xref: /openbmc/qemu/hw/arm/virt.c (revision b8c7ebbb70a316c8ea4a2a31eb071456bcc350b5)
1 /*
2  * ARM mach-virt emulation
3  *
4  * Copyright (c) 2013 Linaro Limited
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms and conditions of the GNU General Public License,
8  * version 2 or later, as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along with
16  * this program.  If not, see <http://www.gnu.org/licenses/>.
17  *
18  * Emulate a virtual board which works by passing Linux all the information
19  * it needs about what devices are present via the device tree.
20  * There are some restrictions about what we can do here:
21  *  + we can only present devices whose Linux drivers will work based
22  *    purely on the device tree with no platform data at all
23  *  + we want to present a very stripped-down minimalist platform,
24  *    both because this reduces the security attack surface from the guest
25  *    and also because it reduces our exposure to being broken when
26  *    the kernel updates its device tree bindings and requires further
27  *    information in a device binding that we aren't providing.
28  * This is essentially the same approach kvmtool uses.
29  */
30 
31 #include "qemu/osdep.h"
32 #include "qemu/datadir.h"
33 #include "qemu/units.h"
34 #include "qemu/option.h"
35 #include "monitor/qdev.h"
36 #include "hw/sysbus.h"
37 #include "hw/arm/boot.h"
38 #include "hw/arm/primecell.h"
39 #include "hw/arm/virt.h"
40 #include "hw/block/flash.h"
41 #include "hw/vfio/vfio-calxeda-xgmac.h"
42 #include "hw/vfio/vfio-amd-xgbe.h"
43 #include "hw/display/ramfb.h"
44 #include "net/net.h"
45 #include "system/device_tree.h"
46 #include "system/numa.h"
47 #include "system/runstate.h"
48 #include "system/tpm.h"
49 #include "system/tcg.h"
50 #include "system/kvm.h"
51 #include "system/hvf.h"
52 #include "system/qtest.h"
53 #include "hw/loader.h"
54 #include "qapi/error.h"
55 #include "qemu/bitops.h"
56 #include "qemu/cutils.h"
57 #include "qemu/error-report.h"
58 #include "qemu/module.h"
59 #include "hw/pci-host/gpex.h"
60 #include "hw/pci-bridge/pci_expander_bridge.h"
61 #include "hw/virtio/virtio-pci.h"
62 #include "hw/core/sysbus-fdt.h"
63 #include "hw/platform-bus.h"
64 #include "hw/qdev-properties.h"
65 #include "hw/arm/fdt.h"
66 #include "hw/intc/arm_gic.h"
67 #include "hw/intc/arm_gicv3_common.h"
68 #include "hw/intc/arm_gicv3_its_common.h"
69 #include "hw/irq.h"
70 #include "kvm_arm.h"
71 #include "hvf_arm.h"
72 #include "hw/firmware/smbios.h"
73 #include "qapi/visitor.h"
74 #include "qapi/qapi-visit-common.h"
75 #include "qobject/qlist.h"
76 #include "standard-headers/linux/input.h"
77 #include "hw/arm/smmuv3.h"
78 #include "hw/acpi/acpi.h"
79 #include "hw/acpi/pcihp.h"
80 #include "target/arm/cpu-qom.h"
81 #include "target/arm/internals.h"
82 #include "target/arm/multiprocessing.h"
83 #include "target/arm/gtimer.h"
84 #include "hw/mem/pc-dimm.h"
85 #include "hw/mem/nvdimm.h"
86 #include "hw/acpi/generic_event_device.h"
87 #include "hw/uefi/var-service-api.h"
88 #include "hw/virtio/virtio-md-pci.h"
89 #include "hw/virtio/virtio-iommu.h"
90 #include "hw/char/pl011.h"
91 #include "hw/cxl/cxl.h"
92 #include "hw/cxl/cxl_host.h"
93 #include "qemu/guest-random.h"
94 
95 static GlobalProperty arm_virt_compat[] = {
96     { TYPE_VIRTIO_IOMMU_PCI, "aw-bits", "48" },
97 };
98 static const size_t arm_virt_compat_len = G_N_ELEMENTS(arm_virt_compat);
99 
100 /*
101  * This cannot be called from the virt_machine_class_init() because
102  * TYPE_VIRT_MACHINE is abstract and mc->compat_props g_ptr_array_new()
103  * only is called on virt non abstract class init.
104  */
105 static void arm_virt_compat_set(MachineClass *mc)
106 {
107     compat_props_add(mc->compat_props, arm_virt_compat,
108                      arm_virt_compat_len);
109 }
110 
111 #define DEFINE_VIRT_MACHINE_IMPL(latest, ...) \
112     static void MACHINE_VER_SYM(class_init, virt, __VA_ARGS__)( \
113         ObjectClass *oc, \
114         const void *data) \
115     { \
116         MachineClass *mc = MACHINE_CLASS(oc); \
117         arm_virt_compat_set(mc); \
118         MACHINE_VER_SYM(options, virt, __VA_ARGS__)(mc); \
119         mc->desc = "QEMU " MACHINE_VER_STR(__VA_ARGS__) " ARM Virtual Machine"; \
120         MACHINE_VER_DEPRECATION(__VA_ARGS__); \
121         if (latest) { \
122             mc->alias = "virt"; \
123         } \
124     } \
125     static const TypeInfo MACHINE_VER_SYM(info, virt, __VA_ARGS__) = \
126     { \
127         .name = MACHINE_VER_TYPE_NAME("virt", __VA_ARGS__), \
128         .parent = TYPE_VIRT_MACHINE, \
129         .class_init = MACHINE_VER_SYM(class_init, virt, __VA_ARGS__), \
130     }; \
131     static void MACHINE_VER_SYM(register, virt, __VA_ARGS__)(void) \
132     { \
133         MACHINE_VER_DELETION(__VA_ARGS__); \
134         type_register_static(&MACHINE_VER_SYM(info, virt, __VA_ARGS__)); \
135     } \
136     type_init(MACHINE_VER_SYM(register, virt, __VA_ARGS__));
137 
138 #define DEFINE_VIRT_MACHINE_AS_LATEST(major, minor) \
139     DEFINE_VIRT_MACHINE_IMPL(true, major, minor)
140 #define DEFINE_VIRT_MACHINE(major, minor) \
141     DEFINE_VIRT_MACHINE_IMPL(false, major, minor)
142 
143 
144 /* Number of external interrupt lines to configure the GIC with */
145 #define NUM_IRQS 256
146 
147 #define PLATFORM_BUS_NUM_IRQS 64
148 
149 /* Legacy RAM limit in GB (< version 4.0) */
150 #define LEGACY_RAMLIMIT_GB 255
151 #define LEGACY_RAMLIMIT_BYTES (LEGACY_RAMLIMIT_GB * GiB)
152 
153 /* Addresses and sizes of our components.
154  * 0..128MB is space for a flash device so we can run bootrom code such as UEFI.
155  * 128MB..256MB is used for miscellaneous device I/O.
156  * 256MB..1GB is reserved for possible future PCI support (ie where the
157  * PCI memory window will go if we add a PCI host controller).
158  * 1GB and up is RAM (which may happily spill over into the
159  * high memory region beyond 4GB).
160  * This represents a compromise between how much RAM can be given to
161  * a 32 bit VM and leaving space for expansion and in particular for PCI.
162  * Note that devices should generally be placed at multiples of 0x10000,
163  * to accommodate guests using 64K pages.
164  */
165 static const MemMapEntry base_memmap[] = {
166     /* Space up to 0x8000000 is reserved for a boot ROM */
167     [VIRT_FLASH] =              {          0, 0x08000000 },
168     [VIRT_CPUPERIPHS] =         { 0x08000000, 0x00020000 },
169     /* GIC distributor and CPU interfaces sit inside the CPU peripheral space */
170     [VIRT_GIC_DIST] =           { 0x08000000, 0x00010000 },
171     [VIRT_GIC_CPU] =            { 0x08010000, 0x00010000 },
172     [VIRT_GIC_V2M] =            { 0x08020000, 0x00001000 },
173     [VIRT_GIC_HYP] =            { 0x08030000, 0x00010000 },
174     [VIRT_GIC_VCPU] =           { 0x08040000, 0x00010000 },
175     /* The space in between here is reserved for GICv3 CPU/vCPU/HYP */
176     [VIRT_GIC_ITS] =            { 0x08080000, 0x00020000 },
177     /* This redistributor space allows up to 2*64kB*123 CPUs */
178     [VIRT_GIC_REDIST] =         { 0x080A0000, 0x00F60000 },
179     [VIRT_UART0] =              { 0x09000000, 0x00001000 },
180     [VIRT_RTC] =                { 0x09010000, 0x00001000 },
181     [VIRT_FW_CFG] =             { 0x09020000, 0x00000018 },
182     [VIRT_GPIO] =               { 0x09030000, 0x00001000 },
183     [VIRT_UART1] =              { 0x09040000, 0x00001000 },
184     [VIRT_SMMU] =               { 0x09050000, 0x00020000 },
185     [VIRT_PCDIMM_ACPI] =        { 0x09070000, MEMORY_HOTPLUG_IO_LEN },
186     [VIRT_ACPI_GED] =           { 0x09080000, ACPI_GED_EVT_SEL_LEN },
187     [VIRT_NVDIMM_ACPI] =        { 0x09090000, NVDIMM_ACPI_IO_LEN},
188     [VIRT_PVTIME] =             { 0x090a0000, 0x00010000 },
189     [VIRT_SECURE_GPIO] =        { 0x090b0000, 0x00001000 },
190     [VIRT_ACPI_PCIHP] =         { 0x090c0000, ACPI_PCIHP_SIZE },
191     [VIRT_MMIO] =               { 0x0a000000, 0x00000200 },
192     /* ...repeating for a total of NUM_VIRTIO_TRANSPORTS, each of that size */
193     [VIRT_PLATFORM_BUS] =       { 0x0c000000, 0x02000000 },
194     [VIRT_SECURE_MEM] =         { 0x0e000000, 0x01000000 },
195     [VIRT_PCIE_MMIO] =          { 0x10000000, 0x2eff0000 },
196     [VIRT_PCIE_PIO] =           { 0x3eff0000, 0x00010000 },
197     [VIRT_PCIE_ECAM] =          { 0x3f000000, 0x01000000 },
198     /* Actual RAM size depends on initial RAM and device memory settings */
199     [VIRT_MEM] =                { GiB, LEGACY_RAMLIMIT_BYTES },
200 };
201 
202 /* Update the docs for highmem-mmio-size when changing this default */
203 #define DEFAULT_HIGH_PCIE_MMIO_SIZE_GB 512
204 #define DEFAULT_HIGH_PCIE_MMIO_SIZE (DEFAULT_HIGH_PCIE_MMIO_SIZE_GB * GiB)
205 
206 /*
207  * Highmem IO Regions: This memory map is floating, located after the RAM.
208  * Each MemMapEntry base (GPA) will be dynamically computed, depending on the
209  * top of the RAM, so that its base get the same alignment as the size,
210  * ie. a 512GiB entry will be aligned on a 512GiB boundary. If there is
211  * less than 256GiB of RAM, the floating area starts at the 256GiB mark.
212  * Note the extended_memmap is sized so that it eventually also includes the
213  * base_memmap entries (VIRT_HIGH_GIC_REDIST2 index is greater than the last
214  * index of base_memmap).
215  *
216  * The memory map for these Highmem IO Regions can be in legacy or compact
217  * layout, depending on 'compact-highmem' property. With legacy layout, the
218  * PA space for one specific region is always reserved, even if the region
219  * has been disabled or doesn't fit into the PA space. However, the PA space
220  * for the region won't be reserved in these circumstances with compact layout.
221  *
222  * Note that the highmem-mmio-size property will update the high PCIE MMIO size
223  * field in this array.
224  */
225 static MemMapEntry extended_memmap[] = {
226     /* Additional 64 MB redist region (can contain up to 512 redistributors) */
227     [VIRT_HIGH_GIC_REDIST2] =   { 0x0, 64 * MiB },
228     [VIRT_CXL_HOST] =           { 0x0, 64 * KiB * 16 }, /* 16 UID */
229     [VIRT_HIGH_PCIE_ECAM] =     { 0x0, 256 * MiB },
230     /* Second PCIe window */
231     [VIRT_HIGH_PCIE_MMIO] =     { 0x0, DEFAULT_HIGH_PCIE_MMIO_SIZE },
232     /* Any CXL Fixed memory windows come here */
233 };
234 
235 static const int a15irqmap[] = {
236     [VIRT_UART0] = 1,
237     [VIRT_RTC] = 2,
238     [VIRT_PCIE] = 3, /* ... to 6 */
239     [VIRT_GPIO] = 7,
240     [VIRT_UART1] = 8,
241     [VIRT_ACPI_GED] = 9,
242     [VIRT_MMIO] = 16, /* ...to 16 + NUM_VIRTIO_TRANSPORTS - 1 */
243     [VIRT_GIC_V2M] = 48, /* ...to 48 + NUM_GICV2M_SPIS - 1 */
244     [VIRT_SMMU] = 74,    /* ...to 74 + NUM_SMMU_IRQS - 1 */
245     [VIRT_PLATFORM_BUS] = 112, /* ...to 112 + PLATFORM_BUS_NUM_IRQS -1 */
246 };
247 
248 static void create_randomness(MachineState *ms, const char *node)
249 {
250     struct {
251         uint64_t kaslr;
252         uint8_t rng[32];
253     } seed;
254 
255     if (qemu_guest_getrandom(&seed, sizeof(seed), NULL)) {
256         return;
257     }
258     qemu_fdt_setprop_u64(ms->fdt, node, "kaslr-seed", seed.kaslr);
259     qemu_fdt_setprop(ms->fdt, node, "rng-seed", seed.rng, sizeof(seed.rng));
260 }
261 
262 /*
263  * The CPU object always exposes the NS EL2 virt timer IRQ line,
264  * but we don't want to advertise it to the guest in the dtb or ACPI
265  * table unless it's really going to do something.
266  */
267 static bool ns_el2_virt_timer_present(void)
268 {
269     ARMCPU *cpu = ARM_CPU(qemu_get_cpu(0));
270     CPUARMState *env = &cpu->env;
271 
272     return arm_feature(env, ARM_FEATURE_AARCH64) &&
273         arm_feature(env, ARM_FEATURE_EL2) && cpu_isar_feature(aa64_vh, cpu);
274 }
275 
276 static void create_fdt(VirtMachineState *vms)
277 {
278     MachineState *ms = MACHINE(vms);
279     int nb_numa_nodes = ms->numa_state->num_nodes;
280     void *fdt = create_device_tree(&vms->fdt_size);
281 
282     if (!fdt) {
283         error_report("create_device_tree() failed");
284         exit(1);
285     }
286 
287     ms->fdt = fdt;
288 
289     /* Header */
290     qemu_fdt_setprop_string(fdt, "/", "compatible", "linux,dummy-virt");
291     qemu_fdt_setprop_cell(fdt, "/", "#address-cells", 0x2);
292     qemu_fdt_setprop_cell(fdt, "/", "#size-cells", 0x2);
293     qemu_fdt_setprop_string(fdt, "/", "model", "linux,dummy-virt");
294 
295     /*
296      * For QEMU, all DMA is coherent. Advertising this in the root node
297      * has two benefits:
298      *
299      * - It avoids potential bugs where we forget to mark a DMA
300      *   capable device as being dma-coherent
301      * - It avoids spurious warnings from the Linux kernel about
302      *   devices which can't do DMA at all
303      */
304     qemu_fdt_setprop(fdt, "/", "dma-coherent", NULL, 0);
305 
306     /* /chosen must exist for load_dtb to fill in necessary properties later */
307     qemu_fdt_add_subnode(fdt, "/chosen");
308     if (vms->dtb_randomness) {
309         create_randomness(ms, "/chosen");
310     }
311 
312     if (vms->secure) {
313         qemu_fdt_add_subnode(fdt, "/secure-chosen");
314         if (vms->dtb_randomness) {
315             create_randomness(ms, "/secure-chosen");
316         }
317     }
318 
319     qemu_fdt_add_subnode(fdt, "/aliases");
320 
321     /* Clock node, for the benefit of the UART. The kernel device tree
322      * binding documentation claims the PL011 node clock properties are
323      * optional but in practice if you omit them the kernel refuses to
324      * probe for the device.
325      */
326     vms->clock_phandle = qemu_fdt_alloc_phandle(fdt);
327     qemu_fdt_add_subnode(fdt, "/apb-pclk");
328     qemu_fdt_setprop_string(fdt, "/apb-pclk", "compatible", "fixed-clock");
329     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "#clock-cells", 0x0);
330     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "clock-frequency", 24000000);
331     qemu_fdt_setprop_string(fdt, "/apb-pclk", "clock-output-names",
332                                 "clk24mhz");
333     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "phandle", vms->clock_phandle);
334 
335     if (nb_numa_nodes > 0 && ms->numa_state->have_numa_distance) {
336         int size = nb_numa_nodes * nb_numa_nodes * 3 * sizeof(uint32_t);
337         uint32_t *matrix = g_malloc0(size);
338         int idx, i, j;
339 
340         for (i = 0; i < nb_numa_nodes; i++) {
341             for (j = 0; j < nb_numa_nodes; j++) {
342                 idx = (i * nb_numa_nodes + j) * 3;
343                 matrix[idx + 0] = cpu_to_be32(i);
344                 matrix[idx + 1] = cpu_to_be32(j);
345                 matrix[idx + 2] =
346                     cpu_to_be32(ms->numa_state->nodes[i].distance[j]);
347             }
348         }
349 
350         qemu_fdt_add_subnode(fdt, "/distance-map");
351         qemu_fdt_setprop_string(fdt, "/distance-map", "compatible",
352                                 "numa-distance-map-v1");
353         qemu_fdt_setprop(fdt, "/distance-map", "distance-matrix",
354                          matrix, size);
355         g_free(matrix);
356     }
357 }
358 
359 static void fdt_add_timer_nodes(const VirtMachineState *vms)
360 {
361     /* On real hardware these interrupts are level-triggered.
362      * On KVM they were edge-triggered before host kernel version 4.4,
363      * and level-triggered afterwards.
364      * On emulated QEMU they are level-triggered.
365      *
366      * Getting the DTB info about them wrong is awkward for some
367      * guest kernels:
368      *  pre-4.8 ignore the DT and leave the interrupt configured
369      *   with whatever the GIC reset value (or the bootloader) left it at
370      *  4.8 before rc6 honour the incorrect data by programming it back
371      *   into the GIC, causing problems
372      *  4.8rc6 and later ignore the DT and always write "level triggered"
373      *   into the GIC
374      *
375      * For backwards-compatibility, virt-2.8 and earlier will continue
376      * to say these are edge-triggered, but later machines will report
377      * the correct information.
