Re: [RFC PATCH] perf/x86/intel/rapl: avoid access unallocate memory

From: "Charles (Chas) Williams" <ciwillia@brocade.com>
To: Thomas Gleixner <tglx@linutronix.de>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
	<x86@kernel.org>, <linux-kernel@vger.kernel.org>,
	"M. Vefa Bicakci" <m.v.b@runbox.com>
Subject: Re: [RFC PATCH] perf/x86/intel/rapl: avoid access unallocate memory
Date: Tue, 8 Nov 2016 09:57:19 -0500	[thread overview]
Message-ID: <86609338-2b45-ed7e-fb07-99421e43a2f1@brocade.com> (raw)
In-Reply-To: <alpine.DEB.2.20.1611081525160.3501@nanos>

On 11/08/2016 09:31 AM, Thomas Gleixner wrote:
> On Tue, 8 Nov 2016, Charles (Chas) Williams wrote:
>> 	[    0.016335] topology_update_package_map:  apicid 0  pkg 0  cpu 0
>> 	[    0.016398] smpboot: APIC(0) Converting physical 0 to logical
>> package 0, cpu 0 (ffff88023fc0a040)
>> 	[    0.016399] topology_update_package_map:  apicid 1  pkg 1  cpu 1
>> 	[    0.016462] smpboot: APIC(1) Converting physical 1 to logical
>> package 1, cpu 1 (ffff88023fd0a040)
>>
>> So, I don't know where apic->cpu_present_to_apicid(cpu) is getting its
>> apicid but it certainly doesn't seem to the match the apicid in the
>> CPU's registers.  For whatever reason, my VMware system is reporting
>> that the second CPU has a local APIC ID of 2:
>
> The initial information comes from MP tables or ACPI.
>
>> 	[    0.009115] identify_cpu: cpu_index 0  phys_proc_id is now 0,
>> apicid 0, initial_apicid 0
>> 	...
>> 	[    0.237401] identify_cpu: cpu_index 1  phys_proc_id is now 2,
>> apicid 2, initial_apicid 2
>
> And the CPUID emulation tells something different. Sigh!
>
>> I was thinking it might be better to call topology_update_package_map()
>> at the bottom of identify_cpu() to setup the secondary CPU's.  The boot
>> cpu could be setup during smp_init_package_map().
>
> Perhaps, but that does not make the inconsistencies go away....

By the time I know it's not consistent, there isn't anything I can do
about it.  I can't update the table to remove the bad information.

The other alternative, is to trust the ACPI and just update the
cpu_data's apicid in identify_cpu() to the value from the table.
The earlier kernels didn't seem to rely as much on this information.
But it does appear to be "wrong" in the APIC table.  From acpidump:

[02Ch 0044   1]                Subtable Type : 00 [Processor Local APIC]
[02Dh 0045   1]                       Length : 08
[02Eh 0046   1]                 Processor ID : 00
[02Fh 0047   1]                Local Apic ID : 00
[030h 0048   4]        Flags (decoded below) : 00000001
                            Processor Enabled : 1

[034h 0052   1]                Subtable Type : 00 [Processor Local APIC]
[035h 0053   1]                       Length : 08
[036h 0054   1]                 Processor ID : 01
[037h 0055   1]                Local Apic ID : 01
[038h 0056   4]        Flags (decoded below) : 00000001
                            Processor Enabled : 1