[patch 00/20] x86, ptrace, bts, hw-branch-tracer: fixes and cleanups

[patch 00/20] x86, ptrace, bts, hw-branch-tracer: fixes and cleanups

[markus.t.metzger]

-- 
Patches 1-5 fix races with context switching code when the branch traced task
is currently running.

In the worst case, this might cause context switch code to access freed
memory or the tracing hardware to continue tracing into a freed buffer.
Both might crash the kernel.

The first 4 patches apply to .29 using the below preparation patch.


The remaining patches fix bugs in the context of per-cpu tracing (i.e.
hw-branch-tracer) and pebs, add more selftest code, and do some cleanups.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---

Index: linux-2.6.29/arch/x86/kernel/ds.c
===================================================================
--- linux-2.6.29.orig/arch/x86/kernel/ds.c	2009-04-01 17:39:55.000000000 +0200
+++ linux-2.6.29/arch/x86/kernel/ds.c	2009-04-01 17:40:42.000000000 +0200
@@ -78,8 +78,8 @@ struct bts_tracer {
 	struct ds_tracer ds;
 	/* the trace including the DS configuration */
 	struct bts_trace trace;
-	/* buffer overflow notification function */
-	bts_ovfl_callback_t ovfl;
+	/* Buffer overflow notification function: */
+	bts_ovfl_callback_t	ovfl;
 };
 
 struct pebs_tracer {
Index: linux-2.6.29/arch/x86/kernel/ptrace.c
===================================================================
--- linux-2.6.29.orig/arch/x86/kernel/ptrace.c	2009-04-01 17:39:55.000000000 +0200
+++ linux-2.6.29/arch/x86/kernel/ptrace.c	2009-04-01 17:40:42.000000000 +0200
@@ -21,6 +21,7 @@
 #include <linux/audit.h>
 #include <linux/seccomp.h>
 #include <linux/signal.h>
+#include <linux/ftrace.h>
 
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 01/20] x86, bts: fix race when bts tracer is removed

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.tracing_flags.patch --]
[-- Type: text/plain, Size: 3670 bytes --]

When the bts tracer is removed while the traced task is running,
the write to clear the bts tracer pointer races with context switch code.

Read the tracer once during a context switch.


When a new tracer is installed, the bts tracer is set in the ds context
before the tracer is initialized in order to claim the context for that
tracer.
This may result in write accesses using an uninitialized trace configuration
when scheduling timestamps have been requested.

Store active tracing flags separately and only set active flags after
the tracing configuration has been initialized.


Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds.c |   58 	37 +	21 -	0 !
 1 file changed, 37 insertions(+), 21 deletions(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -89,6 +89,9 @@ struct bts_tracer {
 
 	/* Buffer overflow notification function: */
 	bts_ovfl_callback_t	ovfl;
+
+	/* Active flags affecting trace collection. */
+	unsigned int		flags;
 };
 
 struct pebs_tracer {
@@ -799,6 +802,8 @@ void ds_suspend_bts(struct bts_tracer *t
 	if (!tracer)
 		return;
 
+	tracer->flags = 0;
+
 	task = tracer->ds.context->task;
 
 	if (!task || (task == current))
@@ -820,6 +825,8 @@ void ds_resume_bts(struct bts_tracer *tr
 	if (!tracer)
 		return;
 
+	tracer->flags = tracer->trace.ds.flags;
+
 	task = tracer->ds.context->task;
 
 	control = ds_cfg.ctl[dsf_bts];
@@ -1037,43 +1044,52 @@ void __cpuinit ds_init_intel(struct cpui
 	}
 }
 
+static inline void ds_take_timestamp(struct ds_context *context,
+				     enum bts_qualifier qualifier,
+				     struct task_struct *task)
+{
+	struct bts_tracer *tracer = context->bts_master;
+	struct bts_struct ts;
+
+	/* Prevent compilers from reading the tracer pointer twice. */
+	barrier();
+
+	if (!tracer || !(tracer->flags & BTS_TIMESTAMPS))
+		return;
+
+	memset(&ts, 0, sizeof(ts));
+	ts.qualifier			= qualifier;
+	ts.variant.timestamp.jiffies	= jiffies_64;
+	ts.variant.timestamp.pid	= task->pid;
+
+	bts_write(tracer, &ts);
+}
+
 /*
  * Change the DS configuration from tracing prev to tracing next.
  */
 void ds_switch_to(struct task_struct *prev, struct task_struct *next)
 {
-	struct ds_context *prev_ctx = prev->thread.ds_ctx;
-	struct ds_context *next_ctx = next->thread.ds_ctx;
+	struct ds_context *prev_ctx	= prev->thread.ds_ctx;
+	struct ds_context *next_ctx	= next->thread.ds_ctx;
+	unsigned long debugctlmsr	= next->thread.debugctlmsr;
+
+	/* Make sure all data is read before we start. */
+	barrier();
 
 	if (prev_ctx) {
 		update_debugctlmsr(0);
 
-		if (prev_ctx->bts_master &&
-		    (prev_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
-			struct bts_struct ts = {
-				.qualifier = bts_task_departs,
-				.variant.timestamp.jiffies = jiffies_64,
-				.variant.timestamp.pid = prev->pid
-			};
-			bts_write(prev_ctx->bts_master, &ts);
-		}
+		ds_take_timestamp(prev_ctx, bts_task_departs, prev);
 	}
 
 	if (next_ctx) {
-		if (next_ctx->bts_master &&
-		    (next_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
-			struct bts_struct ts = {
-				.qualifier = bts_task_arrives,
-				.variant.timestamp.jiffies = jiffies_64,
-				.variant.timestamp.pid = next->pid
-			};
-			bts_write(next_ctx->bts_master, &ts);
-		}
+		ds_take_timestamp(next_ctx, bts_task_arrives, next);
 
 		wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds);
 	}
 
-	update_debugctlmsr(next->thread.debugctlmsr);
+	update_debugctlmsr(debugctlmsr);
 }
 
 void ds_copy_thread(struct task_struct *tsk, struct task_struct *father)

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 02/20] sched: add wait_task_context_switch() fucntion to sched.h

[markus.t.metzger]

[-- Attachment #1: tip.master.sched.task_running.patch --]
[-- Type: text/plain, Size: 2240 bytes --]

Add a function to wait until some other task has been switched out at least
once.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>

---
 include/linux/sched.h |    2 	2 +	0 -	0 !
 kernel/sched.c        |   43 	43 +	0 -	0 !
 2 files changed, 45 insertions(+)

Index: b/include/linux/sched.h
===================================================================
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1988,8 +1988,10 @@ extern void set_task_comm(struct task_st
 extern char *get_task_comm(char *to, struct task_struct *tsk);
 
 #ifdef CONFIG_SMP
+extern void wait_task_context_switch(struct task_struct *p);
 extern unsigned long wait_task_inactive(struct task_struct *, long match_state);
 #else
+static inline void wait_task_context_switch(struct task_struct *p) {}
 static inline unsigned long wait_task_inactive(struct task_struct *p,
 					       long match_state)
 {
Index: b/kernel/sched.c
===================================================================
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -1933,6 +1933,49 @@ migrate_task(struct task_struct *p, int 
 }
 
 /*
+ * wait_task_context_switch -	wait for a thread to complete at least one
+ *				context switch.
+ *
+ * @p must not be current.
+ */
+void wait_task_context_switch(struct task_struct *p)
+{
+	unsigned long nvcsw, nivcsw, flags;
+	int running;
+	struct rq *rq;
+
+	nvcsw	= p->nvcsw;
+	nivcsw	= p->nivcsw;
+	for (;;) {
+		/*
+		 * The runqueue is assigned before the actual context
+		 * switch. We need to take the runqueue lock.
+		 *
+		 * We could check initially without the lock but it is
+		 * very likely that we need to take the lock in every
+		 * iteration.
+		 */
+		rq = task_rq_lock(p, &flags);
+		running = task_running(rq, p);
+		task_rq_unlock(rq, &flags);
+
+		if (likely(!running))
+			break;
+		/*
+		 * The switch count is incremented before the actual
+		 * context switch. We thus wait for two switches to be
+		 * sure at least one completed.
+		 */
+		if ((p->nvcsw - nvcsw) > 1)
+			break;
+		if ((p->nivcsw - nivcsw) > 1)
+			break;
+
+		cpu_relax();
+	}
+}
+
+/*
  * wait_task_inactive - wait for a thread to unschedule.
  *
  * If @match_state is nonzero, it's the @p->state value just checked and

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 03/20] x86, ptrace, bts: defer branch trace stopping

[markus.t.metzger]

[-- Attachment #1: tip.master.ptrace_bts.report_exit.patch --]
[-- Type: text/plain, Size: 14218 bytes --]

When a ptraced task is unlinked, we need to stop branch tracing for
that task.

Since the unlink is called with interrupts disabled, and we need interrupts
enabled to stop branch tracing, we defer the work.

Collect all branch tracing related stuff in a branch tracing context.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/include/asm/processor.h |    4 	0 +	4 -	0 !
 arch/x86/kernel/ptrace.c         |  254 	169 +	85 -	0 !
 include/linux/mm.h               |    3 	2 +	1 -	0 !
 include/linux/sched.h            |    9 	2 +	7 -	0 !
 mm/mlock.c                       |   13 	6 +	7 -	0 !
 5 files changed, 179 insertions(+), 104 deletions(-)

Index: b/arch/x86/kernel/ptrace.c
===================================================================
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -22,6 +22,7 @@
 #include <linux/seccomp.h>
 #include <linux/signal.h>
 #include <linux/ftrace.h>
+#include <linux/workqueue.h>
 
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
@@ -577,17 +578,119 @@ static int ioperm_get(struct task_struct
 }
 
 #ifdef CONFIG_X86_PTRACE_BTS
+/*
+ * A branch trace store context.
+ *
+ * Contexts may only be installed by ptrace_bts_config() and only for
+ * ptraced tasks.
+ *
+ * Contexts are destroyed when the tracee is detached from the tracer.
+ * The actual destruction work requires interrupts enabled, so the
+ * work is deferred and will be scheduled during __ptrace_unlink().
+ *
+ * Contexts hold an additional task_struct reference on the traced
+ * task, as well as a reference on the tracer's mm.
+ *
+ * Ptrace already holds a task_struct for the duration of ptrace operations,
+ * but since destruction is deferred, it may be executed after both
+ * tracer and tracee exited.
+ */
+struct bts_context {
+	/* The branch trace handle. */
+	struct bts_tracer	*tracer;
+
+	/* The buffer used to store the branch trace and its size. */
+	void			*buffer;
+	unsigned int		size;
+
+	/* The mm that paid for the above buffer. */
+	struct mm_struct	*mm;
+
+	/* The task this context belongs to. */
+	struct task_struct	*task;
+
+	/* The signal to send on a bts buffer overflow. */
+	unsigned int		bts_ovfl_signal;
+
+	/* The work struct to destroy a context. */
+	struct work_struct	work;
+};
+
+static inline void alloc_bts_buffer(struct bts_context *context,
+				    unsigned int size)
+{
+	void *buffer;
+
+	buffer = alloc_locked_buffer(size);
+	if (buffer) {
+		context->buffer = buffer;
+		context->size = size;
+		context->mm = get_task_mm(current);
+	}
+}
+
+static inline void free_bts_buffer(struct bts_context *context)
+{
+	if (!context->buffer)
+		return;
+
+	kfree(context->buffer);
+	context->buffer = NULL;
+
+	refund_locked_buffer_memory(context->mm, context->size);
+	context->size = 0;
+
+	mmput(context->mm);
+	context->mm = NULL;
+}
+
+static void free_bts_context_work(struct work_struct *w)
+{
+	struct bts_context *context;
+
+	context = container_of(w, struct bts_context, work);
+
+	ds_release_bts(context->tracer);
+	put_task_struct(context->task);
+	free_bts_buffer(context);
+	kfree(context);
+}
+
+static inline void free_bts_context(struct bts_context *context)
+{
+	INIT_WORK(&context->work, free_bts_context_work);
+	schedule_work(&context->work);
+}
+
+static inline struct bts_context *alloc_bts_context(struct task_struct *task)
+{
+	struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
+	if (context) {
+		context->task = task;
+		task->bts = context;
+
+		get_task_struct(task);
+	}
+
+	return context;
+}
+
 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 				  struct bts_struct __user *out)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 	struct bts_struct bts;
 	const unsigned char *at;
 	int error;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	at = trace->ds.top - ((index + 1) * trace->ds.size);
 	if ((void *)at < trace->ds.begin)
@@ -596,7 +699,7 @@ static int ptrace_bts_read_record(struct
 	if (!trace->read)
 		return -EOPNOTSUPP;
 
-	error = trace->read(child->bts, at, &bts);
+	error = trace->read(context->tracer, at, &bts);
 	if (error < 0)
 		return error;
 
@@ -610,13 +713,18 @@ static int ptrace_bts_drain(struct task_
 			    long size,
 			    struct bts_struct __user *out)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 	const unsigned char *at;
 	int error, drained = 0;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	if (!trace->read)
 		return -EOPNOTSUPP;
@@ -627,9 +735,8 @@ static int ptrace_bts_drain(struct task_
 	for (at = trace->ds.begin; (void *)at < trace->ds.top;
 	     out++, drained++, at += trace->ds.size) {
 		struct bts_struct bts;
-		int error;
 
-		error = trace->read(child->bts, at, &bts);
+		error = trace->read(context->tracer, at, &bts);
 		if (error < 0)
 			return error;
 
@@ -639,35 +746,18 @@ static int ptrace_bts_drain(struct task_
 
 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 
-	error = ds_reset_bts(child->bts);
+	error = ds_reset_bts(context->tracer);
 	if (error < 0)
 		return error;
 
 	return drained;
 }
 
-static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size)
-{
-	child->bts_buffer = alloc_locked_buffer(size);
-	if (!child->bts_buffer)
-		return -ENOMEM;
-
-	child->bts_size = size;
-
-	return 0;
-}
-
-static void ptrace_bts_free_buffer(struct task_struct *child)
-{
-	free_locked_buffer(child->bts_buffer, child->bts_size);
-	child->bts_buffer = NULL;
-	child->bts_size = 0;
-}
-
 static int ptrace_bts_config(struct task_struct *child,
 			     long cfg_size,
 			     const struct ptrace_bts_config __user *ucfg)
 {
+	struct bts_context *context;
 	struct ptrace_bts_config cfg;
 	unsigned int flags = 0;
 
@@ -677,10 +767,11 @@ static int ptrace_bts_config(struct task
 	if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 		return -EFAULT;
 
-	if (child->bts) {
-		ds_release_bts(child->bts);
-		child->bts = NULL;
-	}
+	context = child->bts;
+	if (!context)
+		context = alloc_bts_context(child);
+	if (!context)
+		return -ENOMEM;
 
 	if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 		if (!cfg.signal)
@@ -688,18 +779,20 @@ static int ptrace_bts_config(struct task
 
 		return -EOPNOTSUPP;
 
-		child->thread.bts_ovfl_signal = cfg.signal;
+		context->bts_ovfl_signal = cfg.signal;
 	}
 
-	if ((cfg.flags & PTRACE_BTS_O_ALLOC) &&
-	    (cfg.size != child->bts_size)) {
-		int error;
+	ds_release_bts(context->tracer);
+	context->tracer = NULL;
 
-		ptrace_bts_free_buffer(child);
+	if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
+		free_bts_buffer(context);
+		if (!cfg.size)
+			return 0;
 
-		error = ptrace_bts_allocate_buffer(child, cfg.size);
-		if (error < 0)
-			return error;
+		alloc_bts_buffer(context, cfg.size);
+		if (!context->buffer)
+			return -ENOMEM;
 	}
 
 	if (cfg.flags & PTRACE_BTS_O_TRACE)
@@ -708,15 +801,13 @@ static int ptrace_bts_config(struct task
 	if (cfg.flags & PTRACE_BTS_O_SCHED)
 		flags |= BTS_TIMESTAMPS;
 
-	child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size,
-				    /* ovfl = */ NULL, /* th = */ (size_t)-1,
-				    flags);
-	if (IS_ERR(child->bts)) {
-		int error = PTR_ERR(child->bts);
-
-		ptrace_bts_free_buffer(child);
-		child->bts = NULL;
+	context->tracer = ds_request_bts(child, context->buffer, context->size,
+					 NULL, (size_t)-1, flags);
+	if (unlikely(IS_ERR(context->tracer))) {
+		int error = PTR_ERR(context->tracer);
 
+		free_bts_buffer(context);
+		context->tracer = NULL;
 		return error;
 	}
 
@@ -727,20 +818,25 @@ static int ptrace_bts_status(struct task
 			     long cfg_size,
 			     struct ptrace_bts_config __user *ucfg)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 	struct ptrace_bts_config cfg;
 
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
 	if (cfg_size < sizeof(cfg))
 		return -EIO;
 
-	trace = ds_read_bts(child->bts);
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	memset(&cfg, 0, sizeof(cfg));
-	cfg.size = trace->ds.end - trace->ds.begin;
-	cfg.signal = child->thread.bts_ovfl_signal;
-	cfg.bts_size = sizeof(struct bts_struct);
+	cfg.size	= trace->ds.end - trace->ds.begin;
+	cfg.signal	= context->bts_ovfl_signal;
+	cfg.bts_size	= sizeof(struct bts_struct);
 
 	if (cfg.signal)
 		cfg.flags |= PTRACE_BTS_O_SIGNAL;
@@ -759,67 +855,56 @@ static int ptrace_bts_status(struct task
 
 static int ptrace_bts_clear(struct task_struct *child)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 
-	return ds_reset_bts(child->bts);
+	return ds_reset_bts(context->tracer);
 }
 
 static int ptrace_bts_size(struct task_struct *child)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 }
 
-static void ptrace_bts_fork(struct task_struct *tsk)
+static inline void ptrace_bts_fork(struct task_struct *tsk)
 {
 	tsk->bts = NULL;
-	tsk->bts_buffer = NULL;
-	tsk->bts_size = 0;
-	tsk->thread.bts_ovfl_signal = 0;
 }
 
-static void ptrace_bts_untrace(struct task_struct *child)
+/*
+ * Called from __ptrace_unlink() after the child has been moved back
+ * to its original parent.
+ */
+static inline void ptrace_bts_untrace(struct task_struct *child)
 {
 	if (unlikely(child->bts)) {
-		ds_release_bts(child->bts);
+		free_bts_context(child->bts);
 		child->bts = NULL;
-
-		/* We cannot update total_vm and locked_vm since
-		   child's mm is already gone. But we can reclaim the
-		   memory. */
-		kfree(child->bts_buffer);
-		child->bts_buffer = NULL;
-		child->bts_size = 0;
 	}
 }
-
-static void ptrace_bts_detach(struct task_struct *child)
-{
-	/*
-	 * Ptrace_detach() races with ptrace_untrace() in case
-	 * the child dies and is reaped by another thread.
-	 *
-	 * We only do the memory accounting at this point and
-	 * leave the buffer deallocation and the bts tracer
-	 * release to ptrace_bts_untrace() which will be called
-	 * later on with tasklist_lock held.
-	 */
-	release_locked_buffer(child->bts_buffer, child->bts_size);
-}
 #else
 static inline void ptrace_bts_fork(struct task_struct *tsk) {}
-static inline void ptrace_bts_detach(struct task_struct *child) {}
 static inline void ptrace_bts_untrace(struct task_struct *child) {}
 #endif /* CONFIG_X86_PTRACE_BTS */
 
@@ -844,7 +929,6 @@ void ptrace_disable(struct task_struct *
 #ifdef TIF_SYSCALL_EMU
 	clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 #endif
-	ptrace_bts_detach(child);
 }
 
 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
Index: b/include/linux/sched.h
===================================================================
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -96,7 +96,7 @@ struct exec_domain;
 struct futex_pi_state;
 struct robust_list_head;
 struct bio;
-struct bts_tracer;
+struct bts_context;
 
 /*
  * List of flags we want to share for kernel threads,
@@ -1212,12 +1212,7 @@ struct task_struct {
 	 * This is the tracer handle for the ptrace BTS extension.
 	 * This field actually belongs to the ptracer task.
 	 */
-	struct bts_tracer *bts;
-	/*
-	 * The buffer to hold the BTS data.
-	 */
-	void *bts_buffer;
-	size_t bts_size;
+	struct bts_context *bts;
 #endif /* CONFIG_X86_PTRACE_BTS */
 
 	/* PID/PID hash table linkage. */
Index: b/include/linux/mm.h
===================================================================
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -13,6 +13,7 @@
 #include <linux/prio_tree.h>
 #include <linux/debug_locks.h>
 #include <linux/mm_types.h>
+#include <linux/sched.h>
 
 struct mempolicy;
 struct anon_vma;
@@ -1318,6 +1319,6 @@ void vmemmap_populate_print_last(void);
 
 extern void *alloc_locked_buffer(size_t size);
 extern void free_locked_buffer(void *buffer, size_t size);
-extern void release_locked_buffer(void *buffer, size_t size);
+extern void refund_locked_buffer_memory(struct mm_struct *mm, size_t size);
 #endif /* __KERNEL__ */
 #endif /* _LINUX_MM_H */
Index: b/mm/mlock.c
===================================================================
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -660,21 +660,20 @@ void *alloc_locked_buffer(size_t size)
 	return buffer;
 }
 
-void release_locked_buffer(void *buffer, size_t size)
+void refund_locked_buffer_memory(struct mm_struct *mm, size_t size)
 {
 	unsigned long pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
 
-	down_write(&current->mm->mmap_sem);
+	down_write(&mm->mmap_sem);
 
-	current->mm->total_vm  -= pgsz;
-	current->mm->locked_vm -= pgsz;
+	mm->total_vm  -= pgsz;
+	mm->locked_vm -= pgsz;
 
-	up_write(&current->mm->mmap_sem);
+	up_write(&mm->mmap_sem);
 }
 
 void free_locked_buffer(void *buffer, size_t size)
 {
-	release_locked_buffer(buffer, size);
-
+	refund_locked_buffer_memory(current->mm, size);
 	kfree(buffer);
 }
Index: b/arch/x86/include/asm/processor.h
===================================================================
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -455,10 +455,6 @@ struct thread_struct {
 /* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
 	struct ds_context	*ds_ctx;
 #endif /* CONFIG_X86_DS */
-#ifdef CONFIG_X86_PTRACE_BTS
-/* the signal to send on a bts buffer overflow */
-	unsigned int	bts_ovfl_signal;
-#endif /* CONFIG_X86_PTRACE_BTS */
 };
 
 static inline unsigned long native_get_debugreg(int regno)

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 04/20] x86, bts: wait until traced task has been scheduled out

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.wait_task_inactive.patch --]
[-- Type: text/plain, Size: 2249 bytes --]

In order to stop branch tracing for a running task, we need to first
clear the branch tracing control bits before we may free the tracing
buffer.
If the traced task is running, the cpu might still trace that task
after the branch trace control bits have cleared.

Wait until the traced task has been scheduled out before proceeding.


A similar problem affects the task debug store context. We first remove
the context, then we need to wait until the task has been scheduled
out before we can free the context memory.


Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds.c |   23 	19 +	4 -	0 !
 1 file changed, 19 insertions(+), 4 deletions(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -299,6 +299,7 @@ static inline struct ds_context *ds_get_
 
 static inline void ds_put_context(struct ds_context *context)
 {
+	struct task_struct *task;
 	unsigned long irq;
 
 	if (!context)
@@ -313,14 +314,20 @@ static inline void ds_put_context(struct
 
 	*(context->this) = NULL;
 
-	if (context->task)
-		clear_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
+	task = context->task;
+
+	if (task)
+		clear_tsk_thread_flag(task, TIF_DS_AREA_MSR);
 
-	if (!context->task || (context->task == current))
+	if (!task || (task == current))
 		wrmsrl(MSR_IA32_DS_AREA, 0);
 
 	spin_unlock_irqrestore(&ds_lock, irq);
 
+	/* The context might still be in use for context switching. */
+	if (task && (task != current))
+		wait_task_context_switch(task);
+
 	kfree(context);
 }
 
@@ -781,15 +788,23 @@ struct pebs_tracer *ds_request_pebs(stru
 
 void ds_release_bts(struct bts_tracer *tracer)
 {
+	struct task_struct *task;
+
 	if (!tracer)
 		return;
 
+	task = tracer->ds.context->task;
+
 	ds_suspend_bts(tracer);
 
 	WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
 	tracer->ds.context->bts_master = NULL;
 
-	put_tracer(tracer->ds.context->task);
+	/* Make sure tracing stopped and the tracer is not in use. */
+	if (task && (task != current))
+		wait_task_context_switch(task);
+
+	put_tracer(task);
 	ds_put_context(tracer->ds.context);
 
 	kfree(tracer);

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 05/20] x86, bts: fix race between per-task and per-cpu branch tracing

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.get_tracer.patch --]
[-- Type: text/plain, Size: 4925 bytes --]

Per-task branch tracing installs a debug store context with the traced
task. This immediately results in the branch trace control bits to be
cleared for the next context switch of that task, if not set before.

Either per-cpu or per-task tracing are allowed at the same time.

An active per-cpu tracing would be disabled even if the per-task tracing
request is rejected and the task debug store context removed.


Check the tracing type (per-cpu or per-task) before installing a task
debug store context.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds.c |   72 	41 +	31 -	0 !
 1 file changed, 41 insertions(+), 31 deletions(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -193,12 +193,28 @@ static DEFINE_SPINLOCK(ds_lock);
  */
 static atomic_t tracers = ATOMIC_INIT(0);
 
-static inline void get_tracer(struct task_struct *task)
+static inline int get_tracer(struct task_struct *task)
 {
-	if (task)
+	int error;
+
+	spin_lock_irq(&ds_lock);
+
+	if (task) {
+		error = -EPERM;
+		if (atomic_read(&tracers) < 0)
+			goto out;
 		atomic_inc(&tracers);
-	else
+	} else {
+		error = -EPERM;
+		if (atomic_read(&tracers) > 0)
+			goto out;
 		atomic_dec(&tracers);
+	}
+
+	error = 0;
+out:
+	spin_unlock_irq(&ds_lock);
+	return error;
 }
 
 static inline void put_tracer(struct task_struct *task)
@@ -209,14 +225,6 @@ static inline void put_tracer(struct tas
 		atomic_inc(&tracers);
 }
 
-static inline int check_tracer(struct task_struct *task)
-{
-	return task ?
-		(atomic_read(&tracers) >= 0) :
-		(atomic_read(&tracers) <= 0);
-}
-
-
 /*
  * The DS context is either attached to a thread or to a cpu:
  * - in the former case, the thread_struct contains a pointer to the
@@ -677,6 +685,10 @@ struct bts_tracer *ds_request_bts(struct
 	if (ovfl)
 		goto out;
 
+	error = get_tracer(task);
+	if (error < 0)
+		goto out;
+
 	/*
 	 * Per-cpu tracing is typically requested using smp_call_function().
 	 * We must not sleep.
@@ -684,7 +696,7 @@ struct bts_tracer *ds_request_bts(struct
 	error = -ENOMEM;
 	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
 	if (!tracer)
-		goto out;
+		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
@@ -696,13 +708,8 @@ struct bts_tracer *ds_request_bts(struct
 	spin_lock_irqsave(&ds_lock, irq);
 
 	error = -EPERM;
-	if (!check_tracer(task))
-		goto out_unlock;
-	get_tracer(task);
-
-	error = -EPERM;
 	if (tracer->ds.context->bts_master)
-		goto out_put_tracer;
+		goto out_unlock;
 	tracer->ds.context->bts_master = tracer;
 
 	spin_unlock_irqrestore(&ds_lock, irq);
@@ -716,13 +723,13 @@ struct bts_tracer *ds_request_bts(struct
 
 	return tracer;
 
- out_put_tracer:
-	put_tracer(task);
  out_unlock:
 	spin_unlock_irqrestore(&ds_lock, irq);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
+ out_put_tracer:
+	put_tracer(task);
  out:
 	return ERR_PTR(error);
 }
@@ -741,6 +748,10 @@ struct pebs_tracer *ds_request_pebs(stru
 	if (ovfl)
 		goto out;
 
+	error = get_tracer(task);
+	if (error < 0)
+		goto out;
+
 	/*
 	 * Per-cpu tracing is typically requested using smp_call_function().
 	 * We must not sleep.
@@ -748,7 +759,7 @@ struct pebs_tracer *ds_request_pebs(stru
 	error = -ENOMEM;
 	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
 	if (!tracer)
-		goto out;
+		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
@@ -759,13 +770,8 @@ struct pebs_tracer *ds_request_pebs(stru
 	spin_lock_irqsave(&ds_lock, irq);
 
 	error = -EPERM;
-	if (!check_tracer(task))
-		goto out_unlock;
-	get_tracer(task);
-
-	error = -EPERM;
 	if (tracer->ds.context->pebs_master)
-		goto out_put_tracer;
+		goto out_unlock;
 	tracer->ds.context->pebs_master = tracer;
 
 	spin_unlock_irqrestore(&ds_lock, irq);
@@ -775,13 +781,13 @@ struct pebs_tracer *ds_request_pebs(stru
 
 	return tracer;
 
- out_put_tracer:
-	put_tracer(task);
  out_unlock:
 	spin_unlock_irqrestore(&ds_lock, irq);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
+ out_put_tracer:
+	put_tracer(task);
  out:
 	return ERR_PTR(error);
 }
@@ -804,8 +810,8 @@ void ds_release_bts(struct bts_tracer *t
 	if (task && (task != current))
 		wait_task_context_switch(task);
 
-	put_tracer(task);
 	ds_put_context(tracer->ds.context);
+	put_tracer(task);
 
 	kfree(tracer);
 }
@@ -861,16 +867,20 @@ void ds_resume_bts(struct bts_tracer *tr
 
 void ds_release_pebs(struct pebs_tracer *tracer)
 {
+	struct task_struct *task;
+
 	if (!tracer)
 		return;
 
+	task = tracer->ds.context->task;
+
 	ds_suspend_pebs(tracer);
 
 	WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
 	tracer->ds.context->pebs_master = NULL;
 
-	put_tracer(tracer->ds.context->task);
 	ds_put_context(tracer->ds.context);
+	put_tracer(task);
 
 	kfree(tracer);
 }

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 06/20] x86, bts: use trace_clock_global() for timestamps

[markus.t.metzger]

[-- Attachment #1: tip.master.bts.cpu_clock.patch --]
[-- Type: text/plain, Size: 2539 bytes --]

Rename the bts_struct timestamp field to event.

