xref: /openbmc/qemu/hw/arm/virt.c (revision 5a28fa5ba17254d0398a854657b47af3096bd86a)
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 "target/arm/cpu-qom.h"
80 #include "target/arm/internals.h"
81 #include "target/arm/multiprocessing.h"
82 #include "target/arm/gtimer.h"
83 #include "hw/mem/pc-dimm.h"
84 #include "hw/mem/nvdimm.h"
85 #include "hw/acpi/generic_event_device.h"
86 #include "hw/uefi/var-service-api.h"
87 #include "hw/virtio/virtio-md-pci.h"
88 #include "hw/virtio/virtio-iommu.h"
89 #include "hw/char/pl011.h"
90 #include "hw/cxl/cxl.h"
91 #include "hw/cxl/cxl_host.h"
92 #include "qemu/guest-random.h"
93 
94 static GlobalProperty arm_virt_compat[] = {
95     { TYPE_VIRTIO_IOMMU_PCI, "aw-bits", "48" },
96 };
97 static const size_t arm_virt_compat_len = G_N_ELEMENTS(arm_virt_compat);
98 
99 /*
100  * This cannot be called from the virt_machine_class_init() because
101  * TYPE_VIRT_MACHINE is abstract and mc->compat_props g_ptr_array_new()
102  * only is called on virt non abstract class init.
103  */
104 static void arm_virt_compat_set(MachineClass *mc)
105 {
106     compat_props_add(mc->compat_props, arm_virt_compat,
107                      arm_virt_compat_len);
108 }
109 
110 #define DEFINE_VIRT_MACHINE_IMPL(latest, ...) \
111     static void MACHINE_VER_SYM(class_init, virt, __VA_ARGS__)( \
112         ObjectClass *oc, \
113         const void *data) \
114     { \
115         MachineClass *mc = MACHINE_CLASS(oc); \
116         arm_virt_compat_set(mc); \
117         MACHINE_VER_SYM(options, virt, __VA_ARGS__)(mc); \
118         mc->desc = "QEMU " MACHINE_VER_STR(__VA_ARGS__) " ARM Virtual Machine"; \
119         MACHINE_VER_DEPRECATION(__VA_ARGS__); \
120         if (latest) { \
121             mc->alias = "virt"; \
122         } \
123     } \
124     static const TypeInfo MACHINE_VER_SYM(info, virt, __VA_ARGS__) = \
125     { \
126         .name = MACHINE_VER_TYPE_NAME("virt", __VA_ARGS__), \
127         .parent = TYPE_VIRT_MACHINE, \
128         .class_init = MACHINE_VER_SYM(class_init, virt, __VA_ARGS__), \
129     }; \
130     static void MACHINE_VER_SYM(register, virt, __VA_ARGS__)(void) \
131     { \
132         MACHINE_VER_DELETION(__VA_ARGS__); \
133         type_register_static(&MACHINE_VER_SYM(info, virt, __VA_ARGS__)); \
134     } \
135     type_init(MACHINE_VER_SYM(register, virt, __VA_ARGS__));
136 
137 #define DEFINE_VIRT_MACHINE_AS_LATEST(major, minor) \
138     DEFINE_VIRT_MACHINE_IMPL(true, major, minor)
139 #define DEFINE_VIRT_MACHINE(major, minor) \
140     DEFINE_VIRT_MACHINE_IMPL(false, major, minor)
141 
142 
143 /* Number of external interrupt lines to configure the GIC with */
144 #define NUM_IRQS 256
145 
146 #define PLATFORM_BUS_NUM_IRQS 64
147 
148 /* Legacy RAM limit in GB (< version 4.0) */
149 #define LEGACY_RAMLIMIT_GB 255
150 #define LEGACY_RAMLIMIT_BYTES (LEGACY_RAMLIMIT_GB * GiB)
151 
152 /* Addresses and sizes of our components.
153  * 0..128MB is space for a flash device so we can run bootrom code such as UEFI.
154  * 128MB..256MB is used for miscellaneous device I/O.
155  * 256MB..1GB is reserved for possible future PCI support (ie where the
156  * PCI memory window will go if we add a PCI host controller).
157  * 1GB and up is RAM (which may happily spill over into the
158  * high memory region beyond 4GB).
159  * This represents a compromise between how much RAM can be given to
160  * a 32 bit VM and leaving space for expansion and in particular for PCI.
161  * Note that devices should generally be placed at multiples of 0x10000,
162  * to accommodate guests using 64K pages.
