X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=libringbuffer%2Fring_buffer_frontend.c;h=ea73bcc676351ac76f25eeae85282c1aa17bc25a;hb=d1a996f5524540ebc8caa7c3c1bd3049018d212c;hp=1fc44912b51f2db96eb30348740645e18f6a25bd;hpb=38fae1d354502ffb37f471fbd0fef03412606cc0;p=lttng-ust.git

diff --git a/libringbuffer/ring_buffer_frontend.c b/libringbuffer/ring_buffer_frontend.c
index 1fc44912..ea73bcc6 100644
--- a/libringbuffer/ring_buffer_frontend.c
+++ b/libringbuffer/ring_buffer_frontend.c
@@ -1,7 +1,22 @@
 /*
  * ring_buffer_frontend.c
  *
- * (C) Copyright 2005-2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ * Copyright (C) 2005-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; only
+ * version 2.1 of the License.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ *
  *
  * Ring buffer wait-free buffer synchronization. Producer-consumer and flight
  * recorder (overwrite) modes. See thesis:
@@ -34,27 +49,42 @@
  *   - splice one subbuffer worth of data to a pipe
  *   - splice the data from pipe to disk/network
  *   - put_subbuf
- *
- * Dual LGPL v2.1/GPL v2 license.
  */
 
+#define _GNU_SOURCE
 #include <sys/types.h>
 #include <sys/mman.h>
 #include <sys/stat.h>
+#include <unistd.h>
 #include <fcntl.h>
+#include <signal.h>
+#include <time.h>
 #include <urcu/compiler.h>
 #include <urcu/ref.h>
+#include <urcu/tls-compat.h>
+#include <helper.h>
 
 #include "smp.h"
-#include <ust/ringbuffer-config.h>
+#include <lttng/ringbuffer-config.h>
+#include "vatomic.h"
 #include "backend.h"
 #include "frontend.h"
 #include "shm.h"
+#include "tlsfixup.h"
+#include "../liblttng-ust/compat.h"	/* For ENODATA */
 
 #ifndef max
 #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)
+
+#define LTTNG_UST_RB_SIG_FLUSH		SIGRTMIN
+#define LTTNG_UST_RB_SIG_READ		SIGRTMIN + 1
+#define LTTNG_UST_RB_SIG_TEARDOWN	SIGRTMIN + 2
+#define CLOCKID		CLOCK_MONOTONIC
+
 /*
  * Use POSIX SHM: shm_open(3) and shm_unlink(3).
  * close(2) to close the fd returned by shm_open.
@@ -81,27 +111,32 @@ struct switch_offsets {
 		     switch_old_end:1;
 };
 
-__thread unsigned int lib_ring_buffer_nesting;
+DEFINE_URCU_TLS(unsigned int, lib_ring_buffer_nesting);
+
+/*
+ * wakeup_fd_mutex protects wakeup fd use by timer from concurrent
+ * close.
+ */
+static pthread_mutex_t wakeup_fd_mutex = PTHREAD_MUTEX_INITIALIZER;
 
 static
 void lib_ring_buffer_print_errors(struct channel *chan,
-				  struct lib_ring_buffer *buf, int cpu,
-				  struct lttng_ust_shm_handle *handle);
+				struct lttng_ust_lib_ring_buffer *buf, int cpu,
+				struct lttng_ust_shm_handle *handle);
 
 /*
- * Must be called under cpu hotplug protection.
+ * Handle timer teardown race wrt memory free of private data by
+ * ring buffer signals are handled by a single thread, which permits
+ * a synchronization point between handling of each signal.
+ * Protected by the ust mutex.
  */
-void lib_ring_buffer_free(struct lib_ring_buffer *buf,
-			  struct lttng_ust_shm_handle *handle)
-{
-	struct channel *chan = shmp(handle, buf->backend.chan);
-
-	lib_ring_buffer_print_errors(chan, buf, buf->backend.cpu, handle);
-	/* buf->commit_hot will be freed by shm teardown */
-	/* buf->commit_cold will be freed by shm teardown */
+struct timer_signal_data {
+	pthread_t tid;	/* thread id managing signals */
+	int setup_done;
+	int qs_done;
+};
 
-	lib_ring_buffer_backend_free(&buf->backend);
-}
+static struct timer_signal_data timer_signal;
 
 /**
  * lib_ring_buffer_reset - Reset ring buffer to initial values.
@@ -112,11 +147,11 @@ void lib_ring_buffer_free(struct lib_ring_buffer *buf,
  * should not be using the iterator concurrently with reset. The previous
  * current iterator record is reset.
  */
-void lib_ring_buffer_reset(struct lib_ring_buffer *buf,
+void lib_ring_buffer_reset(struct lttng_ust_lib_ring_buffer *buf,
 			   struct lttng_ust_shm_handle *handle)
 {
 	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned int i;
 
 	/*
@@ -167,17 +202,16 @@ void channel_reset(struct channel *chan)
 /*
  * Must be called under cpu hotplug protection.
  */
-int lib_ring_buffer_create(struct lib_ring_buffer *buf,
+int lib_ring_buffer_create(struct lttng_ust_lib_ring_buffer *buf,
 			   struct channel_backend *chanb, int cpu,
 			   struct lttng_ust_shm_handle *handle,
 			   struct shm_object *shmobj)
 {
-	const struct lib_ring_buffer_config *config = &chanb->config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chanb->config;
 	struct channel *chan = caa_container_of(chanb, struct channel, backend);
-	void *priv = chanb->priv;
-	unsigned int num_subbuf;
+	void *priv = channel_get_private(chan);
 	size_t subbuf_header_size;
-	u64 tsc;
+	uint64_t tsc;
 	int ret;
 
 	/* Test for cpu hotplug */
@@ -207,9 +241,6 @@ int lib_ring_buffer_create(struct lib_ring_buffer *buf,
 		goto free_commit;
 	}
 
-	num_subbuf = chan->backend.num_subbuf;
-	//init_waitqueue_head(&buf->read_wait);
-
 	/*
 	 * Write the subbuffer header for first subbuffer so we know the total
 	 * duration of data gathering.
@@ -235,171 +266,412 @@ free_init:
 free_commit:
 	/* commit_hot will be freed by shm teardown */
 free_chanbuf:
-	lib_ring_buffer_backend_free(&buf->backend);
 	return ret;
 }
 
-#if 0
-static void switch_buffer_timer(unsigned long data)
+static
+void lib_ring_buffer_channel_switch_timer(int sig, siginfo_t *si, void *uc)
 {
-	struct lib_ring_buffer *buf = (struct lib_ring_buffer *)data;
-	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config;
+	struct lttng_ust_shm_handle *handle;
+	struct channel *chan;
+	int cpu;
+
+	assert(CMM_LOAD_SHARED(timer_signal.tid) == pthread_self());
+
+	chan = si->si_value.sival_ptr;
+	handle = chan->handle;
+	config = &chan->backend.config;
+
+	DBG("Switch timer for channel %p\n", chan);
 
 	/*
 	 * Only flush buffers periodically if readers are active.
 	 */
-	if (uatomic_read(&buf->active_readers) || uatomic_read(&buf->active_shadow_readers))
-		lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE, handle);
-
-	//TODO timers
-	//if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
-	//	mod_timer_pinned(&buf->switch_timer,
-	//			 jiffies + chan->switch_timer_interval);
-	//else
-	//	mod_timer(&buf->switch_timer,
-	//		  jiffies + chan->switch_timer_interval);
+	pthread_mutex_lock(&wakeup_fd_mutex);
+	if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
+		for_each_possible_cpu(cpu) {
+			struct lttng_ust_lib_ring_buffer *buf =
+				shmp(handle, chan->backend.buf[cpu].shmp);
+			if (uatomic_read(&buf->active_readers))
+				lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE,
+					chan->handle);
+		}
+	} else {
+		struct lttng_ust_lib_ring_buffer *buf =
+			shmp(handle, chan->backend.buf[0].shmp);
+
+		if (uatomic_read(&buf->active_readers))
+			lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE,
+				chan->handle);
+	}
+	pthread_mutex_unlock(&wakeup_fd_mutex);
+	return;
 }
-#endif //0
 
