[PATCH v4 00/18] KVM: arm64: Confine protected VM vCPU state to EL2

[PATCH v4 00/18] KVM: arm64: Confine protected VM vCPU state to EL2

[Fuad Tabba]

Hi folks,

Changes since v3 [1]:
  - A protected vCPU that EL2 holds powered off reads as
    KVM_MP_STATE_UNINITIALIZED, in place of a new bool. (Marc)
  - The marshalling and PSCI patches access the host copy through the
    vCPU accessors, with every assignment on one line. (Marc)
  - The PC adjustment pair is kvm_adjust_pc_get()/put() and tests for
    the hyp vCPU directly. (Marc)
  - CPU_OFF clears the reset flag before its release of OFF, CPU_ON's
    relaxed cmpxchg is ordered by its control dependency, and the
    CPU_ON rollback stores OFF with a release. pkvm_reset_vcpu()
    carries the CPU_ON/CPU_OFF sequence as a comment and drops its
    WARN_ON(). (Vincent, Will)
  - Collected Reviewed-bys.

Following the vCPU state-sync series [3], this series completes the
job for protected VMs: a protected guest's register state stays at
EL2, and the host sees only what handling each exit requires.

EL2 marshals a protected vCPU's state per exception class instead of
copying the whole context both ways. It owns the vCPU's trap
configuration, system register reset and HVC handling, and implements
PSCI itself: AFFINITY_INFO is left to the host, and CPU_ON and CPU_OFF
are resolved at EL2 with the host only scheduling the target or
stopping it. A protected guest's TRNG calls, which the host handled
until now, return NOT_SUPPORTED until TRNG for protected guests
follows. EL2 implements PSCI 1.1, so a protected guest also loses the
functions the host supports above that version, SYSTEM_OFF2 included.
Host ioctls that would reach the state EL2 owns fail with -EPERM, so a
protected VM's state isn't save/restorable. All of this is scoped
to KVM_VM_TYPE_ARM_PROTECTED, and pkvm.rst describes the resulting API.

The kvmtool changes that go with this are posted separately [4].

Patches 1 to 3 go out separately: the HCR_EL2.VSE fix [5], and the
host vCPU VM read and pin fixes, patches 4 and 5 of the host hypercall
fixes series [2]. None of the three is part of this series. They're
carried so the series applies as is and Sashiko can run on it.

The KVM_ARM_PREFERRED_TARGET documentation fix [6] went out just ahead
of v1. Nothing here depends on it to apply, but patch 18 documents vCPU
feature availability as something the capabilities report, while
api.rst 4.83 still points userspace at a bitmap that has always been
empty.

The series is structured as follows:

  01:     The HCR_EL2.VSE fix, posted separately.
  02-03:  The host vCPU VM read and pin fixes, posted separately.
  04:     Steal time disabled for protected VMs.
  05-06:  Per-exception-class entry handlers; EL2 owns a protected
          vCPU's trap configuration.
  07-09:  Timer state, system register reset and HVC handling at EL2.
  10-11:  PSCI at EL2, and the KVM_ARM_VCPU_INIT and PSCI version
          restrictions.
  12-14:  Host PC adjustments blocked; an UNDEF at EL2 for exit
          classes the host doesn't emulate; per-class state
          marshalling.
  15-16:  Host changes to private state, and host power-on of a vCPU
          EL2 holds powered off, rejected.
  17:     Capability allowlist.
  18:     Documentation.

Still to come: selftests, TRNG, self-hosted debug and SVE for protected
guests, and much more, as separate series.

Based on v7.3-rc3 (fd73f4a665989).

Cheers,
/fuad

[1] https://lore.kernel.org/all/20260914113338.159227-1-fuad.tabba@linux.dev/
[2] https://lore.kernel.org/all/20260915123846.2317931-1-fuad.tabba@linux.dev/
[3] https://lore.kernel.org/all/20260729131823.2021516-1-fuad.tabba@linux.dev/
[4] https://lore.kernel.org/all/20260831192406.1341841-1-fuad.tabba@linux.dev/
[5] https://lore.kernel.org/all/20260829071120.2522788-1-fuad.tabba@linux.dev/
[6] https://lore.kernel.org/all/20260831162815.269851-1-fuad.tabba@linux.dev/

Fuad Tabba (16):
  KVM: arm64: Sync HCR_EL2.VSE back to the host vCPU under pKVM
  KVM: arm64: Validate the host vCPU's VM before reading it under pKVM
  KVM: arm64: Pin the host vCPU before adjusting its PC under pKVM
  KVM: arm64: Disable steal time for protected VMs
  KVM: arm64: Skip fixed-feature state flush for protected vCPUs
  KVM: arm64: Add system register reset framework for protected VMs
  KVM: arm64: Implement HVC handling for protected guests at EL2
  KVM: arm64: Handle PSCI calls for protected VMs at EL2
  KVM: arm64: Restrict KVM_ARM_VCPU_INIT and PSCI version for protected
    VMs
  KVM: arm64: Prevent host PC adjustments for protected vCPUs
  KVM: arm64: Inject an UNDEF at EL2 for unhandled protected guest exits
  KVM: arm64: Add per-EC entry/exit state marshalling for protected
    guests
  KVM: arm64: Reject host access to protected VM private state
  KVM: arm64: Reject host power-on of a vCPU that EL2 holds powered off
  KVM: arm64: Advertise the capabilities that protected VMs support
  KVM: arm64: Document the protected VM userspace API

Marc Zyngier (2):
  KVM: arm64: Introduce per-EC entry handlers for pKVM
  KVM: arm64: Add {flush,sync}_hyp_timer_state() primitives

 Documentation/virt/kvm/api.rst                |  27 +-
 .../virt/kvm/arm/fw-pseudo-registers.rst      |   2 +
 Documentation/virt/kvm/arm/pkvm.rst           | 164 ++++-
 Documentation/virt/kvm/devices/vcpu.rst       |   3 +-
 arch/arm64/include/asm/kvm_host.h             |  22 +-
 arch/arm64/include/asm/kvm_hyp.h              |   4 +
 arch/arm64/include/asm/kvm_pkvm.h             |  33 +
 arch/arm64/kvm/arm.c                          |  76 ++-
 arch/arm64/kvm/guest.c                        |  11 +
 arch/arm64/kvm/hyp/exception.c                |  26 +-
 arch/arm64/kvm/hyp/include/hyp/adjust_pc.h    |  48 ++
 arch/arm64/kvm/hyp/include/nvhe/pkvm.h        |  20 +
 arch/arm64/kvm/hyp/nvhe/hyp-main.c            | 595 +++++++++++++++++-
 arch/arm64/kvm/hyp/nvhe/pkvm.c                | 422 ++++++++++++-
 arch/arm64/kvm/hyp/nvhe/switch.c              |  29 +-
 arch/arm64/kvm/hyp/nvhe/sys_regs.c            |  60 ++
 arch/arm64/kvm/hypercalls.c                   |   7 +
 arch/arm64/kvm/pkvm.c                         |  21 +-
 arch/arm64/kvm/psci.c                         |  11 +-
 arch/arm64/kvm/pvtime.c                       |  10 +-
 20 files changed, 1503 insertions(+), 88 deletions(-)


base-commit: fd73f4a6659897191fa0d40695fe370925dd3780
-- 
2.39.5

[PATCH v4 01/18] KVM: arm64: Sync HCR_EL2.VSE back to the host vCPU under pKVM

[Fuad Tabba]

flush_hyp_vcpu() copies HCR_EL2.VSE from the host vCPU into the hyp
vCPU on every entry, and nothing copies it back. Once the guest takes
the vSError the hyp vCPU's copy clears with the hardware bit while the
host's stays set, so the next entry pends the same SError again,
KVM_GET_VCPU_EVENTS keeps reporting it, and kvm_arch_vcpu_runnable()
never lets the vCPU block in WFI.

Reflect VSE back on every exit, which the flush-side comment already
states.

Fixes: 734dc8c01c838 ("KVM: arm64: Implement lazy vCPU state sync for non-protected guests")
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/hyp-main.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 9a3b92e626adb..b6bfe502bcd04 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -276,6 +276,10 @@ static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 
 	host_vcpu->arch.iflags		= hyp_vcpu->vcpu.arch.iflags;
 
+	/* Cleared by hardware once the guest takes the vSError. */
+	host_vcpu->arch.hcr_el2 &= ~HCR_VSE;
+	host_vcpu->arch.hcr_el2 |= hyp_vcpu->vcpu.arch.hcr_el2 & HCR_VSE;
+
 	sync_hyp_vgic_state(hyp_vcpu);
 }
 
-- 
2.39.5

[PATCH v4 02/18] KVM: arm64: Validate the host vCPU's VM before reading it under pKVM

[Fuad Tabba]

On an MTE-capable host under pKVM, enter_exception64() reads the VM's
MTE flag through vcpu->kvm, which for a host vCPU is a host-writable
pointer nothing validates. The host can point it at any address in the
hyp linear map and read back bit 1 of that word through PSR_TCO in the
vCPU's CPSR, or panic the hypervisor with an unmapped one.

Get the VM through a get/put pair around the read: a loaded vCPU's is
the hyp VM, an unloaded host vCPU's is read once and pinned, and a
pointer the host never shared leaves TCO clear.

Fixes: ea7fc1bb1cd1b ("KVM: arm64: Introduce MTE VM feature")
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://lore.kernel.org/all/20260914070536.877D91F000FF@smtp.kernel.org/
Reviewed-by: Vincent Donnefort <vdonnefort@google.com>
Reviewed-by: Joey Gouly <joey.gouly@arm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/exception.c             |  5 ++-
 arch/arm64/kvm/hyp/include/hyp/adjust_pc.h | 28 ++++++++++++++++
 arch/arm64/kvm/hyp/include/nvhe/pkvm.h     |  2 ++
 arch/arm64/kvm/hyp/nvhe/pkvm.c             | 37 ++++++++++++++++++++++
 4 files changed, 71 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/kvm/hyp/exception.c b/arch/arm64/kvm/hyp/exception.c
index 754e2dc1df54a..6e60d890afa4a 100644
--- a/arch/arm64/kvm/hyp/exception.c
+++ b/arch/arm64/kvm/hyp/exception.c
@@ -70,6 +70,7 @@ static void enter_exception64(struct kvm_vcpu *vcpu, unsigned long target_mode,
 			      enum exception_type type)
 {
 	unsigned long sctlr, vbar, old, new, mode;
+	struct kvm *kvm;
 	u64 exc_offset;
 
 	mode = *vcpu_cpsr(vcpu) & (PSR_MODE_MASK | PSR_MODE32_BIT);
@@ -109,8 +110,10 @@ static void enter_exception64(struct kvm_vcpu *vcpu, unsigned long target_mode,
 	new |= (old & PSR_C_BIT);
 	new |= (old & PSR_V_BIT);
 
-	if (kvm_has_mte(kern_hyp_va(vcpu->kvm)))
+	kvm = vcpu_get_kvm(vcpu);
+	if (kvm && kvm_has_mte(kvm))
 		new |= PSR_TCO_BIT;
+	vcpu_put_kvm(vcpu, kvm);
 
 	new |= (old & PSR_DIT_BIT);
 
diff --git a/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h b/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h
index 4fdfeabefeb43..f55950ee2a7e4 100644
--- a/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h
+++ b/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h
@@ -13,6 +13,34 @@
 #include <asm/kvm_emulate.h>
 #include <asm/kvm_host.h>
 
+#ifdef __KVM_NVHE_HYPERVISOR__
+#include <nvhe/pkvm.h>
+
+/* Under pKVM a host vCPU's ->kvm is host-writable. */
+static inline struct kvm *vcpu_get_kvm(struct kvm_vcpu *vcpu)
+{
+	if (is_protected_kvm_enabled())
+		return pkvm_vcpu_get_kvm(vcpu);
+
+	return kern_hyp_va(vcpu->kvm);
+}
+
+static inline void vcpu_put_kvm(struct kvm_vcpu *vcpu, struct kvm *kvm)
+{
+	if (is_protected_kvm_enabled())
+		pkvm_vcpu_put_kvm(vcpu, kvm);
+}
+#else
+static inline struct kvm *vcpu_get_kvm(struct kvm_vcpu *vcpu)
+{
+	return vcpu->kvm;
+}
+
+static inline void vcpu_put_kvm(struct kvm_vcpu *vcpu, struct kvm *kvm)
+{
+}
+#endif
+
 static inline void kvm_skip_instr(struct kvm_vcpu *vcpu)
 {
 	if (vcpu_mode_is_32bit(vcpu)) {
diff --git a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
index c904647d2f760..5ddb407149667 100644
--- a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
+++ b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
@@ -83,6 +83,8 @@ struct pkvm_hyp_vcpu *pkvm_load_hyp_vcpu(pkvm_handle_t handle,
 					 unsigned int vcpu_idx);
 void pkvm_put_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu);
 struct pkvm_hyp_vcpu *pkvm_get_loaded_hyp_vcpu(void);
+struct kvm *pkvm_vcpu_get_kvm(struct kvm_vcpu *vcpu);
+void pkvm_vcpu_put_kvm(struct kvm_vcpu *vcpu, struct kvm *kvm);
 
 struct pkvm_hyp_vm *get_pkvm_hyp_vm(pkvm_handle_t handle);
 struct pkvm_hyp_vm *get_np_pkvm_hyp_vm(pkvm_handle_t handle);
diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index 459bd9eb7e4bc..0e90de74778cf 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -304,6 +304,43 @@ struct pkvm_hyp_vcpu *pkvm_get_loaded_hyp_vcpu(void)
 
 }
 
+static struct pkvm_hyp_vm *pkvm_get_loaded_hyp_vm(struct kvm_vcpu *vcpu)
+{
+	struct pkvm_hyp_vcpu *hyp_vcpu = pkvm_get_loaded_hyp_vcpu();
+
+	if (hyp_vcpu &&
+	    (vcpu == &hyp_vcpu->vcpu || vcpu == hyp_vcpu->host_vcpu))
+		return pkvm_hyp_vcpu_to_hyp_vm(hyp_vcpu);
+
+	return NULL;
+}
+
+/*
+ * A loaded vCPU's VM is the hyp VM. An unloaded host vCPU's is mapped at
+ * EL2 only while pinned, so it's read once and pinned; the pin fails for
+ * memory the host isn't sharing.
+ */
+struct kvm *pkvm_vcpu_get_kvm(struct kvm_vcpu *vcpu)
+{
+	struct pkvm_hyp_vm *hyp_vm = pkvm_get_loaded_hyp_vm(vcpu);
+	struct kvm *kvm;
+
+	if (hyp_vm)
+		return &hyp_vm->kvm;
+
+	kvm = kern_hyp_va(READ_ONCE(vcpu->kvm));
+	if (hyp_pin_shared_mem(kvm, kvm + 1))
+		return NULL;
+
+	return kvm;
+}
+
+void pkvm_vcpu_put_kvm(struct kvm_vcpu *vcpu, struct kvm *kvm)
+{
+	if (kvm && !pkvm_get_loaded_hyp_vm(vcpu))
+		hyp_unpin_shared_mem(kvm, kvm + 1);
+}
+
 struct pkvm_hyp_vm *get_pkvm_hyp_vm(pkvm_handle_t handle)
 {
 	struct pkvm_hyp_vm *hyp_vm;
-- 
2.39.5

[PATCH v4 03/18] KVM: arm64: Pin the host vCPU before adjusting its PC under pKVM

[Fuad Tabba]

Under pKVM, a page the host shares with EL2 is mapped at EL2 only while
it's pinned, and the host vCPU is pinned at its first KVM_RUN.
KVM_SET_VCPU_EVENTS with ext_dabt_pending reaches __kvm_adjust_pc() at
EL2 before that, and dereferencing the unmapped host vCPU panics the
hypervisor. Any process holding /dev/kvm on a pKVM host can trigger it.

Pin the host vCPU around the adjustment when no hyp vCPU is loaded for
it. A loaded hyp vCPU implies it's pinned. A pin fails only for memory
the host isn't sharing, and the request is then dropped like any other
bad host pointer.

Fixes: efa1368ba9f4b ("KVM: arm64: Commit exceptions from KVM_SET_VCPU_EVENTS immediately")
Reviewed-by: Vincent Donnefort <vdonnefort@google.com>
Reviewed-by: Joey Gouly <joey.gouly@arm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/hyp-main.c | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index b6bfe502bcd04..8b9d76e5336bb 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -546,8 +546,27 @@ static void handle___pkvm_host_mkyoung_guest(struct kvm_cpu_context *host_ctxt)
 static void handle___kvm_adjust_pc(struct kvm_cpu_context *host_ctxt)
 {
 	DECLARE_REG(struct kvm_vcpu *, vcpu, host_ctxt, 1);
+	struct pkvm_hyp_vcpu *hyp_vcpu;
+	struct kvm_vcpu *host_vcpu;
 
-	__kvm_adjust_pc(kern_hyp_va(vcpu));
+	host_vcpu = __get_host_hyp_vcpus(vcpu, &hyp_vcpu);
+	if (host_vcpu) {
+		__kvm_adjust_pc(host_vcpu);
+		return;
+	}
+
+	/*
+	 * With no hyp vCPU loaded for it, the host vCPU may be unpinned,
+	 * and so unmapped at EL2: its first run pins it. A pin fails only
+	 * for memory the host isn't sharing, a bad pointer, so the request
+	 * is dropped.
+	 */
+	host_vcpu = kern_hyp_va(vcpu);
+	if (hyp_pin_shared_mem(host_vcpu, host_vcpu + 1))
+		return;
+
+	__kvm_adjust_pc(host_vcpu);
+	hyp_unpin_shared_mem(host_vcpu, host_vcpu + 1);
 }
 
 static void handle___kvm_flush_vm_context(struct kvm_cpu_context *host_ctxt)
-- 
2.39.5

[PATCH v4 04/18] KVM: arm64: Disable steal time for protected VMs

[Fuad Tabba]

pKVM doesn't support steal time for protected guests, and
KVM_CAP_STEAL_TIME already reads 0 for them on the VM file descriptor,
but kvm_arm_pvtime_supported() takes no VM, so the host still
accepts the KVM_ARM_VCPU_PVTIME_CTRL attribute, whose IPA would point
kvm_update_stolen_time() at the guest's private memory on every vCPU
load. Pass the VM in and return false for a protected one, so the
attribute returns -ENXIO as it does where steal time isn't implemented.