163  */
164 static const MemMapEntry base_memmap[] = {
165     /* Space up to 0x8000000 is reserved for a boot ROM */
166     [VIRT_FLASH] =              {          0, 0x08000000 },
167     [VIRT_CPUPERIPHS] =         { 0x08000000, 0x00020000 },
168     /* GIC distributor and CPU interfaces sit inside the CPU peripheral space */
169     [VIRT_GIC_DIST] =           { 0x08000000, 0x00010000 },
170     [VIRT_GIC_CPU] =            { 0x08010000, 0x00010000 },
171     [VIRT_GIC_V2M] =            { 0x08020000, 0x00001000 },
172     [VIRT_GIC_HYP] =            { 0x08030000, 0x00010000 },
173     [VIRT_GIC_VCPU] =           { 0x08040000, 0x00010000 },
174     /* The space in between here is reserved for GICv3 CPU/vCPU/HYP */
175     [VIRT_GIC_ITS] =            { 0x08080000, 0x00020000 },
176     /* This redistributor space allows up to 2*64kB*123 CPUs */
177     [VIRT_GIC_REDIST] =         { 0x080A0000, 0x00F60000 },
178     [VIRT_UART0] =              { 0x09000000, 0x00001000 },
179     [VIRT_RTC] =                { 0x09010000, 0x00001000 },
180     [VIRT_FW_CFG] =             { 0x09020000, 0x00000018 },
181     [VIRT_GPIO] =               { 0x09030000, 0x00001000 },
182     [VIRT_UART1] =              { 0x09040000, 0x00001000 },
183     [VIRT_SMMU] =               { 0x09050000, 0x00020000 },
184     [VIRT_PCDIMM_ACPI] =        { 0x09070000, MEMORY_HOTPLUG_IO_LEN },
185     [VIRT_ACPI_GED] =           { 0x09080000, ACPI_GED_EVT_SEL_LEN },
186     [VIRT_NVDIMM_ACPI] =        { 0x09090000, NVDIMM_ACPI_IO_LEN},
187     [VIRT_PVTIME] =             { 0x090a0000, 0x00010000 },
188     [VIRT_SECURE_GPIO] =        { 0x090b0000, 0x00001000 },
189     [VIRT_MMIO] =               { 0x0a000000, 0x00000200 },
190     /* ...repeating for a total of NUM_VIRTIO_TRANSPORTS, each of that size */
191     [VIRT_PLATFORM_BUS] =       { 0x0c000000, 0x02000000 },
192     [VIRT_SECURE_MEM] =         { 0x0e000000, 0x01000000 },
193     [VIRT_PCIE_MMIO] =          { 0x10000000, 0x2eff0000 },
194     [VIRT_PCIE_PIO] =           { 0x3eff0000, 0x00010000 },
195     [VIRT_PCIE_ECAM] =          { 0x3f000000, 0x01000000 },
196     /* Actual RAM size depends on initial RAM and device memory settings */
197     [VIRT_MEM] =                { GiB, LEGACY_RAMLIMIT_BYTES },
198 };
199 
200 /* Update the docs for highmem-mmio-size when changing this default */
201 #define DEFAULT_HIGH_PCIE_MMIO_SIZE_GB 512
202 #define DEFAULT_HIGH_PCIE_MMIO_SIZE (DEFAULT_HIGH_PCIE_MMIO_SIZE_GB * GiB)
203 
204 /*
205  * Highmem IO Regions: This memory map is floating, located after the RAM.
206  * Each MemMapEntry base (GPA) will be dynamically computed, depending on the
207  * top of the RAM, so that its base get the same alignment as the size,
208  * ie. a 512GiB entry will be aligned on a 512GiB boundary. If there is
209  * less than 256GiB of RAM, the floating area starts at the 256GiB mark.
210  * Note the extended_memmap is sized so that it eventually also includes the
211  * base_memmap entries (VIRT_HIGH_GIC_REDIST2 index is greater than the last
212  * index of base_memmap).
213  *
214  * The memory map for these Highmem IO Regions can be in legacy or compact
215  * layout, depending on 'compact-highmem' property. With legacy layout, the
216  * PA space for one specific region is always reserved, even if the region
217  * has been disabled or doesn't fit into the PA space. However, the PA space
218  * for the region won't be reserved in these circumstances with compact layout.
219  *
220  * Note that the highmem-mmio-size property will update the high PCIE MMIO size
221  * field in this array.
