Revert "Oprofile Multiplexing Patch"
[safe/jmp/linux-2.6] / arch / x86 / oprofile / nmi_int.c
index 11b7a51..287513a 100644 (file)
@@ -1,10 +1,11 @@
 /**
  * @file nmi_int.c
  *
- * @remark Copyright 2002 OProfile authors
+ * @remark Copyright 2002-2008 OProfile authors
  * @remark Read the file COPYING
  *
  * @author John Levon <levon@movementarian.org>
+ * @author Robert Richter <robert.richter@amd.com>
  */
 
 #include <linux/init.h>
 #include <asm/nmi.h>
 #include <asm/msr.h>
 #include <asm/apic.h>
+
 #include "op_counter.h"
 #include "op_x86_model.h"
 
-static struct op_x86_model_spec const * model;
-static struct op_msrs cpu_msrs[NR_CPUS];
-static unsigned long saved_lvtpc[NR_CPUS];
+static struct op_x86_model_spec const *model;
+static DEFINE_PER_CPU(struct op_msrs, cpu_msrs);
+static DEFINE_PER_CPU(unsigned long, saved_lvtpc);
 
 static int nmi_start(void);
 static void nmi_stop(void);
@@ -41,7 +42,6 @@ static int nmi_suspend(struct sys_device *dev, pm_message_t state)
        return 0;
 }
 
-
 static int nmi_resume(struct sys_device *dev)
 {
        if (nmi_enabled == 1)
@@ -49,29 +49,27 @@ static int nmi_resume(struct sys_device *dev)
        return 0;
 }
 
-
 static struct sysdev_class oprofile_sysclass = {
-       set_kset_name("oprofile"),
+       .name           = "oprofile",
        .resume         = nmi_resume,
        .suspend        = nmi_suspend,
 };
 
-
 static struct sys_device device_oprofile = {
        .id     = 0,
        .cls    = &oprofile_sysclass,
 };
 
-
 static int __init init_sysfs(void)
 {
        int error;
-       if (!(error = sysdev_class_register(&oprofile_sysclass)))
+
+       error = sysdev_class_register(&oprofile_sysclass);
+       if (!error)
                error = sysdev_register(&device_oprofile);
        return error;
 }
 
-
 static void exit_sysfs(void)
 {
        sysdev_unregister(&device_oprofile);
@@ -90,9 +88,9 @@ static int profile_exceptions_notify(struct notifier_block *self,
        int ret = NOTIFY_DONE;
        int cpu = smp_processor_id();
 
-       switch(val) {
+       switch (val) {
        case DIE_NMI:
-               if (model->check_ctrs(args->regs, &cpu_msrs[cpu]))
+               if (model->check_ctrs(args->regs, &per_cpu(cpu_msrs, cpu)))
                        ret = NOTIFY_STOP;
                break;
        default:
@@ -101,24 +99,24 @@ static int profile_exceptions_notify(struct notifier_block *self,
        return ret;
 }
 
-static void nmi_cpu_save_registers(struct op_msrs * msrs)
+static void nmi_cpu_save_registers(struct op_msrs *msrs)
 {
        unsigned int const nr_ctrs = model->num_counters;
-       unsigned int const nr_ctrls = model->num_controls; 
-       struct op_msr * counters = msrs->counters;
-       struct op_msr * controls = msrs->controls;
+       unsigned int const nr_ctrls = model->num_controls;
+       struct op_msr *counters = msrs->counters;
+       struct op_msr *controls = msrs->controls;
        unsigned int i;
 
        for (i = 0; i < nr_ctrs; ++i) {
-               if (counters[i].addr){
+               if (counters[i].addr) {
                        rdmsr(counters[i].addr,
                                counters[i].saved.low,
                                counters[i].saved.high);
                }
        }
+
        for (i = 0; i < nr_ctrls; ++i) {
-               if (controls[i].addr){
+               if (controls[i].addr) {
                        rdmsr(controls[i].addr,
                                controls[i].saved.low,
                                controls[i].saved.high);
@@ -126,27 +124,24 @@ static void nmi_cpu_save_registers(struct op_msrs * msrs)
        }
 }
 
-
-static void nmi_save_registers(void * dummy)
+static void nmi_save_registers(void *dummy)
 {
        int cpu = smp_processor_id();
-       struct op_msrs * msrs = &cpu_msrs[cpu];
+       struct op_msrs *msrs = &per_cpu(cpu_msrs, cpu);
        nmi_cpu_save_registers(msrs);
 }
 
-
 static void free_msrs(void)
 {
        int i;
        for_each_possible_cpu(i) {
-               kfree(cpu_msrs[i].counters);
-               cpu_msrs[i].counters = NULL;
-               kfree(cpu_msrs[i].controls);
-               cpu_msrs[i].controls = NULL;
+               kfree(per_cpu(cpu_msrs, i).counters);
+               per_cpu(cpu_msrs, i).counters = NULL;
+               kfree(per_cpu(cpu_msrs, i).controls);
+               per_cpu(cpu_msrs, i).controls = NULL;
        }
 }
 
