s390/cpumf: Add support for complete counter set extraction

Add support to the CPU Measurement counter facility device driver
to extract complete counter sets per CPU and per counter set from user
space. This includes a new device named /dev/hwctr and support
for the device driver functions open, close and ioctl. Other
functions are not supported.

The ioctl command supports 3 subcommands:
S390_HWCTR_START: enables counter sets on a list of CPUs.
S390_HWCTR_STOP: disables counter sets on a list of CPUs.
S390_HWCTR_READ: reads counter sets on a list of CPUs.

The ioctl(..., S390_HWCTR_READ, ...) is the only subcommand which
returns data.  It requires member data_bytes to be positive and
indicates the maximum amount of data available to store counter set
data. The other ioctl() subcommands do not use this member and it
should be set to zero.
The S390_HWCTR_READ subcommand returns the following data:

The cpuset data is flattened using the following scheme, stored in member
data:

 0x0       0x8   0xc       0x10  0x10      0x18  0x20  0x28         0xU-1
 +---------+-----+---------+-----+---------+-----+-----+------+------+
 | no_cpus | cpu | no_sets | set | no_cnts | cv1 | cv2 | .... | cv_n |
 +---------+-----+---------+-----+---------+-----+-----+------+------+

                           0xU   0xU+4     0xU+8 0xU+10             0xV-1
                           +-----+---------+-----+-----+------+------+
                           | set | no_cnts | cv1 | cv2 | .... | cv_n |
                           +-----+---------+-----+-----+------+------+

           0xV   0xV+4     0xV+8 0xV+c
           +-----+---------+-----+---------+-----+-----+------+------+
           | cpu | no_sets | set | no_cnts | cv1 | cv2 | .... | cv_n |
           +-----+---------+-----+---------+-----+-----+------+------+

U and V denote arbitrary hexadezimal addresses.
The first integer represents the number of CPUs data was extracted
from. This is followed by CPU number and number of counter sets extracted.
Both are two integer values. This is followed by the set identifer
and number of counters extracted. Both are two integer values. This is
followed by the counter values, each element is eight bytes in size.

The S390_HWCTR_READ ioctl subcommand is also limited to one call per
minute. This ensures that an application does not read out the
counter sets too often and reduces the overall CPU performance.
The complete counter set extraction is an expensive operation.

