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

[kbusch@kernel.org (Keith Busch)]

On Thu, May 16, 2019@06:33:56PM +0000, Mario.Limonciello@dell.com wrote:
> > >> Then, you can populate nvme_dev_pm_ops as follows:
> > >>
> > >> static const struct dev_pm_ops = {
> > >>     .suspend = nvme_suspend,
> > >>     .resume = nvme_resume,
> > >>     .freeze = nvme_simple_suspend,
> > >>     .thaw = nvme_simple_resume,
> > >>     .poweroff = nvme_simple_suspend,
> > >>     .restore = nvme_simple_resume,
> > >> };
> > >>
> > >> and it should all work.
> > >
> > > Thanks for the pointers, I'll give that idea a shot.
> > 
> > Thanks for your work, I?ve tested your patch series on top of nvme-5.2 branch. It
> > works well once I apply flag PCI_DEV_FLAGS_NO_D3.
> > After some digging I think it?s another bug though:

....

> > pci_pm_suspend_noirq() gets called twice:
> > [   70.358779] nvme 0000:03:00.0: PCI PM: Suspend power state: D0
> > [   70.496908] nvme 0000:03:00.0: PCI PM: Suspend power state: D3hot
> > So it?s still being put to D3.
> > 
> > I?ll check why this bug happens.
> > 
> > Kai-Heng
> 
> Can you please try this patch:
> https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/1612437
> 
> It hasn't been submitted upstream yet, but should soon and I think it will fix this behavior.
> 
> Thanks,

If we're going to replace our SIMPLE_DEV_PM_OPS as Rafael suggests,
might as well add a .suspend_noirq callback. We can just save the
state again to work around this too.