Skip to contentSkip to conversation

HVF: Add support for platform vGIC and nested virtualisation

PATCH Superseded Mohamed Mediouni 2 participants 27 messages Feb 28 – Mar 6 Arm/Amlogic Meson SoC support

v11 of 21 Superseded Latest is v21 →

lore
Share & tools
lore
#0 lore raw

v11 cover HVF: Add support for platform vGIC and nested virtualisation

Link to branch: https://github.com/mediouni-m/qemu hvf-irqchip-and-nested (tag for this submission: hvf-irqchip-and-nested-v11)

This series adds supports for nested virtualisation when using HVF on arm64 Macs.

Reviews would be very welcome :-) It's maybe a bit unlikely that this will make it to QEMU 11.0 I might as well try.

It has two parts:
- Apple vGICv3 support and necessary infrastructure changes for it
- Nested virtualisation support. Note that the nested virtualisation implementation shipping as of macOS 26.3 is nVHE only (but it _does_ use VNCR as shipped).

This series doesn't contain EL2 physical timer emulation, which is
needed if not leveraging the Apple vGIC. As such, nested virt requires kernel-irqchip=on.

Known issues:
- when nested virt is enabled, no UI response within EDK2 and a permanent wait. Workaround: -boot menu=on,splash-time=0.

Apple Feedback Assistant item: FB21649319

When the VM is running at EL2 at the very moment the virtual timer fires:
HV_EXIT_REASON_VTIMER_ACTIVATED doesn’t fire (when not using the provided vGIC)
- using a GICv2 doesn’t require having transition notifiers which Hypervisor.framework doesn’t have…)

and when using the provided vGIC, the interrupt never gets delivered back to the guest

Linux as a guest OS is fine with this… but the reference ArmVirtQemu edk2 build always uses the virtual timer even when running EFI at EL2, so it gets broken unless this patch is applied to edk2:

diff of ArmVirtPkg/ArmVirt.dsc.inc:

-  ArmGenericTimerCounterLib|ArmPkg/Library/ArmGenericTimerVirtCounterLib/ArmGenericTimerVirtCounterLib.inf
+  ArmGenericTimerCounterLib|ArmPkg/Library/ArmGenericTimerPhyCounterLib/ArmGenericTimerPhyCounterLib.inf^M

Changelog:

v1->v2:
Oops. I did a mistake when preparing my patches.

- Add hvf_arm_el2_enable(_) call to virt_set_virt
- Fix nested virt support check to add HVF

v2->v3:
- LORC_EL1 patch was merged separately, remove from this series.
- fix LPIs when kernel-irqchip disabled and using TCG
- remove spurious if case in vGIC supported version detection (inapplicable now)
- Add hvf_enabled() check in combination with hvf kernel-irqchip checks
- cleanly fail on attempt to use the platform vGIC together with ITS

v3->v4:
- GIC state save improvements, including saving the opaque Apple-specific state
- Saving HVF system register state when using the vGIC and/or EL2

v5:
- oops, fixed up save/restore to be functional
- misc changes otherwise

v6:
- Addressing review comments

v7:
- Address review comments, adapt around Qemu changes and bugfixes.

v8:
- Rebase, and misc fixes

v9:
- Misc fixes from WHPX review comments
- Split enablement commit for EL2 into its own separate commit

v10:

- A bit of changes across the board
- move HVF irqchip save/restore into irqchip code
- block nested virt support when kernel-irqchip=off with a GICv2 as we don't implement an emulated physical timer yet.

v11:
- fixups for the save/restore Apple GIC state patch
- block the same registers as the revert in 28b0ed32b32c7e5094cf2f1ec9c0645c65fad2aa for "hvf: sync registers used at EL2"
- drop patch that warns instead of aborts on mismatch

Mohamed Mediouni (12):
hw/intc: Add hvf vGIC interrupt controller support hw/intc: arm_gicv3_hvf: save/restore Apple GIC state accel, hw/arm, include/system/hvf: infrastructure changes for HVF vGIC target/arm: hvf: instantiate GIC early
hw/arm, target/arm: nested virtualisation on HVF
hvf: only call hvf_sync_vtimer() when running without the platform
vGIC
hvf: gate ARM_FEATURE_PMU register emulation behind not being at EL2
hvf: arm: allow exposing minimal PMU when running with nested virt on target/arm: hvf: add asserts for code paths not leveraged when using
the vGIC
hvf: sync registers used at EL2
target/arm: hvf: pass through CNTHCTL_EL2 and MDCCINT_EL1
hvf: enable nested virtualisation support

 accel/hvf/hvf-all.c                |  51 ++
 accel/stubs/hvf-stub.c             |   2 +
 hw/arm/virt.c                      |  37 +-
 hw/intc/arm_gicv3_common.c         |   4 +
 hw/intc/arm_gicv3_hvf.c            | 822 +++++++++++++++++++++++++++++
 hw/intc/arm_gicv3_hvf_stub.c       |  25 +
 hw/intc/meson.build                |   2 +
 include/hw/intc/arm_gicv3_common.h |   4 +
 include/system/hvf.h               |   8 +
 system/vl.c                        |   2 +
 target/arm/hvf/hvf.c               | 124 ++++-
 target/arm/hvf/sysreg.c.inc        |  45 ++
 12 files changed, 1112 insertions(+), 14 deletions(-)
 create mode 100644 hw/intc/arm_gicv3_hvf.c
 create mode 100644 hw/intc/arm_gicv3_hvf_stub.c
Signature · 1 line
2.50.1 (Apple Git-155)
#1 lore raw

PATCH v11 1/12 hw/intc: Add hvf vGIC interrupt controller support

This opens up the door to nested virtualisation support.