Use trace_clock_global() for time measurement.


Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/include/asm/ds.h |    4 	2 +	2 -	0 !
 arch/x86/kernel/ds.c      |   17 	9 +	8 -	0 !
 2 files changed, 11 insertions(+), 10 deletions(-)

Index: b/arch/x86/include/asm/ds.h
===================================================================
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -170,9 +170,9 @@ struct bts_struct {
 		} lbr;
 		/* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */
 		struct {
-			__u64 jiffies;
+			__u64 clock;
 			pid_t pid;
-		} timestamp;
+		} event;
 	} variant;
 };
 
Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -25,6 +25,7 @@
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/mm.h>
+#include <linux/trace_clock.h>
 
 #include <asm/ds.h>
 
@@ -471,7 +472,7 @@ enum bts_field {
 	bts_flags,
 
 	bts_qual		= bts_from,
-	bts_jiffies		= bts_to,
+	bts_clock		= bts_to,
 	bts_pid			= bts_flags,
 
 	bts_qual_mask		= (bts_qual_max - 1),
@@ -517,8 +518,8 @@ bts_read(struct bts_tracer *tracer, cons
 	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);
+		out->variant.event.clock = bts_get(at, bts_clock);
+		out->variant.event.pid = bts_get(at, bts_pid);
 	} else {
 		out->qualifier = bts_branch;
 		out->variant.lbr.from = bts_get(at, bts_from);
@@ -555,8 +556,8 @@ static int bts_write(struct bts_tracer *
 	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);
+		bts_set(raw, bts_clock, in->variant.event.clock);
+		bts_set(raw, bts_pid, in->variant.event.pid);
 		break;
 	default:
 		return -EINVAL;
@@ -1083,9 +1084,9 @@ static inline void ds_take_timestamp(str
 		return;
 
 	memset(&ts, 0, sizeof(ts));
-	ts.qualifier			= qualifier;
-	ts.variant.timestamp.jiffies	= jiffies_64;
-	ts.variant.timestamp.pid	= task->pid;
+	ts.qualifier		= qualifier;
+	ts.variant.event.clock	= trace_clock_global();
+	ts.variant.event.pid	= task->pid;
 
 	bts_write(tracer, &ts);
 }

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 07/20] x86, debugctlmsr: add _on_cpu variants to debugctlmsr functions

[markus.t.metzger]

[-- Attachment #1: tip.master.update_debugctlmsr_on_cpu.patch --]
[-- Type: text/plain, Size: 1432 bytes --]

Add functions to get and set the debugctlmsr on different cpus.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/include/asm/processor.h |   27 	27 +	0 -	0 !
 1 file changed, 27 insertions(+)

Index: b/arch/x86/include/asm/processor.h
===================================================================
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -781,6 +781,21 @@ static inline unsigned long get_debugctl
     return debugctlmsr;
 }
 
+static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
+{
+	u64 debugctlmsr = 0;
+	u32 val1, val2;
+
+#ifndef CONFIG_X86_DEBUGCTLMSR
+	if (boot_cpu_data.x86 < 6)
+		return 0;
+#endif
+	rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
+	debugctlmsr = val1 | ((u64)val2 << 32);
+
+	return debugctlmsr;
+}
+
 static inline void update_debugctlmsr(unsigned long debugctlmsr)
 {
 #ifndef CONFIG_X86_DEBUGCTLMSR
@@ -790,6 +805,18 @@ static inline void update_debugctlmsr(un
 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
 }
 
+static inline void update_debugctlmsr_on_cpu(int cpu,
+					     unsigned long debugctlmsr)
+{
+#ifndef CONFIG_X86_DEBUGCTLMSR
+	if (boot_cpu_data.x86 < 6)
+		return;
+#endif
+	wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
+		     (u32)((u64)debugctlmsr),
+		     (u32)((u64)debugctlmsr >> 32));
+}
+
 /*
  * from system description table in BIOS. Mostly for MCA use, but
  * others may find it useful:

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 08/20] x86, bts, hw-branch-tracer: add _noirq variants to the debug store interface

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.request_cpu.patch --]
[-- Type: text/plain, Size: 31956 bytes --]

The hw-branch-tracer uses debug store functions from an on_each_cpu()
context, which is simply wrong since the functions may sleep.

Add _noirq variants for most functions, which  may be called with
interrupts disabled.

Separate per-cpu and per-task tracing and allow per-cpu tracing to be
controlled from any cpu.

Make the hw-branch-tracer use the new debug store interface, synchronize
with hotplug cpu event using get/put_online_cpus(), and remove the
unnecessary spinlock.

Make the ptrace bts and the ds selftest code use the new interface.

Defer the ds selftest.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/include/asm/ds.h        |   57 	45 +	12 -	0 !
 arch/x86/kernel/ds.c             |  474 	367 +	107 -	0 !
 arch/x86/kernel/ds_selftest.c    |    9 	5 +	4 -	0 !
 arch/x86/kernel/ptrace.c         |    5 	3 +	2 -	0 !
 kernel/trace/trace_hw_branches.c |  193 	72 +	121 -	0 !
 5 files changed, 492 insertions(+), 246 deletions(-)

Index: b/arch/x86/include/asm/ds.h
===================================================================
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -15,8 +15,8 @@
  * - buffer allocation (memory accounting)
  *
  *
- * Copyright (C) 2007-2008 Intel Corporation.
- * Markus Metzger <markus.t.metzger@intel.com>, 2007-2008
+ * Copyright (C) 2007-2009 Intel Corporation.
+ * Markus Metzger <markus.t.metzger@intel.com>, 2007-2009
  */
 
 #ifndef _ASM_X86_DS_H
@@ -83,8 +83,10 @@ enum ds_feature {
  * The interrupt threshold is independent from the overflow callback
  * to allow users to use their own overflow interrupt handling mechanism.
  *
- * task: the task to request recording for;
- *       NULL for per-cpu recording on the current cpu
+ * The function might sleep.
+ *
+ * task: the task to request recording for
+ * cpu:  the cpu to request recording for
  * base: the base pointer for the (non-pageable) buffer;
  * size: the size of the provided buffer in bytes
  * ovfl: pointer to a function to be called on buffer overflow;
@@ -93,19 +95,28 @@ enum ds_feature {
  *     -1 if no interrupt threshold is requested.
  * flags: a bit-mask of the above flags
  */
-extern struct bts_tracer *ds_request_bts(struct task_struct *task,
-					 void *base, size_t size,
-					 bts_ovfl_callback_t ovfl,
-					 size_t th, unsigned int flags);
-extern struct pebs_tracer *ds_request_pebs(struct task_struct *task,
-					   void *base, size_t size,
-					   pebs_ovfl_callback_t ovfl,
-					   size_t th, unsigned int flags);
+extern struct bts_tracer *ds_request_bts_task(struct task_struct *task,
+					      void *base, size_t size,
+					      bts_ovfl_callback_t ovfl,
+					      size_t th, unsigned int flags);
+extern struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
+					     bts_ovfl_callback_t ovfl,
+					     size_t th, unsigned int flags);
+extern struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
+						void *base, size_t size,
+						pebs_ovfl_callback_t ovfl,
+						size_t th, unsigned int flags);
+extern struct pebs_tracer *ds_request_pebs_cpu(int cpu,
+					       void *base, size_t size,
+					       pebs_ovfl_callback_t ovfl,
+					       size_t th, unsigned int flags);
 
 /*
  * Release BTS or PEBS resources
  * Suspend and resume BTS or PEBS tracing
  *
+ * Must be called with irq's enabled.
+ *
  * tracer: the tracer handle returned from ds_request_~()
  */
 extern void ds_release_bts(struct bts_tracer *tracer);
@@ -115,6 +126,28 @@ extern void ds_release_pebs(struct pebs_
 extern void ds_suspend_pebs(struct pebs_tracer *tracer);
 extern void ds_resume_pebs(struct pebs_tracer *tracer);
 
+/*
+ * Release BTS or PEBS resources
+ * Suspend and resume BTS or PEBS tracing
+ *
+ * Cpu tracers must call this on the traced cpu.
+ * Task tracers must call ds_release_~_noirq() for themselves.
+ *
+ * May be called with irq's disabled.
+ *
+ * Returns 0 if successful;
+ * -EPERM if the cpu tracer does not trace the current cpu.
+ * -EPERM if the task tracer does not trace itself.
+ *
+ * tracer: the tracer handle returned from ds_request_~()
+ */
+extern int ds_release_bts_noirq(struct bts_tracer *tracer);
+extern int ds_suspend_bts_noirq(struct bts_tracer *tracer);
+extern int ds_resume_bts_noirq(struct bts_tracer *tracer);
+extern int ds_release_pebs_noirq(struct pebs_tracer *tracer);
+extern int ds_suspend_pebs_noirq(struct pebs_tracer *tracer);
+extern int ds_resume_pebs_noirq(struct pebs_tracer *tracer);
+
 
 /*
  * The raw DS buffer state as it is used for BTS and PEBS recording.
Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -245,60 +245,50 @@ struct ds_context {
 	struct pebs_tracer	*pebs_master;
 
 	/* Use count: */
-	unsigned long count;
+	unsigned long		count;
 
 	/* Pointer to the context pointer field: */
 	struct ds_context	**this;
 
-	/* The traced task; NULL for current cpu: */
+	/* The traced task; NULL for cpu tracing: */
 	struct task_struct	*task;
-};
 
-static DEFINE_PER_CPU(struct ds_context *, system_context_array);
+	/* The traced cpu; only valid if task is NULL: */
+	int			cpu;
+};
 
-#define system_context per_cpu(system_context_array, smp_processor_id())
+static DEFINE_PER_CPU(struct ds_context *, cpu_context);
 
 
-static inline struct ds_context *ds_get_context(struct task_struct *task)
+static struct ds_context *ds_get_context(struct task_struct *task, int cpu)
 {
 	struct ds_context **p_context =
-		(task ? &task->thread.ds_ctx : &system_context);
+		(task ? &task->thread.ds_ctx : &per_cpu(cpu_context, cpu));
 	struct ds_context *context = NULL;
 	struct ds_context *new_context = NULL;
-	unsigned long irq;
 
-	/*
-	 * Chances are small that we already have a context.
-	 *
-	 * Contexts for per-cpu tracing are allocated using
-	 * smp_call_function(). We must not sleep.
-	 */
-	new_context = kzalloc(sizeof(*new_context), GFP_ATOMIC);
+	/* 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_irq(&ds_lock);
 
 	context = *p_context;
-	if (!context) {
+	if (likely(!context)) {
 		context = new_context;
 
 		context->this = p_context;
 		context->task = task;
+		context->cpu = cpu;
 		context->count = 0;
 
-		if (task)
-			set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
-
-		if (!task || (task == current))
-			wrmsrl(MSR_IA32_DS_AREA, (unsigned long)context->ds);
-
 		*p_context = context;
 	}
 
 	context->count++;
 
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 
 	if (context != new_context)
 		kfree(new_context);
@@ -306,7 +296,7 @@ static inline struct ds_context *ds_get_
 	return context;
 }
 
-static inline void ds_put_context(struct ds_context *context)
+static void ds_put_context(struct ds_context *context)
 {
 	struct task_struct *task;
 	unsigned long irq;
@@ -328,8 +318,15 @@ static inline void ds_put_context(struct
 	if (task)
 		clear_tsk_thread_flag(task, TIF_DS_AREA_MSR);
 
-	if (!task || (task == current))
-		wrmsrl(MSR_IA32_DS_AREA, 0);
+	/*
+	 * We leave the (now dangling) pointer to the DS configuration in
+	 * the DS_AREA msr. This is as good or as bad as replacing it with
+	 * NULL - the hardware would crash if we enabled tracing.
+	 *
+	 * This saves us some problems with having to write an msr on a
+	 * different cpu while preventing others from doing the same for the
+	 * next context for that same cpu.
+	 */
 
 	spin_unlock_irqrestore(&ds_lock, irq);
 
@@ -340,6 +337,31 @@ static inline void ds_put_context(struct
 	kfree(context);
 }
 
+static void ds_install_ds_area(struct ds_context *context)
+{
+	unsigned long ds;
+
+	ds = (unsigned long)context->ds;
+
+	/*
+	 * There is a race between the bts master and the pebs master.
+	 *
+	 * The thread/cpu access is synchronized via get/put_cpu() for
+	 * task tracing and via wrmsr_on_cpu for cpu tracing.
+	 *
+	 * If bts and pebs are collected for the same task or same cpu,
+	 * the same confiuration is written twice.
+	 */
+	if (context->task) {
+		get_cpu();
+		if (context->task == current)
+			wrmsrl(MSR_IA32_DS_AREA, ds);
+		set_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
+		put_cpu();
+	} else
+		wrmsr_on_cpu(context->cpu, MSR_IA32_DS_AREA,
+			     (u32)((u64)ds), (u32)((u64)ds >> 32));
+}
 
 /*
  * Call the tracer's callback on a buffer overflow.
@@ -622,6 +644,7 @@ static void ds_init_ds_trace(struct ds_t
 	 * The value for 'no threshold' is -1, which will set the
 	 * threshold outside of the buffer, just like we want it.
 	 */
+	ith *= ds_cfg.sizeof_rec[qual];
 	trace->ith = (void *)(buffer + size - ith);
 
 	trace->flags = flags;
@@ -630,7 +653,7 @@ static void ds_init_ds_trace(struct ds_t
 
 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)
+		      int cpu, void *base, size_t size, size_t th)
 {
 	struct ds_context *context;
 	int error;
@@ -643,7 +666,7 @@ static int ds_request(struct ds_tracer *
 	if (!base)
 		goto out;
 
-	/* We require some space to do alignment adjustments below. */
+	/* We need space for alignment adjustments in ds_init_ds_trace(). */
 	error = -EINVAL;
 	if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual]))
 		goto out;
@@ -660,25 +683,27 @@ static int ds_request(struct ds_tracer *
 	tracer->size = size;
 
 	error = -ENOMEM;
-	context = ds_get_context(task);
+	context = ds_get_context(task, cpu);
 	if (!context)
 		goto out;
 	tracer->context = context;
 
-	ds_init_ds_trace(trace, qual, base, size, th, flags);
+	/*
+	 * Defer any tracer-specific initialization work for the context until
+	 * context ownership has been clarified.
+	 */
 
 	error = 0;
  out:
 	return error;
 }
 
-struct bts_tracer *ds_request_bts(struct task_struct *task,
-				  void *base, size_t size,
-				  bts_ovfl_callback_t ovfl, size_t th,
-				  unsigned int flags)
+static struct bts_tracer *ds_request_bts(struct task_struct *task, int cpu,
+					 void *base, size_t size,
+					 bts_ovfl_callback_t ovfl, size_t th,
+					 unsigned int flags)
 {
 	struct bts_tracer *tracer;
-	unsigned long irq;
 	int error;
 
 	/* Buffer overflow notification is not yet implemented. */
@@ -690,42 +715,46 @@ struct bts_tracer *ds_request_bts(struct
 	if (error < 0)
 		goto out;
 
-	/*
-	 * Per-cpu tracing is typically requested using smp_call_function().
-	 * We must not sleep.
-	 */
 	error = -ENOMEM;
-	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
+	tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
 	if (!tracer)
 		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
+	/* Do some more error checking and acquire a tracing context. */
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
-			   ds_bts, task, base, size, th, flags);
+			   ds_bts, task, cpu, base, size, th);
 	if (error < 0)
 		goto out_tracer;
 
-
-	spin_lock_irqsave(&ds_lock, irq);
+	/* Claim the bts part of the tracing context we acquired above. */
+	spin_lock_irq(&ds_lock);
 
 	error = -EPERM;
 	if (tracer->ds.context->bts_master)
 		goto out_unlock;
 	tracer->ds.context->bts_master = tracer;
 
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 
+	/*
+	 * Now that we own the bts part of the context, let's complete the
+	 * initialization for that part.
+	 */
+	ds_init_ds_trace(&tracer->trace.ds, ds_bts, base, size, th, flags);
+	ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
+	ds_install_ds_area(tracer->ds.context);
 
 	tracer->trace.read  = bts_read;
 	tracer->trace.write = bts_write;
 
-	ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
+	/* Start tracing. */
 	ds_resume_bts(tracer);
 
 	return tracer;
 
  out_unlock:
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
@@ -735,13 +764,27 @@ struct bts_tracer *ds_request_bts(struct
 	return ERR_PTR(error);
 }
 
-struct pebs_tracer *ds_request_pebs(struct task_struct *task,
-				    void *base, size_t size,
-				    pebs_ovfl_callback_t ovfl, size_t th,
-				    unsigned int flags)
+struct bts_tracer *ds_request_bts_task(struct task_struct *task,
+				       void *base, size_t size,
+				       bts_ovfl_callback_t ovfl,
+				       size_t th, unsigned int flags)
+{
+	return ds_request_bts(task, 0, base, size, ovfl, th, flags);
+}
+
+struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
+				      bts_ovfl_callback_t ovfl,
+				      size_t th, unsigned int flags)
+{
+	return ds_request_bts(NULL, cpu, base, size, ovfl, th, flags);
+}
+
+static struct pebs_tracer *ds_request_pebs(struct task_struct *task, int cpu,
+					   void *base, size_t size,
+					   pebs_ovfl_callback_t ovfl, size_t th,
+					   unsigned int flags)
 {
 	struct pebs_tracer *tracer;
-	unsigned long irq;
 	int error;
 
 	/* Buffer overflow notification is not yet implemented. */
@@ -753,37 +796,43 @@ struct pebs_tracer *ds_request_pebs(stru
 	if (error < 0)
 		goto out;
 
-	/*
-	 * Per-cpu tracing is typically requested using smp_call_function().
-	 * We must not sleep.
-	 */
 	error = -ENOMEM;
-	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
+	tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
 	if (!tracer)
 		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
+	/* Do some more error checking and acquire a tracing context. */
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
-			   ds_pebs, task, base, size, th, flags);
+			   ds_pebs, task, cpu, base, size, th);
 	if (error < 0)
 		goto out_tracer;
 
-	spin_lock_irqsave(&ds_lock, irq);
+	/* Claim the pebs part of the tracing context we acquired above. */
+	spin_lock_irq(&ds_lock);
 
 	error = -EPERM;
 	if (tracer->ds.context->pebs_master)
 		goto out_unlock;
 	tracer->ds.context->pebs_master = tracer;
 
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 
+	/*
+	 * Now that we own the pebs part of the context, let's complete the
+	 * initialization for that part.
+	 */
+	ds_init_ds_trace(&tracer->trace.ds, ds_pebs, base, size, th, flags);
 	ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
+	ds_install_ds_area(tracer->ds.context);
+
+	/* Start tracing. */
 	ds_resume_pebs(tracer);
 
 	return tracer;
 
  out_unlock:
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
@@ -793,16 +842,26 @@ struct pebs_tracer *ds_request_pebs(stru
 	return ERR_PTR(error);
 }
 
-void ds_release_bts(struct bts_tracer *tracer)
+struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
+					 void *base, size_t size,
+					 pebs_ovfl_callback_t ovfl,
+					 size_t th, unsigned int flags)
 {
-	struct task_struct *task;
+	return ds_request_pebs(task, 0, base, size, ovfl, th, flags);
+}
 
-	if (!tracer)
-		return;
+struct pebs_tracer *ds_request_pebs_cpu(int cpu, void *base, size_t size,
+					pebs_ovfl_callback_t ovfl,
+					size_t th, unsigned int flags)
+{
+	return ds_request_pebs(NULL, cpu, base, size, ovfl, th, flags);
+}
 
-	task = tracer->ds.context->task;
+static void ds_free_bts(struct bts_tracer *tracer)
+{
+	struct task_struct *task;
 
-	ds_suspend_bts(tracer);
+	task = tracer->ds.context->task;
 
 	WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
 	tracer->ds.context->bts_master = NULL;
@@ -817,9 +876,69 @@ void ds_release_bts(struct bts_tracer *t
 	kfree(tracer);
 }
 
+void ds_release_bts(struct bts_tracer *tracer)
+{
+	might_sleep();
+
+	if (!tracer)
+		return;
+
+	ds_suspend_bts(tracer);
+	ds_free_bts(tracer);
+}
+
+int ds_release_bts_noirq(struct bts_tracer *tracer)
+{
+	struct task_struct *task;
+	unsigned long irq;
+	int error;
+
+	if (!tracer)
+		return 0;
+
+	task = tracer->ds.context->task;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task &&
+	    (tracer->ds.context->cpu != smp_processor_id()))
+		goto out;
+
+	error = -EPERM;
+	if (task && (task != current))
+		goto out;
+
+	ds_suspend_bts_noirq(tracer);
+	ds_free_bts(tracer);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
+static void update_task_debugctlmsr(struct task_struct *task,
+				    unsigned long debugctlmsr)
+{
+	task->thread.debugctlmsr = debugctlmsr;
+
+	get_cpu();
+	if (task == current)
+		update_debugctlmsr(debugctlmsr);
+
+	if (task->thread.debugctlmsr)
+		set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
+	else
+		clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
+	put_cpu();
+}
+
 void ds_suspend_bts(struct bts_tracer *tracer)
 {
 	struct task_struct *task;
+	unsigned long debugctlmsr;
+	int cpu;
 
 	if (!tracer)
 		return;
@@ -827,29 +946,60 @@ void ds_suspend_bts(struct bts_tracer *t
 	tracer->flags = 0;
 
 	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
 
-	if (!task || (task == current))
-		update_debugctlmsr(get_debugctlmsr() & ~BTS_CONTROL);
+	WARN_ON(!task && irqs_disabled());
 
-	if (task) {
-		task->thread.debugctlmsr &= ~BTS_CONTROL;
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr_on_cpu(cpu));
+	debugctlmsr &= ~BTS_CONTROL;
 
-		if (!task->thread.debugctlmsr)
-			clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
-	}
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr_on_cpu(cpu, debugctlmsr);
 }
 
-void ds_resume_bts(struct bts_tracer *tracer)
+int ds_suspend_bts_noirq(struct bts_tracer *tracer)
 {
 	struct task_struct *task;
-	unsigned long control;
+	unsigned long debugctlmsr, irq;
+	int cpu, error = 0;
 
 	if (!tracer)
-		return;
+		return 0;
 
-	tracer->flags = tracer->trace.ds.flags;
+	tracer->flags = 0;
 
 	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task && (cpu != smp_processor_id()))
+		goto out;
+
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr());
+	debugctlmsr &= ~BTS_CONTROL;
+
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr(debugctlmsr);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
+static unsigned long ds_bts_control(struct bts_tracer *tracer)
+{
+	unsigned long control;
 
 	control = ds_cfg.ctl[dsf_bts];
 	if (!(tracer->trace.ds.flags & BTS_KERNEL))
@@ -857,25 +1007,77 @@ void ds_resume_bts(struct bts_tracer *tr
 	if (!(tracer->trace.ds.flags & BTS_USER))
 		control |= ds_cfg.ctl[dsf_bts_user];
 
-	if (task) {
-		task->thread.debugctlmsr |= control;
-		set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
-	}
-
-	if (!task || (task == current))
-		update_debugctlmsr(get_debugctlmsr() | control);
+	return control;
 }
 
-void ds_release_pebs(struct pebs_tracer *tracer)
+void ds_resume_bts(struct bts_tracer *tracer)
 {
 	struct task_struct *task;
+	unsigned long debugctlmsr;
+	int cpu;
 
 	if (!tracer)
 		return;
 
+	tracer->flags = tracer->trace.ds.flags;
+
 	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
 
-	ds_suspend_pebs(tracer);
+	WARN_ON(!task && irqs_disabled());
+
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr_on_cpu(cpu));
+	debugctlmsr |= ds_bts_control(tracer);
+
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr_on_cpu(cpu, debugctlmsr);
+}
+
+int ds_resume_bts_noirq(struct bts_tracer *tracer)
+{
+	struct task_struct *task;
+	unsigned long debugctlmsr, irq;
+	int cpu, error = 0;
+
+	if (!tracer)
+		return 0;
+
+	tracer->flags = tracer->trace.ds.flags;
+
+	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task && (cpu != smp_processor_id()))
+		goto out;
+
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr());
+	debugctlmsr |= ds_bts_control(tracer);
+
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr(debugctlmsr);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
+static void ds_free_pebs(struct pebs_tracer *tracer)
+{
+	struct task_struct *task;
+
+	task = tracer->ds.context->task;
 
 	WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
 	tracer->ds.context->pebs_master = NULL;
@@ -886,16 +1088,68 @@ void ds_release_pebs(struct pebs_tracer 
 	kfree(tracer);
 }
 
+void ds_release_pebs(struct pebs_tracer *tracer)
+{
+	might_sleep();
+
+	if (!tracer)
+		return;
+
+	ds_suspend_pebs(tracer);
+	ds_free_pebs(tracer);
+}
+
+int ds_release_pebs_noirq(struct pebs_tracer *tracer)
+{
+	struct task_struct *task;
+	unsigned long irq;
+	int error;
+
+	if (!tracer)
+		return 0;
+
+	task = tracer->ds.context->task;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task &&
+	    (tracer->ds.context->cpu != smp_processor_id()))
+		goto out;
+
+	error = -EPERM;
+	if (task && (task != current))
+		goto out;
+
+	ds_suspend_pebs_noirq(tracer);
+	ds_free_pebs(tracer);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
 void ds_suspend_pebs(struct pebs_tracer *tracer)
 {
 
 }
 
+int ds_suspend_pebs_noirq(struct pebs_tracer *tracer)
+{
+	return 0;
+}
+
 void ds_resume_pebs(struct pebs_tracer *tracer)
 {
 
 }
 
+int ds_resume_pebs_noirq(struct pebs_tracer *tracer)
+{
+	return 0;
+}
+
 const struct bts_trace *ds_read_bts(struct bts_tracer *tracer)
 {
 	if (!tracer)
@@ -1004,26 +1258,6 @@ ds_configure(const struct ds_configurati
 		printk(KERN_INFO "[ds] pebs not available\n");
 	}
 
-	if (ds_cfg.sizeof_rec[ds_bts]) {
-		int error;
-
-		error = ds_selftest_bts();
-		if (error) {
-			WARN(1, "[ds] selftest failed. disabling bts.\n");
-			ds_cfg.sizeof_rec[ds_bts] = 0;
-		}
-	}
-
-	if (ds_cfg.sizeof_rec[ds_pebs]) {
-		int error;
-
-		error = ds_selftest_pebs();
-		if (error) {
-			WARN(1, "[ds] selftest failed. disabling pebs.\n");
-			ds_cfg.sizeof_rec[ds_pebs] = 0;
-		}
-	}
-
 	printk(KERN_INFO "[ds] sizes: address: %u bit, ",
 	       8 * ds_cfg.sizeof_ptr_field);
 	printk("bts/pebs record: %u/%u bytes\n",
@@ -1127,3 +1361,29 @@ void ds_copy_thread(struct task_struct *
 void ds_exit_thread(struct task_struct *tsk)
 {
 }
+
+static __init int ds_selftest(void)
+{
+	if (ds_cfg.sizeof_rec[ds_bts]) {
+		int error;
+
+		error = ds_selftest_bts();
+		if (error) {
+			WARN(1, "[ds] selftest failed. disabling bts.\n");
+			ds_cfg.sizeof_rec[ds_bts] = 0;
+		}
+	}
+
+	if (ds_cfg.sizeof_rec[ds_pebs]) {
+		int error;
+
+		error = ds_selftest_pebs();
+		if (error) {
+			WARN(1, "[ds] selftest failed. disabling pebs.\n");
+			ds_cfg.sizeof_rec[ds_pebs] = 0;
+		}
+	}
+
+	return 0;
+}
+device_initcall(ds_selftest);
Index: b/kernel/trace/trace_hw_branches.c
===================================================================
--- a/kernel/trace/trace_hw_branches.c
+++ b/kernel/trace/trace_hw_branches.c
@@ -4,7 +4,6 @@
  * Copyright (C) 2008-2009 Intel Corporation.
  * Markus Metzger <markus.t.metzger@gmail.com>, 2008-2009
  */
-#include <linux/spinlock.h>
 #include <linux/kallsyms.h>
 #include <linux/debugfs.h>
 #include <linux/ftrace.h>
@@ -21,168 +20,113 @@
 
 #define BTS_BUFFER_SIZE (1 << 13)
 
-/*
- * The tracer lock protects the below per-cpu tracer array.
- * It needs to be held to:
- * - start tracing on all cpus
- * - stop tracing on all cpus
- * - start tracing on a single hotplug cpu
- * - stop tracing on a single hotplug cpu
- * - read the trace from all cpus
- * - read the trace from a single cpu
- */
-static DEFINE_SPINLOCK(bts_tracer_lock);
 static DEFINE_PER_CPU(struct bts_tracer *, tracer);
 static DEFINE_PER_CPU(unsigned char[BTS_BUFFER_SIZE], buffer);
 
 #define this_tracer per_cpu(tracer, smp_processor_id())
-#define this_buffer per_cpu(buffer, smp_processor_id())
 
 static int trace_hw_branches_enabled __read_mostly;
 static int trace_hw_branches_suspended __read_mostly;
 static struct trace_array *hw_branch_trace __read_mostly;
 
 
-/*
- * Initialize the tracer for the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_init_cpu(void *arg)
+static void bts_trace_init_cpu(int cpu)
 {
-	if (this_tracer)
-		ds_release_bts(this_tracer);
+	per_cpu(tracer, cpu) =
+		ds_request_bts_cpu(cpu, per_cpu(buffer, cpu), BTS_BUFFER_SIZE,
+				   NULL, (size_t)-1, BTS_KERNEL);
 
-	this_tracer = ds_request_bts(NULL, this_buffer, BTS_BUFFER_SIZE,
-				     NULL, (size_t)-1, BTS_KERNEL);
-	if (IS_ERR(this_tracer)) {
-		this_tracer = NULL;
-		return;
-	}
+	if (IS_ERR(per_cpu(tracer, cpu)))
+		per_cpu(tracer, cpu) = NULL;
 }
 
 static int bts_trace_init(struct trace_array *tr)
 {
-	int cpu, avail;
-
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
 	hw_branch_trace = tr;
+	trace_hw_branches_enabled = 0;
 
-	on_each_cpu(bts_trace_init_cpu, NULL, 1);
-
-	/* Check on how many cpus we could enable tracing */
-	avail = 0;
-	for_each_online_cpu(cpu)
-		if (per_cpu(tracer, cpu))
-			avail++;
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		bts_trace_init_cpu(cpu);
 
-	trace_hw_branches_enabled = (avail ? 1 : 0);
+		if (likely(per_cpu(tracer, cpu)))
+			trace_hw_branches_enabled = 1;
+	}
 	trace_hw_branches_suspended = 0;
-
-	spin_unlock(&bts_tracer_lock);
-
+	put_online_cpus();
 
 	/* If we could not enable tracing on a single cpu, we fail. */
-	return avail ? 0 : -EOPNOTSUPP;
-}
-
-/*
- * Release the tracer for the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_release_cpu(void *arg)
-{
-	if (this_tracer) {
-		ds_release_bts(this_tracer);
-		this_tracer = NULL;
-	}
+	return trace_hw_branches_enabled ? 0 : -EOPNOTSUPP;
 }
 
 static void bts_trace_reset(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
-	on_each_cpu(bts_trace_release_cpu, NULL, 1);
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		if (likely(per_cpu(tracer, cpu))) {
+			ds_release_bts(per_cpu(tracer, cpu));
+			per_cpu(tracer, cpu) = NULL;
+		}
+	}
 	trace_hw_branches_enabled = 0;
 	trace_hw_branches_suspended = 0;
-
-	spin_unlock(&bts_tracer_lock);
-}
-
-/*
- * Resume tracing on the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_resume_cpu(void *arg)
-{
-	if (this_tracer)
-		ds_resume_bts(this_tracer);
+	put_online_cpus();
 }
 
 static void bts_trace_start(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
-	on_each_cpu(bts_trace_resume_cpu, NULL, 1);
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_resume_bts(per_cpu(tracer, cpu));
 	trace_hw_branches_suspended = 0;
-
-	spin_unlock(&bts_tracer_lock);
-}
-
-/*
- * Suspend tracing on the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_suspend_cpu(void *arg)
-{
-	if (this_tracer)
-		ds_suspend_bts(this_tracer);
+	put_online_cpus();
 }
 
 static void bts_trace_stop(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
-	on_each_cpu(bts_trace_suspend_cpu, NULL, 1);
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_suspend_bts(per_cpu(tracer, cpu));
 	trace_hw_branches_suspended = 1;
-
-	spin_unlock(&bts_tracer_lock);
+	put_online_cpus();
 }
 
 static int __cpuinit bts_hotcpu_handler(struct notifier_block *nfb,
 				     unsigned long action, void *hcpu)
 {
-	unsigned int cpu = (unsigned long)hcpu;
-
-	spin_lock(&bts_tracer_lock);
-
-	if (!trace_hw_branches_enabled)
-		goto out;
+	int cpu = (long)hcpu;
 
 	switch (action) {
 	case CPU_ONLINE:
 	case CPU_DOWN_FAILED:
-		smp_call_function_single(cpu, bts_trace_init_cpu, NULL, 1);
-
-		if (trace_hw_branches_suspended)
-			smp_call_function_single(cpu, bts_trace_suspend_cpu,
-						 NULL, 1);
+		/* The notification is sent with interrupts enabled. */
+		if (trace_hw_branches_enabled) {
+			bts_trace_init_cpu(cpu);
+
+			if (trace_hw_branches_suspended &&
+			    likely(per_cpu(tracer, cpu)))
+				ds_suspend_bts(per_cpu(tracer, cpu));
+		}
 		break;
+
 	case CPU_DOWN_PREPARE:
-		smp_call_function_single(cpu, bts_trace_release_cpu, NULL, 1);
-		break;
+		/* The notification is sent with interrupts enabled. */
+		if (likely(per_cpu(tracer, cpu))) {
+			ds_release_bts(per_cpu(tracer, cpu));
+			per_cpu(tracer, cpu) = NULL;
+		}
 	}
 
- out:
-	spin_unlock(&bts_tracer_lock);
 	return NOTIFY_DONE;
 }
 
@@ -274,7 +218,7 @@ static void trace_bts_at(const struct bt
 /*
  * Collect the trace on the current cpu and write it into the ftrace buffer.
  *
- * pre: bts_tracer_lock must be locked
+ * pre: tracing must be suspended on the current cpu
  */
 static void trace_bts_cpu(void *arg)
 {
@@ -291,10 +235,9 @@ static void trace_bts_cpu(void *arg)
 	if (unlikely(!this_tracer))
 		return;
 
-	ds_suspend_bts(this_tracer);
 	trace = ds_read_bts(this_tracer);
 	if (!trace)
-		goto out;
+		return;
 
 	for (at = trace->ds.top; (void *)at < trace->ds.end;
 	     at += trace->ds.size)
@@ -303,18 +246,27 @@ static void trace_bts_cpu(void *arg)
 	for (at = trace->ds.begin; (void *)at < trace->ds.top;
 	     at += trace->ds.size)
 		trace_bts_at(trace, at);
-
-out:
-	ds_resume_bts(this_tracer);
 }
 
 static void trace_bts_prepare(struct trace_iterator *iter)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_suspend_bts(per_cpu(tracer, cpu));
+	/*
+	 * We need to collect the trace on the respective cpu since ftrace
+	 * implicitly adds the record for the current cpu.
+	 * Once that is more flexible, we could collect the data from any cpu.
+	 */
 	on_each_cpu(trace_bts_cpu, iter->tr, 1);
 
-	spin_unlock(&bts_tracer_lock);
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_resume_bts(per_cpu(tracer, cpu));
+	put_online_cpus();
 }
 
 static void trace_bts_close(struct trace_iterator *iter)
@@ -324,12 +276,11 @@ static void trace_bts_close(struct trace
 
 void trace_hw_branch_oops(void)
 {
-	spin_lock(&bts_tracer_lock);
-
-	if (trace_hw_branches_enabled)
+	if (this_tracer) {
+		ds_suspend_bts_noirq(this_tracer);
 		trace_bts_cpu(hw_branch_trace);
-
-	spin_unlock(&bts_tracer_lock);
+		ds_resume_bts_noirq(this_tracer);
+	}
 }
 
 struct tracer bts_tracer __read_mostly =
Index: b/arch/x86/kernel/ptrace.c
===================================================================
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -801,8 +801,9 @@ static int ptrace_bts_config(struct task
 	if (cfg.flags & PTRACE_BTS_O_SCHED)
 		flags |= BTS_TIMESTAMPS;
 
-	context->tracer = ds_request_bts(child, context->buffer, context->size,
-					 NULL, (size_t)-1, flags);
+	context->tracer =
+		ds_request_bts_task(child, context->buffer, context->size,
+				    NULL, (size_t)-1, flags);
 	if (unlikely(IS_ERR(context->tracer))) {
 		int error = PTR_ERR(context->tracer);
 
Index: b/arch/x86/kernel/ds_selftest.c
===================================================================
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -10,11 +10,12 @@
 
 #include <linux/kernel.h>
 #include <linux/string.h>
+#include <linux/smp.h>
 
 #include <asm/ds.h>
 
 
-#define DS_SELFTEST_BUFFER_SIZE 1021 /* Intentionally chose an odd size. */
+#define BUFFER_SIZE 1021 /* Intentionally chose an odd size. */
 
 
 static int ds_selftest_bts_consistency(const struct bts_trace *trace)
@@ -125,12 +126,12 @@ int ds_selftest_bts(void)
 	struct bts_tracer *tracer;
 	int error = 0;
 	void *top;
-	unsigned char buffer[DS_SELFTEST_BUFFER_SIZE];
+	unsigned char buffer[BUFFER_SIZE];
 
 	printk(KERN_INFO "[ds] bts selftest...");
 
-	tracer = ds_request_bts(NULL, buffer, DS_SELFTEST_BUFFER_SIZE,
-				NULL, (size_t)-1, BTS_KERNEL);
+	tracer = ds_request_bts_cpu(smp_processor_id(), buffer, BUFFER_SIZE,
+				    NULL, (size_t)-1, BTS_KERNEL);
 	if (IS_ERR(tracer)) {
 		error = PTR_ERR(tracer);
 		tracer = NULL;

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 09/20] x86, hw-branch-tracer: allocate selftest iterator on heap

[markus.t.metzger]

[-- Attachment #1: tip.master.hw-branch-tracer.selftest_alloc.patch --]
[-- Type: text/plain, Size: 1328 bytes --]

Allocate the trace_iterator for the hw-branch-tracer selftest on the heap.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 kernel/trace/trace_selftest.c |   20 	12 +	8 -	0 !
 1 file changed, 12 insertions(+), 8 deletions(-)

Index: b/kernel/trace/trace_selftest.c
===================================================================
--- a/kernel/trace/trace_selftest.c
+++ b/kernel/trace/trace_selftest.c
@@ -699,7 +699,7 @@ int
 trace_selftest_startup_hw_branches(struct tracer *trace,
 				   struct trace_array *tr)
 {
-	struct trace_iterator iter;
+	struct trace_iterator *iter;
 	struct tracer tracer;
 	unsigned long count;
 	int ret;
@@ -719,17 +719,21 @@ trace_selftest_startup_hw_branches(struc
 	 * The hw-branch tracer needs to collect the trace from the various
 	 * cpu trace buffers - before tracing is stopped.
 	 */
-	memset(&iter, 0, sizeof(iter));
+	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
+	if (!iter)
+		return -ENOMEM;
+
 	memcpy(&tracer, trace, sizeof(tracer));
 
-	iter.trace = &tracer;
-	iter.tr = tr;
-	iter.pos = -1;
-	mutex_init(&iter.mutex);
+	iter->trace = &tracer;
+	iter->tr = tr;
+	iter->pos = -1;
+	mutex_init(&iter->mutex);
 
-	trace->open(&iter);
+	trace->open(iter);
 
-	mutex_destroy(&iter.mutex);
+	mutex_destroy(&iter->mutex);
+	kfree(iter);
 
 	tracing_stop();
 

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 10/20] x86, ds: fix compiler warning

[markus.t.metzger]

[-- Attachment #1: tip.master.ds_selftest.printk.patch --]
[-- Type: text/plain, Size: 728 bytes --]

Size_t is defined differently on i386 and x86_64.
Change type to avoid compiler warning.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds_selftest.c |    2 	1 +	1 -	0 !
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: b/arch/x86/kernel/ds_selftest.c
===================================================================
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -87,7 +87,7 @@ static int ds_selftest_bts_read(struct b
 	/* Now to the test itself. */
 	for (at = from; (void *)at < to; at += trace->ds.size) {
 		struct bts_struct bts;
-		size_t index;
+		unsigned long index;
 		int error;
 
 		if (((void *)at - trace->ds.begin) % trace->ds.size) {

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 11/20] x86, ds: fix bounds check in ds selftest

[markus.t.metzger]

[-- Attachment #1: tip.master.ds_selftest.bounds_check.patch --]
[-- Type: text/plain, Size: 877 bytes --]

Fix a bad bounds check in the debug store selftest.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds_selftest.c |    7 	6 +	1 -	0 !
 1 file changed, 6 insertions(+), 1 deletion(-)

Index: b/arch/x86/kernel/ds_selftest.c
===================================================================
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -47,8 +47,13 @@ static int ds_selftest_bts_consistency(c
 		printk(KERN_CONT "bad bts buffer setup...");
 		error = -1;
 	}
+	/*
+	 * We allow top in [begin; end], since its not clear when the
+	 * overflow adjustment happens: after the increment or before the
+	 * write.
+	 */
 	if ((trace->ds.top < trace->ds.begin) ||
-	    (trace->ds.end <= trace->ds.top)) {
+	    (trace->ds.end < trace->ds.top)) {
 		printk(KERN_CONT "bts top out of bounds...");
 		error = -1;
 	}

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 12/20] x86, ds: selftest each cpu

[markus.t.metzger]

[-- Attachment #1: tip.master.ds_selftest.for_each_cpu.patch --]
[-- Type: text/plain, Size: 6893 bytes --]

Perform debug store selftests on each cpu.

Cover both the normal and the _noirq variant of the debug store interface.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds_selftest.c |  186 	137 +	49 -	0 !
 1 file changed, 137 insertions(+), 49 deletions(-)

Index: b/arch/x86/kernel/ds_selftest.c
===================================================================
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -11,13 +11,21 @@
 #include <linux/kernel.h>
 #include <linux/string.h>
 #include <linux/smp.h>
+#include <linux/cpu.h>
 
 #include <asm/ds.h>
 
 
-#define BUFFER_SIZE 1021 /* Intentionally chose an odd size. */
+#define BUFFER_SIZE	521 /* Intentionally chose an odd size. */
 
 
+struct ds_selftest_bts_conf {
+	struct bts_tracer *tracer;
+	int error;
+	int (*suspend)(struct bts_tracer *);
+	int (*resume)(struct bts_tracer *);
+};
+
 static int ds_selftest_bts_consistency(const struct bts_trace *trace)
 {
 	int error = 0;
@@ -125,36 +133,32 @@ static int ds_selftest_bts_read(struct b
 	return 0;
 }
 
-int ds_selftest_bts(void)
+static void ds_selftest_bts_cpu(void *arg)
 {
+	struct ds_selftest_bts_conf *conf = arg;
 	const struct bts_trace *trace;
-	struct bts_tracer *tracer;
-	int error = 0;
 	void *top;
-	unsigned char buffer[BUFFER_SIZE];
 
-	printk(KERN_INFO "[ds] bts selftest...");
-
-	tracer = ds_request_bts_cpu(smp_processor_id(), buffer, BUFFER_SIZE,
-				    NULL, (size_t)-1, BTS_KERNEL);
-	if (IS_ERR(tracer)) {
-		error = PTR_ERR(tracer);
-		tracer = NULL;
+	if (IS_ERR(conf->tracer)) {
+		conf->error = PTR_ERR(conf->tracer);
+		conf->tracer = NULL;
 
 		printk(KERN_CONT
-		       "initialization failed (err: %d)...", error);
-		goto out;
+		       "initialization failed (err: %d)...", conf->error);
+		return;
 	}
 
-	/* The return should already give us enough trace. */
-	ds_suspend_bts(tracer);
+	/* We should meanwhile have enough trace. */
+	conf->error = conf->suspend(conf->tracer);
+	if (conf->error < 0)
+		return;
 
 	/* Let's see if we can access the trace. */
-	trace = ds_read_bts(tracer);
+	trace = ds_read_bts(conf->tracer);
 
-	error = ds_selftest_bts_consistency(trace);
-	if (error < 0)
-		goto out;
+	conf->error = ds_selftest_bts_consistency(trace);
+	if (conf->error < 0)
+		return;
 
 	/* If everything went well, we should have a few trace entries. */
 	if (trace->ds.top == trace->ds.begin) {
@@ -168,10 +172,11 @@ int ds_selftest_bts(void)
 	}
 
 	/* Let's try to read the trace we collected. */
-	error = ds_selftest_bts_read(tracer, trace,
+	conf->error =
+		ds_selftest_bts_read(conf->tracer, trace,
 				     trace->ds.begin, trace->ds.top);
-	if (error < 0)
-		goto out;
+	if (conf->error < 0)
+		return;
 
 	/*
 	 * Let's read the trace again.
@@ -179,26 +184,31 @@ int ds_selftest_bts(void)
 	 */
 	top = trace->ds.top;
 
-	trace = ds_read_bts(tracer);
-	error = ds_selftest_bts_consistency(trace);
-	if (error < 0)
-		goto out;
+	trace = ds_read_bts(conf->tracer);
+	conf->error = ds_selftest_bts_consistency(trace);
+	if (conf->error < 0)
+		return;
 
 	if (top != trace->ds.top) {
 		printk(KERN_CONT "suspend not working...");
-		error = -1;
-		goto out;
+		conf->error = -1;
+		return;
 	}
 
 	/* Let's collect some more trace - see if resume is working. */
-	ds_resume_bts(tracer);
-	ds_suspend_bts(tracer);
-
-	trace = ds_read_bts(tracer);
-
-	error = ds_selftest_bts_consistency(trace);
-	if (error < 0)
-		goto out;
+	conf->error = conf->resume(conf->tracer);
+	if (conf->error < 0)
+		return;
+
+	conf->error = conf->suspend(conf->tracer);
+	if (conf->error < 0)
+		return;
+
+	trace = ds_read_bts(conf->tracer);
+
+	conf->error = ds_selftest_bts_consistency(trace);
+	if (conf->error < 0)
+		return;
 
 	if (trace->ds.top == top) {
 		/*
@@ -210,35 +220,113 @@ int ds_selftest_bts(void)
 		printk(KERN_CONT
 		       "no resume progress/overflow...");
 
-		error = ds_selftest_bts_read(tracer, trace,
+		conf->error =
+			ds_selftest_bts_read(conf->tracer, trace,
 					     trace->ds.begin, trace->ds.end);
 	} else if (trace->ds.top < top) {
 		/*
 		 * We had a buffer overflow - the entire buffer should
 		 * contain trace records.
 		 */
-		error = ds_selftest_bts_read(tracer, trace,
+		conf->error =
+			ds_selftest_bts_read(conf->tracer, trace,
 					     trace->ds.begin, trace->ds.end);
 	} else {
 		/*
 		 * It is quite likely that the buffer did not overflow.
 		 * Let's just check the delta trace.
 		 */
-		error = ds_selftest_bts_read(tracer, trace,
-					     top, trace->ds.top);
+		conf->error =
+			ds_selftest_bts_read(conf->tracer, trace, top,
+					     trace->ds.top);
 	}
-	if (error < 0)
-		goto out;
+	if (conf->error < 0)
+		return;
 
-	error = 0;
+	conf->error = 0;
+}
 
-	/* The final test: release the tracer while tracing is suspended. */
- out:
-	ds_release_bts(tracer);
+static int ds_suspend_bts_wrap(struct bts_tracer *tracer)
+{
+	ds_suspend_bts(tracer);
+	return 0;
+}
 
-	printk(KERN_CONT "%s.\n", (error ? "failed" : "passed"));
+static int ds_resume_bts_wrap(struct bts_tracer *tracer)
+{
+	ds_resume_bts(tracer);
+	return 0;
+}
 
-	return error;
+static void ds_release_bts_noirq_wrap(void *tracer)
+{
+	(void)ds_release_bts_noirq(tracer);
+}
+
+static int ds_selftest_bts_bad_release_noirq(int cpu,
+					     struct bts_tracer *tracer)
+{
+	int error = -EPERM;
+
+	/* Try to release the tracer on the wrong cpu. */
+	get_cpu();
+	if (cpu != smp_processor_id()) {
+		error = ds_release_bts_noirq(tracer);
+		if (error != -EPERM)
+			printk(KERN_CONT "release on wrong cpu...");
+	}
+	put_cpu();
+
+	return error ? 0 : -1;
+}
+
+int ds_selftest_bts(void)
+{
+	struct ds_selftest_bts_conf conf;
+	unsigned char buffer[BUFFER_SIZE];
+	int cpu;
+
+	printk(KERN_INFO "[ds] bts selftest...");
+	conf.error = 0;
+
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		conf.suspend = ds_suspend_bts_wrap;
+		conf.resume = ds_resume_bts_wrap;
+		conf.tracer =
+			ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
+					   NULL, (size_t)-1, BTS_KERNEL);
+		ds_selftest_bts_cpu(&conf);
+		ds_release_bts(conf.tracer);
+		if (conf.error < 0)
+			goto out;
+
+		conf.suspend = ds_suspend_bts_noirq;
+		conf.resume = ds_resume_bts_noirq;
+		conf.tracer =
+			ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
+					   NULL, (size_t)-1, BTS_KERNEL);
+		smp_call_function_single(cpu, ds_selftest_bts_cpu, &conf, 1);
+		if (conf.error >= 0) {
+			conf.error =
+				ds_selftest_bts_bad_release_noirq(cpu,
+								  conf.tracer);
+			/* We must not release the tracer twice. */
+			if (conf.error < 0)
+				conf.tracer = NULL;
+		}
+		smp_call_function_single(cpu, ds_release_bts_noirq_wrap,
+					 conf.tracer, 1);
+		if (conf.error < 0)
+			goto out;
+	}
+
+	conf.error = 0;
+ out:
+	put_online_cpus();
+	printk(KERN_CONT "%s.\n", (conf.error ? "failed" : "passed"));
+
+	return conf.error;
 }
 
 int ds_selftest_pebs(void)

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 13/20] x86, ds: add task tracing selftest

[markus.t.metzger]

[-- Attachment #1: tip.master.ds_selftest.current.patch --]
[-- Type: text/plain, Size: 3077 bytes --]

Add selftests to cover per-task branch tracing.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds_selftest.c |   71 	71 +	0 -	0 !
 1 file changed, 71 insertions(+)

Index: b/arch/x86/kernel/ds_selftest.c
===================================================================
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -280,10 +280,51 @@ static int ds_selftest_bts_bad_release_n
 	return error ? 0 : -1;
 }
 
+static int ds_selftest_bts_bad_request_cpu(int cpu, void *buffer)
+{
+	struct bts_tracer *tracer;
+	int error;
+
+	/* Try to request cpu tracing while task tracing is active. */
+	tracer = ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE, NULL,
+				    (size_t)-1, BTS_KERNEL);
+	error = PTR_ERR(tracer);
+	if (!IS_ERR(tracer)) {
+		ds_release_bts(tracer);
+		error = 0;
+	}
+
+	if (error != -EPERM)
+		printk(KERN_CONT "cpu/task tracing overlap...");
+
+	return error ? 0 : -1;
+}
+
+static int ds_selftest_bts_bad_request_task(void *buffer)
+{
+	struct bts_tracer *tracer;
+	int error;
+
+	/* Try to request cpu tracing while task tracing is active. */
+	tracer = ds_request_bts_task(current, buffer, BUFFER_SIZE, NULL,
+				    (size_t)-1, BTS_KERNEL);
+	error = PTR_ERR(tracer);
+	if (!IS_ERR(tracer)) {
+		error = 0;
+		ds_release_bts(tracer);
+	}
+
+	if (error != -EPERM)
+		printk(KERN_CONT "task/cpu tracing overlap...");
+
+	return error ? 0 : -1;
+}
+
 int ds_selftest_bts(void)
 {
 	struct ds_selftest_bts_conf conf;
 	unsigned char buffer[BUFFER_SIZE];
+	unsigned long irq;
 	int cpu;
 
 	printk(KERN_INFO "[ds] bts selftest...");
@@ -297,6 +338,8 @@ int ds_selftest_bts(void)
 			ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
 					   NULL, (size_t)-1, BTS_KERNEL);
 		ds_selftest_bts_cpu(&conf);
+		if (conf.error >= 0)
+			conf.error = ds_selftest_bts_bad_request_task(buffer);
 		ds_release_bts(conf.tracer);
 		if (conf.error < 0)
 			goto out;
@@ -315,12 +358,40 @@ int ds_selftest_bts(void)
 			if (conf.error < 0)
 				conf.tracer = NULL;
 		}
+		if (conf.error >= 0)
+			conf.error = ds_selftest_bts_bad_request_task(buffer);
 		smp_call_function_single(cpu, ds_release_bts_noirq_wrap,
 					 conf.tracer, 1);
 		if (conf.error < 0)
 			goto out;
 	}
 
+	conf.suspend = ds_suspend_bts_wrap;
+	conf.resume = ds_resume_bts_wrap;
+	conf.tracer =
+		ds_request_bts_task(current, buffer, BUFFER_SIZE,
+				    NULL, (size_t)-1, BTS_KERNEL);
+	ds_selftest_bts_cpu(&conf);
+	if (conf.error >= 0)
+		conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
+	ds_release_bts(conf.tracer);
+	if (conf.error < 0)
+		goto out;
+
+	conf.suspend = ds_suspend_bts_noirq;
+	conf.resume = ds_resume_bts_noirq;
+	conf.tracer =
+		ds_request_bts_task(current, buffer, BUFFER_SIZE,
+				   NULL, (size_t)-1, BTS_KERNEL);
+	local_irq_save(irq);
+	ds_selftest_bts_cpu(&conf);
+	if (conf.error >= 0)
+		conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
+	ds_release_bts_noirq(conf.tracer);
+	local_irq_restore(irq);
+	if (conf.error < 0)
+		goto out;
+
 	conf.error = 0;
  out:
 	put_online_cpus();

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 14/20] x86, ds: add leakage warning

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.warn_on_exit.patch --]
[-- Type: text/plain, Size: 4496 bytes --]

Add a warning in case a debug store context is not removed before
the task it is attached to is freed.

Remove the old warning at thread exit. It is too early.

Declare the debug store context field in thread_struct unconditionally.

Remove ds_copy_thread() and ds_exit_thread() and do the work directly
in process*.c.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/include/asm/ds.h        |    9 	0 +	9 -	0 !
 arch/x86/include/asm/processor.h |    4 	1 +	3 -	0 !
 arch/x86/kernel/ds.c             |   10 	0 +	10 -	0 !
 arch/x86/kernel/process.c        |    5 	3 +	2 -	0 !
 arch/x86/kernel/process_32.c     |    3 	2 +	1 -	0 !
 arch/x86/kernel/process_64.c     |    3 	2 +	1 -	0 !
 6 files changed, 8 insertions(+), 26 deletions(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -1352,16 +1352,6 @@ void ds_switch_to(struct task_struct *pr
 	update_debugctlmsr(debugctlmsr);
 }
 
-void ds_copy_thread(struct task_struct *tsk, struct task_struct *father)
-{
-	clear_tsk_thread_flag(tsk, TIF_DS_AREA_MSR);
-	tsk->thread.ds_ctx = NULL;
-}
-
-void ds_exit_thread(struct task_struct *tsk)
-{
-}
-
 static __init int ds_selftest(void)
 {
 	if (ds_cfg.sizeof_rec[ds_bts]) {
Index: b/arch/x86/include/asm/ds.h
===================================================================
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -285,21 +285,12 @@ extern void __cpuinit ds_init_intel(stru
  */
 extern void ds_switch_to(struct task_struct *prev, struct task_struct *next);
 
-/*
- * Task clone/init and cleanup work
- */
-extern void ds_copy_thread(struct task_struct *tsk, struct task_struct *father);
-extern void ds_exit_thread(struct task_struct *tsk);
-
 #else /* CONFIG_X86_DS */
 
 struct cpuinfo_x86;
 static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {}
 static inline void ds_switch_to(struct task_struct *prev,
 				struct task_struct *next) {}
-static inline void ds_copy_thread(struct task_struct *tsk,
-				  struct task_struct *father) {}
-static inline void ds_exit_thread(struct task_struct *tsk) {}
 
 #endif /* CONFIG_X86_DS */
 #endif /* _ASM_X86_DS_H */
Index: b/arch/x86/kernel/process.c
===================================================================
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -14,6 +14,7 @@
 #include <asm/idle.h>
 #include <asm/uaccess.h>
 #include <asm/i387.h>
+#include <asm/ds.h>
 
 unsigned long idle_halt;
 EXPORT_SYMBOL(idle_halt);
@@ -45,6 +46,8 @@ void free_thread_xstate(struct task_stru
 		kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
 		tsk->thread.xstate = NULL;
 	}
+
+	WARN(tsk->thread.ds_ctx, "leaking DS context\n");
 }
 
 void free_thread_info(struct thread_info *ti)
@@ -82,8 +85,6 @@ void exit_thread(void)
 		t->io_bitmap_max = 0;
 		put_cpu();
 	}
-
-	ds_exit_thread(current);
 }
 
 void flush_thread(void)
Index: b/arch/x86/include/asm/processor.h
===================================================================
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -451,10 +451,8 @@ struct thread_struct {
 	unsigned		io_bitmap_max;
 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
 	unsigned long	debugctlmsr;
-#ifdef CONFIG_X86_DS
-/* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
+	/* Debug Store context; see asm/ds.h */
 	struct ds_context	*ds_ctx;
-#endif /* CONFIG_X86_DS */
 };
 
 static inline unsigned long native_get_debugreg(int regno)
Index: b/arch/x86/kernel/process_32.c
===================================================================
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -290,7 +290,8 @@ int copy_thread(int nr, unsigned long cl
 		p->thread.io_bitmap_max = 0;
 	}
 
-	ds_copy_thread(p, current);
+	clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
+	p->thread.ds_ctx = NULL;
 
 	clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
 	p->thread.debugctlmsr = 0;
Index: b/arch/x86/kernel/process_64.c
===================================================================
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -335,7 +335,8 @@ int copy_thread(int nr, unsigned long cl
 			goto out;
 	}
 
-	ds_copy_thread(p, me);
+	clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
+	p->thread.ds_ctx = NULL;
 
 	clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
 	p->thread.debugctlmsr = 0;

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 15/20] x86, ds: use single debug store cpu configuration

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.single_configuration.patch --]
[-- Type: text/plain, Size: 1061 bytes --]

Use a single configuration for all cpus.


Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds.c |    7 	5 +	2 -	0 !
 1 file changed, 5 insertions(+), 2 deletions(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -47,9 +47,8 @@ struct ds_configuration {
 	/* Control bit-masks indexed by enum ds_feature: */
 	unsigned long		ctl[dsf_ctl_max];
 };
-static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array);
+static struct ds_configuration ds_cfg __read_mostly;
 
-#define ds_cfg per_cpu(ds_cfg_array, smp_processor_id())
 
 /* Maximal size of a DS configuration: */
 #define MAX_SIZEOF_DS		(12 * 8)
@@ -1268,6 +1267,10 @@ ds_configure(const struct ds_configurati
 
 void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
 {
+	/* Only configure the first cpu. Others are identical. */
+	if (ds_cfg.name)
+		return;
+
 	switch (c->x86) {
 	case 0x6:
 		switch (c->x86_model) {

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 16/20] x86, ptrace: add bts context unconditionally

[markus.t.metzger]

[-- Attachment #1: tip.master.ptrace.arch_init_task.patch --]
[-- Type: text/plain, Size: 5153 bytes --]

Add the ptrace bts context field to task_struct unconditionally.

Initialize the field directly in copy_process().
Remove all the unneeded functionality used to initialize that field.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/include/asm/ptrace.h |    9 	4 +	5 -	0 !
 arch/x86/kernel/ptrace.c      |   20 	1 +	19 -	0 !
 include/linux/ptrace.h        |   10 	0 +	10 -	0 !
 include/linux/sched.h         |    2 	0 +	2 -	0 !
 kernel/fork.c                 |    4 	2 +	2 -	0 !
 kernel/ptrace.c               |   10 	0 +	10 -	0 !
 6 files changed, 7 insertions(+), 48 deletions(-)

Index: b/arch/x86/include/asm/ptrace.h
===================================================================
--- a/arch/x86/include/asm/ptrace.h
+++ b/arch/x86/include/asm/ptrace.h
@@ -235,12 +235,11 @@ extern int do_get_thread_area(struct tas
 extern int do_set_thread_area(struct task_struct *p, int idx,
 			      struct user_desc __user *info, int can_allocate);
 
-extern void x86_ptrace_untrace(struct task_struct *);
-extern void x86_ptrace_fork(struct task_struct *child,
-			    unsigned long clone_flags);
+#ifdef CONFIG_X86_PTRACE_BTS
+extern void ptrace_bts_untrace(struct task_struct *tsk);
 
-#define arch_ptrace_untrace(tsk) x86_ptrace_untrace(tsk)
-#define arch_ptrace_fork(child, flags) x86_ptrace_fork(child, flags)
+#define arch_ptrace_untrace(tsk)	ptrace_bts_untrace(tsk)
+#endif /* CONFIG_X86_PTRACE_BTS */
 
 #endif /* __KERNEL__ */
 
Index: b/include/linux/ptrace.h
===================================================================
--- a/include/linux/ptrace.h
+++ b/include/linux/ptrace.h
@@ -94,7 +94,6 @@ extern void ptrace_notify(int exit_code)
 extern void __ptrace_link(struct task_struct *child,
 			  struct task_struct *new_parent);
 extern void __ptrace_unlink(struct task_struct *child);
-extern void ptrace_fork(struct task_struct *task, unsigned long clone_flags);
 #define PTRACE_MODE_READ   1
 #define PTRACE_MODE_ATTACH 2
 /* Returns 0 on success, -errno on denial. */
@@ -326,15 +325,6 @@ static inline void user_enable_block_ste
 #define arch_ptrace_untrace(task)		do { } while (0)
 #endif
 
-#ifndef arch_ptrace_fork
-/*
- * Do machine-specific work to initialize a new task.
- *
- * This is called from copy_process().
- */
-#define arch_ptrace_fork(child, clone_flags)	do { } while (0)
-#endif
-
 extern int task_current_syscall(struct task_struct *target, long *callno,
 				unsigned long args[6], unsigned int maxargs,
 				unsigned long *sp, unsigned long *pc);
Index: b/kernel/fork.c
===================================================================
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1100,8 +1100,8 @@ static struct task_struct *copy_process(
 #ifdef CONFIG_DEBUG_MUTEXES
 	p->blocked_on = NULL; /* not blocked yet */
 #endif
-	if (unlikely(current->ptrace))
-		ptrace_fork(p, clone_flags);
+
+	p->bts = NULL;
 
 	/* Perform scheduler related setup. Assign this task to a CPU. */
 	sched_fork(p, clone_flags);
Index: b/kernel/ptrace.c
===================================================================
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -27,16 +27,6 @@
 
 
 /*
- * Initialize a new task whose father had been ptraced.
- *
- * Called from copy_process().
- */
-void ptrace_fork(struct task_struct *child, unsigned long clone_flags)
-{
-	arch_ptrace_fork(child, clone_flags);
-}
-
-/*
  * ptrace a task: make the debugger its new parent and
  * move it to the ptrace list.
  *
Index: b/arch/x86/kernel/ptrace.c
===================================================================
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -888,37 +888,19 @@ static int ptrace_bts_size(struct task_s
 	return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 }
 
-static inline void ptrace_bts_fork(struct task_struct *tsk)
-{
-	tsk->bts = NULL;
-}
-
 /*
  * Called from __ptrace_unlink() after the child has been moved back
  * to its original parent.
  */
-static inline void ptrace_bts_untrace(struct task_struct *child)
+void ptrace_bts_untrace(struct task_struct *child)
 {
 	if (unlikely(child->bts)) {
 		free_bts_context(child->bts);
 		child->bts = NULL;
 	}
 }
-#else
-static inline void ptrace_bts_fork(struct task_struct *tsk) {}
-static inline void ptrace_bts_untrace(struct task_struct *child) {}
 #endif /* CONFIG_X86_PTRACE_BTS */
 
-void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
-{
-	ptrace_bts_fork(child);
-}
-
-void x86_ptrace_untrace(struct task_struct *child)
-{
-	ptrace_bts_untrace(child);
-}
-
 /*
  * Called by kernel/ptrace.c when detaching..
  *
Index: b/include/linux/sched.h
===================================================================
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1207,13 +1207,11 @@ struct task_struct {
 	struct list_head ptraced;
 	struct list_head ptrace_entry;
 
-#ifdef CONFIG_X86_PTRACE_BTS
 	/*
 	 * This is the tracer handle for the ptrace BTS extension.
 	 * This field actually belongs to the ptracer task.
 	 */
 	struct bts_context *bts;
-#endif /* CONFIG_X86_PTRACE_BTS */
 
 	/* PID/PID hash table linkage. */
 	struct pid_link pids[PIDTYPE_MAX];

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 17/20] x86, ds: dont use TIF_DEBUGCTLMSR

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.tif_debugctlmsr.patch --]
[-- Type: text/plain, Size: 716 bytes --]

Debug store already uses TIF_DS_AREA_MSR to trigger debug store context
switch handling. No need to use TIF_DEBUGCTLMSR, as well.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds.c |    5 	0 +	5 -	0 !
 1 file changed, 5 deletions(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -925,11 +925,6 @@ static void update_task_debugctlmsr(stru
 	get_cpu();
 	if (task == current)
 		update_debugctlmsr(debugctlmsr);
-
-	if (task->thread.debugctlmsr)
-		set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
-	else
-		clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
 	put_cpu();
 }
 

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 18/20] x86, ds: fix bad ds_reset_pebs()

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.reset_pebs.patch --]
[-- Type: text/plain, Size: 758 bytes --]

Ds_reset_pebs() passed the wrong qualifier to a shared function resulting
in a reset of bts, rather than pebs.


Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds.c |    2 	1 +	1 -	0 !
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -1186,7 +1186,7 @@ int ds_reset_pebs(struct pebs_tracer *tr
 
 	tracer->trace.ds.top = tracer->trace.ds.begin;
 
-	ds_set(tracer->ds.context->ds, ds_bts, ds_index,
+	ds_set(tracer->ds.context->ds, ds_pebs, ds_index,
 	       (unsigned long)tracer->trace.ds.top);
 
 	return 0;

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 19/20] x86, ds: allow small debug store buffers

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.small_buffers.patch --]
[-- Type: text/plain, Size: 2066 bytes --]

Check the buffer size more precisely to allow buffers for exactly one element
provided the base address is already properly aligned.

Add a debug store selftest.


Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/kernel/ds.c          |    9 	7 +	2 -	0 !
 arch/x86/kernel/ds_selftest.c |    6 	3 +	3 -	0 !
 2 files changed, 10 insertions(+), 5 deletions(-)

Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -656,6 +656,7 @@ static int ds_request(struct ds_tracer *
 {
 	struct ds_context *context;
 	int error;
+	size_t req_size;
 
 	error = -EOPNOTSUPP;
 	if (!ds_cfg.sizeof_rec[qual])
@@ -665,9 +666,13 @@ static int ds_request(struct ds_tracer *
 	if (!base)
 		goto out;
 
-	/* We need space for alignment adjustments in ds_init_ds_trace(). */
+	req_size = ds_cfg.sizeof_rec[qual];
+	/* We might need space for alignment adjustments. */
+	if (!IS_ALIGNED((unsigned long)base, DS_ALIGNMENT))
+		req_size += DS_ALIGNMENT;
+
 	error = -EINVAL;
-	if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual]))
+	if (size < req_size)
 		goto out;
 
 	if (th != (size_t)-1) {
Index: b/arch/x86/kernel/ds_selftest.c
===================================================================
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -16,8 +16,8 @@
 #include <asm/ds.h>
 
 
-#define BUFFER_SIZE	521 /* Intentionally chose an odd size. */
-
+#define BUFFER_SIZE		521	/* Intentionally chose an odd size. */
+#define SMALL_BUFFER_SIZE	24	/* A single bts entry. */
 
 struct ds_selftest_bts_conf {
 	struct bts_tracer *tracer;
@@ -381,7 +381,7 @@ int ds_selftest_bts(void)
 	conf.suspend = ds_suspend_bts_noirq;
 	conf.resume = ds_resume_bts_noirq;
 	conf.tracer =
-		ds_request_bts_task(current, buffer, BUFFER_SIZE,
+		ds_request_bts_task(current, buffer, SMALL_BUFFER_SIZE,
 				   NULL, (size_t)-1, BTS_KERNEL);
 	local_irq_save(irq);
 	ds_selftest_bts_cpu(&conf);

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

[patch 20/20] x86, ds: support Core i7

[markus.t.metzger]

[-- Attachment #1: tip.master.ds.i7.patch --]
[-- Type: text/plain, Size: 5599 bytes --]

Add debug store support for Core i7.

Core i7 adds a reset value for each performance counter and a new
PEBS record format.


Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
---
 arch/x86/include/asm/ds.h |   12 	9 +	3 -	0 !
 arch/x86/kernel/ds.c      |   69 	62 +	7 -	0 !
 2 files changed, 71 insertions(+), 10 deletions(-)

Index: b/arch/x86/include/asm/ds.h
===================================================================
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -234,8 +234,12 @@ struct bts_trace {
 struct pebs_trace {
 	struct ds_trace ds;
 
-	/* the PEBS reset value */
-	unsigned long long reset_value;
+	/* the number of valid counters in the below array */
+	unsigned int counters;
+
+#define MAX_PEBS_COUNTERS 4
+	/* the counter reset value */
+	unsigned long long counter_reset[MAX_PEBS_COUNTERS];
 };
 
 
@@ -270,9 +274,11 @@ extern int ds_reset_pebs(struct pebs_tra
  * Returns 0 on success; -Eerrno on error
  *
  * tracer: the tracer handle returned from ds_request_pebs()
+ * counter: the index of the counter
  * value: the new counter reset value
  */
-extern int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value);
+extern int ds_set_pebs_reset(struct pebs_tracer *tracer,
+			     unsigned int counter, u64 value);
 
 /*
  * Initialization
Index: b/arch/x86/kernel/ds.c
===================================================================
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -44,6 +44,9 @@ struct ds_configuration {
 	/* The size of a BTS/PEBS record in bytes: */
 	unsigned char		sizeof_rec[2];
 
+	/* The number of pebs counter reset values in the DS structure. */
+	unsigned char		nr_counter_reset;
+
 	/* Control bit-masks indexed by enum ds_feature: */
 	unsigned long		ctl[dsf_ctl_max];
 };
@@ -51,7 +54,7 @@ static struct ds_configuration ds_cfg __
 
 
 /* Maximal size of a DS configuration: */
-#define MAX_SIZEOF_DS		(12 * 8)
+#define MAX_SIZEOF_DS		0x80
 
 /* Maximal size of a BTS record: */
 #define MAX_SIZEOF_BTS		(3 * 8)
@@ -59,6 +62,12 @@ static struct ds_configuration ds_cfg __
 /* BTS and PEBS buffer alignment: */
 #define DS_ALIGNMENT		(1 << 3)
 
+/* Number of buffer pointers in DS: */
+#define NUM_DS_PTR_FIELDS	8
+
+/* Size of a pebs reset value in DS: */
+#define PEBS_RESET_FIELD_SIZE	8
+
 /* Mask of control bits in the DS MSR register: */
 #define BTS_CONTROL				  \
 	( ds_cfg.ctl[dsf_bts]			| \
@@ -1164,9 +1173,12 @@ const struct pebs_trace *ds_read_pebs(st
 		return NULL;
 
 	ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
-	tracer->trace.reset_value =
-		*(u64 *)(tracer->ds.context->ds +
-			 (ds_cfg.sizeof_ptr_field * 8));
+
+	tracer->trace.counters = ds_cfg.nr_counter_reset;
+	memcpy(tracer->trace.counter_reset,
+	       tracer->ds.context->ds +
+	       (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field),
+	       ds_cfg.nr_counter_reset * PEBS_RESET_FIELD_SIZE);
 
 	return &tracer->trace;
 }
@@ -1197,13 +1209,18 @@ int ds_reset_pebs(struct pebs_tracer *tr
 	return 0;
 }
 
-int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value)
+int ds_set_pebs_reset(struct pebs_tracer *tracer,
+		      unsigned int counter, u64 value)
 {
 	if (!tracer)
 		return -EINVAL;
 
+	if (ds_cfg.nr_counter_reset < counter)
+		return -EINVAL;
+
 	*(u64 *)(tracer->ds.context->ds +
-		 (ds_cfg.sizeof_ptr_field * 8)) = value;
+		 (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field) +
+		 (counter * PEBS_RESET_FIELD_SIZE)) = value;
 
 	return 0;
 }
@@ -1213,16 +1230,26 @@ static const struct ds_configuration ds_
 	.ctl[dsf_bts]		= (1 << 2) | (1 << 3),
 	.ctl[dsf_bts_kernel]	= (1 << 5),
 	.ctl[dsf_bts_user]	= (1 << 6),
+	.nr_counter_reset	= 1,
 };
 static const struct ds_configuration ds_cfg_pentium_m = {
 	.name = "Pentium M",
 	.ctl[dsf_bts]		= (1 << 6) | (1 << 7),
+	.nr_counter_reset	= 1,
 };
 static const struct ds_configuration ds_cfg_core2_atom = {
 	.name = "Core 2/Atom",
 	.ctl[dsf_bts]		= (1 << 6) | (1 << 7),
 	.ctl[dsf_bts_kernel]	= (1 << 9),
 	.ctl[dsf_bts_user]	= (1 << 10),
+	.nr_counter_reset	= 1,
+};
+static const struct ds_configuration ds_cfg_core_i7 = {
+	.name = "Core i7",
+	.ctl[dsf_bts]		= (1 << 6) | (1 << 7),
+	.ctl[dsf_bts_kernel]	= (1 << 9),
+	.ctl[dsf_bts_user]	= (1 << 10),
+	.nr_counter_reset	= 4,
 };
 
 static void
@@ -1239,6 +1266,32 @@ ds_configure(const struct ds_configurati
 	nr_pebs_fields = 18;
 #endif
 
+	/*
+	 * Starting with version 2, architectural performance
+	 * monitoring supports a format specifier.
+	 */
+	if ((cpuid_eax(0xa) & 0xff) > 1) {
+		unsigned long perf_capabilities, format;
+
+		rdmsrl(MSR_IA32_PERF_CAPABILITIES, perf_capabilities);
+
+		format = (perf_capabilities >> 8) & 0xf;
+
+		switch (format) {
+		case 0:
+			nr_pebs_fields = 18;
+			break;
+		case 1:
+			nr_pebs_fields = 22;
+			break;
+		default:
+			printk(KERN_INFO
+			       "[ds] unknown PEBS format: %lu\n", format);
+			nr_pebs_fields = 0;
+			break;
+		}
+	}
+
 	memset(&ds_cfg, 0, sizeof(ds_cfg));
 	ds_cfg = *cfg;
 
@@ -1262,7 +1315,7 @@ ds_configure(const struct ds_configurati
 	printk("bts/pebs record: %u/%u bytes\n",
 	       ds_cfg.sizeof_rec[ds_bts], ds_cfg.sizeof_rec[ds_pebs]);
 
-	WARN_ON_ONCE(MAX_SIZEOF_DS < (12 * ds_cfg.sizeof_ptr_field));
+	WARN_ON_ONCE(MAX_PEBS_COUNTERS < ds_cfg.nr_counter_reset);
 }
 
 void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
@@ -1284,6 +1337,8 @@ void __cpuinit ds_init_intel(struct cpui
 			ds_configure(&ds_cfg_core2_atom, c);
 			break;
 		case 0x1a: /* Core i7 */
+			ds_configure(&ds_cfg_core_i7, c);
+			break;
 		default:
 			/* Sorry, don't know about them. */
 			break;

-- 

[-- Attachment #2: Type: text/plain, Size: 656 bytes --]

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

Re: [patch 03/20] x86, ptrace, bts: defer branch trace stopping

[Peter Zijlstra]

On Fri, 2009-04-03 at 16:43 +0200, markus.t.metzger@intel.com wrote:
> Index: b/include/linux/mm.h
> ===================================================================
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -13,6 +13,7 @@
>  #include <linux/prio_tree.h>
>  #include <linux/debug_locks.h>
>  #include <linux/mm_types.h>
> +#include <linux/sched.h>

Why do we need that new include?

>  struct mempolicy;
>  struct anon_vma;
> @@ -1318,6 +1319,6 @@ void vmemmap_populate_print_last(void);
>  
>  extern void *alloc_locked_buffer(size_t size);
>  extern void free_locked_buffer(void *buffer, size_t size);
> -extern void release_locked_buffer(void *buffer, size_t size);
> +extern void refund_locked_buffer_memory(struct mm_struct *mm, size_t size);

that's a pointer to a struct, so it doesn't need a fully qualified type.

Also, I can't say I like the name, what about something like:

void account_locked_buffer(struct mm_struct *mm, long pages)
{
	down_write(&mm->mmap_sem);

	mm->total_vm += pages;
	mm->locked_vm += pages;

	up_write(&mm->mmap_sem);
}

but looking more closely at that alloc_locked_buffer() stuff, I really
hate it, who in his right mind does a multi-page kmalloc() -- that's
crazy.

Re: [patch 00/20] x86, ptrace, bts, hw-branch-tracer: fixes and cleanups

[Ingo Molnar]

* markus.t.metzger@intel.com <markus.t.metzger@intel.com> wrote:

> Patches 1-5 fix races with context switching code when the branch 
> traced task is currently running.
> 
> In the worst case, this might cause context switch code to access 
> freed memory or the tracing hardware to continue tracing into a 
> freed buffer. Both might crash the kernel.
> 
> The first 4 patches apply to .29 using the below preparation 
> patch.
> 
> The remaining patches fix bugs in the context of per-cpu tracing 
> (i.e. hw-branch-tracer) and pebs, add more selftest code, and do 
> some cleanups.

Thanks - this looks pretty acceptable. Latest -git changed a lot of 
code in the same area (-mm bits went upstream), creating a lot of 
conflicts.

To not prolongue this any longer (20 patches are difficult enough 
already to handle) i picked up your patches and resolved the 
conflicts in situ - mind having a look at the resulting 
tip:tracing/hw-branch-tracing branch - does the end result look sane 
to you?

It would also be nice to address Peter's feedback about the mm.h 
detail and the locked-pages API - but we can do that on top.

Thanks,

	Ingo

Re: [patch 00/20] x86, ptrace, bts, hw-branch-tracer: fixes and cleanups

[Markus Metzger]

On Fri, 2009-04-03 at 17:36 +0200, Ingo Molnar wrote: 
> * markus.t.metzger@intel.com <markus.t.metzger@intel.com> wrote:
> 
> > Patches 1-5 fix races with context switching code when the branch 
> > traced task is currently running.
> > 
> > In the worst case, this might cause context switch code to access 
> > freed memory or the tracing hardware to continue tracing into a 
> > freed buffer. Both might crash the kernel.
> > 
> > The first 4 patches apply to .29 using the below preparation 
> > patch.
> > 
> > The remaining patches fix bugs in the context of per-cpu tracing 
> > (i.e. hw-branch-tracer) and pebs, add more selftest code, and do 
> > some cleanups.
> 
> Thanks - this looks pretty acceptable. Latest -git changed a lot of 
> code in the same area (-mm bits went upstream), creating a lot of 
> conflicts.
> 
> To not prolongue this any longer (20 patches are difficult enough 
> already to handle) i picked up your patches and resolved the 
> conflicts in situ - mind having a look at the resulting 
> tip:tracing/hw-branch-tracing branch - does the end result look sane 
> to you?

Thanks,

The branch tracing bits in tip/tracing/hw-branch-tracing look good.
The kernel boots (after changes, see below) and my tests pass.

A defconfig build fails with a compile error, though:
In file included from net/core/skbuff.c:68:
include/trace/skb.h:7: error: expected ‘)’ before ‘(’ token
.....

Changing TP~ to TP_~ (as in tip/master) seems to do the trick.


> It would also be nice to address Peter's feedback about the mm.h 
> detail and the locked-pages API - but we can do that on top.

I will address them, though not immediately (vacation).


thanks and regards,
markus.

Re: [patch 00/20] x86, ptrace, bts, hw-branch-tracer: fixes and cleanups

[Ingo Molnar]

* Markus Metzger <markus.t.metzger@googlemail.com> wrote:

> On Fri, 2009-04-03 at 17:36 +0200, Ingo Molnar wrote: 
> > * markus.t.metzger@intel.com <markus.t.metzger@intel.com> wrote:
> > 
> > > Patches 1-5 fix races with context switching code when the branch 
> > > traced task is currently running.
> > > 
> > > In the worst case, this might cause context switch code to access 
> > > freed memory or the tracing hardware to continue tracing into a 
> > > freed buffer. Both might crash the kernel.
> > > 
> > > The first 4 patches apply to .29 using the below preparation 
> > > patch.
> > > 
> > > The remaining patches fix bugs in the context of per-cpu tracing 
> > > (i.e. hw-branch-tracer) and pebs, add more selftest code, and do 
> > > some cleanups.
> > 
> > Thanks - this looks pretty acceptable. Latest -git changed a lot of 
> > code in the same area (-mm bits went upstream), creating a lot of 
> > conflicts.
> > 
> > To not prolongue this any longer (20 patches are difficult enough 
> > already to handle) i picked up your patches and resolved the 
> > conflicts in situ - mind having a look at the resulting 
> > tip:tracing/hw-branch-tracing branch - does the end result look sane 
> > to you?
> 
> Thanks,
> 
> The branch tracing bits in tip/tracing/hw-branch-tracing look good.
> The kernel boots (after changes, see below) and my tests pass.
> 
> A defconfig build fails with a compile error, though:
> In file included from net/core/skbuff.c:68:
> include/trace/skb.h:7: error: expected ‘)’ before ‘(’ token
> .....
> 
> Changing TP~ to TP_~ (as in tip/master) seems to do the trick.

Ah. Darn - i should have merged tracing/core into it. That will ruin 
bisectability... I'll redo it.

	Ingo

RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping

[Metzger, Markus T]

>-----Original Message-----
>From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
>Sent: Friday, April 03, 2009 5:01 PM
>To: Metzger, Markus T


>Also, I can't say I like the name, what about something like:
>
>void account_locked_buffer(struct mm_struct *mm, long pages)
>{
>	down_write(&mm->mmap_sem);
>
>	mm->total_vm += pages;
>	mm->locked_vm += pages;
>
>	up_write(&mm->mmap_sem);
>}
>
>but looking more closely at that alloc_locked_buffer() stuff, I really
>hate it, who in his right mind does a multi-page kmalloc() -- that's
>crazy.

I need a non-pageable chunk of memory to give to the cpu to store branch
trace data in. Kmalloc() is easy to use and gives me what I need.

How would I do this correctly?

thanks,
markus.
---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping

[Peter Zijlstra]

On Sat, 2009-04-04 at 08:17 +0100, Metzger, Markus T wrote:
> >-----Original Message-----
> >From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
> >Sent: Friday, April 03, 2009 5:01 PM
> >To: Metzger, Markus T
> 
> 
> >Also, I can't say I like the name, what about something like:
> >
> >void account_locked_buffer(struct mm_struct *mm, long pages)
> >{
> >	down_write(&mm->mmap_sem);
> >
> >	mm->total_vm += pages;
> >	mm->locked_vm += pages;
> >
> >	up_write(&mm->mmap_sem);
> >}
> >
> >but looking more closely at that alloc_locked_buffer() stuff, I really
> >hate it, who in his right mind does a multi-page kmalloc() -- that's
> >crazy.
> 
> I need a non-pageable chunk of memory to give to the cpu to store branch
> trace data in. Kmalloc() is easy to use and gives me what I need.
> 
> How would I do this correctly?

Well, how large should this buffer be and must it be physically
contiguous?

Apparently its large enough to account in pages, that would suggest you
should use the page allocator to get memory.

Furthermore, if it needs to be physically contiguous and its more than
say 8 pages worth, you're basically up shit creek.

RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping

[Metzger, Markus T]

>-----Original Message-----
>From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
>Sent: Saturday, April 04, 2009 1:13 PM
>To: Metzger, Markus T
>Cc: mingo@elte.hu; tglx@linutronix.de; hpa@zytor.com; markus.t.metzger@gmail.com; roland@redhat.com;
>eranian@googlemail.com; oleg@redhat.com; Villacis, Juan; ak@linux.jf.intel.com; linux-
>kernel@vger.kernel.org; Andrew Morton
>Subject: RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping
>
>On Sat, 2009-04-04 at 08:17 +0100, Metzger, Markus T wrote:
>> >-----Original Message-----
>> >From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
>> >Sent: Friday, April 03, 2009 5:01 PM
>> >To: Metzger, Markus T
>>
>>
>> >Also, I can't say I like the name, what about something like:
>> >
>> >void account_locked_buffer(struct mm_struct *mm, long pages)
>> >{
>> >	down_write(&mm->mmap_sem);
>> >
>> >	mm->total_vm += pages;
>> >	mm->locked_vm += pages;
>> >
>> >	up_write(&mm->mmap_sem);
>> >}
>> >
>> >but looking more closely at that alloc_locked_buffer() stuff, I really
>> >hate it, who in his right mind does a multi-page kmalloc() -- that's
>> >crazy.
>>
>> I need a non-pageable chunk of memory to give to the cpu to store branch
>> trace data in. Kmalloc() is easy to use and gives me what I need.
>>
>> How would I do this correctly?
>
>Well, how large should this buffer be and must it be physically
>contiguous?

The size is set by the user. It is limited by (locked memory) RLIMIT.

The buffer need not be physically contiguous, but it must not be paged out
and should be marked accessed and dirty.
(see §18.7.8.2 in vol. 3b, Software Developer's Manual).

>Apparently its large enough to account in pages, that would suggest you
>should use the page allocator to get memory.

Wouldn't kmalloc forward the request to the page allocator for large memory?


>Furthermore, if it needs to be physically contiguous and its more than
>say 8 pages worth, you're basically up shit creek.

