oprofile, s390: Add event interface to the System z hardware sampling module

With this patch the OProfile Basic Mode Sampling support for System z
is enhanced with a counter file system.  That way hardware sampling
can be configured using the user space tools with only little
modifications.

With the patch by default new cpu_types (s390/z10, s390/z196) are
returned in order to indicate that we are running a CPU which provides
the hardware sampling facility.  Existing user space tools will
complain about an unknown cpu type. In order to be compatible with
existing user space tools the `cpu_type' module parameter has been
added.  Setting the parameter to `timer' will force the module to
return `timer' as cpu_type.  The module will still try to use hardware
sampling if available and the hwsampling virtual filesystem will be
also be available for configuration.  So this has a different effect
than using the generic oprofile module parameter `timer=1'.

If the basic mode sampling is enabled on the machine and the
cpu_type=timer parameter is not used the kernel module will provide
the following virtual filesystem:

/dev/oprofile/0/enabled
/dev/oprofile/0/event
/dev/oprofile/0/count
/dev/oprofile/0/unit_mask
/dev/oprofile/0/kernel
/dev/oprofile/0/user

In the counter file system only the values of 'enabled', 'count',
'kernel', and 'user' are evaluated by the kernel module. Everything
else must contain fixed values.

The 'event' value only supports a single event - HWSAMPLING with value
0.

The 'count' value specifies the hardware sampling rate as it is passed
to the CPU measurement facility.

The 'kernel' and 'user' flags can now be used to filter for samples
when using hardware sampling.

Additionally also the following file will be created:
/dev/oprofile/timer/enabled

This will always be the inverted value of /dev/oprofile/0/enabled. 0
is not accepted without hardware sampling.
Signed-off-by: Andreas Krebbel <krebbel@linux.vnet.ibm.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>

Documentation/kernel-parameters.txt

@@ -1851,6 +1851,8 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
 			timer: [X86] Force use of architectural NMI
 				timer mode (see also oprofile.timer
 				for generic hr timer mode)
+				[s390] Force legacy basic mode sampling
+                                (report cpu_type "timer")
 
 	oops=panic	Always panic on oopses. Default is to just kill the
 			process, but there is a small probability of

arch/s390/oprofile/hwsampler.c

@@ -22,6 +22,7 @@
 #include <asm/irq.h>
 
 #include "hwsampler.h"
+#include "op_counter.h"
 
 #define MAX_NUM_SDB 511
 #define MIN_NUM_SDB 1
@@ -896,6 +897,8 @@ static void add_samples_to_oprofile(unsigned int cpu, unsigned long *sdbt,
 		if (sample_data_ptr->P == 1) {
 			/* userspace sample */
 			unsigned int pid = sample_data_ptr->prim_asn;
+			if (!counter_config.user)
+				goto skip_sample;
 			rcu_read_lock();
 			tsk = pid_task(find_vpid(pid), PIDTYPE_PID);
 			if (tsk)
@@ -903,6 +906,8 @@ static void add_samples_to_oprofile(unsigned int cpu, unsigned long *sdbt,
 			rcu_read_unlock();
 		} else {
 			/* kernelspace sample */
+			if (!counter_config.kernel)
+				goto skip_sample;
 			regs = task_pt_regs(current);
 		}
 
@@ -910,7 +915,7 @@ static void add_samples_to_oprofile(unsigned int cpu, unsigned long *sdbt,
 		oprofile_add_ext_hw_sample(sample_data_ptr->ia, regs, 0,
 				!sample_data_ptr->P, tsk);
 		mutex_unlock(&hws_sem);
-
+	skip_sample:
 		sample_data_ptr++;
 	}
 }

arch/s390/oprofile/init.c

@@ -2,10 +2,11 @@
  * arch/s390/oprofile/init.c
  *
  * S390 Version
- *   Copyright (C) 2003 IBM Deutschland Entwicklung GmbH, IBM Corporation
+ *   Copyright (C) 2002-2011 IBM Deutschland Entwicklung GmbH, IBM Corporation
  *   Author(s): Thomas Spatzier (tspat@de.ibm.com)
  *   Author(s): Mahesh Salgaonkar (mahesh@linux.vnet.ibm.com)
  *   Author(s): Heinz Graalfs (graalfs@linux.vnet.ibm.com)
+ *   Author(s): Andreas Krebbel (krebbel@linux.vnet.ibm.com)
  *
  * @remark Copyright 2002-2011 OProfile authors
  */
@@ -14,6 +15,8 @@
 #include <linux/init.h>
 #include <linux/errno.h>
 #include <linux/fs.h>
+#include <linux/module.h>
+#include <asm/processor.h>
 