-static void lib_ring_buffer_start_switch_timer(struct lib_ring_buffer *buf,
-			   struct lttng_ust_shm_handle *handle)
+static
+void lib_ring_buffer_channel_do_read(struct channel *chan)
 {
-	struct channel *chan = shmp(handle, buf->backend.chan);
-	//const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config;
+	struct lttng_ust_shm_handle *handle;
+	int cpu;
+
+	handle = chan->handle;
+	config = &chan->backend.config;
+
+	/*
+	 * Only flush buffers periodically if readers are active.
+	 */
+	pthread_mutex_lock(&wakeup_fd_mutex);
+	if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
+		for_each_possible_cpu(cpu) {
+			struct lttng_ust_lib_ring_buffer *buf =
+				shmp(handle, chan->backend.buf[cpu].shmp);
+
+			if (uatomic_read(&buf->active_readers)
+			    && lib_ring_buffer_poll_deliver(config, buf,
+					chan, handle)) {
+				lib_ring_buffer_wakeup(buf, handle);
+			}
+		}
+	} else {
+		struct lttng_ust_lib_ring_buffer *buf =
+			shmp(handle, chan->backend.buf[0].shmp);
+
+		if (uatomic_read(&buf->active_readers)
+		    && lib_ring_buffer_poll_deliver(config, buf,
+				chan, handle)) {
+			lib_ring_buffer_wakeup(buf, handle);
+		}
+	}
+	pthread_mutex_unlock(&wakeup_fd_mutex);
+}
+
+static
+void lib_ring_buffer_channel_read_timer(int sig, siginfo_t *si, void *uc)
+{
+	struct channel *chan;
+
+	assert(CMM_LOAD_SHARED(timer_signal.tid) == pthread_self());
+	chan = si->si_value.sival_ptr;
+	DBG("Read timer for channel %p\n", chan);
+	lib_ring_buffer_channel_do_read(chan);
+	return;
+}
+
+static
+void rb_setmask(sigset_t *mask)
+{
+	int ret;
+
+	ret = sigemptyset(mask);
+	if (ret) {
+		PERROR("sigemptyset");
+	}
+	ret = sigaddset(mask, LTTNG_UST_RB_SIG_FLUSH);
+	if (ret) {
+		PERROR("sigaddset");
+	}
+	ret = sigaddset(mask, LTTNG_UST_RB_SIG_READ);
+	if (ret) {
+		PERROR("sigaddset");
+	}
+	ret = sigaddset(mask, LTTNG_UST_RB_SIG_TEARDOWN);
+	if (ret) {
+		PERROR("sigaddset");
+	}
+}
 
-	if (!chan->switch_timer_interval || buf->switch_timer_enabled)
+static
+void *sig_thread(void *arg)
+{
+	sigset_t mask;
+	siginfo_t info;
+	int signr;
+
+	/* Only self thread will receive signal mask. */
+	rb_setmask(&mask);
+	CMM_STORE_SHARED(timer_signal.tid, pthread_self());
+
+	for (;;) {
+		signr = sigwaitinfo(&mask, &info);
+		if (signr == -1) {
+			if (errno != EINTR)
+				PERROR("sigwaitinfo");
+			continue;
+		}
+		if (signr == LTTNG_UST_RB_SIG_FLUSH) {
+			lib_ring_buffer_channel_switch_timer(info.si_signo,
+					&info, NULL);
+		} else if (signr == LTTNG_UST_RB_SIG_READ) {
+			lib_ring_buffer_channel_read_timer(info.si_signo,
+					&info, NULL);
+		} else if (signr == LTTNG_UST_RB_SIG_TEARDOWN) {
+			cmm_smp_mb();
+			CMM_STORE_SHARED(timer_signal.qs_done, 1);
+			cmm_smp_mb();
+		} else {
+			ERR("Unexptected signal %d\n", info.si_signo);
+		}
+	}
+	return NULL;
+}
+
+/*
+ * Called with ust_lock() held.
+ * Ensure only a single thread listens on the timer signal.
+ */
+static
+void lib_ring_buffer_setup_timer_thread(void)
+{
+	pthread_t thread;
+	int ret;
+
+	if (timer_signal.setup_done)
 		return;
-	//TODO
-	//init_timer(&buf->switch_timer);
-	//buf->switch_timer.function = switch_buffer_timer;
-	//buf->switch_timer.expires = jiffies + chan->switch_timer_interval;
-	//buf->switch_timer.data = (unsigned long)buf;
-	//if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
-	//	add_timer_on(&buf->switch_timer, buf->backend.cpu);
-	//else
-	//	add_timer(&buf->switch_timer);
-	buf->switch_timer_enabled = 1;
+
+	ret = pthread_create(&thread, NULL, &sig_thread, NULL);
+	if (ret) {
+		errno = ret;
+		PERROR("pthread_create");
+	}
+	ret = pthread_detach(thread);
+	if (ret) {
+		errno = ret;
+		PERROR("pthread_detach");
+	}
+	timer_signal.setup_done = 1;
 }
 
-static void lib_ring_buffer_stop_switch_timer(struct lib_ring_buffer *buf,
-			   struct lttng_ust_shm_handle *handle)
+/*
+ * Called with ust_lock() held.
+ */
+static
+void lib_ring_buffer_channel_switch_timer_start(struct channel *chan)
 {
-	struct channel *chan = shmp(handle, buf->backend.chan);
+	struct sigevent sev;
+	struct itimerspec its;
+	int ret;
 
-	if (!chan->switch_timer_interval || !buf->switch_timer_enabled)
+	if (!chan->switch_timer_interval || chan->switch_timer_enabled)
 		return;
 
-	//TODO
-	//del_timer_sync(&buf->switch_timer);
-	buf->switch_timer_enabled = 0;
+	chan->switch_timer_enabled = 1;
+
+	lib_ring_buffer_setup_timer_thread();
+
+	sev.sigev_notify = SIGEV_SIGNAL;
+	sev.sigev_signo = LTTNG_UST_RB_SIG_FLUSH;
+	sev.sigev_value.sival_ptr = chan;
+	ret = timer_create(CLOCKID, &sev, &chan->switch_timer);
+	if (ret == -1) {
+		PERROR("timer_create");
+	}
+
+	its.it_value.tv_sec = chan->switch_timer_interval / 1000000;
+	its.it_value.tv_nsec = chan->switch_timer_interval % 1000000;
+	its.it_interval.tv_sec = its.it_value.tv_sec;
+	its.it_interval.tv_nsec = its.it_value.tv_nsec;
+
+	ret = timer_settime(chan->switch_timer, 0, &its, NULL);
+	if (ret == -1) {
+		PERROR("timer_settime");
+	}
 }
 
-#if 0
 /*
- * Polling timer to check the channels for data.
+ * Called with ust_lock() held.
  */
-static void read_buffer_timer(unsigned long data)
+static
+void lib_ring_buffer_channel_switch_timer_stop(struct channel *chan)
 {
-	struct lib_ring_buffer *buf = (struct lib_ring_buffer *)data;
-	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	sigset_t pending_set;
+	int ret;
 