---
 hw/intc/arm_gicv3_hvf.c            | 735 +++++++++++++++++++++++++++++
 hw/intc/meson.build                |   1 +
 include/hw/intc/arm_gicv3_common.h |   1 +
 3 files changed, 737 insertions(+)
 create mode 100644 hw/intc/arm_gicv3_hvf.c
hw/intc/arm_gicv3_hvf.c new file +735 -0
new file mode 100644
index 0000000000..d6a46b7d53
@@ -0,0 +1,735 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * ARM Generic Interrupt Controller using HVF platform support
+ *
+ * Copyright (c) 2025 Mohamed Mediouni
+ * Based on vGICv3 KVM code by Pavel Fedin
+ *
+ */
+
+#include "qemu/osdep.h"
+#include "qapi/error.h"
+#include "hw/intc/arm_gicv3_common.h"
+#include "qemu/error-report.h"
+#include "qemu/module.h"
+#include "system/runstate.h"
+#include "system/hvf.h"
+#include "system/hvf_int.h"
+#include "hvf_arm.h"
+#include "gicv3_internal.h"
+#include "vgic_common.h"
+#include "qom/object.h"
+#include "target/arm/cpregs.h"
+#include <Hypervisor/Hypervisor.h>
+
+struct HVFARMGICv3Class {
+    ARMGICv3CommonClass parent_class;
+    DeviceRealize parent_realize;
+    ResettablePhases parent_phases;
+};
+
+typedef struct HVFARMGICv3Class HVFARMGICv3Class;
+
+/* This is reusing the GICv3State typedef from ARM_GICV3_ITS_COMMON */
+DECLARE_OBJ_CHECKERS(GICv3State, HVFARMGICv3Class,
+                     HVF_GICV3, TYPE_HVF_GICV3);
+
+/*
+ * Loop through each distributor IRQ related register; since bits
+ * corresponding to SPIs and PPIs are RAZ/WI when affinity routing
+ * is enabled, we skip those.
+ */
+#define for_each_dist_irq_reg(_irq, _max, _field_width) \
+    for (_irq = GIC_INTERNAL; _irq < _max; _irq += (32 / _field_width))
+
+/*
+ * Wrap calls to the vGIC APIs to assert_hvf_ok()
+ * as a macro to keep the code clean.
+ */
+#define hv_gic_get_distributor_reg(offset, reg) \
+    assert_hvf_ok(hv_gic_get_distributor_reg(offset, reg))
+
+#define hv_gic_set_distributor_reg(offset, reg) \
+    assert_hvf_ok(hv_gic_set_distributor_reg(offset, reg))
+
+#define hv_gic_get_redistributor_reg(vcpu, reg, value) \
+    assert_hvf_ok(hv_gic_get_redistributor_reg(vcpu, reg, value))
+
+#define hv_gic_set_redistributor_reg(vcpu, reg, value) \
+    assert_hvf_ok(hv_gic_set_redistributor_reg(vcpu, reg, value))
+
+#define hv_gic_get_icc_reg(vcpu, reg, value) \
+    assert_hvf_ok(hv_gic_get_icc_reg(vcpu, reg, value))
+
+#define hv_gic_set_icc_reg(vcpu, reg, value) \
+    assert_hvf_ok(hv_gic_set_icc_reg(vcpu, reg, value))
+
+#define hv_gic_get_ich_reg(vcpu, reg, value) \
+    assert_hvf_ok(hv_gic_get_ich_reg(vcpu, reg, value))
+
+#define hv_gic_set_ich_reg(vcpu, reg, value) \
+    assert_hvf_ok(hv_gic_set_ich_reg(vcpu, reg, value))
+
+static void hvf_dist_get_priority(GICv3State *s, hv_gic_distributor_reg_t offset
+    , uint8_t *bmp)
+{
+    uint64_t reg;
+    uint32_t *field;
+    int irq;
+    field = (uint32_t *)(bmp);
+
+    for_each_dist_irq_reg(irq, s->num_irq, 8) {
+        hv_gic_get_distributor_reg(offset, &reg);
+        *field = reg;
+        offset += 4;
+        field++;
+    }
+}
+
+static void hvf_dist_put_priority(GICv3State *s, hv_gic_distributor_reg_t offset
+    , uint8_t *bmp)
+{
+    uint32_t reg, *field;
+    int irq;
+    field = (uint32_t *)(bmp);
+
+    for_each_dist_irq_reg(irq, s->num_irq, 8) {
+        reg = *field;
+        hv_gic_set_distributor_reg(offset, reg);
+        offset += 4;
+        field++;
+    }
+}
+
+static void hvf_dist_get_edge_trigger(GICv3State *s, hv_gic_distributor_reg_t offset,
+                                      uint32_t *bmp)
+{
+    uint64_t reg;
+    int irq;
+
+    for_each_dist_irq_reg(irq, s->num_irq, 2) {
+        hv_gic_get_distributor_reg(offset, &reg);
+        reg = half_unshuffle32(reg >> 1);
+        if (irq % 32 != 0) {
+            reg = (reg << 16);
+        }
+        *gic_bmp_ptr32(bmp, irq) |= reg;
+        offset += 4;
+    }
+}
+
+static void hvf_dist_put_edge_trigger(GICv3State *s, hv_gic_distributor_reg_t offset,
+                                      uint32_t *bmp)
+{
+    uint32_t reg;
+    int irq;
+
+    for_each_dist_irq_reg(irq, s->num_irq, 2) {
+        reg = *gic_bmp_ptr32(bmp, irq);
+        if (irq % 32 != 0) {
+            reg = (reg & 0xffff0000) >> 16;
+        } else {
+            reg = reg & 0xffff;
+        }
+        reg = half_shuffle32(reg) << 1;
+        hv_gic_set_distributor_reg(offset, reg);
+        offset += 4;
+    }
+}
+
+/* Read a bitmap register group from the kernel VGIC. */
+static void hvf_dist_getbmp(GICv3State *s, hv_gic_distributor_reg_t offset, uint32_t *bmp)
+{
+    uint64_t reg;
+    int irq;
+
+    for_each_dist_irq_reg(irq, s->num_irq, 1) {
+
+        hv_gic_get_distributor_reg(offset, &reg);
+        *gic_bmp_ptr32(bmp, irq) = reg;
+        offset += 4;
+    }
+}
+
+static void hvf_dist_putbmp(GICv3State *s, hv_gic_distributor_reg_t offset,
+                            hv_gic_distributor_reg_t clroffset, uint32_t *bmp)
+{
+    uint32_t reg;
+    int irq;
+
+    for_each_dist_irq_reg(irq, s->num_irq, 1) {
+        /*
+         * If this bitmap is a set/clear register pair, first write to the
+         * clear-reg to clear all bits before using the set-reg to write
+         * the 1 bits.
+         */
+        if (clroffset != 0) {
+            reg = 0;
+            hv_gic_set_distributor_reg(clroffset, reg);
+            clroffset += 4;
+        }
+        reg = *gic_bmp_ptr32(bmp, irq);
+        hv_gic_set_distributor_reg(offset, reg);
+        offset += 4;
+    }
+}
+
+static void hvf_gicv3_check(GICv3State *s)
+{
+    uint64_t reg;
+    uint32_t num_irq;
+
+    /* Sanity checking s->num_irq */
+    hv_gic_get_distributor_reg(HV_GIC_DISTRIBUTOR_REG_GICD_TYPER, &reg);
+    num_irq = ((reg & 0x1f) + 1) * 32;
+
+    if (num_irq < s->num_irq) {
+        error_report("Model requests %u IRQs, but HVF supports max %u",
+                     s->num_irq, num_irq);
+        abort();
+    }
+}
+
+static void hvf_gicv3_put_cpu_el2(CPUState *cpu_state, run_on_cpu_data arg)
+{
+    int num_pri_bits;
+
+    /* Redistributor state */
+    GICv3CPUState *c = arg.host_ptr;
+    hv_vcpu_t vcpu = c->cpu->accel->fd;
+
+    hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_VMCR_EL2, c->ich_vmcr_el2);
+    hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_HCR_EL2, c->ich_hcr_el2);
+
+    for (int i = 0; i < GICV3_LR_MAX; i++) {
+        hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_LR0_EL2, c->ich_lr_el2[i]);
+    }
+
+    num_pri_bits = c->vpribits;
+
+    switch (num_pri_bits) {
+    case 7:
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2 + 3,
+                         c->ich_apr[GICV3_G0][3]);
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2 + 2,
+                         c->ich_apr[GICV3_G0][2]);
+      /* fall through */
+    case 6:
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2 + 1,
+                         c->ich_apr[GICV3_G0][1]);
+      /* fall through */
+    default:
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2,
+                         c->ich_apr[GICV3_G0][0]);
+    }
+
+    switch (num_pri_bits) {
+    case 7:
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2 + 3,
+                         c->ich_apr[GICV3_G1NS][3]);
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2 + 2,
+                         c->ich_apr[GICV3_G1NS][2]);
+      /* fall through */
+    case 6:
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2 + 1,
+                         c->ich_apr[GICV3_G1NS][1]);
+      /* fall through */
+    default:
+      hv_gic_set_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2,
+                         c->ich_apr[GICV3_G1NS][0]);
+    }
+}
+
+static void hvf_gicv3_put_cpu(CPUState *cpu_state, run_on_cpu_data arg)
+{
+    uint32_t reg;
+    uint64_t reg64;
+    int i, num_pri_bits;
+
+    /* Redistributor state */
+    GICv3CPUState *c = arg.host_ptr;
+    hv_vcpu_t vcpu = c->cpu->accel->fd;
+
+    reg = c->gicr_waker;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_IGROUPR0, reg);
+
+    reg = c->gicr_igroupr0;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_IGROUPR0, reg);
+
+    reg = ~0;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ICENABLER0, reg);
+    reg = c->gicr_ienabler0;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ISENABLER0, reg);
+
+    /* Restore config before pending so we treat level/edge correctly */
+    reg = half_shuffle32(c->edge_trigger >> 16) << 1;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ICFGR1, reg);
+
+    reg = ~0;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ICPENDR0, reg);
+    reg = c->gicr_ipendr0;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ISPENDR0, reg);
+
+    reg = ~0;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ICACTIVER0, reg);
+    reg = c->gicr_iactiver0;
+    hv_gic_set_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ISACTIVER0, reg);
+
+    for (i = 0; i < GIC_INTERNAL; i += 4) {
+        reg = c->gicr_ipriorityr[i] |
+            (c->gicr_ipriorityr[i + 1] << 8) |
+            (c->gicr_ipriorityr[i + 2] << 16) |
+            (c->gicr_ipriorityr[i + 3] << 24);
+        hv_gic_set_redistributor_reg(vcpu,
+            HV_GIC_REDISTRIBUTOR_REG_GICR_IPRIORITYR0 + i, reg);
+    }
+
+    /* CPU interface state */
+    hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_SRE_EL1, c->icc_sre_el1);
+
+    hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_CTLR_EL1,
+                    c->icc_ctlr_el1[GICV3_NS]);
+    hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_IGRPEN0_EL1,
+                    c->icc_igrpen[GICV3_G0]);
+    hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_IGRPEN1_EL1,
+                    c->icc_igrpen[GICV3_G1NS]);
+    hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_PMR_EL1, c->icc_pmr_el1);
+    hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_BPR0_EL1, c->icc_bpr[GICV3_G0]);
+    hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_BPR1_EL1, c->icc_bpr[GICV3_G1NS]);
+
+    num_pri_bits = ((c->icc_ctlr_el1[GICV3_NS] &
+                    ICC_CTLR_EL1_PRIBITS_MASK) >>
+                    ICC_CTLR_EL1_PRIBITS_SHIFT) + 1;
+
+    switch (num_pri_bits) {
+    case 7:
+           reg64 = c->icc_apr[GICV3_G0][3];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1 + 3, reg64);
+        reg64 = c->icc_apr[GICV3_G0][2];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1 + 2, reg64);
+        /* fall through */
+    case 6:
+        reg64 = c->icc_apr[GICV3_G0][1];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1 + 1, reg64);
+        /* fall through */
+    default:
+        reg64 = c->icc_apr[GICV3_G0][0];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1, reg64);
+    }
+
+    switch (num_pri_bits) {
+    case 7:
+        reg64 = c->icc_apr[GICV3_G1NS][3];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1 + 3, reg64);
+        reg64 = c->icc_apr[GICV3_G1NS][2];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1 + 2, reg64);
+        /* fall through */
+    case 6:
+        reg64 = c->icc_apr[GICV3_G1NS][1];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1 + 1, reg64);
+        /* fall through */
+    default:
+        reg64 = c->icc_apr[GICV3_G1NS][0];
+        hv_gic_set_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1, reg64);
+    }
+
+    /* Registers beyond this point are with nested virt only */
+    if (c->gic->maint_irq) {
+        hvf_gicv3_put_cpu_el2(cpu_state, arg);
+    }
+}
+
+static void hvf_gicv3_put(GICv3State *s)
+{
+    uint32_t reg;
+    int ncpu, i;
+
+    hvf_gicv3_check(s);
+
+    reg = s->gicd_ctlr;
+    hv_gic_set_distributor_reg(HV_GIC_DISTRIBUTOR_REG_GICD_CTLR, reg);
+
+    /* per-CPU state */
+
+    for (ncpu = 0; ncpu < s->num_cpu; ncpu++) {
+        run_on_cpu_data data;
+        data.host_ptr = &s->cpu[ncpu];
+        run_on_cpu(s->cpu[ncpu].cpu, hvf_gicv3_put_cpu, data);
+    }
+
+    /* s->enable bitmap -> GICD_ISENABLERn */
+    hvf_dist_putbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_ISENABLER0
+        , HV_GIC_DISTRIBUTOR_REG_GICD_ICENABLER0, s->enabled);
+
+    /* s->group bitmap -> GICD_IGROUPRn */
+    hvf_dist_putbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_IGROUPR0
+        , 0, s->group);
+
+    /* Restore targets before pending to ensure the pending state is set on
+     * the appropriate CPU interfaces in the kernel
+     */
+
+    /* s->gicd_irouter[irq] -> GICD_IROUTERn */
+    for (i = GIC_INTERNAL; i < s->num_irq; i++) {
+        uint32_t offset = HV_GIC_DISTRIBUTOR_REG_GICD_IROUTER32 + (8 * i)
+            - (8 * GIC_INTERNAL);
+        hv_gic_set_distributor_reg(offset, s->gicd_irouter[i]);
+    }
+
+    /*
+     * s->trigger bitmap -> GICD_ICFGRn
+     * (restore configuration registers before pending IRQs so we treat
+     * level/edge correctly)
+     */
+    hvf_dist_put_edge_trigger(s, HV_GIC_DISTRIBUTOR_REG_GICD_ICFGR0, s->edge_trigger);
+
+    /* s->pending bitmap -> GICD_ISPENDRn */
+    hvf_dist_putbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_ISPENDR0,
+        HV_GIC_DISTRIBUTOR_REG_GICD_ICPENDR0, s->pending);
