Performance: split check deliver fast/slow paths
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Sat, 16 Jul 2016 14:07:55 +0000 (10:07 -0400)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Sun, 21 Aug 2016 20:19:38 +0000 (16:19 -0400)
Performance improvement changelog from lttng-ust, ported back to
lttng-modules:

On ARMv7l (Cubietruck), the compiler generates a function call for each
lib_ring_buffer_check_deliver, even though it typically only do an
unlikely check. Split it into an inline fast path, and a function call
for the slow path. This brings a performance gain of about 500ns/event
on the Cubietruck.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
lib/ringbuffer/frontend_internal.h
lib/ringbuffer/ring_buffer_frontend.c

index 76bf20e79b1ae21ad217bc073c9e9f02a1709ee4..32785853b721bb3a6d0aebedce45607bab8efb0b 100644 (file)
@@ -156,6 +156,15 @@ extern
 void lib_ring_buffer_switch_slow(struct lib_ring_buffer *buf,
                                 enum switch_mode mode);
 
+extern
+void lib_ring_buffer_check_deliver_slow(const struct lib_ring_buffer_config *config,
+                                  struct lib_ring_buffer *buf,
+                                  struct channel *chan,
+                                  unsigned long offset,
+                                  unsigned long commit_count,
+                                  unsigned long idx,
+                                  u64 tsc);
+
 extern
 void lib_ring_buffer_switch_remote(struct lib_ring_buffer *buf);
 
@@ -192,58 +201,6 @@ void lib_ring_buffer_reserve_push_reader(struct lib_ring_buffer *buf,
                                              consumed_new) != consumed_old));
 }
 
-static inline
-void lib_ring_buffer_vmcore_check_deliver(const struct lib_ring_buffer_config *config,
-                                         struct lib_ring_buffer *buf,
-                                         unsigned long commit_count,
-                                         unsigned long idx)
-{
-       if (config->oops == RING_BUFFER_OOPS_CONSISTENCY)
-               v_set(config, &buf->commit_hot[idx].seq, commit_count);
-}
-
-static inline
-int lib_ring_buffer_poll_deliver(const struct lib_ring_buffer_config *config,
-                                struct lib_ring_buffer *buf,
-                                struct channel *chan)
-{
-       unsigned long consumed_old, consumed_idx, commit_count, write_offset;
-
-       consumed_old = atomic_long_read(&buf->consumed);
-       consumed_idx = subbuf_index(consumed_old, chan);
-       commit_count = v_read(config, &buf->commit_cold[consumed_idx].cc_sb);
-       /*
-        * No memory barrier here, since we are only interested
-        * in a statistically correct polling result. The next poll will
-        * get the data is we are racing. The mb() that ensures correct
-        * memory order is in get_subbuf.
-        */
-       write_offset = v_read(config, &buf->offset);
-
-       /*
-        * Check that the subbuffer we are trying to consume has been
-        * already fully committed.
-        */
-
-       if (((commit_count - chan->backend.subbuf_size)
-            & chan->commit_count_mask)
-           - (buf_trunc(consumed_old, chan)
-              >> chan->backend.num_subbuf_order)
-           != 0)
-               return 0;
-
-       /*
-        * Check that we are not about to read the same subbuffer in
-        * which the writer head is.
-        */
-       if (subbuf_trunc(write_offset, chan) - subbuf_trunc(consumed_old, chan)
-           == 0)
-               return 0;
-
-       return 1;
-
-}
-
 static inline
 int lib_ring_buffer_pending_data(const struct lib_ring_buffer_config *config,
                                 struct lib_ring_buffer *buf,
@@ -314,106 +271,9 @@ void lib_ring_buffer_check_deliver(const struct lib_ring_buffer_config *config,
 
