[PATCHv2 6/6] nvme-pci: Use host managed power state for suspend

[kai.heng.feng@canonical.com (Kai Heng Feng)]

at 6:34 PM, Rafael J. Wysocki <rafael@kernel.org> wrote:

> On Fri, May 17, 2019@11:35 AM Christoph Hellwig <hch@lst.de> wrote:
>> On Fri, May 17, 2019@11:17:52AM +0200, Rafael J. Wysocki wrote:
>>> AFAICS, using DMA while creating a hibernation snapshot image of
>>> memory would not be a good idea, so the host memory buffer should be
>>> disabled during the "freeze" stage or hibernation (and it can be
>>> re-enabled when that stage is complete).
>>
>> Once we'd disable the HMB the harm is done, so no point in re-enabling it.
>
> But after the "freeze" stage the image may need to be written to the
> device, so I guess it would be useful to enable the HMB for that?
>
>> What would this mean in terms of ops?  Only disable in freeze_noirq?
>
> It can be disabled in any of the "freeze" callbacks (freeze,
> freeze_late, freeze_irq) where it makes most sense.  I also would
> re-enable it in the complementary "thaw" one, in case the device will
> be used to store the image.
>
>> Or stick to the simple ops and find another helper that indicates we are  
>> doing a hibernation?
>
> There are no helpers for that, it needs to be done manually, sort of.

I've got confirmation from Vendor, disabling HMB isn?t necessary when the  
host stays at D0, which is what we use in S2I.
I?ve done some testing and I can confirm S2I suspend/resume can work  
without disabling HMB.

Kai-Heng