#include "blktrace.h"
#include "rbtree.h"
+#include "jhash.h"
-#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))
-
-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;
-};
-
-struct per_cpu_info {
- int cpu;
- int nelems;
-
- int fd;
- char fname[128];
-
- struct io_stats io_stats;
-};
+static char blkparse_version[] = "0.90";
struct per_dev_info {
- dev_t id;
+ dev_t dev;
char *name;
int backwards;
unsigned long long events;
unsigned long long last_reported_time;
+ unsigned long long last_read_time;
struct io_stats io_stats;
+ unsigned long last_sequence;
+ unsigned long skips;
+
+ struct rb_root rb_last;
+ unsigned long rb_last_entries;
+
+ struct rb_root rb_track;
+ int nfiles;
int ncpus;
struct per_cpu_info *cpus;
};
__u32 pid;
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:"
+#define S_OPTS "i:o:b:stqw:f:F:vnm"
static struct option l_opts[] = {
{
.name = "input",
- .has_arg = 1,
+ .has_arg = required_argument,
.flag = NULL,
.val = 'i'
},
{
.name = "output",
- .has_arg = 1,
+ .has_arg = required_argument,
.flag = NULL,
.val = 'o'
},
{
.name = "batch",
- .has_arg = 1,
+ .has_arg = required_argument,
.flag = NULL,
.val = 'b'
},
{
.name = "per program stats",
- .has_arg = 0,
+ .has_arg = no_argument,
.flag = NULL,
.val = 's'
},
{
.name = "track ios",
- .has_arg = 0,
+ .has_arg = no_argument,
.flag = NULL,
.val = 't'
},
{
.name = "quiet",
- .has_arg = 0,
+ .has_arg = no_argument,
.flag = NULL,
.val = 'q'
},
{
.name = "stopwatch",
- .has_arg = 1,
+ .has_arg = required_argument,
.flag = NULL,
.val = 'w'
},
{
- .name = NULL,
- .has_arg = 0,
+ .name = "format",
+ .has_arg = required_argument,
.flag = NULL,
- .val = 0
- }
+ .val = 'f'
+ },
+ {
+ .name = "format-spec",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'F'
+ },
+ {
+ .name = "hash by name",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'n'
+ },
+ {
+ .name = "missing",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'm'
+ },
+ {
+ .name = "version",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'v'
+ },
};
-static struct rb_root rb_sort_root;
-static struct rb_root rb_track_root;
-
/*
* for sorting the displayed output
*/
struct trace {
struct blk_io_trace *bit;
struct rb_node rb_node;
+ struct trace *next;
};
+static struct rb_root rb_sort_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;
__u64 sector;
__u32 pid;
+ char comm[16];
unsigned long long allocation_time;
unsigned long long queue_time;
unsigned long long dispatch_time;
static struct per_dev_info *devices;
static char *get_dev_name(struct per_dev_info *, char *, int);
-static FILE *ofp;
+FILE *ofp = NULL;
static char *output_name;
static unsigned long long genesis_time;
+static unsigned long long last_allowed_time;
+static unsigned int smallest_seq_read;
static unsigned long long stopwatch_start; /* start from zero by default */
static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */
static int per_process_stats;
static int track_ios;
+static int ppi_hash_by_pid = 1;
+static int print_missing;
+
+static unsigned int t_alloc_cache;
+static unsigned int bit_alloc_cache;
-#define RB_BATCH_DEFAULT (1024)
-static int rb_batch = RB_BATCH_DEFAULT;
+#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)
+
+static inline int ppi_hash_pid(__u32 pid)
{
-#if __WORDSIZE == 32
- val *= 0x9e370001UL;
-#elif __WORDSIZE == 64
- val *= 0x9e37fffffffc0001UL;
-#else
-#error unknown word size
-#endif
+ 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)
+{
+ if (ppi_hash_by_pid)
+ return ppi_hash_pid(ppi->pid);
- return val >> (__WORDSIZE - PPI_HASH_SHIFT);
+ return ppi_hash_name(ppi->name);
}
static inline void add_process_to_hash(struct per_process_info *ppi)
{
- const int hash_idx = hash_long(ppi->pid);
+ const int hash_idx = ppi_hash(ppi);
- ppi->hash_next = ppi_hash[hash_idx];
- ppi_hash[hash_idx] = ppi;
+ ppi->hash_next = ppi_hash_table[hash_idx];
+ ppi_hash_table[hash_idx] = ppi;
}
static inline void add_process_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_name(char *name)
+{
