linux-kernel - Re: [RFC PATCH 2/4] trace: Introduce an output interface from ftrace to STM

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] [day] [month] [year] [list]

Message-ID: <CAG2=9p95EWUTh-SPs38RDAyn3YJ=ADSLba2EM-5TvDt1_tBaGw@mail.gmail.com>
Date:	Mon, 20 Jun 2016 17:22:41 +0800
From:	Chunyan Zhang <zhang.chunyan@...aro.org>
To:	Alexander Shishkin <alexander.shishkin@...ux.intel.com>
Cc:	Steven Rostedt <rostedt@...dmis.org>,
	Mathieu Poirier <mathieu.poirier@...aro.org>, mingo@...hat.com,
	Mike Leach <mike.leach@....com>, Tor Jeremiassen <tor@...com>,
	Maxime Coquelin <maxime.coquelin@...com>,
	philippe.langlais@...com, Nicolas GUION <nicolas.guion@...com>,
	Lyra Zhang <zhang.lyra@...il.com>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>
Subject: Re: [RFC PATCH 2/4] trace: Introduce an output interface from ftrace
 to STM

On Tue, Jun 7, 2016 at 6:04 PM, Alexander Shishkin
<alexander.shishkin@...ux.intel.com> wrote:
> Chunyan Zhang <zhang.chunyan@...aro.org> writes:
>
>> This patch is introducing a new function to print Ftrace messages
>> to STM buffer when the traces happen.  In order to reduce the
>> effect on timing overhead as much as possible, only the current
>> function and its parent ip address will be recorded into STM in
>> this patch.  This idea was first introduced by Philippe Langlais
>> at ST-Microelectronics a long time ago[1].
>
> So why is this useful? The value of trace points is in their payload.

>From what I understand, we have no way to log as much information as
Ftrace has done if we want to control the overhead of writing STM in a
limited/acceptable range.  Exporting information to STM is much more
time-consuming than writing ring buffer, the thought was simply that
export as much useful information as possible while taking a limited
bytes.

Thanks,
Chunyan

>
>> +#define STM_FTRACE_CHAN 0
>
> This is why we have the stm master/channel allocation policy, which
> should already assign a channel to your stm_source device when it is
> linked to an stm device.
>
> Also, why is this a separate compilation unit from the stm_ftrace.o?
>
>> +
>> +void ftrace_stm_func(unsigned long ip, unsigned long parent_ip)
>> +{
>> +     unsigned long ip_array[2] = {ip, parent_ip};
>> +
>> +     stm_ftrace_write((char *)ip_array, sizeof(unsigned long) * 2,
>> +                      STM_FTRACE_CHAN);
>> +}
>> +EXPORT_SYMBOL_GPL(ftrace_stm_func);
>> diff --git a/kernel/trace/trace_output_stm.h b/kernel/trace/trace_output_stm.h
>> new file mode 100644
>> index 0000000..fc3f989
>> --- /dev/null
>> +++ b/kernel/trace/trace_output_stm.h
>> @@ -0,0 +1,14 @@
>> +#ifndef __TRACE_OUTPUT_STM_H
>> +#define __TRACE_OUTPUT_STM_H
>> +
>> +#include <linux/module.h>
>> +
>> +#ifdef CONFIG_STM_FTRACE
>> +extern void stm_ftrace_write(const char *buf, unsigned int len,
>> +                          unsigned int chan);
>> +extern void ftrace_stm_func(unsigned long ip, unsigned long parent_ip);
>> +#else
>> +static inline void ftrace_stm_func(unsigned long ip, unsigned long parent_ip) {}
>> +#endif
>> +
>> +#endif /* __TRACE_OUTPUT_STM_H */
>> --
>> 1.9.1