linux-trace-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Francis Laniel <flaniel@linux.microsoft.com>
To: Masami Hiramatsu <mhiramat@kernel.org>
Cc: linux-kernel@vger.kernel.org,
	Steven Rostedt <rostedt@goodmis.org>,
	linux-trace-kernel@vger.kernel.org
Subject: Re: [RFC PATCH v1 1/1] tracing/kprobe: Add multi-probe support for 'perf_kprobe' PMU
Date: Fri, 18 Aug 2023 20:12:11 +0200	[thread overview]
Message-ID: <5702263.DvuYhMxLoT@pwmachine> (raw)
In-Reply-To: <20230818220537.75ce8210c6a4c80a5a8d16f8@kernel.org>

Hi.

Le vendredi 18 août 2023, 15:05:37 CEST Masami Hiramatsu a écrit :
> On Thu, 17 Aug 2023 13:06:20 +0200
> 
> Francis Laniel <flaniel@linux.microsoft.com> wrote:
> > Hi.
> > 
> > Le jeudi 17 août 2023, 09:50:57 CEST Masami Hiramatsu a écrit :
> > > Hi,
> > > 
> > > On Wed, 16 Aug 2023 18:35:17 +0200
> > > 
> > > Francis Laniel <flaniel@linux.microsoft.com> wrote:
> > > > When using sysfs, it is possible to create kprobe for several kernel
> > > > functions sharing the same name, but of course with different
> > > > addresses,
> > > > by writing their addresses in kprobe_events file.
> > > > 
> > > > When using PMU, if only the symbol name is given, the event will be
> > > > created for the first address which matches the symbol, as returned by
> > > > kallsyms_lookup_name().
> > > 
> > > Do you mean probing the same name symbols? Yes, it is intended behavior,
> > > since it is not always true that the same name function has the same
> > > prototype (it is mostly true but is not ensured), it is better to leave
> > > user to decide which one is what you want to probe.
> > 
> > This is what I meant.
> > I also share your mind regarding leaving the users deciding which one they
> > want to probe but in my case (which I agree is a bit a corner one) it
> > leaded me to misunderstanding as the PMU kprobe was only added to the
> > first ntfs_file_write_iter() which is not the one for ntfs3.
> 
> Hmm, OK. I think in that case (multiple same-name symbols exist) the default
> behavior is rejecting with error message. And optionally, it will probe all
> or them like your patch.

I am not sure to understand.
Can you please precise the default behavior of which software component?

> > > Have you used 'perf probe' tool? It tries to find the appropriate
> > > function
> > > by line number and creates the probe by 'text+OFFSET' style, not by
> > > symbol.
> > > I think this is the correct way to do that, because user will not know
> > > which 'address' of the symbol is what the user want.
> > 
> > 'perf probe' perfectly does the trick, as it would find all the kernel
> > addresses which correspond to the symbol name and create as many probes as
> > corresponding symbols [1]:
> > root@vm-amd64:~# perf probe --add ntfs_file_write_iter
> 
> If you can specify the (last part of) file path as below,
> 
> perf probe --add ntfs_file_write_iter@ntfs3/file.c
> 
> Then it will choose correct one. :)

Nice! TIL thank you! perf is really powerful!

> > Added new events:
> >   probe:ntfs_file_write_iter (on ntfs_file_write_iter)
> >   probe:ntfs_file_write_iter (on ntfs_file_write_iter)
> > 
> > You can now use it in all perf tools, such as:
> >         perf record -e probe:ntfs_file_write_iter -aR sleep 1
> > 
> > root@vm-amd64:~# cat /sys/kernel/tracing/kprobe_events
> > p:probe/ntfs_file_write_iter _text+5088544
> > p:probe/ntfs_file_write_iter _text+5278560
> > 
> > > Thought?
> > 
> > This contribution is basically here to sort of mimic what perf does but
> > with PMU kprobes, as this is not possible to write in a sysfs file with
> > this type of probe.
> 
> OK, I see it is for BPF only. Maybe BPF program can filter correct one
> to access the argument etc.