If not enough memory is available, I would expect users to request smaller
buffers until the request can be satisfied.


thanks and regards,
markus.
---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping

[Peter Zijlstra]

On Tue, 2009-04-07 at 09:12 +0100, Metzger, Markus T wrote:
> >-----Original Message-----
> >From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
> >Sent: Saturday, April 04, 2009 1:13 PM
> >To: Metzger, Markus T
> >Cc: mingo@elte.hu; tglx@linutronix.de; hpa@zytor.com; markus.t.metzger@gmail.com; roland@redhat.com;
> >eranian@googlemail.com; oleg@redhat.com; Villacis, Juan; ak@linux.jf.intel..com; linux-
> >kernel@vger.kernel.org; Andrew Morton
> >Subject: RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping
> >
> >On Sat, 2009-04-04 at 08:17 +0100, Metzger, Markus T wrote:
> >> >-----Original Message-----
> >> >From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
> >> >Sent: Friday, April 03, 2009 5:01 PM
> >> >To: Metzger, Markus T
> >>
> >>
> >> >Also, I can't say I like the name, what about something like:
> >> >
> >> >void account_locked_buffer(struct mm_struct *mm, long pages)
> >> >{
> >> >	down_write(&mm->mmap_sem);
> >> >
> >> >	mm->total_vm += pages;
> >> >	mm->locked_vm += pages;
> >> >
> >> >	up_write(&mm->mmap_sem);
> >> >}
> >> >
> >> >but looking more closely at that alloc_locked_buffer() stuff, I really
> >> >hate it, who in his right mind does a multi-page kmalloc() -- that's
> >> >crazy.
> >>
> >> I need a non-pageable chunk of memory to give to the cpu to store branch
> >> trace data in. Kmalloc() is easy to use and gives me what I need.
> >>
> >> How would I do this correctly?
> >
> >Well, how large should this buffer be and must it be physically
> >contiguous?
> 
> The size is set by the user. It is limited by (locked memory) RLIMIT.
> 
> The buffer need not be physically contiguous, but it must not be paged out
> and should be marked accessed and dirty.
> (see §18.7.8.2 in vol. 3b, Software Developer's Manual).

The Intel one I presume?

Ok, if you must mark the pages accessed and dirty you should get whole
pages, otherwise you cannot make that promise on the ptes, right?

> >Apparently its large enough to account in pages, that would suggest you
> >should use the page allocator to get memory.
> 
> Wouldn't kmalloc forward the request to the page allocator for large memory?

It may, but doesn't need to.

> >Furthermore, if it needs to be physically contiguous and its more than
> >say 8 pages worth, you're basically up shit creek.
> 
> If not enough memory is available, I would expect users to request smaller
> buffers until the request can be satisfied.

Sounds like a very bad interface, esp since you don't need physically
contiguous memory.

So what you need to do is allocate pages, order-0, grab a reference,
then vmap them into a linear piece, poke at the ptes to set the young
and dirty bits and provide that to your hardware.

RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping

[Metzger, Markus T]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 4323 bytes --]

>-----Original Message-----
>From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
>Sent: Tuesday, April 07, 2009 10:29 AM
>To: Metzger, Markus T
>Cc: mingo@elte.hu; tglx@linutronix.de; hpa@zytor.com; markus.t.metzger@gmail.com; roland@redhat.com;
>eranian@googlemail.com; oleg@redhat.com; Villacis, Juan; ak@linux.jf.intel.com; linux-
>kernel@vger.kernel.org; Andrew Morton
>Subject: RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping
>
>On Tue, 2009-04-07 at 09:12 +0100, Metzger, Markus T wrote:
>> >-----Original Message-----
>> >From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
>> >Sent: Saturday, April 04, 2009 1:13 PM
>> >To: Metzger, Markus T
>> >Cc: mingo@elte.hu; tglx@linutronix.de; hpa@zytor.com; markus.t.metzger@gmail.com;
>roland@redhat.com;
>> >eranian@googlemail.com; oleg@redhat.com; Villacis, Juan; ak@linux.jf.intel..com; linux-
>> >kernel@vger.kernel.org; Andrew Morton
>> >Subject: RE: [patch 03/20] x86, ptrace, bts: defer branch trace stopping
>> >
>> >On Sat, 2009-04-04 at 08:17 +0100, Metzger, Markus T wrote:
>> >> >-----Original Message-----
>> >> >From: Peter Zijlstra [mailto:a.p.zijlstra@chello.nl]
>> >> >Sent: Friday, April 03, 2009 5:01 PM
>> >> >To: Metzger, Markus T
>> >>
>> >>
>> >> >Also, I can't say I like the name, what about something like:
>> >> >
>> >> >void account_locked_buffer(struct mm_struct *mm, long pages)
>> >> >{
>> >> >	down_write(&mm->mmap_sem);
>> >> >
>> >> >	mm->total_vm += pages;
>> >> >	mm->locked_vm += pages;
>> >> >
>> >> >	up_write(&mm->mmap_sem);
>> >> >}
>> >> >
>> >> >but looking more closely at that alloc_locked_buffer() stuff, I really
>> >> >hate it, who in his right mind does a multi-page kmalloc() -- that's
>> >> >crazy.
>> >>
>> >> I need a non-pageable chunk of memory to give to the cpu to store branch
>> >> trace data in. Kmalloc() is easy to use and gives me what I need.
>> >>
>> >> How would I do this correctly?
>> >
>> >Well, how large should this buffer be and must it be physically
>> >contiguous?
>>
>> The size is set by the user. It is limited by (locked memory) RLIMIT.
>>
>> The buffer need not be physically contiguous, but it must not be paged out
>> and should be marked accessed and dirty.
>> (see §18.7.8.2 in vol. 3b, Software Developer's Manual).
>
>The Intel one I presume?

Yes. For x86 Architecture.


>Ok, if you must mark the pages accessed and dirty you should get whole
>pages, otherwise you cannot make that promise on the ptes, right?

The feedback I got so far was that kmalloc'ed memory is OK to use
in that case.


>> >Apparently its large enough to account in pages, that would suggest you
>> >should use the page allocator to get memory.
>>
>> Wouldn't kmalloc forward the request to the page allocator for large memory?
>
>It may, but doesn't need to.
>
>> >Furthermore, if it needs to be physically contiguous and its more than
>> >say 8 pages worth, you're basically up shit creek.
>>
>> If not enough memory is available, I would expect users to request smaller
>> buffers until the request can be satisfied.
>
>Sounds like a very bad interface, esp since you don't need physically
>contiguous memory.
>
>So what you need to do is allocate pages, order-0, grab a reference,
>then vmap them into a linear piece, poke at the ptes to set the young
>and dirty bits and provide that to your hardware.

OK.

Would a simple vmalloc() do the job?
Looks like it uses PAGE_KERNEL which has the present, accessed, and dirty bits
set. It also does the page allocation and vmap() for me.

thanks,
markus.

---------------------------------------------------------------------Intel GmbHDornacher Strasse 185622 Feldkirchen/Muenchen GermanySitz der Gesellschaft: Feldkirchen bei MuenchenGeschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes SchwadererRegistergericht: Muenchen HRB 47456 Ust.-IdNr.VAT Registration No.: DE129385895Citibank Frankfurt (BLZ 502 109 00) 600119052
This e-mail and any attachments may contain confidential material forthe sole use of the intended recipient(s). Any review or distributionby others is strictly prohibited. If you are not the intendedrecipient, please contact the sender and delete all copies.ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥

[tip:tracing/core] x86, bts: fix race when bts tracer is removed

[tip-bot for Markus Metzger]

Commit-ID:  cac94f979326212831c0ea44ed9ea1622b4f4e93
Gitweb:     http://git.kernel.org/tip/cac94f979326212831c0ea44ed9ea1622b4f4e93
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:33 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:10 +0200

x86, bts: fix race when bts tracer is removed

When the bts tracer is removed while the traced task is running,
the write to clear the bts tracer pointer races with context switch code.

Read the tracer once during a context switch.

When a new tracer is installed, the bts tracer is set in the ds context
before the tracer is initialized in order to claim the context for that
tracer.

This may result in write accesses using an uninitialized trace configuration
when scheduling timestamps have been requested.

Store active tracing flags separately and only set active flags after
the tracing configuration has been initialized.

Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144548.881338000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds.c |   58 +++++++++++++++++++++++++++++++------------------
 1 files changed, 37 insertions(+), 21 deletions(-)

diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index b1d6e1f..c730155 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -89,6 +89,9 @@ struct bts_tracer {
 
 	/* Buffer overflow notification function: */
 	bts_ovfl_callback_t	ovfl;
+
+	/* Active flags affecting trace collection. */
+	unsigned int		flags;
 };
 
 struct pebs_tracer {
@@ -799,6 +802,8 @@ void ds_suspend_bts(struct bts_tracer *tracer)
 	if (!tracer)
 		return;
 
+	tracer->flags = 0;
+
 	task = tracer->ds.context->task;
 
 	if (!task || (task == current))
@@ -820,6 +825,8 @@ void ds_resume_bts(struct bts_tracer *tracer)
 	if (!tracer)
 		return;
 
+	tracer->flags = tracer->trace.ds.flags;
+
 	task = tracer->ds.context->task;
 
 	control = ds_cfg.ctl[dsf_bts];
@@ -1037,43 +1044,52 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
 	}
 }
 
+static inline void ds_take_timestamp(struct ds_context *context,
+				     enum bts_qualifier qualifier,
+				     struct task_struct *task)
+{
+	struct bts_tracer *tracer = context->bts_master;
+	struct bts_struct ts;
+
+	/* Prevent compilers from reading the tracer pointer twice. */
+	barrier();
+
+	if (!tracer || !(tracer->flags & BTS_TIMESTAMPS))
+		return;
+
+	memset(&ts, 0, sizeof(ts));
+	ts.qualifier			= qualifier;
+	ts.variant.timestamp.jiffies	= jiffies_64;
+	ts.variant.timestamp.pid	= task->pid;
+
+	bts_write(tracer, &ts);
+}
+
 /*
  * Change the DS configuration from tracing prev to tracing next.
  */
 void ds_switch_to(struct task_struct *prev, struct task_struct *next)
 {
-	struct ds_context *prev_ctx = prev->thread.ds_ctx;
-	struct ds_context *next_ctx = next->thread.ds_ctx;
+	struct ds_context *prev_ctx	= prev->thread.ds_ctx;
+	struct ds_context *next_ctx	= next->thread.ds_ctx;
+	unsigned long debugctlmsr	= next->thread.debugctlmsr;
+
+	/* Make sure all data is read before we start. */
+	barrier();
 
 	if (prev_ctx) {
 		update_debugctlmsr(0);
 
-		if (prev_ctx->bts_master &&
-		    (prev_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
-			struct bts_struct ts = {
-				.qualifier = bts_task_departs,
-				.variant.timestamp.jiffies = jiffies_64,
-				.variant.timestamp.pid = prev->pid
-			};
-			bts_write(prev_ctx->bts_master, &ts);
-		}
+		ds_take_timestamp(prev_ctx, bts_task_departs, prev);
 	}
 
 	if (next_ctx) {
-		if (next_ctx->bts_master &&
-		    (next_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
-			struct bts_struct ts = {
-				.qualifier = bts_task_arrives,
-				.variant.timestamp.jiffies = jiffies_64,
-				.variant.timestamp.pid = next->pid
-			};
-			bts_write(next_ctx->bts_master, &ts);
-		}
+		ds_take_timestamp(next_ctx, bts_task_arrives, next);
 
 		wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds);
 	}
 
-	update_debugctlmsr(next->thread.debugctlmsr);
+	update_debugctlmsr(debugctlmsr);
 }
 
 void ds_copy_thread(struct task_struct *tsk, struct task_struct *father)

[tip:tracing/core] sched, hw-branch-tracer: add wait_task_context_switch() function to sched.h

[tip-bot for Markus Metzger]

Commit-ID:  a26b89f05d194413c7238e0bea071054f6b5d3c8
Gitweb:     http://git.kernel.org/tip/a26b89f05d194413c7238e0bea071054f6b5d3c8
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:34 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:12 +0200

sched, hw-branch-tracer: add wait_task_context_switch() function to sched.h

Add a function to wait until some other task has been
switched out at least once.

This differs from wait_task_inactive() subtly, in that the
latter will wait until the task has left the CPU.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: markus.t.metzger@gmail.com
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144549.794157000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 include/linux/sched.h |    2 ++
 kernel/sched.c        |   43 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 45 insertions(+), 0 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index b94f354..a5b9a83 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1993,8 +1993,10 @@ extern void set_task_comm(struct task_struct *tsk, char *from);
 extern char *get_task_comm(char *to, struct task_struct *tsk);
 
 #ifdef CONFIG_SMP
+extern void wait_task_context_switch(struct task_struct *p);
 extern unsigned long wait_task_inactive(struct task_struct *, long match_state);
 #else
+static inline void wait_task_context_switch(struct task_struct *p) {}
 static inline unsigned long wait_task_inactive(struct task_struct *p,
 					       long match_state)
 {
diff --git a/kernel/sched.c b/kernel/sched.c
index 6cc1fd5..f91bc81 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -2003,6 +2003,49 @@ migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req)
 }
 
 /*
+ * wait_task_context_switch -	wait for a thread to complete at least one
+ *				context switch.
+ *
+ * @p must not be current.
+ */
+void wait_task_context_switch(struct task_struct *p)
+{
+	unsigned long nvcsw, nivcsw, flags;
+	int running;
+	struct rq *rq;
+
+	nvcsw	= p->nvcsw;
+	nivcsw	= p->nivcsw;
+	for (;;) {
+		/*
+		 * The runqueue is assigned before the actual context
+		 * switch. We need to take the runqueue lock.
+		 *
+		 * We could check initially without the lock but it is
+		 * very likely that we need to take the lock in every
+		 * iteration.
+		 */
+		rq = task_rq_lock(p, &flags);
+		running = task_running(rq, p);
+		task_rq_unlock(rq, &flags);
+
+		if (likely(!running))
+			break;
+		/*
+		 * The switch count is incremented before the actual
+		 * context switch. We thus wait for two switches to be
+		 * sure at least one completed.
+		 */
+		if ((p->nvcsw - nvcsw) > 1)
+			break;
+		if ((p->nivcsw - nivcsw) > 1)
+			break;
+
+		cpu_relax();
+	}
+}
+
+/*
  * wait_task_inactive - wait for a thread to unschedule.
  *
  * If @match_state is nonzero, it's the @p->state value just checked and

[tip:tracing/core] mm, x86, ptrace, bts: defer branch trace stopping

[tip-bot for Markus Metzger]

Commit-ID:  e2b371f00a6f529f6362654239bdec8dcd510760
Gitweb:     http://git.kernel.org/tip/e2b371f00a6f529f6362654239bdec8dcd510760
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:35 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:13 +0200

mm, x86, ptrace, bts: defer branch trace stopping

When a ptraced task is unlinked, we need to stop branch tracing for
that task.

Since the unlink is called with interrupts disabled, and we need
interrupts enabled to stop branch tracing, we defer the work.

Collect all branch tracing related stuff in a branch tracing context.

Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144550.712401000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/include/asm/processor.h |    4 -
 arch/x86/kernel/ptrace.c         |  254 +++++++++++++++++++++++++-------------
 include/linux/mm.h               |    3 +-
 include/linux/sched.h            |    9 +-
 mm/mlock.c                       |   13 +-
 5 files changed, 179 insertions(+), 104 deletions(-)

diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 34c5237..2483807 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -458,10 +458,6 @@ struct thread_struct {
 /* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
 	struct ds_context	*ds_ctx;
 #endif /* CONFIG_X86_DS */
-#ifdef CONFIG_X86_PTRACE_BTS
-/* the signal to send on a bts buffer overflow */
-	unsigned int	bts_ovfl_signal;
-#endif /* CONFIG_X86_PTRACE_BTS */
 };
 
 static inline unsigned long native_get_debugreg(int regno)
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index fe9345c..7c21d1e 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -22,6 +22,7 @@
 #include <linux/seccomp.h>
 #include <linux/signal.h>
 #include <linux/ftrace.h>
+#include <linux/workqueue.h>
 
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
@@ -577,17 +578,119 @@ static int ioperm_get(struct task_struct *target,
 }
 
 #ifdef CONFIG_X86_PTRACE_BTS
+/*
+ * A branch trace store context.
+ *
+ * Contexts may only be installed by ptrace_bts_config() and only for
+ * ptraced tasks.
+ *
+ * Contexts are destroyed when the tracee is detached from the tracer.
+ * The actual destruction work requires interrupts enabled, so the
+ * work is deferred and will be scheduled during __ptrace_unlink().
+ *
+ * Contexts hold an additional task_struct reference on the traced
+ * task, as well as a reference on the tracer's mm.
+ *
+ * Ptrace already holds a task_struct for the duration of ptrace operations,
+ * but since destruction is deferred, it may be executed after both
+ * tracer and tracee exited.
+ */
+struct bts_context {
+	/* The branch trace handle. */
+	struct bts_tracer	*tracer;
+
+	/* The buffer used to store the branch trace and its size. */
+	void			*buffer;
+	unsigned int		size;
+
+	/* The mm that paid for the above buffer. */
+	struct mm_struct	*mm;
+
+	/* The task this context belongs to. */
+	struct task_struct	*task;
+
+	/* The signal to send on a bts buffer overflow. */
+	unsigned int		bts_ovfl_signal;
+
+	/* The work struct to destroy a context. */
+	struct work_struct	work;
+};
+
+static inline void alloc_bts_buffer(struct bts_context *context,
+				    unsigned int size)
+{
+	void *buffer;
+
+	buffer = alloc_locked_buffer(size);
+	if (buffer) {
+		context->buffer = buffer;
+		context->size = size;
+		context->mm = get_task_mm(current);
+	}
+}
+
+static inline void free_bts_buffer(struct bts_context *context)
+{
+	if (!context->buffer)
+		return;
+
+	kfree(context->buffer);
+	context->buffer = NULL;
+
+	refund_locked_buffer_memory(context->mm, context->size);
+	context->size = 0;
+
+	mmput(context->mm);
+	context->mm = NULL;
+}
+
+static void free_bts_context_work(struct work_struct *w)
+{
+	struct bts_context *context;
+
+	context = container_of(w, struct bts_context, work);
+
+	ds_release_bts(context->tracer);
+	put_task_struct(context->task);
+	free_bts_buffer(context);
+	kfree(context);
+}
+
+static inline void free_bts_context(struct bts_context *context)
+{
+	INIT_WORK(&context->work, free_bts_context_work);
+	schedule_work(&context->work);
+}
+
+static inline struct bts_context *alloc_bts_context(struct task_struct *task)
+{
+	struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
+	if (context) {
+		context->task = task;
+		task->bts = context;
+
+		get_task_struct(task);
+	}
+
+	return context;
+}
+
 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 				  struct bts_struct __user *out)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 	struct bts_struct bts;
 	const unsigned char *at;
 	int error;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	at = trace->ds.top - ((index + 1) * trace->ds.size);
 	if ((void *)at < trace->ds.begin)
@@ -596,7 +699,7 @@ static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 	if (!trace->read)
 		return -EOPNOTSUPP;
 
-	error = trace->read(child->bts, at, &bts);
+	error = trace->read(context->tracer, at, &bts);
 	if (error < 0)
 		return error;
 
@@ -610,13 +713,18 @@ static int ptrace_bts_drain(struct task_struct *child,
 			    long size,
 			    struct bts_struct __user *out)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 	const unsigned char *at;
 	int error, drained = 0;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	if (!trace->read)
 		return -EOPNOTSUPP;
@@ -627,9 +735,8 @@ static int ptrace_bts_drain(struct task_struct *child,
 	for (at = trace->ds.begin; (void *)at < trace->ds.top;
 	     out++, drained++, at += trace->ds.size) {
 		struct bts_struct bts;
-		int error;
 
-		error = trace->read(child->bts, at, &bts);
+		error = trace->read(context->tracer, at, &bts);
 		if (error < 0)
 			return error;
 
@@ -639,35 +746,18 @@ static int ptrace_bts_drain(struct task_struct *child,
 
 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 
-	error = ds_reset_bts(child->bts);
+	error = ds_reset_bts(context->tracer);
 	if (error < 0)
 		return error;
 
 	return drained;
 }
 
-static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size)
-{
-	child->bts_buffer = alloc_locked_buffer(size);
-	if (!child->bts_buffer)
-		return -ENOMEM;
-
-	child->bts_size = size;
-
-	return 0;
-}
-
-static void ptrace_bts_free_buffer(struct task_struct *child)
-{
-	free_locked_buffer(child->bts_buffer, child->bts_size);
-	child->bts_buffer = NULL;
-	child->bts_size = 0;
-}
-
 static int ptrace_bts_config(struct task_struct *child,
 			     long cfg_size,
 			     const struct ptrace_bts_config __user *ucfg)
 {
+	struct bts_context *context;
 	struct ptrace_bts_config cfg;
 	unsigned int flags = 0;
 
@@ -677,28 +767,31 @@ static int ptrace_bts_config(struct task_struct *child,
 	if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 		return -EFAULT;
 
-	if (child->bts) {
-		ds_release_bts(child->bts);
-		child->bts = NULL;
-	}
+	context = child->bts;
+	if (!context)
+		context = alloc_bts_context(child);
+	if (!context)
+		return -ENOMEM;
 
 	if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 		if (!cfg.signal)
 			return -EINVAL;
 
-		child->thread.bts_ovfl_signal = cfg.signal;
 		return -EOPNOTSUPP;
+		context->bts_ovfl_signal = cfg.signal;
 	}
 
-	if ((cfg.flags & PTRACE_BTS_O_ALLOC) &&
-	    (cfg.size != child->bts_size)) {
-		int error;
+	ds_release_bts(context->tracer);
+	context->tracer = NULL;
 
-		ptrace_bts_free_buffer(child);
+	if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
+		free_bts_buffer(context);
+		if (!cfg.size)
+			return 0;
 
-		error = ptrace_bts_allocate_buffer(child, cfg.size);
-		if (error < 0)
-			return error;
+		alloc_bts_buffer(context, cfg.size);
+		if (!context->buffer)
+			return -ENOMEM;
 	}
 
 	if (cfg.flags & PTRACE_BTS_O_TRACE)
@@ -707,15 +800,13 @@ static int ptrace_bts_config(struct task_struct *child,
 	if (cfg.flags & PTRACE_BTS_O_SCHED)
 		flags |= BTS_TIMESTAMPS;
 
-	child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size,
-				    /* ovfl = */ NULL, /* th = */ (size_t)-1,
-				    flags);
-	if (IS_ERR(child->bts)) {
-		int error = PTR_ERR(child->bts);
-
-		ptrace_bts_free_buffer(child);
-		child->bts = NULL;
+	context->tracer = ds_request_bts(child, context->buffer, context->size,
+					 NULL, (size_t)-1, flags);
+	if (unlikely(IS_ERR(context->tracer))) {
+		int error = PTR_ERR(context->tracer);
 
+		free_bts_buffer(context);
+		context->tracer = NULL;
 		return error;
 	}
 
@@ -726,20 +817,25 @@ static int ptrace_bts_status(struct task_struct *child,
 			     long cfg_size,
 			     struct ptrace_bts_config __user *ucfg)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 	struct ptrace_bts_config cfg;
 
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
 	if (cfg_size < sizeof(cfg))
 		return -EIO;
 
-	trace = ds_read_bts(child->bts);
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	memset(&cfg, 0, sizeof(cfg));
-	cfg.size = trace->ds.end - trace->ds.begin;
-	cfg.signal = child->thread.bts_ovfl_signal;
-	cfg.bts_size = sizeof(struct bts_struct);
+	cfg.size	= trace->ds.end - trace->ds.begin;
+	cfg.signal	= context->bts_ovfl_signal;
+	cfg.bts_size	= sizeof(struct bts_struct);
 
 	if (cfg.signal)
 		cfg.flags |= PTRACE_BTS_O_SIGNAL;
@@ -758,67 +854,56 @@ static int ptrace_bts_status(struct task_struct *child,
 
 static int ptrace_bts_clear(struct task_struct *child)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 
-	return ds_reset_bts(child->bts);
+	return ds_reset_bts(context->tracer);
 }
 
 static int ptrace_bts_size(struct task_struct *child)
 {
+	struct bts_context *context;
 	const struct bts_trace *trace;
 
-	trace = ds_read_bts(child->bts);
+	context = child->bts;
+	if (!context)
+		return -ESRCH;
+
+	trace = ds_read_bts(context->tracer);
 	if (!trace)
-		return -EPERM;
+		return -ESRCH;
 
 	return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 }
 
-static void ptrace_bts_fork(struct task_struct *tsk)
+static inline void ptrace_bts_fork(struct task_struct *tsk)
 {
 	tsk->bts = NULL;
-	tsk->bts_buffer = NULL;
-	tsk->bts_size = 0;
-	tsk->thread.bts_ovfl_signal = 0;
 }
 
-static void ptrace_bts_untrace(struct task_struct *child)
+/*
+ * Called from __ptrace_unlink() after the child has been moved back
+ * to its original parent.
+ */
+static inline void ptrace_bts_untrace(struct task_struct *child)
 {
 	if (unlikely(child->bts)) {
-		ds_release_bts(child->bts);
+		free_bts_context(child->bts);
 		child->bts = NULL;
-
-		/* We cannot update total_vm and locked_vm since
-		   child's mm is already gone. But we can reclaim the
-		   memory. */
-		kfree(child->bts_buffer);
-		child->bts_buffer = NULL;
-		child->bts_size = 0;
 	}
 }
-
-static void ptrace_bts_detach(struct task_struct *child)
-{
-	/*
-	 * Ptrace_detach() races with ptrace_untrace() in case
-	 * the child dies and is reaped by another thread.
-	 *
-	 * We only do the memory accounting at this point and
-	 * leave the buffer deallocation and the bts tracer
-	 * release to ptrace_bts_untrace() which will be called
-	 * later on with tasklist_lock held.
-	 */
-	release_locked_buffer(child->bts_buffer, child->bts_size);
-}
 #else
 static inline void ptrace_bts_fork(struct task_struct *tsk) {}
-static inline void ptrace_bts_detach(struct task_struct *child) {}
 static inline void ptrace_bts_untrace(struct task_struct *child) {}
 #endif /* CONFIG_X86_PTRACE_BTS */
 
@@ -843,7 +928,6 @@ void ptrace_disable(struct task_struct *child)
 #ifdef TIF_SYSCALL_EMU
 	clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 #endif
-	ptrace_bts_detach(child);
 }
 
 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
diff --git a/include/linux/mm.h b/include/linux/mm.h
index bff1f0d..64d8ed2 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -13,6 +13,7 @@
 #include <linux/prio_tree.h>
 #include <linux/debug_locks.h>
 #include <linux/mm_types.h>
+#include <linux/sched.h>
 
 struct mempolicy;
 struct anon_vma;
@@ -1321,6 +1322,6 @@ void vmemmap_populate_print_last(void);
 
 extern void *alloc_locked_buffer(size_t size);
 extern void free_locked_buffer(void *buffer, size_t size);
-extern void release_locked_buffer(void *buffer, size_t size);
+extern void refund_locked_buffer_memory(struct mm_struct *mm, size_t size);
 #endif /* __KERNEL__ */
 #endif /* _LINUX_MM_H */
diff --git a/include/linux/sched.h b/include/linux/sched.h
index a5b9a83..52b8cd0 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -96,8 +96,8 @@ struct exec_domain;
 struct futex_pi_state;
 struct robust_list_head;
 struct bio;
-struct bts_tracer;
 struct fs_struct;
+struct bts_context;
 
 /*
  * List of flags we want to share for kernel threads,
@@ -1210,12 +1210,7 @@ struct task_struct {
 	 * This is the tracer handle for the ptrace BTS extension.
 	 * This field actually belongs to the ptracer task.
 	 */
-	struct bts_tracer *bts;
-	/*
-	 * The buffer to hold the BTS data.
-	 */
-	void *bts_buffer;
-	size_t bts_size;
+	struct bts_context *bts;
 #endif /* CONFIG_X86_PTRACE_BTS */
 
 	/* PID/PID hash table linkage. */
diff --git a/mm/mlock.c b/mm/mlock.c
index cbe9e05..749383b 100644
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -660,21 +660,20 @@ void *alloc_locked_buffer(size_t size)
 	return buffer;
 }
 
-void release_locked_buffer(void *buffer, size_t size)
+void refund_locked_buffer_memory(struct mm_struct *mm, size_t size)
 {
 	unsigned long pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
 
-	down_write(&current->mm->mmap_sem);
+	down_write(&mm->mmap_sem);
 
-	current->mm->total_vm  -= pgsz;
-	current->mm->locked_vm -= pgsz;
+	mm->total_vm  -= pgsz;
+	mm->locked_vm -= pgsz;
 
-	up_write(&current->mm->mmap_sem);
+	up_write(&mm->mmap_sem);
 }
 
 void free_locked_buffer(void *buffer, size_t size)
 {
-	release_locked_buffer(buffer, size);
-
+	refund_locked_buffer_memory(current->mm, size);
 	kfree(buffer);
 }

[tip:tracing/core] x86, bts: wait until traced task has been scheduled out

[tip-bot for Markus Metzger]

Commit-ID:  8d99b3ac2726e5edd97ad147fa5c1f2acb63a745
Gitweb:     http://git.kernel.org/tip/8d99b3ac2726e5edd97ad147fa5c1f2acb63a745
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:36 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:15 +0200

x86, bts: wait until traced task has been scheduled out

In order to stop branch tracing for a running task, we need to first
clear the branch tracing control bits before we may free the tracing
buffer.