-	CHAN_WARN_ON(chan, !buf->backend.allocated);
+	if (!chan->switch_timer_interval || !chan->switch_timer_enabled)
+		return;
 
-	if (uatomic_read(&buf->active_readers) || uatomic_read(&buf->active_shadow_readers))
-	    && lib_ring_buffer_poll_deliver(config, buf, chan)) {
-		//TODO
-		//wake_up_interruptible(&buf->read_wait);
-		//wake_up_interruptible(&chan->read_wait);
+	ret = timer_delete(chan->switch_timer);
+	if (ret == -1) {
+		PERROR("timer_delete");
 	}
 
-	//TODO
-	//if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
-	//	mod_timer_pinned(&buf->read_timer,
-	//			 jiffies + chan->read_timer_interval);
-	//else
-	//	mod_timer(&buf->read_timer,
-	//		  jiffies + chan->read_timer_interval);
+	/*
+	 * Ensure we don't have any signal queued for this channel.
+	 */
+	for (;;) {
+		ret = sigemptyset(&pending_set);
+		if (ret == -1) {
+			PERROR("sigemptyset");
+		}
+		ret = sigpending(&pending_set);
+		if (ret == -1) {
+			PERROR("sigpending");
+		}
+		if (!sigismember(&pending_set, LTTNG_UST_RB_SIG_FLUSH))
+			break;
+		caa_cpu_relax();
+	}
+
+	/*
+	 * From this point, no new signal handler will be fired that
+	 * would try to access "chan". However, we still need to wait
+	 * for any currently executing handler to complete.
+	 */
+	cmm_smp_mb();
+	CMM_STORE_SHARED(timer_signal.qs_done, 0);
+	cmm_smp_mb();
+
+	/*
+	 * Kill with LTTNG_UST_RB_SIG_TEARDOWN, so signal management
+	 * thread wakes up.
+	 */
+	kill(getpid(), LTTNG_UST_RB_SIG_TEARDOWN);
+
+	while (!CMM_LOAD_SHARED(timer_signal.qs_done))
+		caa_cpu_relax();
+	cmm_smp_mb();
+
+	chan->switch_timer = 0;
+	chan->switch_timer_enabled = 0;
 }
-#endif //0
 
-static void lib_ring_buffer_start_read_timer(struct lib_ring_buffer *buf,
-			   struct lttng_ust_shm_handle *handle)
+/*
+ * Called with ust_lock() held.
+ */
+static
+void lib_ring_buffer_channel_read_timer_start(struct channel *chan)
 {
-	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
+	struct sigevent sev;
+	struct itimerspec its;
+	int ret;
 
 	if (config->wakeup != RING_BUFFER_WAKEUP_BY_TIMER
-	    || !chan->read_timer_interval
-	    || buf->read_timer_enabled)
+			|| !chan->read_timer_interval || chan->read_timer_enabled)
 		return;
 
-	//TODO
-	//init_timer(&buf->read_timer);
-	//buf->read_timer.function = read_buffer_timer;
-	//buf->read_timer.expires = jiffies + chan->read_timer_interval;
-	//buf->read_timer.data = (unsigned long)buf;
-
-	//if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
-	//	add_timer_on(&buf->read_timer, buf->backend.cpu);
-	//else
-	//	add_timer(&buf->read_timer);
-	buf->read_timer_enabled = 1;
+	chan->read_timer_enabled = 1;
+
+	lib_ring_buffer_setup_timer_thread();
+
+	sev.sigev_notify = SIGEV_SIGNAL;
+	sev.sigev_signo = LTTNG_UST_RB_SIG_READ;
+	sev.sigev_value.sival_ptr = chan;
+	ret = timer_create(CLOCKID, &sev, &chan->read_timer);
+	if (ret == -1) {
+		PERROR("timer_create");
+	}
+
+	its.it_value.tv_sec = chan->read_timer_interval / 1000000;
+	its.it_value.tv_nsec = chan->read_timer_interval % 1000000;
+	its.it_interval.tv_sec = its.it_value.tv_sec;
+	its.it_interval.tv_nsec = its.it_value.tv_nsec;
+
+	ret = timer_settime(chan->read_timer, 0, &its, NULL);
+	if (ret == -1) {
+		PERROR("timer_settime");
+	}
 }
 
-static void lib_ring_buffer_stop_read_timer(struct lib_ring_buffer *buf,
-			   struct lttng_ust_shm_handle *handle)
+/*
+ * Called with ust_lock() held.
+ */
+static
+void lib_ring_buffer_channel_read_timer_stop(struct channel *chan)
 {
-	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
+	sigset_t pending_set;
+	int ret;
 
 	if (config->wakeup != RING_BUFFER_WAKEUP_BY_TIMER
-	    || !chan->read_timer_interval
-	    || !buf->read_timer_enabled)
+			|| !chan->read_timer_interval || !chan->read_timer_enabled)
 		return;
 
-	//TODO
-	//del_timer_sync(&buf->read_timer);
+	ret = timer_delete(chan->read_timer);
+	if (ret == -1) {
+		PERROR("timer_delete");
+	}
+
 	/*
 	 * do one more check to catch data that has been written in the last
 	 * timer period.
 	 */
-	if (lib_ring_buffer_poll_deliver(config, buf, chan, handle)) {
-		//TODO
-		//wake_up_interruptible(&buf->read_wait);
-		//wake_up_interruptible(&chan->read_wait);
+	lib_ring_buffer_channel_do_read(chan);
+
+
+	/*
+	 * Ensure we don't have any signal queued for this channel.
+	 */
+	for (;;) {
+		ret = sigemptyset(&pending_set);
+		if (ret == -1) {
+			PERROR("sigemptyset");
+		}
+		ret = sigpending(&pending_set);
+		if (ret == -1) {
+			PERROR("sigpending");
+		}
+		if (!sigismember(&pending_set, LTTNG_UST_RB_SIG_READ))
+			break;
+		caa_cpu_relax();
 	}
-	buf->read_timer_enabled = 0;
+
+	/*
+	 * From this point, no new signal handler will be fired that
+	 * would try to access "chan". However, we still need to wait
+	 * for any currently executing handler to complete.
+	 */
+	cmm_smp_mb();
+	CMM_STORE_SHARED(timer_signal.qs_done, 0);
+	cmm_smp_mb();
+
+	/*
+	 * Kill with LTTNG_UST_RB_SIG_TEARDOWN, so signal management
+	 * thread wakes up.
+	 */
+	kill(getpid(), LTTNG_UST_RB_SIG_TEARDOWN);
+
+	while (!CMM_LOAD_SHARED(timer_signal.qs_done))
+		caa_cpu_relax();
+	cmm_smp_mb();
+
+	chan->read_timer = 0;
+	chan->read_timer_enabled = 0;
 }
 
 static void channel_unregister_notifiers(struct channel *chan,
 			   struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	lib_ring_buffer_channel_switch_timer_stop(chan);
+	lib_ring_buffer_channel_read_timer_stop(chan);
+}
+
+static void channel_print_errors(struct channel *chan,
+		struct lttng_ust_shm_handle *handle)
+{
+	const struct lttng_ust_lib_ring_buffer_config *config =
+			&chan->backend.config;
 	int cpu;
 
 	if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 		for_each_possible_cpu(cpu) {
-			struct lib_ring_buffer *buf = shmp(handle, chan->backend.buf[cpu].shmp);
-
-			lib_ring_buffer_stop_switch_timer(buf, handle);
-			lib_ring_buffer_stop_read_timer(buf, handle);
+			struct lttng_ust_lib_ring_buffer *buf =
+				shmp(handle, chan->backend.buf[cpu].shmp);
+			lib_ring_buffer_print_errors(chan, buf, cpu, handle);
 		}
 	} else {
-		struct lib_ring_buffer *buf = shmp(handle, chan->backend.buf[0].shmp);
+		struct lttng_ust_lib_ring_buffer *buf =
+			shmp(handle, chan->backend.buf[0].shmp);
 