378      */
379     ARMCPU *armcpu;
380     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
381     MachineState *ms = MACHINE(vms);
382 
383     if (vms->gic_version == VIRT_GIC_VERSION_2) {
384         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
385                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
386                              (1 << MACHINE(vms)->smp.cpus) - 1);
387     }
388 
389     qemu_fdt_add_subnode(ms->fdt, "/timer");
390 
391     armcpu = ARM_CPU(qemu_get_cpu(0));
392     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
393         const char compat[] = "arm,armv8-timer\0arm,armv7-timer";
394         qemu_fdt_setprop(ms->fdt, "/timer", "compatible",
395                          compat, sizeof(compat));
396     } else {
397         qemu_fdt_setprop_string(ms->fdt, "/timer", "compatible",
398                                 "arm,armv7-timer");
399     }
400     qemu_fdt_setprop(ms->fdt, "/timer", "always-on", NULL, 0);
401     if (vms->ns_el2_virt_timer_irq) {
402         qemu_fdt_setprop_cells(ms->fdt, "/timer", "interrupts",
403                                GIC_FDT_IRQ_TYPE_PPI,
404                                INTID_TO_PPI(ARCH_TIMER_S_EL1_IRQ), irqflags,
405                                GIC_FDT_IRQ_TYPE_PPI,
406                                INTID_TO_PPI(ARCH_TIMER_NS_EL1_IRQ), irqflags,
407                                GIC_FDT_IRQ_TYPE_PPI,
408                                INTID_TO_PPI(ARCH_TIMER_VIRT_IRQ), irqflags,
409                                GIC_FDT_IRQ_TYPE_PPI,
410                                INTID_TO_PPI(ARCH_TIMER_NS_EL2_IRQ), irqflags,
411                                GIC_FDT_IRQ_TYPE_PPI,
412                                INTID_TO_PPI(ARCH_TIMER_NS_EL2_VIRT_IRQ), irqflags);
413     } else {
414         qemu_fdt_setprop_cells(ms->fdt, "/timer", "interrupts",
415                                GIC_FDT_IRQ_TYPE_PPI,
416                                INTID_TO_PPI(ARCH_TIMER_S_EL1_IRQ), irqflags,
417                                GIC_FDT_IRQ_TYPE_PPI,
418                                INTID_TO_PPI(ARCH_TIMER_NS_EL1_IRQ), irqflags,
419                                GIC_FDT_IRQ_TYPE_PPI,
420                                INTID_TO_PPI(ARCH_TIMER_VIRT_IRQ), irqflags,
421                                GIC_FDT_IRQ_TYPE_PPI,
422                                INTID_TO_PPI(ARCH_TIMER_NS_EL2_IRQ), irqflags);
423     }
424 }
425 
426 static void fdt_add_cpu_nodes(const VirtMachineState *vms)
427 {
428     int cpu;
429     int addr_cells = 1;
430     const MachineState *ms = MACHINE(vms);
431     const VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
432     int smp_cpus = ms->smp.cpus;
433 
434     /*
435      * See Linux Documentation/devicetree/bindings/arm/cpus.yaml
436      * On ARM v8 64-bit systems value should be set to 2,
437      * that corresponds to the MPIDR_EL1 register size.
438      * If MPIDR_EL1[63:32] value is equal to 0 on all CPUs
439      * in the system, #address-cells can be set to 1, since
440      * MPIDR_EL1[63:32] bits are not used for CPUs
441      * identification.
442      *
443      * Here we actually don't know whether our system is 32- or 64-bit one.
444      * The simplest way to go is to examine affinity IDs of all our CPUs. If
445      * at least one of them has Aff3 populated, we set #address-cells to 2.
446      */
447     for (cpu = 0; cpu < smp_cpus; cpu++) {
448         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
449 
450         if (arm_cpu_mp_affinity(armcpu) & ARM_AFF3_MASK) {
451             addr_cells = 2;
452             break;
453         }
454     }
455 
456     qemu_fdt_add_subnode(ms->fdt, "/cpus");
457     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#address-cells", addr_cells);
458     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#size-cells", 0x0);
459 
460     for (cpu = smp_cpus - 1; cpu >= 0; cpu--) {
461         char *nodename = g_strdup_printf("/cpus/cpu@%d", cpu);
462         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
463         CPUState *cs = CPU(armcpu);
464 
465         qemu_fdt_add_subnode(ms->fdt, nodename);
466         qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "cpu");
467         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
468                                     armcpu->dtb_compatible);
469 
470         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED && smp_cpus > 1) {
471             qemu_fdt_setprop_string(ms->fdt, nodename,
472                                         "enable-method", "psci");
473         }
474 
475         if (addr_cells == 2) {
476             qemu_fdt_setprop_u64(ms->fdt, nodename, "reg",
477                                  arm_cpu_mp_affinity(armcpu));
478         } else {
479             qemu_fdt_setprop_cell(ms->fdt, nodename, "reg",
480                                   arm_cpu_mp_affinity(armcpu));
481         }
482 
483         if (ms->possible_cpus->cpus[cs->cpu_index].props.has_node_id) {
484             qemu_fdt_setprop_cell(ms->fdt, nodename, "numa-node-id",
485                 ms->possible_cpus->cpus[cs->cpu_index].props.node_id);
486         }
487 
488         if (!vmc->no_cpu_topology) {
489             qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle",
490                                   qemu_fdt_alloc_phandle(ms->fdt));
491         }
492 
493         g_free(nodename);
494     }
495 
496     if (!vmc->no_cpu_topology) {
497         /*
498          * Add vCPU topology description through fdt node cpu-map.
499          *
500          * See Linux Documentation/devicetree/bindings/cpu/cpu-topology.txt
501          * In a SMP system, the hierarchy of CPUs can be defined through
502          * four entities that are used to describe the layout of CPUs in
503          * the system: socket/cluster/core/thread.
504          *
505          * A socket node represents the boundary of system physical package
506          * and its child nodes must be one or more cluster nodes. A system
507          * can contain several layers of clustering within a single physical
508          * package and cluster nodes can be contained in parent cluster nodes.
509          *
510          * Note: currently we only support one layer of clustering within
511          * each physical package.
512          */
513         qemu_fdt_add_subnode(ms->fdt, "/cpus/cpu-map");
514 
515         for (cpu = smp_cpus - 1; cpu >= 0; cpu--) {
516             char *cpu_path = g_strdup_printf("/cpus/cpu@%d", cpu);
517             char *map_path;
518 
519             if (ms->smp.threads > 1) {
520                 map_path = g_strdup_printf(
521                     "/cpus/cpu-map/socket%d/cluster%d/core%d/thread%d",
522                     cpu / (ms->smp.clusters * ms->smp.cores * ms->smp.threads),
523                     (cpu / (ms->smp.cores * ms->smp.threads)) % ms->smp.clusters,
524                     (cpu / ms->smp.threads) % ms->smp.cores,
525                     cpu % ms->smp.threads);
526             } else {
527                 map_path = g_strdup_printf(
528                     "/cpus/cpu-map/socket%d/cluster%d/core%d",
529                     cpu / (ms->smp.clusters * ms->smp.cores),
530                     (cpu / ms->smp.cores) % ms->smp.clusters,
531                     cpu % ms->smp.cores);
532             }
533             qemu_fdt_add_path(ms->fdt, map_path);
534             qemu_fdt_setprop_phandle(ms->fdt, map_path, "cpu", cpu_path);
535 
536             g_free(map_path);
537             g_free(cpu_path);
538         }
539     }
540 }
541 
542 static void fdt_add_its_gic_node(VirtMachineState *vms)
543 {
544     char *nodename;
545     MachineState *ms = MACHINE(vms);
546 
547     vms->msi_phandle = qemu_fdt_alloc_phandle(ms->fdt);
548     nodename = g_strdup_printf("/intc/its@%" PRIx64,
549                                vms->memmap[VIRT_GIC_ITS].base);
550     qemu_fdt_add_subnode(ms->fdt, nodename);
551     qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
552                             "arm,gic-v3-its");
553     qemu_fdt_setprop(ms->fdt, nodename, "msi-controller", NULL, 0);
554     qemu_fdt_setprop_cell(ms->fdt, nodename, "#msi-cells", 1);
555     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
556                                  2, vms->memmap[VIRT_GIC_ITS].base,
557                                  2, vms->memmap[VIRT_GIC_ITS].size);
558     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->msi_phandle);
559     g_free(nodename);
560 }
561 
562 static void fdt_add_v2m_gic_node(VirtMachineState *vms)
563 {
564     MachineState *ms = MACHINE(vms);
565     char *nodename;
566 
567     nodename = g_strdup_printf("/intc/v2m@%" PRIx64,
568                                vms->memmap[VIRT_GIC_V2M].base);
569     vms->msi_phandle = qemu_fdt_alloc_phandle(ms->fdt);
570     qemu_fdt_add_subnode(ms->fdt, nodename);
571     qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
572                             "arm,gic-v2m-frame");
573     qemu_fdt_setprop(ms->fdt, nodename, "msi-controller", NULL, 0);
574     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
575                                  2, vms->memmap[VIRT_GIC_V2M].base,
576                                  2, vms->memmap[VIRT_GIC_V2M].size);
577     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->msi_phandle);
578     g_free(nodename);
579 }
580 
581 static void fdt_add_gic_node(VirtMachineState *vms)
582 {
583     MachineState *ms = MACHINE(vms);
584     char *nodename;
585 
586     vms->gic_phandle = qemu_fdt_alloc_phandle(ms->fdt);
587     qemu_fdt_setprop_cell(ms->fdt, "/", "interrupt-parent", vms->gic_phandle);
588 
589     nodename = g_strdup_printf("/intc@%" PRIx64,
590                                vms->memmap[VIRT_GIC_DIST].base);
591     qemu_fdt_add_subnode(ms->fdt, nodename);
592     qemu_fdt_setprop_cell(ms->fdt, nodename, "#interrupt-cells", 3);
593     qemu_fdt_setprop(ms->fdt, nodename, "interrupt-controller", NULL, 0);
594     qemu_fdt_setprop_cell(ms->fdt, nodename, "#address-cells", 0x2);
595     qemu_fdt_setprop_cell(ms->fdt, nodename, "#size-cells", 0x2);
596     qemu_fdt_setprop(ms->fdt, nodename, "ranges", NULL, 0);
597     if (vms->gic_version != VIRT_GIC_VERSION_2) {
598         int nb_redist_regions = virt_gicv3_redist_region_count(vms);
599 
600         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
601                                 "arm,gic-v3");
602 
603         qemu_fdt_setprop_cell(ms->fdt, nodename,
604                               "#redistributor-regions", nb_redist_regions);
605 
606         if (nb_redist_regions == 1) {
607             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
608                                          2, vms->memmap[VIRT_GIC_DIST].base,
609                                          2, vms->memmap[VIRT_GIC_DIST].size,
610                                          2, vms->memmap[VIRT_GIC_REDIST].base,
611                                          2, vms->memmap[VIRT_GIC_REDIST].size);
612         } else {
613             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
614                                  2, vms->memmap[VIRT_GIC_DIST].base,
615                                  2, vms->memmap[VIRT_GIC_DIST].size,
616                                  2, vms->memmap[VIRT_GIC_REDIST].base,
617                                  2, vms->memmap[VIRT_GIC_REDIST].size,
618                                  2, vms->memmap[VIRT_HIGH_GIC_REDIST2].base,
619                                  2, vms->memmap[VIRT_HIGH_GIC_REDIST2].size);
620         }
621 
622         if (vms->virt) {
623             qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
624                                    GIC_FDT_IRQ_TYPE_PPI,
625                                    INTID_TO_PPI(ARCH_GIC_MAINT_IRQ),
626                                    GIC_FDT_IRQ_FLAGS_LEVEL_HI);
627         }
628     } else {
629         /* 'cortex-a15-gic' means 'GIC v2' */
630         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
631                                 "arm,cortex-a15-gic");
632         if (!vms->virt) {
633             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
634                                          2, vms->memmap[VIRT_GIC_DIST].base,
635                                          2, vms->memmap[VIRT_GIC_DIST].size,
636                                          2, vms->memmap[VIRT_GIC_CPU].base,
637                                          2, vms->memmap[VIRT_GIC_CPU].size);
638         } else {
639             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
640                                          2, vms->memmap[VIRT_GIC_DIST].base,
641                                          2, vms->memmap[VIRT_GIC_DIST].size,
642                                          2, vms->memmap[VIRT_GIC_CPU].base,
643                                          2, vms->memmap[VIRT_GIC_CPU].size,
644                                          2, vms->memmap[VIRT_GIC_HYP].base,
645                                          2, vms->memmap[VIRT_GIC_HYP].size,
646                                          2, vms->memmap[VIRT_GIC_VCPU].base,
647                                          2, vms->memmap[VIRT_GIC_VCPU].size);
648             qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
649                                    GIC_FDT_IRQ_TYPE_PPI,
650                                    INTID_TO_PPI(ARCH_GIC_MAINT_IRQ),
651                                    GIC_FDT_IRQ_FLAGS_LEVEL_HI);
652         }
653     }
654 
655     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->gic_phandle);
656     g_free(nodename);
657 }
658 
659 static void fdt_add_pmu_nodes(const VirtMachineState *vms)
660 {
661     ARMCPU *armcpu = ARM_CPU(first_cpu);
662     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
663     MachineState *ms = MACHINE(vms);
664 
665     if (!arm_feature(&armcpu->env, ARM_FEATURE_PMU)) {
666         assert(!object_property_get_bool(OBJECT(armcpu), "pmu", NULL));
667         return;
668     }
669 
670     if (vms->gic_version == VIRT_GIC_VERSION_2) {
671         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
672                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
673                              (1 << MACHINE(vms)->smp.cpus) - 1);
674     }
675 
676     qemu_fdt_add_subnode(ms->fdt, "/pmu");
677     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
678         const char compat[] = "arm,armv8-pmuv3";
679         qemu_fdt_setprop(ms->fdt, "/pmu", "compatible",
680                          compat, sizeof(compat));
681         qemu_fdt_setprop_cells(ms->fdt, "/pmu", "interrupts",
682                                GIC_FDT_IRQ_TYPE_PPI,
683                                INTID_TO_PPI(VIRTUAL_PMU_IRQ), irqflags);
684     }
685 }
686 
687 static inline DeviceState *create_acpi_ged(VirtMachineState *vms)
688 {
689     DeviceState *dev;
690     MachineState *ms = MACHINE(vms);
691     int irq = vms->irqmap[VIRT_ACPI_GED];
692     uint32_t event = ACPI_GED_PWR_DOWN_EVT;
693 
694     if (ms->ram_slots) {
695         event |= ACPI_GED_MEM_HOTPLUG_EVT;
696     }
697 
698     if (ms->nvdimms_state->is_enabled) {
699         event |= ACPI_GED_NVDIMM_HOTPLUG_EVT;
700     }
701 
702     dev = qdev_new(TYPE_ACPI_GED);
703     qdev_prop_set_uint32(dev, "ged-event", event);
704     object_property_set_link(OBJECT(dev), "bus", OBJECT(vms->bus), &error_abort);
705     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
706 
707     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_ACPI_GED].base);
708     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 1, vms->memmap[VIRT_PCDIMM_ACPI].base);
709     sysbus_connect_irq(SYS_BUS_DEVICE(dev), 0, qdev_get_gpio_in(vms->gic, irq));
710 
711     return dev;
712 }
713 
714 static void create_its(VirtMachineState *vms)
715 {
716     DeviceState *dev;
717 
718     assert(vms->its);
719     if (!kvm_irqchip_in_kernel() && !vms->tcg_its) {
720         /*
721          * Do nothing if ITS is neither supported by the host nor emulated by
722          * the machine.
723          */
724         return;
725     }
726 
727     dev = qdev_new(its_class_name());
728 
729     object_property_set_link(OBJECT(dev), "parent-gicv3", OBJECT(vms->gic),
730                              &error_abort);
731     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
732     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_ITS].base);
733 
734     fdt_add_its_gic_node(vms);
735     vms->msi_controller = VIRT_MSI_CTRL_ITS;
736 }
737 
738 static void create_v2m(VirtMachineState *vms)
739 {
740     int i;
741     int irq = vms->irqmap[VIRT_GIC_V2M];
742     DeviceState *dev;
743 
744     dev = qdev_new("arm-gicv2m");
745     qdev_prop_set_uint32(dev, "base-spi", irq);
746     qdev_prop_set_uint32(dev, "num-spi", NUM_GICV2M_SPIS);
747     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
748     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_V2M].base);
749 
750     for (i = 0; i < NUM_GICV2M_SPIS; i++) {
751         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
752                            qdev_get_gpio_in(vms->gic, irq + i));
753     }
754 
755     fdt_add_v2m_gic_node(vms);
756     vms->msi_controller = VIRT_MSI_CTRL_GICV2M;
757 }
758 
759 /*
760  * If the CPU has FEAT_NMI, then turn on the NMI support in the GICv3 too.
761  * It's permitted to have a configuration with NMI in the CPU (and thus the
762  * GICv3 CPU interface) but not in the distributor/redistributors, but it's
763  * not very useful.
764  */
765 static bool gicv3_nmi_present(VirtMachineState *vms)
766 {
767     ARMCPU *cpu = ARM_CPU(qemu_get_cpu(0));
768 
769     return tcg_enabled() && cpu_isar_feature(aa64_nmi, cpu) &&
770            (vms->gic_version != VIRT_GIC_VERSION_2);
771 }
772 
773 static void create_gic(VirtMachineState *vms, MemoryRegion *mem)
774 {
775     MachineState *ms = MACHINE(vms);
776     /* We create a standalone GIC */
777     SysBusDevice *gicbusdev;
778     const char *gictype;
779     int i;
780     unsigned int smp_cpus = ms->smp.cpus;
781     uint32_t nb_redist_regions = 0;
782     int revision;
783 
784     if (vms->gic_version == VIRT_GIC_VERSION_2) {
785         gictype = gic_class_name();
786     } else {
787         gictype = gicv3_class_name();
788     }
789 
790     switch (vms->gic_version) {
791     case VIRT_GIC_VERSION_2:
792         revision = 2;
793         break;
794     case VIRT_GIC_VERSION_3:
795         revision = 3;
796         break;
797     case VIRT_GIC_VERSION_4:
798         revision = 4;
799         break;
800     default:
801         g_assert_not_reached();
802     }
803 
804     if (kvm_enabled() && vms->virt &&
805         (revision != 3 || !kvm_irqchip_in_kernel())) {
806         error_report("KVM EL2 is only supported with in-kernel GICv3");
807         exit(1);
808     }
809 
810     vms->gic = qdev_new(gictype);
811     qdev_prop_set_uint32(vms->gic, "revision", revision);
812     qdev_prop_set_uint32(vms->gic, "num-cpu", smp_cpus);
813     /* Note that the num-irq property counts both internal and external
814      * interrupts; there are always 32 of the former (mandated by GIC spec).
