RISC-V: KVM: Use SBI sync SRET call when available

Implement an optimized KVM world-switch using SBI sync SRET call
when SBI nested acceleration extension is available. This improves
KVM world-switch when KVM RISC-V is running as a Guest under some
other hypervisor.

Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Anup Patel <[email protected]>

Author: avpatel

arch/riscv/include/asm/kvm_nacl.h

@@ -12,6 +12,8 @@
 #include <asm/csr.h>
 #include <asm/sbi.h>
 
+struct kvm_vcpu_arch;
+
 DECLARE_STATIC_KEY_FALSE(kvm_riscv_nacl_available);
 #define kvm_riscv_nacl_available() \
 	static_branch_unlikely(&kvm_riscv_nacl_available)
@@ -43,6 +45,10 @@ void __kvm_riscv_nacl_hfence(void *shmem,
 			     unsigned long page_num,
 			     unsigned long page_count);
 
+void __kvm_riscv_nacl_switch_to(struct kvm_vcpu_arch *vcpu_arch,
+				unsigned long sbi_ext_id,
+				unsigned long sbi_func_id);
+
 int kvm_riscv_nacl_enable(void);
 
 void kvm_riscv_nacl_disable(void);

arch/riscv/kvm/vcpu.c

@@ -766,19 +766,59 @@ static __always_inline void kvm_riscv_vcpu_swap_in_host_state(struct kvm_vcpu *v
  */
 static void noinstr kvm_riscv_vcpu_enter_exit(struct kvm_vcpu *vcpu)
 {
+	void *nsh;
 	struct kvm_cpu_context *gcntx = &vcpu->arch.guest_context;
 	struct kvm_cpu_context *hcntx = &vcpu->arch.host_context;
 
 	kvm_riscv_vcpu_swap_in_guest_state(vcpu);
 	guest_state_enter_irqoff();
 
-	hcntx->hstatus = ncsr_swap(CSR_HSTATUS, gcntx->hstatus);
+	if (kvm_riscv_nacl_sync_sret_available()) {
+		nsh = nacl_shmem();
 
-	nsync_csr(-1UL);
+		if (kvm_riscv_nacl_autoswap_csr_available()) {
+			hcntx->hstatus =
+				nacl_csr_read(nsh, CSR_HSTATUS);
+			nacl_scratch_write_long(nsh,
+						SBI_NACL_SHMEM_AUTOSWAP_OFFSET +
+						SBI_NACL_SHMEM_AUTOSWAP_HSTATUS,
+						gcntx->hstatus);
+			nacl_scratch_write_long(nsh,
+						SBI_NACL_SHMEM_AUTOSWAP_OFFSET,
+						SBI_NACL_SHMEM_AUTOSWAP_FLAG_HSTATUS);
+		} else if (kvm_riscv_nacl_sync_csr_available()) {
+			hcntx->hstatus = nacl_csr_swap(nsh,
+						       CSR_HSTATUS, gcntx->hstatus);
+		} else {
+			hcntx->hstatus = csr_swap(CSR_HSTATUS, gcntx->hstatus);
+		}
 
-	__kvm_riscv_switch_to(&vcpu->arch);
+		nacl_scratch_write_longs(nsh,
+					 SBI_NACL_SHMEM_SRET_OFFSET +
+					 SBI_NACL_SHMEM_SRET_X(1),
+					 &gcntx->ra,
+					 SBI_NACL_SHMEM_SRET_X_LAST);
+
+		__kvm_riscv_nacl_switch_to(&vcpu->arch, SBI_EXT_NACL,
+					   SBI_EXT_NACL_SYNC_SRET);
+
+		if (kvm_riscv_nacl_autoswap_csr_available()) {
+			nacl_scratch_write_long(nsh,
+						SBI_NACL_SHMEM_AUTOSWAP_OFFSET,
+						0);
+			gcntx->hstatus = nacl_scratch_read_long(nsh,
+								SBI_NACL_SHMEM_AUTOSWAP_OFFSET +
+								SBI_NACL_SHMEM_AUTOSWAP_HSTATUS);
+		} else {
+			gcntx->hstatus = csr_swap(CSR_HSTATUS, hcntx->hstatus);
+		}
+	} else {
+		hcntx->hstatus = csr_swap(CSR_HSTATUS, gcntx->hstatus);
 
-	gcntx->hstatus = csr_swap(CSR_HSTATUS, hcntx->hstatus);
+		__kvm_riscv_switch_to(&vcpu->arch);
+
+		gcntx->hstatus = csr_swap(CSR_HSTATUS, hcntx->hstatus);
+	}
 
 	vcpu->arch.last_exit_cpu = vcpu->cpu;
 	guest_state_exit_irqoff();

arch/riscv/kvm/vcpu_switch.S

@@ -218,6 +218,35 @@ SYM_FUNC_START(__kvm_riscv_switch_to)
 	ret
 SYM_FUNC_END(__kvm_riscv_switch_to)
 
+	/*
+	 * Parameters:
+	 * A0 <= Pointer to struct kvm_vcpu_arch
+	 * A1 <= SBI extension ID
+	 * A2 <= SBI function ID
+	 */
+SYM_FUNC_START(__kvm_riscv_nacl_switch_to)
+	SAVE_HOST_GPRS
+
+	SAVE_HOST_AND_RESTORE_GUEST_CSRS .Lkvm_nacl_switch_return
+
+	/* Resume Guest using SBI nested acceleration */
+	add	a6, a2, zero
+	add	a7, a1, zero
+	ecall
+
+	/* Back to Host */
+	.align 2
+.Lkvm_nacl_switch_return:
+	SAVE_GUEST_GPRS
+
+	SAVE_GUEST_AND_RESTORE_HOST_CSRS
+
+	RESTORE_HOST_GPRS
+
+	/* Return to C code */
+	ret
+SYM_FUNC_END(__kvm_riscv_nacl_switch_to)
+
 SYM_CODE_START(__kvm_riscv_unpriv_trap)
 	/*
 	 * We assume that faulting unpriv load/store instruction is