[IA64] xenoprof ia64 xen side

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
Signed-off-by: Alex Williamson <alex.williamson@hp.com>

xen/arch/ia64/Rules.mk

@@ -3,6 +3,7 @@
 
 HAS_ACPI := y
 HAS_VGA  := y
+xenoprof := y
 VALIDATE_VT	?= n
 no_warns ?= n
 xen_ia64_expose_p2m	?= y

xen/arch/ia64/linux-xen/Makefile

@@ -18,3 +18,5 @@ obj-y += unwind.o
 obj-y += iosapic.o
 obj-y += numa.o
 obj-y += mm_numa.o
+obj-y += perfmon.o
+obj-y += perfmon_default_smpl.o

xen/arch/ia64/linux-xen/perfmon_default_smpl.c

@@ -16,9 +16,15 @@
 #include <asm/perfmon.h>
 #include <asm/perfmon_default_smpl.h>
 
+#ifndef XEN
 MODULE_AUTHOR("Stephane Eranian <eranian@hpl.hp.com>");
 MODULE_DESCRIPTION("perfmon default sampling format");
 MODULE_LICENSE("GPL");
+#endif
+
+#ifdef XEN
+#define pid	vcpu_id
+#endif
 
 #define DEFAULT_DEBUG 1
 
@@ -157,7 +163,9 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 	 * system-wide:
 	 * 	- this is not necessarily the task controlling the session
 	 */
+#ifndef XEN
 	ent->pid            = current->pid;
+#endif
 	ent->ovfl_pmd  	    = ovfl_pmd;
 	ent->last_reset_val = arg->pmd_last_reset; //pmd[0].reg_last_reset_val;
 
@@ -169,7 +177,9 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 	ent->tstamp    = stamp;
 	ent->cpu       = smp_processor_id();
 	ent->set       = arg->active_set;
+#ifndef XEN
 	ent->tgid      = current->tgid;
+#endif
 
 	/*
 	 * selectively store PMDs in increasing index number
@@ -263,6 +273,7 @@ static pfm_buffer_fmt_t default_fmt={
  	.fmt_exit	    = default_exit,
 };
 
+#ifndef XEN
 static int __init
 pfm_default_smpl_init_module(void)
 {
@@ -282,6 +293,7 @@ pfm_default_smpl_init_module(void)
 
 	return ret;
 }
+#endif
 
 static void __exit
 pfm_default_smpl_cleanup_module(void)
@@ -292,6 +304,8 @@ pfm_default_smpl_cleanup_module(void)
 	printk("perfmon_default_smpl: unregister %s=%d\n", default_fmt.fmt_name, ret);
 }
 
+#ifndef XEN
 module_init(pfm_default_smpl_init_module);
 module_exit(pfm_default_smpl_cleanup_module);
+#endif
 

xen/arch/ia64/linux/Makefile

@@ -22,6 +22,7 @@ obj-y += __divdi3.o
 obj-y += __udivdi3.o
 obj-y += __moddi3.o
 obj-y += __umoddi3.o
+obj-y += carta_random.o
 
 ## variants of divide/modulo
 ## see files in xen/arch/ia64/linux/lib (linux/arch/ia64/lib)

xen/arch/ia64/xen/Makefile

@@ -1,3 +1,5 @@
+subdir-y += oprofile
+
 obj-y += acpi.o
 obj-y += dom0_ops.o
 obj-y += domain.o

xen/arch/ia64/xen/dom0_ops.c

@@ -343,6 +343,12 @@ do_dom0vp_op(unsigned long cmd,
     case IA64_DOM0VP_expose_p2m:
         ret = dom0vp_expose_p2m(d, arg0, arg1, arg2, arg3);
         break;
+    case IA64_DOM0VP_perfmon: {
+        XEN_GUEST_HANDLE(void) hnd;
+        set_xen_guest_handle(hnd, (void*)arg1);
+        ret = do_perfmon_op(arg0, hnd, arg2);
+        break;
+    }
     default:
         ret = -1;
 		printk("unknown dom0_vp_op 0x%lx\n", cmd);

xen/arch/ia64/xen/domain.c

@@ -48,6 +48,7 @@
 #include <asm/shadow.h>
 #include <xen/guest_access.h>
 #include <asm/tlb_track.h>
+#include <asm/perfmon.h>
 
 unsigned long dom0_size = 512*1024*1024;
 
