[lttng-modules.git] / ltt-tracer.h

/*
 * Copyright (C) 2005,2006,2008 Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca)
 *
 * This contains the definitions for the Linux Trace Toolkit tracer.
 *
 * Dual LGPL v2.1/GPL v2 license.
 */

#ifndef _LTT_TRACER_H
#define _LTT_TRACER_H

#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
/* Align data on its natural alignment */
#define RING_BUFFER_ALIGN
#endif

#include <linux/ringbuffer/config.h>

#include <stdarg.h>
#include <linux/types.h>
#include <linux/limits.h>
#include <linux/list.h>
#include <linux/cache.h>
#include <linux/kernel.h>
#include <linux/timex.h>
#include <linux/wait.h>
#include <linux/marker.h>
#include <linux/trace-clock.h>
#include <asm/atomic.h>
#include <asm/local.h>

#include "ltt-tracer-core.h"
#include "ltt-channels.h"

/* Number of bytes to log with a read/write event */
#define LTT_LOG_RW_SIZE			32L

struct ltt_active_marker;

/* Maximum number of callbacks per marker */
#define LTT_NR_CALLBACKS	10

struct ltt_serialize_closure {
	ltt_serialize_cb *callbacks;
	long cb_args[LTT_NR_CALLBACKS];
	unsigned int cb_idx;
};

size_t ltt_serialize_data(struct ltt_chanbuf *buf, size_t buf_offset,
			  struct ltt_serialize_closure *closure,
			  void *serialize_private, unsigned int stack_pos_ctx,
			  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 */
};

enum ltt_channels {
	LTT_CHANNEL_METADATA,
	LTT_CHANNEL_FD_STATE,
	LTT_CHANNEL_GLOBAL_STATE,
	LTT_CHANNEL_IRQ_STATE,
	LTT_CHANNEL_MODULE_STATE,
	LTT_CHANNEL_NETIF_STATE,
	LTT_CHANNEL_SOFTIRQ_STATE,
	LTT_CHANNEL_SWAP_STATE,
	LTT_CHANNEL_SYSCALL_STATE,
	LTT_CHANNEL_TASK_STATE,
	LTT_CHANNEL_VM_STATE,
	LTT_CHANNEL_FS,
	LTT_CHANNEL_INPUT,
	LTT_CHANNEL_IPC,
	LTT_CHANNEL_KERNEL,
	LTT_CHANNEL_MM,
	LTT_CHANNEL_RCU,
	LTT_CHANNEL_DEFAULT,
};

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;
};

extern void ltt_vtrace(const struct marker *mdata, void *probe_data,
		       void *call_data, const char *fmt, va_list *args);
extern void ltt_trace(const struct marker *mdata, void *probe_data,
		      void *call_data, const char *fmt, ...);

size_t ltt_serialize_printf(struct ltt_chanbuf *buf, unsigned long buf_offset,
			    size_t *msg_size, char *output, size_t outlen,
			    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;
};

enum trace_mode { LTT_TRACE_NORMAL, LTT_TRACE_FLIGHT, LTT_TRACE_HYBRID };

#define CHANNEL_FLAG_ENABLE	(1U<<0)
#define CHANNEL_FLAG_OVERWRITE	(1U<<1)

#if 0
/* Per-trace information - each trace/flight recorder represented by one */
struct ltt_trace {
	/* First 32 bytes cache-hot cacheline */
	struct list_head list;
	struct ltt_chan **channels;
	unsigned int nr_channels;
	int active;
	/* Second 32 bytes cache-hot cacheline */
	struct ltt_trace_ops *ops;
	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;
		struct dentry			*ascii_root;
	} dentry;
	struct kref kref; /* Each channel has a kref of the trace struct */
	struct ltt_transport *transport;
	struct kref ltt_transport_kref;
	wait_queue_head_t kref_wq; /* Place for ltt_trace_destroy to sleep */
	char trace_name[NAME_MAX];
} ____cacheline_aligned;
#endif //0

/*
 * 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 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)

#define LTT_MAX_SMALL_SIZE		0xFFFFU

/*
 * 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 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 data_size;		/* Size of data in subbuffer */
	uint32_t sb_size;		/* Subbuffer size (include padding) */
	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 */
};

static inline notrace u64 lib_ring_buffer_clock_read(struct channel *chan)
{
	return trace_clock_read64();
}

/*
 * record_header_size - Calculate the header size and padding necessary.
 * @config: ring buffer instance configuration
 * @chan: channel
 * @offset: offset in the write buffer
 * @data_size: size of the payload
 * @pre_header_padding: padding to add before the header (output)
 * @rflags: reservation flags
 * @ctx: reservation context
 *
 * Returns the event header size (including padding).
 *
 * 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 == LTT_MAX_SMALL_SIZE, 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 record_header_size(const struct lib_ring_buffer_config *config,
				 struct channel *chan, size_t offset,
				 size_t data_size, size_t *pre_header_padding,
				 unsigned int rflags,
				 struct lib_ring_buffer_ctx *ctx)
{
	size_t orig_offset = offset;
	size_t padding;

