Attachment 159969 Details for Bug 245582 – Very rough first patch for xenoprof support

[patch] Very rough first patch for xenoprof support

linux-2.6.9-xenoprof.patch (text/plain), 30.97 KB, created by Chris Lalancette on 2007-07-25 18:27:07 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Chris Lalancette

Created: 2007-07-25 18:27:07 UTC

Size: 30.97 KB

patch

obsolete

>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/i386/Kconfig /root/linux-2.6.9/arch/i386/Kconfig
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/i386/Kconfig	2007-07-17 13:09:56.000000000 -0400
>+++ /root/linux-2.6.9/arch/i386/Kconfig	2007-07-17 12:56:40.000000000 -0400
>@@ -1295,9 +1295,7 @@ source "drivers/Kconfig"
> 
> source "fs/Kconfig"
> 
>-if !X86_XEN
> source "arch/i386/oprofile/Kconfig"
>-endif
> 
> source "arch/i386/Kconfig.debug"
> 
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/i386/oprofile/Makefile /root/linux-2.6.9/arch/i386/oprofile/Makefile
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/i386/oprofile/Makefile	2004-10-18 17:55:06.000000000 -0400
>+++ /root/linux-2.6.9/arch/i386/oprofile/Makefile	2007-07-17 12:56:40.000000000 -0400
>@@ -6,7 +6,11 @@ DRIVER_OBJS = $(addprefix ../../../drive
> 		oprofilefs.o oprofile_stats.o  \
> 		timer_int.o )
> 
>+ifdef CONFIG_XEN
>+oprofile-y				:= $(DRIVER_OBJS) xenoprof.o
>+else 
> oprofile-y				:= $(DRIVER_OBJS) init.o
> oprofile-$(CONFIG_X86_LOCAL_APIC) 	+= nmi_int.o op_model_athlon.o \
> 					   op_model_ppro.o op_model_p4.o
> oprofile-$(CONFIG_X86_IO_APIC)		+= nmi_timer_int.o
>+endif
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/i386/oprofile/xenoprof.c /root/linux-2.6.9/arch/i386/oprofile/xenoprof.c
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/i386/oprofile/xenoprof.c	1969-12-31 19:00:00.000000000 -0500
>+++ /root/linux-2.6.9/arch/i386/oprofile/xenoprof.c	2007-07-17 13:04:53.000000000 -0400
>@@ -0,0 +1,582 @@
>+/**
>+ * @file xenoprof.c
>+ *
>+ * @remark Copyright 2002 OProfile authors
>+ * @remark Read the file COPYING
>+ *
>+ * @author John Levon <levon@movementarian.org>
>+ *
>+ * Modified by Aravind Menon and Jose Renato Santos for Xen
>+ * These modifications are:
>+ * Copyright (C) 2005 Hewlett-Packard Co.
>+ */
>+
>+#include <linux/init.h>
>+#include <linux/notifier.h>
>+#include <linux/smp.h>
>+#include <linux/oprofile.h>
>+#include <linux/sysdev.h>
>+#include <linux/slab.h>
>+#include <linux/interrupt.h>
>+#include <linux/vmalloc.h>
>+#include <asm/nmi.h>
>+#include <asm/msr.h>
>+#include <asm/apic.h>
>+#include <asm/pgtable.h>
>+#include <xen/evtchn.h>
>+#include "op_counter.h"
>+
>+#include <xen/driver_util.h>
>+#include <xen/interface/xen.h>
>+#include <xen/interface/xenoprof.h>
>+#include <../../../drivers/oprofile/cpu_buffer.h>
>+#include <../../../drivers/oprofile/event_buffer.h>
>+
>+#define MAX_XENOPROF_SAMPLES 16
>+
>+static int xenoprof_start(void);
>+static void xenoprof_stop(void);
>+
>+static int xenoprof_enabled = 0;
>+static unsigned int num_events = 0;
>+static int is_primary = 0;
>+static int active_defined;
>+
>+/* sample buffers shared with Xen */
>+xenoprof_buf_t * xenoprof_buf[MAX_VIRT_CPUS];
>+/* Shared buffer area */
>+char * shared_buffer = NULL;
>+/* Number of buffers in shared area (one per VCPU) */
>+int nbuf;
>+/* Mappings of VIRQ_XENOPROF to irq number (per cpu) */
>+int ovf_irq[NR_CPUS];
>+/* cpu model type string - copied from Xen memory space on XENOPROF_init command */
>+char cpu_type[XENOPROF_CPU_TYPE_SIZE];
>+
>+/* Passive sample buffers shared with Xen */
>+xenoprof_buf_t *p_xenoprof_buf[MAX_OPROF_DOMAINS][MAX_VIRT_CPUS];
>+/* Passive shared buffer area */
>+char *p_shared_buffer[MAX_OPROF_DOMAINS];
>+
>+#ifdef CONFIG_PM
>+
>+static int xenoprof_suspend(struct sys_device * dev, pm_message_t state)
>+{
>+	if (xenoprof_enabled == 1)
>+		xenoprof_stop();
>+	return 0;
>+}
>+
>+
>+static int xenoprof_resume(struct sys_device * dev)
>+{
>+	if (xenoprof_enabled == 1)
>+		xenoprof_start();
>+	return 0;
>+}
>+
>+
>+static struct sysdev_class oprofile_sysclass = {
>+	set_kset_name("oprofile"),
>+	.resume		= xenoprof_resume,
>+	.suspend	= xenoprof_suspend
>+};
>+
>+
