[PATCH] KVM: arm64: Ensure CPU PMU probes before pKVM host de-privilege

[Will Deacon <will@kernel.org>]

Although pKVM supports CPU PMU emulation for non-protected guests since
722625c6f4c5 ("KVM: arm64: Reenable pmu in Protected Mode"), this relies
on the PMU driver probing before the host has de-privileged so that the
'kvm_arm_pmu_available' static key can still be enabled by patching the
hypervisor text.

As it happens, both of these events hang off device_initcall() but the
PMU consistently won the race until 7755cec63ade ("arm64: perf: Move
PMUv3 driver to drivers/perf"). Since then, the host will fail to boot
when pKVM is enabled:

  | hw perfevents: enabled with armv8_pmuv3_0 PMU driver, 7 counters available
  | kvm [1]: nVHE hyp BUG at: [<ffff8000090366e0>] __kvm_nvhe_handle_host_mem_abort+0x270/0x284!
  | kvm [1]: Cannot dump pKVM nVHE stacktrace: !CONFIG_PROTECTED_NVHE_STACKTRACE
  | kvm [1]: Hyp Offset: 0xfffea41fbdf70000
  | Kernel panic - not syncing: HYP panic:
  | PS:a00003c9 PC:0000dbe04b0c66e0 ESR:00000000f2000800
  | FAR:fffffbfffddfcf00 HPFAR:00000000010b0bf0 PAR:0000000000000000
  | VCPU:0000000000000000
  | CPU: 2 PID: 1 Comm: swapper/0 Not tainted 6.3.0-rc7-00083-g0bce6746d154 #1
  | Hardware name: QEMU QEMU Virtual Machine, BIOS 0.0.0 02/06/2015
  | Call trace:
  |  dump_backtrace+0xec/0x108
  |  show_stack+0x18/0x2c
  |  dump_stack_lvl+0x50/0x68
  |  dump_stack+0x18/0x24
  |  panic+0x13c/0x33c
  |  nvhe_hyp_panic_handler+0x10c/0x190
  |  aarch64_insn_patch_text_nosync+0x64/0xc8
  |  arch_jump_label_transform+0x4c/0x5c
  |  __jump_label_update+0x84/0xfc
  |  jump_label_update+0x100/0x134
  |  static_key_enable_cpuslocked+0x68/0xac
  |  static_key_enable+0x20/0x34
  |  kvm_host_pmu_init+0x88/0xa4
  |  armpmu_register+0xf0/0xf4
  |  arm_pmu_acpi_probe+0x2ec/0x368
  |  armv8_pmu_driver_init+0x38/0x44
  |  do_one_initcall+0xcc/0x240

Fix the race properly by deferring the de-privilege step to
device_initcall_sync(). This will also be needed in future when probing
IOMMU devices and allows us to separate the pKVM de-privilege logic from
the core hypervisor initialisation path.

Cc: Oliver Upton <oliver.upton@linux.dev>
Cc: Fuad Tabba <tabba@google.com>
Cc: Marc Zyngier <maz@kernel.org>
Fixes: 7755cec63ade ("arm64: perf: Move PMUv3 driver to drivers/perf")
Signed-off-by: Will Deacon <will@kernel.org>
---

Marc, Oliver -- in practice, this issue only crops with the patches
moving the CPU PMU driver out into drivers/perf/ and so the arm64
for-next/core branch is broken. Please can I queue this in the arm64
tree for 6.4 with your Ack? Thanks.

 arch/arm64/kvm/arm.c  | 45 -----------------------------------------
 arch/arm64/kvm/pkvm.c | 47 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 47 insertions(+), 45 deletions(-)

diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index 3bd732eaf087..890f730bc3ab 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -16,7 +16,6 @@
 #include <linux/fs.h>
 #include <linux/mman.h>
 #include <linux/sched.h>
-#include <linux/kmemleak.h>
 #include <linux/kvm.h>
 #include <linux/kvm_irqfd.h>
 #include <linux/irqbypass.h>
@@ -46,7 +45,6 @@
 #include <kvm/arm_psci.h>
 
