install tracer.h, add new ust/ust.h
[ust.git] / include / ust / tracer.h
diff --git a/include/ust/tracer.h b/include/ust/tracer.h
new file mode 100644 (file)
index 0000000..502cdcc
--- /dev/null
@@ -0,0 +1,623 @@
+/*
+ * Copyright (C) 2005,2006,2008 Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca)
+ * Copyright (C) 2009 Pierre-Marc Fournier
+ *
+ * This contains the definitions for the Linux Trace Toolkit tracer.
+ *
+ * Ported to userspace by Pierre-Marc Fournier.
+ *
+ * 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; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * 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 _LTT_TRACER_H
+#define _LTT_TRACER_H
+
+#include <sys/types.h>
+#include <stdarg.h>
+//#include "list.h"
+#include <ust/kernelcompat.h>
+#include "buffer.h"
+#include "relay.h"
+#include "channels.h"
+#include "tracercore.h"
+#include <ust/marker.h>
+
+/* Number of bytes to log with a read/write event */
+#define LTT_LOG_RW_SIZE                        32L
+
+/* Interval (in jiffies) at which the LTT per-CPU timer fires */
+#define LTT_PERCPU_TIMER_INTERVAL      1
+
+#ifndef LTT_ARCH_TYPE
+#define LTT_ARCH_TYPE                  LTT_ARCH_TYPE_UNDEFINED
+#endif
+
+#ifndef LTT_ARCH_VARIANT
+#define LTT_ARCH_VARIANT               LTT_ARCH_VARIANT_NONE
+#endif
+
+struct ltt_active_marker;
+
+/* Maximum number of callbacks per marker */
+#define LTT_NR_CALLBACKS       10
+
+struct ltt_serialize_closure;
+struct ltt_probe_private_data;
+
+/* Serialization callback '%k' */
+typedef size_t (*ltt_serialize_cb)(struct rchan_buf *buf, size_t buf_offset,
+                       struct ltt_serialize_closure *closure,
+                       void *serialize_private, int *largest_align,
+                       const char *fmt, va_list *args);
+
+struct ltt_serialize_closure {
+       ltt_serialize_cb *callbacks;
+       long cb_args[LTT_NR_CALLBACKS];
+       unsigned int cb_idx;
+};
+
+extern size_t ltt_serialize_data(struct rchan_buf *buf, size_t buf_offset,
+                       struct ltt_serialize_closure *closure,
+                       void *serialize_private,
+                       int *largest_align, const char *fmt, va_list *args);
+
+struct ltt_available_probe {
+       const char *name;               /* probe name */
+       const char *format;
+       marker_probe_func *probe_func;
+       ltt_serialize_cb callbacks[LTT_NR_CALLBACKS];
+       struct list_head node;          /* registered probes list */
+};
+
+struct ltt_probe_private_data {
+       struct ltt_trace_struct *trace; /*
+                                        * Target trace, for metadata
+                                        * or statedump.
+                                        */
+       ltt_serialize_cb serializer;    /*
+                                        * Serialization function override.
+                                        */
+       void *serialize_private;        /*
+                                        * Private data for serialization
+                                        * functions.
+                                        */
+};
+
+enum ltt_channels {
+       LTT_CHANNEL_METADATA,
+       LTT_CHANNEL_UST,
+};
+
+struct ltt_active_marker {
+       struct list_head node;          /* active markers list */
+       const char *channel;
+       const char *name;
+       const char *format;
+       struct ltt_available_probe *probe;
+};
+
+struct marker; //ust//
+extern void ltt_vtrace(const struct marker *mdata, void *probe_data,
+       struct registers *regs, void *call_data, const char *fmt, va_list *args);
+extern void ltt_trace(const struct marker *mdata, void *probe_data,
+       struct registers *regs, void *call_data, const char *fmt, ...);
+
+/*
+ * Unique ID assigned to each registered probe.
