Received-SPF: pass (google.com: domain of linux-kernel+bounces-35085-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Message-ID: <0279a4cb-ced0-447a-a06f-37c38650ed5b@quicinc.com>
Date: Tue, 23 Jan 2024 18:23:58 +0800
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v3] tracing: Support to dump instance traces by
 ftrace_dump_on_oops
To: Steven Rostedt <rostedt@goodmis.org>
CC: <mhiramat@kernel.org>, <mark.rutland@arm.com>, <mcgrof@kernel.org>,
        <keescook@chromium.org>, <j.granados@samsung.com>,
        <mathieu.desnoyers@efficios.com>, <corbet@lwn.net>,
        <linux-kernel@vger.kernel.org>, <linux-trace-kernel@vger.kernel.org>,
        <linux-fsdevel@vger.kernel.org>, <linux-doc@vger.kernel.org>,
        <quic_bjorande@quicinc.com>, <quic_tsoni@quicinc.com>,
        <quic_satyap@quicinc.com>, <quic_aiquny@quicinc.com>,
        <kernel@quicinc.com>, Ross Zwisler <zwisler@google.com>,
        Joel Fernandes <joel@joelfernandes.org>
References: <20240119080824.907101-1-quic_hyiwei@quicinc.com>
 <20240119115625.603188d1@gandalf.local.home>
Content-Language: en-US
From: Huang Yiwei <quic_hyiwei@quicinc.com>
In-Reply-To: <20240119115625.603188d1@gandalf.local.home>
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Transfer-Encoding: 7bit


On 1/20/2024 12:56 AM, Steven Rostedt wrote:
> On Fri, 19 Jan 2024 16:08:24 +0800
> Huang Yiwei <quic_hyiwei@quicinc.com> wrote:
> 
>> -	ftrace_dump_on_oops[=orig_cpu]
>> +	ftrace_dump_on_oops[=orig_cpu | =<instance>]
> 
> I wonder if we should have it be:
> 
> 	ftrace_dump_on_oops[=orig_cpu | =<instance> | =<instance>:orig_cpu ]
> 
> Then last would be to only print out a specific CPU trace of the given instance.
> 
> And if we really want to be fancy!
> 
> 	ftrace_dump_on_opps[=orig_cpu | =<instance> | =orig_cpu:<instance> ][,<instance> | ,<instance>:orig_cpu]
> 
Yeah, I agree to make the parameter more flexible.

"=orig_cpu:<instance>" means to dump global and another instance?

I'm thinking of the following format:

ftrace_dump_on_opps[=orig_cpu | =<instance>][,<instance> | 
,<instance>=orig_cpu]

Here list some possible situations:

1. Dump global on orig_cpu:
ftrace_dump_on_oops=orig_cpu

2. Dump global and instance1 on all cpu, instance2 on orig_cpu:
ftrace_dump_on_opps,<instance1>,<instance2>=orig_cpu

3. Dump global and instance1 on orig_cpu, instance2 on all cpu:
ftrace_dump_on_opps=orig_cpu,<instance1>=orig_cpu,<instance2>

4. Dump instance1 on all cpu, instance2 on orig_cpu:
ftrace_dump_on_opps=<instance1>,<instance2>=orig_cpu

5. Dump instance1 and instance2 on orig_cpu:
ftrace_dump_on_opps=<instance1>=orig_cpu,<instance2>=orig_cpu

This makes orig_cpu dump for global same as instance, the parameter may 
seems more unified and users don't need to remember another markers to 
request orig_cpu dump.

But one problem here is if there's an instance named "orig_cpu", then we 
may not dump it correctly.

Regards,
Huang Yiwei

> That would allow dumping more than one instance.
> 
> If you want to dump the main buffer and an instance foo:
> 
> 	ftrace_dump_on_opps,foo
> 
> Where the ',' says to dump the top instance as well as the foo instance.
> 
> -- Steve
> 
> 
>>   			[FTRACE] will dump the trace buffers on oops.
>> -			If no parameter is passed, ftrace will dump
>> -			buffers of all CPUs, but if you pass orig_cpu, it will
>> +			If no parameter is passed, ftrace will dump global
>> +			buffers of all CPUs, if you pass orig_cpu, it will
>>   			dump only the buffer of the CPU that triggered the
>> -			oops.
>> +			oops, or specific instance will be dumped if instance
>> +			name is passed.
>>