815      */
816     qdev_prop_set_uint32(vms->gic, "num-irq", NUM_IRQS + 32);
817     if (!kvm_irqchip_in_kernel()) {
818         qdev_prop_set_bit(vms->gic, "has-security-extensions", vms->secure);
819     }
820 
821     if (vms->gic_version != VIRT_GIC_VERSION_2) {
822         QList *redist_region_count;
823         uint32_t redist0_capacity = virt_redist_capacity(vms, VIRT_GIC_REDIST);
824         uint32_t redist0_count = MIN(smp_cpus, redist0_capacity);
825 
826         nb_redist_regions = virt_gicv3_redist_region_count(vms);
827 
828         redist_region_count = qlist_new();
829         qlist_append_int(redist_region_count, redist0_count);
830         if (nb_redist_regions == 2) {
831             uint32_t redist1_capacity =
832                 virt_redist_capacity(vms, VIRT_HIGH_GIC_REDIST2);
833 
834             qlist_append_int(redist_region_count,
835                 MIN(smp_cpus - redist0_count, redist1_capacity));
836         }
837         qdev_prop_set_array(vms->gic, "redist-region-count",
838                             redist_region_count);
839 
840         if (!kvm_irqchip_in_kernel()) {
841             if (vms->tcg_its) {
842                 object_property_set_link(OBJECT(vms->gic), "sysmem",
843                                          OBJECT(mem), &error_fatal);
844                 qdev_prop_set_bit(vms->gic, "has-lpi", true);
845             }
846         } else if (vms->virt) {
847             qdev_prop_set_uint32(vms->gic, "maintenance-interrupt-id",
848                                  ARCH_GIC_MAINT_IRQ);
849         }
850     } else {
851         if (!kvm_irqchip_in_kernel()) {
852             qdev_prop_set_bit(vms->gic, "has-virtualization-extensions",
853                               vms->virt);
854         }
855     }
856 
857     if (gicv3_nmi_present(vms)) {
858         qdev_prop_set_bit(vms->gic, "has-nmi", true);
859     }
860 
861     gicbusdev = SYS_BUS_DEVICE(vms->gic);
862     sysbus_realize_and_unref(gicbusdev, &error_fatal);
863     sysbus_mmio_map(gicbusdev, 0, vms->memmap[VIRT_GIC_DIST].base);
864     if (vms->gic_version != VIRT_GIC_VERSION_2) {
865         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_REDIST].base);
866         if (nb_redist_regions == 2) {
867             sysbus_mmio_map(gicbusdev, 2,
868                             vms->memmap[VIRT_HIGH_GIC_REDIST2].base);
869         }
870     } else {
871         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base);
872         if (vms->virt) {
873             sysbus_mmio_map(gicbusdev, 2, vms->memmap[VIRT_GIC_HYP].base);
874             sysbus_mmio_map(gicbusdev, 3, vms->memmap[VIRT_GIC_VCPU].base);
875         }
876     }
877 
878     /* Wire the outputs from each CPU's generic timer and the GICv3
879      * maintenance interrupt signal to the appropriate GIC PPI inputs,
880      * and the GIC's IRQ/FIQ/VIRQ/VFIQ/NMI/VINMI interrupt outputs to the
881      * CPU's inputs.
882      */
883     for (i = 0; i < smp_cpus; i++) {
884         DeviceState *cpudev = DEVICE(qemu_get_cpu(i));
885         int intidbase = NUM_IRQS + i * GIC_INTERNAL;
886         /* Mapping from the output timer irq lines from the CPU to the
887          * GIC PPI inputs we use for the virt board.
888          */
889         const int timer_irq[] = {
890             [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ,
891             [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ,
892             [GTIMER_HYP]  = ARCH_TIMER_NS_EL2_IRQ,
893             [GTIMER_SEC]  = ARCH_TIMER_S_EL1_IRQ,
894             [GTIMER_HYPVIRT] = ARCH_TIMER_NS_EL2_VIRT_IRQ,
895             [GTIMER_S_EL2_PHYS] = ARCH_TIMER_S_EL2_IRQ,
896             [GTIMER_S_EL2_VIRT] = ARCH_TIMER_S_EL2_VIRT_IRQ,
897         };
898 
899         for (unsigned irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) {
900             qdev_connect_gpio_out(cpudev, irq,
901                                   qdev_get_gpio_in(vms->gic,
902                                                    intidbase + timer_irq[irq]));
903         }
904 
905         if (vms->gic_version != VIRT_GIC_VERSION_2) {
906             qemu_irq irq = qdev_get_gpio_in(vms->gic,
907                                             intidbase + ARCH_GIC_MAINT_IRQ);
908             qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt",
909                                         0, irq);
910         } else if (vms->virt) {
911             qemu_irq irq = qdev_get_gpio_in(vms->gic,
912                                             intidbase + ARCH_GIC_MAINT_IRQ);
913             sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus, irq);
914         }
915 
916         qdev_connect_gpio_out_named(cpudev, "pmu-interrupt", 0,
917                                     qdev_get_gpio_in(vms->gic, intidbase
918                                                      + VIRTUAL_PMU_IRQ));
919 
920         sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ));
921         sysbus_connect_irq(gicbusdev, i + smp_cpus,
922                            qdev_get_gpio_in(cpudev, ARM_CPU_FIQ));
923         sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus,
924                            qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ));
925         sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus,
926                            qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ));
927 
928         if (vms->gic_version != VIRT_GIC_VERSION_2) {
929             sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus,
930                                qdev_get_gpio_in(cpudev, ARM_CPU_NMI));
931             sysbus_connect_irq(gicbusdev, i + 5 * smp_cpus,
932                                qdev_get_gpio_in(cpudev, ARM_CPU_VINMI));
933         }
934     }
935 
936     fdt_add_gic_node(vms);
937 
938     if (vms->gic_version != VIRT_GIC_VERSION_2 && vms->its) {
939         create_its(vms);
940     } else if (vms->gic_version == VIRT_GIC_VERSION_2) {
941         create_v2m(vms);
942     }
943 }
944 
945 static void create_uart(const VirtMachineState *vms, int uart,
946                         MemoryRegion *mem, Chardev *chr, bool secure)
947 {
948     char *nodename;
949     hwaddr base = vms->memmap[uart].base;
950     hwaddr size = vms->memmap[uart].size;
951     int irq = vms->irqmap[uart];
952     const char compat[] = "arm,pl011\0arm,primecell";
953     const char clocknames[] = "uartclk\0apb_pclk";
954     DeviceState *dev = qdev_new(TYPE_PL011);
955     SysBusDevice *s = SYS_BUS_DEVICE(dev);
956     MachineState *ms = MACHINE(vms);
957 
958     qdev_prop_set_chr(dev, "chardev", chr);
959     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
960     memory_region_add_subregion(mem, base,
961                                 sysbus_mmio_get_region(s, 0));
962     sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
963 
964     nodename = g_strdup_printf("/pl011@%" PRIx64, base);
965     qemu_fdt_add_subnode(ms->fdt, nodename);
966     /* Note that we can't use setprop_string because of the embedded NUL */
967     qemu_fdt_setprop(ms->fdt, nodename, "compatible",
968                          compat, sizeof(compat));
969     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
970                                      2, base, 2, size);
971     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
972                                GIC_FDT_IRQ_TYPE_SPI, irq,
973                                GIC_FDT_IRQ_FLAGS_LEVEL_HI);
974     qemu_fdt_setprop_cells(ms->fdt, nodename, "clocks",
975                                vms->clock_phandle, vms->clock_phandle);
976     qemu_fdt_setprop(ms->fdt, nodename, "clock-names",
977                          clocknames, sizeof(clocknames));
978 
979     if (uart == VIRT_UART0) {
980         qemu_fdt_setprop_string(ms->fdt, "/chosen", "stdout-path", nodename);
981         qemu_fdt_setprop_string(ms->fdt, "/aliases", "serial0", nodename);
982     } else {
983         qemu_fdt_setprop_string(ms->fdt, "/aliases", "serial1", nodename);
984     }
985     if (secure) {
986         /* Mark as not usable by the normal world */
987         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
988         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
989 
990         qemu_fdt_setprop_string(ms->fdt, "/secure-chosen", "stdout-path",
991                                 nodename);
992     }
993 
994     g_free(nodename);
995 }
996 
997 static void create_rtc(const VirtMachineState *vms)
998 {
999     char *nodename;
1000     hwaddr base = vms->memmap[VIRT_RTC].base;
1001     hwaddr size = vms->memmap[VIRT_RTC].size;
1002     int irq = vms->irqmap[VIRT_RTC];
1003     const char compat[] = "arm,pl031\0arm,primecell";
1004     MachineState *ms = MACHINE(vms);
1005 
1006     sysbus_create_simple("pl031", base, qdev_get_gpio_in(vms->gic, irq));
1007 
1008     nodename = g_strdup_printf("/pl031@%" PRIx64, base);
1009     qemu_fdt_add_subnode(ms->fdt, nodename);
1010     qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
1011     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1012                                  2, base, 2, size);
1013     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1014                            GIC_FDT_IRQ_TYPE_SPI, irq,
1015                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
1016     qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
1017     qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
1018     g_free(nodename);
1019 }
1020 
1021 static DeviceState *gpio_key_dev;
1022 static void virt_powerdown_req(Notifier *n, void *opaque)
1023 {
1024     VirtMachineState *s = container_of(n, VirtMachineState, powerdown_notifier);
1025 
1026     if (s->acpi_dev) {
1027         acpi_send_event(s->acpi_dev, ACPI_POWER_DOWN_STATUS);
1028     } else {
1029         /* use gpio Pin for power button event */
1030         qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1);
1031     }
1032 }
1033 
1034 static void create_gpio_keys(char *fdt, DeviceState *pl061_dev,
1035                              uint32_t phandle)
1036 {
1037     gpio_key_dev = sysbus_create_simple("gpio-key", -1,
1038                                         qdev_get_gpio_in(pl061_dev,
1039                                                          GPIO_PIN_POWER_BUTTON));
1040 
1041     qemu_fdt_add_subnode(fdt, "/gpio-keys");
1042     qemu_fdt_setprop_string(fdt, "/gpio-keys", "compatible", "gpio-keys");
1043 
1044     qemu_fdt_add_subnode(fdt, "/gpio-keys/poweroff");
1045     qemu_fdt_setprop_string(fdt, "/gpio-keys/poweroff",
1046                             "label", "GPIO Key Poweroff");
1047     qemu_fdt_setprop_cell(fdt, "/gpio-keys/poweroff", "linux,code",
1048                           KEY_POWER);
1049     qemu_fdt_setprop_cells(fdt, "/gpio-keys/poweroff",
1050                            "gpios", phandle, GPIO_PIN_POWER_BUTTON, 0);
1051 }
1052 
1053 #define SECURE_GPIO_POWEROFF 0
1054 #define SECURE_GPIO_RESET    1
1055 
1056 static void create_secure_gpio_pwr(char *fdt, DeviceState *pl061_dev,
1057                                    uint32_t phandle)
1058 {
1059     DeviceState *gpio_pwr_dev;
1060 
1061     /* gpio-pwr */
1062     gpio_pwr_dev = sysbus_create_simple("gpio-pwr", -1, NULL);
1063 
1064     /* connect secure pl061 to gpio-pwr */
1065     qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_RESET,
1066                           qdev_get_gpio_in_named(gpio_pwr_dev, "reset", 0));
1067     qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_POWEROFF,
1068                           qdev_get_gpio_in_named(gpio_pwr_dev, "shutdown", 0));
1069 
1070     qemu_fdt_add_subnode(fdt, "/gpio-poweroff");
1071     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "compatible",
1072                             "gpio-poweroff");
1073     qemu_fdt_setprop_cells(fdt, "/gpio-poweroff",
1074                            "gpios", phandle, SECURE_GPIO_POWEROFF, 0);
1075     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "status", "disabled");
1076     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "secure-status",
1077                             "okay");
1078 
1079     qemu_fdt_add_subnode(fdt, "/gpio-restart");
1080     qemu_fdt_setprop_string(fdt, "/gpio-restart", "compatible",
1081                             "gpio-restart");
1082     qemu_fdt_setprop_cells(fdt, "/gpio-restart",
1083                            "gpios", phandle, SECURE_GPIO_RESET, 0);
1084     qemu_fdt_setprop_string(fdt, "/gpio-restart", "status", "disabled");
1085     qemu_fdt_setprop_string(fdt, "/gpio-restart", "secure-status",
1086                             "okay");
1087 }
1088 
1089 static void create_gpio_devices(const VirtMachineState *vms, int gpio,
1090                                 MemoryRegion *mem)
1091 {
1092     char *nodename;
1093     DeviceState *pl061_dev;
1094     hwaddr base = vms->memmap[gpio].base;
1095     hwaddr size = vms->memmap[gpio].size;
1096     int irq = vms->irqmap[gpio];
1097     const char compat[] = "arm,pl061\0arm,primecell";
1098     SysBusDevice *s;
1099     MachineState *ms = MACHINE(vms);
1100 
1101     pl061_dev = qdev_new("pl061");
1102     /* Pull lines down to 0 if not driven by the PL061 */
1103     qdev_prop_set_uint32(pl061_dev, "pullups", 0);
1104     qdev_prop_set_uint32(pl061_dev, "pulldowns", 0xff);
1105     s = SYS_BUS_DEVICE(pl061_dev);
1106     sysbus_realize_and_unref(s, &error_fatal);
1107     memory_region_add_subregion(mem, base, sysbus_mmio_get_region(s, 0));
1108     sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
1109 
1110     uint32_t phandle = qemu_fdt_alloc_phandle(ms->fdt);
1111     nodename = g_strdup_printf("/pl061@%" PRIx64, base);
1112     qemu_fdt_add_subnode(ms->fdt, nodename);
1113     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1114                                  2, base, 2, size);
1115     qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
1116     qemu_fdt_setprop_cell(ms->fdt, nodename, "#gpio-cells", 2);
1117     qemu_fdt_setprop(ms->fdt, nodename, "gpio-controller", NULL, 0);
1118     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1119                            GIC_FDT_IRQ_TYPE_SPI, irq,
1120                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
1121     qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
1122     qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
1123     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", phandle);
1124 
1125     if (gpio != VIRT_GPIO) {
1126         /* Mark as not usable by the normal world */
1127         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1128         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1129     }
1130     g_free(nodename);
1131 
1132     /* Child gpio devices */
1133     if (gpio == VIRT_GPIO) {
1134         create_gpio_keys(ms->fdt, pl061_dev, phandle);
1135     } else {
1136         create_secure_gpio_pwr(ms->fdt, pl061_dev, phandle);
1137     }
1138 }
1139 
1140 static void create_virtio_devices(const VirtMachineState *vms)
1141 {
1142     int i;
1143     hwaddr size = vms->memmap[VIRT_MMIO].size;
1144     MachineState *ms = MACHINE(vms);
1145 
1146     /* We create the transports in forwards order. Since qbus_realize()
1147      * prepends (not appends) new child buses, the incrementing loop below will
1148      * create a list of virtio-mmio buses with decreasing base addresses.
1149      *
1150      * When a -device option is processed from the command line,
1151      * qbus_find_recursive() picks the next free virtio-mmio bus in forwards
1152      * order. The upshot is that -device options in increasing command line
1153      * order are mapped to virtio-mmio buses with decreasing base addresses.
1154      *
1155      * When this code was originally written, that arrangement ensured that the
1156      * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to
1157      * the first -device on the command line. (The end-to-end order is a
1158      * function of this loop, qbus_realize(), qbus_find_recursive(), and the
1159      * guest kernel's name-to-address assignment strategy.)
1160      *
1161      * Meanwhile, the kernel's traversal seems to have been reversed; see eg.
1162      * the message, if not necessarily the code, of commit 70161ff336.
1163      * Therefore the loop now establishes the inverse of the original intent.
1164      *
1165      * Unfortunately, we can't counteract the kernel change by reversing the
1166      * loop; it would break existing command lines.
1167      *
1168      * In any case, the kernel makes no guarantee about the stability of
1169      * enumeration order of virtio devices (as demonstrated by it changing
1170      * between kernel versions). For reliable and stable identification
1171      * of disks users must use UUIDs or similar mechanisms.
1172      */
1173     for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) {
1174         int irq = vms->irqmap[VIRT_MMIO] + i;
1175         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1176 
1177         sysbus_create_simple("virtio-mmio", base,
1178                              qdev_get_gpio_in(vms->gic, irq));
1179     }
1180 
1181     /* We add dtb nodes in reverse order so that they appear in the finished
1182      * device tree lowest address first.
1183      *
1184      * Note that this mapping is independent of the loop above. The previous
1185      * loop influences virtio device to virtio transport assignment, whereas
1186      * this loop controls how virtio transports are laid out in the dtb.
1187      */
1188     for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) {
1189         char *nodename;
1190         int irq = vms->irqmap[VIRT_MMIO] + i;
1191         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1192 
1193         nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base);
1194         qemu_fdt_add_subnode(ms->fdt, nodename);
1195         qemu_fdt_setprop_string(ms->fdt, nodename,
1196                                 "compatible", "virtio,mmio");
1197         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1198                                      2, base, 2, size);
1199         qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1200                                GIC_FDT_IRQ_TYPE_SPI, irq,
1201                                GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1202         qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1203         g_free(nodename);
1204     }
1205 }
1206 
1207 #define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
1208 
1209 static PFlashCFI01 *virt_flash_create1(VirtMachineState *vms,
1210                                         const char *name,
1211                                         const char *alias_prop_name)
1212 {
1213     /*
1214      * Create a single flash device.  We use the same parameters as
1215      * the flash devices on the Versatile Express board.