+ */
+enum marker_id {
+       MARKER_ID_SET_MARKER_ID = 0,    /* Static IDs available (range 0-7) */
+       MARKER_ID_SET_MARKER_FORMAT,
+       MARKER_ID_COMPACT,              /* Compact IDs (range: 8-127)       */
+       MARKER_ID_DYNAMIC,              /* Dynamic IDs (range: 128-65535)   */
+};
+
+/* static ids 0-1 reserved for internal use. */
+#define MARKER_CORE_IDS                2
+static inline enum marker_id marker_id_type(uint16_t id)
+{
+       if (id < MARKER_CORE_IDS)
+               return (enum marker_id)id;
+       else
+               return MARKER_ID_DYNAMIC;
+}
+
+struct user_dbg_data {
+       unsigned long avail_size;
+       unsigned long write;
+       unsigned long read;
+};
+
+struct ltt_trace_ops {
+       /* First 32 bytes cache-hot cacheline */
+       int (*reserve_slot) (struct ltt_trace_struct *trace,
+                               struct ltt_channel_struct *channel,
+                               void **transport_data, size_t data_size,
+                               size_t *slot_size, long *buf_offset, u64 *tsc,
+                               unsigned int *rflags,
+                               int largest_align);
+//ust//        void (*commit_slot) (struct ltt_channel_struct *channel,
+//ust//                                void **transport_data, long buf_offset,
+//ust//                                size_t slot_size);
+       void (*wakeup_channel) (struct ltt_channel_struct *ltt_channel);
+       int (*user_blocking) (struct ltt_trace_struct *trace,
+                               unsigned int index, size_t data_size,
+                               struct user_dbg_data *dbg);
+       /* End of first 32 bytes cacheline */
+       int (*create_dirs) (struct ltt_trace_struct *new_trace);
+       void (*remove_dirs) (struct ltt_trace_struct *new_trace);
+       int (*create_channel) (const char *trace_name,
+                               struct ltt_trace_struct *trace,
+                               struct dentry *dir, const char *channel_name,
+                               struct ltt_channel_struct *ltt_chan,
+                               unsigned int subbuf_size,
+                               unsigned int n_subbufs, int overwrite);
+       void (*finish_channel) (struct ltt_channel_struct *channel);
+       void (*remove_channel) (struct ltt_channel_struct *channel);
+       void (*user_errors) (struct ltt_trace_struct *trace,
+                               unsigned int index, size_t data_size,
+                               struct user_dbg_data *dbg);
+} ____cacheline_aligned;
+
+struct ltt_transport {
+       char *name;
+       struct module *owner;
+       struct list_head node;
+       struct ltt_trace_ops ops;
+};
+
+enum trace_mode { LTT_TRACE_NORMAL, LTT_TRACE_FLIGHT, LTT_TRACE_HYBRID };
+
+#define CHANNEL_FLAG_ENABLE    (1U<<0)
+#define CHANNEL_FLAG_OVERWRITE (1U<<1)
+
+/* Per-trace information - each trace/flight recorder represented by one */
+struct ltt_trace_struct {
+       /* First 32 bytes cache-hot cacheline */
+       struct list_head list;
+       struct ltt_trace_ops *ops;
+       int active;
+       /* Second 32 bytes cache-hot cacheline */
+       struct ltt_channel_struct *channels;
+       unsigned int nr_channels;
+       u32 freq_scale;
+       u64 start_freq;
+       u64 start_tsc;
+       unsigned long long start_monotonic;
+       struct timeval          start_time;
+       struct ltt_channel_setting *settings;
+       struct {
+               struct dentry                   *trace_root;
+       } dentry;
+       struct kref kref; /* Each channel has a kref of the trace struct */
+       struct ltt_transport *transport;
+       struct kref ltt_transport_kref;
+       char trace_name[NAME_MAX];
+} ____cacheline_aligned;
+
+/* Hardcoded event headers
+ *
+ * event header for a trace with active heartbeat : 27 bits timestamps
+ *
+ * headers are 32-bits aligned. In order to insure such alignment, a dynamic per
+ * trace alignment value must be done.
+ *
+ * Remember that the C compiler does align each member on the boundary
+ * equivalent to their own size.
+ *
+ * As relay subbuffers are aligned on pages, we are sure that they are 4 and 8
+ * bytes aligned, so the buffer header and trace header are aligned.
+ *
+ * Event headers are aligned depending on the trace alignment option.
+ *
+ * Note using C structure bitfields for cross-endianness and portability
+ * concerns.
