- Add support to atomically set/clear individual bits of a MSR register

  via cpuctl(4) driver.  Two new CPUCTL_MSRSBIT and CPUCTL_MSRCBIT ioctl(2)
  calls treat the data field of the argument struct passed as a mask
  and set/clear bits of the MSR register according to the mask value.
- Allow user to perform atomic bitwise AND and OR operaions on MSR registers
  via cpucontrol(8) utility.  Two new operations ("&=" and "|=") have been
  added.  The first one applies bitwise AND operaion between the current
  contents of the MSR register and the mask, and the second performs bitwise
  OR.  The argument can be optionally prefixed with "~" inversion operator.
  This allows one to mimic the "clear bit" behavior by using the command
  like this:
      cpucontrol -m 0x10&=~0x02		# clear the second bit of TSC MSR

  Inversion operator support in all modes (assignment, OR, AND).

Approved by:	re (kib)
MFC after:	1 month

5 files changed, 162 insertions, 28 deletions

diff --git a/share/man/man4/cpuctl.4 b/share/man/man4/cpuctl.4
index 9dd149a..09b2d4e 100644
--- a/share/man/man4/cpuctl.4
+++ b/share/man/man4/cpuctl.4
@@ -24,7 +24,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd May 31, 2008
+.Dd June 30, 2009
 .Dt CPUCTL 4
 .Os
 .Sh NAME
@@ -81,6 +81,11 @@ typedef struct {
 	uint64_t	data;
 } cpuctl_msr_args_t;
 .Ed
+.It Dv CPUCTL_MSRSBIT Fa cpuctl_msr_args_t *args
+.It Dv CPUCTL_MSRCBIT Fa cpuctl_msr_args_t *args
+Set/clear MSR bits according to the mask given in the
+.Va data
+field.
 .It Dv CPUCTL_CPUID Fa cpuctl_cpuid_args_t *args
 Retrieve CPUID information.
 Arguments are supplied in
diff --git a/sys/dev/cpuctl/cpuctl.c b/sys/dev/cpuctl/cpuctl.c
index f325797..c3a2434 100644
--- a/sys/dev/cpuctl/cpuctl.c
+++ b/sys/dev/cpuctl/cpuctl.c
@@ -158,6 +158,8 @@ cpuctl_ioctl(struct cdev *dev, u_long cmd, caddr_t data,
 	case CPUCTL_RDMSR:
 		ret = cpuctl_do_msr(cpu, (cpuctl_msr_args_t *)data, cmd, td);
 		break;
+	case CPUCTL_MSRSBIT:
+	case CPUCTL_MSRCBIT:
 	case CPUCTL_WRMSR:
 		ret = priv_check(td, PRIV_CPUCTL_WRMSR);
 		if (ret != 0)
@@ -211,6 +213,7 @@ cpuctl_do_cpuid(int cpu, cpuctl_cpuid_args_t *data, struct thread *td)
 static int
 cpuctl_do_msr(int cpu, cpuctl_msr_args_t *data, u_long cmd, struct thread *td)
 {
+	uint64_t reg;
 	int is_bound = 0;
 	int oldcpu;
 	int ret;
@@ -230,9 +233,22 @@ cpuctl_do_msr(int cpu, cpuctl_msr_args_t *data, u_long cmd, struct thread *td)
 	if (cmd == CPUCTL_RDMSR) {
 		data->data = 0;
 		ret = rdmsr_safe(data->msr, &data->data);
-	} else {
+	} else if (cmd == CPUCTL_WRMSR) {
 		ret = wrmsr_safe(data->msr, data->data);
-	}
+	} else if (cmd == CPUCTL_MSRSBIT) {
+		critical_enter();
+		ret = rdmsr_safe(data->msr, &reg);
+		if (ret == 0)
+			ret = wrmsr_safe(data->msr, reg | data->data);
+		critical_exit();
+	} else if (cmd == CPUCTL_MSRCBIT) {
+		critical_enter();
+		ret = rdmsr_safe(data->msr, &reg);
+		if (ret == 0)
+			ret = wrmsr_safe(data->msr, reg & ~data->data);
+		critical_exit();
+	} else
+		panic("[cpuctl,%d]: unknown operation requested: %lu", __LINE__, cmd);
 	restore_cpu(oldcpu, is_bound, td);
 	return (ret);
 }
diff --git a/sys/sys/cpuctl.h b/sys/sys/cpuctl.h
index 57b5acf..4bddb34 100644
--- a/sys/sys/cpuctl.h
+++ b/sys/sys/cpuctl.h
@@ -48,5 +48,7 @@ typedef struct {
 #define	CPUCTL_WRMSR	_IOWR('c', 2, cpuctl_msr_args_t)
 #define	CPUCTL_CPUID	_IOWR('c', 3, cpuctl_cpuid_args_t)
 #define	CPUCTL_UPDATE	_IOWR('c', 4, cpuctl_update_args_t)
+#define	CPUCTL_MSRSBIT	_IOWR('c', 5, cpuctl_msr_args_t)
+#define	CPUCTL_MSRCBIT	_IOWR('c', 6, cpuctl_msr_args_t)
 