>+static struct sys_device device_oprofile = {
>+	.id	= 0,
>+	.cls	= &oprofile_sysclass,
>+};
>+
>+
>+static int __init init_driverfs(void)
>+{
>+	int error;
>+	if (!(error = sysdev_class_register(&oprofile_sysclass)))
>+		error = sysdev_register(&device_oprofile);
>+	return error;
>+}
>+
>+
>+static void __exit exit_driverfs(void)
>+{
>+	sysdev_unregister(&device_oprofile);
>+	sysdev_class_unregister(&oprofile_sysclass);
>+}
>+
>+#else
>+#define init_driverfs() do { } while (0)
>+#define exit_driverfs() do { } while (0)
>+#endif /* CONFIG_PM */
>+
>+unsigned long long oprofile_samples = 0;
>+unsigned long long p_oprofile_samples = 0;
>+
>+unsigned int pdomains;
>+struct xenoprof_passive passive_domains[MAX_OPROF_DOMAINS];
>+
>+static void xenoprof_add_pc(xenoprof_buf_t *buf, int is_passive)
>+{
>+	int head, tail, size;
>+
>+	head = buf->event_head;
>+	tail = buf->event_tail;
>+	size = buf->event_size;
>+
>+	if (tail > head) {
>+		while (tail < size) {
>+			oprofile_add_pc(buf->event_log[tail].eip,
>+					buf->event_log[tail].mode,
>+					buf->event_log[tail].event);
>+			if (!is_passive)
>+				oprofile_samples++;
>+			else
>+				p_oprofile_samples++;
>+			tail++;
>+		}
>+		tail = 0;
>+	}
>+	while (tail < head) {
>+		oprofile_add_pc(buf->event_log[tail].eip,
>+				buf->event_log[tail].mode,
>+				buf->event_log[tail].event);
>+		if (!is_passive)
>+			oprofile_samples++;
>+		else
>+			p_oprofile_samples++;
>+		tail++;
>+	}
>+
>+	buf->event_tail = tail;
>+}
>+
>+static void xenoprof_handle_passive(void)
>+{
>+	int i, j;
>+	int flag_domain, flag_switch = 0;
>+	
>+	for (i = 0; i < pdomains; i++) {
>+		flag_domain = 0;
>+		for (j = 0; j < passive_domains[i].nbuf; j++) {
>+			xenoprof_buf_t *buf = p_xenoprof_buf[i][j];
>+			if (buf->event_head == buf->event_tail)
>+				continue;
>+			if (!flag_domain) {
>+				if (!oprofile_add_domain_switch(passive_domains[i].
>+								domain_id))
>+					goto done;
>+				flag_domain = 1;
>+			}
>+			xenoprof_add_pc(buf, 1);
>+			flag_switch = 1;
>+		}
>+	}
>+done:
>+	if (flag_switch)
>+		oprofile_add_domain_switch(COORDINATOR_DOMAIN);
>+}
>+
>+static irqreturn_t 
>+xenoprof_ovf_interrupt(int irq, void * dev_id, struct pt_regs * regs)
>+{
>+	struct xenoprof_buf * buf;
>+	int cpu;
>+	static unsigned long flag;
>+
>+	cpu = smp_processor_id();
>+	buf = xenoprof_buf[cpu];
>+
>+	xenoprof_add_pc(buf, 0);
>+
>+	if (is_primary && !test_and_set_bit(0, &flag)) {
>+		xenoprof_handle_passive();
>+		smp_mb__before_clear_bit();
>+		clear_bit(0, &flag);
>+	}
>+
>+	return IRQ_HANDLED;
>+}
>+
>+
>+static void unbind_virq(void)
>+{
>+	int i;
>+
>+	for_each_possible_cpu(i) {
>+		if (ovf_irq[i] >= 0) {
>+			unbind_from_irqhandler(ovf_irq[i], NULL);
>+			ovf_irq[i] = -1;
>+		}
>+	}
>+}
>+
>+
>+static int bind_virq(void)
>+{
>+	int i, result;
>+
>+	for_each_possible_cpu(i) {
>+		result = bind_virq_to_irqhandler(VIRQ_XENOPROF,
>+						 i,
>+						 xenoprof_ovf_interrupt,
>+						 SA_INTERRUPT,
>+						 "xenoprof",
>+						 NULL);
>+
>+		if (result < 0) {
>+			unbind_virq();
>+			return result;
>+		}
>+
>+		ovf_irq[i] = result;
>+	}
>+		
>+	return 0;
>+}
>+
>+
>+static int map_xenoprof_buffer(int max_samples)
>+{
>+	struct xenoprof_get_buffer get_buffer;
>+	struct xenoprof_buf *buf;
>+	int npages, ret, i;
>+	struct vm_struct *area;
>+
>+	if ( shared_buffer )
>+		return 0;
>+
>+	get_buffer.max_samples = max_samples;
>+
>+	if ( (ret = HYPERVISOR_xenoprof_op(XENOPROF_get_buffer, &get_buffer)) )
>+		return ret;
>+
>+	nbuf = get_buffer.nbuf;
>+	npages = (get_buffer.bufsize * nbuf - 1) / PAGE_SIZE + 1;
>+
>+	area = alloc_vm_area(npages * PAGE_SIZE);
>+	if (area == NULL)
>+		return -ENOMEM;
>+
>+	if ( (ret = direct_kernel_remap_pfn_range(
>+		      (unsigned long)area->addr,
>+		      get_buffer.buf_maddr >> PAGE_SHIFT,
>+		      npages * PAGE_SIZE, __pgprot(_KERNPG_TABLE), DOMID_SELF)) ) {
>+		vunmap(area->addr);
>+		return ret;
>+	}
>+
>+	shared_buffer = area->addr;
>+	for (i=0; i< nbuf; i++) {
>+		buf = (struct xenoprof_buf*) 
>+			&shared_buffer[i * get_buffer.bufsize];
>+		BUG_ON(buf->vcpu_id >= MAX_VIRT_CPUS);
>+		xenoprof_buf[buf->vcpu_id] = buf;
>+	}
>+
>+	return 0;
>+}
>+
>+
