Merge branch 'tracing-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'tracing-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (241 commits)
  sched, trace: update trace_sched_wakeup()
  tracing/ftrace: don't trace on early stage of a secondary cpu boot, v3
  Revert "x86: disable X86_PTRACE_BTS"
  ring-buffer: prevent false positive warning
  ring-buffer: fix dangling commit race
  ftrace: enable format arguments checking
  x86, bts: memory accounting
  x86, bts: add fork and exit handling
  ftrace: introduce tracing_reset_online_cpus() helper
  tracing: fix warnings in kernel/trace/trace_sched_switch.c
  tracing: fix warning in kernel/trace/trace.c
  tracing/ring-buffer: remove unused ring_buffer size
  trace: fix task state printout
  ftrace: add not to regex on filtering functions
  trace: better use of stack_trace_enabled for boot up code
  trace: add a way to enable or disable the stack tracer
  x86: entry_64 - introduce FTRACE_ frame macro v2
  tracing/ftrace: add the printk-msg-only option
  tracing/ftrace: use preempt_enable_no_resched_notrace in ring_buffer_time_stamp()
  x86, bts: correctly report invalid bts records
  ...

Fixed up trivial conflict in scripts/recordmcount.pl due to SH bits
being already partly merged by the SH merge.

23 files changed, 1465 insertions, 923 deletions

diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 1f208aaee780..88dd768eab6d 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -66,6 +66,7 @@ obj-$(CONFIG_X86_LOCAL_APIC)	+= apic.o nmi.o
 obj-$(CONFIG_X86_IO_APIC)	+= io_apic.o
 obj-$(CONFIG_X86_REBOOTFIXUPS)	+= reboot_fixups_32.o
 obj-$(CONFIG_DYNAMIC_FTRACE)	+= ftrace.o
+obj-$(CONFIG_FUNCTION_GRAPH_TRACER)	+= ftrace.o
 obj-$(CONFIG_KEXEC)		+= machine_kexec_$(BITS).o
 obj-$(CONFIG_KEXEC)		+= relocate_kernel_$(BITS).o crash.o
 obj-$(CONFIG_CRASH_DUMP)	+= crash_dump_$(BITS).o
diff --git a/arch/x86/kernel/apic.c b/arch/x86/kernel/apic.c
index 7397911f8478..b5229affb953 100644
--- a/arch/x86/kernel/apic.c
+++ b/arch/x86/kernel/apic.c
@@ -30,6 +30,7 @@
 #include <linux/module.h>
 #include <linux/dmi.h>
 #include <linux/dmar.h>
+#include <linux/ftrace.h>
 
 #include <asm/atomic.h>
 #include <asm/smp.h>
@@ -790,7 +791,7 @@ static void local_apic_timer_interrupt(void)
  * [ if a single-CPU system runs an SMP kernel then we call the local
  *   interrupt as well. Thus we cannot inline the local irq ... ]
  */
-void smp_apic_timer_interrupt(struct pt_regs *regs)
+void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
 {
 	struct pt_regs *old_regs = set_irq_regs(regs);
 
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index a5c04e88777e..82db7f45e2de 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -2,6 +2,11 @@
 # Makefile for x86-compatible CPU details and quirks
 #
 
+# Don't trace early stages of a secondary CPU boot
+ifdef CONFIG_FUNCTION_TRACER
+CFLAGS_REMOVE_common.o = -pg
+endif
+
 obj-y			:= intel_cacheinfo.o addon_cpuid_features.o
 obj-y			+= proc.o capflags.o powerflags.o common.o
 obj-y			+= vmware.o hypervisor.o
diff --git a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
index 8e48c5d4467d..88ea02dcb622 100644
--- a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -33,6 +33,7 @@
 #include <linux/cpufreq.h>
 #include <linux/compiler.h>
 #include <linux/dmi.h>
+#include <linux/ftrace.h>
 