	BUILD_BUG_ON(sizeof(struct event_header) != sizeof(u32));

	padding = lib_ring_buffer_align(config, offset,
					sizeof(struct event_header));
	offset += padding;
	offset += sizeof(struct event_header);

	if (unlikely(rflags)) {
		switch (rflags) {
		case LTT_RFLAG_ID_SIZE_TSC:
			offset += sizeof(u16) + sizeof(u16);
			if (data_size >= LTT_MAX_SMALL_SIZE)
				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 >= LTT_MAX_SMALL_SIZE)
				offset += sizeof(u32);
			break;
		case LTT_RFLAG_ID:
			offset += sizeof(u16);
			break;
		}
	}

	*pre_header_padding = padding;
	return offset - orig_offset;
}

#include <linux/ringbuffer/api.h>

extern
size_t ltt_write_event_header_slow(struct ltt_chanbuf_alloc *bufa,
				   struct ltt_chan_alloc *chana,
				   long buf_offset, u16 eID, u32 event_size,
				   u64 tsc, unsigned int rflags);

/*
 * ltt_write_event_header
 *
 * Writes the event header to the offset (already aligned on 32-bits).
 *
 * @config: ring buffer instance configuration
 * @ctx: reservation context
 * @eID : event ID
 * @event_size : size of the event, excluding the event header.
 */
static __inline__
void ltt_write_event_header(const struct lib_ring_buffer_config *config,
			    struct lib_ring_buffer_ctx *ctx,
			    u16 eID, u32 event_size)
{
	struct event_header header;

	if (unlikely(ctx->rflags))
		goto slow_path;

	header.id_time = eID << LTT_TSC_BITS;
	header.id_time |= (u32)ctx->tsc & LTT_TSC_MASK;
	lib_ring_buffer_write(config, ctx, &header, sizeof(header));

slow_path:
	ltt_write_event_header_slow(config, ctx, eID, event_size);
}

#if 0
/*
 * ltt_read_event_header
 * buf_offset must aligned on 32 bits
 */
static __inline__
size_t ltt_read_event_header(struct ltt_chanbuf_alloc *bufa, long buf_offset,
			     u64 *tsc, u32 *event_size, u16 *eID,
			     unsigned int *rflags)
{
	struct ltt_event_header header;
	u16 small_size;

	ltt_relay_read(bufa, buf_offset, &header, sizeof(header));
	buf_offset += sizeof(header);

	*event_size = INT_MAX;
	*eID = header.id_time >> LTT_TSC_BITS;
	*tsc = header.id_time & LTT_TSC_MASK;

	switch (*eID) {
	case 29:
		*rflags = LTT_RFLAG_ID_SIZE_TSC;
		ltt_relay_read(bufa, buf_offset, eID, sizeof(u16));
		buf_offset += sizeof(u16);
		ltt_relay_read(bufa, buf_offset, &small_size, sizeof(u16));
		buf_offset += sizeof(u16);
		if (small_size == LTT_MAX_SMALL_SIZE) {
			ltt_relay_read(bufa, buf_offset, event_size,
					sizeof(u32));
			buf_offset += sizeof(u32);
		} else
			*event_size = small_size;
		buf_offset += ltt_align(buf_offset, sizeof(u64));
		ltt_relay_read(bufa, buf_offset, tsc, sizeof(u64));
		buf_offset += sizeof(u64);
		break;
	case 30:
		*rflags = LTT_RFLAG_ID_SIZE;
		ltt_relay_read(bufa, buf_offset, eID, sizeof(u16));
		buf_offset += sizeof(u16);
		ltt_relay_read(bufa, buf_offset, &small_size, sizeof(u16));
		buf_offset += sizeof(u16);
		if (small_size == LTT_MAX_SMALL_SIZE) {
			ltt_relay_read(bufa, buf_offset, event_size,
					sizeof(u32));
			buf_offset += sizeof(u32);
		} else
			*event_size = small_size;
		break;
	case 31:
		*rflags = LTT_RFLAG_ID;
		ltt_relay_read(bufa, buf_offset, eID, sizeof(u16));
		buf_offset += sizeof(u16);
		break;
	default:
		*rflags = 0;
		break;
	}

	return buf_offset;
}
#endif //0

/*
 * Control channels :
 * control/metadata
 * control/interrupts
 * control/...
 *
 * cpu channel :
 * cpu
 */
#define LTT_RELAY_ROOT			"ltt"