1216      */
1217     DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
1218 
1219     qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
1220     qdev_prop_set_uint8(dev, "width", 4);
1221     qdev_prop_set_uint8(dev, "device-width", 2);
1222     qdev_prop_set_bit(dev, "big-endian", false);
1223     qdev_prop_set_uint16(dev, "id0", 0x89);
1224     qdev_prop_set_uint16(dev, "id1", 0x18);
1225     qdev_prop_set_uint16(dev, "id2", 0x00);
1226     qdev_prop_set_uint16(dev, "id3", 0x00);
1227     qdev_prop_set_string(dev, "name", name);
1228     object_property_add_child(OBJECT(vms), name, OBJECT(dev));
1229     object_property_add_alias(OBJECT(vms), alias_prop_name,
1230                               OBJECT(dev), "drive");
1231     return PFLASH_CFI01(dev);
1232 }
1233 
1234 static void virt_flash_create(VirtMachineState *vms)
1235 {
1236     vms->flash[0] = virt_flash_create1(vms, "virt.flash0", "pflash0");
1237     vms->flash[1] = virt_flash_create1(vms, "virt.flash1", "pflash1");
1238 }
1239 
1240 static void virt_flash_map1(PFlashCFI01 *flash,
1241                             hwaddr base, hwaddr size,
1242                             MemoryRegion *sysmem)
1243 {
1244     DeviceState *dev = DEVICE(flash);
1245 
1246     assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
1247     assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
1248     qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
1249     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1250 
1251     memory_region_add_subregion(sysmem, base,
1252                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
1253                                                        0));
1254 }
1255 
1256 static void virt_flash_map(VirtMachineState *vms,
1257                            MemoryRegion *sysmem,
1258                            MemoryRegion *secure_sysmem)
1259 {
1260     /*
1261      * Map two flash devices to fill the VIRT_FLASH space in the memmap.
1262      * sysmem is the system memory space. secure_sysmem is the secure view
1263      * of the system, and the first flash device should be made visible only
1264      * there. The second flash device is visible to both secure and nonsecure.
1265      * If sysmem == secure_sysmem this means there is no separate Secure
1266      * address space and both flash devices are generally visible.
1267      */
1268     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1269     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1270 
1271     virt_flash_map1(vms->flash[0], flashbase, flashsize,
1272                     secure_sysmem);
1273     virt_flash_map1(vms->flash[1], flashbase + flashsize, flashsize,
1274                     sysmem);
1275 }
1276 
1277 static void virt_flash_fdt(VirtMachineState *vms,
1278                            MemoryRegion *sysmem,
1279                            MemoryRegion *secure_sysmem)
1280 {
1281     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1282     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1283     MachineState *ms = MACHINE(vms);
1284     char *nodename;
1285 
1286     if (sysmem == secure_sysmem) {
1287         /* Report both flash devices as a single node in the DT */
1288         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
1289         qemu_fdt_add_subnode(ms->fdt, nodename);
1290         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1291         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1292                                      2, flashbase, 2, flashsize,
1293                                      2, flashbase + flashsize, 2, flashsize);
1294         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1295         g_free(nodename);
1296     } else {
1297         /*
1298          * Report the devices as separate nodes so we can mark one as
1299          * only visible to the secure world.
1300          */
1301         nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase);
1302         qemu_fdt_add_subnode(ms->fdt, nodename);
1303         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1304         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1305                                      2, flashbase, 2, flashsize);
1306         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1307         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1308         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1309         g_free(nodename);
1310 
1311         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase + flashsize);
1312         qemu_fdt_add_subnode(ms->fdt, nodename);
1313         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1314         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1315                                      2, flashbase + flashsize, 2, flashsize);
1316         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1317         g_free(nodename);
1318     }
1319 }
1320 
1321 static bool virt_firmware_init(VirtMachineState *vms,
1322                                MemoryRegion *sysmem,
1323                                MemoryRegion *secure_sysmem)
1324 {
1325     int i;
1326     const char *bios_name;
1327     BlockBackend *pflash_blk0;
1328 
1329     /* Map legacy -drive if=pflash to machine properties */
1330     for (i = 0; i < ARRAY_SIZE(vms->flash); i++) {
1331         pflash_cfi01_legacy_drive(vms->flash[i],
1332                                   drive_get(IF_PFLASH, 0, i));
1333     }
1334 
1335     virt_flash_map(vms, sysmem, secure_sysmem);
1336 
1337     pflash_blk0 = pflash_cfi01_get_blk(vms->flash[0]);
1338 
1339     bios_name = MACHINE(vms)->firmware;
1340     if (bios_name) {
1341         char *fname;
1342         MemoryRegion *mr;
1343         int image_size;
1344 
1345         if (pflash_blk0) {
1346             error_report("The contents of the first flash device may be "
1347                          "specified with -bios or with -drive if=pflash... "
1348                          "but you cannot use both options at once");
1349             exit(1);
1350         }
1351 
1352         /* Fall back to -bios */
1353 
1354         fname = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1355         if (!fname) {
1356             error_report("Could not find ROM image '%s'", bios_name);
1357             exit(1);
1358         }
1359         mr = sysbus_mmio_get_region(SYS_BUS_DEVICE(vms->flash[0]), 0);
1360         image_size = load_image_mr(fname, mr);
1361         g_free(fname);
1362         if (image_size < 0) {
1363             error_report("Could not load ROM image '%s'", bios_name);
1364             exit(1);
1365         }
1366     }
1367 
1368     return pflash_blk0 || bios_name;
1369 }
1370 
1371 static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as)
1372 {
1373     MachineState *ms = MACHINE(vms);
1374     hwaddr base = vms->memmap[VIRT_FW_CFG].base;
1375     hwaddr size = vms->memmap[VIRT_FW_CFG].size;
1376     FWCfgState *fw_cfg;
1377     char *nodename;
1378 
1379     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as);
1380     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)ms->smp.cpus);
1381 
1382     nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
1383     qemu_fdt_add_subnode(ms->fdt, nodename);
1384     qemu_fdt_setprop_string(ms->fdt, nodename,
1385                             "compatible", "qemu,fw-cfg-mmio");
1386     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1387                                  2, base, 2, size);
1388     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1389     g_free(nodename);
1390     return fw_cfg;
1391 }
1392 
1393 static void create_pcie_irq_map(const MachineState *ms,
1394                                 uint32_t gic_phandle,
1395                                 int first_irq, const char *nodename)
1396 {
1397     int devfn, pin;
1398     uint32_t full_irq_map[4 * 4 * 10] = { 0 };
1399     uint32_t *irq_map = full_irq_map;
1400 
1401     for (devfn = 0; devfn <= 0x18; devfn += 0x8) {
1402         for (pin = 0; pin < 4; pin++) {
1403             int irq_type = GIC_FDT_IRQ_TYPE_SPI;
1404             int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS);
1405             int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
1406             int i;
1407 
1408             uint32_t map[] = {
1409                 devfn << 8, 0, 0,                           /* devfn */
1410                 pin + 1,                                    /* PCI pin */
1411                 gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */
1412 
1413             /* Convert map to big endian */
1414             for (i = 0; i < 10; i++) {
1415                 irq_map[i] = cpu_to_be32(map[i]);
1416             }
1417             irq_map += 10;
1418         }
1419     }
1420 
1421     qemu_fdt_setprop(ms->fdt, nodename, "interrupt-map",
1422                      full_irq_map, sizeof(full_irq_map));
1423 
1424     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupt-map-mask",
1425                            cpu_to_be16(PCI_DEVFN(3, 0)), /* Slot 3 */
1426                            0, 0,
1427                            0x7           /* PCI irq */);
1428 }
1429 
1430 static void create_smmu(const VirtMachineState *vms,
1431                         PCIBus *bus)
1432 {
1433     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1434     char *node;
1435     const char compat[] = "arm,smmu-v3";
1436     int irq =  vms->irqmap[VIRT_SMMU];
1437     int i;
1438     hwaddr base = vms->memmap[VIRT_SMMU].base;
1439     hwaddr size = vms->memmap[VIRT_SMMU].size;
1440     const char irq_names[] = "eventq\0priq\0cmdq-sync\0gerror";
1441     DeviceState *dev;
1442     MachineState *ms = MACHINE(vms);
1443 
1444     if (vms->iommu != VIRT_IOMMU_SMMUV3 || !vms->iommu_phandle) {
1445         return;
1446     }
1447 
1448     dev = qdev_new(TYPE_ARM_SMMUV3);
1449 
1450     if (!vmc->no_nested_smmu) {
1451         object_property_set_str(OBJECT(dev), "stage", "nested", &error_fatal);
1452     }
1453     object_property_set_link(OBJECT(dev), "primary-bus", OBJECT(bus),
1454                              &error_abort);
1455     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1456     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, base);
1457     for (i = 0; i < NUM_SMMU_IRQS; i++) {
1458         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1459                            qdev_get_gpio_in(vms->gic, irq + i));
1460     }
1461 
1462     node = g_strdup_printf("/smmuv3@%" PRIx64, base);
1463     qemu_fdt_add_subnode(ms->fdt, node);
1464     qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1465     qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg", 2, base, 2, size);
1466 
1467     qemu_fdt_setprop_cells(ms->fdt, node, "interrupts",
1468             GIC_FDT_IRQ_TYPE_SPI, irq    , GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1469             GIC_FDT_IRQ_TYPE_SPI, irq + 1, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1470             GIC_FDT_IRQ_TYPE_SPI, irq + 2, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1471             GIC_FDT_IRQ_TYPE_SPI, irq + 3, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1472 
1473     qemu_fdt_setprop(ms->fdt, node, "interrupt-names", irq_names,
1474                      sizeof(irq_names));
1475 
1476     qemu_fdt_setprop(ms->fdt, node, "dma-coherent", NULL, 0);
1477 
1478     qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1479 
1480     qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1481     g_free(node);
1482 }
1483 
1484 static void create_virtio_iommu_dt_bindings(VirtMachineState *vms)
1485 {
1486     const char compat[] = "virtio,pci-iommu\0pci1af4,1057";
1487     uint16_t bdf = vms->virtio_iommu_bdf;
1488     MachineState *ms = MACHINE(vms);
1489     char *node;
1490 
1491     vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1492 
1493     node = g_strdup_printf("%s/virtio_iommu@%x,%x", vms->pciehb_nodename,
1494                            PCI_SLOT(bdf), PCI_FUNC(bdf));
1495     qemu_fdt_add_subnode(ms->fdt, node);
1496     qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1497     qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg",
1498                                  1, bdf << 8, 1, 0, 1, 0,
1499                                  1, 0, 1, 0);
1500 
1501     qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1502     qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1503     g_free(node);
1504 
1505     if (!vms->default_bus_bypass_iommu) {
1506         qemu_fdt_setprop_cells(ms->fdt, vms->pciehb_nodename, "iommu-map",
1507                                0x0, vms->iommu_phandle, 0x0, bdf,
1508                                bdf + 1, vms->iommu_phandle, bdf + 1,
1509                                0xffff - bdf);
1510     }
1511 }
1512 
1513 static void create_pcie(VirtMachineState *vms)
1514 {
1515     hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base;
1516     hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size;
1517     hwaddr base_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].base;
1518     hwaddr size_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].size;
1519     hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base;
1520     hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size;
1521     hwaddr base_ecam, size_ecam;
1522     hwaddr base = base_mmio;
1523     int nr_pcie_buses;
1524     int irq = vms->irqmap[VIRT_PCIE];
1525     MemoryRegion *mmio_alias;
1526     MemoryRegion *mmio_reg;
1527     MemoryRegion *ecam_alias;
1528     MemoryRegion *ecam_reg;
1529     DeviceState *dev;
1530     char *nodename;
1531     int i, ecam_id;
1532     PCIHostState *pci;
1533     MachineState *ms = MACHINE(vms);
1534     MachineClass *mc = MACHINE_GET_CLASS(ms);
1535 
1536     dev = qdev_new(TYPE_GPEX_HOST);
1537     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1538 
1539     ecam_id = VIRT_ECAM_ID(vms->highmem_ecam);
1540     base_ecam = vms->memmap[ecam_id].base;
1541     size_ecam = vms->memmap[ecam_id].size;
1542     nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN;
1543     /* Map only the first size_ecam bytes of ECAM space */
1544     ecam_alias = g_new0(MemoryRegion, 1);
1545     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1546     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1547                              ecam_reg, 0, size_ecam);
1548     memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias);
1549 
1550     /* Map the MMIO window into system address space so as to expose
1551      * the section of PCI MMIO space which starts at the same base address
1552      * (ie 1:1 mapping for that part of PCI MMIO space visible through
1553      * the window).
1554      */
1555     mmio_alias = g_new0(MemoryRegion, 1);
1556     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1557     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1558                              mmio_reg, base_mmio, size_mmio);
1559     memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias);
1560 
1561     if (vms->highmem_mmio) {
1562         /* Map high MMIO space */
1563         MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1);
1564 
1565         memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1566                                  mmio_reg, base_mmio_high, size_mmio_high);
1567         memory_region_add_subregion(get_system_memory(), base_mmio_high,
1568                                     high_mmio_alias);
1569     }
1570 
1571     /* Map IO port space */
1572     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio);
1573 
1574     for (i = 0; i < PCI_NUM_PINS; i++) {
1575         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1576                            qdev_get_gpio_in(vms->gic, irq + i));
1577         gpex_set_irq_num(GPEX_HOST(dev), i, irq + i);
1578     }
1579 
1580     pci = PCI_HOST_BRIDGE(dev);
1581     pci->bypass_iommu = vms->default_bus_bypass_iommu;
1582     vms->bus = pci->bus;
1583     if (vms->bus) {
1584         pci_init_nic_devices(pci->bus, mc->default_nic);
1585     }
1586 
1587     nodename = vms->pciehb_nodename = g_strdup_printf("/pcie@%" PRIx64, base);
1588     qemu_fdt_add_subnode(ms->fdt, nodename);
1589     qemu_fdt_setprop_string(ms->fdt, nodename,
1590                             "compatible", "pci-host-ecam-generic");
1591     qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "pci");
1592     qemu_fdt_setprop_cell(ms->fdt, nodename, "#address-cells", 3);
1593     qemu_fdt_setprop_cell(ms->fdt, nodename, "#size-cells", 2);
1594     qemu_fdt_setprop_cell(ms->fdt, nodename, "linux,pci-domain", 0);
1595     qemu_fdt_setprop_cells(ms->fdt, nodename, "bus-range", 0,
1596                            nr_pcie_buses - 1);
1597     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1598 
1599     if (vms->msi_phandle) {
1600         qemu_fdt_setprop_cells(ms->fdt, nodename, "msi-map",
1601                                0, vms->msi_phandle, 0, 0x10000);
1602     }
1603 
1604     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1605                                  2, base_ecam, 2, size_ecam);
1606 
1607     if (vms->highmem_mmio) {
1608         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1609                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1610                                      2, base_pio, 2, size_pio,
1611                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1612                                      2, base_mmio, 2, size_mmio,
1613                                      1, FDT_PCI_RANGE_MMIO_64BIT,
1614                                      2, base_mmio_high,
1615                                      2, base_mmio_high, 2, size_mmio_high);
1616     } else {
1617         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1618                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1619                                      2, base_pio, 2, size_pio,
1620                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1621                                      2, base_mmio, 2, size_mmio);
1622     }
1623 
1624     qemu_fdt_setprop_cell(ms->fdt, nodename, "#interrupt-cells", 1);
1625     create_pcie_irq_map(ms, vms->gic_phandle, irq, nodename);
1626 
1627     if (vms->iommu) {
1628         vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1629 
1630         switch (vms->iommu) {
1631         case VIRT_IOMMU_SMMUV3:
1632             create_smmu(vms, vms->bus);
1633             if (!vms->default_bus_bypass_iommu) {
1634                 qemu_fdt_setprop_cells(ms->fdt, nodename, "iommu-map",
1635                                        0x0, vms->iommu_phandle, 0x0, 0x10000);
1636             }
1637             break;
1638         default:
1639             g_assert_not_reached();
1640         }
1641     }
1642 }
1643 
1644 static void create_cxl_host_reg_region(VirtMachineState *vms)
1645 {
1646     MemoryRegion *sysmem = get_system_memory();
1647     MemoryRegion *mr = &vms->cxl_devices_state.host_mr;
1648 
1649     memory_region_init(mr, OBJECT(vms), "cxl_host_reg",
1650                        vms->memmap[VIRT_CXL_HOST].size);
1651     memory_region_add_subregion(sysmem, vms->memmap[VIRT_CXL_HOST].base, mr);
1652     vms->highmem_cxl = true;
1653 }
1654 
1655 static void create_platform_bus(VirtMachineState *vms)
1656 {
1657     DeviceState *dev;
1658     SysBusDevice *s;
1659     int i;
1660     MemoryRegion *sysmem = get_system_memory();
1661 
1662     dev = qdev_new(TYPE_PLATFORM_BUS_DEVICE);
1663     dev->id = g_strdup(TYPE_PLATFORM_BUS_DEVICE);
1664     qdev_prop_set_uint32(dev, "num_irqs", PLATFORM_BUS_NUM_IRQS);
1665     qdev_prop_set_uint32(dev, "mmio_size", vms->memmap[VIRT_PLATFORM_BUS].size);
1666     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1667     vms->platform_bus_dev = dev;
1668 
1669     s = SYS_BUS_DEVICE(dev);
1670     for (i = 0; i < PLATFORM_BUS_NUM_IRQS; i++) {
1671         int irq = vms->irqmap[VIRT_PLATFORM_BUS] + i;
1672         sysbus_connect_irq(s, i, qdev_get_gpio_in(vms->gic, irq));
1673     }
1674 
1675     memory_region_add_subregion(sysmem,
1676                                 vms->memmap[VIRT_PLATFORM_BUS].base,
1677                                 sysbus_mmio_get_region(s, 0));
1678 }
1679 
1680 static void create_tag_ram(MemoryRegion *tag_sysmem,
1681                            hwaddr base, hwaddr size,
1682                            const char *name)
1683 {
1684     MemoryRegion *tagram = g_new(MemoryRegion, 1);
1685 
1686     memory_region_init_ram(tagram, NULL, name, size / 32, &error_fatal);
1687     memory_region_add_subregion(tag_sysmem, base / 32, tagram);
1688 }
1689 
1690 static void create_secure_ram(VirtMachineState *vms,
1691                               MemoryRegion *secure_sysmem,
1692                               MemoryRegion *secure_tag_sysmem)
1693 {
1694     MemoryRegion *secram = g_new(MemoryRegion, 1);
1695     char *nodename;
1696     hwaddr base = vms->memmap[VIRT_SECURE_MEM].base;
1697     hwaddr size = vms->memmap[VIRT_SECURE_MEM].size;
1698     MachineState *ms = MACHINE(vms);
1699 
1700     memory_region_init_ram(secram, NULL, "virt.secure-ram", size,
1701                            &error_fatal);
1702     memory_region_add_subregion(secure_sysmem, base, secram);
1703 
1704     nodename = g_strdup_printf("/secram@%" PRIx64, base);
1705     qemu_fdt_add_subnode(ms->fdt, nodename);
1706     qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "memory");
1707     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg", 2, base, 2, size);
1708     qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1709     qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1710 
1711     if (secure_tag_sysmem) {
1712         create_tag_ram(secure_tag_sysmem, base, size, "mach-virt.secure-tag");
1713     }
1714 
1715     g_free(nodename);
1716 }
1717 
1718 static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size)
1719 {
1720     const VirtMachineState *board = container_of(binfo, VirtMachineState,
1721                                                  bootinfo);
1722     MachineState *ms = MACHINE(board);
1723 
1724 
1725     *fdt_size = board->fdt_size;
1726     return ms->fdt;
1727 }
1728 
1729 static void virt_build_smbios(VirtMachineState *vms)
1730 {
1731     MachineClass *mc = MACHINE_GET_CLASS(vms);
1732     MachineState *ms = MACHINE(vms);
1733     uint8_t *smbios_tables, *smbios_anchor;
1734     size_t smbios_tables_len, smbios_anchor_len;
1735     struct smbios_phys_mem_area mem_array;
1736     const char *product = "QEMU Virtual Machine";
1737 
1738     if (kvm_enabled()) {
1739         product = "KVM Virtual Machine";
1740     }
1741 
1742     smbios_set_defaults("QEMU", product, mc->name);
1743 
1744     /* build the array of physical mem area from base_memmap */
1745     mem_array.address = vms->memmap[VIRT_MEM].base;
1746     mem_array.length = ms->ram_size;
1747 
1748     smbios_get_tables(ms, SMBIOS_ENTRY_POINT_TYPE_64, &mem_array, 1,
1749                       &smbios_tables, &smbios_tables_len,
1750                       &smbios_anchor, &smbios_anchor_len,
1751                       &error_fatal);
1752 
1753     if (smbios_anchor) {
1754         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables",
1755                         smbios_tables, smbios_tables_len);
1756         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor",
1757                         smbios_anchor, smbios_anchor_len);
1758     }
1759 }
1760 
1761 static
1762 void virt_machine_done(Notifier *notifier, void *data)
1763 {
1764     VirtMachineState *vms = container_of(notifier, VirtMachineState,
1765                                          machine_done);
1766     MachineState *ms = MACHINE(vms);
1767     ARMCPU *cpu = ARM_CPU(first_cpu);
1768     struct arm_boot_info *info = &vms->bootinfo;
1769     AddressSpace *as = arm_boot_address_space(cpu, info);
1770 
1771     cxl_hook_up_pxb_registers(vms->bus, &vms->cxl_devices_state,
1772                               &error_fatal);
1773 
1774     if (vms->cxl_devices_state.is_enabled) {
1775         cxl_fmws_link_targets(&error_fatal);
1776     }
1777     /*
1778      * If the user provided a dtb, we assume the dynamic sysbus nodes
1779      * already are integrated there. This corresponds to a use case where
1780      * the dynamic sysbus nodes are complex and their generation is not yet
1781      * supported. In that case the user can take charge of the guest dt
1782      * while qemu takes charge of the qom stuff.
