Remove deprecated files
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Fri, 26 Aug 2011 10:29:29 +0000 (06:29 -0400)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Fri, 26 Aug 2011 10:29:29 +0000 (06:29 -0400)
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
libust/buffers.c [deleted file]
libust/buffers.h [deleted file]

diff --git a/libust/buffers.c b/libust/buffers.c
deleted file mode 100644 (file)
index 5e67d54..0000000
+++ /dev/null
@@ -1,1140 +0,0 @@
-/*
- * buffers.c
- * LTTng userspace tracer buffering system
- *
- * Copyright (C) 2009 - Pierre-Marc Fournier (pierre-marc dot fournier at polymtl dot ca)
- * Copyright (C) 2008-2011 - Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca)
- *
- * 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; 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
- */
-
-/*
- * Note: this code does not support the ref/noref flag and reader-owned
- * subbuffer scheme. Therefore, flight recorder mode uses a mechanism
- * where the reader can read corrupted data (and detect this), thus
- * returning -EIO.
- */
-
-#include <unistd.h>
-#include <sys/mman.h>
-#include <sys/ipc.h>
-#include <sys/shm.h>
-#include <fcntl.h>
-#include <stdlib.h>
-
-#include <ust/clock.h>
-
-#include "buffers.h"
-#include "channels.h"
-#include "tracer.h"
-#include "tracercore.h"
-#include <ust/usterr-signal-safe.h>
-
-struct ltt_reserve_switch_offsets {
-       long begin, end, old;
-       long begin_switch, end_switch_current, end_switch_old;
-       size_t before_hdr_pad, size;
-};
-
-
-static DEFINE_MUTEX(ust_buffers_channels_mutex);
-static CDS_LIST_HEAD(ust_buffers_channels);
-
-static void ltt_force_switch(struct ust_buffer *buf,
-               enum force_switch_mode mode);
-
-/**
- * _ust_buffers_strncpy_fixup - Fix an incomplete string in a ltt_relay buffer.
- * @buf : buffer
- * @offset : offset within the buffer
- * @len : length to write
- * @copied: string actually copied
- * @terminated: does string end with \0
- *
- * Fills string with "X" if incomplete.
- */
-void _ust_buffers_strncpy_fixup(struct ust_buffer *buf, size_t offset,
-                               size_t len, size_t copied, int terminated)
-{
-       size_t buf_offset, cpy;
-
-       if (copied == len) {
-               /*
-                * Deal with non-terminated string.
-                */
-               assert(!terminated);
-               offset += copied - 1;
-               buf_offset = BUFFER_OFFSET(offset, buf->chan);
-               /*
-                * Underlying layer should never ask for writes across
-                * subbuffers.
-                */
-               assert(buf_offset
-                      < buf->chan->subbuf_size*buf->chan->subbuf_cnt);
-               ust_buffers_do_memset(buf->buf_data + buf_offset, '\0', 1);
-               return;
-       }
-
-       /*
-        * Deal with incomplete string.
-        * Overwrite string's \0 with X too.
-        */
-       cpy = copied - 1;
-       assert(terminated);
-       len -= cpy;
-       offset += cpy;
-       buf_offset = BUFFER_OFFSET(offset, buf->chan);
-
-       /*
-        * Underlying layer should never ask for writes across subbuffers.
-        */
-       assert(buf_offset
-              < buf->chan->subbuf_size*buf->chan->subbuf_cnt);
-
-       ust_buffers_do_memset(buf->buf_data + buf_offset,
-                             'X', len);
-
-       /*
-        * Overwrite last 'X' with '\0'.
-        */
-       offset += len - 1;
-       buf_offset = BUFFER_OFFSET(offset, buf->chan);
-       /*
-        * Underlying layer should never ask for writes across subbuffers.
-        */
-       assert(buf_offset
-              < buf->chan->subbuf_size*buf->chan->subbuf_cnt);
-       ust_buffers_do_memset(buf->buf_data + buf_offset, '\0', 1);
-}
-
-static void ltt_buffer_begin(struct ust_buffer *buf,
-                            u64 tsc, unsigned int subbuf_idx)
-{
-       struct ust_channel *channel = buf->chan;
-       struct ltt_subbuffer_header *header =
-               (struct ltt_subbuffer_header *)
-                       ust_buffers_offset_address(buf,
-                               subbuf_idx * buf->chan->subbuf_size);
-
-       header->cycle_count_begin = tsc;
-       header->data_size = 0xFFFFFFFF; /* for recognizing crashed buffers */
-       header->sb_size = 0xFFFFFFFF; /* for recognizing crashed buffers */
-       /*
-        * No memory barrier needed to order data_data/sb_size vs commit count
-        * update, because commit count update contains a compiler barrier that
-        * ensures the order of the writes are OK from a program POV. It only
-        * matters for crash dump recovery which is not executed concurrently,
-        * so memory write order does not matter.
-        */
-       ltt_write_trace_header(channel->trace, header);
-}
-
-static int map_buf_data(struct ust_buffer *buf, size_t *size)
-{
-       void *ptr;
-       int result;
-
-       *size = PAGE_ALIGN(*size);
-
-       result = buf->shmid = shmget(getpid(), *size, IPC_CREAT | IPC_EXCL | 0700);
-       if (result < 0 && errno == EINVAL) {
-               ERR("shmget() returned EINVAL; maybe /proc/sys/kernel/shmmax should be increased.");
-               return -1;
-       } else if (result < 0) {
-               PERROR("shmget");
-               return -1;
-       }
-
-       ptr = shmat(buf->shmid, NULL, 0);
-       if (ptr == (void *) -1) {
-               perror("shmat");
-               goto destroy_shmem;
-       }
-
-       /* Already mark the shared memory for destruction. This will occur only
-         * when all users have detached.
-        */
-       result = shmctl(buf->shmid, IPC_RMID, NULL);
-       if(result == -1) {
-               perror("shmctl");
-               return -1;
-       }
-
-       buf->buf_data = ptr;
-       buf->buf_size = *size;
-
-       return 0;
-
-destroy_shmem:
-       result = shmctl(buf->shmid, IPC_RMID, NULL);
-       if(result == -1) {
-               perror("shmctl");
-       }
-
-       return -1;
-}
-
-static int open_buf(struct ust_channel *chan, int cpu)
-{
-       int result, fds[2];
-       unsigned int j;
-       struct ust_trace *trace = chan->trace;
-       struct ust_buffer *buf = chan->buf[cpu];
-       unsigned int n_subbufs = chan->subbuf_cnt;
-
-
-       result = map_buf_data(buf, &chan->alloc_size);
-       if (result < 0)
-               return -1;
-
-       buf->commit_count =
-               zmalloc(sizeof(*buf->commit_count) * n_subbufs);
-       if (!buf->commit_count)
-               goto unmap_buf;
-
-       result = pipe(fds);
-       if (result < 0) {
-               PERROR("pipe");
-               goto free_commit_count;
-       }
-       buf->data_ready_fd_read = fds[0];
-       buf->data_ready_fd_write = fds[1];
-
-       buf->cpu = cpu;
-       buf->chan = chan;
-
-       uatomic_set(&buf->offset, ltt_subbuffer_header_size());
-       uatomic_set(&buf->consumed, 0);
-       uatomic_set(&buf->active_readers, 0);
-       for (j = 0; j < n_subbufs; j++) {
-               uatomic_set(&buf->commit_count[j].cc, 0);
-               uatomic_set(&buf->commit_count[j].cc_sb, 0);
-       }
-
-       ltt_buffer_begin(buf, trace->start_tsc, 0);
-
-       uatomic_add(&buf->commit_count[0].cc, ltt_subbuffer_header_size());
-
-       uatomic_set(&buf->events_lost, 0);
-       uatomic_set(&buf->corrupted_subbuffers, 0);
-
-       memset(buf->commit_seq, 0, sizeof(buf->commit_seq[0]) * n_subbufs);
-
-       return 0;
-
-free_commit_count:
-       free(buf->commit_count);
-
-unmap_buf:
-       if (shmdt(buf->buf_data) < 0) {
-               PERROR("shmdt failed");
-       }
-
-       return -1;
-}
-
-static void close_buf(struct ust_buffer *buf)
-{
-       int result;
-
-       result = shmdt(buf->buf_data);
-       if (result < 0) {
-               PERROR("shmdt");
-       }
-
-       result = close(buf->data_ready_fd_read);
-       if (result < 0) {
-               PERROR("close");
-       }
-
-       result = close(buf->data_ready_fd_write);
-       if (result < 0 && errno != EBADF) {
-               PERROR("close");
-       }
-}
-
-
-static int open_channel(struct ust_channel *chan, size_t subbuf_size,
-                       size_t subbuf_cnt)
-{
-       int i;
-       int result;
-
-       if(subbuf_size == 0 || subbuf_cnt == 0)
-               return -1;
-
-       /* Check that the subbuffer size is larger than a page. */
-       WARN_ON_ONCE(subbuf_size < PAGE_SIZE);
-
-       /*
-        * Make sure the number of subbuffers and subbuffer size are power of 2.
