/*
* 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:
* - 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 "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)
+
/*
* Use POSIX SHM: shm_open(3) and shm_unlink(3).
* close(2) to close the fd returned by shm_open.
/*
* Only flush buffers periodically if readers are active.
*/
- if (uatomic_read(&buf->active_readers) || uatomic_read(&buf->active_shadow_readers))
+ if (uatomic_read(&buf->active_readers))
lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE, handle);
//TODO timers
CHAN_WARN_ON(chan, !buf->backend.allocated);
- if (uatomic_read(&buf->active_readers) || uatomic_read(&buf->active_shadow_readers))
+ if (uatomic_read(&buf->active_readers))
&& lib_ring_buffer_poll_deliver(config, buf, chan)) {
//TODO
//wake_up_interruptible(&buf->read_wait);
//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);
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, 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))
/* Calculate the shm allocation layout */
shmsize = sizeof(struct channel);
shmsize += offset_align(shmsize, __alignof__(struct lttng_ust_lib_ring_buffer_shmp));
- if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
- shmsize += sizeof(struct lttng_ust_lib_ring_buffer_shmp) * num_possible_cpus();
- else
- shmsize += sizeof(struct lttng_ust_lib_ring_buffer_shmp);
+ shmsize += sizeof(struct lttng_ust_lib_ring_buffer_shmp) * nr_streams;
chansize = shmsize;
- shmsize += offset_align(shmsize, priv_data_align);
+ if (priv_data_align)
+ shmsize += offset_align(shmsize, priv_data_align);
shmsize += priv_data_size;
- shmobj = shm_object_table_append(handle->table, shmsize);
+ /* 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) */
chan = shmp(handle, handle->chan);
if (!chan)
goto error_append;
+ chan->nr_streams = nr_streams;
/* space for private data */
if (priv_data_size) {
memcpy(*priv_data, priv_data_init, priv_data_size);
} else {
chan->priv_data_offset = -1;
- *priv_data = NULL;
+ if (priv_data)
+ *priv_data = NULL;
}
ret = channel_backend_init(&chan->backend, name, config,
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:
return NULL;
}
-struct lttng_ust_shm_handle *channel_handle_create(int shm_fd, int wait_fd,
+struct lttng_ust_shm_handle *channel_handle_create(void *data,
uint64_t memory_map_size)
{
struct lttng_ust_shm_handle *handle;
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);
if (!object)
goto error_table_object;
/* struct channel is at object 0, offset 0 (hardcoded) */
}
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);
}
/**
* They should release their handle at that point.
*/
void channel_destroy(struct channel *chan, struct lttng_ust_shm_handle *handle,
- int shadow)
+ int consumer)
{
- if (shadow) {
- channel_release(chan, handle, shadow);
- return;
+ if (consumer) {
+ /*
+ * Note: the consumer takes care of finalizing and
+ * switching the buffers.
+ */
+ channel_unregister_notifiers(chan, handle);
}
- channel_unregister_notifiers(chan, handle);
-
- /*
- * Note: the consumer takes care of finalizing and switching the
- * buffers.
- */
-
/*
* sessiond/consumer are keeping a reference on the shm file
* descriptor directly. No need to refcount.
*/
- channel_release(chan, handle, shadow);
+ channel_release(chan, handle);
return;
}
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,
+ 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 lttng_ust_lib_ring_buffer *buf,
- struct lttng_ust_shm_handle *handle,
- int shadow)
+int ring_buffer_close_wait_fd(const struct lttng_ust_lib_ring_buffer_config *config,
+ struct channel *chan,
+ struct lttng_ust_shm_handle *handle,
+ int cpu)
{
- if (shadow) {
- if (uatomic_cmpxchg(&buf->active_shadow_readers, 0, 1) != 0)
- return -EBUSY;
- cmm_smp_mb();
- return 0;
+ 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_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;
+
+ 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_wakeup_fd(handle, ref);
+}
+
+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();
}
void lib_ring_buffer_release_read(struct lttng_ust_lib_ring_buffer *buf,
- struct lttng_ust_shm_handle *handle,
- int shadow)
+ 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);
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.
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) {
/*
const struct lttng_ust_lib_ring_buffer_config *config = &chan->backend.config;
void *priv = channel_get_private(chan);
- 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));
-
+ 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);
}
- 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 {
/*
*/
}
} 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 =
+ 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 {
/*
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" (lib_ring_buffer_nesting));
+}