 #include <linux/acpi.h>
 #include <acpi/processor.h>
@@ -391,6 +392,7 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 	unsigned int next_perf_state = 0; /* Index into perf table */
 	unsigned int i;
 	int result = 0;
+	struct power_trace it;
 
 	dprintk("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
 
@@ -427,6 +429,8 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 		}
 	}
 
+	trace_power_mark(&it, POWER_PSTATE, next_perf_state);
+
 	switch (data->cpu_feature) {
 	case SYSTEM_INTEL_MSR_CAPABLE:
 		cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index ccfd2047630c..8ea6929e974c 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -11,7 +11,6 @@
 #include <asm/pgtable.h>
 #include <asm/msr.h>
 #include <asm/uaccess.h>
-#include <asm/ptrace.h>
 #include <asm/ds.h>
 #include <asm/bugs.h>
 
@@ -326,9 +325,6 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 		set_cpu_cap(c, X86_FEATURE_P3);
 #endif
 
-	if (cpu_has_bts)
-		ptrace_bts_init_intel(c);
-
 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
 		/*
 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index d6938d9351cf..da91701a2348 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -6,14 +6,13 @@
  * precise-event based sampling (PEBS).
  *
  * It manages:
- * - per-thread and per-cpu allocation of BTS and PEBS
- * - buffer memory allocation (optional)
- * - buffer overflow handling
+ * - DS and BTS hardware configuration
+ * - buffer overflow handling (to be done)
  * - buffer access
  *
- * It assumes:
- * - get_task_struct on all parameter tasks
- * - current is allowed to trace parameter tasks
+ * It does not do:
+ * - security checking (is the caller allowed to trace the task)
+ * - buffer allocation (memory accounting)
  *
  *
  * Copyright (C) 2007-2008 Intel Corporation.
@@ -28,22 +27,69 @@
 #include <linux/slab.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
+#include <linux/kernel.h>
 
 
 /*
  * The configuration for a particular DS hardware implementation.
  */
 struct ds_configuration {
-	/* the size of the DS structure in bytes */
-	unsigned char  sizeof_ds;
-	/* the size of one pointer-typed field in the DS structure in bytes;
-	   this covers the first 8 fields related to buffer management. */
+	/* the name of the configuration */
+	const char *name;
+	/* the size of one pointer-typed field in the DS structure and
+	   in the BTS and PEBS buffers in bytes;
+	   this covers the first 8 DS fields related to buffer management. */
 	unsigned char  sizeof_field;
 	/* the size of a BTS/PEBS record in bytes */
 	unsigned char  sizeof_rec[2];
+	/* a series of bit-masks to control various features indexed
+	 * by enum ds_feature */
+	unsigned long ctl[dsf_ctl_max];
 };
-static struct ds_configuration ds_cfg;
+static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array);
 
+#define ds_cfg per_cpu(ds_cfg_array, smp_processor_id())
+
+#define MAX_SIZEOF_DS (12 * 8)	/* maximal size of a DS configuration */
+#define MAX_SIZEOF_BTS (3 * 8)	/* maximal size of a BTS record */
+#define DS_ALIGNMENT (1 << 3)	/* BTS and PEBS buffer alignment */
+
+#define BTS_CONTROL \
+ (ds_cfg.ctl[dsf_bts] | ds_cfg.ctl[dsf_bts_kernel] | ds_cfg.ctl[dsf_bts_user] |\
+  ds_cfg.ctl[dsf_bts_overflow])
+
+
+/*
+ * A BTS or PEBS tracer.
+ *
+ * This holds the configuration of the tracer and serves as a handle
+ * to identify tracers.
+ */
+struct ds_tracer {
+	/* the DS context (partially) owned by this tracer */
+	struct ds_context *context;
+	/* the buffer provided on ds_request() and its size in bytes */
+	void *buffer;
+	size_t size;
+};
+
+struct bts_tracer {
+	/* the common DS part */
+	struct ds_tracer ds;
+	/* the trace including the DS configuration */
+	struct bts_trace trace;
+	/* buffer overflow notification function */
+	bts_ovfl_callback_t ovfl;
+};
+
+struct pebs_tracer {
+	/* the common DS part */
+	struct ds_tracer ds;
+	/* the trace including the DS configuration */
+	struct pebs_trace trace;
+	/* buffer overflow notification function */
+	pebs_ovfl_callback_t ovfl;
+};
 
 /*
  * Debug Store (DS) save area configuration (see Intel64 and IA32
@@ -109,32 +155,9 @@ static inline void ds_set(unsigned char *base, enum ds_qualifier qual,
 
 
 /*
- * Locking is done only for allocating BTS or PEBS resources and for
- * guarding context and buffer memory allocation.
- *
- * Most functions require the current task to own the ds context part
- * they are going to access. All the locking is done when validating
- * access to the context.
+ * Locking is done only for allocating BTS or PEBS resources.
  */
-static spinlock_t ds_lock = __SPIN_LOCK_UNLOCKED(ds_lock);
-
-/*
- * Validate that the current task is allowed to access the BTS/PEBS
- * buffer of the parameter task.
- *
- * Returns 0, if access is granted; -Eerrno, otherwise.
- */
-static inline int ds_validate_access(struct ds_context *context,
-				     enum ds_qualifier qual)
-{
-	if (!context)
-		return -EPERM;
-
-	if (context->owner[qual] == current)
-		return 0;
-
-	return -EPERM;
-}
+static DEFINE_SPINLOCK(ds_lock);
 
 
 /*
@@ -150,27 +173,32 @@ static inline int ds_validate_access(struct ds_context *context,
  *   >0  number of per-thread tracers
  *   <0  number of per-cpu tracers
  *
- * The below functions to get and put tracers and to check the
- * allocation type require the ds_lock to be held by the caller.
- *
  * Tracers essentially gives the number of ds contexts for a certain
  * type of allocation.
  */
-static long tracers;
+static atomic_t tracers = ATOMIC_INIT(0);
 
 static inline void get_tracer(struct task_struct *task)
 {
-	tracers += (task ? 1 : -1);
+	if (task)
+		atomic_inc(&tracers);
+	else
+		atomic_dec(&tracers);
 }
 
 static inline void put_tracer(struct task_struct *task)
 {
-	tracers -= (task ? 1 : -1);
+	if (task)
+		atomic_dec(&tracers);
+	else
+		atomic_inc(&tracers);
 }
 
 static inline int check_tracer(struct task_struct *task)
 {
-	return (task ? (tracers >= 0) : (tracers <= 0));
+	return task ?
+		(atomic_read(&tracers) >= 0) :
+		(atomic_read(&tracers) <= 0);
 }
 
 
@@ -183,99 +211,70 @@ static inline int check_tracer(struct task_struct *task)
  *
  * Contexts are use-counted. They are allocated on first access and
  * deallocated when the last user puts the context.
- *
- * We distinguish between an allocating and a non-allocating get of a
- * context:
- * - the allocating get is used for requesting BTS/PEBS resources. It
- *   requires the caller to hold the global ds_lock.
- * - the non-allocating get is used for all other cases. A
- *   non-existing context indicates an error. It acquires and releases
- *   the ds_lock itself for obtaining the context.
- *
- * A context and its DS configuration are allocated and deallocated
- * together. A context always has a DS configuration of the
- * appropriate size.
  */
-static DEFINE_PER_CPU(struct ds_context *, system_context);
-
-#define this_system_context per_cpu(system_context, smp_processor_id())
-
-/*
- * Returns the pointer to the parameter task's context or to the
- * system-wide context, if task is NULL.
- *
- * Increases the use count of the returned context, if not NULL.
- */
-static inline struct ds_context *ds_get_context(struct task_struct *task)
-{
-	struct ds_context *context;
-	unsigned long irq;
+struct ds_context {
+	/* pointer to the DS configuration; goes into MSR_IA32_DS_AREA */
+	unsigned char ds[MAX_SIZEOF_DS];
+	/* the owner of the BTS and PEBS configuration, respectively */
+	struct bts_tracer *bts_master;
+	struct pebs_tracer *pebs_master;
+	/* use count */
+	unsigned long count;
+	/* a pointer to the context location inside the thread_struct
+	 * or the per_cpu context array */
+	struct ds_context **this;
+	/* a pointer to the task owning this context, or NULL, if the
+	 * context is owned by a cpu */
+	struct task_struct *task;
+};
 