 #endif /* _CPUCTL_H_ */
diff --git a/usr.sbin/cpucontrol/cpucontrol.8 b/usr.sbin/cpucontrol/cpucontrol.8
index 01bc9d3..165e67b 100644
--- a/usr.sbin/cpucontrol/cpucontrol.8
+++ b/usr.sbin/cpucontrol/cpucontrol.8
@@ -24,7 +24,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd August 4, 2008
+.Dd June 30, 2009
 .Dt CPUCONTROL 8
 .Os
 .Sh NAME
@@ -35,7 +35,25 @@ device.
 .Sh SYNOPSIS
 .Nm
 .Op Fl vh
-.Fl m Ar msr Ns Op = Ns Ar value
+.Fl m Ar msr
+.Bk
+.Ar device
+.Ek
+.Nm
+.Op Fl vh
+.Fl m Ar msr Ns = Ns Ar value
+.Bk
+.Ar device
+.Ek
+.Nm
+.Op Fl vh
+.Fl m Ar msr Ns &= Ns Ar mask
+.Bk
+.Ar device
+.Ek
+.Nm
+.Op Fl vh
+.Fl m Ar msr Ns |= Ns Ar mask
 .Bk
 .Ar device
 .Ek
@@ -67,8 +85,32 @@ The following options are available:
 Where to look for microcode images.
 The option can be specified multiple times.
 .It Fl m Ar msr Ns Op = Ns Ar value
-Read/write the specified MSR.
-Both the MSR and the value should be given as a hex number.
+Show value of the specified MSR.
+MSR register number should be given as a hexadecimal number.
+.It Fl m Ar msr Ns = Ns Ar value
+Store the
+.Ar value
+in the specified MSR register.
+The
+.Ar value
+argument can be prefixed with ~ operator.
+In this case the inverted value of argument will be stored in the register.
+.It Fl m Ar msr Ns &= Ns Ar mask
+Store the result of bitwise AND operation between
+.Ar mask
+and the current MSR value in the MSR register.
+The
+.Ar mask
+argument can be prefixed with ~ operator.
+In this case the inverted value of mask will be used.
+.It Fl m Ar msr Ns |= Ns Ar mask
+Store the result of bitwise OR operation between
+.Ar mask
+and the current MSR value in the MSR register.
+The
+.Ar mask
+argument can be prefixed with ~ operator.
+In this case the inverted value of mask will be used.
 .It Fl i Ar level
 Retrieve CPUID info.
 Level should be given as a hex number.
@@ -94,7 +136,15 @@ will read the contents of TSC MSR from CPU 0.
 .Pp
 To set the CPU 0 TSC MSR register value to 0x1 issue
 .Pp
-.Dq Li "cpucontrol -m 0x10=0x1 /dev/cpuctl0"
+.Dq Li "cpucontrol -m 0x10=0x1 /dev/cpuctl0" .
+.Pp
+The following command will clear the second bit of TSC register:
+.Pp
+.Dq Li "cpucontrol -m 0x10&=~0x02 /dev/cpuctl0" .
+.Pp
+The following command will set the forth and second bit of TSC register:
+.Pp
+.Dq Li "cpucontrol -m 0x10|=0x0a /dev/cpuctl0" .
 .Pp
 The command
 .Pp
diff --git a/usr.sbin/cpucontrol/cpucontrol.c b/usr.sbin/cpucontrol/cpucontrol.c
index 5469ee2..8b3ca1d 100644
--- a/usr.sbin/cpucontrol/cpucontrol.c
+++ b/usr.sbin/cpucontrol/cpucontrol.c
@@ -60,6 +60,12 @@ int	verbosity_level = 0;
 #define	FLAG_M	0x02
 #define	FLAG_U	0x04
 