Reviewed-by: Vincent Donnefort <vdonnefort@google.com>
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/include/asm/kvm_host.h |  2 +-
 arch/arm64/kvm/arm.c              |  2 +-
 arch/arm64/kvm/pvtime.c           | 10 +++++-----
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index 27fe0cd5b2d7a..286489a69dff5 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -1346,7 +1346,7 @@ long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
 
-bool kvm_arm_pvtime_supported(void);
+bool kvm_arm_pvtime_supported(struct kvm *kvm);
 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
 			    struct kvm_device_attr *attr);
 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index 8b080804bc90b..db36815630790 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -447,7 +447,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 		r = system_supports_mte();
 		break;
 	case KVM_CAP_STEAL_TIME:
-		r = kvm_arm_pvtime_supported();
+		r = kvm_arm_pvtime_supported(kvm);
 		break;
 	case KVM_CAP_ARM_EL1_32BIT:
 		r = cpus_have_final_cap(ARM64_HAS_32BIT_EL1);
diff --git a/arch/arm64/kvm/pvtime.c b/arch/arm64/kvm/pvtime.c
index 4ceabaa4c30bd..053fe831fa541 100644
--- a/arch/arm64/kvm/pvtime.c
+++ b/arch/arm64/kvm/pvtime.c
@@ -67,9 +67,9 @@ gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu)
 	return base;
 }
 
-bool kvm_arm_pvtime_supported(void)
+bool kvm_arm_pvtime_supported(struct kvm *kvm)
 {
-	return !!sched_info_on();
+	return !!sched_info_on() && (!kvm || !kvm_vm_is_protected(kvm));
 }
 
 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
@@ -81,7 +81,7 @@ int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
 	int ret = 0;
 	int idx;
 
-	if (!kvm_arm_pvtime_supported() ||
+	if (!kvm_arm_pvtime_supported(kvm) ||
 	    attr->attr != KVM_ARM_VCPU_PVTIME_IPA)
 		return -ENXIO;
 
@@ -110,7 +110,7 @@ int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
 	u64 __user *user = (u64 __user *)attr->addr;
 	u64 ipa;
 
-	if (!kvm_arm_pvtime_supported() ||
+	if (!kvm_arm_pvtime_supported(vcpu->kvm) ||
 	    attr->attr != KVM_ARM_VCPU_PVTIME_IPA)
 		return -ENXIO;
 
@@ -126,7 +126,7 @@ int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
 {
 	switch (attr->attr) {
 	case KVM_ARM_VCPU_PVTIME_IPA:
-		if (kvm_arm_pvtime_supported())
+		if (kvm_arm_pvtime_supported(vcpu->kvm))
 			return 0;
 	}
 	return -ENXIO;
-- 
2.39.5

[PATCH v4 05/18] KVM: arm64: Introduce per-EC entry handlers for pKVM

[Fuad Tabba]

From: Marc Zyngier <maz@kernel.org>

Add an ESR_EL2.EC-indexed handler table, entry_hyp_vm_handlers[],
read from flush_hyp_vcpu() on re-entry when the previous exit was
a trap: sync_hyp_vcpu() records the exit reason in the hyp vCPU as
exit_code, and the trap's exception class comes from its ESR_EL2. Its
single generic handler stands in until the per-EC marshalling patch
replaces the table with protected-only ones.

Entry carries the host's requested PC updates to the hyp vCPU: add
vcpu_copy_flag() to copy a masked set of flags between the two vCPU
structures, with PC_UPDATE_REQ covering INCREMENT_PC and the
pending-exception flags. The wholesale iflags copy moves into the
non-protected branch; a protected vCPU takes only PC_UPDATE_REQ.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Joey Gouly <joey.gouly@arm.com>
Co-developed-by: Fuad Tabba <fuad.tabba@linux.dev>
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/include/asm/kvm_host.h      | 19 +++++++++++++
 arch/arm64/kvm/hyp/include/nvhe/pkvm.h |  3 ++
 arch/arm64/kvm/hyp/nvhe/hyp-main.c     | 39 +++++++++++++++++++++++---
 3 files changed, 57 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index 286489a69dff5..37d0721d39a45 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -1033,11 +1033,28 @@ struct kvm_vcpu_arch {
 		set;						\
 	})
 
+#define __vcpu_copy_flag(vt, vs, flagset, f, m)			\
+	do {							\
+		typeof(vs->arch.flagset) tmp, val;		\
+								\
+		__build_check_flag(vs, flagset, f, m);		\
+								\
+		val = READ_ONCE(vs->arch.flagset);		\
+		val &= (m);					\
+		__vcpu_flags_preempt_disable();			\
+		tmp = READ_ONCE(vt->arch.flagset);		\
+		tmp &= ~(m);					\
+		tmp |= val;					\
+		WRITE_ONCE(vt->arch.flagset, tmp);		\
+		__vcpu_flags_preempt_enable();			\
+	} while (0)
+
 #define vcpu_get_flag(v, ...)	__vcpu_get_flag((v), __VA_ARGS__)
 #define vcpu_set_flag(v, ...)	__vcpu_set_flag((v), __VA_ARGS__)
 #define vcpu_clear_flag(v, ...)	__vcpu_clear_flag((v), __VA_ARGS__)
 #define vcpu_test_and_clear_flag(v, ...)			\
 	__vcpu_test_and_clear_flag((v), __VA_ARGS__)
+#define vcpu_copy_flag(vt, vs, ...) __vcpu_copy_flag((vt), (vs), __VA_ARGS__)
 
 /* KVM_ARM_VCPU_INIT completed */
 #define VCPU_INITIALIZED	__vcpu_single_flag(cflags, BIT(0))
@@ -1055,6 +1072,8 @@ struct kvm_vcpu_arch {
 #define INCREMENT_PC		__vcpu_single_flag(iflags, BIT(1))
 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
 #define EXCEPT_MASK		__vcpu_single_flag(iflags, GENMASK(3, 1))
+/* Cover both PENDING_EXCEPTION and EXCEPT_MASK for global operations */
+#define PC_UPDATE_REQ		__vcpu_single_flag(iflags, GENMASK(3, 0))
 /* Host-set: the hyp flushes the non-protected vCPU state in on entry */
 #define PKVM_HOST_STATE_DIRTY	__vcpu_single_flag(iflags, BIT(4))
 
diff --git a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
index 5ddb407149667..ad42abaf31c5c 100644
--- a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
+++ b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
@@ -26,6 +26,9 @@ struct pkvm_hyp_vcpu {
 	 * per-cpu pointer tracking us. Otherwise, NULL if not loaded.
 	 */
 	struct pkvm_hyp_vcpu **loaded_hyp_vcpu;
+
+	/* The previous exit's ARM_EXCEPTION_* code. */
+	u32 exit_code;
 };
 
 /*
diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 8b9d76e5336bb..e3f4bbb623492 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -31,6 +31,17 @@ unsigned int hyp_gicv3_nr_lr;
 
 void __kvm_hyp_host_forward_smc(struct kvm_cpu_context *host_ctxt);
 
+typedef void (*hyp_entry_exit_handler_fn)(struct pkvm_hyp_vcpu *);
+
+static void handle_vm_entry_generic(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	vcpu_copy_flag(&hyp_vcpu->vcpu, hyp_vcpu->host_vcpu, PC_UPDATE_REQ);
+}
+
+static const hyp_entry_exit_handler_fn entry_hyp_vm_handlers[] = {
+	[0 ... ESR_ELx_EC_MAX]		= handle_vm_entry_generic,
+};
+
 static void __hyp_sve_save_guest(struct kvm_vcpu *vcpu)
 {
 	__vcpu_assign_sys_reg(vcpu, ZCR_EL1, read_sysreg_el1(SYS_ZCR));
@@ -216,6 +227,8 @@ static void sync_debug_state(struct pkvm_hyp_vcpu *hyp_vcpu)
 static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 {
 	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	hyp_entry_exit_handler_fn ec_handler;
+	u8 esr_ec;
 
 	fpsimd_sve_flush();
 	flush_debug_state(hyp_vcpu);
@@ -228,6 +241,7 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 	if (!pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
 		if (vcpu_get_flag(host_vcpu, PKVM_HOST_STATE_DIRTY))
 			flush_hyp_vcpu_state(hyp_vcpu);
+		hyp_vcpu->vcpu.arch.iflags = host_vcpu->arch.iflags;
 	} else {
 		hyp_vcpu->vcpu.arch.ctxt = host_vcpu->arch.ctxt;
 	}
@@ -245,16 +259,31 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 	hyp_vcpu->vcpu.arch.hcr_el2 |= READ_ONCE(host_vcpu->arch.hcr_el2) &
 						 (HCR_TWI | HCR_TWE | HCR_VSE);
 
-	hyp_vcpu->vcpu.arch.iflags	= host_vcpu->arch.iflags;
-
 	hyp_vcpu->vcpu.arch.vsesr_el2	= host_vcpu->arch.vsesr_el2;
 
 	flush_hyp_vgic_state(hyp_vcpu);
 
 	hyp_vcpu->vcpu.arch.pid = host_vcpu->arch.pid;
+
+	switch (ARM_EXCEPTION_CODE(hyp_vcpu->exit_code)) {
+	case ARM_EXCEPTION_IRQ:
+	case ARM_EXCEPTION_EL1_SERROR:
+	case ARM_EXCEPTION_IL:
+		break;
+	case ARM_EXCEPTION_TRAP:
+		esr_ec = ESR_ELx_EC(kvm_vcpu_get_esr(&hyp_vcpu->vcpu));
+		ec_handler = entry_hyp_vm_handlers[esr_ec];
+		if (ec_handler)
+			ec_handler(hyp_vcpu);
+		break;
+	default:
+		BUG();
+	}
+
+	hyp_vcpu->exit_code = 0;
 }
 
-static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
+static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu, u32 exit_reason)
 {
 	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
 
@@ -281,6 +310,8 @@ static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 	host_vcpu->arch.hcr_el2 |= hyp_vcpu->vcpu.arch.hcr_el2 & HCR_VSE;
 
 	sync_hyp_vgic_state(hyp_vcpu);
+
+	hyp_vcpu->exit_code = exit_reason;
 }
 
 static void handle___pkvm_vcpu_load(struct kvm_cpu_context *host_ctxt)
@@ -395,7 +426,7 @@ static void handle___kvm_vcpu_run(struct kvm_cpu_context *host_ctxt)
 
 		ret = __kvm_vcpu_run(&hyp_vcpu->vcpu);
 
-		sync_hyp_vcpu(hyp_vcpu);
+		sync_hyp_vcpu(hyp_vcpu, ret);
 	} else {
 		/* The host is fully trusted, run its vCPU directly. */
 		fpsimd_lazy_switch_to_guest(host_vcpu);
-- 
2.39.5

[PATCH v4 06/18] KVM: arm64: Skip fixed-feature state flush for protected vCPUs

[Fuad Tabba]

flush_hyp_vcpu() copies the host's mdcr_el2, TWI/TWE and debug state
into the hyp vCPU on every entry, so for a protected vCPU the host's
trap configuration overrides the one pkvm_vcpu_init_traps() computed
at EL2. Move the mdcr_el2 and TWI/TWE copies into the non-protected
branch, and return early from flush_debug_state() and
sync_debug_state() for a protected vCPU: its debug registers are
hypervisor-owned, and it takes the host's TWI/TWE at vCPU load.
MDCR_EL2.TDA traps only the guest's own accesses; it doesn't stop the
world switch from loading host-supplied breakpoints and MDSCR_EL1.

EL2's mdcr_el2 starts from 0, and HPMN == 0 is reserved without
FEAT_HPMN0. Set HPMN at vCPU load to PMCR_EL0.N on a PE whose
ID_AA64DFR0_EL1 implements PMUv3, the value the host's copy carried,
read as kvm_init_host_debug_data() reads it: the PMUv3 cpucap is
system-wide and off without CONFIG_HW_PERF_EVENTS. A protected guest's
PMU accesses trap regardless.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/hyp-main.c | 34 ++++++++++++++++++++++--------
 arch/arm64/kvm/hyp/nvhe/pkvm.c     |  2 +-
 2 files changed, 26 insertions(+), 10 deletions(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index e3f4bbb623492..b7b9f5afb673b 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -200,6 +200,9 @@ static void flush_debug_state(struct pkvm_hyp_vcpu *hyp_vcpu)
 {
 	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
 
+	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu))
+		return;
+
 	hyp_vcpu->vcpu.arch.debug_owner = host_vcpu->arch.debug_owner;
 
 	if (kvm_guest_owns_debug_regs(&hyp_vcpu->vcpu)) {
@@ -218,6 +221,9 @@ static void sync_debug_state(struct pkvm_hyp_vcpu *hyp_vcpu)
 {
 	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
 
+	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu))
+		return;
+
 	if (kvm_guest_owns_debug_regs(&hyp_vcpu->vcpu))
 		host_vcpu->arch.vcpu_debug_state = hyp_vcpu->vcpu.arch.vcpu_debug_state;
 	else if (kvm_host_owns_debug_regs(&hyp_vcpu->vcpu))
@@ -241,6 +247,12 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 	if (!pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
 		if (vcpu_get_flag(host_vcpu, PKVM_HOST_STATE_DIRTY))
 			flush_hyp_vcpu_state(hyp_vcpu);
+
+		hyp_vcpu->vcpu.arch.hcr_el2 &= ~(HCR_TWI | HCR_TWE);
+		hyp_vcpu->vcpu.arch.hcr_el2 |= READ_ONCE(host_vcpu->arch.hcr_el2) &
+							 (HCR_TWI | HCR_TWE);
+
+		hyp_vcpu->vcpu.arch.mdcr_el2 = host_vcpu->arch.mdcr_el2;
 		hyp_vcpu->vcpu.arch.iflags = host_vcpu->arch.iflags;
 	} else {
 		hyp_vcpu->vcpu.arch.ctxt = host_vcpu->arch.ctxt;
@@ -249,17 +261,13 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 	/* __hyp_running_vcpu must be NULL in a guest context. */
 	hyp_vcpu->vcpu.arch.ctxt.__hyp_running_vcpu = NULL;
 
-	hyp_vcpu->vcpu.arch.mdcr_el2	= host_vcpu->arch.mdcr_el2;
 	/*
-	 * HCR_EL2.VSE is host-owned (a pending virtual SError to inject), not a
-	 * trap-control bit, so it must flow to the hyp vCPU alongside TWI/TWE
-	 * for the vSError to be delivered. sync_hyp_vcpu() reflects it back.
+	 * A host-injected vSError is masked by the guest's own PSTATE.A, so it
+	 * applies to protected guests too.
 	 */
-	hyp_vcpu->vcpu.arch.hcr_el2 &= ~(HCR_TWI | HCR_TWE | HCR_VSE);
-	hyp_vcpu->vcpu.arch.hcr_el2 |= READ_ONCE(host_vcpu->arch.hcr_el2) &
-						 (HCR_TWI | HCR_TWE | HCR_VSE);
-
-	hyp_vcpu->vcpu.arch.vsesr_el2	= host_vcpu->arch.vsesr_el2;
+	hyp_vcpu->vcpu.arch.hcr_el2 &= ~HCR_VSE;
+	hyp_vcpu->vcpu.arch.hcr_el2 |= READ_ONCE(host_vcpu->arch.hcr_el2) & HCR_VSE;
+	hyp_vcpu->vcpu.arch.vsesr_el2 = host_vcpu->arch.vsesr_el2;
 
 	flush_hyp_vgic_state(hyp_vcpu);
 
@@ -326,9 +334,17 @@ static void handle___pkvm_vcpu_load(struct kvm_cpu_context *host_ctxt)
 		return;
 
 	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
+		u64 dfr0 = read_sysreg(id_aa64dfr0_el1);
+
 		/* Propagate WFx trapping flags */
 		hyp_vcpu->vcpu.arch.hcr_el2 &= ~(HCR_TWE | HCR_TWI);
 		hyp_vcpu->vcpu.arch.hcr_el2 |= hcr_el2 & (HCR_TWE | HCR_TWI);
+
+		/* HPMN == 0 is reserved without FEAT_HPMN0. */
+		if (pmuv3_implemented(SYS_FIELD_GET(ID_AA64DFR0_EL1, PMUVer, dfr0)))
+			u64p_replace_bits(&hyp_vcpu->vcpu.arch.mdcr_el2,
+					  FIELD_GET(ARMV8_PMU_PMCR_N, read_sysreg(pmcr_el0)),
+					  MDCR_EL2_HPMN);
 	} else {
 		memcpy(&hyp_vcpu->vcpu.arch.fgt, hyp_vcpu->host_vcpu->arch.fgt,
 		       sizeof(hyp_vcpu->vcpu.arch.fgt));
diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index 0e90de74778cf..8f3d9c293d186 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -102,7 +102,7 @@ static void pvm_init_traps_mdcr(struct kvm_vcpu *vcpu)
 
 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMUVer, IMP)) {
 		val |= MDCR_EL2_TPM | MDCR_EL2_TPMCR;
-		val &= ~(MDCR_EL2_HPME | MDCR_EL2_MTPME | MDCR_EL2_HPMN_MASK);
+		val &= ~(MDCR_EL2_HPME | MDCR_EL2_MTPME);
 	}
 
 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, DebugVer, IMP))
-- 
2.39.5

[PATCH v4 07/18] KVM: arm64: Add {flush,sync}_hyp_timer_state() primitives

[Fuad Tabba]

From: Marc Zyngier <maz@kernel.org>

A protected guest's virtual timer is programmed from EL2's own copy of
the state and read back there, so the host can't change the guest's
deadline. cntvoff_el2 is zeroed, so the guest's virtual time is the
physical time.

The flush also preserves the guest's CNTV and CNTP CVAL/CTL across the
host's full-context copy into the hyp vCPU, until the per-EC
marshalling patch removes that copy.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Co-developed-by: Fuad Tabba <fuad.tabba@linux.dev>
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/hyp-main.c | 41 ++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)

diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index b7b9f5afb673b..6f068d0f78d25 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -154,6 +154,35 @@ static void sync_hyp_vgic_state(struct pkvm_hyp_vcpu *hyp_vcpu)
 		host_cpu_if->vgic_lr[i] = hyp_cpu_if->vgic_lr[i];
 }
 
+static void flush_hyp_timer_state(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+
+	if (!pkvm_hyp_vcpu_is_protected(hyp_vcpu))
+		return;
+
+	/* A hyp vcpu has no offset, and sees vtime == ptime. */
+	write_sysreg(0, cntvoff_el2);
+	write_sysreg_el0(__vcpu_sys_reg(vcpu, CNTV_CVAL_EL0), SYS_CNTV_CVAL);
+	isb();
+	write_sysreg_el0(__vcpu_sys_reg(vcpu, CNTV_CTL_EL0), SYS_CNTV_CTL);
+}
+
+static void sync_hyp_timer_state(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+
+	if (!pkvm_hyp_vcpu_is_protected(hyp_vcpu))
+		return;
+
+	/*
+	 * Preserve the vtimer state so that it is always correct,
+	 * even if the host tries to make a mess.
+	 */
+	__vcpu_assign_sys_reg(vcpu, CNTV_CVAL_EL0, read_sysreg_el0(SYS_CNTV_CVAL));
+	__vcpu_assign_sys_reg(vcpu, CNTV_CTL_EL0, read_sysreg_el0(SYS_CNTV_CTL));
+}
+
 static void __copy_vcpu_state(const struct kvm_vcpu *from_vcpu,
 			      struct kvm_vcpu *to_vcpu)
 {
@@ -255,7 +284,17 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 		hyp_vcpu->vcpu.arch.mdcr_el2 = host_vcpu->arch.mdcr_el2;
 		hyp_vcpu->vcpu.arch.iflags = host_vcpu->arch.iflags;
 	} else {
+		u64 v_cval = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CVAL_EL0];
+		u64 v_ctl = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CTL_EL0];
+		u64 p_cval = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CVAL_EL0];
+		u64 p_ctl = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CTL_EL0];
+
 		hyp_vcpu->vcpu.arch.ctxt = host_vcpu->arch.ctxt;
+
+		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CVAL_EL0] = v_cval;
+		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CTL_EL0] = v_ctl;
+		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CVAL_EL0] = p_cval;
+		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CTL_EL0] = p_ctl;
 	}
 
 	/* __hyp_running_vcpu must be NULL in a guest context. */
@@ -270,6 +309,7 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 	hyp_vcpu->vcpu.arch.vsesr_el2 = host_vcpu->arch.vsesr_el2;
 
 	flush_hyp_vgic_state(hyp_vcpu);
+	flush_hyp_timer_state(hyp_vcpu);
 
 	hyp_vcpu->vcpu.arch.pid = host_vcpu->arch.pid;
 
@@ -318,6 +358,7 @@ static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu, u32 exit_reason)
 	host_vcpu->arch.hcr_el2 |= hyp_vcpu->vcpu.arch.hcr_el2 & HCR_VSE;
 
 	sync_hyp_vgic_state(hyp_vcpu);
+	sync_hyp_timer_state(hyp_vcpu);
 
 	hyp_vcpu->exit_code = exit_reason;
 }
-- 
2.39.5

[PATCH v4 08/18] KVM: arm64: Add system register reset framework for protected VMs

[Fuad Tabba]

Add kvm_reset_pvm_sys_regs() and the pvm_sys_reg_reset_vals[] table
that drives it, and call it from init_pkvm_hyp_vcpu() for protected
vCPUs. The table holds the registers KVM resets that EL2 loads for a
protected vCPU: the ones __sysreg_restore_state_nvhe() restores, and
CNTV_CTL_EL0, which flush_hyp_timer_state() loads from a copy the
host's kvm_timer_vcpu_reset() doesn't write. The ptrauth keys, which
entry.S loads, are not in the table: sys_regs.c resets them to
UNKNOWN, and a vCPU that CPU_ON brings back keeps the keys the guest
last set, one UNKNOWN value in place of another.