-	spin_lock_irqsave(&ds_lock, irq);
+static DEFINE_PER_CPU(struct ds_context *, system_context_array);
 
-	context = (task ? task->thread.ds_ctx : this_system_context);
-	if (context)
-		context->count++;
+#define system_context per_cpu(system_context_array, smp_processor_id())
 
-	spin_unlock_irqrestore(&ds_lock, irq);
-
-	return context;
-}
 
-/*
- * Same as ds_get_context, but allocates the context and it's DS
- * structure, if necessary; returns NULL; if out of memory.
- */
-static inline struct ds_context *ds_alloc_context(struct task_struct *task)
+static inline struct ds_context *ds_get_context(struct task_struct *task)
 {
 	struct ds_context **p_context =
-		(task ? &task->thread.ds_ctx : &this_system_context);
-	struct ds_context *context = *p_context;
+		(task ? &task->thread.ds_ctx : &system_context);
+	struct ds_context *context = NULL;
+	struct ds_context *new_context = NULL;
 	unsigned long irq;
 
-	if (!context) {
-		context = kzalloc(sizeof(*context), GFP_KERNEL);
-		if (!context)
-			return NULL;
-
-		context->ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL);
-		if (!context->ds) {
-			kfree(context);
-			return NULL;
-		}
+	/* Chances are small that we already have a context. */
+	new_context = kzalloc(sizeof(*new_context), GFP_KERNEL);
+	if (!new_context)
+		return NULL;
 
-		spin_lock_irqsave(&ds_lock, irq);
+	spin_lock_irqsave(&ds_lock, irq);
 
-		if (*p_context) {
-			kfree(context->ds);
-			kfree(context);
+	context = *p_context;
+	if (!context) {
+		context = new_context;
 
-			context = *p_context;
-		} else {
-			*p_context = context;
+		context->this = p_context;
+		context->task = task;
+		context->count = 0;
 
-			context->this = p_context;
-			context->task = task;
+		if (task)
+			set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
 
-			if (task)
-				set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
+		if (!task || (task == current))
+			wrmsrl(MSR_IA32_DS_AREA, (unsigned long)context->ds);
 
-			if (!task || (task == current))
-				wrmsrl(MSR_IA32_DS_AREA,
-				       (unsigned long)context->ds);
-		}
-		spin_unlock_irqrestore(&ds_lock, irq);
+		*p_context = context;
 	}
 
 	context->count++;
 
+	spin_unlock_irqrestore(&ds_lock, irq);
+
+	if (context != new_context)
+		kfree(new_context);
+
 	return context;
 }
 
-/*
- * Decreases the use count of the parameter context, if not NULL.
- * Deallocates the context, if the use count reaches zero.
- */
 static inline void ds_put_context(struct ds_context *context)
 {
 	unsigned long irq;
@@ -285,8 +284,10 @@ static inline void ds_put_context(struct ds_context *context)
 
 	spin_lock_irqsave(&ds_lock, irq);
 
-	if (--context->count)
-		goto out;
+	if (--context->count) {
+		spin_unlock_irqrestore(&ds_lock, irq);
+		return;
+	}
 