@@ -231,11 +232,35 @@ void continue_running(struct vcpu *same)
 	/* nothing to do */
 }
 
+#ifdef CONFIG_PERFMON
+static int pal_halt        = 1;
+static int can_do_pal_halt = 1;
+
+static int __init nohalt_setup(char * str)
+{
+       pal_halt = can_do_pal_halt = 0;
+       return 1;
+}
+__setup("nohalt", nohalt_setup);
+
+void
+update_pal_halt_status(int status)
+{
+       can_do_pal_halt = pal_halt && status;
+}
+#else
+#define can_do_pal_halt	(1)
+#endif
+
 static void default_idle(void)
 {
 	local_irq_disable();
-	if ( !softirq_pending(smp_processor_id()) )
-	        safe_halt();
+	if ( !softirq_pending(smp_processor_id()) ) {
+		if (can_do_pal_halt)
+			safe_halt();
+		else
+			cpu_relax();
+	}
 	local_irq_enable();
 }
 
@@ -628,6 +653,9 @@ void domain_relinquish_resources(struct domain *d)
 
     if (d->arch.is_vti && d->arch.sal_data)
 	    xfree(d->arch.sal_data);
+
+    /* Free page used by xen oprofile buffer */
+    free_xenoprof_pages(d);
 }
 
 unsigned long

xen/arch/ia64/xen/hypercall.c

@@ -68,7 +68,7 @@ const hypercall_t ia64_hypercall_table[NR_hypercalls] =
 	(hypercall_t)do_ni_hypercall,		/* do_nmi_op */
 	(hypercall_t)do_sched_op,
 	(hypercall_t)do_callback_op,		/*  */                 /* 30 */
-	(hypercall_t)do_ni_hypercall,		/*  */
+	(hypercall_t)do_xenoprof_op,		/*  */
 	(hypercall_t)do_event_channel_op,
 	(hypercall_t)do_physdev_op,
 	(hypercall_t)do_hvm_op,			/*  */

xen/arch/ia64/xen/oprofile/Makefile

@@ -0,0 +1 @@
+obj-y += perfmon.o xenoprof.o

xen/arch/ia64/xen/oprofile/perfmon.c

@@ -1,3 +1,25 @@
+/******************************************************************************
+ * perfmon.c for xenoprof
+ * This is based linux/arch/ia64/oprofile/perfmon.c, but heavily rewritten.
+ *
+ * Copyright (c) 2006 Isaku Yamahata <yamahata at valinux co jp>
+ *                    VA Linux Systems Japan K.K.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ */
 /**
  * @file perfmon.c
  *
@@ -7,94 +29,172 @@
  * @author John Levon <levon@movementarian.org>
  */
 
-#include <linux/kernel.h>
-#include <linux/config.h>
-#include <linux/oprofile.h>
-#include <linux/sched.h>
+#include <xen/config.h>
+#include <xen/sched.h>
+#include <xen/event.h>
+#include <xen/xenoprof.h>
 #include <asm/perfmon.h>
 #include <asm/ptrace.h>
-#include <asm/errno.h>
 
+// XXX move them to an appropriate header file
+extern void xenoprof_log_event(struct vcpu *vcpu,
+                               unsigned long eip, int mode, int event);
+extern int is_active(struct domain *d);
+
+static int allow_virq;
 static int allow_ints;
 
 static int
-perfmon_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg,
-                struct pt_regs *regs, unsigned long stamp)
+xenoprof_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg,
+                 struct pt_regs *regs, unsigned long stamp)
 {
-	int event = arg->pmd_eventid;
+    unsigned long ip = regs->cr_iip;
+    int event = arg->pmd_eventid;
  
-	arg->ovfl_ctrl.bits.reset_ovfl_pmds = 1;
+    arg->ovfl_ctrl.bits.reset_ovfl_pmds = 1;
+    if (!allow_virq || !allow_ints)
+        return 0;
 