#define LTT_METADATA_CHANNEL		"metadata_state"
#define LTT_FD_STATE_CHANNEL		"fd_state"
#define LTT_GLOBAL_STATE_CHANNEL	"global_state"
#define LTT_IRQ_STATE_CHANNEL		"irq_state"
#define LTT_MODULE_STATE_CHANNEL	"module_state"
#define LTT_NETIF_STATE_CHANNEL		"netif_state"
#define LTT_SOFTIRQ_STATE_CHANNEL	"softirq_state"
#define LTT_SWAP_STATE_CHANNEL		"swap_state"
#define LTT_SYSCALL_STATE_CHANNEL	"syscall_state"
#define LTT_TASK_STATE_CHANNEL		"task_state"
#define LTT_VM_STATE_CHANNEL		"vm_state"
#define LTT_FS_CHANNEL			"fs"
#define LTT_INPUT_CHANNEL		"input"
#define LTT_IPC_CHANNEL			"ipc"
#define LTT_KERNEL_CHANNEL		"kernel"
#define LTT_MM_CHANNEL			"mm"
#define LTT_RCU_CHANNEL			"rcu"

#define LTT_FLIGHT_PREFIX		"flight-"

#define LTT_ASCII			"ascii"

/* Tracer properties */
#define LTT_DEFAULT_SUBBUF_SIZE_LOW	65536
#define LTT_DEFAULT_N_SUBBUFS_LOW	2
#define LTT_DEFAULT_SUBBUF_SIZE_MED	262144
#define LTT_DEFAULT_N_SUBBUFS_MED	2
#define LTT_DEFAULT_SUBBUF_SIZE_HIGH	1048576
#define LTT_DEFAULT_N_SUBBUFS_HIGH	2
#define LTT_TRACER_MAGIC_NUMBER		0x00D6B7ED
#define LTT_TRACER_VERSION_MAJOR	2
#define LTT_TRACER_VERSION_MINOR	6

/**
 * ltt_write_trace_header - Write trace header
 * @priv: Private data (struct trace)
 * @header: Memory address where the information must be written to
 */
static __inline__
void ltt_write_trace_header(void *priv,
			    struct subbuffer_header *header)
{
	struct ltt_trace *trace = priv;

	header->magic_number = LTT_TRACER_MAGIC_NUMBER;
	header->major_version = LTT_TRACER_VERSION_MAJOR;
	header->minor_version = LTT_TRACER_VERSION_MINOR;
	header->arch_size = sizeof(void *);
	header->alignment = lib_ring_buffer_get_alignment();
	header->start_time_sec = trace->start_time.tv_sec;
	header->start_time_usec = trace->start_time.tv_usec;
	header->start_freq = trace->start_freq;
	header->freq_scale = trace->freq_scale;
}

/*
 * 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 int ltt_module_register(enum ltt_module_function name, void *function,
			       struct module *owner);
extern void ltt_module_unregister(enum ltt_module_function name);

/* Exported control function */

enum ltt_control_msg {
	LTT_CONTROL_START,
	LTT_CONTROL_STOP,
	LTT_CONTROL_CREATE_TRACE,
	LTT_CONTROL_DESTROY_TRACE
};

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;
};

void ltt_core_register(int (*function)(u8, void *));

void ltt_core_unregister(void);

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 *trace);

extern
void ltt_statedump_register_kprobes_dump(void (*callback)(void *call_data));
extern
void ltt_statedump_unregister_kprobes_dump(void (*callback)(void *call_data));

extern void ltt_dump_softirq_vec(void *call_data);

#ifdef CONFIG_HAVE_LTT_DUMP_TABLES
extern void ltt_dump_sys_call_table(void *call_data);
extern void ltt_dump_idt_table(void *call_data);
#else
static inline void ltt_dump_sys_call_table(void *call_data)
{
}