+ const int hash_idx = ppi_hash_name(name);
+ struct per_process_info *ppi;
+
+ ppi = ppi_hash_table[hash_idx];
+ while (ppi) {
+ if (!strcmp(ppi->name, name))
+ return ppi;
+
+ ppi = ppi->hash_next;
+ }
+
+ return NULL;
}
static struct per_process_info *find_process_by_pid(__u32 pid)
{
- const int hash_idx = hash_long(pid);
+ const int hash_idx = ppi_hash_pid(pid);
struct per_process_info *ppi;
- ppi = ppi_hash[hash_idx];
+ ppi = ppi_hash_table[hash_idx];
while (ppi) {
if (ppi->pid == pid)
return ppi;
return NULL;
}
-static inline int trace_rb_insert(struct trace *t)
+static struct per_process_info *find_process(__u32 pid, char *name)
+{
+ struct per_process_info *ppi;
+
+ if (ppi_hash_by_pid)
+ return find_process_by_pid(pid);
+
+ ppi = find_process_by_name(name);
+ if (ppi && ppi->pid != pid)
+ ppi->more_than_one = 1;
+
+ return ppi;
+}
+
+static inline int trace_rb_insert(struct trace *t, struct rb_root *root,
+ int check_time)
{
- struct rb_node **p = &rb_sort_root.rb_node;
+ 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_left;
- else if (t->bit->time > __t->bit->time)
- p = &(*p)->rb_right;
- else if (t->bit->device < __t->bit->device)
+ if (check_time) {
+ if (t->bit->time < __t->bit->time) {
+ p = &(*p)->rb_left;
+ continue;
+ } else if (t->bit->time > __t->bit->time) {
+ p = &(*p)->rb_right;
+ continue;
+ }
+ }
+ if (t->bit->device < __t->bit->device)
p = &(*p)->rb_left;
else if (t->bit->device > __t->bit->device)
p = &(*p)->rb_right;
}
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, 1)) {
+ rb_sort_entries++;
+ return 0;
+ }
+
+ return 1;
+}
+
+static inline int trace_rb_insert_last(struct per_dev_info *pdi,struct trace *t)
+{
+ if (!trace_rb_insert(t, &pdi->rb_last, 1)) {
+ pdi->rb_last_entries++;
+ return 0;
+ }
+
+ return 1;
+}
+
+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_sort(dev_t dev, unsigned long seq)
+{
+ return trace_rb_find(dev, seq, &rb_sort_root, 1);
+}
+
+static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi,
+ unsigned long seq)
+{
+ return trace_rb_find(pdi->dev, seq, &pdi->rb_last, 0);
+}
+
+static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot)
{
- struct rb_node **p = &rb_track_root.rb_node;
+ 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;
else {
fprintf(stderr,
- "sector alias (%llu) on device %d,%d!\n",
- iot->sector,
- MAJOR(iot->device), MINOR(iot->device));
+ "sector alias (%Lu) on device %d,%d!\n",
+ (unsigned long long) iot->sector,
+ 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, __u32 pid,
+ char *comm, __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;
+ memcpy(iot->comm, comm, sizeof(iot->comm));
iot->sector = sector;
- track_rb_insert(iot);
+ track_rb_insert(pdi, iot);
}
return iot;
}
-static void log_track_merge(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;
- if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
- return;
- iot = __find_track(t->device, t->sector - (t->bytes >> 10));
+ iot = __find_track(pdi, t->sector + (t->bytes >> 9));
if (!iot) {
- fprintf(stderr, "Trying to merge on non-existing request\n");
+ fprintf(stderr, "merge not found for (%d,%d): %llu\n",
+ MAJOR(pdi->dev), MINOR(pdi->dev),
+ (unsigned long long) t->sector + (t->bytes >> 9));
return;
}
- rb_erase(&iot->rb_node, &rb_track_root);
- iot->sector -= t->bytes >> 10;
- track_rb_insert(iot);
+ rb_erase(&iot->rb_node, &pdi->rb_track);
+ iot->sector -= t->bytes >> 9;
+ 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->comm, t->sector);
iot->allocation_time = t->time;
}
-
/*
- * return time between rq allocation and queue
+ * return time between rq allocation and insertion
*/
-static unsigned long long log_track_queue(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->comm, 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_process(iot->pid,iot->comm);
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, "Trying to issue on non-existing request\n");
+ 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_process(iot->pid,iot->comm);
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 ((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, "Trying to dispatch on non-existing request\n");
+ 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_process(iot->pid,iot->comm);
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)
{
- struct per_process_info *ppi = find_process_by_pid(pid);
+ struct per_process_info *ppi = find_process(pid, name);
if (!ppi) {
ppi = malloc(sizeof(*ppi));
memset(ppi, 0, sizeof(*ppi));
- strncpy(ppi->name, name, sizeof(ppi->name));
+ memcpy(ppi->name, name, 16);
ppi->pid = pid;
add_process_to_hash(ppi);
add_process_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;
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);
- return &pdi->cpus[cpu];
+
+ pci = &pdi->cpus[cpu];
+ pci->cpu = cpu;
+ return pci;
}
return 0;
}
-static struct per_dev_info *get_dev_info(dev_t id, int create)
+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].id == id)
+ for (i = 0; i < ndevices; i++) {
+ if (!devices[i].dev)
+ devices[i].dev = dev;
+ if (devices[i].dev == dev)
return &devices[i];
- if (!create)
- return NULL;
- if (resize_devices(NULL) != 0)
+ }
+
+ if (resize_devices(NULL))
return NULL;
- return &devices[ndevices-1];
+
+ pdi = &devices[ndevices - 1];
+ pdi->dev = dev;
+ pdi->last_sequence = -1;
+ pdi->last_read_time = 0;
+ memset(&pdi->rb_last, 0, sizeof(pdi->rb_last));
+ pdi->rb_last_entries = 0;
+ 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;
}
-
static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
{
unsigned long long this = bit->time;
pdi->last_reported_time = this;
}
-
static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
int rw)
{
}
}
-static inline void __account_q(struct io_stats *ios, struct blk_io_trace *t,
- int rw)
+static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t,
+ int rw)
{
if (rw) {
ios->qwrites++;
}
}
-static inline void account_q(struct blk_io_trace *t, struct per_cpu_info *pci,
- int rw)
+static inline void account_queue(struct blk_io_trace *t,
+ struct per_cpu_info *pci, int rw)
{
- __account_q(&pci->io_stats, t, rw);
+ __account_queue(&pci->io_stats, t, rw);
if (per_process_stats) {
struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
- __account_q(ios, t, rw);
+ __account_queue(ios, t, rw);
}
}
}
}
-static inline void __account_i(struct io_stats *ios, int rw, unsigned int bytes)
+static inline void __account_issue(struct io_stats *ios, int rw,
+ unsigned int bytes)
{
if (rw) {
ios->iwrites++;
}
}
-static inline void account_i(struct blk_io_trace *t, struct per_cpu_info *pci,
- int rw)
+static inline void account_issue(struct blk_io_trace *t,
+ struct per_cpu_info *pci, int rw)
{
- __account_i(&pci->io_stats, rw, t->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);
- __account_i(ios, rw, t->bytes);
+ __account_issue(ios, rw, t->bytes);
}
}
-static void output(struct per_cpu_info *pci, char *s)
+static inline void __account_unplug(struct io_stats *ios, int timer)
{
- fprintf(ofp, "%s", s);
+ if (timer)
+ ios->timer_unplugs++;
+ else
+ ios->io_unplugs++;
}
-static char hstring[256];
-static char tstring[256];
-
-static inline char *setup_header(struct per_cpu_info *pci,
- struct blk_io_trace *t, char act)
+static inline void account_unplug(struct blk_io_trace *t,
+ struct per_cpu_info *pci, int timer)
{
- 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);
- char rwbs[4];
- int i = 0;
-
- if (w)
- rwbs[i++] = 'W';
- else
- rwbs[i++] = 'R';
- if (b)
- rwbs[i++] = 'B';
- if (s)
- rwbs[i++] = 'S';
+ __account_unplug(&pci->io_stats, timer);
- rwbs[i] = '\0';
-
- sprintf(hstring, "%3d,%-3d %2d %8ld %5Lu.%09Lu %5u %c %3s",
- MAJOR(t->device), MINOR(t->device), pci->cpu,
- (unsigned long)t->sequence, SECONDS(t->time),
- NANO_SECONDS(t->time), t->pid, act, rwbs);
+ if (per_process_stats) {
+ struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
- return hstring;
+ __account_unplug(ios, timer);
+ }
}
-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)
{
- unsigned long long elapsed = log_track_complete(t);
-
- if (elapsed != -1ULL) {
