oprofile: rename kernel-wide identifiers
[linux-2.6.git] / drivers / oprofile / cpu_buffer.c
index 7ba39fe..fcf96f6 100644 (file)
 #include <linux/oprofile.h>
 #include <linux/vmalloc.h>
 #include <linux/errno.h>
+
 #include "event_buffer.h"
 #include "cpu_buffer.h"
 #include "buffer_sync.h"
 #include "oprof.h"
 
+#define OP_BUFFER_FLAGS        0
+
+/*
+ * Read and write access is using spin locking. Thus, writing to the
+ * buffer by NMI handler (x86) could occur also during critical
+ * sections when reading the buffer. To avoid this, there are 2
+ * buffers for independent read and write access. Read access is in
+ * process context only, write access only in the NMI handler. If the
+ * read buffer runs empty, both buffers are swapped atomically. There
+ * is potentially a small window during swapping where the buffers are
+ * disabled and samples could be lost.
+ *
+ * Using 2 buffers is a little bit overhead, but the solution is clear
+ * and does not require changes in the ring buffer implementation. It
+ * can be changed to a single buffer solution when the ring buffer
+ * access is implemented as non-locking atomic code.
+ */
+struct ring_buffer *op_ring_buffer_read;
+struct ring_buffer *op_ring_buffer_write;
 DEFINE_PER_CPU(struct oprofile_cpu_buffer, cpu_buffer);
 
 static void wq_sync_buffer(struct work_struct *work);
@@ -37,17 +56,17 @@ static int work_enabled;
 
 void free_cpu_buffers(void)
 {
-       int i;
-
-       for_each_online_cpu(i) {
-               vfree(per_cpu(cpu_buffer, i).buffer);
-               per_cpu(cpu_buffer, i).buffer = NULL;
-       }
+       if (op_ring_buffer_read)
+               ring_buffer_free(op_ring_buffer_read);
+       op_ring_buffer_read = NULL;
+       if (op_ring_buffer_write)
+               ring_buffer_free(op_ring_buffer_write);
+       op_ring_buffer_write = NULL;
 }
 
 unsigned long oprofile_get_cpu_buffer_size(void)
 {
-       return fs_cpu_buffer_size;
+       return oprofile_cpu_buffer_size;
 }
 
 void oprofile_cpu_buffer_inc_smpl_lost(void)
@@ -61,17 +80,19 @@ void oprofile_cpu_buffer_inc_smpl_lost(void)
 int alloc_cpu_buffers(void)
 {
        int i;
-       unsigned long buffer_size = fs_cpu_buffer_size;
-       for_each_online_cpu(i) {
+
+       unsigned long buffer_size = oprofile_cpu_buffer_size;
+
+       op_ring_buffer_read = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
+       if (!op_ring_buffer_read)
+               goto fail;
+       op_ring_buffer_write = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
+       if (!op_ring_buffer_write)
+               goto fail;
+
+       for_each_possible_cpu(i) {
                struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
-               b->buffer = vmalloc_node(sizeof(struct op_sample) * buffer_size,
-                       cpu_to_node(i));
-               if (!b->buffer)
-                       goto fail;
+
                b->last_task = NULL;
                b->last_is_kernel = -1;
                b->tracing = 0;
@@ -124,57 +145,31 @@ void end_cpu_work(void)
        flush_scheduled_work();
 }
 
-/* Resets the cpu buffer to a sane state. */
-void cpu_buffer_reset(struct oprofile_cpu_buffer * cpu_buf)
+static inline int
+add_sample(struct oprofile_cpu_buffer *cpu_buf,
+          unsigned long pc, unsigned long event)
 {
-       /* reset these to invalid values; the next sample
-        * collected will populate the buffer with proper
-        * values to initialize the buffer
-        */
-       cpu_buf->last_is_kernel = -1;
-       cpu_buf->last_task = NULL;
-}
+       struct op_entry entry;
+       int ret;
 
-/* compute number of available slots in cpu_buffer queue */
-static unsigned long nr_available_slots(struct oprofile_cpu_buffer const * b)
-{
-       unsigned long head = b->head_pos;
-       unsigned long tail = b->tail_pos;
+       ret = cpu_buffer_write_entry(&entry);
+       if (ret)
+               return ret;
 
-       if (tail > head)
-               return (tail - head) - 1;
+       entry.sample->eip = pc;
+       entry.sample->event = event;
 