+ */
+
+#define LTT_RESERVED_EVENTS    3
+#define LTT_EVENT_BITS         5
+#define LTT_FREE_EVENTS                ((1 << LTT_EVENT_BITS) - LTT_RESERVED_EVENTS)
+#define LTT_TSC_BITS           27
+#define LTT_TSC_MASK           ((1 << LTT_TSC_BITS) - 1)
+
+struct ltt_event_header {
+       u32 id_time;            /* 5 bits event id (MSB); 27 bits time (LSB) */
+};
+
+/* Reservation flags */
+#define        LTT_RFLAG_ID                    (1 << 0)
+#define        LTT_RFLAG_ID_SIZE               (1 << 1)
+#define        LTT_RFLAG_ID_SIZE_TSC           (1 << 2)
+
+/*
+ * We use asm/timex.h : cpu_khz/HZ variable in here : we might have to deal
+ * specifically with CPU frequency scaling someday, so using an interpolation
+ * between the start and end of buffer values is not flexible enough. Using an
+ * immediate frequency value permits to calculate directly the times for parts
+ * of a buffer that would be before a frequency change.
+ *
+ * Keep the natural field alignment for _each field_ within this structure if
+ * you ever add/remove a field from this header. Packed attribute is not used
+ * because gcc generates poor code on at least powerpc and mips. Don't ever
+ * let gcc add padding between the structure elements.
+ */
+struct ltt_subbuffer_header {
+       uint64_t cycle_count_begin;     /* Cycle count at subbuffer start */
+       uint64_t cycle_count_end;       /* Cycle count at subbuffer end */
+       uint32_t magic_number;          /*
+                                        * Trace magic number.
+                                        * contains endianness information.
+                                        */
+       uint8_t major_version;
+       uint8_t minor_version;
+       uint8_t arch_size;              /* Architecture pointer size */
+       uint8_t alignment;              /* LTT data alignment */
+       uint64_t start_time_sec;        /* NTP-corrected start time */
+       uint64_t start_time_usec;
+       uint64_t start_freq;            /*
+                                        * Frequency at trace start,
+                                        * used all along the trace.
+                                        */
+       uint32_t freq_scale;            /* Frequency scaling (divisor) */
+       uint32_t lost_size;             /* Size unused at end of subbuffer */
+       uint32_t buf_size;              /* Size of this subbuffer */
+       uint32_t events_lost;           /*
+                                        * Events lost in this subbuffer since
+                                        * the beginning of the trace.
+                                        * (may overflow)
+                                        */
+       uint32_t subbuf_corrupt;        /*
+                                        * Corrupted (lost) subbuffers since
+                                        * the begginig of the trace.
+                                        * (may overflow)
+                                        */
+       uint8_t header_end[0];          /* End of header */
+};
+
+/**
+ * ltt_subbuffer_header_size - called on buffer-switch to a new sub-buffer
+ *
+ * Return header size without padding after the structure. Don't use packed
+ * structure because gcc generates inefficient code on some architectures
+ * (powerpc, mips..)
+ */
+static inline size_t ltt_subbuffer_header_size(void)
+{
+       return offsetof(struct ltt_subbuffer_header, header_end);
+}
+
+/*
+ * ltt_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 ltt_get_header_size(
+               struct ltt_channel_struct *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);
+
+       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;
+}
+
+/*
+ * ltt_write_event_header
+ *
+ * Writes the event header to the offset (already aligned on 32-bits).
+ *
+ * @trace : trace to write to.
+ * @channel : pointer to the channel structure..
+ * @buf : buffer to write to.
+ * @buf_offset : buffer offset to write to (aligned on 32 bits).
+ * @eID : event ID
+ * @event_size : size of the event, excluding the event header.
+ * @tsc : time stamp counter.
+ * @rflags : reservation flags.
+ *
+ * returns : offset where the event data must be written.