>+static int xenoprof_setup(void)
>+{
>+	int ret;
>+	int i;
>+
>+	if ( (ret = map_xenoprof_buffer(MAX_XENOPROF_SAMPLES)) )
>+		return ret;
>+
>+	if ( (ret = bind_virq()) )
>+		return ret;
>+
>+	if (is_primary) {
>+		struct xenoprof_counter counter;
>+
>+		/* Define dom0 as an active domain if not done yet */
>+		if (!active_defined) {
>+			domid_t domid;
>+			ret = HYPERVISOR_xenoprof_op(XENOPROF_reset_active_list, NULL);
>+			if (ret)
>+				goto err;
>+			domid = 0;
>+			ret = HYPERVISOR_xenoprof_op(XENOPROF_set_active, &domid);
>+			if (ret)
>+				goto err;
>+			active_defined = 1;
>+		}
>+
>+		ret = HYPERVISOR_xenoprof_op(XENOPROF_reserve_counters, NULL);
>+		if (ret)
>+			goto err;
>+		for (i=0; i<num_events; i++) {
>+			counter.ind       = i;
>+			counter.count     = (uint64_t)counter_config[i].count;
>+			counter.enabled   = (uint32_t)counter_config[i].enabled;
>+			counter.event     = (uint32_t)counter_config[i].event;
>+			counter.kernel    = (uint32_t)counter_config[i].kernel;
>+			counter.user      = (uint32_t)counter_config[i].user;
>+			counter.unit_mask = (uint64_t)counter_config[i].unit_mask;
>+			HYPERVISOR_xenoprof_op(XENOPROF_counter, 
>+					       &counter);
>+		}
>+		ret = HYPERVISOR_xenoprof_op(XENOPROF_setup_events, NULL);
>+
>+		if (ret)
>+			goto err;
>+	}
>+
>+	ret = HYPERVISOR_xenoprof_op(XENOPROF_enable_virq, NULL);
>+	if (ret)
>+		goto err;
>+
>+	xenoprof_enabled = 1;
>+	return 0;
>+ err:
>+	unbind_virq();
>+	return ret;
>+}
>+
>+
>+static void xenoprof_shutdown(void)
>+{
>+	xenoprof_enabled = 0;
>+
>+	HYPERVISOR_xenoprof_op(XENOPROF_disable_virq, NULL);
>+
>+	if (is_primary) {
>+		HYPERVISOR_xenoprof_op(XENOPROF_release_counters, NULL);
>+		active_defined = 0;
>+	}
>+
>+	unbind_virq();
>+
>+}
>+
>+
>+static int xenoprof_start(void)
>+{
>+	int ret = 0;
>+
>+	if (is_primary)
>+		ret = HYPERVISOR_xenoprof_op(XENOPROF_start, NULL);
>+
>+	return ret;
>+}
>+
>+
>+static void xenoprof_stop(void)
>+{
>+	if (is_primary)
>+		HYPERVISOR_xenoprof_op(XENOPROF_stop, NULL);
>+}
>+
>+
>+int xenoprof_set_active(int * active_domains,
>+			unsigned int adomains)
>+{
>+	int ret = 0;
>+	int i;
>+	int set_dom0 = 0;
>+	domid_t domid;
>+
>+	if (!is_primary)
>+		return 0;
>+
>+	if (adomains > MAX_OPROF_DOMAINS)
>+		return -E2BIG;
>+
>+	ret = HYPERVISOR_xenoprof_op(XENOPROF_reset_active_list, NULL);
>+	if (ret)
>+		return ret;
>+
>+	for (i=0; i<adomains; i++) {
>+		domid = active_domains[i];
>+		if (domid != active_domains[i]) {
>+			ret = -EINVAL;
>+			goto out;
>+		}
>+		ret = HYPERVISOR_xenoprof_op(XENOPROF_set_active, &domid);
>+		if (ret)
>+			goto out;
>+		if (active_domains[i] == 0)
>+			set_dom0 = 1;
>+	}
>+	/* dom0 must always be active but may not be in the list */ 
>+	if (!set_dom0) {
>+		domid = 0;
>+		ret = HYPERVISOR_xenoprof_op(XENOPROF_set_active, &domid);
>+	}
>+
>+out:
>+	if (ret)
>+		HYPERVISOR_xenoprof_op(XENOPROF_reset_active_list, NULL);
>+	active_defined = !ret;
>+	return ret;
>+}
>+
>+int xenoprof_set_passive(int * p_domains,
>+			 unsigned int pdoms)
>+{
>+	int ret;
>+	int i, j;
>+	int npages;
>+	struct xenoprof_buf *buf;
>+	struct vm_struct *area;
>+	pgprot_t prot = __pgprot(_KERNPG_TABLE);
>+
>+	if (!is_primary)
>+        	return 0;
>+
>+	if (pdoms > MAX_OPROF_DOMAINS)
>+		return -E2BIG;
>+
>+	ret = HYPERVISOR_xenoprof_op(XENOPROF_reset_passive_list, NULL);
>+	if (ret)
>+		return ret;
>+
>+	for (i = 0; i < pdoms; i++) {
>+		passive_domains[i].domain_id = p_domains[i];
>+		passive_domains[i].max_samples = 2048;
>+		ret = HYPERVISOR_xenoprof_op(XENOPROF_set_passive,
>+					     &passive_domains[i]);
>+		if (ret)
>+			goto out;
>+
>+		npages = (passive_domains[i].bufsize * passive_domains[i].nbuf - 1) / PAGE_SIZE + 1;
>+
>+		area = alloc_vm_area(npages * PAGE_SIZE);
>+		if (area == NULL) {
>+			ret = -ENOMEM;
>+			goto out;
>+		}
>+
>+		ret = direct_kernel_remap_pfn_range(
>+			(unsigned long)area->addr,
>+			passive_domains[i].buf_maddr >> PAGE_SHIFT,
>+			npages * PAGE_SIZE, prot, DOMID_SELF);
>+		if (ret) {
>+			vunmap(area->addr);
>+			goto out;
>+		}
>+
>+		p_shared_buffer[i] = area->addr;
>+
>+		for (j = 0; j < passive_domains[i].nbuf; j++) {
