From: Dexuan Cui <decui@xxxxxxxxxxxxx> Sent: Sunday, November 27, 2022 4:59 PM > > > From: Michael Kelley (LINUX) <mikelley@xxxxxxxxxxxxx> > > Sent: Wednesday, November 23, 2022 6:45 AM > > To: Dexuan Cui <decui@xxxxxxxxxxxxx>; ak@xxxxxxxxxxxxxxx; arnd@xxxxxxxx; > > > > Two thoughts: > > > > 1) The #ifdef CONFIG_INTEL_TDX_GUEST could probably be removed entirely > > with a tweak. hv_isolation_type_tdx() already doesn't need the #ifdef asthere's > > already a stub that returns 'false'. Then you just need a way to handle > > __tdx_ms_hv_hypercall(), or whatever it becomes based on the other discussion. > > As long as you can provide a stub that does nothing, the #ifdef won't be needed. > > > > 2) Assuming that we end up with some kind of Hyper-V specific version of > > __tdx_hypercall(), and hopefully as a "C" function, could you move the handling > > of ms_hyperv.shared_gpa_boundary into that function? Then you won't need > > to break out a separate include file for struct ms_hyperv. The Hyper-V TDX > > hypercall function must handle both normal and "fast" hypercalls, and the > > shared_gpa_boundary adjustment is needed only for normal hypercalls, > > but you can check the "fast" bit in the control word to decide. > > > > I haven't coded these ideas, so maybe there are snags I haven't thought of. > > But I'm really hoping we can avoid having to create a separate include > > file for struct ms_hyperv. > > > > Michael > > Thanks for the great suggestions! Now the code looks like this: > (the full list of v2 patches are still WIP: > > https://github.com/dcui/tdx/commits/decui/hyperv-next/2022-1121/v6.1-rc5/v2 > > diff --git a/arch/x86/hyperv/ivm.c b/arch/x86/hyperv/ivm.c > index 13ccb52eecd7..00e5c84e380b 100644 > --- a/arch/x86/hyperv/ivm.c > +++ b/arch/x86/hyperv/ivm.c > @@ -276,6 +276,27 @@ bool hv_isolation_type_tdx(void) > { > return static_branch_unlikely(&isolation_type_tdx); > } > + > +u64 hv_tdx_hypercall(u64 control, u64 input_addr, u64 output_addr) > +{ > + struct tdx_hypercall_args args = { }; > + > + if (!(control & HV_HYPERCALL_FAST_BIT)) { > + if (input_addr) > + input_addr += ms_hyperv.shared_gpa_boundary; At one point when working with the vTOM code, I realized that or'ing in the shared_gpa_boundary is probably safer than add'ing it, just in case the physical address already has vTOM set. I don't know if that possibility exists here, but it's something to consider as being slightly more robust. > + > + if (output_addr) > + output_addr += ms_hyperv.shared_gpa_boundary; Same here. > + } > + > + args.r10 = control; > + args.rdx = input_addr; > + args.r8 = output_addr; > + > + (void)__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT); > + > + return args.r11; > +} > #endif > > /* > > diff --git a/arch/x86/include/asm/mshyperv.h b/arch/x86/include/asm/mshyperv.h > index 8a2cafec4675..1be7bcf0d7d1 100644 > --- a/arch/x86/include/asm/mshyperv.h > +++ b/arch/x86/include/asm/mshyperv.h > @@ -39,6 +39,8 @@ int hv_call_deposit_pages(int node, u64 partition_id, u32 > num_pages); > int hv_call_add_logical_proc(int node, u32 lp_index, u32 acpi_id); > int hv_call_create_vp(int node, u64 partition_id, u32 vp_index, u32 flags); > > +u64 hv_tdx_hypercall(u64 control, u64 input_addr, u64 output_addr); > + > static inline u64 hv_do_hypercall(u64 control, void *input, void *output) > { > u64 input_address = input ? virt_to_phys(input) : 0; > @@ -46,6 +48,9 @@ static inline u64 hv_do_hypercall(u64 control, void *input, void > *output) > u64 hv_status; > > #ifdef CONFIG_X86_64 > + if (hv_isolation_type_tdx()) > + return hv_tdx_hypercall(control, input_address, output_address); > + > if (!hv_hypercall_pg) > return U64_MAX; > > @@ -83,6 +88,9 @@ static inline u64 hv_do_fast_hypercall8(u16 code, u64 input1) > u64 hv_status, control = (u64)code | HV_HYPERCALL_FAST_BIT; > > #ifdef CONFIG_X86_64 > + if (hv_isolation_type_tdx()) > + return hv_tdx_hypercall(control, input1, 0); > + > { > __asm__ __volatile__(CALL_NOSPEC > : "=a" (hv_status), ASM_CALL_CONSTRAINT, > @@ -114,6 +122,9 @@ static inline u64 hv_do_fast_hypercall16(u16 code, u64 input1, > u64 input2) > u64 hv_status, control = (u64)code | HV_HYPERCALL_FAST_BIT; > > #ifdef CONFIG_X86_64 > + if (hv_isolation_type_tdx()) > + return hv_tdx_hypercall(control, input1, input2); > + > { > __asm__ __volatile__("mov %4, %%r8\n" > CALL_NOSPEC Yes. This new structure LGTM. Michael
