Re: [PATCH v3] hwrng: core: Freeze khwrng thread during suspend

[Maciej S. Szmigiero]

On 29.10.2019 15:44, Stephen Boyd wrote:
> Quoting Maciej S. Szmigiero (2019-10-28 16:45:31)
>> Hi Stephen,
>>
>> On 06.08.2019 01:32, Stephen Boyd wrote:
>>> The hwrng_fill() function can run while devices are suspending and
>>> resuming. If the hwrng is behind a bus such as i2c or SPI and that bus
>>> is suspended, the hwrng may hang the bus while attempting to add some
>>> randomness. It's been observed on ChromeOS devices with suspend-to-idle
>>> (s2idle) and an i2c based hwrng that this kthread may run and ask the
>>> hwrng device for randomness before the i2c bus has been resumed.
>>>
>>> Let's make this kthread freezable so that we don't try to touch the
>>> hwrng during suspend/resume. This ensures that we can't cause the hwrng
>>> backing driver to get into a bad state because the device is guaranteed
>>> to be resumed before the hwrng kthread is thawed.
>>
>> This patch broke suspend with virtio-rng loaded (it hangs).
>>
>> The problematic call chain is:
>> virtrng_freeze() -> remove_common() -> hwrng_unregister() ->
>> kthread_stop().
>>
>> It looks like kthread_stop() can't finish on a frozen khwrng thread.
> 
> Can you provide the suspend/resume logs?

There isn't much in the kernel log, the closest thing I can get is
with dyndbg="file drivers/base/power/main.c +p":
[   58.441073][ T3511] virtio-pci 0000:00:06.0: bus freeze
[   58.448744][ T3511] virtio-pci 0000:00:05.0: bus freeze
[   58.454500][ T3511] virtio-pci 0000:00:04.0: bus freeze
[   58.456873][ T3511] virtio-pci 0000:00:03.0: bus freeze

And then the VM hangs.

The 0000:00:03.0 pci device is virtio-rng.

If I add printks around that kthread_stop() in hwrng_unregister()
only the first one gets printed.

>>
>> Reverting this commit makes a VM with virtio-rng driver loaded
>> suspend and resume correctly again.
> 
> Which kernel are you testing on? There was a fix to this commit, i.e.
> ff296293b353 ("random: Support freezable kthreads in
> add_hwgenerator_randomness()"), which was fixed again by 59b569480dc8
> ("random: Use wait_event_freezable() in add_hwgenerator_randomness()").
> There was a problem with suspend/resume that I tried to fix with the
> first patch and then the second patch fixed the first one. See this
> thread[1] for some more background. You'll want all three.
> 
> [1] https://lkml.kernel.org/r/49fc7c64-88c0-74d0-2cb3-07986490941d@xxxxxx

The kernel under test is current torvalds/master, I can see that it
contains both commit ff296293b353 and commit 59b569480dc8.

I assume that the third commit you mention is the original one that this
e-mail message Subject line refers to (03a3bb7ae63150).

Thanks,
Maciej