Implement Dynamic Profiling, a ddb(4) based & gprof compatible kernel

profiling framework, for i386.

Code patching is used to enable probes when entering functions.  The
probes will call a mcount()-like function to match the behavior of a
GPROF kernel.

A new sysctl knob, ddb.profile, need to be set to 1 in securelevel 0
to be able to use this feature.

ok jasper@, guenther@, mlarkin@

6 files changed, 113 insertions, 13 deletions

diff --git a/sys/arch/amd64/amd64/db_trace.c b/sys/arch/amd64/amd64/db_trace.c
index 3615b2c84df..a24b8ba0371 100644
--- a/sys/arch/amd64/amd64/db_trace.c
+++ b/sys/arch/amd64/amd64/db_trace.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: db_trace.c,v 1.24 2017/01/24 00:58:55 mpi Exp $	*/
+/*	$OpenBSD: db_trace.c,v 1.25 2017/02/06 09:13:41 mpi Exp $	*/
 /*	$NetBSD: db_trace.c,v 1.1 2003/04/26 18:39:27 fvdl Exp $	*/
 
 /*
@@ -325,9 +325,9 @@ db_stack_trace_print(db_expr_t addr, boolean_t have_addr, db_expr_t count,
 vaddr_t
 db_get_pc(struct trapframe *tf)
 {
-	struct callframe *cf = (struct callframe *)(tf->tf_rsp - 8);
+	struct callframe *cf = (struct callframe *)(tf->tf_rsp - sizeof(long));
 
-	return db_get_value((db_addr_t)&cf->f_retaddr, 8, 0);
+	return db_get_value((db_addr_t)&cf->f_retaddr, sizeof(long), 0);
 }
 
 vaddr_t
diff --git a/sys/arch/i386/i386/db_trace.c b/sys/arch/i386/i386/db_trace.c
index 0a49cdbae50..4ef17ce2333 100644
--- a/sys/arch/i386/i386/db_trace.c
+++ b/sys/arch/i386/i386/db_trace.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: db_trace.c,v 1.23 2017/01/24 00:58:55 mpi Exp $	*/
+/*	$OpenBSD: db_trace.c,v 1.24 2017/02/06 09:13:41 mpi Exp $	*/
 /*	$NetBSD: db_trace.c,v 1.18 1996/05/03 19:42:01 christos Exp $	*/
 
 /*
@@ -319,3 +319,22 @@ db_stack_trace_print(db_expr_t addr, boolean_t have_addr, db_expr_t count,
 		(*pr)(":\n");
 	}
 }
+
+vaddr_t
+db_get_pc(struct trapframe *tf)
+{
+	struct callframe *cf;
+
+	if (KERNELMODE(tf->tf_cs, tf->tf_eflags))
+		cf = (struct callframe *)((long)&tf->tf_esp - sizeof(long));
+	else
+		cf = (struct callframe *)(tf->tf_esp - sizeof(long));
+
+	return db_get_value((db_addr_t)&cf->f_retaddr, sizeof(long), 0);
+}
+
+vaddr_t
+db_get_probe_addr(struct trapframe *tf)
+{
+	return tf->tf_eip - BKPT_SIZE;
+}
diff --git a/sys/arch/i386/i386/locore.s b/sys/arch/i386/i386/locore.s
index 82e4a1df64a..9969b8e782c 100644
--- a/sys/arch/i386/i386/locore.s
+++ b/sys/arch/i386/i386/locore.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.s,v 1.171 2016/10/13 09:01:04 mlarkin Exp $	*/
+/*	$OpenBSD: locore.s,v 1.172 2017/02/06 09:13:41 mpi Exp $	*/
 /*	$NetBSD: locore.s,v 1.145 1996/05/03 19:41:19 christos Exp $	*/
 
 /*-
@@ -133,7 +133,8 @@
 	pushl	%fs		; \
 	movl	$GSEL(GCPU_SEL, SEL_KPL),%eax	; \
 	movw	%ax,%fs
-#define	INTRFASTEXIT \
+
+#define	INTR_RESTORE_ALL \
 	popl	%fs		; \
 	popl	%gs		; \
 	popl	%es		; \
@@ -144,10 +145,14 @@
 	popl	%ebx		; \
 	popl	%edx		; \
 	popl	%ecx		; \
-	popl	%eax		; \
+	popl	%eax
+
+#define	INTRFASTEXIT \
+	INTR_RESTORE_ALL	;\
 	addl	$8,%esp		; \
 	iret
 
