chiark / gitweb /
tools: add gdb command to dump hashmap information
authorMichal Schmidt <mschmidt@redhat.com>
Tue, 14 Oct 2014 23:31:23 +0000 (01:31 +0200)
committerMichal Schmidt <mschmidt@redhat.com>
Thu, 30 Oct 2014 18:50:51 +0000 (19:50 +0100)
$ sudo gdb -p 1
...
(gdb) source gdb-sd_dump_hashmaps.py
(gdb) sd_dump_hashmaps
  ... lists allocated hashmaps ...
(gdb) sd_dump_hashmaps 1
  ... lists allocated hashmaps, their DIB histograms and contiguous
      blocks statistics ...

Makefile.am
tools/gdb-sd_dump_hashmaps.py [new file with mode: 0644]

index 946e46f..3b273d4 100644 (file)
@@ -6247,3 +6247,6 @@ print-%:
 
 git-contrib:
        @git shortlog -s `git describe --abbrev=0`.. | cut -c8- | awk '{ print $$0 "," }' | sort -u
+
+EXTRA_DIST += \
+        tools/gdb-sd_dump_hashmaps.py
diff --git a/tools/gdb-sd_dump_hashmaps.py b/tools/gdb-sd_dump_hashmaps.py
new file mode 100644 (file)
index 0000000..9ee81fb
--- /dev/null
@@ -0,0 +1,94 @@
+#  -*- Mode: python; coding: utf-8; indent-tabs-mode: nil -*- */
+#
+#  This file is part of systemd.
+#
+#  Copyright 2014 Michal Schmidt
+#
+#  systemd is free software; you can redistribute it and/or modify it
+#  under the terms of the GNU Lesser General Public License as published by
+#  the Free Software Foundation; either version 2.1 of the License, or
+#  (at your option) any later version.
+#
+#  systemd is distributed in the hope that it will be useful, but
+#  WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+#  Lesser General Public License for more details.
+#
+#  You should have received a copy of the GNU Lesser General Public License
+#  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+
+import gdb
+
+class sd_dump_hashmaps(gdb.Command):
+        "dump systemd's hashmaps"
+
+        def __init__(self):
+                super(sd_dump_hashmaps, self).__init__("sd_dump_hashmaps", gdb.COMMAND_DATA, gdb.COMPLETE_NONE)
+
+        def invoke(self, arg, from_tty):
+                d = gdb.parse_and_eval("hashmap_debug_list")
+                all_entry_sizes = gdb.parse_and_eval("all_entry_sizes")
+                all_direct_buckets = gdb.parse_and_eval("all_direct_buckets")
+                hashmap_base_t = gdb.lookup_type("HashmapBase")
+                uchar_t = gdb.lookup_type("unsigned char")
+                ulong_t = gdb.lookup_type("unsigned long")
+                debug_offset = gdb.parse_and_eval("(unsigned long)&((HashmapBase*)0)->debug")
+
+                print "type, hash, indirect, entries, max_entries, buckets, creator"
+                while d:
+                        h = gdb.parse_and_eval("(HashmapBase*)((char*)%d - %d)" % (int(d.cast(ulong_t)), debug_offset))
+
+                        if h["has_indirect"]:
+                                storage_ptr = h["indirect"]["storage"].cast(uchar_t.pointer())
+                                n_entries = h["indirect"]["n_entries"]
+                                n_buckets = h["indirect"]["n_buckets"]
+                        else:
+                                storage_ptr = h["direct"]["storage"].cast(uchar_t.pointer())
+                                n_entries = h["n_direct_entries"]
+                                n_buckets = all_direct_buckets[int(h["type"])];
+
+                        t = ["plain", "ordered", "set"][int(h["type"])]
+
+                        print "%s, %s, %s, %d, %d, %d, %s (%s:%d)" % (t, h["hash_ops"], bool(h["has_indirect"]), n_entries, d["max_entries"], n_buckets, d["func"], d["file"], d["line"])
+
+                        if arg != "" and n_entries > 0:
+                                dib_raw_addr = storage_ptr + (all_entry_sizes[h["type"]] * n_buckets)
+
+                                histogram = {}
+                                for i in xrange(0, n_buckets):
+                                        dib = int(dib_raw_addr[i])
+                                        histogram[dib] = histogram.get(dib, 0) + 1
+
+                                for dib in sorted(iter(histogram)):
+                                        if dib != 255:
+                                                print "%3d %8d %f%% of entries" % (dib, histogram[dib], 100.0*histogram[dib]/n_entries)
+                                        else:
+                                                print "%3d %8d %f%% of slots" % (dib, histogram[dib], 100.0*histogram[dib]/n_buckets)
+                                print "mean DIB of entries: %f" % (sum([dib*histogram[dib] for dib in iter(histogram) if dib != 255])*1.0/n_entries)
+
+                                blocks = []
+                                current_len = 1
+                                prev = int(dib_raw_addr[0])
+                                for i in xrange(1, n_buckets):
+                                        dib = int(dib_raw_addr[i])
+                                        if (dib == 255) != (prev == 255):
+                                                if prev != 255:
+                                                        blocks += [[i, current_len]]
+                                                current_len = 1
+                                        else:
+                                                current_len += 1
+
+                                        prev = dib
+                                if prev != 255:
+                                        blocks += [[i, current_len]]
+                                # a block may be wrapped around
+                                if len(blocks) > 1 and blocks[0][0] == blocks[0][1] and blocks[-1][0] == n_buckets - 1:
+                                        blocks[0][1] += blocks[-1][1]
+                                        blocks = blocks[0:-1]
+                                print "max block: %s" % max(blocks, key=lambda a: a[1])
+                                print "sum block lens: %d" % sum(b[1] for b in blocks)
+                                print "mean block len: %f" % (1.0 * sum(b[1] for b in blocks) / len(blocks))
+
+                        d = d["debug_list_next"]
+
+sd_dump_hashmaps()