#include <errno.h>
#include <signal.h>
#include <locale.h>
-#include <limits.h>
-#include <ctype.h>
+#include <libgen.h>
#include "blktrace.h"
#include "rbtree.h"
+#include "jhash.h"
-static char blkparse_version[] = "0.90";
+static char blkparse_version[] = "0.99";
-#define SECONDS(x) ((unsigned long long)(x) / 1000000000)
-#define NANO_SECONDS(x) ((unsigned long long)(x) % 1000000000)
-#define DOUBLE_TO_NANO_ULL(d) ((unsigned long long)((d) * 1000000000))
-
-#define MINORBITS 20
-#define MINORMASK ((1U << MINORBITS) - 1)
-#define MAJOR(dev) ((unsigned int) ((dev) >> MINORBITS))
-#define MINOR(dev) ((unsigned int) ((dev) & MINORMASK))
-
-#define min(a, b) ((a) < (b) ? (a) : (b))
-
-#define HEADER "%D %2c %8s %5T.%9t %5p %2a %3d "
-
-struct io_stats {
- unsigned long qreads, qwrites, creads, cwrites, mreads, mwrites;
- unsigned long ireads, iwrites;
- unsigned long long qread_kb, qwrite_kb, cread_kb, cwrite_kb;
- unsigned long long iread_kb, iwrite_kb;
- unsigned long io_unplugs, timer_unplugs;
-};
-
-struct per_cpu_info {
- int cpu;
- int nelems;
-
- int fd;
- char fname[128];
-
- struct io_stats io_stats;
+struct skip_info {
+ unsigned long start, end;
+ struct skip_info *prev, *next;
};
struct per_dev_info {
- dev_t id;
+ dev_t dev;
char *name;
int backwards;
unsigned long long events;
+ unsigned long long first_reported_time;
unsigned long long last_reported_time;
+ unsigned long long last_read_time;
struct io_stats io_stats;
- unsigned long last_sequence;
unsigned long skips;
+ unsigned long long seq_skips;
+ unsigned int max_depth[2];
+ unsigned int cur_depth[2];
+ struct rb_root rb_track;
+
+ int nfiles;
int ncpus;
+
+ unsigned long *cpu_map;
+ unsigned int cpu_map_max;
+
struct per_cpu_info *cpus;
};
+/*
+ * some duplicated effort here, we can unify this hash and the ppi hash later
+ */
+struct process_pid_map {
+ pid_t pid;
+ char comm[16];
+ struct process_pid_map *hash_next, *list_next;
+};
+
+#define PPM_HASH_SHIFT (8)
+#define PPM_HASH_SIZE (1 << PPM_HASH_SHIFT)
+#define PPM_HASH_MASK (PPM_HASH_SIZE - 1)
+static struct process_pid_map *ppm_hash_table[PPM_HASH_SIZE];
+
struct per_process_info {
- char name[16];
- __u32 pid;
+ struct process_pid_map *ppm;
struct io_stats io_stats;
struct per_process_info *hash_next, *list_next;
+ int more_than_one;
/*
* individual io stats
};
#define PPI_HASH_SHIFT (8)
-static struct per_process_info *ppi_hash[1 << PPI_HASH_SHIFT];
+#define PPI_HASH_SIZE (1 << PPI_HASH_SHIFT)
+#define PPI_HASH_MASK (PPI_HASH_SIZE - 1)
+static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE];
static struct per_process_info *ppi_list;
static int ppi_list_entries;
-#define S_OPTS "i:o:b:stqw:f:F:v"
+#define S_OPTS "a:A:i:o:b:stqw:f:F:vVhD:"
static struct option l_opts[] = {
+ {
+ .name = "act-mask",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'a'
+ },
+ {
+ .name = "set-mask",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'A'
+ },
{
.name = "input",
.has_arg = required_argument,
.val = 'b'
},
{
- .name = "per program stats",
+ .name = "per-program-stats",
.has_arg = no_argument,
.flag = NULL,
.val = 's'
},
{
- .name = "track ios",
+ .name = "track-ios",
.has_arg = no_argument,
.flag = NULL,
.val = 't'
.val = 'F'
},
{
- .name = "version",
+ .name = "hash-by-name",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'h'
+ },
+ {
+ .name = "verbose",
.has_arg = no_argument,
.flag = NULL,
.val = 'v'
},
+ {
+ .name = "version",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'V'
+ },
+ {
+ .name = "input-directory",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'D'
+ },
+ {
+ .name = NULL,
+ }
};
/*
struct blk_io_trace *bit;
struct rb_node rb_node;
struct trace *next;
- int skipped;
+ unsigned long read_sequence;
};
static struct rb_root rb_sort_root;
-static struct rb_root rb_track_root;
+static unsigned long rb_sort_entries;
static struct trace *trace_list;
+/*
+ * allocation cache
+ */
+static struct blk_io_trace *bit_alloc_list;
+static struct trace *t_alloc_list;
+
/*
* for tracking individual ios
*/
struct io_track {
struct rb_node rb_node;
- dev_t device;
+ struct process_pid_map *ppm;
__u64 sector;
- __u32 pid;
unsigned long long allocation_time;
unsigned long long queue_time;
unsigned long long dispatch_time;
static int ndevices;
static struct per_dev_info *devices;
static char *get_dev_name(struct per_dev_info *, char *, int);
+static int trace_rb_insert_last(struct per_dev_info *, struct trace *);
-static FILE *ofp;
+FILE *ofp = NULL;
static char *output_name;
+static char *input_dir;
static unsigned long long genesis_time;
+static unsigned long long last_allowed_time;
static unsigned long long stopwatch_start; /* start from zero by default */
-static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */
+static unsigned long long stopwatch_end = -1ULL; /* "infinity" */
+static unsigned long read_sequence;
static int per_process_stats;
+static int per_device_and_cpu_stats = 1;
static int track_ios;
+static int ppi_hash_by_pid = 1;
+static int verbose;
+static unsigned int act_mask = -1U;
+static int stats_printed;
+int data_is_native = -1;
-#define RB_BATCH_DEFAULT (1024)
-static int rb_batch = RB_BATCH_DEFAULT;
+static unsigned int t_alloc_cache;
+static unsigned int bit_alloc_cache;
+
+#define RB_BATCH_DEFAULT (512)
+static unsigned int rb_batch = RB_BATCH_DEFAULT;
static int pipeline;
#define is_done() (*(volatile int *)(&done))
static volatile int done;
-static inline unsigned long hash_long(unsigned long val)
+#define JHASH_RANDOM (0x3af5f2ee)
+
+#define CPUS_PER_LONG (8 * sizeof(unsigned long))
+#define CPU_IDX(cpu) ((cpu) / CPUS_PER_LONG)
+#define CPU_BIT(cpu) ((cpu) & (CPUS_PER_LONG - 1))
+
+static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
{
-#if __WORDSIZE == 32
- val *= 0x9e370001UL;
-#elif __WORDSIZE == 64
- val *= 0x9e37fffffffc0001UL;
-#else
-#error unknown word size
-#endif
+ struct per_cpu_info *cpus = pdi->cpus;
+ int ncpus = pdi->ncpus;
+ int new_count = cpu + 1;
+ int new_space, size;
+ char *new_start;
- return val >> (__WORDSIZE - PPI_HASH_SHIFT);
+ size = new_count * sizeof(struct per_cpu_info);
+ cpus = realloc(cpus, size);
+ if (!cpus) {
+ char name[20];
+ fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
+ get_dev_name(pdi, name, sizeof(name)), size);
+ exit(1);
+ }
+
+ new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
+ new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
+ memset(new_start, 0, new_space);
+
+ pdi->ncpus = new_count;
+ pdi->cpus = cpus;
+
+ for (new_count = 0; new_count < pdi->ncpus; new_count++) {
+ struct per_cpu_info *pci = &pdi->cpus[new_count];
+
+ if (!pci->fd) {
+ pci->fd = -1;
+ memset(&pci->rb_last, 0, sizeof(pci->rb_last));
+ pci->rb_last_entries = 0;
+ pci->last_sequence = -1;
+ }
+ }
}
-static inline void add_process_to_hash(struct per_process_info *ppi)
+static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
{
- const int hash_idx = hash_long(ppi->pid);
+ struct per_cpu_info *pci;
- ppi->hash_next = ppi_hash[hash_idx];
- ppi_hash[hash_idx] = ppi;
+ if (cpu >= pdi->ncpus)
+ resize_cpu_info(pdi, cpu);
+
+ pci = &pdi->cpus[cpu];
+ pci->cpu = cpu;
+ return pci;
}
-static inline void add_process_to_list(struct per_process_info *ppi)
+
+static int resize_devices(char *name)
+{
+ int size = (ndevices + 1) * sizeof(struct per_dev_info);
+
+ devices = realloc(devices, size);
+ if (!devices) {
+ fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
+ return 1;
+ }
+ memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
+ devices[ndevices].name = name;
+ ndevices++;
+ return 0;
+}
+
+static struct per_dev_info *get_dev_info(dev_t dev)
+{
+ struct per_dev_info *pdi;
+ int i;
+
+ for (i = 0; i < ndevices; i++) {
+ if (!devices[i].dev)
+ devices[i].dev = dev;
+ if (devices[i].dev == dev)
+ return &devices[i];
+ }
+
+ if (resize_devices(NULL))
