Merge branch 'perf/fast' into perf/core
[linux-3.10.git] / tools / perf / builtin-kmem.c
index 2071d24..39104c0 100644 (file)
@@ -7,21 +7,19 @@
 #include "util/thread.h"
 #include "util/header.h"
 #include "util/session.h"
+#include "util/tool.h"
 
 #include "util/parse-options.h"
 #include "util/trace-event.h"
 
 #include "util/debug.h"
-#include "util/data_map.h"
 
 #include <linux/rbtree.h>
 
 struct alloc_stat;
 typedef int (*sort_fn_t)(struct alloc_stat *, struct alloc_stat *);
 
-static char const              *input_name = "perf.data";
-
-static u64                     sample_type;
+static const char              *input_name;
 
 static int                     alloc_flag;
 static int                     caller_flag;
@@ -95,27 +93,24 @@ static void setup_cpunode_map(void)
        if (!dir1)
                return;
 
-       while (true) {
-               dent1 = readdir(dir1);
-               if (!dent1)
-                       break;
-
-               if (sscanf(dent1->d_name, "node%u", &mem) < 1)
+       while ((dent1 = readdir(dir1)) != NULL) {
+               if (dent1->d_type != DT_DIR ||
+                   sscanf(dent1->d_name, "node%u", &mem) < 1)
                        continue;
 
                snprintf(buf, PATH_MAX, "%s/%s", PATH_SYS_NODE, dent1->d_name);
                dir2 = opendir(buf);
                if (!dir2)
                        continue;
-               while (true) {
-                       dent2 = readdir(dir2);
-                       if (!dent2)
-                               break;
-                       if (sscanf(dent2->d_name, "cpu%u", &cpu) < 1)
+               while ((dent2 = readdir(dir2)) != NULL) {
+                       if (dent2->d_type != DT_LNK ||
+                           sscanf(dent2->d_name, "cpu%u", &cpu) < 1)
                                continue;
                        cpunode_map[cpu] = mem;
                }
+               closedir(dir2);
        }
+       closedir(dir1);
 }
 
 static void insert_alloc_stat(unsigned long call_site, unsigned long ptr,
@@ -140,7 +135,7 @@ static void insert_alloc_stat(unsigned long call_site, unsigned long ptr,
        if (data && data->ptr == ptr) {
                data->hit++;
                data->bytes_req += bytes_req;
-               data->bytes_alloc += bytes_req;
+               data->bytes_alloc += bytes_alloc;
        } else {
                data = malloc(sizeof(*data));
                if (!data)
@@ -180,7 +175,7 @@ static void insert_caller_stat(unsigned long call_site,
        if (data && data->call_site == call_site) {
                data->hit++;
                data->bytes_req += bytes_req;
-               data->bytes_alloc += bytes_req;
+               data->bytes_alloc += bytes_alloc;
        } else {
                data = malloc(sizeof(*data));
                if (!data)
@@ -283,9 +278,8 @@ static void process_free_event(void *data,
        s_alloc->alloc_cpu = -1;
 }
 
-static void
-process_raw_event(event_t *raw_event __used, void *data,
-                 int cpu, u64 timestamp, struct thread *thread)
+static void process_raw_event(union perf_event *raw_event __used, void *data,
+                             int cpu, u64 timestamp, struct thread *thread)
 {
        struct event *event;
        int type;
@@ -312,25 +306,14 @@ process_raw_event(event_t *raw_event __used, void *data,
        }
 }
 
-static int process_sample_event(event_t *event)
+static int process_sample_event(struct perf_tool *tool __used,
+                               union perf_event *event,
+                               struct perf_sample *sample,
+                               struct perf_evsel *evsel __used,
+                               struct machine *machine)
 {
-       struct sample_data data;
-       struct thread *thread;
-
-       memset(&data, 0, sizeof(data));
-       data.time = -1;
-       data.cpu = -1;
-       data.period = 1;
+       struct thread *thread = machine__findnew_thread(machine, event->ip.pid);
 