+
+    /* s->active bitmap -> GICD_ISACTIVERn */
+    hvf_dist_putbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_ISACTIVER0,
+        HV_GIC_DISTRIBUTOR_REG_GICD_ICACTIVER0, s->active);
+
+    /* s->gicd_ipriority[] -> GICD_IPRIORITYRn */
+    hvf_dist_put_priority(s, HV_GIC_DISTRIBUTOR_REG_GICD_IPRIORITYR0, s->gicd_ipriority);
+}
+
+static void hvf_gicv3_get_cpu_el2(CPUState *cpu_state, run_on_cpu_data arg)
+{
+    int num_pri_bits;
+
+    /* Redistributor state */
+    GICv3CPUState *c = arg.host_ptr;
+    hv_vcpu_t vcpu = c->cpu->accel->fd;
+
+    hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_VMCR_EL2, &c->ich_vmcr_el2);
+    hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_HCR_EL2, &c->ich_hcr_el2);
+
+    for (int i = 0; i < GICV3_LR_MAX; i++) {
+        hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_LR0_EL2, &c->ich_lr_el2[i]);
+    }
+
+    num_pri_bits = c->vpribits;
+
+    switch (num_pri_bits) {
+    case 7:
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2 + 3,
+                         &c->ich_apr[GICV3_G0][3]);
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2 + 2,
+                         &c->ich_apr[GICV3_G0][2]);
+      /* fall through */
+    case 6:
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2 + 1,
+                         &c->ich_apr[GICV3_G0][1]);
+      /* fall through */
+    default:
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP0R0_EL2,
+                         &c->ich_apr[GICV3_G0][0]);
+    }
+
+    switch (num_pri_bits) {
+    case 7:
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2 + 3,
+                         &c->ich_apr[GICV3_G1NS][3]);
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2 + 2,
+                         &c->ich_apr[GICV3_G1NS][2]);
+      /* fall through */
+    case 6:
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2 + 1,
+                         &c->ich_apr[GICV3_G1NS][1]);
+      /* fall through */
+    default:
+      hv_gic_get_ich_reg(vcpu, HV_GIC_ICH_REG_AP1R0_EL2,
+                         &c->ich_apr[GICV3_G1NS][0]);
+    }
+}
+
+static void hvf_gicv3_get_cpu(CPUState *cpu_state, run_on_cpu_data arg)
+{
+    uint64_t reg;
+    int i, num_pri_bits;
+
+    /* Redistributor state */
+    GICv3CPUState *c = arg.host_ptr;
+    hv_vcpu_t vcpu = c->cpu->accel->fd;
+
+    hv_gic_get_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_IGROUPR0,
+                                 &reg);
+    c->gicr_igroupr0 = reg;
+    hv_gic_get_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ISENABLER0,
+                                 &reg);
+    c->gicr_ienabler0 = reg;
+    hv_gic_get_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ICFGR1,
+                                 &reg);
+    c->edge_trigger = half_unshuffle32(reg >> 1) << 16;
+    hv_gic_get_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ISPENDR0,
+                                 &reg);
+    c->gicr_ipendr0 = reg;
+    hv_gic_get_redistributor_reg(vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_ISACTIVER0,
+                                 &reg);
+    c->gicr_iactiver0 = reg;
+
+    for (i = 0; i < GIC_INTERNAL; i += 4) {
+        hv_gic_get_redistributor_reg(
+          vcpu, HV_GIC_REDISTRIBUTOR_REG_GICR_IPRIORITYR0 + i, &reg);
+        c->gicr_ipriorityr[i] = extract32(reg, 0, 8);
+        c->gicr_ipriorityr[i + 1] = extract32(reg, 8, 8);
+        c->gicr_ipriorityr[i + 2] = extract32(reg, 16, 8);
+        c->gicr_ipriorityr[i + 3] = extract32(reg, 24, 8);
+    }
+
+    /* CPU interface */
+    hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_SRE_EL1, &c->icc_sre_el1);
+
+    hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_CTLR_EL1,
+                       &c->icc_ctlr_el1[GICV3_NS]);
+    hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_IGRPEN0_EL1,
+                       &c->icc_igrpen[GICV3_G0]);
+    hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_IGRPEN1_EL1,
+                       &c->icc_igrpen[GICV3_G1NS]);
+    hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_PMR_EL1, &c->icc_pmr_el1);
+    hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_BPR0_EL1, &c->icc_bpr[GICV3_G0]);
+    hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_BPR1_EL1, &c->icc_bpr[GICV3_G1NS]);
+    num_pri_bits = ((c->icc_ctlr_el1[GICV3_NS] & ICC_CTLR_EL1_PRIBITS_MASK) >>
+                    ICC_CTLR_EL1_PRIBITS_SHIFT) +
+                   1;
+
+    switch (num_pri_bits) {
+    case 7:
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1 + 3,
+                         &c->icc_apr[GICV3_G0][3]);
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1 + 2,
+                         &c->icc_apr[GICV3_G0][2]);
+      /* fall through */
+    case 6:
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1 + 1,
+                         &c->icc_apr[GICV3_G0][1]);
+      /* fall through */
+    default:
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP0R0_EL1,
+                         &c->icc_apr[GICV3_G0][0]);
+    }
+
+    switch (num_pri_bits) {
+    case 7:
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1 + 3,
+                         &c->icc_apr[GICV3_G1NS][3]);
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1 + 2,
+                         &c->icc_apr[GICV3_G1NS][2]);
+      /* fall through */
+    case 6:
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1 + 1,
+                         &c->icc_apr[GICV3_G1NS][1]);
+      /* fall through */
+    default:
+      hv_gic_get_icc_reg(vcpu, HV_GIC_ICC_REG_AP1R0_EL1,
+                         &c->icc_apr[GICV3_G1NS][0]);
+    }
+
+    /* Registers beyond this point are with nested virt only */
+    if (c->gic->maint_irq) {
+        hvf_gicv3_get_cpu_el2(cpu_state, arg);
+    }
+}
+
+static void hvf_gicv3_get(GICv3State *s)
+{
+    uint64_t reg;
+    int ncpu, i;
+
+    hvf_gicv3_check(s);
+
+    hv_gic_get_distributor_reg(HV_GIC_DISTRIBUTOR_REG_GICD_CTLR, &reg);
+    s->gicd_ctlr = reg;
+
+    /* Redistributor state (one per CPU) */
+
+    for (ncpu = 0; ncpu < s->num_cpu; ncpu++) {
+        run_on_cpu_data data;
+        data.host_ptr = &s->cpu[ncpu];
+        run_on_cpu(s->cpu[ncpu].cpu, hvf_gicv3_get_cpu, data);
+    }
+
+    /* GICD_IGROUPRn -> s->group bitmap */
+    hvf_dist_getbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_IGROUPR0, s->group);
+
+    /* GICD_ISENABLERn -> s->enabled bitmap */
+    hvf_dist_getbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_ISENABLER0, s->enabled);
+
+    /* GICD_ISPENDRn -> s->pending bitmap */
+    hvf_dist_getbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_ISPENDR0, s->pending);
+
+    /* GICD_ISACTIVERn -> s->active bitmap */
+    hvf_dist_getbmp(s, HV_GIC_DISTRIBUTOR_REG_GICD_ISACTIVER0, s->active);
+
+    /* GICD_ICFGRn -> s->trigger bitmap */
+    hvf_dist_get_edge_trigger(s, HV_GIC_DISTRIBUTOR_REG_GICD_ICFGR0
+        , s->edge_trigger);
+
+    /* GICD_IPRIORITYRn -> s->gicd_ipriority[] */
+    hvf_dist_get_priority(s, HV_GIC_DISTRIBUTOR_REG_GICD_IPRIORITYR0
+        , s->gicd_ipriority);
+
+    /* GICD_IROUTERn -> s->gicd_irouter[irq] */
+    for (i = GIC_INTERNAL; i < s->num_irq; i++) {
+        uint32_t offset = HV_GIC_DISTRIBUTOR_REG_GICD_IROUTER32
+            + (8 * i) - (8 * GIC_INTERNAL);
+        hv_gic_get_distributor_reg(offset, &s->gicd_irouter[i]);
+    }
+}
+
+static void hvf_gicv3_set_irq(void *opaque, int irq, int level)
+{
+    GICv3State *s = opaque;
+    if (irq > s->num_irq) {
+        return;
+    }
+    hv_gic_set_spi(GIC_INTERNAL + irq, !!level);
+}
+
+static void hvf_gicv3_icc_reset(CPUARMState *env, const ARMCPRegInfo *ri)
+{
+    GICv3CPUState *c;
+
+    c = env->gicv3state;
+    c->icc_pmr_el1 = 0;
+    /*
+     * Architecturally the reset value of the ICC_BPR registers
+     * is UNKNOWN. We set them all to 0 here; when the kernel
+     * uses these values to program the ICH_VMCR_EL2 fields that
+     * determine the guest-visible ICC_BPR register values, the
+     * hardware's "writing a value less than the minimum sets
+     * the field to the minimum value" behaviour will result in
+     * them effectively resetting to the correct minimum value
+     * for the host GIC.
+     */
+    c->icc_bpr[GICV3_G0] = 0;
+    c->icc_bpr[GICV3_G1] = 0;
+    c->icc_bpr[GICV3_G1NS] = 0;
+
+    c->icc_sre_el1 = 0x7;
+    memset(c->icc_apr, 0, sizeof(c->icc_apr));
+    memset(c->icc_igrpen, 0, sizeof(c->icc_igrpen));
+}
+
+static void hvf_gicv3_reset_hold(Object *obj, ResetType type)
+{
+    GICv3State *s = ARM_GICV3_COMMON(obj);
+    HVFARMGICv3Class *kgc = HVF_GICV3_GET_CLASS(s);
+
+    if (kgc->parent_phases.hold) {
+        kgc->parent_phases.hold(obj, type);
+    }
+
+    hvf_gicv3_put(s);
+}
+
+
+/*
+ * CPU interface registers of GIC needs to be reset on CPU reset.
+ * For the calling arm_gicv3_icc_reset() on CPU reset, we register
+ * below ARMCPRegInfo. As we reset the whole cpu interface under single
+ * register reset, we define only one register of CPU interface instead
+ * of defining all the registers.
+ */
+static const ARMCPRegInfo gicv3_cpuif_reginfo[] = {
+    { .name = "ICC_CTLR_EL1", .state = ARM_CP_STATE_BOTH,
+      .opc0 = 3, .opc1 = 0, .crn = 12, .crm = 12, .opc2 = 4,
+      /*
+       * If ARM_CP_NOP is used, resetfn is not called,
+       * So ARM_CP_NO_RAW is appropriate type.
+       */
+      .type = ARM_CP_NO_RAW,
+      .access = PL1_RW,
+      .readfn = arm_cp_read_zero,
+      .writefn = arm_cp_write_ignore,
+      /*
+       * We hang the whole cpu interface reset routine off here
+       * rather than parcelling it out into one little function
+       * per register
+       */
+      .resetfn = hvf_gicv3_icc_reset,
+    },
+};
+
+static void hvf_gicv3_realize(DeviceState *dev, Error **errp)
+{
+    ERRP_GUARD();
+    GICv3State *s = HVF_GICV3(dev);
+    HVFARMGICv3Class *kgc = HVF_GICV3_GET_CLASS(s);
+    int i;
+
+    kgc->parent_realize(dev, errp);
+    if (*errp) {
+        return;
+    }
+
+    if (s->revision != 3) {
+        error_setg(errp, "unsupported GIC revision %d for platform GIC",
+                   s->revision);
+    }
+
+    if (s->security_extn) {
+        error_setg(errp, "the platform vGICv3 does not implement the "
+                   "security extensions");
+        return;
+    }
+
+    if (s->nmi_support) {
+        error_setg(errp, "NMI is not supported with the platform GIC");
+        return;
+    }
+
+    if (s->nb_redist_regions > 1) {
+        error_setg(errp, "Multiple VGICv3 redistributor regions are not "
+                   "supported by HVF");
+        error_append_hint(errp, "A maximum of %d VCPUs can be used",
+                          s->redist_region_count[0]);
+        return;
+    }
+
+    gicv3_init_irqs_and_mmio(s, hvf_gicv3_set_irq, NULL);
+
+    for (i = 0; i < s->num_cpu; i++) {
+        ARMCPU *cpu = ARM_CPU(qemu_get_cpu(i));
+
+        define_arm_cp_regs(cpu, gicv3_cpuif_reginfo);
+    }
+
+    if (s->maint_irq && s->maint_irq != HV_GIC_INT_MAINTENANCE) {
+        error_setg(errp, "vGIC maintenance IRQ mismatch with the hardcoded one in HVF.");
+        return;
+    }
+}
+
+static void hvf_gicv3_class_init(ObjectClass *klass, const void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    ResettableClass *rc = RESETTABLE_CLASS(klass);
+    ARMGICv3CommonClass *agcc = ARM_GICV3_COMMON_CLASS(klass);
+    HVFARMGICv3Class *kgc = HVF_GICV3_CLASS(klass);
+
+    agcc->pre_save = hvf_gicv3_get;
+    agcc->post_load = hvf_gicv3_put;
+
+    device_class_set_parent_realize(dc, hvf_gicv3_realize,
+                                    &kgc->parent_realize);
+    resettable_class_set_parent_phases(rc, NULL, hvf_gicv3_reset_hold, NULL,
+                                       &kgc->parent_phases);
+}
+
+static const TypeInfo hvf_arm_gicv3_info = {
+    .name = TYPE_HVF_GICV3,
+    .parent = TYPE_ARM_GICV3_COMMON,
+    .instance_size = sizeof(GICv3State),
+    .class_init = hvf_gicv3_class_init,
+    .class_size = sizeof(HVFARMGICv3Class),
+};
+
+static void hvf_gicv3_register_types(void)
+{
+    type_register_static(&hvf_arm_gicv3_info);
+}
+
+type_init(hvf_gicv3_register_types)
hw/intc/meson.build +1 -0
index 96742df090..b7baf8a0f6 100644
@@ -42,6 +42,7 @@ arm_common_ss.add(when: 'CONFIG_ARM_GIC', if_true: files('arm_gicv3_cpuif_common
 arm_common_ss.add(when: 'CONFIG_ARM_GICV3', if_true: files('arm_gicv3_cpuif.c'))
 specific_ss.add(when: 'CONFIG_ARM_GIC_KVM', if_true: files('arm_gic_kvm.c'))
 specific_ss.add(when: ['CONFIG_WHPX', 'TARGET_AARCH64'], if_true: files('arm_gicv3_whpx.c'))
+specific_ss.add(when: ['CONFIG_HVF', 'CONFIG_ARM_GICV3'], if_true: files('arm_gicv3_hvf.c'))
 specific_ss.add(when: ['CONFIG_ARM_GIC_KVM', 'TARGET_AARCH64'], if_true: files('arm_gicv3_kvm.c', 'arm_gicv3_its_kvm.c'))
 arm_common_ss.add(when: 'CONFIG_ARM_V7M', if_true: files('armv7m_nvic.c'))
 specific_ss.add(when: 'CONFIG_GRLIB', if_true: files('grlib_irqmp.c'))
include/hw/intc/arm_gicv3_common.h +1 -0
index c55cf18120..9adcab0a0c 100644
@@ -315,6 +315,7 @@ DECLARE_OBJ_CHECKERS(GICv3State, ARMGICv3CommonClass,
 