222  */
223 static MemMapEntry extended_memmap[] = {
224     /* Additional 64 MB redist region (can contain up to 512 redistributors) */
225     [VIRT_HIGH_GIC_REDIST2] =   { 0x0, 64 * MiB },
226     [VIRT_CXL_HOST] =           { 0x0, 64 * KiB * 16 }, /* 16 UID */
227     [VIRT_HIGH_PCIE_ECAM] =     { 0x0, 256 * MiB },
228     /* Second PCIe window */
229     [VIRT_HIGH_PCIE_MMIO] =     { 0x0, DEFAULT_HIGH_PCIE_MMIO_SIZE },
230     /* Any CXL Fixed memory windows come here */
231 };
232 
233 static const int a15irqmap[] = {
234     [VIRT_UART0] = 1,
235     [VIRT_RTC] = 2,
236     [VIRT_PCIE] = 3, /* ... to 6 */
237     [VIRT_GPIO] = 7,
238     [VIRT_UART1] = 8,
239     [VIRT_ACPI_GED] = 9,
240     [VIRT_MMIO] = 16, /* ...to 16 + NUM_VIRTIO_TRANSPORTS - 1 */
241     [VIRT_GIC_V2M] = 48, /* ...to 48 + NUM_GICV2M_SPIS - 1 */
242     [VIRT_SMMU] = 74,    /* ...to 74 + NUM_SMMU_IRQS - 1 */
243     [VIRT_PLATFORM_BUS] = 112, /* ...to 112 + PLATFORM_BUS_NUM_IRQS -1 */
244 };
245 
246 static void create_randomness(MachineState *ms, const char *node)
247 {
248     struct {
249         uint64_t kaslr;
250         uint8_t rng[32];
251     } seed;
252 
253     if (qemu_guest_getrandom(&seed, sizeof(seed), NULL)) {
254         return;
255     }
256     qemu_fdt_setprop_u64(ms->fdt, node, "kaslr-seed", seed.kaslr);
257     qemu_fdt_setprop(ms->fdt, node, "rng-seed", seed.rng, sizeof(seed.rng));
258 }
259 
260 /*
261  * The CPU object always exposes the NS EL2 virt timer IRQ line,
262  * but we don't want to advertise it to the guest in the dtb or ACPI
263  * table unless it's really going to do something.
264  */
265 static bool ns_el2_virt_timer_present(void)
266 {
267     ARMCPU *cpu = ARM_CPU(qemu_get_cpu(0));
268     CPUARMState *env = &cpu->env;
269 
270     return arm_feature(env, ARM_FEATURE_AARCH64) &&
271         arm_feature(env, ARM_FEATURE_EL2) && cpu_isar_feature(aa64_vh, cpu);
272 }
273 
274 static void create_fdt(VirtMachineState *vms)
275 {
276     MachineState *ms = MACHINE(vms);
277     int nb_numa_nodes = ms->numa_state->num_nodes;
278     void *fdt = create_device_tree(&vms->fdt_size);
279 
280     if (!fdt) {
281         error_report("create_device_tree() failed");
282         exit(1);
283     }
284 
285     ms->fdt = fdt;
286 
287     /* Header */
288     qemu_fdt_setprop_string(fdt, "/", "compatible", "linux,dummy-virt");
289     qemu_fdt_setprop_cell(fdt, "/", "#address-cells", 0x2);
290     qemu_fdt_setprop_cell(fdt, "/", "#size-cells", 0x2);
291     qemu_fdt_setprop_string(fdt, "/", "model", "linux,dummy-virt");
292 
293     /*
294      * For QEMU, all DMA is coherent. Advertising this in the root node
295      * has two benefits:
296      *
297      * - It avoids potential bugs where we forget to mark a DMA
298      *   capable device as being dma-coherent
299      * - It avoids spurious warnings from the Linux kernel about
300      *   devices which can't do DMA at all
301      */
302     qemu_fdt_setprop(fdt, "/", "dma-coherent", NULL, 0);
303 
304     /* /chosen must exist for load_dtb to fill in necessary properties later */
305     qemu_fdt_add_subnode(fdt, "/chosen");
306     if (vms->dtb_randomness) {
307         create_randomness(ms, "/chosen");
308     }
309 
310     if (vms->secure) {
311         qemu_fdt_add_subnode(fdt, "/secure-chosen");
312         if (vms->dtb_randomness) {
313             create_randomness(ms, "/secure-chosen");
314         }
315     }
316 
317     qemu_fdt_add_subnode(fdt, "/aliases");
318 
319     /* Clock node, for the benefit of the UART. The kernel device tree
320      * binding documentation claims the PL011 node clock properties are
321      * optional but in practice if you omit them the kernel refuses to
322      * probe for the device.