A register that resets to 0 still takes an entry: a later patch calls
kvm_reset_pvm_sys_regs() again for PSCI CPU_ON at EL2, on a context
that has run.

A poison value stands in where sys_regs.c resets to UNKNOWN, and for
VBAR_EL1 and CONTEXTIDR_EL1 in place of the 0 it uses.

MPIDR_EL1 is derived from vcpu_id, as for any KVM guest.

AMAIR_EL1 resets to 0 where sys_regs.c reads the hardware: the host
writes it freely before the hypercall, and it's loaded on every guest
entry.

Until the per-EC marshalling patch removes the host's full-context
copy from flush_hyp_vcpu(), that copy overwrites these values on every
entry. It preserves CNTV_CTL_EL0, whose entry writes the zero the page
already holds. So this patch has no observable effect on its own.

No functional change intended.

Reviewed-by: Joey Gouly <joey.gouly@arm.com>
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/include/nvhe/pkvm.h |  1 +
 arch/arm64/kvm/hyp/nvhe/pkvm.c         |  5 +++
 arch/arm64/kvm/hyp/nvhe/sys_regs.c     | 60 ++++++++++++++++++++++++++
 3 files changed, 66 insertions(+)

diff --git a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
index ad42abaf31c5c..fcb13d8b4cb69 100644
--- a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
+++ b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
@@ -97,6 +97,7 @@ bool kvm_handle_pvm_hvc64(struct kvm_vcpu *vcpu, u64 *exit_code);
 bool kvm_handle_pvm_sysreg(struct kvm_vcpu *vcpu, u64 *exit_code);
 bool kvm_handle_pvm_restricted(struct kvm_vcpu *vcpu, u64 *exit_code);
 void kvm_init_pvm_id_regs(struct kvm_vcpu *vcpu);
+void kvm_reset_pvm_sys_regs(struct kvm_vcpu *vcpu);
 int kvm_check_pvm_sysreg_table(void);
 
 #endif /* __ARM64_KVM_NVHE_PKVM_H__ */
diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index 8f3d9c293d186..6ff08d195ff56 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -588,6 +588,11 @@ static int init_pkvm_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu,
 		goto done;
 
 	ret = pkvm_vcpu_init_sve(hyp_vcpu, host_vcpu);
+	if (ret)
+		goto done;
+
+	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu))
+		kvm_reset_pvm_sys_regs(&hyp_vcpu->vcpu);
 done:
 	if (ret)
 		unpin_host_vcpu(host_vcpu);
diff --git a/arch/arm64/kvm/hyp/nvhe/sys_regs.c b/arch/arm64/kvm/hyp/nvhe/sys_regs.c
index 8758c68017765..87557d1b3bf07 100644
--- a/arch/arm64/kvm/hyp/nvhe/sys_regs.c
+++ b/arch/arm64/kvm/hyp/nvhe/sys_regs.c
@@ -525,6 +525,66 @@ static const struct sys_reg_desc pvm_sys_reg_descs[] = {
 	/* Performance Monitoring Registers are restricted. */
 };
 
+struct sys_reg_desc_reset {
+	int reg;
+	void (*reset)(struct kvm_vcpu *vcpu, const struct sys_reg_desc_reset *rd);
+	u64 value;
+};
+
+static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc_reset *r)
+{
+	__vcpu_assign_sys_reg(vcpu, r->reg, kvm_calculate_mpidr(vcpu));
+}
+
+static void reset_value(struct kvm_vcpu *vcpu, const struct sys_reg_desc_reset *r)
+{
+	__vcpu_assign_sys_reg(vcpu, r->reg, r->value);
+}
+
+#define RESET_VAL(REG, VAL) {  REG, reset_value, VAL }
+
+#define RESET_ZERO(REG) RESET_VAL(REG, 0)
+
+#define RESET_UNKNOWN(REG) RESET_VAL(REG, 0x1de7ec7edbadc0deULL)
+
+#define RESET_FUNC(REG, FN) {  REG, FN, 0 }
+
+static const struct sys_reg_desc_reset pvm_sys_reg_reset_vals[] = {
+	RESET_FUNC(MPIDR_EL1, reset_mpidr),
+	RESET_UNKNOWN(TPIDR_EL0),
+	RESET_UNKNOWN(TPIDRRO_EL0),
+	RESET_UNKNOWN(TPIDR_EL1),
+	RESET_ZERO(CNTKCTL_EL1),
+	RESET_UNKNOWN(PAR_EL1),
+	RESET_ZERO(DISR_EL1),
+	RESET_ZERO(CPACR_EL1),
+	RESET_VAL(CONTEXTIDR_EL1, 0x00000000dbadc0deULL),
+	RESET_VAL(SCTLR_EL1, 0x00C50078ULL),
+	RESET_ZERO(TCR_EL1),
+	RESET_UNKNOWN(AFSR0_EL1),
+	RESET_UNKNOWN(AFSR1_EL1),
+	RESET_UNKNOWN(ESR_EL1),
+	RESET_UNKNOWN(MAIR_EL1),
+	RESET_ZERO(AMAIR_EL1),
+	RESET_ZERO(MDSCR_EL1),
+	RESET_ZERO(CNTV_CTL_EL0),
+	RESET_UNKNOWN(TTBR0_EL1),
+	RESET_UNKNOWN(TTBR1_EL1),
+	RESET_UNKNOWN(FAR_EL1),
+	RESET_VAL(VBAR_EL1, 0x1de7ec7edbadc000ULL),
+};
+
+void kvm_reset_pvm_sys_regs(struct kvm_vcpu *vcpu)
+{
+	unsigned long i;
+
+	for (i = 0; i < ARRAY_SIZE(pvm_sys_reg_reset_vals); i++) {
+		const struct sys_reg_desc_reset *r = &pvm_sys_reg_reset_vals[i];
+
+		r->reset(vcpu, r);
+	}
+}
+
 /*
  * Initializes feature registers for protected vms.
  */
-- 
2.39.5

[PATCH v4 09/18] KVM: arm64: Implement HVC handling for protected guests at EL2

[Fuad Tabba]

Extend kvm_handle_pvm_hvc64() to handle SMCCC_VERSION,
SMCCC_ARCH_FEATURES and the vendor hypervisor call UID at EL2, so
these queries don't reach the host. ARCH_FEATURES is mandatory from
SMCCC 1.1, the version EL2 implements: it returns SUCCESS for itself and
for SMCCC_VERSION, NOT_SUPPORTED for anything else, and handles the
ARCH_WORKAROUND_1/2/3 queries as the host does, from a copy of the
host's mitigation state taken at init with the sanitised ID registers.
The workaround calls themselves are already handled on hyp entry.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/include/asm/kvm_hyp.h |  4 ++
 arch/arm64/kvm/arm.c             |  4 ++
 arch/arm64/kvm/hyp/nvhe/pkvm.c   | 68 ++++++++++++++++++++++++++++++++
 3 files changed, 76 insertions(+)

diff --git a/arch/arm64/include/asm/kvm_hyp.h b/arch/arm64/include/asm/kvm_hyp.h
index 4974492744cc8..7609c78c68f34 100644
--- a/arch/arm64/include/asm/kvm_hyp.h
+++ b/arch/arm64/include/asm/kvm_hyp.h
@@ -10,6 +10,7 @@
 #include <linux/compiler.h>
 #include <linux/kvm_host.h>
 #include <asm/alternative.h>
+#include <asm/spectre.h>
 #include <asm/sysreg.h>
 
 DECLARE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
@@ -150,6 +151,9 @@ extern u64 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val);
 
 extern unsigned long kvm_nvhe_sym(__icache_flags);
 extern unsigned int kvm_nvhe_sym(kvm_arm_vmid_bits);
+extern enum mitigation_state kvm_nvhe_sym(spectre_v2_state);
+extern enum mitigation_state kvm_nvhe_sym(spectre_v4_state);
+extern enum mitigation_state kvm_nvhe_sym(spectre_bhb_state);
 extern unsigned int kvm_nvhe_sym(kvm_host_sve_max_vl);
 extern unsigned long kvm_nvhe_sym(hyp_nr_cpus);
 extern unsigned int kvm_nvhe_sym(hyp_gicv3_nr_lr);
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index db36815630790..cceeeee266902 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -2622,6 +2622,10 @@ static void kvm_hyp_init_symbols(void)
 	kvm_nvhe_sym(__icache_flags) = __icache_flags;
 	kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
 
+	kvm_nvhe_sym(spectre_v2_state) = arm64_get_spectre_v2_state();
+	kvm_nvhe_sym(spectre_v4_state) = arm64_get_spectre_v4_state();
+	kvm_nvhe_sym(spectre_bhb_state) = arm64_get_spectre_bhb_state();
+
 	/* Propagate the FGT state to the nVHE side */
 	kvm_nvhe_sym(hfgrtr_masks)  = hfgrtr_masks;
 	kvm_nvhe_sym(hfgwtr_masks)  = hfgwtr_masks;
diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index 6ff08d195ff56..10ff2ba6c6c1b 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -10,12 +10,18 @@
 #include <linux/mm.h>
 
 #include <asm/kvm_emulate.h>
+#include <asm/spectre.h>
 
 #include <nvhe/mem_protect.h>
 #include <nvhe/memory.h>
 #include <nvhe/pkvm.h>
 #include <nvhe/trap_handler.h>
 
+/* The host's Spectre mitigation state, for SMCCC_ARCH_FEATURES. */
+enum mitigation_state spectre_v2_state;
+enum mitigation_state spectre_v4_state;
+enum mitigation_state spectre_bhb_state;
+
 /* Used by icache_is_aliasing(). */
 unsigned long __icache_flags;
 
@@ -1181,8 +1187,70 @@ bool kvm_handle_pvm_hvc64(struct kvm_vcpu *vcpu, u64 *exit_code)
 {
 	u64 val[4] = { SMCCC_RET_INVALID_PARAMETER };
 	bool handled = true;
+	u32 feature;
+	uuid_t uuid;
 
 	switch (smccc_get_function(vcpu)) {
+	case ARM_SMCCC_VERSION_FUNC_ID:
+		/* Nothing to be handled by the host. Go back to the guest. */
+		val[0] = ARM_SMCCC_VERSION_1_1;
+		break;
+	case ARM_SMCCC_ARCH_FEATURES_FUNC_ID:
+		/* The workaround calls themselves are handled on hyp entry. */
+		val[0] = SMCCC_RET_NOT_SUPPORTED;
+		feature = smccc_get_arg1(vcpu);
+		switch (feature) {
+		case ARM_SMCCC_VERSION_FUNC_ID:
+		case ARM_SMCCC_ARCH_FEATURES_FUNC_ID:
+			val[0] = SMCCC_RET_SUCCESS;
+			break;
+		case ARM_SMCCC_ARCH_WORKAROUND_1:
+			switch (spectre_v2_state) {
+			case SPECTRE_VULNERABLE:
+				break;
+			case SPECTRE_MITIGATED:
+				val[0] = SMCCC_RET_SUCCESS;
+				break;
+			case SPECTRE_UNAFFECTED:
+				val[0] = SMCCC_ARCH_WORKAROUND_RET_UNAFFECTED;
+				break;
+			}
+			break;
+		case ARM_SMCCC_ARCH_WORKAROUND_2:
+			switch (spectre_v4_state) {
+			case SPECTRE_VULNERABLE:
+				break;
+			case SPECTRE_MITIGATED:
+				/* With SSBS advertised the guest's default is safe. */
+				if (kvm_has_feat(vcpu->kvm, ID_AA64PFR1_EL1, SSBS, IMP))
+					break;
+				fallthrough;
+			case SPECTRE_UNAFFECTED:
+				val[0] = SMCCC_RET_NOT_REQUIRED;
+				break;
+			}
+			break;
+		case ARM_SMCCC_ARCH_WORKAROUND_3:
+			switch (spectre_bhb_state) {
+			case SPECTRE_VULNERABLE:
+				break;
+			case SPECTRE_MITIGATED:
+				val[0] = SMCCC_RET_SUCCESS;
+				break;
+			case SPECTRE_UNAFFECTED:
+				val[0] = SMCCC_ARCH_WORKAROUND_RET_UNAFFECTED;
+				break;
+			}
+			break;
+		}
+		break;
+	case ARM_SMCCC_VENDOR_HYP_CALL_UID_FUNC_ID:
+		uuid = ARM_SMCCC_VENDOR_HYP_UID_KVM;
+		val[0] = smccc_uuid_to_reg(&uuid, 0);
+		val[1] = smccc_uuid_to_reg(&uuid, 1);
+		val[2] = smccc_uuid_to_reg(&uuid, 2);
+		val[3] = smccc_uuid_to_reg(&uuid, 3);
+		break;
 	case ARM_SMCCC_VENDOR_HYP_KVM_FEATURES_FUNC_ID:
 		val[0] = BIT(ARM_SMCCC_KVM_FUNC_FEATURES);
 		val[0] |= BIT(ARM_SMCCC_KVM_FUNC_HYP_MEMINFO);
-- 
2.39.5

[PATCH v4 10/18] KVM: arm64: Handle PSCI calls for protected VMs at EL2

[Fuad Tabba]

EL2 implements PSCI 1.1 for protected VMs: CPU_ON, CPU_OFF,
PSCI_VERSION and PSCI_FEATURES are resolved at EL2 (CPU_ON and CPU_OFF
still exit to the host, which only schedules the target or stops
scheduling it), AFFINITY_INFO, CPU_SUSPEND and the platform power
operations are forwarded to the host, and anything else returns
NOT_SUPPORTED, including the TRNG calls and the functions above 1.1,
SYSTEM_OFF2 among them, that the host handled for a protected guest
until now. TRNG for protected guests is a follow-up. AFFINITY_INFO
stays with the host, which returns OFF only once it has stopped
scheduling the target: the host is what a guest polls to see a CPU_OFF
complete before it issues the next CPU_ON, as Linux does on hotplug.

Three consequences follow:

  - A protected VM has one primary vCPU, the first whose hyp vCPU is
    created with mp_state RUNNABLE. A second one, or an mp_state other
    than RUNNABLE or STOPPED, fails that vCPU's first KVM_RUN with
    -EINVAL.

  - CPU_ON finds its target among the hyp vCPUs, which exist from the
    target's first KVM_RUN; before that the guest gets
    INVALID_PARAMETERS.

  - A vCPU EL2 holds powered off doesn't run: handle___kvm_vcpu_run()
    returns ARM_EXCEPTION_IL, reported as KVM_EXIT_FAIL_ENTRY. Its
    existing bail-outs return the same code, for every hyp vCPU, instead
    of an -EINVAL whose bit 31 the host read as a pending SError.

Non-protected VMs keep power_state ON and accept any mp_state.

Each protected vCPU is OFF, ON_PENDING or ON. CPU_ON moves the target
to ON_PENDING, and the target's next run resets it and moves it to ON.
The racing transitions are cmpxchg, and the reset state is published
with a release/acquire pair, documented at each site. CPU_OFF clears
the reset flag and stores OFF with a release, and CPU_ON's writes are
control dependent on its cmpxchg reading OFF, so the target's accesses
to reset_state are ordered before the winner's writes. Rolling a CPU_ON
the host failed back to OFF requires the host's return value, which
the per-EC marshalling patch delivers along with the rollback. Until
then such a target stays ON_PENDING, and the reset has no observable
effect: flush_hyp_vcpu() copies the host's context in on every entry
until that patch removes the copy, so the target enters on the host's
values rather than the ones EL2 reset.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/include/nvhe/pkvm.h |  14 ++
 arch/arm64/kvm/hyp/nvhe/hyp-main.c     |  25 ++-
 arch/arm64/kvm/hyp/nvhe/pkvm.c         | 293 ++++++++++++++++++++++++-
 3 files changed, 321 insertions(+), 11 deletions(-)

diff --git a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
index fcb13d8b4cb69..0623ec9d3a931 100644
--- a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
+++ b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
@@ -29,6 +29,12 @@ struct pkvm_hyp_vcpu {
 
 	/* The previous exit's ARM_EXCEPTION_* code. */
 	u32 exit_code;
+
+	/*
+	 * PSCI_0_2_AFFINITY_LEVEL_{OFF, ON_PENDING, ON}. A non-protected
+	 * vCPU is always ON.
+	 */
+	int power_state;
 };
 
 /*
@@ -46,6 +52,12 @@ struct pkvm_hyp_vm {
 	struct hyp_pool pool;
 	hyp_spinlock_t lock;
 
+	/*
+	 * The vCPU initialised RUNNABLE: claimed under vm_table_lock,
+	 * released only if its own init fails.
+	 */
+	struct pkvm_hyp_vcpu *primary_vcpu;
+
 	/* Array of the hyp vCPU structures for this VM. */
 	struct pkvm_hyp_vcpu *vcpus[];
 };
@@ -100,4 +112,6 @@ void kvm_init_pvm_id_regs(struct kvm_vcpu *vcpu);
 void kvm_reset_pvm_sys_regs(struct kvm_vcpu *vcpu);
 int kvm_check_pvm_sysreg_table(void);
 
+int pkvm_reset_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu);
+struct pkvm_hyp_vcpu *pkvm_mpidr_to_hyp_vcpu(struct pkvm_hyp_vm *vm, unsigned long mpidr);
 #endif /* __ARM64_KVM_NVHE_PKVM_H__ */
diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 6f068d0f78d25..00038a162d09b 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -8,6 +8,7 @@
 #include <hyp/switch.h>
 
 #include <linux/irqchip/arm-gic-v3.h>
+#include <uapi/linux/psci.h>
 
 #include <asm/pgtable-types.h>
 #include <asm/kvm_asm.h>
@@ -458,14 +459,12 @@ static void handle___kvm_vcpu_run(struct kvm_cpu_context *host_ctxt)
 {
 	struct pkvm_hyp_vcpu *hyp_vcpu;
 	struct kvm_vcpu *host_vcpu;
-	int ret;
+	int ret = ARM_EXCEPTION_IL;
 
 	host_vcpu = get_host_hyp_vcpus(host_ctxt, 1, &hyp_vcpu);
 
-	if (!host_vcpu) {
-		ret = -EINVAL;
+	if (!host_vcpu)
 		goto out;
-	}
 
 	if (unlikely(hyp_vcpu)) {
 		/*
@@ -474,8 +473,22 @@ static void handle___kvm_vcpu_run(struct kvm_cpu_context *host_ctxt)
 		 * loading a vcpu. Therefore, if SME features enabled the host
 		 * is misbehaving.
 		 */
-		if (unlikely(system_supports_sme() && read_sysreg_s(SYS_SVCR))) {
-			ret = -EINVAL;
+		if (unlikely(system_supports_sme() && read_sysreg_s(SYS_SVCR)))
+			goto out;
+
+		/*
+		 * ON has a single writer, pkvm_reset_vcpu() on this CPU, so
+		 * READ_ONCE suffices. ON_PENDING takes the reset; -ECANCELED
+		 * is a rollback that raced it.
+		 */
+		switch (READ_ONCE(hyp_vcpu->power_state)) {
+		case PSCI_0_2_AFFINITY_LEVEL_ON:
+			break;
+		case PSCI_0_2_AFFINITY_LEVEL_ON_PENDING:
+			if (pkvm_reset_vcpu(hyp_vcpu))
+				goto out;
+			break;
+		default:
 			goto out;
 		}
 
diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index 10ff2ba6c6c1b..dcc1db6201f14 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -5,6 +5,7 @@
  */
 