 /* Types for GICv3 kernel-irqchip */
 #define TYPE_WHPX_GICV3 "whpx-arm-gicv3"
+#define TYPE_HVF_GICV3 "hvf-arm-gicv3"
 
 struct ARMGICv3CommonClass {
     /*< private >*/
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 1/12
#2 lore raw

Re: [PATCH v11 01/12] hw/intc: Add hvf vGIC interrupt controller support

Quoted 9 lines · Mohamed Mediouni

This opens up the door to nested virtualisation support.

---
  hw/intc/arm_gicv3_hvf.c            | 735 +++++++++++++++++++++++++++++
  hw/intc/meson.build                |   1 +
  include/hw/intc/arm_gicv3_common.h |   1 +
  3 files changed, 737 insertions(+)

create mode 100644 hw/intc/arm_gicv3_hvf.c

This patch is huge, so I surely missed details; but the overall LGTM (it it won't break any existing feature), so:

#3 lore raw

PATCH v11 2/12 hw/intc: arm_gicv3_hvf: save/restore Apple GIC state

In reply to Mohamed Mediouni

On HVF, some of the GIC state is in an opaque Apple-provided structure.

Save/restore that state to be able to save/restore VMs that use the hardware GIC.

---
 hw/intc/arm_gicv3_common.c         |  1 +
 hw/intc/arm_gicv3_hvf.c            | 95 ++++++++++++++++++++++++++++--
 hw/intc/arm_gicv3_hvf_stub.c       | 25 ++++++++
 hw/intc/meson.build                |  1 +
 include/hw/intc/arm_gicv3_common.h |  3 +
 5 files changed, 121 insertions(+), 4 deletions(-)
 create mode 100644 hw/intc/arm_gicv3_hvf_stub.c
hw/intc/arm_gicv3_common.c +1 -0
index 9200671c7a..9c3fb2f4bf 100644
@@ -305,6 +305,7 @@ static const VMStateDescription vmstate_gicv3 = {
     .subsections = (const VMStateDescription * const []) {
         &vmstate_gicv3_gicd_no_migration_shift_bug,
         &vmstate_gicv3_gicd_nmi,
+        &vmstate_gicv3_hvf,
         NULL
     }
 };
hw/intc/arm_gicv3_hvf.c +91 -4

125 lines not on this page · raw message

hw/intc/arm_gicv3_hvf_stub.c new file +25 -0
new file mode 100644
index 0000000000..a587332c7c
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * ARM Generic Interrupt Controller using HVF platform support stub
+ *
+ * Copyright (c) 2026 Mohamed Mediouni
+ *
+ */
+#include "qemu/osdep.h"
+#include "hw/intc/arm_gicv3_common.h"
+#include "migration/vmstate.h"
+#include "qemu/typedefs.h"
+
+static bool needed_never(void *opaque)
+{
+    return false;
+}
+
+const VMStateDescription vmstate_gicv3_hvf = {
+    .name = "arm_gicv3/hvf_gic_state",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = needed_never,
+    .version_id = 1,
+    .minimum_version_id = 1,
+};
hw/intc/meson.build +1 -0
index b7baf8a0f6..c6de2d9d00 100644
@@ -43,6 +43,7 @@ arm_common_ss.add(when: 'CONFIG_ARM_GICV3', if_true: files('arm_gicv3_cpuif.c'))
 specific_ss.add(when: 'CONFIG_ARM_GIC_KVM', if_true: files('arm_gic_kvm.c'))
 specific_ss.add(when: ['CONFIG_WHPX', 'TARGET_AARCH64'], if_true: files('arm_gicv3_whpx.c'))
 specific_ss.add(when: ['CONFIG_HVF', 'CONFIG_ARM_GICV3'], if_true: files('arm_gicv3_hvf.c'))
+specific_ss.add(when: ['CONFIG_HVF', 'CONFIG_ARM_GICV3'], if_false: files('arm_gicv3_hvf_stub.c'))
 specific_ss.add(when: ['CONFIG_ARM_GIC_KVM', 'TARGET_AARCH64'], if_true: files('arm_gicv3_kvm.c', 'arm_gicv3_its_kvm.c'))
 arm_common_ss.add(when: 'CONFIG_ARM_V7M', if_true: files('armv7m_nvic.c'))
 specific_ss.add(when: 'CONFIG_GRLIB', if_true: files('grlib_irqmp.c'))
include/hw/intc/arm_gicv3_common.h +3 -0
index 9adcab0a0c..03ab3e8f2f 100644
@@ -339,4 +339,7 @@ void gicv3_init_irqs_and_mmio(GICv3State *s, qemu_irq_handler handler,
  */
 const char *gicv3_class_name(void);
 
+/* HVF vGIC-specific state: stubbed out on a build with HVF disabled */
+extern const VMStateDescription vmstate_gicv3_hvf;
+
 #endif
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 2/12
#4 lore raw

Re: [PATCH v11 02/12] hw/intc: arm_gicv3_hvf: save/restore Apple GIC state

Quoted 5 lines · Mohamed Mediouni

On HVF, some of the GIC state is in an opaque Apple-provided structure.

Save/restore that state to be able to save/restore VMs that use the hardware GIC.

Quoted 8 lines
---
  hw/intc/arm_gicv3_common.c         |  1 +
  hw/intc/arm_gicv3_hvf.c            | 95 ++++++++++++++++++++++++++++--
  hw/intc/arm_gicv3_hvf_stub.c       | 25 ++++++++
  hw/intc/meson.build                |  1 +
  include/hw/intc/arm_gicv3_common.h |  3 +
  5 files changed, 121 insertions(+), 4 deletions(-)

create mode 100644 hw/intc/arm_gicv3_hvf_stub.c

#5 lore raw

PATCH v11 3/12 accel, hw/arm, include/system/hvf: infrastructure changes for HVF vGIC

In reply to Mohamed Mediouni

Misc changes needed for HVF vGIC enablement.

Note: x86_64 macOS exposes interrupt controller virtualisation since macOS 12. Keeping an #ifdef here in case we end up supporting that...

However, given that x86_64 macOS is on its way out, it'll probably (?) not be supported in Qemu.

---
 accel/hvf/hvf-all.c        | 50 ++++++++++++++++++++++++++++++++++++++
 accel/stubs/hvf-stub.c     |  1 +
 hw/arm/virt.c              | 23 +++++++++++++++---
 hw/intc/arm_gicv3_common.c |  3 +++
 include/system/hvf.h       |  3 +++
 system/vl.c                |  2 ++
 6 files changed, 78 insertions(+), 4 deletions(-)
accel/hvf/hvf-all.c +50 -0
index 033c677b6f..929f53fd37 100644
@@ -10,6 +10,8 @@
 
 #include "qemu/osdep.h"
 #include "qemu/error-report.h"
+#include "qapi/error.h"
+#include "qapi/qapi-visit-common.h"
 #include "accel/accel-ops.h"
 #include "exec/cpu-common.h"
 #include "system/address-spaces.h"
@@ -22,6 +24,7 @@
 #include "trace.h"
 
 bool hvf_allowed;
+bool hvf_kernel_irqchip;
 
 const char *hvf_return_string(hv_return_t ret)
 {
@@ -217,6 +220,43 @@ static int hvf_gdbstub_sstep_flags(AccelState *as)
     return SSTEP_ENABLE | SSTEP_NOIRQ;
 }
 
+static void hvf_set_kernel_irqchip(Object *obj, Visitor *v,
+                                   const char *name, void *opaque,
+                                   Error **errp)
+{
+    OnOffSplit mode;
+    if (!visit_type_OnOffSplit(v, name, &mode, errp)) {
+        return;
+    }
+
+    switch (mode) {
+    case ON_OFF_SPLIT_ON:
+#ifdef HOST_X86_64
+        /* macOS 12 onwards exposes an HVF virtual APIC. */
+        error_setg(errp, "HVF: kernel irqchip is not currently implemented for x86.");
+        break;
+#else
+        hvf_kernel_irqchip = true;
+        break;
+#endif
+
+    case ON_OFF_SPLIT_OFF:
+        hvf_kernel_irqchip = false;
+        break;
+
+    case ON_OFF_SPLIT_SPLIT:
+        error_setg(errp, "HVF: split irqchip is not supported on HVF.");
+        break;
+
+    default:
+        /*
+         * The value was checked in visit_type_OnOffSplit() above. If
+         * we get here, then something is wrong in QEMU.
+         */
+        abort();
+    }
+}
+
 static void hvf_accel_class_init(ObjectClass *oc, const void *data)
 {
     AccelClass *ac = ACCEL_CLASS(oc);
@@ -224,6 +264,16 @@ static void hvf_accel_class_init(ObjectClass *oc, const void *data)
     ac->init_machine = hvf_accel_init;
     ac->allowed = &hvf_allowed;
     ac->gdbstub_supported_sstep_flags = hvf_gdbstub_sstep_flags;
+#ifdef HOST_X86_64
+    hvf_kernel_irqchip = false;
+#else
+    hvf_kernel_irqchip = true;
+#endif
+    object_class_property_add(oc, "kernel-irqchip", "on|off|split",
+        NULL, hvf_set_kernel_irqchip,
+        NULL, NULL);
+    object_class_property_set_description(oc, "kernel-irqchip",
+        "Configure HVF irqchip");
 }
 
 static const TypeInfo hvf_accel_type = {
accel/stubs/hvf-stub.c +1 -0
index 42eadc5ca9..6bd08759ba 100644
@@ -10,3 +10,4 @@
 #include "system/hvf.h"
 
 bool hvf_allowed;
+bool hvf_kernel_irqchip;
hw/arm/virt.c +19 -4
index cab2e21e8a..8db5611f7f 100644
@@ -838,7 +838,7 @@ static void create_gic(VirtMachineState *vms, MemoryRegion *mem)
      * interrupts; there are always 32 of the former (mandated by GIC spec).
      */
     qdev_prop_set_uint32(vms->gic, "num-irq", NUM_IRQS + 32);
-    if (!kvm_irqchip_in_kernel()) {
+    if (!kvm_irqchip_in_kernel() && !hvf_irqchip_in_kernel()) {
         qdev_prop_set_bit(vms->gic, "has-security-extensions", vms->secure);
     }
 