>+			buf = (struct xenoprof_buf *)
>+				&p_shared_buffer[i][j * passive_domains[i].bufsize];
>+			BUG_ON(buf->vcpu_id >= MAX_VIRT_CPUS);
>+			p_xenoprof_buf[i][buf->vcpu_id] = buf;
>+		}
>+
>+	}
>+
>+	pdomains = pdoms;
>+	return 0;
>+
>+out:
>+	for (j = 0; j < i; j++) {
>+		vunmap(p_shared_buffer[j]);
>+		p_shared_buffer[j] = NULL;
>+	}
>+
>+ 	return ret;
>+}
>+
>+struct op_counter_config counter_config[OP_MAX_COUNTER];
>+
>+static int xenoprof_create_files(struct super_block * sb, struct dentry * root)
>+{
>+	unsigned int i;
>+
>+	for (i = 0; i < num_events; ++i) {
>+		struct dentry * dir;
>+		char buf[2];
>+ 
>+		snprintf(buf, 2, "%d", i);
>+		dir = oprofilefs_mkdir(sb, root, buf);
>+		oprofilefs_create_ulong(sb, dir, "enabled",
>+					&counter_config[i].enabled);
>+		oprofilefs_create_ulong(sb, dir, "event",
>+					&counter_config[i].event);
>+		oprofilefs_create_ulong(sb, dir, "count",
>+					&counter_config[i].count);
>+		oprofilefs_create_ulong(sb, dir, "unit_mask",
>+					&counter_config[i].unit_mask);
>+		oprofilefs_create_ulong(sb, dir, "kernel",
>+					&counter_config[i].kernel);
>+		oprofilefs_create_ulong(sb, dir, "user",
>+					&counter_config[i].user);
>+	}
>+
>+	return 0;
>+}
>+
>+
>+struct oprofile_operations xenoprof_ops = {
>+	.create_files 	= xenoprof_create_files,
>+	.setup 		= xenoprof_setup,
>+	.shutdown	= xenoprof_shutdown,
>+	.start		= xenoprof_start,
>+	.stop		= xenoprof_stop
>+};
>+
>+
>+/* in order to get driverfs right */
>+static int using_xenoprof;
>+
>+int __init oprofile_arch_init(struct oprofile_operations ** ops)
>+{
>+	struct xenoprof_init init;
>+	int ret, i;
>+
>+	ret = HYPERVISOR_xenoprof_op(XENOPROF_init, &init);
>+
>+	if (!ret) {
>+		num_events = init.num_events;
>+		is_primary = init.is_primary;
>+
>+		/* just in case - make sure we do not overflow event list 
>+		   (i.e. counter_config list) */
>+		if (num_events > OP_MAX_COUNTER)
>+			num_events = OP_MAX_COUNTER;
>+
>+		/*  cpu_type is detected by Xen */
>+		cpu_type[XENOPROF_CPU_TYPE_SIZE-1] = 0;
>+		strncpy(cpu_type, init.cpu_type, XENOPROF_CPU_TYPE_SIZE - 1);
>+		xenoprof_ops.cpu_type = cpu_type;
>+
>+		init_driverfs();
>+		using_xenoprof = 1;
>+		*ops = &xenoprof_ops;
>+
>+		for (i=0; i<NR_CPUS; i++)
>+			ovf_irq[i] = -1;
>+
>+		active_defined = 0;
>+	}
>+	printk(KERN_INFO "oprofile_arch_init: ret %d, events %d, "
>+	       "is_primary %d\n", ret, num_events, is_primary);
>+	return ret;
>+}
>+
>+
>+void __exit oprofile_arch_exit(void)
>+{
>+	int i;
>+
>+	if (using_xenoprof)
>+		exit_driverfs();
>+
>+	if (shared_buffer) {
>+		vunmap(shared_buffer);
>+		shared_buffer = NULL;
>+	}
>+	if (is_primary) {
>+		for (i = 0; i < pdomains; i++)
>+			if (p_shared_buffer[i]) {
>+		                vunmap(p_shared_buffer[i]);
>+                		p_shared_buffer[i] = NULL;
>+			}
>+		HYPERVISOR_xenoprof_op(XENOPROF_shutdown, NULL);
>+        }
>+
>+}
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/x86_64/oprofile/Makefile /root/linux-2.6.9/arch/x86_64/oprofile/Makefile
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/arch/x86_64/oprofile/Makefile	2004-10-18 17:53:43.000000000 -0400
>+++ /root/linux-2.6.9/arch/x86_64/oprofile/Makefile	2007-07-17 12:56:40.000000000 -0400
>@@ -11,9 +11,13 @@ DRIVER_OBJS = $(addprefix ../../../drive
> 	oprofilefs.o oprofile_stats.o \
> 	timer_int.o )
> 
>+ifdef CONFIG_XEN
>+OPROFILE-y := xenoprof.o
>+else
> OPROFILE-y := init.o
> OPROFILE-$(CONFIG_X86_LOCAL_APIC) += nmi_int.o op_model_athlon.o op_model_p4.o \
> 				     op_model_ppro.o
> OPROFILE-$(CONFIG_X86_IO_APIC)    += nmi_timer_int.o 
>+endif
> 
> oprofile-y = $(DRIVER_OBJS) $(addprefix ../../i386/oprofile/, $(OPROFILE-y))
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/buffer_sync.c /root/linux-2.6.9/drivers/oprofile/buffer_sync.c
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/buffer_sync.c	2004-10-18 17:53:45.000000000 -0400
>+++ /root/linux-2.6.9/drivers/oprofile/buffer_sync.c	2007-07-17 12:56:40.000000000 -0400
>@@ -6,6 +6,10 @@
>  *
>  * @author John Levon <levon@movementarian.org>
>  *
>+ * Modified by Aravind Menon for Xen
>+ * These modifications are:
>+ * Copyright (C) 2005 Hewlett-Packard Co.