 #include "../../../drivers/oprofile/oprof.h"
 
@@ -22,6 +25,7 @@ extern void s390_backtrace(struct pt_regs * const regs, unsigned int depth);
 #ifdef CONFIG_64BIT
 
 #include "hwsampler.h"
+#include "op_counter.h"
 
 #define DEFAULT_INTERVAL	4127518
 
@@ -35,16 +39,41 @@ static unsigned long oprofile_max_interval;
 static unsigned long oprofile_sdbt_blocks = DEFAULT_SDBT_BLOCKS;
 static unsigned long oprofile_sdb_blocks = DEFAULT_SDB_BLOCKS;
 
-static int hwsampler_file;
+static int hwsampler_enabled;
 static int hwsampler_running;	/* start_mutex must be held to change */
+static int hwsampler_available;
 
 static struct oprofile_operations timer_ops;
 
+struct op_counter_config counter_config;
+
+enum __force_cpu_type {
+	reserved = 0,		/* do not force */
+	timer,
+};
+static int force_cpu_type;
+
+static int set_cpu_type(const char *str, struct kernel_param *kp)
+{
+	if (!strcmp(str, "timer")) {
+		force_cpu_type = timer;
+		printk(KERN_INFO "oprofile: forcing timer to be returned "
+		                 "as cpu type\n");
+	} else {
+		force_cpu_type = 0;
+	}
+
+	return 0;
+}
+module_param_call(cpu_type, set_cpu_type, NULL, NULL, 0);
+MODULE_PARM_DESC(cpu_type, "Force legacy basic mode sampling"
+		           "(report cpu_type \"timer\"");
+
 static int oprofile_hwsampler_start(void)
 {
 	int retval;
 
-	hwsampler_running = hwsampler_file;
+	hwsampler_running = hwsampler_enabled;
 
 	if (!hwsampler_running)
 		return timer_ops.start();
@@ -72,10 +101,16 @@ static void oprofile_hwsampler_stop(void)
 	return;
 }
 
+/*
+ * File ops used for:
+ * /dev/oprofile/0/enabled
+ * /dev/oprofile/hwsampling/hwsampler  (cpu_type = timer)
+ */
+
 static ssize_t hwsampler_read(struct file *file, char __user *buf,
 		size_t count, loff_t *offset)
 {
-	return oprofilefs_ulong_to_user(hwsampler_file, buf, count, offset);
+	return oprofilefs_ulong_to_user(hwsampler_enabled, buf, count, offset);
 }
 
 static ssize_t hwsampler_write(struct file *file, char const __user *buf,
@@ -91,6 +126,9 @@ static ssize_t hwsampler_write(struct file *file, char const __user *buf,
 	if (retval)
 		return retval;
 
+	if (val != 0 && val != 1)
+		return -EINVAL;
+
 	if (oprofile_started)
 		/*
 		 * save to do without locking as we set
@@ -99,7 +137,7 @@ static ssize_t hwsampler_write(struct file *file, char const __user *buf,
 		 */
 		return -EBUSY;
 
-	hwsampler_file = val;
+	hwsampler_enabled = val;
 
 	return count;
 }
@@ -109,38 +147,311 @@ static const struct file_operations hwsampler_fops = {
 	.write		= hwsampler_write,
 };
 