-       return tail + (b->buffer_size - head) - 1;
-}
+       ret = cpu_buffer_write_commit(&entry);
+       if (ret)
+               return ret;
 
-static void increment_head(struct oprofile_cpu_buffer * b)
-{
-       unsigned long new_head = b->head_pos + 1;
-
-       /* Ensure anything written to the slot before we
-        * increment is visible */
-       wmb();
-
-       if (new_head < b->buffer_size)
-               b->head_pos = new_head;
-       else
-               b->head_pos = 0;
-}
-
-static inline void
-add_sample(struct oprofile_cpu_buffer * cpu_buf,
-           unsigned long pc, unsigned long event)
-{
-       struct op_sample * entry = &cpu_buf->buffer[cpu_buf->head_pos];
-       entry->eip = pc;
-       entry->event = event;
-       increment_head(cpu_buf);
+       return 0;
 }
 
-static inline void
-add_code(struct oprofile_cpu_buffer * buffer, unsigned long value)
+static inline int
+add_code(struct oprofile_cpu_buffer *buffer, unsigned long value)
 {
-       add_sample(buffer, ESCAPE_CODE, value);
+       return add_sample(buffer, ESCAPE_CODE, value);
 }
 
 /* This must be safe from any context. It's safe writing here
@@ -186,10 +181,10 @@ add_code(struct oprofile_cpu_buffer * buffer, unsigned long value)
  * 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,
+static int log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
                      int is_kernel, unsigned long event)
 {
-       struct task_struct * task;
+       struct task_struct *task;
 
        cpu_buf->sample_received++;
 
@@ -198,11 +193,6 @@ static int log_sample(struct oprofile_cpu_buffer * cpu_buf, unsigned long pc,
                return 0;
        }
 
-       if (nr_available_slots(cpu_buf) < 3) {
-               cpu_buf->sample_lost_overflow++;
-               return 0;
-       }
-
        is_kernel = !!is_kernel;
 
        task = current;
@@ -210,32 +200,35 @@ static int log_sample(struct oprofile_cpu_buffer * cpu_buf, unsigned long pc,
        /* notice a switch from user->kernel or vice versa */
        if (cpu_buf->last_is_kernel != is_kernel) {
                cpu_buf->last_is_kernel = is_kernel;
-               add_code(cpu_buf, 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;
-               add_code(cpu_buf, (unsigned long)task);
+               if (add_code(cpu_buf, (unsigned long)task))
+                       goto fail;
        }
-       add_sample(cpu_buf, pc, event);
+
+       if (add_sample(cpu_buf, pc, event))
+               goto fail;
+
        return 1;
+
+fail:
+       cpu_buf->sample_lost_overflow++;
+       return 0;
 }
 
 static int oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
 {
-       if (nr_available_slots(cpu_buf) < 4) {
-               cpu_buf->sample_lost_overflow++;
-               return 0;
-       }
-
        add_code(cpu_buf, CPU_TRACE_BEGIN);
        cpu_buf->tracing = 1;
        return 1;
 }
 
-static void oprofile_end_trace(struct oprofile_cpu_buffer * cpu_buf)
+static void oprofile_end_trace(struct oprofile_cpu_buffer *cpu_buf)
 {
        cpu_buf->tracing = 0;
 }
@@ -245,7 +238,7 @@ void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
 {
        struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
 
-       if (!backtrace_depth) {
+       if (!oprofile_backtrace_depth) {
                log_sample(cpu_buf, pc, is_kernel, event);
                return;
        }
@@ -253,10 +246,12 @@ void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
        if (!oprofile_begin_trace(cpu_buf))
                return;
 
-       /* if log_sample() fail we can't backtrace since we lost the source
-        * of this event */
+       /*
+        * 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, backtrace_depth);
+               oprofile_ops.backtrace(regs, oprofile_backtrace_depth);
        oprofile_end_trace(cpu_buf);
 }
 
@@ -270,69 +265,57 @@ void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
 
 #ifdef CONFIG_OPROFILE_IBS
 
-#define MAX_IBS_SAMPLE_SIZE    14
-static int log_ibs_sample(struct oprofile_cpu_buffer *cpu_buf,
-       unsigned long pc, int is_kernel, unsigned  int *ibs, int ibs_code)
+#define MAX_IBS_SAMPLE_SIZE 14
+
+void oprofile_add_ibs_sample(struct pt_regs * const regs,
+                            unsigned int * const ibs_sample, int ibs_code)
 {
+       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++;
 