I am not sure I understand, can you please precise?
The eBPF program will be run when the kprobe will be triggered, so if the 
kprobe is armed for the function (e.g. old ntfs_file_write_iter()), the eBPF 
program will never be called.

> 
> Thank you,
> 
> > > Thank you,
> > > 
> > > > The idea here is to search all kernel functions which match this
> > > > symbol
> > > > and
> > > > create a trace_kprobe for each of them.
> > > > All these trace_kprobes are linked together by sharing the same
> > > > trace_probe.
> > > > 
> > > > Signed-off-by: Francis Laniel <flaniel@linux.microsoft.com>
> > > > ---
> > > > 
> > > >  kernel/trace/trace_kprobe.c | 86
> > > >  +++++++++++++++++++++++++++++++++++++
> > > >  1 file changed, 86 insertions(+)
> > > > 
> > > > diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
> > > > index 1b3fa7b854aa..08580f1466c7 100644
> > > > --- a/kernel/trace/trace_kprobe.c
> > > > +++ b/kernel/trace/trace_kprobe.c
> > > > @@ -1682,13 +1682,42 @@ static int unregister_kprobe_event(struct
> > > > trace_kprobe *tk)>
> > > > 
> > > >  }
> > > >  
> > > >  #ifdef CONFIG_PERF_EVENTS
> > > > 
> > > > +
> > > > +struct address_array {
> > > > +	unsigned long *addrs;
> > > > +	size_t size;
> > > > +};
> > > > +
> > > > +static int add_addr(void *data, unsigned long addr)
> > > > +{
> > > > +	struct address_array *array = data;
> > > > +	unsigned long *p;
> > > > +
> > > > +	array->size++;
> > > > +	p = krealloc(array->addrs,
> > > > +				sizeof(*array->addrs) * array->size,
> > > > +				GFP_KERNEL);
> > > > +	if (!p) {
> > > > +		kfree(array->addrs);
> > > > +		return -ENOMEM;
> > > > +	}
> > > > +
> > > > +	array->addrs = p;
> > > > +	array->addrs[array->size - 1] = addr;
> > > > +
> > > > +	return 0;
> > > > +}
> > > > +
> > > > 
> > > >  /* create a trace_kprobe, but don't add it to global lists */
> > > >  struct trace_event_call *
> > > >  create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
> > > >  
> > > >  			  bool is_return)
> > > >  
> > > >  {
> > > >  
> > > >  	enum probe_print_type ptype;
> > > > 
> > > > +	struct address_array array;
> > > > 
> > > >  	struct trace_kprobe *tk;
> > > > 
> > > > +	unsigned long func_addr;
> > > > +	unsigned int i;
> > > > 
> > > >  	int ret;
> > > >  	char *event;
> > > > 
> > > > @@ -1722,7 +1751,64 @@ create_local_trace_kprobe(char *func, void
> > > > *addr,
> > > > unsigned long offs,>
> > > > 
> > > >  	if (ret < 0)
> > > >  	
> > > >  		goto error;
> > > > 
> > > > +	array.addrs = NULL;
> > > > +	array.size = 0;
> > > > +	ret = kallsyms_on_each_match_symbol(add_addr, func, &array);
> > > > +	if (ret)
> > > > +		goto error_free;
> > > > +
> > > > +	if (array.size == 1)
> > > > +		goto end;
> > > > +
> > > > +	/*
> > > > +	 * Below loop allocates a trace_kprobe for each function with the
> > > > same
> > > > +	 * name in kernel source code.
> > > > +	 * All this differente trace_kprobes will be linked together through
> > > > +	 * append_trace_kprobe().
> > > > +	 * NOTE append_trace_kprobe() is called in register_trace_kprobe()
> > 
> > which
> > 
> > > > +	 * is called when a kprobe is added through sysfs.
> > > > +	 */
> > > > +	func_addr = kallsyms_lookup_name(func);
> > > > +	for (i = 0; i < array.size; i++) {
> > > > +		struct trace_kprobe *tk_same_name;
> > > > +		unsigned long address;
> > > > +
> > > > +		address = array.addrs[i];
> > > > +		/* Skip the function address as we already registered it. */
> > > > +		if (address == func_addr)
> > > > +			continue;
> > > > +
> > > > +		/*
> > > > +		 * alloc_trace_kprobe() first considers symbol name, so we set
> > > > +		 * this to NULL to allocate this kprobe on the given address.
> > > > +		 */
> > > > +		tk_same_name = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event,
> > > > +						  (void *)address, NULL, offs,
> > > > +						  0 /* maxactive */,
> > > > +						  0 /* nargs */, is_return);
> > > > +
> > > > +		if (IS_ERR(tk_same_name)) {
> > > > +			ret = -ENOMEM;
> > > > +			goto error_free;
> > > > +		}
> > > > +
> > > > +		init_trace_event_call(tk_same_name);
> > > > +
> > > > +		if (traceprobe_set_print_fmt(&tk_same_name->tp, ptype) < 0) {
> > > > +			ret = -ENOMEM;
> > > > +			goto error_free;
> > > > +		}
> > > > +
> > > > +		ret = append_trace_kprobe(tk_same_name, tk);
> > > > +		if (ret)
> > > > +			goto error_free;
> > > > +	}
> > > > +
> > > > +end:
> > > > +	kfree(array.addrs);
> > > > 
> > > >  	return trace_probe_event_call(&tk->tp);
> > > > 
> > > > +error_free:
> > > > +	kfree(array.addrs);
> > > > 
> > > >  error:
> > > >  	free_trace_kprobe(tk);
> > > >  	return ERR_PTR(ret);
> > 
> > ---
> > [1]: https://github.com/torvalds/linux/blob/
> > 57012c57536f8814dec92e74197ee96c3498d24e/tools/perf/util/probe-event.c#L29
> > 89- L2993





