RISC-V: KVM: FP lazy save/restore

This patch adds floating point (F and D extension) context save/restore
for guest VCPUs. The FP context is saved and restored lazily only when
kernel enter/exits the in-kernel run loop and not during the KVM world
switch. This way FP save/restore has minimal impact on KVM performance.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>

2 files changed, 265 insertions, 0 deletions

diff --git a/arch/riscv/kvm/vcpu.c b/arch/riscv/kvm/vcpu.c
index 840f4586796f..cf6832365345 100644
--- a/arch/riscv/kvm/vcpu.c
+++ b/arch/riscv/kvm/vcpu.c
@@ -38,6 +38,86 @@ const struct kvm_stats_header kvm_vcpu_stats_header = {
 		       sizeof(kvm_vcpu_stats_desc),
 };
 
+#ifdef CONFIG_FPU
+static void kvm_riscv_vcpu_fp_reset(struct kvm_vcpu *vcpu)
+{
+	unsigned long isa = vcpu->arch.isa;
+	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
+
+	cntx->sstatus &= ~SR_FS;
+	if (riscv_isa_extension_available(&isa, f) ||
+	    riscv_isa_extension_available(&isa, d))
+		cntx->sstatus |= SR_FS_INITIAL;
+	else
+		cntx->sstatus |= SR_FS_OFF;
+}
+
+static void kvm_riscv_vcpu_fp_clean(struct kvm_cpu_context *cntx)
+{
+	cntx->sstatus &= ~SR_FS;
+	cntx->sstatus |= SR_FS_CLEAN;
+}
+
+static void kvm_riscv_vcpu_guest_fp_save(struct kvm_cpu_context *cntx,
+					 unsigned long isa)
+{
+	if ((cntx->sstatus & SR_FS) == SR_FS_DIRTY) {
+		if (riscv_isa_extension_available(&isa, d))
+			__kvm_riscv_fp_d_save(cntx);
+		else if (riscv_isa_extension_available(&isa, f))
+			__kvm_riscv_fp_f_save(cntx);
+		kvm_riscv_vcpu_fp_clean(cntx);
+	}
+}
+
+static void kvm_riscv_vcpu_guest_fp_restore(struct kvm_cpu_context *cntx,
+					    unsigned long isa)
+{
+	if ((cntx->sstatus & SR_FS) != SR_FS_OFF) {
+		if (riscv_isa_extension_available(&isa, d))
+			__kvm_riscv_fp_d_restore(cntx);
+		else if (riscv_isa_extension_available(&isa, f))
+			__kvm_riscv_fp_f_restore(cntx);
+		kvm_riscv_vcpu_fp_clean(cntx);
+	}
+}
+
+static void kvm_riscv_vcpu_host_fp_save(struct kvm_cpu_context *cntx)
+{
+	/* No need to check host sstatus as it can be modified outside */
+	if (riscv_isa_extension_available(NULL, d))
+		__kvm_riscv_fp_d_save(cntx);
+	else if (riscv_isa_extension_available(NULL, f))
+		__kvm_riscv_fp_f_save(cntx);
+}
+
+static void kvm_riscv_vcpu_host_fp_restore(struct kvm_cpu_context *cntx)
+{
+	if (riscv_isa_extension_available(NULL, d))
+		__kvm_riscv_fp_d_restore(cntx);
+	else if (riscv_isa_extension_available(NULL, f))
+		__kvm_riscv_fp_f_restore(cntx);
+}
+#else
+static void kvm_riscv_vcpu_fp_reset(struct kvm_vcpu *vcpu)
+{
+}
+static void kvm_riscv_vcpu_guest_fp_save(struct kvm_cpu_context *cntx,
+					 unsigned long isa)
+{
+}
+static void kvm_riscv_vcpu_guest_fp_restore(struct kvm_cpu_context *cntx,
+					    unsigned long isa)
+{
+}
+static void kvm_riscv_vcpu_host_fp_save(struct kvm_cpu_context *cntx)
+{
+}
+static void kvm_riscv_vcpu_host_fp_restore(struct kvm_cpu_context *cntx)
+{
+}
+#endif
+
 #define KVM_RISCV_ISA_ALLOWED	(riscv_isa_extension_mask(a) | \
 				 riscv_isa_extension_mask(c) | \
 				 riscv_isa_extension_mask(d) | \
@@ -58,6 +138,8 @@ static void kvm_riscv_reset_vcpu(struct kvm_vcpu *vcpu)
 
 	memcpy(cntx, reset_cntx, sizeof(*cntx));
 