 	*(context->this) = NULL;
 
@@ -296,135 +297,263 @@ static inline void ds_put_context(struct ds_context *context)
 	if (!context->task || (context->task == current))
 		wrmsrl(MSR_IA32_DS_AREA, 0);
 
-	put_tracer(context->task);
+	spin_unlock_irqrestore(&ds_lock, irq);
 
-	/* free any leftover buffers from tracers that did not
-	 * deallocate them properly. */
-	kfree(context->buffer[ds_bts]);
-	kfree(context->buffer[ds_pebs]);
-	kfree(context->ds);
 	kfree(context);
- out:
-	spin_unlock_irqrestore(&ds_lock, irq);
 }
 
 
 /*
- * Handle a buffer overflow
+ * Call the tracer's callback on a buffer overflow.
  *
- * task: the task whose buffers are overflowing;
- *       NULL for a buffer overflow on the current cpu
  * context: the ds context
  * qual: the buffer type
  */
-static void ds_overflow(struct task_struct *task, struct ds_context *context,
-			enum ds_qualifier qual)
+static void ds_overflow(struct ds_context *context, enum ds_qualifier qual)
 {
-	if (!context)
-		return;
-
-	if (context->callback[qual])
-		(*context->callback[qual])(task);
-
-	/* todo: do some more overflow handling */
+	switch (qual) {
+	case ds_bts:
+		if (context->bts_master &&
+		    context->bts_master->ovfl)
+			context->bts_master->ovfl(context->bts_master);
+		break;
+	case ds_pebs:
+		if (context->pebs_master &&
+		    context->pebs_master->ovfl)
+			context->pebs_master->ovfl(context->pebs_master);
+		break;
+	}
 }
 
 
 /*
- * Allocate a non-pageable buffer of the parameter size.
- * Checks the memory and the locked memory rlimit.
+ * Write raw data into the BTS or PEBS buffer.
  *
- * Returns the buffer, if successful;
- *         NULL, if out of memory or rlimit exceeded.
+ * The remainder of any partially written record is zeroed out.
  *
- * size: the requested buffer size in bytes
- * pages (out): if not NULL, contains the number of pages reserved
+ * context: the DS context
+ * qual: the buffer type
+ * record: the data to write
+ * size: the size of the data
  */
-static inline void *ds_allocate_buffer(size_t size, unsigned int *pages)
+static int ds_write(struct ds_context *context, enum ds_qualifier qual,
+		    const void *record, size_t size)
 {
-	unsigned long rlim, vm, pgsz;
-	void *buffer;
+	int bytes_written = 0;
 
-	pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
+	if (!record)
+		return -EINVAL;
 
-	rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
-	vm   = current->mm->total_vm  + pgsz;
-	if (rlim < vm)
-		return NULL;
+	while (size) {
+		unsigned long base, index, end, write_end, int_th;
+		unsigned long write_size, adj_write_size;
 
-	rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
-	vm   = current->mm->locked_vm  + pgsz;
-	if (rlim < vm)
-		return NULL;
+		/*
+		 * write as much as possible without producing an
+		 * overflow interrupt.
+		 *
+		 * interrupt_threshold must either be
+		 * - bigger than absolute_maximum or
+		 * - point to a record between buffer_base and absolute_maximum
+		 *
+		 * index points to a valid record.
+		 */
+		base   = ds_get(context->ds, qual, ds_buffer_base);
+		index  = ds_get(context->ds, qual, ds_index);
+		end    = ds_get(context->ds, qual, ds_absolute_maximum);
+		int_th = ds_get(context->ds, qual, ds_interrupt_threshold);
 
-	buffer = kzalloc(size, GFP_KERNEL);
-	if (!buffer)
-		return NULL;
+		write_end = min(end, int_th);
 
-	current->mm->total_vm  += pgsz;
-	current->mm->locked_vm += pgsz;
+		/* if we are already beyond the interrupt threshold,
+		 * we fill the entire buffer */
+		if (write_end <= index)
+			write_end = end;
 
