Fix: only print event errors every 1048576 hits

[lttng-ust.git] / libringbuffer / ring_buffer_frontend.c

diff --git a/libringbuffer/ring_buffer_frontend.c b/libringbuffer/ring_buffer_frontend.c
index 7de1d60c8961df072704e9698cbbf2bbf7296296..0996ecefc0313ff5ef8b955ce7dffd3a6a793849 100644 (file)
--- a/libringbuffer/ring_buffer_frontend.c
+++ b/libringbuffer/ring_buffer_frontend.c
@@ -38,6 +38,7 @@
  * Dual LGPL v2.1/GPL v2 license.
  */
 
+#define _GNU_SOURCE
 #include <sys/types.h>
 #include <sys/mman.h>
 #include <sys/stat.h>
@@ -57,6 +58,9 @@
 #define max(a, b)      ((a) > (b) ? (a) : (b))
 #endif
 
+/* Print DBG() messages about events lost only every 1048576 hits */
+#define DBG_PRINT_NR_LOST      (1UL << 20)
+
 /*
  * Use POSIX SHM: shm_open(3) and shm_unlink(3).
  * close(2) to close the fd returned by shm_open.
@@ -92,7 +96,8 @@ __thread unsigned int lib_ring_buffer_nesting;
 static
 void lib_ring_buffer_print_errors(struct channel *chan,
                                  struct lttng_ust_lib_ring_buffer *buf, int cpu,
-                                 struct lttng_ust_shm_handle *handle);
+                                 struct lttng_ust_shm_handle *handle)
+       __attribute__((unused));
 
 /**
  * lib_ring_buffer_reset - Reset ring buffer to initial values.
@@ -420,7 +425,7 @@ struct lttng_ust_shm_handle *channel_create(const struct lttng_ust_lib_ring_buff
                   void *buf_addr, size_t subbuf_size,
                   size_t num_subbuf, unsigned int switch_timer_interval,
                   unsigned int read_timer_interval,
-                  int *shm_fd, int *wait_fd, uint64_t *memory_map_size)
+                  int **shm_fd, int **wait_fd, uint64_t **memory_map_size)
 {
        int ret, cpu;
        size_t shmsize, chansize;
@@ -610,8 +615,8 @@ struct lttng_ust_lib_ring_buffer *channel_get_ring_buffer(
                                        const struct lttng_ust_lib_ring_buffer_config *config,
                                        struct channel *chan, int cpu,
                                        struct lttng_ust_shm_handle *handle,
-                                       int *shm_fd, int *wait_fd,
-                                       uint64_t *memory_map_size)
+                                       int **shm_fd, int **wait_fd,
+                                       uint64_t **memory_map_size)
 {
        struct shm_ref *ref;
 
@@ -946,12 +951,6 @@ void lib_ring_buffer_print_buffer_errors(struct lttng_ust_lib_ring_buffer *buf,
        const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
        unsigned long write_offset, cons_offset;
 
-       /*
-        * Can be called in the error path of allocation when
-        * trans_channel_data is not yet set.
-        */
-       if (!chan)
-               return;
        /*
         * No need to order commit_count, write_offset and cons_offset reads
         * because we execute at teardown when no more writer nor reader
@@ -1340,11 +1339,19 @@ int lib_ring_buffer_try_reserve_slow(struct lttng_ust_lib_ring_buffer *buf,
                                 - subbuf_trunc((unsigned long)
                                     uatomic_read(&buf->consumed), chan)
                                >= chan->backend.buf_size)) {
+                               unsigned long nr_lost;
+
                                /*
                                 * We do not overwrite non consumed buffers
                                 * and we are full : record is lost.
                                 */
+                               nr_lost = v_read(config, &buf->records_lost_full);
                                v_inc(config, &buf->records_lost_full);
+                               if ((nr_lost & (DBG_PRINT_NR_LOST - 1)) == 0) {
+                                       DBG("%lu or more records lost in (%s:%d) (buffer full)\n",
+                                               nr_lost + 1, chan->backend.name,
+                                               buf->backend.cpu);
+                               }
                                return -ENOBUFS;
                        } else {
                                /*
@@ -1355,13 +1362,21 @@ int lib_ring_buffer_try_reserve_slow(struct lttng_ust_lib_ring_buffer *buf,
                                 */
                        }
                } else {
+                       unsigned long nr_lost;
+
                        /*
                         * Next subbuffer reserve offset does not match the
                         * commit offset. Drop record in producer-consumer and
                         * overwrite mode. Caused by either a writer OOPS or too
                         * many nested writes over a reserve/commit pair.
                         */
+                       nr_lost = v_read(config, &buf->records_lost_wrap);
                        v_inc(config, &buf->records_lost_wrap);
+                       if ((nr_lost & (DBG_PRINT_NR_LOST - 1)) == 0) {
+                               DBG("%lu or more records lost in (%s:%d) (wrap-around)\n",
+                                       nr_lost + 1, chan->backend.name,
+                                       buf->backend.cpu);
+                       }
                        return -EIO;
                }
                offsets->size =
@@ -1375,11 +1390,20 @@ int lib_ring_buffer_try_reserve_slow(struct lttng_ust_lib_ring_buffer *buf,
                        + ctx->data_size;
                if (caa_unlikely(subbuf_offset(offsets->begin, chan)
                             + offsets->size > chan->backend.subbuf_size)) {
+                       unsigned long nr_lost;
+
                        /*
                         * Record too big for subbuffers, report error, don't
                         * complete the sub-buffer switch.
                         */
+                       nr_lost = v_read(config, &buf->records_lost_big);
                        v_inc(config, &buf->records_lost_big);
+                       if ((nr_lost & (DBG_PRINT_NR_LOST - 1)) == 0) {
+                               DBG("%lu or more records lost in (%s:%d) record size "
+                                       " of %zu bytes is too large for buffer\n",
+                                       nr_lost + 1, chan->backend.name,
+                                       buf->backend.cpu, offsets->size);
+                       }
                        return -ENOSPC;
                } else {
                        /*