+ return NULL;
+
+ pdi = &devices[ndevices - 1];
+ pdi->dev = dev;
+ pdi->first_reported_time = 0;
+ pdi->last_read_time = 0;
+
+ return pdi;
+}
+
+static void insert_skip(struct per_cpu_info *pci, unsigned long start,
+ unsigned long end)
+{
+ struct skip_info *sip;
+
+ for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
+ if (end == (sip->start - 1)) {
+ sip->start = start;
+ return;
+ } else if (start == (sip->end + 1)) {
+ sip->end = end;
+ return;
+ }
+ }
+
+ sip = malloc(sizeof(struct skip_info));
+ sip->start = start;
+ sip->end = end;
+ sip->prev = sip->next = NULL;
+ if (pci->skips_tail == NULL)
+ pci->skips_head = pci->skips_tail = sip;
+ else {
+ sip->prev = pci->skips_tail;
+ pci->skips_tail->next = sip;
+ pci->skips_tail = sip;
+ }
+}
+
+static void remove_sip(struct per_cpu_info *pci, struct skip_info *sip)
+{
+ if (sip->prev == NULL) {
+ if (sip->next == NULL)
+ pci->skips_head = pci->skips_tail = NULL;
+ else {
+ pci->skips_head = sip->next;
+ sip->next->prev = NULL;
+ }
+ } else if (sip->next == NULL) {
+ pci->skips_tail = sip->prev;
+ sip->prev->next = NULL;
+ } else {
+ sip->prev->next = sip->next;
+ sip->next->prev = sip->prev;
+ }
+
+ sip->prev = sip->next = NULL;
+ free(sip);
+}
+
+#define IN_SKIP(sip,seq) (((sip)->start <= (seq)) && ((seq) <= sip->end))
+static int check_current_skips(struct per_cpu_info *pci, unsigned long seq)
+{
+ struct skip_info *sip;
+
+ for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
+ if (IN_SKIP(sip, seq)) {
+ if (sip->start == seq) {
+ if (sip->end == seq)
+ remove_sip(pci, sip);
+ else
+ sip->start += 1;
+ } else if (sip->end == seq)
+ sip->end -= 1;
+ else {
+ sip->end = seq - 1;
+ insert_skip(pci, seq + 1, sip->end);
+ }
+ return 1;
+ }
+ }
+
+ return 0;
+}
+
+static void collect_pdi_skips(struct per_dev_info *pdi)
+{
+ struct skip_info *sip;
+ int cpu;
+
+ pdi->skips = 0;
+ pdi->seq_skips = 0;
+
+ for (cpu = 0; cpu < pdi->ncpus; cpu++) {
+ struct per_cpu_info *pci = &pdi->cpus[cpu];
+
+ for (sip = pci->skips_head; sip != NULL; sip = sip->next) {
+ pdi->skips++;
+ pdi->seq_skips += (sip->end - sip->start + 1);
+ if (verbose)
+ fprintf(stderr,"(%d,%d): skipping %lu -> %lu\n",
+ MAJOR(pdi->dev), MINOR(pdi->dev),
+ sip->start, sip->end);
+ }
+ }
+}
+
+static void cpu_mark_online(struct per_dev_info *pdi, unsigned int cpu)
+{
+ if (cpu >= pdi->cpu_map_max || !pdi->cpu_map) {
+ int new_max = (cpu + CPUS_PER_LONG) & ~(CPUS_PER_LONG - 1);
+ unsigned long *map = malloc(new_max / sizeof(long));
+
+ memset(map, 0, new_max / sizeof(long));
+
+ if (pdi->cpu_map) {
+ memcpy(map, pdi->cpu_map, pdi->cpu_map_max / sizeof(long));
+ free(pdi->cpu_map);
+ }
+
+ pdi->cpu_map = map;
+ pdi->cpu_map_max = new_max;
+ }
+
+ pdi->cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
+}
+
+static inline void cpu_mark_offline(struct per_dev_info *pdi, int cpu)
+{
+ pdi->cpu_map[CPU_IDX(cpu)] &= ~(1UL << CPU_BIT(cpu));
+}
+
+static inline int cpu_is_online(struct per_dev_info *pdi, int cpu)
+{
+ return (pdi->cpu_map[CPU_IDX(cpu)] & (1UL << CPU_BIT(cpu))) != 0;
+}
+
+static inline int ppm_hash_pid(pid_t pid)
+{
+ return jhash_1word(pid, JHASH_RANDOM) & PPM_HASH_MASK;
+}
+
+static struct process_pid_map *find_ppm(pid_t pid)
+{
+ const int hash_idx = ppm_hash_pid(pid);
+ struct process_pid_map *ppm;
+
+ ppm = ppm_hash_table[hash_idx];
+ while (ppm) {
+ if (ppm->pid == pid)
+ return ppm;
+
+ ppm = ppm->hash_next;
+ }
+
+ return NULL;
+}
+
+static void add_ppm_hash(pid_t pid, const char *name)
+{
+ const int hash_idx = ppm_hash_pid(pid);
+ struct process_pid_map *ppm;
+
+ ppm = find_ppm(pid);
+ if (!ppm) {
+ ppm = malloc(sizeof(*ppm));
+ memset(ppm, 0, sizeof(*ppm));
+ ppm->pid = pid;
+ strcpy(ppm->comm, name);
+ ppm->hash_next = ppm_hash_table[hash_idx];
+ ppm_hash_table[hash_idx] = ppm;
+ }
+}
+
+char *find_process_name(pid_t pid)
+{
+ struct process_pid_map *ppm = find_ppm(pid);
+
+ if (ppm)
+ return ppm->comm;
+
+ return NULL;
+}
+
+static inline int ppi_hash_pid(pid_t pid)
+{
+ return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK;
+}
+
+static inline int ppi_hash_name(const char *name)
+{
+ return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK;
+}
+
+static inline int ppi_hash(struct per_process_info *ppi)
+{
+ struct process_pid_map *ppm = ppi->ppm;
+
+ if (ppi_hash_by_pid)
+ return ppi_hash_pid(ppm->pid);
+
+ return ppi_hash_name(ppm->comm);
+}
+
+static inline void add_ppi_to_hash(struct per_process_info *ppi)
+{
+ const int hash_idx = ppi_hash(ppi);
+
+ ppi->hash_next = ppi_hash_table[hash_idx];
+ ppi_hash_table[hash_idx] = ppi;
+}
+
+static inline void add_ppi_to_list(struct per_process_info *ppi)
{
ppi->list_next = ppi_list;
ppi_list = ppi;
ppi_list_entries++;
}
-static struct per_process_info *find_process_by_pid(__u32 pid)
+static struct per_process_info *find_ppi_by_name(char *name)
{
- const int hash_idx = hash_long(pid);
+ const int hash_idx = ppi_hash_name(name);
struct per_process_info *ppi;
- ppi = ppi_hash[hash_idx];
+ ppi = ppi_hash_table[hash_idx];
while (ppi) {
- if (ppi->pid == pid)
+ struct process_pid_map *ppm = ppi->ppm;
+
+ if (!strcmp(ppm->comm, name))
return ppi;
ppi = ppi->hash_next;
return NULL;
}
-static inline int trace_rb_insert(struct trace *t)
+static struct per_process_info *find_ppi_by_pid(pid_t pid)
+{
+ const int hash_idx = ppi_hash_pid(pid);
+ struct per_process_info *ppi;
+
+ ppi = ppi_hash_table[hash_idx];
+ while (ppi) {
+ struct process_pid_map *ppm = ppi->ppm;
+
+ if (ppm->pid == pid)
+ return ppi;
+
+ ppi = ppi->hash_next;
+ }
+
+ return NULL;
+}
+
+static struct per_process_info *find_ppi(pid_t pid)
+{
+ struct per_process_info *ppi;
+ char *name;
+
+ if (ppi_hash_by_pid)
+ return find_ppi_by_pid(pid);
+
+ name = find_process_name(pid);
+ if (!name)
+ return NULL;
+
+ ppi = find_ppi_by_name(name);
+ if (ppi && ppi->ppm->pid != pid)
+ ppi->more_than_one = 1;
+
+ return ppi;
+}
+
+/*
+ * struct trace and blktrace allocation cache, we do potentially
+ * millions of mallocs for these structures while only using at most
+ * a few thousand at the time
+ */
+static inline void t_free(struct trace *t)
+{
+ if (t_alloc_cache < 1024) {
+ t->next = t_alloc_list;
+ t_alloc_list = t;
+ t_alloc_cache++;
+ } else
+ free(t);
+}
+
+static inline struct trace *t_alloc(void)
{
- struct rb_node **p = &rb_sort_root.rb_node;
+ struct trace *t = t_alloc_list;
+
+ if (t) {
+ t_alloc_list = t->next;
+ t_alloc_cache--;
+ return t;
+ }
+
+ return malloc(sizeof(*t));
+}
+
+static inline void bit_free(struct blk_io_trace *bit)
+{
+ if (bit_alloc_cache < 1024 && !bit->pdu_len) {
+ /*
+ * abuse a 64-bit field for a next pointer for the free item
+ */
+ bit->time = (__u64) (unsigned long) bit_alloc_list;
+ bit_alloc_list = (struct blk_io_trace *) bit;
+ bit_alloc_cache++;
+ } else
+ free(bit);
+}
+
+static inline struct blk_io_trace *bit_alloc(void)
+{
+ struct blk_io_trace *bit = bit_alloc_list;
+
+ if (bit) {
+ bit_alloc_list = (struct blk_io_trace *) (unsigned long) \
+ bit->time;
+ bit_alloc_cache--;
+ return bit;
+ }
+
+ return malloc(sizeof(*bit));
+}
+
+static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t)
+{
+ struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
+
+ rb_erase(&t->rb_node, &pci->rb_last);
+ pci->rb_last_entries--;
+
+ bit_free(t->bit);
+ t_free(t);
+}
+
+static void put_trace(struct per_dev_info *pdi, struct trace *t)
+{
+ rb_erase(&t->rb_node, &rb_sort_root);
+ rb_sort_entries--;
+
+ trace_rb_insert_last(pdi, t);
+}
+
+static inline int trace_rb_insert(struct trace *t, struct rb_root *root)
+{
+ struct rb_node **p = &root->rb_node;
struct rb_node *parent = NULL;
struct trace *__t;
- if (genesis_time == 0 || t->bit->time < genesis_time)
- genesis_time = t->bit->time;
-
while (*p) {
parent = *p;
+
__t = rb_entry(parent, struct trace, rb_node);
if (t->bit->time < __t->bit->time)