-       event__parse_sample(event, sample_type, &data);
-
-       dump_printf("(IP, %d): %d/%d: %p period: %Ld\n",
-               event->header.misc,
-               data.pid, data.tid,
-               (void *)(long)data.ip,
-               (long long)data.period);
-
-       thread = threads__findnew(event->ip.pid);
        if (thread == NULL) {
                pr_debug("problem processing %d event, skipping it.\n",
                         event->header.type);
@@ -339,48 +322,18 @@ static int process_sample_event(event_t *event)
 
        dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
 
-       process_raw_event(event, data.raw_data, data.cpu,
-                         data.time, thread);
-
-       return 0;
-}
-
-static int sample_type_check(u64 type)
-{
-       sample_type = type;
-
-       if (!(sample_type & PERF_SAMPLE_RAW)) {
-               fprintf(stderr,
-                       "No trace sample to read. Did you call perf record "
-                       "without -R?");
-               return -1;
-       }
+       process_raw_event(event, sample->raw_data, sample->cpu,
+                         sample->time, thread);
 
        return 0;
 }
 
-static struct perf_file_handler file_handler = {
-       .process_sample_event   = process_sample_event,
-       .process_comm_event     = event__process_comm,
-       .sample_type_check      = sample_type_check,
+static struct perf_tool perf_kmem = {
+       .sample                 = process_sample_event,
+       .comm                   = perf_event__process_comm,
+       .ordered_samples        = true,
 };
 
-static int read_events(void)
-{
-       int err;
-       struct perf_session *session = perf_session__new(input_name, O_RDONLY, 0);
-
-       if (session == NULL)
-               return -ENOMEM;
-
-       register_idle_thread();
-       register_perf_file_handler(&file_handler);
-
-       err = perf_session__process_events(session, 0, &event__cwdlen, &event__cwd);
-       perf_session__delete(session);
-       return err;
-}
-
 static double fragmentation(unsigned long n_req, unsigned long n_alloc)
 {
        if (n_alloc == 0)
@@ -389,39 +342,47 @@ static double fragmentation(unsigned long n_req, unsigned long n_alloc)
                return 100.0 - (100.0 * n_req / n_alloc);
 }
 
-static void __print_result(struct rb_root *root, int n_lines, int is_caller)
+static void __print_result(struct rb_root *root, struct perf_session *session,
+                          int n_lines, int is_caller)
 {
        struct rb_node *next;
+       struct machine *machine;
 
        printf("%.102s\n", graph_dotted_line);
        printf(" %-34s |",  is_caller ? "Callsite": "Alloc Ptr");
-       printf(" Total_alloc/Per | Total_req/Per   | Hit   | Ping-pong | Frag\n");
+       printf(" Total_alloc/Per | Total_req/Per   | Hit      | Ping-pong | Frag\n");
        printf("%.102s\n", graph_dotted_line);
 
        next = rb_first(root);
 
+       machine = perf_session__find_host_machine(session);
+       if (!machine) {
+               pr_err("__print_result: couldn't find kernel information\n");
+               return;
+       }
        while (next && n_lines--) {
                struct alloc_stat *data = rb_entry(next, struct alloc_stat,
                                                   node);
                struct symbol *sym = NULL;
+               struct map *map;
                char buf[BUFSIZ];
                u64 addr;
 
                if (is_caller) {
                        addr = data->call_site;
                        if (!raw_ip)
-                               sym = map_groups__find_function(kmaps, addr, NULL);
+                               sym = machine__find_kernel_function(machine, addr, &map, NULL);
                } else
                        addr = data->ptr;
 
                if (sym != NULL)
-                       snprintf(buf, sizeof(buf), "%s+%Lx", sym->name,
-                                addr - sym->start);
+                       snprintf(buf, sizeof(buf), "%s+%" PRIx64 "", sym->name,
+                                addr - map->unmap_ip(map, sym->start));
                else
-                       snprintf(buf, sizeof(buf), "%#Lx", addr);
+                       snprintf(buf, sizeof(buf), "%#" PRIx64 "", addr);
                printf(" %-34s |", buf);
 