-        */
-       WARN_ON_ONCE(hweight32(subbuf_size) != 1);
-       WARN_ON(hweight32(subbuf_cnt) != 1);
-
-       chan->version = UST_CHANNEL_VERSION;
-       chan->subbuf_cnt = subbuf_cnt;
-       chan->subbuf_size = subbuf_size;
-       chan->subbuf_size_order = get_count_order(subbuf_size);
-       chan->alloc_size = subbuf_size * subbuf_cnt;
-
-       pthread_mutex_lock(&ust_buffers_channels_mutex);
-       for (i=0; i < chan->n_cpus; i++) {
-               result = open_buf(chan, i);
-               if (result == -1)
-                       goto error;
-       }
-       cds_list_add(&chan->list, &ust_buffers_channels);
-       pthread_mutex_unlock(&ust_buffers_channels_mutex);
-
-       return 0;
-
-       /* Error handling */
-error:
-       for(i--; i >= 0; i--)
-               close_buf(chan->buf[i]);
-
-       pthread_mutex_unlock(&ust_buffers_channels_mutex);
-       return -1;
-}
-
-static void close_channel(struct ust_channel *chan)
-{
-       int i;
-       if(!chan)
-               return;
-
-       pthread_mutex_lock(&ust_buffers_channels_mutex);
-       /*
-        * checking for chan->buf[i] being NULL or not is useless in
-        * practice because we allocate buffers for all possible cpus.
-        * However, should we decide to change this and only allocate
-        * for online cpus, this check becomes useful.
-        */
-       for (i=0; i<chan->n_cpus; i++) {
-               if (chan->buf[i])
-                       close_buf(chan->buf[i]);
-       }
-
-       cds_list_del(&chan->list);
-
-       pthread_mutex_unlock(&ust_buffers_channels_mutex);
-}
-
-/*
- * offset is assumed to never be 0 here : never deliver a completely empty
- * subbuffer. The lost size is between 0 and subbuf_size-1.
- */
-static notrace void ltt_buffer_end(struct ust_buffer *buf,
-               u64 tsc, unsigned int offset, unsigned int subbuf_idx)
-{
-       struct ltt_subbuffer_header *header =
-               (struct ltt_subbuffer_header *)
-                       ust_buffers_offset_address(buf,
-                               subbuf_idx * buf->chan->subbuf_size);
-       u32 data_size = SUBBUF_OFFSET(offset - 1, buf->chan) + 1;
-
-       header->sb_size = PAGE_ALIGN(data_size);
-       header->cycle_count_end = tsc;
-       header->events_lost = uatomic_read(&buf->events_lost);
-       header->subbuf_corrupt = uatomic_read(&buf->corrupted_subbuffers);
-       if(unlikely(header->events_lost > 0)) {
-               DBG("Some events (%d) were lost in %s_%d", header->events_lost, buf->chan->channel_name, buf->cpu);
-       }
-       /*
-        * Makes sure data_size write happens after write of the rest of the
-        * buffer end data, because data_size is used to identify a completely
-        * written subbuffer in a crash dump.
-        */
-       cmm_barrier();
-       header->data_size = data_size;
-}
-
-/*
- * This function should not be called from NMI interrupt context
- */
-static notrace void ltt_buf_unfull(struct ust_buffer *buf,
-               unsigned int subbuf_idx,
-               long offset)
-{
-}
-
-/*
- * Promote compiler cmm_barrier to a smp_mb().
- * For the specific LTTng case, this IPI call should be removed if the
- * architecture does not reorder writes.  This should eventually be provided by
- * a separate architecture-specific infrastructure.
- */
-//ust// static void remote_mb(void *info)
-//ust// {
-//ust//        smp_mb();
-//ust// }
-
-int ust_buffers_get_subbuf(struct ust_buffer *buf, long *consumed)
-{
-       struct ust_channel *channel = buf->chan;
-       long consumed_old, consumed_idx, commit_count, write_offset;
-//ust//        int retval;
-
-       consumed_old = uatomic_read(&buf->consumed);
-       consumed_idx = SUBBUF_INDEX(consumed_old, buf->chan);
-       commit_count = uatomic_read(&buf->commit_count[consumed_idx].cc_sb);
-       /*
-        * Make sure we read the commit count before reading the buffer
-        * data and the write offset. Correct consumed offset ordering
-        * wrt commit count is insured by the use of cmpxchg to update
-        * the consumed offset.
-        */
-
-       /*
-        * Local rmb to match the remote wmb to read the commit count before the
-        * buffer data and the write offset.
-        */
-       cmm_smp_rmb();
-
-       write_offset = uatomic_read(&buf->offset);
-       /*
-        * Check that the subbuffer we are trying to consume has been
-        * already fully committed.
-        */
-       if (((commit_count - buf->chan->subbuf_size)
-            & channel->commit_count_mask)
-           - (BUFFER_TRUNC(consumed_old, buf->chan)
-              >> channel->n_subbufs_order)
-           != 0) {
-               return -EAGAIN;
-       }
-       /*
-        * Check that we are not about to read the same subbuffer in
-        * which the writer head is.
-        */
-       if ((SUBBUF_TRUNC(write_offset, buf->chan)
-          - SUBBUF_TRUNC(consumed_old, buf->chan))
-          == 0) {
-               return -EAGAIN;
-       }
-       *consumed = consumed_old;
-
-       return 0;
-}
-
-int ust_buffers_put_subbuf(struct ust_buffer *buf, unsigned long uconsumed_old)
-{
-       long consumed_new, consumed_old;
-
-       consumed_old = uatomic_read(&buf->consumed);
-       consumed_old = consumed_old & (~0xFFFFFFFFL);
-       consumed_old = consumed_old | uconsumed_old;
-       consumed_new = SUBBUF_ALIGN(consumed_old, buf->chan);
-
-       if (uatomic_cmpxchg(&buf->consumed, consumed_old,
-                               consumed_new)
-           != consumed_old) {
-               /* We have been pushed by the writer : the last
-                * buffer read _is_ corrupted! It can also
-                * happen if this is a buffer we never got. */
-               return -EIO;
-       } else {
-               /* tell the client that buffer is now unfull */
-               int index;
-               long data;
-               index = SUBBUF_INDEX(consumed_old, buf->chan);
-               data = BUFFER_OFFSET(consumed_old, buf->chan);
-               ltt_buf_unfull(buf, index, data);
-       }
-       return 0;
-}
-
-static int map_buf_structs(struct ust_channel *chan)
-{
-       void *ptr;
-       int result;
-       size_t size;
-       int i;
-
-       size = PAGE_ALIGN(1);
-
-       for(i=0; i<chan->n_cpus; i++) {
-
-               result = chan->buf_struct_shmids[i] = shmget(getpid(), size, IPC_CREAT | IPC_EXCL | 0700);
-               if(result == -1) {
-                       PERROR("shmget");
-                       goto destroy_previous;
-               }
-
-               ptr = shmat(chan->buf_struct_shmids[i], NULL, 0);
-               if(ptr == (void *) -1) {
-                       perror("shmat");
-                       goto destroy_shm;
-               }
-
-               /* Already mark the shared memory for destruction. This will occur only
-                * when all users have detached.
