perf: mmap 512 kiB by default
Frederic Weisbecker [Thu, 31 Mar 2011 01:35:24 +0000 (03:35 +0200)]
The default setting of perf record is to mmap 128 pages if the user
did not override with -m.

However the page size may vary accross different architecture
settings, giving different default size between each.

Moreover the kernel side still has a default max number of mlocked
pages of 512 kiB + 1 page for unprivileged users. 128 + 1 pages
with page size > 4096 overlaps this threshold.

Thus, better adapt to this limitation and set the default number of
pages to fit those 512 kiB + 1 page.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Stephane Eranian <eranian@google.com>
LKML-Reference: <1301535324-9735-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tools/perf/builtin-record.c

index db6adec..17d1dcb 100644 (file)
@@ -41,7 +41,7 @@ static u64                    user_interval                   = ULLONG_MAX;
 static u64                     default_interval                =      0;
 
 static unsigned int            page_size;
-static unsigned int            mmap_pages                      =    128;
+static unsigned int            mmap_pages                      = UINT_MAX;
 static unsigned int            user_freq                       = UINT_MAX;
 static int                     freq                            =   1000;
 static int                     output;
@@ -513,6 +513,10 @@ static int __cmd_record(int argc, const char **argv)
        if (have_tracepoints(&evsel_list->entries))
                perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
 
+       /* 512 kiB: default amount of unprivileged mlocked memory */
+       if (mmap_pages == UINT_MAX)
+               mmap_pages = (512 * 1024) / page_size;
+
        if (forks) {
                child_pid = fork();
                if (child_pid < 0) {