-
 static int allocate_msrs(void)
 {
        int success = 1;
@@ -155,13 +150,15 @@ static int allocate_msrs(void)
 
        int i;
        for_each_possible_cpu(i) {
-               cpu_msrs[i].counters = kmalloc(counters_size, GFP_KERNEL);
-               if (!cpu_msrs[i].counters) {
+               per_cpu(cpu_msrs, i).counters = kmalloc(counters_size,
+                                                               GFP_KERNEL);
+               if (!per_cpu(cpu_msrs, i).counters) {
                        success = 0;
                        break;
                }
-               cpu_msrs[i].controls = kmalloc(controls_size, GFP_KERNEL);
-               if (!cpu_msrs[i].controls) {
+               per_cpu(cpu_msrs, i).controls = kmalloc(controls_size,
+                                                               GFP_KERNEL);
+               if (!per_cpu(cpu_msrs, i).controls) {
                        success = 0;
                        break;
                }
@@ -173,15 +170,14 @@ static int allocate_msrs(void)
        return success;
 }
 
-
-static void nmi_cpu_setup(void * dummy)
+static void nmi_cpu_setup(void *dummy)
 {
        int cpu = smp_processor_id();
-       struct op_msrs * msrs = &cpu_msrs[cpu];
+       struct op_msrs *msrs = &per_cpu(cpu_msrs, cpu);
        spin_lock(&oprofilefs_lock);
        model->setup_ctrs(msrs);
        spin_unlock(&oprofilefs_lock);
-       saved_lvtpc[cpu] = apic_read(APIC_LVTPC);
+       per_cpu(saved_lvtpc, cpu) = apic_read(APIC_LVTPC);
        apic_write(APIC_LVTPC, APIC_DM_NMI);
 }
 
@@ -193,13 +189,14 @@ static struct notifier_block profile_exceptions_nb = {
 
 static int nmi_setup(void)
 {
-       int err=0;
+       int err = 0;
        int cpu;
 
        if (!allocate_msrs())
                return -ENOMEM;
 
-       if ((err = register_die_notifier(&profile_exceptions_nb))){
+       err = register_die_notifier(&profile_exceptions_nb);
+       if (err) {
                free_msrs();
                return err;
        }
@@ -209,56 +206,56 @@ static int nmi_setup(void)
         */
 
        /* Assume saved/restored counters are the same on all CPUs */
-       model->fill_in_addresses(&cpu_msrs[0]);
-       for_each_possible_cpu (cpu) {
+       model->fill_in_addresses(&per_cpu(cpu_msrs, 0));
+       for_each_possible_cpu(cpu) {
                if (cpu != 0) {
-                       memcpy(cpu_msrs[cpu].counters, cpu_msrs[0].counters,
+                       memcpy(per_cpu(cpu_msrs, cpu).counters,
+                               per_cpu(cpu_msrs, 0).counters,
                                sizeof(struct op_msr) * model->num_counters);
 
-                       memcpy(cpu_msrs[cpu].controls, cpu_msrs[0].controls,
+                       memcpy(per_cpu(cpu_msrs, cpu).controls,
+                               per_cpu(cpu_msrs, 0).controls,
                                sizeof(struct op_msr) * model->num_controls);
                }
 
        }
-       on_each_cpu(nmi_save_registers, NULL, 0, 1);
-       on_each_cpu(nmi_cpu_setup, NULL, 0, 1);
+       on_each_cpu(nmi_save_registers, NULL, 1);
+       on_each_cpu(nmi_cpu_setup, NULL, 1);
        nmi_enabled = 1;
        return 0;
 }
 
-
-static void nmi_restore_registers(struct op_msrs * msrs)
+static void nmi_restore_registers(struct op_msrs *msrs)
 {
        unsigned int const nr_ctrs = model->num_counters;
-       unsigned int const nr_ctrls = model->num_controls; 
-       struct op_msr * counters = msrs->counters;
-       struct op_msr * controls = msrs->controls;
+       unsigned int const nr_ctrls = model->num_controls;
+       struct op_msr *counters = msrs->counters;
+       struct op_msr *controls = msrs->controls;
        unsigned int i;
 
        for (i = 0; i < nr_ctrls; ++i) {
-               if (controls[i].addr){
+               if (controls[i].addr) {
                        wrmsr(controls[i].addr,
                                controls[i].saved.low,
                                controls[i].saved.high);
                }
        }
+
        for (i = 0; i < nr_ctrs; ++i) {
-               if (counters[i].addr){
+               if (counters[i].addr) {
                        wrmsr(counters[i].addr,
                                counters[i].saved.low,
                                counters[i].saved.high);
                }
        }
 }
 