-                */
-               result = shmctl(chan->buf_struct_shmids[i], IPC_RMID, NULL);
-               if(result == -1) {
-                       perror("shmctl");
-                       goto destroy_previous;
-               }
-
-               chan->buf[i] = ptr;
-       }
-
-       return 0;
-
-       /* Jumping inside this loop occurs from within the other loop above with i as
-        * counter, so it unallocates the structures for the cpu = current_i down to
-        * zero. */
-       for(; i>=0; i--) {
-               destroy_shm:
-               result = shmctl(chan->buf_struct_shmids[i], IPC_RMID, NULL);
-               if(result == -1) {
-                       perror("shmctl");
-               }
-
-               destroy_previous:
-               continue;
-       }
-
-       return -1;
-}
-
-static int unmap_buf_structs(struct ust_channel *chan)
-{
-       int i;
-
-       for (i=0; i < chan->n_cpus; i++) {
-               if (shmdt(chan->buf[i]) < 0) {
-                       PERROR("shmdt");
-               }
-       }
-       return 0;
-}
-
-/*
- * Create channel.
- */
-static int create_channel(const char *trace_name, struct ust_trace *trace,
-       const char *channel_name, struct ust_channel *chan,
-       unsigned int subbuf_size, unsigned int n_subbufs, int overwrite)
-{
-       int i, result;
-
-       chan->trace = trace;
-       chan->overwrite = overwrite;
-       chan->n_subbufs_order = get_count_order(n_subbufs);
-       chan->commit_count_mask = (~0UL >> chan->n_subbufs_order);
-       chan->n_cpus = get_n_cpus();
-
-       /* These mappings should ideall be per-cpu, if somebody can do that
-        * from userspace, that would be cool!
-        */
-       chan->buf = (void *) zmalloc(chan->n_cpus * sizeof(void *));
-       if(chan->buf == NULL) {
-               goto error;
-       }
-       chan->buf_struct_shmids = (int *) zmalloc(chan->n_cpus * sizeof(int));
-       if(chan->buf_struct_shmids == NULL)
-               goto free_buf;
-
-       result = map_buf_structs(chan);
-       if(result != 0) {
-               goto free_buf_struct_shmids;
-       }
-
-       result = open_channel(chan, subbuf_size, n_subbufs);
-       if (result != 0) {
-               ERR("Cannot open channel for trace %s", trace_name);
-               goto unmap_buf_structs;
-       }
-
-       return 0;
-
-unmap_buf_structs:
-       for (i=0; i < chan->n_cpus; i++) {
-               if (shmdt(chan->buf[i]) < 0) {
-                       PERROR("shmdt bufstruct");
-               }
-       }
-
-free_buf_struct_shmids:
-       free(chan->buf_struct_shmids);
-
-free_buf:
-       free(chan->buf);
-
-error:
-       return -1;
-}
-
-
-static void remove_channel(struct ust_channel *chan)
-{
-       close_channel(chan);
-
-       unmap_buf_structs(chan);
-
-       free(chan->buf_struct_shmids);
-
-       free(chan->buf);
-}
-
-static void ltt_relay_async_wakeup_chan(struct ust_channel *ltt_channel)
-{
-}
-
-static void ltt_relay_finish_buffer(struct ust_channel *channel, unsigned int cpu)
-{
-       if (channel->buf[cpu]) {
-               struct ust_buffer *buf = channel->buf[cpu];
-               ltt_force_switch(buf, FORCE_FLUSH);
-
-               /* closing the pipe tells the consumer the buffer is finished */
-               close(buf->data_ready_fd_write);
-       }
-}
-
-
-static void finish_channel(struct ust_channel *channel)
-{
-       unsigned int i;
-
-       for (i=0; i<channel->n_cpus; i++) {
-               ltt_relay_finish_buffer(channel, i);
-       }
-}
-
-
-/*
- * ltt_reserve_switch_old_subbuf: switch old subbuffer
- *
- * Concurrency safe because we are the last and only thread to alter this
- * sub-buffer. As long as it is not delivered and read, no other thread can
- * alter the offset, alter the reserve_count or call the
- * client_buffer_end_callback on this sub-buffer.
- *
- * The only remaining threads could be the ones with pending commits. They will
- * have to do the deliver themselves.  Not concurrency safe in overwrite mode.
- * We detect corrupted subbuffers with commit and reserve counts. We keep a
- * corrupted sub-buffers count and push the readers across these sub-buffers.
- *
- * Not concurrency safe if a writer is stalled in a subbuffer and another writer
- * switches in, finding out it's corrupted.  The result will be than the old
- * (uncommited) subbuffer will be declared corrupted, and that the new subbuffer
- * will be declared corrupted too because of the commit count adjustment.
- *
- * Note : offset_old should never be 0 here.
- */
-static void ltt_reserve_switch_old_subbuf(
-               struct ust_channel *chan, struct ust_buffer *buf,
-               struct ltt_reserve_switch_offsets *offsets, u64 *tsc)
-{
-       long oldidx = SUBBUF_INDEX(offsets->old - 1, chan);
-       long commit_count, padding_size;
-
-       padding_size = chan->subbuf_size
-                       - (SUBBUF_OFFSET(offsets->old - 1, chan) + 1);
-       ltt_buffer_end(buf, *tsc, offsets->old, oldidx);
-
-       /*
-        * Must write slot data before incrementing commit count.
-        * This compiler barrier is upgraded into a cmm_smp_wmb() by the IPI
-        * sent by get_subbuf() when it does its cmm_smp_rmb().
-        */
-       cmm_smp_wmb();
-       uatomic_add(&buf->commit_count[oldidx].cc, padding_size);
-       commit_count = uatomic_read(&buf->commit_count[oldidx].cc);
-       ltt_check_deliver(chan, buf, offsets->old - 1, commit_count, oldidx);
-       ltt_write_commit_counter(chan, buf, oldidx,
-               offsets->old, commit_count, padding_size);
-}
-
-/*
- * ltt_reserve_switch_new_subbuf: Populate new subbuffer.
- *
- * This code can be executed unordered : writers may already have written to the
- * sub-buffer before this code gets executed, caution.  The commit makes sure
- * that this code is executed before the deliver of this sub-buffer.
- */
-static void ltt_reserve_switch_new_subbuf(
-               struct ust_channel *chan, struct ust_buffer *buf,
-               struct ltt_reserve_switch_offsets *offsets, u64 *tsc)
-{
-       long beginidx = SUBBUF_INDEX(offsets->begin, chan);
-       long commit_count;
-
-       ltt_buffer_begin(buf, *tsc, beginidx);
-
-       /*
-        * Must write slot data before incrementing commit count.
-        * This compiler barrier is upgraded into a cmm_smp_wmb() by the IPI
-        * sent by get_subbuf() when it does its cmm_smp_rmb().
-        */
-       cmm_smp_wmb();
-       uatomic_add(&buf->commit_count[beginidx].cc, ltt_subbuffer_header_size());
-       commit_count = uatomic_read(&buf->commit_count[beginidx].cc);
-       /* Check if the written buffer has to be delivered */
-       ltt_check_deliver(chan, buf, offsets->begin, commit_count, beginidx);
-       ltt_write_commit_counter(chan, buf, beginidx,
-               offsets->begin, commit_count, ltt_subbuffer_header_size());
-}
-
-/*
- * ltt_reserve_end_switch_current: finish switching current subbuffer
- *
- * Concurrency safe because we are the last and only thread to alter this
- * sub-buffer. As long as it is not delivered and read, no other thread can
- * alter the offset, alter the reserve_count or call the
- * client_buffer_end_callback on this sub-buffer.
- *
- * The only remaining threads could be the ones with pending commits. They will
- * have to do the deliver themselves.  Not concurrency safe in overwrite mode.
- * We detect corrupted subbuffers with commit and reserve counts. We keep a
- * corrupted sub-buffers count and push the readers across these sub-buffers.
- *
- * Not concurrency safe if a writer is stalled in a subbuffer and another writer
- * switches in, finding out it's corrupted.  The result will be than the old
- * (uncommited) subbuffer will be declared corrupted, and that the new subbuffer
- * will be declared corrupted too because of the commit count adjustment.
