]> git.karo-electronics.de Git - linux-beck.git/commitdiff
x86, perfcounters: prepare for fixed-mode PMCs
authorIngo Molnar <mingo@elte.hu>
Wed, 17 Dec 2008 08:09:13 +0000 (09:09 +0100)
committerIngo Molnar <mingo@elte.hu>
Tue, 23 Dec 2008 11:45:10 +0000 (12:45 +0100)
Impact: refactor the x86 code for fixed-mode PMCs

Extend the data structures and rename the existing facilities
to allow for a 'generic' versus 'fixed' counter distinction.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/include/asm/perf_counter.h
arch/x86/kernel/cpu/perf_counter.c
include/linux/perf_counter.h

index 9dadce1124ee64acba6486c9ae918d6e6b339268..dd5a4a559e2dcb6fffb8297dc874a75eda5d5489 100644 (file)
@@ -1,6 +1,13 @@
 #ifndef _ASM_X86_PERF_COUNTER_H
 #define _ASM_X86_PERF_COUNTER_H
 
+/*
+ * Performance counter hw details:
+ */
+
+#define X86_PMC_MAX_GENERIC                                    8
+#define X86_PMC_MAX_FIXED                                      3
+
 #define MSR_ARCH_PERFMON_PERFCTR0                            0xc1
 #define MSR_ARCH_PERFMON_PERFCTR1                            0xc2
 
 
 #define ARCH_PERFMON_BRANCH_MISSES_RETIRED                      6
 
+/*
+ * Intel "Architectural Performance Monitoring" CPUID
+ * detection/enumeration details:
+ */
 union cpuid10_eax {
        struct {
                unsigned int version_id:8;
index a4a3a09a654b5e1353bebe395a0ce74b6b752b01..fc3af868823261bce57e1b424a33d1aead22621d 100644 (file)
@@ -27,13 +27,12 @@ static bool perf_counters_initialized __read_mostly;
 static int nr_hw_counters __read_mostly;
 static u32 perf_counter_mask __read_mostly;
 
-/* No support for fixed function counters yet */
-
-#define MAX_HW_COUNTERS                8
-
 struct cpu_hw_counters {
-       struct perf_counter     *counters[MAX_HW_COUNTERS];
-       unsigned long           used[BITS_TO_LONGS(MAX_HW_COUNTERS)];
+       struct perf_counter     *generic[X86_PMC_MAX_GENERIC];
+       unsigned long           used[BITS_TO_LONGS(X86_PMC_MAX_GENERIC)];
+
+       struct perf_counter     *fixed[X86_PMC_MAX_FIXED];
+       unsigned long           used_fixed[BITS_TO_LONGS(X86_PMC_MAX_FIXED)];
 };
 
 /*
@@ -185,7 +184,7 @@ void hw_perf_restore(u64 ctrl)
 EXPORT_SYMBOL_GPL(hw_perf_restore);
 
 static inline void
-__x86_perf_counter_disable(struct perf_counter *counter,
+__pmc_generic_disable(struct perf_counter *counter,
                           struct hw_perf_counter *hwc, unsigned int idx)
 {
        int err;
@@ -193,7 +192,7 @@ __x86_perf_counter_disable(struct perf_counter *counter,
        err = wrmsr_safe(hwc->config_base + idx, hwc->config, 0);
 }
 
-static DEFINE_PER_CPU(u64, prev_left[MAX_HW_COUNTERS]);
+static DEFINE_PER_CPU(u64, prev_left[X86_PMC_MAX_GENERIC]);
 
 /*
  * Set the next IRQ period, based on the hwc->period_left value.
@@ -231,7 +230,7 @@ __hw_perf_counter_set_period(struct perf_counter *counter,
 }
 
 static void
-__x86_perf_counter_enable(struct perf_counter *counter,
+__pmc_generic_enable(struct perf_counter *counter,
                          struct hw_perf_counter *hwc, int idx)
 {
        wrmsr(hwc->config_base + idx,
@@ -241,7 +240,7 @@ __x86_perf_counter_enable(struct perf_counter *counter,
 /*
  * Find a PMC slot for the freshly enabled / scheduled in counter:
  */
-static void x86_perf_counter_enable(struct perf_counter *counter)
+static void pmc_generic_enable(struct perf_counter *counter)
 {
        struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
        struct hw_perf_counter *hwc = &counter->hw;
@@ -256,12 +255,12 @@ static void x86_perf_counter_enable(struct perf_counter *counter)
 
        perf_counters_lapic_init(hwc->nmi);
 
-       __x86_perf_counter_disable(counter, hwc, idx);
+       __pmc_generic_disable(counter, hwc, idx);
 