>+ *
>  * This is the core of the buffer management. Each
>  * CPU buffer is processed and entered into the
>  * global event buffer. Such processing is necessary
>@@ -265,13 +269,29 @@ static void add_cpu_switch(int i)
> 	last_cookie = ~0UL;
> }
> 
>-static void add_kernel_ctx_switch(unsigned int in_kernel)
>+static void add_cpu_mode_switch(unsigned int cpu_mode)
> {
> 	add_event_entry(ESCAPE_CODE);
>-	if (in_kernel)
>-		add_event_entry(KERNEL_ENTER_SWITCH_CODE); 
>-	else
>-		add_event_entry(KERNEL_EXIT_SWITCH_CODE); 
>+	switch (cpu_mode) {
>+	case CPU_MODE_USER:
>+		add_event_entry(KERNEL_EXIT_SWITCH_CODE);
>+		break;
>+	case CPU_MODE_KERNEL:
>+		add_event_entry(KERNEL_ENTER_SWITCH_CODE);
>+		break;
>+	case CPU_MODE_XEN:
>+		add_event_entry(XEN_ENTER_SWITCH_CODE);
>+	  	break;
>+	default:
>+		break;
>+	}
>+}
>+
>+static void add_domain_switch(unsigned long domain_id)
>+{
>+	add_event_entry(ESCAPE_CODE);
>+	add_event_entry(DOMAIN_SWITCH_CODE);
>+	add_event_entry(domain_id);
> }
>  
> static void
>@@ -303,7 +323,7 @@ static void add_sample_entry(unsigned lo
> }
> 
> 
>-static void add_us_sample(struct mm_struct * mm, struct op_sample * s)
>+static int add_us_sample(struct mm_struct * mm, struct op_sample * s)
> {
> 	unsigned long cookie;
> 	off_t offset;
>@@ -312,7 +332,7 @@ static void add_us_sample(struct mm_stru
>  
> 	if (!cookie) {
> 		atomic_inc(&oprofile_stats.sample_lost_no_mapping);
>-		return;
>+		return 0;
> 	}
> 
> 	if (cookie != last_cookie) {
>@@ -321,6 +341,8 @@ static void add_us_sample(struct mm_stru
> 	}
> 
> 	add_sample_entry(offset, s->event);
>+
>+	return 1;
> }
> 
>  
>@@ -328,15 +350,17 @@ static void add_us_sample(struct mm_stru
>  * sample is converted into a persistent dentry/offset pair
>  * for later lookup from userspace.
>  */
>-static void add_sample(struct mm_struct * mm, struct op_sample * s, int in_kernel)
>+static int add_sample(struct mm_struct * mm, struct op_sample * s, int cpu_mode)
> {
>-	if (in_kernel) {
>+	if (cpu_mode >= CPU_MODE_KERNEL) {
> 		add_sample_entry(s->eip, s->event);
>+		return 1;
> 	} else if (mm) {
>-		add_us_sample(mm, s);
>+		return add_us_sample(mm, s);
> 	} else {
> 		atomic_inc(&oprofile_stats.sample_lost_no_mm);
> 	}
>+	return 0;
> }
>  
> 
>@@ -358,9 +382,9 @@ static struct mm_struct * take_tasks_mm(
> }
> 
> 
>-static inline int is_ctx_switch(unsigned long val)
>+static inline int is_code(unsigned long val)
> {
>-	return val == ~0UL;
>+	return val == ESCAPE_CODE;
> }
>  
> 
>@@ -453,6 +477,12 @@ static void mark_done(int cpu)
> 	cpus_clear(marked_cpus);
> }
> 
>+typedef enum {
>+	sb_bt_ignore = -2,
>+	sb_buffer_start,
>+	sb_bt_start,
>+	sb_sample_start,
>+} sync_buffer_state;
> 
> /* Sync one of the CPU's buffers into the global event buffer.
