Fix: timestamp_end field should include all events within sub-buffer
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Tue, 30 Apr 2019 15:23:44 +0000 (11:23 -0400)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Tue, 30 Apr 2019 19:00:54 +0000 (15:00 -0400)
Fix for timestamp_end not including all events within sub-buffer. This
happens if a thread is preempted/interrupted for a long time between
reserve and commit (e.g. in the middle of a packet), which causes the
timestamp used for timestamp_end field of the packet header to be lower
than the timestamp of the last events in the buffer (those following the
event that was preempted/interrupted between reserve and commit).

The fix involves sampling the timestamp when doing the last space
reservation in a sub-buffer (which necessarily happens before doing the
delivery after its last commit). Save this timestamp temporarily in a
per-sub-buffer control area (we have exclusive access to that area until
we increment the commit counter).

Then, that timestamp value will be read when delivering the sub-buffer,
whichever event or switch happens to be the last to increment the commit
counter to perform delivery. The timestamp value can be read without
worrying about concurrent access, because at that point sub-buffer
delivery has exclusive access to the sub-buffer.

This ensures the timestamp_end value is always larger or equal to the
timestamp of the last event, always below or equal the timestamp_begin
of the following packet, and always below or equal the timestamp of the
first event in the following packet.

This changes the layout of the ring buffer shared memory area, so we
need to bump the LTTNG_UST_ABI version from 7.2 to 8.0, thus requiring
locked-step upgrade between liblttng-ust in applications, session
daemon, and consumer daemon. This fix therefore cannot be backported
to existing stable releases.

Fixes: #1183
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
include/lttng/ust-abi.h
libringbuffer/frontend_types.h
libringbuffer/ring_buffer_frontend.c

index 01e9daf44a4dcf7f6915d3b20cc43a7b27ce2c8c..c1e13085bbf44cfa327aeabf5025e0c4f94c0aa6 100644 (file)
@@ -46,8 +46,8 @@
 #define LTTNG_UST_COMM_MAGIC                   0xC57C57C5
 
 /* Version for ABI between liblttng-ust, sessiond, consumerd */
-#define LTTNG_UST_ABI_MAJOR_VERSION            7
-#define LTTNG_UST_ABI_MINOR_VERSION            2
+#define LTTNG_UST_ABI_MAJOR_VERSION            8
+#define LTTNG_UST_ABI_MINOR_VERSION            0
 
 enum lttng_ust_instrumentation {
        LTTNG_UST_TRACEPOINT            = 0,
index 629abcbd5b4186592176f2609b03d4820a56b7b8..d0890408eb9bfa4965fda6f1326d00693530a6d8 100644 (file)
@@ -202,6 +202,20 @@ struct lttng_ust_lib_ring_buffer {
 
        DECLARE_SHMP(struct commit_counters_cold, commit_cold);
                                        /* Commit count per sub-buffer */
+       DECLARE_SHMP(uint64_t, ts_end); /*
+                                        * timestamp_end per sub-buffer.
+                                        * Time is sampled by the
+                                        * switch_*_end() callbacks
+                                        * which are the last space
+                                        * reservation performed in the
+                                        * sub-buffer before it can be
+                                        * fully committed and
+                                        * delivered. This time value is
+                                        * then read by the deliver
+                                        * callback, performed by the
+                                        * last commit before the buffer
+                                        * becomes readable.
+                                        */
        long active_readers;            /*
                                         * Active readers count
                                         * standard atomic access (shared)
index 9721df1634585f028316d13f96b184af7e99cc08..ce759ff18aa76ce638596b6f25d59d779a868edf 100644 (file)
@@ -194,6 +194,7 @@ void lib_ring_buffer_reset(struct lttng_ust_lib_ring_buffer *buf,
        for (i = 0; i < chan->backend.num_subbuf; i++) {
                struct commit_counters_hot *cc_hot;
                struct commit_counters_cold *cc_cold;
+               uint64_t *ts_end;
 
                cc_hot = shmp_index(handle, buf->commit_hot, i);
                if (!cc_hot)
@@ -201,9 +202,13 @@ void lib_ring_buffer_reset(struct lttng_ust_lib_ring_buffer *buf,
                cc_cold = shmp_index(handle, buf->commit_cold, i);
                if (!cc_cold)
                        return;
+               ts_end = shmp_index(handle, buf->ts_end, i);
+               if (!ts_end)
+                       return;
                v_set(config, &cc_hot->cc, 0);
                v_set(config, &cc_hot->seq, 0);
                v_set(config, &cc_cold->cc_sb, 0);
+               *ts_end = 0;
        }
        uatomic_set(&buf->consumed, 0);
        uatomic_set(&buf->record_disabled, 0);
@@ -368,6 +373,16 @@ int lib_ring_buffer_create(struct lttng_ust_lib_ring_buffer *buf,
                goto free_commit;
        }
 