 #include <kvm/arm_hypercalls.h>
+#include <kvm/arm_psci.h>
 
 #include <linux/kvm_host.h>
 #include <linux/mm.h>
@@ -431,6 +432,40 @@ static void pkvm_init_features_from_host(struct pkvm_hyp_vm *hyp_vm, const struc
 		   allowed_features, KVM_VCPU_MAX_FEATURES);
 }
 
+static int pkvm_vcpu_init_psci(struct pkvm_hyp_vcpu *hyp_vcpu, u32 mp_state)
+{
+	struct vcpu_reset_state *reset_state = &hyp_vcpu->vcpu.arch.reset_state;
+	struct pkvm_hyp_vm *hyp_vm = pkvm_hyp_vcpu_to_hyp_vm(hyp_vcpu);
+	struct kvm_vcpu *host_vcpu;
+
+	if (!pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
+		/* The host manages a non-protected vCPU: always ON at EL2. */
+		hyp_vcpu->power_state = PSCI_0_2_AFFINITY_LEVEL_ON;
+		return 0;
+	}
+
+	if (mp_state != KVM_MP_STATE_RUNNABLE && mp_state != KVM_MP_STATE_STOPPED)
+		return -EINVAL;
+
+	if (mp_state == KVM_MP_STATE_STOPPED) {
+		reset_state->reset = false;
+		hyp_vcpu->power_state = PSCI_0_2_AFFINITY_LEVEL_OFF;
+		return 0;
+	}
+
+	hyp_assert_lock_held(&vm_table_lock);
+	if (hyp_vm->primary_vcpu)
+		return -EINVAL;
+	hyp_vm->primary_vcpu = hyp_vcpu;
+
+	host_vcpu = hyp_vcpu->host_vcpu;
+	reset_state->pc = READ_ONCE(*vcpu_pc(host_vcpu));
+	reset_state->r0 = READ_ONCE(vcpu_gp_regs(host_vcpu)->regs[0]);
+	reset_state->reset = true;
+	hyp_vcpu->power_state = PSCI_0_2_AFFINITY_LEVEL_ON_PENDING;
+	return 0;
+}
+
 static void unpin_host_vcpu(struct kvm_vcpu *host_vcpu)
 {
 	if (host_vcpu)
@@ -445,6 +480,9 @@ static void unpin_host_sve_state(struct pkvm_hyp_vcpu *hyp_vcpu)
 		return;
 
 	sve_state = hyp_vcpu->vcpu.arch.sve_state;
+	if (!sve_state)
+		return;
+
 	hyp_unpin_shared_mem(sve_state,
 			     sve_state + vcpu_sve_state_size(&hyp_vcpu->vcpu));
 }
@@ -557,10 +595,12 @@ static int init_pkvm_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu,
 			      struct kvm_vcpu *host_vcpu)
 {
 	int ret = 0;
+	u32 mp_state;
 
 	if (hyp_pin_shared_mem(host_vcpu, host_vcpu + 1))
 		return -EBUSY;
 
+	mp_state = READ_ONCE(host_vcpu->arch.mp_state.mp_state);
 	hyp_vcpu->host_vcpu = host_vcpu;
 
 	hyp_vcpu->vcpu.kvm = &hyp_vm->kvm;
@@ -569,7 +609,6 @@ static int init_pkvm_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu,
 
 	hyp_vcpu->vcpu.arch.hw_mmu = &hyp_vm->kvm.arch.mmu;
 	hyp_vcpu->vcpu.arch.cflags = READ_ONCE(host_vcpu->arch.cflags);
-	hyp_vcpu->vcpu.arch.mp_state.mp_state = KVM_MP_STATE_STOPPED;
 
 	if (!pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
 		/*
@@ -599,9 +638,12 @@ static int init_pkvm_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu,
 
 	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu))
 		kvm_reset_pvm_sys_regs(&hyp_vcpu->vcpu);
+	ret = pkvm_vcpu_init_psci(hyp_vcpu, mp_state);
 done:
-	if (ret)
+	if (ret) {
 		unpin_host_vcpu(host_vcpu);
+		unpin_host_sve_state(hyp_vcpu);
+	}
 	return ret;
 }
 
@@ -978,13 +1020,22 @@ int __pkvm_init_vcpu(pkvm_handle_t handle, struct kvm_vcpu *host_vcpu,
 
 	ret = init_pkvm_hyp_vcpu(hyp_vcpu, hyp_vm, host_vcpu);
 	if (ret)
-		goto unlock;
+		goto unclaim;
 
 	ret = register_hyp_vcpu(hyp_vm, hyp_vcpu);
 	if (ret) {
 		unpin_host_vcpu(host_vcpu);
 		unpin_host_sve_state(hyp_vcpu);
+		goto unclaim;
 	}
+	goto unlock;
+unclaim:
+	/*
+	 * Under vm_table_lock, so no other claim can have landed: undo
+	 * this one.
+	 */
+	if (hyp_vm->primary_vcpu == hyp_vcpu)
+		hyp_vm->primary_vcpu = NULL;
 unlock:
 	hyp_spin_unlock(&vm_table_lock);
 
@@ -1176,6 +1227,238 @@ static void pkvm_memunshare_call(u64 *ret, struct kvm_vcpu *vcpu)
 		ret[0] = SMCCC_RET_SUCCESS;
 }
 
+/*
+ * Reset the vCPU to its power-on state and commit ON_PENDING -> ON, on the
+ * target's own CPU. Returns -ECANCELED, with no side effects, if a rollback
+ * raced the reset.
+ *
+ * The requestor's PSCI CPU_ON, in pvm_psci_vcpu_on():
+ *
+ *	cmpxchg_relaxed(power_state): OFF -> ON_PENDING
+ *	<ctrl>
+ *	write_reset_state();
+ *	smp_store_release(reset_state->reset, true);
+ *
+ * The target coming online here, running, then its PSCI CPU_OFF in
+ * pvm_psci_vcpu_off():
+ *
+ *	smp_load_acquire(reset_state->reset) == true;
+ *	cmpxchg_relaxed(power_state): ON_PENDING -> ON;
+ *	read_reset_state();
+ *
+ *	<run guest>
+ *
+ *	WRITE_ONCE(reset_state->reset, false);
+ *	smp_store_release(power_state, OFF);
+ */
+int pkvm_reset_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct vcpu_reset_state *reset_state = &hyp_vcpu->vcpu.arch.reset_state;
+
+	/*
+	 * Pairs with smp_store_release(&reset_state->reset, true) in
+	 * pvm_psci_vcpu_on(). The acquire must precede the cmpxchg: reversed, a
+	 * winning cmpxchg with a false acquire would leave power_state == ON
+	 * with the reset skipped.
+	 */
+	if (!smp_load_acquire(&reset_state->reset))
+		return -ECANCELED;
+
+	if (cmpxchg_relaxed(&hyp_vcpu->power_state, PSCI_0_2_AFFINITY_LEVEL_ON_PENDING,
+			    PSCI_0_2_AFFINITY_LEVEL_ON) != PSCI_0_2_AFFINITY_LEVEL_ON_PENDING)
+		return -ECANCELED;
+
+	kvm_reset_vcpu_core(&hyp_vcpu->vcpu);
+	kvm_reset_pvm_sys_regs(&hyp_vcpu->vcpu);
+
+	/* Must be done after resetting sys registers. */
+	kvm_reset_vcpu_psci(&hyp_vcpu->vcpu, reset_state);
+
+	hyp_vcpu->exit_code = 0;
+	return 0;
+}
+
+struct pkvm_hyp_vcpu *pkvm_mpidr_to_hyp_vcpu(struct pkvm_hyp_vm *hyp_vm, unsigned long mpidr)
+{
+	struct pkvm_hyp_vcpu *hyp_vcpu;
+	int i;
+
+	mpidr &= MPIDR_HWID_BITMASK;
+
+	for (i = 0; i < hyp_vm->kvm.created_vcpus; i++) {
+		/* Pairs with smp_store_release() in register_hyp_vcpu(). */
+		hyp_vcpu = smp_load_acquire(&hyp_vm->vcpus[i]);
+
+		if (!hyp_vcpu)
+			continue;
+
+		if (mpidr == kvm_vcpu_get_mpidr_aff(&hyp_vcpu->vcpu))
+			return hyp_vcpu;
+	}
+
+	return NULL;
+}
+
+/*
+ * Returns true when handled at EL2, false when the host must wake the target
+ * vCPU.
+ */
+static bool pvm_psci_vcpu_on(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct pkvm_hyp_vm *hyp_vm = pkvm_hyp_vcpu_to_hyp_vm(hyp_vcpu);
+	struct vcpu_reset_state *reset_state;
+	struct pkvm_hyp_vcpu *target;
+	unsigned long cpu_id, ret;
+	int power_state;
+
+	cpu_id = smccc_get_arg1(&hyp_vcpu->vcpu);
+	if (!kvm_psci_valid_affinity(&hyp_vcpu->vcpu, cpu_id)) {
+		ret = PSCI_RET_INVALID_PARAMS;
+		goto error;
+	}
+
+	target = pkvm_mpidr_to_hyp_vcpu(hyp_vm, cpu_id);
+	if (!target) {
+		ret = PSCI_RET_INVALID_PARAMS;
+		goto error;
+	}
+
+	/*
+	 * vCPUs race to power on the same target. The writes below depend on
+	 * the cmpxchg reading OFF, and that control dependency with the release
+	 * of OFF in pvm_psci_vcpu_off() orders them after the target's accesses
+	 * to reset_state, so relaxed suffices.
+	 */
+	power_state = cmpxchg_relaxed(&target->power_state,
+				      PSCI_0_2_AFFINITY_LEVEL_OFF,
+				      PSCI_0_2_AFFINITY_LEVEL_ON_PENDING);
+	switch (power_state) {
+	case PSCI_0_2_AFFINITY_LEVEL_ON_PENDING:
+		ret = PSCI_RET_ON_PENDING;
+		goto error;
+	case PSCI_0_2_AFFINITY_LEVEL_ON:
+		ret = PSCI_RET_ALREADY_ON;
+		goto error;
+	case PSCI_0_2_AFFINITY_LEVEL_OFF:
+		break;
+	default:
+		ret = PSCI_RET_INTERNAL_FAILURE;
+		goto error;
+	}
+
+	reset_state = &target->vcpu.arch.reset_state;
+	WRITE_ONCE(reset_state->pc, smccc_get_arg2(&hyp_vcpu->vcpu));
+	WRITE_ONCE(reset_state->r0, smccc_get_arg3(&hyp_vcpu->vcpu));
+	WRITE_ONCE(reset_state->be, kvm_vcpu_is_be(&hyp_vcpu->vcpu));
+	/*
+	 * Publish reset_state.{pc, r0, be} to the target vCPU. Pairs with
+	 * smp_load_acquire(&reset_state->reset) in pkvm_reset_vcpu().
+	 */
+	smp_store_release(&reset_state->reset, true);
+
+	/* The host requests KVM_REQ_VCPU_RESET and wakes the target. */
+	return false;
+
+error:
+	smccc_set_retval(&hyp_vcpu->vcpu, ret, 0, 0, 0);
+	return true;
+}
+
+/*
+ * Returns true when handled at EL2, false when the host must stop scheduling
+ * the vCPU.
+ */
+static bool pvm_psci_vcpu_off(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	/* No other writer runs while this vCPU is ON and executing. */
+	WARN_ON(READ_ONCE(hyp_vcpu->power_state) != PSCI_0_2_AFFINITY_LEVEL_ON);
+
+	WRITE_ONCE(hyp_vcpu->vcpu.arch.reset_state.reset, false);
+
+	/*
+	 * Orders pkvm_reset_vcpu()'s accesses to reset_state, and the clear
+	 * above, before OFF. Pairs with the control dependency in
+	 * pvm_psci_vcpu_on().
+	 */
+	smp_store_release(&hyp_vcpu->power_state, PSCI_0_2_AFFINITY_LEVEL_OFF);
+
+	/* Return to the host so that it can finish powering off the vcpu. */
+	return false;
+}
+
+static bool pvm_psci_version(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	/* Nothing to be handled by the host. Go back to the guest. */
+	smccc_set_retval(&hyp_vcpu->vcpu, KVM_ARM_PSCI_1_1, 0, 0, 0);
+	return true;
+}
+
+static bool pvm_psci_features(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	u32 feature = smccc_get_arg1(vcpu);
+	unsigned long val;
+
+	switch (feature) {
+	case PSCI_0_2_FN_PSCI_VERSION:
+	case PSCI_0_2_FN_CPU_SUSPEND:
+	case PSCI_0_2_FN64_CPU_SUSPEND:
+	case PSCI_0_2_FN_CPU_OFF:
+	case PSCI_0_2_FN_CPU_ON:
+	case PSCI_0_2_FN64_CPU_ON:
+	case PSCI_0_2_FN_AFFINITY_INFO:
+	case PSCI_0_2_FN64_AFFINITY_INFO:
+	case PSCI_0_2_FN_SYSTEM_OFF:
+	case PSCI_0_2_FN_SYSTEM_RESET:
+	case PSCI_1_0_FN_PSCI_FEATURES:
+	case PSCI_1_1_FN_SYSTEM_RESET2:
+	case PSCI_1_1_FN64_SYSTEM_RESET2:
+	case ARM_SMCCC_VERSION_FUNC_ID:
+		val = PSCI_RET_SUCCESS;
+		break;
+	default:
+		val = PSCI_RET_NOT_SUPPORTED;
+		break;
+	}
+
+	/* Nothing to be handled by the host. Go back to the guest. */
+	smccc_set_retval(vcpu, val, 0, 0, 0);
+	return true;
+}
+
+static bool pkvm_handle_psci(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	u32 psci_fn = smccc_get_function(vcpu);
+
+	switch (psci_fn) {
+	case PSCI_0_2_FN_CPU_ON:
+		kvm_psci_narrow_to_32bit(vcpu);
+		fallthrough;
+	case PSCI_0_2_FN64_CPU_ON:
+		return pvm_psci_vcpu_on(hyp_vcpu);
+	case PSCI_0_2_FN_CPU_OFF:
+		return pvm_psci_vcpu_off(hyp_vcpu);
+	case PSCI_0_2_FN_PSCI_VERSION:
+		return pvm_psci_version(hyp_vcpu);
+	case PSCI_1_0_FN_PSCI_FEATURES:
+		return pvm_psci_features(hyp_vcpu);
+	case PSCI_0_2_FN_AFFINITY_INFO:
+	case PSCI_0_2_FN64_AFFINITY_INFO:
+	case PSCI_0_2_FN_SYSTEM_RESET:
+	case PSCI_0_2_FN_CPU_SUSPEND:
+	case PSCI_0_2_FN64_CPU_SUSPEND:
+	case PSCI_0_2_FN_SYSTEM_OFF:
+	case PSCI_1_1_FN_SYSTEM_RESET2:
+	case PSCI_1_1_FN64_SYSTEM_RESET2:
+		return false; /* Handled by the host. */
+	default:
+		/* Unknown PSCI calls are handled here, not forwarded. */
+		smccc_set_retval(vcpu, PSCI_RET_NOT_SUPPORTED, 0, 0, 0);
+		return true;
+	}
+}
+
 /*
  * Handler for protected VM HVC calls.
  *
@@ -1185,6 +1468,7 @@ static void pkvm_memunshare_call(u64 *ret, struct kvm_vcpu *vcpu)
  */
 bool kvm_handle_pvm_hvc64(struct kvm_vcpu *vcpu, u64 *exit_code)
 {
+	struct pkvm_hyp_vcpu *hyp_vcpu = container_of(vcpu, struct pkvm_hyp_vcpu, vcpu);
 	u64 val[4] = { SMCCC_RET_INVALID_PARAMETER };
 	bool handled = true;
 	u32 feature;
@@ -1283,8 +1567,7 @@ bool kvm_handle_pvm_hvc64(struct kvm_vcpu *vcpu, u64 *exit_code)
 		pkvm_memunshare_call(val, vcpu);
 		break;
 	default:
-		/* Punt everything else back to the host, for now. */
-		handled = false;
+		return pkvm_handle_psci(hyp_vcpu);
 	}
 
 	if (handled)
-- 
2.39.5

[PATCH v4 11/18] KVM: arm64: Restrict KVM_ARM_VCPU_INIT and PSCI version for protected VMs

[Fuad Tabba]

EL2 masks a protected VM's vCPU features down to the set pKVM allows,
so a KVM_ARM_VCPU_INIT that requests more succeeds and produces a
guest without them. Reject such an init with -EINVAL. The host and
EL2 both take the allowed set from kvm_pkvm_vcpu_allowed_features().