1783      */
1784     if (info->dtb_filename == NULL) {
1785         platform_bus_add_all_fdt_nodes(ms->fdt, "/intc",
1786                                        vms->memmap[VIRT_PLATFORM_BUS].base,
1787                                        vms->memmap[VIRT_PLATFORM_BUS].size,
1788                                        vms->irqmap[VIRT_PLATFORM_BUS]);
1789     }
1790     if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms, cpu) < 0) {
1791         exit(1);
1792     }
1793 
1794     pci_bus_add_fw_cfg_extra_pci_roots(vms->fw_cfg, vms->bus,
1795                                        &error_abort);
1796 
1797     virt_acpi_setup(vms);
1798     virt_build_smbios(vms);
1799 }
1800 
1801 static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx)
1802 {
1803     uint8_t clustersz;
1804 
1805     /*
1806      * Adjust MPIDR to make TCG consistent (with 64-bit KVM hosts)
1807      * and to improve SGI efficiency.
1808      */
1809     if (vms->gic_version == VIRT_GIC_VERSION_2) {
1810         clustersz = GIC_TARGETLIST_BITS;
1811     } else {
1812         clustersz = GICV3_TARGETLIST_BITS;
1813     }
1814 
1815     return arm_build_mp_affinity(idx, clustersz);
1816 }
1817 
1818 static inline bool *virt_get_high_memmap_enabled(VirtMachineState *vms,
1819                                                  int index)
1820 {
1821     bool *enabled_array[] = {
1822         &vms->highmem_redists,
1823         &vms->highmem_cxl,
1824         &vms->highmem_ecam,
1825         &vms->highmem_mmio,
1826     };
1827 
1828     assert(ARRAY_SIZE(extended_memmap) - VIRT_LOWMEMMAP_LAST ==
1829            ARRAY_SIZE(enabled_array));
1830     assert(index - VIRT_LOWMEMMAP_LAST < ARRAY_SIZE(enabled_array));
1831 
1832     return enabled_array[index - VIRT_LOWMEMMAP_LAST];
1833 }
1834 
1835 static void virt_set_high_memmap(VirtMachineState *vms,
1836                                  hwaddr base, int pa_bits)
1837 {
1838     hwaddr region_base, region_size;
1839     bool *region_enabled, fits;
1840     int i;
1841 
1842     for (i = VIRT_LOWMEMMAP_LAST; i < ARRAY_SIZE(extended_memmap); i++) {
1843         region_enabled = virt_get_high_memmap_enabled(vms, i);
1844         region_base = ROUND_UP(base, extended_memmap[i].size);
1845         region_size = extended_memmap[i].size;
1846 
1847         vms->memmap[i].base = region_base;
1848         vms->memmap[i].size = region_size;
1849 
1850         /*
1851          * Check each device to see if it fits in the PA space,
1852          * moving highest_gpa as we go. For compatibility, move
1853          * highest_gpa for disabled fitting devices as well, if
1854          * the compact layout has been disabled.
1855          *
1856          * For each device that doesn't fit, disable it.
1857          */
1858         fits = (region_base + region_size) <= BIT_ULL(pa_bits);
1859         *region_enabled &= fits;
1860         if (vms->highmem_compact && !*region_enabled) {
1861             continue;
1862         }
1863 
1864         base = region_base + region_size;
1865         if (fits) {
1866             vms->highest_gpa = base - 1;
1867         }
1868     }
1869 }
1870 
1871 static void virt_set_memmap(VirtMachineState *vms, int pa_bits)
1872 {
1873     MachineState *ms = MACHINE(vms);
1874     hwaddr base, device_memory_base, device_memory_size, memtop;
1875     int i;
1876 
1877     vms->memmap = extended_memmap;
1878 
1879     for (i = 0; i < ARRAY_SIZE(base_memmap); i++) {
1880         vms->memmap[i] = base_memmap[i];
1881     }
1882 
1883     if (ms->ram_slots > ACPI_MAX_RAM_SLOTS) {
1884         error_report("unsupported number of memory slots: %"PRIu64,
1885                      ms->ram_slots);
1886         exit(EXIT_FAILURE);
1887     }
1888 
1889     /*
1890      * !highmem is exactly the same as limiting the PA space to 32bit,
1891      * irrespective of the underlying capabilities of the HW.
1892      */
1893     if (!vms->highmem) {
1894         pa_bits = 32;
1895     }
1896 
1897     /*
1898      * We compute the base of the high IO region depending on the
1899      * amount of initial and device memory. The device memory start/size
1900      * is aligned on 1GiB. We never put the high IO region below 256GiB
1901      * so that if maxram_size is < 255GiB we keep the legacy memory map.
1902      * The device region size assumes 1GiB page max alignment per slot.
1903      */
1904     device_memory_base =
1905         ROUND_UP(vms->memmap[VIRT_MEM].base + ms->ram_size, GiB);
1906     device_memory_size = ms->maxram_size - ms->ram_size + ms->ram_slots * GiB;
1907 
1908     /* Base address of the high IO region */
1909     memtop = base = device_memory_base + ROUND_UP(device_memory_size, GiB);
1910     if (memtop > BIT_ULL(pa_bits)) {
1911         error_report("Addressing limited to %d bits, but memory exceeds it by %llu bytes",
1912                      pa_bits, memtop - BIT_ULL(pa_bits));
1913         exit(EXIT_FAILURE);
1914     }
1915     if (base < device_memory_base) {
1916         error_report("maxmem/slots too huge");
1917         exit(EXIT_FAILURE);
1918     }
1919     if (base < vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES) {
1920         base = vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES;
1921     }
1922 
1923     /* We know for sure that at least the memory fits in the PA space */
1924     vms->highest_gpa = memtop - 1;
1925 
1926     virt_set_high_memmap(vms, base, pa_bits);
1927 
1928     if (device_memory_size > 0) {
1929         machine_memory_devices_init(ms, device_memory_base, device_memory_size);
1930     }
1931     vms->highest_gpa = cxl_fmws_set_memmap(ROUND_UP(vms->highest_gpa + 1,
1932                                                     256 * MiB),
1933                                            BIT_ULL(pa_bits)) - 1;
1934 }
1935 
1936 static VirtGICType finalize_gic_version_do(const char *accel_name,
1937                                            VirtGICType gic_version,
1938                                            int gics_supported,
1939                                            unsigned int max_cpus)
1940 {
1941     /* Convert host/max/nosel to GIC version number */
1942     switch (gic_version) {
1943     case VIRT_GIC_VERSION_HOST:
1944         if (!kvm_enabled()) {
1945             error_report("gic-version=host requires KVM");
1946             exit(1);
1947         }
1948 
1949         /* For KVM, gic-version=host means gic-version=max */
1950         return finalize_gic_version_do(accel_name, VIRT_GIC_VERSION_MAX,
1951                                        gics_supported, max_cpus);
1952     case VIRT_GIC_VERSION_MAX:
1953         if (gics_supported & VIRT_GIC_VERSION_4_MASK) {
1954             gic_version = VIRT_GIC_VERSION_4;
1955         } else if (gics_supported & VIRT_GIC_VERSION_3_MASK) {
1956             gic_version = VIRT_GIC_VERSION_3;
1957         } else {
1958             gic_version = VIRT_GIC_VERSION_2;
1959         }
1960         break;
1961     case VIRT_GIC_VERSION_NOSEL:
1962         if ((gics_supported & VIRT_GIC_VERSION_2_MASK) &&
1963             max_cpus <= GIC_NCPU) {
1964             gic_version = VIRT_GIC_VERSION_2;
1965         } else if (gics_supported & VIRT_GIC_VERSION_3_MASK) {
1966             /*
1967              * in case the host does not support v2 emulation or
1968              * the end-user requested more than 8 VCPUs we now default
1969              * to v3. In any case defaulting to v2 would be broken.
1970              */
1971             gic_version = VIRT_GIC_VERSION_3;
1972         } else if (max_cpus > GIC_NCPU) {
1973             error_report("%s only supports GICv2 emulation but more than 8 "
1974                          "vcpus are requested", accel_name);
1975             exit(1);
1976         }
1977         break;
1978     case VIRT_GIC_VERSION_2:
1979     case VIRT_GIC_VERSION_3:
1980     case VIRT_GIC_VERSION_4:
1981         break;
1982     }
1983 
1984     /* Check chosen version is effectively supported */
1985     switch (gic_version) {
1986     case VIRT_GIC_VERSION_2:
1987         if (!(gics_supported & VIRT_GIC_VERSION_2_MASK)) {
1988             error_report("%s does not support GICv2 emulation", accel_name);
1989             exit(1);
1990         }
1991         break;
1992     case VIRT_GIC_VERSION_3:
1993         if (!(gics_supported & VIRT_GIC_VERSION_3_MASK)) {
1994             error_report("%s does not support GICv3 emulation", accel_name);
1995             exit(1);
1996         }
1997         break;
1998     case VIRT_GIC_VERSION_4:
1999         if (!(gics_supported & VIRT_GIC_VERSION_4_MASK)) {
2000             error_report("%s does not support GICv4 emulation, is virtualization=on?",
2001                          accel_name);
2002             exit(1);
2003         }
2004         break;
2005     default:
2006         error_report("logic error in finalize_gic_version");
2007         exit(1);
2008         break;
2009     }
2010 
2011     return gic_version;
2012 }
2013 
2014 /*
2015  * finalize_gic_version - Determines the final gic_version
2016  * according to the gic-version property
2017  *
2018  * Default GIC type is v2
2019  */
2020 static void finalize_gic_version(VirtMachineState *vms)
2021 {
2022     const char *accel_name = current_accel_name();
2023     unsigned int max_cpus = MACHINE(vms)->smp.max_cpus;
2024     int gics_supported = 0;
2025 
2026     /* Determine which GIC versions the current environment supports */
2027     if (kvm_enabled() && kvm_irqchip_in_kernel()) {
2028         int probe_bitmap = kvm_arm_vgic_probe();
2029 
2030         if (!probe_bitmap) {
2031             error_report("Unable to determine GIC version supported by host");
2032             exit(1);
2033         }
2034 
2035         if (probe_bitmap & KVM_ARM_VGIC_V2) {
2036             gics_supported |= VIRT_GIC_VERSION_2_MASK;
2037         }
2038         if (probe_bitmap & KVM_ARM_VGIC_V3) {
2039             gics_supported |= VIRT_GIC_VERSION_3_MASK;
2040         }
2041     } else if (kvm_enabled() && !kvm_irqchip_in_kernel()) {
2042         /* KVM w/o kernel irqchip can only deal with GICv2 */
2043         gics_supported |= VIRT_GIC_VERSION_2_MASK;
2044         accel_name = "KVM with kernel-irqchip=off";
2045     } else if (tcg_enabled() || hvf_enabled() || qtest_enabled())  {
2046         gics_supported |= VIRT_GIC_VERSION_2_MASK;
2047         if (module_object_class_by_name("arm-gicv3")) {
2048             gics_supported |= VIRT_GIC_VERSION_3_MASK;
2049             if (vms->virt) {
2050                 /* GICv4 only makes sense if CPU has EL2 */
2051                 gics_supported |= VIRT_GIC_VERSION_4_MASK;
2052             }
2053         }
2054     } else {
2055         error_report("Unsupported accelerator, can not determine GIC support");
2056         exit(1);
2057     }
2058 
2059     /*
2060      * Then convert helpers like host/max to concrete GIC versions and ensure
2061      * the desired version is supported
2062      */
2063     vms->gic_version = finalize_gic_version_do(accel_name, vms->gic_version,
2064                                                gics_supported, max_cpus);
2065 }
2066 
2067 /*
2068  * virt_post_cpus_gic_realized() must be called after the CPUs and
2069  * the GIC have both been realized.
2070  */
2071 static void virt_post_cpus_gic_realized(VirtMachineState *vms,
2072                                         MemoryRegion *sysmem)
2073 {
2074     int max_cpus = MACHINE(vms)->smp.max_cpus;
2075     bool aarch64, pmu, steal_time;
2076     CPUState *cpu;
2077 
2078     aarch64 = object_property_get_bool(OBJECT(first_cpu), "aarch64", NULL);
2079     pmu = object_property_get_bool(OBJECT(first_cpu), "pmu", NULL);
2080     steal_time = object_property_get_bool(OBJECT(first_cpu),
2081                                           "kvm-steal-time", NULL);
2082 
2083     if (kvm_enabled()) {
2084         hwaddr pvtime_reg_base = vms->memmap[VIRT_PVTIME].base;
2085         hwaddr pvtime_reg_size = vms->memmap[VIRT_PVTIME].size;
2086 
2087         if (steal_time) {
2088             MemoryRegion *pvtime = g_new(MemoryRegion, 1);
2089             hwaddr pvtime_size = max_cpus * PVTIME_SIZE_PER_CPU;
2090 
2091             /* The memory region size must be a multiple of host page size. */
2092             pvtime_size = REAL_HOST_PAGE_ALIGN(pvtime_size);
2093 
2094             if (pvtime_size > pvtime_reg_size) {
2095                 error_report("pvtime requires a %" HWADDR_PRId
2096                              " byte memory region for %d CPUs,"
2097                              " but only %" HWADDR_PRId " has been reserved",
2098                              pvtime_size, max_cpus, pvtime_reg_size);
2099                 exit(1);
2100             }
2101 
2102             memory_region_init_ram(pvtime, NULL, "pvtime", pvtime_size, NULL);
2103             memory_region_add_subregion(sysmem, pvtime_reg_base, pvtime);
2104         }
2105         if (!aarch64 && vms->virt) {
2106             error_report("KVM does not support EL2 on an AArch32 vCPU");
2107             exit(1);
2108         }
2109 
2110         CPU_FOREACH(cpu) {
2111             if (pmu) {
2112                 assert(arm_feature(&ARM_CPU(cpu)->env, ARM_FEATURE_PMU));
2113                 if (kvm_irqchip_in_kernel()) {
2114                     kvm_arm_pmu_set_irq(ARM_CPU(cpu), VIRTUAL_PMU_IRQ);
2115                 }
2116                 kvm_arm_pmu_init(ARM_CPU(cpu));
2117             }
2118             if (steal_time) {
2119                 kvm_arm_pvtime_init(ARM_CPU(cpu), pvtime_reg_base
2120                                                   + cpu->cpu_index
2121                                                     * PVTIME_SIZE_PER_CPU);
2122             }
2123         }
2124     } else {
2125         if (aarch64 && vms->highmem) {
2126             int requested_pa_size = 64 - clz64(vms->highest_gpa);
2127             int pamax = arm_pamax(ARM_CPU(first_cpu));
2128 
2129             if (pamax < requested_pa_size) {
2130                 error_report("VCPU supports less PA bits (%d) than "
2131                              "requested by the memory map (%d)",
2132                              pamax, requested_pa_size);
2133                 exit(1);
2134             }
2135         }
2136     }
2137 }
2138 
2139 static void machvirt_init(MachineState *machine)
2140 {
2141     VirtMachineState *vms = VIRT_MACHINE(machine);
2142     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine);
2143     MachineClass *mc = MACHINE_GET_CLASS(machine);
2144     const CPUArchIdList *possible_cpus;
2145     MemoryRegion *sysmem = get_system_memory();
2146     MemoryRegion *secure_sysmem = NULL;
2147     MemoryRegion *tag_sysmem = NULL;
2148     MemoryRegion *secure_tag_sysmem = NULL;
2149     int n, virt_max_cpus;
2150     bool firmware_loaded;
2151     bool aarch64 = true;
2152     bool has_ged = !vmc->no_ged;
2153     unsigned int smp_cpus = machine->smp.cpus;
2154     unsigned int max_cpus = machine->smp.max_cpus;
2155 
2156     possible_cpus = mc->possible_cpu_arch_ids(machine);
2157 
2158     /*
2159      * In accelerated mode, the memory map is computed earlier in kvm_type()
2160      * for Linux, or hvf_get_physical_address_range() for macOS to create a
2161      * VM with the right number of IPA bits.
