen't need to know that they are
+ * included.
+ */
+
+#ifdef IA32
 	.data
-	ALIGN_DATA
+	.align	4
+	.text
+	SUPERALIGN_TEXT
+
+#include <amd64/ia32/ia32_exception.S>
+#endif
+
+	.data
+	.align	4
+	.text
+	SUPERALIGN_TEXT
+MCOUNT_LABEL(bintr)
+
+#include <amd64/amd64/apic_vector.S>
+
+#ifdef DEV_ATPIC
+	.data
+	.align	4
+	.text
+	SUPERALIGN_TEXT
+
+#include <amd64/isa/atpic_vector.S>
+#endif
+
+	.text
+MCOUNT_LABEL(eintr)
 
 /*
  * void doreti(struct trapframe)
@@ -304,7 +343,6 @@ ENTRY(fork_trampoline)
  */
 	.text
 	SUPERALIGN_TEXT
-	.globl	doreti
 	.type	doreti,@function
 doreti:
 	FAKE_MCOUNT($bintr)		/* init "from" bintr -> doreti */

sys/amd64/isa/atpic_vector.S

@@ -72,7 +72,6 @@ IDTVEC(vec_name) ;							\
 	MEXITCOUNT ;							\
 	jmp	doreti
 
-MCOUNT_LABEL(bintr)
 	INTR(0, atpic_intr0)
 	INTR(1, atpic_intr1)
 	INTR(2, atpic_intr2)
@@ -89,4 +88,3 @@ MCOUNT_LABEL(bintr)
 	INTR(13, atpic_intr13)
 	INTR(14, atpic_intr14)
 	INTR(15, atpic_intr15)
-MCOUNT_LABEL(eintr)

sys/conf/files.amd64

@@ -41,7 +41,7 @@ amd64/acpica/acpi_wakeup.c	optional	acpi
 amd64/acpica/madt.c		optional	acpi
 amd64/amd64/amd64-gdbstub.c	optional	ddb
 amd64/amd64/amd64_mem.c		standard
-amd64/amd64/apic_vector.S	standard
+#amd64/amd64/apic_vector.S	standard
 amd64/amd64/atomic.c		standard
 amd64/amd64/autoconf.c		standard
 amd64/amd64/busdma_machdep.c	standard
@@ -80,7 +80,7 @@ amd64/amd64/uio_machdep.c	standard
 amd64/amd64/uma_machdep.c	standard
 amd64/amd64/vm_machdep.c	standard
 amd64/isa/atpic.c		optional	atpic isa
-amd64/isa/atpic_vector.S	optional	atpic isa
+#amd64/isa/atpic_vector.S	optional	atpic isa
 amd64/isa/clock.c		standard
 amd64/isa/elcr.c		standard
 amd64/isa/isa.c			standard
@@ -140,7 +140,7 @@ pci/agp_via.c			optional	agp
 #
 # IA32 binary support
 #
-amd64/ia32/ia32_exception.S	optional	ia32
+#amd64/ia32/ia32_exception.S	optional	ia32
 amd64/ia32/ia32_signal.c	optional	ia32
 amd64/ia32/ia32_sigtramp.S	optional	ia32
 amd64/ia32/ia32_syscall.c	optional	ia32