- */
-static void ltt_reserve_end_switch_current(
-               struct ust_channel *chan,
-               struct ust_buffer *buf,
-               struct ltt_reserve_switch_offsets *offsets, u64 *tsc)
-{
-       long endidx = SUBBUF_INDEX(offsets->end - 1, chan);
-       long commit_count, padding_size;
-
-       padding_size = chan->subbuf_size
-                       - (SUBBUF_OFFSET(offsets->end - 1, chan) + 1);
-
-       ltt_buffer_end(buf, *tsc, offsets->end, endidx);
-
-       /*
-        * Must write slot data before incrementing commit count.
-        * This compiler barrier is upgraded into a cmm_smp_wmb() by the IPI
-        * sent by get_subbuf() when it does its cmm_smp_rmb().
-        */
-       cmm_smp_wmb();
-       uatomic_add(&buf->commit_count[endidx].cc, padding_size);
-       commit_count = uatomic_read(&buf->commit_count[endidx].cc);
-       ltt_check_deliver(chan, buf,
-               offsets->end - 1, commit_count, endidx);
-       ltt_write_commit_counter(chan, buf, endidx,
-               offsets->end, commit_count, padding_size);
-}
-
-/*
- * Returns :
- * 0 if ok
- * !0 if execution must be aborted.
- */
-static int ltt_relay_try_switch_slow(
-               enum force_switch_mode mode,
-               struct ust_channel *chan,
-               struct ust_buffer *buf,
-               struct ltt_reserve_switch_offsets *offsets,
-               u64 *tsc)
-{
-       long subbuf_index;
-       long reserve_commit_diff;
-
-       offsets->begin = uatomic_read(&buf->offset);
-       offsets->old = offsets->begin;
-       offsets->begin_switch = 0;
-       offsets->end_switch_old = 0;
-
-       *tsc = trace_clock_read64();
-
-       if (SUBBUF_OFFSET(offsets->begin, buf->chan) != 0) {
-               offsets->begin = SUBBUF_ALIGN(offsets->begin, buf->chan);
-               offsets->end_switch_old = 1;
-       } else {
-               /* we do not have to switch : buffer is empty */
-               return -1;
-       }
-       if (mode == FORCE_ACTIVE)
-               offsets->begin += ltt_subbuffer_header_size();
-       /*
-        * Always begin_switch in FORCE_ACTIVE mode.
-        * Test new buffer integrity
-        */
-       subbuf_index = SUBBUF_INDEX(offsets->begin, buf->chan);
-       reserve_commit_diff =
-               (BUFFER_TRUNC(offsets->begin, buf->chan)
-                >> chan->n_subbufs_order)
-               - (uatomic_read(&buf->commit_count[subbuf_index].cc_sb)
-                       & chan->commit_count_mask);
-       if (reserve_commit_diff == 0) {
-               /* Next buffer not corrupted. */
-               if (mode == FORCE_ACTIVE
-                   && !chan->overwrite
-                   && offsets->begin - uatomic_read(&buf->consumed)
-                      >= chan->alloc_size) {
-                       /*
-                        * We do not overwrite non consumed buffers and we are
-                        * full : ignore switch while tracing is active.
-                        */
-                       return -1;
-               }
-       } else {
-               /*
-                * Next subbuffer corrupted. Force pushing reader even in normal
-                * mode
-                */
-       }
-       offsets->end = offsets->begin;
-       return 0;
-}
-
-/*
- * Force a sub-buffer switch for a per-cpu buffer. This operation is
- * completely reentrant : can be called while tracing is active with
- * absolutely no lock held.
- */
-void ltt_force_switch_lockless_slow(struct ust_buffer *buf,
-               enum force_switch_mode mode)
-{
-       struct ust_channel *chan = buf->chan;
-       struct ltt_reserve_switch_offsets offsets;
-       u64 tsc;
-
-       offsets.size = 0;
-
-       DBG("Switching (forced) %s_%d", chan->channel_name, buf->cpu);
-       /*
-        * Perform retryable operations.
-        */
-       do {
-               if (ltt_relay_try_switch_slow(mode, chan, buf,
-                               &offsets, &tsc))
-                       return;
-       } while (uatomic_cmpxchg(&buf->offset, offsets.old,
-                       offsets.end) != offsets.old);
-
-       /*
-        * Atomically update last_tsc. This update races against concurrent
-        * atomic updates, but the race will always cause supplementary full TSC
-        * events, never the opposite (missing a full TSC event when it would be
-        * needed).
-        */
-       save_last_tsc(buf, tsc);
-
-       /*
-        * Push the reader if necessary
-        */
-       if (mode == FORCE_ACTIVE) {
-               ltt_reserve_push_reader(chan, buf, offsets.end - 1);
-       }
-
-       /*
-        * Switch old subbuffer if needed.
-        */
-       if (offsets.end_switch_old) {
-               ltt_reserve_switch_old_subbuf(chan, buf, &offsets, &tsc);
-       }
-
-       /*
-        * Populate new subbuffer.
-        */
-       if (mode == FORCE_ACTIVE)
-               ltt_reserve_switch_new_subbuf(chan, buf, &offsets, &tsc);
-}
-
-/*
- * Returns :
- * 0 if ok
- * !0 if execution must be aborted.
- */
-static int ltt_relay_try_reserve_slow(struct ust_channel *chan, struct ust_buffer *buf,
-               struct ltt_reserve_switch_offsets *offsets, size_t data_size,
-               u64 *tsc, unsigned int *rflags, int largest_align)
-{
-       long reserve_commit_diff;
-
-       offsets->begin = uatomic_read(&buf->offset);
-       offsets->old = offsets->begin;
-       offsets->begin_switch = 0;
-       offsets->end_switch_current = 0;
-       offsets->end_switch_old = 0;
-
-       *tsc = trace_clock_read64();
-       if (last_tsc_overflow(buf, *tsc))
-               *rflags = LTT_RFLAG_ID_SIZE_TSC;
-
-       if (unlikely(SUBBUF_OFFSET(offsets->begin, buf->chan) == 0)) {
-               offsets->begin_switch = 1;              /* For offsets->begin */
-       } else {
-               offsets->size = ust_get_header_size(chan,
-                                       offsets->begin, data_size,
-                                       &offsets->before_hdr_pad, *rflags);
-               offsets->size += ltt_align(offsets->begin + offsets->size,
-                                          largest_align)
-                                + data_size;
-               if (unlikely((SUBBUF_OFFSET(offsets->begin, buf->chan) +
-                            offsets->size) > buf->chan->subbuf_size)) {
-                       offsets->end_switch_old = 1;    /* For offsets->old */
-                       offsets->begin_switch = 1;      /* For offsets->begin */
-               }
-       }
-       if (unlikely(offsets->begin_switch)) {
-               long subbuf_index;
-
-               /*
-                * We are typically not filling the previous buffer completely.
-                */
-               if (likely(offsets->end_switch_old))
-                       offsets->begin = SUBBUF_ALIGN(offsets->begin,
-                                                     buf->chan);
-               offsets->begin = offsets->begin + ltt_subbuffer_header_size();
-               /* Test new buffer integrity */
-               subbuf_index = SUBBUF_INDEX(offsets->begin, buf->chan);
-               reserve_commit_diff =
-                 (BUFFER_TRUNC(offsets->begin, buf->chan)
-                  >> chan->n_subbufs_order)
-                 - (uatomic_read(&buf->commit_count[subbuf_index].cc_sb)
-                               & chan->commit_count_mask);
-               if (likely(reserve_commit_diff == 0)) {
-                       /* Next buffer not corrupted. */
-                       if (unlikely(!chan->overwrite &&
-                               (SUBBUF_TRUNC(offsets->begin, buf->chan)
-                                - SUBBUF_TRUNC(uatomic_read(
-                                                       &buf->consumed),
-                                               buf->chan))
-                               >= chan->alloc_size)) {
-                               /*
-                                * We do not overwrite non consumed buffers
-                                * and we are full : event is lost.
-                                */
-                               uatomic_inc(&buf->events_lost);
-                               return -1;
-                       } else {
-                               /*
-                                * next buffer not corrupted, we are either in
-                                * overwrite mode or the buffer is not full.