EL2 returns 1.1 for a protected guest's PSCI_VERSION whatever the
host is configured for, and forwards the platform calls to the host.
Require KVM_ARM_VCPU_PSCI_0_2: without it the host dispatches the
forwarded calls as PSCI 0.1. Reject a KVM_REG_ARM_PSCI_VERSION write
below 1.1: below it the host rejects the SYSTEM_RESET2 that EL2
advertises.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/include/asm/kvm_pkvm.h | 24 ++++++++++++++++++++++++
 arch/arm64/kvm/arm.c              | 13 +++++++++++++
 arch/arm64/kvm/hyp/nvhe/pkvm.c    | 17 ++---------------
 arch/arm64/kvm/hypercalls.c       |  7 +++++++
 4 files changed, 46 insertions(+), 15 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_pkvm.h b/arch/arm64/include/asm/kvm_pkvm.h
index beea00e693a0a..fbea052fa3e16 100644
--- a/arch/arm64/include/asm/kvm_pkvm.h
+++ b/arch/arm64/include/asm/kvm_pkvm.h
@@ -53,6 +53,30 @@ static inline bool kvm_pkvm_ext_allowed(struct kvm *kvm, long ext)
 	}
 }
 
+/*
+ * The vCPU features a protected VM may use: checked by the host at
+ * KVM_ARM_VCPU_INIT, applied by EL2 when the hyp VM is created.
+ */
+static inline void kvm_pkvm_vcpu_allowed_features(struct kvm *kvm,
+						  unsigned long *allowed)
+{
+	bitmap_zero(allowed, KVM_VCPU_MAX_FEATURES);
+
+	set_bit(KVM_ARM_VCPU_PSCI_0_2, allowed);
+
+	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_PMU_V3))
+		set_bit(KVM_ARM_VCPU_PMU_V3, allowed);
+
+	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_PTRAUTH_ADDRESS))
+		set_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, allowed);
+
+	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_PTRAUTH_GENERIC))
+		set_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, allowed);
+
+	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_SVE))
+		set_bit(KVM_ARM_VCPU_SVE, allowed);
+}
+
 /*
  * Check whether the KVM VM IOCTL is allowed in pKVM.
  *
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index cceeeee266902..ca6e109b3e5d6 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -1620,6 +1620,19 @@ static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
 	if (features & ~system_supported_vcpu_features())
 		return -EINVAL;
 
+	/* Reject features EL2 would drop when it creates the hyp VM. */
+	if (vcpu_is_protected(vcpu)) {
+		DECLARE_BITMAP(allowed, KVM_VCPU_MAX_FEATURES);
+
+		kvm_pkvm_vcpu_allowed_features(vcpu->kvm, allowed);
+		if (!bitmap_subset(&features, allowed, KVM_VCPU_MAX_FEATURES))
+			return -EINVAL;
+
+		/* EL2 implements PSCI 1.1; the host must not dispatch as 0.1. */
+		if (!test_bit(KVM_ARM_VCPU_PSCI_0_2, &features))
+			return -EINVAL;
+	}
+
 	/*
 	 * For now make sure that both address/generic pointer authentication
 	 * features are requested by the userspace together.
diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
index dcc1db6201f14..696f999996b13 100644
--- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
+++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
@@ -410,23 +410,10 @@ static void pkvm_init_features_from_host(struct pkvm_hyp_vm *hyp_vm, const struc
 	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_MTE))
 		kvm->arch.flags |= host_arch_flags & BIT(KVM_ARCH_FLAG_MTE_ENABLED);
 
-	bitmap_zero(allowed_features, KVM_VCPU_MAX_FEATURES);
+	kvm_pkvm_vcpu_allowed_features(kvm, allowed_features);
 
-	set_bit(KVM_ARM_VCPU_PSCI_0_2, allowed_features);
-
-	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_PMU_V3))
-		set_bit(KVM_ARM_VCPU_PMU_V3, allowed_features);
-
-	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_PTRAUTH_ADDRESS))
-		set_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, allowed_features);
-
-	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_PTRAUTH_GENERIC))
-		set_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, allowed_features);
-
-	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_SVE)) {
-		set_bit(KVM_ARM_VCPU_SVE, allowed_features);
+	if (kvm_pkvm_ext_allowed(kvm, KVM_CAP_ARM_SVE))
 		kvm->arch.flags |= host_arch_flags & BIT(KVM_ARCH_FLAG_GUEST_HAS_SVE);
-	}
 
 	bitmap_and(kvm->arch.vcpu_features, host_kvm->arch.vcpu_features,
 		   allowed_features, KVM_VCPU_MAX_FEATURES);
diff --git a/arch/arm64/kvm/hypercalls.c b/arch/arm64/kvm/hypercalls.c
index b11b8821c9fbc..e43ed13c82bec 100644
--- a/arch/arm64/kvm/hypercalls.c
+++ b/arch/arm64/kvm/hypercalls.c
@@ -580,6 +580,13 @@ int kvm_arm_set_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
 
 		wants_02 = vcpu_has_feature(vcpu, KVM_ARM_VCPU_PSCI_0_2);
 
+		/*
+		 * EL2 advertises PSCI 1.1; the host handles the forwarded calls
+		 * by this version.
+		 */
+		if (vcpu_is_protected(vcpu) && val < KVM_ARM_PSCI_1_1)
+			return -EINVAL;
+
 		switch (val) {
 		case KVM_ARM_PSCI_0_1:
 			if (wants_02)
-- 
2.39.5

[PATCH v4 12/18] KVM: arm64: Prevent host PC adjustments for protected vCPUs

[Fuad Tabba]

__kvm_adjust_pc() lets the host advance a vCPU's PC or inject an
exception, which for a protected vCPU would let the host redirect
guest execution. Leave the request on the host copy there: the entry
handlers apply the host's PC_UPDATE_REQ at the next entry, where EL2
allows it.

__kvm_adjust_pc() adjusts the vCPU a get/put pair returns, the one it
was given outside pKVM. Both host callers hold the vCPU mutex, so a
hyp vCPU loaded for the vCPU is loaded on the calling CPU, and EL2's
own calls pass the hyp vCPU itself. For a loaded protected vCPU,
kvm_adjust_pc_get() returns NULL and the request stays on the host
copy. For a loaded non-protected vCPU, PKVM_HOST_STATE_DIRTY selects
the copy, since adjusting the hyp vCPU while the host copy is
authoritative loses the update at the next flush. Adjusting the hyp
vCPU copies PC_UPDATE_REQ in and back out again: without the copy
back, INCREMENT_PC outlives the adjustment and the next
KVM_SET_VCPU_EVENTS trips WARN_ON(INCREMENT_PC) in
kvm_pend_exception(). With no hyp vCPU loaded, as under
KVM_SET_VCPU_EVENTS, the host copy is adjusted as before.

Until the marshalling patch clears PC_UPDATE_REQ on the host copy at
exit, a KVM_RUN that returns to userspace with INCREMENT_PC set leaves
it on the host copy of a loaded protected vCPU, and a
KVM_SET_VCPU_EVENTS before the next run then trips that WARN_ON.

Suggested-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/exception.c             | 21 +++++++++-----
 arch/arm64/kvm/hyp/include/hyp/adjust_pc.h | 20 ++++++++++++++
 arch/arm64/kvm/hyp/nvhe/hyp-main.c         | 32 ++++++++++++++++++++++
 3 files changed, 66 insertions(+), 7 deletions(-)

diff --git a/arch/arm64/kvm/hyp/exception.c b/arch/arm64/kvm/hyp/exception.c
index 6e60d890afa4a..93d68b1d38be8 100644
--- a/arch/arm64/kvm/hyp/exception.c
+++ b/arch/arm64/kvm/hyp/exception.c
@@ -353,12 +353,19 @@ static void kvm_inject_exception(struct kvm_vcpu *vcpu)
  */
 void __kvm_adjust_pc(struct kvm_vcpu *vcpu)
 {
-	if (vcpu_get_flag(vcpu, PENDING_EXCEPTION)) {
-		kvm_inject_exception(vcpu);
-		vcpu_clear_flag(vcpu, PENDING_EXCEPTION);
-		vcpu_clear_flag(vcpu, EXCEPT_MASK);
-	} else if (vcpu_get_flag(vcpu, INCREMENT_PC)) {
-		kvm_skip_instr(vcpu);
-		vcpu_clear_flag(vcpu, INCREMENT_PC);
+	struct kvm_vcpu *target = kvm_adjust_pc_get(vcpu);
+
+	if (!target)
+		return;
+
+	if (vcpu_get_flag(target, PENDING_EXCEPTION)) {
+		kvm_inject_exception(target);
+		vcpu_clear_flag(target, PENDING_EXCEPTION);
+		vcpu_clear_flag(target, EXCEPT_MASK);
+	} else if (vcpu_get_flag(target, INCREMENT_PC)) {
+		kvm_skip_instr(target);
+		vcpu_clear_flag(target, INCREMENT_PC);
 	}
+
+	kvm_adjust_pc_put(vcpu, target);
 }
diff --git a/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h b/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h
index f55950ee2a7e4..a07936f87af83 100644
--- a/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h
+++ b/arch/arm64/kvm/hyp/include/hyp/adjust_pc.h
@@ -78,4 +78,24 @@ static inline void kvm_skip_host_instr(void)
 	write_sysreg_el2(read_sysreg_el2(SYS_ELR) + 4, SYS_ELR);
 }
 
+/*
+ * Under pKVM, the vCPU __kvm_adjust_pc() adjusts for @vcpu (NULL leaves the
+ * request on @vcpu for its next entry), and the copy of the consumed
+ * PC_UPDATE_REQ back to @vcpu.
+ */
+#ifdef __KVM_NVHE_HYPERVISOR__
+struct kvm_vcpu *kvm_adjust_pc_get(struct kvm_vcpu *vcpu);
+void kvm_adjust_pc_put(struct kvm_vcpu *vcpu, struct kvm_vcpu *target);
+#else
+static inline struct kvm_vcpu *kvm_adjust_pc_get(struct kvm_vcpu *vcpu)
+{
+	return vcpu;
+}
+
+static inline void kvm_adjust_pc_put(struct kvm_vcpu *vcpu,
+				     struct kvm_vcpu *target)
+{
+}
+#endif
+
 #endif
diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 00038a162d09b..2fcd3f8cc8fc4 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -644,6 +644,38 @@ static void handle___pkvm_host_mkyoung_guest(struct kvm_cpu_context *host_ctxt)
 	cpu_reg(host_ctxt, 1) =  ret;
 }
 
+/*
+ * PKVM_HOST_STATE_DIRTY names the authoritative copy, the host's when set.
+ * A loaded protected vCPU takes the request at its next entry instead.
+ */
+struct kvm_vcpu *kvm_adjust_pc_get(struct kvm_vcpu *vcpu)
+{
+	struct pkvm_hyp_vcpu *hyp_vcpu;
+
+	if (!is_protected_kvm_enabled())
+		return vcpu;
+
+	hyp_vcpu = pkvm_get_loaded_hyp_vcpu();
+	if (!hyp_vcpu || vcpu == &hyp_vcpu->vcpu)
+		return vcpu;
+
+	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu))
+		return NULL;
+
+	if (vcpu_get_flag(vcpu, PKVM_HOST_STATE_DIRTY))
+		return vcpu;
+
+	vcpu_copy_flag(&hyp_vcpu->vcpu, vcpu, PC_UPDATE_REQ);
+	return &hyp_vcpu->vcpu;
+}
+
+/* Reflect the consumed request back, otherwise it stays pending. */
+void kvm_adjust_pc_put(struct kvm_vcpu *vcpu, struct kvm_vcpu *target)
+{
+	if (target != vcpu)
+		vcpu_copy_flag(vcpu, target, PC_UPDATE_REQ);
+}
+
 static void handle___kvm_adjust_pc(struct kvm_cpu_context *host_ctxt)
 {
 	DECLARE_REG(struct kvm_vcpu *, vcpu, host_ctxt, 1);
-- 
2.39.5

[PATCH v4 13/18] KVM: arm64: Inject an UNDEF at EL2 for unhandled protected guest exits

[Fuad Tabba]

Make pvm_exit_handlers[] an allow-list: a protected guest's exit class
is handled at EL2, forwarded where the host emulates it, or results in
an UNDEF at EL2. Any other class, once forwarded, would re-execute
forever. The host resolves such an exit by injecting an exception or
incrementing the PC. The marshalling that follows allows neither: a
protected vCPU takes only the UNDEF on a SYS64 trap and the external
abort on a forwarded IABT or DABT that EL2 builds for it.

WFxT takes the UNDEF, as it isn't advertised to protected guests.
FP/SIMD keeps the lazy switch, and the watchpoint entry is dropped: a
protected vCPU runs with TDE clear.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/switch.c | 29 +++++++++++++++++++++++++----
 1 file changed, 25 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/switch.c b/arch/arm64/kvm/hyp/nvhe/switch.c
index 7318e3e6a5f36..91e1edf40bcd2 100644
--- a/arch/arm64/kvm/hyp/nvhe/switch.c
+++ b/arch/arm64/kvm/hyp/nvhe/switch.c
@@ -203,15 +203,36 @@ static const exit_handler_fn hyp_exit_handlers[] = {
 	[ESR_ELx_EC_MOPS]		= kvm_hyp_handle_mops,
 };
 
+/* WFI/WFE exit to the host, which emulates them. WFxT is not advertised. */
+static bool kvm_handle_pvm_wfx(struct kvm_vcpu *vcpu, u64 *exit_code)
+{
+	if (kvm_vcpu_get_esr(vcpu) & ESR_ELx_WFx_ISS_WFxT)
+		return kvm_handle_pvm_restricted(vcpu, exit_code);
+
+	return false;
+}
+
+/* Lazy FP/SIMD switch, or the UNDEF that would otherwise be left to the host. */
+static bool kvm_handle_pvm_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
+{
+	if (kvm_hyp_handle_fpsimd(vcpu, exit_code))
+		return true;
+
+	return kvm_handle_pvm_restricted(vcpu, exit_code);
+}
+
+/*
+ * A class not listed takes an UNDEF at EL2: the host has no way to
+ * inject one into a protected vCPU.
+ */
 static const exit_handler_fn pvm_exit_handlers[] = {
-	[0 ... ESR_ELx_EC_MAX]		= NULL,
+	[0 ... ESR_ELx_EC_MAX]		= kvm_handle_pvm_restricted,
+	[ESR_ELx_EC_WFx]		= kvm_handle_pvm_wfx,
 	[ESR_ELx_EC_HVC64]		= kvm_handle_pvm_hvc64,
 	[ESR_ELx_EC_SYS64]		= kvm_handle_pvm_sys64,
-	[ESR_ELx_EC_SVE]		= kvm_handle_pvm_restricted,
-	[ESR_ELx_EC_FP_ASIMD]		= kvm_hyp_handle_fpsimd,
+	[ESR_ELx_EC_FP_ASIMD]		= kvm_handle_pvm_fpsimd,
 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
-	[ESR_ELx_EC_WATCHPT_LOW]	= kvm_hyp_handle_watchpt_low,
 	[ESR_ELx_EC_MOPS]		= kvm_hyp_handle_mops,
 };
 
-- 
2.39.5

[PATCH v4 14/18] KVM: arm64: Add per-EC entry/exit state marshalling for protected guests

[Fuad Tabba]

Move a protected guest's state between the hyp vCPU and the host per
exception class instead of copying the whole context. Add
entry_hyp_pvm_handlers[] and exit_hyp_pvm_handlers[] for WFx, SYS64,
IABT, DABT and HVC64, and route protected guests through them: on exit
each handler copies out only what its class requires, and on re-entry
only what the host may have changed: the PC update or exception it
requested (PC_UPDATE_REQ), the value of a read it emulated, and a
forwarded PSCI call's return value. entry_hyp_vm_handlers[] is
removed: a non-protected vCPU's iflags are copied wholesale.

The host's fault view is EL2's own syndrome with the guest register
index withheld, the deferred SError syndrome (DISR_EL1), and the
addresses each class requires. The value of a written register is
passed in r0. MMIO data is clamped to the access width and, for a load,
sign-extended at EL2 from EL2's syndrome. Endianness stays with the
host. A data abort's PC update is taken for a completed MMIO access,
and for a cache maintenance operation the host skips on unbacked
memory, as it does for any guest.

The host's copy of a protected vCPU's PSTATE is a view the guest never
runs from: its reset value has PSTATE.A set, and EL2 sets PSTATE.A in
the mode it copies out regardless of the guest's, so, unless the VMM
wrote the host copy's PSTATE or SCTLR2_EL1 before the first run,
serror_is_masked() is true and kvm_inject_serror_esr() pends a
host-injected SError through HCR_EL2.VSE, masked by the guest's own
PSTATE.A, instead of emulating the entry on that copy.

Neither dispatch runs for a trap taken with an SError pending: EL2
doesn't handle it, and the guest replays it once the host has
injected the SError. The exit handlers would otherwise marshal a trap
EL2 never handled, and handle_pvm_exit_hvc64() would panic on an
unfiltered function id.

For HVC64, only the PSCI calls EL2 forwards reach the host: the exit
handler passes the function id and the arguments each call takes, and
the entry handler returns the host's result. A CPU_ON the host failed
is rolled back to OFF and returned as INTERNAL_FAILURE, or as
ALREADY_ON when that's what the host returned: PSCI defines it as the
retry signal for a CPU_ON that reaches the implementation before the
target's CPU_OFF has been processed (DEN0022 section 6.6), which a
guest that doesn't poll AFFINITY_INFO first can do. A target that
already ran returns SUCCESS. The rollback leaves the published reset
state in place: clearing it races a fresh CPU_ON's publication and
leaves the target stuck at ON_PENDING, and the entry point left behind
is one the guest supplied. The rollback stores OFF with a release, as
CPU_OFF does, so this vCPU's writes to the reset state are ordered
before the next winner's. The rollback's cmpxchg carries no generation,
so one that lands after a later CPU_ON has republished ON_PENDING
cancels that cycle too. The target is then OFF at EL2 while the host
has it runnable, and its CPU_ONs return ALREADY_ON until the VMM stops
it again.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/hyp/nvhe/hyp-main.c | 435 +++++++++++++++++++++++++++--
 1 file changed, 412 insertions(+), 23 deletions(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 2fcd3f8cc8fc4..eb73374bbf87b 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -4,6 +4,8 @@
  * Author: Andrew Scull <ascull@google.com>
  */
 
+#include <kvm/arm_hypercalls.h>
+
 #include <hyp/adjust_pc.h>
 #include <hyp/switch.h>
 
@@ -34,13 +36,371 @@ void __kvm_hyp_host_forward_smc(struct kvm_cpu_context *host_ctxt);
 
 typedef void (*hyp_entry_exit_handler_fn)(struct pkvm_hyp_vcpu *);
 
-static void handle_vm_entry_generic(struct pkvm_hyp_vcpu *hyp_vcpu)
+static bool pvm_sys64_is_write(u64 esr)
 {
-	vcpu_copy_flag(&hyp_vcpu->vcpu, hyp_vcpu->host_vcpu, PC_UPDATE_REQ);
+	return (esr & ESR_ELx_SYS64_ISS_DIR_MASK) == ESR_ELx_SYS64_ISS_DIR_WRITE;
 }
 
