[PATCH 4/16] LTTng : Linux Trace Toolkit Next Generation 0.5.95, kernel 2.6.17

[Mathieu Desnoyers]

4- LTTng facility infrastructure : A facility contains the metainformation
on a group of events. It is dynamically loadable.
patch-2.6.17-lttng-0.5.95-facilities.diff

OpenPGP public key:              http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint:     8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68 
--- /dev/null
+++ b/include/linux/ltt-facilities.h
@@ -0,0 +1,113 @@
+/*
+ * linux/include/linux/ltt-facilities.h
+ *
+ * Copyright (C) 2005 Mathieu Desnoyers (mathieu.desnoyers@xxxxxxxxxx)
+ *
+ * This contains the definitions for the Linux Trace Toolkit facilities.
+ *
+ * Each facility must store its facility index number (returned by the register)
+ * in an exported symbol named :
+ *
+ * ltt_facility_name_checksum
+ *
+ * Where name is the name of the facility, and checksum is the text that
+ * corresponds to the checksum of the facility in hexadecimal notation.
+ */
+
+#ifndef _LTT_FACILITIES_H
+#define _LTT_FACILITIES_H
+
+#include <linux/config.h>
+#include <linux/types.h>
+#include <asm/atomic.h>
+#include <linux/types.h>
+
+#define FACNAME_LEN 32
+
+struct user_facility_info {
+	char name[FACNAME_LEN];
+	unsigned int num_events;
+	size_t alignment;
+	u32 checksum;
+	size_t int_size;
+	size_t long_size;
+	size_t pointer_size;
+	size_t size_t_size;
+};
+
+#ifdef __KERNEL__
+
+/* Is kernel tracing enabled */
+#if defined(CONFIG_LTT)
+
+#define LTT_FAC_PER_PROCESS 16
+
+#define LTT_MAX_NUM_FACILITIES	256
+#define LTT_RESERVED_FACILITIES	1
+
+typedef unsigned long ltt_facility_t;
+
+enum ltt_facility_type {
+	LTT_FACILITY_TYPE_KERNEL,
+	LTT_FACILITY_TYPE_USER
+};
+
+struct ltt_facility {
+	const char *name;
+	const unsigned int num_events;
+	const u32 checksum;
+	const char *symbol;
+};
+
+struct ltt_facility_info {
+	char name[FACNAME_LEN];
+	enum ltt_facility_type type;
+	unsigned int num_events;
+	size_t alignment;
+	u32 checksum;
+	size_t int_size;
+	size_t long_size;
+	size_t pointer_size;
+	size_t size_t_size;
+	atomic_t ref;
+};
+
+struct ltt_trace_struct;
+
+void ltt_facility_ref(ltt_facility_t facility_id);
+
+int ltt_facility_unregister(ltt_facility_t facility_id);
+
+
+int ltt_facility_register(enum ltt_facility_type type,
+		const char *name,
+		const unsigned int num_events,
+		u32 checksum,
+		size_t int_size,
+		size_t long_size,
+		size_t pointer_size,
+		size_t size_t_size,
+		size_t alignment);
+
+int ltt_facility_verify(enum ltt_facility_type type,
+		const char *name,
+		const unsigned int num_events,
+		const u32 checksum,
+		size_t int_size,
+		size_t long_size,
+		size_t pointer_size,
+		size_t size_t_size,
+		size_t alignment);
+
+unsigned int ltt_facility_kernel_register(struct ltt_facility *facility);
+
+int ltt_facility_user_access_ok(ltt_facility_t fac_id);
+
+void ltt_facility_free_unused(void);
+
+void ltt_facility_state_dump(struct ltt_trace_struct *trace);
+
+#endif //__KERNEL__
+
+#endif /* defined(CONFIG_LTT) */
+#endif /* _LTT_FACILITIES_H */
diff --git a/include/linux/ltt-relayfs.h b/include/linux/ltt-relayfs.h
new file mode 100644
index 0000000..804ad19
--- /dev/null
+++ b/kernel/ltt-facilities.c
@@ -0,0 +1,318 @@
+/*
+ * ltt-facilities.c
+ *
+ * (C) Copyright	2005 -
+ *		Mathieu Desnoyers (mathieu.desnoyers@xxxxxxxxxx)
+ *
+ * Contains the kernel core code for Linux Trace Toolkit facilities.
+ *
+ * Facilities are a group of events that can be recorded by instrumentation
+ * points to a trace. We keep track of the active facilities to know which type
+ * of information can be present in a trace.
+ *
+ * We never reuse a facility id.
+ * 
+ * Author:
+ *	Mathieu Desnoyers (mathieu.desnoyers@xxxxxxxxxx)
+ *
+ */
+
+#include <linux/ltt/ltt-facility-core.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/ltt-facilities.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+
+static struct ltt_facility_info facilities[LTT_MAX_NUM_FACILITIES];
+static spinlock_t	facilities_lock;
+
+/* Facility registration :
+ * hash based on the checksum, except for the core facility, which is 0.
+ * This function assumes that the facility has never been registered before.