- unsigned long usec = elapsed / 1000;
+ process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL);
+}
- sprintf(tstring,"%s %Lu + %u (%8lu) [%d]\n",
- setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9,
- usec, t->error);
- } else {
- sprintf(tstring,"%s %Lu + %u [%d]\n", setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9, t->error);
- }
-
- output(pci, tstring);
+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(pdi, t), 0, NULL);
}
static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
- char act)
+ char *act)
{
- unsigned long long elapsed = log_track_queue(t);
-
- if (elapsed != -1ULL) {
- unsigned long usec = elapsed / 1000;
-
- sprintf(tstring,"%s %Lu + %u (%8lu) [%s]\n",
- setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9,
- usec, t->comm);
- } else {
- sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9, t->comm);
- }
- output(pci, tstring);
+ 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)
{
- unsigned long long elapsed = log_track_issue(t);
-
- if (elapsed != -1ULL) {
- double usec = (double) elapsed / 1000;
+ process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL);
+}
- sprintf(tstring,"%s %Lu + %u (%8.2f) [%s]\n",
- setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9,
- usec, t->comm);
- } else {
- sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9, t->comm);
- }
+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(pdi, t);
- output(pci, tstring);
+ process_fmt(act, pci, t, -1ULL, 0, NULL);
}
-static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t,
- char act)
+static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
+ char *act)
{
- log_track_merge(t);
-
- sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9, t->comm);
- output(pci, tstring);
+ process_fmt(act, pci, t, -1ULL, 0, NULL);
}
static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
- char act)
+ char *act)
{
- sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
- (unsigned long long)t->sector, t->bytes >> 9, t->comm);
- output(pci, tstring);
+ process_fmt(act, pci, t, -1ULL, 0, NULL);
}
-static int log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char act)
+static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
+ char *act)
{
- unsigned char *buf;
- int i;
+ process_fmt(act, pci, t, -1ULL, 0, NULL);
+}
- sprintf(tstring,"%s ", setup_header(pci, t, act));
- output(pci, tstring);
+static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t,
+ char *act)
+{
+ process_fmt(act, pci, t, -1ULL, 0, NULL);
+}
- buf = (unsigned char *) t + sizeof(*t);
- for (i = 0; i < t->pdu_len; i++) {
- sprintf(tstring,"%02x ", buf[i]);
- output(pci, tstring);
- }
+static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
+{
+ unsigned char *buf = (unsigned char *) t + sizeof(*t);
- if (act == 'C') {
- sprintf(tstring,"[%d]\n", t->error);
- output(pci, tstring);
- } else {
- sprintf(tstring,"[%s]\n", t->comm);
- output(pci, tstring);
- }
- return 0;
+ process_fmt(act, pci, t, -1ULL, t->pdu_len, buf);
}
-static int dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci)
+static void dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci)
{
- int ret = 0;
+ int act = t->action & 0xffff;
- switch (t->action & 0xffff) {
+ switch (act) {
case __BLK_TA_QUEUE:
- log_generic(pci, t, 'Q');
+ log_generic(pci, t, "Q");
break;
case __BLK_TA_GETRQ:
- log_generic(pci, t, 'G');
+ log_generic(pci, t, "G");
break;
case __BLK_TA_SLEEPRQ:
- log_generic(pci, t, 'S');
+ log_generic(pci, t, "S");
break;
case __BLK_TA_REQUEUE:
- log_generic(pci, t, 'R');
+ log_generic(pci, t, "R");
break;
case __BLK_TA_ISSUE:
- ret = log_pc(pci, t, 'D');
+ log_pc(pci, t, "D");
break;
case __BLK_TA_COMPLETE:
- log_pc(pci, t, 'C');
+ log_pc(pci, t, "C");
+ break;
+ case __BLK_TA_INSERT:
+ log_pc(pci, t, "I");
break;
default:
- fprintf(stderr, "Bad pc action %x\n", t->action);
- ret = 1;
+ fprintf(stderr, "Bad pc action %x\n", act);
break;
}
-
- return ret;
}
-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 act = t->action & 0xffff;
switch (act) {