-	/* the owner of the oprofile event buffer may have exited
-	 * without perfmon being shutdown (e.g. SIGSEGV)
-	 */
-	if (allow_ints)
-		oprofile_add_sample(regs, event);
-	return 0;
+    xenoprof_log_event(current, ip, xenoprofile_get_mode(task, regs), event);
+    
+    // send VIRQ_XENOPROF
+    if (is_active(current->domain) && !ring_0(regs))
+        send_guest_vcpu_virq(current, VIRQ_XENOPROF);
+
+    return 0;
 }
 
+// same as linux OPROFILE_FMT_UUID
+#define XENOPROF_FMT_UUID { \
+    0x77, 0x7a, 0x6e, 0x61, 0x20, 0x65, 0x73, 0x69, 0x74, 0x6e, 0x72, 0x20, 0x61, 0x65, 0x0a, 0x6c }
 
-static int perfmon_start(void)
+static pfm_buffer_fmt_t xenoprof_fmt = {
+    .fmt_name    = "xenoprof_format",
+    .fmt_uuid    = XENOPROF_FMT_UUID,
+    .fmt_handler = xenoprof_handler,
+};
+
+static char * get_cpu_type(void)
+{
+    __u8 family = local_cpu_data->family;
+
+    switch (family) {
+        case 0x07:
+            return "ia64/itanium";
+        case 0x1f:
+            return "ia64/itanium2";
+        default:
+            return "ia64/ia64";
+    }
+}
+
+static int using_xenoprof;
+
+int __init
+xenprof_perfmon_init(void)
+{
+    int ret = pfm_register_buffer_fmt(&xenoprof_fmt);
+    if (ret)
+        return -ENODEV;
+    using_xenoprof = 1;
+    printk("xenoprof: using perfmon.\n");
+    return 0;
+}
+__initcall(xenprof_perfmon_init);
+
+#ifdef notyet
+void xenoprof_perfmon_exit(void)
+{
+    if (!using_xenoprof)
+        return;
+
+    pfm_unregister_buffer_fmt(xenoprof_fmt.fmt_uuid);
+}
+__exitcall(xenoprof_perfmon_exit);
+#endif
+
+///////////////////////////////////////////////////////////////////////////
+// glue methods for xenoprof and perfmon.
+int
+xenoprof_arch_init(int *num_events, int *is_primary, char *cpu_type)
+{
+    *num_events = 0;
+    strncpy(cpu_type, get_cpu_type(), XENOPROF_CPU_TYPE_SIZE - 1);
+    cpu_type[XENOPROF_CPU_TYPE_SIZE - 1] = '\0';
+
+    *is_primary = 0;
+    if (xenoprof_primary_profiler == NULL) {
+        /* For now, only dom0 can be the primary profiler */
+        if (current->domain->domain_id == 0) {
+            *is_primary = 1;
+        }
+    } else if (xenoprof_primary_profiler == current->domain)
+        *is_primary = 1;
+    return 0;
+}
+
+int
+xenoprof_arch_reserve_counters(void)
+{
+    // perfmon takes care
+    return 0;
+}
+
+int
+xenoprof_arch_counter(XEN_GUEST_HANDLE(void) arg)
+{
+    return -ENOSYS;
+}
+
+int
+xenoprof_arch_setup_events(void)
+{
+    // perfmon takes care
+    return 0;
+}
+
+//XXX SMP: sync by IPI?
+int
+xenoprof_arch_enable_virq(void)
+{
+    allow_virq = 1;
+    return 0;
+}
+
+//XXX SMP: sync by IPI?
+int
+xenoprof_arch_start(void)
 {
 	allow_ints = 1;
-	return 0;
+    return 0;
 }
 
-
-static void perfmon_stop(void)
+//XXX SMP: sync by IPI?
+void
+xenoprof_arch_stop(void)
 {
 	allow_ints = 0;
 }
 
-
-#define OPROFILE_FMT_UUID { \
-	0x77, 0x7a, 0x6e, 0x61, 0x20, 0x65, 0x73, 0x69, 0x74, 0x6e, 0x72, 0x20, 0x61, 0x65, 0x0a, 0x6c }
-
-static pfm_buffer_fmt_t oprofile_fmt = {
- 	.fmt_name 	    = "oprofile_format",
- 	.fmt_uuid	    = OPROFILE_FMT_UUID,
- 	.fmt_handler	    = perfmon_handler,
-};
-
-
-static char * get_cpu_type(void)
+//XXX SMP: sync by IPI?
+void
+xenoprof_arch_disable_virq(void)
 {
-	__u8 family = local_cpu_data->family;
-
-	switch (family) {
-		case 0x07:
-			return "ia64/itanium";
-		case 0x1f:
-			return "ia64/itanium2";
-		default:
-			return "ia64/ia64";
-	}
+    allow_virq = 0;
 }
 
