linux-perf-users.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Masami Hiramatsu <mhiramat@kernel.org>
To: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com>,
	"linux-perf-use." <linux-perf-users@vger.kernel.org>,
	Hendrik Brueckner <brueckner@linux.vnet.ibm.com>,
	Masami Hiramatsu <mhiramat@kernel.org>
Subject: Re: perf buildid-cache -p <path> question
Date: Sat, 21 Oct 2017 01:12:59 +0900	[thread overview]
Message-ID: <20171021011259.5d6368a7fd1f54004b6ae1da@kernel.org> (raw)
In-Reply-To: <20171016141222.GC5311@kernel.org>

On Mon, 16 Oct 2017 11:12:22 -0300
Arnaldo Carvalho de Melo <acme@kernel.org> wrote:

> Em Mon, Oct 16, 2017 at 02:24:11PM +0200, Thomas-Mich Richter escreveu:
> > Maybe its me misunderstanding the buildid cache completely, but I ran into the following issue:
> > 
> > I can add and remove files using the perf buildid-cache command. For example 
> > perf buildid-cache -a /usr/bin/tar adds the tar executable to the buildid-cache directory
> > .debug/usr/bin/tar 
> > and creates the <buildid> subdirectory
> > .debug/usr/bin/tar/e54c9946802bbbcb85760ffeb80700a5fd35ebe7/elf file.
> 
> That is correct.
> 
> > Also a symbolic link from the directory
> > .debug/.buildid/e5/4c9946802bbbcb85760ffeb80700a5fd35ebe7 --> ../../usr/bin/tar/<buildid>/
> > is created.
> > 
> > Command perf buildid-cache -a /usr/bin/tar
> 
> Nope, command 'perf buildid-cache -r /usr/bin/tar' does it:

Wait, -r is remove command.

> 
> [root@jouet ~]# ls -la ~/.debug/usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> [root@jouet ~]# ls -la ~/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea
> lrwxrwxrwx. 1 root root 58 Oct 16 10:50 /root/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea -> ../../usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> [root@jouet ~]# 
> 
> [root@jouet ~]# perf buildid-cache -r /usr/bin/tar
> [root@jouet ~]# ls -la ~/.debug/usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> ls: cannot access '/root/.debug/usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea': No such file or directory
> [root@jouet ~]# ls -la ~/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea
> ls: cannot access '/root/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea': No such file or directory
> [root@jouet ~]# 
> 
> > removes these entries.
> > 
> > Now when I run ./perf buildid-cache -p /usr/bin
> > nothing happens and success is reported:
> 
> When you do '-p' after a '-a' or after a '-r'? From what you wrote
> below, that is after you do a '-a /usr/bin/tar', as it is being purged
> as well.
> 
> Masami?

Yeah, should be. 

OK, -p and -r is slightly different. That is for handling "versioning" caches.

For example, you have a /usr/bin/tar (this is old tar) and its buildid-cache
in cache directory by perf buildid-cache -a /usr/bin/tar.
And at some point, you will update the /usr/bin/tar (this is new tar).
This new tar binary is not same as old one, so if you also do
perf buildid-cache -a /usr/bin/tar, it creates another version of buildid
cache in cache directory.

At this moment,
$ perf buildid-cache -r /usr/bin/tar
will remove only new tar's buildid cache, but old one remains.

On the other hand,
$ perf buildid-cache -p /usr/bin/tar
will remove all the caches related to /usr/bin/tar from buildid cache.

And since the -p is supposed to be used for cleanup, it doesn't return
error even if there is no buildid cache remaining.

> 
> > [root@s35lp76 perf]# ./perf buildid-cache -vp /usr/bin/
> > Removing bash /usr/bin/: Ok
> > Removing dbus-daemon /usr/bin/: Ok
> > Removing ls /usr/bin/: Ok
> > Removing readlink /usr/bin/: Ok
> > Removing sleep /usr/bin/: Ok
> > Removing tar /usr/bin/: Ok
> > Removing time /usr/bin/: Ok
> > Removing vim /usr/bin/: Ok
> > Purging /usr/bin/: Ok

Hmm, curious...