-		lib_ring_buffer_stop_switch_timer(buf, handle);
-		lib_ring_buffer_stop_read_timer(buf, handle);
+		lib_ring_buffer_print_errors(chan, buf, -1, handle);
 	}
-	//channel_backend_unregister_notifiers(&chan->backend);
 }
 
-static void channel_free(struct channel *chan, struct lttng_ust_shm_handle *handle,
-		int shadow)
+static void channel_free(struct channel *chan,
+		struct lttng_ust_shm_handle *handle)
 {
-	if (!shadow)
-		channel_backend_free(&chan->backend, handle);
+	channel_backend_free(&chan->backend, handle);
 	/* chan is freed by shm teardown */
 	shm_object_table_destroy(handle->table);
 	free(handle);
@@ -409,7 +681,9 @@ static void channel_free(struct channel *chan, struct lttng_ust_shm_handle *hand
  * channel_create - Create channel.
  * @config: ring buffer instance configuration
  * @name: name of the channel
- * @priv: ring buffer client private data
+ * @priv_data: ring buffer client private data area pointer (output)
+ * @priv_data_size: length, in bytes, of the private data area.
+ * @priv_data_init: initialization data for private data.
  * @buf_addr: pointer the the beginning of the preallocated buffer contiguous
  *            address mapping. It is used only by RING_BUFFER_STATIC
  *            configuration. It can be set to NULL for other backends.
@@ -423,19 +697,27 @@ static void channel_free(struct channel *chan, struct lttng_ust_shm_handle *hand
  * Holds cpu hotplug.
  * Returns NULL on failure.
  */
-struct lttng_ust_shm_handle *channel_create(const struct lib_ring_buffer_config *config,
-		   const char *name, void *priv, void *buf_addr,
-		   size_t subbuf_size,
+struct lttng_ust_shm_handle *channel_create(const struct lttng_ust_lib_ring_buffer_config *config,
+		   const char *name,
+		   void **priv_data,
+		   size_t priv_data_align,
+		   size_t priv_data_size,
+		   void *priv_data_init,
+		   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)
+		   unsigned int read_timer_interval)
 {
-	int ret, cpu;
-	size_t shmsize;
+	int ret;
+	size_t shmsize, chansize;
 	struct channel *chan;
 	struct lttng_ust_shm_handle *handle;
 	struct shm_object *shmobj;
-	struct shm_ref *ref;
+	unsigned int nr_streams;
+
+	if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
+		nr_streams = num_possible_cpus();
+	else
+		nr_streams = 1;
 
 	if (lib_ring_buffer_check_config(config, switch_timer_interval,
 					 read_timer_interval))
@@ -452,54 +734,56 @@ struct lttng_ust_shm_handle *channel_create(const struct lib_ring_buffer_config
 
 	/* Calculate the shm allocation layout */
 	shmsize = sizeof(struct channel);
-	if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
-		shmsize += sizeof(struct lib_ring_buffer_shmp) * num_possible_cpus();
-	else
-		shmsize += sizeof(struct lib_ring_buffer_shmp);
-
-	shmobj = shm_object_table_append(handle->table, shmsize);
+	shmsize += offset_align(shmsize, __alignof__(struct lttng_ust_lib_ring_buffer_shmp));
+	shmsize += sizeof(struct lttng_ust_lib_ring_buffer_shmp) * nr_streams;
+	chansize = shmsize;
+	if (priv_data_align)
+		shmsize += offset_align(shmsize, priv_data_align);
+	shmsize += priv_data_size;
+
+	/* Allocate normal memory for channel (not shared) */
+	shmobj = shm_object_table_alloc(handle->table, shmsize, SHM_OBJECT_MEM);
 	if (!shmobj)
 		goto error_append;
 	/* struct channel is at object 0, offset 0 (hardcoded) */
-	set_shmp(handle->chan, zalloc_shm(shmobj, sizeof(struct channel)));
+	set_shmp(handle->chan, zalloc_shm(shmobj, chansize));
 	assert(handle->chan._ref.index == 0);
 	assert(handle->chan._ref.offset == 0);
 	chan = shmp(handle, handle->chan);
 	if (!chan)
 		goto error_append;
+	chan->nr_streams = nr_streams;
+
+	/* space for private data */
+	if (priv_data_size) {
+		DECLARE_SHMP(void, priv_data_alloc);
+
+		align_shm(shmobj, priv_data_align);
+		chan->priv_data_offset = shmobj->allocated_len;
+		set_shmp(priv_data_alloc, zalloc_shm(shmobj, priv_data_size));
+		if (!shmp(handle, priv_data_alloc))
+			goto error_append;
+		*priv_data = channel_get_private(chan);
+		memcpy(*priv_data, priv_data_init, priv_data_size);
+	} else {
+		chan->priv_data_offset = -1;
+		if (priv_data)
+			*priv_data = NULL;
+	}
 
-	ret = channel_backend_init(&chan->backend, name, config, priv,
+	ret = channel_backend_init(&chan->backend, name, config,
 				   subbuf_size, num_subbuf, handle);
 	if (ret)
 		goto error_backend_init;
 
+	chan->handle = handle;
 	chan->commit_count_mask = (~0UL >> chan->backend.num_subbuf_order);
-	//TODO
-	//chan->switch_timer_interval = usecs_to_jiffies(switch_timer_interval);
-	//chan->read_timer_interval = usecs_to_jiffies(read_timer_interval);
-	//TODO
-	//init_waitqueue_head(&chan->read_wait);
-	//init_waitqueue_head(&chan->hp_wait);
 
-	if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
-		/*
-		 * In case of non-hotplug cpu, if the ring-buffer is allocated
-		 * in early initcall, it will not be notified of secondary cpus.
-		 * In that off case, we need to allocate for all possible cpus.
-		 */
-		for_each_possible_cpu(cpu) {
-			struct lib_ring_buffer *buf = shmp(handle, chan->backend.buf[cpu].shmp);
-			lib_ring_buffer_start_switch_timer(buf, handle);
-			lib_ring_buffer_start_read_timer(buf, handle);
-		}
-	} else {
-		struct lib_ring_buffer *buf = shmp(handle, chan->backend.buf[0].shmp);
+	chan->switch_timer_interval = switch_timer_interval;
+	chan->read_timer_interval = read_timer_interval;
+	lib_ring_buffer_channel_switch_timer_start(chan);
+	lib_ring_buffer_channel_read_timer_start(chan);
 
-		lib_ring_buffer_start_switch_timer(buf, handle);
-		lib_ring_buffer_start_read_timer(buf, handle);
-	}
-	ref = &handle->chan._ref;
-	shm_get_object_data(handle, ref, shm_fd, wait_fd, memory_map_size);
 	return handle;
 
 error_backend_init:
@@ -510,8 +794,9 @@ error_table_alloc:
 	return NULL;
 }
 
-struct lttng_ust_shm_handle *channel_handle_create(int shm_fd, int wait_fd,
-					uint64_t memory_map_size)
+struct lttng_ust_shm_handle *channel_handle_create(void *data,
+					uint64_t memory_map_size,
+					int wakeup_fd)
 {
 	struct lttng_ust_shm_handle *handle;
 	struct shm_object *object;
@@ -525,8 +810,8 @@ struct lttng_ust_shm_handle *channel_handle_create(int shm_fd, int wait_fd,
 	if (!handle->table)
 		goto error_table_alloc;
 	/* Add channel object */
-	object = shm_object_table_append_shadow(handle->table,
-			shm_fd, wait_fd, memory_map_size);
+	object = shm_object_table_append_mem(handle->table, data,
+			memory_map_size, wakeup_fd);
 	if (!object)
 		goto error_table_object;
 	/* struct channel is at object 0, offset 0 (hardcoded) */
@@ -542,23 +827,30 @@ error_table_alloc:
 }
 
 int channel_handle_add_stream(struct lttng_ust_shm_handle *handle,
-		int shm_fd, int wait_fd, uint64_t memory_map_size)
+		int shm_fd, int wakeup_fd, uint32_t stream_nr,
+		uint64_t memory_map_size)
 {
 	struct shm_object *object;
 