p = &(*p)->rb_right;
else if (t->bit->sequence < __t->bit->sequence)
p = &(*p)->rb_left;
- else if (t->bit->sequence > __t->bit->sequence)
+ else /* >= sequence */
p = &(*p)->rb_right;
- else if (t->bit->device == __t->bit->device) {
- fprintf(stderr,
- "sequence alias (%d) on device %d,%d!\n",
- t->bit->sequence,
- MAJOR(t->bit->device), MINOR(t->bit->device));
- return 1;
- }
}
rb_link_node(&t->rb_node, parent, p);
- rb_insert_color(&t->rb_node, &rb_sort_root);
+ rb_insert_color(&t->rb_node, root);
return 0;
}
-static inline int track_rb_insert(struct io_track *iot)
+static inline int trace_rb_insert_sort(struct trace *t)
+{
+ if (!trace_rb_insert(t, &rb_sort_root)) {
+ rb_sort_entries++;
+ return 0;
+ }
+
+ return 1;
+}
+
+static int trace_rb_insert_last(struct per_dev_info *pdi, struct trace *t)
{
- struct rb_node **p = &rb_track_root.rb_node;
+ struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
+
+ if (trace_rb_insert(t, &pci->rb_last))
+ return 1;
+
+ pci->rb_last_entries++;
+
+ if (pci->rb_last_entries > rb_batch * pdi->nfiles) {
+ struct rb_node *n = rb_first(&pci->rb_last);
+
+ t = rb_entry(n, struct trace, rb_node);
+ __put_trace_last(pdi, t);
+ }
+
+ return 0;
+}
+
+static struct trace *trace_rb_find(dev_t device, unsigned long sequence,
+ struct rb_root *root, int order)
+{
+ struct rb_node *n = root->rb_node;
+ struct rb_node *prev = NULL;
+ struct trace *__t;
+
+ while (n) {
+ __t = rb_entry(n, struct trace, rb_node);
+ prev = n;
+
+ if (device < __t->bit->device)
+ n = n->rb_left;
+ else if (device > __t->bit->device)
+ n = n->rb_right;
+ else if (sequence < __t->bit->sequence)
+ n = n->rb_left;
+ else if (sequence > __t->bit->sequence)
+ n = n->rb_right;
+ else
+ return __t;
+ }
+
+ /*
+ * hack - the list may not be sequence ordered because some
+ * events don't have sequence and time matched. so we end up
+ * being a little off in the rb lookup here, because we don't
+ * know the time we are looking for. compensate by browsing
+ * a little ahead from the last entry to find the match
+ */
+ if (order && prev) {
+ int max = 5;
+
+ while (((n = rb_next(prev)) != NULL) && max--) {
+ __t = rb_entry(n, struct trace, rb_node);
+
+ if (__t->bit->device == device &&
+ __t->bit->sequence == sequence)
+ return __t;
+
+ prev = n;
+ }
+ }
+
+ return NULL;
+}
+
+static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi,
+ struct per_cpu_info *pci,
+ unsigned long seq)
+{
+ return trace_rb_find(pdi->dev, seq, &pci->rb_last, 0);
+}
+
+static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot)
+{
+ struct rb_node **p = &pdi->rb_track.rb_node;
struct rb_node *parent = NULL;
struct io_track *__iot;
while (*p) {
parent = *p;
-
__iot = rb_entry(parent, struct io_track, rb_node);
- if (iot->device < __iot->device)
- p = &(*p)->rb_left;
- else if (iot->device > __iot->device)
- p = &(*p)->rb_right;
- else if (iot->sector < __iot->sector)
+ if (iot->sector < __iot->sector)
p = &(*p)->rb_left;
else if (iot->sector > __iot->sector)
p = &(*p)->rb_right;
fprintf(stderr,
"sector alias (%Lu) on device %d,%d!\n",
(unsigned long long) iot->sector,
- MAJOR(iot->device), MINOR(iot->device));
+ MAJOR(pdi->dev), MINOR(pdi->dev));
return 1;
}
}
rb_link_node(&iot->rb_node, parent, p);
- rb_insert_color(&iot->rb_node, &rb_track_root);
+ rb_insert_color(&iot->rb_node, &pdi->rb_track);
return 0;
}
-static struct io_track *__find_track(dev_t device, __u64 sector)
+static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector)
{
- struct rb_node **p = &rb_track_root.rb_node;
- struct rb_node *parent = NULL;
+ struct rb_node *n = pdi->rb_track.rb_node;
struct io_track *__iot;
- while (*p) {
- parent = *p;
-
- __iot = rb_entry(parent, struct io_track, rb_node);
+ while (n) {
+ __iot = rb_entry(n, struct io_track, rb_node);
- if (device < __iot->device)
- p = &(*p)->rb_left;
- else if (device > __iot->device)
- p = &(*p)->rb_right;
- else if (sector < __iot->sector)
- p = &(*p)->rb_left;
+ if (sector < __iot->sector)
+ n = n->rb_left;
else if (sector > __iot->sector)
- p = &(*p)->rb_right;
+ n = n->rb_right;
else
return __iot;
}
return NULL;
}
-static struct io_track *find_track(__u32 pid, dev_t device, __u64 sector)
+static struct io_track *find_track(struct per_dev_info *pdi, pid_t pid,
+ __u64 sector)
{
struct io_track *iot;
- iot = __find_track(device, sector);
+ iot = __find_track(pdi, sector);
if (!iot) {
iot = malloc(sizeof(*iot));
- iot->pid = pid;
- iot->device = device;
+ iot->ppm = find_ppm(pid);
iot->sector = sector;
- track_rb_insert(iot);
+ track_rb_insert(pdi, iot);
}
return iot;
}
-static void log_track_frontmerge(struct blk_io_trace *t)
+static void log_track_frontmerge(struct per_dev_info *pdi,
+ struct blk_io_trace *t)
{
struct io_track *iot;
if (!track_ios)
return;
- iot = __find_track(t->device, t->sector + (t->bytes >> 9));
+ iot = __find_track(pdi, t->sector + t_sec(t));
if (!iot) {
- fprintf(stderr, "failed to find mergeable event\n");
+ if (verbose)
+ fprintf(stderr, "merge not found for (%d,%d): %llu\n",
+ MAJOR(pdi->dev), MINOR(pdi->dev),
+ (unsigned long long) t->sector + t_sec(t));
return;
}
- rb_erase(&iot->rb_node, &rb_track_root);
- iot->sector -= t->bytes >> 9;
- track_rb_insert(iot);
+ rb_erase(&iot->rb_node, &pdi->rb_track);
+ iot->sector -= t_sec(t);
+ track_rb_insert(pdi, iot);
}
-static void log_track_getrq(struct blk_io_trace *t)
+static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t)
{
struct io_track *iot;
if (!track_ios)
return;
- iot = find_track(t->pid, t->device, t->sector);
+ iot = find_track(pdi, t->pid, t->sector);
iot->allocation_time = t->time;
}
+static inline int is_remapper(struct per_dev_info *pdi)
+{
+ int major = MAJOR(pdi->dev);
+
+ return (major == 253 || major == 9);
+}
+
+/*
+ * for md/dm setups, the interesting cycle is Q -> C. So track queueing
+ * time here, as dispatch time
+ */
+static void log_track_queue(struct per_dev_info *pdi, struct blk_io_trace *t)
+{
+ struct io_track *iot;
+
+ if (!track_ios)
+ return;
+ if (!is_remapper(pdi))
+ return;
+
+ iot = find_track(pdi, t->pid, t->sector);
+ iot->dispatch_time = t->time;
+}
/*
* return time between rq allocation and insertion
*/
-static unsigned long long log_track_insert(struct blk_io_trace *t)
+static unsigned long long log_track_insert(struct per_dev_info *pdi,
+ struct blk_io_trace *t)
{
unsigned long long elapsed;
struct io_track *iot;
if (!track_ios)
return -1;
- iot = find_track(t->pid, t->device, t->sector);
+ iot = find_track(pdi, t->pid, t->sector);
iot->queue_time = t->time;
+
+ if (!iot->allocation_time)
+ return -1;
+
elapsed = iot->queue_time - iot->allocation_time;
if (per_process_stats) {
- struct per_process_info *ppi = find_process_by_pid(iot->pid);
+ struct per_process_info *ppi = find_ppi(iot->ppm->pid);
int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
if (ppi && elapsed > ppi->longest_allocation_wait[w])
/*
* return time between queue and issue
*/
-static unsigned long long log_track_issue(struct blk_io_trace *t)
+static unsigned long long log_track_issue(struct per_dev_info *pdi,
+ struct blk_io_trace *t)
{
unsigned long long elapsed;
struct io_track *iot;
if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
return -1;
- iot = __find_track(t->device, t->sector);
+ iot = __find_track(pdi, t->sector);
if (!iot) {
- fprintf(stderr, "failed to find issue event\n");
+ if (verbose)
+ fprintf(stderr, "issue not found for (%d,%d): %llu\n",
+ MAJOR(pdi->dev), MINOR(pdi->dev),
+ (unsigned long long) t->sector);
return -1;
}
elapsed = iot->dispatch_time - iot->queue_time;
if (per_process_stats) {
- struct per_process_info *ppi = find_process_by_pid(iot->pid);
+ struct per_process_info *ppi = find_ppi(iot->ppm->pid);
int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
if (ppi && elapsed > ppi->longest_dispatch_wait[w])
/*