Reviewed-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

2 files changed, 576 insertions, 23 deletions

diff --git a/arch/s390/include/uapi/asm/perf_cpum_cf_diag.h b/arch/s390/include/uapi/asm/perf_cpum_cf_diag.h
new file mode 100644
index 000000000000..3d8284b95f87
--- /dev/null
+++ b/arch/s390/include/uapi/asm/perf_cpum_cf_diag.h
@@ -0,0 +1,51 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/*
+ * Copyright IBM Corp. 2021
+ * Interface implementation for communication with the CPU Measurement
+ * counter facility device driver.
+ *
+ * Author(s): Thomas Richter <tmricht@linux.ibm.com>
+ *
+ * Define for ioctl() commands to communicate with the CPU Measurement
+ * counter facility device driver.
+ */
+
+#ifndef _PERF_CPUM_CF_DIAG_H
+#define _PERF_CPUM_CF_DIAG_H
+
+#include <linux/ioctl.h>
+#include <linux/types.h>
+
+#define S390_HWCTR_DEVICE		"hwctr"
+#define S390_HWCTR_START_VERSION	1
+
+struct s390_ctrset_start {		/* Set CPUs to operate on */
+	__u64 version;			/* Version of interface */
+	__u64 data_bytes;		/* # of bytes required */
+	__u64 cpumask_len;		/* Length of CPU mask in bytes */
+	__u64 *cpumask;			/* Pointer to CPU mask */
+	__u64 counter_sets;		/* Bit mask of counter sets to get */
+};
+
+struct s390_ctrset_setdata {		/* Counter set data */
+	__u32 set;			/* Counter set number */
+	__u32 no_cnts;			/* # of counters stored in cv[] */
+	__u64 cv[0];			/* Counter values (variable length) */
+};
+
+struct s390_ctrset_cpudata {		/* Counter set data per CPU */
+	__u32 cpu_nr;			/* CPU number */
+	__u32 no_sets;			/* # of counters sets in data[] */
+	struct s390_ctrset_setdata data[0];
+};
+
+struct s390_ctrset_read {		/* Structure to get all ctr sets */
+	__u64 no_cpus;			/* Total # of CPUs data taken from */
+	struct s390_ctrset_cpudata data[0];
+};
+
+#define S390_HWCTR_MAGIC	'C'	/* Random magic # for ioctls */
+#define	S390_HWCTR_START	_IOWR(S390_HWCTR_MAGIC, 1, struct s390_ctrset_start)
+#define	S390_HWCTR_STOP		_IO(S390_HWCTR_MAGIC, 2)
+#define	S390_HWCTR_READ		_IOWR(S390_HWCTR_MAGIC, 3, struct s390_ctrset_read)
+#endif
diff --git a/arch/s390/kernel/perf_cpum_cf_diag.c b/arch/s390/kernel/perf_cpum_cf_diag.c
index b5c86fb70d63..db4877bbb9aa 100644
--- a/arch/s390/kernel/perf_cpum_cf_diag.c
+++ b/arch/s390/kernel/perf_cpum_cf_diag.c
@@ -2,7 +2,7 @@
 /*
  * Performance event support for s390x - CPU-measurement Counter Sets
  *
- *  Copyright IBM Corp. 2019
+ *  Copyright IBM Corp. 2019, 2021
  *  Author(s): Hendrik Brueckner <brueckner@linux.ibm.com>
  *	       Thomas Richer <tmricht@linux.ibm.com>
  */
@@ -17,6 +17,8 @@
 #include <linux/export.h>
 #include <linux/slab.h>
 #include <linux/processor.h>
+#include <linux/miscdevice.h>
+#include <linux/mutex.h>
 
 #include <asm/ctl_reg.h>
 #include <asm/irq.h>
@@ -24,15 +26,20 @@
 #include <asm/timex.h>
 #include <asm/debug.h>
 
-#define	CF_DIAG_CTRSET_DEF		0xfeef	/* Counter set header mark */
+#include <asm/perf_cpum_cf_diag.h>
 
+#define	CF_DIAG_CTRSET_DEF		0xfeef	/* Counter set header mark */
+#define CF_DIAG_MIN_INTERVAL		60	/* Minimum counter set read */
+						/* interval in seconds */
+static unsigned long cf_diag_interval = CF_DIAG_MIN_INTERVAL;
 static unsigned int cf_diag_cpu_speed;
 static debug_info_t *cf_diag_dbg;
 
-struct cf_diag_csd {		/* Counter set data per CPU */
+struct cf_diag_csd {			/* Counter set data per CPU */
 	size_t used;			/* Bytes used in data/start */
 	unsigned char start[PAGE_SIZE];	/* Counter set at event start */
 	unsigned char data[PAGE_SIZE];	/* Counter set at event delete */
+	unsigned int sets;		/* # Counter set saved in data */
 };
 static DEFINE_PER_CPU(struct cf_diag_csd, cf_diag_csd);
 
@@ -178,18 +185,35 @@ static void cf_diag_disable(struct pmu *pmu)
 
 /* Number of perf events counting hardware events */
 static atomic_t cf_diag_events = ATOMIC_INIT(0);
+/* Used to avoid races in calling reserve/release_cpumf_hardware */
+static DEFINE_MUTEX(cf_diag_reserve_mutex);
 
 /* Release the PMU if event is the last perf event */
 static void cf_diag_perf_event_destroy(struct perf_event *event)
 {
 	debug_sprintf_event(cf_diag_dbg, 5,
 			    "%s event %p cpu %d cf_diag_events %d\n",
-			    __func__, event, event->cpu,
+			    __func__, event, smp_processor_id(),
 			    atomic_read(&cf_diag_events));
 	if (atomic_dec_return(&cf_diag_events) == 0)
 		__kernel_cpumcf_end();
 }
 