-static const hyp_entry_exit_handler_fn entry_hyp_vm_handlers[] = {
-	[0 ... ESR_ELx_EC_MAX]		= handle_vm_entry_generic,
+static void handle_pvm_entry_wfx(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+
+	/* Exceptions have priority; the host injects none on WFx. */
+	if (vcpu_get_flag(host_vcpu, PENDING_EXCEPTION))
+		return;
+
+	if (vcpu_get_flag(host_vcpu, INCREMENT_PC)) {
+		vcpu_clear_flag(vcpu, PC_UPDATE_REQ);
+		kvm_incr_pc(vcpu);
+	}
+}
+
+static void handle_pvm_entry_sys64(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	bool pc_update;
+
+	/* Exceptions have priority over anything else */
+	if (vcpu_get_flag(host_vcpu, PENDING_EXCEPTION)) {
+		/* A host-requested exception on SYS64 is always an UNDEF. */
+		u32 esr = (ESR_ELx_EC_UNKNOWN << ESR_ELx_EC_SHIFT) | ESR_ELx_IL;
+
+		__vcpu_assign_sys_reg(vcpu, ESR_EL1, esr);
+		kvm_pend_exception(vcpu, EXCEPT_AA64_EL1_SYNC);
+		return;
+	}
+
+	/* Handle PC increment on a host-emulated access */
+	pc_update = vcpu_get_flag(host_vcpu, INCREMENT_PC);
+	if (pc_update) {
+		vcpu_clear_flag(vcpu, PC_UPDATE_REQ);
+		kvm_incr_pc(vcpu);
+	}
+
+	/* If the host emulated a read access, update the register */
+	if (pc_update && !pvm_sys64_is_write(kvm_vcpu_get_esr(vcpu))) {
+		/* r0 as transfer register between the guest and the host. */
+		u64 rt_val = READ_ONCE(vcpu_gp_regs(host_vcpu)->regs[0]);
+		int rt = kvm_vcpu_sys_get_rt(vcpu);
+
+		vcpu_set_reg(vcpu, rt, rt_val);
+	}
+}
+
+static void handle_pvm_entry_iabt(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	unsigned long cpsr = *vcpu_cpsr(vcpu);
+	u32 esr = ESR_ELx_IL;
+
+	if (!vcpu_get_flag(host_vcpu, PENDING_EXCEPTION))
+		return;
+
+	/* The host's only IABT injection: an external abort. */
+	if ((cpsr & PSR_MODE_MASK) == PSR_MODE_EL0t)
+		esr |= (ESR_ELx_EC_IABT_LOW << ESR_ELx_EC_SHIFT);
+	else
+		esr |= (ESR_ELx_EC_IABT_CUR << ESR_ELx_EC_SHIFT);
+
+	esr |= ESR_ELx_FSC_EXTABT;
+
+	__vcpu_assign_sys_reg(vcpu, ESR_EL1, esr);
+	__vcpu_assign_sys_reg(vcpu, FAR_EL1, kvm_vcpu_get_hfar(vcpu));
+
+	/* Injected by __kvm_adjust_pc() on entry. */
+	kvm_pend_exception(vcpu, EXCEPT_AA64_EL1_SYNC);
+}
+
+/*
+ * Clamp MMIO data to the access width, so a write does not leak the
+ * register's upper bits and a read takes no bits beyond the load. The
+ * host applies endianness.
+ */
+static inline u64 kvm_mmio_clamp_data(struct kvm_vcpu *vcpu, u64 val)
+{
+	unsigned int len = kvm_vcpu_dabt_get_as(vcpu);
+
+	return val & GENMASK_U64(len * 8 - 1, 0);
+}
+
+/*
+ * Complete an MMIO load: sign-extend from EL2's own syndrome, as the
+ * architecture does.
+ */
+static inline u64 kvm_mmio_read_data(struct kvm_vcpu *vcpu, u64 val)
+{
+	val = kvm_mmio_clamp_data(vcpu, val);
+
+	if (kvm_vcpu_dabt_issext(vcpu))
+		val = sign_extend64(val, kvm_vcpu_dabt_get_as(vcpu) * 8 - 1);
+
+	if (!kvm_vcpu_dabt_issf(vcpu))
+		val &= GENMASK_U64(31, 0);
+
+	return val;
+}
+
+static void handle_pvm_entry_dabt(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	bool pc_update;
+
+	/* Exceptions have priority over anything else */
+	if (vcpu_get_flag(host_vcpu, PENDING_EXCEPTION)) {
+		unsigned long cpsr = *vcpu_cpsr(vcpu);
+		u32 esr = ESR_ELx_IL;
+
+		if ((cpsr & PSR_MODE_MASK) == PSR_MODE_EL0t)
+			esr |= (ESR_ELx_EC_DABT_LOW << ESR_ELx_EC_SHIFT);
+		else
+			esr |= (ESR_ELx_EC_DABT_CUR << ESR_ELx_EC_SHIFT);
+
+		esr |= ESR_ELx_FSC_EXTABT;
+
+		__vcpu_assign_sys_reg(vcpu, ESR_EL1, esr);
+		__vcpu_assign_sys_reg(vcpu, FAR_EL1, kvm_vcpu_get_hfar(vcpu));
+
+		/* Injected by __kvm_adjust_pc() on entry. */
+		kvm_pend_exception(vcpu, EXCEPT_AA64_EL1_SYNC);
+
+		/* Cancel any in-flight MMIO */
+		vcpu->mmio_needed = false;
+		return;
+	}
+
+	/* Handle PC increment on MMIO, or on a CMO the host skipped */
+	pc_update = vcpu_get_flag(host_vcpu, INCREMENT_PC) &&
+		    (vcpu->mmio_needed || kvm_vcpu_dabt_is_cm(vcpu));
+	if (pc_update) {
+		vcpu_clear_flag(vcpu, PC_UPDATE_REQ);
+		kvm_incr_pc(vcpu);
+	}
+
+	/* If the host emulated an MMIO read, update the register */
+	if (pc_update && vcpu->mmio_needed && !kvm_vcpu_dabt_iswrite(vcpu)) {
+		/* r0 as transfer register between the guest and the host. */
+		u64 rd_val = READ_ONCE(vcpu_gp_regs(host_vcpu)->regs[0]);
+		int rd = kvm_vcpu_dabt_get_rd(vcpu);
+
+		rd_val = kvm_mmio_read_data(vcpu, rd_val);
+		vcpu_set_reg(vcpu, rd, rd_val);
+	}
+
+	vcpu->mmio_needed = false;
+}
+
+static void handle_pvm_entry_hvc64(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	u64 ret = READ_ONCE(vcpu_gp_regs(host_vcpu)->regs[0]);
+	u32 psci_fn = smccc_get_function(vcpu);
+
+	switch (psci_fn) {
+	case PSCI_0_2_FN_CPU_ON:
+	case PSCI_0_2_FN64_CPU_ON:
+		/*
+		 * Roll back a CPU_ON the host failed, unless the target
+		 * already reached ON: it is running, and the guest sees
+		 * SUCCESS.
+		 */
+		if (ret != PSCI_RET_SUCCESS) {
+			unsigned long cpu_id = smccc_get_arg1(vcpu);
+			struct pkvm_hyp_vcpu *target_vcpu;
+			struct pkvm_hyp_vm *hyp_vm;
+			int prev;
+
+			hyp_vm = pkvm_hyp_vcpu_to_hyp_vm(hyp_vcpu);
+			target_vcpu = pkvm_mpidr_to_hyp_vcpu(hyp_vm, cpu_id);
+
+			/*
+			 * pvm_psci_vcpu_on() resolved this MPIDR and vcpus[]
+			 * entries are never removed, so the lookup cannot miss.
+			 * The release orders this vCPU's reset_state writes
+			 * before OFF, for the next CPU_ON.
+			 */
+			prev = cmpxchg_release(&target_vcpu->power_state,
+					       PSCI_0_2_AFFINITY_LEVEL_ON_PENDING,
+					       PSCI_0_2_AFFINITY_LEVEL_OFF);
+			switch (prev) {
+			case PSCI_0_2_AFFINITY_LEVEL_ON_PENDING:
+				/*
+				 * Leave reset_state.reset set: a clear races a
+				 * fresh CPU_ON's publish. The stale pc/r0/be are
+				 * the guest's own. ALREADY_ON is PSCI's retry
+				 * signal for a CPU_ON that raced the CPU_OFF.
+				 */
+				if (ret != PSCI_RET_ALREADY_ON)
+					ret = PSCI_RET_INTERNAL_FAILURE;
+				break;
+			case PSCI_0_2_AFFINITY_LEVEL_ON:
+			case PSCI_0_2_AFFINITY_LEVEL_OFF:
+				/* Target already ran (and may have stopped). */
+				ret = PSCI_RET_SUCCESS;
+				break;
+			default:
+				ret = PSCI_RET_INTERNAL_FAILURE;
+				break;
+			}
+		}
+
+		break;
+	default:
+		break;
+	}
+
+	vcpu_set_reg(vcpu, 0, ret);
+}
+
+/* The host's view of a syndrome: the guest register index is withheld. */
+static u64 pvm_host_esr(u64 esr)
+{
+	switch (ESR_ELx_EC(esr)) {
+	case ESR_ELx_EC_WFx:
+		return esr & ~ESR_ELx_WFx_ISS_RN;
+	case ESR_ELx_EC_SYS64:
+		return esr & ~ESR_ELx_SYS64_ISS_RT_MASK;
+	case ESR_ELx_EC_DABT_LOW:
+		return esr & ~ESR_ELx_SRT_MASK;
+	default:
+		return esr;
+	}
+}
+
+/*
+ * The host's view of PSTATE: the mode, with SErrors masked so that the host
+ * pends an SError through HCR_EL2.VSE rather than emulating the entry.
+ */
+static unsigned long pvm_host_pstate(unsigned long pstate)
+{
+	return (pstate & PSR_MODE_MASK) | PSR_A_BIT;
+}
+
+static void handle_pvm_exit_wfx(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+
+	*vcpu_cpsr(host_vcpu) = pvm_host_pstate(*vcpu_cpsr(vcpu));
+}
+
+static void handle_pvm_exit_sys64(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	u64 esr = kvm_vcpu_get_esr(vcpu);
+
+	/* The mode is required for the host to emulate some sysregs */
+	*vcpu_cpsr(host_vcpu) = pvm_host_pstate(*vcpu_cpsr(vcpu));
+
+	/* r0 as transfer register between the guest and the host. */
+	if (pvm_sys64_is_write(esr)) {
+		int rt = kvm_vcpu_sys_get_rt(vcpu);
+		u64 rt_val = vcpu_get_reg(vcpu, rt);
+
+		vcpu_set_reg(host_vcpu, 0, rt_val);
+	}
+}
+
+static void handle_pvm_exit_iabt(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+
+	host_vcpu->arch.fault.hpfar_el2 = vcpu->arch.fault.hpfar_el2;
+}
+
+static void handle_pvm_exit_dabt(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	u64 sctlr;
+
+	/*
+	 * EL2 has no memslot view: a decodable data abort is prepared as MMIO
+	 * for the host to resolve. On unbacked memory the host injects an SEA
+	 * for one with ISV clear (LDP/STP, atomics), which EL2 does not
+	 * decode, and skips a cache maintenance operation, as for any guest.
+	 */
+	vcpu->mmio_needed = kvm_vcpu_dabt_isvalid(vcpu);
+
+	/* r0 as transfer register between the guest and the host. */
+	if (vcpu->mmio_needed && kvm_vcpu_dabt_iswrite(vcpu)) {
+		int rt = kvm_vcpu_dabt_get_rd(vcpu);
+		u64 rt_val = vcpu_get_reg(vcpu, rt);
+
+		rt_val = kvm_mmio_clamp_data(vcpu, rt_val);
+		vcpu_set_reg(host_vcpu, 0, rt_val);
+	}
+
+	*vcpu_cpsr(host_vcpu) = pvm_host_pstate(*vcpu_cpsr(vcpu));
+	host_vcpu->arch.fault.far_el2 = vcpu->arch.fault.far_el2 & GENMASK(11, 0);
+	host_vcpu->arch.fault.hpfar_el2 = vcpu->arch.fault.hpfar_el2;
+	sctlr = __vcpu_sys_reg(vcpu, SCTLR_EL1) & (SCTLR_ELx_EE | SCTLR_EL1_E0E);
+	__vcpu_assign_sys_reg(host_vcpu, SCTLR_EL1, sctlr);
+}
+
+static void handle_pvm_exit_hvc64(struct pkvm_hyp_vcpu *hyp_vcpu)
+{
+	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	struct kvm_vcpu *vcpu = &hyp_vcpu->vcpu;
+	int n, i;
+
+	switch (smccc_get_function(vcpu)) {
+	/*
+	 * CPU_ON: the host uses only the target MPIDR (x1). EL2 resets the
+	 * target from its own copy of the entry point and context id.
+	 */
+	case PSCI_0_2_FN_CPU_ON:
+	case PSCI_0_2_FN64_CPU_ON:
+		n = 2;
+		break;
+
+	case PSCI_0_2_FN_CPU_OFF:
+	case PSCI_0_2_FN_SYSTEM_OFF:
+	case PSCI_0_2_FN_SYSTEM_RESET:
+	case PSCI_0_2_FN_CPU_SUSPEND:
+	case PSCI_0_2_FN64_CPU_SUSPEND:
+		n = 1;
+		break;
+
+	case PSCI_0_2_FN_AFFINITY_INFO:
+	case PSCI_0_2_FN64_AFFINITY_INFO:
+	case PSCI_1_1_FN_SYSTEM_RESET2:
+	case PSCI_1_1_FN64_SYSTEM_RESET2:
+		n = 3;
+		break;
+
+	/* Unreachable: kvm_handle_pvm_hvc64() forwards only the calls above. */
+	default:
+		hyp_panic();
+	}
+
+	/* Pass the HVC function id (r0) and its arguments. */
+	for (i = 0; i < n; i++)
+		vcpu_set_reg(host_vcpu, i, vcpu_get_reg(vcpu, i));
+}
+
+static const hyp_entry_exit_handler_fn entry_hyp_pvm_handlers[] = {
+	[0 ... ESR_ELx_EC_MAX]		= NULL,
+	[ESR_ELx_EC_WFx]		= handle_pvm_entry_wfx,
+	[ESR_ELx_EC_SYS64]		= handle_pvm_entry_sys64,
+	[ESR_ELx_EC_IABT_LOW]		= handle_pvm_entry_iabt,
+	[ESR_ELx_EC_DABT_LOW]		= handle_pvm_entry_dabt,
+	[ESR_ELx_EC_HVC64]		= handle_pvm_entry_hvc64,
+};
+
+static const hyp_entry_exit_handler_fn exit_hyp_pvm_handlers[] = {
+	[0 ... ESR_ELx_EC_MAX]		= NULL,
+	[ESR_ELx_EC_WFx]		= handle_pvm_exit_wfx,
+	[ESR_ELx_EC_SYS64]		= handle_pvm_exit_sys64,
+	[ESR_ELx_EC_IABT_LOW]		= handle_pvm_exit_iabt,
+	[ESR_ELx_EC_DABT_LOW]		= handle_pvm_exit_dabt,
+	[ESR_ELx_EC_HVC64]		= handle_pvm_exit_hvc64,
 };
 
 static void __hyp_sve_save_guest(struct kvm_vcpu *vcpu)
@@ -284,18 +644,6 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 
 		hyp_vcpu->vcpu.arch.mdcr_el2 = host_vcpu->arch.mdcr_el2;
 		hyp_vcpu->vcpu.arch.iflags = host_vcpu->arch.iflags;
-	} else {
-		u64 v_cval = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CVAL_EL0];
-		u64 v_ctl = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CTL_EL0];
-		u64 p_cval = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CVAL_EL0];
-		u64 p_ctl = hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CTL_EL0];
-
-		hyp_vcpu->vcpu.arch.ctxt = host_vcpu->arch.ctxt;
-
-		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CVAL_EL0] = v_cval;
-		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTV_CTL_EL0] = v_ctl;
-		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CVAL_EL0] = p_cval;
-		hyp_vcpu->vcpu.arch.ctxt.sys_regs[CNTP_CTL_EL0] = p_ctl;
 	}
 
 	/* __hyp_running_vcpu must be NULL in a guest context. */
@@ -320,10 +668,16 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 	case ARM_EXCEPTION_IL:
 		break;
 	case ARM_EXCEPTION_TRAP:
-		esr_ec = ESR_ELx_EC(kvm_vcpu_get_esr(&hyp_vcpu->vcpu));
-		ec_handler = entry_hyp_vm_handlers[esr_ec];
-		if (ec_handler)
-			ec_handler(hyp_vcpu);
+		/* Nothing was marshalled for this trap, see sync_hyp_vcpu(). */
+		if (ARM_SERROR_PENDING(hyp_vcpu->exit_code))
+			break;
+
+		if (pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
+			esr_ec = ESR_ELx_EC(kvm_vcpu_get_esr(&hyp_vcpu->vcpu));
+			ec_handler = entry_hyp_pvm_handlers[esr_ec];
+			if (ec_handler)
+				ec_handler(hyp_vcpu);
+		}
 		break;
 	default:
 		BUG();
@@ -335,13 +689,26 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
 static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu, u32 exit_reason)
 {
 	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
+	hyp_entry_exit_handler_fn ec_handler;
+	u8 esr_ec;
 
 	fpsimd_sve_sync(&hyp_vcpu->vcpu);
 	sync_debug_state(hyp_vcpu);
 
 	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
-		host_vcpu->arch.ctxt = hyp_vcpu->vcpu.arch.ctxt;
+		/*
+		 * Protected: the host sees ESR_EL2 as EL2 took it, register
+		 * index withheld; the fault addresses stay withheld unless the
+		 * EC handler below adds them.
+		 */
+		host_vcpu->arch.fault = (struct kvm_vcpu_fault_info) {
+			.esr_el2 = pvm_host_esr(hyp_vcpu->vcpu.arch.fault.esr_el2),
+			.disr_el1 = hyp_vcpu->vcpu.arch.fault.disr_el1,
+		};
 	} else {
+		/* Non-protected: the host gets the full fault. */
+		host_vcpu->arch.fault = hyp_vcpu->vcpu.arch.fault;
+		host_vcpu->arch.iflags = hyp_vcpu->vcpu.arch.iflags;
 		/*
 		 * PC feeds trace_kvm_exit(), PSTATE.SS the host software-step
 		 * machine, and both run before the next on-demand ctxt sync.
@@ -350,9 +717,28 @@ static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu, u32 exit_reason)
 		host_vcpu->arch.ctxt.regs.pstate = hyp_vcpu->vcpu.arch.ctxt.regs.pstate;
 	}
 
-	host_vcpu->arch.fault		= hyp_vcpu->vcpu.arch.fault;
+	switch (ARM_EXCEPTION_CODE(exit_reason)) {
+	case ARM_EXCEPTION_IRQ:
+		break;
+	case ARM_EXCEPTION_TRAP:
+		/* SError pending: not handled at EL2, the guest replays it. */
+		if (ARM_SERROR_PENDING(exit_reason))
+			break;
 
-	host_vcpu->arch.iflags		= hyp_vcpu->vcpu.arch.iflags;
+		/* Per-EC marshalling is for protected guests only. */
+		if (pkvm_hyp_vcpu_is_protected(hyp_vcpu)) {
+			esr_ec = ESR_ELx_EC(kvm_vcpu_get_esr(&hyp_vcpu->vcpu));
+			ec_handler = exit_hyp_pvm_handlers[esr_ec];
+			if (ec_handler)
+				ec_handler(hyp_vcpu);
+		}
+		break;
+	case ARM_EXCEPTION_EL1_SERROR:
+	case ARM_EXCEPTION_IL:
+		break;
+	default:
+		BUG();
+	}
 
 	/* Cleared by hardware once the guest takes the vSError. */
 	host_vcpu->arch.hcr_el2 &= ~HCR_VSE;
@@ -361,6 +747,9 @@ static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu, u32 exit_reason)
 	sync_hyp_vgic_state(hyp_vcpu);
 	sync_hyp_timer_state(hyp_vcpu);
 
+	if (pkvm_hyp_vcpu_is_protected(hyp_vcpu))
+		vcpu_clear_flag(host_vcpu, PC_UPDATE_REQ);
+
 	hyp_vcpu->exit_code = exit_reason;
 }
 
-- 
2.39.5

[PATCH v4 15/18] KVM: arm64: Reject host access to protected VM private state

[Fuad Tabba]

A protected vCPU's register and debug state is no longer exposed to
the host, and the host ioctls that would reach it fail instead:

  - KVM_GET_ONE_REG and KVM_SET_ONE_REG return -EPERM once the vCPU has
    run: the host copy then holds the boot state the VMM wrote plus
    what the exit handlers copy out, which the VMM can't tell apart.
    Pre-run access still builds the guest's boot state.
  - KVM_ARM_VCPU_INIT returns -EPERM once the vCPU has run: it would
    reset the host copy alone and rewrite mp_state, which EL2 reads
    only at hyp vCPU creation, so a vCPU the guest powered off would
    come back RUNNABLE.
  - KVM_SET_VCPU_EVENTS rejects external-abort injection with -EPERM;
    SError injection is forwarded and stays permitted.
  - KVM_SET_GUEST_DEBUG returns -EPERM: a protected guest's debug state
    is hypervisor-owned.

The KVM_{GET,SET}_ONE_REG and KVM_ARM_VCPU_INIT checks are one filter
on the ioctl number in kvm_arch_vcpu_ioctl(), the only caller of the
three functions they were in. Its -EPERM now precedes the cases'
-EFAULT and the ONE_REG case's pending-reset handling, which the next
KVM_RUN performs. The external-abort check reads the payload, and
KVM_SET_GUEST_DEBUG has its own case in kvm_vcpu_ioctl(), so it never
reaches kvm_arch_vcpu_ioctl(): those two stay in their handlers.