static inline void ltt_dump_idt_table(void *call_data)
{
}
#endif

#endif /* _LTT_TRACER_H */
Commit	Line	Data
1c8284eb MD	1	/*
	2	* Copyright (C) 2005,2006,2008 Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca)
	3	*
	4	* This contains the definitions for the Linux Trace Toolkit tracer.
	5	*
	6	* Dual LGPL v2.1/GPL v2 license.
	7	*/
	8
	9	#ifndef _LTT_TRACER_H
	10	#define _LTT_TRACER_H
	11
7514523f MD	12	#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
	13	/* Align data on its natural alignment */
	14	#define RING_BUFFER_ALIGN
	15	#endif
	16
	17	#include <linux/ringbuffer/config.h>
	18
1c8284eb MD	19	#include <stdarg.h>
	20	#include <linux/types.h>
	21	#include <linux/limits.h>
	22	#include <linux/list.h>
	23	#include <linux/cache.h>
	24	#include <linux/kernel.h>
	25	#include <linux/timex.h>
	26	#include <linux/wait.h>
	27	#include <linux/marker.h>
	28	#include <linux/trace-clock.h>
1c8284eb MD	29	#include <asm/atomic.h>
	30	#include <asm/local.h>
	31
	32	#include "ltt-tracer-core.h"
7514523f	33	#include "ltt-channels.h"
1c8284eb MD	34
	35	/* Number of bytes to log with a read/write event */
	36	#define LTT_LOG_RW_SIZE 32L
	37
1c8284eb MD	38	struct ltt_active_marker;
	39
	40	/* Maximum number of callbacks per marker */
	41	#define LTT_NR_CALLBACKS 10
	42
	43	struct ltt_serialize_closure {
	44	ltt_serialize_cb *callbacks;
	45	long cb_args[LTT_NR_CALLBACKS];
	46	unsigned int cb_idx;
	47	};
	48
	49	size_t ltt_serialize_data(struct ltt_chanbuf *buf, size_t buf_offset,
	50	struct ltt_serialize_closure *closure,
	51	void *serialize_private, unsigned int stack_pos_ctx,
	52	int largest_align, const char fmt, va_list *args);
	53
	54	struct ltt_available_probe {
	55	const char name; / probe name */
	56	const char *format;
	57	marker_probe_func *probe_func;
	58	ltt_serialize_cb callbacks[LTT_NR_CALLBACKS];
	59	struct list_head node; /* registered probes list */
	60	};
	61
	62	enum ltt_channels {
	63	LTT_CHANNEL_METADATA,
	64	LTT_CHANNEL_FD_STATE,
	65	LTT_CHANNEL_GLOBAL_STATE,
	66	LTT_CHANNEL_IRQ_STATE,
	67	LTT_CHANNEL_MODULE_STATE,
	68	LTT_CHANNEL_NETIF_STATE,
	69	LTT_CHANNEL_SOFTIRQ_STATE,
	70	LTT_CHANNEL_SWAP_STATE,
	71	LTT_CHANNEL_SYSCALL_STATE,
	72	LTT_CHANNEL_TASK_STATE,
	73	LTT_CHANNEL_VM_STATE,
	74	LTT_CHANNEL_FS,
	75	LTT_CHANNEL_INPUT,
	76	LTT_CHANNEL_IPC,
	77	LTT_CHANNEL_KERNEL,
	78	LTT_CHANNEL_MM,
	79	LTT_CHANNEL_RCU,
	80	LTT_CHANNEL_DEFAULT,
	81	};
	82
	83	struct ltt_active_marker {
	84	struct list_head node; /* active markers list */
	85	const char *channel;
	86	const char *name;
	87	const char *format;
	88	struct ltt_available_probe *probe;
	89	};
	90
	91	extern void ltt_vtrace(const struct marker mdata, void probe_data,
	92	void call_data, const char fmt, va_list *args);
	93	extern void ltt_trace(const struct marker mdata, void probe_data,
	94	void call_data, const char fmt, ...);
	95
	96	size_t ltt_serialize_printf(struct ltt_chanbuf *buf, unsigned long buf_offset,
	97	size_t msg_size, char output, size_t outlen,
	98	const char *fmt);
	99
	100	/*
	101	* Unique ID assigned to each registered probe.
102	*/
103	enum marker_id {
104	MARKER_ID_SET_MARKER_ID = 0, /* Static IDs available (range 0-7) */
105	MARKER_ID_SET_MARKER_FORMAT,
106	MARKER_ID_COMPACT, /* Compact IDs (range: 8-127) */
107	MARKER_ID_DYNAMIC, /* Dynamic IDs (range: 128-65535) */
108	};
109
110	/* static ids 0-1 reserved for internal use. */
111	#define MARKER_CORE_IDS 2
112	static __inline__ enum marker_id marker_id_type(uint16_t id)
113	{
114	if (id < MARKER_CORE_IDS)
115	return (enum marker_id)id;
116	else
117	return MARKER_ID_DYNAMIC;
118	}
119
120	struct user_dbg_data {
121	unsigned long avail_size;
122	unsigned long write;
123	unsigned long read;
124	};
125