+static int get_authctrsets(void)
+{
+	struct cpu_cf_events *cpuhw;
+	unsigned long auth = 0;
+	enum cpumf_ctr_set i;
+
+	cpuhw = &get_cpu_var(cpu_cf_events);
+	for (i = CPUMF_CTR_SET_BASIC; i < CPUMF_CTR_SET_MAX; ++i) {
+		if (cpuhw->info.auth_ctl & cpumf_ctr_ctl[i])
+			auth |= cpumf_ctr_ctl[i];
+	}
+	put_cpu_var(cpu_cf_events);
+	return auth;
+}
+
 /* Setup the event. Test for authorized counter sets and only include counter
  * sets which are authorized at the time of the setup. Including unauthorized
  * counter sets result in specification exception (and panic).
@@ -197,15 +221,12 @@ static void cf_diag_perf_event_destroy(struct perf_event *event)
 static int __hw_perf_event_init(struct perf_event *event)
 {
 	struct perf_event_attr *attr = &event->attr;
-	struct cpu_cf_events *cpuhw;
-	enum cpumf_ctr_set i;
 	int err = 0;
 
 	debug_sprintf_event(cf_diag_dbg, 5, "%s event %p cpu %d\n", __func__,
 			    event, event->cpu);
 
 	event->hw.config = attr->config;
-	event->hw.config_base = 0;
 
 	/* Add all authorized counter sets to config_base. The
 	 * the hardware init function is either called per-cpu or just once
@@ -215,11 +236,7 @@ static int __hw_perf_event_init(struct perf_event *event)
 	 * Checking the authorization on any CPU is fine as the hardware
 	 * applies the same authorization settings to all CPUs.
 	 */
-	cpuhw = &get_cpu_var(cpu_cf_events);
-	for (i = CPUMF_CTR_SET_BASIC; i < CPUMF_CTR_SET_MAX; ++i)
-		if (cpuhw->info.auth_ctl & cpumf_ctr_ctl[i])
-			event->hw.config_base |= cpumf_ctr_ctl[i];
-	put_cpu_var(cpu_cf_events);
+	event->hw.config_base = get_authctrsets();
 
 	/* No authorized counter sets, nothing to count/sample */
 	if (!event->hw.config_base) {
@@ -237,6 +254,25 @@ out:
 	return err;
 }
 
+/* Return 0 if the CPU-measurement counter facility is currently free
+ * and an error otherwise.
+ */
+static int cf_diag_perf_event_inuse(void)
+{
+	int err = 0;
+
+	if (!atomic_inc_not_zero(&cf_diag_events)) {
+		mutex_lock(&cf_diag_reserve_mutex);
+		if (atomic_read(&cf_diag_events) == 0 &&
+		    __kernel_cpumcf_begin())
+			err = -EBUSY;
+		else
+			err = atomic_inc_return(&cf_diag_events);
+		mutex_unlock(&cf_diag_reserve_mutex);
+	}
+	return err;
+}
+
 static int cf_diag_event_init(struct perf_event *event)
 {
 	struct perf_event_attr *attr = &event->attr;
@@ -264,13 +300,9 @@ static int cf_diag_event_init(struct perf_event *event)
 	}
 
 	/* Initialize for using the CPU-measurement counter facility */
-	if (atomic_inc_return(&cf_diag_events) == 1) {
-		if (__kernel_cpumcf_begin()) {
-			atomic_dec(&cf_diag_events);
-			err = -EBUSY;
-			goto out;
-		}
-	}
+	err = cf_diag_perf_event_inuse();
+	if (err < 0)
+		goto out;
 	event->destroy = cf_diag_perf_event_destroy;
 
 	err = __hw_perf_event_init(event);
@@ -599,6 +631,8 @@ static void cf_diag_del(struct perf_event *event, int flags)
 	cpuhw->flags &= ~PMU_F_IN_USE;
 }
 
+/* Default counter set events and format attribute groups */
+
 CPUMF_EVENT_ATTR(CF_DIAG, CF_DIAG, PERF_EVENT_CPUM_CF_DIAG);
 
 static struct attribute *cf_diag_events_attr[] = {
@@ -663,6 +697,452 @@ static void cf_diag_get_cpu_speed(void)
 	}
 }
 