2162      */
2163     if (!vms->memmap) {
2164         Object *cpuobj;
2165         ARMCPU *armcpu;
2166         int pa_bits;
2167 
2168         /*
2169          * Instantiate a temporary CPU object to find out about what
2170          * we are about to deal with. Once this is done, get rid of
2171          * the object.
2172          */
2173         cpuobj = object_new(possible_cpus->cpus[0].type);
2174         armcpu = ARM_CPU(cpuobj);
2175 
2176         pa_bits = arm_pamax(armcpu);
2177 
2178         object_unref(cpuobj);
2179 
2180         virt_set_memmap(vms, pa_bits);
2181     }
2182 
2183     /* We can probe only here because during property set
2184      * KVM is not available yet
2185      */
2186     finalize_gic_version(vms);
2187 
2188     if (vms->secure) {
2189         /*
2190          * The Secure view of the world is the same as the NonSecure,
2191          * but with a few extra devices. Create it as a container region
2192          * containing the system memory at low priority; any secure-only
2193          * devices go in at higher priority and take precedence.
2194          */
2195         secure_sysmem = g_new(MemoryRegion, 1);
2196         memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory",
2197                            UINT64_MAX);
2198         memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1);
2199     }
2200 
2201     firmware_loaded = virt_firmware_init(vms, sysmem,
2202                                          secure_sysmem ?: sysmem);
2203 
2204     /* If we have an EL3 boot ROM then the assumption is that it will
2205      * implement PSCI itself, so disable QEMU's internal implementation
2206      * so it doesn't get in the way. Instead of starting secondary
2207      * CPUs in PSCI powerdown state we will start them all running and
2208      * let the boot ROM sort them out.
2209      * The usual case is that we do use QEMU's PSCI implementation;
2210      * if the guest has EL2 then we will use SMC as the conduit,
2211      * and otherwise we will use HVC (for backwards compatibility and
2212      * because if we're using KVM then we must use HVC).
2213      */
2214     if (vms->secure && firmware_loaded) {
2215         vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
2216     } else if (vms->virt) {
2217         vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC;
2218     } else {
2219         vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC;
2220     }
2221 
2222     /*
2223      * The maximum number of CPUs depends on the GIC version, or on how
2224      * many redistributors we can fit into the memory map (which in turn
2225      * depends on whether this is a GICv3 or v4).
2226      */
2227     if (vms->gic_version == VIRT_GIC_VERSION_2) {
2228         virt_max_cpus = GIC_NCPU;
2229     } else {
2230         virt_max_cpus = virt_redist_capacity(vms, VIRT_GIC_REDIST);
2231         if (vms->highmem_redists) {
2232             virt_max_cpus += virt_redist_capacity(vms, VIRT_HIGH_GIC_REDIST2);
2233         }
2234     }
2235 
2236     if (max_cpus > virt_max_cpus) {
2237         error_report("Number of SMP CPUs requested (%d) exceeds max CPUs "
2238                      "supported by machine 'mach-virt' (%d)",
2239                      max_cpus, virt_max_cpus);
2240         if (vms->gic_version != VIRT_GIC_VERSION_2 && !vms->highmem_redists) {
2241             error_printf("Try 'highmem-redists=on' for more CPUs\n");
2242         }
2243 
2244         exit(1);
2245     }
2246 
2247     if (vms->secure && !tcg_enabled() && !qtest_enabled()) {
2248         error_report("mach-virt: %s does not support providing "
2249                      "Security extensions (TrustZone) to the guest CPU",
2250                      current_accel_name());
2251         exit(1);
2252     }
2253 
2254     if (vms->virt && kvm_enabled() && !kvm_arm_el2_supported()) {
2255         error_report("mach-virt: host kernel KVM does not support providing "
2256                      "Virtualization extensions to the guest CPU");
2257         exit(1);
2258     }
2259 
2260     if (vms->virt && !kvm_enabled() && !tcg_enabled() && !qtest_enabled()) {
2261         error_report("mach-virt: %s does not support providing "
2262                      "Virtualization extensions to the guest CPU",
2263                      current_accel_name());
2264         exit(1);
2265     }
2266 
2267     if (vms->mte && hvf_enabled()) {
2268         error_report("mach-virt: %s does not support providing "
2269                      "MTE to the guest CPU",
2270                      current_accel_name());
2271         exit(1);
2272     }
2273 
2274     create_fdt(vms);
2275 
2276     assert(possible_cpus->len == max_cpus);
2277     for (n = 0; n < possible_cpus->len; n++) {
2278         Object *cpuobj;
2279         CPUState *cs;
2280 
2281         if (n >= smp_cpus) {
2282             break;
2283         }
2284 
2285         cpuobj = object_new(possible_cpus->cpus[n].type);
2286         object_property_set_int(cpuobj, "mp-affinity",
2287                                 possible_cpus->cpus[n].arch_id, NULL);
2288 
2289         cs = CPU(cpuobj);
2290         cs->cpu_index = n;
2291 
2292         numa_cpu_pre_plug(&possible_cpus->cpus[cs->cpu_index], DEVICE(cpuobj),
2293                           &error_fatal);
2294 
2295         aarch64 &= object_property_get_bool(cpuobj, "aarch64", NULL);
2296 
2297         if (!vms->secure) {
2298             object_property_set_bool(cpuobj, "has_el3", false, NULL);
2299         }
2300 
2301         if (!vms->virt && object_property_find(cpuobj, "has_el2")) {
2302             object_property_set_bool(cpuobj, "has_el2", false, NULL);
2303         }
2304 
2305         if (vmc->kvm_no_adjvtime &&
2306             object_property_find(cpuobj, "kvm-no-adjvtime")) {
2307             object_property_set_bool(cpuobj, "kvm-no-adjvtime", true, NULL);
2308         }
2309 
2310         if (vmc->no_kvm_steal_time &&
2311             object_property_find(cpuobj, "kvm-steal-time")) {
2312             object_property_set_bool(cpuobj, "kvm-steal-time", false, NULL);
2313         }
2314 
2315         if (vmc->no_tcg_lpa2 && object_property_find(cpuobj, "lpa2")) {
2316             object_property_set_bool(cpuobj, "lpa2", false, NULL);
2317         }
2318 
2319         if (object_property_find(cpuobj, "reset-cbar")) {
2320             object_property_set_int(cpuobj, "reset-cbar",
2321                                     vms->memmap[VIRT_CPUPERIPHS].base,
2322                                     &error_abort);
2323         }
2324 
2325         object_property_set_link(cpuobj, "memory", OBJECT(sysmem),
2326                                  &error_abort);
2327         if (vms->secure) {
2328             object_property_set_link(cpuobj, "secure-memory",
2329                                      OBJECT(secure_sysmem), &error_abort);
2330         }
2331 
2332         if (vms->mte) {
2333             if (tcg_enabled()) {
2334                 /* Create the memory region only once, but link to all cpus. */
2335                 if (!tag_sysmem) {
2336                     /*
2337                      * The property exists only if MemTag is supported.
2338                      * If it is, we must allocate the ram to back that up.
2339                      */
2340                     if (!object_property_find(cpuobj, "tag-memory")) {
2341                         error_report("MTE requested, but not supported "
2342                                      "by the guest CPU");
2343                         exit(1);
2344                     }
2345 
2346                     tag_sysmem = g_new(MemoryRegion, 1);
2347                     memory_region_init(tag_sysmem, OBJECT(machine),
2348                                        "tag-memory", UINT64_MAX / 32);
2349 
2350                     if (vms->secure) {
2351                         secure_tag_sysmem = g_new(MemoryRegion, 1);
2352                         memory_region_init(secure_tag_sysmem, OBJECT(machine),
2353                                            "secure-tag-memory",
2354                                            UINT64_MAX / 32);
2355 
2356                         /* As with ram, secure-tag takes precedence over tag. */
2357                         memory_region_add_subregion_overlap(secure_tag_sysmem,
2358                                                             0, tag_sysmem, -1);
2359                     }
2360                 }
2361 
2362                 object_property_set_link(cpuobj, "tag-memory",
2363                                          OBJECT(tag_sysmem), &error_abort);
2364                 if (vms->secure) {
2365                     object_property_set_link(cpuobj, "secure-tag-memory",
2366                                              OBJECT(secure_tag_sysmem),
2367                                              &error_abort);
2368                 }
2369             } else if (kvm_enabled()) {
2370                 if (!kvm_arm_mte_supported()) {
2371                     error_report("MTE requested, but not supported by KVM");
2372                     exit(1);
2373                 }
2374                 kvm_arm_enable_mte(cpuobj, &error_abort);
2375             } else {
2376                     error_report("MTE requested, but not supported ");
2377                     exit(1);
2378             }
2379         }
2380 
2381         qdev_realize(DEVICE(cpuobj), NULL, &error_fatal);
2382         object_unref(cpuobj);
2383     }
2384 
2385     /* Now we've created the CPUs we can see if they have the hypvirt timer */
2386     vms->ns_el2_virt_timer_irq = ns_el2_virt_timer_present() &&
2387         !vmc->no_ns_el2_virt_timer_irq;
2388 
2389     fdt_add_timer_nodes(vms);
2390     fdt_add_cpu_nodes(vms);
2391 
2392     memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base,
2393                                 machine->ram);
2394 
2395     cxl_fmws_update_mmio();
2396 
2397     virt_flash_fdt(vms, sysmem, secure_sysmem ?: sysmem);
2398 
2399     create_gic(vms, sysmem);
2400 
2401     virt_post_cpus_gic_realized(vms, sysmem);
2402 
2403     fdt_add_pmu_nodes(vms);
2404 
2405     /*
2406      * The first UART always exists. If the security extensions are
2407      * enabled, the second UART also always exists. Otherwise, it only exists
2408      * if a backend is configured explicitly via '-serial <backend>'.
2409      * This avoids potentially breaking existing user setups that expect
2410      * only one NonSecure UART to be present (for instance, older EDK2
2411      * binaries).
2412      *
2413      * The nodes end up in the DTB in reverse order of creation, so we must
2414      * create UART0 last to ensure it appears as the first node in the DTB,
2415      * for compatibility with guest software that just iterates through the
2416      * DTB to find the first UART, as older versions of EDK2 do.
2417      * DTB readers that follow the spec, as Linux does, should honour the
2418      * aliases node information and /chosen/stdout-path regardless of
2419      * the order that nodes appear in the DTB.
2420      *
2421      * For similar back-compatibility reasons, if UART1 is the secure UART
2422      * we create it second (and so it appears first in the DTB), because
2423      * that's what QEMU has always done.
2424      */
2425     if (!vms->secure) {
2426         Chardev *serial1 = serial_hd(1);
2427 
2428         if (serial1) {
2429             vms->second_ns_uart_present = true;
2430             create_uart(vms, VIRT_UART1, sysmem, serial1, false);
2431         }
2432     }
2433     create_uart(vms, VIRT_UART0, sysmem, serial_hd(0), false);
2434     if (vms->secure) {
2435         create_uart(vms, VIRT_UART1, secure_sysmem, serial_hd(1), true);
2436     }
2437 
2438     if (vms->secure) {
2439         create_secure_ram(vms, secure_sysmem, secure_tag_sysmem);
2440     }
2441 
2442     if (tag_sysmem) {
2443         create_tag_ram(tag_sysmem, vms->memmap[VIRT_MEM].base,
2444                        machine->ram_size, "mach-virt.tag");
2445     }
2446 
2447     vms->highmem_ecam &= (!firmware_loaded || aarch64);
2448 
2449     create_rtc(vms);
2450 
2451     create_pcie(vms);
2452     create_cxl_host_reg_region(vms);
2453 
2454     if (has_ged && aarch64 && firmware_loaded && virt_is_acpi_enabled(vms)) {
2455         vms->acpi_dev = create_acpi_ged(vms);
2456     } else {
2457         create_gpio_devices(vms, VIRT_GPIO, sysmem);
2458     }
2459 
2460     if (vms->secure && !vmc->no_secure_gpio) {
2461         create_gpio_devices(vms, VIRT_SECURE_GPIO, secure_sysmem);
2462     }
2463 
2464      /* connect powerdown request */
2465      vms->powerdown_notifier.notify = virt_powerdown_req;
2466      qemu_register_powerdown_notifier(&vms->powerdown_notifier);
2467 
2468     /* Create mmio transports, so the user can create virtio backends
2469      * (which will be automatically plugged in to the transports). If
2470      * no backend is created the transport will just sit harmlessly idle.
2471      */
2472     create_virtio_devices(vms);
2473 
2474     vms->fw_cfg = create_fw_cfg(vms, &address_space_memory);
2475     rom_set_fw(vms->fw_cfg);
2476 
2477     create_platform_bus(vms);
2478 
2479     if (machine->nvdimms_state->is_enabled) {
2480         const struct AcpiGenericAddress arm_virt_nvdimm_acpi_dsmio = {
2481             .space_id = AML_AS_SYSTEM_MEMORY,
2482             .address = vms->memmap[VIRT_NVDIMM_ACPI].base,
2483             .bit_width = NVDIMM_ACPI_IO_LEN << 3
2484         };
2485 
2486         nvdimm_init_acpi_state(machine->nvdimms_state, sysmem,
2487                                arm_virt_nvdimm_acpi_dsmio,
2488                                vms->fw_cfg, OBJECT(vms));
2489     }
2490 
2491     vms->bootinfo.ram_size = machine->ram_size;
2492     vms->bootinfo.board_id = -1;
2493     vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base;
2494     vms->bootinfo.get_dtb = machvirt_dtb;
2495     vms->bootinfo.skip_dtb_autoload = true;
2496     vms->bootinfo.firmware_loaded = firmware_loaded;
2497     vms->bootinfo.psci_conduit = vms->psci_conduit;
2498     arm_load_kernel(ARM_CPU(first_cpu), machine, &vms->bootinfo);
2499 
2500     vms->machine_done.notify = virt_machine_done;
2501     qemu_add_machine_init_done_notifier(&vms->machine_done);
2502 }
2503 
2504 static bool virt_get_secure(Object *obj, Error **errp)
2505 {
2506     VirtMachineState *vms = VIRT_MACHINE(obj);
2507 
2508     return vms->secure;
2509 }
2510 
2511 static void virt_set_secure(Object *obj, bool value, Error **errp)
2512 {
2513     VirtMachineState *vms = VIRT_MACHINE(obj);
2514 
2515     vms->secure = value;
2516 }
2517 
2518 static bool virt_get_virt(Object *obj, Error **errp)
2519 {
2520     VirtMachineState *vms = VIRT_MACHINE(obj);
2521 
2522     return vms->virt;
2523 }
2524 
2525 static void virt_set_virt(Object *obj, bool value, Error **errp)
2526 {
2527     VirtMachineState *vms = VIRT_MACHINE(obj);
2528 
2529     vms->virt = value;
2530 }
2531 
2532 static bool virt_get_highmem(Object *obj, Error **errp)
2533 {
2534     VirtMachineState *vms = VIRT_MACHINE(obj);
2535 
2536     return vms->highmem;
2537 }
2538 
2539 static void virt_set_highmem(Object *obj, bool value, Error **errp)
2540 {
2541     VirtMachineState *vms = VIRT_MACHINE(obj);
2542 
2543     vms->highmem = value;
2544 }
2545 
2546 static bool virt_get_compact_highmem(Object *obj, Error **errp)
2547 {
2548     VirtMachineState *vms = VIRT_MACHINE(obj);
2549 
2550     return vms->highmem_compact;
2551 }
2552 
2553 static void virt_set_compact_highmem(Object *obj, bool value, Error **errp)
2554 {
2555     VirtMachineState *vms = VIRT_MACHINE(obj);
2556 
2557     vms->highmem_compact = value;
2558 }
2559 
2560 static bool virt_get_highmem_redists(Object *obj, Error **errp)
2561 {
2562     VirtMachineState *vms = VIRT_MACHINE(obj);
2563 
2564     return vms->highmem_redists;
2565 }
2566 
2567 static void virt_set_highmem_redists(Object *obj, bool value, Error **errp)
2568 {
2569     VirtMachineState *vms = VIRT_MACHINE(obj);
2570 
2571     vms->highmem_redists = value;
2572 }
2573 
2574 static bool virt_get_highmem_ecam(Object *obj, Error **errp)
2575 {
2576     VirtMachineState *vms = VIRT_MACHINE(obj);
2577 
2578     return vms->highmem_ecam;
2579 }
2580 
2581 static void virt_set_highmem_ecam(Object *obj, bool value, Error **errp)
2582 {
2583     VirtMachineState *vms = VIRT_MACHINE(obj);
2584 
2585     vms->highmem_ecam = value;
2586 }
2587 
2588 static bool virt_get_highmem_mmio(Object *obj, Error **errp)
2589 {
2590     VirtMachineState *vms = VIRT_MACHINE(obj);
2591 
2592     return vms->highmem_mmio;
2593 }
2594 
2595 static void virt_set_highmem_mmio(Object *obj, bool value, Error **errp)
2596 {
2597     VirtMachineState *vms = VIRT_MACHINE(obj);
2598 
2599     vms->highmem_mmio = value;
2600 }
2601 
2602 static void virt_get_highmem_mmio_size(Object *obj, Visitor *v,
2603                                        const char *name, void *opaque,
2604                                        Error **errp)
2605 {
2606     uint64_t size = extended_memmap[VIRT_HIGH_PCIE_MMIO].size;
2607 
2608     visit_type_size(v, name, &size, errp);
2609 }
2610 
2611 static void virt_set_highmem_mmio_size(Object *obj, Visitor *v,
2612                                        const char *name, void *opaque,
2613                                        Error **errp)
2614 {
2615     uint64_t size;
2616 
2617     if (!visit_type_size(v, name, &size, errp)) {
2618         return;
2619     }
2620 
2621     if (!is_power_of_2(size)) {
2622         error_setg(errp, "highmem-mmio-size is not a power of 2");
2623         return;
2624     }
2625 
2626     if (size < DEFAULT_HIGH_PCIE_MMIO_SIZE) {
2627         char *sz = size_to_str(DEFAULT_HIGH_PCIE_MMIO_SIZE);
2628         error_setg(errp, "highmem-mmio-size cannot be set to a lower value "
2629                          "than the default (%s)", sz);
2630         g_free(sz);
2631         return;
2632     }
2633 
2634     extended_memmap[VIRT_HIGH_PCIE_MMIO].size = size;
2635 }
2636 