-	if (pages)
-		*pages = pgsz;
+		if (write_end <= index)
+			break;
+
+		write_size = min((unsigned long) size, write_end - index);
+		memcpy((void *)index, record, write_size);
+
+		record = (const char *)record + write_size;
+		size -= write_size;
+		bytes_written += write_size;
+
+		adj_write_size = write_size / ds_cfg.sizeof_rec[qual];
+		adj_write_size *= ds_cfg.sizeof_rec[qual];
 
-	return buffer;
+		/* zero out trailing bytes */
+		memset((char *)index + write_size, 0,
+		       adj_write_size - write_size);
+		index += adj_write_size;
+
+		if (index >= end)
+			index = base;
+		ds_set(context->ds, qual, ds_index, index);
+
+		if (index >= int_th)
+			ds_overflow(context, qual);
+	}
+
+	return bytes_written;
 }
 
-static int ds_request(struct task_struct *task, void *base, size_t size,
-		      ds_ovfl_callback_t ovfl, enum ds_qualifier qual)
+
+/*
+ * Branch Trace Store (BTS) uses the following format. Different
+ * architectures vary in the size of those fields.
+ * - source linear address
+ * - destination linear address
+ * - flags
+ *
+ * Later architectures use 64bit pointers throughout, whereas earlier
+ * architectures use 32bit pointers in 32bit mode.
+ *
+ * We compute the base address for the first 8 fields based on:
+ * - the field size stored in the DS configuration
+ * - the relative field position
+ *
+ * In order to store additional information in the BTS buffer, we use
+ * a special source address to indicate that the record requires
+ * special interpretation.
+ *
+ * Netburst indicated via a bit in the flags field whether the branch
+ * was predicted; this is ignored.
+ *
+ * We use two levels of abstraction:
+ * - the raw data level defined here
+ * - an arch-independent level defined in ds.h
+ */
+
+enum bts_field {
+	bts_from,
+	bts_to,
+	bts_flags,
+
+	bts_qual = bts_from,
+	bts_jiffies = bts_to,
+	bts_pid = bts_flags,
+
+	bts_qual_mask = (bts_qual_max - 1),
+	bts_escape = ((unsigned long)-1 & ~bts_qual_mask)
+};
+
+static inline unsigned long bts_get(const char *base, enum bts_field field)
 {
-	struct ds_context *context;
-	unsigned long buffer, adj;
-	const unsigned long alignment = (1 << 3);
-	unsigned long irq;
-	int error = 0;
+	base += (ds_cfg.sizeof_field * field);
+	return *(unsigned long *)base;
+}
 
-	if (!ds_cfg.sizeof_ds)
-		return -EOPNOTSUPP;
+static inline void bts_set(char *base, enum bts_field field, unsigned long val)
+{
+	base += (ds_cfg.sizeof_field * field);;
+	(*(unsigned long *)base) = val;
+}
 
-	/* we require some space to do alignment adjustments below */
-	if (size < (alignment + ds_cfg.sizeof_rec[qual]))
+
+/*
+ * The raw BTS data is architecture dependent.
+ *
+ * For higher-level users, we give an arch-independent view.
+ * - ds.h defines struct bts_struct
+ * - bts_read translates one raw bts record into a bts_struct
+ * - bts_write translates one bts_struct into the raw format and
+ *   writes it into the top of the parameter tracer's buffer.
+ *
+ * return: bytes read/written on success; -Eerrno, otherwise
+ */
+static int bts_read(struct bts_tracer *tracer, const void *at,
+		    struct bts_struct *out)
+{
+	if (!tracer)
 		return -EINVAL;
 
-	/* buffer overflow notification is not yet implemented */
-	if (ovfl)
-		return -EOPNOTSUPP;
+	if (at < tracer->trace.ds.begin)
+		return -EINVAL;
 