@@ -861,7 +861,8 @@ static void create_gic(VirtMachineState *vms, MemoryRegion *mem)
         qdev_prop_set_array(vms->gic, "redist-region-count",
                             redist_region_count);
 
-        if (!kvm_irqchip_in_kernel()) {
+        if (!kvm_irqchip_in_kernel() &&
+         !(hvf_enabled() && hvf_irqchip_in_kernel())) {
             if (vms->tcg_its) {
                 object_property_set_link(OBJECT(vms->gic), "sysmem",
                                          OBJECT(mem), &error_fatal);
@@ -872,7 +873,7 @@ static void create_gic(VirtMachineState *vms, MemoryRegion *mem)
                                  ARCH_GIC_MAINT_IRQ);
         }
     } else {
-        if (!kvm_irqchip_in_kernel()) {
+        if (!kvm_irqchip_in_kernel() && !hvf_irqchip_in_kernel()) {
             qdev_prop_set_bit(vms->gic, "has-virtualization-extensions",
                               vms->virt);
         }
@@ -2119,7 +2120,15 @@ static void finalize_gic_version(VirtMachineState *vms)
         accel_name = "KVM with kernel-irqchip=off";
     } else if (whpx_enabled()) {
         gics_supported |= VIRT_GIC_VERSION_3_MASK;
-    } else if (tcg_enabled() || hvf_enabled() || qtest_enabled())  {
+    } else if (hvf_enabled()) {
+        if (!hvf_irqchip_in_kernel()) {
+            gics_supported |= VIRT_GIC_VERSION_2_MASK;
+        }
+        /* Hypervisor.framework doesn't expose EL2<->1 transition notifiers */
+        if (!(!hvf_irqchip_in_kernel() && vms->virt)) {
+            gics_supported |= VIRT_GIC_VERSION_3_MASK;
+        }
+    } else if (tcg_enabled() || qtest_enabled())  {
         gics_supported |= VIRT_GIC_VERSION_2_MASK;
         if (module_object_class_by_name("arm-gicv3")) {
             gics_supported |= VIRT_GIC_VERSION_3_MASK;
@@ -2161,6 +2170,8 @@ static void finalize_msi_controller(VirtMachineState *vms)
             vms->msi_controller = VIRT_MSI_CTRL_GICV2M;
         } else if (whpx_enabled()) {
             vms->msi_controller = VIRT_MSI_CTRL_GICV2M;
+        }  else if (hvf_enabled() && hvf_irqchip_in_kernel()) {
+            vms->msi_controller = VIRT_MSI_CTRL_GICV2M;
         } else {
             vms->msi_controller = VIRT_MSI_CTRL_ITS;
         }
@@ -2180,6 +2191,10 @@ static void finalize_msi_controller(VirtMachineState *vms)
             error_report("ITS not supported on WHPX.");
             exit(1);
         }
+        if (hvf_enabled() && hvf_irqchip_in_kernel()) {
+            error_report("ITS not supported on HVF when using the hardware vGIC.");
+            exit(1);
+        }
     }
 
     assert(vms->msi_controller != VIRT_MSI_CTRL_AUTO);
hw/intc/arm_gicv3_common.c +3 -0
index 9c3fb2f4bf..f7ba74e6d5 100644
@@ -33,6 +33,7 @@
 #include "hw/arm/linux-boot-if.h"
 #include "system/kvm.h"
 #include "system/whpx.h"
+#include "system/hvf.h"
 
 
 static void gicv3_gicd_no_migration_shift_bug_post_load(GICv3State *cs)
@@ -659,6 +660,8 @@ const char *gicv3_class_name(void)
         return "kvm-arm-gicv3";
     } else if (whpx_enabled()) {
         return TYPE_WHPX_GICV3;
+    } else if (hvf_enabled() && hvf_irqchip_in_kernel()) {
+        return TYPE_HVF_GICV3;
     } else {
         if (kvm_enabled()) {
             error_report("Userspace GICv3 is not supported with KVM");
include/system/hvf.h +3 -0
index d3dcf088b3..dc8da85979 100644
@@ -26,8 +26,11 @@
 #ifdef CONFIG_HVF_IS_POSSIBLE
 extern bool hvf_allowed;
 #define hvf_enabled() (hvf_allowed)
+extern bool hvf_kernel_irqchip;
+#define hvf_irqchip_in_kernel()  (hvf_kernel_irqchip)
 #else /* !CONFIG_HVF_IS_POSSIBLE */
 #define hvf_enabled() 0
+#define hvf_irqchip_in_kernel() 0
 #endif /* !CONFIG_HVF_IS_POSSIBLE */
 
 #define TYPE_HVF_ACCEL ACCEL_CLASS_NAME("hvf")
system/vl.c +2 -0
index 3e341142a0..7a3db97ee6 100644
@@ -1778,6 +1778,8 @@ static void qemu_apply_legacy_machine_options(QDict *qdict)
                                    false);
         object_register_sugar_prop(ACCEL_CLASS_NAME("whpx"), "kernel-irqchip", value,
                                    false);
+        object_register_sugar_prop(ACCEL_CLASS_NAME("hvf"), "kernel-irqchip", value,
+                                   false);
         qdict_del(qdict, "kernel-irqchip");
     }
 
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 3/12
#6 lore raw

Re: [PATCH v11 03/12] accel, hw/arm, include/system/hvf: infrastructure changes for HVF vGIC

Quoted 8 lines · Mohamed Mediouni

Misc changes needed for HVF vGIC enablement.

Note: x86_64 macOS exposes interrupt controller virtualisation since macOS 12. Keeping an #ifdef here in case we end up supporting that...

However, given that x86_64 macOS is on its way out, it'll probably (?) not be supported in Qemu.

Quoted 8 lines
---
  accel/hvf/hvf-all.c        | 50 ++++++++++++++++++++++++++++++++++++++
  accel/stubs/hvf-stub.c     |  1 +
  hw/arm/virt.c              | 23 +++++++++++++++---
  hw/intc/arm_gicv3_common.c |  3 +++
  include/system/hvf.h       |  3 +++
  system/vl.c                |  2 ++
  6 files changed, 78 insertions(+), 4 deletions(-)
#7 lore raw

PATCH v11 4/12 target/arm: hvf: instantiate GIC early

In reply to Mohamed Mediouni

While figuring out a better spot for it, put it in hv_arch_vm_create().

After hv_vcpu_create is documented as too late, and deferring vCPU initialization isn't enough either.

---
 target/arm/hvf/hvf.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
target/arm/hvf/hvf.c +16 -0
index d79469ca27..a22b544afe 100644
@@ -961,6 +961,22 @@ hv_return_t hvf_arch_vm_create(MachineState *ms, uint32_t pa_range)
     chosen_ipa_bit_size = pa_range;
 
     ret = hv_vm_create(config);
+    if (hvf_irqchip_in_kernel()) {
+        /*
+         * Instantiate GIC.
+         * This must be done prior to the creation of any vCPU
+         * but past hv_vm_create()
+         */
+        hv_gic_config_t cfg = hv_gic_config_create();
+        hv_gic_config_set_distributor_base(cfg, 0x08000000);
+        hv_gic_config_set_redistributor_base(cfg, 0x080A0000);
+        hv_return_t err = hv_gic_create(cfg);
+        if (err != HV_SUCCESS) {
+            error_report("error creating platform VGIC");
+            goto cleanup;
+        }
+        os_release(cfg);
+    }
 
 cleanup:
     os_release(config);
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 4/12
#8 lore raw

Re: [PATCH v11 04/12] target/arm: hvf: instantiate GIC early

Quoted 9 lines · Mohamed Mediouni

While figuring out a better spot for it, put it in hv_arch_vm_create().

After hv_vcpu_create is documented as too late, and deferring vCPU initialization isn't enough either.

---
  target/arm/hvf/hvf.c | 16 ++++++++++++++++
  1 file changed, 16 insertions(+)
#9 lore raw

PATCH v11 5/12 hw/arm, target/arm: nested virtualisation on HVF

In reply to Mohamed Mediouni
---
 accel/hvf/hvf-all.c    |  1 +
 accel/stubs/hvf-stub.c |  1 +
 hw/arm/virt.c          |  5 +++++
 include/system/hvf.h   |  5 +++++
 target/arm/hvf/hvf.c   | 36 ++++++++++++++++++++++++++++++++++--
 5 files changed, 46 insertions(+), 2 deletions(-)
accel/hvf/hvf-all.c +1 -0
index 929f53fd37..88f6d5accb 100644
@@ -25,6 +25,7 @@
 
 bool hvf_allowed;
 bool hvf_kernel_irqchip;
+bool hvf_nested_virt;
 
 const char *hvf_return_string(hv_return_t ret)
 {
accel/stubs/hvf-stub.c +1 -0
index 6bd08759ba..cec1cbb056 100644
@@ -11,3 +11,4 @@
 
 bool hvf_allowed;
 bool hvf_kernel_irqchip;
+bool hvf_nested_virt;
hw/arm/virt.c +5 -0
index 8db5611f7f..3fa4f41821 100644
@@ -2662,6 +2662,11 @@ static void virt_set_virt(Object *obj, bool value, Error **errp)
     VirtMachineState *vms = VIRT_MACHINE(obj);
 
     vms->virt = value;
+    /*
+     * At this point, HVF is not initialised yet.
+     * However, it needs to know if nested virt is enabled at init time.
+     */
+    hvf_nested_virt_enable(value);
 }
 
 static bool virt_get_highmem(Object *obj, Error **errp)
include/system/hvf.h +5 -0
index dc8da85979..0f0632f7ae 100644
@@ -28,9 +28,14 @@ extern bool hvf_allowed;
 #define hvf_enabled() (hvf_allowed)
 extern bool hvf_kernel_irqchip;
 #define hvf_irqchip_in_kernel()  (hvf_kernel_irqchip)
+extern bool hvf_nested_virt;
+#define hvf_nested_virt_enabled()  (hvf_nested_virt)
+#define hvf_nested_virt_enable(enable) hvf_nested_virt = enable
 #else /* !CONFIG_HVF_IS_POSSIBLE */
 #define hvf_enabled() 0
 #define hvf_irqchip_in_kernel() 0
+#define hvf_nested_virt_enabled() 0
+#define hvf_nested_virt_enable(enable) 0
 #endif /* !CONFIG_HVF_IS_POSSIBLE */
 
 #define TYPE_HVF_ACCEL ACCEL_CLASS_NAME("hvf")
target/arm/hvf/hvf.c +34 -2
index a22b544afe..a80c5f8384 100644
@@ -27,6 +27,7 @@
 #include "system/memory.h"
 #include "hw/core/boards.h"
 #include "hw/core/irq.h"
+#include "hw/arm/virt.h"
 #include "qemu/main-loop.h"
 #include "system/cpus.h"
 #include "arm-powerctl.h"
@@ -842,6 +843,10 @@ static bool hvf_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
                      (1ULL << ARM_FEATURE_PMU) |
                      (1ULL << ARM_FEATURE_GENERIC_TIMER);
 