  reply	other threads:[~2023-08-18 18:12 UTC|newest]

Thread overview: 35+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <20230816163517.112518-1-flaniel@linux.microsoft.com>
2023-08-16 16:35 ` [RFC PATCH v1 1/1] tracing/kprobe: Add multi-probe support for 'perf_kprobe' PMU Francis Laniel
2023-08-16 18:42   ` Steven Rostedt
2023-08-17 10:59     ` Francis Laniel
2023-08-17 15:13       ` Steven Rostedt
2023-08-18  9:01         ` Francis Laniel
2023-08-18 12:37         ` Masami Hiramatsu
2023-08-18 15:41           ` Steven Rostedt
2023-08-18 18:13             ` Francis Laniel
2023-08-18 18:20               ` Steven Rostedt
2023-08-19  1:15                 ` Masami Hiramatsu
2023-08-19 15:22                   ` Song Liu
2023-08-20  9:32                     ` Masami Hiramatsu
2023-08-20 10:02                       ` Song Liu
2023-08-20 13:16                         ` Masami Hiramatsu
2023-08-21  6:09                           ` Song Liu
2023-08-21 10:01                             ` Masami Hiramatsu
2023-08-21 14:45                               ` Steven Rostedt
2023-08-21 18:07                                 ` Kees Cook
2023-08-21 14:29                         ` Steven Rostedt
2023-08-21 15:19                           ` Masami Hiramatsu
2023-08-21 15:28                             ` Steven Rostedt
2023-08-17  7:50   ` Masami Hiramatsu
2023-08-17 11:06     ` Francis Laniel
2023-08-18 13:05       ` Masami Hiramatsu
2023-08-18 18:12         ` Francis Laniel [this message]
2023-08-19  1:11           ` Masami Hiramatsu
2023-08-20 20:23             ` Jiri Olsa
2023-08-21 12:22               ` Francis Laniel
2023-08-20 20:34             ` Jiri Olsa
2023-08-21 12:24               ` Francis Laniel
2023-08-22 13:13                 ` Jiri Olsa
2023-08-21 12:55             ` Francis Laniel
2023-08-23  0:36               ` Masami Hiramatsu
2023-08-23  9:54                 ` Francis Laniel
2023-08-23 13:45                   ` 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=5702263.DvuYhMxLoT@pwmachine \
    --to=flaniel@linux.microsoft.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-trace-kernel@vger.kernel.org \
    --cc=mhiramat@kernel.org \
    --cc=rostedt@goodmis.org \
    /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).