If the traced task is running, the cpu might still trace that task
after the branch trace control bits have cleared.

Wait until the traced task has been scheduled out before proceeding.

A similar problem affects the task debug store context. We first remove
the context, then we need to wait until the task has been scheduled
out before we can free the context memory.

Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144551.919636000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds.c |   23 +++++++++++++++++++----
 1 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index c730155..5cd137a 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -299,6 +299,7 @@ static inline struct ds_context *ds_get_context(struct task_struct *task)
 
 static inline void ds_put_context(struct ds_context *context)
 {
+	struct task_struct *task;
 	unsigned long irq;
 
 	if (!context)
@@ -313,14 +314,20 @@ static inline void ds_put_context(struct ds_context *context)
 
 	*(context->this) = NULL;
 
-	if (context->task)
-		clear_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
+	task = context->task;
+
+	if (task)
+		clear_tsk_thread_flag(task, TIF_DS_AREA_MSR);
 
-	if (!context->task || (context->task == current))
+	if (!task || (task == current))
 		wrmsrl(MSR_IA32_DS_AREA, 0);
 
 	spin_unlock_irqrestore(&ds_lock, irq);
 
+	/* The context might still be in use for context switching. */
+	if (task && (task != current))
+		wait_task_context_switch(task);
+
 	kfree(context);
 }
 
@@ -781,15 +788,23 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 
 void ds_release_bts(struct bts_tracer *tracer)
 {
+	struct task_struct *task;
+
 	if (!tracer)
 		return;
 
+	task = tracer->ds.context->task;
+
 	ds_suspend_bts(tracer);
 
 	WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
 	tracer->ds.context->bts_master = NULL;
 
-	put_tracer(tracer->ds.context->task);
+	/* Make sure tracing stopped and the tracer is not in use. */
+	if (task && (task != current))
+		wait_task_context_switch(task);
+
+	put_tracer(task);
 	ds_put_context(tracer->ds.context);
 
 	kfree(tracer);

[tip:tracing/core] x86, bts: fix race between per-task and per-cpu branch tracing

[tip-bot for Markus Metzger]

Commit-ID:  38f801129ad07b9afa7f9bd3779f61b805416d8c
Gitweb:     http://git.kernel.org/tip/38f801129ad07b9afa7f9bd3779f61b805416d8c
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:37 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:16 +0200

x86, bts: fix race between per-task and per-cpu branch tracing

Per-task branch tracing installs a debug store context with the traced
task. This immediately results in the branch trace control bits to be
cleared for the next context switch of that task, if not set before.

Either per-cpu or per-task tracing are allowed at the same time.

An active per-cpu tracing would be disabled even if the per-task tracing
request is rejected and the task debug store context removed.

Check the tracing type (per-cpu or per-task) before installing a task
debug store context.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144552.856000000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds.c |   72 ++++++++++++++++++++++++++++---------------------
 1 files changed, 41 insertions(+), 31 deletions(-)

diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 5cd137a..f03f117 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -193,12 +193,28 @@ static DEFINE_SPINLOCK(ds_lock);
  */
 static atomic_t tracers = ATOMIC_INIT(0);
 
-static inline void get_tracer(struct task_struct *task)
+static inline int get_tracer(struct task_struct *task)
 {
-	if (task)
+	int error;
+
+	spin_lock_irq(&ds_lock);
+
+	if (task) {
+		error = -EPERM;
+		if (atomic_read(&tracers) < 0)
+			goto out;
 		atomic_inc(&tracers);
-	else
+	} else {
+		error = -EPERM;
+		if (atomic_read(&tracers) > 0)
+			goto out;
 		atomic_dec(&tracers);
+	}
+
+	error = 0;
+out:
+	spin_unlock_irq(&ds_lock);
+	return error;
 }
 
 static inline void put_tracer(struct task_struct *task)
@@ -209,14 +225,6 @@ static inline void put_tracer(struct task_struct *task)
 		atomic_inc(&tracers);
 }
 
-static inline int check_tracer(struct task_struct *task)
-{
-	return task ?
-		(atomic_read(&tracers) >= 0) :
-		(atomic_read(&tracers) <= 0);
-}
-
-
 /*
  * The DS context is either attached to a thread or to a cpu:
  * - in the former case, the thread_struct contains a pointer to the
@@ -677,6 +685,10 @@ struct bts_tracer *ds_request_bts(struct task_struct *task,
 	if (ovfl)
 		goto out;
 
+	error = get_tracer(task);
+	if (error < 0)
+		goto out;
+
 	/*
 	 * Per-cpu tracing is typically requested using smp_call_function().
 	 * We must not sleep.
@@ -684,7 +696,7 @@ struct bts_tracer *ds_request_bts(struct task_struct *task,
 	error = -ENOMEM;
 	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
 	if (!tracer)
-		goto out;
+		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
@@ -696,13 +708,8 @@ struct bts_tracer *ds_request_bts(struct task_struct *task,
 	spin_lock_irqsave(&ds_lock, irq);
 
 	error = -EPERM;
-	if (!check_tracer(task))
-		goto out_unlock;
-	get_tracer(task);
-
-	error = -EPERM;
 	if (tracer->ds.context->bts_master)
-		goto out_put_tracer;
+		goto out_unlock;
 	tracer->ds.context->bts_master = tracer;
 
 	spin_unlock_irqrestore(&ds_lock, irq);
@@ -716,13 +723,13 @@ struct bts_tracer *ds_request_bts(struct task_struct *task,
 
 	return tracer;
 
- out_put_tracer:
-	put_tracer(task);
  out_unlock:
 	spin_unlock_irqrestore(&ds_lock, irq);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
+ out_put_tracer:
+	put_tracer(task);
  out:
 	return ERR_PTR(error);
 }
@@ -741,6 +748,10 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 	if (ovfl)
 		goto out;
 
+	error = get_tracer(task);
+	if (error < 0)
+		goto out;
+
 	/*
 	 * Per-cpu tracing is typically requested using smp_call_function().
 	 * We must not sleep.
@@ -748,7 +759,7 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 	error = -ENOMEM;
 	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
 	if (!tracer)
-		goto out;
+		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
@@ -759,13 +770,8 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 	spin_lock_irqsave(&ds_lock, irq);
 
 	error = -EPERM;
-	if (!check_tracer(task))
-		goto out_unlock;
-	get_tracer(task);
-
-	error = -EPERM;
 	if (tracer->ds.context->pebs_master)
-		goto out_put_tracer;
+		goto out_unlock;
 	tracer->ds.context->pebs_master = tracer;
 
 	spin_unlock_irqrestore(&ds_lock, irq);
@@ -775,13 +781,13 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 
 	return tracer;
 
- out_put_tracer:
-	put_tracer(task);
  out_unlock:
 	spin_unlock_irqrestore(&ds_lock, irq);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
+ out_put_tracer:
+	put_tracer(task);
  out:
 	return ERR_PTR(error);
 }
@@ -804,8 +810,8 @@ void ds_release_bts(struct bts_tracer *tracer)
 	if (task && (task != current))
 		wait_task_context_switch(task);
 
-	put_tracer(task);
 	ds_put_context(tracer->ds.context);
+	put_tracer(task);
 
 	kfree(tracer);
 }
@@ -861,16 +867,20 @@ void ds_resume_bts(struct bts_tracer *tracer)
 
 void ds_release_pebs(struct pebs_tracer *tracer)
 {
+	struct task_struct *task;
+
 	if (!tracer)
 		return;
 
+	task = tracer->ds.context->task;
+
 	ds_suspend_pebs(tracer);
 
 	WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
 	tracer->ds.context->pebs_master = NULL;
 
-	put_tracer(tracer->ds.context->task);
 	ds_put_context(tracer->ds.context);
+	put_tracer(task);
 
 	kfree(tracer);
 }

[tip:tracing/core] x86, bts: use trace_clock_global() for timestamps

[tip-bot for Markus Metzger]

Commit-ID:  15879d042164650b93d83281ad5f87ad323bfbfe
Gitweb:     http://git.kernel.org/tip/15879d042164650b93d83281ad5f87ad323bfbfe
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:38 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:17 +0200

x86, bts: use trace_clock_global() for timestamps

Rename the bts_struct timestamp field to event.

Use trace_clock_global() for time measurement.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144553.773216000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/include/asm/ds.h |    4 ++--
 arch/x86/kernel/ds.c      |   17 +++++++++--------
 2 files changed, 11 insertions(+), 10 deletions(-)

diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h
index a8f672b..772f141 100644
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -170,9 +170,9 @@ struct bts_struct {
 		} lbr;
 		/* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */
 		struct {
-			__u64 jiffies;
+			__u64 clock;
 			pid_t pid;
-		} timestamp;
+		} event;
 	} variant;
 };
 
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index f03f117..2071b99 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -25,6 +25,7 @@
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/mm.h>
+#include <linux/trace_clock.h>
 
 #include <asm/ds.h>
 
@@ -471,7 +472,7 @@ enum bts_field {
 	bts_flags,
 
 	bts_qual		= bts_from,
-	bts_jiffies		= bts_to,
+	bts_clock		= bts_to,
 	bts_pid			= bts_flags,
 
 	bts_qual_mask		= (bts_qual_max - 1),
@@ -517,8 +518,8 @@ bts_read(struct bts_tracer *tracer, const void *at, struct bts_struct *out)
 	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);
+		out->variant.event.clock = bts_get(at, bts_clock);
+		out->variant.event.pid = bts_get(at, bts_pid);
 	} else {
 		out->qualifier = bts_branch;
 		out->variant.lbr.from = bts_get(at, bts_from);
@@ -555,8 +556,8 @@ static int bts_write(struct bts_tracer *tracer, const struct bts_struct *in)
 	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);
+		bts_set(raw, bts_clock, in->variant.event.clock);
+		bts_set(raw, bts_pid, in->variant.event.pid);
 		break;
 	default:
 		return -EINVAL;
@@ -1083,9 +1084,9 @@ static inline void ds_take_timestamp(struct ds_context *context,
 		return;
 
 	memset(&ts, 0, sizeof(ts));
-	ts.qualifier			= qualifier;
-	ts.variant.timestamp.jiffies	= jiffies_64;
-	ts.variant.timestamp.pid	= task->pid;
+	ts.qualifier		= qualifier;
+	ts.variant.event.clock	= trace_clock_global();
+	ts.variant.event.pid	= task->pid;
 
 	bts_write(tracer, &ts);
 }

[tip:tracing/core] x86, debugctlmsr: add _on_cpu variants to debugctlmsr functions

[tip-bot for Markus Metzger]

Commit-ID:  35bb7600c17762bb129588c1877d2717fe325289
Gitweb:     http://git.kernel.org/tip/35bb7600c17762bb129588c1877d2717fe325289
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:39 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:19 +0200

x86, debugctlmsr: add _on_cpu variants to debugctlmsr functions

Add functions to get and set the debugctlmsr on different cpus.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144554.738772000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/include/asm/processor.h |   27 +++++++++++++++++++++++++++
 1 files changed, 27 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 2483807..1efeb49 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -785,6 +785,21 @@ static inline unsigned long get_debugctlmsr(void)
     return debugctlmsr;
 }
 
+static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
+{
+	u64 debugctlmsr = 0;
+	u32 val1, val2;
+
+#ifndef CONFIG_X86_DEBUGCTLMSR
+	if (boot_cpu_data.x86 < 6)
+		return 0;
+#endif
+	rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
+	debugctlmsr = val1 | ((u64)val2 << 32);
+
+	return debugctlmsr;
+}
+
 static inline void update_debugctlmsr(unsigned long debugctlmsr)
 {
 #ifndef CONFIG_X86_DEBUGCTLMSR
@@ -794,6 +809,18 @@ static inline void update_debugctlmsr(unsigned long debugctlmsr)
 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
 }
 
+static inline void update_debugctlmsr_on_cpu(int cpu,
+					     unsigned long debugctlmsr)
+{
+#ifndef CONFIG_X86_DEBUGCTLMSR
+	if (boot_cpu_data.x86 < 6)
+		return;
+#endif
+	wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
+		     (u32)((u64)debugctlmsr),
+		     (u32)((u64)debugctlmsr >> 32));
+}
+
 /*
  * from system description table in BIOS. Mostly for MCA use, but
  * others may find it useful:

[tip:tracing/core] x86, bts, hw-branch-tracer: add _noirq variants to the debug store interface

[tip-bot for Markus Metzger]

Commit-ID:  de79f54f5347ad7ec6ff55ccbb6d4ab2a21f6a93
Gitweb:     http://git.kernel.org/tip/de79f54f5347ad7ec6ff55ccbb6d4ab2a21f6a93
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:40 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:20 +0200

x86, bts, hw-branch-tracer: add _noirq variants to the debug store interface

The hw-branch-tracer uses debug store functions from an on_each_cpu()
context, which is simply wrong since the functions may sleep.

Add _noirq variants for most functions, which  may be called with
interrupts disabled.

Separate per-cpu and per-task tracing and allow per-cpu tracing to be
controlled from any cpu.

Make the hw-branch-tracer use the new debug store interface, synchronize
with hotplug cpu event using get/put_online_cpus(), and remove the
unnecessary spinlock.

Make the ptrace bts and the ds selftest code use the new interface.

Defer the ds selftest.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144555.658136000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/include/asm/ds.h        |   57 ++++-
 arch/x86/kernel/ds.c             |  474 +++++++++++++++++++++++++++++---------
 arch/x86/kernel/ds_selftest.c    |    9 +-
 arch/x86/kernel/ptrace.c         |    5 +-
 kernel/trace/trace_hw_branches.c |  193 ++++++----------
 5 files changed, 492 insertions(+), 246 deletions(-)

diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h
index 772f141..413e127 100644
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -15,8 +15,8 @@
  * - buffer allocation (memory accounting)
  *
  *
- * Copyright (C) 2007-2008 Intel Corporation.
- * Markus Metzger <markus.t.metzger@intel.com>, 2007-2008
+ * Copyright (C) 2007-2009 Intel Corporation.
+ * Markus Metzger <markus.t.metzger@intel.com>, 2007-2009
  */
 
 #ifndef _ASM_X86_DS_H
@@ -83,8 +83,10 @@ enum ds_feature {
  * The interrupt threshold is independent from the overflow callback
  * to allow users to use their own overflow interrupt handling mechanism.
  *
- * task: the task to request recording for;
- *       NULL for per-cpu recording on the current cpu
+ * The function might sleep.
+ *
+ * task: the task to request recording for
+ * cpu:  the cpu to request recording for
  * base: the base pointer for the (non-pageable) buffer;
  * size: the size of the provided buffer in bytes
  * ovfl: pointer to a function to be called on buffer overflow;
@@ -93,19 +95,28 @@ enum ds_feature {
  *     -1 if no interrupt threshold is requested.
  * flags: a bit-mask of the above flags
  */
-extern struct bts_tracer *ds_request_bts(struct task_struct *task,
-					 void *base, size_t size,
-					 bts_ovfl_callback_t ovfl,
-					 size_t th, unsigned int flags);
-extern struct pebs_tracer *ds_request_pebs(struct task_struct *task,
-					   void *base, size_t size,
-					   pebs_ovfl_callback_t ovfl,
-					   size_t th, unsigned int flags);
+extern struct bts_tracer *ds_request_bts_task(struct task_struct *task,
+					      void *base, size_t size,
+					      bts_ovfl_callback_t ovfl,
+					      size_t th, unsigned int flags);
+extern struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
+					     bts_ovfl_callback_t ovfl,
+					     size_t th, unsigned int flags);
+extern struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
+						void *base, size_t size,
+						pebs_ovfl_callback_t ovfl,
+						size_t th, unsigned int flags);
+extern struct pebs_tracer *ds_request_pebs_cpu(int cpu,
+					       void *base, size_t size,
+					       pebs_ovfl_callback_t ovfl,
+					       size_t th, unsigned int flags);
 
 /*
  * Release BTS or PEBS resources
  * Suspend and resume BTS or PEBS tracing
  *
+ * Must be called with irq's enabled.
+ *
  * tracer: the tracer handle returned from ds_request_~()
  */
 extern void ds_release_bts(struct bts_tracer *tracer);
@@ -115,6 +126,28 @@ extern void ds_release_pebs(struct pebs_tracer *tracer);
 extern void ds_suspend_pebs(struct pebs_tracer *tracer);
 extern void ds_resume_pebs(struct pebs_tracer *tracer);
 
+/*
+ * Release BTS or PEBS resources
+ * Suspend and resume BTS or PEBS tracing
+ *
+ * Cpu tracers must call this on the traced cpu.
+ * Task tracers must call ds_release_~_noirq() for themselves.
+ *
+ * May be called with irq's disabled.
+ *
+ * Returns 0 if successful;
+ * -EPERM if the cpu tracer does not trace the current cpu.
+ * -EPERM if the task tracer does not trace itself.
+ *
+ * tracer: the tracer handle returned from ds_request_~()
+ */
+extern int ds_release_bts_noirq(struct bts_tracer *tracer);
+extern int ds_suspend_bts_noirq(struct bts_tracer *tracer);
+extern int ds_resume_bts_noirq(struct bts_tracer *tracer);
+extern int ds_release_pebs_noirq(struct pebs_tracer *tracer);
+extern int ds_suspend_pebs_noirq(struct pebs_tracer *tracer);
+extern int ds_resume_pebs_noirq(struct pebs_tracer *tracer);
+
 
 /*
  * The raw DS buffer state as it is used for BTS and PEBS recording.
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 2071b99..21a3852 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -245,60 +245,50 @@ struct ds_context {
 	struct pebs_tracer	*pebs_master;
 
 	/* Use count: */
-	unsigned long count;
+	unsigned long		count;
 
 	/* Pointer to the context pointer field: */
 	struct ds_context	**this;
 
-	/* The traced task; NULL for current cpu: */
+	/* The traced task; NULL for cpu tracing: */
 	struct task_struct	*task;
-};
 
-static DEFINE_PER_CPU(struct ds_context *, system_context_array);
+	/* The traced cpu; only valid if task is NULL: */
+	int			cpu;
+};
 
-#define system_context per_cpu(system_context_array, smp_processor_id())
+static DEFINE_PER_CPU(struct ds_context *, cpu_context);
 
 
-static inline struct ds_context *ds_get_context(struct task_struct *task)
+static struct ds_context *ds_get_context(struct task_struct *task, int cpu)
 {
 	struct ds_context **p_context =
-		(task ? &task->thread.ds_ctx : &system_context);
+		(task ? &task->thread.ds_ctx : &per_cpu(cpu_context, cpu));
 	struct ds_context *context = NULL;
 	struct ds_context *new_context = NULL;
-	unsigned long irq;
 
-	/*
-	 * Chances are small that we already have a context.
-	 *
-	 * Contexts for per-cpu tracing are allocated using
-	 * smp_call_function(). We must not sleep.
-	 */
-	new_context = kzalloc(sizeof(*new_context), GFP_ATOMIC);
+	/* 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_irq(&ds_lock);
 
 	context = *p_context;
-	if (!context) {
+	if (likely(!context)) {
 		context = new_context;
 
 		context->this = p_context;
 		context->task = task;
+		context->cpu = cpu;
 		context->count = 0;
 
-		if (task)
-			set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
-
-		if (!task || (task == current))
-			wrmsrl(MSR_IA32_DS_AREA, (unsigned long)context->ds);
-
 		*p_context = context;
 	}
 
 	context->count++;
 
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 
 	if (context != new_context)
 		kfree(new_context);
@@ -306,7 +296,7 @@ static inline struct ds_context *ds_get_context(struct task_struct *task)
 	return context;
 }
 
-static inline void ds_put_context(struct ds_context *context)
+static void ds_put_context(struct ds_context *context)
 {
 	struct task_struct *task;
 	unsigned long irq;
@@ -328,8 +318,15 @@ static inline void ds_put_context(struct ds_context *context)
 	if (task)
 		clear_tsk_thread_flag(task, TIF_DS_AREA_MSR);
 
-	if (!task || (task == current))
-		wrmsrl(MSR_IA32_DS_AREA, 0);
+	/*
+	 * We leave the (now dangling) pointer to the DS configuration in
+	 * the DS_AREA msr. This is as good or as bad as replacing it with
+	 * NULL - the hardware would crash if we enabled tracing.
+	 *
+	 * This saves us some problems with having to write an msr on a
+	 * different cpu while preventing others from doing the same for the
+	 * next context for that same cpu.
+	 */
 
 	spin_unlock_irqrestore(&ds_lock, irq);
 
@@ -340,6 +337,31 @@ static inline void ds_put_context(struct ds_context *context)
 	kfree(context);
 }
 
+static void ds_install_ds_area(struct ds_context *context)
+{
+	unsigned long ds;
+
+	ds = (unsigned long)context->ds;
+
+	/*
+	 * There is a race between the bts master and the pebs master.
+	 *
+	 * The thread/cpu access is synchronized via get/put_cpu() for
+	 * task tracing and via wrmsr_on_cpu for cpu tracing.
+	 *
+	 * If bts and pebs are collected for the same task or same cpu,
+	 * the same confiuration is written twice.
+	 */
+	if (context->task) {
+		get_cpu();
+		if (context->task == current)
+			wrmsrl(MSR_IA32_DS_AREA, ds);
+		set_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
+		put_cpu();
+	} else
+		wrmsr_on_cpu(context->cpu, MSR_IA32_DS_AREA,
+			     (u32)((u64)ds), (u32)((u64)ds >> 32));
+}
 
 /*
  * Call the tracer's callback on a buffer overflow.
@@ -622,6 +644,7 @@ static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
 	 * The value for 'no threshold' is -1, which will set the
 	 * threshold outside of the buffer, just like we want it.
 	 */
+	ith *= ds_cfg.sizeof_rec[qual];
 	trace->ith = (void *)(buffer + size - ith);
 
 	trace->flags = flags;
@@ -630,7 +653,7 @@ static void ds_init_ds_trace(struct ds_trace *trace, 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)
+		      int cpu, void *base, size_t size, size_t th)
 {
 	struct ds_context *context;
 	int error;
@@ -643,7 +666,7 @@ static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
 	if (!base)
 		goto out;
 
-	/* We require some space to do alignment adjustments below. */
+	/* We need space for alignment adjustments in ds_init_ds_trace(). */
 	error = -EINVAL;
 	if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual]))
 		goto out;
@@ -660,25 +683,27 @@ static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
 	tracer->size = size;
 
 	error = -ENOMEM;
-	context = ds_get_context(task);
+	context = ds_get_context(task, cpu);
 	if (!context)
 		goto out;
 	tracer->context = context;
 
-	ds_init_ds_trace(trace, qual, base, size, th, flags);
+	/*
+	 * Defer any tracer-specific initialization work for the context until
+	 * context ownership has been clarified.
+	 */
 
 	error = 0;
  out:
 	return error;
 }
 
-struct bts_tracer *ds_request_bts(struct task_struct *task,
-				  void *base, size_t size,
-				  bts_ovfl_callback_t ovfl, size_t th,
-				  unsigned int flags)
+static struct bts_tracer *ds_request_bts(struct task_struct *task, int cpu,
+					 void *base, size_t size,
+					 bts_ovfl_callback_t ovfl, size_t th,
+					 unsigned int flags)
 {
 	struct bts_tracer *tracer;
-	unsigned long irq;
 	int error;
 
 	/* Buffer overflow notification is not yet implemented. */
@@ -690,42 +715,46 @@ struct bts_tracer *ds_request_bts(struct task_struct *task,
 	if (error < 0)
 		goto out;
 
-	/*
-	 * Per-cpu tracing is typically requested using smp_call_function().
-	 * We must not sleep.
-	 */
 	error = -ENOMEM;
-	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
+	tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
 	if (!tracer)
 		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
+	/* Do some more error checking and acquire a tracing context. */
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
-			   ds_bts, task, base, size, th, flags);
+			   ds_bts, task, cpu, base, size, th);
 	if (error < 0)
 		goto out_tracer;
 
-
-	spin_lock_irqsave(&ds_lock, irq);
+	/* Claim the bts part of the tracing context we acquired above. */
+	spin_lock_irq(&ds_lock);
 
 	error = -EPERM;
 	if (tracer->ds.context->bts_master)
 		goto out_unlock;
 	tracer->ds.context->bts_master = tracer;
 
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 
+	/*
+	 * Now that we own the bts part of the context, let's complete the
+	 * initialization for that part.
+	 */
+	ds_init_ds_trace(&tracer->trace.ds, ds_bts, base, size, th, flags);
+	ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
+	ds_install_ds_area(tracer->ds.context);
 
 	tracer->trace.read  = bts_read;
 	tracer->trace.write = bts_write;
 
-	ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
+	/* Start tracing. */
 	ds_resume_bts(tracer);
 
 	return tracer;
 
  out_unlock:
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
@@ -735,13 +764,27 @@ struct bts_tracer *ds_request_bts(struct task_struct *task,
 	return ERR_PTR(error);
 }
 
-struct pebs_tracer *ds_request_pebs(struct task_struct *task,
-				    void *base, size_t size,
-				    pebs_ovfl_callback_t ovfl, size_t th,
-				    unsigned int flags)
+struct bts_tracer *ds_request_bts_task(struct task_struct *task,
+				       void *base, size_t size,
+				       bts_ovfl_callback_t ovfl,
+				       size_t th, unsigned int flags)
+{
+	return ds_request_bts(task, 0, base, size, ovfl, th, flags);
+}
+
+struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
+				      bts_ovfl_callback_t ovfl,
+				      size_t th, unsigned int flags)
+{
+	return ds_request_bts(NULL, cpu, base, size, ovfl, th, flags);
+}
+
+static struct pebs_tracer *ds_request_pebs(struct task_struct *task, int cpu,
+					   void *base, size_t size,
+					   pebs_ovfl_callback_t ovfl, size_t th,
+					   unsigned int flags)
 {
 	struct pebs_tracer *tracer;
-	unsigned long irq;
 	int error;
 
 	/* Buffer overflow notification is not yet implemented. */
@@ -753,37 +796,43 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 	if (error < 0)
 		goto out;
 
-	/*
-	 * Per-cpu tracing is typically requested using smp_call_function().
-	 * We must not sleep.
-	 */
 	error = -ENOMEM;
-	tracer = kzalloc(sizeof(*tracer), GFP_ATOMIC);
+	tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
 	if (!tracer)
 		goto out_put_tracer;
 	tracer->ovfl = ovfl;
 
+	/* Do some more error checking and acquire a tracing context. */
 	error = ds_request(&tracer->ds, &tracer->trace.ds,
-			   ds_pebs, task, base, size, th, flags);
+			   ds_pebs, task, cpu, base, size, th);
 	if (error < 0)
 		goto out_tracer;
 
-	spin_lock_irqsave(&ds_lock, irq);
+	/* Claim the pebs part of the tracing context we acquired above. */
+	spin_lock_irq(&ds_lock);
 
 	error = -EPERM;
 	if (tracer->ds.context->pebs_master)
 		goto out_unlock;
 	tracer->ds.context->pebs_master = tracer;
 
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 
+	/*
+	 * Now that we own the pebs part of the context, let's complete the
+	 * initialization for that part.
+	 */
+	ds_init_ds_trace(&tracer->trace.ds, ds_pebs, base, size, th, flags);
 	ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
+	ds_install_ds_area(tracer->ds.context);
+
+	/* Start tracing. */
 	ds_resume_pebs(tracer);
 
 	return tracer;
 
  out_unlock:
-	spin_unlock_irqrestore(&ds_lock, irq);
+	spin_unlock_irq(&ds_lock);
 	ds_put_context(tracer->ds.context);
  out_tracer:
 	kfree(tracer);
@@ -793,16 +842,26 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 	return ERR_PTR(error);
 }
 
-void ds_release_bts(struct bts_tracer *tracer)
+struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
+					 void *base, size_t size,
+					 pebs_ovfl_callback_t ovfl,
+					 size_t th, unsigned int flags)
 {
-	struct task_struct *task;
+	return ds_request_pebs(task, 0, base, size, ovfl, th, flags);
+}
 
-	if (!tracer)
-		return;
+struct pebs_tracer *ds_request_pebs_cpu(int cpu, void *base, size_t size,
+					pebs_ovfl_callback_t ovfl,
+					size_t th, unsigned int flags)
+{
+	return ds_request_pebs(NULL, cpu, base, size, ovfl, th, flags);
+}
 
-	task = tracer->ds.context->task;
+static void ds_free_bts(struct bts_tracer *tracer)
+{
+	struct task_struct *task;
 
-	ds_suspend_bts(tracer);
+	task = tracer->ds.context->task;
 
 	WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
 	tracer->ds.context->bts_master = NULL;
@@ -817,9 +876,69 @@ void ds_release_bts(struct bts_tracer *tracer)
 	kfree(tracer);
 }
 