+       align_shm(shmobj, __alignof__(uint64_t));
+       set_shmp(buf->ts_end,
+                zalloc_shm(shmobj,
+                       sizeof(uint64_t) * chan->backend.num_subbuf));
+       if (!shmp(handle, buf->ts_end)) {
+               ret = -ENOMEM;
+               goto free_commit_cold;
+       }
+
+
        ret = lib_ring_buffer_backend_create(&buf->backend, &chan->backend,
                        cpu, handle, shmobj);
        if (ret) {
@@ -414,6 +429,8 @@ int lib_ring_buffer_create(struct lttng_ust_lib_ring_buffer *buf,
 
        /* Error handling */
 free_init:
+       /* ts_end will be freed by shm teardown */
+free_commit_cold:
        /* commit_cold will be freed by shm teardown */
 free_commit:
        /* commit_hot will be freed by shm teardown */
@@ -1795,15 +1812,29 @@ void lib_ring_buffer_switch_old_end(struct lttng_ust_lib_ring_buffer *buf,
        unsigned long oldidx = subbuf_index(offsets->old - 1, chan);
        unsigned long commit_count, padding_size, data_size;
        struct commit_counters_hot *cc_hot;
+       uint64_t *ts_end;
 
        data_size = subbuf_offset(offsets->old - 1, chan) + 1;
        padding_size = chan->backend.subbuf_size - data_size;
        subbuffer_set_data_size(config, &buf->backend, oldidx, data_size,
                                handle);
 
+       ts_end = shmp_index(handle, buf->ts_end, oldidx);
+       if (!ts_end)
+               return;
        /*
-        * Order all writes to buffer before the commit count update that will
-        * determine that the subbuffer is full.
+        * This is the last space reservation in that sub-buffer before
+        * it gets delivered. This provides exclusive access to write to
+        * this sub-buffer's ts_end. There are also no concurrent
+        * readers of that ts_end because delivery of that sub-buffer is
+        * postponed until the commit counter is incremented for the
+        * current space reservation.
+        */
+       *ts_end = tsc;
+
+       /*
+        * Order all writes to buffer and store to ts_end before the commit
+        * count update that will determine that the subbuffer is full.
         */
        cmm_smp_wmb();
        cc_hot = shmp_index(handle, buf->commit_hot, oldidx);
@@ -1874,11 +1905,24 @@ void lib_ring_buffer_switch_new_end(struct lttng_ust_lib_ring_buffer *buf,
 {
        const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
        unsigned long endidx, data_size;
+       uint64_t *ts_end;
 
        endidx = subbuf_index(offsets->end - 1, chan);
        data_size = subbuf_offset(offsets->end - 1, chan) + 1;
        subbuffer_set_data_size(config, &buf->backend, endidx, data_size,
                                handle);
+       ts_end = shmp_index(handle, buf->ts_end, endidx);
+       if (!ts_end)
+               return;
+       /*
+        * This is the last space reservation in that sub-buffer before
+        * it gets delivered. This provides exclusive access to write to
+        * this sub-buffer's ts_end. There are also no concurrent
+        * readers of that ts_end because delivery of that sub-buffer is
+        * postponed until the commit counter is incremented for the
+        * current space reservation.
+        */
+       *ts_end = tsc;
 }
 
 /*
@@ -2445,14 +2489,26 @@ void lib_ring_buffer_check_deliver_slow(const struct lttng_ust_lib_ring_buffer_c
        if (caa_likely(v_cmpxchg(config, &cc_cold->cc_sb,
                                 old_commit_count, old_commit_count + 1)
                   == old_commit_count)) {
+               uint64_t *ts_end;
+
                /*
                 * Start of exclusive subbuffer access. We are
                 * guaranteed to be the last writer in this subbuffer
                 * and any other writer trying to access this subbuffer
                 * in this state is required to drop records.
+                *
+                * We can read the ts_end for the current sub-buffer
+                * which has been saved by the very last space
+                * reservation for the current sub-buffer.
+                *
+                * Order increment of commit counter before reading ts_end.
                 */
+               cmm_smp_mb();
+               ts_end = shmp_index(handle, buf->ts_end, idx);
+               if (!ts_end)
+                       return;
                deliver_count_events(config, buf, idx, handle);
-               config->cb.buffer_end(buf, tsc, idx,
+               config->cb.buffer_end(buf, *ts_end, idx,
                                      lib_ring_buffer_get_data_size(config,
                                                                buf,
                                                                idx,
This page took 0.029756 seconds and 4 git commands to generate.