KVM_CHECK_EXTENSION returns 0 for KVM_CAP_ARM_INJECT_EXT_DABT and
KVM_CAP_SET_GUEST_DEBUG on a protected VM: kvm_pkvm_ext_allowed()
returns false on every capability it doesn't list, and the patch that
advertises the capabilities protected VMs support leaves these two
out. The two ioctl checks stay for a VMM that doesn't query
KVM_CHECK_EXTENSION.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/kvm/arm.c   | 22 ++++++++++++++++++++++
 arch/arm64/kvm/guest.c | 11 +++++++++++
 2 files changed, 33 insertions(+)

diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index ca6e109b3e5d6..14fe4e7275796 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -1865,6 +1865,24 @@ static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
 	return __kvm_arm_vcpu_set_events(vcpu, events);
 }
 
+/*
+ * Once a protected vCPU has run, the host copy holds the boot state the VMM
+ * wrote plus what the exit handlers copy out, and EL2 has read mp_state,
+ * which it does only at hyp vCPU creation.
+ */
+static long pkvm_filter_vcpu_ioctl(struct kvm_vcpu *vcpu, unsigned int ioctl)
+{
+	switch (ioctl) {
+	case KVM_ARM_VCPU_INIT:
+	case KVM_SET_ONE_REG:
+	case KVM_GET_ONE_REG:
+		if (vcpu_is_protected(vcpu) && vcpu_has_run_once(vcpu))
+			return -EPERM;
+	}
+
+	return 0;
+}
+
 long kvm_arch_vcpu_ioctl(struct file *filp,
 			 unsigned int ioctl, unsigned long arg)
 {
@@ -1873,6 +1891,10 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
 	struct kvm_device_attr attr;
 	long r;
 
+	r = pkvm_filter_vcpu_ioctl(vcpu, ioctl);
+	if (r)
+		return r;
+
 	switch (ioctl) {
 	case KVM_ARM_VCPU_INIT: {
 		struct kvm_vcpu_init init;
diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c
index b01d6622b8720..d5e2e08f05461 100644
--- a/arch/arm64/kvm/guest.c
+++ b/arch/arm64/kvm/guest.c
@@ -786,6 +786,13 @@ int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
 	u64 esr = events->exception.serror_esr;
 	int ret = 0;
 
+	/*
+	 * EL2 injects an external abort only to complete a forwarded abort.
+	 * SError injection is forwarded.
+	 */
+	if (vcpu_is_protected(vcpu) && ext_dabt_pending)
+		return -EPERM;
+
 	/*
 	 * Immediately commit the pending SEA to the vCPU's architectural
 	 * state which is necessary since we do not return a pending SEA
@@ -883,6 +890,10 @@ int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 {
 	trace_kvm_set_guest_debug(vcpu, dbg->control);
 
+	/* A protected guest's debug state is not exposed to the host. */
+	if (vcpu_is_protected(vcpu))
+		return -EPERM;
+
 	if (dbg->control & ~KVM_GUESTDBG_VALID_MASK)
 		return -EINVAL;
 
-- 
2.39.5

[PATCH v4 16/18] KVM: arm64: Reject host power-on of a vCPU that EL2 holds powered off

[Fuad Tabba]

A protected vCPU's power state belongs to EL2, which changes it only
on the guest's own PSCI calls. KVM_SET_MP_STATE(RUNNABLE) on a vCPU
EL2 holds powered off changes the host's copy alone, and the guest's
next CPU_ON of that vCPU then fails: mp_state is no longer STOPPED, so
the host returns ALREADY_ON, and the guest's retry fails the same way
until the VMM stops the vCPU again. SUSPENDED has the same effect.

Record a vCPU EL2 holds powered off as KVM_MP_STATE_UNINITIALIZED, an
encoding arm64 didn't use: the guest's CPU_OFF writes it, as does hyp
vCPU creation for a vCPU that starts STOPPED, and only the guest's
CPU_ON writes RUNNABLE over it. RUNNABLE and SUSPENDED from the VMM
return -EPERM on it, the VMM can't write the value itself, and
KVM_ARM_VCPU_INIT, which would write STOPPED or RUNNABLE over it, is
rejected once the vCPU has run.

The VMM can still pause and resume a vCPU EL2 has powered on. Neither
STOPPED nor kvm_prepare_system_event() overwrites the value: a vCPU at
it is already off, and a STOPPED in its place would turn the guest's
next CPU_ON into ALREADY_ON, after a SYSTEM_RESET exit the VMM runs on
from. kvm_arm_vcpu_stopped() covers it, so the run loop sleeps the
vCPU as before.

The guest's CPU_ON is gated on the value rather than on STOPPED. A
STOPPED written between the target's CPU_OFF exit and the host's
power-off of it, which kvm_prepare_system_event() can do from another
vCPU's SYSTEM_OFF or SYSTEM_RESET, would otherwise let the CPU_ON
through, and the power-off would then record the target powered off
while EL2 holds it ON_PENDING, with no way back before VM teardown.
Gated on the value, that CPU_ON returns ALREADY_ON and the retry
succeeds once the power-off has run.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/include/asm/kvm_host.h |  1 +
 arch/arm64/kvm/arm.c              | 35 +++++++++++++++++++++++++------
 arch/arm64/kvm/pkvm.c             | 21 +++++++++++++++----
 arch/arm64/kvm/psci.c             | 11 ++++++++--
 4 files changed, 56 insertions(+), 12 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index 37d0721d39a45..e99b706c879cf 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -1564,6 +1564,7 @@ static inline void kvm_hyp_reserve(void) { }
 
 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
+bool kvm_pkvm_vcpu_is_powered_off(struct kvm_vcpu *vcpu);
 
 static inline u64 *__vm_id_reg(struct kvm_arch *ka, u32 reg)
 {
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index 14fe4e7275796..a89087c83e27f 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -770,23 +770,28 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 	vcpu->cpu = -1;
 }
 
-static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
+static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu, u32 mp_state)
 {
-	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
+	WRITE_ONCE(vcpu->arch.mp_state.mp_state, mp_state);
 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
 	kvm_vcpu_kick(vcpu);
 }
 
+/* The guest's own CPU_OFF: EL2 has already powered the vCPU off. */
 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
 {
+	u32 mp_state = vcpu_is_protected(vcpu) ? KVM_MP_STATE_UNINITIALIZED : KVM_MP_STATE_STOPPED;
+
 	spin_lock(&vcpu->arch.mp_state_lock);
-	__kvm_arm_vcpu_power_off(vcpu);
+	__kvm_arm_vcpu_power_off(vcpu, mp_state);
 	spin_unlock(&vcpu->arch.mp_state_lock);
 }
 
 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
 {
-	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
+	u32 mp_state = READ_ONCE(vcpu->arch.mp_state.mp_state);
+
+	return mp_state == KVM_MP_STATE_STOPPED || mp_state == KVM_MP_STATE_UNINITIALIZED;
 }
 
 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
@@ -801,6 +806,12 @@ static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
 	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
 }
 
+/* Only the guest's CPU_ON may start a vCPU EL2 holds powered off. */
+bool kvm_pkvm_vcpu_is_powered_off(struct kvm_vcpu *vcpu)
+{
+	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_UNINITIALIZED;
+}
+
 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 				    struct kvm_mp_state *mp_state)
 {
@@ -818,12 +829,24 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 
 	switch (mp_state->mp_state) {
 	case KVM_MP_STATE_RUNNABLE:
+		if (kvm_pkvm_vcpu_is_powered_off(vcpu)) {
+			ret = -EPERM;
+			break;
+		}
+
 		WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
 		break;
 	case KVM_MP_STATE_STOPPED:
-		__kvm_arm_vcpu_power_off(vcpu);
+		/* Keep EL2's off record: STOPPED over it would let RUNNABLE through. */
+		if (!kvm_pkvm_vcpu_is_powered_off(vcpu))
+			__kvm_arm_vcpu_power_off(vcpu, KVM_MP_STATE_STOPPED);
 		break;
 	case KVM_MP_STATE_SUSPENDED:
+		if (kvm_pkvm_vcpu_is_powered_off(vcpu)) {
+			ret = -EPERM;
+			break;
+		}
+
 		kvm_arm_vcpu_suspend(vcpu);
 		break;
 	default:
@@ -1788,7 +1811,7 @@ static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 	spin_lock(&vcpu->arch.mp_state_lock);
 
 	if (power_off)
-		__kvm_arm_vcpu_power_off(vcpu);
+		__kvm_arm_vcpu_power_off(vcpu, KVM_MP_STATE_STOPPED);
 	else
 		WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
 
diff --git a/arch/arm64/kvm/pkvm.c b/arch/arm64/kvm/pkvm.c
index 8e4c6e4bec123..99a5d9a913129 100644
--- a/arch/arm64/kvm/pkvm.c
+++ b/arch/arm64/kvm/pkvm.c
@@ -118,12 +118,25 @@ static int __pkvm_create_hyp_vcpu(struct kvm_vcpu *vcpu)
 		return -ENOMEM;
 
 	ret = kvm_call_hyp_nvhe(__pkvm_init_vcpu, handle, vcpu, hyp_vcpu);
-	if (!ret)
-		vcpu_set_flag(vcpu, VCPU_PKVM_FINALIZED);
-	else
+	if (ret) {
 		free_pages_exact(hyp_vcpu, hyp_vcpu_sz);
+		return ret;
+	}
 
-	return ret;
+	/*
+	 * Mirror EL2's seeding of power_state from mp_state. The hyp vCPU is
+	 * published, so take mp_state_lock against kvm_psci_vcpu_on().
+	 */
+	if (vcpu_is_protected(vcpu)) {
+		spin_lock(&vcpu->arch.mp_state_lock);
+		if (kvm_arm_vcpu_stopped(vcpu))
+			WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_UNINITIALIZED);
+		spin_unlock(&vcpu->arch.mp_state_lock);
+	}
+
+	vcpu_set_flag(vcpu, VCPU_PKVM_FINALIZED);
+
+	return 0;
 }
 
 /*
diff --git a/arch/arm64/kvm/psci.c b/arch/arm64/kvm/psci.c
index e3db84400d1f8..b01f8d44033f1 100644
--- a/arch/arm64/kvm/psci.c
+++ b/arch/arm64/kvm/psci.c
@@ -63,7 +63,12 @@ static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
 		return PSCI_RET_INVALID_PARAMS;
 
 	spin_lock(&vcpu->arch.mp_state_lock);
-	if (!kvm_arm_vcpu_stopped(vcpu)) {
+	/*
+	 * A protected vCPU is off once the host has acted on EL2's CPU_OFF.
+	 * STOPPED before that is the VMM's pause.
+	 */
+	if (!kvm_arm_vcpu_stopped(vcpu) ||
+	    (vcpu_is_protected(vcpu) && !kvm_pkvm_vcpu_is_powered_off(vcpu))) {
 		if (kvm_psci_version(source_vcpu) != KVM_ARM_PSCI_0_1)
 			ret = PSCI_RET_ALREADY_ON;
 		else
@@ -161,7 +166,9 @@ static void kvm_prepare_system_event(struct kvm_vcpu *vcpu, u32 type, u64 flags)
 	 */
 	kvm_for_each_vcpu(i, tmp, vcpu->kvm) {
 		spin_lock(&tmp->arch.mp_state_lock);
-		WRITE_ONCE(tmp->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
+		/* Keep EL2's off record for the guest's next CPU_ON. */
+		if (!kvm_pkvm_vcpu_is_powered_off(tmp))
+			WRITE_ONCE(tmp->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
 		spin_unlock(&tmp->arch.mp_state_lock);
 	}
 	kvm_make_all_cpus_request(vcpu->kvm, KVM_REQ_SLEEP);
-- 
2.39.5

[PATCH v4 17/18] KVM: arm64: Advertise the capabilities that protected VMs support

[Fuad Tabba]

For a protected VM, kvm_pkvm_ext_allowed() returns false on every
capability it doesn't name, so KVM_CHECK_EXTENSION returns 0 for
interfaces the VM implements: it denies KVM_CAP_ONE_REG while
KVM_{GET,SET}_ONE_REG stage the guest's boot state.

Allow the capabilities that work for a protected guest under the
restrictions the preceding patches add, the vGIC and the I/O bus being
host-managed.

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 arch/arm64/include/asm/kvm_pkvm.h | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/arch/arm64/include/asm/kvm_pkvm.h b/arch/arm64/include/asm/kvm_pkvm.h
index fbea052fa3e16..a165af0d75689 100644
--- a/arch/arm64/include/asm/kvm_pkvm.h
+++ b/arch/arm64/include/asm/kvm_pkvm.h
@@ -42,6 +42,15 @@ static inline bool kvm_pkvm_ext_allowed(struct kvm *kvm, long ext)
 	case KVM_CAP_ARM_VM_IPA_SIZE:
 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
+	case KVM_CAP_ONE_REG:
+	case KVM_CAP_MP_STATE:
+	case KVM_CAP_VCPU_EVENTS:
+	case KVM_CAP_VCPU_ATTRIBUTES:
+	case KVM_CAP_IMMEDIATE_EXIT:
+	case KVM_CAP_IOEVENTFD:
+	case KVM_CAP_IRQFD_RESAMPLE:
+	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
+	case KVM_CAP_ARM_INJECT_SERROR_ESR:
 		return true;
 	case KVM_CAP_ARM_MTE:
 		return false;
-- 
2.39.5

[PATCH v4 18/18] KVM: arm64: Document the protected VM userspace API

[Fuad Tabba]

EL2 owns a protected vCPU's state once it has run, and the ioctls that
would reach it fail rather than operate on the host copy, which holds
the boot state the VMM wrote plus what the exit handlers copy out.
Describe the resulting API in pkvm.rst, point api.rst at it from each
ioctl that behaves differently, and say beside the PVTIME attribute
in devices/vcpu.rst and the PSCI version register in
fw-pseudo-registers.rst how protected VMs differ. The overview also
states the GIC support pKVM enforces: GICv2 is not supported in
protected mode (vgic-v2.c) and GICv5-based guests aren't registered
under pKVM (vgic-v5.c).

Signed-off-by: Fuad Tabba <fuad.tabba@linux.dev>
---
 Documentation/virt/kvm/api.rst                |  27 ++-
 .../virt/kvm/arm/fw-pseudo-registers.rst      |   2 +
 Documentation/virt/kvm/arm/pkvm.rst           | 164 +++++++++++++++++-
 Documentation/virt/kvm/devices/vcpu.rst       |   3 +-
 4 files changed, 185 insertions(+), 11 deletions(-)

diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
index e0430cc750c9e..7fc0fed0edb18 100644
--- a/Documentation/virt/kvm/api.rst
+++ b/Documentation/virt/kvm/api.rst
@@ -1322,6 +1322,9 @@ guests, across different userspace implementations. Nevertheless, userspace
 can still emulate all Arm exceptions by manipulating individual registers
 using the KVM_SET_ONE_REG API.
 
+For a protected VM, setting ext_dabt_pending returns -EPERM; see
+Documentation/virt/kvm/arm/pkvm.rst. Injecting an SError is unaffected.
+
 See KVM_GET_VCPU_EVENTS for the data structure.
 
 Calling this ioctl on a vCPU that hasn't been initialized will return
@@ -1545,7 +1548,9 @@ Possible values are:
    KVM_MP_STATE_RUNNABLE         the vcpu is currently running
                                  [x86,arm64,riscv,loongarch]
    KVM_MP_STATE_UNINITIALIZED    the vcpu is an application processor (AP)
-                                 which has not yet received an INIT signal [x86]
+                                 which has not yet received an INIT signal
+                                 [x86], or a protected VM's vcpu that the
+                                 guest holds powered off [arm64]
    KVM_MP_STATE_INIT_RECEIVED    the vcpu has received an INIT signal, and is
                                  now ready for a SIPI [x86]
    KVM_MP_STATE_HALTED           the vcpu has executed a HLT instruction and
@@ -1624,6 +1629,12 @@ For arm64/riscv:
 The only states that are valid are KVM_MP_STATE_STOPPED and
 KVM_MP_STATE_RUNNABLE which reflect if the vcpu should be paused or not.
 
+On arm64, once a protected VM's vcpu has run, it reads as
+KVM_MP_STATE_UNINITIALIZED while the guest holds it powered off, with
+CPU_OFF or by not yet bringing it online with CPU_ON, and
+KVM_MP_STATE_RUNNABLE and KVM_MP_STATE_SUSPENDED return -EPERM on it: only
+the guest can power it on. See Documentation/virt/kvm/arm/pkvm.rst.
+
 On LoongArch, only the KVM_MP_STATE_RUNNABLE state is used to reflect
 whether the vcpu is runnable.
 
@@ -2324,7 +2335,8 @@ Errors:
   ENOENT   no such register
   EINVAL   invalid register ID, or no such register or used with VMs in
            protected virtualization mode on s390
-  EPERM    (arm64) register access not allowed before vcpu finalization
+  EPERM    (arm64) register access not allowed before vcpu
+           finalization, or after a protected VM's vcpu has run
   EBUSY    (riscv) changing register value not allowed after the vcpu
            has run at least once
   ======   ============================================================
@@ -2949,7 +2961,8 @@ Errors include:
   ENOENT   no such register
   EINVAL   invalid register ID, or no such register or used with VMs in
            protected virtualization mode on s390
-  EPERM    (arm64) register access not allowed before vcpu finalization
+  EPERM    (arm64) register access not allowed before vcpu
+           finalization, or after a protected VM's vcpu has run
   ======== ============================================================
 
 (These error codes are indicative only: do not rely on a specific error
@@ -3484,6 +3497,7 @@ Errors:
   ======     =================================================================
   EINVAL     the target is unknown, or the combination of features is invalid.
   ENOENT     a features bit specified is unknown.
+  EPERM      the vcpu belongs to a protected VM and has already run.
   ======     =================================================================
 
 This tells KVM what type of CPU to present to the guest, and what
@@ -3512,6 +3526,10 @@ after the vcpu has been run. This will reset the vcpu to its initial
 state. All calls to this function after the initial call must use the same
 target and same set of feature flags, otherwise EINVAL will be returned.
 
+For a protected VM this ioctl returns EPERM once the vcpu has run, and the
+features it accepts are restricted: KVM_ARM_VCPU_PSCI_0_2 is required. See
+Documentation/virt/kvm/arm/pkvm.rst.
+
 Possible features:
 
 	- KVM_ARM_VCPU_POWER_OFF: Starts the CPU in a power-off state.
@@ -3790,6 +3808,9 @@ can be determined by querying the KVM_CAP_GUEST_DEBUG_HW_BPS and
 KVM_CAP_GUEST_DEBUG_HW_WPS capabilities which return a positive number
 indicating the number of supported registers.
 
+On arm64, this ioctl returns -EPERM for a protected VM: debugging a
+protected guest is not supported. See Documentation/virt/kvm/arm/pkvm.rst.
+
 For ppc, the KVM_CAP_PPC_GUEST_DEBUG_SSTEP capability indicates whether
 the single-step debug event (KVM_GUESTDBG_SINGLESTEP) is supported.
 
diff --git a/Documentation/virt/kvm/arm/fw-pseudo-registers.rst b/Documentation/virt/kvm/arm/fw-pseudo-registers.rst
index d78b53b05dfcf..c07471e258ea2 100644
--- a/Documentation/virt/kvm/arm/fw-pseudo-registers.rst
+++ b/Documentation/virt/kvm/arm/fw-pseudo-registers.rst
@@ -33,6 +33,8 @@ The following registers are defined:
     highest PSCI version implemented by KVM and compatible with v0.2)
   - Allows any PSCI version implemented by KVM and compatible with
     v0.2 to be set with SET_ONE_REG
+  - Returns -EINVAL for a version below v1.1 on a protected VM (see
+    Documentation/virt/kvm/arm/pkvm.rst)
   - Affects the whole VM (even if the register view is per-vcpu)
 
 * KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
diff --git a/Documentation/virt/kvm/arm/pkvm.rst b/Documentation/virt/kvm/arm/pkvm.rst
index 514992a79a833..b75bb8ad6f26d 100644
--- a/Documentation/virt/kvm/arm/pkvm.rst
+++ b/Documentation/virt/kvm/arm/pkvm.rst
@@ -15,7 +15,9 @@ Overview
 Booting a host kernel with '``kvm-arm.mode=protected``' enables
 "Protected KVM" (pKVM). During boot, pKVM installs a stage-2 identity
 map page-table for the host and uses it to isolate the hypervisor
-running at EL2 from the rest of the host running at EL1/0.
+running at EL2 from the rest of the host running at EL1/0. pKVM supports
+GICv3 only: GICv2 is not supported in protected mode, and neither are
+GICv5-based guests.
 
 pKVM permits creation of protected virtual machines (pVMs) by passing
 the ``KVM_VM_TYPE_ARM_PROTECTED`` machine type identifier to the
@@ -31,7 +33,7 @@ See hypercalls.rst for more details.
 Isolation mechanisms
 ====================
 
-pKVM relies on a number of mechanisms to isolate PVMs from the host:
+pKVM relies on a number of mechanisms to isolate pVMs from the host:
 
 CPU memory isolation
 --------------------
@@ -50,8 +52,9 @@ stage-2 identity map as part of the donation procedure. This gives rise
 to some user-visible differences when compared to non-protected VMs,
 largely due to the lack of MMU notifiers:
 
-* Memslots cannot be moved or deleted once the pVM has started running.
-* Read-only memslots and dirty logging are not supported.
+* Memslots cannot be moved or deleted once the pVM has started running
+  (``-EPERM``).
+* Read-only memslots and dirty logging are not supported (``-EPERM``).
 * With the exception of swap, file-backed pages cannot be mapped into a
   pVM.
 * Donated pages are accounted against ``RLIMIT_MLOCK`` and so the VMM
@@ -67,12 +70,19 @@ largely due to the lack of MMU notifiers:
   then it will either return ``-EFAULT`` or forcefully reclaim the
   memory pages. Reclaimed memory is zeroed by the hypervisor and a
   subsequent attempt to access it in the pVM will return ``-EFAULT``
-  from the ``VCPU_RUN`` ioctl().
+  from the ``KVM_RUN`` ioctl().
 