-                                * It's safe to write in this new subbuffer.
-                                */
-                       }
-               } else {
-                       /*
-                        * Next subbuffer corrupted. Drop event in normal and
-                        * overwrite mode. Caused by either a writer OOPS or
-                        * too many nested writes over a reserve/commit pair.
-                        */
-                       uatomic_inc(&buf->events_lost);
-                       return -1;
-               }
-               offsets->size = ust_get_header_size(chan,
-                                       offsets->begin, data_size,
-                                       &offsets->before_hdr_pad, *rflags);
-               offsets->size += ltt_align(offsets->begin + offsets->size,
-                                          largest_align)
-                                + data_size;
-               if (unlikely((SUBBUF_OFFSET(offsets->begin, buf->chan)
-                            + offsets->size) > buf->chan->subbuf_size)) {
-                       /*
-                        * Event too big for subbuffers, report error, don't
-                        * complete the sub-buffer switch.
-                        */
-                       uatomic_inc(&buf->events_lost);
-                       return -1;
-               } else {
-                       /*
-                        * We just made a successful buffer switch and the event
-                        * fits in the new subbuffer. Let's write.
-                        */
-               }
-       } else {
-               /*
-                * Event fits in the current buffer and we are not on a switch
-                * boundary. It's safe to write.
-                */
-       }
-       offsets->end = offsets->begin + offsets->size;
-
-       if (unlikely((SUBBUF_OFFSET(offsets->end, buf->chan)) == 0)) {
-               /*
-                * The offset_end will fall at the very beginning of the next
-                * subbuffer.
-                */
-               offsets->end_switch_current = 1;        /* For offsets->begin */
-       }
-       return 0;
-}
-
-/**
- * ltt_relay_reserve_slot_lockless_slow - Atomic slot reservation in a buffer.
- * @trace: the trace structure to log to.
- * @ltt_channel: channel structure
- * @transport_data: data structure specific to ltt relay
- * @data_size: size of the variable length data to log.
- * @slot_size: pointer to total size of the slot (out)
- * @buf_offset : pointer to reserved buffer offset (out)
- * @tsc: pointer to the tsc at the slot reservation (out)
- * @cpu: cpuid
- *
- * Return : -ENOSPC if not enough space, else returns 0.
- * It will take care of sub-buffer switching.
- */
-int ltt_reserve_slot_lockless_slow(struct ust_channel *chan,
-               struct ust_trace *trace, size_t data_size,
-               int largest_align, int cpu,
-               struct ust_buffer **ret_buf,
-               size_t *slot_size, long *buf_offset,
-               u64 *tsc, unsigned int *rflags)
-{
-       struct ust_buffer *buf = *ret_buf = chan->buf[cpu];
-       struct ltt_reserve_switch_offsets offsets;
-
-       offsets.size = 0;
-
-       do {
-               if (unlikely(ltt_relay_try_reserve_slow(chan, buf, &offsets,
-                               data_size, tsc, rflags, largest_align)))
-                       return -ENOSPC;
-       } while (unlikely(uatomic_cmpxchg(&buf->offset, offsets.old,
-                       offsets.end) != offsets.old));
-
-       /*
-        * Atomically update last_tsc. This update races against concurrent
-        * atomic updates, but the race will always cause supplementary full TSC
-        * events, never the opposite (missing a full TSC event when it would be
-        * needed).
-        */
-       save_last_tsc(buf, *tsc);
-
-       /*
-        * Push the reader if necessary
-        */
-       ltt_reserve_push_reader(chan, buf, offsets.end - 1);
-
-       /*
-        * Switch old subbuffer if needed.
-        */
-       if (unlikely(offsets.end_switch_old)) {
-               ltt_reserve_switch_old_subbuf(chan, buf, &offsets, tsc);
-               DBG("Switching %s_%d", chan->channel_name, cpu);
-       }
-
-       /*
-        * Populate new subbuffer.
-        */
-       if (unlikely(offsets.begin_switch))
-               ltt_reserve_switch_new_subbuf(chan, buf, &offsets, tsc);
-
-       if (unlikely(offsets.end_switch_current))
-               ltt_reserve_end_switch_current(chan, buf, &offsets, tsc);
-
-       *slot_size = offsets.size;
-       *buf_offset = offsets.begin + offsets.before_hdr_pad;
-       return 0;
-}
-
-static struct ltt_transport ust_relay_transport = {
-       .name = "ustrelay",
-       .ops = {
-               .create_channel = create_channel,
-               .finish_channel = finish_channel,
-               .remove_channel = remove_channel,
-               .wakeup_channel = ltt_relay_async_wakeup_chan,
-       },
-};
-
-static char initialized = 0;
-
-void __attribute__((constructor)) init_ustrelay_transport(void)
-{
-       init_usterr();
-       if(!initialized) {
-               ltt_transport_register(&ust_relay_transport);
-               initialized = 1;
-       }
-}
-
-static void __attribute__((destructor)) ust_buffers_exit(void)
-{
-       ltt_transport_unregister(&ust_relay_transport);
-}
-
-size_t ltt_write_event_header_slow(struct ust_channel *channel,
-               struct ust_buffer *buf, long buf_offset,
-               u16 eID, u32 event_size,
-               u64 tsc, unsigned int rflags)
-{
-       struct ltt_event_header header;
-       u16 small_size;
-
-       switch (rflags) {
-       case LTT_RFLAG_ID_SIZE_TSC:
-               header.id_time = 29 << LTT_TSC_BITS;
-               break;
-       case LTT_RFLAG_ID_SIZE:
-               header.id_time = 30 << LTT_TSC_BITS;
-               break;
-       case LTT_RFLAG_ID:
-               header.id_time = 31 << LTT_TSC_BITS;
-               break;
-       default:
-               WARN_ON_ONCE(1);
-               header.id_time = 0;
-               break;
-       }
-
-       header.id_time |= (u32)tsc & LTT_TSC_MASK;
-       ust_buffers_write(buf, buf_offset, &header, sizeof(header));
-       buf_offset += sizeof(header);
-
-       switch (rflags) {
-       case LTT_RFLAG_ID_SIZE_TSC:
-               small_size = (u16)min_t(u32, event_size, LTT_MAX_SMALL_SIZE);
-               ust_buffers_write(buf, buf_offset,
-                       &eID, sizeof(u16));
-               buf_offset += sizeof(u16);
-               ust_buffers_write(buf, buf_offset,
-                       &small_size, sizeof(u16));
-               buf_offset += sizeof(u16);
-               if (small_size == LTT_MAX_SMALL_SIZE) {
-                       ust_buffers_write(buf, buf_offset,
-                               &event_size, sizeof(u32));
-                       buf_offset += sizeof(u32);
-               }
-               buf_offset += ltt_align(buf_offset, sizeof(u64));
-               ust_buffers_write(buf, buf_offset,
-                       &tsc, sizeof(u64));
-               buf_offset += sizeof(u64);
-               break;
-       case LTT_RFLAG_ID_SIZE:
-               small_size = (u16)min_t(u32, event_size, LTT_MAX_SMALL_SIZE);
-               ust_buffers_write(buf, buf_offset,
-                       &eID, sizeof(u16));
-               buf_offset += sizeof(u16);
-               ust_buffers_write(buf, buf_offset,
-                       &small_size, sizeof(u16));
-               buf_offset += sizeof(u16);
-               if (small_size == LTT_MAX_SMALL_SIZE) {
-                       ust_buffers_write(buf, buf_offset,
-                               &event_size, sizeof(u32));
-                       buf_offset += sizeof(u32);
-               }
-               break;
-       case LTT_RFLAG_ID:
-               ust_buffers_write(buf, buf_offset,
-                       &eID, sizeof(u16));
-               buf_offset += sizeof(u16);
-               break;
-       }
-
-       return buf_offset;
-}
diff --git a/libust/buffers.h b/libust/buffers.h
deleted file mode 100644 (file)
index a838644..0000000
+++ /dev/null
@@ -1,699 +0,0 @@
-/*
- * buffers.h
- * LTTng userspace tracer buffering system
- *
- * Copyright (C) 2009 - Pierre-Marc Fournier (pierre-marc dot fournier at polymtl dot ca)
- * Copyright (C) 2008 - Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca)
- *
- * 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; 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
- */
-
-#ifndef _UST_BUFFERS_H
-#define _UST_BUFFERS_H
-
-#include <assert.h>
-
-#include <ust/core.h>
-#include <ust/clock.h>
-
-#include "usterr_signal_safe.h"
-#include "channels.h"
-#include "tracerconst.h"
-#include "tracercore.h"
-
-/*
- * BUFFER_TRUNC zeroes the subbuffer offset and the subbuffer number parts of
- * the offset, which leaves only the buffer number.