+#define	INTR_FAKE_TRAP	0xbadabada
 
 /*
  * PTmap is recursive pagemap at top of virtual address space.
@@ -1445,6 +1450,24 @@ calltrap:
 #ifdef DIAGNOSTIC
 	movl	CPL,%ebx
 #endif /* DIAGNOSTIC */
+#if !defined(GPROF) && defined(DDBPROF)
+	cmpl	$T_BPTFLT,TF_TRAPNO(%esp)
+	jne	.Lreal_trap
+
+	pushl	%esp
+	call	_C_LABEL(db_prof_hook)
+	addl	$4,%esp
+	cmpl	$1,%eax
+	jne	.Lreal_trap
+
+	/*
+	 * Abuse the error field to indicate that INTRFASTEXIT needs
+	 * to emulate the patched instruction.
+	 */
+	movl	$INTR_FAKE_TRAP, TF_ERR(%esp)
+	jz	2f
+.Lreal_trap:
+#endif /* !defined(GPROF) && defined(DDBPROF) */
 	pushl	%esp
 	call	_C_LABEL(trap)
 	addl	$4,%esp
@@ -1464,10 +1487,22 @@ calltrap:
 	call	_C_LABEL(ast)
 	addl	$4,%esp
 	jmp	2b
+1:
+#if !defined(GPROF) && defined(DDBPROF)
+	/*
+	 * If we are returning from a probe trap we need to fix the
+	 * stack layout and emulate the patched instruction.
+	 *
+	 * The code below does that by trashing %eax, so it MUST be
+	 * restored afterward.
+	 */
+	cmpl	$INTR_FAKE_TRAP, TF_ERR(%esp)
+	je	.Lprobe_fixup
+#endif /* !defined(GPROF) && defined(DDBPROF) */
 #ifndef DIAGNOSTIC
-1:	INTRFASTEXIT
+	INTRFASTEXIT
 #else
-1:	cmpl	CPL,%ebx
+	cmpl	CPL,%ebx
 	jne	3f
 	INTRFASTEXIT
 3:	sti
@@ -1482,6 +1517,33 @@ calltrap:
 4:	.asciz	"WARNING: SPL NOT LOWERED ON TRAP EXIT\n"
 #endif /* DIAGNOSTIC */
 
+#if !defined(GPROF) && defined(DDBPROF)
+.Lprobe_fixup:
+	/* Restore all register unwinding the stack. */
+	INTR_RESTORE_ALL
+
+	/*
+	 * Use the space left by ``err'' and ``trapno'' to emulate
+	 * "pushl %ebp".
+	 *
+	 * Temporarily save %eax.
+	 */
+	movl	%eax,0(%esp)
+
+	/* Shift hardware-saved registers: eip, cs, eflags */
+	movl	8(%esp),%eax
+	movl	%eax,4(%esp)
+	movl	12(%esp),%eax
+	movl	%eax,8(%esp)
+	movl	16(%esp),%eax
+	movl	%eax,12(%esp)
+
+	/* Store %ebp in the expected location to finish the emulation. */
+	movl	%ebp,16(%esp)
+
+	popl	%eax
+	iret
+#endif /* !defined(GPROF) && defined(DDBPROF) */
 /*
  * Trap gate entry for syscall
  */
diff --git a/sys/arch/i386/include/cpu.h b/sys/arch/i386/include/cpu.h
index 27980979b6d..bde738031e7 100644
--- a/sys/arch/i386/include/cpu.h
+++ b/sys/arch/i386/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.151 2017/01/03 09:48:15 mlarkin Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.152 2017/02/06 09:13:41 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.35 1996/05/05 19:29:26 christos Exp $	*/
 
 /*-
@@ -185,7 +185,7 @@ struct cpu_info {
 
 	struct ksensordev	ci_sensordev;
 	struct ksensor		ci_sensor;
-#ifdef GPROF
+#if defined(GPROF) || defined(DDBPROF)
 	struct gmonparam	*ci_gmon;
 #endif
 	u_int32_t		ci_vmm_flags;
diff --git a/sys/arch/i386/include/cpufunc.h b/sys/arch/i386/include/cpufunc.h
index 7b025d1bbd7..2262237c229 100644
--- a/sys/arch/i386/include/cpufunc.h
+++ b/sys/arch/i386/include/cpufunc.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpufunc.h,v 1.22 2014/03/29 18:09:29 guenther Exp $	*/
+/*	$OpenBSD: cpufunc.h,v 1.23 2017/02/06 09:13:41 mpi Exp $	*/
 /*	$NetBSD: cpufunc.h,v 1.8 1994/10/27 04:15:59 cgd Exp $	*/
 
 /*
@@ -215,6 +215,22 @@ write_eflags(u_int ef)
 	__asm volatile("pushl %0; popfl" : : "r" (ef));
 }
 
+static inline u_long
+intr_disable(void)
+{
+	u_long ef;
+
+	ef = read_eflags();
+	disable_intr();
+	return (ef);
+}
+
+static inline void
+intr_restore(u_long ef)
+{
+	write_eflags(ef);
+}
+
 static __inline void
 wbinvd(void)
 {
diff --git a/sys/arch/i386/include/db_machdep.h b/sys/arch/i386/include/db_machdep.h
index 6866b8b316d..5d9fb9bd5e8 100644
--- a/sys/arch/i386/include/db_machdep.h
+++ b/sys/arch/i386/include/db_machdep.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: db_machdep.h,v 1.25 2016/04/27 11:10:48 mpi Exp $	*/
+/*	$OpenBSD: db_machdep.h,v 1.26 2017/02/06 09:13:41 mpi Exp $	*/
 /*	$NetBSD: db_machdep.h,v 1.9 1996/05/03 19:23:59 christos Exp $	*/
 
 /* 
@@ -51,6 +51,9 @@ extern db_regs_t	ddb_regs;	/* register state */
 #define	BKPT_SIZE	(1)		/* size of breakpoint inst */
 #define	BKPT_SET(inst)	(BKPT_INST)
 
+#define	SSF_INST	0x55
+#define	SSF_SIZE	(1)
+
 #define	FIXUP_PC_AFTER_BREAK(regs)	((regs)->tf_eip -= BKPT_SIZE)
 
 #define	db_clear_single_step(regs)	((regs)->tf_eflags &= ~PSL_T)