Re: [PATCH 3/4] x86/kexec(): Reset TDX private memory on platforms with TDX erratum

[Huang, Kai]

On 2/02/2024 12:57 am, Dave Hansen wrote:
> On 2/1/24 06:39, Kirill A. Shutemov wrote:
> > > On 1/02/2024 6:03 am, Kirill A. Shutemov wrote:
> > > > On Wed, Jan 31, 2024 at 01:21:39PM -0800, Dave Hansen wrote:
> > > > > >    #ifdef CONFIG_KEXEC_JUMP
> > > > > >    	if (image->preserve_context)
> > > > > >    		save_processor_state();
> > > > > > +	else
> > > > > > +		tdx_reset_memory();
> > > > > > +#else
> > > > > > +	tdx_reset_memory();
> > > > > >    #endif
> ...
> > +void save_processor_state(void);
> > +void restore_processor_state(void);
> > +
> >   #ifdef CONFIG_SUSPEND
> >   extern suspend_state_t pm_suspend_target_state;
> >   extern suspend_state_t mem_sleep_current;
> > @@ -491,8 +494,6 @@ static inline int is_hibernate_resume_dev(dev_t dev) { return 0; }
> >   extern struct mutex system_transition_mutex;
> >   
> >   #ifdef CONFIG_PM_SLEEP
> > -void save_processor_state(void);
> > -void restore_processor_state(void);
>
> It's a little funky that we've got a #ifdef CONFIG_KEXEC_JUMP in the .c
> file and then we're dodging around an #ifdef CONFIG_PM_SLEEP in the
> header.  This is one of the reasons we shouldn't be putting #ifdefs in
> .c files in the first place.  But I digress...
>
> Either way, if you focus on getting the dang #ifdef out of the main code
> flow, the rest will fall in place easily.  Heck, if you even do this in
> the x86 kexec code:
>
> static void kexec_save_processor_start(image)
> {
> #ifdef CONFIG_KEXEC_JUMP
> 	if (image->preserve_context)
> 		save_processor_state();
> #endif
> }
>
> it'll leave you with:
>
> 	kexec_save_processor_start(image);
>
> 	/* Give a good reason here */
> 	if (!image->preserve_context)
> 		tdx_reset_memory();
>
> which is *FINE*.  No funky #ifdefs, indentation or else's dangling about.

Thanks for the insight and explanation!  I'll use this in the new version.