-static void nmi_cpu_shutdown(void * dummy)
+static void nmi_cpu_shutdown(void *dummy)
 {
        unsigned int v;
        int cpu = smp_processor_id();
-       struct op_msrs * msrs = &cpu_msrs[cpu];
+       struct op_msrs *msrs = &__get_cpu_var(cpu_msrs);
+
        /* restoring APIC_LVTPC can trigger an apic error because the delivery
         * mode and vector nr combination can be illegal. That's by design: on
         * power on apic lvt contain a zero vector nr which are legal only for
@@ -266,60 +263,56 @@ static void nmi_cpu_shutdown(void * dummy)
         */
        v = apic_read(APIC_LVTERR);
        apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
-       apic_write(APIC_LVTPC, saved_lvtpc[cpu]);
+       apic_write(APIC_LVTPC, per_cpu(saved_lvtpc, cpu));
        apic_write(APIC_LVTERR, v);
        nmi_restore_registers(msrs);
-       model->shutdown(msrs);
 }
 
 static void nmi_shutdown(void)
 {
+       struct op_msrs *msrs = &get_cpu_var(cpu_msrs);
        nmi_enabled = 0;
-       on_each_cpu(nmi_cpu_shutdown, NULL, 0, 1);
+       on_each_cpu(nmi_cpu_shutdown, NULL, 1);
        unregister_die_notifier(&profile_exceptions_nb);
+       model->shutdown(msrs);
        free_msrs();
+       put_cpu_var(cpu_msrs);
 }
 
-static void nmi_cpu_start(void * dummy)
+static void nmi_cpu_start(void *dummy)
 {
-       struct op_msrs const * msrs = &cpu_msrs[smp_processor_id()];
+       struct op_msrs const *msrs = &__get_cpu_var(cpu_msrs);
        model->start(msrs);
 }
 
 static int nmi_start(void)
 {
-       on_each_cpu(nmi_cpu_start, NULL, 0, 1);
+       on_each_cpu(nmi_cpu_start, NULL, 1);
        return 0;
 }
-static void nmi_cpu_stop(void * dummy)
+
+static void nmi_cpu_stop(void *dummy)
 {
-       struct op_msrs const * msrs = &cpu_msrs[smp_processor_id()];
+       struct op_msrs const *msrs = &__get_cpu_var(cpu_msrs);
        model->stop(msrs);
 }
+
 static void nmi_stop(void)
 {
-       on_each_cpu(nmi_cpu_stop, NULL, 0, 1);
+       on_each_cpu(nmi_cpu_stop, NULL, 1);
 }
 
-
 struct op_counter_config counter_config[OP_MAX_COUNTER];
 
-static int nmi_create_files(struct super_block * sb, struct dentry * root)
+static int nmi_create_files(struct super_block *sb, struct dentry *root)
 {
        unsigned int i;
 
        for (i = 0; i < model->num_counters; ++i) {
-               struct dentry * dir;
+               struct dentry *dir;
                char buf[4];
-               /* quick little hack to _not_ expose a counter if it is not
+
+               /* quick little hack to _not_ expose a counter if it is not
                 * available for use.  This should protect userspace app.
                 * NOTE:  assumes 1:1 mapping here (that counters are organized
                 *        sequentially in their struct assignment).
@@ -329,21 +322,21 @@ static int nmi_create_files(struct super_block * sb, struct dentry * root)
 
                snprintf(buf,  sizeof(buf), "%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); 
+               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;
 }
+
 static int p4force;
 module_param(p4force, int, 0);
-static int __init p4_init(char ** cpu_type)
+
+static int __init p4_init(char **cpu_type)
 {
        __u8 cpu_model = boot_cpu_data.x86_model;
 
@@ -356,15 +349,15 @@ static int __init p4_init(char ** cpu_type)
        return 1;
 #else
        switch (smp_num_siblings) {
-               case 1:
-                       *cpu_type = "i386/p4";
-                       model = &op_p4_spec;
-                       return 1;
-
-               case 2:
-                       *cpu_type = "i386/p4-ht";
-                       model = &op_p4_ht2_spec;
-                       return 1;
+       case 1:
+               *cpu_type = "i386/p4";
+               model = &op_p4_spec;
+               return 1;
+
+       case 2:
+               *cpu_type = "i386/p4-ht";
+               model = &op_p4_ht2_spec;
+               return 1;
        }
 #endif
 
@@ -373,25 +366,38 @@ static int __init p4_init(char ** cpu_type)
        return 0;
 }
 
-
-static int __init ppro_init(char ** cpu_type)
+static int __init ppro_init(char **cpu_type)
 {
        __u8 cpu_model = boot_cpu_data.x86_model;
 
