[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <8722b425-4f20-20e5-7828-3feb0a3c4109@gmail.com>
Date: Mon, 30 Jan 2017 11:36:53 -0500
From: Yasuaki Ishimatsu <yasu.isimatu@...il.com>
To: Thomas Gleixner <tglx@...utronix.de>
Cc: linux-kernel@...r.kernel.org, x86@...nel.org
Subject: Re: [PATCH] perf/x86/intel/rapl: Rename rapl_cpu_prepare() to
rapl_cpu_starting()
Hi Thomas,
Do you have any idea to fix the issue?
If you have the idea, please send me the patch.
Thanks,
Yasuaki Ishimatsu
On 01/24/2017 02:54 PM, Thomas Gleixner wrote:
> On Tue, 24 Jan 2017, Yasuaki Ishimatsu wrote:
>> rapl_cpu_prepare() must be called after logical package id of CPU
>> is set by topology_update_package_map().
>>
>> But when onlining hot-added CPU, rapl_cpu_prepare() is called before
>> setting logical package id of the hot-added CPU. So cpu_to_rapl_pmu()
>> in rapl_cpu_prepare() finds a rapl_pmu of wrong logical package id and
>> rapl_cpu_prepare() initializes the wrong rapl_pmu.
>>
>> After that logical package id of the hot-added CPU is set by
>> topology_update_package_map(). But rapl_cpu_prepare() does
>> not initialize pmu of the logical package id of the hot-added CPU.
>> So when calling rapl_cpu_online(), cpu_to_rapl_pmu() returns NULL and
>> the following NULL pointer dereference occurs.
>>
>> BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
>> IP: rapl_cpu_online+0x8d/0xb0
>> <snip>
>> Call Trace:
>> ? rapl_cpu_offline+0xc0/0xc0
>> cpuhp_invoke_callback+0x8d/0x3f0
>> cpuhp_up_callbacks+0x37/0xb0
>> cpuhp_thread_fun+0xc9/0xe0
>> smpboot_thread_fn+0x110/0x160
>> kthread+0x101/0x140
>> ? sort_range+0x30/0x30
>> ? kthread_park+0x90/0x90
>> ret_from_fork+0x25/0x30
>>
>> The patch renames rapl_cpu_prepare() to rapl_cpu_starting() and changes
>> the position of cpuhp_state so that rapl_cpu_starting() is called
>> after topology_update_package_map().
>
> Does not work. You cannot call that callback in the starting context. It
> does allocations. This needs be fixed in a different way. I'll have a look
> tomorrow.
>
> Thanks,
>
> tglx
>
Powered by blists - more mailing lists