diff --git a/sys/amd64/amd64/genassym.c b/sys/amd64/amd64/genassym.c
index e4fd1961ac19..705e3fd1c230 100644
--- a/sys/amd64/amd64/genassym.c
+++ b/sys/amd64/amd64/genassym.c
@@ -185,6 +185,7 @@ ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread));
 ASSYM(PC_FPCURTHREAD, offsetof(struct pcpu, pc_fpcurthread));
 ASSYM(PC_IDLETHREAD, offsetof(struct pcpu, pc_idlethread));
 ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
+ASSYM(PC_CONS_BUFR, offsetof(struct pcpu, pc_cons_bufr));
 ASSYM(PC_CPUID, offsetof(struct pcpu, pc_cpuid));
 ASSYM(PC_SCRATCH_RSP, offsetof(struct pcpu, pc_scratch_rsp));
 ASSYM(PC_CURPMAP, offsetof(struct pcpu, pc_curpmap));
diff --git a/sys/arm/arm/genassym.c b/sys/arm/arm/genassym.c
index b179e7475ed4..f61e1f718cab 100644
--- a/sys/arm/arm/genassym.c
+++ b/sys/arm/arm/genassym.c
@@ -69,6 +69,7 @@ ASSYM(PCB_R12, offsetof(struct pcb, un_32.pcb32_r12));
 ASSYM(PCB_PC, offsetof(struct pcb, un_32.pcb32_pc));
 ASSYM(PCB_SP, offsetof(struct pcb, un_32.pcb32_sp));
 
+ASSYM(PC_CONS_BUFR, offsetof(struct pcpu, pc_cons_bufr));
 ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
 ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread));
 ASSYM(M_LEN, offsetof(struct mbuf, m_len));
diff --git a/sys/i386/i386/genassym.c b/sys/i386/i386/genassym.c
index ca24402ae486..01fb5fad50b5 100644
--- a/sys/i386/i386/genassym.c
+++ b/sys/i386/i386/genassym.c
@@ -195,6 +195,7 @@ ASSYM(PC_COMMON_TSSD, offsetof(struct pcpu, pc_common_tssd));
 ASSYM(PC_TSS_GDT, offsetof(struct pcpu, pc_tss_gdt));
 ASSYM(PC_FSGS_GDT, offsetof(struct pcpu, pc_fsgs_gdt));
 ASSYM(PC_CURRENTLDT, offsetof(struct pcpu, pc_currentldt));
+ASSYM(PC_CONS_BUFR, offsetof(struct pcpu, pc_cons_bufr));
 ASSYM(PC_CPUID, offsetof(struct pcpu, pc_cpuid));
 ASSYM(PC_CURPMAP, offsetof(struct pcpu, pc_curpmap));
 ASSYM(PC_PRIVATE_TSS, offsetof(struct pcpu, pc_private_tss));
diff --git a/sys/ia64/ia64/genassym.c b/sys/ia64/ia64/genassym.c
index 8e988b675452..cee754ce215b 100644
--- a/sys/ia64/ia64/genassym.c
+++ b/sys/ia64/ia64/genassym.c
@@ -91,6 +91,7 @@ ASSYM(MC_SPECIAL_RNAT,	offsetof(mcontext_t, mc_special.rnat));
 ASSYM(PAGE_SHIFT,	PAGE_SHIFT);
 ASSYM(PAGE_SIZE,	PAGE_SIZE);
 
+ASSYM(PC_CONS_BUFR,	offsetof(struct pcpu, pc_cons_bufr));
 ASSYM(PC_CPUID,		offsetof(struct pcpu, pc_cpuid));
 ASSYM(PC_CURRENT_PMAP,	offsetof(struct pcpu, pc_current_pmap));
 ASSYM(PC_CURTHREAD,	offsetof(struct pcpu, pc_curthread));
diff --git a/sys/kern/subr_prf.c b/sys/kern/subr_prf.c
index bef5566994b6..b6ffc91fc1ac 100644
--- a/sys/kern/subr_prf.c
+++ b/sys/kern/subr_prf.c
@@ -78,6 +78,10 @@ struct putchar_arg {
 	int	flags;
 	int	pri;
 	struct	tty *tty;
+	char	*p_bufr;
+	size_t	n_bufr;
+	char	*p_next;
+	size_t	remain;
 };
 
 struct snprintf_arg {
@@ -92,7 +96,6 @@ static void  putchar(int ch, void *arg);
 static char *ksprintn(char *nbuf, uintmax_t num, int base, int *len, int upper);
 static void  snprintf_func(int ch, void *arg);
 
-static int consintr = 1;		/* Ok to handle console interrupts? */
 static int msgbufmapped;		/* Set when safe to use msgbuf */
 int msgbuftrigger;
 
