diff mbox series

[v2,15/94] KVM: arm64: nv: Handle SPSR_EL2 specially

Message ID 20200211174938.27809-16-maz@kernel.org (mailing list archive)
State New, archived
Headers show
Series KVM: arm64: ARMv8.3/8.4 Nested Virtualization support | expand

Commit Message

Marc Zyngier Feb. 11, 2020, 5:48 p.m. UTC
SPSR_EL2 needs special attention when running nested on ARMv8.3:

If taking an exception while running at vEL2 (actually EL1), the
HW will update the SPSR_EL1 register with the EL1 mode. We need
to track this in order to make sure that accesses to the virtual
view of SPSR_EL2 is correct.

To do so, we place an illegal value in SPSR_EL1.M, and patch it
accordingly if required when accessing it.

Signed-off-by: Marc Zyngier <maz@kernel.org>
---
 arch/arm64/include/asm/kvm_emulate.h | 45 ++++++++++++++++++++++++++++
 arch/arm64/kvm/sys_regs.c            | 23 ++++++++++++--
 2 files changed, 66 insertions(+), 2 deletions(-)

Comments

Mark Rutland Feb. 17, 2020, 3:04 p.m. UTC | #1
On Tue, Feb 11, 2020 at 05:48:19PM +0000, Marc Zyngier wrote:
> SPSR_EL2 needs special attention when running nested on ARMv8.3:
> 
> If taking an exception while running at vEL2 (actually EL1), the
> HW will update the SPSR_EL1 register with the EL1 mode. We need
> to track this in order to make sure that accesses to the virtual
> view of SPSR_EL2 is correct.
> 
> To do so, we place an illegal value in SPSR_EL1.M, and patch it
> accordingly if required when accessing it.
> 
> Signed-off-by: Marc Zyngier <maz@kernel.org>
> ---
>  arch/arm64/include/asm/kvm_emulate.h | 45 ++++++++++++++++++++++++++++
>  arch/arm64/kvm/sys_regs.c            | 23 ++++++++++++--
>  2 files changed, 66 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
> index 486978d0346b..26552c8571cb 100644
> --- a/arch/arm64/include/asm/kvm_emulate.h
> +++ b/arch/arm64/include/asm/kvm_emulate.h
> @@ -277,11 +277,51 @@ static inline bool is_hyp_ctxt(const struct kvm_vcpu *vcpu)
>  	return __is_hyp_ctxt(&vcpu->arch.ctxt);
>  }
>  
> +static inline u64 __fixup_spsr_el2_write(struct kvm_cpu_context *ctxt, u64 val)
> +{
> +	if (!__vcpu_el2_e2h_is_set(ctxt)) {
> +		/*
> +		 * Clear the .M field when writing SPSR to the CPU, so that we
> +		 * can detect when the CPU clobbered our SPSR copy during a
> +		 * local exception.
> +		 */
> +		val &= ~0xc;

Probably worth making that UL(0xc) so that we consistently treat the
SPSRs as 64-bits.

Perhaps:

#define SPSR_EL2_M_3_2	UL(0xc)

... so that we can use it consistently below?

> +	}
> +
> +	return val;
> +}
> +
> +static inline u64 __fixup_spsr_el2_read(const struct kvm_cpu_context *ctxt, u64 val)
> +{
> +	if (__vcpu_el2_e2h_is_set(ctxt))
> +		return val;
> +
> +	/*
> +	 * SPSR.M == 0 means the CPU has not touched the SPSR, so the
> +	 * register has still the value we saved on the last write.
> +	 */
> +	if ((val & 0xc) == 0)
> +		return ctxt->sys_regs[SPSR_EL2];
> +
> +	/*
> +	 * Otherwise there was a "local" exception on the CPU,

Perhaps "exit-less" rather than "local"?

Thanks,
Mark.

