Re: [PATCH v14 21/23] x86/virt/tdx: Handle TDX interaction with ACPI S3 and deeper states

["Rafael J. Wysocki" <rafael@kernel.org>]

On Wed, Oct 18, 2023 at 12:51 PM Huang, Kai <kai.huang@intel.com> wrote:
>
> On Wed, 2023-10-18 at 12:15 +0200, Rafael J. Wysocki wrote:
> > On Wed, Oct 18, 2023 at 5:22 AM Huang, Kai <kai.huang@intel.com> wrote:
> > >
> > > Hi Rafael,
> > > Thanks for feedback!
> > > >
> > >
> > >
> > > > > @@ -1427,6 +1429,22 @@ static int __init tdx_init(void)
> > > > >                 return -ENODEV;
> > > > >         }
> > > > >
> > > > > +#define HIBERNATION_MSG                \
> > > > > +       "Disable TDX due to hibernation is available. Use 'nohibernate'
> > > command line to disable hibernation."
> > > >
> > > > I'm not sure if this new symbol is really necessary.
> > > >
> > > > The message could be as simple as "Initialization failed: Hibernation
> > > > support is enabled" (assuming a properly defined pr_fmt()), because
> > > > that carries enough information about the reason for the failure IMO.
> > > >
> > > > How to address it can be documented elsewhere.
> > >
> > >
> > > The last patch of this series is the documentation patch to add TDX host.  We
> > > can add a sentence to suggest the user to use 'nohibernate' kernel command line
> > > when one sees TDX gets disabled because of hibernation being available.
> > >
> > > But isn't better to just provide such information together in the dmesg so the
> > > user can immediately know how to resolve this issue?
> > >
> > > If user only sees "... failed: Hibernation support is enabled", then the user
> > > will need additional knowledge to know where to look for the solution first, and
> > > only after that, the user can know how to resolve this.
> >
> > I would expect anyone interested in a given feature to get familiar
> > with its documentation in the first place.  If they neglect to do that
> > and then find this message, it is absolutely fair to expect them to go
> > and look into the documentation after all.
>
> OK.  I'll remove HIBERNATION_MSG and just print the message suggested by you.
>
> And in the documentation patch, add one sentence to tell user when this happens,
> add 'nohibernate' to resolve.
>
>
> [...]
>
> > >
> > > -/* Low-level suspend routine. */
> > > -extern int (*acpi_suspend_lowlevel)(void);
> > > +typedef int (*acpi_suspend_lowlevel_t)(void);
> > > +
> > > +/* Set up low-level suspend routine. */
> > > +void acpi_set_suspend_lowlevel(acpi_suspend_lowlevel_t func);
> >
> > I'm not sure about the typededf, but I have no strong opinion against it either.
> >
> > >
> > >  /* Physical address to resume after wakeup */
> > >  unsigned long acpi_get_wakeup_address(void);
> > > diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
> > > index 2a0ea38955df..95be371305c6 100644
> > > --- a/arch/x86/kernel/acpi/boot.c
> > > +++ b/arch/x86/kernel/acpi/boot.c
> > > @@ -779,11 +779,17 @@ int (*__acpi_register_gsi)(struct device *dev, u32 gsi,
> > >  void (*__acpi_unregister_gsi)(u32 gsi) = NULL;
> > >
> > >  #ifdef CONFIG_ACPI_SLEEP
> > > -int (*acpi_suspend_lowlevel)(void) = x86_acpi_suspend_lowlevel;
> > > +static int (*acpi_suspend_lowlevel)(void) = x86_acpi_suspend_lowlevel;
> > >  #else
> > > -int (*acpi_suspend_lowlevel)(void);
> > > +static int (*acpi_suspend_lowlevel)(void);
> >
> > For the sake of consistency, either use the typedef here, or don't use
> > it at all.
>
> Ah right.
>
> Since you don't prefer the typedef, I'll abandon it:
>
> E.g,:
>
> void acpi_set_suspend_lowlevel(int (*suspend_lowlevel)(void))
> {
>         acpi_suspend_lowlevel = suspend_lowlevel;
> }
>
> Let me know whether this looks good to you?

Yes, this is fine with me.

> [...]
>
> >
> > Otherwise LGTM.
>
> Thanks. I'll split the helper patch out and include it to the next version of
> this series.
>