+ */
+static inline size_t ltt_write_event_header(struct ltt_trace_struct *trace,
+               struct ltt_channel_struct *channel,
+               struct rchan_buf *buf, long buf_offset,
+               u16 eID, size_t event_size,
+               u64 tsc, unsigned int rflags)
+{
+       struct ltt_event_header header;
+       size_t 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:
+               header.id_time = eID << LTT_TSC_BITS;
+               break;
+       }
+       header.id_time |= (u32)tsc & LTT_TSC_MASK;
+       ltt_relay_write(buf, buf_offset, &header, sizeof(header));
+       buf_offset += sizeof(header);
+
+       switch (rflags) {
+       case LTT_RFLAG_ID_SIZE_TSC:
+               small_size = min_t(size_t, event_size, 0xFFFFU);
+               ltt_relay_write(buf, buf_offset,
+                       (u16[]){ (u16)eID }, sizeof(u16));
+               buf_offset += sizeof(u16);
+               ltt_relay_write(buf, buf_offset,
+                       (u16[]){ (u16)small_size }, sizeof(u16));
+               buf_offset += sizeof(u16);
+               if (small_size == 0xFFFFU) {
+                       ltt_relay_write(buf, buf_offset,
+                               (u32[]){ (u32)event_size }, sizeof(u32));
+                       buf_offset += sizeof(u32);
+               }
+               buf_offset += ltt_align(buf_offset, sizeof(u64));
+               ltt_relay_write(buf, buf_offset,
+                       (u64[]){ (u64)tsc }, sizeof(u64));
+               buf_offset += sizeof(u64);
+               break;
+       case LTT_RFLAG_ID_SIZE:
+               small_size = min_t(size_t, event_size, 0xFFFFU);
+               ltt_relay_write(buf, buf_offset,
+                       (u16[]){ (u16)eID }, sizeof(u16));
+               buf_offset += sizeof(u16);
+               ltt_relay_write(buf, buf_offset,
+                       (u16[]){ (u16)small_size }, sizeof(u16));
+               buf_offset += sizeof(u16);
+               if (small_size == 0xFFFFU) {
+                       ltt_relay_write(buf, buf_offset,
+                               (u32[]){ (u32)event_size }, sizeof(u32));
+                       buf_offset += sizeof(u32);
+               }
+               break;
+       case LTT_RFLAG_ID:
+               ltt_relay_write(buf, buf_offset,
+                       (u16[]){ (u16)eID }, sizeof(u16));
+               buf_offset += sizeof(u16);
+               break;
+       default:
+               break;
+       }
+
+       return buf_offset;
+}
+
+/* Lockless LTTng */
+
+/*
+ * ltt_reserve_slot
+ *
+ * Atomic slot reservation in a LTTng buffer. It will take care of
+ * sub-buffer switching.
+ *
+ * Parameters:
+ *
+ * @trace : the trace structure to log to.
+ * @channel : the chanel to reserve space into.
+ * @transport_data : specific transport data.
+ * @data_size : size of the variable length data to log.
+ * @slot_size : pointer to total size of the slot (out)
+ * @buf_offset : pointer to reserve offset (out)
+ * @tsc : pointer to the tsc at the slot reservation (out)
+ * @rflags : reservation flags (header specificity)
+ * @cpu : cpu id
+ *
+ * Return : -ENOSPC if not enough space, else 0.
+ */
+static inline int ltt_reserve_slot(
+               struct ltt_trace_struct *trace,
+               struct ltt_channel_struct *channel,
+               void **transport_data,
+               size_t data_size,
+               size_t *slot_size,
+               long *buf_offset,
+               u64 *tsc,
+               unsigned int *rflags,
+               int largest_align)
+{
+       return trace->ops->reserve_slot(trace, channel, transport_data,
+                       data_size, slot_size, buf_offset, tsc, rflags,
+                       largest_align);
+}
+
+
+///*
+// * ltt_commit_slot
+// *
+// * Atomic unordered slot commit. Increments the commit count in the
+// * specified sub-buffer, and delivers it if necessary.
+// *
+// * Parameters:
+// *
+// * @channel : the chanel to reserve space into.
+// * @transport_data : specific transport data.
+// * @buf_offset : offset of beginning of reserved slot
+// * @slot_size : size of the reserved slot.
+// */
+//static inline void ltt_commit_slot(
+//             struct ltt_channel_struct *channel,
+//             void **transport_data,
+//             long buf_offset,
+//             size_t slot_size)
+//{
+//     struct ltt_trace_struct *trace = channel->trace;
+//
+//     trace->ops->commit_slot(channel, transport_data, buf_offset, slot_size);
+//}
+
+/*
+ * Control channels :
+ * control/metadata
+ * control/interrupts
+ * control/...