+    if (hvf_nested_virt_enabled()) {
+        ahcf->features |= 1ULL << ARM_FEATURE_EL2;
+    }
+
     for (i = 0; i < ARRAY_SIZE(regs); i++) {
         r |= hv_vcpu_config_get_feature_reg(config, regs[i].reg,
                                             &host_isar.idregs[regs[i].index]);
@@ -949,6 +954,15 @@ void hvf_arch_vcpu_destroy(CPUState *cpu)
     assert_hvf_ok(ret);
 }
 
+static bool hvf_arm_el2_supported(void)
+{
+    bool is_nested_virt_supported;
+    hv_return_t ret = hv_vm_config_get_el2_supported(&is_nested_virt_supported);
+    assert_hvf_ok(ret);
+    return is_nested_virt_supported;
+}
+
+
 hv_return_t hvf_arch_vm_create(MachineState *ms, uint32_t pa_range)
 {
     hv_return_t ret;
@@ -960,6 +974,18 @@ hv_return_t hvf_arch_vm_create(MachineState *ms, uint32_t pa_range)
     }
     chosen_ipa_bit_size = pa_range;
 
+    if (hvf_nested_virt_enabled()) {
+        if (!hvf_arm_el2_supported()) {
+            error_report("Nested virtualization not supported on this system.");
+            goto cleanup;
+        }
+        ret = hv_vm_config_set_el2_enabled(config, true);
+        if (ret != HV_SUCCESS) {
+            error_report("Failed to enable nested virtualization.");
+            goto cleanup;
+        }
+    }
+
     ret = hv_vm_create(config);
     if (hvf_irqchip_in_kernel()) {
         /*
@@ -1110,6 +1136,13 @@ static void hvf_psci_cpu_off(ARMCPU *arm_cpu)
     assert(ret == QEMU_ARM_POWERCTL_RET_SUCCESS);
 }
 
+static int hvf_psci_get_target_el(void)
+{
+    if (hvf_nested_virt_enabled()) {
+        return 2;
+    }
+    return 1;
+}
 /*
  * Handle a PSCI call.
  *
@@ -1131,7 +1164,6 @@ static bool hvf_handle_psci_call(CPUState *cpu, int *excp_ret)
     CPUState *target_cpu_state;
     ARMCPU *target_cpu;
     target_ulong entry;
-    int target_el = 1;
     int32_t ret = 0;
 
     trace_arm_psci_call(param[0], param[1], param[2], param[3],
@@ -1185,7 +1217,7 @@ static bool hvf_handle_psci_call(CPUState *cpu, int *excp_ret)
         entry = param[2];
         context_id = param[3];
         ret = arm_set_cpu_on(mpidr, entry, context_id,
-                             target_el, target_aarch64);
+                             hvf_psci_get_target_el(), target_aarch64);
         break;
     case QEMU_PSCI_0_1_FN_CPU_OFF:
     case QEMU_PSCI_0_2_FN_CPU_OFF:
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 5/12
#10 lore raw

Re: [PATCH v11 05/12] hw/arm, target/arm: nested virtualisation on HVF

Quoted 8 lines · Mohamed Mediouni
---
  accel/hvf/hvf-all.c    |  1 +
  accel/stubs/hvf-stub.c |  1 +
  hw/arm/virt.c          |  5 +++++
  include/system/hvf.h   |  5 +++++
  target/arm/hvf/hvf.c   | 36 ++++++++++++++++++++++++++++++++++--
  5 files changed, 46 insertions(+), 2 deletions(-)
#11 lore raw

PATCH v11 6/12 hvf: only call hvf_sync_vtimer() when running without the platform vGIC

In reply to Mohamed Mediouni

When running with the Apple vGIC, the EL1 vtimer is handled by the platform.

---
 target/arm/hvf/hvf.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
target/arm/hvf/hvf.c +3 -1
index a80c5f8384..e0a51faf28 100644
@@ -2052,7 +2052,9 @@ static int hvf_handle_vmexit(CPUState *cpu, hv_vcpu_exit_t *exit)
 
     switch (exit->reason) {
     case HV_EXIT_REASON_EXCEPTION:
-        hvf_sync_vtimer(cpu);
+        if (!hvf_irqchip_in_kernel()) {
+            hvf_sync_vtimer(cpu);
+        }
         ret = hvf_handle_exception(cpu, &exit->exception);
         break;
     case HV_EXIT_REASON_VTIMER_ACTIVATED:
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 6/12
#12 lore raw

Re: [PATCH v11 06/12] hvf: only call hvf_sync_vtimer() when running without the platform vGIC

Quoted 7 lines · Mohamed Mediouni

When running with the Apple vGIC, the EL1 vtimer is handled by the platform.

---
  target/arm/hvf/hvf.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
#13 lore raw

PATCH v11 7/12 hvf: gate ARM_FEATURE_PMU register emulation behind not being at EL2

In reply to Mohamed Mediouni

From Apple documentation:

Quoted 10 lines

When EL2 is disabled, PMU register accesses trigger "Trapped MSR, MRS, or System Instruction" exceptions. When this happens, hv_vcpu_run() returns, and the hv_vcpu_exit_t object contains the information about this exception.

When EL2 is enabled, the handling of PMU register accesses is determined by the PMUVer field of ID_AA64DFR0_EL1 register.
If the PMUVer field value is zero or is invalid, PMU register accesses generate "Undefined" exceptions, which are sent to the guest.
If the PMUVer field value is non-zero and valid, PMU register accesses are emulated by the framework. The ID_AA64DFR0_EL1 register can be modified via hv_vcpu_set_sys_reg API.

---
 target/arm/hvf/hvf.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/arm/hvf/hvf.c +2 -2
index e0a51faf28..be1ff93e0e 100644
@@ -1285,7 +1285,7 @@ static int hvf_sysreg_read(CPUState *cpu, uint32_t reg, uint64_t *val)
     ARMCPU *arm_cpu = ARM_CPU(cpu);
     CPUARMState *env = &arm_cpu->env;
 
-    if (arm_feature(env, ARM_FEATURE_PMU)) {
+    if (!hvf_nested_virt_enabled() && arm_feature(env, ARM_FEATURE_PMU)) {
         switch (reg) {
         case SYSREG_PMCR_EL0:
             *val = env->cp15.c9_pmcr;
@@ -1546,7 +1546,7 @@ static int hvf_sysreg_write(CPUState *cpu, uint32_t reg, uint64_t val)
                            SYSREG_OP2(reg),
                            val);
 
-    if (arm_feature(env, ARM_FEATURE_PMU)) {
+    if (!hvf_nested_virt_enabled() && arm_feature(env, ARM_FEATURE_PMU)) {
         switch (reg) {
         case SYSREG_PMCCNTR_EL0:
             pmu_op_start(env);
Signature · 1 line
2.50.1 (Apple Git-155)
#14 lore raw

PATCH v11 8/12 hvf: arm: allow exposing minimal PMU when running with nested virt on

In reply to Mohamed Mediouni

When running with nested virt on, a minimum PMU is exposed by Hypervisor.framework if a valid PMUVer register value is set. That PMU isn't exposed otherwise.

---
 target/arm/hvf/hvf.c | 4 ++++
 1 file changed, 4 insertions(+)
target/arm/hvf/hvf.c +4 -0
index be1ff93e0e..0f45e22221 100644
@@ -879,6 +879,10 @@ static bool hvf_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
     SET_IDREG(&host_isar, ID_AA64PFR1,
               GET_IDREG(&host_isar, ID_AA64PFR1) & ~R_ID_AA64PFR1_SME_MASK);
 
+    if (hvf_nested_virt_enabled()) {
+        FIELD_DP64_IDREG(&host_isar, ID_AA64DFR0, PMUVER, 0x1);
+    }
+
     ahcf->isar = host_isar;
 
     /*
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 8/12
#15 lore raw

Re: [PATCH v11 08/12] hvf: arm: allow exposing minimal PMU when running with nested virt on

Quoted 7 lines · Mohamed Mediouni

When running with nested virt on, a minimum PMU is exposed by Hypervisor.framework if a valid PMUVer register value is set. That PMU isn't exposed otherwise.

---
  target/arm/hvf/hvf.c | 4 ++++
  1 file changed, 4 insertions(+)
#16 lore raw

PATCH v11 9/12 target/arm: hvf: add asserts for code paths not leveraged when using the vGIC

In reply to Mohamed Mediouni

When using the vGIC, timers are directly handled by the platform. No vmexits ought to happen in that case. Abort if reaching those code paths.

---
 target/arm/hvf/hvf.c | 3 +++
 1 file changed, 3 insertions(+)
target/arm/hvf/hvf.c +3 -0
index 0f45e22221..0aae421ba6 100644
@@ -1365,6 +1365,7 @@ static int hvf_sysreg_read(CPUState *cpu, uint32_t reg, uint64_t *val)
     case SYSREG_ICC_SGI1R_EL1:
     case SYSREG_ICC_SRE_EL1:
     case SYSREG_ICC_CTLR_EL1:
+        assert(!hvf_irqchip_in_kernel());
         /* Call the TCG sysreg handler. This is only safe for GICv3 regs. */
         if (hvf_sysreg_read_cp(cpu, "GICv3", reg, val)) {
             return 0;
@@ -1655,6 +1656,7 @@ static int hvf_sysreg_write(CPUState *cpu, uint32_t reg, uint64_t val)
     case SYSREG_ICC_SGI0R_EL1:
     case SYSREG_ICC_SGI1R_EL1:
     case SYSREG_ICC_SRE_EL1:
+        assert(!hvf_irqchip_in_kernel());
         /* Call the TCG sysreg handler. This is only safe for GICv3 regs. */
         if (hvf_sysreg_write_cp(cpu, "GICv3", reg, val)) {
             return 0;
@@ -2062,6 +2064,7 @@ static int hvf_handle_vmexit(CPUState *cpu, hv_vcpu_exit_t *exit)
         ret = hvf_handle_exception(cpu, &exit->exception);
         break;
     case HV_EXIT_REASON_VTIMER_ACTIVATED:
+        assert(!hvf_irqchip_in_kernel());
         qemu_set_irq(arm_cpu->gt_timer_outputs[GTIMER_VIRT], 1);
         cpu->accel->vtimer_masked = true;
         break;
Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 9/12
#17 lore raw

Re: [PATCH v11 09/12] target/arm: hvf: add asserts for code paths not leveraged when using the vGIC

Quoted 8 lines · Mohamed Mediouni

When using the vGIC, timers are directly handled by the platform. No vmexits ought to happen in that case. Abort if reaching those code paths.

---
  target/arm/hvf/hvf.c | 3 +++
  1 file changed, 3 insertions(+)
#18 lore raw

PATCH v11 10/12 hvf: sync registers used at EL2

In reply to Mohamed Mediouni

When starting up the VM at EL2, more sysregs are available. Sync the state of those.

In addition, sync the state of the EL1 physical timer when the vGIC is used, even if running at EL1. However, no OS running at EL1 is expected to use those registers.