 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
-DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
 
 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
 
@@ -2105,41 +2103,6 @@ static int __init init_hyp_mode(void)
 	return err;
 }
 
-static void __init _kvm_host_prot_finalize(void *arg)
-{
-	int *err = arg;
-
-	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
-		WRITE_ONCE(*err, -EINVAL);
-}
-
-static int __init pkvm_drop_host_privileges(void)
-{
-	int ret = 0;
-
-	/*
-	 * Flip the static key upfront as that may no longer be possible
-	 * once the host stage 2 is installed.
-	 */
-	static_branch_enable(&kvm_protected_mode_initialized);
-	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
-	return ret;
-}
-
-static int __init finalize_hyp_mode(void)
-{
-	if (!is_protected_kvm_enabled())
-		return 0;
-
-	/*
-	 * Exclude HYP sections from kmemleak so that they don't get peeked
-	 * at, which would end badly once inaccessible.
-	 */
-	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
-	kmemleak_free_part_phys(hyp_mem_base, hyp_mem_size);
-	return pkvm_drop_host_privileges();
-}
-
 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
 {
 	struct kvm_vcpu *vcpu;
@@ -2257,14 +2220,6 @@ static __init int kvm_arm_init(void)
 	if (err)
 		goto out_hyp;
 
-	if (!in_hyp_mode) {
-		err = finalize_hyp_mode();
-		if (err) {
-			kvm_err("Failed to finalize Hyp protection\n");
-			goto out_subs;
-		}
-	}
-
 	if (is_protected_kvm_enabled()) {
 		kvm_info("Protected nVHE mode initialized successfully\n");
 	} else if (in_hyp_mode) {
diff --git a/arch/arm64/kvm/pkvm.c b/arch/arm64/kvm/pkvm.c
index cf56958b1492..6e9ece1ebbe7 100644
--- a/arch/arm64/kvm/pkvm.c
+++ b/arch/arm64/kvm/pkvm.c
@@ -4,6 +4,8 @@
  * Author: Quentin Perret <qperret@google.com>
  */
 
+#include <linux/init.h>
+#include <linux/kmemleak.h>
 #include <linux/kvm_host.h>
 #include <linux/memblock.h>
 #include <linux/mutex.h>
@@ -13,6 +15,8 @@
 
 #include "hyp_constants.h"
 
+DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
+
 static struct memblock_region *hyp_memory = kvm_nvhe_sym(hyp_memory);
 static unsigned int *hyp_memblock_nr_ptr = &kvm_nvhe_sym(hyp_memblock_nr);
 
@@ -213,3 +217,46 @@ int pkvm_init_host_vm(struct kvm *host_kvm)
 	mutex_init(&host_kvm->lock);
 	return 0;
 }
+
+static void __init _kvm_host_prot_finalize(void *arg)
+{
+	int *err = arg;
+
+	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
+		WRITE_ONCE(*err, -EINVAL);
+}
+
+static int __init pkvm_drop_host_privileges(void)
+{
+	int ret = 0;
+
+	/*
+	 * Flip the static key upfront as that may no longer be possible
+	 * once the host stage 2 is installed.
+	 */
+	static_branch_enable(&kvm_protected_mode_initialized);
+	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
+	return ret;
+}
+
+static int __init finalize_pkvm(void)
+{
+	int ret;
+
+	if (!is_protected_kvm_enabled())
+		return 0;
+
+	/*
+	 * Exclude HYP sections from kmemleak so that they don't get peeked
+	 * at, which would end badly once inaccessible.
+	 */
+	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
+	kmemleak_free_part_phys(hyp_mem_base, hyp_mem_size);
+
+	ret = pkvm_drop_host_privileges();
+	if (ret)
+		pr_err("Failed to finalize Hyp protection: %d\n", ret);
+
+	return ret;
+}
+device_initcall_sync(finalize_pkvm);
-- 
2.40.0.634.g4ca3ef3211-goog