>  * Here we need to go through each batch of samples punctuated
>@@ -466,9 +496,11 @@ void sync_buffer(int cpu)
> 	struct mm_struct *mm = NULL;
> 	struct task_struct * new;
> 	unsigned long cookie = 0;
>-	int in_kernel = 1;
>+	int cpu_mode = 1;
> 	unsigned int i;
>+	sync_buffer_state state = sb_buffer_start;
> 	unsigned long available;
>+	int domain_switch = 0;
> 
> 	down(&buffer_sem);
>  
>@@ -481,11 +513,19 @@ void sync_buffer(int cpu)
> 	for (i=0; i < available; ++i) {
> 		struct op_sample * s = &cpu_buf->buffer[cpu_buf->tail_pos];
>  
>-		if (is_ctx_switch(s->eip)) {
>-			if (s->event <= 1) {
>+		if (is_code(s->eip) && !domain_switch) {
>+			if (s->event <= CPU_MODE_XEN) {
> 				/* kernel/userspace switch */
>-				in_kernel = s->event;
>-				add_kernel_ctx_switch(s->event);
>+				cpu_mode = s->event;
>+				if (state == sb_buffer_start)
>+					state = sb_sample_start;
>+				add_cpu_mode_switch(s->event);
>+			} else if (s->event == CPU_TRACE_BEGIN) {
>+				state = sb_bt_start;
>+				add_event_entry(ESCAPE_CODE);
>+				add_event_entry(TRACE_BEGIN_CODE);
>+			} else if (s->event == CPU_DOMAIN_SWITCH) {
>+				domain_switch = 1;
> 			} else {
> 				struct mm_struct * oldmm = mm;
> 
>@@ -499,7 +539,17 @@ void sync_buffer(int cpu)
> 				add_user_ctx_switch(new, cookie);
> 			}
> 		} else {
>-			add_sample(mm, s, in_kernel);
>+			if (domain_switch) {
>+				add_domain_switch(s->eip);
>+				domain_switch = 0;
>+			} else {
>+				if (state >= sb_bt_start &&
>+				    !add_sample(mm, s, cpu_mode)) {
>+					if (state == sb_bt_start) {
>+						state = sb_bt_ignore;
>+					}
>+				}
>+			}
> 		}
> 
> 		increment_tail(cpu_buf);
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/cpu_buffer.c /root/linux-2.6.9/drivers/oprofile/cpu_buffer.c
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/cpu_buffer.c	2004-10-18 17:53:06.000000000 -0400
>+++ /root/linux-2.6.9/drivers/oprofile/cpu_buffer.c	2007-07-17 12:56:40.000000000 -0400
>@@ -6,6 +6,10 @@
>  *
>  * @author John Levon <levon@movementarian.org>
>  *
>+ * Modified by Aravind Menon for Xen
>+ * These modifications are:
>+ * Copyright (C) 2005 Hewlett-Packard Co.
>+ *
>  * Each CPU has a local buffer that stores PC value/event
>  * pairs. We also log context switches when we notice them.
>  * Eventually each CPU's buffer is processed into the global
>@@ -21,6 +25,7 @@
> #include <linux/vmalloc.h>
> #include <linux/errno.h>
>  
>+#include "event_buffer.h"
> #include "cpu_buffer.h"
> #include "buffer_sync.h"
> #include "oprof.h"
>@@ -32,6 +37,8 @@ static void wq_sync_buffer(void *);
> #define DEFAULT_TIMER_EXPIRE (HZ / 10)
> int work_enabled;
> 
>+static int32_t current_domain = COORDINATOR_DOMAIN;
>+
> static void __free_cpu_buffers(int num)
> {
> 	int i;
>@@ -148,8 +155,8 @@ static void increment_head(struct oprofi
>  *
>  * is_kernel is needed because on some architectures you cannot
>  * tell if you are in kernel or user space simply by looking at
>- * eip. We tag this in the buffer by generating kernel enter/exit
>- * events whenever is_kernel changes
>+ * pc. We tag this in the buffer by generating kernel/user (and xen)
>+ * enter events whenever cpu_mode changes
>  */
> void oprofile_add_sample(unsigned long eip, unsigned int is_kernel, 
> 	unsigned long event, int cpu)
>@@ -157,7 +164,7 @@ void oprofile_add_sample(unsigned long e
> 	struct oprofile_cpu_buffer * cpu_buf = &cpu_buffer[cpu];
> 	struct task_struct * task;
> 
>-	is_kernel = !!is_kernel;
>+	WARN_ON(is_kernel > CPU_MODE_XEN);
> 
> 	cpu_buf->sample_received++;
>  
>@@ -178,7 +185,9 @@ void oprofile_add_sample(unsigned long e
> 	}
> 
> 	/* notice a task switch */
>-	if (cpu_buf->last_task != task) {
>+	/* if not processing other domain samples */
>+	if (cpu_buf->last_task != task &&
>+	    (current_domain == COORDINATOR_DOMAIN)) {
> 		cpu_buf->last_task = task;
> 		cpu_buf->buffer[cpu_buf->head_pos].eip = ~0UL;
> 		cpu_buf->buffer[cpu_buf->head_pos].event = (unsigned long)task;
>@@ -202,6 +211,68 @@ void cpu_buffer_reset(struct oprofile_cp
> 	cpu_buf->last_task = NULL;
> }
> 
>+/* CRL - this is almost a complete copy of oprofile_add_sample; switch
>+   to log_sample like RHEL-5 does */
>+void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
>+{
>+	struct oprofile_cpu_buffer * cpu_buf = &cpu_buffer[smp_processor_id()];
>+	struct task_struct * task;
>+
>+	cpu_buf->sample_received++;
>+
>+	if (nr_available_slots(cpu_buf) < 3) {
>+		cpu_buf->sample_lost_overflow++;
>+		return;
>+	}
>+
>+	task = current;
>+
>+	/* notice a switch from user->kernel or vice versa */
>+	if (cpu_buf->last_is_kernel != is_kernel) {
>+		cpu_buf->last_is_kernel = is_kernel;
>+		cpu_buf->buffer[cpu_buf->head_pos].eip = ~0UL;
