cpqarray: switch to seq_file
[safe/jmp/linux-2.6] / drivers / oprofile / cpu_buffer.c
index 934ff15..a7aae24 100644 (file)
@@ -21,7 +21,6 @@
 
 #include <linux/sched.h>
 #include <linux/oprofile.h>
-#include <linux/vmalloc.h>
 #include <linux/errno.h>
 
 #include "event_buffer.h"
@@ -78,16 +77,20 @@ void free_cpu_buffers(void)
        op_ring_buffer_write = NULL;
 }
 
+#define RB_EVENT_HDR_SIZE 4
+
 int alloc_cpu_buffers(void)
 {
        int i;
 
        unsigned long buffer_size = oprofile_cpu_buffer_size;
+       unsigned long byte_size = buffer_size * (sizeof(struct op_sample) +
+                                                RB_EVENT_HDR_SIZE);
 
-       op_ring_buffer_read = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
+       op_ring_buffer_read = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS);
        if (!op_ring_buffer_read)
                goto fail;
-       op_ring_buffer_write = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
+       op_ring_buffer_write = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS);
        if (!op_ring_buffer_write)
                goto fail;
 
@@ -161,7 +164,7 @@ struct op_sample
 {
        entry->event = ring_buffer_lock_reserve
                (op_ring_buffer_write, sizeof(struct op_sample) +
-                size * sizeof(entry->sample->data[0]), &entry->irq_flags);
+                size * sizeof(entry->sample->data[0]));
        if (entry->event)
                entry->sample = ring_buffer_event_data(entry->event);
        else
@@ -178,24 +181,31 @@ struct op_sample
 
 int op_cpu_buffer_write_commit(struct op_entry *entry)
 {
-       return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event,
-                                        entry->irq_flags);
+       return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event);
 }
 
-struct op_sample *op_cpu_buffer_read_entry(int cpu)
+struct op_sample *op_cpu_buffer_read_entry(struct op_entry *entry, int cpu)
 {
        struct ring_buffer_event *e;
        e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
        if (e)
-               return ring_buffer_event_data(e);
+               goto event;
        if (ring_buffer_swap_cpu(op_ring_buffer_read,
                                 op_ring_buffer_write,
                                 cpu))
                return NULL;
        e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
        if (e)
-               return ring_buffer_event_data(e);
+               goto event;
        return NULL;
+
+event:
+       entry->event = e;
+       entry->sample = ring_buffer_event_data(e);
+       entry->size = (ring_buffer_event_length(e) - sizeof(struct op_sample))
+               / sizeof(entry->sample->data[0]);
+       entry->data = entry->sample->data;
+       return entry->sample;
 }
 
 unsigned long op_cpu_buffer_entries(int cpu)
@@ -204,6 +214,59 @@ unsigned long op_cpu_buffer_entries(int cpu)
                + ring_buffer_entries_cpu(op_ring_buffer_write, cpu);
 }
 
+static int
+op_add_code(struct oprofile_cpu_buffer *cpu_buf, unsigned long backtrace,
+           int is_kernel, struct task_struct *task)
+{
+       struct op_entry entry;
+       struct op_sample *sample;
+       unsigned long flags;
+       int size;
+
+       flags = 0;
+
+       if (backtrace)
+               flags |= TRACE_BEGIN;
+
+       /* notice a switch from user->kernel or vice versa */
+       is_kernel = !!is_kernel;
+       if (cpu_buf->last_is_kernel != is_kernel) {
+               cpu_buf->last_is_kernel = is_kernel;
+               flags |= KERNEL_CTX_SWITCH;
+               if (is_kernel)
+                       flags |= IS_KERNEL;
+       }
+
+       /* notice a task switch */
+       if (cpu_buf->last_task != task) {
+               cpu_buf->last_task = task;
+               flags |= USER_CTX_SWITCH;
+       }
+
+       if (!flags)
+               /* nothing to do */
+               return 0;
+
+       if (flags & USER_CTX_SWITCH)
+               size = 1;
+       else
+               size = 0;
+
+       sample = op_cpu_buffer_write_reserve(&entry, size);
+       if (!sample)
+               return -ENOMEM;
+
+       sample->eip = ESCAPE_CODE;
+       sample->event = flags;
+
+       if (size)
+               op_cpu_buffer_add_data(&entry, (unsigned long)task);
+
+       op_cpu_buffer_write_commit(&entry);
+
+       return 0;
+}
+
 static inline int
 op_add_sample(struct oprofile_cpu_buffer *cpu_buf,
              unsigned long pc, unsigned long event)
@@ -221,26 +284,18 @@ op_add_sample(struct oprofile_cpu_buffer *cpu_buf,
        return op_cpu_buffer_write_commit(&entry);
 }
 
-static inline int
-add_code(struct oprofile_cpu_buffer *buffer, unsigned long value)
-{
-       return op_add_sample(buffer, ESCAPE_CODE, value);
-}
-
-/* This must be safe from any context. It's safe writing here
- * because of the head/tail separation of the writer and reader
- * of the CPU buffer.
+/*
+ * This must be safe from any context.
  *
  * is_kernel is needed because on some architectures you cannot
  * tell if you are in kernel or user space simply by looking at
  * pc. We tag this in the buffer by generating kernel enter/exit
  * events whenever is_kernel changes
  */
-static int log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
-                     int is_kernel, unsigned long event)
+static int
+log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
+          unsigned long backtrace, int is_kernel, unsigned long event)
 {
-       struct task_struct *task;
-
        cpu_buf->sample_received++;
 
        if (pc == ESCAPE_CODE) {
@@ -248,23 +303,8 @@ static int log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
                return 0;
        }
 