+/*
+ * File ops used for:
+ * /dev/oprofile/0/count
+ * /dev/oprofile/hwsampling/hw_interval  (cpu_type = timer)
+ *
+ * Make sure that the value is within the hardware range.
+ */
+
+static ssize_t hw_interval_read(struct file *file, char __user *buf,
+				size_t count, loff_t *offset)
+{
+	return oprofilefs_ulong_to_user(oprofile_hw_interval, buf,
+					count, offset);
+}
+
+static ssize_t hw_interval_write(struct file *file, char const __user *buf,
+				 size_t count, loff_t *offset)
+{
+	unsigned long val;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+	retval = oprofilefs_ulong_from_user(&val, buf, count);
+	if (retval)
+		return retval;
+	if (val < oprofile_min_interval)
+		oprofile_hw_interval = oprofile_min_interval;
+	else if (val > oprofile_max_interval)
+		oprofile_hw_interval = oprofile_max_interval;
+	else
+		oprofile_hw_interval = val;
+
+	return count;
+}
+
+static const struct file_operations hw_interval_fops = {
+	.read		= hw_interval_read,
+	.write		= hw_interval_write,
+};
+
+/*
+ * File ops used for:
+ * /dev/oprofile/0/event
+ * Only a single event with number 0 is supported with this counter.
+ *
+ * /dev/oprofile/0/unit_mask
+ * This is a dummy file needed by the user space tools.
+ * No value other than 0 is accepted or returned.
+ */
+
+static ssize_t hwsampler_zero_read(struct file *file, char __user *buf,
+				    size_t count, loff_t *offset)
+{
+	return oprofilefs_ulong_to_user(0, buf, count, offset);
+}
+
+static ssize_t hwsampler_zero_write(struct file *file, char const __user *buf,
+				     size_t count, loff_t *offset)
+{
+	unsigned long val;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = oprofilefs_ulong_from_user(&val, buf, count);
+	if (retval)
+		return retval;
+	if (val != 0)
+		return -EINVAL;
+	return count;
+}
+
+static const struct file_operations zero_fops = {
+	.read		= hwsampler_zero_read,
+	.write		= hwsampler_zero_write,
+};
+
+/* /dev/oprofile/0/kernel file ops.  */
+
+static ssize_t hwsampler_kernel_read(struct file *file, char __user *buf,
+				     size_t count, loff_t *offset)
+{
+	return oprofilefs_ulong_to_user(counter_config.kernel,
+					buf, count, offset);
+}
+
+static ssize_t hwsampler_kernel_write(struct file *file, char const __user *buf,
+				      size_t count, loff_t *offset)
+{
+	unsigned long val;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = oprofilefs_ulong_from_user(&val, buf, count);
+	if (retval)
+		return retval;
+
+	if (val != 0 && val != 1)
+		return -EINVAL;
+
+	counter_config.kernel = val;
+
+	return count;
+}
+
+static const struct file_operations kernel_fops = {
+	.read		= hwsampler_kernel_read,
+	.write		= hwsampler_kernel_write,
+};
+
+/* /dev/oprofile/0/user file ops. */
+
+static ssize_t hwsampler_user_read(struct file *file, char __user *buf,
+				   size_t count, loff_t *offset)
+{
+	return oprofilefs_ulong_to_user(counter_config.user,
+					buf, count, offset);
+}
+
+static ssize_t hwsampler_user_write(struct file *file, char const __user *buf,
+				    size_t count, loff_t *offset)
+{
+	unsigned long val;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = oprofilefs_ulong_from_user(&val, buf, count);
+	if (retval)
+		return retval;
+
+	if (val != 0 && val != 1)
+		return -EINVAL;
+
+	counter_config.user = val;
+
+	return count;
+}
+
+static const struct file_operations user_fops = {
+	.read		= hwsampler_user_read,
+	.write		= hwsampler_user_write,
+};
+
+
+/*
+ * File ops used for: /dev/oprofile/timer/enabled
+ * The value always has to be the inverted value of hwsampler_enabled. So
+ * no separate variable is created. That way we do not need locking.
+ */
+
+static ssize_t timer_enabled_read(struct file *file, char __user *buf,
+				  size_t count, loff_t *offset)
+{
+	return oprofilefs_ulong_to_user(!hwsampler_enabled, buf, count, offset);
+}
+
+static ssize_t timer_enabled_write(struct file *file, char const __user *buf,
+				   size_t count, loff_t *offset)
+{
+	unsigned long val;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = oprofilefs_ulong_from_user(&val, buf, count);
+	if (retval)
+		return retval;
+
+	if (val != 0 && val != 1)
+		return -EINVAL;
+
+	/* Timer cannot be disabled without having hardware sampling.  */
+	if (val == 0 && !hwsampler_available)
+		return -EINVAL;
+
+	if (oprofile_started)
+		/*
+		 * save to do without locking as we set
+		 * hwsampler_running in start() when start_mutex is
+		 * held
+		 */
+		return -EBUSY;
+
+	hwsampler_enabled = !val;
+
+	return count;
+}
+
+static const struct file_operations timer_enabled_fops = {
+	.read		= timer_enabled_read,
+	.write		= timer_enabled_write,
+};
+
+
 static int oprofile_create_hwsampling_files(struct super_block *sb,
-						struct dentry *root)
+					    struct dentry *root)
 {
-	struct dentry *hw_dir;
+	struct dentry *dir;
+
+	dir = oprofilefs_mkdir(sb, root, "timer");
+	if (!dir)
+		return -EINVAL;
+
+	oprofilefs_create_file(sb, dir, "enabled", &timer_enabled_fops);
+
+	if (!hwsampler_available)
+		return 0;
 