+void ds_release_bts(struct bts_tracer *tracer)
+{
+	might_sleep();
+
+	if (!tracer)
+		return;
+
+	ds_suspend_bts(tracer);
+	ds_free_bts(tracer);
+}
+
+int ds_release_bts_noirq(struct bts_tracer *tracer)
+{
+	struct task_struct *task;
+	unsigned long irq;
+	int error;
+
+	if (!tracer)
+		return 0;
+
+	task = tracer->ds.context->task;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task &&
+	    (tracer->ds.context->cpu != smp_processor_id()))
+		goto out;
+
+	error = -EPERM;
+	if (task && (task != current))
+		goto out;
+
+	ds_suspend_bts_noirq(tracer);
+	ds_free_bts(tracer);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
+static void update_task_debugctlmsr(struct task_struct *task,
+				    unsigned long debugctlmsr)
+{
+	task->thread.debugctlmsr = debugctlmsr;
+
+	get_cpu();
+	if (task == current)
+		update_debugctlmsr(debugctlmsr);
+
+	if (task->thread.debugctlmsr)
+		set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
+	else
+		clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
+	put_cpu();
+}
+
 void ds_suspend_bts(struct bts_tracer *tracer)
 {
 	struct task_struct *task;
+	unsigned long debugctlmsr;
+	int cpu;
 
 	if (!tracer)
 		return;
@@ -827,29 +946,60 @@ void ds_suspend_bts(struct bts_tracer *tracer)
 	tracer->flags = 0;
 
 	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
 
-	if (!task || (task == current))
-		update_debugctlmsr(get_debugctlmsr() & ~BTS_CONTROL);
+	WARN_ON(!task && irqs_disabled());
 
-	if (task) {
-		task->thread.debugctlmsr &= ~BTS_CONTROL;
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr_on_cpu(cpu));
+	debugctlmsr &= ~BTS_CONTROL;
 
-		if (!task->thread.debugctlmsr)
-			clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
-	}
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr_on_cpu(cpu, debugctlmsr);
 }
 
-void ds_resume_bts(struct bts_tracer *tracer)
+int ds_suspend_bts_noirq(struct bts_tracer *tracer)
 {
 	struct task_struct *task;
-	unsigned long control;
+	unsigned long debugctlmsr, irq;
+	int cpu, error = 0;
 
 	if (!tracer)
-		return;
+		return 0;
 
-	tracer->flags = tracer->trace.ds.flags;
+	tracer->flags = 0;
 
 	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task && (cpu != smp_processor_id()))
+		goto out;
+
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr());
+	debugctlmsr &= ~BTS_CONTROL;
+
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr(debugctlmsr);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
+static unsigned long ds_bts_control(struct bts_tracer *tracer)
+{
+	unsigned long control;
 
 	control = ds_cfg.ctl[dsf_bts];
 	if (!(tracer->trace.ds.flags & BTS_KERNEL))
@@ -857,25 +1007,77 @@ void ds_resume_bts(struct bts_tracer *tracer)
 	if (!(tracer->trace.ds.flags & BTS_USER))
 		control |= ds_cfg.ctl[dsf_bts_user];
 
-	if (task) {
-		task->thread.debugctlmsr |= control;
-		set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
-	}
-
-	if (!task || (task == current))
-		update_debugctlmsr(get_debugctlmsr() | control);
+	return control;
 }
 
-void ds_release_pebs(struct pebs_tracer *tracer)
+void ds_resume_bts(struct bts_tracer *tracer)
 {
 	struct task_struct *task;
+	unsigned long debugctlmsr;
+	int cpu;
 
 	if (!tracer)
 		return;
 
+	tracer->flags = tracer->trace.ds.flags;
+
 	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
 
-	ds_suspend_pebs(tracer);
+	WARN_ON(!task && irqs_disabled());
+
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr_on_cpu(cpu));
+	debugctlmsr |= ds_bts_control(tracer);
+
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr_on_cpu(cpu, debugctlmsr);
+}
+
+int ds_resume_bts_noirq(struct bts_tracer *tracer)
+{
+	struct task_struct *task;
+	unsigned long debugctlmsr, irq;
+	int cpu, error = 0;
+
+	if (!tracer)
+		return 0;
+
+	tracer->flags = tracer->trace.ds.flags;
+
+	task = tracer->ds.context->task;
+	cpu  = tracer->ds.context->cpu;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task && (cpu != smp_processor_id()))
+		goto out;
+
+	debugctlmsr = (task ?
+		       task->thread.debugctlmsr :
+		       get_debugctlmsr());
+	debugctlmsr |= ds_bts_control(tracer);
+
+	if (task)
+		update_task_debugctlmsr(task, debugctlmsr);
+	else
+		update_debugctlmsr(debugctlmsr);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
+static void ds_free_pebs(struct pebs_tracer *tracer)
+{
+	struct task_struct *task;
+
+	task = tracer->ds.context->task;
 
 	WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
 	tracer->ds.context->pebs_master = NULL;
@@ -886,16 +1088,68 @@ void ds_release_pebs(struct pebs_tracer *tracer)
 	kfree(tracer);
 }
 
+void ds_release_pebs(struct pebs_tracer *tracer)
+{
+	might_sleep();
+
+	if (!tracer)
+		return;
+
+	ds_suspend_pebs(tracer);
+	ds_free_pebs(tracer);
+}
+
+int ds_release_pebs_noirq(struct pebs_tracer *tracer)
+{
+	struct task_struct *task;
+	unsigned long irq;
+	int error;
+
+	if (!tracer)
+		return 0;
+
+	task = tracer->ds.context->task;
+
+	local_irq_save(irq);
+
+	error = -EPERM;
+	if (!task &&
+	    (tracer->ds.context->cpu != smp_processor_id()))
+		goto out;
+
+	error = -EPERM;
+	if (task && (task != current))
+		goto out;
+
+	ds_suspend_pebs_noirq(tracer);
+	ds_free_pebs(tracer);
+
+	error = 0;
+ out:
+	local_irq_restore(irq);
+	return error;
+}
+
 void ds_suspend_pebs(struct pebs_tracer *tracer)
 {
 
 }
 
+int ds_suspend_pebs_noirq(struct pebs_tracer *tracer)
+{
+	return 0;
+}
+
 void ds_resume_pebs(struct pebs_tracer *tracer)
 {
 
 }
 
+int ds_resume_pebs_noirq(struct pebs_tracer *tracer)
+{
+	return 0;
+}
+
 const struct bts_trace *ds_read_bts(struct bts_tracer *tracer)
 {
 	if (!tracer)
@@ -1004,26 +1258,6 @@ ds_configure(const struct ds_configuration *cfg,
 		printk(KERN_INFO "[ds] pebs not available\n");
 	}
 
-	if (ds_cfg.sizeof_rec[ds_bts]) {
-		int error;
-
-		error = ds_selftest_bts();
-		if (error) {
-			WARN(1, "[ds] selftest failed. disabling bts.\n");
-			ds_cfg.sizeof_rec[ds_bts] = 0;
-		}
-	}
-
-	if (ds_cfg.sizeof_rec[ds_pebs]) {
-		int error;
-
-		error = ds_selftest_pebs();
-		if (error) {
-			WARN(1, "[ds] selftest failed. disabling pebs.\n");
-			ds_cfg.sizeof_rec[ds_pebs] = 0;
-		}
-	}
-
 	printk(KERN_INFO "[ds] sizes: address: %u bit, ",
 	       8 * ds_cfg.sizeof_ptr_field);
 	printk("bts/pebs record: %u/%u bytes\n",
@@ -1127,3 +1361,29 @@ void ds_copy_thread(struct task_struct *tsk, struct task_struct *father)
 void ds_exit_thread(struct task_struct *tsk)
 {
 }
+
+static __init int ds_selftest(void)
+{
+	if (ds_cfg.sizeof_rec[ds_bts]) {
+		int error;
+
+		error = ds_selftest_bts();
+		if (error) {
+			WARN(1, "[ds] selftest failed. disabling bts.\n");
+			ds_cfg.sizeof_rec[ds_bts] = 0;
+		}
+	}
+
+	if (ds_cfg.sizeof_rec[ds_pebs]) {
+		int error;
+
+		error = ds_selftest_pebs();
+		if (error) {
+			WARN(1, "[ds] selftest failed. disabling pebs.\n");
+			ds_cfg.sizeof_rec[ds_pebs] = 0;
+		}
+	}
+
+	return 0;
+}
+device_initcall(ds_selftest);
diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
index 8c46fbf..e5a263c 100644
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -10,11 +10,12 @@
 
 #include <linux/kernel.h>
 #include <linux/string.h>
+#include <linux/smp.h>
 
 #include <asm/ds.h>
 
 
-#define DS_SELFTEST_BUFFER_SIZE 1021 /* Intentionally chose an odd size. */
+#define BUFFER_SIZE 1021 /* Intentionally chose an odd size. */
 
 
 static int ds_selftest_bts_consistency(const struct bts_trace *trace)
@@ -125,12 +126,12 @@ int ds_selftest_bts(void)
 	struct bts_tracer *tracer;
 	int error = 0;
 	void *top;
-	unsigned char buffer[DS_SELFTEST_BUFFER_SIZE];
+	unsigned char buffer[BUFFER_SIZE];
 
 	printk(KERN_INFO "[ds] bts selftest...");
 
-	tracer = ds_request_bts(NULL, buffer, DS_SELFTEST_BUFFER_SIZE,
-				NULL, (size_t)-1, BTS_KERNEL);
+	tracer = ds_request_bts_cpu(smp_processor_id(), buffer, BUFFER_SIZE,
+				    NULL, (size_t)-1, BTS_KERNEL);
 	if (IS_ERR(tracer)) {
 		error = PTR_ERR(tracer);
 		tracer = NULL;
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index 7c21d1e..adbb243 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -800,8 +800,9 @@ static int ptrace_bts_config(struct task_struct *child,
 	if (cfg.flags & PTRACE_BTS_O_SCHED)
 		flags |= BTS_TIMESTAMPS;
 
-	context->tracer = ds_request_bts(child, context->buffer, context->size,
-					 NULL, (size_t)-1, flags);
+	context->tracer =
+		ds_request_bts_task(child, context->buffer, context->size,
+				    NULL, (size_t)-1, flags);
 	if (unlikely(IS_ERR(context->tracer))) {
 		int error = PTR_ERR(context->tracer);
 
diff --git a/kernel/trace/trace_hw_branches.c b/kernel/trace/trace_hw_branches.c
index 8b2109a..50565d8 100644
--- a/kernel/trace/trace_hw_branches.c
+++ b/kernel/trace/trace_hw_branches.c
@@ -4,7 +4,6 @@
  * Copyright (C) 2008-2009 Intel Corporation.
  * Markus Metzger <markus.t.metzger@gmail.com>, 2008-2009
  */
-#include <linux/spinlock.h>
 #include <linux/kallsyms.h>
 #include <linux/debugfs.h>
 #include <linux/ftrace.h>
@@ -21,168 +20,113 @@
 
 #define BTS_BUFFER_SIZE (1 << 13)
 
-/*
- * The tracer lock protects the below per-cpu tracer array.
- * It needs to be held to:
- * - start tracing on all cpus
- * - stop tracing on all cpus
- * - start tracing on a single hotplug cpu
- * - stop tracing on a single hotplug cpu
- * - read the trace from all cpus
- * - read the trace from a single cpu
- */
-static DEFINE_SPINLOCK(bts_tracer_lock);
 static DEFINE_PER_CPU(struct bts_tracer *, tracer);
 static DEFINE_PER_CPU(unsigned char[BTS_BUFFER_SIZE], buffer);
 
 #define this_tracer per_cpu(tracer, smp_processor_id())
-#define this_buffer per_cpu(buffer, smp_processor_id())
 
 static int trace_hw_branches_enabled __read_mostly;
 static int trace_hw_branches_suspended __read_mostly;
 static struct trace_array *hw_branch_trace __read_mostly;
 
 
-/*
- * Initialize the tracer for the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_init_cpu(void *arg)
+static void bts_trace_init_cpu(int cpu)
 {
-	if (this_tracer)
-		ds_release_bts(this_tracer);
+	per_cpu(tracer, cpu) =
+		ds_request_bts_cpu(cpu, per_cpu(buffer, cpu), BTS_BUFFER_SIZE,
+				   NULL, (size_t)-1, BTS_KERNEL);
 
-	this_tracer = ds_request_bts(NULL, this_buffer, BTS_BUFFER_SIZE,
-				     NULL, (size_t)-1, BTS_KERNEL);
-	if (IS_ERR(this_tracer)) {
-		this_tracer = NULL;
-		return;
-	}
+	if (IS_ERR(per_cpu(tracer, cpu)))
+		per_cpu(tracer, cpu) = NULL;
 }
 
 static int bts_trace_init(struct trace_array *tr)
 {
-	int cpu, avail;
-
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
 	hw_branch_trace = tr;
+	trace_hw_branches_enabled = 0;
 
-	on_each_cpu(bts_trace_init_cpu, NULL, 1);
-
-	/* Check on how many cpus we could enable tracing */
-	avail = 0;
-	for_each_online_cpu(cpu)
-		if (per_cpu(tracer, cpu))
-			avail++;
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		bts_trace_init_cpu(cpu);
 
-	trace_hw_branches_enabled = (avail ? 1 : 0);
+		if (likely(per_cpu(tracer, cpu)))
+			trace_hw_branches_enabled = 1;
+	}
 	trace_hw_branches_suspended = 0;
-
-	spin_unlock(&bts_tracer_lock);
-
+	put_online_cpus();
 
 	/* If we could not enable tracing on a single cpu, we fail. */
-	return avail ? 0 : -EOPNOTSUPP;
-}
-
-/*
- * Release the tracer for the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_release_cpu(void *arg)
-{
-	if (this_tracer) {
-		ds_release_bts(this_tracer);
-		this_tracer = NULL;
-	}
+	return trace_hw_branches_enabled ? 0 : -EOPNOTSUPP;
 }
 
 static void bts_trace_reset(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
-	on_each_cpu(bts_trace_release_cpu, NULL, 1);
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		if (likely(per_cpu(tracer, cpu))) {
+			ds_release_bts(per_cpu(tracer, cpu));
+			per_cpu(tracer, cpu) = NULL;
+		}
+	}
 	trace_hw_branches_enabled = 0;
 	trace_hw_branches_suspended = 0;
-
-	spin_unlock(&bts_tracer_lock);
-}
-
-/*
- * Resume tracing on the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_resume_cpu(void *arg)
-{
-	if (this_tracer)
-		ds_resume_bts(this_tracer);
+	put_online_cpus();
 }
 
 static void bts_trace_start(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
-	on_each_cpu(bts_trace_resume_cpu, NULL, 1);
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_resume_bts(per_cpu(tracer, cpu));
 	trace_hw_branches_suspended = 0;
-
-	spin_unlock(&bts_tracer_lock);
-}
-
-/*
- * Suspend tracing on the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_suspend_cpu(void *arg)
-{
-	if (this_tracer)
-		ds_suspend_bts(this_tracer);
+	put_online_cpus();
 }
 
 static void bts_trace_stop(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
-	on_each_cpu(bts_trace_suspend_cpu, NULL, 1);
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_suspend_bts(per_cpu(tracer, cpu));
 	trace_hw_branches_suspended = 1;
-
-	spin_unlock(&bts_tracer_lock);
+	put_online_cpus();
 }
 
 static int __cpuinit bts_hotcpu_handler(struct notifier_block *nfb,
 				     unsigned long action, void *hcpu)
 {
-	unsigned int cpu = (unsigned long)hcpu;
-
-	spin_lock(&bts_tracer_lock);
-
-	if (!trace_hw_branches_enabled)
-		goto out;
+	int cpu = (long)hcpu;
 
 	switch (action) {
 	case CPU_ONLINE:
 	case CPU_DOWN_FAILED:
-		smp_call_function_single(cpu, bts_trace_init_cpu, NULL, 1);
-
-		if (trace_hw_branches_suspended)
-			smp_call_function_single(cpu, bts_trace_suspend_cpu,
-						 NULL, 1);
+		/* The notification is sent with interrupts enabled. */
+		if (trace_hw_branches_enabled) {
+			bts_trace_init_cpu(cpu);
+
+			if (trace_hw_branches_suspended &&
+			    likely(per_cpu(tracer, cpu)))
+				ds_suspend_bts(per_cpu(tracer, cpu));
+		}
 		break;
+
 	case CPU_DOWN_PREPARE:
-		smp_call_function_single(cpu, bts_trace_release_cpu, NULL, 1);
-		break;
+		/* The notification is sent with interrupts enabled. */
+		if (likely(per_cpu(tracer, cpu))) {
+			ds_release_bts(per_cpu(tracer, cpu));
+			per_cpu(tracer, cpu) = NULL;
+		}
 	}
 
- out:
-	spin_unlock(&bts_tracer_lock);
 	return NOTIFY_DONE;
 }
 
@@ -274,7 +218,7 @@ static void trace_bts_at(const struct bts_trace *trace, void *at)
 /*
  * Collect the trace on the current cpu and write it into the ftrace buffer.
  *
- * pre: bts_tracer_lock must be locked
+ * pre: tracing must be suspended on the current cpu
  */
 static void trace_bts_cpu(void *arg)
 {
@@ -291,10 +235,9 @@ static void trace_bts_cpu(void *arg)
 	if (unlikely(!this_tracer))
 		return;
 
-	ds_suspend_bts(this_tracer);
 	trace = ds_read_bts(this_tracer);
 	if (!trace)
-		goto out;
+		return;
 
 	for (at = trace->ds.top; (void *)at < trace->ds.end;
 	     at += trace->ds.size)
@@ -303,18 +246,27 @@ static void trace_bts_cpu(void *arg)
 	for (at = trace->ds.begin; (void *)at < trace->ds.top;
 	     at += trace->ds.size)
 		trace_bts_at(trace, at);
-
-out:
-	ds_resume_bts(this_tracer);
 }
 
 static void trace_bts_prepare(struct trace_iterator *iter)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_suspend_bts(per_cpu(tracer, cpu));
+	/*
+	 * We need to collect the trace on the respective cpu since ftrace
+	 * implicitly adds the record for the current cpu.
+	 * Once that is more flexible, we could collect the data from any cpu.
+	 */
 	on_each_cpu(trace_bts_cpu, iter->tr, 1);
 
-	spin_unlock(&bts_tracer_lock);
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_resume_bts(per_cpu(tracer, cpu));
+	put_online_cpus();
 }
 
 static void trace_bts_close(struct trace_iterator *iter)
@@ -324,12 +276,11 @@ static void trace_bts_close(struct trace_iterator *iter)
 
 void trace_hw_branch_oops(void)
 {
-	spin_lock(&bts_tracer_lock);
-
-	if (trace_hw_branches_enabled)
+	if (this_tracer) {
+		ds_suspend_bts_noirq(this_tracer);
 		trace_bts_cpu(hw_branch_trace);
-
-	spin_unlock(&bts_tracer_lock);
+		ds_resume_bts_noirq(this_tracer);
+	}
 }
 
 struct tracer bts_tracer __read_mostly =

[tip:tracing/core] x86, hw-branch-tracer: allocate selftest iterator on heap

[tip-bot for Markus Metzger]

Commit-ID:  4d657e51dfc042216febd4a007c6f36881f9256d
Gitweb:     http://git.kernel.org/tip/4d657e51dfc042216febd4a007c6f36881f9256d
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:41 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:21 +0200

x86, hw-branch-tracer: allocate selftest iterator on heap

Allocate the trace_iterator for the hw-branch-tracer selftest on the heap.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144556.578777000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 kernel/trace/trace_selftest.c |   20 ++++++++++++--------
 1 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c
index 499d01c..00dd648 100644
--- a/kernel/trace/trace_selftest.c
+++ b/kernel/trace/trace_selftest.c
@@ -757,7 +757,7 @@ int
 trace_selftest_startup_hw_branches(struct tracer *trace,
 				   struct trace_array *tr)
 {
-	struct trace_iterator iter;
+	struct trace_iterator *iter;
 	struct tracer tracer;
 	unsigned long count;
 	int ret;
@@ -777,17 +777,21 @@ trace_selftest_startup_hw_branches(struct tracer *trace,
 	 * The hw-branch tracer needs to collect the trace from the various
 	 * cpu trace buffers - before tracing is stopped.
 	 */
-	memset(&iter, 0, sizeof(iter));
+	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
+	if (!iter)
+		return -ENOMEM;
+
 	memcpy(&tracer, trace, sizeof(tracer));
 
-	iter.trace = &tracer;
-	iter.tr = tr;
-	iter.pos = -1;
-	mutex_init(&iter.mutex);
+	iter->trace = &tracer;
+	iter->tr = tr;
+	iter->pos = -1;
+	mutex_init(&iter->mutex);
 
-	trace->open(&iter);
+	trace->open(iter);
 
-	mutex_destroy(&iter.mutex);
+	mutex_destroy(&iter->mutex);
+	kfree(iter);
 
 	tracing_stop();
 

[tip:tracing/core] x86, ds: fix compiler warning

[tip-bot for Markus Metzger]

Commit-ID:  353afeea24cc51aafc0ff21a72ec740b6f0af50c
Gitweb:     http://git.kernel.org/tip/353afeea24cc51aafc0ff21a72ec740b6f0af50c
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:42 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:23 +0200

x86, ds: fix compiler warning

Size_t is defined differently on i386 and x86_64.
Change type to avoid compiler warning.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144557.523964000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds_selftest.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
index e5a263c..e1ba510 100644
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -87,7 +87,7 @@ static int ds_selftest_bts_read(struct bts_tracer *tracer,
 	/* Now to the test itself. */
 	for (at = from; (void *)at < to; at += trace->ds.size) {
 		struct bts_struct bts;
-		size_t index;
+		unsigned long index;
 		int error;
 
 		if (((void *)at - trace->ds.begin) % trace->ds.size) {

[tip:tracing/core] x86, ds: fix bounds check in ds selftest

[tip-bot for Markus Metzger]

Commit-ID:  84f201139245c30777ff858e71b8d7e134b8c3ed
Gitweb:     http://git.kernel.org/tip/84f201139245c30777ff858e71b8d7e134b8c3ed
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:43 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:24 +0200

x86, ds: fix bounds check in ds selftest

Fix a bad bounds check in the debug store selftest.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144558.450027000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds_selftest.c |    7 ++++++-
 1 files changed, 6 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
index e1ba510..cccc19a 100644
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -47,8 +47,13 @@ static int ds_selftest_bts_consistency(const struct bts_trace *trace)
 		printk(KERN_CONT "bad bts buffer setup...");
 		error = -1;
 	}
+	/*
+	 * We allow top in [begin; end], since its not clear when the
+	 * overflow adjustment happens: after the increment or before the
+	 * write.
+	 */
 	if ((trace->ds.top < trace->ds.begin) ||
-	    (trace->ds.end <= trace->ds.top)) {
+	    (trace->ds.end < trace->ds.top)) {
 		printk(KERN_CONT "bts top out of bounds...");
 		error = -1;
 	}

[tip:tracing/core] x86, ds: selftest each cpu

[tip-bot for Markus Metzger]

Commit-ID:  01f6569ece6915616f6cae1d7d8b46ab8da9c1bd
Gitweb:     http://git.kernel.org/tip/01f6569ece6915616f6cae1d7d8b46ab8da9c1bd
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:44 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:25 +0200

x86, ds: selftest each cpu

Perform debug store selftests on each cpu.

Cover both the normal and the _noirq variant of the debug store interface.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144559.394583000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds_selftest.c |  182 ++++++++++++++++++++++++++++++-----------
 1 files changed, 135 insertions(+), 47 deletions(-)

diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
index cccc19a..599a963 100644
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -11,13 +11,21 @@
 #include <linux/kernel.h>
 #include <linux/string.h>
 #include <linux/smp.h>
+#include <linux/cpu.h>
 
 #include <asm/ds.h>
 
 
-#define BUFFER_SIZE 1021 /* Intentionally chose an odd size. */
+#define BUFFER_SIZE	521 /* Intentionally chose an odd size. */
 
 
+struct ds_selftest_bts_conf {
+	struct bts_tracer *tracer;
+	int error;
+	int (*suspend)(struct bts_tracer *);
+	int (*resume)(struct bts_tracer *);
+};
+
 static int ds_selftest_bts_consistency(const struct bts_trace *trace)
 {
 	int error = 0;
@@ -125,36 +133,32 @@ static int ds_selftest_bts_read(struct bts_tracer *tracer,
 	return 0;
 }
 
-int ds_selftest_bts(void)
+static void ds_selftest_bts_cpu(void *arg)
 {
+	struct ds_selftest_bts_conf *conf = arg;
 	const struct bts_trace *trace;
-	struct bts_tracer *tracer;
-	int error = 0;
 	void *top;
-	unsigned char buffer[BUFFER_SIZE];
 
-	printk(KERN_INFO "[ds] bts selftest...");
-
-	tracer = ds_request_bts_cpu(smp_processor_id(), buffer, BUFFER_SIZE,
-				    NULL, (size_t)-1, BTS_KERNEL);
-	if (IS_ERR(tracer)) {
-		error = PTR_ERR(tracer);
-		tracer = NULL;
+	if (IS_ERR(conf->tracer)) {
+		conf->error = PTR_ERR(conf->tracer);
+		conf->tracer = NULL;
 
 		printk(KERN_CONT
-		       "initialization failed (err: %d)...", error);
-		goto out;
+		       "initialization failed (err: %d)...", conf->error);
+		return;
 	}
 
-	/* The return should already give us enough trace. */
-	ds_suspend_bts(tracer);
+	/* We should meanwhile have enough trace. */
+	conf->error = conf->suspend(conf->tracer);
+	if (conf->error < 0)
+		return;
 
 	/* Let's see if we can access the trace. */
-	trace = ds_read_bts(tracer);
+	trace = ds_read_bts(conf->tracer);
 
-	error = ds_selftest_bts_consistency(trace);
-	if (error < 0)
-		goto out;
+	conf->error = ds_selftest_bts_consistency(trace);
+	if (conf->error < 0)
+		return;
 
 	/* If everything went well, we should have a few trace entries. */
 	if (trace->ds.top == trace->ds.begin) {
@@ -168,10 +172,11 @@ int ds_selftest_bts(void)
 	}
 
 	/* Let's try to read the trace we collected. */
-	error = ds_selftest_bts_read(tracer, trace,
+	conf->error =
+		ds_selftest_bts_read(conf->tracer, trace,
 				     trace->ds.begin, trace->ds.top);
-	if (error < 0)
-		goto out;
+	if (conf->error < 0)
+		return;
 
 	/*
 	 * Let's read the trace again.
@@ -179,26 +184,31 @@ int ds_selftest_bts(void)
 	 */
 	top = trace->ds.top;
 
-	trace = ds_read_bts(tracer);
-	error = ds_selftest_bts_consistency(trace);
-	if (error < 0)
-		goto out;
+	trace = ds_read_bts(conf->tracer);
+	conf->error = ds_selftest_bts_consistency(trace);
+	if (conf->error < 0)
+		return;
 
 	if (top != trace->ds.top) {
 		printk(KERN_CONT "suspend not working...");
-		error = -1;
-		goto out;
+		conf->error = -1;
+		return;
 	}
 
 	/* Let's collect some more trace - see if resume is working. */
-	ds_resume_bts(tracer);
-	ds_suspend_bts(tracer);
+	conf->error = conf->resume(conf->tracer);
+	if (conf->error < 0)
+		return;
+
+	conf->error = conf->suspend(conf->tracer);
+	if (conf->error < 0)
+		return;
 
-	trace = ds_read_bts(tracer);
+	trace = ds_read_bts(conf->tracer);
 
-	error = ds_selftest_bts_consistency(trace);
-	if (error < 0)
-		goto out;
+	conf->error = ds_selftest_bts_consistency(trace);
+	if (conf->error < 0)
+		return;
 
 	if (trace->ds.top == top) {
 		/*
@@ -210,35 +220,113 @@ int ds_selftest_bts(void)
 		printk(KERN_CONT
 		       "no resume progress/overflow...");
 
-		error = ds_selftest_bts_read(tracer, trace,
+		conf->error =
+			ds_selftest_bts_read(conf->tracer, trace,
 					     trace->ds.begin, trace->ds.end);
 	} else if (trace->ds.top < top) {
 		/*
 		 * We had a buffer overflow - the entire buffer should
 		 * contain trace records.
 		 */
-		error = ds_selftest_bts_read(tracer, trace,
+		conf->error =
+			ds_selftest_bts_read(conf->tracer, trace,
 					     trace->ds.begin, trace->ds.end);
 	} else {
 		/*
 		 * It is quite likely that the buffer did not overflow.
 		 * Let's just check the delta trace.
 		 */
-		error = ds_selftest_bts_read(tracer, trace,
-					     top, trace->ds.top);
+		conf->error =
+			ds_selftest_bts_read(conf->tracer, trace, top,
+					     trace->ds.top);
 	}
-	if (error < 0)
-		goto out;
+	if (conf->error < 0)
+		return;
 
-	error = 0;
+	conf->error = 0;
+}
 
-	/* The final test: release the tracer while tracing is suspended. */
- out:
-	ds_release_bts(tracer);
+static int ds_suspend_bts_wrap(struct bts_tracer *tracer)
+{
+	ds_suspend_bts(tracer);
+	return 0;
+}
+
+static int ds_resume_bts_wrap(struct bts_tracer *tracer)
+{
+	ds_resume_bts(tracer);
+	return 0;
+}
 
-	printk(KERN_CONT "%s.\n", (error ? "failed" : "passed"));
+static void ds_release_bts_noirq_wrap(void *tracer)
+{
+	(void)ds_release_bts_noirq(tracer);
+}
 