 CPU state isolation
 -------------------
 
-Status: **Unimplemented.**
+Status: CPU register state of protected vCPUs is managed entirely at EL2.
+
+pKVM performs the complete context switch for protected vCPUs within the
+hypervisor. The hypervisor sets a protected vCPU's initial state, not the
+host, and only what each exit needs is synchronised back.
+
+The user-visible consequences are described under `API behaviour for
+protected VMs`_.
 
 DMA isolation using an IOMMU
 ----------------------------
@@ -89,13 +99,153 @@ The FF-A proxy ensures that the host cannot share pVM or hypervisor
 memory with Trustzone as part of a "confused deputy" attack.
 
 The PSCI proxy ensures that CPUs always have the stage-2 identity map
-installed when they are executing in the host.
+installed when they are executing in the host. This proxy is distinct
+from the PSCI handling provided to protected guests, which is described
+under `API behaviour for protected VMs`_.
 
 Protected VM firmware (pvmfw)
 -----------------------------
 
 Status: **Unimplemented.**
 
+API behaviour for protected VMs
+===============================
+
+Protected vCPU state is owned by EL2 (see `CPU state isolation`_). The VMM
+configures a vCPU before its first ``KVM_RUN``; afterwards the state is
+private to the guest and the ioctls that access it return ``-EPERM``. The
+vCPU ioctls' errors follow one rule: ``-EPERM`` means the host asked for
+state that the guest owns, ``-EINVAL`` means the request is not valid for a
+protected VM, and a feature pKVM doesn't support for protected VMs is
+rejected the way KVM rejects one that isn't implemented (``-ENXIO`` from
+``KVM_ARM_VCPU_PVTIME_CTRL``, for example). The ioctls themselves are
+described in Documentation/virt/kvm/api.rst.
+
+Boot
+----
+
+A protected VM boots from a single primary vCPU. Before the first
+``KVM_RUN``, the VMM prepares the boot state:
+
+* Set ``KVM_MP_STATE_RUNNABLE`` on the primary vCPU and
+  ``KVM_MP_STATE_STOPPED`` on every other vCPU. EL2 allows only one
+  RUNNABLE primary per protected VM. A second RUNNABLE vCPU fails at its
+  first ``KVM_RUN``.
+* Set the primary vCPU's boot state with ``KVM_SET_ONE_REG``: the kernel
+  entry address in ``PC`` and the DTB pointer in ``x0``.
+
+``PC`` and ``x0`` are the only registers EL2 takes from the host. Other
+pre-run writes are accepted, but the guest starts from the state the
+hypervisor resets it to.
+
+Secondary vCPUs are started by the guest itself through PSCI ``CPU_ON``
+(see `Power state`_), which supplies their entry point and context ID.
+The VMM cannot choose where they boot.
+
+vCPU state
+----------
+
+* ``KVM_GET_ONE_REG`` and ``KVM_SET_ONE_REG`` return ``-EPERM`` once the
+  vCPU has run. Before that, they access the host-side copy from which
+  EL2 builds the guest's boot state (see `Boot`_).
+* ``KVM_ARM_VCPU_INIT`` accepts only the vCPU features that a protected
+  guest supports and returns ``-EINVAL`` otherwise.
+  ``KVM_ARM_VCPU_PSCI_0_2`` is required, as EL2 implements PSCI 1.1 for
+  the guest (see `Power state`_). ``KVM_ARM_VCPU_EL1_32BIT`` is not
+  supported: protected guests run in AArch64 only and see no AArch32
+  support in ``ID_AA64PFR0_EL1``. Once the vCPU has run,
+  ``KVM_ARM_VCPU_INIT`` returns ``-EPERM``, as re-initialising it would
+  reset the host-side copy alone.
+* ``KVM_SET_VCPU_EVENTS`` returns ``-EPERM`` for external-abort injection
+  (``ext_dabt_pending``), and ``KVM_CAP_ARM_INJECT_EXT_DABT`` is not
+  advertised to a protected VM. SError injection is unaffected.
+* ``KVM_SET_GUEST_DEBUG`` returns ``-EPERM``, and ``KVM_CAP_SET_GUEST_DEBUG``
+  is not advertised to a protected VM (see `Debug`_).
+
+Power state
+-----------
+
+EL2 implements PSCI 1.1 for a protected guest. The calls that move a
+vCPU's power state, ``CPU_ON`` and ``CPU_OFF``, are handled at EL2, and
+the host cannot change the outcome: for ``CPU_ON`` the host only
+schedules the target, which EL2 has already reset to the entry point
+the guest chose, and for ``CPU_OFF`` it only stops scheduling it.
+``AFFINITY_INFO`` is handled by the host, which returns ``OFF`` once
+it has stopped scheduling every vCPU the query covers, as for a
+non-protected VM. A vCPU becomes a valid ``CPU_ON`` target at its first
+``KVM_RUN``, and before that EL2 returns ``INVALID_PARAMETERS``. The
+platform calls, ``CPU_SUSPEND``, ``SYSTEM_OFF``, ``SYSTEM_RESET`` and
+``SYSTEM_RESET2``, are forwarded to the host and behave as for a
+non-protected VM, with the ``SYSTEM_*`` calls exiting to the VMM as
+``KVM_EXIT_SYSTEM_EVENT``; the ``SYSTEM_RESET2`` reset type and cookie
+are in the guest's registers, which ``KVM_GET_ONE_REG`` rejects once
+the vCPU has run. Any other function returns ``NOT_SUPPORTED``, and
+``PSCI_FEATURES`` reports the same set. Because the host handles those
+forwarded calls, ``KVM_SET_ONE_REG`` on ``KVM_REG_ARM_PSCI_VERSION``
+returns ``-EINVAL`` for a version below 1.1.
+
+Once a vCPU has run, its power state follows the guest's PSCI calls, not
+the VMM's. A vCPU that the guest has powered off with ``CPU_OFF``, or has
+not yet brought online with ``CPU_ON``, reads as
+``KVM_MP_STATE_UNINITIALIZED``, and only an in-guest ``CPU_ON`` can power
+it on: ``KVM_SET_MP_STATE`` returns ``-EPERM`` for ``KVM_MP_STATE_RUNNABLE``
+and ``KVM_MP_STATE_SUSPENDED`` on it, accepts ``KVM_MP_STATE_STOPPED`` and
+leaves it as it is, and never takes ``KVM_MP_STATE_UNINITIALIZED`` from
+the VMM. ``KVM_MP_STATE_STOPPED`` still stops a vCPU EL2 has powered on,
+and ``KVM_MP_STATE_RUNNABLE`` resumes it, so the VMM can pause it.
+
+Other interface differences
+---------------------------
+
+* ``KVM_CHECK_EXTENSION`` reports only the arm64 capabilities that pKVM
+  supports for protected guests, and ``KVM_ENABLE_CAP`` accepts only
+  those. Query them on the VM file descriptor: the system file descriptor
+  has no VM to filter against. The capabilities the generic KVM code
+  answers itself (``KVM_CAP_SYNC_MMU`` or ``KVM_CAP_DEVICE_CTRL``, for
+  example) are reported as for any VM, and don't mean the feature works
+  for a protected one. The filter does not cover every interface either:
+  device-fd configuration (for example the VGIC after
+  ``KVM_CREATE_DEVICE``) and vCPU attributes are unfiltered, so a feature
+  the filter excludes is rejected by the attribute itself.
+  ``KVM_ARM_VCPU_PVTIME_CTRL`` returns ``-ENXIO``, for example, as it does
+  where steal time isn't implemented.
+* The vGIC of a protected VM remains host-managed: device creation,
+  configuration and interrupt injection all work as they do for a
+  non-protected VM.
+* ``KVM_ARM_SET_COUNTER_OFFSET`` and ``KVM_ARM_GET_REG_WRITABLE_MASKS``
+  return ``-EINVAL``: their capabilities are not advertised to a protected
+  VM, whose counter offset and ID registers are set by EL2. A protected
+  guest sees the physical timebase.
+* A protected guest's first access to each page of memory exits to the
+  host, since the hypervisor cannot tell memory from a device before the
+  page is mapped. For a store, the host sees the value of the register
+  the syndrome names, clamped to the access width, and nothing else from
+  the register file. For a load, the register the syndrome names takes
+  the value the host returns, as for any emulated access.
+* A protected guest that uses a feature it was not given, or executes an
+  ``SMC``, takes an undefined instruction exception from the hypervisor;
+  the host is not involved.
+* Apart from PSCI, the hypervisor handles a protected guest's SMCCC
+  calls itself and does not involve the host. A function it does not
+  implement returns ``NOT_SUPPORTED``, and the host's SMCCC filter does
+  not apply: the VM device attribute ioctls (``KVM_ARM_VM_SMCCC_CTRL``)
+  return ``-EINVAL`` for a protected VM.
+* The hypervisor can decode a trapped guest access only from the CPU's
+  instruction syndrome, which is provided only for a load or store of a
+  single general-purpose register. An access without one (for example a
+  load/store pair or a SIMD/FP access) cannot be decoded. For a
+  non-protected VM it can exit to the VMM as ``KVM_EXIT_ARM_NISV``. For a
+  protected VM it cannot be emulated by the VMM, so the guest takes a
+  synchronous external abort instead.
+
+Debug
+-----
+
+Hardware-assisted debugging is not available to protected guests.
+``MDSCR_EL1``, ``OSLAR_EL1``, ``OSLSR_EL1``, ``OSDLR_EL1``, and breakpoint
+and watchpoint 0 are RAZ/WI; any other debug register access takes an
+undefined instruction exception.
+
 Resources
 =========
 
diff --git a/Documentation/virt/kvm/devices/vcpu.rst b/Documentation/virt/kvm/devices/vcpu.rst
index deb5c51bc00c8..916407835acdf 100644
--- a/Documentation/virt/kvm/devices/vcpu.rst
+++ b/Documentation/virt/kvm/devices/vcpu.rst
@@ -231,7 +231,8 @@ Returns:
 Specifies the base address of the stolen time structure for this VCPU. The
 base address must be 64 byte aligned and exist within a valid guest memory
 region. See Documentation/virt/kvm/arm/pvtime.rst for more information
-including the layout of the stolen time structure.
+including the layout of the stolen time structure. Stolen time is not
+supported for protected VMs (see Documentation/virt/kvm/arm/pkvm.rst).
 
 4. GROUP: KVM_VCPU_TSC_CTRL
 ===========================
-- 
2.39.5

Re: [PATCH v4 00/18] KVM: arm64: Confine protected VM vCPU state to EL2

[Fuad Tabba]

Hi Marc,

Here's a link to the series based on kvmarm/next, conflicts resolved:

https://android-kvm.googlesource.com/linux/+/refs/heads/tabba/pkvm-core-b-v4-kvmarm-next

Cheers,
/fuad

On Thu, 1 Oct 2026 at 14:57, Fuad Tabba <fuad.tabba@linux.dev> wrote:
>
> Hi folks,
>
> Changes since v3 [1]:
>   - A protected vCPU that EL2 holds powered off reads as
>     KVM_MP_STATE_UNINITIALIZED, in place of a new bool. (Marc)
>   - The marshalling and PSCI patches access the host copy through the
>     vCPU accessors, with every assignment on one line. (Marc)
>   - The PC adjustment pair is kvm_adjust_pc_get()/put() and tests for
>     the hyp vCPU directly. (Marc)
>   - CPU_OFF clears the reset flag before its release of OFF, CPU_ON's
>     relaxed cmpxchg is ordered by its control dependency, and the
>     CPU_ON rollback stores OFF with a release. pkvm_reset_vcpu()
>     carries the CPU_ON/CPU_OFF sequence as a comment and drops its
>     WARN_ON(). (Vincent, Will)
>   - Collected Reviewed-bys.
>
> Following the vCPU state-sync series [3], this series completes the
> job for protected VMs: a protected guest's register state stays at
> EL2, and the host sees only what handling each exit requires.
>
> EL2 marshals a protected vCPU's state per exception class instead of
> copying the whole context both ways. It owns the vCPU's trap
> configuration, system register reset and HVC handling, and implements
> PSCI itself: AFFINITY_INFO is left to the host, and CPU_ON and CPU_OFF
> are resolved at EL2 with the host only scheduling the target or
> stopping it. A protected guest's TRNG calls, which the host handled
> until now, return NOT_SUPPORTED until TRNG for protected guests
> follows. EL2 implements PSCI 1.1, so a protected guest also loses the
> functions the host supports above that version, SYSTEM_OFF2 included.
> Host ioctls that would reach the state EL2 owns fail with -EPERM, so a
> protected VM's state isn't save/restorable. All of this is scoped
> to KVM_VM_TYPE_ARM_PROTECTED, and pkvm.rst describes the resulting API.
>
> The kvmtool changes that go with this are posted separately [4].
>
> Patches 1 to 3 go out separately: the HCR_EL2.VSE fix [5], and the
> host vCPU VM read and pin fixes, patches 4 and 5 of the host hypercall
> fixes series [2]. None of the three is part of this series. They're
> carried so the series applies as is and Sashiko can run on it.
>
> The KVM_ARM_PREFERRED_TARGET documentation fix [6] went out just ahead
> of v1. Nothing here depends on it to apply, but patch 18 documents vCPU
> feature availability as something the capabilities report, while
> api.rst 4.83 still points userspace at a bitmap that has always been
> empty.
>
> The series is structured as follows:
>
>   01:     The HCR_EL2.VSE fix, posted separately.
>   02-03:  The host vCPU VM read and pin fixes, posted separately.
>   04:     Steal time disabled for protected VMs.
>   05-06:  Per-exception-class entry handlers; EL2 owns a protected
>           vCPU's trap configuration.
>   07-09:  Timer state, system register reset and HVC handling at EL2.
>   10-11:  PSCI at EL2, and the KVM_ARM_VCPU_INIT and PSCI version
>           restrictions.
>   12-14:  Host PC adjustments blocked; an UNDEF at EL2 for exit
>           classes the host doesn't emulate; per-class state
>           marshalling.
>   15-16:  Host changes to private state, and host power-on of a vCPU
>           EL2 holds powered off, rejected.
>   17:     Capability allowlist.
>   18:     Documentation.
>
> Still to come: selftests, TRNG, self-hosted debug and SVE for protected
> guests, and much more, as separate series.
>
> Based on v7.3-rc3 (fd73f4a665989).
>
> Cheers,
> /fuad
>
> [1] https://lore.kernel.org/all/20260914113338.159227-1-fuad.tabba@linux.dev/
> [2] https://lore.kernel.org/all/20260915123846.2317931-1-fuad.tabba@linux.dev/
> [3] https://lore.kernel.org/all/20260729131823.2021516-1-fuad.tabba@linux.dev/
> [4] https://lore.kernel.org/all/20260831192406.1341841-1-fuad.tabba@linux.dev/
> [5] https://lore.kernel.org/all/20260829071120.2522788-1-fuad.tabba@linux.dev/
> [6] https://lore.kernel.org/all/20260831162815.269851-1-fuad.tabba@linux.dev/
>
> Fuad Tabba (16):
>   KVM: arm64: Sync HCR_EL2.VSE back to the host vCPU under pKVM
>   KVM: arm64: Validate the host vCPU's VM before reading it under pKVM
>   KVM: arm64: Pin the host vCPU before adjusting its PC under pKVM
>   KVM: arm64: Disable steal time for protected VMs
>   KVM: arm64: Skip fixed-feature state flush for protected vCPUs
>   KVM: arm64: Add system register reset framework for protected VMs
>   KVM: arm64: Implement HVC handling for protected guests at EL2
>   KVM: arm64: Handle PSCI calls for protected VMs at EL2
>   KVM: arm64: Restrict KVM_ARM_VCPU_INIT and PSCI version for protected
>     VMs
>   KVM: arm64: Prevent host PC adjustments for protected vCPUs
>   KVM: arm64: Inject an UNDEF at EL2 for unhandled protected guest exits
>   KVM: arm64: Add per-EC entry/exit state marshalling for protected
>     guests
>   KVM: arm64: Reject host access to protected VM private state
>   KVM: arm64: Reject host power-on of a vCPU that EL2 holds powered off
>   KVM: arm64: Advertise the capabilities that protected VMs support
>   KVM: arm64: Document the protected VM userspace API
>
> Marc Zyngier (2):
>   KVM: arm64: Introduce per-EC entry handlers for pKVM
>   KVM: arm64: Add {flush,sync}_hyp_timer_state() primitives
>
>  Documentation/virt/kvm/api.rst                |  27 +-
>  .../virt/kvm/arm/fw-pseudo-registers.rst      |   2 +
>  Documentation/virt/kvm/arm/pkvm.rst           | 164 ++++-
>  Documentation/virt/kvm/devices/vcpu.rst       |   3 +-
>  arch/arm64/include/asm/kvm_host.h             |  22 +-
>  arch/arm64/include/asm/kvm_hyp.h              |   4 +
>  arch/arm64/include/asm/kvm_pkvm.h             |  33 +
>  arch/arm64/kvm/arm.c                          |  76 ++-
>  arch/arm64/kvm/guest.c                        |  11 +
>  arch/arm64/kvm/hyp/exception.c                |  26 +-
>  arch/arm64/kvm/hyp/include/hyp/adjust_pc.h    |  48 ++
>  arch/arm64/kvm/hyp/include/nvhe/pkvm.h        |  20 +
>  arch/arm64/kvm/hyp/nvhe/hyp-main.c            | 595 +++++++++++++++++-
>  arch/arm64/kvm/hyp/nvhe/pkvm.c                | 422 ++++++++++++-
>  arch/arm64/kvm/hyp/nvhe/switch.c              |  29 +-
>  arch/arm64/kvm/hyp/nvhe/sys_regs.c            |  60 ++
>  arch/arm64/kvm/hypercalls.c                   |   7 +
>  arch/arm64/kvm/pkvm.c                         |  21 +-
>  arch/arm64/kvm/psci.c                         |  11 +-
>  arch/arm64/kvm/pvtime.c                       |  10 +-
>  20 files changed, 1503 insertions(+), 88 deletions(-)
>
>
> base-commit: fd73f4a6659897191fa0d40695fe370925dd3780
> --
> 2.39.5
>