2637 static bool virt_get_its(Object *obj, Error **errp)
2638 {
2639     VirtMachineState *vms = VIRT_MACHINE(obj);
2640 
2641     return vms->its;
2642 }
2643 
2644 static void virt_set_its(Object *obj, bool value, Error **errp)
2645 {
2646     VirtMachineState *vms = VIRT_MACHINE(obj);
2647 
2648     vms->its = value;
2649 }
2650 
2651 static bool virt_get_dtb_randomness(Object *obj, Error **errp)
2652 {
2653     VirtMachineState *vms = VIRT_MACHINE(obj);
2654 
2655     return vms->dtb_randomness;
2656 }
2657 
2658 static void virt_set_dtb_randomness(Object *obj, bool value, Error **errp)
2659 {
2660     VirtMachineState *vms = VIRT_MACHINE(obj);
2661 
2662     vms->dtb_randomness = value;
2663 }
2664 
2665 static char *virt_get_oem_id(Object *obj, Error **errp)
2666 {
2667     VirtMachineState *vms = VIRT_MACHINE(obj);
2668 
2669     return g_strdup(vms->oem_id);
2670 }
2671 
2672 static void virt_set_oem_id(Object *obj, const char *value, Error **errp)
2673 {
2674     VirtMachineState *vms = VIRT_MACHINE(obj);
2675     size_t len = strlen(value);
2676 
2677     if (len > 6) {
2678         error_setg(errp,
2679                    "User specified oem-id value is bigger than 6 bytes in size");
2680         return;
2681     }
2682 
2683     strncpy(vms->oem_id, value, 6);
2684 }
2685 
2686 static char *virt_get_oem_table_id(Object *obj, Error **errp)
2687 {
2688     VirtMachineState *vms = VIRT_MACHINE(obj);
2689 
2690     return g_strdup(vms->oem_table_id);
2691 }
2692 
2693 static void virt_set_oem_table_id(Object *obj, const char *value,
2694                                   Error **errp)
2695 {
2696     VirtMachineState *vms = VIRT_MACHINE(obj);
2697     size_t len = strlen(value);
2698 
2699     if (len > 8) {
2700         error_setg(errp,
2701                    "User specified oem-table-id value is bigger than 8 bytes in size");
2702         return;
2703     }
2704     strncpy(vms->oem_table_id, value, 8);
2705 }
2706 
2707 
2708 bool virt_is_acpi_enabled(VirtMachineState *vms)
2709 {
2710     if (vms->acpi == ON_OFF_AUTO_OFF) {
2711         return false;
2712     }
2713     return true;
2714 }
2715 
2716 static void virt_get_acpi(Object *obj, Visitor *v, const char *name,
2717                           void *opaque, Error **errp)
2718 {
2719     VirtMachineState *vms = VIRT_MACHINE(obj);
2720     OnOffAuto acpi = vms->acpi;
2721 
2722     visit_type_OnOffAuto(v, name, &acpi, errp);
2723 }
2724 
2725 static void virt_set_acpi(Object *obj, Visitor *v, const char *name,
2726                           void *opaque, Error **errp)
2727 {
2728     VirtMachineState *vms = VIRT_MACHINE(obj);
2729 
2730     visit_type_OnOffAuto(v, name, &vms->acpi, errp);
2731 }
2732 
2733 static bool virt_get_ras(Object *obj, Error **errp)
2734 {
2735     VirtMachineState *vms = VIRT_MACHINE(obj);
2736 
2737     return vms->ras;
2738 }
2739 
2740 static void virt_set_ras(Object *obj, bool value, Error **errp)
2741 {
2742     VirtMachineState *vms = VIRT_MACHINE(obj);
2743 
2744     vms->ras = value;
2745 }
2746 
2747 static bool virt_get_mte(Object *obj, Error **errp)
2748 {
2749     VirtMachineState *vms = VIRT_MACHINE(obj);
2750 
2751     return vms->mte;
2752 }
2753 
2754 static void virt_set_mte(Object *obj, bool value, Error **errp)
2755 {
2756     VirtMachineState *vms = VIRT_MACHINE(obj);
2757 
2758     vms->mte = value;
2759 }
2760 
2761 static char *virt_get_gic_version(Object *obj, Error **errp)
2762 {
2763     VirtMachineState *vms = VIRT_MACHINE(obj);
2764     const char *val;
2765 
2766     switch (vms->gic_version) {
2767     case VIRT_GIC_VERSION_4:
2768         val = "4";
2769         break;
2770     case VIRT_GIC_VERSION_3:
2771         val = "3";
2772         break;
2773     default:
2774         val = "2";
2775         break;
2776     }
2777     return g_strdup(val);
2778 }
2779 
2780 static void virt_set_gic_version(Object *obj, const char *value, Error **errp)
2781 {
2782     VirtMachineState *vms = VIRT_MACHINE(obj);
2783 
2784     if (!strcmp(value, "4")) {
2785         vms->gic_version = VIRT_GIC_VERSION_4;
2786     } else if (!strcmp(value, "3")) {
2787         vms->gic_version = VIRT_GIC_VERSION_3;
2788     } else if (!strcmp(value, "2")) {
2789         vms->gic_version = VIRT_GIC_VERSION_2;
2790     } else if (!strcmp(value, "host")) {
2791         vms->gic_version = VIRT_GIC_VERSION_HOST; /* Will probe later */
2792     } else if (!strcmp(value, "max")) {
2793         vms->gic_version = VIRT_GIC_VERSION_MAX; /* Will probe later */
2794     } else {
2795         error_setg(errp, "Invalid gic-version value");
2796         error_append_hint(errp, "Valid values are 3, 2, host, max.\n");
2797     }
2798 }
2799 
2800 static char *virt_get_iommu(Object *obj, Error **errp)
2801 {
2802     VirtMachineState *vms = VIRT_MACHINE(obj);
2803 
2804     switch (vms->iommu) {
2805     case VIRT_IOMMU_NONE:
2806         return g_strdup("none");
2807     case VIRT_IOMMU_SMMUV3:
2808         return g_strdup("smmuv3");
2809     default:
2810         g_assert_not_reached();
2811     }
2812 }
2813 
2814 static void virt_set_iommu(Object *obj, const char *value, Error **errp)
2815 {
2816     VirtMachineState *vms = VIRT_MACHINE(obj);
2817 
2818     if (!strcmp(value, "smmuv3")) {
2819         vms->iommu = VIRT_IOMMU_SMMUV3;
2820     } else if (!strcmp(value, "none")) {
2821         vms->iommu = VIRT_IOMMU_NONE;
2822     } else {
2823         error_setg(errp, "Invalid iommu value");
2824         error_append_hint(errp, "Valid values are none, smmuv3.\n");
2825     }
2826 }
2827 
2828 static bool virt_get_default_bus_bypass_iommu(Object *obj, Error **errp)
2829 {
2830     VirtMachineState *vms = VIRT_MACHINE(obj);
2831 
2832     return vms->default_bus_bypass_iommu;
2833 }
2834 
2835 static void virt_set_default_bus_bypass_iommu(Object *obj, bool value,
2836                                               Error **errp)
2837 {
2838     VirtMachineState *vms = VIRT_MACHINE(obj);
2839 
2840     vms->default_bus_bypass_iommu = value;
2841 }
2842 
2843 static CpuInstanceProperties
2844 virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
2845 {
2846     MachineClass *mc = MACHINE_GET_CLASS(ms);
2847     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
2848 
2849     assert(cpu_index < possible_cpus->len);
2850     return possible_cpus->cpus[cpu_index].props;
2851 }
2852 
2853 static int64_t virt_get_default_cpu_node_id(const MachineState *ms, int idx)
2854 {
2855     int64_t socket_id = ms->possible_cpus->cpus[idx].props.socket_id;
2856 
2857     return socket_id % ms->numa_state->num_nodes;
2858 }
2859 
2860 static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms)
2861 {
2862     int n;
2863     unsigned int max_cpus = ms->smp.max_cpus;
2864     VirtMachineState *vms = VIRT_MACHINE(ms);
2865     MachineClass *mc = MACHINE_GET_CLASS(vms);
2866 
2867     if (ms->possible_cpus) {
2868         assert(ms->possible_cpus->len == max_cpus);
2869         return ms->possible_cpus;
2870     }
2871 
2872     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2873                                   sizeof(CPUArchId) * max_cpus);
2874     ms->possible_cpus->len = max_cpus;
2875     for (n = 0; n < ms->possible_cpus->len; n++) {
2876         ms->possible_cpus->cpus[n].type = ms->cpu_type;
2877         ms->possible_cpus->cpus[n].arch_id =
2878             virt_cpu_mp_affinity(vms, n);
2879 
2880         assert(!mc->smp_props.dies_supported);
2881         ms->possible_cpus->cpus[n].props.has_socket_id = true;
2882         ms->possible_cpus->cpus[n].props.socket_id =
2883             n / (ms->smp.clusters * ms->smp.cores * ms->smp.threads);
2884         ms->possible_cpus->cpus[n].props.has_cluster_id = true;
2885         ms->possible_cpus->cpus[n].props.cluster_id =
2886             (n / (ms->smp.cores * ms->smp.threads)) % ms->smp.clusters;
2887         ms->possible_cpus->cpus[n].props.has_core_id = true;
2888         ms->possible_cpus->cpus[n].props.core_id =
2889             (n / ms->smp.threads) % ms->smp.cores;
2890         ms->possible_cpus->cpus[n].props.has_thread_id = true;
2891         ms->possible_cpus->cpus[n].props.thread_id =
2892             n % ms->smp.threads;
2893     }
2894     return ms->possible_cpus;
2895 }
2896 
2897 static void virt_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2898                                  Error **errp)
2899 {
2900     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2901     const MachineState *ms = MACHINE(hotplug_dev);
2902     const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2903 
2904     if (!vms->acpi_dev) {
2905         error_setg(errp,
2906                    "memory hotplug is not enabled: missing acpi-ged device");
2907         return;
2908     }
2909 
2910     if (vms->mte) {
2911         error_setg(errp, "memory hotplug is not enabled: MTE is enabled");
2912         return;
2913     }
2914 
2915     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
2916         error_setg(errp, "nvdimm is not enabled: add 'nvdimm=on' to '-M'");
2917         return;
2918     }
2919 
2920     pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev), errp);
2921 }
2922 
2923 static void virt_memory_plug(HotplugHandler *hotplug_dev,
2924                              DeviceState *dev, Error **errp)
2925 {
2926     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2927     MachineState *ms = MACHINE(hotplug_dev);
2928     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2929 
2930     pc_dimm_plug(PC_DIMM(dev), MACHINE(vms));
2931 
2932     if (is_nvdimm) {
2933         nvdimm_plug(ms->nvdimms_state);
2934     }
2935 
2936     hotplug_handler_plug(HOTPLUG_HANDLER(vms->acpi_dev),
2937                          dev, &error_abort);
2938 }
2939 
2940 static void virt_machine_device_pre_plug_cb(HotplugHandler *hotplug_dev,
2941                                             DeviceState *dev, Error **errp)
2942 {
2943     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2944 
2945     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2946         virt_memory_pre_plug(hotplug_dev, dev, errp);
2947     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
2948         virtio_md_pci_pre_plug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
2949     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
2950         hwaddr db_start = 0, db_end = 0;
2951         QList *reserved_regions;
2952         char *resv_prop_str;
2953 
2954         if (vms->iommu != VIRT_IOMMU_NONE) {
2955             error_setg(errp, "virt machine does not support multiple IOMMUs");
2956             return;
2957         }
2958 
2959         switch (vms->msi_controller) {
2960         case VIRT_MSI_CTRL_NONE:
2961             return;
2962         case VIRT_MSI_CTRL_ITS:
2963             /* GITS_TRANSLATER page */
2964             db_start = base_memmap[VIRT_GIC_ITS].base + 0x10000;
2965             db_end = base_memmap[VIRT_GIC_ITS].base +
2966                      base_memmap[VIRT_GIC_ITS].size - 1;
2967             break;
2968         case VIRT_MSI_CTRL_GICV2M:
2969             /* MSI_SETSPI_NS page */
2970             db_start = base_memmap[VIRT_GIC_V2M].base;
2971             db_end = db_start + base_memmap[VIRT_GIC_V2M].size - 1;
2972             break;
2973         }
2974         resv_prop_str = g_strdup_printf("0x%"PRIx64":0x%"PRIx64":%u",
2975                                         db_start, db_end,
2976                                         VIRTIO_IOMMU_RESV_MEM_T_MSI);
2977 
2978         reserved_regions = qlist_new();
2979         qlist_append_str(reserved_regions, resv_prop_str);
2980         qdev_prop_set_array(dev, "reserved-regions", reserved_regions);
2981         g_free(resv_prop_str);
2982     }
2983 }
2984 
2985 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
2986                                         DeviceState *dev, Error **errp)
2987 {
2988     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2989 
2990     if (vms->platform_bus_dev) {
2991         MachineClass *mc = MACHINE_GET_CLASS(vms);
2992 
2993         if (device_is_dynamic_sysbus(mc, dev)) {
2994             platform_bus_link_device(PLATFORM_BUS_DEVICE(vms->platform_bus_dev),
2995                                      SYS_BUS_DEVICE(dev));
2996         }
2997     }
2998 
2999     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3000         virt_memory_plug(hotplug_dev, dev, errp);
3001     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
3002         virtio_md_pci_plug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
3003     }
3004 
3005     if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
3006         PCIDevice *pdev = PCI_DEVICE(dev);
3007 
3008         vms->iommu = VIRT_IOMMU_VIRTIO;
3009         vms->virtio_iommu_bdf = pci_get_bdf(pdev);
3010         create_virtio_iommu_dt_bindings(vms);
3011     }
3012 }
3013 
3014 static void virt_dimm_unplug_request(HotplugHandler *hotplug_dev,
3015                                      DeviceState *dev, Error **errp)
3016 {
3017     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
3018 
3019     if (!vms->acpi_dev) {
3020         error_setg(errp,
3021                    "memory hotplug is not enabled: missing acpi-ged device");
3022         return;
3023     }
3024 
3025     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3026         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3027         return;
3028     }
3029 
3030     hotplug_handler_unplug_request(HOTPLUG_HANDLER(vms->acpi_dev), dev,
3031                                    errp);
3032 }
3033 
3034 static void virt_dimm_unplug(HotplugHandler *hotplug_dev,
3035                              DeviceState *dev, Error **errp)
3036 {
3037     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
3038     Error *local_err = NULL;
3039 
3040     hotplug_handler_unplug(HOTPLUG_HANDLER(vms->acpi_dev), dev, &local_err);
3041     if (local_err) {
3042         goto out;
3043     }
3044 
3045     pc_dimm_unplug(PC_DIMM(dev), MACHINE(vms));
3046     qdev_unrealize(dev);
3047 
3048 out:
3049     error_propagate(errp, local_err);
3050 }
3051 
3052 static void virt_machine_device_unplug_request_cb(HotplugHandler *hotplug_dev,
3053                                           DeviceState *dev, Error **errp)
3054 {
3055     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3056         virt_dimm_unplug_request(hotplug_dev, dev, errp);
3057     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
3058         virtio_md_pci_unplug_request(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev),
3059                                      errp);
3060     } else {
3061         error_setg(errp, "device unplug request for unsupported device"
3062                    " type: %s", object_get_typename(OBJECT(dev)));
3063     }
3064 }
3065 
3066 static void virt_machine_device_unplug_cb(HotplugHandler *hotplug_dev,
3067                                           DeviceState *dev, Error **errp)
3068 {
3069     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3070         virt_dimm_unplug(hotplug_dev, dev, errp);
3071     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
3072         virtio_md_pci_unplug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
3073     } else {
3074         error_setg(errp, "virt: device unplug for unsupported device"
3075                    " type: %s", object_get_typename(OBJECT(dev)));
3076     }
3077 }
3078 
3079 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
3080                                                         DeviceState *dev)
3081 {
3082     MachineClass *mc = MACHINE_GET_CLASS(machine);
3083 
3084     if (device_is_dynamic_sysbus(mc, dev) ||
3085         object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3086         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI) ||
3087         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
3088         return HOTPLUG_HANDLER(machine);
3089     }
3090     return NULL;
3091 }
3092 
3093 /*
3094  * for arm64 kvm_type [7-0] encodes the requested number of bits
3095  * in the IPA address space
3096  */
3097 static int virt_kvm_type(MachineState *ms, const char *type_str)
3098 {
3099     VirtMachineState *vms = VIRT_MACHINE(ms);
3100     int max_vm_pa_size, requested_pa_size;
3101     bool fixed_ipa;
3102 
3103     max_vm_pa_size = kvm_arm_get_max_vm_ipa_size(ms, &fixed_ipa);
3104 
3105     /* we freeze the memory map to compute the highest gpa */
3106     virt_set_memmap(vms, max_vm_pa_size);
3107 
3108     requested_pa_size = 64 - clz64(vms->highest_gpa);
3109 
3110     /*
3111      * KVM requires the IPA size to be at least 32 bits.
3112      */
3113     if (requested_pa_size < 32) {
3114         requested_pa_size = 32;
3115     }
3116 
3117     if (requested_pa_size > max_vm_pa_size) {
3118         error_report("-m and ,maxmem option values "
3119                      "require an IPA range (%d bits) larger than "
3120                      "the one supported by the host (%d bits)",
3121                      requested_pa_size, max_vm_pa_size);
3122         return -1;
3123     }
3124     /*
3125      * We return the requested PA log size, unless KVM only supports
3126      * the implicit legacy 40b IPA setting, in which case the kvm_type
3127      * must be 0.
3128      */
3129     return fixed_ipa ? 0 : requested_pa_size;
3130 }
3131 
3132 static int virt_hvf_get_physical_address_range(MachineState *ms)
3133 {
3134     VirtMachineState *vms = VIRT_MACHINE(ms);
3135 
3136     int default_ipa_size = hvf_arm_get_default_ipa_bit_size();
3137     int max_ipa_size = hvf_arm_get_max_ipa_bit_size();
3138 
3139     /* We freeze the memory map to compute the highest gpa */
3140     virt_set_memmap(vms, max_ipa_size);
3141 
3142     int requested_ipa_size = 64 - clz64(vms->highest_gpa);
3143 
3144     /*
3145      * If we're <= the default IPA size just use the default.
3146      * If we're above the default but below the maximum, round up to
3147      * the maximum. hvf_arm_get_max_ipa_bit_size() conveniently only
3148      * returns values that are valid ARM PARange values.
