Re: [PATCH v2 12/13] x86_64,hyperv: Use direct call to hypercall-page

[Peter Zijlstra]

On Thu, May 01, 2025 at 02:36:26AM +0000, Michael Kelley wrote:
> From: Peter Zijlstra <peterz@xxxxxxxxxxxxx> Sent: Wednesday, April 30, 2025 4:08 AM
> > @@ -528,8 +546,8 @@ void __init hyperv_init(void)
> >  	if (hv_isolation_type_tdx() && !ms_hyperv.paravisor_present)
> >  		goto skip_hypercall_pg_init;
> > 
> > -	hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, VMALLOC_START,
> > -			VMALLOC_END, GFP_KERNEL, PAGE_KERNEL_ROX,
> > +	hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, MODULES_VADDR,
> > +			MODULES_END, GFP_KERNEL, PAGE_KERNEL_ROX,
> 
> Curiosity question (which I forgot ask about in v1):  Is this change so that the
> hypercall page kernel address is "close enough" for the direct call to work from
> built-in code and from module code?  Or is there some other reason?

No, you nailed it. Because we want to do a direct CALL, the hypercall
page must be in the disp32 range relative to the call site. The module
address space ensures this.

> >  			VM_FLUSH_RESET_PERMS, NUMA_NO_NODE,
> >  			__builtin_return_address(0));
> >  	if (hv_hypercall_pg == NULL)
> > @@ -567,27 +585,9 @@ void __init hyperv_init(void)
> >  		wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
> >  	}
> > 
> > -skip_hypercall_pg_init:
> > -	/*
> > -	 * Some versions of Hyper-V that provide IBT in guest VMs have a bug
> > -	 * in that there's no ENDBR64 instruction at the entry to the
> > -	 * hypercall page. Because hypercalls are invoked via an indirect call
> > -	 * to the hypercall page, all hypercall attempts fail when IBT is
> > -	 * enabled, and Linux panics. For such buggy versions, disable IBT.
> > -	 *
> > -	 * Fixed versions of Hyper-V always provide ENDBR64 on the hypercall
> > -	 * page, so if future Linux kernel versions enable IBT for 32-bit
> > -	 * builds, additional hypercall page hackery will be required here
> > -	 * to provide an ENDBR32.
> > -	 */
> > -#ifdef CONFIG_X86_KERNEL_IBT
> > -	if (cpu_feature_enabled(X86_FEATURE_IBT) &&
> > -	    *(u32 *)hv_hypercall_pg != gen_endbr()) {
> > -		setup_clear_cpu_cap(X86_FEATURE_IBT);
> > -		pr_warn("Disabling IBT because of Hyper-V bug\n");
> > -	}
> > -#endif
> 
> Nit: With this IBT code removed, the #include <asm/ibt.h> at the top
> of this source code file should be removed.

Indeed so.

> 
> > +	hv_set_hypercall_pg(hv_hypercall_pg);
> > 
> > +skip_hypercall_pg_init:
> >  	/*
> >  	 * hyperv_init() is called before LAPIC is initialized: see
> >  	 * apic_intr_mode_init() -> x86_platform.apic_post_init() and
> > 
> > 
> 
> The nit notwithstanding,
> 
> Reviewed-by: Michael Kelley <mhklinux@xxxxxxxxxxx>

Thanks!