-
-/* all the ops are handled via userspace for IA64 perfmon */
-
-static int using_perfmon;
-
-int perfmon_init(struct oprofile_operations * ops)
+void
+xenoprof_arch_release_counters(void)
 {
-	int ret = pfm_register_buffer_fmt(&oprofile_fmt);
-	if (ret)
-		return -ENODEV;
-
-	ops->cpu_type = get_cpu_type();
-	ops->start = perfmon_start;
-	ops->stop = perfmon_stop;
-	using_perfmon = 1;
-	printk(KERN_INFO "oprofile: using perfmon.\n");
-	return 0;
+    // perfmon takes care
 }
 
-
-void perfmon_exit(void)
-{
-	if (!using_perfmon)
-		return;
-
-	pfm_unregister_buffer_fmt(oprofile_fmt.fmt_uuid);
-}
+/*
+ * Local variables:
+ * mode: C
+ * c-set-style: "BSD"
+ * c-basic-offset: 4
+ * tab-width: 4
+ * indent-tabs-mode: nil
+ * End:
+ */

xen/arch/ia64/xen/oprofile/xenoprof.c

@@ -0,0 +1,56 @@
+/******************************************************************************
+ * xenoprof.c 
+ *
+ * Copyright (c) 2006 Isaku Yamahata <yamahata at valinux co jp>
+ *                    VA Linux Systems Japan K.K.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ */
+
+#include <xen/config.h>
+#include <xen/sched.h>
+#include <public/xen.h>
+#include <xen/xenoprof.h>
+
+int
+xenoprofile_get_mode(struct vcpu *v, struct cpu_user_regs * const regs)
+{
+    int mode = 0;
+
+    // mode
+    // 0: user, 1: kernel, 2: xen
+    // Xen/IA64 uses ring2 for kernel, and doesn't use ring1.
+    if (ring_2(regs))
+        mode = 1;
+    else if (ring_0(regs))
+        mode = 2;
+    else if (ring_1(regs)) {
+        gdprintk(XENLOG_ERR, "%s:%d ring1 is used!\n", __func__, __LINE__);
+        mode = 1;// fall back to kernel mode.
+    }
+
+    return mode;
+}
+
+/*
+ * Local variables:
+ * mode: C
+ * c-set-style: "BSD"
+ * c-basic-offset: 4
+ * tab-width: 4
+ * indent-tabs-mode: nil
+ * End:
+ */

xen/include/asm-ia64/config.h

@@ -125,6 +125,7 @@ extern char _end[]; /* standard ELF symbol */
 // from include/asm-ia64/smp.h
 #define	get_cpu()	smp_processor_id()
 #define put_cpu()	do {} while(0)
+#define put_cpu_no_resched()	do{} while (0)
 
 // needed for common/dom0_ops.c until hyperthreading is supported
 #ifdef CONFIG_SMP
@@ -166,6 +167,8 @@ extern int smp_num_siblings;
 #define ____cacheline_aligned_in_smp ____cacheline_aligned
 #endif
 
+#define CONFIG_PERFMON
+
 #ifndef __ASSEMBLY__
 #include "asm/types.h"	// for u64
 #include "linux/linkage.h"	// for asmlinkage which is used by

xen/include/asm-ia64/domain.h

@@ -211,6 +211,10 @@ struct arch_vcpu {
 #define IO_PORTS_PADDR          0x00000ffffc000000UL
 #define IO_PORTS_SIZE           0x0000000004000000UL
 
+int
+do_perfmon_op(unsigned long cmd,
+              XEN_GUEST_HANDLE(void) arg1, unsigned long arg2);
+
 #endif /* __ASM_DOMAIN_H__ */
 