>+		cpu_buf->buffer[cpu_buf->head_pos].event = is_kernel;
>+		increment_head(cpu_buf);
>+	}
>+
>+	/* notice a task switch */
>+	/* if not processing other domain samples */
>+	if (cpu_buf->last_task != task &&
>+	    (current_domain == COORDINATOR_DOMAIN)) {
>+		cpu_buf->last_task = task;
>+		cpu_buf->buffer[cpu_buf->head_pos].eip = ~0UL;
>+		cpu_buf->buffer[cpu_buf->head_pos].event = (unsigned long)task;
>+		increment_head(cpu_buf);
>+	}
>+ 
>+	cpu_buf->buffer[cpu_buf->head_pos].eip = pc;
>+	cpu_buf->buffer[cpu_buf->head_pos].event = event;
>+	increment_head(cpu_buf);
>+}
>+
>+int oprofile_add_domain_switch(int32_t domain_id)
>+{
>+	struct oprofile_cpu_buffer * cpu_buf = &cpu_buffer[smp_processor_id()];
>+
>+	/* should have space for switching into and out of domain 
>+	   (2 slots each) plus one sample and one cpu mode switch */
>+	if (((nr_available_slots(cpu_buf) < 6) && 
>+	     (domain_id != COORDINATOR_DOMAIN)) ||
>+	    (nr_available_slots(cpu_buf) < 2))
>+		return 0;
>+
>+	cpu_buf->buffer[cpu_buf->head_pos].eip = ESCAPE_CODE;
>+	cpu_buf->buffer[cpu_buf->head_pos].event = CPU_DOMAIN_SWITCH;
>+	increment_head(cpu_buf);
>+
>+	cpu_buf->buffer[cpu_buf->head_pos].eip = domain_id;
>+	cpu_buf->buffer[cpu_buf->head_pos].event = 0;
>+	increment_head(cpu_buf);
>+
>+	current_domain = domain_id;
>+
>+	return 1;
>+}
> 
> /*
>  * This serves to avoid cpu buffer overflow, and makes sure
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/cpu_buffer.h /root/linux-2.6.9/drivers/oprofile/cpu_buffer.h
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/cpu_buffer.h	2004-10-18 17:55:27.000000000 -0400
>+++ /root/linux-2.6.9/drivers/oprofile/cpu_buffer.h	2007-07-17 12:56:40.000000000 -0400
>@@ -48,4 +48,10 @@ extern struct oprofile_cpu_buffer cpu_bu
> 
> void cpu_buffer_reset(struct oprofile_cpu_buffer * cpu_buf);
> 
>+#define CPU_MODE_USER           0
>+#define CPU_MODE_KERNEL         1
>+#define CPU_MODE_XEN            2
>+#define CPU_TRACE_BEGIN         3
>+#define CPU_DOMAIN_SWITCH       4
>+
> #endif /* OPROFILE_CPU_BUFFER_H */
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/event_buffer.h /root/linux-2.6.9/drivers/oprofile/event_buffer.h
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/event_buffer.h	2004-10-18 17:53:21.000000000 -0400
>+++ /root/linux-2.6.9/drivers/oprofile/event_buffer.h	2007-07-17 12:56:40.000000000 -0400
>@@ -32,7 +32,13 @@ void wake_up_buffer_waiter(void);
> #define KERNEL_EXIT_SWITCH_CODE		5
> #define MODULE_LOADED_CODE		6
> #define CTX_TGID_CODE			7
>- 
>+#define TRACE_BEGIN_CODE                8
>+#define XEN_ENTER_SWITCH_CODE           10
>+#define DOMAIN_SWITCH_CODE              11 
>+
>+/* Constant used to refer to coordinator domain (Xen) */
>+#define COORDINATOR_DOMAIN -1
>+
> /* add data to the event buffer */
> void add_event_entry(unsigned long data);
>  
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/oprofile_files.c /root/linux-2.6.9/drivers/oprofile/oprofile_files.c
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/drivers/oprofile/oprofile_files.c	2004-10-18 17:54:08.000000000 -0400
>+++ /root/linux-2.6.9/drivers/oprofile/oprofile_files.c	2007-07-17 13:04:08.000000000 -0400
>@@ -5,10 +5,18 @@
>  * @remark Read the file COPYING
>  *
>  * @author John Levon <levon@movementarian.org>
>+ *
>+ * Modified by Aravind Menon for Xen
>+ * These modifications are:
>+ * Copyright (C) 2005 Hewlett-Packard Co.	
>  */
> 
> #include <linux/fs.h>
> #include <linux/oprofile.h>
>+#include <linux/mutex.h>
>+#include <linux/ctype.h>
>+
>+#include <asm/uaccess.h>
> 
> #include "event_buffer.h"
> #include "oprofile_stats.h"
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/include/linux/oprofile.h /root/linux-2.6.9/include/linux/oprofile.h
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/include/linux/oprofile.h	2004-10-18 17:54:37.000000000 -0400
>+++ /root/linux-2.6.9/include/linux/oprofile.h	2007-07-17 13:01:37.000000000 -0400
>@@ -16,7 +16,11 @@
> #include <linux/types.h>
> #include <linux/spinlock.h>
> #include <asm/atomic.h>
>- 
>+
>+#ifdef CONFIG_XEN
>+#include <xen/interface/xenoprof.h>
>+#endif
>+
> struct super_block;
> struct dentry;
> struct file_operations;
>@@ -59,6 +63,13 @@ void oprofile_arch_exit(void);
> extern void oprofile_add_sample(unsigned long eip, unsigned int is_kernel, 
> 	unsigned long event, int cpu);
> 
>+/* Use this instead when the PC value is not from the regs. Doesn't
>+ * backtrace. */
>+void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event);
>+
>+/* add a domain switch entry */
>+int oprofile_add_domain_switch(int32_t domain_id);
>+
> /**
>  * Create a file of the given name as a child of the given root, with
>  * the specified file operations.