-               printf(" %9llu/%-5lu | %9llu/%-5lu | %6lu | %8lu | %6.3f%%\n",
+               printf(" %9llu/%-5lu | %9llu/%-5lu | %8lu | %8lu | %6.3f%%\n",
                       (unsigned long long)data->bytes_alloc,
                       (unsigned long)data->bytes_alloc / data->hit,
                       (unsigned long long)data->bytes_req,
@@ -451,12 +412,12 @@ static void print_summary(void)
        printf("Cross CPU allocations: %lu/%lu\n", nr_cross_allocs, nr_allocs);
 }
 
-static void print_result(void)
+static void print_result(struct perf_session *session)
 {
        if (caller_flag)
-               __print_result(&root_caller_sorted, caller_lines, 1);
+               __print_result(&root_caller_sorted, session, caller_lines, 1);
        if (alloc_flag)
-               __print_result(&root_alloc_sorted, alloc_lines, 0);
+               __print_result(&root_alloc_sorted, session, alloc_lines, 0);
        print_summary();
 }
 
@@ -524,12 +485,27 @@ static void sort_result(void)
 
 static int __cmd_kmem(void)
 {
+       int err = -EINVAL;
+       struct perf_session *session = perf_session__new(input_name, O_RDONLY,
+                                                        0, false, &perf_kmem);
+       if (session == NULL)
+               return -ENOMEM;
+
+       if (perf_session__create_kernel_maps(session) < 0)
+               goto out_delete;
+
+       if (!perf_session__has_traces(session, "kmem record"))
+               goto out_delete;
+
        setup_pager();
-       read_events();
+       err = perf_session__process_events(session, &perf_kmem);
+       if (err != 0)
+               goto out_delete;
        sort_result();
-       print_result();
-
-       return 0;
+       print_result(session);
+out_delete:
+       perf_session__delete(session);
+       return err;
 }
 
 static const char * const kmem_usage[] = {
@@ -671,6 +647,7 @@ static int setup_sorting(struct list_head *sort_list, const char *arg)
                        break;
                if (sort_dimension__add(tok, sort_list) < 0) {
                        error("Unknown --sort key: '%s'", tok);
+                       free(str);
                        return -1;
                }
        }
@@ -748,7 +725,6 @@ static const char *record_args[] = {
        "record",
        "-a",
        "-R",
-       "-M",
        "-f",
        "-c", "1",
        "-e", "kmem:kmalloc",
@@ -767,6 +743,9 @@ static int __cmd_record(int argc, const char **argv)
        rec_argc = ARRAY_SIZE(record_args) + argc - 1;
        rec_argv = calloc(rec_argc + 1, sizeof(char *));
 
+       if (rec_argv == NULL)
+               return -ENOMEM;
+
        for (i = 0; i < ARRAY_SIZE(record_args); i++)
                rec_argv[i] = strdup(record_args[i]);
 
@@ -778,13 +757,13 @@ static int __cmd_record(int argc, const char **argv)
 
 int cmd_kmem(int argc, const char **argv, const char *prefix __used)
 {
-       symbol__init(0);
-
        argc = parse_options(argc, argv, kmem_options, kmem_usage, 0);
 
        if (!argc)
                usage_with_options(kmem_usage, kmem_options);
 
+       symbol__init();
+
        if (!strncmp(argv[0], "rec", 3)) {
                return __cmd_record(argc, argv);
        } else if (!strcmp(argv[0], "stat")) {
@@ -796,7 +775,8 @@ int cmd_kmem(int argc, const char **argv, const char *prefix __used)
                        setup_sorting(&alloc_sort, default_sort_order);
 
                return __cmd_kmem();
-       }
+       } else
+               usage_with_options(kmem_usage, kmem_options);
 
        return 0;
 }