 	/* reinitialize default values */
-	hwsampler_file = 1;
+	hwsampler_enabled = 1;
+	counter_config.kernel = 1;
+	counter_config.user = 1;
 
-	hw_dir = oprofilefs_mkdir(sb, root, "hwsampling");
-	if (!hw_dir)
-		return -EINVAL;
+	if (!force_cpu_type) {
+		/*
+		 * Create the counter file system.  A single virtual
+		 * counter is created which can be used to
+		 * enable/disable hardware sampling dynamically from
+		 * user space.  The user space will configure a single
+		 * counter with a single event.  The value of 'event'
+		 * and 'unit_mask' are not evaluated by the kernel code
+		 * and can only be set to 0.
+		 */
+
+		dir = oprofilefs_mkdir(sb, root, "0");
+		if (!dir)
+			return -EINVAL;
 
-	oprofilefs_create_file(sb, hw_dir, "hwsampler", &hwsampler_fops);
-	oprofilefs_create_ulong(sb, hw_dir, "hw_interval",
-				&oprofile_hw_interval);
-	oprofilefs_create_ro_ulong(sb, hw_dir, "hw_min_interval",
-				&oprofile_min_interval);
-	oprofilefs_create_ro_ulong(sb, hw_dir, "hw_max_interval",
-				&oprofile_max_interval);
-	oprofilefs_create_ulong(sb, hw_dir, "hw_sdbt_blocks",
-				&oprofile_sdbt_blocks);
+		oprofilefs_create_file(sb, dir, "enabled", &hwsampler_fops);
+		oprofilefs_create_file(sb, dir, "event", &zero_fops);
+		oprofilefs_create_file(sb, dir, "count", &hw_interval_fops);
+		oprofilefs_create_file(sb, dir, "unit_mask", &zero_fops);
+		oprofilefs_create_file(sb, dir, "kernel", &kernel_fops);
+		oprofilefs_create_file(sb, dir, "user", &user_fops);
+		oprofilefs_create_ulong(sb, dir, "hw_sdbt_blocks",
+					&oprofile_sdbt_blocks);
 