+	kvm_riscv_vcpu_fp_reset(vcpu);
+
 	kvm_riscv_vcpu_timer_reset(vcpu);
 
 	WRITE_ONCE(vcpu->arch.irqs_pending, 0);
@@ -192,6 +274,7 @@ static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
 			vcpu->arch.isa = reg_val;
 			vcpu->arch.isa &= riscv_isa_extension_base(NULL);
 			vcpu->arch.isa &= KVM_RISCV_ISA_ALLOWED;
+			kvm_riscv_vcpu_fp_reset(vcpu);
 		} else {
 			return -EOPNOTSUPP;
 		}
@@ -593,6 +676,10 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 
 	kvm_riscv_vcpu_timer_restore(vcpu);
 
+	kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context);
+	kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context,
+					vcpu->arch.isa);
+
 	vcpu->cpu = cpu;
 }
 
@@ -602,6 +689,10 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 
 	vcpu->cpu = -1;
 
+	kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context,
+				     vcpu->arch.isa);
+	kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context);
+
 	csr_write(CSR_HGATP, 0);
 
 	csr->vsstatus = csr_read(CSR_VSSTATUS);
diff --git a/arch/riscv/kvm/vcpu_switch.S b/arch/riscv/kvm/vcpu_switch.S
index e22721e1b892..029a28a195c6 100644
--- a/arch/riscv/kvm/vcpu_switch.S
+++ b/arch/riscv/kvm/vcpu_switch.S
@@ -224,3 +224,177 @@ ENTRY(__kvm_riscv_unpriv_trap)
 	REG_S	a1, (KVM_ARCH_TRAP_HTINST)(a0)
 	sret
 ENDPROC(__kvm_riscv_unpriv_trap)
+
+#ifdef	CONFIG_FPU
+	.align 3
+	.global __kvm_riscv_fp_f_save
+__kvm_riscv_fp_f_save:
+	csrr t2, CSR_SSTATUS
+	li t1, SR_FS
+	csrs CSR_SSTATUS, t1
+	frcsr t0
+	fsw f0,  KVM_ARCH_FP_F_F0(a0)
+	fsw f1,  KVM_ARCH_FP_F_F1(a0)
+	fsw f2,  KVM_ARCH_FP_F_F2(a0)
+	fsw f3,  KVM_ARCH_FP_F_F3(a0)
+	fsw f4,  KVM_ARCH_FP_F_F4(a0)
+	fsw f5,  KVM_ARCH_FP_F_F5(a0)
+	fsw f6,  KVM_ARCH_FP_F_F6(a0)
+	fsw f7,  KVM_ARCH_FP_F_F7(a0)
+	fsw f8,  KVM_ARCH_FP_F_F8(a0)
+	fsw f9,  KVM_ARCH_FP_F_F9(a0)
+	fsw f10, KVM_ARCH_FP_F_F10(a0)
+	fsw f11, KVM_ARCH_FP_F_F11(a0)
+	fsw f12, KVM_ARCH_FP_F_F12(a0)
+	fsw f13, KVM_ARCH_FP_F_F13(a0)
+	fsw f14, KVM_ARCH_FP_F_F14(a0)
+	fsw f15, KVM_ARCH_FP_F_F15(a0)
+	fsw f16, KVM_ARCH_FP_F_F16(a0)
+	fsw f17, KVM_ARCH_FP_F_F17(a0)
+	fsw f18, KVM_ARCH_FP_F_F18(a0)