> 
> In my test it works as advertised:
> 
> [root@jouet ~]# perf buildid-cache -a /usr/bin/tar
> [root@jouet ~]# ls -la ~/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea
> lrwxrwxrwx. 1 root root 58 Oct 16 10:56 /root/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea -> ../../usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> [root@jouet ~]# ls -la ~/.debug/usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> total 424
> drwxr-xr-x. 2 root root   4096 Oct 16 10:56 .
> drwxr-xr-x. 3 root root   4096 Oct 16 10:56 ..
> -rwxr-xr-x. 2 root root 425032 Jul 19 05:55 elf
> -rw-r--r--. 1 root root      0 Oct 16 10:56 probes
> [root@jouet ~]# perf buildid-cache -vp /usr/bin/tar
> Removing 4c10f253c0cb8459dd846235c0bc18929e4758ea /usr/bin/tar: Ok
> Purging /usr/bin/tar: Ok
> [root@jouet ~]# ls -la ~/.debug/usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> ls: cannot access '/root/.debug/usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea': No such file or directory
> [root@jouet ~]# ls -la ~/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea
> ls: cannot access '/root/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea': No such file or directory
> [root@jouet ~]#
> 
> It was purged.

OK. it seems good.

> 
> Now lets try what you did, which is to ask for a path that will match
> with multiple binaries:
> 
> [root@jouet ~]# perf buildid-cache -v -a /usr/bin/tar
> Adding 4c10f253c0cb8459dd846235c0bc18929e4758ea /usr/bin/tar: Ok
> [root@jouet ~]# ls -la ~/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea
> lrwxrwxrwx. 1 root root 58 Oct 16 10:58 /root/.debug/.build-id/4c/10f253c0cb8459dd846235c0bc18929e4758ea -> ../../usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> [root@jouet ~]# ls -la ~/.debug/usr/bin/tar/4c10f253c0cb8459dd846235c0bc18929e4758ea
> total 424
> drwxr-xr-x. 2 root root   4096 Oct 16 10:58 .
> drwxr-xr-x. 3 root root   4096 Oct 16 10:58 ..
> -rwxr-xr-x. 2 root root 425032 Jul 19 05:55 elf
> -rw-r--r--. 1 root root      0 Oct 16 10:58 probes
> [root@jouet ~]# perf buildid-cache -vp /usr/bin
> Removing as /usr/bin: FAIL
> Purging /usr/bin: FAIL
> /usr/bin wasn't in the cache

I think -p does only support a file not a path like that.

> [root@jouet ~]# perf buildid-cache -vp /usr/bin/
> Removing as /usr/bin/: FAIL
> Purging /usr/bin/: FAIL
> /usr/bin/ wasn't in the cache
> [root@jouet ~]#
> 
> This is when --purge was added:
> 
> commit 8d8c8e4cb3014fcc51f0e127b4316043306f5bb0
> Author: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> Date:   Fri Feb 27 13:50:26 2015 +0900
> 
>     perf buildid-cache: Add --purge FILE to remove all caches of FILE
> 
> -----------------------
>  
> > I have done some more debugging and there is something wrong. 
> > The function
> > build_id_cache__purge_path() is called and 
> > build_id_cache__list_build_ids() creates a list of file names located in /usr/bin of the buildid-cache.
> > build_id_cache__remove_s() is called for each name in the list and tries to locate each file name
> > 	in directory .buildid/YY/ZZZ..ZZZ which fails because 
> > build_id_cache__linkname() expects a buildid and gets a file name.
> > The file name bash is treated as .buildid/ba/sh which does not exist.
> 
> probably this got broken when the /elf, /probe separation was added,
> humm, unsure if a 'git bisect' would help here...
> 
> Masami?

Should it work for a directory instead of a file too?
I need to investigate it too.

Thanks,

>  
> > This seems to be a simple mistake so I wonder if I do something wrong.
> 
> - Arnaldo


-- 
Masami Hiramatsu <mhiramat@kernel.org>

  reply	other threads:[~2017-10-20 16:13 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-10-16 12:24 perf buildid-cache -p <path> question Thomas-Mich Richter
2017-10-16 13:45 ` Thomas-Mich Richter
2017-10-16 14:12 ` Arnaldo Carvalho de Melo
2017-10-20 16:12   ` Masami Hiramatsu [this message]
2017-10-20 16:20     ` Arnaldo Carvalho de Melo
2017-10-24  7:56       ` Thomas-Mich Richter
2017-10-25 17:04         ` Masami Hiramatsu
2017-10-26  7:05           ` Thomas-Mich Richter
2017-10-21 15:17 ` Masami Hiramatsu
2017-10-24  6:50   ` Thomas-Mich Richter
2017-10-25 16:57     ` Masami Hiramatsu
2017-10-26  7:07       ` Thomas-Mich Richter
2017-10-26 14:41         ` Masami Hiramatsu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20171021011259.5d6368a7fd1f54004b6ae1da@kernel.org \
    --to=mhiramat@kernel.org \
    --cc=acme@kernel.org \
    --cc=brueckner@linux.vnet.ibm.com \
    --cc=linux-perf-users@vger.kernel.org \
    --cc=tmricht@linux.vnet.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).