        /* Check if all commits have been done */
        if (unlikely((buf_trunc(offset, chan) >> chan->backend.num_subbuf_order)
-                    - (old_commit_count & chan->commit_count_mask) == 0)) {
-               /*
-                * If we succeeded at updating cc_sb below, we are the subbuffer
-                * writer delivering the subbuffer. Deals with concurrent
-                * updates of the "cc" value without adding a add_return atomic
-                * operation to the fast path.
-                *
-                * We are doing the delivery in two steps:
-                * - First, we cmpxchg() cc_sb to the new value
-                *   old_commit_count + 1. This ensures that we are the only
-                *   subbuffer user successfully filling the subbuffer, but we
-                *   do _not_ set the cc_sb value to "commit_count" yet.
-                *   Therefore, other writers that would wrap around the ring
-                *   buffer and try to start writing to our subbuffer would
-                *   have to drop records, because it would appear as
-                *   non-filled.
-                *   We therefore have exclusive access to the subbuffer control
-                *   structures.  This mutual exclusion with other writers is
-                *   crucially important to perform record overruns count in
-                *   flight recorder mode locklessly.
-                * - When we are ready to release the subbuffer (either for
-                *   reading or for overrun by other writers), we simply set the
-                *   cc_sb value to "commit_count" and perform delivery.
-                *
-                * The subbuffer size is least 2 bytes (minimum size: 1 page).
-                * This guarantees that old_commit_count + 1 != commit_count.
-                */
-
-               /*
-                * Order prior updates to reserve count prior to the
-                * commit_cold cc_sb update.
-                */
-               smp_wmb();
-               if (likely(v_cmpxchg(config, &buf->commit_cold[idx].cc_sb,
-                                        old_commit_count, old_commit_count + 1)
-                          == old_commit_count)) {
-                       /*
-                        * 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.
-                        */
-                       v_add(config,
-                             subbuffer_get_records_count(config,
-                                                         &buf->backend, idx),
-                             &buf->records_count);
-                       v_add(config,
-                             subbuffer_count_records_overrun(config,
-                                                             &buf->backend,
-                                                             idx),
-                             &buf->records_overrun);
-                       config->cb.buffer_end(buf, tsc, idx,
-                                             lib_ring_buffer_get_data_size(config,
-                                                                       buf,
-                                                                       idx));
-
-                       /*
-                        * Increment the packet counter while we have exclusive
-                        * access.
-                        */
-                       subbuffer_inc_packet_count(config, &buf->backend, idx);
-
-                       /*
-                        * Set noref flag and offset for this subbuffer id.
-                        * Contains a memory barrier that ensures counter stores
-                        * are ordered before set noref and offset.
-                        */
-                       lib_ring_buffer_set_noref_offset(config, &buf->backend, idx,
-                                                        buf_trunc_val(offset, chan));
-
-                       /*
-                        * Order set_noref and record counter updates before the
-                        * end of subbuffer exclusive access. Orders with
-                        * respect to writers coming into the subbuffer after
-                        * wrap around, and also order wrt concurrent readers.
-                        */
-                       smp_mb();
-                       /* End of exclusive subbuffer access */
-                       v_set(config, &buf->commit_cold[idx].cc_sb,
-                             commit_count);
-                       /*
-                        * Order later updates to reserve count after
-                        * the commit_cold cc_sb update.
-                        */
-                       smp_wmb();
-                       lib_ring_buffer_vmcore_check_deliver(config, buf,
-                                                        commit_count, idx);
-
-                       /*
-                        * RING_BUFFER_WAKEUP_BY_WRITER wakeup is not lock-free.
-                        */
-                       if (config->wakeup == RING_BUFFER_WAKEUP_BY_WRITER
-                           && atomic_long_read(&buf->active_readers)
-                           && lib_ring_buffer_poll_deliver(config, buf, chan)) {
-                               wake_up_interruptible(&buf->read_wait);
-                               wake_up_interruptible(&chan->read_wait);
-                       }
-
-               }
-       }
+                    - (old_commit_count & chan->commit_count_mask) == 0))
+               lib_ring_buffer_check_deliver_slow(config, buf, chan, offset,
+                       commit_count, idx, tsc);
 }
 
 /*
index 89eefe96bdd05baafe93ddb776aff84e84c24d15..c5123a6415e39afc75cc55b8208b9b6924fbf975 100644 (file)
@@ -97,6 +97,47 @@ static
 void _lib_ring_buffer_switch_remote(struct lib_ring_buffer *buf,
                enum switch_mode mode);
 
+static
+int lib_ring_buffer_poll_deliver(const struct lib_ring_buffer_config *config,
+                                struct lib_ring_buffer *buf,
+                                struct channel *chan)
+{
+       unsigned long consumed_old, consumed_idx, commit_count, write_offset;
+
+       consumed_old = atomic_long_read(&buf->consumed);
+       consumed_idx = subbuf_index(consumed_old, chan);
+       commit_count = v_read(config, &buf->commit_cold[consumed_idx].cc_sb);
+       /*
+        * No memory barrier here, since we are only interested
+        * in a statistically correct polling result. The next poll will
+        * get the data is we are racing. The mb() that ensures correct
+        * memory order is in get_subbuf.
+        */
+       write_offset = v_read(config, &buf->offset);
+
+       /*
+        * Check that the subbuffer we are trying to consume has been
+        * already fully committed.
+        */
+
+       if (((commit_count - chan->backend.subbuf_size)
+            & chan->commit_count_mask)
+           - (buf_trunc(consumed_old, chan)
+              >> chan->backend.num_subbuf_order)
+           != 0)
+               return 0;
+
+       /*
+        * Check that we are not about to read the same subbuffer in
+        * which the writer head is.
+        */
+       if (subbuf_trunc(write_offset, chan) - subbuf_trunc(consumed_old, chan)
+           == 0)
+               return 0;
+
+       return 1;
+}
+
 /*
  * Must be called under cpu hotplug protection.
  */
@@ -1929,6 +1970,128 @@ int lib_ring_buffer_reserve_slow(struct lib_ring_buffer_ctx *ctx)
 }
 EXPORT_SYMBOL_GPL(lib_ring_buffer_reserve_slow);
 
+static
+void lib_ring_buffer_vmcore_check_deliver(const struct lib_ring_buffer_config *config,
+                                         struct lib_ring_buffer *buf,
+                                         unsigned long commit_count,
+                                         unsigned long idx)
+{
+       if (config->oops == RING_BUFFER_OOPS_CONSISTENCY)
+               v_set(config, &buf->commit_hot[idx].seq, commit_count);
+}
+
+void lib_ring_buffer_check_deliver_slow(const struct lib_ring_buffer_config *config,
+                                  struct lib_ring_buffer *buf,
+                                  struct channel *chan,
+                                  unsigned long offset,
+                                  unsigned long commit_count,
+                                  unsigned long idx,
+                                  u64 tsc)
+{
+       unsigned long old_commit_count = commit_count
+                                        - chan->backend.subbuf_size;
+
+       /*
+        * If we succeeded at updating cc_sb below, we are the subbuffer
+        * writer delivering the subbuffer. Deals with concurrent
+        * updates of the "cc" value without adding a add_return atomic
+        * operation to the fast path.
+        *
+        * We are doing the delivery in two steps:
+        * - First, we cmpxchg() cc_sb to the new value
+        *   old_commit_count + 1. This ensures that we are the only
+        *   subbuffer user successfully filling the subbuffer, but we
+        *   do _not_ set the cc_sb value to "commit_count" yet.
+        *   Therefore, other writers that would wrap around the ring
+        *   buffer and try to start writing to our subbuffer would
+        *   have to drop records, because it would appear as
+        *   non-filled.
+        *   We therefore have exclusive access to the subbuffer control
+        *   structures.  This mutual exclusion with other writers is
+        *   crucially important to perform record overruns count in
+        *   flight recorder mode locklessly.
+        * - When we are ready to release the subbuffer (either for
+        *   reading or for overrun by other writers), we simply set the
+        *   cc_sb value to "commit_count" and perform delivery.
+        *
+        * The subbuffer size is least 2 bytes (minimum size: 1 page).
+        * This guarantees that old_commit_count + 1 != commit_count.
+        */
+
+       /*
+        * Order prior updates to reserve count prior to the
+        * commit_cold cc_sb update.
+        */
+       smp_wmb();
+       if (likely(v_cmpxchg(config, &buf->commit_cold[idx].cc_sb,
+                                old_commit_count, old_commit_count + 1)
+                  == old_commit_count)) {
+               /*
+                * 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.
+                */
+               v_add(config,
+                     subbuffer_get_records_count(config,
+                                                 &buf->backend, idx),
+                     &buf->records_count);
+               v_add(config,
+                     subbuffer_count_records_overrun(config,
+                                                     &buf->backend,
+                                                     idx),
+                     &buf->records_overrun);
+               config->cb.buffer_end(buf, tsc, idx,
+                                     lib_ring_buffer_get_data_size(config,
+                                                               buf,
+                                                               idx));
+
+               /*
+                * Increment the packet counter while we have exclusive
+                * access.
+                */
+               subbuffer_inc_packet_count(config, &buf->backend, idx);
+
+               /*
+                * Set noref flag and offset for this subbuffer id.
+                * Contains a memory barrier that ensures counter stores
+                * are ordered before set noref and offset.
+                */
+               lib_ring_buffer_set_noref_offset(config, &buf->backend, idx,
+                                                buf_trunc_val(offset, chan));
+
+               /*
+                * Order set_noref and record counter updates before the
+                * end of subbuffer exclusive access. Orders with
+                * respect to writers coming into the subbuffer after
+                * wrap around, and also order wrt concurrent readers.
+                */
+               smp_mb();
+               /* End of exclusive subbuffer access */
+               v_set(config, &buf->commit_cold[idx].cc_sb,
+                     commit_count);
+               /*
+                * Order later updates to reserve count after
+                * the commit_cold cc_sb update.
+                */
+               smp_wmb();
+               lib_ring_buffer_vmcore_check_deliver(config, buf,
+                                                commit_count, idx);
+
+               /*
+                * RING_BUFFER_WAKEUP_BY_WRITER wakeup is not lock-free.
+                */
+               if (config->wakeup == RING_BUFFER_WAKEUP_BY_WRITER
+                   && atomic_long_read(&buf->active_readers)
+                   && lib_ring_buffer_poll_deliver(config, buf, chan)) {
+                       wake_up_interruptible(&buf->read_wait);
+                       wake_up_interruptible(&chan->read_wait);
+               }
+
+       }
+}
+EXPORT_SYMBOL_GPL(lib_ring_buffer_check_deliver_slow);
+
 int __init init_lib_ring_buffer_frontend(void)
 {
        int cpu;
This page took 0.030212 seconds and 4 git commands to generate.