Re: [PATCH v2 0/3] Freezer, CPU hotplug, x86 Microcode: Fix task freezing failures

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 10/10/2011 08:53 PM, Alan Stern wrote:
> On Mon, 10 Oct 2011, Srivatsa S. Bhat wrote:
> 
>> When CPU hotplug is run along with suspend/hibernate tests using
>> the pm_test framework, even at the freezer level, we hit task freezing
>> failures. One such failure was reported here:
>> https://lkml.org/lkml/2011/9/5/28
>>
>> An excerpt of the log:
>>
>>   Freezing of tasks failed after 20.01 seconds (2 tasks refusing to
>>   freeze, wq_busy=0):
>>   invert_cpu_stat D 0000000000000000  5304 20435  17329 0x00000084
>>    ffff8801f367bab8 0000000000000046 ffff8801f367bfd8 00000000001d3a00
>>    ffff8801f367a010 00000000001d3a00 00000000001d3a00 00000000001d3a00
>>    ffff8801f367bfd8 00000000001d3a00 ffff880414cc6840 ffff8801f36783c0
>>   Call Trace:
>>    [<ffffffff81532de5>] schedule_timeout+0x235/0x320
>>    [<ffffffff81532a0b>] wait_for_common+0x11b/0x170
>>    [<ffffffff81532b3d>] wait_for_completion+0x1d/0x20
>>    [<ffffffff81364486>] _request_firmware+0x156/0x2c0
>>    [<ffffffff81364686>] request_firmware+0x16/0x20
>>    [<ffffffffa01f0da0>] request_microcode_fw+0x70/0xf0 [microcode]
>>    [<ffffffffa01f0390>] microcode_init_cpu+0xc0/0x100 [microcode]
>>    [<ffffffffa01f14b4>] mc_cpu_callback+0x7c/0x11f [microcode]
>>    [<ffffffff815393a4>] notifier_call_chain+0x94/0xd0
>>    [<ffffffff8109770e>] __raw_notifier_call_chain+0xe/0x10
>>    [<ffffffff8106d000>] __cpu_notify+0x20/0x40
>>    [<ffffffff8152cf5b>] _cpu_up+0xc7/0x10e
>>    [<ffffffff8152d07b>] cpu_up+0xd9/0xec
>>    [<ffffffff8151e599>] store_online+0x99/0xd0
>>    [<ffffffff81355eb0>] sysdev_store+0x20/0x30
>>    [<ffffffff811f3096>] sysfs_write_file+0xe6/0x170
>>    [<ffffffff8117ee50>] vfs_write+0xd0/0x1a0
>>    [<ffffffff8117f024>] sys_write+0x54/0xa0
>>    [<ffffffff8153df02>] system_call_fastpath+0x16/0x1b
>>
>>
>> The reason behind this failure is explained below:
>>
>> The x86 microcode update driver has callbacks registered for CPU hotplug
>> events such as a CPU getting offlined or onlined. Things go wrong when a
>> CPU hotplug stress test is carried out along with a suspend/resume operation
>> running simultaneously. Upon getting a CPU_DEAD notification (for example,
>> when a CPU offline occurs with tasks not frozen), the microcode callback
>> frees up the microcode and invalidates it. Later, when that CPU gets onlined
>> with tasks being frozen, the microcode callback (for the CPU_ONLINE_FROZEN
>> event) tries to apply the microcode to the CPU; doesn't find it and hence
>> depends on the (currently frozen) userspace to get the microcode again. This
>> leads to the numerous "WARNING"s at drivers/base/firmware_class.c which
>> eventually leads to task freezing failures in the suspend code path, as has
>> been reported.
>>
>> So, this patch series addresses this issue by ensuring that CPU hotplug and
>> suspend/hibernate don't run in parallel, thereby fixing the task freezing
>> failures.
> 
> The seems like entirely the wrong way to go about solving this problem.
> 
> The kernel shouldn't be responsible for making hotplug stress tests 
> exclusive with system sleep.  Whoever is running those tests should be 
> smart enough to realize what's wrong if system sleep interferes with a 
> test.
> 
> Furthermore, if the entire problem is lack of CPU microcode, hasn't 
> that been fixed already?  There recently was a patch to avoid releasing 
> microcode after it was first loaded -- the idea being that there would 
> then be no need to get the microcode from userspace again at awkward 
> times while the system is resuming.
> 

Well, that was the first version of this patch itself :)
I forgot to give a link to it in the patch description:
http://thread.gmane.org/gmane.linux.kernel/1198291/focus=1200591

That was my first idea: to avoid releasing microcode after it was first
loaded. But Tejun and Borislav felt that a better way to fix the problem
would be to mutually exclude CPU hotplug and suspend/hibernate.
And later on, Borislav Acked that one-line patch on the grounds that even
though that was not the best solution for the bug, it is an optimization
in its own right.
And then I posted that one-line patch with a revised motivation:
http://thread.gmane.org/gmane.linux.kernel/1200882

-- 
Regards,
Srivatsa S. Bhat  <srivatsa.bhat@linux.vnet.ibm.com>
Linux Technology Center,
IBM India Systems and Technology Lab