case __BLK_TA_QUEUE:
- account_q(t, pci, w);
- log_queue(pci, t, 'Q');
+ account_queue(t, pci, w);
+ log_queue(pci, t, "Q");
+ break;
+ case __BLK_TA_INSERT:
+ 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_generic(pci, t, 'G');
+ log_track_getrq(pdi, t);
+ log_generic(pci, t, "G");
break;
case __BLK_TA_SLEEPRQ:
- log_generic(pci, t, 'S');
+ log_generic(pci, t, "S");
break;
case __BLK_TA_REQUEUE:
account_c(t, pci, w, -t->bytes);
- log_queue(pci, t, 'R');
+ log_queue(pci, t, "R");
break;
case __BLK_TA_ISSUE:
- account_i(t, pci, w);
- log_issue(pci, t, 'D');
+ account_issue(t, pci, w);
+ log_issue(pdi, pci, t, "D");
break;
case __BLK_TA_COMPLETE:
account_c(t, pci, w, t->bytes);
- log_complete(pci, t, 'C');
+ log_complete(pdi, pci, t, "C");
break;
case __BLK_TA_PLUG:
- log_generic(pci, t, 'P');
+ log_action(pci, t, "P");
+ break;
+ case __BLK_TA_UNPLUG_IO:
+ account_unplug(t, pci, 0);
+ log_unplug(pci, t, "U");
break;
- case __BLK_TA_UNPLUG:
- log_generic(pci, t, 'U');
+ case __BLK_TA_UNPLUG_TIMER:
+ account_unplug(t, pci, 1);
+ log_unplug(pci, t, "UT");
+ break;
+ case __BLK_TA_SPLIT:
+ log_split(pci, t, "X");
+ break;
+ case __BLK_TA_BOUNCE:
+ log_generic(pci, t, "B");
break;
default:
fprintf(stderr, "Bad fs action %x\n", t->action);
}
}
-static int dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
- struct per_dev_info *pdi)
+static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
+ struct per_dev_info *pdi)
{
- int ret = 0;
-
if (t->action & BLK_TC_ACT(BLK_TC_PC))
- ret = dump_trace_pc(t, pci);
+ dump_trace_pc(t, pci);
else
- dump_trace_fs(t, pci);
+ dump_trace_fs(t, pdi, pci);
pdi->events++;
- return ret;
}
static void dump_io_stats(struct io_stats *ios, char *msg)
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 Merges: %'8lu%8c\t", ios->mreads, ' ');
-
fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites);
+ fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' ');
+ fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs);
}
static void dump_wait_stats(struct per_process_info *ppi)
fprintf(ofp, " Completion wait: %'8lu\n", wcwait);
}
+static int ppi_name_compare(const void *p1, const void *p2)
+{
+ struct per_process_info *ppi1 = *((struct per_process_info **) p1);
+ struct per_process_info *ppi2 = *((struct per_process_info **) p2);
+ int res;
+
+ res = strverscmp(ppi1->name, ppi2->name);
+ if (!res)
+ res = ppi1->pid > ppi2->pid;
+
+ return res;
+}
+
+static void sort_process_list(void)
+{
+ struct per_process_info **ppis;
+ struct per_process_info *ppi;
+ int i = 0;
+
+ ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *));
+
+ ppi = ppi_list;
+ while (ppi) {
+ ppis[i++] = ppi;
+ ppi = ppi->list_next;
+ }
+
+ qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare);
+
+ i = ppi_list_entries - 1;
+ ppi_list = NULL;
+ while (i >= 0) {
+ ppi = ppis[i];
+
+ ppi->list_next = ppi_list;
+ ppi_list = ppi;
+ i--;
+ }
+
+ free(ppis);
+}
+
static void show_process_stats(void)
{
struct per_process_info *ppi;
+ sort_process_list();
+
ppi = ppi_list;
while (ppi) {
- dump_io_stats(&ppi->io_stats, ppi->name);
+ char name[64];
+
+ if (ppi->more_than_one)
+ sprintf(name, "%s (%u, ...)", ppi->name, ppi->pid);
+ else
+ sprintf(name, "%s (%u)", ppi->name, ppi->pid);
+
+ dump_io_stats(&ppi->io_stats, name);
dump_wait_stats(ppi);
ppi = ppi->list_next;
}
total.cwrite_kb += ios->cwrite_kb;
total.iread_kb += ios->iread_kb;
total.iwrite_kb += ios->iwrite_kb;
+ total.timer_unplugs += ios->timer_unplugs;
+ total.io_unplugs += ios->io_unplugs;
snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
j, get_dev_name(pdi, name, sizeof(name)));
dump_io_stats(&total, line);
}
- fprintf(ofp, "Events (%s): %'Lu\n",
- get_dev_name(pdi, line, sizeof(line)), pdi->events);
+ fprintf(ofp, "\nEvents (%s): %'Lu entries, %'lu skips\n",
+ get_dev_name(pdi, line, sizeof(line)), pdi->events,
+ pdi->skips);
}
}
-static struct blk_io_trace *find_trace(void *p, unsigned long offset, int nr)
+/*
+ * 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)
{