-       if (nr_available_slots(cpu_buf) < MAX_IBS_SAMPLE_SIZE) {
-               cpu_buf->sample_lost_overflow++;
-               return 0;
-       }
-
-       is_kernel = !!is_kernel;
-
        /* 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;
-               add_code(cpu_buf, 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;
-                       add_code(cpu_buf, (unsigned long)task);
                }
        }
 
-       add_code(cpu_buf, ibs_code);
-       add_sample(cpu_buf, ibs[0], ibs[1]);
-       add_sample(cpu_buf, ibs[2], ibs[3]);
-       add_sample(cpu_buf, ibs[4], ibs[5]);
+       fail = fail || add_code(cpu_buf, ibs_code);
+       fail = fail || add_sample(cpu_buf, ibs_sample[0], ibs_sample[1]);
+       fail = fail || add_sample(cpu_buf, ibs_sample[2], ibs_sample[3]);
+       fail = fail || add_sample(cpu_buf, ibs_sample[4], ibs_sample[5]);
 
        if (ibs_code == IBS_OP_BEGIN) {
-       add_sample(cpu_buf, ibs[6], ibs[7]);
-       add_sample(cpu_buf, ibs[8], ibs[9]);
-       add_sample(cpu_buf, ibs[10], ibs[11]);
+               fail = fail || add_sample(cpu_buf, ibs_sample[6], ibs_sample[7]);
+               fail = fail || add_sample(cpu_buf, ibs_sample[8], ibs_sample[9]);
+               fail = fail || add_sample(cpu_buf, ibs_sample[10], ibs_sample[11]);
        }
 
-       return 1;
-}
-
-void oprofile_add_ibs_sample(struct pt_regs *const regs,
-                               unsigned int * const ibs_sample, u8 code)
-{
-       int is_kernel = !user_mode(regs);
-       unsigned long pc = profile_pc(regs);
+       if (fail)
+               goto fail;
 
-       struct oprofile_cpu_buffer *cpu_buf =
-                        &per_cpu(cpu_buffer, smp_processor_id());
+       if (oprofile_backtrace_depth)
+               oprofile_ops.backtrace(regs, oprofile_backtrace_depth);
 
-       if (!backtrace_depth) {
-               log_ibs_sample(cpu_buf, pc, is_kernel, ibs_sample, code);
-               return;
-       }
+       return;
 
-       /* if log_sample() fails we can't backtrace since we lost the source
-       * of this event */
-       if (log_ibs_sample(cpu_buf, pc, is_kernel, ibs_sample, code))
-               oprofile_ops.backtrace(regs, backtrace_depth);
+fail:
+       cpu_buf->sample_lost_overflow++;
+       return;
 }
 
 #endif
@@ -350,21 +333,21 @@ void oprofile_add_trace(unsigned long pc)
        if (!cpu_buf->tracing)
                return;
 
-       if (nr_available_slots(cpu_buf) < 1) {
-               cpu_buf->tracing = 0;
-               cpu_buf->sample_lost_overflow++;
-               return;
-       }
+       /*
+        * broken frame can give an eip with the same value as an
+        * escape code, abort the trace if we get it
+        */
+       if (pc == ESCAPE_CODE)
+               goto fail;
 
-       /* broken frame can give an eip with the same value as an escape code,
-        * abort the trace if we get it */
-       if (pc == ESCAPE_CODE) {
-               cpu_buf->tracing = 0;
-               cpu_buf->backtrace_aborted++;
-               return;
-       }
+       if (add_sample(cpu_buf, pc, 0))
+               goto fail;
 
-       add_sample(cpu_buf, pc, 0);
+       return;
+fail:
+       cpu_buf->tracing = 0;
+       cpu_buf->backtrace_aborted++;
+       return;
 }
 
 /*
@@ -376,11 +359,16 @@ void oprofile_add_trace(unsigned long pc)
  */
 static void wq_sync_buffer(struct work_struct *work)
 {
-       struct oprofile_cpu_buffer * b =
+       struct oprofile_cpu_buffer *b =
                container_of(work, struct oprofile_cpu_buffer, work.work);
        if (b->cpu != smp_processor_id()) {
                printk(KERN_DEBUG "WQ on CPU%d, prefer CPU%d\n",
                       smp_processor_id(), b->cpu);
+
+               if (!cpu_online(b->cpu)) {
+                       cancel_delayed_work(&b->work);
+                       return;
+               }
        }
        sync_buffer(b->cpu);