- */
-#define BUFFER_TRUNC(offset, chan) \
-       ((offset) & (~((chan)->alloc_size-1)))
-#define BUFFER_OFFSET(offset, chan) ((offset) & ((chan)->alloc_size - 1))
-#define SUBBUF_OFFSET(offset, chan) ((offset) & ((chan)->subbuf_size - 1))
-#define SUBBUF_ALIGN(offset, chan) \
-       (((offset) + (chan)->subbuf_size) & (~((chan)->subbuf_size - 1)))
-#define SUBBUF_TRUNC(offset, chan) \
-       ((offset) & (~((chan)->subbuf_size - 1)))
-#define SUBBUF_INDEX(offset, chan) \
-       (BUFFER_OFFSET((offset), chan) >> (chan)->subbuf_size_order)
-
-/*
- * Tracks changes to rchan/rchan_buf structs
- */
-#define UST_CHANNEL_VERSION            8
-
-/**************************************/
-
-/*
- * TODO: using "long" type for struct ust_buffer (control structure
- * shared between traced apps and the consumer) is a very bad idea when
- * we get to systems with mixed 32/64-bit processes.
- *
- * But on 64-bit system, we want the full power of 64-bit counters,
- * which wraps less often. Therefore, it's not as easy as "use 32-bit
- * types everywhere".
- *
- * One way to deal with this is to:
- * 1) Design 64-bit consumer so it can detect 32-bit and 64-bit apps.
- * 2) The 32-bit consumer only supports 32-bit apps.
- */
-
-struct commit_counters {
-       long cc;                        /* ATOMIC */
-       long cc_sb;                     /* ATOMIC - Incremented _once_ at sb switch */
-};
-
-struct ust_buffer {
-       /* First 32 bytes cache-hot cacheline */
-       long offset;                    /* Current offset in the buffer *atomic* */
-       struct commit_counters *commit_count;   /* Commit count per sub-buffer */
-       long consumed;                  /* Current offset in the buffer *atomic* access (shared) */
-       unsigned long last_tsc;         /*
-                                        * Last timestamp written in the buffer.
-                                        */
-       /* End of first 32 bytes cacheline */
-       long active_readers;    /* ATOMIC - Active readers count standard atomic access (shared) */
-       long events_lost;       /* ATOMIC */
-       long corrupted_subbuffers; /* *ATOMIC* */
-       /* one byte is written to this pipe when data is available, in order
-           to wake the consumer */
-       /* portability: Single byte writes must be as quick as possible. The kernel-side
-          buffer must be large enough so the writer doesn't block. From the pipe(7)
-           man page: Since linux 2.6.11, the pipe capacity is 65536 bytes. */
-       int data_ready_fd_write;
-       /* the reading end of the pipe */
-       int data_ready_fd_read;
-       /*
-        * List of buffers with an open pipe, used for fork and forced subbuffer
-        * switch.
-        */
-       struct cds_list_head open_buffers_list;
-
-       unsigned int finalized;
-//ust//        struct timer_list switch_timer; /* timer for periodical switch */
-       unsigned long switch_timer_interval; /* 0 = unset */
-
-       struct ust_channel *chan;
-
-       struct urcu_ref urcu_ref;
-       void *buf_data;
-       size_t buf_size;
-       int shmid;
-       unsigned int cpu;
-
-       /* commit count per subbuffer; must be at end of struct */
-       long commit_seq[0]; /* ATOMIC */
-} ____cacheline_aligned;
-
-/*
- * A switch is either done during tracing (FORCE_ACTIVE) or as a final
- * flush after tracing (with FORCE_FLUSH). FORCE_FLUSH ensures we won't
- * write in the new sub-buffer).
- */
-enum force_switch_mode { FORCE_ACTIVE, FORCE_FLUSH };
-
-extern int ltt_reserve_slot_lockless_slow(struct ust_channel *chan,
-               struct ust_trace *trace, size_t data_size,
-               int largest_align, int cpu,
-               struct ust_buffer **ret_buf,
-               size_t *slot_size, long *buf_offset,
-               u64 *tsc, unsigned int *rflags);
-
-extern void ltt_force_switch_lockless_slow(struct ust_buffer *buf,
-               enum force_switch_mode mode);
-
-#ifndef HAVE_EFFICIENT_UNALIGNED_ACCESS
-
-/*
- * Calculate the offset needed to align the type.
- * size_of_type must be non-zero.
- */
-static inline unsigned int ltt_align(size_t align_drift, size_t size_of_type)
-{
-       size_t alignment = min(sizeof(void *), size_of_type);
-       return (alignment - align_drift) & (alignment - 1);
-}
-/* Default arch alignment */
-#define LTT_ALIGN
-
-static inline int ltt_get_alignment(void)
-{
-       return sizeof(void *);
-}
-
-#else /* HAVE_EFFICIENT_UNALIGNED_ACCESS */
-
-static inline unsigned int ltt_align(size_t align_drift,
-                size_t size_of_type)
-{
-       return 0;
-}
-
-#define LTT_ALIGN __attribute__((packed))
-
-static inline int ltt_get_alignment(void)
-{
-       return 0;
-}
-#endif /* HAVE_EFFICIENT_UNALIGNED_ACCESS */
-
-static __inline__ void ust_buffers_do_copy(void *dest, const void *src, size_t len)
-{
-       union {
-               const void *src;
-               const u8 *src8;
-               const u16 *src16;
-               const u32 *src32;
-               const u64 *src64;
-       } u = { .src = src };
-
-       switch (len) {
-       case 0: break;
-       case 1: *(u8 *)dest = *u.src8;
-               break;
-       case 2: *(u16 *)dest = *u.src16;
-               break;
-       case 4: *(u32 *)dest = *u.src32;
-               break;
-       case 8: *(u64 *)dest = *u.src64;
-               break;
-       default:
-               memcpy(dest, src, len);
-       }
-}
-
-static __inline__ void *ust_buffers_offset_address(struct ust_buffer *buf, size_t offset)
-{
-       return ((char *)buf->buf_data)+offset;
-}
-
-/*
- * Last TSC comparison functions. Check if the current TSC overflows
- * LTT_TSC_BITS bits from the last TSC read. Reads and writes last_tsc
- * atomically.
- */
-
-/* FIXME: does this test work properly? */
-#if (BITS_PER_LONG == 32)
-static __inline__ void save_last_tsc(struct ust_buffer *ltt_buf,
-                                       u64 tsc)
-{
-       ltt_buf->last_tsc = (unsigned long)(tsc >> LTT_TSC_BITS);
-}
-
-static __inline__ int last_tsc_overflow(struct ust_buffer *ltt_buf,
-                                       u64 tsc)
-{
-       unsigned long tsc_shifted = (unsigned long)(tsc >> LTT_TSC_BITS);
-
-       if (unlikely((tsc_shifted - ltt_buf->last_tsc)))
-               return 1;
-       else
-               return 0;
-}
-#else
-static __inline__ void save_last_tsc(struct ust_buffer *ltt_buf,
-                                       u64 tsc)
-{
-       ltt_buf->last_tsc = (unsigned long)tsc;
-}
-
-static __inline__ int last_tsc_overflow(struct ust_buffer *ltt_buf,
-                                       u64 tsc)
-{
-       if (unlikely((tsc - ltt_buf->last_tsc) >> LTT_TSC_BITS))
-               return 1;
-       else
-               return 0;
-}
-#endif
-
-/*
- * ust_get_header_size
- *
- * Calculate alignment offset to 32-bits. This is the alignment offset of the
- * event header.
