[<prev] [next>] [day] [month] [year] [list]
Message-ID: <1643082961-8535-1-git-send-email-xiaojiangfeng@huawei.com>
Date: Tue, 25 Jan 2022 11:56:01 +0800
From: Jiangfeng Xiao <xiaojiangfeng@...wei.com>
To: <peterz@...radead.org>, <mingo@...hat.com>, <acme@...nel.org>,
<mark.rutland@....com>, <alexander.shishkin@...ux.intel.com>,
<jolsa@...hat.com>, <namhyung@...nel.org>, <james.clark@....com>,
<leo.yan@...aro.org>, <qiuxi1@...wei.com>, <wangbing6@...wei.com>,
<nixiaoming@...wei.com>, <xiaojiangfeng@...wei.com>
CC: <linux-perf-users@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: [PATCH] [PING] perf dso: Avoid copying redundant kallsyms in th buildid_dir
The issue is occurs in following step:
(1)copy the kernel module (such as test.ko) to /root directory
(2)load the kernel module (such as insmod test.ko)
(3)run perf-record as: perf record sleep 1
Then /proc/kallsyms will be copied not only to
/root/.debug/[kernel.kallsyms]/<buildid>/kallsyms,
but also be copied to
/root/.debug/[test]/<buildid>/kallsyms(this is redundant).
Say a kallsyms file content takes 5MB memory,
100 out-of-tree kernel modules will take 500MB,
which is not bearable on embedded environment.
__cmd_record():
__perf_session__new():
perf_session__create_kernel_maps():
machine__create_kernel_maps():
machine__create_modules():
machine__set_modules_path(): perf will only search for kernel
modules in "/lib/modules/`uname -r`/"
/root/test.ko is not in scope.
maps__set_modules_path_dir():
maps__set_module_path():
dso__set_long_name(): the dso->long_name of test.ko has
not changed, it is still [test].
record__finish_output():
perf_session__write_header():
[...]
dso__cache_build_id():
bool is_kallsyms = dso__is_kallsyms(dso);
dso__is_kallsyms(): return dso->kernel&&dso->long_name[0] != '/';
[test] will be wrongly judged as kallsyms,
and then the value of is_kallsyms is true.
build_id_cache__add_b(, is_kallsyms,):
build_id_cache__add_s(, is_kallsyms,):
build_id_cache__add(, is_kallsyms,):
if (is_kallsyms)
copyfile("/proc/kallsyms", filename): /proc/kallsyms will
be copied to /root/.debug/[test]/<buildid>/kallsyms.
Why can this modification can avoid the redundancy of kallsyms:
if (dso->kernel == DSO_SPACE__KERNEL) there are only two possibilities,
one is dso__is_kallsyms, and the other is dso__is_kmod. After
modification, [test] is no longer incorrectly judged as kallsyms.
Why can dso__is_kmod judge correctly?
__cmd_record():
__perf_session__new():
perf_session__create_kernel_maps():
machine__create_kernel_maps():
machine__create_modules():
modules__parse(): for each line in /proc/modules,
not just in "/lib/modules/`uname -r`/".
machine__create_module():
machine__addnew_module_map():
machine__findnew_module_dso():
dso__set_module_info(): assign dso->symtab_type to
DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE and so on.
Reported-by: Lexi Shao <shaolexi@...wei.com>
Signed-off-by: Jiangfeng Xiao <xiaojiangfeng@...wei.com>
---
tools/perf/util/dso.h | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/tools/perf/util/dso.h b/tools/perf/util/dso.h
index 011da39..57a6ab5 100644
--- a/tools/perf/util/dso.h
+++ b/tools/perf/util/dso.h
@@ -380,9 +380,17 @@ static inline bool dso__is_kcore(struct dso *dso)
dso->binary_type == DSO_BINARY_TYPE__GUEST_KCORE;
}
+static inline bool dso__is_kmod(struct dso *dso)
+{
+ return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
+ dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP ||
+ dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE ||
+ dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE_COMP;
+}
+
static inline bool dso__is_kallsyms(struct dso *dso)
{
- return dso->kernel && dso->long_name[0] != '/';
+ return dso->kernel && !dso__is_kmod(dso);
}
void dso__free_a2l(struct dso *dso);
--
1.8.5.6
Powered by blists - more mailing lists