323      */
324     vms->clock_phandle = qemu_fdt_alloc_phandle(fdt);
325     qemu_fdt_add_subnode(fdt, "/apb-pclk");
326     qemu_fdt_setprop_string(fdt, "/apb-pclk", "compatible", "fixed-clock");
327     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "#clock-cells", 0x0);
328     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "clock-frequency", 24000000);
329     qemu_fdt_setprop_string(fdt, "/apb-pclk", "clock-output-names",
330                                 "clk24mhz");
331     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "phandle", vms->clock_phandle);
332 
333     if (nb_numa_nodes > 0 && ms->numa_state->have_numa_distance) {
334         int size = nb_numa_nodes * nb_numa_nodes * 3 * sizeof(uint32_t);
335         uint32_t *matrix = g_malloc0(size);
336         int idx, i, j;
337 
338         for (i = 0; i < nb_numa_nodes; i++) {
339             for (j = 0; j < nb_numa_nodes; j++) {
340                 idx = (i * nb_numa_nodes + j) * 3;
341                 matrix[idx + 0] = cpu_to_be32(i);
342                 matrix[idx + 1] = cpu_to_be32(j);
343                 matrix[idx + 2] =
344                     cpu_to_be32(ms->numa_state->nodes[i].distance[j]);
345             }
346         }
347 
348         qemu_fdt_add_subnode(fdt, "/distance-map");
349         qemu_fdt_setprop_string(fdt, "/distance-map", "compatible",
350                                 "numa-distance-map-v1");
351         qemu_fdt_setprop(fdt, "/distance-map", "distance-matrix",
352                          matrix, size);
353         g_free(matrix);
354     }
355 }
356 
357 static void fdt_add_timer_nodes(const VirtMachineState *vms)
358 {
359     /* On real hardware these interrupts are level-triggered.
360      * On KVM they were edge-triggered before host kernel version 4.4,
361      * and level-triggered afterwards.
362      * On emulated QEMU they are level-triggered.
363      *
364      * Getting the DTB info about them wrong is awkward for some
365      * guest kernels:
366      *  pre-4.8 ignore the DT and leave the interrupt configured
367      *   with whatever the GIC reset value (or the bootloader) left it at
368      *  4.8 before rc6 honour the incorrect data by programming it back
369      *   into the GIC, causing problems
370      *  4.8rc6 and later ignore the DT and always write "level triggered"
371      *   into the GIC
372      *
373      * For backwards-compatibility, virt-2.8 and earlier will continue
374      * to say these are edge-triggered, but later machines will report
375      * the correct information.
376      */
377     ARMCPU *armcpu;
378     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
379     MachineState *ms = MACHINE(vms);
380 
381     if (vms->gic_version == VIRT_GIC_VERSION_2) {
382         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
383                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
384                              (1 << MACHINE(vms)->smp.cpus) - 1);
385     }
386 
387     qemu_fdt_add_subnode(ms->fdt, "/timer");
388 
389     armcpu = ARM_CPU(qemu_get_cpu(0));
390     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
391         const char compat[] = "arm,armv8-timer\0arm,armv7-timer";
392         qemu_fdt_setprop(ms->fdt, "/timer", "compatible",
393                          compat, sizeof(compat));
394     } else {
395         qemu_fdt_setprop_string(ms->fdt, "/timer", "compatible",
396                                 "arm,armv7-timer");
397     }
398     qemu_fdt_setprop(ms->fdt, "/timer", "always-on", NULL, 0);
399     if (vms->ns_el2_virt_timer_irq) {
400         qemu_fdt_setprop_cells(ms->fdt, "/timer", "interrupts",
401                                GIC_FDT_IRQ_TYPE_PPI,
402                                INTID_TO_PPI(ARCH_TIMER_S_EL1_IRQ), irqflags,
403                                GIC_FDT_IRQ_TYPE_PPI,
404                                INTID_TO_PPI(ARCH_TIMER_NS_EL1_IRQ), irqflags,
405                                GIC_FDT_IRQ_TYPE_PPI,
406                                INTID_TO_PPI(ARCH_TIMER_VIRT_IRQ), irqflags,
407                                GIC_FDT_IRQ_TYPE_PPI,
408                                INTID_TO_PPI(ARCH_TIMER_NS_EL2_IRQ), irqflags,
409                                GIC_FDT_IRQ_TYPE_PPI,
410                                INTID_TO_PPI(ARCH_TIMER_NS_EL2_VIRT_IRQ), irqflags);
411     } else {