* return time between dispatch and complete
*/
-static unsigned long long log_track_complete(struct blk_io_trace *t)
+static unsigned long long log_track_complete(struct per_dev_info *pdi,
+ struct blk_io_trace *t)
{
unsigned long long elapsed;
struct io_track *iot;
if (!track_ios)
return -1;
- if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
- return -1;
- iot = __find_track(t->device, t->sector);
+ iot = __find_track(pdi, t->sector);
if (!iot) {
- fprintf(stderr, "failed to find complete event\n");
+ if (verbose)
+ fprintf(stderr,"complete not found for (%d,%d): %llu\n",
+ MAJOR(pdi->dev), MINOR(pdi->dev),
+ (unsigned long long) t->sector);
return -1;
}
elapsed = iot->completion_time - iot->dispatch_time;
if (per_process_stats) {
- struct per_process_info *ppi = find_process_by_pid(iot->pid);
+ struct per_process_info *ppi = find_ppi(iot->ppm->pid);
int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
if (ppi && elapsed > ppi->longest_completion_wait[w])
/*
* kill the trace, we don't need it after completion
*/
- rb_erase(&iot->rb_node, &rb_track_root);
+ rb_erase(&iot->rb_node, &pdi->rb_track);
free(iot);
return elapsed;
}
-static struct io_stats *find_process_io_stats(__u32 pid, char *name)
+static struct io_stats *find_process_io_stats(pid_t pid)
{
- struct per_process_info *ppi = find_process_by_pid(pid);
+ struct per_process_info *ppi = find_ppi(pid);
if (!ppi) {
ppi = malloc(sizeof(*ppi));
memset(ppi, 0, sizeof(*ppi));
- strncpy(ppi->name, name, sizeof(ppi->name));
- ppi->pid = pid;
- add_process_to_hash(ppi);
- add_process_to_list(ppi);
+ ppi->ppm = find_ppm(pid);
+ add_ppi_to_hash(ppi);
+ add_ppi_to_list(ppi);
}
return &ppi->io_stats;
}
-
-static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
-{
- struct per_cpu_info *cpus = pdi->cpus;
- int ncpus = pdi->ncpus;
- int new_count = cpu + 1;
- int new_space, size;
- char *new_start;
-
- size = new_count * sizeof(struct per_cpu_info);
- cpus = realloc(cpus, size);
- if (!cpus) {
- char name[20];
- fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
- get_dev_name(pdi, name, sizeof(name)), size);
- exit(1);
- }
-
- new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
- new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
- memset(new_start, 0, new_space);
-
- pdi->ncpus = new_count;
- pdi->cpus = cpus;
-}
-
-static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
-{
- struct per_cpu_info *pci;
-
- if (cpu >= pdi->ncpus)
- resize_cpu_info(pdi, cpu);
-
- pci = &pdi->cpus[cpu];
- pci->cpu = cpu;
- return pci;
-}
-
-
-static int resize_devices(char *name)
-{
- int size = (ndevices + 1) * sizeof(struct per_dev_info);
-
- devices = realloc(devices, size);
- if (!devices) {
- fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
- return 1;
- }
- memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
- devices[ndevices].name = name;
- ndevices++;
- return 0;
-}
-
-static struct per_dev_info *get_dev_info(dev_t id)
-{
- struct per_dev_info *pdi;
- int i;
-
- for (i = 0; i < ndevices; i++)
- if (devices[i].id == id)
- return &devices[i];
-
- if (resize_devices(NULL) != 0)
- return NULL;
-
- pdi = &devices[ndevices - 1];
- pdi->id = id;
- pdi->last_sequence = -1;
- return pdi;
-}
-
static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
{
if (pdi->name)
snprintf(buffer, size, "%s", pdi->name);
else
- snprintf(buffer, size, "%d,%d", MAJOR(pdi->id), MINOR(pdi->id));
+ snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev));
return buffer;
}
{
if (rw) {
ios->mwrites++;
- ios->qwrite_kb += t->bytes >> 10;
+ ios->qwrite_kb += t_kb(t);
} else {
ios->mreads++;
- ios->qread_kb += t->bytes >> 10;
+ ios->qread_kb += t_kb(t);
}
}
__account_m(&pci->io_stats, t, rw);
if (per_process_stats) {
- struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
+ struct io_stats *ios = find_process_io_stats(t->pid);
__account_m(ios, t, rw);
}
{
if (rw) {
ios->qwrites++;
- ios->qwrite_kb += t->bytes >> 10;
+ ios->qwrite_kb += t_kb(t);
} else {
ios->qreads++;
- ios->qread_kb += t->bytes >> 10;
+ ios->qread_kb += t_kb(t);
}
}
__account_queue(&pci->io_stats, t, rw);
if (per_process_stats) {
- struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
+ struct io_stats *ios = find_process_io_stats(t->pid);
__account_queue(ios, t, rw);
}
}
-static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes)
+static inline void __account_c(struct io_stats *ios, int rw, int bytes)
{
if (rw) {
ios->cwrites++;
__account_c(&pci->io_stats, rw, bytes);
if (per_process_stats) {
- struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
+ struct io_stats *ios = find_process_io_stats(t->pid);
__account_c(ios, rw, bytes);
}
__account_issue(&pci->io_stats, rw, t->bytes);
if (per_process_stats) {
- struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
+ struct io_stats *ios = find_process_io_stats(t->pid);
__account_issue(ios, rw, t->bytes);
}
__account_unplug(&pci->io_stats, timer);
if (per_process_stats) {
- struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
+ struct io_stats *ios = find_process_io_stats(t->pid);
__account_unplug(ios, timer);
}
}
-#define VALID_SPECS "BCDFGMPQRSTU"
-static char *override_format[256];
-static inline int valid_spec(int spec)
-{
- return strchr(VALID_SPECS, spec) != NULL;
-}
-
-static void set_all_format_specs(char *optarg)
-{
- char *p;
-
- for (p = VALID_SPECS; *p; p++)
- if (override_format[(int)(*p)] == NULL)
- override_format[(int)(*p)] = strdup(optarg);
-}
-
-static int add_format_spec(char *optarg)
-{
- int spec = optarg[0];
-
- if (!valid_spec(spec)) {
- fprintf(stderr,"Bad format specifier %c\n", spec);
- return 1;
- }
- if (optarg[1] != ',') {
- fprintf(stderr,"Bad format specifier - need ',' %s\n", optarg);
- return 1;
- }
- optarg += 2;
- if (*optarg == '\0') {
- fprintf(stderr,"Bad format specifier - need fmt %s\n", optarg);
- return 1;
- }
-
- /*
- * Set both merges (front and back)
- */
- if (spec == 'M') {
- override_format['B'] = strdup(optarg);
- override_format['M'] = strdup(optarg);
- } else
- override_format[spec] = strdup(optarg);
-
- return 0;
-}
-
-static void print_field(char *act, struct per_cpu_info *pci,
- struct blk_io_trace *t, unsigned long long elapsed,
- int pdu_len, unsigned char *pdu_buf, char field,
- int minus, int has_w, int width)
-{
- char format[64];
-
- if (has_w) {
- if (minus)
- sprintf(format, "%%-%d", width);
- else
- sprintf(format, "%%%d", width);
- } else
- sprintf(format, "%%");
-
- switch (field) {
- case 'a':
- fprintf(ofp, strcat(format, "s"), act);
- break;
- case 'c':
- fprintf(ofp, strcat(format, "d"), pci->cpu);
- break;
- case 'C':
- fprintf(ofp, strcat(format, "s"), t->comm);
- break;
- case 'd': {
- char rwbs[4];
- int i = 0;
- int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
- int b = t->action & BLK_TC_ACT(BLK_TC_BARRIER);
- int s = t->action & BLK_TC_ACT(BLK_TC_SYNC);
- if (w)
- rwbs[i++] = 'W';
- else
- rwbs[i++] = 'R';
- if (b)
- rwbs[i++] = 'B';
- if (s)
- rwbs[i++] = 'S';
- rwbs[i] = '\0';
- fprintf(ofp, strcat(format, "s"), rwbs);
- break;
- }
- case 'D': /* format width ignored */
- fprintf(ofp,"%3d,%-3d", MAJOR(t->device), MINOR(t->device));
- break;
- case 'e':
- fprintf(ofp, strcat(format, "d"), t->error);
- break;
- case 'M':
- fprintf(ofp, strcat(format, "d"), MAJOR(t->device));
- break;
- case 'm':
- fprintf(ofp, strcat(format, "d"), MINOR(t->device));
- break;
- case 'n':
- fprintf(ofp, strcat(format, "u"), t->bytes >> 9);
- break;
- case 'p':
- fprintf(ofp, strcat(format, "u"), t->pid);
- break;
- case 'P': /* format width ignored */
- if ((pdu_len > 0) && (pdu_buf != NULL)) {
- int i;
- unsigned char *p = pdu_buf;
- for (i = 0; i < pdu_len; i++) {
- if (i)
- fprintf(ofp, " ");
-
- fprintf(ofp, "%02x", *p++);
- }
- }
- break;
- case 's':
- fprintf(ofp, strcat(format, "ld"), t->sequence);
- break;
- case 'S':