+ *
+ * cpu channel :
+ * cpu
+ */
+
+#define LTT_METADATA_CHANNEL           "metadata_state"
+#define LTT_UST_CHANNEL                        "ust"
+
+#define LTT_FLIGHT_PREFIX      "flight-"
+
+/* Tracer properties */
+//#define LTT_DEFAULT_SUBBUF_SIZE_LOW  134217728
+#define LTT_DEFAULT_SUBBUF_SIZE_LOW    65536
+//#define LTT_DEFAULT_SUBBUF_SIZE_LOW  4096
+#define LTT_DEFAULT_N_SUBBUFS_LOW      2
+//#define LTT_DEFAULT_SUBBUF_SIZE_MED  134217728
+#define LTT_DEFAULT_SUBBUF_SIZE_MED    262144
+//#define LTT_DEFAULT_SUBBUF_SIZE_MED  4096
+#define LTT_DEFAULT_N_SUBBUFS_MED      2
+//#define LTT_DEFAULT_SUBBUF_SIZE_HIGH 134217728
+#define LTT_DEFAULT_SUBBUF_SIZE_HIGH   1048576
+//#define LTT_DEFAULT_SUBBUF_SIZE_HIGH 4096
+#define LTT_DEFAULT_N_SUBBUFS_HIGH     2
+#define LTT_TRACER_MAGIC_NUMBER                0x00D6B7ED
+#define LTT_TRACER_VERSION_MAJOR       2
+#define LTT_TRACER_VERSION_MINOR       3
+
+/*
+ * Size reserved for high priority events (interrupts, NMI, BH) at the end of a
+ * nearly full buffer. User space won't use this last amount of space when in
+ * blocking mode. This space also includes the event header that would be
+ * written by this user space event.
+ */
+#define LTT_RESERVE_CRITICAL           4096
+
+/* Register and unregister function pointers */
+
+enum ltt_module_function {
+       LTT_FUNCTION_RUN_FILTER,
+       LTT_FUNCTION_FILTER_CONTROL,
+       LTT_FUNCTION_STATEDUMP
+};
+
+extern void ltt_transport_register(struct ltt_transport *transport);
+extern void ltt_transport_unregister(struct ltt_transport *transport);
+
+/* Exported control function */
+
+union ltt_control_args {
+       struct {
+               enum trace_mode mode;
+               unsigned int subbuf_size_low;
+               unsigned int n_subbufs_low;
+               unsigned int subbuf_size_med;
+               unsigned int n_subbufs_med;
+               unsigned int subbuf_size_high;
+               unsigned int n_subbufs_high;
+       } new_trace;
+};
+
+extern int _ltt_trace_setup(const char *trace_name);
+extern int ltt_trace_setup(const char *trace_name);
+extern struct ltt_trace_struct *_ltt_trace_find_setup(const char *trace_name);
+extern int ltt_trace_set_type(const char *trace_name, const char *trace_type);
+extern int ltt_trace_set_channel_subbufsize(const char *trace_name,
+               const char *channel_name, unsigned int size);
+extern int ltt_trace_set_channel_subbufcount(const char *trace_name,
+               const char *channel_name, unsigned int cnt);
+extern int ltt_trace_set_channel_enable(const char *trace_name,
+               const char *channel_name, unsigned int enable);
+extern int ltt_trace_set_channel_overwrite(const char *trace_name,
+               const char *channel_name, unsigned int overwrite);
+extern int ltt_trace_alloc(const char *trace_name);
+extern int ltt_trace_destroy(const char *trace_name);
+extern int ltt_trace_start(const char *trace_name);
+extern int ltt_trace_stop(const char *trace_name);
+
+enum ltt_filter_control_msg {
+       LTT_FILTER_DEFAULT_ACCEPT,
+       LTT_FILTER_DEFAULT_REJECT
+};
+
+extern int ltt_filter_control(enum ltt_filter_control_msg msg,
+               const char *trace_name);
+
+extern struct dentry *get_filter_root(void);
+
+extern void ltt_write_trace_header(struct ltt_trace_struct *trace,
+               struct ltt_subbuffer_header *header);
+extern void ltt_buffer_destroy(struct ltt_channel_struct *ltt_chan);
+
+extern void ltt_core_register(int (*function)(u8, void *));
+
+extern void ltt_core_unregister(void);
+
+extern void ltt_release_trace(struct kref *kref);
+extern void ltt_release_transport(struct kref *kref);
+
+extern int ltt_probe_register(struct ltt_available_probe *pdata);
+extern int ltt_probe_unregister(struct ltt_available_probe *pdata);
+extern int ltt_marker_connect(const char *channel, const char *mname,
+               const char *pname);
+extern int ltt_marker_disconnect(const char *channel, const char *mname,
+               const char *pname);
+extern void ltt_dump_marker_state(struct ltt_trace_struct *trace);
+
+extern void ltt_lock_traces(void);
+extern void ltt_unlock_traces(void);
+
+extern struct ltt_trace_struct *_ltt_trace_find(const char *trace_name);
+
+#endif /* _LTT_TRACER_H */
This page took 0.027801 seconds and 4 git commands to generate.