+	} else {
+		/*
+		 * Hardware sampling can be used but the cpu_type is
+		 * forced to timer in order to deal with legacy user
+		 * space tools.  The /dev/oprofile/hwsampling fs is
+		 * provided in that case.
+		 */
+		dir = oprofilefs_mkdir(sb, root, "hwsampling");
+		if (!dir)
+			return -EINVAL;
+
+		oprofilefs_create_file(sb, dir, "hwsampler",
+				       &hwsampler_fops);
+		oprofilefs_create_file(sb, dir, "hw_interval",
+				       &hw_interval_fops);
+		oprofilefs_create_ro_ulong(sb, dir, "hw_min_interval",
+					   &oprofile_min_interval);
+		oprofilefs_create_ro_ulong(sb, dir, "hw_max_interval",
+					   &oprofile_max_interval);
+		oprofilefs_create_ulong(sb, dir, "hw_sdbt_blocks",
+					&oprofile_sdbt_blocks);
+	}
 	return 0;
 }
 
 static int oprofile_hwsampler_init(struct oprofile_operations *ops)
 {
+	/*
+	 * Initialize the timer mode infrastructure as well in order
+	 * to be able to switch back dynamically.  oprofile_timer_init
+	 * is not supposed to fail.
+	 */
+	if (oprofile_timer_init(ops))
+		BUG();
+
+	memcpy(&timer_ops, ops, sizeof(timer_ops));
+	ops->create_files = oprofile_create_hwsampling_files;
+
+	/*
+	 * If the user space tools do not support newer cpu types,
+	 * the force_cpu_type module parameter
+	 * can be used to always return \"timer\" as cpu type.
+	 */
+	if (force_cpu_type != timer) {
+		struct cpuid id;
+
+		get_cpu_id (&id);
+
+		switch (id.machine) {
+		case 0x2097: case 0x2098: ops->cpu_type = "s390/z10"; break;
+		case 0x2817: case 0x2818: ops->cpu_type = "s390/z196"; break;
+		default: return -ENODEV;
+		}
+	}
+
 	if (hwsampler_setup())
 		return -ENODEV;
 
 	/*
-	 * create hwsampler files only if hwsampler_setup() succeeds.
+	 * Query the range for the sampling interval from the
+	 * hardware.
 	 */
 	oprofile_min_interval = hwsampler_query_min_interval();
 	if (oprofile_min_interval == 0)
@@ -155,16 +466,11 @@ static int oprofile_hwsampler_init(struct oprofile_operations *ops)
 	if (oprofile_hw_interval > oprofile_max_interval)
 		oprofile_hw_interval = oprofile_max_interval;
 
-	if (oprofile_timer_init(ops))
-		return -ENODEV;
-
-	printk(KERN_INFO "oprofile: using hardware sampling\n");
-
-	memcpy(&timer_ops, ops, sizeof(timer_ops));
+	printk(KERN_INFO "oprofile: System z hardware sampling "
+	       "facility found.\n");
 
 	ops->start = oprofile_hwsampler_start;
 	ops->stop = oprofile_hwsampler_stop;
-	ops->create_files = oprofile_create_hwsampling_files;
 
 	return 0;
 }
@@ -181,7 +487,15 @@ int __init oprofile_arch_init(struct oprofile_operations *ops)
 	ops->backtrace = s390_backtrace;
 
 #ifdef CONFIG_64BIT
-	return oprofile_hwsampler_init(ops);
+
+	/*
+	 * -ENODEV is not reported to the caller.  The module itself
+         * will use the timer mode sampling as fallback and this is
+         * always available.
+	 */
+	hwsampler_available = oprofile_hwsampler_init(ops) == 0;
+
+	return 0;
 #else
 	return -ENODEV;
 #endif

arch/s390/oprofile/op_counter.h

+/**
+ * arch/s390/oprofile/op_counter.h
+ *
+ *   Copyright (C) 2011 IBM Deutschland Entwicklung GmbH, IBM Corporation
+ *   Author(s): Andreas Krebbel (krebbel@linux.vnet.ibm.com)
+ *
+ * @remark Copyright 2011 OProfile authors
+ */
+
+#ifndef OP_COUNTER_H
+#define OP_COUNTER_H
+
+struct op_counter_config {
+	/* `enabled' maps to the hwsampler_file variable.  */
+	/* `count' maps to the oprofile_hw_interval variable.  */
+	/* `event' and `unit_mask' are unused. */
+	unsigned long kernel;
+	unsigned long user;
+};
+
+extern struct op_counter_config counter_config;
+
+#endif /* OP_COUNTER_H */