 	/* Add stream object */
-	object = shm_object_table_append_shadow(handle->table,
-			shm_fd, wait_fd, memory_map_size);
+	object = shm_object_table_append_shm(handle->table,
+			shm_fd, wakeup_fd, stream_nr,
+			memory_map_size);
 	if (!object)
-		return -1;
+		return -EINVAL;
 	return 0;
 }
 
+unsigned int channel_handle_get_nr_streams(struct lttng_ust_shm_handle *handle)
+{
+	assert(handle->table);
+	return handle->table->allocated_len - 1;
+}
+
 static
-void channel_release(struct channel *chan, struct lttng_ust_shm_handle *handle,
-		int shadow)
+void channel_release(struct channel *chan, struct lttng_ust_shm_handle *handle)
 {
-	channel_free(chan, handle, shadow);
+	channel_free(chan, handle);
 }
 
 /**
@@ -569,120 +861,128 @@ void channel_release(struct channel *chan, struct lttng_ust_shm_handle *handle,
  * Call "destroy" callback, finalize channels, decrement the channel
  * reference count. Note that when readers have completed data
  * consumption of finalized channels, get_subbuf() will return -ENODATA.
- * They should release their handle at that point.  Returns the private
- * data pointer.
+ * They should release their handle at that point. 
  */
-void *channel_destroy(struct channel *chan, struct lttng_ust_shm_handle *handle,
-		int shadow)
+void channel_destroy(struct channel *chan, struct lttng_ust_shm_handle *handle,
+		int consumer)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
-	void *priv;
-	int cpu;
-
-	if (shadow) {
-		channel_release(chan, handle, shadow);
-		return NULL;
-	}
-
-	channel_unregister_notifiers(chan, handle);
-
-	if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
-		for_each_channel_cpu(cpu, chan) {
-			struct lib_ring_buffer *buf = shmp(handle, chan->backend.buf[cpu].shmp);
-
-			if (config->cb.buffer_finalize)
-				config->cb.buffer_finalize(buf,
-							   chan->backend.priv,
-							   cpu, handle);
-			if (buf->backend.allocated)
-				lib_ring_buffer_switch_slow(buf, SWITCH_FLUSH,
-						handle);
-			/*
-			 * Perform flush before writing to finalized.
-			 */
-			cmm_smp_wmb();
-			CMM_ACCESS_ONCE(buf->finalized) = 1;
-			//wake_up_interruptible(&buf->read_wait);
-		}
-	} else {
-		struct lib_ring_buffer *buf = shmp(handle, chan->backend.buf[0].shmp);
-
-		if (config->cb.buffer_finalize)
-			config->cb.buffer_finalize(buf, chan->backend.priv, -1, handle);
-		if (buf->backend.allocated)
-			lib_ring_buffer_switch_slow(buf, SWITCH_FLUSH,
-						handle);
+	if (consumer) {
 		/*
-		 * Perform flush before writing to finalized.
+		 * Note: the consumer takes care of finalizing and
+		 * switching the buffers.
 		 */
-		cmm_smp_wmb();
-		CMM_ACCESS_ONCE(buf->finalized) = 1;
-		//wake_up_interruptible(&buf->read_wait);
+		channel_unregister_notifiers(chan, handle);
+		/*
+		 * The consumer prints errors.
+		 */
+		channel_print_errors(chan, handle);
 	}
-	CMM_ACCESS_ONCE(chan->finalized) = 1;
-	//wake_up_interruptible(&chan->hp_wait);
-	//wake_up_interruptible(&chan->read_wait);
+
 	/*
 	 * sessiond/consumer are keeping a reference on the shm file
 	 * descriptor directly. No need to refcount.
 	 */
-	priv = chan->backend.priv;
-	channel_release(chan, handle, shadow);
-	return priv;
+	channel_release(chan, handle);
+	return;
 }
 
-struct lib_ring_buffer *channel_get_ring_buffer(
-					const struct lib_ring_buffer_config *config,
+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,
+					int *wakeup_fd,
 					uint64_t *memory_map_size)
 {
 	struct shm_ref *ref;
 
 	if (config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
-		ref = &chan->backend.buf[0].shmp._ref;
-		shm_get_object_data(handle, ref, shm_fd, wait_fd,
-			memory_map_size);
-		return shmp(handle, chan->backend.buf[0].shmp);
+		cpu = 0;
 	} else {
 		if (cpu >= num_possible_cpus())
 			return NULL;
-		ref = &chan->backend.buf[cpu].shmp._ref;
-		shm_get_object_data(handle, ref, shm_fd, wait_fd,
-			memory_map_size);
-		return shmp(handle, chan->backend.buf[cpu].shmp);
 	}
+	ref = &chan->backend.buf[cpu].shmp._ref;
+	*shm_fd = shm_get_shm_fd(handle, ref);
+	*wait_fd = shm_get_wait_fd(handle, ref);
+	*wakeup_fd = shm_get_wakeup_fd(handle, ref);
+	if (shm_get_shm_size(handle, ref, memory_map_size))
+		return NULL;
+	return shmp(handle, chan->backend.buf[cpu].shmp);
 }
 
-int lib_ring_buffer_open_read(struct lib_ring_buffer *buf,
-			      struct lttng_ust_shm_handle *handle,
-			      int shadow)
+int ring_buffer_channel_close_wait_fd(const struct lttng_ust_lib_ring_buffer_config *config,
+			struct channel *chan,
+			struct lttng_ust_shm_handle *handle)
 {
-	if (shadow) {
-		if (uatomic_cmpxchg(&buf->active_shadow_readers, 0, 1) != 0)
-			return -EBUSY;
-		cmm_smp_mb();
-		return 0;
+	struct shm_ref *ref;
+
+	ref = &handle->chan._ref;
+	return shm_close_wait_fd(handle, ref);
+}
+
+int ring_buffer_channel_close_wakeup_fd(const struct lttng_ust_lib_ring_buffer_config *config,
+			struct channel *chan,
+			struct lttng_ust_shm_handle *handle)
+{
+	struct shm_ref *ref;
+
+	ref = &handle->chan._ref;
+	return shm_close_wakeup_fd(handle, ref);
+}
+
+int ring_buffer_stream_close_wait_fd(const struct lttng_ust_lib_ring_buffer_config *config,
+			struct channel *chan,
+			struct lttng_ust_shm_handle *handle,
+			int cpu)
+{
+	struct shm_ref *ref;
+
+	if (config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
+		cpu = 0;
+	} else {
+		if (cpu >= num_possible_cpus())
+			return -EINVAL;
 	}
+	ref = &chan->backend.buf[cpu].shmp._ref;
+	return shm_close_wait_fd(handle, ref);
+}
+
+int ring_buffer_stream_close_wakeup_fd(const struct lttng_ust_lib_ring_buffer_config *config,
+			struct channel *chan,
+			struct lttng_ust_shm_handle *handle,
+			int cpu)
+{
+	struct shm_ref *ref;
+	int ret;
+
+	if (config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
+		cpu = 0;
+	} else {
+		if (cpu >= num_possible_cpus())
+			return -EINVAL;
+	}
+	ref = &chan->backend.buf[cpu].shmp._ref;
+	pthread_mutex_lock(&wakeup_fd_mutex);
+	ret = shm_close_wakeup_fd(handle, ref);
+	pthread_mutex_unlock(&wakeup_fd_mutex);
+	return ret;
+}
+
+int lib_ring_buffer_open_read(struct lttng_ust_lib_ring_buffer *buf,
+			      struct lttng_ust_shm_handle *handle)
+{
 	if (uatomic_cmpxchg(&buf->active_readers, 0, 1) != 0)
 		return -EBUSY;
 	cmm_smp_mb();
 	return 0;
 }
 
