perf evlist: Update mmap related APIs and helpers
authorWang Nan <wangnan0@huawei.com>
Thu, 14 Jul 2016 08:34:35 +0000 (08:34 +0000)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 15 Jul 2016 20:27:46 +0000 (17:27 -0300)
Currently, the evlist mmap related helpers and APIs accept evlist and
idx, and dereference 'struct perf_mmap' by evlist->mmap[idx]. This is
unnecessary, and force each evlist contains only one mmap array.

Following commits are going to introduce multiple mmap arrays to a
evlist.  This patch refators these APIs and helpers, introduces
functions accept perf_mmap pointer directly. New helpers and APIs are
decoupled with perf_evlist, and become perf_mmap functions (so they have
perf_mmap prefix).

Old functions are reimplemented with new functions. Some of them will be
removed in following commits.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: He Kuang <hekuang@huawei.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Nilay Vaish <nilayvaish@gmail.com>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1468485287-33422-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/evlist.c
tools/perf/util/evlist.h

index 6803f5ccd15e5412b7e6d75a299a8797e438aea8..a4137e02eab8e3e0b246e10812f758d8c0446df0 100644 (file)
@@ -29,6 +29,7 @@
 
 static void perf_evlist__mmap_put(struct perf_evlist *evlist, int idx);
 static void __perf_evlist__munmap(struct perf_evlist *evlist, int idx);
+static void perf_mmap__munmap(struct perf_mmap *map);
 
 #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
 #define SID(e, x, y) xyarray__entry(e->sample_id, x, y)
@@ -781,9 +782,8 @@ broken_event:
        return event;
 }
 
-union perf_event *perf_evlist__mmap_read_forward(struct perf_evlist *evlist, int idx)
+union perf_event *perf_mmap__read_forward(struct perf_mmap *md, bool check_messup)
 {
-       struct perf_mmap *md = &evlist->mmap[idx];
        u64 head;
        u64 old = md->prev;
 
@@ -795,13 +795,12 @@ union perf_event *perf_evlist__mmap_read_forward(struct perf_evlist *evlist, int
 
        head = perf_mmap__read_head(md);
 
-       return perf_mmap__read(md, evlist->overwrite, old, head, &md->prev);
+       return perf_mmap__read(md, check_messup, old, head, &md->prev);
 }
 
 union perf_event *
-perf_evlist__mmap_read_backward(struct perf_evlist *evlist, int idx)
+perf_mmap__read_backward(struct perf_mmap *md)
 {
-       struct perf_mmap *md = &evlist->mmap[idx];
        u64 head, end;
        u64 start = md->prev;
 
@@ -836,6 +835,31 @@ perf_evlist__mmap_read_backward(struct perf_evlist *evlist, int idx)
        return perf_mmap__read(md, false, start, end, &md->prev);
 }
 
+union perf_event *perf_evlist__mmap_read_forward(struct perf_evlist *evlist, int idx)
+{
+       struct perf_mmap *md = &evlist->mmap[idx];
+
+       /*
+        * Check messup is required for forward overwritable ring buffer:
+        * memory pointed by md->prev can be overwritten in this case.
+        * No need for read-write ring buffer: kernel stop outputting when
+        * it hit md->prev (perf_mmap__consume()).
+        */
+       return perf_mmap__read_forward(md, evlist->overwrite);
+}
+
+union perf_event *perf_evlist__mmap_read_backward(struct perf_evlist *evlist, int idx)
+{
+       struct perf_mmap *md = &evlist->mmap[idx];
+
+       /*
+        * No need to check messup for backward ring buffer:
+        * We can always read arbitrary long data from a backward
+        * ring buffer unless we forget to pause it before reading.
+        */
+       return perf_mmap__read_backward(md);
+}
+
 union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx)
 {
        if (!evlist->backward)
@@ -843,9 +867,8 @@ union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx)
        return perf_evlist__mmap_read_backward(evlist, idx);
 }
 
-void perf_evlist__mmap_read_catchup(struct perf_evlist *evlist, int idx)
+void perf_mmap__read_catchup(struct perf_mmap *md)
 {
-       struct perf_mmap *md = &evlist->mmap[idx];
        u64 head;
 
        if (!atomic_read(&md->refcnt))
@@ -855,38 +878,54 @@ void perf_evlist__mmap_read_catchup(struct perf_evlist *evlist, int idx)
        md->prev = head;
 }
 
