]> git.baikalelectronics.ru Git - kernel.git/commitdiff
KVM: SVM: Add register operand to vmsave call in sev_es_vcpu_load
authorNathan Chancellor <natechancellor@gmail.com>
Sat, 19 Dec 2020 06:37:11 +0000 (23:37 -0700)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 7 Jan 2021 23:07:27 +0000 (18:07 -0500)
When using LLVM's integrated assembler (LLVM_IAS=1) while building
x86_64_defconfig + CONFIG_KVM=y + CONFIG_KVM_AMD=y, the following build
error occurs:

 $ make LLVM=1 LLVM_IAS=1 arch/x86/kvm/svm/sev.o
 arch/x86/kvm/svm/sev.c:2004:15: error: too few operands for instruction
         asm volatile(__ex("vmsave") : : "a" (__sme_page_pa(sd->save_area)) : "memory");
                      ^
 arch/x86/kvm/svm/sev.c:28:17: note: expanded from macro '__ex'
 #define __ex(x) __kvm_handle_fault_on_reboot(x)
                 ^
 ./arch/x86/include/asm/kvm_host.h:1646:10: note: expanded from macro '__kvm_handle_fault_on_reboot'
         "666: \n\t"                                                     \
                 ^
 <inline asm>:2:2: note: instantiated into assembly here
         vmsave
         ^
 1 error generated.

This happens because LLVM currently does not support calling vmsave
without the fixed register operand (%rax for 64-bit and %eax for
32-bit). This will be fixed in LLVM 12 but the kernel currently supports
LLVM 10.0.1 and newer so this needs to be handled.

Add the proper register using the _ASM_AX macro, which matches the
vmsave call in vmenter.S.

Fixes: f5c4f4c8011d ("KVM: SVM: Provide support for SEV-ES vCPU loading")
Link: https://reviews.llvm.org/D93524
Link: https://github.com/ClangBuiltLinux/linux/issues/1216
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Message-Id: <20201219063711.3526947-1-natechancellor@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/svm/sev.c

index 9858d5ae9dddde0dae70bc8359526eecdc375ecd..563ced07b0b85da361172d76ed1231208bef1590 100644 (file)
@@ -2001,7 +2001,7 @@ void sev_es_vcpu_load(struct vcpu_svm *svm, int cpu)
         * of which one step is to perform a VMLOAD. Since hardware does not
         * perform a VMSAVE on VMRUN, the host savearea must be updated.
         */
-       asm volatile(__ex("vmsave") : : "a" (__sme_page_pa(sd->save_area)) : "memory");
+       asm volatile(__ex("vmsave %0") : : "a" (__sme_page_pa(sd->save_area)) : "memory");
 
        /*
         * Certain MSRs are restored on VMEXIT, only save ones that aren't