1c8284eb MD	126	enum trace_mode { LTT_TRACE_NORMAL, LTT_TRACE_FLIGHT, LTT_TRACE_HYBRID };
	127
	128	#define CHANNEL_FLAG_ENABLE (1U<<0)
	129	#define CHANNEL_FLAG_OVERWRITE (1U<<1)
	130
c0e31d2e	131	#if 0
1c8284eb MD	132	/* Per-trace information - each trace/flight recorder represented by one */
	133	struct ltt_trace {
	134	/* First 32 bytes cache-hot cacheline */
	135	struct list_head list;
7514523f	136	struct ltt_chan **channels;
1c8284eb MD	137	unsigned int nr_channels;
	138	int active;
	139	/* Second 32 bytes cache-hot cacheline */
	140	struct ltt_trace_ops *ops;
	141	u32 freq_scale;
	142	u64 start_freq;
	143	u64 start_tsc;
	144	unsigned long long start_monotonic;
	145	struct timeval start_time;
	146	struct ltt_channel_setting *settings;
	147	struct {
	148	struct dentry *trace_root;
	149	struct dentry *ascii_root;
	150	} dentry;
	151	struct kref kref; /* Each channel has a kref of the trace struct */
	152	struct ltt_transport *transport;
	153	struct kref ltt_transport_kref;
	154	wait_queue_head_t kref_wq; /* Place for ltt_trace_destroy to sleep */
	155	char trace_name[NAME_MAX];
	156	} ____cacheline_aligned;
c0e31d2e	157	#endif //0
1c8284eb	158
7514523f MD	159	/*
7514523f MD	160	* Hardcoded event headers
1c8284eb MD	161	*
	162	* event header for a trace with active heartbeat : 27 bits timestamps
	163	*
	164	* headers are 32-bits aligned. In order to insure such alignment, a dynamic per
	165	* trace alignment value must be done.
	166	*
	167	* Remember that the C compiler does align each member on the boundary
	168	* equivalent to their own size.
	169	*
	170	* As relay subbuffers are aligned on pages, we are sure that they are 4 and 8
	171	* bytes aligned, so the buffer header and trace header are aligned.
	172	*
	173	* Event headers are aligned depending on the trace alignment option.
	174	*
	175	* Note using C structure bitfields for cross-endianness and portability
	176	* concerns.
	177	*/
	178
	179	#define LTT_RESERVED_EVENTS 3
	180	#define LTT_EVENT_BITS 5
	181	#define LTT_FREE_EVENTS ((1 << LTT_EVENT_BITS) - LTT_RESERVED_EVENTS)
	182	#define LTT_TSC_BITS 27
	183	#define LTT_TSC_MASK ((1 << LTT_TSC_BITS) - 1)
	184
7514523f	185	struct event_header {
1c8284eb MD	186	u32 id_time; /* 5 bits event id (MSB); 27 bits time (LSB) */
	187	};
	188
	189	/* Reservation flags */
	190	#define LTT_RFLAG_ID (1 << 0)
	191	#define LTT_RFLAG_ID_SIZE (1 << 1)
	192	#define LTT_RFLAG_ID_SIZE_TSC (1 << 2)
	193
	194	#define LTT_MAX_SMALL_SIZE 0xFFFFU
	195
	196	/*
	197	* We use asm/timex.h : cpu_khz/HZ variable in here : we might have to deal
	198	* specifically with CPU frequency scaling someday, so using an interpolation
	199	* between the start and end of buffer values is not flexible enough. Using an
	200	* immediate frequency value permits to calculate directly the times for parts
	201	* of a buffer that would be before a frequency change.
	202	*
	203	* Keep the natural field alignment for _each field_ within this structure if
	204	* you ever add/remove a field from this header. Packed attribute is not used
	205	* because gcc generates poor code on at least powerpc and mips. Don't ever
	206	* let gcc add padding between the structure elements.
	207	*/
7514523f	208	struct subbuffer_header {
1c8284eb MD	209	uint64_t cycle_count_begin; /* Cycle count at subbuffer start */
	210	uint64_t cycle_count_end; /* Cycle count at subbuffer end */
	211	uint32_t magic_number; /*
	212	* Trace magic number.
	213	* contains endianness information.
	214	*/
	215	uint8_t major_version;
	216	uint8_t minor_version;
	217	uint8_t arch_size; /* Architecture pointer size */
	218	uint8_t alignment; /* LTT data alignment */
	219	uint64_t start_time_sec; /* NTP-corrected start time */
	220	uint64_t start_time_usec;
	221	uint64_t start_freq; /*