- fprintf(ofp, strcat(format, "lu"), t->sector);
- break;
- case 't':
- sprintf(format, "%%0%dlu", has_w ? width : 9);
- fprintf(ofp, format, NANO_SECONDS(t->time));
- break;
- case 'T':
- fprintf(ofp, strcat(format, "d"), SECONDS(t->time));
- break;
- case 'u':
- if (elapsed == -1ULL) {
- fprintf(stderr, "Expecting elapsed value\n");
- exit(1);
- }
- fprintf(ofp, strcat(format, "llu"), elapsed / 1000);
- break;
- case 'U': {
- __u64 *depth = (__u64 *) ((char *) t + sizeof(*t));
- fprintf(ofp, strcat(format, "u"),
- (unsigned int) be64_to_cpu(*depth));
- break;
- }
- default:
- fprintf(ofp,strcat(format, "c"), field);
- break;
- }
-}
-
-static char *parse_field(char *act, struct per_cpu_info *pci,
- struct blk_io_trace *t, unsigned long long elapsed,
- int pdu_len, unsigned char *pdu_buf,
- char *master_format)
+static inline void __account_requeue(struct io_stats *ios,
+ struct blk_io_trace *t, int rw)
{
- int minus = 0;
- int has_w = 0;
- int width = 0;
- char *p = master_format;
-
- if (*p == '-') {
- minus = 1;
- p++;
- }
- if (isdigit(*p)) {
- has_w = 1;
- do {
- width = (width * 10) + (*p++ - '0');
- } while ((*p) && (isdigit(*p)));
- }
- if (*p) {
- print_field(act, pci, t, elapsed, pdu_len, pdu_buf, *p++,
- minus, has_w, width);
+ if (rw) {
+ ios->wrqueue++;
+ ios->iwrite_kb -= t_kb(t);
+ } else {
+ ios->rrqueue++;
+ ios->iread_kb -= t_kb(t);
}
- return p;
}
-static char *fmt_select(int fmt_spec, struct blk_io_trace *t,
- unsigned long long elapsed)
+static inline void account_requeue(struct blk_io_trace *t,
+ struct per_cpu_info *pci, int rw)
{
- char *fmt;
- char scratch_format[1024];
-
- if (override_format[fmt_spec] != NULL)
- return override_format[fmt_spec];
-
- switch (fmt_spec) {
- case 'C': /* Complete */
- if (t->action & BLK_TC_ACT(BLK_TC_PC)) {
- strcpy(scratch_format, HEADER);
- strcat(scratch_format, "(%P) ");
- } else {
- strcpy(scratch_format, HEADER "%S + %n ");
- if (elapsed != -1ULL)
- strcat(scratch_format, "(%8u) ");
- }
- strcat(scratch_format, "[%e]\n");
- fmt = scratch_format;
- break;
-
- case 'D': /* Issue */
- if (t->action & BLK_TC_ACT(BLK_TC_PC)) {
- strcpy(scratch_format, HEADER);
- strcat(scratch_format, "%n (%P) ");
- } else {
- strcpy(scratch_format, HEADER "%S + %n ");
- if (elapsed != -1ULL)
- strcat(scratch_format, "(%8u) ");
- }
- strcat(scratch_format,"[%C]\n");
- fmt = scratch_format;
- break;
-
- case 'I': /* Insert */
- if (t->action & BLK_TC_ACT(BLK_TC_PC)) {
- strcpy(scratch_format, HEADER);
- strcat(scratch_format, "%n (%P) ");
- } else {
- strcpy(scratch_format, HEADER "%S + %n ");
- if (elapsed != -1ULL)
- strcat(scratch_format, "(%8u) ");
- }
- strcat(scratch_format,"[%C]\n");
- fmt = scratch_format;
- break;
-
- case 'Q': /* Queue */
- case 'W': /* Bounce */
- strcpy(scratch_format, HEADER "%S + %n ");
- if (elapsed != -1ULL)
- strcat(scratch_format, "(%8u) ");
- strcat(scratch_format,"[%C]\n");
- fmt = scratch_format;
- break;
-
- case 'B': /* Back merge */
- case 'F': /* Front merge */
- case 'M': /* Front or back merge */
- fmt = HEADER "%S + %n [%C]\n";
- break;
-
- case 'P': /* Plug */
- fmt = HEADER "[%C]\n";
- break;
-
- case 'G': /* Get request */
- case 'S': /* Sleep request */
- fmt = HEADER "%S + %n [%C]\n";
- break;
-
- case 'U': /* Unplug IO */
- case 'T': /* Unplug timer */
- fmt = HEADER "[%C] %U\n";
- break;
-
- case 'X': /* Split */
- strcpy(scratch_format, HEADER "%S / %U ");
- strcat(scratch_format,"[%C]\n");
- fmt = scratch_format;
- break;
-
- default:
- fprintf(stderr,"FATAL: Invalid format spec %c\n", fmt_spec);
- exit(1);
- /*NOTREACHED*/
- }
-
- return fmt;
-}
+ __account_requeue(&pci->io_stats, t, rw);
-static void process_fmt(char *act, struct per_cpu_info *pci,
- struct blk_io_trace *t, unsigned long long elapsed,
- int pdu_len, unsigned char *pdu_buf)
-{
- char *p = fmt_select(act[0], t, elapsed);
+ if (per_process_stats) {
+ struct io_stats *ios = find_process_io_stats(t->pid);
- while (*p) {
- switch (*p) {
- case '%': /* Field specifier */
- p++;
- if (*p == '%')
- fprintf(ofp, "%c", *p++);
- else if (!*p)
- fprintf(ofp, "%c", '%');
- else
- p = parse_field(act, pci, t, elapsed,
- pdu_len, pdu_buf, p);
- break;
- case '\\': { /* escape */
- switch (p[1]) {
- case 'b': fprintf(ofp, "\b"); break;
- case 'n': fprintf(ofp, "\n"); break;
- case 'r': fprintf(ofp, "\r"); break;
- case 't': fprintf(ofp, "\t"); break;
- default:
- fprintf(stderr,
- "Invalid escape char in format %c\n",
- p[1]);
- exit(1);
- /*NOTREACHED*/
- }
- p += 2;
- break;
- }
- default:
- fprintf(ofp, "%c", *p++);
- break;
- }
+ __account_requeue(ios, t, rw);
}
}
-static void log_complete(struct per_cpu_info *pci, struct blk_io_trace *t,
- char *act)
+static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci,
+ struct blk_io_trace *t, char *act)
{
- process_fmt(act, pci, t, log_track_complete(t), 0, NULL);
+ process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL);
}
-static void log_insert(struct per_cpu_info *pci, struct blk_io_trace *t,
- char *act)
+static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci,
+ struct blk_io_trace *t, char *act)
{
- process_fmt(act, pci, t, log_track_insert(t), 0, NULL);
+ process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL);
}
static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
process_fmt(act, pci, t, -1, 0, NULL);
}
-static void log_issue(struct per_cpu_info *pci, struct blk_io_trace *t,
- char *act)
+static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci,
+ struct blk_io_trace *t, char *act)
{
- process_fmt(act, pci, t, log_track_issue(t), 0, NULL);
+ process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL);
}
-static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t,
- char *act)
+static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci,
+ struct blk_io_trace *t, char *act)
{
if (act[0] == 'F')
- log_track_frontmerge(t);
+ log_track_frontmerge(pdi, t);
process_fmt(act, pci, t, -1ULL, 0, NULL);
}
}
}
-static void dump_trace_fs(struct blk_io_trace *t, struct per_cpu_info *pci)
+static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi,
+ struct per_cpu_info *pci)
{
- int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
+ int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
int act = t->action & 0xffff;
switch (act) {
case __BLK_TA_QUEUE:
+ log_track_queue(pdi, t);
account_queue(t, pci, w);
log_queue(pci, t, "Q");
break;
case __BLK_TA_INSERT:
- log_insert(pci, t, "I");
+ log_insert(pdi, pci, t, "I");
break;
case __BLK_TA_BACKMERGE:
account_m(t, pci, w);
- log_merge(pci, t, "M");
+ log_merge(pdi, pci, t, "M");
break;
case __BLK_TA_FRONTMERGE:
account_m(t, pci, w);
- log_merge(pci, t, "F");
+ log_merge(pdi, pci, t, "F");
break;
case __BLK_TA_GETRQ:
- log_track_getrq(t);
+ log_track_getrq(pdi, t);
log_generic(pci, t, "G");
break;
case __BLK_TA_SLEEPRQ:
log_generic(pci, t, "S");
break;
case __BLK_TA_REQUEUE:
- account_c(t, pci, w, -t->bytes);
+ /*
+ * can happen if we miss traces, don't let it go
+ * below zero
+ */
+ if (pdi->cur_depth[w])
+ pdi->cur_depth[w]--;
+ account_requeue(t, pci, w);
log_queue(pci, t, "R");
break;
case __BLK_TA_ISSUE:
account_issue(t, pci, w);
- log_issue(pci, t, "D");
+ pdi->cur_depth[w]++;
+ if (pdi->cur_depth[w] > pdi->max_depth[w])
+ pdi->max_depth[w] = pdi->cur_depth[w];
+ log_issue(pdi, pci, t, "D");
break;
case __BLK_TA_COMPLETE:
+ if (pdi->cur_depth[w])
+ pdi->cur_depth[w]--;
account_c(t, pci, w, t->bytes);
- log_complete(pci, t, "C");
+ log_complete(pdi, pci, t, "C");
break;
case __BLK_TA_PLUG:
log_action(pci, t, "P");
case __BLK_TA_BOUNCE:
log_generic(pci, t, "B");
break;
+ case __BLK_TA_REMAP:
+ log_generic(pci, t, "A");
+ break;
default:
fprintf(stderr, "Bad fs action %x\n", t->action);
break;
if (t->action & BLK_TC_ACT(BLK_TC_PC))
dump_trace_pc(t, pci);
else
- dump_trace_fs(t, pci);
+ dump_trace_fs(t, pdi, pci);
+