-       is_kernel = !!is_kernel;
-
-       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;
-               if (add_code(cpu_buf, is_kernel))
-                       goto fail;
-       }
-
-       /* notice a task switch */
-       if (cpu_buf->last_task != task) {
-               cpu_buf->last_task = task;
-               if (add_code(cpu_buf, (unsigned long)task))
-                       goto fail;
-       }
+       if (op_add_code(cpu_buf, backtrace, is_kernel, current))
+               goto fail;
 
        if (op_add_sample(cpu_buf, pc, event))
                goto fail;
@@ -278,7 +318,6 @@ fail:
 
 static inline void oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
 {
-       add_code(cpu_buf, CPU_TRACE_BEGIN);
        cpu_buf->tracing = 1;
 }
 
@@ -292,21 +331,21 @@ __oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
                          unsigned long event, int is_kernel)
 {
        struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
-
-       if (!oprofile_backtrace_depth) {
-               log_sample(cpu_buf, pc, is_kernel, event);
-               return;
-       }
-
-       oprofile_begin_trace(cpu_buf);
+       unsigned long backtrace = oprofile_backtrace_depth;
 
        /*
         * if log_sample() fail we can't backtrace since we lost the
         * source of this event
         */
-       if (log_sample(cpu_buf, pc, is_kernel, event))
-               oprofile_ops.backtrace(regs, oprofile_backtrace_depth);
+       if (!log_sample(cpu_buf, pc, backtrace, is_kernel, event))
+               /* failed */
+               return;
+
+       if (!backtrace)
+               return;
 
+       oprofile_begin_trace(cpu_buf);
+       oprofile_ops.backtrace(regs, backtrace);
        oprofile_end_trace(cpu_buf);
 }
 
@@ -324,59 +363,75 @@ void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
        __oprofile_add_ext_sample(pc, regs, event, is_kernel);
 }
 
-#ifdef CONFIG_OPROFILE_IBS
-
-void oprofile_add_ibs_sample(struct pt_regs * const regs,
-                            unsigned int * const ibs_sample, int ibs_code)
+/*
+ * Add samples with data to the ring buffer.
+ *
+ * Use oprofile_add_data(&entry, val) to add data and
+ * oprofile_write_commit(&entry) to commit the sample.
+ */
+void
+oprofile_write_reserve(struct op_entry *entry, struct pt_regs * const regs,
+                      unsigned long pc, int code, int size)
 {
+       struct op_sample *sample;
        int is_kernel = !user_mode(regs);
        struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
-       struct task_struct *task;
-       int fail = 0;
 
        cpu_buf->sample_received++;
 
-       /* notice a switch from user->kernel or vice versa */
-       if (cpu_buf->last_is_kernel != is_kernel) {
-               if (add_code(cpu_buf, is_kernel))
-                       goto fail;
-               cpu_buf->last_is_kernel = is_kernel;
-       }
-
-       /* notice a task switch */
-       if (!is_kernel) {
-               task = current;
-               if (cpu_buf->last_task != task) {
-                       if (add_code(cpu_buf, (unsigned long)task))
-                               goto fail;
-                       cpu_buf->last_task = task;
-               }
-       }
+       /* no backtraces for samples with data */
+       if (op_add_code(cpu_buf, 0, is_kernel, current))
+               goto fail;
 
-       fail = fail || add_code(cpu_buf, ibs_code);
-       fail = fail || op_add_sample(cpu_buf, ibs_sample[0], ibs_sample[1]);
-       fail = fail || op_add_sample(cpu_buf, ibs_sample[2], ibs_sample[3]);
-       fail = fail || op_add_sample(cpu_buf, ibs_sample[4], ibs_sample[5]);
+       sample = op_cpu_buffer_write_reserve(entry, size + 2);
+       if (!sample)
+               goto fail;
+       sample->eip = ESCAPE_CODE;
+       sample->event = 0;              /* no flags */
 
-       if (ibs_code == IBS_OP_BEGIN) {
-               fail = fail || op_add_sample(cpu_buf, ibs_sample[6], ibs_sample[7]);
-               fail = fail || op_add_sample(cpu_buf, ibs_sample[8], ibs_sample[9]);
-               fail = fail || op_add_sample(cpu_buf, ibs_sample[10], ibs_sample[11]);
-       }
+       op_cpu_buffer_add_data(entry, code);
+       op_cpu_buffer_add_data(entry, pc);
 
-       if (!fail)
-               return;
+       return;
 
 fail:
+       entry->event = NULL;
        cpu_buf->sample_lost_overflow++;
 }
 
-#endif
+int oprofile_add_data(struct op_entry *entry, unsigned long val)
+{
+       if (!entry->event)
+               return 0;
+       return op_cpu_buffer_add_data(entry, val);
+}
+
+int oprofile_add_data64(struct op_entry *entry, u64 val)
+{
+       if (!entry->event)
+               return 0;
+       if (op_cpu_buffer_get_size(entry) < 2)
+               /*
+                * the function returns 0 to indicate a too small
+                * buffer, even if there is some space left
+                */
+               return 0;
+       if (!op_cpu_buffer_add_data(entry, (u32)val))
+               return 0;
+       return op_cpu_buffer_add_data(entry, (u32)(val >> 32));
+}
+
+int oprofile_write_commit(struct op_entry *entry)
+{
+       if (!entry->event)
+               return -EINVAL;
+       return op_cpu_buffer_write_commit(entry);
+}
 
 void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
 {
        struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
-       log_sample(cpu_buf, pc, is_kernel, event);
+       log_sample(cpu_buf, pc, 0, is_kernel, event);
 }
 
 void oprofile_add_trace(unsigned long pc)