---
 target/arm/hvf/hvf.c        | 37 ++++++++++++++++++++++++++----
 target/arm/hvf/sysreg.c.inc | 45 +++++++++++++++++++++++++++++++++++++
 2 files changed, 78 insertions(+), 4 deletions(-)
target/arm/hvf/hvf.c +33 -4
index 0aae421ba6..f286e6a014 100644
@@ -412,13 +412,34 @@ static const struct hvf_reg_match hvf_fpreg_match[] = {
 #define DEF_SYSREG(HVF_ID, ...) \
   QEMU_BUILD_BUG_ON(HVF_ID != KVMID_TO_HVF(KVMID_AA64_SYS_REG64(__VA_ARGS__)));
 
+#define DEF_SYSREG_EL2(HVF_ID, ...) \
+  QEMU_BUILD_BUG_ON(HVF_ID != KVMID_TO_HVF(KVMID_AA64_SYS_REG64(__VA_ARGS__)));
+
+#define DEF_SYSREG_VGIC(HVF_ID, ...) \
+  QEMU_BUILD_BUG_ON(HVF_ID != KVMID_TO_HVF(KVMID_AA64_SYS_REG64(__VA_ARGS__)));
+
+#define DEF_SYSREG_VGIC_EL2(HVF_ID, ...) \
+  QEMU_BUILD_BUG_ON(HVF_ID != KVMID_TO_HVF(KVMID_AA64_SYS_REG64(__VA_ARGS__)));
+
 #include "sysreg.c.inc"
 
 #undef DEF_SYSREG
+#undef DEF_SYSREG_EL2
+#undef DEF_SYSREG_VGIC
+#undef DEF_SYSREG_VGIC_EL2
+
+#define DEF_SYSREG(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID},
+#define DEF_SYSREG_EL2(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID, .el2 = true},
+#define DEF_SYSREG_VGIC(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID, .vgic = true},
+#define DEF_SYSREG_VGIC_EL2(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID, true, true},
+
+struct hvf_sreg {
+    hv_sys_reg_t sreg;
+    bool vgic;
+    bool el2;
+};
 
-#define DEF_SYSREG(HVF_ID, op0, op1, crn, crm, op2)  HVF_ID,
-
-static const hv_sys_reg_t hvf_sreg_list[] = {
+static struct hvf_sreg hvf_sreg_list[] = {
 #include "sysreg.c.inc"
 };
 
@@ -1052,11 +1073,19 @@ int hvf_arch_init_vcpu(CPUState *cpu)
 
     /* Populate cp list for all known sysregs */
     for (i = 0; i < sregs_match_len; i++) {
-        hv_sys_reg_t hvf_id = hvf_sreg_list[i];
+        hv_sys_reg_t hvf_id = hvf_sreg_list[i].sreg;
         uint64_t kvm_id = HVF_TO_KVMID(hvf_id);
         uint32_t key = kvm_to_cpreg_id(kvm_id);
         const ARMCPRegInfo *ri = get_arm_cp_reginfo(arm_cpu->cp_regs, key);
 
+        if (hvf_sreg_list[i].vgic && !hvf_irqchip_in_kernel()) {
+            continue;
+        }
+
+        if (hvf_sreg_list[i].el2 && !hvf_nested_virt_enabled()) {
+            continue;
+        }
+
         if (ri) {
             assert(!(ri->type & ARM_CP_NO_RAW));
             arm_cpu->cpreg_indexes[sregs_cnt++] = kvm_id;
target/arm/hvf/sysreg.c.inc +45 -0

48 lines not on this page · raw message

Signature · 1 line
2.50.1 (Apple Git-155)
Reviewed-by on v11 10/12
#19 lore raw

Re: [PATCH v11 10/12] hvf: sync registers used at EL2

Quoted 37 lines · Mohamed Mediouni

When starting up the VM at EL2, more sysregs are available. Sync the state of those.

In addition, sync the state of the EL1 physical timer when the vGIC is used, even if running at EL1. However, no OS running at EL1 is expected to use those registers.

---
  target/arm/hvf/hvf.c        | 37 ++++++++++++++++++++++++++----
  target/arm/hvf/sysreg.c.inc | 45 +++++++++++++++++++++++++++++++++++++
  2 files changed, 78 insertions(+), 4 deletions(-)
target/arm/hvf/hvf.c +12 -4
+#define DEF_SYSREG(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID},
+#define DEF_SYSREG_EL2(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID, .el2 = true},
+#define DEF_SYSREG_VGIC(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID, .vgic = true},
+#define DEF_SYSREG_VGIC_EL2(HVF_ID, op0, op1, crn, crm, op2)  {HVF_ID, true, true},
+
+struct hvf_sreg {
+    hv_sys_reg_t sreg;
+    bool vgic;
+    bool el2;
+};
 
-#define DEF_SYSREG(HVF_ID, op0, op1, crn, crm, op2)  HVF_ID,
-
-static const hv_sys_reg_t hvf_sreg_list[] = {
+static struct hvf_sreg hvf_sreg_list[] = {
  #include "sysreg.c.inc"
  };
@@ -1052,11 +1073,19 @@ int hvf_arch_init_vcpu(CPUState *cpu)
 
      /* Populate cp list for all known sysregs */
      for (i = 0; i < sregs_match_len; i++) {
-        hv_sys_reg_t hvf_id = hvf_sreg_list[i];
+        hv_sys_reg_t hvf_id = hvf_sreg_list[i].sreg;
          uint64_t kvm_id = HVF_TO_KVMID(hvf_id);

target/arm/hvf/hvf.c int hvf_arch_init_vcpu(CPUState *cpu) · line 1087

Quoted 10 lines
target/arm/hvf/hvf.c +7 -0
          uint32_t key = kvm_to_cpreg_id(kvm_id);
          const ARMCPRegInfo *ri = get_arm_cp_reginfo(arm_cpu->cp_regs, key);
 
+        if (hvf_sreg_list[i].vgic && !hvf_irqchip_in_kernel()) {
+            continue;
+        }
+
+        if (hvf_sreg_list[i].el2 && !hvf_nested_virt_enabled()) {
+            continue;
+        }

Nice :)

#20 lore raw

PATCH v11 11/12 target/arm: hvf: pass through CNTHCTL_EL2 and MDCCINT_EL1

In reply to Mohamed Mediouni

HVF traps accesses to CNTHCTL_EL2. For nested guests, HVF traps accesses to MDCCINT_EL1. Pass through those accesses to the Hypervisor.framework library.

---
 target/arm/hvf/hvf.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
target/arm/hvf/hvf.c +20 -0
index f286e6a014..8d537b6201 100644
@@ -296,6 +296,10 @@ void hvf_arm_init_debug(void)
 #define SYSREG_DBGWVR15_EL1   SYSREG(2, 0, 0, 15, 6)
 #define SYSREG_DBGWCR15_EL1   SYSREG(2, 0, 0, 15, 7)
 
+/* EL2 registers */
+#define SYSREG_CNTHCTL_EL2    SYSREG(3, 4, 14, 1, 0)
+#define SYSREG_MDCCINT_EL1    SYSREG(2, 0, 0, 2, 0)
+
 #define WFX_IS_WFE (1 << 0)
 
 #define TMR_CTL_ENABLE  (1 << 0)
@@ -1368,6 +1372,14 @@ static int hvf_sysreg_read(CPUState *cpu, uint32_t reg, uint64_t *val)
     case SYSREG_OSDLR_EL1:
         /* Dummy register */
         return 0;
+    case SYSREG_CNTHCTL_EL2:
+        if (__builtin_available(macOS 15.0, *)) {
+            assert_hvf_ok(hv_vcpu_get_sys_reg(cpu->accel->fd, HV_SYS_REG_CNTHCTL_EL2, val));
+        }
+        return 0;
+    case SYSREG_MDCCINT_EL1:
+        assert_hvf_ok(hv_vcpu_get_sys_reg(cpu->accel->fd, HV_SYS_REG_MDCCINT_EL1, val));
+        return 0;
     case SYSREG_ICC_AP0R0_EL1:
     case SYSREG_ICC_AP0R1_EL1:
     case SYSREG_ICC_AP0R2_EL1:
@@ -1656,6 +1668,14 @@ static int hvf_sysreg_write(CPUState *cpu, uint32_t reg, uint64_t val)
     case SYSREG_OSDLR_EL1:
         /* Dummy register */
         return 0;
+    case SYSREG_CNTHCTL_EL2:
+        if (__builtin_available(macOS 15.0, *)) {
+            assert_hvf_ok(hv_vcpu_set_sys_reg(cpu->accel->fd, HV_SYS_REG_CNTHCTL_EL2, val));
+        }
+        return 0;
+    case SYSREG_MDCCINT_EL1:
+        assert_hvf_ok(hv_vcpu_set_sys_reg(cpu->accel->fd, HV_SYS_REG_MDCCINT_EL1, val));
+        return 0;
     case SYSREG_LORC_EL1:
         /* Dummy register */
         return 0;
Signature · 1 line
2.50.1 (Apple Git-155)
#21 lore raw

PATCH v11 12/12 hvf: enable nested virtualisation support

In reply to Mohamed Mediouni
---
 hw/arm/virt.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
hw/arm/virt.c +8 -1
index 3fa4f41821..731ada5d28 100644
@@ -2394,13 +2394,20 @@ static void machvirt_init(MachineState *machine)
         exit(1);
     }
 
-    if (vms->virt && !kvm_enabled() && !tcg_enabled() && !qtest_enabled()) {
+    if (vms->virt && !kvm_enabled() && !tcg_enabled()
+       && !hvf_enabled() && !qtest_enabled()) {
         error_report("mach-virt: %s does not support providing "
                      "Virtualization extensions to the guest CPU",
                      current_accel_name());
         exit(1);
     }
 
+    if (vms->virt && hvf_enabled() && !hvf_irqchip_in_kernel()) {
+        error_report("mach-virt: For HVF and Virtualization extensions "
+                     "only kernel-irqchip=on is currently supported");
+        exit(1);
+    }
+
     if (vms->mte && hvf_enabled()) {
         error_report("mach-virt: %s does not support providing "
                      "MTE to the guest CPU",
Signature · 1 line
2.50.1 (Apple Git-155)
#23 lore raw
In reply to Mohamed Mediouni
Quoted 11 lines · Mohamed Mediouni

Link to branch: https://github.com/mediouni-m/qemu hvf-irqchip-and-nested (tag for this submission: hvf-irqchip-and-nested-v11)

This series adds supports for nested virtualisation when using HVF on arm64 Macs.

Reviews would be very welcome :-) It's maybe a bit unlikely that this will make it to QEMU 11.0 I might as well try.

The patches remaining for review:

1, 2, 3, 4, 5, 8, 10, 12

Quoted 5 lines

It has two parts:
- Apple vGICv3 support and necessary infrastructure changes for it
- Nested virtualisation support. Note that the nested virtualisation implementation shipping as of macOS 26.3 is nVHE only (but it _does_ use VNCR as shipped).

Quoted 7 lines

This series doesn't contain EL2 physical timer emulation, which is
needed if not leveraging the Apple vGIC. As such, nested virt requires kernel-irqchip=on.

Known issues:
- when nested virt is enabled, no UI response within EDK2 and a permanent wait. Workaround: -boot menu=on,splash-time=0.