	222	* Frequency at trace start,
	223	* used all along the trace.
	224	*/
	225	uint32_t freq_scale; /* Frequency scaling (divisor) */
	226	uint32_t data_size; /* Size of data in subbuffer */
	227	uint32_t sb_size; /* Subbuffer size (include padding) */
	228	uint32_t events_lost; /*
	229	* Events lost in this subbuffer since
	230	* the beginning of the trace.
	231	* (may overflow)
	232	*/
	233	uint32_t subbuf_corrupt; /*
	234	* Corrupted (lost) subbuffers since
	235	* the begginig of the trace.
	236	* (may overflow)
	237	*/
	238	uint8_t header_end[0]; /* End of header */
	239	};
	240
7514523f	241	static inline notrace u64 lib_ring_buffer_clock_read(struct channel *chan)
1c8284eb	242	{
7514523f	243	return trace_clock_read64();
1c8284eb MD	244	}
	245
	246	/*
7514523f MD	247	* record_header_size - Calculate the header size and padding necessary.
	248	* @config: ring buffer instance configuration
	249	* @chan: channel
	250	* @offset: offset in the write buffer
	251	* @data_size: size of the payload
	252	* @pre_header_padding: padding to add before the header (output)
	253	* @rflags: reservation flags
	254	* @ctx: reservation context
1c8284eb	255	*
7514523f	256	* Returns the event header size (including padding).
1c8284eb MD	257	*
	258	* Important note :
	259	* The event header must be 32-bits. The total offset calculated here :
	260	*
	261	* Alignment of header struct on 32 bits (min arch size, header size)
	262	* + sizeof(header struct) (32-bits)
	263	* + (opt) u16 (ext. event id)
	264	* + (opt) u16 (event_size)
	265	* (if event_size == LTT_MAX_SMALL_SIZE, has ext. event size)
	266	* + (opt) u32 (ext. event size)
	267	* + (opt) u64 full TSC (aligned on min(64-bits, arch size))
	268	*
	269	* The payload must itself determine its own alignment from the biggest type it
	270	* contains.
7514523f	271	*/
1c8284eb	272	static __inline__
7514523f MD	273	unsigned char record_header_size(const struct lib_ring_buffer_config *config,
	274	struct channel *chan, size_t offset,
	275	size_t data_size, size_t *pre_header_padding,
	276	unsigned int rflags,
	277	struct lib_ring_buffer_ctx *ctx)
1c8284eb MD	278	{
	279	size_t orig_offset = offset;
	280	size_t padding;
	281
7514523f	282	BUILD_BUG_ON(sizeof(struct event_header) != sizeof(u32));
1c8284eb	283
7514523f MD	284	padding = lib_ring_buffer_align(config, offset,
7514523f MD	285	sizeof(struct event_header));
1c8284eb	286	offset += padding;
7514523f	287	offset += sizeof(struct event_header);
1c8284eb MD	288
	289	if (unlikely(rflags)) {
	290	switch (rflags) {
	291	case LTT_RFLAG_ID_SIZE_TSC:
	292	offset += sizeof(u16) + sizeof(u16);
	293	if (data_size >= LTT_MAX_SMALL_SIZE)
	294	offset += sizeof(u32);
	295	offset += ltt_align(offset, sizeof(u64));
	296	offset += sizeof(u64);
	297	break;
	298	case LTT_RFLAG_ID_SIZE:
	299	offset += sizeof(u16) + sizeof(u16);
	300	if (data_size >= LTT_MAX_SMALL_SIZE)
	301	offset += sizeof(u32);
	302	break;
	303	case LTT_RFLAG_ID:
	304	offset += sizeof(u16);
	305	break;
	306	}
	307	}
	308
7514523f	309	*pre_header_padding = padding;
1c8284eb MD	310	return offset - orig_offset;
	311	}
	312
7514523f MD	313	#include <linux/ringbuffer/api.h>
7514523f MD	314
1c8284eb MD	315	extern
	316	size_t ltt_write_event_header_slow(struct ltt_chanbuf_alloc *bufa,
	317	struct ltt_chan_alloc *chana,
	318	long buf_offset, u16 eID, u32 event_size,
	319	u64 tsc, unsigned int rflags);
	320
	321	/*
	322	* ltt_write_event_header
	323	*
	324	* Writes the event header to the offset (already aligned on 32-bits).
	325	*
7514523f MD	326	* @config: ring buffer instance configuration
7514523f MD	327	* @ctx: reservation context
1c8284eb MD	328	* @eID : event ID
1c8284eb MD	329	* @event_size : size of the event, excluding the event header.