@@ -234,6 +237,7 @@ log(int level, const char *fmt, ...)
 	pca.tty = NULL;
 	pca.pri = level;
 	pca.flags = log_open ? TOLOG : TOCONS;
+	pca.p_bufr = NULL;
 
 	va_start(ap, fmt);
 	kvprintf(fmt, putchar, &pca, 10, ap);
@@ -284,43 +288,96 @@ int
 printf(const char *fmt, ...)
 {
 	va_list ap;
-	int savintr;
 	struct putchar_arg pca;
 	int retval;
 
-	savintr = consintr;		/* disable interrupts */
-	consintr = 0;
+	critical_enter();
+
 	va_start(ap, fmt);
 	pca.tty = NULL;
 	pca.flags = TOCONS | TOLOG;
 	pca.pri = -1;
+	pca.p_bufr = (char *) PCPU_PTR(cons_bufr);
+	pca.p_next = pca.p_bufr;
+	pca.n_bufr = PCPU_CONS_BUFR;
+	pca.remain = PCPU_CONS_BUFR;
+	*pca.p_next = '\0';
+
 	retval = kvprintf(fmt, putchar, &pca, 10, ap);
 	va_end(ap);
+
+	/* Write any buffered console output: */
+	if (*pca.p_bufr != '\0')
+		cnputs(pca.p_bufr);
+
 	if (!panicstr)
 		msgbuftrigger = 1;
-	consintr = savintr;		/* reenable interrupts */
+
+	critical_exit();
+
 	return (retval);
 }
 
 int
 vprintf(const char *fmt, va_list ap)
 {
-	int savintr;
 	struct putchar_arg pca;
 	int retval;
 
-	savintr = consintr;		/* disable interrupts */
-	consintr = 0;
+	critical_enter();
+
 	pca.tty = NULL;
 	pca.flags = TOCONS | TOLOG;
 	pca.pri = -1;
+	pca.p_bufr = (char *) PCPU_PTR(cons_bufr);
+	pca.p_next = pca.p_bufr;
+	pca.n_bufr = PCPU_CONS_BUFR;
+	pca.remain = PCPU_CONS_BUFR;
+	*pca.p_next = '\0';
+
 	retval = kvprintf(fmt, putchar, &pca, 10, ap);
+
+	/* Write any buffered console output: */
+	if (*pca.p_bufr != '\0')
+		cnputs(pca.p_bufr);
+
 	if (!panicstr)
 		msgbuftrigger = 1;
-	consintr = savintr;		/* reenable interrupts */
+
+	critical_exit();
+
 	return (retval);
 }
 
+static void
+putcons(int c, struct putchar_arg *ap)
+{
+	/* Check if no console output buffer was provided. */
+	if (ap->p_bufr == NULL)
+		/* Output direct to the console. */
+		cnputc(c);
+	else {
+		/* Buffer the character: */
+		if (c == '\n') {
+			*ap->p_next++ = '\r';
+			ap->remain--;
+		}
+		*ap->p_next++ = c;
+		ap->remain--;
+
+		/* Always leave the buffer zero terminated. */
+		*ap->p_next = '\0';
+
+		/* Check if the buffer needs to be flushed. */
+		if (ap->remain < 3 || c == '\n') {
+			cnputs(ap->p_bufr);
+			ap->p_next = ap->p_bufr;
+			ap->remain = ap->n_bufr;
+			*ap->p_next = '\0';
+		}
+	}
+}
+
 /*
  * Print a character on console or users terminal.  If destination is
  * the console then the last bunch of characters are saved in msgbuf for
@@ -331,17 +388,15 @@ putchar(int c, void *arg)
 {
 	struct putchar_arg *ap = (struct putchar_arg*) arg;
 	struct tty *tp = ap->tty;
-	int consdirect, flags = ap->flags;
+	int flags = ap->flags;
 
-	consdirect = ((flags & TOCONS) && constty == NULL);
 	/* Don't use the tty code after a panic or while in ddb. */
-	if (panicstr)
-		consdirect = 1;
-	if (kdb_active)
-		consdirect = 1;
-	if (consdirect) {
+	if (kdb_active) {
 		if (c != '\0')
 			cnputc(c);
+	} else if (panicstr || ((flags & TOCONS) && constty == NULL)) {
+		if (c != '\0')
+			putcons(c, ap);
 	} else {
 		if ((flags & TOTTY) && tp != NULL)
 			tputchar(c, tp);
@@ -349,7 +404,7 @@ putchar(int c, void *arg)
 			if (constty != NULL)
 				msgbuf_addchar(&consmsgbuf, c);
 			if (always_console_output && c != '\0')
-				cnputc(c);
+				putcons(c, ap);
 		}
 	}
 	if ((flags & TOLOG))
diff --git a/sys/kern/tty_cons.c b/sys/kern/tty_cons.c
index f65f7ad10f57..c0f86920e352 100644
--- a/sys/kern/tty_cons.c
+++ b/sys/kern/tty_cons.c
@@ -41,6 +41,8 @@ __FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
 #include <sys/systm.h>
+#include <sys/lock.h>
+#include <sys/mutex.h>
 #include <sys/conf.h>
 #include <sys/cons.h>
 #include <sys/fcntl.h>
@@ -117,6 +119,8 @@ static u_char console_pausing;		/* pause after each line during probe */
 static char *console_pausestr=
 "<pause; press any key to proceed to next line or '.' to end pause mode>";
 struct tty *constty;			/* pointer to console "window" tty */
+static struct mtx cnputs_mtx;		/* Mutex for cnputs(). */
+static int use_cnputs_mtx = 0;		/* != 0 if cnputs_mtx locking reqd. */
 
