Re: [Pv-drivers] [PATCH 29/36] cpuidle, xenpv: Make more PARAVIRT_XXL noinstr clean

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Jun 13, 2022 at 07:23:13PM +0000, Nadav Amit wrote:
> On Jun 13, 2022, at 11:48 AM, Srivatsa S. Bhat <srivatsa@csail.mit.edu> wrote:
> 
> > ⚠ External Email
> > 
> > On 6/8/22 4:27 PM, Peter Zijlstra wrote:
> >> vmlinux.o: warning: objtool: acpi_idle_enter_s2idle+0xde: call to wbinvd() leaves .noinstr.text section
> >> vmlinux.o: warning: objtool: default_idle+0x4: call to arch_safe_halt() leaves .noinstr.text section
> >> vmlinux.o: warning: objtool: xen_safe_halt+0xa: call to HYPERVISOR_sched_op.constprop.0() leaves .noinstr.text section
> >> 
> >> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > 
> > Reviewed-by: Srivatsa S. Bhat (VMware) <srivatsa@csail.mit.edu>
> > 
> >> 
> >> -static inline void wbinvd(void)
> >> +extern noinstr void pv_native_wbinvd(void);
> >> +
> >> +static __always_inline void wbinvd(void)
> >> {
> >>      PVOP_ALT_VCALL0(cpu.wbinvd, "wbinvd", ALT_NOT(X86_FEATURE_XENPV));
> >> }
> 
> I guess it is yet another instance of wrong accounting of GCC for
> the assembly blocks’ weight. I guess it is not a solution for older
> GCCs, but presumably ____PVOP_ALT_CALL() and friends should have
> used asm_inline or some new “asm_volatile_inline” variant.

Partially, some of the *SAN options also generate a metric ton of
nonsense when enabled and skew the compilers towards not inlining
things.