> +	 * which from the guest's point of view was being taken from
> +	 * EL2 to EL2, although it actually happened to be from
> +	 * EL1 to EL1.
> +	 * So we need to fix the .M field in SPSR, to make it look
> +	 * like EL2, which is what the guest would expect.
> +	 */
> +	return (val & ~0x0c) | CurrentEL_EL2;
> +}
> +
>  static inline unsigned long vcpu_read_spsr(const struct kvm_vcpu *vcpu)
>  {
>  	if (vcpu_mode_is_32bit(vcpu))
>  		return vcpu_read_spsr32(vcpu);
>  
> +	if (unlikely(vcpu_mode_el2(vcpu)))
> +		return vcpu_read_sys_reg(vcpu, SPSR_EL2);
> +
>  	if (vcpu->arch.sysregs_loaded_on_cpu)
>  		return read_sysreg_el1(SYS_SPSR);
>  	else
> @@ -295,6 +335,11 @@ static inline void vcpu_write_spsr(struct kvm_vcpu *vcpu, unsigned long v)
>  		return;
>  	}
>  
> +	if (unlikely(vcpu_mode_el2(vcpu))) {
> +		vcpu_write_sys_reg(vcpu, v, SPSR_EL2);
> +		return;
> +	}
> +
>  	if (vcpu->arch.sysregs_loaded_on_cpu)
>  		write_sysreg_el1(v, SYS_SPSR);
>  	else
> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> index 64be9f452ad6..8c7d3d410689 100644
> --- a/arch/arm64/kvm/sys_regs.c
> +++ b/arch/arm64/kvm/sys_regs.c
> @@ -258,11 +258,14 @@ u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
>  			goto memory_read;
>  
>  		/*
> -		 * ELR_EL2 is special cased for now.
> +		 * ELR_EL2 and SPSR_EL2 are special cased for now.
>  		 */
>  		switch (reg) {
>  		case ELR_EL2:
>  			return read_sysreg_el1(SYS_ELR);
> +		case SPSR_EL2:
> +			val = read_sysreg_el1(SYS_SPSR);
> +			return __fixup_spsr_el2_read(&vcpu->arch.ctxt, val);
>  		}
>  
>  		/*
> @@ -319,6 +322,10 @@ void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
>  		case ELR_EL2:
>  			write_sysreg_el1(val, SYS_ELR);
>  			return;
> +		case SPSR_EL2:
> +			val = __fixup_spsr_el2_write(&vcpu->arch.ctxt, val);
> +			write_sysreg_el1(val, SYS_SPSR);
> +			return;
>  		}
>  
>  		/* No EL1 counterpart? We're done here.? */
> @@ -1589,6 +1596,18 @@ static bool access_sp_el1(struct kvm_vcpu *vcpu,
>  	return true;
>  }
>  
> +static bool access_spsr_el2(struct kvm_vcpu *vcpu,
> +			    struct sys_reg_params *p,
> +			    const struct sys_reg_desc *r)
> +{
> +	if (p->is_write)
> +		vcpu_write_sys_reg(vcpu, p->regval, SPSR_EL2);
> +	else
> +		p->regval = vcpu_read_sys_reg(vcpu, SPSR_EL2);
> +
> +	return true;
> +}
> +
>  /*
>   * Architected system registers.
>   * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
> @@ -1899,7 +1918,7 @@ static const struct sys_reg_desc sys_reg_descs[] = {
>  	{ SYS_DESC(SYS_VTCR_EL2), access_rw, reset_val, VTCR_EL2, 0 },
>  
>  	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
> -	{ SYS_DESC(SYS_SPSR_EL2), access_rw, reset_val, SPSR_EL2, 0 },
> +	{ SYS_DESC(SYS_SPSR_EL2), access_spsr_el2, reset_val, SPSR_EL2, 0 },
>  	{ SYS_DESC(SYS_ELR_EL2), access_rw, reset_val, ELR_EL2, 0 },
>  	{ SYS_DESC(SYS_SP_EL1), access_sp_el1},
>  
> -- 
> 2.20.1
> 
> _______________________________________________
> kvmarm mailing list
> kvmarm@lists.cs.columbia.edu
> https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
diff mbox series

Patch

diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
index 486978d0346b..26552c8571cb 100644
--- a/arch/arm64/include/asm/kvm_emulate.h
+++ b/arch/arm64/include/asm/kvm_emulate.h
@@ -277,11 +277,51 @@  static inline bool is_hyp_ctxt(const struct kvm_vcpu *vcpu)
 	return __is_hyp_ctxt(&vcpu->arch.ctxt);
 }
 