+	if (tracer->trace.ds.end < (at + tracer->trace.ds.size))
+		return -EINVAL;
 
-	context = ds_alloc_context(task);
-	if (!context)
-		return -ENOMEM;
+	memset(out, 0, sizeof(*out));
+	if ((bts_get(at, bts_qual) & ~bts_qual_mask) == bts_escape) {
+		out->qualifier = (bts_get(at, bts_qual) & bts_qual_mask);
+		out->variant.timestamp.jiffies = bts_get(at, bts_jiffies);
+		out->variant.timestamp.pid = bts_get(at, bts_pid);
+	} else {
+		out->qualifier = bts_branch;
+		out->variant.lbr.from = bts_get(at, bts_from);
+		out->variant.lbr.to   = bts_get(at, bts_to);
+
+		if (!out->variant.lbr.from && !out->variant.lbr.to)
+			out->qualifier = bts_invalid;
+	}
 
-	spin_lock_irqsave(&ds_lock, irq);
+	return ds_cfg.sizeof_rec[ds_bts];
+}
 
-	error = -EPERM;
-	if (!check_tracer(task))
-		goto out_unlock;
+static int bts_write(struct bts_tracer *tracer, const struct bts_struct *in)
+{
+	unsigned char raw[MAX_SIZEOF_BTS];
 
-	get_tracer(task);
+	if (!tracer)
+		return -EINVAL;
 
-	error = -EALREADY;
-	if (context->owner[qual] == current)
-		goto out_put_tracer;
-	error = -EPERM;
-	if (context->owner[qual] != NULL)
-		goto out_put_tracer;
-	context->owner[qual] = current;
+	if (MAX_SIZEOF_BTS < ds_cfg.sizeof_rec[ds_bts])
+		return -EOVERFLOW;
 
-	spin_unlock_irqrestore(&ds_lock, irq);
+	switch (in->qualifier) {
+	case bts_invalid:
+		bts_set(raw, bts_from, 0);
+		bts_set(raw, bts_to, 0);
+		bts_set(raw, bts_flags, 0);
+		break;
+	case bts_branch:
+		bts_set(raw, bts_from, in->variant.lbr.from);
+		bts_set(raw, bts_to,   in->variant.lbr.to);
+		bts_set(raw, bts_flags, 0);
+		break;
+	case bts_task_arrives:
+	case bts_task_departs:
+		bts_set(raw, bts_qual, (bts_escape | in->qualifier));
+		bts_set(raw, bts_jiffies, in->variant.timestamp.jiffies);
+		bts_set(raw, bts_pid, in->variant.timestamp.pid);
+		break;
+	default:
+		return -EINVAL;
+	}
 
+	return ds_write(tracer->ds.context, ds_bts, raw,
+			ds_cfg.sizeof_rec[ds_bts]);
+}
 
-	error = -ENOMEM;
-	if (!base) {
-		base = ds_allocate_buffer(size, &context->pages[qual]);
-		if (!base)
-			goto out_release;
 
-		context->buffer[qual]   = base;
-	}
-	error = 0;
+static void ds_write_config(struct ds_context *context,
+			    struct ds_trace *cfg, enum ds_qualifier qual)
+{
+	unsigned char *ds = context->ds;
 
-	context->callback[qual] = ovfl;
+	ds_set(ds, qual, ds_buffer_base, (unsigned long)cfg->begin);
+	ds_set(ds, qual, ds_index, (unsigned long)cfg->top);
+	ds_set(ds, qual, ds_absolute_maximum, (unsigned long)cfg->end);
+	ds_set(ds, qual, ds_interrupt_threshold, (unsigned long)cfg->ith);
+}
+
+static void ds_read_config(struct ds_context *context,
+			   struct ds_trace *cfg, enum ds_qualifier qual)
+{
+	unsigned char *ds = context->ds;
+
+	cfg->begin = (void *)ds_get(ds, qual, ds_buffer_base);
+	cfg->top = (void *)ds_get(ds, qual, ds_index);
+	cfg->end = (void *)ds_get(ds, qual, ds_absolute_maximum);
+	cfg->ith = (void *)ds_get(ds, qual, ds_interrupt_threshold);
+}
+
+static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
+			     void *base, size_t size, size_t ith,
+			     unsigned int flags) {
+	unsigned long buffer, adj;
 