+ if (!pdi->events)
+ pdi->first_reported_time = t->time;
pdi->events++;
}
-static void dump_io_stats(struct io_stats *ios, char *msg)
+/*
+ * print in a proper way, not too small and not too big. if more than
+ * 1000,000K, turn into M and so on
+ */
+static char *size_cnv(char *dst, unsigned long long num, int in_kb)
+{
+ char suff[] = { '\0', 'K', 'M', 'G', 'P' };
+ unsigned int i = 0;
+
+ if (in_kb)
+ i++;
+
+ while (num > 1000 * 1000ULL && (i < sizeof(suff) - 1)) {
+ i++;
+ num /= 1000;
+ }
+
+ sprintf(dst, "%'8Lu%c", num, suff[i]);
+ return dst;
+}
+
+static void dump_io_stats(struct per_dev_info *pdi, struct io_stats *ios,
+ char *msg)
{
+ static char x[256], y[256];
+
fprintf(ofp, "%s\n", msg);
- fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb);
- fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb);
+ fprintf(ofp, " Reads Queued: %s, %siB\t", size_cnv(x, ios->qreads, 0), size_cnv(y, ios->qread_kb, 1));
+ fprintf(ofp, " Writes Queued: %s, %siB\n", size_cnv(x, ios->qwrites, 0), size_cnv(y, ios->qwrite_kb, 1));
- fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb);
- fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb);
- fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb);
- fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb);
+ fprintf(ofp, " Read Dispatches: %s, %siB\t", size_cnv(x, ios->ireads, 0), size_cnv(y, ios->iread_kb, 1));
+ fprintf(ofp, " Write Dispatches: %s, %siB\n", size_cnv(x, ios->iwrites, 0), size_cnv(y, ios->iwrite_kb, 1));
+ fprintf(ofp, " Reads Requeued: %s\t\t", size_cnv(x, ios->rrqueue, 0));
+ fprintf(ofp, " Writes Requeued: %s\n", size_cnv(x, ios->wrqueue, 0));
+ fprintf(ofp, " Reads Completed: %s, %siB\t", size_cnv(x, ios->creads, 0), size_cnv(y, ios->cread_kb, 1));
+ fprintf(ofp, " Writes Completed: %s, %siB\n", size_cnv(x, ios->cwrites, 0), size_cnv(y, ios->cwrite_kb, 1));
fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' ');
fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites);
+ if (pdi) {
+ fprintf(ofp, " Read depth: %'8u%8c\t", pdi->max_depth[0], ' ');
+ fprintf(ofp, " Write depth: %'8u\n", pdi->max_depth[1]);
+ }
fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' ');
fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs);
}
struct per_process_info *ppi2 = *((struct per_process_info **) p2);
int res;
- res = strverscmp(ppi1->name, ppi2->name);
+ res = strverscmp(ppi1->ppm->comm, ppi2->ppm->comm);
if (!res)
- res = ppi1->pid > ppi2->pid;
+ res = ppi1->ppm->pid > ppi2->ppm->pid;
return res;
}
ppi = ppi_list;
while (ppi) {
+ struct process_pid_map *ppm = ppi->ppm;
char name[64];
- snprintf(name, sizeof(name)-1, "%s (%u)", ppi->name, ppi->pid);
- dump_io_stats(&ppi->io_stats, name);
+ if (ppi->more_than_one)
+ sprintf(name, "%s (%u, ...)", ppm->comm, ppm->pid);
+ else
+ sprintf(name, "%s (%u)", ppm->comm, ppm->pid);
+
+ dump_io_stats(NULL, &ppi->io_stats, name);
dump_wait_stats(ppi);
ppi = ppi->list_next;
}
struct per_dev_info *pdi;
struct per_cpu_info *pci;
struct io_stats total, *ios;
+ unsigned long long rrate, wrate, msec;
int i, j, pci_events;
char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
char name[32];
total.mwrites += ios->mwrites;
total.ireads += ios->ireads;
total.iwrites += ios->iwrites;
+ total.rrqueue += ios->rrqueue;
+ total.wrqueue += ios->wrqueue;
total.qread_kb += ios->qread_kb;
total.qwrite_kb += ios->qwrite_kb;
total.cread_kb += ios->cread_kb;
snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
j, get_dev_name(pdi, name, sizeof(name)));
- dump_io_stats(ios, line);
+ dump_io_stats(pdi, ios, line);
pci_events++;
}
fprintf(ofp, "\n");
snprintf(line, sizeof(line) - 1, "Total (%s):",
get_dev_name(pdi, name, sizeof(name)));
- dump_io_stats(&total, line);
+ dump_io_stats(NULL, &total, line);
+ }
+
+ wrate = rrate = 0;
+ msec = (pdi->last_reported_time - pdi->first_reported_time) / 1000000;
+ if (msec) {
+ rrate = 1000 * total.cread_kb / msec;
+ wrate = 1000 * total.cwrite_kb / msec;
}
- fprintf(ofp, "\nEvents (%s): %'Lu entries, %'lu skips\n",
- get_dev_name(pdi, line, sizeof(line)), pdi->events,
- pdi->skips);
+ fprintf(ofp, "\nThroughput (R/W): %'LuKiB/s / %'LuKiB/s\n",
+ rrate, wrate);
+ fprintf(ofp, "Events (%s): %'Lu entries\n",
+ get_dev_name(pdi, line, sizeof(line)), pdi->events);
+
+ collect_pdi_skips(pdi);
+ fprintf(ofp, "Skips: %'lu forward (%'llu - %5.1lf%%)\n",
+ pdi->skips,pdi->seq_skips,
+ 100.0 * ((double)pdi->seq_skips /
+ (double)(pdi->events + pdi->seq_skips)));
}
}
-static struct blk_io_trace *find_trace(void *p, unsigned long offset)
+static void find_genesis(void)
{
- unsigned long max_offset = offset;
- unsigned long off;
- struct blk_io_trace *bit;
- __u32 magic;
+ struct trace *t = trace_list;
- for (off = 0; off < max_offset; off++) {
- bit = p + off;
+ genesis_time = -1ULL;
+ while (t != NULL) {
+ if (t->bit->time < genesis_time)
+ genesis_time = t->bit->time;
- magic = be32_to_cpu(bit->magic);
- if ((magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
- return bit;
+ t = t->next;
}
-
- return NULL;
}
-static inline int verify_and_add_trace(struct trace *t)
+static inline int check_stopwatch(struct blk_io_trace *bit)
{
- if (verify_trace(t->bit))
- return 1;
- if (trace_rb_insert(t))
- return 1;
+ if (bit->time < stopwatch_end &&
+ bit->time >= stopwatch_start)
+ return 0;
- return 0;
+ return 1;
}
-static int sort_entries(void)
+/*
+ * return youngest entry read
+ */
+static int sort_entries(unsigned long long *youngest)
{
+ struct per_dev_info *pdi = NULL;
+ struct per_cpu_info *pci = NULL;
struct trace *t;
- int nr = 0;
+ if (!genesis_time)
+ find_genesis();
+
+ *youngest = 0;
while ((t = trace_list) != NULL) {
+ struct blk_io_trace *bit = t->bit;
+
trace_list = t->next;
- verify_and_add_trace(t);
- nr++;
+
+ bit->time -= genesis_time;
+
+ if (bit->time < *youngest || !*youngest)
+ *youngest = bit->time;
+
+ if (!pdi || pdi->dev != bit->device) {
+ pdi = get_dev_info(bit->device);
+ pci = NULL;
+ }
+
+ if (!pci || pci->cpu != bit->cpu)
+ pci = get_cpu_info(pdi, bit->cpu);
+
+ if (bit->sequence < pci->smallest_seq_read)
+ pci->smallest_seq_read = bit->sequence;
+
+ if (check_stopwatch(bit)) {
+ bit_free(bit);
+ t_free(t);
+ continue;
+ }
+
+ if (trace_rb_insert_sort(t))
+ return -1;
+ }
+
+ return 0;
+}
+
+/*
+ * to continue, we must have traces from all online cpus in the tree
+ */
+static int check_cpu_map(struct per_dev_info *pdi)
+{
+ unsigned long *cpu_map;
+ struct rb_node *n;
+ struct trace *__t;
+ unsigned int i;
+ int ret, cpu;
+
+ /*
+ * create a map of the cpus we have traces for
+ */
+ cpu_map = malloc(pdi->cpu_map_max / sizeof(long));
+ n = rb_first(&rb_sort_root);
+ while (n) {
+ __t = rb_entry(n, struct trace, rb_node);
+ cpu = __t->bit->cpu;
+
+ cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
+ n = rb_next(n);
+ }
+
+ /*
+ * we can't continue if pdi->cpu_map has entries set that we don't
+ * have in the sort rbtree. the opposite is not a problem, though
+ */
+ ret = 0;
+ for (i = 0; i < pdi->cpu_map_max / CPUS_PER_LONG; i++) {
+ if (pdi->cpu_map[i] & ~(cpu_map[i])) {
+ ret = 1;
+ break;
+ }
}
- return nr;
+ free(cpu_map);
+ return ret;
+}
+
+static int check_sequence(struct per_dev_info *pdi, struct trace *t, int force)
+{
+ struct blk_io_trace *bit = t->bit;
+ unsigned long expected_sequence;
+ struct per_cpu_info *pci;
+ struct trace *__t;
+
+ pci = get_cpu_info(pdi, bit->cpu);
+ expected_sequence = pci->last_sequence + 1;
+
+ if (!expected_sequence) {
+ /*
+ * 1 should be the first entry, just allow it
+ */
+ if (bit->sequence == 1)
+ return 0;
+ if (bit->sequence == pci->smallest_seq_read)
+ return 0;
+
+ return check_cpu_map(pdi);
+ }
+
+ if (bit->sequence == expected_sequence)
+ return 0;
+
+ /*
+ * we may not have seen that sequence yet. if we are not doing
+ * the final run, break and wait for more entries.
