sparc64: Use NMI oprofile profiling on cheetah and derivative cpus.

We use clock cycle counter, adjusted to HZ.

This can be extended to sun4v based processors as well, as they
also have a proper overflow interrupt facility for the performance
counters.

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 109 insertions, 1 deletions

diff --git a/arch/sparc/oprofile/init.c b/arch/sparc/oprofile/init.c
index 17bb6035069b..e979dcd77726 100644
--- a/arch/sparc/oprofile/init.c
+++ b/arch/sparc/oprofile/init.c
@@ -12,9 +12,117 @@
 #include <linux/errno.h>
 #include <linux/init.h>
  
+#ifdef CONFIG_SPARC64
+#include <asm/spitfire.h>
+#include <asm/cpudata.h>
+#include <asm/irq.h>
+
+static int nmi_enabled;
+
+static u64 picl_value(void)
+{
+        u32 delta = local_cpu_data().clock_tick / HZ;
+
+        return (0 - delta) & 0xffffffff;
+}
+
+#define PCR_PIC_PRIV    0x1 /* PIC access is privileged */
+#define PCR_STRACE      0x2 /* Trace supervisor events  */
+#define PCR_UTRACE      0x4 /* Trace user events        */
+
+static void nmi_handler(struct pt_regs *regs)
+{
+        write_pcr(PCR_PIC_PRIV);
+
+        if (nmi_enabled) {
+                oprofile_add_sample(regs, 0);
+
+                write_pic(picl_value());
+                write_pcr(PCR_PIC_PRIV | PCR_STRACE | PCR_UTRACE);
+        }
+}
+
+/* We count "clock cycle" events in the lower 32-bit PIC.
+ * Then configure it such that it overflows every HZ, and thus
+ * generates a level 15 interrupt at that frequency.
+ */
+static void cpu_nmi_start(void *_unused)
+{
+        write_pcr(PCR_PIC_PRIV);
+        write_pic(picl_value());
+
+        /* Bit 0: PIC access is privileged
+         * Bit 1: Supervisor Trace
+         * Bit 2: User Trace
+         *
+         * And the event selection code for cpu cycles is zero.
+         */
+        write_pcr(PCR_PIC_PRIV | PCR_STRACE | PCR_UTRACE);
+}
+
+static void cpu_nmi_stop(void *_unused)
+{
+        write_pcr(PCR_PIC_PRIV);
+}
+
+static int nmi_start(void)
+{
+        int err = register_perfctr_intr(nmi_handler);
+
+        if (!err) {
+                nmi_enabled = 1;
+                wmb();
+                err = on_each_cpu(cpu_nmi_start, NULL, 1);
+                if (err) {
+                        nmi_enabled = 0;
+                        wmb();
+                        on_each_cpu(cpu_nmi_stop, NULL, 1);
+                        release_perfctr_intr(nmi_handler);
+                }
+        }
+
+        return err;
+}
+
+static void nmi_stop(void)
+{
+        nmi_enabled = 0;
+        wmb();
+
+        on_each_cpu(cpu_nmi_stop, NULL, 1);
+        release_perfctr_intr(nmi_handler);
+        synchronize_sched();
+}
+
+static int oprofile_nmi_init(struct oprofile_operations *ops)
+{
+        if (tlb_type != cheetah && tlb_type != cheetah_plus)
+                return -ENODEV;
+
+        ops->create_files = NULL;
+        ops->setup = NULL;
+        ops->shutdown = NULL;
+        ops->start = nmi_start;
+        ops->stop = nmi_stop;
+        ops->cpu_type = "timer";
+
+        printk(KERN_INFO "oprofile: Using perfctr based NMI timer interrupt.\n");
+
+        return 0;
+}
+#endif
+
 int __init oprofile_arch_init(struct oprofile_operations *ops)
 {
-        return -ENODEV;
+        int ret = -ENODEV;
+
+#ifdef CONFIG_SPARC64
+        ret = oprofile_nmi_init(ops);
+        if (!ret)
+                return ret;
+#endif
+
+        return ret;
 }