From: Markus Armbruster <armbru@redhat.com>
Date: Wed, 19 Dec 2007 11:06:55 +0100
Subject: [xen] xenoprof: loses samples for passive domains
Message-id: 87mys7dmm8.fsf@pike.pond.sub.org
O-Subject: [PATCH RHEL-5.2] 426200: Fix Xenoprof not to lose samples for passive domains
Bugzilla: 426200

Xenoprof can get confused an lose samples for passive domains.  This
leads to inaccurate profiles.

This is actually an attempt to salvage a bug fix from the shipwrecked
bug 241982.  That bug has all the acks, but I expect it to be
dev-acked shortly.  I'm posting the relevant part of its patch now,
before the new bug I created for it has all the acks.  Hope that's
okay.

This patch has been tested as part of the various incarnations of bug
241982's patch.  It's been the same in all of them.  It's been
upstream for a year.

It has no real effect unless domain_switch gets set, which I believe
can happen only under Xen (CPU_DOMAIN_SWITCH event).

Please ACK.

Derived from
# HG changeset patch
# User kfraser@localhost.localdomain
# Date 1163072862 0
# Node ID eea9247ad5a06965dc46cbcab236402dd8f7431b
# Parent  b8a2db59150a084e4e1a5a1cae660dbf7cc14c2d
[XENOPROF] Oprofile user level samples for passive domains are being lost.
The number of of lost samples is most significant when dom0 is idle.

From: joserenato.santos@hp.com
Signed-off-by: Keir Fraser <keir@xensource.com>

Acked-by: "Stephen C. Tweedie" <sct@redhat.com>
Acked-by: Chris Lalancette <clalance@redhat.com>

diff --git a/drivers/oprofile/buffer_sync.c b/drivers/oprofile/buffer_sync.c
index 13416ad..1b30909 100644
--- a/drivers/oprofile/buffer_sync.c
+++ b/drivers/oprofile/buffer_sync.c
@@ -42,6 +42,7 @@ static cpumask_t marked_cpus = CPU_MASK_NONE;
 static DEFINE_SPINLOCK(task_mortuary);
 static void process_task_mortuary(void);
 
+static int cpu_current_domain[NR_CPUS];
 
 /* Take ownership of the task struct and place it on the
  * list for processing. Only after two full buffer syncs
@@ -150,6 +151,11 @@ static void end_sync(void)
 int sync_start(void)
 {
 	int err;
+	int i;
+
+	for (i = 0; i < NR_CPUS; i++) {
+		cpu_current_domain[i] = COORDINATOR_DOMAIN;
+	}
 
 	start_cpu_work();
 
@@ -526,6 +532,11 @@ void sync_buffer(int cpu)
  
 	add_cpu_switch(cpu);
 
+	/* We need to assign the first samples in this CPU buffer to the
+	   same domain that we were processing at the last sync_buffer */
+	if (cpu_current_domain[cpu] != COORDINATOR_DOMAIN) {
+		add_domain_switch(cpu_current_domain[cpu]);
+	}
 	/* Remember, only we can modify tail_pos */
 
 	available = get_slots(cpu_buf);
@@ -559,10 +570,15 @@ void sync_buffer(int cpu)
 			}
 		} else {
 			if (domain_switch) {
+				cpu_current_domain[cpu] = s->eip;
 				add_domain_switch(s->eip);
 				domain_switch = 0;
 			} else {
-				if (state >= sb_bt_start &&
+				if (cpu_current_domain[cpu] !=
+				    COORDINATOR_DOMAIN) {
+					add_sample_entry(s->eip, s->event);
+				}
+				else  if (state >= sb_bt_start &&
 				    !add_sample(mm, s, cpu_mode)) {
 					if (state == sb_bt_start) {
 						state = sb_bt_ignore;
@@ -576,6 +592,11 @@ void sync_buffer(int cpu)
 	}
 	release_mm(mm);
 
+	/* We reset domain to COORDINATOR at each CPU switch */
+	if (cpu_current_domain[cpu] != COORDINATOR_DOMAIN) {
+		add_domain_switch(COORDINATOR_DOMAIN);
+	}
+
 	mark_done(cpu);
 
 	mutex_unlock(&buffer_mutex);