- unsigned long max_offset = min(offset,nr * sizeof(struct blk_io_trace));
- unsigned long off;
- struct blk_io_trace *bit;
- __u32 magic;
+ if (t_alloc_cache < 1024) {
+ t->next = t_alloc_list;
+ t_alloc_list = t;
+ t_alloc_cache++;
+ } else
+ free(t);
+}
- for (off = 0; off < max_offset; off++) {
- bit = p + off;
+static inline struct trace *t_alloc(void)
+{
+ struct trace *t = t_alloc_list;
- magic = be32_to_cpu(bit->magic);
- if ((magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
- return bit;
+ if (t) {
+ t_alloc_list = t->next;
+ t_alloc_cache--;
+ return t;
}
- return NULL;
+ return malloc(sizeof(*t));
}
-static int sort_entries(void *traces, unsigned long offset, int nr,
- struct per_dev_info *fpdi, struct per_cpu_info *fpci)
+static inline void bit_free(struct blk_io_trace *bit)
+{
+ if (bit_alloc_cache < 1024) {
+ /*
+ * 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 void find_genesis(void)
+{
+ struct trace *t = trace_list;
+
+ genesis_time = -1ULL;
+ while (t != NULL) {
+ if (t->bit->time < genesis_time)
+ genesis_time = t->bit->time;
+
+ t = t->next;
+ }
+}
+
+static inline int check_stopwatch(struct blk_io_trace *bit)
+{
+ if (bit->time < stopwatch_end &&
+ bit->time >= stopwatch_start)
+ return 0;
+
+ return 1;
+}
+
+/*
+ * return youngest entry read
+ */
+static int sort_entries(unsigned long long *youngest)
{
- struct per_dev_info *pdi;
- struct per_cpu_info *pci;
- struct blk_io_trace *bit;
struct trace *t;
- void *start = traces;
- while (traces - start <= offset - sizeof(*bit)) {
- if (!nr)
- break;
+ if (!genesis_time)
+ find_genesis();
- bit = find_trace(traces, offset - (traces - start), nr);
- if (!bit)
- break;
+ *youngest = 0;
+ while ((t = trace_list) != NULL) {
+ struct blk_io_trace *bit = t->bit;
- t = malloc(sizeof(*t));
- if (!t) {
- fprintf(stderr, "Out of memory, seq %d on dev %d,%d\n",
- bit->sequence,
- MAJOR(bit->device), MINOR(bit->device));
- return -1;
- }
- t->bit = bit;
- memset(&t->rb_node, 0, sizeof(t->rb_node));
+ trace_list = t->next;
- trace_to_cpu(bit);
+ bit->time -= genesis_time;
- if (verify_trace(bit)) {
- free(t);
- break;
- }
+ if (bit->time < *youngest || !*youngest)
+ *youngest = bit->time;
- pdi = fpdi ? fpdi : get_dev_info(bit->device, 1);
- pdi->id = bit->device;
- pci = fpci ? fpci : get_cpu_info(pdi, bit->cpu);
- pci->cpu = bit->cpu;
- pci->nelems++;
+ if (check_stopwatch(bit)) {
+ bit_free(bit);
+ t_free(t);
+ continue;
+ }
- if (trace_rb_insert(t)) {
- free(t);
+ if (trace_rb_insert_sort(t))
return -1;
- }
- traces += sizeof(*bit) + bit->pdu_len;
- nr--;
+ if (bit->sequence < smallest_seq_read)
+ smallest_seq_read = bit->sequence;
}
return 0;
}
-static void free_entries_rb(void)
+static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t)
{
- struct rb_node *n;
+ rb_erase(&t->rb_node, &pdi->rb_last);
+ pdi->rb_last_entries--;
- while ((n = rb_first(&rb_sort_root)) != NULL) {
- struct trace *t = rb_entry(n, struct trace, rb_node);
+ bit_free(t->bit);
+ t_free(t);
+}
- rb_erase(&t->rb_node, &rb_sort_root);
- 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);
+
+ if (pdi->rb_last_entries > rb_batch * pdi->nfiles) {
+ struct rb_node *n = rb_first(&pdi->rb_last);
+
+ t = rb_entry(n, struct trace, rb_node);
+ __put_trace_last(pdi, t);
}
}
-static void show_entries_rb(void)
+static int check_sequence(struct per_dev_info *pdi, struct blk_io_trace *bit,
+ int force)
{
- struct per_dev_info *pdi;
+ unsigned long expected_sequence = pdi->last_sequence + 1;
+ struct trace *t;
+
+ /*
+ * first entry, always ok
+ */
+ if (!expected_sequence)
+ return 0;
+
+ 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 < smallest_seq_read) {
+ t = trace_rb_find_last(pdi, expected_sequence);
+ if (!t)
+ goto skip;
+
+ __put_trace_last(pdi, t);
+ return 0;
+ } else if (!force)
+ return 1;
+ else {
+skip:
+ if (print_missing) {
+ fprintf(stderr, "(%d,%d): skipping %lu -> %u\n",
+ MAJOR(pdi->dev), MINOR(pdi->dev),
+ pdi->last_sequence, bit->sequence);
+ }
+ pdi->skips++;
+ return 0;
+ }
+}
+