 	/* adjust the buffer address and size to meet alignment
 	 * constraints:
@@ -436,410 +565,383 @@ static int ds_request(struct task_struct *task, void *base, size_t size,
 	 */
 	buffer = (unsigned long)base;
 
-	adj = ALIGN(buffer, alignment) - buffer;
+	adj = ALIGN(buffer, DS_ALIGNMENT) - buffer;
 	buffer += adj;
 	size   -= adj;
 
-	size /= ds_cfg.sizeof_rec[qual];
-	size *= ds_cfg.sizeof_rec[qual];
-
-	ds_set(context->ds, qual, ds_buffer_base, buffer);
-	ds_set(context->ds, qual, ds_index, buffer);
-	ds_set(context->ds, qual, ds_absolute_maximum, buffer + size);
-
-	if (ovfl) {
-		/* todo: select a suitable interrupt threshold */
-	} else
-		ds_set(context->ds, qual,
-		       ds_interrupt_threshold, buffer + size + 1);
+	trace->n = size / ds_cfg.sizeof_rec[qual];
+	trace->size = ds_cfg.sizeof_rec[qual];
 
-	/* we keep the context until ds_release */
-	return error;
-
- out_release:
-	context->owner[qual] = NULL;
-	ds_put_context(context);
-	put_tracer(task);
-	return error;
+	size = (trace->n * trace->size);
 
- out_put_tracer:
-	spin_unlock_irqrestore(&ds_lock, irq);
-	ds_put_context(context);
-	put_tracer(task);
-	return error;
+	trace->begin = (void *)buffer;
+	trace->top = trace->begin;
+	trace->end = (void *)(buffer + size);
+	/* The value for 'no threshold' is -1, which will set the
+	 * threshold outside of the buffer, just like we want it.
+	 */
+	trace->ith = (void *)(buffer + size - ith);
 
- out_unlock:
-	spin_unlock_irqrestore(&ds_lock, irq);
-	ds_put_context(context);
-	return error;
+	trace->flags = flags;
 }
 
-int ds_request_bts(struct task_struct *task, void *base, size_t size,
-		   ds_ovfl_callback_t ovfl)
-{
-	return ds_request(task, base, size, ovfl, ds_bts);
-}
 
-int ds_request_pebs(struct task_struct *task, void *base, size_t size,
-		    ds_ovfl_callback_t ovfl)
-{
-	return ds_request(task, base, size, ovfl, ds_pebs);
-}
-
-static int ds_release(struct task_struct *task, enum ds_qualifier qual)
+static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
+		      enum ds_qualifier qual, struct task_struct *task,
+		      void *base, size_t size, size_t th, unsigned int flags)
 {
 	struct ds_context *context;
 	int error;
 
-	context = ds_get_context(task);
-	error = ds_validate_access(context, qual);
-	if (error < 0)
+	error = -EINVAL;
+	if (!base)
 		goto out;
 
-	kfree(context->buffer[qual]);
-	context->buffer[qual] = NULL;
-
-	current->mm->total_vm  -= context->pages[qual];
-	current->mm->locked_vm -= context->pages[qual];
-	context->pages[qual] = 0;
-	context->owner[qual] = NULL;
-
-	/*
-	 * we put the context twice:
-	 *   once for the ds_get_context
-	 *   once for the corresponding ds_request
-	 */
-	ds_put_context(context);
- out:
-	d