1c8284eb MD	330	*/
1c8284eb MD	331	static __inline__
7514523f MD	332	void ltt_write_event_header(const struct lib_ring_buffer_config *config,
	333	struct lib_ring_buffer_ctx *ctx,
	334	u16 eID, u32 event_size)
1c8284eb	335	{
7514523f	336	struct event_header header;
1c8284eb	337
7514523f	338	if (unlikely(ctx->rflags))
1c8284eb MD	339	goto slow_path;
	340
	341	header.id_time = eID << LTT_TSC_BITS;
7514523f MD	342	header.id_time \|= (u32)ctx->tsc & LTT_TSC_MASK;
7514523f MD	343	lib_ring_buffer_write(config, ctx, &header, sizeof(header));
1c8284eb MD	344
1c8284eb MD	345	slow_path:
7514523f	346	ltt_write_event_header_slow(config, ctx, eID, event_size);
1c8284eb MD	347	}
1c8284eb MD	348
7514523f	349	#if 0
1c8284eb MD	350	/*
	351	* ltt_read_event_header
	352	* buf_offset must aligned on 32 bits
	353	*/
	354	static __inline__
	355	size_t ltt_read_event_header(struct ltt_chanbuf_alloc *bufa, long buf_offset,
	356	u64 tsc, u32 event_size, u16 *eID,
	357	unsigned int *rflags)
	358	{
	359	struct ltt_event_header header;
	360	u16 small_size;
	361
	362	ltt_relay_read(bufa, buf_offset, &header, sizeof(header));
	363	buf_offset += sizeof(header);
	364
	365	*event_size = INT_MAX;
	366	*eID = header.id_time >> LTT_TSC_BITS;
	367	*tsc = header.id_time & LTT_TSC_MASK;
	368
	369	switch (*eID) {
	370	case 29:
	371	*rflags = LTT_RFLAG_ID_SIZE_TSC;
	372	ltt_relay_read(bufa, buf_offset, eID, sizeof(u16));
	373	buf_offset += sizeof(u16);
	374	ltt_relay_read(bufa, buf_offset, &small_size, sizeof(u16));
	375	buf_offset += sizeof(u16);
	376	if (small_size == LTT_MAX_SMALL_SIZE) {
	377	ltt_relay_read(bufa, buf_offset, event_size,
	378	sizeof(u32));
	379	buf_offset += sizeof(u32);
	380	} else
	381	*event_size = small_size;
	382	buf_offset += ltt_align(buf_offset, sizeof(u64));
	383	ltt_relay_read(bufa, buf_offset, tsc, sizeof(u64));
	384	buf_offset += sizeof(u64);
	385	break;
	386	case 30:
	387	*rflags = LTT_RFLAG_ID_SIZE;
	388	ltt_relay_read(bufa, buf_offset, eID, sizeof(u16));
	389	buf_offset += sizeof(u16);
	390	ltt_relay_read(bufa, buf_offset, &small_size, sizeof(u16));
	391	buf_offset += sizeof(u16);
	392	if (small_size == LTT_MAX_SMALL_SIZE) {
	393	ltt_relay_read(bufa, buf_offset, event_size,
	394	sizeof(u32));
	395	buf_offset += sizeof(u32);
	396	} else
	397	*event_size = small_size;
	398	break;
	399	case 31:
	400	*rflags = LTT_RFLAG_ID;
	401	ltt_relay_read(bufa, buf_offset, eID, sizeof(u16));
	402	buf_offset += sizeof(u16);
	403	break;
	404	default:
	405	*rflags = 0;
	406	break;
	407	}
	408
	409	return buf_offset;
	410	}
7514523f	411	#endif //0
1c8284eb MD	412
	413	/*
	414	* Control channels :
	415	* control/metadata
	416	* control/interrupts
	417	* control/...
	418	*
	419	* cpu channel :
	420	* cpu
	421	*/
	422	#define LTT_RELAY_ROOT "ltt"
1c8284eb MD	423
	424	#define LTT_METADATA_CHANNEL "metadata_state"
	425	#define LTT_FD_STATE_CHANNEL "fd_state"
	426	#define LTT_GLOBAL_STATE_CHANNEL "global_state"
	427	#define LTT_IRQ_STATE_CHANNEL "irq_state"
	428	#define LTT_MODULE_STATE_CHANNEL "module_state"
	429	#define LTT_NETIF_STATE_CHANNEL "netif_state"
	430	#define LTT_SOFTIRQ_STATE_CHANNEL "softirq_state"
	431	#define LTT_SWAP_STATE_CHANNEL "swap_state"
	432	#define LTT_SYSCALL_STATE_CHANNEL "syscall_state"
	433	#define LTT_TASK_STATE_CHANNEL "task_state"
	434	#define LTT_VM_STATE_CHANNEL "vm_state"
	435	#define LTT_FS_CHANNEL "fs"
	436	#define LTT_INPUT_CHANNEL "input"
	437	#define LTT_IPC_CHANNEL "ipc"
	438	#define LTT_KERNEL_CHANNEL "kernel"
	439	#define LTT_MM_CHANNEL "mm"
	440	#define LTT_RCU_CHANNEL "rcu"
	441
	442	#define LTT_FLIGHT_PREFIX "flight-"
	443
	444	#define LTT_ASCII "ascii"
	445
	446	/* Tracer properties */
	447	#define LTT_DEFAULT_SUBBUF_SIZE_LOW 65536