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/include/xen/interface/xen.h /root/linux-2.6.9/include/xen/interface/xen.h
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/include/xen/interface/xen.h	2007-07-17 13:09:56.000000000 -0400
>+++ /root/linux-2.6.9/include/xen/interface/xen.h	2007-07-17 12:56:40.000000000 -0400
>@@ -116,6 +116,7 @@
> #define VIRQ_CONSOLE    2  /* (DOM0) Bytes received on emergency console. */
> #define VIRQ_DOM_EXC    3  /* (DOM0) Exceptional event for some domain.   */
> #define VIRQ_DEBUGGER   6  /* (DOM0) A domain has paused for debugging.   */
>+#define VIRQ_XENOPROF   7  /* V. XenOprofile interrupt: new sample available */
> #define NR_VIRQS        8
> 
> /*
>diff -Nurp /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/include/xen/interface/xenoprof.h /root/linux-2.6.9/include/xen/interface/xenoprof.h
>--- /tmp/kernel-RHEL-4/kernel/kernel-2.6.9/linux-2.6.9/include/xen/interface/xenoprof.h	1969-12-31 19:00:00.000000000 -0500
>+++ /root/linux-2.6.9/include/xen/interface/xenoprof.h	2007-07-17 12:56:40.000000000 -0400
>@@ -0,0 +1,110 @@
>+/******************************************************************************
>+ * xenoprof.h
>+ * 
>+ * Interface for enabling system wide profiling based on hardware performance
>+ * counters
>+ * 
>+ * Copyright (C) 2005 Hewlett-Packard Co.
>+ * Written by Aravind Menon & Jose Renato Santos
>+ */
>+
>+#ifndef __XEN_PUBLIC_XENOPROF_H__
>+#define __XEN_PUBLIC_XENOPROF_H__
>+
>+/*
>+ * Commands to HYPERVISOR_xenoprof_op().
>+ */
>+#define XENOPROF_init                0
>+#define XENOPROF_reset_active_list   1
>+#define XENOPROF_reset_passive_list  2
>+#define XENOPROF_set_active          3
>+#define XENOPROF_set_passive         4
>+#define XENOPROF_reserve_counters    5
>+#define XENOPROF_counter             6
>+#define XENOPROF_setup_events        7
>+#define XENOPROF_enable_virq         8
>+#define XENOPROF_start               9
>+#define XENOPROF_stop               10
>+#define XENOPROF_disable_virq       11
>+#define XENOPROF_release_counters   12
>+#define XENOPROF_shutdown           13
>+#define XENOPROF_get_buffer         14
>+#define XENOPROF_last_op            14
>+
>+#define MAX_OPROF_EVENTS    32
>+#define MAX_OPROF_DOMAINS   25	
>+#define XENOPROF_CPU_TYPE_SIZE 64
>+
>+/* Xenoprof performance events (not Xen events) */
>+struct event_log {
>+    uint64_t eip;
>+    uint8_t mode;
>+    uint8_t event;
>+};
>+
>+/* Xenoprof buffer shared between Xen and domain - 1 per VCPU */
>+struct xenoprof_buf {
>+    uint32_t event_head;
>+    uint32_t event_tail;
>+    uint32_t event_size;
>+    uint32_t vcpu_id;
>+    uint64_t xen_samples;
>+    uint64_t kernel_samples;
>+    uint64_t user_samples;
>+    uint64_t lost_samples;
>+    struct event_log event_log[1];
>+};
>+typedef struct xenoprof_buf xenoprof_buf_t;
>+DEFINE_XEN_GUEST_HANDLE(xenoprof_buf_t);
>+
>+struct xenoprof_init {
>+    int32_t  num_events;
>+    int32_t  is_primary;
>+    char cpu_type[XENOPROF_CPU_TYPE_SIZE];
>+};
>+typedef struct xenoprof_init xenoprof_init_t;
>+DEFINE_XEN_GUEST_HANDLE(xenoprof_init_t);
>+
>+struct xenoprof_get_buffer {
>+    int32_t  max_samples;
>+    int32_t  nbuf;
>+    int32_t  bufsize;
>+    uint64_t buf_maddr;
>+};
>+typedef struct xenoprof_get_buffer xenoprof_get_buffer_t;
>+DEFINE_XEN_GUEST_HANDLE(xenoprof_get_buffer_t);
>+
>+struct xenoprof_counter {
>+    uint32_t ind;
>+    uint64_t count;
>+    uint32_t enabled;
>+    uint32_t event;
>+    uint32_t hypervisor;
>+    uint32_t kernel;
>+    uint32_t user;
>+    uint64_t unit_mask;
>+};
>+typedef struct xenoprof_counter xenoprof_counter_t;
>+DEFINE_XEN_GUEST_HANDLE(xenoprof_counter_t);
>+
>+typedef struct xenoprof_passive {
>+    uint16_t domain_id;
>+    int32_t  max_samples;
>+    int32_t  nbuf;
>+    int32_t  bufsize;
>+    uint64_t buf_maddr;
>+} xenoprof_passive_t;
>+DEFINE_XEN_GUEST_HANDLE(xenoprof_passive_t);
>+
>+
>+#endif /* __XEN_PUBLIC_XENOPROF_H__ */
>+
>+/*
>+ * Local variables:
>+ * mode: C
>+ * c-set-style: "BSD"
>+ * c-basic-offset: 4
>+ * tab-width: 4
>+ * indent-tabs-mode: nil
>+ * End:
>+ */

Actions: View | Diff

Attachments on bug 245582: 159969