+/* Code to create device and file I/O operations */
+static atomic_t ctrset_opencnt = ATOMIC_INIT(0);	/* Excl. access */
+
+static int cf_diag_open(struct inode *inode, struct file *file)
+{
+	int err = 0;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+	if (atomic_xchg(&ctrset_opencnt, 1))
+		return -EBUSY;
+
+	/* Avoid concurrent access with perf_event_open() system call */
+	mutex_lock(&cf_diag_reserve_mutex);
+	if (atomic_read(&cf_diag_events) || __kernel_cpumcf_begin())
+		err = -EBUSY;
+	mutex_unlock(&cf_diag_reserve_mutex);
+	if (err) {
+		atomic_set(&ctrset_opencnt, 0);
+		return err;
+	}
+	file->private_data = NULL;
+	debug_sprintf_event(cf_diag_dbg, 2, "%s\n", __func__);
+	/* nonseekable_open() never fails */
+	return nonseekable_open(inode, file);
+}
+
+/* Variables for ioctl() interface support */
+static DEFINE_MUTEX(cf_diag_ctrset_mutex);
+static struct cf_diag_ctrset {
+	unsigned long ctrset;		/* Bit mask of counter set to read */
+	cpumask_t mask;			/* CPU mask to read from */
+	time64_t lastread;		/* Epoch counter set last read */
+} cf_diag_ctrset;
+
+static void cf_diag_ctrset_clear(void)
+{
+	cpumask_clear(&cf_diag_ctrset.mask);
+	cf_diag_ctrset.ctrset = 0;
+}
+
+static void cf_diag_release_cpu(void *p)
+{
+	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
+
+	debug_sprintf_event(cf_diag_dbg, 3, "%s cpu %d\n", __func__,
+			    smp_processor_id());
+	lcctl(0);		/* Reset counter sets */
+	cpuhw->state = 0;	/* Save state in CPU hardware state */
+}
+
+/* Release function is also called when application gets terminated without
+ * doing a proper ioctl(..., S390_HWCTR_STOP, ...) command.
+ * Since only one application is allowed to open the device, simple stop all
+ * CPU counter sets.
+ */
+static int cf_diag_release(struct inode *inode, struct file *file)
+{
+	on_each_cpu(cf_diag_release_cpu, NULL, 1);
+	cf_diag_ctrset_clear();
+	atomic_set(&ctrset_opencnt, 0);
+	__kernel_cpumcf_end();
+	debug_sprintf_event(cf_diag_dbg, 2, "%s\n", __func__);
+	return 0;
+}
+
+struct cf_diag_call_on_cpu_parm {	/* Parm struct for smp_call_on_cpu */
+	unsigned int sets;		/* Counter set bit mask */
+	atomic_t cpus_ack;		/* # CPUs successfully executed func */
+};
+
+static int cf_diag_all_copy(unsigned long arg, cpumask_t *mask)
+{
+	struct s390_ctrset_read __user *ctrset_read;
+	unsigned int cpu, cpus, rc;
+	void __user *uptr;
+
+	ctrset_read = (struct s390_ctrset_read __user *)arg;
+	uptr = ctrset_read->data;
+	for_each_cpu(cpu, mask) {
+		struct cf_diag_csd *csd = per_cpu_ptr(&cf_diag_csd, cpu);
+		struct s390_ctrset_cpudata __user *ctrset_cpudata;
+
+		ctrset_cpudata = uptr;
+		debug_sprintf_event(cf_diag_dbg, 5, "%s cpu %d used %zd\n",
+				    __func__, cpu, csd->used);
+		rc  = put_user(cpu, &ctrset_cpudata->cpu_nr);
+		rc |= put_user(csd->sets, &ctrset_cpudata->no_sets);
+		rc |= copy_to_user(ctrset_cpudata->data, csd->data, csd->used);
+		if (rc)
+			return -EFAULT;
+		uptr += sizeof(struct s390_ctrset_cpudata) + csd->used;
+		cond_resched();
+	}
+	cpus = cpumask_weight(mask);
+	if (put_user(cpus, &ctrset_read->no_cpus))
+		return -EFAULT;
+	debug_sprintf_event(cf_diag_dbg, 5, "%s copied %ld\n",
+			    __func__, uptr - (void __user *)ctrset_read->data);
+	return 0;
+}
+
+static size_t cf_diag_cpuset_read(struct s390_ctrset_setdata *p, int ctrset,
+				  int ctrset_size, size_t room)
+{
+	size_t need = 0;