+static inline u64 __fixup_spsr_el2_write(struct kvm_cpu_context *ctxt, u64 val)
+{
+	if (!__vcpu_el2_e2h_is_set(ctxt)) {
+		/*
+		 * Clear the .M field when writing SPSR to the CPU, so that we
+		 * can detect when the CPU clobbered our SPSR copy during a
+		 * local exception.
+		 */
+		val &= ~0xc;
+	}
+
+	return val;
+}
+
+static inline u64 __fixup_spsr_el2_read(const struct kvm_cpu_context *ctxt, u64 val)
+{
+	if (__vcpu_el2_e2h_is_set(ctxt))
+		return val;
+
+	/*
+	 * SPSR.M == 0 means the CPU has not touched the SPSR, so the
+	 * register has still the value we saved on the last write.
+	 */
+	if ((val & 0xc) == 0)
+		return ctxt->sys_regs[SPSR_EL2];
+
+	/*
+	 * Otherwise there was a "local" exception on the CPU,
+	 * which from the guest's point of view was being taken from
+	 * EL2 to EL2, although it actually happened to be from
+	 * EL1 to EL1.
+	 * So we need to fix the .M field in SPSR, to make it look
+	 * like EL2, which is what the guest would expect.
+	 */
+	return (val & ~0x0c) | CurrentEL_EL2;
+}
+
 static inline unsigned long vcpu_read_spsr(const struct kvm_vcpu *vcpu)
 {
 	if (vcpu_mode_is_32bit(vcpu))
 		return vcpu_read_spsr32(vcpu);
 
+	if (unlikely(vcpu_mode_el2(vcpu)))
+		return vcpu_read_sys_reg(vcpu, SPSR_EL2);
+
 	if (vcpu->arch.sysregs_loaded_on_cpu)
 		return read_sysreg_el1(SYS_SPSR);
 	else
@@ -295,6 +335,11 @@  static inline void vcpu_write_spsr(struct kvm_vcpu *vcpu, unsigned long v)
 		return;
 	}
 
+	if (unlikely(vcpu_mode_el2(vcpu))) {
+		vcpu_write_sys_reg(vcpu, v, SPSR_EL2);
+		return;
+	}
+
 	if (vcpu->arch.sysregs_loaded_on_cpu)
 		write_sysreg_el1(v, SYS_SPSR);
 	else
diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
index 64be9f452ad6..8c7d3d410689 100644
--- a/arch/arm64/kvm/sys_regs.c
+++ b/arch/arm64/kvm/sys_regs.c
@@ -258,11 +258,14 @@  u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
 			goto memory_read;
 
 		/*
-		 * ELR_EL2 is special cased for now.
+		 * ELR_EL2 and SPSR_EL2 are special cased for now.
 		 */
 		switch (reg) {
 		case ELR_EL2:
 			return read_sysreg_el1(SYS_ELR);
+		case SPSR_EL2:
+			val = read_sysreg_el1(SYS_SPSR);
+			return __fixup_spsr_el2_read(&vcpu->arch.ctxt, val);
 		}
 
 		/*
@@ -319,6 +322,10 @@  void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
 		case ELR_EL2:
 			write_sysreg_el1(val, SYS_ELR);
 			return;
+		case SPSR_EL2:
+			val = __fixup_spsr_el2_write(&vcpu->arch.ctxt, val);
+			write_sysreg_el1(val, SYS_SPSR);
+			return;
 		}
 
 		/* No EL1 counterpart? We're done here.? */
@@ -1589,6 +1596,18 @@  static bool access_sp_el1(struct kvm_vcpu *vcpu,
 	return true;
 }
 
+static bool access_spsr_el2(struct kvm_vcpu *vcpu,
+			    struct sys_reg_params *p,
+			    const struct sys_reg_desc *r)
+{
+	if (p->is_write)
+		vcpu_write_sys_reg(vcpu, p->regval, SPSR_EL2);
+	else
+		p->regval = vcpu_read_sys_reg(vcpu, SPSR_EL2);
+
+	return true;
+}
+
 /*
  * Architected system registers.
  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
@@ -1899,7 +1918,7 @@  static const struct sys_reg_desc sys_reg_descs[] = {
 	{ SYS_DESC(SYS_VTCR_EL2), access_rw, reset_val, VTCR_EL2, 0 },
 
 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
-	{ SYS_DESC(SYS_SPSR_EL2), access_rw, reset_val, SPSR_EL2, 0 },
+	{ SYS_DESC(SYS_SPSR_EL2), access_spsr_el2, reset_val, SPSR_EL2, 0 },
 	{ SYS_DESC(SYS_ELR_EL2), access_rw, reset_val, ELR_EL2, 0 },
 	{ SYS_DESC(SYS_SP_EL1), access_sp_el1},