 /*

xen/include/asm-ia64/linux-xen/asm/perfmon.h

@@ -6,6 +6,14 @@
 #ifndef _ASM_IA64_PERFMON_H
 #define _ASM_IA64_PERFMON_H
 
+#ifdef XEN
+#include <asm/config.h>
+#ifndef pt_regs
+#define pt_regs cpu_user_regs
+#endif
+struct cpu_user_regs;
+#endif
+
 /*
  * perfmon comamnds supported on all CPU models
  */

xen/include/asm-ia64/xenoprof.h

@@ -0,0 +1,64 @@
+/******************************************************************************
+ * asm-ia64/xenoprof.h
+ * xenoprof ia64 arch specific header file
+ *
+ * Copyright (c) 2006 Isaku Yamahata <yamahata at valinux co jp>
+ *                    VA Linux Systems Japan K.K.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ */
+
+#ifndef __ASM_XENOPROF_H__
+#define __ASM_XENOPROF_H__
+
+int xenoprof_arch_init(int *num_events, int *is_primary, char *cpu_type);
+int xenoprof_arch_reserve_counters(void);
+int xenoprof_arch_counter(XEN_GUEST_HANDLE(void) arg);
+int xenoprof_arch_setup_events(void);
+int xenoprof_arch_enable_virq(void);
+int xenoprof_arch_start(void);
+void xenoprof_arch_stop(void);
+void xenoprof_arch_disable_virq(void);
+void xenoprof_arch_release_counters(void);
+
+struct vcpu;
+struct cpu_user_regs;
+int xenoprofile_get_mode(struct vcpu *v, struct cpu_user_regs * const regs);
+
+#define xenoprof_shared_gmfn(d, gmaddr, maddr)  \
+    assign_domain_page((d), (gmaddr), (maddr));
+
+static inline int
+ring(const struct pt_regs* regs)
+{
+    return ((struct ia64_psr*)(&(regs)->cr_ipsr))->cpl;
+}
+#define ring_0(r)       (ring(r) == 0)
+#define ring_1(r)       (ring(r) == 1)
+#define ring_2(r)       (ring(r) == 2)
+#define ring_3(r)       (ring(r) == 3)
+
+#endif /* __ASM_XENOPROF_H__ */
+
+/*
+ * Local variables:
+ * mode: C
+ * c-set-style: "BSD"
+ * c-basic-offset: 4
+ * tab-width: 4
+ * indent-tabs-mode: nil
+ * End:
+ */

xen/include/public/arch-ia64.h

@@ -376,6 +376,9 @@ DEFINE_XEN_GUEST_HANDLE(vcpu_guest_context_t);
 /* expose the p2m table into domain */
 #define IA64_DOM0VP_expose_p2m          7
 
+/* xen perfmon */
+#define IA64_DOM0VP_perfmon             8
+
 // flags for page assignement to pseudo physical address space
 #define _ASSIGN_readonly                0
 #define ASSIGN_readonly                 (1UL << _ASSIGN_readonly)
@@ -462,6 +465,25 @@ struct xen_ia64_boot_param {
   (((unsigned long)(addr) & XENCOMM_INLINE_MASK) == XENCOMM_INLINE_FLAG)
 #define XENCOMM_INLINE_ADDR(addr) \
   ((unsigned long)(addr) & ~XENCOMM_INLINE_MASK)
+
+/* xen perfmon */
+#ifdef XEN
+#ifndef __ASSEMBLY__
+#ifndef _ASM_IA64_PERFMON_H
+
+#include <xen/list.h>   // asm/perfmon.h requires struct list_head
+#include <asm/perfmon.h>
+// for PFM_xxx and pfarg_features_t, pfarg_context_t, pfarg_reg_t, pfarg_load_t
+
+#endif /* _ASM_IA64_PERFMON_H */
+
+DEFINE_XEN_GUEST_HANDLE(pfarg_features_t);
+DEFINE_XEN_GUEST_HANDLE(pfarg_context_t);
+DEFINE_XEN_GUEST_HANDLE(pfarg_reg_t);
+DEFINE_XEN_GUEST_HANDLE(pfarg_load_t);
+#endif /* __ASSEMBLY__ */
+#endif /* XEN */
+
 #endif /* __HYPERVISOR_IF_IA64_H__ */
 
 /*