+	int rc = -1;
+
+	need = sizeof(*p) + sizeof(u64) * ctrset_size;
+	debug_sprintf_event(cf_diag_dbg, 5,
+			    "%s room %zd need %zd set %#x set_size %d\n",
+			    __func__, room, need, ctrset, ctrset_size);
+	if (need <= room) {
+		p->set = cpumf_ctr_ctl[ctrset];
+		p->no_cnts = ctrset_size;
+		rc = ctr_stcctm(ctrset, ctrset_size, (u64 *)p->cv);
+		if (rc == 3)		/* Nothing stored */
+			need = 0;
+	}
+	debug_sprintf_event(cf_diag_dbg, 5, "%s need %zd rc %d\n", __func__,
+			    need, rc);
+	return need;
+}
+
+/* Read all counter sets. Since the perf_event_open() system call with
+ * event cpum_cf_diag/.../ is blocked when this interface is active, reuse
+ * the perf_event_open() data buffer to store the counter sets.
+ */
+static void cf_diag_cpu_read(void *parm)
+{
+	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
+	struct cf_diag_csd *csd = this_cpu_ptr(&cf_diag_csd);
+	struct cf_diag_call_on_cpu_parm *p = parm;
+	int set, set_size;
+	size_t space;
+
+	debug_sprintf_event(cf_diag_dbg, 5,
+			    "%s new %#x flags %#x state %#llx\n",
+			    __func__, p->sets, cpuhw->flags,
+			    cpuhw->state);
+	/* No data saved yet */
+	csd->used = 0;
+	csd->sets = 0;
+	memset(csd->data, 0, sizeof(csd->data));
+
+	/* Scan the counter sets */
+	for (set = CPUMF_CTR_SET_BASIC; set < CPUMF_CTR_SET_MAX; ++set) {
+		struct s390_ctrset_setdata *sp = (void *)csd->data + csd->used;
+
+		if (!(p->sets & cpumf_ctr_ctl[set]))
+			continue;	/* Counter set not in list */
+		set_size = cf_diag_ctrset_size(set, &cpuhw->info);
+		space = sizeof(csd->data) - csd->used;
+		space = cf_diag_cpuset_read(sp, set, set_size, space);
+		if (space) {
+			csd->used += space;
+			csd->sets += 1;
+		}
+		debug_sprintf_event(cf_diag_dbg, 5, "%s sp %px space %zd\n",
+				    __func__, sp, space);
+	}
+	debug_sprintf_event(cf_diag_dbg, 5, "%s sets %d used %zd\n", __func__,
+			    csd->sets, csd->used);
+}
+
+static int cf_diag_all_read(unsigned long arg)
+{
+	struct cf_diag_call_on_cpu_parm p;
+	cpumask_var_t mask;
+	time64_t now;
+	int rc = 0;
+
+	debug_sprintf_event(cf_diag_dbg, 5, "%s\n", __func__);
+	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
+		return -ENOMEM;
+	now = ktime_get_seconds();
+	if (cf_diag_ctrset.lastread + cf_diag_interval > now) {
+		debug_sprintf_event(cf_diag_dbg, 5, "%s now %lld "
+				    " lastread %lld\n", __func__, now,
+				    cf_diag_ctrset.lastread);
+		rc = -EAGAIN;
+		goto out;
+	} else {
+		cf_diag_ctrset.lastread = now;
+	}
+	p.sets = cf_diag_ctrset.ctrset;
+	cpumask_and(mask, &cf_diag_ctrset.mask, cpu_online_mask);
+	on_each_cpu_mask(mask, cf_diag_cpu_read, &p, 1);
+	rc = cf_diag_all_copy(arg, mask);
+out:
+	free_cpumask_var(mask);
+	debug_sprintf_event(cf_diag_dbg, 5, "%s rc %d\n", __func__, rc);
+	return rc;
+}
+
+/* Stop all counter sets via ioctl interface */
+static void cf_diag_ioctl_off(void *parm)
+{
+	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
+	struct cf_diag_call_on_cpu_parm *p = parm;
+	int rc;
+
+	debug_sprintf_event(cf_diag_dbg, 5,
+			    "%s new %#x flags %#x state %#llx\n",
+			    __func__, p->sets, cpuhw->flags,
+			    cpuhw->state);
+
+	ctr_set_multiple_disable(&cpuhw->state, p->sets);
+	ctr_set_multiple_stop(&cpuhw->state, p->sets);
+	rc = lcctl(cpuhw->state);		/* Stop counter sets */
+	if (!cpuhw->state)
+		cpuhw->flags &= ~PMU_F_IN_USE;
+	debug_sprintf_event(cf_diag_dbg, 5,