+ */
+ if (expected_sequence < pci->smallest_seq_read) {
+ __t = trace_rb_find_last(pdi, pci, expected_sequence);
+ if (!__t)
+ goto skip;
+
+ __put_trace_last(pdi, __t);
+ return 0;
+ } else if (!force) {
+ return 1;
+ } else {
+skip:
+ if (check_current_skips(pci, bit->sequence))
+ return 0;
+
+ if (expected_sequence < bit->sequence)
+ insert_skip(pci, expected_sequence, bit->sequence - 1);
+ return 0;
+ }
}
-static void show_entries_rb(int piped)
+static void show_entries_rb(int force)
{
struct per_dev_info *pdi = NULL;
struct per_cpu_info *pci = NULL;
struct blk_io_trace *bit;
struct rb_node *n;
struct trace *t;
- __u32 device = 0;
- int cpu = 0;
-
- n = rb_first(&rb_sort_root);
- while (n != NULL) {
- if (done)
+ while ((n = rb_first(&rb_sort_root)) != NULL) {
+ if (is_done() && !force && !pipeline)
break;
t = rb_entry(n, struct trace, rb_node);
bit = t->bit;
- if (!pdi || device != bit->device) {
- device = bit->device;
- pdi = get_dev_info(device);
+ if (read_sequence - t->read_sequence < 1 && !force)
+ break;
+
+ if (!pdi || pdi->dev != bit->device) {
+ pdi = get_dev_info(bit->device);
+ pci = NULL;
}
if (!pdi) {
break;
}
- if (bit->cpu > pdi->ncpus) {
- fprintf(stderr, "Unknown CPU ID? (%d, device %d,%d)\n",
- cpu, MAJOR(bit->device), MINOR(bit->device));
+ if (check_sequence(pdi, t, force))
break;
- }
- if (!pci || cpu != bit->cpu) {
- cpu = bit->cpu;
- pci = get_cpu_info(pdi, cpu);
- }
+ if (!force && bit->time > last_allowed_time)
+ break;
- /*
- * back off displaying more info if we are out of sync
- * on SMP systems. to prevent stalling on lost events,
- * only allow an event to us a few times
- */
- if (bit->sequence != (pdi->last_sequence + 1)
- && pdi->last_sequence != -1) {
- if (piped && t->skipped < 5) {
- t->skipped++;
- break;
- } else {
- fprintf(stderr, "skipping from %lu to %u\n", pdi->last_sequence, bit->sequence);
- pdi->skips++;
- }
- }
+ check_time(pdi, bit);
- pdi->last_sequence = bit->sequence;
+ if (!pci || pci->cpu != bit->cpu)
+ pci = get_cpu_info(pdi, bit->cpu);
- bit->time -= genesis_time;
- if (bit->time >= stopwatch_end)
- break;
+ pci->last_sequence = bit->sequence;
- if (bit->time >= stopwatch_start) {
- check_time(pdi, bit);
+ pci->nelems++;
+ if (bit->action & (act_mask << BLK_TC_SHIFT))
dump_trace(bit, pci, pdi);
- }
- if (piped) {
- rb_erase(&t->rb_node, &rb_sort_root);
- free(bit);
- free(t);
- n = rb_first(&rb_sort_root);
- } else
- n = rb_next(n);
+ put_trace(pdi, t);
}
}
-static int read_data(int fd, void *buffer, int bytes, int block)
+static int read_data(int fd, void *buffer, int bytes, int block, int *fdblock)
{
int ret, bytes_left, fl;
void *p;
- fl = fcntl(fd, F_GETFL);
+ if (block != *fdblock) {
+ fl = fcntl(fd, F_GETFL);
- if (!block)
- fcntl(fd, F_SETFL, fl | O_NONBLOCK);
- else
- fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
+ if (!block) {
+ *fdblock = 0;
+ fcntl(fd, F_SETFL, fl | O_NONBLOCK);
+ } else {
+ *fdblock = 1;
+ fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
+ }
+ }
bytes_left = bytes;
p = buffer;
if (!ret)
return 1;
else if (ret < 0) {
- if (errno != EAGAIN)
+ if (errno != EAGAIN) {
perror("read");
- return -1;
+ return -1;
+ }
+
+ /*
+ * never do partial reads. we can return if we
+ * didn't read anything and we should not block,
+ * otherwise wait for data
+ */
+ if ((bytes_left == bytes) && !block)
+ return 1;
+
+ usleep(10);
+ continue;
} else {
p += ret;
bytes_left -= ret;
return 0;
}
-/*
- * Find the traces in 'tb' and add them to the list for sorting and
- * displaying
- */
-static int find_sort_entries(void *tb, unsigned long size)
+static inline __u16 get_pdulen(struct blk_io_trace *bit)
{
- struct blk_io_trace *bit;
- struct trace *t;
- void *start = tb;
- int nr = 0;
+ if (data_is_native)
+ return bit->pdu_len;
+
+ return __bswap_16(bit->pdu_len);
+}
+
+static inline __u32 get_magic(struct blk_io_trace *bit)
+{
+ if (data_is_native)
+ return bit->magic;
+
+ return __bswap_32(bit->magic);
+}
+
+static int read_events(int fd, int always_block, int *fdblock)
+{
+ struct per_dev_info *pdi = NULL;
+ unsigned int events = 0;
+
+ while (!is_done() && events < rb_batch) {
+ struct blk_io_trace *bit;
+ struct trace *t;
+ int pdu_len, should_block, ret;
+ __u32 magic;
- while (tb - start <= size - sizeof(*bit)) {
- bit = find_trace(tb, size - (tb - start));
- if (!bit)
+ bit = bit_alloc();
+
+ should_block = !events || always_block;
+
+ ret = read_data(fd, bit, sizeof(*bit), should_block, fdblock);
+ if (ret) {
+ bit_free(bit);
+ if (!events && ret < 0)
+ events = ret;
+ break;
+ }
+
+ /*
+ * look at first trace to check whether we need to convert
+ * data in the future
+ */
+ if (data_is_native == -1 && check_data_endianness(bit->magic))
+ break;
+
+ magic = get_magic(bit);
+ if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
+ fprintf(stderr, "Bad magic %x\n", magic);
break;
+ }
+
+ pdu_len = get_pdulen(bit);
+ if (pdu_len) {
+ void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
+
+ if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1, fdblock)) {
+ bit_free(ptr);
+ break;
+ }
+
+ bit = ptr;
+ }
+
+ trace_to_cpu(bit);
+
+ if (verify_trace(bit)) {
+ bit_free(bit);
+ continue;
+ }
+
+ /*
+ * not a real trace, so grab and handle it here
+ */
+ if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY)) {
+ add_ppm_hash(bit->pid, (char *) bit + sizeof(*bit));
+ continue;
+ }
- t = malloc(sizeof(*t));
+ t = t_alloc();
memset(t, 0, sizeof(*t));
t->bit = bit;
+ t->read_sequence = read_sequence;
- trace_to_cpu(bit);
+ t->next = trace_list;
+ trace_list = t;
- verify_and_add_trace(t);
+ if (!pdi || pdi->dev != bit->device)
+ pdi = get_dev_info(bit->device);
- tb += sizeof(*bit) + bit->pdu_len;
- nr++;
+ if (bit->time > pdi->last_read_time)
+ pdi->last_read_time = bit->time;
+
+ events++;
}
- return nr;
+ return events;
}
static int do_file(void)
{
- int i, j, nfiles = 0;
+ struct per_cpu_info *pci;
+ struct per_dev_info *pdi;
+ int i, j, events, events_added;
+ /*
+ * first prepare all files for reading
+ */
for (i = 0; i < ndevices; i++) {
- for (j = 0;; j++, nfiles++) {
- struct per_dev_info *pdi;
- struct per_cpu_info *pci;
+ pdi = &devices[i];
+ pdi->nfiles = 0;
+
+ for (j = 0;; j++) {
struct stat st;
- void *tb;
+ int len = 0;
+ char *p, *dname;
- pdi = &devices[i];
- pdi->last_sequence = -1;
pci = get_cpu_info(pdi, j);
pci->cpu = j;
-
- snprintf(pci->fname, sizeof(pci->fname)-1,
- "%s.blktrace.%d", pdi->name, j);
- if (stat(pci->fname, &st) < 0)
- break;
- if (!st.st_size)
- continue;
-
- printf("Processing %s\n", pci->fname);
-
- tb = malloc(st.st_size);
- if (!tb) {
- fprintf(stderr, "Out of memory, skip file %s\n",
- pci->fname);
- continue;
+ pci->fd = -1;