412         qemu_fdt_setprop_cells(ms->fdt, "/timer", "interrupts",
413                                GIC_FDT_IRQ_TYPE_PPI,
414                                INTID_TO_PPI(ARCH_TIMER_S_EL1_IRQ), irqflags,
415                                GIC_FDT_IRQ_TYPE_PPI,
416                                INTID_TO_PPI(ARCH_TIMER_NS_EL1_IRQ), irqflags,
417                                GIC_FDT_IRQ_TYPE_PPI,
418                                INTID_TO_PPI(ARCH_TIMER_VIRT_IRQ), irqflags,
419                                GIC_FDT_IRQ_TYPE_PPI,
420                                INTID_TO_PPI(ARCH_TIMER_NS_EL2_IRQ), irqflags);
421     }
422 }
423 
424 static void fdt_add_cpu_nodes(const VirtMachineState *vms)
425 {
426     int cpu;
427     int addr_cells = 1;
428     const MachineState *ms = MACHINE(vms);
429     const VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
430     int smp_cpus = ms->smp.cpus;
431 
432     /*
433      * See Linux Documentation/devicetree/bindings/arm/cpus.yaml
434      * On ARM v8 64-bit systems value should be set to 2,
435      * that corresponds to the MPIDR_EL1 register size.
436      * If MPIDR_EL1[63:32] value is equal to 0 on all CPUs
437      * in the system, #address-cells can be set to 1, since
438      * MPIDR_EL1[63:32] bits are not used for CPUs
439      * identification.
440      *
441      * Here we actually don't know whether our system is 32- or 64-bit one.
442      * The simplest way to go is to examine affinity IDs of all our CPUs. If
443      * at least one of them has Aff3 populated, we set #address-cells to 2.
444      */
445     for (cpu = 0; cpu < smp_cpus; cpu++) {
446         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
447 
448         if (arm_cpu_mp_affinity(armcpu) & ARM_AFF3_MASK) {
449             addr_cells = 2;
450             break;
451         }
452     }
453 
454     qemu_fdt_add_subnode(ms->fdt, "/cpus");
455     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#address-cells", addr_cells);
456     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#size-cells", 0x0);
457 
458     for (cpu = smp_cpus - 1; cpu >= 0; cpu--) {
459         char *nodename = g_strdup_printf("/cpus/cpu@%d", cpu);
460         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
461         CPUState *cs = CPU(armcpu);
462 
463         qemu_fdt_add_subnode(ms->fdt, nodename);
464         qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "cpu");
465         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
466                                     armcpu->dtb_compatible);
467 
468         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED && smp_cpus > 1) {
469             qemu_fdt_setprop_string(ms->fdt, nodename,
470                                         "enable-method", "psci");
471         }
472 
473         if (addr_cells == 2) {
474             qemu_fdt_setprop_u64(ms->fdt, nodename, "reg",
475                                  arm_cpu_mp_affinity(armcpu));
476         } else {
477             qemu_fdt_setprop_cell(ms->fdt, nodename, "reg",
478                                   arm_cpu_mp_affinity(armcpu));
479         }
480 
481         if (ms->possible_cpus->cpus[cs->cpu_index].props.has_node_id) {
482             qemu_fdt_setprop_cell(ms->fdt, nodename, "numa-node-id",
483                 ms->possible_cpus->cpus[cs->cpu_index].props.node_id);
484         }
485 
486         if (!vmc->no_cpu_topology) {
487             qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle",
488                                   qemu_fdt_alloc_phandle(ms->fdt));
489         }
490 
491         g_free(nodename);
492     }
493 
494     if (!vmc->no_cpu_topology) {
495         /*
496          * Add vCPU topology description through fdt node cpu-map.
497          *
498          * See Linux Documentation/devicetree/bindings/cpu/cpu-topology.txt
499          * In a SMP system, the hierarchy of CPUs can be defined through
500          * four entities that are used to describe the layout of CPUs in
501          * the system: socket/cluster/core/thread.
502          *
503          * A socket node represents the boundary of system physical package
504          * and its child nodes must be one or more cluster nodes. A system
505          * can contain several layers of clustering within a single physical
506          * package and cluster nodes can be contained in parent cluster nodes.