For reference this is a platform bug pending resolution an OS update. Would it be good to add a specific warning about it? (In a separate commit)

Quoted 100 lines

Apple Feedback Assistant item: FB21649319

When the VM is running at EL2 at the very moment the virtual timer fires:
HV_EXIT_REASON_VTIMER_ACTIVATED doesn’t fire (when not using the provided vGIC)
- using a GICv2 doesn’t require having transition notifiers which Hypervisor.framework doesn’t have…)

and when using the provided vGIC, the interrupt never gets delivered back to the guest

Linux as a guest OS is fine with this… but the reference ArmVirtQemu edk2 build always uses the virtual timer even when running EFI at EL2, so it gets broken unless this patch is applied to edk2:

diff of ArmVirtPkg/ArmVirt.dsc.inc:

(quoted diff) +1 -1
-  ArmGenericTimerCounterLib|ArmPkg/Library/ArmGenericTimerVirtCounterLib/ArmGenericTimerVirtCounterLib.inf
+  ArmGenericTimerCounterLib|ArmPkg/Library/ArmGenericTimerPhyCounterLib/ArmGenericTimerPhyCounterLib.inf^M

Changelog:

v1->v2:
Oops. I did a mistake when preparing my patches.

- Add hvf_arm_el2_enable(_) call to virt_set_virt
- Fix nested virt support check to add HVF

v2->v3:
- LORC_EL1 patch was merged separately, remove from this series.
- fix LPIs when kernel-irqchip disabled and using TCG
- remove spurious if case in vGIC supported version detection (inapplicable now)
- Add hvf_enabled() check in combination with hvf kernel-irqchip checks
- cleanly fail on attempt to use the platform vGIC together with ITS

v3->v4:
- GIC state save improvements, including saving the opaque Apple-specific state
- Saving HVF system register state when using the vGIC and/or EL2

v5:
- oops, fixed up save/restore to be functional
- misc changes otherwise

v6:
- Addressing review comments

v7:
- Address review comments, adapt around Qemu changes and bugfixes.

v8:
- Rebase, and misc fixes

v9:
- Misc fixes from WHPX review comments
- Split enablement commit for EL2 into its own separate commit

v10:

- A bit of changes across the board
- move HVF irqchip save/restore into irqchip code
- block nested virt support when kernel-irqchip=off with a GICv2 as we don't implement an emulated physical timer yet.

v11:
- fixups for the save/restore Apple GIC state patch
- block the same registers as the revert in 28b0ed32b32c7e5094cf2f1ec9c0645c65fad2aa for "hvf: sync registers used at EL2"
- drop patch that warns instead of aborts on mismatch

Mohamed Mediouni (12):
hw/intc: Add hvf vGIC interrupt controller support
hw/intc: arm_gicv3_hvf: save/restore Apple GIC state
accel, hw/arm, include/system/hvf: infrastructure changes for HVF vGIC target/arm: hvf: instantiate GIC early
hw/arm, target/arm: nested virtualisation on HVF
hvf: only call hvf_sync_vtimer() when running without the platform
vGIC
hvf: gate ARM_FEATURE_PMU register emulation behind not being at EL2
hvf: arm: allow exposing minimal PMU when running with nested virt on target/arm: hvf: add asserts for code paths not leveraged when using
the vGIC
hvf: sync registers used at EL2
target/arm: hvf: pass through CNTHCTL_EL2 and MDCCINT_EL1
hvf: enable nested virtualisation support

accel/hvf/hvf-all.c                |  51 ++
accel/stubs/hvf-stub.c             |   2 +
hw/arm/virt.c                      |  37 +-
hw/intc/arm_gicv3_common.c         |   4 +
hw/intc/arm_gicv3_hvf.c            | 822 +++++++++++++++++++++++++++++
hw/intc/arm_gicv3_hvf_stub.c       |  25 +
hw/intc/meson.build                |   2 +
include/hw/intc/arm_gicv3_common.h |   4 +
include/system/hvf.h               |   8 +
system/vl.c                        |   2 +
target/arm/hvf/hvf.c               | 124 ++++-
target/arm/hvf/sysreg.c.inc        |  45 ++
12 files changed, 1112 insertions(+), 14 deletions(-)

create mode 100644 hw/intc/arm_gicv3_hvf.c
create mode 100644 hw/intc/arm_gicv3_hvf_stub.c

Signature · 1 line
2.50.1 (Apple Git-155)
#24 lore raw
Quoted 8 lines · Mohamed Mediouni
Quoted 8 lines · Mohamed Mediouni

Link to branch: https://github.com/mediouni-m/qemu hvf-irqchip-and-nested (tag for this submission: hvf-irqchip-and-nested-v11)

This series adds supports for nested virtualisation when using HVF on arm64 Macs.

Quoted 7 lines
Quoted 3 lines

Reviews would be very welcome :-) It's maybe a bit unlikely that this will make it to QEMU 11.0 I might as well try.

The patches remaining for review:

1, 2, 3, 4, 5, 8, 10, 12

Cc'ing Manos who recently worked on SME2 in this area, hopefully he has some review time available.

Quoted 119 lines
Quoted 6 lines

It has two parts:
- Apple vGICv3 support and necessary infrastructure changes for it
- Nested virtualisation support. Note that the nested virtualisation implementation shipping as of macOS 26.3 is nVHE only (but it _does_ use VNCR as shipped).

Quoted 7 lines

This series doesn't contain EL2 physical timer emulation, which is
needed if not leveraging the Apple vGIC. As such, nested virt requires kernel-irqchip=on.

Known issues:
- when nested virt is enabled, no UI response within EDK2 and a permanent wait. Workaround: -boot menu=on,splash-time=0.

For reference this is a platform bug pending resolution an OS update. Would it be good to add a specific warning about it? (In a separate commit)

Quoted 100 lines

Apple Feedback Assistant item: FB21649319

When the VM is running at EL2 at the very moment the virtual timer fires:
HV_EXIT_REASON_VTIMER_ACTIVATED doesn’t fire (when not using the provided vGIC)
- using a GICv2 doesn’t require having transition notifiers which Hypervisor.framework doesn’t have…)

and when using the provided vGIC, the interrupt never gets delivered back to the guest

Linux as a guest OS is fine with this… but the reference ArmVirtQemu edk2 build always uses the virtual timer even when running EFI at EL2, so it gets broken unless this patch is applied to edk2:

diff of ArmVirtPkg/ArmVirt.dsc.inc:

(quoted diff) +1 -1
-  ArmGenericTimerCounterLib|ArmPkg/Library/ArmGenericTimerVirtCounterLib/ArmGenericTimerVirtCounterLib.inf
+  ArmGenericTimerCounterLib|ArmPkg/Library/ArmGenericTimerPhyCounterLib/ArmGenericTimerPhyCounterLib.inf^M

Changelog:

v1->v2:
Oops. I did a mistake when preparing my patches.

- Add hvf_arm_el2_enable(_) call to virt_set_virt
- Fix nested virt support check to add HVF

v2->v3:
- LORC_EL1 patch was merged separately, remove from this series.
- fix LPIs when kernel-irqchip disabled and using TCG
- remove spurious if case in vGIC supported version detection (inapplicable now)
- Add hvf_enabled() check in combination with hvf kernel-irqchip checks
- cleanly fail on attempt to use the platform vGIC together with ITS

v3->v4:
- GIC state save improvements, including saving the opaque Apple-specific state
- Saving HVF system register state when using the vGIC and/or EL2

v5:
- oops, fixed up save/restore to be functional
- misc changes otherwise

v6:
- Addressing review comments

v7:
- Address review comments, adapt around Qemu changes and bugfixes.

v8:
- Rebase, and misc fixes

v9:
- Misc fixes from WHPX review comments
- Split enablement commit for EL2 into its own separate commit

v10:

- A bit of changes across the board
- move HVF irqchip save/restore into irqchip code
- block nested virt support when kernel-irqchip=off with a GICv2 as we don't implement an emulated physical timer yet.

v11:
- fixups for the save/restore Apple GIC state patch
- block the same registers as the revert in 28b0ed32b32c7e5094cf2f1ec9c0645c65fad2aa for "hvf: sync registers used at EL2"
- drop patch that warns instead of aborts on mismatch

Mohamed Mediouni (12):
hw/intc: Add hvf vGIC interrupt controller support hw/intc: arm_gicv3_hvf: save/restore Apple GIC state accel, hw/arm, include/system/hvf: infrastructure changes for HVF vGIC target/arm: hvf: instantiate GIC early
hw/arm, target/arm: nested virtualisation on HVF
hvf: only call hvf_sync_vtimer() when running without the platform
vGIC
hvf: gate ARM_FEATURE_PMU register emulation behind not being at EL2
hvf: arm: allow exposing minimal PMU when running with nested virt on target/arm: hvf: add asserts for code paths not leveraged when using
the vGIC
hvf: sync registers used at EL2
target/arm: hvf: pass through CNTHCTL_EL2 and MDCCINT_EL1
hvf: enable nested virtualisation support

accel/hvf/hvf-all.c                |  51 ++
accel/stubs/hvf-stub.c             |   2 +
hw/arm/virt.c                      |  37 +-
hw/intc/arm_gicv3_common.c         |   4 +
hw/intc/arm_gicv3_hvf.c            | 822 +++++++++++++++++++++++++++++
hw/intc/arm_gicv3_hvf_stub.c       |  25 +
hw/intc/meson.build                |   2 +
include/hw/intc/arm_gicv3_common.h |   4 +
include/system/hvf.h               |   8 +
system/vl.c                        |   2 +
target/arm/hvf/hvf.c               | 124 ++++-
target/arm/hvf/sysreg.c.inc        |  45 ++
12 files changed, 1112 insertions(+), 14 deletions(-)

create mode 100644 hw/intc/arm_gicv3_hvf.c
create mode 100644 hw/intc/arm_gicv3_hvf_stub.c

Signature · 1 line
2.50.1 (Apple Git-155)
#25 lore raw
Quoted 9 lines · Philippe Mathieu-Daudé
Quoted 8 lines · Mohamed Mediouni
Quoted 8 lines · Mohamed Mediouni

Link to branch: https://github.com/mediouni-m/qemu hvf-irqchip-and- nested (tag for this submission: hvf-irqchip-and-nested-v11)

This series adds supports for nested virtualisation when using HVF on arm64 Macs.

Quoted 7 lines
Quoted 7 lines
Quoted 3 lines

Reviews would be very welcome :-) It's maybe a bit unlikely that this will make it to QEMU 11.0 I might as well try.

The patches remaining for review:

1, 2, 3, 4, 5, 8, 10, 12

Series fully reviewed! Hopefully it still applies on /master.

#26 lore raw
Quoted 18 lines · Philippe Mathieu-Daudé
Quoted 13 lines · Philippe Mathieu-Daudé
Quoted 12 lines · Mohamed Mediouni
Quoted 11 lines · Mohamed Mediouni

Link to branch: https://github.com/mediouni-m/qemu hvf-irqchip-and- nested (tag for this submission: hvf-irqchip-and-nested-v11)

This series adds supports for nested virtualisation when using HVF on arm64 Macs.

Reviews would be very welcome :-) It's maybe a bit unlikely that this will make it to QEMU 11.0 I might as well try.

Quoted 6 lines
Quoted 3 lines
Quoted 3 lines

The patches remaining for review:

1, 2, 3, 4, 5, 8, 10, 12

Series fully reviewed! Hopefully it still applies on /master.

Hello,

It does apply on master but not together with Manos Pitsidianakis’s SME2 support patches.

Sent over a new revision with those patches at the start of the series and then the nested virt patchset taking them into account.

The patch modified was "hvf: sync registers used at EL2" and a new patch is added: "hvf: arm: SME2 availability fixes”