+ pci->fdblock = -1;
+
+ p = strdup(pdi->name);
+ dname = dirname(p);
+ if (strcmp(dname, ".")) {
+ input_dir = dname;
+ p = strdup(pdi->name);
+ strcpy(pdi->name, basename(p));
}
+ free(p);
- pci->fd = open(pci->fname, O_RDONLY);
- if (pci->fd < 0) {
- perror(pci->fname);
- free(tb);
- continue;
- }
+ if (input_dir)
+ len = sprintf(pci->fname, "%s/", input_dir);
- if (read_data(pci->fd, tb, st.st_size, 1)) {
- close(pci->fd);
- free(tb);
- continue;
+ snprintf(pci->fname + len, sizeof(pci->fname)-1-len,
+ "%s.blktrace.%d", pdi->name, pci->cpu);
+ if (stat(pci->fname, &st) < 0)
+ break;
+ if (st.st_size) {
+ pci->fd = open(pci->fname, O_RDONLY);
+ if (pci->fd < 0) {
+ perror(pci->fname);
+ continue;
+ }
}
- pci->nelems = find_sort_entries(tb, st.st_size);
-
- printf("Completed %s (CPU%d %d, entries)\n",
- pci->fname, j, pci->nelems);
- close(pci->fd);
+ printf("Input file %s added\n", pci->fname);
+ pdi->nfiles++;
+ cpu_mark_online(pdi, pci->cpu);
}
}
- if (!nfiles) {
- fprintf(stderr, "No files found\n");
- return 1;
- }
+ /*
+ * now loop over the files reading in the data
+ */
+ do {
+ unsigned long long youngest;
- show_entries_rb(0);
- return 0;
-}
+ events_added = 0;
+ last_allowed_time = -1ULL;
+ read_sequence++;
-static int read_sort_events(int fd)
-{
- int events = 0;
+ for (i = 0; i < ndevices; i++) {
+ pdi = &devices[i];
+ pdi->last_read_time = -1ULL;
- do {
- struct blk_io_trace *bit;
- struct trace *t;
- int pdu_len;
- __u32 magic;
+ for (j = 0; j < pdi->nfiles; j++) {
- bit = malloc(sizeof(*bit));
+ pci = get_cpu_info(pdi, j);
- if (read_data(fd, bit, sizeof(*bit), !events))
- break;
+ if (pci->fd == -1)
+ continue;
- magic = be32_to_cpu(bit->magic);
- if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
- fprintf(stderr, "Bad magic %x\n", magic);
- break;
- }
+ pci->smallest_seq_read = -1;
- pdu_len = be16_to_cpu(bit->pdu_len);
- if (pdu_len) {
- void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
+ events = read_events(pci->fd, 1, &pci->fdblock);
+ if (events <= 0) {
+ cpu_mark_offline(pdi, pci->cpu);
+ close(pci->fd);
+ pci->fd = -1;
+ continue;
+ }
- if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1))
- break;
+ if (pdi->last_read_time < last_allowed_time)
+ last_allowed_time = pdi->last_read_time;
- bit = ptr;
+ events_added += events;
+ }
}
- t = malloc(sizeof(*t));
- memset(t, 0, sizeof(*t));
- t->bit = bit;
+ if (sort_entries(&youngest))
+ break;
- trace_to_cpu(bit);
- t->next = trace_list;
- trace_list = t;
+ if (youngest > stopwatch_end)
+ break;
- events++;
- } while (!is_done() && events < rb_batch);
+ show_entries_rb(0);
- return events;
+ } while (events_added);
+
+ if (rb_sort_entries)
+ show_entries_rb(1);
+
+ return 0;
}
static int do_stdin(void)
{
- int fd;
+ unsigned long long youngest;
+ int fd, events, fdblock;
+ last_allowed_time = -1ULL;
fd = dup(STDIN_FILENO);
- do {
- int events;
+ if (fd == -1) {
+ perror("dup stdin");
+ return -1;
+ }
- events = read_sort_events(fd);
- if (!events)
- break;
+ fdblock = -1;
+ while ((events = read_events(fd, 0, &fdblock)) > 0) {
+ read_sequence++;
- if (sort_entries() == -1)
+#if 0
+ smallest_seq_read = -1U;
+#endif
+
+ if (sort_entries(&youngest))
+ break;
+
+ if (youngest > stopwatch_end)
break;
+ show_entries_rb(0);
+ }
+
+ if (rb_sort_entries)
show_entries_rb(1);
- } while (1);
close(fd);
return 0;
}
-static void flush_output(void)
+static void show_stats(void)
{
+ if (!ofp)
+ return;
+ if (stats_printed)
+ return;
+
+ stats_printed = 1;
+
+ if (per_process_stats)
+ show_process_stats();
+
+ if (per_device_and_cpu_stats)
+ show_device_and_cpu_stats();
+
fflush(ofp);
}
-static void handle_sigint(int sig)
+static void handle_sigint(__attribute__((__unused__)) int sig)
{
done = 1;
- flush_output();
}
/*
"[ -i <input name> ] [-o <output name> [ -s ] [ -t ] [ -q ]\n" \
"[ -w start:stop ] [ -f output format ] [ -F format spec ] [ -v] \n\n" \
"\t-i Input file containing trace data, or '-' for stdin\n" \
+ "\t-D Directory to prepend to input file names\n" \
"\t-o Output file. If not given, output is stdout\n" \
"\t-b stdin read batching\n" \
"\t-s Show per-program io statistics\n" \
+ "\t-h Hash processes by name, not pid\n" \
"\t-t Track individual ios. Will tell you the time a request took\n" \
"\t to get queued, to get dispatched, and to get completed\n" \
"\t-q Quiet. Don't display any stats at the end of the trace\n" \
"\t-w Only parse data between the given time interval in seconds.\n" \
"\t If 'start' isn't given, blkparse defaults the start time to 0\n" \
- "\t -f Output format. Customize the output format. The format field\n" \
- "\t identifies can be found in the documentation\n" \
+ "\t-f Output format. Customize the output format. The format field\n" \
+ "\t identifies can be found in the documentation\n" \
"\t-F Format specification. Can be found in the documentation\n" \
- "\t-v Print program version info\n\n";
+ "\t-v More verbose for marginal errors\n" \
+ "\t-V Print program version info\n\n";
static void usage(char *prog)
{
int main(int argc, char *argv[])
{
char *ofp_buffer;
- int c, ret, mode;
- int per_device_and_cpu_stats = 1;
+ int i, c, ret, mode;
+ int act_mask_tmp = 0;
while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
switch (c) {
+ case 'a':
+ i = find_mask_map(optarg);
+ if (i < 0) {
+ fprintf(stderr,"Invalid action mask %s\n",
+ optarg);
+ return 1;
+ }
+ act_mask_tmp |= i;
+ break;
+
+ case 'A':
+ if ((sscanf(optarg, "%x", &i) != 1) ||
+ !valid_act_opt(i)) {
+ fprintf(stderr,
+ "Invalid set action mask %s/0x%x\n",
+ optarg, i);
+ return 1;
+ }
+ act_mask_tmp = i;
+ break;
case 'i':
if (!strcmp(optarg, "-") && !pipeline)
pipeline = 1;
else if (resize_devices(optarg) != 0)
return 1;
break;
+ case 'D':
+ input_dir = optarg;
+ break;
case 'o':
output_name = optarg;
break;
if (add_format_spec(optarg) != 0)
return 1;
break;
+ case 'h':
+ ppi_hash_by_pid = 0;
+ break;
case 'v':
+ verbose++;
+ break;
+ case 'V':
printf("%s version %s\n", argv[0], blkparse_version);
return 0;
default:
return 1;
}
+ if (act_mask_tmp != 0)
+ act_mask = act_mask_tmp;
+
memset(&rb_sort_root, 0, sizeof(rb_sort_root));
- memset(&rb_track_root, 0, sizeof(rb_track_root));
signal(SIGINT, handle_sigint);
signal(SIGHUP, handle_sigint);
} else {
char ofname[128];
- snprintf(ofname, sizeof(ofname) - 1, "%s.log", output_name);
+ snprintf(ofname, sizeof(ofname) - 1, "%s", output_name);
ofp = fopen(ofname, "w");
mode = _IOFBF;
}
else
ret = do_file();
- if (per_process_stats)
- show_process_stats();
-
- if (per_device_and_cpu_stats)
- show_device_and_cpu_stats();
-
- flush_output();
+ show_stats();
+ free(ofp_buffer);
return ret;
}
projects / blktrace.git / blobdiff