-void lib_ring_buffer_release_read(struct lib_ring_buffer *buf,
-				  struct lttng_ust_shm_handle *handle,
-				  int shadow)
+void lib_ring_buffer_release_read(struct lttng_ust_lib_ring_buffer *buf,
+				  struct lttng_ust_shm_handle *handle)
 {
 	struct channel *chan = shmp(handle, buf->backend.chan);
 
-	if (shadow) {
-		CHAN_WARN_ON(chan, uatomic_read(&buf->active_shadow_readers) != 1);
-		cmm_smp_mb();
-		uatomic_dec(&buf->active_shadow_readers);
-		return;
-	}
 	CHAN_WARN_ON(chan, uatomic_read(&buf->active_readers) != 1);
 	cmm_smp_mb();
 	uatomic_dec(&buf->active_readers);
@@ -698,12 +998,12 @@ void lib_ring_buffer_release_read(struct lib_ring_buffer *buf,
  * data to read at consumed position, or 0 if the get operation succeeds.
  */
 
-int lib_ring_buffer_snapshot(struct lib_ring_buffer *buf,
+int lib_ring_buffer_snapshot(struct lttng_ust_lib_ring_buffer *buf,
 			     unsigned long *consumed, unsigned long *produced,
 			     struct lttng_ust_shm_handle *handle)
 {
 	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long consumed_cur, write_offset;
 	int finalized;
 
@@ -748,20 +1048,19 @@ nodata:
 }
 
 /**
- * lib_ring_buffer_put_snapshot - move consumed counter forward
+ * lib_ring_buffer_move_consumer - move consumed counter forward
  * @buf: ring buffer
  * @consumed_new: new consumed count value
  */
-void lib_ring_buffer_move_consumer(struct lib_ring_buffer *buf,
+void lib_ring_buffer_move_consumer(struct lttng_ust_lib_ring_buffer *buf,
 				   unsigned long consumed_new,
 				   struct lttng_ust_shm_handle *handle)
 {
-	struct lib_ring_buffer_backend *bufb = &buf->backend;
+	struct lttng_ust_lib_ring_buffer_backend *bufb = &buf->backend;
 	struct channel *chan = shmp(handle, bufb->chan);
 	unsigned long consumed;
 
-	CHAN_WARN_ON(chan, uatomic_read(&buf->active_readers) != 1
-			&& uatomic_read(&buf->active_shadow_readers) != 1);
+	CHAN_WARN_ON(chan, uatomic_read(&buf->active_readers) != 1);
 
 	/*
 	 * Only push the consumed value forward.
@@ -782,12 +1081,12 @@ void lib_ring_buffer_move_consumer(struct lib_ring_buffer *buf,
  * Returns -ENODATA if buffer is finalized, -EAGAIN if there is currently no
  * data to read at consumed position, or 0 if the get operation succeeds.
  */
-int lib_ring_buffer_get_subbuf(struct lib_ring_buffer *buf,
+int lib_ring_buffer_get_subbuf(struct lttng_ust_lib_ring_buffer *buf,
 			       unsigned long consumed,
 			       struct lttng_ust_shm_handle *handle)
 {
 	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long consumed_cur, consumed_idx, commit_count, write_offset;
 	int ret;
 	int finalized;
@@ -876,16 +1175,15 @@ nodata:
  * lib_ring_buffer_put_subbuf - release exclusive subbuffer access
  * @buf: ring buffer
  */
-void lib_ring_buffer_put_subbuf(struct lib_ring_buffer *buf,
+void lib_ring_buffer_put_subbuf(struct lttng_ust_lib_ring_buffer *buf,
 				struct lttng_ust_shm_handle *handle)
 {
-	struct lib_ring_buffer_backend *bufb = &buf->backend;
+	struct lttng_ust_lib_ring_buffer_backend *bufb = &buf->backend;
 	struct channel *chan = shmp(handle, bufb->chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long read_sb_bindex, consumed_idx, consumed;
 
-	CHAN_WARN_ON(chan, uatomic_read(&buf->active_readers) != 1
-			&& uatomic_read(&buf->active_shadow_readers) != 1);
+	CHAN_WARN_ON(chan, uatomic_read(&buf->active_readers) != 1);
 
 	if (!buf->get_subbuf) {
 		/*
@@ -936,13 +1234,13 @@ void lib_ring_buffer_put_subbuf(struct lib_ring_buffer *buf,
  * position and the writer position. (inclusive)
  */
 static
-void lib_ring_buffer_print_subbuffer_errors(struct lib_ring_buffer *buf,
+void lib_ring_buffer_print_subbuffer_errors(struct lttng_ust_lib_ring_buffer *buf,
 					    struct channel *chan,
 					    unsigned long cons_offset,
 					    int cpu,
 					    struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long cons_idx, commit_count, commit_count_sb;
 
 	cons_idx = subbuf_index(cons_offset, chan);
@@ -950,7 +1248,7 @@ void lib_ring_buffer_print_subbuffer_errors(struct lib_ring_buffer *buf,
 	commit_count_sb = v_read(config, &shmp_index(handle, buf->commit_cold, cons_idx)->cc_sb);
 
 	if (subbuf_offset(commit_count, chan) != 0)
-		ERRMSG("ring buffer %s, cpu %d: "
+		DBG("ring buffer %s, cpu %d: "
 		       "commit count in subbuffer %lu,\n"
 		       "expecting multiples of %lu bytes\n"
 		       "  [ %lu bytes committed, %lu bytes reader-visible ]\n",
@@ -958,25 +1256,19 @@ void lib_ring_buffer_print_subbuffer_errors(struct lib_ring_buffer *buf,
 		       chan->backend.subbuf_size,
 		       commit_count, commit_count_sb);
 