+#define	OP_INVAL	0x00
+#define	OP_READ		0x01
+#define	OP_WRITE	0x02
+#define	OP_OR		0x04
+#define	OP_AND		0x08
+
 #define	HIGH(val)	(uint32_t)(((val) >> 32) & 0xffffffff)
 #define	LOW(val)	(uint32_t)((val) & 0xffffffff)
 
@@ -166,28 +172,64 @@ do_msr(const char *cmdarg, const char *dev)
 {
 	unsigned int msr;
 	cpuctl_msr_args_t args;
+	size_t len;
+	uint64_t data = 0;
+	unsigned long command;
+	int do_invert = 0, op;
 	int fd, error;
-	int wr = 0;
-	char *p;
 	char *endptr;
+	char *p;
 
 	assert(cmdarg != NULL);
 	assert(dev != NULL);
+	len = strlen(cmdarg);
+	if (len == 0) {
+		WARNX(0, "MSR register expected");
+		usage();
+		/* NOTREACHED */
+	}
 
-	p = strchr(cmdarg, '=');
-	if (p != NULL) {
-		wr = 1;
-		*p++ = '\0';
-		args.data = strtoull(p, &endptr, 16);
-		if (*p == '\0' || *endptr != '\0') {
-			WARNX(0, "incorrect MSR value: %s", p);
-			usage();
-			/* NOTREACHED */
+	/*
+	 * Parse command string.
+	 */
+	msr = strtoul(cmdarg, &endptr, 16);
+	switch (*endptr) {
+	case '\0':
+		op = OP_READ;
+		break;
+	case '=':
+		op = OP_WRITE;
+		break;
+	case '&':
+		op = OP_AND;
+		endptr++;
+		break;
+	case '|':
+		op = OP_OR;
+		endptr++;
+		break;
+	default:
+		op = OP_INVAL;
+	}
+	if (op != OP_READ) {	/* Complex operation. */
+		if (*endptr != '=')
+			op = OP_INVAL;
+		else {
+			p = ++endptr;
+			if (*p == '~') {
+				do_invert = 1;
+				p++;
+			}
+			data = strtoull(p, &endptr, 16);
+			if (*p == '\0' || *endptr != '\0') {
+				WARNX(0, "argument required: %s", cmdarg);
+				usage();
+				/* NOTREACHED */
+			}
 		}
 	}
-	msr = strtoul(cmdarg, &endptr, 16);
-	if (*cmdarg == '\0' || *endptr != '\0') {
-		WARNX(0, "incorrect MSR register: %s", cmdarg);
+	if (op == OP_INVAL) {
+		WARNX(0, "invalid operator: %s", cmdarg);
 		usage();
 		/* NOTREACHED */
 	}
@@ -196,20 +238,39 @@ do_msr(const char *cmdarg, const char *dev)
 	 * Fill ioctl argument structure.
 	 */
 	args.msr = msr;
-	fd = open(dev, wr == 0 ? O_RDONLY : O_WRONLY);
+	if ((do_invert != 0) ^ (op == OP_AND))
+		args.data = ~data;
+	else
+		args.data = data;
+	switch (op) {
+	case OP_READ:
+		command = CPUCTL_RDMSR;
+		break;
+	case OP_WRITE:
+		command = CPUCTL_WRMSR;
+		break;
+	case OP_OR:
+		command = CPUCTL_MSRSBIT;
+		break;
+	case OP_AND:
+		command = CPUCTL_MSRCBIT;
+		break;
+	default:
+		abort();
+	}
+	fd = open(dev, op == OP_READ ? O_RDONLY : O_WRONLY);
 	if (fd < 0) {
 		WARN(0, "error opening %s for %s", dev,
-		    wr == 0 ? "reading" : "writing");
+		    op == OP_READ ? "reading" : "writing");
 		return (1);
 	}
-	error = ioctl(fd, wr == 0 ? CPUCTL_RDMSR : CPUCTL_WRMSR, &args);
+	error = ioctl(fd, command, &args);
 	if (error < 0) {
-		WARN(0, "ioctl(%s, %s)", dev,
-		    wr == 0 ? "CPUCTL_RDMSR" : "CPUCTL_WRMSR");
+		WARN(0, "ioctl(%s, %lu)", dev, command);
 		close(fd);
 		return (1);
 	}
-	if (wr == 0)
+	if (op == OP_READ)
 		fprintf(stdout, "MSR 0x%x: 0x%.8x 0x%.8x\n", msr,
 		    HIGH(args.data), LOW(args.data));
 	close(fd);