+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;
- int cpu;
- n = rb_first(&rb_sort_root);
- if (!n)
- return;
+ if (force) {
+ n = rb_first(&rb_sort_root);
+ t = rb_entry(n, struct trace, rb_node);
+ fprintf(stderr, "first force %u\n", t->bit->sequence);
+ }
+
+ while ((n = rb_first(&rb_sort_root)) != NULL) {
+ if (done)
+ break;
- do {
t = rb_entry(n, struct trace, rb_node);
bit = t->bit;
- pdi = get_dev_info(bit->device, 0);
+ if (!pdi || pdi->dev != bit->device)
+ pdi = get_dev_info(bit->device);
+
if (!pdi) {
fprintf(stderr, "Unknown device ID? (%d,%d)\n",
MAJOR(bit->device), MINOR(bit->device));
break;
}
- cpu = bit->cpu;
- if (cpu > pdi->ncpus) {
- fprintf(stderr, "Unknown CPU ID? (%d, device %d,%d)\n",
- cpu, MAJOR(bit->device), MINOR(bit->device));
+
+ if (check_sequence(pdi, bit, force))
break;
- }
- bit->time -= genesis_time;
- if (bit->time < stopwatch_start)
- continue;
- if (bit->time >= stopwatch_end)
+ if (!force && bit->time > last_allowed_time)
break;
+ pdi->last_sequence = bit->sequence;
+
check_time(pdi, bit);
- if (dump_trace(bit, &pdi->cpus[cpu], pdi))
- break;
+ if (!pci || pci->cpu != bit->cpu)
+ pci = get_cpu_info(pdi, bit->cpu);
- } while ((n = rb_next(n)) != NULL);
+ dump_trace(bit, pci, pdi);
+
+ put_trace(pdi, t);
+ }
}
static int read_data(int fd, void *buffer, int bytes, int block)
else if (ret < 0) {
if (errno != EAGAIN)
perror("read");
+
return -1;
} else {
p += ret;
return 0;
}
-static int do_file(void)
+static int read_events(int fd, int always_block)
{
- struct per_dev_info *pdi;
- int i, j, nfiles = 0;
+ struct per_dev_info *pdi = NULL;
+ unsigned int events = 0;
- for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
- for (j = 0;; j++, nfiles++) {
- struct per_cpu_info *pci;
- struct stat st;
- void *tb;
+ while (!is_done() && events < rb_batch) {
+ struct blk_io_trace *bit;
+ struct trace *t;
+ int pdu_len;
+ __u32 magic;
- pci = get_cpu_info(pdi, j);
- pci->cpu = j;
+ bit = bit_alloc();
- snprintf(pci->fname, sizeof(pci->fname)-1,
- "%s_out.%d", pdi->name, j);
- if (stat(pci->fname, &st) < 0)
- break;
- if (!st.st_size)
- continue;
+ if (read_data(fd, bit, sizeof(*bit), !events || always_block))
+ break;
- printf("Processing %s\n", pci->fname);
+ magic = be32_to_cpu(bit->magic);
+ if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
+ fprintf(stderr, "Bad magic %x\n", magic);
+ break;
+ }
- tb = malloc(st.st_size);
- if (!tb) {
- fprintf(stderr, "Out of memory, skip file %s\n",
- pci->fname);
- continue;
- }
+ pdu_len = be16_to_cpu(bit->pdu_len);
+ if (pdu_len) {
+ void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
- pci->fd = open(pci->fname, O_RDONLY);
- if (pci->fd < 0) {
- perror(pci->fname);
- free(tb);
- continue;
- }
+ if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1))
+ break;
- if (read_data(pci->fd, tb, st.st_size, 1)) {
- close(pci->fd);
- free(tb);
- continue;
- }
+ bit = ptr;
+ }
- if (sort_entries(tb, st.st_size, ~0U, pdi, pci) == -1) {
- close(pci->fd);
- free(tb);
- continue;
- }
+ trace_to_cpu(bit);
- printf("Completed %s (CPU%d %d, entries)\n",
- pci->fname, j, pci->nelems);
- close(pci->fd);
+ if (verify_trace(bit)) {
+ bit_free(bit);
+ continue;
}
- }
- if (!nfiles) {
- fprintf(stderr, "No files found\n");
- return 1;
+ t = t_alloc();
+ memset(t, 0, sizeof(*t));
+ t->bit = bit;
+
+ t->next = trace_list;
+ trace_list = t;
+
+ if (!pdi || pdi->dev != bit->device)
+ pdi = get_dev_info(bit->device);
+
+ if (bit->time > pdi->last_read_time)
+ pdi->last_read_time = bit->time;
+
+ events++;
}
- show_entries_rb();
- return 0;
+ return events;
}
-static void resize_buffer(void **buffer, long *size, long offset)
+static int do_file(void)
{
- long old_size = *size;
+ struct per_cpu_info *pci;
+ struct per_dev_info *pdi;
+ int i, j, events, events_added;
- if (*size == 0)
- *size = 64 * sizeof(struct blk_io_trace);
+ /*
+ * first prepare all files for reading
+ */
+ for (i = 0; i < ndevices; i++) {
+ pdi = &devices[i];
+ pdi->nfiles = 0;
+ pdi->last_sequence = -1;
- *size *= 2;
- *buffer = realloc(*buffer, *size);