-	ERRMSG("ring buffer: %s, cpu %d: %lu bytes committed\n",
+	DBG("ring buffer: %s, cpu %d: %lu bytes committed\n",
 	       chan->backend.name, cpu, commit_count);
 }
 
 static
-void lib_ring_buffer_print_buffer_errors(struct lib_ring_buffer *buf,
+void lib_ring_buffer_print_buffer_errors(struct lttng_ust_lib_ring_buffer *buf,
 					 struct channel *chan,
 					 void *priv, int cpu,
 					 struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	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
@@ -985,7 +1277,7 @@ void lib_ring_buffer_print_buffer_errors(struct lib_ring_buffer *buf,
 	write_offset = v_read(config, &buf->offset);
 	cons_offset = uatomic_read(&buf->consumed);
 	if (write_offset != cons_offset)
-		ERRMSG("ring buffer %s, cpu %d: "
+		DBG("ring buffer %s, cpu %d: "
 		       "non-consumed data\n"
 		       "  [ %lu bytes written, %lu bytes read ]\n",
 		       chan->backend.name, cpu, write_offset, cons_offset);
@@ -1001,29 +1293,36 @@ void lib_ring_buffer_print_buffer_errors(struct lib_ring_buffer *buf,
 
 static
 void lib_ring_buffer_print_errors(struct channel *chan,
-				  struct lib_ring_buffer *buf, int cpu,
-				  struct lttng_ust_shm_handle *handle)
+				struct lttng_ust_lib_ring_buffer *buf, int cpu,
+				struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
-	void *priv = chan->backend.priv;
-
-	ERRMSG("ring buffer %s, cpu %d: %lu records written, "
-			  "%lu records overrun\n",
-			  chan->backend.name, cpu,
-			  v_read(config, &buf->records_count),
-			  v_read(config, &buf->records_overrun));
-
-	if (v_read(config, &buf->records_lost_full)
-	    || v_read(config, &buf->records_lost_wrap)
-	    || v_read(config, &buf->records_lost_big))
-		ERRMSG("ring buffer %s, cpu %d: records were lost. Caused by:\n"
-		       "  [ %lu buffer full, %lu nest buffer wrap-around, "
-		       "%lu event too big ]\n",
-		       chan->backend.name, cpu,
-		       v_read(config, &buf->records_lost_full),
-		       v_read(config, &buf->records_lost_wrap),
-		       v_read(config, &buf->records_lost_big));
-
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
+	void *priv = channel_get_private(chan);
+
+	if (!strcmp(chan->backend.name, "relay-metadata-mmap")) {
+		DBG("ring buffer %s: %lu records written, "
+			"%lu records overrun\n",
+			chan->backend.name,
+			v_read(config, &buf->records_count),
+			v_read(config, &buf->records_overrun));
+	} else {
+		DBG("ring buffer %s, cpu %d: %lu records written, "
+			"%lu records overrun\n",
+			chan->backend.name, cpu,
+			v_read(config, &buf->records_count),
+			v_read(config, &buf->records_overrun));
+
+		if (v_read(config, &buf->records_lost_full)
+		    || v_read(config, &buf->records_lost_wrap)
+		    || v_read(config, &buf->records_lost_big))
+			DBG("ring buffer %s, cpu %d: records were lost. Caused by:\n"
+				"  [ %lu buffer full, %lu nest buffer wrap-around, "
+				"%lu event too big ]\n",
+				chan->backend.name, cpu,
+				v_read(config, &buf->records_lost_full),
+				v_read(config, &buf->records_lost_wrap),
+				v_read(config, &buf->records_lost_big));
+	}
 	lib_ring_buffer_print_buffer_errors(buf, chan, priv, cpu, handle);
 }
 
@@ -1033,13 +1332,13 @@ void lib_ring_buffer_print_errors(struct channel *chan,
  * Only executed when the buffer is finalized, in SWITCH_FLUSH.
  */
 static
-void lib_ring_buffer_switch_old_start(struct lib_ring_buffer *buf,
+void lib_ring_buffer_switch_old_start(struct lttng_ust_lib_ring_buffer *buf,
 				      struct channel *chan,
 				      struct switch_offsets *offsets,
-				      u64 tsc,
+				      uint64_t tsc,
 				      struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long oldidx = subbuf_index(offsets->old, chan);
 	unsigned long commit_count;
 
@@ -1071,13 +1370,13 @@ void lib_ring_buffer_switch_old_start(struct lib_ring_buffer *buf,
  * subbuffer.
  */
 static
-void lib_ring_buffer_switch_old_end(struct lib_ring_buffer *buf,
+void lib_ring_buffer_switch_old_end(struct lttng_ust_lib_ring_buffer *buf,
 				    struct channel *chan,
 				    struct switch_offsets *offsets,
-				    u64 tsc,
+				    uint64_t tsc,
 				    struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long oldidx = subbuf_index(offsets->old - 1, chan);
 	unsigned long commit_count, padding_size, data_size;
 
@@ -1108,13 +1407,13 @@ void lib_ring_buffer_switch_old_end(struct lib_ring_buffer *buf,
  * that this code is executed before the deliver of this sub-buffer.
  */
 static
-void lib_ring_buffer_switch_new_start(struct lib_ring_buffer *buf,
+void lib_ring_buffer_switch_new_start(struct lttng_ust_lib_ring_buffer *buf,
 				      struct channel *chan,
 				      struct switch_offsets *offsets,
-				      u64 tsc,
+				      uint64_t tsc,
 				      struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long beginidx = subbuf_index(offsets->begin, chan);
 	unsigned long commit_count;
 
@@ -1144,13 +1443,13 @@ void lib_ring_buffer_switch_new_start(struct lib_ring_buffer *buf,
  * have to do the deliver themselves.
  */
 static
-void lib_ring_buffer_switch_new_end(struct lib_ring_buffer *buf,
+void lib_ring_buffer_switch_new_end(struct lttng_ust_lib_ring_buffer *buf,
 				    struct channel *chan,
 				    struct switch_offsets *offsets,
-				    u64 tsc,
+				    uint64_t tsc,
 				    struct lttng_ust_shm_handle *handle)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long endidx = subbuf_index(offsets->end - 1, chan);
 	unsigned long commit_count, padding_size, data_size;
 
@@ -1180,12 +1479,12 @@ void lib_ring_buffer_switch_new_end(struct lib_ring_buffer *buf,
  */
 static
 int lib_ring_buffer_try_switch_slow(enum switch_mode mode,
-				    struct lib_ring_buffer *buf,
+				    struct lttng_ust_lib_ring_buffer *buf,
 				    struct channel *chan,
 				    struct switch_offsets *offsets,
-				    u64 *tsc)
+				    uint64_t *tsc)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	unsigned long off;
 
 	offsets->begin = v_read(config, &buf->offset);
@@ -1212,7 +1511,20 @@ int lib_ring_buffer_try_switch_slow(enum switch_mode mode,
 	 * quiescence guarantees for the fusion merge.
 	 */
 	if (mode == SWITCH_FLUSH || off > 0) {
-		if (unlikely(off == 0)) {
+		if (caa_unlikely(off == 0)) {
+			/*
+			 * A final flush that encounters an empty
+			 * sub-buffer cannot switch buffer if a
+			 * reader is located within this sub-buffer.
+			 * Anyway, the purpose of final flushing of a
+			 * sub-buffer at offset 0 is to handle the case
+			 * of entirely empty stream.
+			 */
+			if (caa_unlikely(subbuf_trunc(offsets->begin, chan)
+					 - subbuf_trunc((unsigned long)
+					     uatomic_read(&buf->consumed), chan)
+					>= chan->backend.buf_size))
+				return -1;
 			/*
 			 * The client does not save any header information.
 			 * Don't switch empty subbuffer on finalize, because it
@@ -1241,14 +1553,14 @@ int lib_ring_buffer_try_switch_slow(enum switch_mode mode,
  * operations, this function must be called from the CPU which owns the buffer
  * for a ACTIVE flush.
  */
-void lib_ring_buffer_switch_slow(struct lib_ring_buffer *buf, enum switch_mode mode,
+void lib_ring_buffer_switch_slow(struct lttng_ust_lib_ring_buffer *buf, enum switch_mode mode,
 				 struct lttng_ust_shm_handle *handle)
 {
 	struct channel *chan = shmp(handle, buf->backend.chan);
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	struct switch_offsets offsets;
 	unsigned long oldidx;
-	u64 tsc;
+	uint64_t tsc;
 
 	offsets.size = 0;
 
