Re: [PATCH v6 07/12] x86/cpu/keylocker: Load an internal wrapping key at boot-time

[Chang S. Bae]

On 5/8/2023 2:56 PM, Dave Hansen wrote:
> On 5/8/23 11:18, Chang S. Bae wrote:
> > On 5/5/2023 4:05 PM, Eric Biggers wrote:
> > > On Mon, Apr 10, 2023 at 03:59:31PM -0700, Chang S. Bae wrote:
> > > >    +#ifdef CONFIG_X86_KEYLOCKER
> > > > +void setup_keylocker(struct cpuinfo_x86 *c);
> > > > +void destroy_keylocker_data(void);
> > > > +#else
> > > > +#define setup_keylocker(c) do { } while (0)
> > > > +#define destroy_keylocker_data() do { } while (0)
> > > > +#endif
> > >
> > > Shouldn't the !CONFIG_X86_KEYLOCKER stubs be static inline functions
> > > instead of
> > > macros, so that type checking works?
> >
> > I think either way works here. This macro is just for nothing.
>
> Chang, I do prefer the 'static inline' as a general rule.  Think of this:
>
> static inline void setup_keylocker(struct cpuinfo_x86 *c) {}
>
> versus:
>
> #define setup_keylocker(c) do { } while (0)
>
> Imagine some dope does:
>
> 	char c;
> 	...
> 	setup_keylocker(c);
>
> With the macro, they'll get no type warning.  The inline actually makes
> it easier to find bugs because folks will get _some_ type checking no
> matter how they compile the code.

Ah, when the prototype with one or more arguments, 'static inline' 
allows the check. Then it is not an 'either-way' thing.

Looking at the x86 code, there are some seemingly related:

$ git grep "do { } while (0)" arch/x86 | grep -v "()"
arch/x86/include/asm/kprobes.h:#define flush_insn_slot(p)       do { } 
while (0)
arch/x86/include/asm/mc146818rtc.h:#define lock_cmos(reg) do { } while (0)
arch/x86/include/asm/pgtable.h:#define flush_tlb_fix_spurious_fault(vma, 
address, ptep) do { } while (0)
arch/x86/include/asm/preempt.h:#define init_task_preempt_count(p) do { } 
while (0)
arch/x86/kvm/ioapic.h:#define ASSERT(x) do { } while (0)
arch/x86/kvm/mmu/mmu_internal.h:#define pgprintk(x...) do { } while (0)
arch/x86/kvm/mmu/mmu_internal.h:#define rmap_printk(x...) do { } while (0)
arch/x86/kvm/mmu/mmu_internal.h:#define MMU_WARN_ON(x) do { } while (0)

Now I feel owed for some potential cleanup work.

Thanks,
Chang