perf session: Create kernel maps in the constructor
in [Kernel]
|
Prev: [PATCH] PM / i915: Skip kernel VT switch during suspend/resume if KMS is used
Next: Warning: dev (ttyS1) tty->count(6) != #fd's(5) in tty_release_dev
From: Masami Hiramatsu on 27 Jan 2010 18:20 Arnaldo Carvalho de Melo wrote: > From: Arnaldo Carvalho de Melo <acme(a)redhat.com> > > Removing one extra step needed in the tools that need this, fixing a bug > in 'perf probe' where this was not being done. Thanks, BTW, when O_WRONLY should be used? I guess I might misunderstand something in builtin-probe.c, and it should use O_RDONLY ... -- Masami Hiramatsu Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division e-mail: mhiramat(a)redhat.com -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Arnaldo Carvalho de Melo on 27 Jan 2010 18:40 Em Wed, Jan 27, 2010 at 06:17:58PM -0500, Masami Hiramatsu escreveu: > Arnaldo Carvalho de Melo wrote: > > From: Arnaldo Carvalho de Melo <acme(a)redhat.com> > > > > Removing one extra step needed in the tools that need this, fixing a bug > > in 'perf probe' where this was not being done. > > Thanks, > BTW, when O_WRONLY should be used? I guess I might misunderstand something > in builtin-probe.c, and it should use O_RDONLY ... O_RDONLY only when you have a perf.data file, O_WRONLY was thought for 'perf record', then for tools that want only to do symbol lookup it was reused... I think we should have a proper enum and more clearly specify these semantics, as well as adding a MMAP mode so that we can lift the code now in perf top somehow. But for now using O_WRONLY provides what 'perf probe' needs, I guess :-) - Arnaldo -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Masami Hiramatsu on 28 Jan 2010 11:30 Arnaldo Carvalho de Melo wrote: > Em Wed, Jan 27, 2010 at 06:17:58PM -0500, Masami Hiramatsu escreveu: >> Arnaldo Carvalho de Melo wrote: >>> From: Arnaldo Carvalho de Melo <acme(a)redhat.com> >>> >>> Removing one extra step needed in the tools that need this, fixing a bug >>> in 'perf probe' where this was not being done. >> >> Thanks, >> BTW, when O_WRONLY should be used? I guess I might misunderstand something >> in builtin-probe.c, and it should use O_RDONLY ... > > O_RDONLY only when you have a perf.data file, O_WRONLY was thought for > 'perf record', then for tools that want only to do symbol lookup it was > reused... Ah, so that is the mode for perf.data. > I think we should have a proper enum and more clearly specify these > semantics, as well as adding a MMAP mode so that we can lift the code > now in perf top somehow. Yeah, and if you can separate an interface only for handling symbols from debug/elf binaries, it will be helpful for me too. > But for now using O_WRONLY provides what 'perf probe' needs, I guess :-) I think so ;-) Acked-by: Masami Hiramatsu <mhiramat(a)redhat.com> Thank you, -- Masami Hiramatsu Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division e-mail: mhiramat(a)redhat.com -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Arnaldo Carvalho de Melo on 28 Jan 2010 13:40 Em Thu, Jan 28, 2010 at 11:28:29AM -0500, Masami Hiramatsu escreveu: > Yeah, and if you can separate an interface only for handling symbols > from debug/elf binaries, it will be helpful for me too. It was done on this same series, now you can do as the patch at the end of this message, that will be on my next series, does. Testing with this patch: [root(a)doppio ~]# perf probe --list [root(a)doppio ~]# perf probe schedule Added new event: probe:schedule (on schedule+0) You can now use it on all perf tools, such as: perf record -e probe:schedule -a sleep 1 [root(a)doppio ~]# perf record -e probe:schedule -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.158 MB perf.data (~6923 samples) ] [root(a)doppio ~]# perf report | head -9 # Samples: 4 # # Overhead Command Shared Object Symbol # ........ ............... ..................... ...... # 25.00% perf libc-2.10.2.so [.] __GI___libc_read 25.00% kondemand/1 [kernel.kallsyms] [k] kernel_thread_helper 25.00% kondemand/0 [kernel.kallsyms] [k] kernel_thread_helper [root(a)doppio ~]# And what about support for modules? You have everything you need there as well :-) If you do: struct map *mod; struct symbol *sym; mod = map_groups__find_by_name(&session.kmap_groups, MAP__FUNCTION, "[ipv6]"); sym = map__find_symbol_by_name(module, "ip6_xmit"); you'll get ipv6.ko located and loaded, mod->dso->long_name thus will have the file to use with libdwarf and sym->start will have the unrelocated address, and mod->start will have where ipv6 is loaded in kernel space. Also you should be able to find variables by name or address too, just use MAP__VARIABLE where above you see MAP__FUNCTION. - Arnaldo From 387ab05301cce32c8b091ba9feb94c906124e8a8 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo <acme(a)redhat.com> Date: Thu, 28 Jan 2010 16:12:31 -0200 Subject: [PATCH 1/1] perf probe: Don't use a perf_session instance just to resolve symbols MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit With the recent modifications done to untie the session and symbol layers, 'perf probe' now can use just the symbols layer. Cc: Fr�d�ric Weisbecker <fweisbec(a)gmail.com> Cc: Masami Hiramatsu <mhiramat(a)redhat.com> Cc: Mike Galbraith <efault(a)gmx.de> Cc: Peter Zijlstra <a.p.zijlstra(a)chello.nl> Cc: Paul Mackerras <paulus(a)samba.org> Signed-off-by: Arnaldo Carvalho de Melo <acme(a)redhat.com> --- tools/perf/builtin-probe.c | 15 +++++++++------ 1 files changed, 9 insertions(+), 6 deletions(-) diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c index 4fa73ec..cd45e02 100644 --- a/tools/perf/builtin-probe.c +++ b/tools/perf/builtin-probe.c @@ -41,7 +41,6 @@ #include "util/debugfs.h" #include "util/symbol.h" #include "util/thread.h" -#include "util/session.h" #include "util/parse-options.h" #include "util/parse-events.h" /* For debugfs_path */ #include "util/probe-finder.h" @@ -59,7 +58,8 @@ static struct { int nr_probe; struct probe_point probes[MAX_PROBES]; struct strlist *dellist; - struct perf_session *psession; + struct map_groups kmap_groups; + struct map *kmaps[MAP__NR_TYPES]; struct map *kmap; struct line_range line_range; } session; @@ -212,10 +212,13 @@ static void init_vmlinux(void) pr_debug("Use vmlinux: %s\n", symbol_conf.vmlinux_name); if (symbol__init() < 0) die("Failed to init symbol map."); - session.psession = perf_session__new(NULL, O_WRONLY, false); - if (session.psession == NULL) - die("Failed to init perf_session."); - session.kmap = session.psession->vmlinux_maps[MAP__FUNCTION]; + + map_groups__init(&session.kmap_groups); + if (map_groups__create_kernel_maps(&session.kmap_groups, + session.kmaps) < 0) + die("Failed to create kernel maps."); + + session.kmap = session.kmaps[MAP__FUNCTION]; if (!session.kmap) die("Could not find kernel map.\n"); } -- 1.6.2.5 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Masami Hiramatsu on 28 Jan 2010 16:10
Arnaldo Carvalho de Melo wrote: > Em Thu, Jan 28, 2010 at 11:28:29AM -0500, Masami Hiramatsu escreveu: > >> Yeah, and if you can separate an interface only for handling symbols >> from debug/elf binaries, it will be helpful for me too. > > It was done on this same series, now you can do as the patch at the end > of this message, that will be on my next series, does. Great! thank you! > > Testing with this patch: > > [root(a)doppio ~]# perf probe --list > [root(a)doppio ~]# perf probe schedule > Added new event: > probe:schedule (on schedule+0) > > You can now use it on all perf tools, such as: > > perf record -e probe:schedule -a sleep 1 > > [root(a)doppio ~]# perf record -e probe:schedule -a sleep 1 > [ perf record: Woken up 1 times to write data ] > [ perf record: Captured and wrote 0.158 MB perf.data (~6923 samples) ] > [root(a)doppio ~]# perf report | head -9 > # Samples: 4 > # > # Overhead Command Shared Object Symbol > # ........ ............... ..................... ...... > # > 25.00% perf libc-2.10.2.so [.] __GI___libc_read > 25.00% kondemand/1 [kernel.kallsyms] [k] kernel_thread_helper > 25.00% kondemand/0 [kernel.kallsyms] [k] kernel_thread_helper > [root(a)doppio ~]# > > And what about support for modules? You have everything you need there > as well :-) > > If you do: > > struct map *mod; > struct symbol *sym; > > mod = map_groups__find_by_name(&session.kmap_groups, > MAP__FUNCTION, "[ipv6]"); > > sym = map__find_symbol_by_name(module, "ip6_xmit"); > > you'll get ipv6.ko located and loaded, mod->dso->long_name thus will > have the file to use with libdwarf and sym->start will have the > unrelocated address, and mod->start will have where ipv6 is loaded in > kernel space Oh, nice :-) Is that available for the modules which aren't loaded? And yeah, I'd like to support modules and it will requires some enhancement in kprobe-tracer too. .. > Also you should be able to find variables by name or address too, just > use MAP__VARIABLE where above you see MAP__FUNCTION. > > - Arnaldo > > From 387ab05301cce32c8b091ba9feb94c906124e8a8 Mon Sep 17 00:00:00 2001 > From: Arnaldo Carvalho de Melo <acme(a)redhat.com> > Date: Thu, 28 Jan 2010 16:12:31 -0200 > Subject: [PATCH 1/1] perf probe: Don't use a perf_session instance just to resolve symbols > MIME-Version: 1.0 > Content-Type: text/plain; charset=utf-8 > Content-Transfer-Encoding: 8bit > > With the recent modifications done to untie the session and symbol > layers, 'perf probe' now can use just the symbols layer. Could you remove session.kmap too? other parts look good to me. Thank you, > > Cc: Fr�d�ric Weisbecker <fweisbec(a)gmail.com> > Cc: Masami Hiramatsu <mhiramat(a)redhat.com> > Cc: Mike Galbraith <efault(a)gmx.de> > Cc: Peter Zijlstra <a.p.zijlstra(a)chello.nl> > Cc: Paul Mackerras <paulus(a)samba.org> > Signed-off-by: Arnaldo Carvalho de Melo <acme(a)redhat.com> > --- > tools/perf/builtin-probe.c | 15 +++++++++------ > 1 files changed, 9 insertions(+), 6 deletions(-) > > diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c > index 4fa73ec..cd45e02 100644 > --- a/tools/perf/builtin-probe.c > +++ b/tools/perf/builtin-probe.c > @@ -41,7 +41,6 @@ > #include "util/debugfs.h" > #include "util/symbol.h" > #include "util/thread.h" > -#include "util/session.h" > #include "util/parse-options.h" > #include "util/parse-events.h" /* For debugfs_path */ > #include "util/probe-finder.h" > @@ -59,7 +58,8 @@ static struct { > int nr_probe; > struct probe_point probes[MAX_PROBES]; > struct strlist *dellist; > - struct perf_session *psession; > + struct map_groups kmap_groups; > + struct map *kmaps[MAP__NR_TYPES]; > struct map *kmap; > struct line_range line_range; > } session; > @@ -212,10 +212,13 @@ static void init_vmlinux(void) > pr_debug("Use vmlinux: %s\n", symbol_conf.vmlinux_name); > if (symbol__init() < 0) > die("Failed to init symbol map."); > - session.psession = perf_session__new(NULL, O_WRONLY, false); > - if (session.psession == NULL) > - die("Failed to init perf_session."); > - session.kmap = session.psession->vmlinux_maps[MAP__FUNCTION]; > + > + map_groups__init(&session.kmap_groups); > + if (map_groups__create_kernel_maps(&session.kmap_groups, > + session.kmaps) < 0) > + die("Failed to create kernel maps."); > + > + session.kmap = session.kmaps[MAP__FUNCTION]; > if (!session.kmap) > die("Could not find kernel map.\n"); > } -- Masami Hiramatsu Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division e-mail: mhiramat(a)redhat.com -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |