Re: [GIT PULL (sort of)] KVM: x86: Static call changes for 6.11

[Sean Christopherson <seanjc@google.com>]

On Tue, Jul 16, 2024, Paolo Bonzini wrote:
> On 7/13/24 01:56, Sean Christopherson wrote:
> > Here's a massage pull request for the static_call() changes, just in case you
> > want to go this route instead of applying patches directly after merging
> > everything else for 6.11 (it was easy to generate this).  If you want to go the
> > patches route, I'll post 'em next week.
> > 
> > The following changes since commit c1c8a908a5f4c372f8a8dca0501b56ffc8d260fe:
> > 
> >    Merge branch 'vmx' (2024-06-28 22:22:53 +0000)
> > 
> > are available in the Git repository at:
> > 
> >    https://github.com/kvm-x86/linux.git  tags/kvm-x86-static_calls-6.11
> > 
> > for you to fetch changes up to b528de209c858f61953023b405a4abbf9a9933da:
> > 
> >    KVM: x86/pmu: Add kvm_pmu_call() to simplify static calls of kvm_pmu_ops (2024-06-28 15:23:49 -0700)
> 
> Thanks, indeed there was no straggler static_call() after applying
> this.  However, there might be a problem: static_call_cond() is equal
> to static_call() only if CONFIG_HAVE_STATIC_CALL_INLINE,

No, I think you misread the #if-#elif-#else.  It's only the !HAVE_STATIC_CALL
case that requires use of static_call_cond().  From include/linux/static_call.h:

  #ifdef CONFIG_HAVE_STATIC_CALL_INLINE
  #define static_call_cond(name)	(void)__static_call(name)

  #elif defined(CONFIG_HAVE_STATIC_CALL)
  #define static_call_cond(name)	(void)__static_call(name)

  #else
  #define static_call_cond(name)	(void)__static_call_cond(name)

  #endif

And per Josh, from an old RFC[*] to yank out static_call_cond():

 : Static calling a NULL pointer is a NOP, unless you're one of those poor
 : souls running on an arch (or backported x86 monstrosity) with
 : CONFIG_HAVE_STATIC_CALL=n, then it's a panic.

I double checked that 32-bit KVM works on Intel (which is guaranteed to have a
NULL guest_memory_reclaimed()).  I also verified that the generated code is
identical for both static_call() and static_call_cond(), i.e. the READ_ONCE() of
the func at runtime that's present in __static_call_cond() isn't showing up.

Dump of assembler code for function kvm_arch_guest_memory_reclaimed:
   0xc1042094 <+0>:	call   0xc10ce650 <__fentry__>
   0xc1042099 <+5>:	push   %ebp
   0xc104209a <+6>:	mov    %esp,%ebp
   0xc104209c <+8>:	call   0xc1932d8c <__SCT__kvm_x86_guest_memory_reclaimed>
   0xc10420a1 <+13>:	pop    %ebp
   0xc10420a2 <+14>:	ret    
End of assembler dump.

Dump of assembler code for function __SCT__kvm_x86_guest_memory_reclaimed:
   0xc1932d8c <+0>:	ret    
   0xc1932d8d <+1>:	int3   
   0xc1932d8e <+2>:	nop
   0xc1932d8f <+3>:	nop
   0xc1932d90 <+4>:	nop
   0xc1932d91 <+5>:	ud1    %esp,%ecx
End of assembler dump.

[*] https://lore.kernel.org/all/cover.1678474914.git.jpoimboe@kernel.org