+			    "%s rc %d flags %#x state %#llx\n", __func__,
+			     rc, cpuhw->flags, cpuhw->state);
+}
+
+/* Start counter sets on particular CPU */
+static void cf_diag_ioctl_on(void *parm)
+{
+	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
+	struct cf_diag_call_on_cpu_parm *p = parm;
+	int rc;
+
+	debug_sprintf_event(cf_diag_dbg, 5,
+			    "%s new %#x flags %#x state %#llx\n",
+			    __func__, p->sets, cpuhw->flags,
+			    cpuhw->state);
+
+	if (!(cpuhw->flags & PMU_F_IN_USE))
+		cpuhw->state = 0;
+	cpuhw->flags |= PMU_F_IN_USE;
+	rc = lcctl(cpuhw->state);		/* Reset unused counter sets */
+	ctr_set_multiple_enable(&cpuhw->state, p->sets);
+	ctr_set_multiple_start(&cpuhw->state, p->sets);
+	rc |= lcctl(cpuhw->state);		/* Start counter sets */
+	if (!rc)
+		atomic_inc(&p->cpus_ack);
+	debug_sprintf_event(cf_diag_dbg, 5, "%s rc %d state %#llx\n",
+			    __func__, rc, cpuhw->state);
+}
+
+static int cf_diag_all_stop(void)
+{
+	struct cf_diag_call_on_cpu_parm p = {
+		.sets = cf_diag_ctrset.ctrset,
+	};
+	cpumask_var_t mask;
+
+	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
+		return -ENOMEM;
+	cpumask_and(mask, &cf_diag_ctrset.mask, cpu_online_mask);
+	on_each_cpu_mask(mask, cf_diag_ioctl_off, &p, 1);
+	free_cpumask_var(mask);
+	return 0;
+}
+
+static int cf_diag_all_start(void)
+{
+	struct cf_diag_call_on_cpu_parm p = {
+		.sets = cf_diag_ctrset.ctrset,
+		.cpus_ack = ATOMIC_INIT(0),
+	};
+	cpumask_var_t mask;
+	int rc = 0;
+
+	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
+		return -ENOMEM;
+	cpumask_and(mask, &cf_diag_ctrset.mask, cpu_online_mask);
+	on_each_cpu_mask(mask, cf_diag_ioctl_on, &p, 1);
+	if (atomic_read(&p.cpus_ack) != cpumask_weight(mask)) {
+		on_each_cpu_mask(mask, cf_diag_ioctl_off, &p, 1);
+		rc = -EIO;
+	}
+	free_cpumask_var(mask);
+	return rc;
+}
+
+/* Return the maximum required space for all possible CPUs in case one
+ * CPU will be onlined during the START, READ, STOP cycles.
+ * To find out the size of the counter sets, any one CPU will do. They
+ * all have the same counter sets.
+ */
+static size_t cf_diag_needspace(unsigned int sets)
+{
+	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
+	size_t bytes = 0;
+	int i;
+
+	for (i = CPUMF_CTR_SET_BASIC; i < CPUMF_CTR_SET_MAX; ++i) {
+		if (!(sets & cpumf_ctr_ctl[i]))
+			continue;
+		bytes += cf_diag_ctrset_size(i, &cpuhw->info) * sizeof(u64) +
+			 sizeof(((struct s390_ctrset_setdata *)0)->set) +
+			 sizeof(((struct s390_ctrset_setdata *)0)->no_cnts);
+	}
+	bytes = sizeof(((struct s390_ctrset_read *)0)->no_cpus) + nr_cpu_ids *
+		(bytes + sizeof(((struct s390_ctrset_cpudata *)0)->cpu_nr) +
+		     sizeof(((struct s390_ctrset_cpudata *)0)->no_sets));
+	debug_sprintf_event(cf_diag_dbg, 5, "%s bytes %ld\n", __func__,
+			    bytes);
+	return bytes;
+}
+
+static long cf_diag_ioctl_read(unsigned long arg)
+{
+	struct s390_ctrset_read read;
+	int ret = 0;
+
+	debug_sprintf_event(cf_diag_dbg, 5, "%s\n", __func__);
+	if (copy_from_user(&read, (char __user *)arg, sizeof(read)))
+		return -EFAULT;
+	ret = cf_diag_all_read(arg);
+	debug_sprintf_event(cf_diag_dbg, 5, "%s ret %d\n", __func__, ret);
+	return ret;
+}
+
+static long cf_diag_ioctl_stop(void)
+{
+	int ret;
+
+	debug_sprintf_event(cf_diag_dbg, 5, "%s\n", __func__);
+	ret = cf_diag_all_stop();
+	cf_diag_ctrset_clear();
+	debug_sprintf_event(cf_diag_dbg, 5, "%s ret %d\n", __func__, ret);