-       cpuc->counters[idx] = counter;
+       cpuc->generic[idx] = counter;
 
        __hw_perf_counter_set_period(counter, hwc, idx);
-       __x86_perf_counter_enable(counter, hwc, idx);
+       __pmc_generic_enable(counter, hwc, idx);
 }
 
 void perf_counter_print_debug(void)
@@ -301,16 +300,16 @@ void perf_counter_print_debug(void)
        local_irq_enable();
 }
 
-static void x86_perf_counter_disable(struct perf_counter *counter)
+static void pmc_generic_disable(struct perf_counter *counter)
 {
        struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
        struct hw_perf_counter *hwc = &counter->hw;
        unsigned int idx = hwc->idx;
 
-       __x86_perf_counter_disable(counter, hwc, idx);
+       __pmc_generic_disable(counter, hwc, idx);
 
        clear_bit(idx, cpuc->used);
-       cpuc->counters[idx] = NULL;
+       cpuc->generic[idx] = NULL;
 
        /*
         * Drain the remaining delta count out of a counter
@@ -349,7 +348,7 @@ static void perf_save_and_restart(struct perf_counter *counter)
        __hw_perf_counter_set_period(counter, hwc, idx);
 
        if (pmc_ctrl & ARCH_PERFMON_EVENTSEL0_ENABLE)
-               __x86_perf_counter_enable(counter, hwc, idx);
+               __pmc_generic_enable(counter, hwc, idx);
 }
 
 static void
@@ -392,7 +391,7 @@ static void __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi)
 again:
        ack = status;
        for_each_bit(bit, (unsigned long *) &status, nr_hw_counters) {
-               struct perf_counter *counter = cpuc->counters[bit];
+               struct perf_counter *counter = cpuc->generic[bit];
 
                clear_bit(bit, (unsigned long *) &status);
                if (!counter)
@@ -412,7 +411,7 @@ again:
                }
                /*
                 * From NMI context we cannot call into the scheduler to
-                * do a task wakeup - but we mark these counters as
+                * do a task wakeup - but we mark these generic as
                 * wakeup_pending and initate a wakeup callback:
                 */
                if (nmi) {
@@ -462,7 +461,7 @@ void perf_counter_notify(struct pt_regs *regs)
        cpuc = &per_cpu(cpu_hw_counters, cpu);
 
        for_each_bit(bit, cpuc->used, nr_hw_counters) {
-               struct perf_counter *counter = cpuc->counters[bit];
+               struct perf_counter *counter = cpuc->generic[bit];
 
                if (!counter)
                        continue;
@@ -539,10 +538,10 @@ void __init init_hw_perf_counters(void)
        printk(KERN_INFO "... version:      %d\n", eax.split.version_id);
        printk(KERN_INFO "... num_counters: %d\n", eax.split.num_counters);
        nr_hw_counters = eax.split.num_counters;
-       if (nr_hw_counters > MAX_HW_COUNTERS) {
-               nr_hw_counters = MAX_HW_COUNTERS;
+       if (nr_hw_counters > X86_PMC_MAX_GENERIC) {
+               nr_hw_counters = X86_PMC_MAX_GENERIC;
                WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!",
-                       nr_hw_counters, MAX_HW_COUNTERS);
+                       nr_hw_counters, X86_PMC_MAX_GENERIC);
        }
        perf_counter_mask = (1 << nr_hw_counters) - 1;
        perf_max_counters = nr_hw_counters;
@@ -556,15 +555,15 @@ void __init init_hw_perf_counters(void)
        register_die_notifier(&perf_counter_nmi_notifier);
 }
 
-static void x86_perf_counter_read(struct perf_counter *counter)
+static void pmc_generic_read(struct perf_counter *counter)
 {
        x86_perf_counter_update(counter, &counter->hw, counter->hw.idx);
 }
 
 static const struct hw_perf_counter_ops x86_perf_counter_ops = {
-       .hw_perf_counter_enable         = x86_perf_counter_enable,
-       .hw_perf_counter_disable        = x86_perf_counter_disable,
-       .hw_perf_counter_read           = x86_perf_counter_read,
+       .hw_perf_counter_enable         = pmc_generic_enable,
+       .hw_perf_counter_disable        = pmc_generic_disable,
+       .hw_perf_counter_read           = pmc_generic_read,
 };
 
 const struct hw_perf_counter_ops *
index d038450de87d5210614d5b0663ce301bb6edb38b..984da540224b61e4e1b3d01c32ba0803bc97629b 100644 (file)
@@ -14,6 +14,7 @@
 #define _LINUX_PERF_COUNTER_H
 
 #include <asm/atomic.h>
+#include <asm/perf_counter.h>
 
 #include <linux/list.h>
 #include <linux/mutex.h>