507          *
508          * Note: currently we only support one layer of clustering within
509          * each physical package.
510          */
511         qemu_fdt_add_subnode(ms->fdt, "/cpus/cpu-map");
512 
513         for (cpu = smp_cpus - 1; cpu >= 0; cpu--) {
514             char *cpu_path = g_strdup_printf("/cpus/cpu@%d", cpu);
515             char *map_path;
516 
517             if (ms->smp.threads > 1) {
518                 map_path = g_strdup_printf(
519                     "/cpus/cpu-map/socket%d/cluster%d/core%d/thread%d",
520                     cpu / (ms->smp.clusters * ms->smp.cores * ms->smp.threads),
521                     (cpu / (ms->smp.cores * ms->smp.threads)) % ms->smp.clusters,
522                     (cpu / ms->smp.threads) % ms->smp.cores,
523                     cpu % ms->smp.threads);
524             } else {
525                 map_path = g_strdup_printf(
526                     "/cpus/cpu-map/socket%d/cluster%d/core%d",
527                     cpu / (ms->smp.clusters * ms->smp.cores),
528                     (cpu / ms->smp.cores) % ms->smp.clusters,
529                     cpu % ms->smp.cores);
530             }
531             qemu_fdt_add_path(ms->fdt, map_path);
532             qemu_fdt_setprop_phandle(ms->fdt, map_path, "cpu", cpu_path);
533 
534             g_free(map_path);
535             g_free(cpu_path);
536         }
537     }
538 }
539 
540 static void fdt_add_its_gic_node(VirtMachineState *vms)
541 {
542     char *nodename;
543     MachineState *ms = MACHINE(vms);
544 
545     vms->msi_phandle = qemu_fdt_alloc_phandle(ms->fdt);
546     nodename = g_strdup_printf("/intc/its@%" PRIx64,
547                                vms->memmap[VIRT_GIC_ITS].base);
548     qemu_fdt_add_subnode(ms->fdt, nodename);
549     qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
550                             "arm,gic-v3-its");
551     qemu_fdt_setprop(ms->fdt, nodename, "msi-controller", NULL, 0);
552     qemu_fdt_setprop_cell(ms->fdt, nodename, "#msi-cells", 1);
553     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
554                                  2, vms->memmap[VIRT_GIC_ITS].base,
555                                  2, vms->memmap[VIRT_GIC_ITS].size);
556     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->msi_phandle);
557     g_free(nodename);
558 }
559 
560 static void fdt_add_v2m_gic_node(VirtMachineState *vms)
561 {
562     MachineState *ms = MACHINE(vms);
563     char *nodename;
564 
565     nodename = g_strdup_printf("/intc/v2m@%" PRIx64,
566                                vms->memmap[VIRT_GIC_V2M].base);
567     vms->msi_phandle = qemu_fdt_alloc_phandle(ms->fdt);
568     qemu_fdt_add_subnode(ms->fdt, nodename);
569     qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
570                             "arm,gic-v2m-frame");
571     qemu_fdt_setprop(ms->fdt, nodename, "msi-controller", NULL, 0);
572     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
573                                  2, vms->memmap[VIRT_GIC_V2M].base,
574                                  2, vms->memmap[VIRT_GIC_V2M].size);
575     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->msi_phandle);
576     g_free(nodename);
577 }
578 
579 static void fdt_add_gic_node(VirtMachineState *vms)
580 {
581     MachineState *ms = MACHINE(vms);
582     char *nodename;
583 
584     vms->gic_phandle = qemu_fdt_alloc_phandle(ms->fdt);
585     qemu_fdt_setprop_cell(ms->fdt, "/", "interrupt-parent", vms->gic_phandle);
586 
587     nodename = g_strdup_printf("/intc@%" PRIx64,
588                                vms->memmap[VIRT_GIC_DIST].base);
589     qemu_fdt_add_subnode(ms->fdt, nodename);
590     qemu_fdt_setprop_cell(ms->fdt, nodename, "#interrupt-cells", 3);
591     qemu_fdt_setprop(ms->fdt, nodename, "interrupt-controller", NULL, 0);
592     qemu_fdt_setprop_cell(ms->fdt, nodename, "#address-cells", 0x2);
593     qemu_fdt_setprop_cell(ms->fdt, nodename, "#size-cells", 0x2);
594     qemu_fdt_setprop(ms->fdt, nodename, "ranges", NULL, 0);
595     if (vms->gic_version != VIRT_GIC_VERSION_2) {
596         int nb_redist_regions = virt_gicv3_redist_region_count(vms);
597 
598         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
599                                 "arm,gic-v3");
600 
601         qemu_fdt_setprop_cell(ms->fdt, nodename,
602                               "#redistributor-regions", nb_redist_regions);
603 
604         if (nb_redist_regions == 1) {
605             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
606                                          2, vms->memmap[VIRT_GIC_DIST].base,
607                                          2, vms->memmap[VIRT_GIC_DIST].size,
608                                          2, vms->memmap[VIRT_GIC_REDIST].base,
609                                          2, vms->memmap[VIRT_GIC_REDIST].size);
610         } else {
611             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