 static void constty_timeout(void *arg);
 
@@ -639,6 +643,24 @@ cnputc(int c)
 	}
 }
 
+void
+cnputs(char *p)
+{
+	int c;
+	int unlock_reqd = 0;
+
+	if (use_cnputs_mtx) {
+		mtx_lock_spin(&cnputs_mtx);
+		unlock_reqd = 1;
+	}
+
+	while ((c = *p++) != '\0')
+		cnputc(c);
+
+	if (unlock_reqd)
+		mtx_unlock_spin(&cnputs_mtx);
+}
+
 static int consmsgbuf_size = 8192;
 SYSCTL_INT(_kern, OID_AUTO, consmsgbuf_size, CTLFLAG_RW, &consmsgbuf_size, 0,
     "");
@@ -708,6 +730,9 @@ cn_drvinit(void *unused)
 {
 
 	make_dev(&cn_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "console");
+
+	mtx_init(&cnputs_mtx, "cnputs_mtx", NULL, MTX_SPIN | MTX_NOWITNESS);
+	use_cnputs_mtx = 1;
 }
 
 SYSINIT(cndev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, cn_drvinit, NULL)
diff --git a/sys/powerpc/powerpc/genassym.c b/sys/powerpc/powerpc/genassym.c
index 63719cc12d20..80deb8eef22a 100644
--- a/sys/powerpc/powerpc/genassym.c
+++ b/sys/powerpc/powerpc/genassym.c
@@ -55,6 +55,7 @@
 #include <machine/pmap.h>
 #include <machine/sigframe.h>
 
+ASSYM(PC_CONS_BUFR, offsetof(struct pcpu, pc_cons_bufr));
 ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread));
 ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
 ASSYM(PC_CURPMAP, offsetof(struct pcpu, pc_curpmap));
diff --git a/sys/sparc64/sparc64/genassym.c b/sys/sparc64/sparc64/genassym.c
index fc79c6b1f954..5639e3ec92f9 100644
--- a/sys/sparc64/sparc64/genassym.c
+++ b/sys/sparc64/sparc64/genassym.c
@@ -198,6 +198,7 @@ ASSYM(HASH_ENTRY_SHIFT, HASH_ENTRY_SHIFT);
 
 ASSYM(V_INTR, offsetof(struct vmmeter, v_intr));
 
+ASSYM(PC_CONS_BUFR, offsetof(struct pcpu, pc_cons_bufr));
 ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread));
 ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
 ASSYM(PC_CPUID, offsetof(struct pcpu, pc_cpuid));
diff --git a/sys/sys/cons.h b/sys/sys/cons.h
index 116aa15416cb..a3229c3f5c81 100644
--- a/sys/sys/cons.h
+++ b/sys/sys/cons.h
@@ -106,6 +106,7 @@ void	cnselect(struct consdev *);
 int	cncheckc(void);
 int	cngetc(void);
 void	cnputc(int);
+void	cnputs(char *);
 int	cnunavailable(void);
 
 #endif /* _KERNEL */
diff --git a/sys/sys/pcpu.h b/sys/sys/pcpu.h
index 00a3cac97c3e..fa54bc4e2a53 100644
--- a/sys/sys/pcpu.h
+++ b/sys/sys/pcpu.h
@@ -48,6 +48,9 @@
 struct pcb;
 struct thread;
 
+/* Size of the per-cpu console buffer for printf(). */
+#define	PCPU_CONS_BUFR	128
+
 /*
  * This structure maps out the global data that needs to be kept on a
  * per-cpu basis.  The members are accessed via the PCPU_GET/SET/PTR
@@ -71,6 +74,8 @@ struct pcpu {
 	int		pc_ktr_idx;		/* Index into trace table */
 	char		*pc_ktr_buf;
 #endif
+	char		pc_cons_bufr[PCPU_CONS_BUFR];
+						/* Console buffer */
 	PCPU_MD_FIELDS;
 	struct vmmeter	pc_cnt;			/* VM stats counters */
 	struct device	*pc_device;