+	fsw f19, KVM_ARCH_FP_F_F19(a0)
+	fsw f20, KVM_ARCH_FP_F_F20(a0)
+	fsw f21, KVM_ARCH_FP_F_F21(a0)
+	fsw f22, KVM_ARCH_FP_F_F22(a0)
+	fsw f23, KVM_ARCH_FP_F_F23(a0)
+	fsw f24, KVM_ARCH_FP_F_F24(a0)
+	fsw f25, KVM_ARCH_FP_F_F25(a0)
+	fsw f26, KVM_ARCH_FP_F_F26(a0)
+	fsw f27, KVM_ARCH_FP_F_F27(a0)
+	fsw f28, KVM_ARCH_FP_F_F28(a0)
+	fsw f29, KVM_ARCH_FP_F_F29(a0)
+	fsw f30, KVM_ARCH_FP_F_F30(a0)
+	fsw f31, KVM_ARCH_FP_F_F31(a0)
+	sw t0, KVM_ARCH_FP_F_FCSR(a0)
+	csrw CSR_SSTATUS, t2
+	ret
+
+	.align 3
+	.global __kvm_riscv_fp_d_save
+__kvm_riscv_fp_d_save:
+	csrr t2, CSR_SSTATUS
+	li t1, SR_FS
+	csrs CSR_SSTATUS, t1
+	frcsr t0
+	fsd f0,  KVM_ARCH_FP_D_F0(a0)
+	fsd f1,  KVM_ARCH_FP_D_F1(a0)
+	fsd f2,  KVM_ARCH_FP_D_F2(a0)
+	fsd f3,  KVM_ARCH_FP_D_F3(a0)
+	fsd f4,  KVM_ARCH_FP_D_F4(a0)
+	fsd f5,  KVM_ARCH_FP_D_F5(a0)
+	fsd f6,  KVM_ARCH_FP_D_F6(a0)
+	fsd f7,  KVM_ARCH_FP_D_F7(a0)
+	fsd f8,  KVM_ARCH_FP_D_F8(a0)
+	fsd f9,  KVM_ARCH_FP_D_F9(a0)
+	fsd f10, KVM_ARCH_FP_D_F10(a0)
+	fsd f11, KVM_ARCH_FP_D_F11(a0)
+	fsd f12, KVM_ARCH_FP_D_F12(a0)
+	fsd f13, KVM_ARCH_FP_D_F13(a0)
+	fsd f14, KVM_ARCH_FP_D_F14(a0)
+	fsd f15, KVM_ARCH_FP_D_F15(a0)
+	fsd f16, KVM_ARCH_FP_D_F16(a0)
+	fsd f17, KVM_ARCH_FP_D_F17(a0)
+	fsd f18, KVM_ARCH_FP_D_F18(a0)
+	fsd f19, KVM_ARCH_FP_D_F19(a0)
+	fsd f20, KVM_ARCH_FP_D_F20(a0)
+	fsd f21, KVM_ARCH_FP_D_F21(a0)
+	fsd f22, KVM_ARCH_FP_D_F22(a0)
+	fsd f23, KVM_ARCH_FP_D_F23(a0)
+	fsd f24, KVM_ARCH_FP_D_F24(a0)
+	fsd f25, KVM_ARCH_FP_D_F25(a0)
+	fsd f26, KVM_ARCH_FP_D_F26(a0)
+	fsd f27, KVM_ARCH_FP_D_F27(a0)
+	fsd f28, KVM_ARCH_FP_D_F28(a0)
+	fsd f29, KVM_ARCH_FP_D_F29(a0)
+	fsd f30, KVM_ARCH_FP_D_F30(a0)
+	fsd f31, KVM_ARCH_FP_D_F31(a0)
+	sw t0, KVM_ARCH_FP_D_FCSR(a0)
+	csrw CSR_SSTATUS, t2
+	ret
+
+	.align 3
+	.global __kvm_riscv_fp_f_restore
+__kvm_riscv_fp_f_restore:
+	csrr t2, CSR_SSTATUS
+	li t1, SR_FS
+	lw t0, KVM_ARCH_FP_F_FCSR(a0)
+	csrs CSR_SSTATUS, t1
+	flw f0,  KVM_ARCH_FP_F_F0(a0)
+	flw f1,  KVM_ARCH_FP_F_F1(a0)
+	flw f2,  KVM_ARCH_FP_F_F2(a0)
+	flw f3,  KVM_ARCH_FP_F_F3(a0)
+	flw f4,  KVM_ARCH_FP_F_F4(a0)
+	flw f5,  KVM_ARCH_FP_F_F5(a0)
+	flw f6,  KVM_ARCH_FP_F_F6(a0)