@@ -1300,12 +1612,12 @@ void lib_ring_buffer_switch_slow(struct lib_ring_buffer *buf, enum switch_mode m
  * -EIO if data cannot be written into the buffer for any other reason.
  */
 static
-int lib_ring_buffer_try_reserve_slow(struct lib_ring_buffer *buf,
+int lib_ring_buffer_try_reserve_slow(struct lttng_ust_lib_ring_buffer *buf,
 				     struct channel *chan,
 				     struct switch_offsets *offsets,
-				     struct lib_ring_buffer_ctx *ctx)
+				     struct lttng_ust_lib_ring_buffer_ctx *ctx)
 {
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
 	struct lttng_ust_shm_handle *handle = ctx->handle;
 	unsigned long reserve_commit_diff;
 
@@ -1323,7 +1635,7 @@ int lib_ring_buffer_try_reserve_slow(struct lib_ring_buffer *buf,
 	if (last_tsc_overflow(config, buf, ctx->tsc))
 		ctx->rflags |= RING_BUFFER_RFLAG_FULL_TSC;
 
-	if (unlikely(subbuf_offset(offsets->begin, ctx->chan) == 0)) {
+	if (caa_unlikely(subbuf_offset(offsets->begin, ctx->chan) == 0)) {
 		offsets->switch_new_start = 1;		/* For offsets->begin */
 	} else {
 		offsets->size = config->cb.record_header_size(config, chan,
@@ -1334,19 +1646,19 @@ int lib_ring_buffer_try_reserve_slow(struct lib_ring_buffer *buf,
 			lib_ring_buffer_align(offsets->begin + offsets->size,
 					      ctx->largest_align)
 			+ ctx->data_size;
-		if (unlikely(subbuf_offset(offsets->begin, chan) +
+		if (caa_unlikely(subbuf_offset(offsets->begin, chan) +
 			     offsets->size > chan->backend.subbuf_size)) {
 			offsets->switch_old_end = 1;	/* For offsets->old */
 			offsets->switch_new_start = 1;	/* For offsets->begin */
 		}
 	}
-	if (unlikely(offsets->switch_new_start)) {
+	if (caa_unlikely(offsets->switch_new_start)) {
 		unsigned long sb_index;
 
 		/*
 		 * We are typically not filling the previous buffer completely.
 		 */
-		if (likely(offsets->switch_old_end))
+		if (caa_likely(offsets->switch_old_end))
 			offsets->begin = subbuf_align(offsets->begin, chan);
 		offsets->begin = offsets->begin
 				 + config->cb.subbuffer_header_size();
@@ -1358,18 +1670,26 @@ int lib_ring_buffer_try_reserve_slow(struct lib_ring_buffer *buf,
 		  - ((unsigned long) v_read(config,
 					    &shmp_index(handle, buf->commit_cold, sb_index)->cc_sb)
 		     & chan->commit_count_mask);
-		if (likely(reserve_commit_diff == 0)) {
+		if (caa_likely(reserve_commit_diff == 0)) {
 			/* Next subbuffer not being written to. */
-			if (unlikely(config->mode != RING_BUFFER_OVERWRITE &&
+			if (caa_unlikely(config->mode != RING_BUFFER_OVERWRITE &&
 				subbuf_trunc(offsets->begin, chan)
 				 - 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 {
 				/*
@@ -1380,13 +1700,21 @@ int lib_ring_buffer_try_reserve_slow(struct 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 =
@@ -1398,13 +1726,22 @@ int lib_ring_buffer_try_reserve_slow(struct lib_ring_buffer *buf,
 			lib_ring_buffer_align(offsets->begin + offsets->size,
 					      ctx->largest_align)
 			+ ctx->data_size;
-		if (unlikely(subbuf_offset(offsets->begin, chan)
+		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 {
 			/*
@@ -1420,7 +1757,7 @@ int lib_ring_buffer_try_reserve_slow(struct lib_ring_buffer *buf,
 	}
 	offsets->end = offsets->begin + offsets->size;
 
-	if (unlikely(subbuf_offset(offsets->end, chan) == 0)) {
+	if (caa_unlikely(subbuf_offset(offsets->end, chan) == 0)) {
 		/*
 		 * The offset_end will fall at the very beginning of the next
 		 * subbuffer.
@@ -1438,12 +1775,12 @@ int lib_ring_buffer_try_reserve_slow(struct lib_ring_buffer *buf,
  * -EIO for other errors, else returns 0.
  * It will take care of sub-buffer switching.
  */
-int lib_ring_buffer_reserve_slow(struct lib_ring_buffer_ctx *ctx)
+int lib_ring_buffer_reserve_slow(struct lttng_ust_lib_ring_buffer_ctx *ctx)
 {
 	struct channel *chan = ctx->chan;
 	struct lttng_ust_shm_handle *handle = ctx->handle;
-	const struct lib_ring_buffer_config *config = &chan->backend.config;
-	struct lib_ring_buffer *buf;
+	const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
+	struct lttng_ust_lib_ring_buffer *buf;
 	struct switch_offsets offsets;
 	int ret;
 
@@ -1458,9 +1795,9 @@ int lib_ring_buffer_reserve_slow(struct lib_ring_buffer_ctx *ctx)
 	do {
 		ret = lib_ring_buffer_try_reserve_slow(buf, chan, &offsets,
 						       ctx);
-		if (unlikely(ret))
+		if (caa_unlikely(ret))
 			return ret;
-	} while (unlikely(v_cmpxchg(config, &buf->offset, offsets.old,
+	} while (caa_unlikely(v_cmpxchg(config, &buf->offset, offsets.old,
 				    offsets.end)
 			  != offsets.old));
 
@@ -1487,7 +1824,7 @@ int lib_ring_buffer_reserve_slow(struct lib_ring_buffer_ctx *ctx)
 	/*
 	 * Switch old subbuffer if needed.
 	 */
-	if (unlikely(offsets.switch_old_end)) {
+	if (caa_unlikely(offsets.switch_old_end)) {
 		lib_ring_buffer_clear_noref(config, &buf->backend,
 					    subbuf_index(offsets.old - 1, chan),
 					    handle);
@@ -1497,10 +1834,10 @@ int lib_ring_buffer_reserve_slow(struct lib_ring_buffer_ctx *ctx)
 	/*
 	 * Populate new subbuffer.
 	 */
-	if (unlikely(offsets.switch_new_start))
+	if (caa_unlikely(offsets.switch_new_start))
 		lib_ring_buffer_switch_new_start(buf, chan, &offsets, ctx->tsc, handle);
 
-	if (unlikely(offsets.switch_new_end))
+	if (caa_unlikely(offsets.switch_new_end))
 		lib_ring_buffer_switch_new_end(buf, chan, &offsets, ctx->tsc, handle);
 
 	ctx->slot_size = offsets.size;
@@ -1508,3 +1845,28 @@ int lib_ring_buffer_reserve_slow(struct lib_ring_buffer_ctx *ctx)
 	ctx->buf_offset = offsets.begin + offsets.pre_header_padding;
 	return 0;
 }
+
+/*
+ * Force a read (imply TLS fixup for dlopen) of TLS variables.
+ */
+void lttng_fixup_ringbuffer_tls(void)
+{
+	asm volatile ("" : : "m" (URCU_TLS(lib_ring_buffer_nesting)));
+}
+
+void lib_ringbuffer_signal_init(void)
+{
+	sigset_t mask;
+	int ret;
+
+	/*
+	 * Block signal for entire process, so only our thread processes
+	 * it.
+	 */
+	rb_setmask(&mask);
+	ret = pthread_sigmask(SIG_BLOCK, &mask, NULL);
+	if (ret) {
+		errno = ret;
+		PERROR("pthread_sigmask");
+	}
+}