612                                  2, vms->memmap[VIRT_GIC_DIST].base,
613                                  2, vms->memmap[VIRT_GIC_DIST].size,
614                                  2, vms->memmap[VIRT_GIC_REDIST].base,
615                                  2, vms->memmap[VIRT_GIC_REDIST].size,
616                                  2, vms->memmap[VIRT_HIGH_GIC_REDIST2].base,
617                                  2, vms->memmap[VIRT_HIGH_GIC_REDIST2].size);
618         }
619 
620         if (vms->virt) {
621             qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
622                                    GIC_FDT_IRQ_TYPE_PPI,
623                                    INTID_TO_PPI(ARCH_GIC_MAINT_IRQ),
624                                    GIC_FDT_IRQ_FLAGS_LEVEL_HI);
625         }
626     } else {
627         /* 'cortex-a15-gic' means 'GIC v2' */
628         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
629                                 "arm,cortex-a15-gic");
630         if (!vms->virt) {
631             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
632                                          2, vms->memmap[VIRT_GIC_DIST].base,
633                                          2, vms->memmap[VIRT_GIC_DIST].size,
634                                          2, vms->memmap[VIRT_GIC_CPU].base,
635                                          2, vms->memmap[VIRT_GIC_CPU].size);
636         } else {
637             qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
638                                          2, vms->memmap[VIRT_GIC_DIST].base,
639                                          2, vms->memmap[VIRT_GIC_DIST].size,
640                                          2, vms->memmap[VIRT_GIC_CPU].base,
641                                          2, vms->memmap[VIRT_GIC_CPU].size,
642                                          2, vms->memmap[VIRT_GIC_HYP].base,
643                                          2, vms->memmap[VIRT_GIC_HYP].size,
644                                          2, vms->memmap[VIRT_GIC_VCPU].base,
645                                          2, vms->memmap[VIRT_GIC_VCPU].size);
646             qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
647                                    GIC_FDT_IRQ_TYPE_PPI,
648                                    INTID_TO_PPI(ARCH_GIC_MAINT_IRQ),
649                                    GIC_FDT_IRQ_FLAGS_LEVEL_HI);
650         }
651     }
652 
653     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->gic_phandle);
654     g_free(nodename);
655 }
656 
657 static void fdt_add_pmu_nodes(const VirtMachineState *vms)
658 {
659     ARMCPU *armcpu = ARM_CPU(first_cpu);
660     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
661     MachineState *ms = MACHINE(vms);
662 
663     if (!arm_feature(&armcpu->env, ARM_FEATURE_PMU)) {
664         assert(!object_property_get_bool(OBJECT(armcpu), "pmu", NULL));
665         return;
666     }
667 
668     if (vms->gic_version == VIRT_GIC_VERSION_2) {
669         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
670                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
671                              (1 << MACHINE(vms)->smp.cpus) - 1);
672     }
673 
674     qemu_fdt_add_subnode(ms->fdt, "/pmu");
675     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
676         const char compat[] = "arm,armv8-pmuv3";
677         qemu_fdt_setprop(ms->fdt, "/pmu", "compatible",
678                          compat, sizeof(compat));
679         qemu_fdt_setprop_cells(ms->fdt, "/pmu", "interrupts",
680                                GIC_FDT_IRQ_TYPE_PPI,
681                                INTID_TO_PPI(VIRTUAL_PMU_IRQ), irqflags);
682     }
683 }
684 
685 static inline DeviceState *create_acpi_ged(VirtMachineState *vms)
686 {
687     DeviceState *dev;
688     MachineState *ms = MACHINE(vms);
689     int irq = vms->irqmap[VIRT_ACPI_GED];
690     uint32_t event = ACPI_GED_PWR_DOWN_EVT;
691 
692     if (ms->ram_slots) {
693         event |= ACPI_GED_MEM_HOTPLUG_EVT;
694     }
695 
696     if (ms->nvdimms_state->is_enabled) {
697         event |= ACPI_GED_NVDIMM_HOTPLUG_EVT;
698     }
699 
700     dev = qdev_new(TYPE_ACPI_GED);
701     qdev_prop_set_uint32(dev, "ged-event", event);
702     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
703 
704     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_ACPI_GED].base);
705     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 1, vms->memmap[VIRT_PCDIMM_ACPI].base);
706     sysbus_connect_irq(SYS_BUS_DEVICE(dev), 0, qdev_get_gpio_in(vms->gic, irq));
707 
708     return dev;
709 }
710 
711 static void create_its(VirtMachineState *vms)
712 {
713     DeviceState *dev;
714 
715     assert(vms->its);
716     if (!kvm_irqchip_in_kernel() && !vms->tcg_its) {
717         /*
718          * Do nothing if ITS is neither supported by the host nor emulated by
719          * the machine.