+	return ret;
+}
+
+static long cf_diag_ioctl_start(unsigned long arg)
+{
+	struct s390_ctrset_start __user *ustart;
+	struct s390_ctrset_start start;
+	void __user *umask;
+	unsigned int len;
+	int ret = 0;
+	size_t need;
+
+	if (cf_diag_ctrset.ctrset)
+		return -EBUSY;
+	ustart = (struct s390_ctrset_start __user *)arg;
+	if (copy_from_user(&start, ustart, sizeof(start)))
+		return -EFAULT;
+	if (start.version != S390_HWCTR_START_VERSION)
+		return -EINVAL;
+	if (start.counter_sets & ~(cpumf_ctr_ctl[CPUMF_CTR_SET_BASIC] |
+				   cpumf_ctr_ctl[CPUMF_CTR_SET_USER] |
+				   cpumf_ctr_ctl[CPUMF_CTR_SET_CRYPTO] |
+				   cpumf_ctr_ctl[CPUMF_CTR_SET_EXT] |
+				   cpumf_ctr_ctl[CPUMF_CTR_SET_MT_DIAG]))
+		return -EINVAL;		/* Invalid counter set */
+	if (!start.counter_sets)
+		return -EINVAL;		/* No counter set at all? */
+	cpumask_clear(&cf_diag_ctrset.mask);
+	len = min_t(u64, start.cpumask_len, cpumask_size());
+	umask = (void __user *)start.cpumask;
+	if (copy_from_user(&cf_diag_ctrset.mask, umask, len))
+		return -EFAULT;
+	if (cpumask_empty(&cf_diag_ctrset.mask))
+		return -EINVAL;
+	need = cf_diag_needspace(start.counter_sets);
+	if (put_user(need, &ustart->data_bytes))
+		ret = -EFAULT;
+	if (ret)
+		goto out;
+	cf_diag_ctrset.ctrset = start.counter_sets;
+	ret = cf_diag_all_start();
+out:
+	if (ret)
+		cf_diag_ctrset_clear();
+	debug_sprintf_event(cf_diag_dbg, 2, "%s sets %#lx need %ld ret %d\n",
+			    __func__, cf_diag_ctrset.ctrset, need, ret);
+	return ret;
+}
+
+static long cf_diag_ioctl(struct file *file, unsigned int cmd,
+			  unsigned long arg)
+{
+	int ret;
+
+	debug_sprintf_event(cf_diag_dbg, 2, "%s cmd %#x arg %lx\n", __func__,
+			    cmd, arg);
+	get_online_cpus();
+	mutex_lock(&cf_diag_ctrset_mutex);
+	switch (cmd) {
+	case S390_HWCTR_START:
+		ret = cf_diag_ioctl_start(arg);
+		break;
+	case S390_HWCTR_STOP:
+		ret = cf_diag_ioctl_stop();
+		break;
+	case S390_HWCTR_READ:
+		ret = cf_diag_ioctl_read(arg);
+		break;
+	default:
+		ret = -ENOTTY;
+		break;
+	}
+	mutex_unlock(&cf_diag_ctrset_mutex);
+	put_online_cpus();
+	debug_sprintf_event(cf_diag_dbg, 2, "%s ret %d\n", __func__, ret);
+	return ret;
+}
+
+static const struct file_operations cf_diag_fops = {
+	.owner = THIS_MODULE,
+	.open = cf_diag_open,
+	.release = cf_diag_release,
+	.unlocked_ioctl	= cf_diag_ioctl,
+	.compat_ioctl = cf_diag_ioctl,
+	.llseek = no_llseek
+};
+
+static struct miscdevice cf_diag_dev = {
+	.name	= S390_HWCTR_DEVICE,
+	.minor	= MISC_DYNAMIC_MINOR,
+	.fops	= &cf_diag_fops,
+};
+
+static int cf_diag_online_cpu(unsigned int cpu)
+{
+	struct cf_diag_call_on_cpu_parm p;
+
+	mutex_lock(&cf_diag_ctrset_mutex);
+	if (!cf_diag_ctrset.ctrset)
+		goto out;
+	p.sets = cf_diag_ctrset.ctrset;
+	cf_diag_ioctl_on(&p);
+out:
+	mutex_unlock(&cf_diag_ctrset_mutex);
+	return 0;
+}
+
+static int cf_diag_offline_cpu(unsigned int cpu)
+{
+	struct cf_diag_call_on_cpu_parm p;
+
+	mutex_lock(&cf_diag_ctrset_mutex);
+	if (!cf_diag_ctrset.ctrset)
+		goto out;
+	p.sets = cf_diag_ctrset.ctrset;
+	cf_diag_ioctl_off(&p);
+out:
+	mutex_unlock(&cf_diag_ctrset_mutex);
+	return 0;
+}
+
 /* Initialize the counter set PMU to generate complete counter set data as
  * event raw data. This relies on the CPU Measurement Counter Facility device
  * already being loaded and initialized.
@@ -685,21 +1165,43 @@ static int __init cf_diag_init(void)
 		return -ENOMEM;
 	}
 
