UST synchronization fix

[ust.git] / libust / buffers.h

diff --git a/libust/buffers.h b/libust/buffers.h
index d3267e102ca5e56a8b6da731ce778790d002f3f7..4fa6262af1715da97ebd69afecfa4084d61566dd 100644 (file)
--- a/libust/buffers.h
+++ b/libust/buffers.h
@@ -24,7 +24,9 @@
 #define _UST_BUFFERS_H
 
 #include <assert.h>
-#include <ust/kernelcompat.h>
+
+#include <ust/core.h>
+
 #include "usterr.h"
 #include "channels.h"
 #include "tracerconst.h"
@@ -104,10 +106,12 @@ struct ust_buffer {
  */
 enum force_switch_mode { FORCE_ACTIVE, FORCE_FLUSH };
 
-extern int ltt_reserve_slot_lockless_slow(struct ust_trace *trace,
-               struct ust_channel *ltt_channel, void **transport_data,
-               size_t data_size, size_t *slot_size, long *buf_offset, u64 *tsc,
-               unsigned int *rflags, int largest_align, int cpu);
+extern int ltt_reserve_slot_lockless_slow(struct ust_channel *chan,
+               struct ust_trace *trace, size_t data_size,
+               int largest_align, int cpu,
+               struct ust_buffer **ret_buf,
+               size_t *slot_size, long *buf_offset,
+               u64 *tsc, unsigned int *rflags);
 
 extern void ltt_force_switch_lockless_slow(struct ust_buffer *buf,
                enum force_switch_mode mode);
@@ -351,19 +355,21 @@ static __inline__ int ltt_relay_try_reserve(
        return 0;
 }
 
-static __inline__ int ltt_reserve_slot(struct ust_trace *trace,
-               struct ust_channel *chan, void **transport_data,
-               size_t data_size, size_t *slot_size, long *buf_offset, u64 *tsc,
-               unsigned int *rflags, int largest_align, int cpu)
+static __inline__ int ltt_reserve_slot(struct ust_channel *chan,
+                                      struct ust_trace *trace, size_t data_size,
+                                      int largest_align, int cpu,
+                                      struct ust_buffer **ret_buf,
+                                      size_t *slot_size, long *buf_offset, u64 *tsc,
+                                      unsigned int *rflags)
 {
-       struct ust_buffer *buf = chan->buf[cpu];
+       struct ust_buffer *buf = *ret_buf = chan->buf[cpu];
        long o_begin, o_end, o_old;
        size_t before_hdr_pad;
 
        /*
         * Perform retryable operations.
         */
-       /* FIXME: make this rellay per cpu? */
+       /* FIXME: make this really per cpu? */
        if (unlikely(LOAD_SHARED(ltt_nesting) > 4)) {
                DBG("Dropping event because nesting is too deep.");
                uatomic_inc(&buf->events_lost);
@@ -400,9 +406,10 @@ static __inline__ int ltt_reserve_slot(struct ust_trace *trace,
        *buf_offset = o_begin + before_hdr_pad;
        return 0;
 slow_path:
-       return ltt_reserve_slot_lockless_slow(trace, chan,
-               transport_data, data_size, slot_size, buf_offset, tsc,
-               rflags, largest_align, cpu);
+       return ltt_reserve_slot_lockless_slow(chan, trace, data_size,
+                                             largest_align, cpu, ret_buf,
+                                             slot_size, buf_offset, tsc,
+                                             rflags);
 }
 
 /*
@@ -476,16 +483,8 @@ static __inline__ void ltt_commit_slot(
        long endidx = SUBBUF_INDEX(offset_end - 1, chan);
        long commit_count;
 
-#ifdef LTT_NO_IPI_BARRIER
        smp_wmb();
-#else
-       /*
-        * Must write slot data before incrementing commit count.
-        * This compiler barrier is upgraded into a smp_mb() by the IPI
-        * sent by get_subbuf().
-        */
-       barrier();
-#endif
+
        uatomic_add(&buf->commit_count[endidx].cc, slot_size);
        /*
         * commit count read can race with concurrent OOO commit count updates.