Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210113091657.1456216-1-glider@google.com> <20210113091657.1456216-2-glider@google.com>
 <20210113161044.43bc1c1a@gandalf.local.home> <CAG_fn=XSkOChCwBp=Vg6jWhZ8K44seCo=0Zu38iUpAj6eCUxjQ@mail.gmail.com>
In-Reply-To: <CAG_fn=XSkOChCwBp=Vg6jWhZ8K44seCo=0Zu38iUpAj6eCUxjQ@mail.gmail.com>
From:   Marco Elver <elver@google.com>
Date:   Thu, 14 Jan 2021 09:59:13 +0100
Message-ID: <CANpmjNP+MC4DYAE6K07sa_mcQ3c59zZ-g5yFotMVFdY+jPUyGw@mail.gmail.com>
Subject: Re: [PATCH 1/4] tracing: add error_report trace points
To:     Alexander Potapenko <glider@google.com>
Cc:     Steven Rostedt <rostedt@goodmis.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Andrey Konovalov <andreyknvl@gmail.com>,
        Dmitry Vyukov <dvyukov@google.com>,
        Ingo Molnar <mingo@redhat.com>, Petr Mladek <pmladek@suse.com>,
        Sergey Senozhatsky <sergey.senozhatsky@gmail.com>,
        Linux Memory Management List <linux-mm@kvack.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, 14 Jan 2021 at 08:50, Alexander Potapenko <glider@google.com> wrote:
>
> On Wed, Jan 13, 2021 at 10:10 PM Steven Rostedt <rostedt@goodmis.org> wrote:
> >
> > On Wed, 13 Jan 2021 10:16:54 +0100
> > Alexander Potapenko <glider@google.com> wrote:
> >
> > > +DECLARE_EVENT_CLASS(error_report_template,
> > > +                 TP_PROTO(const char *error_detector, unsigned long id),
> >
> > Instead of having a random string, as this should be used by a small finite
> > set of subsystems, why not make the above into an enum?
>
> You're probably right.
> I just thought it might be a good idea to minimize the effort needed
> from tools' authors to add these tracepoints to the tools (see the
> following two patches), and leave room for some extensibility (e.g.
> passing bug type together with the tool name etc.)
>
> > > +                 TP_ARGS(error_detector, id),
> > > +                 TP_STRUCT__entry(__field(const char *, error_detector)
> > > +                                          __field(unsigned long, id)),
> > > +                 TP_fast_assign(__entry->error_detector = error_detector;
> > > +                                __entry->id = id;),
> > > +                 TP_printk("[%s] %lx", __entry->error_detector,
> >
> > Then the [%s] portion of this could also be just a __print_symbolic().
>
> We'll need to explicitly list the enum values once again in
> __print_symbolic(), right? E.g.:
>
> enum debugging_tool {

We need to use TRACE_DEFINE_ENUM().

>          TOOL_KFENCE,

For consistency I would call the enum simply "ERROR_DETECTOR" as well.
(Hypothetically, there could also be an "error detector" that is not a
"debugging tool".)

>          TOOL_KASAN,
>          ...
> }
>
> TP_printk(__print_symbolic(__entry->error_detector, TOOL_KFENCE,
> TOOL_KASAN, ...),

It takes a list of val -> str. E.g.
__print_symbolic(__entry->error_detector, { TOOL_KFENCE, "KFENCE" }, {
TOOL_KASAN, "KASAN" }).

Looking around the kernel, sometimes this is simplified with macros,
but not sure if it's worth it. Typing the same thing 3 times is fine,
given this list won't grow really fast.

Thanks,
-- Marco