- *
- * Important note :
- * The event header must be 32-bits. The total offset calculated here :
- *
- * Alignment of header struct on 32 bits (min arch size, header size)
- * + sizeof(header struct)  (32-bits)
- * + (opt) u16 (ext. event id)
- * + (opt) u16 (event_size) (if event_size == 0xFFFFUL, has ext. event size)
- * + (opt) u32 (ext. event size)
- * + (opt) u64 full TSC (aligned on min(64-bits, arch size))
- *
- * The payload must itself determine its own alignment from the biggest type it
- * contains.
- * */
-static __inline__ unsigned char ust_get_header_size(
-               struct ust_channel *channel,
-               size_t offset,
-               size_t data_size,
-               size_t *before_hdr_pad,
-               unsigned int rflags)
-{
-       size_t orig_offset = offset;
-       size_t padding;
-
-       padding = ltt_align(offset, sizeof(struct ltt_event_header));
-       offset += padding;
-       offset += sizeof(struct ltt_event_header);
-
-       if(unlikely(rflags)) {
-               switch (rflags) {
-               case LTT_RFLAG_ID_SIZE_TSC:
-                       offset += sizeof(u16) + sizeof(u16);
-                       if (data_size >= 0xFFFFU)
-                               offset += sizeof(u32);
-                       offset += ltt_align(offset, sizeof(u64));
-                       offset += sizeof(u64);
-                       break;
-               case LTT_RFLAG_ID_SIZE:
-                       offset += sizeof(u16) + sizeof(u16);
-                       if (data_size >= 0xFFFFU)
-                               offset += sizeof(u32);
-                       break;
-               case LTT_RFLAG_ID:
-                       offset += sizeof(u16);
-                       break;
-               }
-       }
-
-       *before_hdr_pad = padding;
-       return offset - orig_offset;
-}
-
-static __inline__ void ltt_reserve_push_reader(
-               struct ust_channel *rchan,
-               struct ust_buffer *buf,
-               long offset)
-{
-       long consumed_old, consumed_new;
-
-       do {
-               consumed_old = uatomic_read(&buf->consumed);
-               /*
-                * If buffer is in overwrite mode, push the reader consumed
-                * count if the write position has reached it and we are not
-                * at the first iteration (don't push the reader farther than
-                * the writer). This operation can be done concurrently by many
-                * writers in the same buffer, the writer being at the farthest
-                * write position sub-buffer index in the buffer being the one
-                * which will win this loop.
-                * If the buffer is not in overwrite mode, pushing the reader
-                * only happens if a sub-buffer is corrupted.
-                */
-               if (unlikely((SUBBUF_TRUNC(offset, buf->chan)
-                  - SUBBUF_TRUNC(consumed_old, buf->chan))
-                  >= rchan->alloc_size))
-                       consumed_new = SUBBUF_ALIGN(consumed_old, buf->chan);
-               else
-                       return;
-       } while (unlikely(uatomic_cmpxchg(&buf->consumed, consumed_old,
-                       consumed_new) != consumed_old));
-}
-
-static __inline__ void ltt_vmcore_check_deliver(
-               struct ust_buffer *buf,
-               long commit_count, long idx)
-{
-       uatomic_set(&buf->commit_seq[idx], commit_count);
-}
-
-static __inline__ void ltt_check_deliver(struct ust_channel *chan,
-               struct ust_buffer *buf,
-               long offset, long commit_count, long idx)
-{
-       long old_commit_count = commit_count - chan->subbuf_size;
-
-       /* Check if all commits have been done */
-       if (unlikely((BUFFER_TRUNC(offset, chan)
-                       >> chan->n_subbufs_order)
-                       - (old_commit_count
-                          & chan->commit_count_mask) == 0)) {
-               /*
-                * If we succeeded in updating the cc_sb, we are delivering
-                * the subbuffer. Deals with concurrent updates of the "cc"
-                * value without adding a add_return atomic operation to the
-                * fast path.
-                */
-               if (likely(uatomic_cmpxchg(&buf->commit_count[idx].cc_sb,
-                                        old_commit_count, commit_count)
-                          == old_commit_count)) {
-                       int result;
-
-                       /*
-                        * Set noref flag for this subbuffer.
-                        */
-//ust//                        ltt_set_noref_flag(rchan, buf, idx);
-                       ltt_vmcore_check_deliver(buf, commit_count, idx);
-
-                       /* wakeup consumer */
-                       result = write(buf->data_ready_fd_write, "1", 1);
-                       if(result == -1) {
-                               PERROR("write (in ltt_relay_buffer_flush)");
-                               ERR("this should never happen!");
-                       }
-               }
-       }
-}
-
-static __inline__ int ltt_poll_deliver(struct ust_channel *chan, struct ust_buffer *buf)
-{
-       long consumed_old, consumed_idx, commit_count, write_offset;
-
-       consumed_old = uatomic_read(&buf->consumed);
-       consumed_idx = SUBBUF_INDEX(consumed_old, buf->chan);
-       commit_count = uatomic_read(&buf->commit_count[consumed_idx].cc_sb);
-       /*
-        * No memory cmm_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 = uatomic_read(&buf->offset);
-
-       /*
-        * Check that the subbuffer we are trying to consume has been
-        * already fully committed.
-        */
-
-       if (((commit_count - chan->subbuf_size)
-            & chan->commit_count_mask)
-           - (BUFFER_TRUNC(consumed_old, buf->chan)
-              >> chan->n_subbufs_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, buf->chan)
-          - SUBBUF_TRUNC(consumed_old, buf->chan))
-          == 0)
-               return 0;
-
-       return 1;
-
-}
-
-/*
- * returns 0 if reserve ok, or 1 if the slow path must be taken.
- */
-static __inline__ int ltt_relay_try_reserve(
-               struct ust_channel *chan,
-               struct ust_buffer *buf,
-               size_t data_size,
-               u64 *tsc, unsigned int *rflags, int largest_align,
-               long *o_begin, long *o_end, long *o_old,
-               size_t *before_hdr_pad, size_t *size)
-{
-       *o_begin = uatomic_read(&buf->offset);
-       *o_old = *o_begin;
-
-       *tsc = trace_clock_read64();
-
-       if (last_tsc_overflow(buf, *tsc))
-               *rflags = LTT_RFLAG_ID_SIZE_TSC;
-
-       if (unlikely(SUBBUF_OFFSET(*o_begin, buf->chan) == 0))
-               return 1;
-
-       *size = ust_get_header_size(chan,
-                               *o_begin, data_size,
-                               before_hdr_pad, *rflags);
-       *size += ltt_align(*o_begin + *size, largest_align) + data_size;
-       if (unlikely((SUBBUF_OFFSET(*o_begin, buf->chan) + *size)
-                    > buf->chan->subbuf_size))
-               return 1;
-
-       /*
-        * Event fits in the current buffer and we are not on a switch
-        * boundary. It's safe to write.
-        */
-       *o_end = *o_begin + *size;
-
-       if (unlikely((SUBBUF_OFFSET(*o_end, buf->chan)) == 0))
-               /*
-                * The offset_end will fall at the very beginning of the next
-                * subbuffer.
-                */
-               return 1;
-
-       return 0;
-}
-
-static __inline__ int ltt_reserve_slot(struct ust_channel *chan,
-                                      struct ust_trace *trace, size_t data_size,
-                                      int largest_align, int cpu,
-                                      struct ust_buffer **ret_buf,
-                                      size_t *slot_size, long *buf_offset, u64 *tsc,
-                                      unsigned int *rflags)
-{
-       struct ust_buffer *buf = *ret_buf = chan->buf[cpu];
-       long o_begin, o_end, o_old;
-       size_t before_hdr_pad;
-
-       /*
-        * Perform retryable operations.
-        */
-       if (unlikely(CMM_LOAD_SHARED(ltt_nesting) > 4)) {
-               DBG("Dropping event because nesting is too deep.");
-               uatomic_inc(&buf->events_lost);
-               return -EPERM;
-       }
-
-       if (unlikely(ltt_relay_try_reserve(chan, buf,
-                       data_size, tsc, rflags,
-                       largest_align, &o_begin, &o_end, &o_old,
-                       &before_hdr_pad, slot_size)))
-               goto slow_path;
-
-       if (unlikely(uatomic_cmpxchg(&buf->offset, o_old, o_end) != o_old))
-               goto slow_path;
-
-       /*
-        * Atomically update last_tsc. This update races against concurrent
-        * atomic updates, but the race will always cause supplementary full TSC
-        * events, never the opposite (missing a full TSC event when it would be
-        * needed).
-        */
-       save_last_tsc(buf, *tsc);
-
-       /*
-        * Push the reader if necessary
-        */
-       ltt_reserve_push_reader(chan, buf, o_end - 1);
-
-       *buf_offset = o_begin + before_hdr_pad;
-       return 0;
-slow_path:
-       return ltt_reserve_slot_lockless_slow(chan, trace, data_size,
-                                             largest_align, cpu, ret_buf,
-                                             slot_size, buf_offset, tsc,
-                                             rflags);
-}
-
-/*
- * Force a sub-buffer switch for a per-cpu buffer. This operation is
- * completely reentrant : can be called while tracing is active with
- * absolutely no lock held.
- */
-static __inline__ void ltt_force_switch(struct ust_buffer *buf,
-               enum force_switch_mode mode)
-{
-       return ltt_force_switch_lockless_slow(buf, mode);
-}
-
-/*
- * for flight recording. must be called after relay_commit.
- * This function increments the subbuffers's commit_seq counter each time the
- * commit count reaches back the reserve offset (module subbuffer size). It is
- * useful for crash dump.
- */
-static __inline__ void ltt_write_commit_counter(struct ust_channel *chan,
-               struct ust_buffer *buf, long idx, long buf_offset,
-               long commit_count, size_t data_size)
-{
-       long offset;
-       long commit_seq_old;
-
-       offset = buf_offset + data_size;
-
-       /*
-        * SUBBUF_OFFSET includes commit_count_mask. We can simply
-        * compare the offsets within the subbuffer without caring about
-        * buffer full/empty mismatch because offset is never zero here
-        * (subbuffer header and event headers have non-zero length).
-        */
-       if (unlikely(SUBBUF_OFFSET(offset - commit_count, buf->chan)))
-               return;
-
-       commit_seq_old = uatomic_read(&buf->commit_seq[idx]);
-       while (commit_seq_old < commit_count)
-               commit_seq_old = uatomic_cmpxchg(&buf->commit_seq[idx],
-                                        commit_seq_old, commit_count);
-
-       DBG("commit_seq for channel %s_%d, subbuf %ld is now %ld", buf->chan->channel_name, buf->cpu, idx, commit_count);
-}
-
-/*
- * Atomic unordered slot commit. Increments the commit count in the
- * specified sub-buffer, and delivers it if necessary.
- *
- * Parameters:
- *
- * @ltt_channel : channel structure
- * @transport_data: transport-specific data
- * @buf_offset : offset following the event header.
- * @data_size : size of the event data.
- * @slot_size : size of the reserved slot.
- */
-static __inline__ void ltt_commit_slot(
-               struct ust_channel *chan,
-               struct ust_buffer *buf, long buf_offset,
-               size_t data_size, size_t slot_size)
-{
-       long offset_end = buf_offset;
-       long endidx = SUBBUF_INDEX(offset_end - 1, chan);
-       long commit_count;
-
-       cmm_smp_wmb();
-
-       uatomic_add(&buf->commit_count[endidx].cc, slot_size);
-       /*
-        * commit count read can race with concurrent OOO commit count updates.
-        * This is only needed for ltt_check_deliver (for non-polling delivery
-        * only) and for ltt_write_commit_counter. The race can only cause the
-        * counter to be read with the same value more than once, which could
-        * cause :
-        * - Multiple delivery for the same sub-buffer (which is handled
-        *   gracefully by the reader code) if the value is for a full
-        *   sub-buffer. It's important that we can never miss a sub-buffer
-        *   delivery. Re-reading the value after the uatomic_add ensures this.
-        * - Reading a commit_count with a higher value that what was actually
-        *   added to it for the ltt_write_commit_counter call (again caused by
-        *   a concurrent committer). It does not matter, because this function
-        *   is interested in the fact that the commit count reaches back the
-        *   reserve offset for a specific sub-buffer, which is completely
-        *   independent of the order.
-        */
-       commit_count = uatomic_read(&buf->commit_count[endidx].cc);
-
-       ltt_check_deliver(chan, buf, offset_end - 1, commit_count, endidx);
-       /*
-        * Update data_size for each commit. It's needed only for extracting
-        * ltt buffers from vmcore, after crash.
-        */
-       ltt_write_commit_counter(chan, buf, endidx, buf_offset, commit_count, data_size);
-}
-
-void _ust_buffers_strncpy_fixup(struct ust_buffer *buf, size_t offset,
-                               size_t len, size_t copied, int terminated);
-
-static __inline__ int ust_buffers_write(struct ust_buffer *buf, size_t offset,
-        const void *src, size_t len)
-{
-       size_t buf_offset = BUFFER_OFFSET(offset, buf->chan);
-
-       assert(buf_offset < buf->chan->subbuf_size*buf->chan->subbuf_cnt);
-       assert(buf_offset + len
-              <= buf->chan->subbuf_size*buf->chan->subbuf_cnt);
-
-       ust_buffers_do_copy(buf->buf_data + buf_offset, src, len);
-
-       return len;
-}
-
-/*
- * ust_buffers_do_memset - write character into dest.
- * @dest: destination
- * @src: source character
- * @len: length to write
- */
-static __inline__
-void ust_buffers_do_memset(void *dest, char src, size_t len)
-{
-       /*
-        * What we really want here is an __inline__ memset, but we
-        * don't have constants, so gcc generally uses a function call.
-        */
-       for (; len > 0; len--)
-               *(u8 *)dest++ = src;
-}
-
-/*
- * ust_buffers_do_strncpy - copy a string up to a certain number of bytes
- * @dest: destination
- * @src: source
- * @len: max. length to copy
- * @terminated: output string ends with \0 (output)
- *
- * returns the number of bytes copied. Does not finalize with \0 if len is
- * reached.
- */
-static __inline__
-size_t ust_buffers_do_strncpy(void *dest, const void *src, size_t len,
-                             int *terminated)
-{
-       size_t orig_len = len;
-
-       *terminated = 0;
-       /*
-        * What we really want here is an __inline__ strncpy, but we
-        * don't have constants, so gcc generally uses a function call.
-        */
-       for (; len > 0; len--) {
-               *(u8 *)dest = CMM_LOAD_SHARED(*(const u8 *)src);
-               /* Check with dest, because src may be modified concurrently */
-               if (*(const u8 *)dest == '\0') {
-                       len--;
-                       *terminated = 1;
-                       break;
-               }
-               dest++;
-               src++;
-       }
-       return orig_len - len;
-}
-
-static __inline__
-int ust_buffers_strncpy(struct ust_buffer *buf, size_t offset, const void *src,
-                       size_t len)
-{
-       size_t buf_offset = BUFFER_OFFSET(offset, buf->chan);
-       ssize_t copied;
-       int terminated;
-
-       assert(buf_offset < buf->chan->subbuf_size*buf->chan->subbuf_cnt);
-       assert(buf_offset + len
-              <= buf->chan->subbuf_size*buf->chan->subbuf_cnt);
-
-       copied = ust_buffers_do_strncpy(buf->buf_data + buf_offset,
-                                       src, len, &terminated);
-       if (unlikely(copied < len || !terminated))
-               _ust_buffers_strncpy_fixup(buf, offset, len, copied,
-                                          terminated);
-       return len;
-}
-
-extern int ust_buffers_get_subbuf(struct ust_buffer *buf, long *consumed);
-extern int ust_buffers_put_subbuf(struct ust_buffer *buf, unsigned long uconsumed_old);
-
-extern void init_ustrelay_transport(void);
-
-#endif /* _UST_BUFFERS_H */
This page took 0.041006 seconds and 4 git commands to generate.