3149      */
3150     if (requested_ipa_size <= default_ipa_size) {
3151         requested_ipa_size = default_ipa_size;
3152     } else if (requested_ipa_size <= max_ipa_size) {
3153         requested_ipa_size = max_ipa_size;
3154     } else {
3155         error_report("-m and ,maxmem option values "
3156                      "require an IPA range (%d bits) larger than "
3157                      "the one supported by the host (%d bits)",
3158                      requested_ipa_size, max_ipa_size);
3159         return -1;
3160     }
3161 
3162     return requested_ipa_size;
3163 }
3164 
3165 static void virt_machine_class_init(ObjectClass *oc, const void *data)
3166 {
3167     MachineClass *mc = MACHINE_CLASS(oc);
3168     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3169     static const char * const valid_cpu_types[] = {
3170 #ifdef CONFIG_TCG
3171         ARM_CPU_TYPE_NAME("cortex-a7"),
3172         ARM_CPU_TYPE_NAME("cortex-a15"),
3173 #ifdef TARGET_AARCH64
3174         ARM_CPU_TYPE_NAME("cortex-a35"),
3175         ARM_CPU_TYPE_NAME("cortex-a55"),
3176         ARM_CPU_TYPE_NAME("cortex-a72"),
3177         ARM_CPU_TYPE_NAME("cortex-a76"),
3178         ARM_CPU_TYPE_NAME("cortex-a710"),
3179         ARM_CPU_TYPE_NAME("a64fx"),
3180         ARM_CPU_TYPE_NAME("neoverse-n1"),
3181         ARM_CPU_TYPE_NAME("neoverse-v1"),
3182         ARM_CPU_TYPE_NAME("neoverse-n2"),
3183 #endif /* TARGET_AARCH64 */
3184 #endif /* CONFIG_TCG */
3185 #ifdef TARGET_AARCH64
3186         ARM_CPU_TYPE_NAME("cortex-a53"),
3187         ARM_CPU_TYPE_NAME("cortex-a57"),
3188 #if defined(CONFIG_KVM) || defined(CONFIG_HVF)
3189         ARM_CPU_TYPE_NAME("host"),
3190 #endif /* CONFIG_KVM || CONFIG_HVF */
3191 #endif /* TARGET_AARCH64 */
3192         ARM_CPU_TYPE_NAME("max"),
3193         NULL
3194     };
3195 
3196     mc->init = machvirt_init;
3197     /* Start with max_cpus set to 512, which is the maximum supported by KVM.
3198      * The value may be reduced later when we have more information about the
3199      * configuration of the particular instance.
3200      */
3201     mc->max_cpus = 512;
3202     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_CALXEDA_XGMAC);
3203     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_AMD_XGBE);
3204     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
3205     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_PLATFORM);
3206     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_UEFI_VARS_SYSBUS);
3207 #ifdef CONFIG_TPM
3208     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_TPM_TIS_SYSBUS);
3209 #endif
3210     mc->block_default_type = IF_VIRTIO;
3211     mc->no_cdrom = 1;
3212     mc->pci_allow_0_address = true;
3213     /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */
3214     mc->minimum_page_bits = 12;
3215     mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids;
3216     mc->cpu_index_to_instance_props = virt_cpu_index_to_props;
3217 #ifdef CONFIG_TCG
3218     mc->default_cpu_type = ARM_CPU_TYPE_NAME("cortex-a15");
3219 #else
3220     mc->default_cpu_type = ARM_CPU_TYPE_NAME("max");
3221 #endif
3222     mc->valid_cpu_types = valid_cpu_types;
3223     mc->get_default_cpu_node_id = virt_get_default_cpu_node_id;
3224     mc->kvm_type = virt_kvm_type;
3225     mc->hvf_get_physical_address_range = virt_hvf_get_physical_address_range;
3226     assert(!mc->get_hotplug_handler);
3227     mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
3228     hc->pre_plug = virt_machine_device_pre_plug_cb;
3229     hc->plug = virt_machine_device_plug_cb;
3230     hc->unplug_request = virt_machine_device_unplug_request_cb;
3231     hc->unplug = virt_machine_device_unplug_cb;
3232     mc->nvdimm_supported = true;
3233     mc->smp_props.clusters_supported = true;
3234     mc->auto_enable_numa_with_memhp = true;
3235     mc->auto_enable_numa_with_memdev = true;
3236     /* platform instead of architectural choice */
3237     mc->cpu_cluster_has_numa_boundary = true;
3238     mc->default_ram_id = "mach-virt.ram";
3239     mc->default_nic = "virtio-net-pci";
3240 
3241     object_class_property_add(oc, "acpi", "OnOffAuto",
3242         virt_get_acpi, virt_set_acpi,
3243         NULL, NULL);
3244     object_class_property_set_description(oc, "acpi",
3245         "Enable ACPI");
3246     object_class_property_add_bool(oc, "secure", virt_get_secure,
3247                                    virt_set_secure);
3248     object_class_property_set_description(oc, "secure",
3249                                                 "Set on/off to enable/disable the ARM "
3250                                                 "Security Extensions (TrustZone)");
3251 
3252     object_class_property_add_bool(oc, "virtualization", virt_get_virt,
3253                                    virt_set_virt);
3254     object_class_property_set_description(oc, "virtualization",
3255                                           "Set on/off to enable/disable emulating a "
3256                                           "guest CPU which implements the ARM "
3257                                           "Virtualization Extensions");
3258 
3259     object_class_property_add_bool(oc, "highmem", virt_get_highmem,
3260                                    virt_set_highmem);
3261     object_class_property_set_description(oc, "highmem",
3262                                           "Set on/off to enable/disable using "
3263                                           "physical address space above 32 bits");
3264 
3265     object_class_property_add_bool(oc, "compact-highmem",
3266                                    virt_get_compact_highmem,
3267                                    virt_set_compact_highmem);
3268     object_class_property_set_description(oc, "compact-highmem",
3269                                           "Set on/off to enable/disable compact "
3270                                           "layout for high memory regions");
3271 
3272     object_class_property_add_bool(oc, "highmem-redists",
3273                                    virt_get_highmem_redists,
3274                                    virt_set_highmem_redists);
3275     object_class_property_set_description(oc, "highmem-redists",
3276                                           "Set on/off to enable/disable high "
3277                                           "memory region for GICv3 or GICv4 "
3278                                           "redistributor");
3279 
3280     object_class_property_add_bool(oc, "highmem-ecam",
3281                                    virt_get_highmem_ecam,
3282                                    virt_set_highmem_ecam);
3283     object_class_property_set_description(oc, "highmem-ecam",
3284                                           "Set on/off to enable/disable high "
3285                                           "memory region for PCI ECAM");
3286 
3287     object_class_property_add_bool(oc, "highmem-mmio",
3288                                    virt_get_highmem_mmio,
3289                                    virt_set_highmem_mmio);
3290     object_class_property_set_description(oc, "highmem-mmio",
3291                                           "Set on/off to enable/disable high "
3292                                           "memory region for PCI MMIO");
3293 
3294     object_class_property_add(oc, "highmem-mmio-size", "size",
3295                                    virt_get_highmem_mmio_size,
3296                                    virt_set_highmem_mmio_size,
3297                                    NULL, NULL);
3298     object_class_property_set_description(oc, "highmem-mmio-size",
3299                                           "Set the high memory region size "
3300                                           "for PCI MMIO");
3301 
3302     object_class_property_add_str(oc, "gic-version", virt_get_gic_version,
3303                                   virt_set_gic_version);
3304     object_class_property_set_description(oc, "gic-version",
3305                                           "Set GIC version. "
3306                                           "Valid values are 2, 3, 4, host and max");
3307 
3308     object_class_property_add_str(oc, "iommu", virt_get_iommu, virt_set_iommu);
3309     object_class_property_set_description(oc, "iommu",
3310                                           "Set the IOMMU type. "
3311                                           "Valid values are none and smmuv3");
3312 
3313     object_class_property_add_bool(oc, "default-bus-bypass-iommu",
3314                                    virt_get_default_bus_bypass_iommu,
3315                                    virt_set_default_bus_bypass_iommu);
3316     object_class_property_set_description(oc, "default-bus-bypass-iommu",
3317                                           "Set on/off to enable/disable "
3318                                           "bypass_iommu for default root bus");
3319 
3320     object_class_property_add_bool(oc, "ras", virt_get_ras,
3321                                    virt_set_ras);
3322     object_class_property_set_description(oc, "ras",
3323                                           "Set on/off to enable/disable reporting host memory errors "
3324                                           "to a KVM guest using ACPI and guest external abort exceptions");
3325 
3326     object_class_property_add_bool(oc, "mte", virt_get_mte, virt_set_mte);
3327     object_class_property_set_description(oc, "mte",
3328                                           "Set on/off to enable/disable emulating a "
3329                                           "guest CPU which implements the ARM "
3330                                           "Memory Tagging Extension");
3331 
3332     object_class_property_add_bool(oc, "its", virt_get_its,
3333                                    virt_set_its);
3334     object_class_property_set_description(oc, "its",
3335                                           "Set on/off to enable/disable "
3336                                           "ITS instantiation");
3337 
3338     object_class_property_add_bool(oc, "dtb-randomness",
3339                                    virt_get_dtb_randomness,
3340                                    virt_set_dtb_randomness);
3341     object_class_property_set_description(oc, "dtb-randomness",
3342                                           "Set off to disable passing random or "
3343                                           "non-deterministic dtb nodes to guest");
3344 
3345     object_class_property_add_bool(oc, "dtb-kaslr-seed",
3346                                    virt_get_dtb_randomness,
3347                                    virt_set_dtb_randomness);
3348     object_class_property_set_description(oc, "dtb-kaslr-seed",
3349                                           "Deprecated synonym of dtb-randomness");
3350 
3351     object_class_property_add_str(oc, "x-oem-id",
3352                                   virt_get_oem_id,
3353                                   virt_set_oem_id);
3354     object_class_property_set_description(oc, "x-oem-id",
3355                                           "Override the default value of field OEMID "
3356                                           "in ACPI table header."
3357                                           "The string may be up to 6 bytes in size");
3358 
3359 
3360     object_class_property_add_str(oc, "x-oem-table-id",
3361                                   virt_get_oem_table_id,
3362                                   virt_set_oem_table_id);
3363     object_class_property_set_description(oc, "x-oem-table-id",
3364                                           "Override the default value of field OEM Table ID "
3365                                           "in ACPI table header."
3366                                           "The string may be up to 8 bytes in size");
3367 
3368 }
3369 
3370 static void virt_instance_init(Object *obj)
3371 {
3372     VirtMachineState *vms = VIRT_MACHINE(obj);
3373     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
3374 
3375     /* EL3 is disabled by default on virt: this makes us consistent
3376      * between KVM and TCG for this board, and it also allows us to
3377      * boot UEFI blobs which assume no TrustZone support.
3378      */
3379     vms->secure = false;
3380 
3381     /* EL2 is also disabled by default, for similar reasons */
3382     vms->virt = false;
3383 
3384     /* High memory is enabled by default */
3385     vms->highmem = true;
3386     vms->highmem_compact = !vmc->no_highmem_compact;
3387     vms->gic_version = VIRT_GIC_VERSION_NOSEL;
3388 
3389     vms->highmem_ecam = true;
3390     vms->highmem_mmio = true;
3391     vms->highmem_redists = true;
3392 
3393     /* Default allows ITS instantiation */
3394     vms->its = true;
3395     /* Allow ITS emulation if the machine version supports it */
3396     vms->tcg_its = !vmc->no_tcg_its;
3397 
3398     /* Default disallows iommu instantiation */
3399     vms->iommu = VIRT_IOMMU_NONE;
3400 
3401     /* The default root bus is attached to iommu by default */
3402     vms->default_bus_bypass_iommu = false;
3403 
3404     /* Default disallows RAS instantiation */
3405     vms->ras = false;
3406 
3407     /* MTE is disabled by default.  */
3408     vms->mte = false;
3409 
3410     /* Supply kaslr-seed and rng-seed by default */
3411     vms->dtb_randomness = true;
3412 
3413     vms->irqmap = a15irqmap;
3414 
3415     virt_flash_create(vms);
3416 
3417     vms->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
3418     vms->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
3419     cxl_machine_init(obj, &vms->cxl_devices_state);
3420 }
3421 
3422 static const TypeInfo virt_machine_info = {
3423     .name          = TYPE_VIRT_MACHINE,
3424     .parent        = TYPE_MACHINE,
3425     .abstract      = true,
3426     .instance_size = sizeof(VirtMachineState),
3427     .class_size    = sizeof(VirtMachineClass),
3428     .class_init    = virt_machine_class_init,
3429     .instance_init = virt_instance_init,
3430     .interfaces = (const InterfaceInfo[]) {
3431          { TYPE_HOTPLUG_HANDLER },
3432          { }
3433     },
3434 };
3435 
3436 static void machvirt_machine_init(void)
3437 {
3438     type_register_static(&virt_machine_info);
3439 }
3440 type_init(machvirt_machine_init);
3441 
3442 static void virt_machine_10_1_options(MachineClass *mc)
3443 {
3444 }
3445 DEFINE_VIRT_MACHINE_AS_LATEST(10, 1)
3446 
3447 static void virt_machine_10_0_options(MachineClass *mc)
3448 {
3449     virt_machine_10_1_options(mc);
3450     compat_props_add(mc->compat_props, hw_compat_10_0, hw_compat_10_0_len);
3451 }
3452 DEFINE_VIRT_MACHINE(10, 0)
3453 
3454 static void virt_machine_9_2_options(MachineClass *mc)
3455 {
3456     virt_machine_10_0_options(mc);
3457     compat_props_add(mc->compat_props, hw_compat_9_2, hw_compat_9_2_len);
3458 }
3459 DEFINE_VIRT_MACHINE(9, 2)
3460 
3461 static void virt_machine_9_1_options(MachineClass *mc)
3462 {
3463     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3464 
3465     virt_machine_9_2_options(mc);
3466     compat_props_add(mc->compat_props, hw_compat_9_1, hw_compat_9_1_len);
3467     /* 9.1 and earlier have only a stage-1 SMMU, not a nested s1+2 one */
3468     vmc->no_nested_smmu = true;
3469 }
3470 DEFINE_VIRT_MACHINE(9, 1)
3471 
3472 static void virt_machine_9_0_options(MachineClass *mc)
3473 {
3474     virt_machine_9_1_options(mc);
3475     mc->smbios_memory_device_size = 16 * GiB;
3476     compat_props_add(mc->compat_props, hw_compat_9_0, hw_compat_9_0_len);
3477 }
3478 DEFINE_VIRT_MACHINE(9, 0)
3479 
3480 static void virt_machine_8_2_options(MachineClass *mc)
3481 {
3482     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3483 
3484     virt_machine_9_0_options(mc);
3485     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
3486     /*
3487      * Don't expose NS_EL2_VIRT timer IRQ in DTB on ACPI on 8.2 and
3488      * earlier machines. (Exposing it tickles a bug in older EDK2
3489      * guest BIOS binaries.)
3490      */
3491     vmc->no_ns_el2_virt_timer_irq = true;
3492 }
3493 DEFINE_VIRT_MACHINE(8, 2)
3494 
3495 static void virt_machine_8_1_options(MachineClass *mc)
3496 {
3497     virt_machine_8_2_options(mc);
3498     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
3499 }
3500 DEFINE_VIRT_MACHINE(8, 1)
3501 
3502 static void virt_machine_8_0_options(MachineClass *mc)
3503 {
3504     virt_machine_8_1_options(mc);
3505     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
3506 }
3507 DEFINE_VIRT_MACHINE(8, 0)
3508 
3509 static void virt_machine_7_2_options(MachineClass *mc)
3510 {
3511     virt_machine_8_0_options(mc);
3512     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
3513 }
3514 DEFINE_VIRT_MACHINE(7, 2)
3515 
3516 static void virt_machine_7_1_options(MachineClass *mc)
3517 {
3518     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3519 
3520     virt_machine_7_2_options(mc);
3521     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
3522     /* Compact layout for high memory regions was introduced with 7.2 */
3523     vmc->no_highmem_compact = true;
3524 }
3525 DEFINE_VIRT_MACHINE(7, 1)
3526 
3527 static void virt_machine_7_0_options(MachineClass *mc)
3528 {
3529     virt_machine_7_1_options(mc);
3530     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
3531 }
3532 DEFINE_VIRT_MACHINE(7, 0)
3533 
3534 static void virt_machine_6_2_options(MachineClass *mc)
3535 {
3536     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3537 
3538     virt_machine_7_0_options(mc);
3539     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
3540     vmc->no_tcg_lpa2 = true;
3541 }
3542 DEFINE_VIRT_MACHINE(6, 2)
3543 
3544 static void virt_machine_6_1_options(MachineClass *mc)
3545 {
3546     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3547 
3548     virt_machine_6_2_options(mc);
3549     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
3550     mc->smp_props.prefer_sockets = true;
3551     vmc->no_cpu_topology = true;
3552 
3553     /* qemu ITS was introduced with 6.2 */
3554     vmc->no_tcg_its = true;
3555 }
3556 DEFINE_VIRT_MACHINE(6, 1)
3557 
3558 static void virt_machine_6_0_options(MachineClass *mc)
3559 {
3560     virt_machine_6_1_options(mc);
3561     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
3562 }
3563 DEFINE_VIRT_MACHINE(6, 0)
3564 
3565 static void virt_machine_5_2_options(MachineClass *mc)
3566 {
3567     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3568 
3569     virt_machine_6_0_options(mc);
3570     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
3571     vmc->no_secure_gpio = true;
3572 }
3573 DEFINE_VIRT_MACHINE(5, 2)
3574 
3575 static void virt_machine_5_1_options(MachineClass *mc)
3576 {
3577     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3578 
3579     virt_machine_5_2_options(mc);
3580     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
3581     vmc->no_kvm_steal_time = true;
3582 }
3583 DEFINE_VIRT_MACHINE(5, 1)
3584 
3585 static void virt_machine_5_0_options(MachineClass *mc)
3586 {
3587     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3588 
3589     virt_machine_5_1_options(mc);
3590     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
3591     mc->numa_mem_supported = true;
3592     vmc->acpi_expose_flash = true;
3593     mc->auto_enable_numa_with_memdev = false;
3594 }
3595 DEFINE_VIRT_MACHINE(5, 0)
3596 
3597 static void virt_machine_4_2_options(MachineClass *mc)
3598 {
3599     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3600 
3601     virt_machine_5_0_options(mc);
3602     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
3603     vmc->kvm_no_adjvtime = true;
3604 }
3605 DEFINE_VIRT_MACHINE(4, 2)
3606 
3607 static void virt_machine_4_1_options(MachineClass *mc)
3608 {
3609     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3610 
3611     virt_machine_4_2_options(mc);
3612     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
3613     vmc->no_ged = true;
3614     mc->auto_enable_numa_with_memhp = false;
3615 }
3616 DEFINE_VIRT_MACHINE(4, 1)
3617