-       if (cpu_model == 14)
+       switch (cpu_model) {
+       case 0 ... 2:
+               *cpu_type = "i386/ppro";
+               break;
+       case 3 ... 5:
+               *cpu_type = "i386/pii";
+               break;
+       case 6 ... 8:
+               *cpu_type = "i386/piii";
+               break;
+       case 9:
+               *cpu_type = "i386/p6_mobile";
+               break;
+       case 10 ... 13:
+               *cpu_type = "i386/p6";
+               break;
+       case 14:
                *cpu_type = "i386/core";
-       else if (cpu_model == 15)
+               break;
+       case 15: case 23:
+               *cpu_type = "i386/core_2";
+               break;
+       case 26:
                *cpu_type = "i386/core_2";
-       else if (cpu_model > 0xd)
+               break;
+       default:
+               /* Unknown */
                return 0;
-       else if (cpu_model == 9) {
-               *cpu_type = "i386/p6_mobile";
-       } else if (cpu_model > 5) {
-               *cpu_type = "i386/piii";
-       } else if (cpu_model > 2) {
-               *cpu_type = "i386/pii";
-       } else {
-               *cpu_type = "i386/ppro";
        }
 
        model = &op_ppro_spec;
@@ -406,72 +412,85 @@ int __init op_nmi_init(struct oprofile_operations *ops)
        __u8 vendor = boot_cpu_data.x86_vendor;
        __u8 family = boot_cpu_data.x86;
        char *cpu_type;
+       int ret = 0;
 
        if (!cpu_has_apic)
                return -ENODEV;
+
        switch (vendor) {
-               case X86_VENDOR_AMD:
-                       /* Needs to be at least an Athlon (or hammer in 32bit mode) */
+       case X86_VENDOR_AMD:
+               /* Needs to be at least an Athlon (or hammer in 32bit mode) */
 
-                       switch (family) {
-                       default:
+               switch (family) {
+               default:
+                       return -ENODEV;
+               case 6:
+                       model = &op_amd_spec;
+                       cpu_type = "i386/athlon";
+                       break;
+               case 0xf:
+                       model = &op_amd_spec;
+                       /* Actually it could be i386/hammer too, but give
+                        user space an consistent name. */
+                       cpu_type = "x86-64/hammer";
+                       break;
+               case 0x10:
+                       model = &op_amd_spec;
+                       cpu_type = "x86-64/family10";
+                       break;
+               case 0x11:
+                       model = &op_amd_spec;
+                       cpu_type = "x86-64/family11h";
+                       break;
+               }
+               break;
+
+       case X86_VENDOR_INTEL:
+               switch (family) {
+                       /* Pentium IV */
+               case 0xf:
+                       if (!p4_init(&cpu_type))
                                return -ENODEV;
-                       case 6:
-                               model = &op_athlon_spec;
-                               cpu_type = "i386/athlon";
-                               break;
-                       case 0xf:
-                               model = &op_athlon_spec;
-                               /* Actually it could be i386/hammer too, but give
-                                  user space an consistent name. */
-                               cpu_type = "x86-64/hammer";
-                               break;
-                       case 0x10:
-                               model = &op_athlon_spec;
-                               cpu_type = "x86-64/family10";
-                               break;
-                       }
                        break;
-               case X86_VENDOR_INTEL:
-                       switch (family) {
-                               /* Pentium IV */
-                               case 0xf:
-                                       if (!p4_init(&cpu_type))
-                                               return -ENODEV;
-                                       break;
-
-                               /* A P6-class processor */
-                               case 6:
-                                       if (!ppro_init(&cpu_type))
-                                               return -ENODEV;
-                                       break;
-
-                               default:
-                                       return -ENODEV;
-                       }
+
+                       /* A P6-class processor */
+               case 6:
+                       if (!ppro_init(&cpu_type))
+                               return -ENODEV;
                        break;
 
                default:
                        return -ENODEV;
+               }
+               break;
+
+       default:
+               return -ENODEV;
        }
 
-       init_sysfs();
-       using_nmi = 1;
+       /* default values, can be overwritten by model */
        ops->create_files = nmi_create_files;
        ops->setup = nmi_setup;
        ops->shutdown = nmi_shutdown;
        ops->start = nmi_start;
        ops->stop = nmi_stop;
        ops->cpu_type = cpu_type;
+
+       if (model->init)
+               ret = model->init(ops);
+       if (ret)
+               return ret;
+
+       init_sysfs();
+       using_nmi = 1;
        printk(KERN_INFO "oprofile: using NMI interrupt.\n");
        return 0;
 }
 
-
 void op_nmi_exit(void)
 {
        if (using_nmi)
                exit_sysfs();
+       if (model->exit)
+               model->exit();
 }