+void perf_evlist__mmap_read_catchup(struct perf_evlist *evlist, int idx)
+{
+       perf_mmap__read_catchup(&evlist->mmap[idx]);
+}
+
 static bool perf_mmap__empty(struct perf_mmap *md)
 {
        return perf_mmap__read_head(md) == md->prev && !md->auxtrace_mmap.base;
 }
 
-static void perf_evlist__mmap_get(struct perf_evlist *evlist, int idx)
+static void perf_mmap__get(struct perf_mmap *map)
 {
-       atomic_inc(&evlist->mmap[idx].refcnt);
+       atomic_inc(&map->refcnt);
 }
 
-static void perf_evlist__mmap_put(struct perf_evlist *evlist, int idx)
+static void perf_mmap__put(struct perf_mmap *md)
 {
-       struct perf_mmap *md = &evlist->mmap[idx];
-
        BUG_ON(md->base && atomic_read(&md->refcnt) == 0);
 
        if (atomic_dec_and_test(&md->refcnt))
-               __perf_evlist__munmap(evlist, idx);
+               perf_mmap__munmap(md);
 }
 
-void perf_evlist__mmap_consume(struct perf_evlist *evlist, int idx)
+static void perf_evlist__mmap_get(struct perf_evlist *evlist, int idx)
 {
-       struct perf_mmap *md = &evlist->mmap[idx];
+       perf_mmap__get(&evlist->mmap[idx]);
+}
+
+static void perf_evlist__mmap_put(struct perf_evlist *evlist, int idx)
+{
+       perf_mmap__put(&evlist->mmap[idx]);
+}
 
-       if (!evlist->overwrite) {
+void perf_mmap__consume(struct perf_mmap *md, bool overwrite)
+{
+       if (!overwrite) {
                u64 old = md->prev;
 
                perf_mmap__write_tail(md, old);
        }
 
        if (atomic_read(&md->refcnt) == 1 && perf_mmap__empty(md))
-               perf_evlist__mmap_put(evlist, idx);
+               perf_mmap__put(md);
+}
+
+void perf_evlist__mmap_consume(struct perf_evlist *evlist, int idx)
+{
+       perf_mmap__consume(&evlist->mmap[idx], evlist->overwrite);
 }
 
 int __weak auxtrace_mmap__mmap(struct auxtrace_mmap *mm __maybe_unused,
@@ -917,15 +956,20 @@ void __weak auxtrace_mmap_params__set_idx(
 {
 }
 
-static void __perf_evlist__munmap(struct perf_evlist *evlist, int idx)
+static void perf_mmap__munmap(struct perf_mmap *map)
 {
-       if (evlist->mmap[idx].base != NULL) {
-               munmap(evlist->mmap[idx].base, evlist->mmap_len);
-               evlist->mmap[idx].base = NULL;
-               evlist->mmap[idx].fd = -1;
-               atomic_set(&evlist->mmap[idx].refcnt, 0);
+       if (map->base != NULL) {
+               munmap(map->base, perf_mmap__mmap_len(map));
+               map->base = NULL;
+               map->fd = -1;
+               atomic_set(&map->refcnt, 0);
        }
-       auxtrace_mmap__munmap(&evlist->mmap[idx].auxtrace_mmap);
+       auxtrace_mmap__munmap(&map->auxtrace_mmap);
+}
+
+static void __perf_evlist__munmap(struct perf_evlist *evlist, int idx)
+{
+       perf_mmap__munmap(&evlist->mmap[idx]);
 }
 
 void perf_evlist__munmap(struct perf_evlist *evlist)
@@ -941,20 +985,21 @@ void perf_evlist__munmap(struct perf_evlist *evlist)
        zfree(&evlist->mmap);
 }
 
-static int perf_evlist__alloc_mmap(struct perf_evlist *evlist)
+static struct perf_mmap *perf_evlist__alloc_mmap(struct perf_evlist *evlist)
 {
        int i;
+       struct perf_mmap *map;
 
        evlist->nr_mmaps = cpu_map__nr(evlist->cpus);
        if (cpu_map__empty(evlist->cpus))
                evlist->nr_mmaps = thread_map__nr(evlist->threads);
-       evlist->mmap = zalloc(evlist->nr_mmaps * sizeof(struct perf_mmap));
-       if (!evlist->mmap)
-               return -ENOMEM;
+       map = zalloc(evlist->nr_mmaps * sizeof(struct perf_mmap));
+       if (!map)
+               return NULL;
 
        for (i = 0; i < evlist->nr_mmaps; i++)
-               evlist->mmap[i].fd = -1;
-       return 0;
+               map[i].fd = -1;
+       return map;
 }
 
 struct mmap_params {
@@ -963,8 +1008,8 @@ struct mmap_params {
        struct auxtrace_mmap_params auxtrace_mp;
 };
 