	448	#define LTT_DEFAULT_N_SUBBUFS_LOW 2
	449	#define LTT_DEFAULT_SUBBUF_SIZE_MED 262144
	450	#define LTT_DEFAULT_N_SUBBUFS_MED 2
	451	#define LTT_DEFAULT_SUBBUF_SIZE_HIGH 1048576
	452	#define LTT_DEFAULT_N_SUBBUFS_HIGH 2
	453	#define LTT_TRACER_MAGIC_NUMBER 0x00D6B7ED
	454	#define LTT_TRACER_VERSION_MAJOR 2
	455	#define LTT_TRACER_VERSION_MINOR 6
	456
	457	/**
	458	* ltt_write_trace_header - Write trace header
7514523f	459	* @priv: Private data (struct trace)
1c8284eb MD	460	* @header: Memory address where the information must be written to
	461	*/
	462	static __inline__
7514523f MD	463	void ltt_write_trace_header(void *priv,
7514523f MD	464	struct subbuffer_header *header)
1c8284eb	465	{
7514523f MD	466	struct ltt_trace *trace = priv;
7514523f MD	467
1c8284eb MD	468	header->magic_number = LTT_TRACER_MAGIC_NUMBER;
	469	header->major_version = LTT_TRACER_VERSION_MAJOR;
	470	header->minor_version = LTT_TRACER_VERSION_MINOR;
	471	header->arch_size = sizeof(void *);
7514523f	472	header->alignment = lib_ring_buffer_get_alignment();
1c8284eb MD	473	header->start_time_sec = trace->start_time.tv_sec;
	474	header->start_time_usec = trace->start_time.tv_usec;
	475	header->start_freq = trace->start_freq;
	476	header->freq_scale = trace->freq_scale;
	477	}
	478
	479	/*
	480	* Size reserved for high priority events (interrupts, NMI, BH) at the end of a
	481	* nearly full buffer. User space won't use this last amount of space when in
	482	* blocking mode. This space also includes the event header that would be
	483	* written by this user space event.
	484	*/
	485	#define LTT_RESERVE_CRITICAL 4096
	486
	487	/* Register and unregister function pointers */
	488
	489	enum ltt_module_function {
	490	LTT_FUNCTION_RUN_FILTER,
	491	LTT_FUNCTION_FILTER_CONTROL,
	492	LTT_FUNCTION_STATEDUMP
	493	};
	494
	495	extern int ltt_module_register(enum ltt_module_function name, void *function,
	496	struct module *owner);
	497	extern void ltt_module_unregister(enum ltt_module_function name);
	498
1c8284eb MD	499	/* Exported control function */
	500
	501	enum ltt_control_msg {
	502	LTT_CONTROL_START,
	503	LTT_CONTROL_STOP,
	504	LTT_CONTROL_CREATE_TRACE,
	505	LTT_CONTROL_DESTROY_TRACE
	506	};
	507
	508	union ltt_control_args {
	509	struct {
	510	enum trace_mode mode;
	511	unsigned int subbuf_size_low;
	512	unsigned int n_subbufs_low;
	513	unsigned int subbuf_size_med;
	514	unsigned int n_subbufs_med;
	515	unsigned int subbuf_size_high;
	516	unsigned int n_subbufs_high;
	517	} new_trace;
	518	};
	519
1c8284eb MD	520	void ltt_core_register(int (function)(u8, void ));
	521
	522	void ltt_core_unregister(void);
	523
1c8284eb MD	524	extern int ltt_probe_register(struct ltt_available_probe *pdata);
	525	extern int ltt_probe_unregister(struct ltt_available_probe *pdata);
	526	extern int ltt_marker_connect(const char channel, const char mname,
	527	const char *pname);
	528	extern int ltt_marker_disconnect(const char channel, const char mname,
	529	const char *pname);
	530	extern void ltt_dump_marker_state(struct ltt_trace *trace);
	531
1c8284eb MD	532	extern
	533	void ltt_statedump_register_kprobes_dump(void (callback)(void call_data));
	534	extern
	535	void ltt_statedump_unregister_kprobes_dump(void (callback)(void call_data));
	536
	537	extern void ltt_dump_softirq_vec(void *call_data);
	538
f277b4c1	539	#ifdef CONFIG_HAVE_LTT_DUMP_TABLES
1c8284eb MD	540	extern void ltt_dump_sys_call_table(void *call_data);
1c8284eb MD	541	extern void ltt_dump_idt_table(void *call_data);
f277b4c1 MD	542	#else
	543	static inline void ltt_dump_sys_call_table(void *call_data)
	544	{
	545	}
	546
	547	static inline void ltt_dump_idt_table(void *call_data)
	548	{
	549	}
	550	#endif
1c8284eb	551
1c8284eb	552	#endif /* _LTT_TRACER_H */