+ for (j = 0;; j++) {
+ struct stat st;
- if (old_size)
- memset(*buffer + offset, 0, *size - old_size);
-}
+ pci = get_cpu_info(pdi, j);
+ pci->cpu = j;
+ pci->fd = -1;
-static int read_sort_events(int fd, void **buffer, long *max_offset)
-{
- long offset;
- int events;
+ snprintf(pci->fname, sizeof(pci->fname)-1,
+ "%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;
+ }
+ }
+
+ printf("Input file %s added\n", pci->fname);
+ pdi->nfiles++;
+ }
+ }
- events = offset = 0;
+ /*
+ * now loop over the files reading in the data
+ */
do {
- struct blk_io_trace *t;
- int pdu_len;
- __u32 magic;
+ unsigned long long youngest;
- if (*max_offset - offset < sizeof(*t))
- resize_buffer(buffer, max_offset, offset);
+ events_added = 0;
+ last_allowed_time = -1ULL;
+ smallest_seq_read = -1U;
- if (read_data(fd, *buffer + offset, sizeof(*t), !events))
- break;
+ for (i = 0; i < ndevices; i++) {
+ pdi = &devices[i];
- t = *buffer + offset;
- offset += sizeof(*t);
+ for (j = 0; j < pdi->nfiles; j++) {
- magic = be32_to_cpu(t->magic);
- if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
- fprintf(stderr, "Bad magic %x\n", magic);
- break;
- }
+ pci = get_cpu_info(pdi, j);
- pdu_len = be16_to_cpu(t->pdu_len);
- if (pdu_len) {
- if (*max_offset - offset <= pdu_len)
- resize_buffer(buffer, max_offset, offset);
+ if (pci->fd == -1)
+ continue;
- if (read_data(fd, *buffer + offset, pdu_len, 1))
- break;
+ events = read_events(pci->fd, 1);
+ if (!events) {
+ close(pci->fd);
+ pci->fd = -1;
+ continue;
+ }
+
+ if (pdi->last_read_time < last_allowed_time)
+ last_allowed_time = pdi->last_read_time;
- offset += pdu_len;
+ events_added += events;
+ }
}
- events++;
- } while (!is_done() && events < rb_batch);
+ if (sort_entries(&youngest))
+ break;
- return events;
+ if (youngest > stopwatch_end)
+ break;
+
+ show_entries_rb(0);
+
+ } while (events_added);
+
+ if (rb_sort_entries)
+ show_entries_rb(1);
+
+ return 0;
}
static int do_stdin(void)
{
+ unsigned long long youngest;
int fd;
- void *ptr = NULL;
- long max_offset;
+ last_allowed_time = -1ULL;
fd = dup(STDIN_FILENO);
- max_offset = 0;
do {
int events;
- events = read_sort_events(fd, &ptr, &max_offset);
+ events = read_events(fd, 0);
if (!events)
break;
- if (sort_entries(ptr, ~0UL, events, NULL, NULL) == -1)
+ if (sort_entries(&youngest))
break;
- show_entries_rb();
- free_entries_rb();
+ if (youngest > stopwatch_end)
+ break;
+
+ show_entries_rb(0);
} while (1);
- if (ptr)
- free(ptr);
+ if (rb_sort_entries)
+ show_entries_rb(1);
close(fd);
return 0;
fflush(ofp);
}
-static void handle_sigint(int sig)
+static void handle_sigint(__attribute__((__unused__)) int sig)
{
done = 1;
flush_output();
fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
return 1;
}
- stopwatch_end = stopwatch_start + DOUBLE_TO_NANO_ULL(value);
+ stopwatch_end = DOUBLE_TO_NANO_ULL(value);
+ if (stopwatch_end <= stopwatch_start) {
+ fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n",
+ stopwatch_start, stopwatch_end);
+ return 1;
+ }
+
return 0;
}
+static char usage_str[] = \
+ "[ -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-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-n 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 Format specification. Can be found in the documentation\n" \
+ "\t-m Print missing entries\n" \
+ "\t-v Print program version info\n\n";
+
static void usage(char *prog)
{
- fprintf(stderr, "Usage: %s "
- "[-i <name>] [-o <output>] [-s] [-w N[:n]] <name>...\n",
- prog);
+ fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str);
}
int main(int argc, char *argv[])
if (find_stopwatch_interval(optarg) != 0)
return 1;
break;
+ case 'f':
+ set_all_format_specs(optarg);
+ break;
+ case 'F':
+ if (add_format_spec(optarg) != 0)
+ return 1;
+ break;
+ case 'n':
+ ppi_hash_by_pid = 0;
+ break;
+ case 'm':
+ print_missing = 1;
+ break;
+ case 'v':
+ printf("%s version %s\n", argv[0], blkparse_version);
+ return 0;
default:
usage(argv[0]);
return 1;
}
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;
}