-static int __perf_evlist__mmap(struct perf_evlist *evlist, int idx,
-                              struct mmap_params *mp, int fd)
+static int perf_mmap__mmap(struct perf_mmap *map,
+                          struct mmap_params *mp, int fd)
 {
        /*
         * The last one will be done at perf_evlist__mmap_consume(), so that we
@@ -979,26 +1024,32 @@ static int __perf_evlist__mmap(struct perf_evlist *evlist, int idx,
         * evlist layer can't just drop it when filtering events in
         * perf_evlist__filter_pollfd().
         */
-       atomic_set(&evlist->mmap[idx].refcnt, 2);
-       evlist->mmap[idx].prev = 0;
-       evlist->mmap[idx].mask = mp->mask;
-       evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, mp->prot,
-                                     MAP_SHARED, fd, 0);
-       if (evlist->mmap[idx].base == MAP_FAILED) {
+       atomic_set(&map->refcnt, 2);
+       map->prev = 0;
+       map->mask = mp->mask;
+       map->base = mmap(NULL, perf_mmap__mmap_len(map), mp->prot,
+                        MAP_SHARED, fd, 0);
+       if (map->base == MAP_FAILED) {
                pr_debug2("failed to mmap perf event ring buffer, error %d\n",
                          errno);
-               evlist->mmap[idx].base = NULL;
+               map->base = NULL;
                return -1;
        }
-       evlist->mmap[idx].fd = fd;
+       map->fd = fd;
 
-       if (auxtrace_mmap__mmap(&evlist->mmap[idx].auxtrace_mmap,
-                               &mp->auxtrace_mp, evlist->mmap[idx].base, fd))
+       if (auxtrace_mmap__mmap(&map->auxtrace_mmap,
+                               &mp->auxtrace_mp, map->base, fd))
                return -1;
 
        return 0;
 }
 
+static int __perf_evlist__mmap(struct perf_evlist *evlist, int idx,
+                              struct mmap_params *mp, int fd)
+{
+       return perf_mmap__mmap(&evlist->mmap[idx], mp, fd);
+}
+
 static bool
 perf_evlist__should_poll(struct perf_evlist *evlist __maybe_unused,
                         struct perf_evsel *evsel)
@@ -1248,7 +1299,9 @@ int perf_evlist__mmap_ex(struct perf_evlist *evlist, unsigned int pages,
                .prot = PROT_READ | (overwrite ? 0 : PROT_WRITE),
        };
 
-       if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0)
+       if (!evlist->mmap)
+               evlist->mmap = perf_evlist__alloc_mmap(evlist);
+       if (!evlist->mmap)
                return -ENOMEM;
 
        if (evlist->pollfd.entries == NULL && perf_evlist__alloc_pollfd(evlist) < 0)
index afd087761a475b284ccb571e022391689d15c4fb..9e680c62c9800472af3c76d96ffec1ca84c6434c 100644 (file)
@@ -35,6 +35,12 @@ struct perf_mmap {
        char             event_copy[PERF_SAMPLE_MAX_SIZE] __attribute__((aligned(8)));
 };
 
+static inline size_t
+perf_mmap__mmap_len(struct perf_mmap *map)
+{
+       return map->mask + 1 + page_size;
+}
+
 struct perf_evlist {
        struct list_head entries;
        struct hlist_head heads[PERF_EVLIST__HLIST_SIZE];
@@ -129,6 +135,12 @@ struct perf_evsel *perf_evlist__id2evsel_strict(struct perf_evlist *evlist,
 
 struct perf_sample_id *perf_evlist__id2sid(struct perf_evlist *evlist, u64 id);
 
+union perf_event *perf_mmap__read_forward(struct perf_mmap *map, bool check_messup);
+union perf_event *perf_mmap__read_backward(struct perf_mmap *map);
+
+void perf_mmap__read_catchup(struct perf_mmap *md);
+void perf_mmap__consume(struct perf_mmap *md, bool overwrite);
+
 union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx);
 
 union perf_event *perf_evlist__mmap_read_forward(struct perf_evlist *evlist,