-	return error;
+static int ds_selftest_bts_bad_release_noirq(int cpu,
+					     struct bts_tracer *tracer)
+{
+	int error = -EPERM;
+
+	/* Try to release the tracer on the wrong cpu. */
+	get_cpu();
+	if (cpu != smp_processor_id()) {
+		error = ds_release_bts_noirq(tracer);
+		if (error != -EPERM)
+			printk(KERN_CONT "release on wrong cpu...");
+	}
+	put_cpu();
+
+	return error ? 0 : -1;
+}
+
+int ds_selftest_bts(void)
+{
+	struct ds_selftest_bts_conf conf;
+	unsigned char buffer[BUFFER_SIZE];
+	int cpu;
+
+	printk(KERN_INFO "[ds] bts selftest...");
+	conf.error = 0;
+
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		conf.suspend = ds_suspend_bts_wrap;
+		conf.resume = ds_resume_bts_wrap;
+		conf.tracer =
+			ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
+					   NULL, (size_t)-1, BTS_KERNEL);
+		ds_selftest_bts_cpu(&conf);
+		ds_release_bts(conf.tracer);
+		if (conf.error < 0)
+			goto out;
+
+		conf.suspend = ds_suspend_bts_noirq;
+		conf.resume = ds_resume_bts_noirq;
+		conf.tracer =
+			ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
+					   NULL, (size_t)-1, BTS_KERNEL);
+		smp_call_function_single(cpu, ds_selftest_bts_cpu, &conf, 1);
+		if (conf.error >= 0) {
+			conf.error =
+				ds_selftest_bts_bad_release_noirq(cpu,
+								  conf.tracer);
+			/* We must not release the tracer twice. */
+			if (conf.error < 0)
+				conf.tracer = NULL;
+		}
+		smp_call_function_single(cpu, ds_release_bts_noirq_wrap,
+					 conf.tracer, 1);
+		if (conf.error < 0)
+			goto out;
+	}
+
+	conf.error = 0;
+ out:
+	put_online_cpus();
+	printk(KERN_CONT "%s.\n", (conf.error ? "failed" : "passed"));
+
+	return conf.error;
 }
 
 int ds_selftest_pebs(void)

[tip:tracing/core] x86, ds: add task tracing selftest

[tip-bot for Markus Metzger]

Commit-ID:  3a68eef945b234f286406d96dc690fe17863c203
Gitweb:     http://git.kernel.org/tip/3a68eef945b234f286406d96dc690fe17863c203
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:45 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:27 +0200

x86, ds: add task tracing selftest

Add selftests to cover per-task branch tracing.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144600.329346000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds_selftest.c |   71 +++++++++++++++++++++++++++++++++++++++++
 1 files changed, 71 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
index 599a963..a40b253 100644
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -280,10 +280,51 @@ static int ds_selftest_bts_bad_release_noirq(int cpu,
 	return error ? 0 : -1;
 }
 
+static int ds_selftest_bts_bad_request_cpu(int cpu, void *buffer)
+{
+	struct bts_tracer *tracer;
+	int error;
+
+	/* Try to request cpu tracing while task tracing is active. */
+	tracer = ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE, NULL,
+				    (size_t)-1, BTS_KERNEL);
+	error = PTR_ERR(tracer);
+	if (!IS_ERR(tracer)) {
+		ds_release_bts(tracer);
+		error = 0;
+	}
+
+	if (error != -EPERM)
+		printk(KERN_CONT "cpu/task tracing overlap...");
+
+	return error ? 0 : -1;
+}
+
+static int ds_selftest_bts_bad_request_task(void *buffer)
+{
+	struct bts_tracer *tracer;
+	int error;
+
+	/* Try to request cpu tracing while task tracing is active. */
+	tracer = ds_request_bts_task(current, buffer, BUFFER_SIZE, NULL,
+				    (size_t)-1, BTS_KERNEL);
+	error = PTR_ERR(tracer);
+	if (!IS_ERR(tracer)) {
+		error = 0;
+		ds_release_bts(tracer);
+	}
+
+	if (error != -EPERM)
+		printk(KERN_CONT "task/cpu tracing overlap...");
+
+	return error ? 0 : -1;
+}
+
 int ds_selftest_bts(void)
 {
 	struct ds_selftest_bts_conf conf;
 	unsigned char buffer[BUFFER_SIZE];
+	unsigned long irq;
 	int cpu;
 
 	printk(KERN_INFO "[ds] bts selftest...");
@@ -297,6 +338,8 @@ int ds_selftest_bts(void)
 			ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
 					   NULL, (size_t)-1, BTS_KERNEL);
 		ds_selftest_bts_cpu(&conf);
+		if (conf.error >= 0)
+			conf.error = ds_selftest_bts_bad_request_task(buffer);
 		ds_release_bts(conf.tracer);
 		if (conf.error < 0)
 			goto out;
@@ -315,12 +358,40 @@ int ds_selftest_bts(void)
 			if (conf.error < 0)
 				conf.tracer = NULL;
 		}
+		if (conf.error >= 0)
+			conf.error = ds_selftest_bts_bad_request_task(buffer);
 		smp_call_function_single(cpu, ds_release_bts_noirq_wrap,
 					 conf.tracer, 1);
 		if (conf.error < 0)
 			goto out;
 	}
 
+	conf.suspend = ds_suspend_bts_wrap;
+	conf.resume = ds_resume_bts_wrap;
+	conf.tracer =
+		ds_request_bts_task(current, buffer, BUFFER_SIZE,
+				    NULL, (size_t)-1, BTS_KERNEL);
+	ds_selftest_bts_cpu(&conf);
+	if (conf.error >= 0)
+		conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
+	ds_release_bts(conf.tracer);
+	if (conf.error < 0)
+		goto out;
+
+	conf.suspend = ds_suspend_bts_noirq;
+	conf.resume = ds_resume_bts_noirq;
+	conf.tracer =
+		ds_request_bts_task(current, buffer, BUFFER_SIZE,
+				   NULL, (size_t)-1, BTS_KERNEL);
+	local_irq_save(irq);
+	ds_selftest_bts_cpu(&conf);
+	if (conf.error >= 0)
+		conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
+	ds_release_bts_noirq(conf.tracer);
+	local_irq_restore(irq);
+	if (conf.error < 0)
+		goto out;
+
 	conf.error = 0;
  out:
 	put_online_cpus();

[tip:tracing/core] x86, ds: add leakage warning

[tip-bot for Markus Metzger]

Commit-ID:  2311f0de21c17b2a8b960677a9cccfbfa52beb35
Gitweb:     http://git.kernel.org/tip/2311f0de21c17b2a8b960677a9cccfbfa52beb35
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:46 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:28 +0200

x86, ds: add leakage warning

Add a warning in case a debug store context is not removed before
the task it is attached to is freed.

Remove the old warning at thread exit. It is too early.

Declare the debug store context field in thread_struct unconditionally.

Remove ds_copy_thread() and ds_exit_thread() and do the work directly
in process*.c.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144601.254472000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/include/asm/ds.h        |    9 ---------
 arch/x86/include/asm/processor.h |    4 +---
 arch/x86/kernel/ds.c             |   10 ----------
 arch/x86/kernel/process.c        |    5 +++--
 arch/x86/kernel/process_32.c     |    3 ++-
 arch/x86/kernel/process_64.c     |    3 ++-
 6 files changed, 8 insertions(+), 26 deletions(-)

diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h
index 413e127..149e520 100644
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -285,21 +285,12 @@ extern void __cpuinit ds_init_intel(struct cpuinfo_x86 *);
  */
 extern void ds_switch_to(struct task_struct *prev, struct task_struct *next);
 
-/*
- * Task clone/init and cleanup work
- */
-extern void ds_copy_thread(struct task_struct *tsk, struct task_struct *father);
-extern void ds_exit_thread(struct task_struct *tsk);
-
 #else /* CONFIG_X86_DS */
 
 struct cpuinfo_x86;
 static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {}
 static inline void ds_switch_to(struct task_struct *prev,
 				struct task_struct *next) {}
-static inline void ds_copy_thread(struct task_struct *tsk,
-				  struct task_struct *father) {}
-static inline void ds_exit_thread(struct task_struct *tsk) {}
 
 #endif /* CONFIG_X86_DS */
 #endif /* _ASM_X86_DS_H */
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 1efeb49..7c39de7 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -454,10 +454,8 @@ struct thread_struct {
 	unsigned		io_bitmap_max;
 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
 	unsigned long	debugctlmsr;
-#ifdef CONFIG_X86_DS
-/* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
+	/* Debug Store context; see asm/ds.h */
 	struct ds_context	*ds_ctx;
-#endif /* CONFIG_X86_DS */
 };
 
 static inline unsigned long native_get_debugreg(int regno)
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 21a3852..71cab3b 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -1352,16 +1352,6 @@ void ds_switch_to(struct task_struct *prev, struct task_struct *next)
 	update_debugctlmsr(debugctlmsr);
 }
 
-void ds_copy_thread(struct task_struct *tsk, struct task_struct *father)
-{
-	clear_tsk_thread_flag(tsk, TIF_DS_AREA_MSR);
-	tsk->thread.ds_ctx = NULL;
-}
-
-void ds_exit_thread(struct task_struct *tsk)
-{
-}
-
 static __init int ds_selftest(void)
 {
 	if (ds_cfg.sizeof_rec[ds_bts]) {
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index ca98915..fb5dfb8 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -14,6 +14,7 @@
 #include <asm/idle.h>
 #include <asm/uaccess.h>
 #include <asm/i387.h>
+#include <asm/ds.h>
 
 unsigned long idle_halt;
 EXPORT_SYMBOL(idle_halt);
@@ -45,6 +46,8 @@ void free_thread_xstate(struct task_struct *tsk)
 		kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
 		tsk->thread.xstate = NULL;
 	}
+
+	WARN(tsk->thread.ds_ctx, "leaking DS context\n");
 }
 
 void free_thread_info(struct thread_info *ti)
@@ -83,8 +86,6 @@ void exit_thread(void)
 		put_cpu();
 		kfree(bp);
 	}
-
-	ds_exit_thread(current);
 }
 
 void flush_thread(void)
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 76f8f84..b5e4bfe 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -290,7 +290,8 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
 		p->thread.io_bitmap_max = 0;
 	}
 
-	ds_copy_thread(p, current);
+	clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
+	p->thread.ds_ctx = NULL;
 
 	clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
 	p->thread.debugctlmsr = 0;
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index b751a41..5a1a1de 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -335,7 +335,8 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
 			goto out;
 	}
 
-	ds_copy_thread(p, me);
+	clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
+	p->thread.ds_ctx = NULL;
 
 	clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
 	p->thread.debugctlmsr = 0;

[tip:tracing/core] x86, ds: use single debug store cpu configuration

[tip-bot for Markus Metzger]

Commit-ID:  ee811517a5604aa63fae803b7c044712699e1303
Gitweb:     http://git.kernel.org/tip/ee811517a5604aa63fae803b7c044712699e1303
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:47 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:29 +0200

x86, ds: use single debug store cpu configuration

Use a single configuration for all cpus.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144602.191165000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds.c |    7 +++++--
 1 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 71cab3b..443f415 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -47,9 +47,8 @@ struct ds_configuration {
 	/* Control bit-masks indexed by enum ds_feature: */
 	unsigned long		ctl[dsf_ctl_max];
 };
-static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array);
+static struct ds_configuration ds_cfg __read_mostly;
 
-#define ds_cfg per_cpu(ds_cfg_array, smp_processor_id())
 
 /* Maximal size of a DS configuration: */
 #define MAX_SIZEOF_DS		(12 * 8)
@@ -1268,6 +1267,10 @@ ds_configure(const struct ds_configuration *cfg,
 
 void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
 {
+	/* Only configure the first cpu. Others are identical. */
+	if (ds_cfg.name)
+		return;
+
 	switch (c->x86) {
 	case 0x6:
 		switch (c->x86_model) {

[tip:tracing/core] x86, ptrace: add bts context unconditionally

[tip-bot for Markus Metzger]

Commit-ID:  0f4814065ff8c24ca8bfd75c9b73502be152c287
Gitweb:     http://git.kernel.org/tip/0f4814065ff8c24ca8bfd75c9b73502be152c287
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:48 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:31 +0200

x86, ptrace: add bts context unconditionally

Add the ptrace bts context field to task_struct unconditionally.

Initialize the field directly in copy_process().
Remove all the unneeded functionality used to initialize that field.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144603.292754000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/include/asm/ptrace.h |    9 ++++-----
 arch/x86/kernel/ptrace.c      |   20 +-------------------
 include/linux/ptrace.h        |   10 ----------
 include/linux/sched.h         |    2 --
 kernel/fork.c                 |    4 ++--
 kernel/ptrace.c               |   10 ----------
 6 files changed, 7 insertions(+), 48 deletions(-)

diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h
index e304b66..5cdd19f 100644
--- a/arch/x86/include/asm/ptrace.h
+++ b/arch/x86/include/asm/ptrace.h
@@ -235,12 +235,11 @@ extern int do_get_thread_area(struct task_struct *p, int idx,
 extern int do_set_thread_area(struct task_struct *p, int idx,
 			      struct user_desc __user *info, int can_allocate);
 
-extern void x86_ptrace_untrace(struct task_struct *);
-extern void x86_ptrace_fork(struct task_struct *child,
-			    unsigned long clone_flags);
+#ifdef CONFIG_X86_PTRACE_BTS
+extern void ptrace_bts_untrace(struct task_struct *tsk);
 
-#define arch_ptrace_untrace(tsk) x86_ptrace_untrace(tsk)
-#define arch_ptrace_fork(child, flags) x86_ptrace_fork(child, flags)
+#define arch_ptrace_untrace(tsk)	ptrace_bts_untrace(tsk)
+#endif /* CONFIG_X86_PTRACE_BTS */
 
 #endif /* __KERNEL__ */
 
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index adbb243..b32a8ee 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -887,37 +887,19 @@ static int ptrace_bts_size(struct task_struct *child)
 	return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 }
 
-static inline void ptrace_bts_fork(struct task_struct *tsk)
-{
-	tsk->bts = NULL;
-}
-
 /*
  * Called from __ptrace_unlink() after the child has been moved back
  * to its original parent.
  */
-static inline void ptrace_bts_untrace(struct task_struct *child)
+void ptrace_bts_untrace(struct task_struct *child)
 {
 	if (unlikely(child->bts)) {
 		free_bts_context(child->bts);
 		child->bts = NULL;
 	}
 }
-#else
-static inline void ptrace_bts_fork(struct task_struct *tsk) {}
-static inline void ptrace_bts_untrace(struct task_struct *child) {}
 #endif /* CONFIG_X86_PTRACE_BTS */
 
-void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
-{
-	ptrace_bts_fork(child);
-}
-
-void x86_ptrace_untrace(struct task_struct *child)
-{
-	ptrace_bts_untrace(child);
-}
-
 /*
  * Called by kernel/ptrace.c when detaching..
  *
diff --git a/include/linux/ptrace.h b/include/linux/ptrace.h
index 67c1565..59e133d 100644
--- a/include/linux/ptrace.h
+++ b/include/linux/ptrace.h
@@ -95,7 +95,6 @@ extern void __ptrace_link(struct task_struct *child,
 			  struct task_struct *new_parent);
 extern void __ptrace_unlink(struct task_struct *child);
 extern void exit_ptrace(struct task_struct *tracer);
-extern void ptrace_fork(struct task_struct *task, unsigned long clone_flags);
 #define PTRACE_MODE_READ   1
 #define PTRACE_MODE_ATTACH 2
 /* Returns 0 on success, -errno on denial. */
@@ -327,15 +326,6 @@ static inline void user_enable_block_step(struct task_struct *task)
 #define arch_ptrace_untrace(task)		do { } while (0)
 #endif
 
-#ifndef arch_ptrace_fork
-/*
- * Do machine-specific work to initialize a new task.
- *
- * This is called from copy_process().
- */
-#define arch_ptrace_fork(child, clone_flags)	do { } while (0)
-#endif
-
 extern int task_current_syscall(struct task_struct *target, long *callno,
 				unsigned long args[6], unsigned int maxargs,
 				unsigned long *sp, unsigned long *pc);
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 52b8cd0..451186a 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1205,13 +1205,11 @@ struct task_struct {
 	struct list_head ptraced;
 	struct list_head ptrace_entry;
 
-#ifdef CONFIG_X86_PTRACE_BTS
 	/*
 	 * This is the tracer handle for the ptrace BTS extension.
 	 * This field actually belongs to the ptracer task.
 	 */
 	struct bts_context *bts;
-#endif /* CONFIG_X86_PTRACE_BTS */
 
 	/* PID/PID hash table linkage. */
 	struct pid_link pids[PIDTYPE_MAX];
diff --git a/kernel/fork.c b/kernel/fork.c
index 660c2b8..69bde7a 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1086,8 +1086,8 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 #ifdef CONFIG_DEBUG_MUTEXES
 	p->blocked_on = NULL; /* not blocked yet */
 #endif
-	if (unlikely(current->ptrace))
-		ptrace_fork(p, clone_flags);
+
+	p->bts = NULL;
 
 	/* Perform scheduler related setup. Assign this task to a CPU. */
 	sched_fork(p, clone_flags);
diff --git a/kernel/ptrace.c b/kernel/ptrace.c
index aaad0ec..321127d 100644
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -27,16 +27,6 @@
 
 
 /*
- * Initialize a new task whose father had been ptraced.
- *
- * Called from copy_process().
- */
-void ptrace_fork(struct task_struct *child, unsigned long clone_flags)
-{
-	arch_ptrace_fork(child, clone_flags);
-}
-
-/*
  * ptrace a task: make the debugger its new parent and
  * move it to the ptrace list.
  *

[tip:tracing/core] x86, ds: dont use TIF_DEBUGCTLMSR

[tip-bot for Markus Metzger]

Commit-ID:  6047550d3d26fed88b18a208b31f8b90b5ef3e9b
Gitweb:     http://git.kernel.org/tip/6047550d3d26fed88b18a208b31f8b90b5ef3e9b
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:49 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:32 +0200

x86, ds: dont use TIF_DEBUGCTLMSR

Debug store already uses TIF_DS_AREA_MSR to trigger debug store context
switch handling. No need to use TIF_DEBUGCTLMSR, as well.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144604.256645000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds.c |    5 -----
 1 files changed, 0 insertions(+), 5 deletions(-)

diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 443f415..cab2832 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -925,11 +925,6 @@ static void update_task_debugctlmsr(struct task_struct *task,
 	get_cpu();
 	if (task == current)
 		update_debugctlmsr(debugctlmsr);
-
-	if (task->thread.debugctlmsr)
-		set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
-	else
-		clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
 	put_cpu();
 }
 

[tip:tracing/core] x86, ds: fix bad ds_reset_pebs()

[tip-bot for Markus Metzger]

Commit-ID:  608780a9048efa3e85fbc4d8649b26805cc588aa
Gitweb:     http://git.kernel.org/tip/608780a9048efa3e85fbc4d8649b26805cc588aa
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:50 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:34 +0200

x86, ds: fix bad ds_reset_pebs()

Ds_reset_pebs() passed the wrong qualifier to a shared function resulting
in a reset of bts, rather than pebs.

Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144605.206510000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index cab2832..ebfb0fd 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -1186,7 +1186,7 @@ int ds_reset_pebs(struct pebs_tracer *tracer)
 
 	tracer->trace.ds.top = tracer->trace.ds.begin;
 
-	ds_set(tracer->ds.context->ds, ds_bts, ds_index,
+	ds_set(tracer->ds.context->ds, ds_pebs, ds_index,
 	       (unsigned long)tracer->trace.ds.top);
 
 	return 0;

[tip:tracing/core] x86, ds: allow small debug store buffers

[tip-bot for Markus Metzger]

Commit-ID:  150f5164c1258e05b7dea16f29e592f354c48f34
Gitweb:     http://git.kernel.org/tip/150f5164c1258e05b7dea16f29e592f354c48f34
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:51 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:35 +0200

x86, ds: allow small debug store buffers

Check the buffer size more precisely to allow buffers for exactly
one element provided the base address is already properly aligned.

Add a debug store selftest.

Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144606.139137000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/kernel/ds.c          |    9 +++++++--
 arch/x86/kernel/ds_selftest.c |    6 +++---
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index ebfb0fd..4e05157 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -656,6 +656,7 @@ static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
 {
 	struct ds_context *context;
 	int error;
+	size_t req_size;
 
 	error = -EOPNOTSUPP;
 	if (!ds_cfg.sizeof_rec[qual])
@@ -665,9 +666,13 @@ static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
 	if (!base)
 		goto out;
 
-	/* We need space for alignment adjustments in ds_init_ds_trace(). */
+	req_size = ds_cfg.sizeof_rec[qual];
+	/* We might need space for alignment adjustments. */
+	if (!IS_ALIGNED((unsigned long)base, DS_ALIGNMENT))
+		req_size += DS_ALIGNMENT;
+
 	error = -EINVAL;
-	if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual]))
+	if (size < req_size)
 		goto out;
 
 	if (th != (size_t)-1) {
diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
index a40b253..5f104a0 100644
--- a/arch/x86/kernel/ds_selftest.c
+++ b/arch/x86/kernel/ds_selftest.c
@@ -16,8 +16,8 @@
 #include <asm/ds.h>
 
 
-#define BUFFER_SIZE	521 /* Intentionally chose an odd size. */
-
+#define BUFFER_SIZE		521	/* Intentionally chose an odd size. */
+#define SMALL_BUFFER_SIZE	24	/* A single bts entry. */
 
 struct ds_selftest_bts_conf {
 	struct bts_tracer *tracer;
@@ -381,7 +381,7 @@ int ds_selftest_bts(void)
 	conf.suspend = ds_suspend_bts_noirq;
 	conf.resume = ds_resume_bts_noirq;
 	conf.tracer =
-		ds_request_bts_task(current, buffer, BUFFER_SIZE,
+		ds_request_bts_task(current, buffer, SMALL_BUFFER_SIZE,
 				   NULL, (size_t)-1, BTS_KERNEL);
 	local_irq_save(irq);
 	ds_selftest_bts_cpu(&conf);

[tip:tracing/core] x86, ds: support Core i7

[tip-bot for Markus Metzger]

Commit-ID:  017bc617657c928cb9a0c45a7a7e9f4e66695347
Gitweb:     http://git.kernel.org/tip/017bc617657c928cb9a0c45a7a7e9f4e66695347
Author:     Markus Metzger <markus.t.metzger@intel.com>
AuthorDate: Fri, 3 Apr 2009 16:43:52 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 7 Apr 2009 13:36:36 +0200

x86, ds: support Core i7

Add debug store support for Core i7.

Core i7 adds a reset value for each performance counter and a new
PEBS record format.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Cc: roland@redhat.com
Cc: eranian@googlemail.com
Cc: oleg@redhat.com
Cc: juan.villacis@intel.com
Cc: ak@linux.jf.intel.com
LKML-Reference: <20090403144607.088997000@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 arch/x86/include/asm/ds.h |   12 ++++++--
 arch/x86/kernel/ds.c      |   69 ++++++++++++++++++++++++++++++++++++++++----
 2 files changed, 71 insertions(+), 10 deletions(-)

diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h
index 149e520..70dac19 100644
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -234,8 +234,12 @@ struct bts_trace {
 struct pebs_trace {
 	struct ds_trace ds;
 
-	/* the PEBS reset value */
-	unsigned long long reset_value;
+	/* the number of valid counters in the below array */
+	unsigned int counters;
+
+#define MAX_PEBS_COUNTERS 4
+	/* the counter reset value */
+	unsigned long long counter_reset[MAX_PEBS_COUNTERS];
 };
 
 
@@ -270,9 +274,11 @@ extern int ds_reset_pebs(struct pebs_tracer *tracer);
  * Returns 0 on success; -Eerrno on error
  *
  * tracer: the tracer handle returned from ds_request_pebs()
+ * counter: the index of the counter
  * value: the new counter reset value
  */
-extern int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value);
+extern int ds_set_pebs_reset(struct pebs_tracer *tracer,
+			     unsigned int counter, u64 value);
 
 /*
  * Initialization
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 4e05157..48bfe13 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -44,6 +44,9 @@ struct ds_configuration {
 	/* The size of a BTS/PEBS record in bytes: */
 	unsigned char		sizeof_rec[2];
 
+	/* The number of pebs counter reset values in the DS structure. */
+	unsigned char		nr_counter_reset;
+
 	/* Control bit-masks indexed by enum ds_feature: */
 	unsigned long		ctl[dsf_ctl_max];
 };
@@ -51,7 +54,7 @@ static struct ds_configuration ds_cfg __read_mostly;
 
 
 /* Maximal size of a DS configuration: */
-#define MAX_SIZEOF_DS		(12 * 8)
+#define MAX_SIZEOF_DS		0x80
 
 /* Maximal size of a BTS record: */
 #define MAX_SIZEOF_BTS		(3 * 8)
@@ -59,6 +62,12 @@ static struct ds_configuration ds_cfg __read_mostly;
 /* BTS and PEBS buffer alignment: */
 #define DS_ALIGNMENT		(1 << 3)
 
+/* Number of buffer pointers in DS: */
+#define NUM_DS_PTR_FIELDS	8
+
+/* Size of a pebs reset value in DS: */
+#define PEBS_RESET_FIELD_SIZE	8
+
 /* Mask of control bits in the DS MSR register: */
 #define BTS_CONTROL				  \
 	( ds_cfg.ctl[dsf_bts]			| \
@@ -1164,9 +1173,12 @@ const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer)
 		return NULL;
 
 	ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
-	tracer->trace.reset_value =
-		*(u64 *)(tracer->ds.context->ds +
-			 (ds_cfg.sizeof_ptr_field * 8));
+
+	tracer->trace.counters = ds_cfg.nr_counter_reset;
+	memcpy(tracer->trace.counter_reset,
+	       tracer->ds.context->ds +
+	       (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field),
+	       ds_cfg.nr_counter_reset * PEBS_RESET_FIELD_SIZE);
 
 	return &tracer->trace;
 }
@@ -1197,13 +1209,18 @@ int ds_reset_pebs(struct pebs_tracer *tracer)
 	return 0;
 }
 
-int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value)
+int ds_set_pebs_reset(struct pebs_tracer *tracer,
+		      unsigned int counter, u64 value)
 {
 	if (!tracer)
 		return -EINVAL;
 
+	if (ds_cfg.nr_counter_reset < counter)
+		return -EINVAL;
+
 	*(u64 *)(tracer->ds.context->ds +
-		 (ds_cfg.sizeof_ptr_field * 8)) = value;
+		 (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field) +
+		 (counter * PEBS_RESET_FIELD_SIZE)) = value;
 
 	return 0;
 }
@@ -1213,16 +1230,26 @@ static const struct ds_configuration ds_cfg_netburst = {
 	.ctl[dsf_bts]		= (1 << 2) | (1 << 3),
 	.ctl[dsf_bts_kernel]	= (1 << 5),
 	.ctl[dsf_bts_user]	= (1 << 6),
+	.nr_counter_reset	= 1,
 };
 static const struct ds_configuration ds_cfg_pentium_m = {
 	.name = "Pentium M",
 	.ctl[dsf_bts]		= (1 << 6) | (1 << 7),
+	.nr_counter_reset	= 1,
 };
 static const struct ds_configuration ds_cfg_core2_atom = {
 	.name = "Core 2/Atom",
 	.ctl[dsf_bts]		= (1 << 6) | (1 << 7),
 	.ctl[dsf_bts_kernel]	= (1 << 9),
 	.ctl[dsf_bts_user]	= (1 << 10),
+	.nr_counter_reset	= 1,
+};
+static const struct ds_configuration ds_cfg_core_i7 = {
+	.name = "Core i7",
+	.ctl[dsf_bts]		= (1 << 6) | (1 << 7),
+	.ctl[dsf_bts_kernel]	= (1 << 9),
+	.ctl[dsf_bts_user]	= (1 << 10),
+	.nr_counter_reset	= 4,
 };
 
 static void
@@ -1239,6 +1266,32 @@ ds_configure(const struct ds_configuration *cfg,
 	nr_pebs_fields = 18;
 #endif
 
+	/*
+	 * Starting with version 2, architectural performance
+	 * monitoring supports a format specifier.
+	 */
+	if ((cpuid_eax(0xa) & 0xff) > 1) {
+		unsigned long perf_capabilities, format;
+
+		rdmsrl(MSR_IA32_PERF_CAPABILITIES, perf_capabilities);
+
+		format = (perf_capabilities >> 8) & 0xf;
+
+		switch (format) {
+		case 0:
+			nr_pebs_fields = 18;
+			break;
+		case 1:
+			nr_pebs_fields = 22;
+			break;
+		default:
+			printk(KERN_INFO
+			       "[ds] unknown PEBS format: %lu\n", format);
+			nr_pebs_fields = 0;
+			break;
+		}
+	}
+
 	memset(&ds_cfg, 0, sizeof(ds_cfg));
 	ds_cfg = *cfg;
 
@@ -1262,7 +1315,7 @@ ds_configure(const struct ds_configuration *cfg,
 	printk("bts/pebs record: %u/%u bytes\n",
 	       ds_cfg.sizeof_rec[ds_bts], ds_cfg.sizeof_rec[ds_pebs]);
 
-	WARN_ON_ONCE(MAX_SIZEOF_DS < (12 * ds_cfg.sizeof_ptr_field));
+	WARN_ON_ONCE(MAX_PEBS_COUNTERS < ds_cfg.nr_counter_reset);
 }
 
 void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
@@ -1284,6 +1337,8 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
 			ds_configure(&ds_cfg_core2_atom, c);
 			break;
 		case 0x1a: /* Core i7 */
+			ds_configure(&ds_cfg_core_i7, c);
+			break;
 		default:
 			/* Sorry, don't know about them. */
 			break;