+	rc = misc_register(&cf_diag_dev);
+	if (rc) {
+		pr_err("Registration of /dev/" S390_HWCTR_DEVICE
+		       "failed rc=%d\n", rc);
+		goto out;
+	}
+
 	/* Setup s390dbf facility */
 	cf_diag_dbg = debug_register(KMSG_COMPONENT, 2, 1, 128);
 	if (!cf_diag_dbg) {
 		pr_err("Registration of s390dbf(cpum_cf_diag) failed\n");
-		return -ENOMEM;
+		rc = -ENOMEM;
+		goto out_dbf;
 	}
 	debug_register_view(cf_diag_dbg, &debug_sprintf_view);
 
 	rc = perf_pmu_register(&cf_diag, "cpum_cf_diag", -1);
 	if (rc) {
-		debug_unregister_view(cf_diag_dbg, &debug_sprintf_view);
-		debug_unregister(cf_diag_dbg);
 		pr_err("Registration of PMU(cpum_cf_diag) failed with rc=%i\n",
 		       rc);
+		goto out_perf;
 	}
+	rc = cpuhp_setup_state_nocalls(CPUHP_AP_PERF_S390_CFD_ONLINE,
+				       "perf/s390/cfd:online",
+				       cf_diag_online_cpu, cf_diag_offline_cpu);
+	if (!rc)
+		goto out;
+
+	pr_err("Registration of CPUHP_AP_PERF_S390_CFD_ONLINE failed rc=%i\n",
+	       rc);
+	perf_pmu_unregister(&cf_diag);
+out_perf:
+	debug_unregister_view(cf_diag_dbg, &debug_sprintf_view);
+	debug_unregister(cf_diag_dbg);
+out_dbf:
+	misc_deregister(&cf_diag_dev);
+out:
 	return rc;
 }
-arch_initcall(cf_diag_init);
+device_initcall(cf_diag_init);