linux-kernel - Re: [PATCH 1/1] tracing: Fix event_trace_printk loss on printk

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 13 Feb 2019 14:49:45 +0800
From:   Mars Cheng <mars.cheng@...iatek.com>
To:     Steven Rostedt <rostedt@...dmis.org>
CC:     Andress Kuo <andress.kuo@...iatek.com>,
        CC Hwang <cc.hwang@...iatek.com>,
        <linux-mediatek@...ts.infradead.org>,
        <linux-kernel@...r.kernel.org>,
        <linux-arm-kernel@...ts.infradead.org>
Subject: Re: [PATCH 1/1] tracing: Fix event_trace_printk loss on
 printk_format.

Hi Steve

On Wed, 2019-02-13 at 12:54 +0800, Mars Cheng wrote:
> Hi Steve
> 
> On Tue, 2019-02-12 at 20:41 -0500, Steven Rostedt wrote:
> > On Tue, 12 Feb 2019 19:41:57 +0800
> > Mars Cheng <mars.cheng@...iatek.com> wrote:
> > 
> > > From: Andress Kuo <andress.kuo@...iatek.com>
> > > 
> > > If fmt on event_trace_printk is not a constant, It means that something not
> > > guaranteed, so the compiler optimizes the fmt out, and then the
> > > __trace_printk_fmt section is not filled. if __trace_printk_fmt is not
> > > filled, the kernel will not allocate the special buffers needed for the
> > > event_trace_printk() and then not shown in the file output
> > > sys/kernel/debug/tracing/print_formats.
> > > 
> > > Adding a "__used" to the variable in the __trace_printk_fmt section on
> > > event_trace_printk() will keep it around, even though it is set to NULL.
> > > This will keep the string from being printed in the
> > > sys/kernel/debug/tracing/printk_formats section.
> > > 
> > > We can also refer to commit 3debb0a9ddb1 ("tracing: Fix trace_printk()
> > > to print when not using bprintk()")that it had similar issue on path of
> > > trace_printk().
> > 
> > Honestly, I don't even remember why that macro was created. I think
> > it's a remnant from the creation of the trace events. I think the best
> > solution is just to nuke it. It shouldn't be used anymore.
> > 
> > I'll dig a bit deeper into the history of that macro, but I'm thinking
> > it shouldn't exist anymore.
> > 
> > [ /me returns from walking down memory lane ]
> > 
> > Wow, that brings back some wild memories. Yes, that macro must die, and
> > I wish the history of it could die along with it ;-) The horror of the
> > old ways I tell you. The event_trace_printk() was the original way we
> > displayed events! No real formatting, no parsing my userspace tools. It
> > was just a glamorized printk. It was called TRACE_FORMAT() which was
> > deprecated by today's TRACE_EVENT().
> > 
> > That macro should have been removed by commit b8e65554d80b4.
> > 
> > Please just send a patch to delete that macro. Let's not be maintaining
> > it. It gives me nightmares.
> > 
> > -- Steve
> 
> Got it, I will send another patch to remove the nightmares. :-)
> 
> Thanks.

After some grep, I found event_trace_printk() not used in 5.0-rc*. But
trace_printk() is still used in several places.

kernel/trace/ring_buffer_benchmark.c:415: trace_printk("Sleeping for 10
secs\n");
...
drivers/gpu/drm/i915/i915_gem.h:66:#define GEM_TRACE(...)
trace_printk(__VA_ARGS__)
drivers/hwtracing/stm/dummy_stm.c:30:   trace_printk("[%u:%u] [pkt: %x/%
x] (%llx)\n", master, channel,
...

since they are similar functions, do you prefer remove both of them or
just remove event_trace_printk()? the former approach might affect some
modules, and the latter approach keeps nightmares, right?

Thanks.
>