+	flw f7,  KVM_ARCH_FP_F_F7(a0)
+	flw f8,  KVM_ARCH_FP_F_F8(a0)
+	flw f9,  KVM_ARCH_FP_F_F9(a0)
+	flw f10, KVM_ARCH_FP_F_F10(a0)
+	flw f11, KVM_ARCH_FP_F_F11(a0)
+	flw f12, KVM_ARCH_FP_F_F12(a0)
+	flw f13, KVM_ARCH_FP_F_F13(a0)
+	flw f14, KVM_ARCH_FP_F_F14(a0)
+	flw f15, KVM_ARCH_FP_F_F15(a0)
+	flw f16, KVM_ARCH_FP_F_F16(a0)
+	flw f17, KVM_ARCH_FP_F_F17(a0)
+	flw f18, KVM_ARCH_FP_F_F18(a0)
+	flw f19, KVM_ARCH_FP_F_F19(a0)
+	flw f20, KVM_ARCH_FP_F_F20(a0)
+	flw f21, KVM_ARCH_FP_F_F21(a0)
+	flw f22, KVM_ARCH_FP_F_F22(a0)
+	flw f23, KVM_ARCH_FP_F_F23(a0)
+	flw f24, KVM_ARCH_FP_F_F24(a0)
+	flw f25, KVM_ARCH_FP_F_F25(a0)
+	flw f26, KVM_ARCH_FP_F_F26(a0)
+	flw f27, KVM_ARCH_FP_F_F27(a0)
+	flw f28, KVM_ARCH_FP_F_F28(a0)
+	flw f29, KVM_ARCH_FP_F_F29(a0)
+	flw f30, KVM_ARCH_FP_F_F30(a0)
+	flw f31, KVM_ARCH_FP_F_F31(a0)
+	fscsr t0
+	csrw CSR_SSTATUS, t2
+	ret
+
+	.align 3
+	.global __kvm_riscv_fp_d_restore
+__kvm_riscv_fp_d_restore:
+	csrr t2, CSR_SSTATUS
+	li t1, SR_FS
+	lw t0, KVM_ARCH_FP_D_FCSR(a0)
+	csrs CSR_SSTATUS, t1
+	fld f0,  KVM_ARCH_FP_D_F0(a0)
+	fld f1,  KVM_ARCH_FP_D_F1(a0)
+	fld f2,  KVM_ARCH_FP_D_F2(a0)
+	fld f3,  KVM_ARCH_FP_D_F3(a0)
+	fld f4,  KVM_ARCH_FP_D_F4(a0)
+	fld f5,  KVM_ARCH_FP_D_F5(a0)
+	fld f6,  KVM_ARCH_FP_D_F6(a0)
+	fld f7,  KVM_ARCH_FP_D_F7(a0)
+	fld f8,  KVM_ARCH_FP_D_F8(a0)
+	fld f9,  KVM_ARCH_FP_D_F9(a0)
+	fld f10, KVM_ARCH_FP_D_F10(a0)
+	fld f11, KVM_ARCH_FP_D_F11(a0)
+	fld f12, KVM_ARCH_FP_D_F12(a0)
+	fld f13, KVM_ARCH_FP_D_F13(a0)
+	fld f14, KVM_ARCH_FP_D_F14(a0)
+	fld f15, KVM_ARCH_FP_D_F15(a0)
+	fld f16, KVM_ARCH_FP_D_F16(a0)
+	fld f17, KVM_ARCH_FP_D_F17(a0)
+	fld f18, KVM_ARCH_FP_D_F18(a0)
+	fld f19, KVM_ARCH_FP_D_F19(a0)
+	fld f20, KVM_ARCH_FP_D_F20(a0)
+	fld f21, KVM_ARCH_FP_D_F21(a0)
+	fld f22, KVM_ARCH_FP_D_F22(a0)
+	fld f23, KVM_ARCH_FP_D_F23(a0)
+	fld f24, KVM_ARCH_FP_D_F24(a0)
+	fld f25, KVM_ARCH_FP_D_F25(a0)
+	fld f26, KVM_ARCH_FP_D_F26(a0)
+	fld f27, KVM_ARCH_FP_D_F27(a0)
+	fld f28, KVM_ARCH_FP_D_F28(a0)
+	fld f29, KVM_ARCH_FP_D_F29(a0)
+	fld f30, KVM_ARCH_FP_D_F30(a0)
+	fld f31, KVM_ARCH_FP_D_F31(a0)
+	fscsr t0
+	csrw CSR_SSTATUS, t2
+	ret
+#endif