+ * User ltt_facility_verify for this. */
+int ltt_facility_register(enum ltt_facility_type type,
+		const char *name,
+		const unsigned int num_events,
+		const u32 checksum,
+		size_t int_size,
+		size_t long_size,
+		size_t pointer_size,
+		size_t size_t_size,
+		size_t alignment)
+{
+	int fac_id;
+	int chk_fac_id;
+	int found=0;
+  
+	spin_lock(&facilities_lock);
+	if(type == LTT_FACILITY_TYPE_KERNEL &&
+			strncmp(name, "core", sizeof("core")) == 0) {
+		fac_id = 0;
+	} else {
+		/* fac_id based on checksum%LTT_MAX_NUM_FACILITIES
+		 * find an empty slot */
+		chk_fac_id = fac_id = checksum % LTT_MAX_NUM_FACILITIES;
+		do {
+			if(atomic_read(&facilities[fac_id].ref) == 0) {
+				found = 1;
+				break;
+			}
+			fac_id = (fac_id+1) % LTT_MAX_NUM_FACILITIES;
+		} while(fac_id != chk_fac_id);
+
+		if(!found) {
+			fac_id = -EPERM;
+			goto unlock;
+		}
+	}
+	switch(type) {
+		case LTT_FACILITY_TYPE_USER:
+			if(strncmp(name, "user_", sizeof("user_")-1) != 0) {
+				fac_id = -EPERM;
+				goto unlock;
+			}
+			break;
+		case LTT_FACILITY_TYPE_KERNEL:
+			break;
+	}
+	strncpy(facilities[fac_id].name, name, FACNAME_LEN-1);
+	facilities[fac_id].name[FACNAME_LEN-1] = '\0';
+	facilities[fac_id].type = type;
+	facilities[fac_id].num_events = num_events;
+	facilities[fac_id].alignment = alignment;
+	facilities[fac_id].checksum = checksum;
+	facilities[fac_id].int_size = int_size;
+	facilities[fac_id].long_size = long_size;
+	facilities[fac_id].pointer_size = pointer_size;
+	facilities[fac_id].size_t_size = size_t_size;
+
+	if(atomic_read(&facilities[fac_id].ref) == 0) {
+		atomic_add(2, &facilities[fac_id].ref);
+		trace_core_facility_load(
+				facilities[fac_id].name,
+				checksum,
+				fac_id,
+				int_size,
+				long_size,
+				pointer_size,
+				size_t_size,
+				alignment);
+	} else {
+		atomic_inc(&facilities[fac_id].ref);
+	}
+unlock:
+	spin_unlock(&facilities_lock);
+	return fac_id;
+}
+
+/* Verifies if a facility is already registered. If it is,
+ * it returns the facility id. If it is not registered, it returns 0. (0 is the
+ * core facility which must never use ltt_facility_verify).
+ * If the facility is already registered, increment its refcount. */
+int ltt_facility_verify(enum ltt_facility_type type,
+		const char *name,
+		const unsigned int num_events,
+		const u32 checksum,
+		size_t int_size,
+		size_t long_size,
+		size_t pointer_size,
+		size_t size_t_size,
+		size_t alignment)
+{
+	int fac_id;
+	int chk_fac_id;
+	int found=0;
+	
+	spin_lock(&facilities_lock);
+	if(type == LTT_FACILITY_TYPE_KERNEL && 
+			strncmp(name, "core", sizeof("core")) == 0) {
+		fac_id = 0; /* Core facility */
+		goto unlock;
+	} else {
+		switch(type) {
+			case LTT_FACILITY_TYPE_USER:
+				if(strncmp(name,
+					"user_", sizeof("user_")-1) != 0) {
+					fac_id = 0;
+					goto unlock;
+				}
+				break;
+			case LTT_FACILITY_TYPE_KERNEL:
+				break;
+		}
+		/* fac_id based on checksum%LTT_MAX_NUM_FACILITIES */
+		chk_fac_id = fac_id = checksum % LTT_MAX_NUM_FACILITIES;
+		do {
+			if(facilities[fac_id].checksum == checksum) {
+				/* Possibly found : check carefully */
+				if((atomic_read(&facilities[fac_id].ref) > 0) &&
+					strncmp(facilities[fac_id].name,
+						name, FACNAME_LEN-1) == 0 &&
+				facilities[fac_id].type == type &&
+				facilities[fac_id].num_events == num_events &&
+				facilities[fac_id].alignment == alignment &&
+				facilities[fac_id].checksum == checksum &&
+				facilities[fac_id].int_size == int_size &&
+				facilities[fac_id].long_size == long_size &&
+				facilities[fac_id].pointer_size
+					== pointer_size &&
+				facilities[fac_id].size_t_size == size_t_size) {
+					found = 1;
+					break;
+				}
+			}
+			fac_id = (fac_id+1) % LTT_MAX_NUM_FACILITIES;
+		} while(fac_id != chk_fac_id);
+		
+		if(!found) {
+			fac_id = 0;
+			goto unlock;
+		}
+		atomic_inc(&facilities[fac_id].ref);
+	}
+unlock:
+	spin_unlock(&facilities_lock);
+	return fac_id;
+}
+
+
+unsigned int ltt_facility_kernel_register(struct ltt_facility *facility)
+{
+	size_t alignment;
+#ifdef CONFIG_LTT_ALIGNMENT
+	alignment = sizeof(void*);
+#else
+	alignment = 0;
+#endif
+	return ltt_facility_register(LTT_FACILITY_TYPE_KERNEL,
+			facility->name, facility->num_events,
+			facility->checksum,
+			sizeof(int), sizeof(long), sizeof(void*),
+			sizeof(size_t), alignment);
+}
+
+void ltt_facility_ref(ltt_facility_t facility_id)
+{
+	atomic_inc(&facilities[facility_id].ref);
+}
+
+int ltt_facility_unregister(ltt_facility_t facility_id)
+{
+	int ret;
+	int freed = 0;
+
+	if(facility_id >= LTT_MAX_NUM_FACILITIES) return -EPERM;
+	
+	spin_lock(&facilities_lock);
+	if(atomic_read(&facilities[facility_id].ref) == 0) {
+		ret = -EPERM;
+		goto unlock;
+	}
+	printk(KERN_DEBUG "LTT unregister facility %lu\n", facility_id);
+	atomic_dec(&facilities[facility_id].ref);
+
+	/* Disable preemption because we read the traces list, and want it to be
+	 * RCU coherent. */
+	preempt_disable();
+	ltt_nesting[smp_processor_id()]++;
+	barrier();
+
+	/* If no more trace in the list, we can free the unused facility,
+	 * otherwise it will be freed later when the last trace is destroyed (by
+	 * ltt_facility_free_unused()). */
+	if(list_empty(&ltt_traces.head))
+		if(atomic_read(&facilities[facility_id].ref) == 1) {
+			atomic_dec(&facilities[facility_id].ref);
+			freed = 1;
+		}
+
+	barrier();
+	ltt_nesting[smp_processor_id()]--;
+	preempt_enable();
+
+	/* Ok, if we think about it, it's never going to be traced as there are
+	 * no traces in the list. In fact, we never really want to free a
+	 * facility id when there is tracing active.
+	 * FIXME : this unload could go away. */
+	if(freed)
+		trace_core_facility_unload(facility_id);
+	ret = 0;
+unlock:
+	spin_unlock(&facilities_lock);
+	return ret;
+}
+
+int ltt_facility_user_access_ok(ltt_facility_t fac_id)
+{
+	if(atomic_read(&facilities[fac_id].ref) == 0) return 0;
+	if(facilities[fac_id].type == LTT_FACILITY_TYPE_KERNEL) return 0;
+
+	return 1;
+}
+
+/* Cleanup all the unregistered facilities. This must be done when all the
+ * traces are destroyed.
+ */
+void ltt_facility_free_unused(void)
+{
+	int fac_id;
+	
+	spin_lock(&facilities_lock);
+	for(fac_id=0;fac_id<LTT_MAX_NUM_FACILITIES;fac_id++)
+		if(atomic_read(&facilities[fac_id].ref) == 1)
+			atomic_dec(&facilities[fac_id].ref);
+	spin_unlock(&facilities_lock);
+}
+
+void ltt_facility_state_dump(struct ltt_trace_struct *trace)
+{
+	int fac_id;
+	struct ltt_facility_info *facility;
+	u32 int_size, long_size, pointer_size, size_t_size;
+
+	int_size = sizeof(int);
+	long_size = sizeof(long);
+	pointer_size = sizeof(void*);
+	size_t_size = sizeof(size_t);
+
+	spin_lock(&facilities_lock);
+	for(fac_id=0;fac_id<LTT_MAX_NUM_FACILITIES;fac_id++) {
+		if(atomic_read(&facilities[fac_id].ref) > 1) {
+			facility = &facilities[fac_id];
+			printk(KERN_DEBUG "Dumping facility %s\n",
+					facility->name);
+			trace_core_state_dump_facility_load(trace,
+					facility->name,
+					facility->checksum,
+					fac_id,
+					facility->int_size,
+					facility->long_size,
+					facility->pointer_size,
+					facility->size_t_size,
+					facility->alignment);
+		}
+	}
+	spin_unlock(&facilities_lock);
+}
+
+EXPORT_SYMBOL(ltt_facility_kernel_register);
+EXPORT_SYMBOL(ltt_facility_ref);
+EXPORT_SYMBOL(ltt_facility_unregister);
+EXPORT_SYMBOL(ltt_facility_free_unused);
+EXPORT_SYMBOL(ltt_facility_state_dump);
+
+static int __init ltt_facilities_init(void)
+{
+	int i;
+	printk(KERN_INFO "LTT : ltt-facilities init\n");
+
+	spin_lock_init(&facilities_lock);
+	for(i=0; i<LTT_MAX_NUM_FACILITIES; i++) {
+		atomic_set(&facilities[i].ref, 0);
+	}
+	
+	return 0;
+}
+__initcall(ltt_facilities_init);
+
diff --git a/kernel/ltt-heartbeat.c b/kernel/ltt-heartbeat.c
new file mode 100644
index 0000000..a88f1b2