720          */
721         return;
722     }
723 
724     dev = qdev_new(its_class_name());
725 
726     object_property_set_link(OBJECT(dev), "parent-gicv3", OBJECT(vms->gic),
727                              &error_abort);
728     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
729     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_ITS].base);
730 
731     fdt_add_its_gic_node(vms);
732     vms->msi_controller = VIRT_MSI_CTRL_ITS;
733 }
734 
735 static void create_v2m(VirtMachineState *vms)
736 {
737     int i;
738     int irq = vms->irqmap[VIRT_GIC_V2M];
739     DeviceState *dev;
740 
741     dev = qdev_new("arm-gicv2m");
742     qdev_prop_set_uint32(dev, "base-spi", irq);
743     qdev_prop_set_uint32(dev, "num-spi", NUM_GICV2M_SPIS);
744     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
745     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_V2M].base);
746 
747     for (i = 0; i < NUM_GICV2M_SPIS; i++) {
748         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
749                            qdev_get_gpio_in(vms->gic, irq + i));
750     }
751 
752     fdt_add_v2m_gic_node(vms);
753     vms->msi_controller = VIRT_MSI_CTRL_GICV2M;
754 }
755 
756 /*
757  * If the CPU has FEAT_NMI, then turn on the NMI support in the GICv3 too.
758  * It's permitted to have a configuration with NMI in the CPU (and thus the
759  * GICv3 CPU interface) but not in the distributor/redistributors, but it's
760  * not very useful.
761  */
762 static bool gicv3_nmi_present(VirtMachineState *vms)
763 {
764     ARMCPU *cpu = ARM_CPU(qemu_get_cpu(0));
765 
766     return tcg_enabled() && cpu_isar_feature(aa64_nmi, cpu) &&
767            (vms->gic_version != VIRT_GIC_VERSION_2);
768 }
769 
770 static void create_gic(VirtMachineState *vms, MemoryRegion *mem)
771 {
772     MachineState *ms = MACHINE(vms);
773     /* We create a standalone GIC */
774     SysBusDevice *gicbusdev;
775     const char *gictype;
776     int i;
777     unsigned int smp_cpus = ms->smp.cpus;
778     uint32_t nb_redist_regions = 0;
779     int revision;
780 
781     if (vms->gic_version == VIRT_GIC_VERSION_2) {
782         gictype = gic_class_name();
783     } else {
784         gictype = gicv3_class_name();
785     }
786 
787     switch (vms->gic_version) {
788     case VIRT_GIC_VERSION_2:
789         revision = 2;
790         break;
791     case VIRT_GIC_VERSION_3:
792         revision = 3;
793         break;
794     case VIRT_GIC_VERSION_4:
795         revision = 4;
796         break;
797     default:
798         g_assert_not_reached();
799     }
800 
801     if (kvm_enabled() && vms->virt &&
802         (revision != 3 || !kvm_irqchip_in_kernel())) {
803         error_report("KVM EL2 is only supported with in-kernel GICv3");
804         exit(1);
805     }
806 
807     vms->gic = qdev_new(gictype);
808     qdev_prop_set_uint32(vms->gic, "revision", revision);
809     qdev_prop_set_uint32(vms->gic, "num-cpu", smp_cpus);
810     /* Note that the num-irq property counts both internal and external
811      * interrupts; there are always 32 of the former (mandated by GIC spec).
812      */
813     qdev_prop_set_uint32(vms->gic, "num-irq", NUM_IRQS + 32);
814     if (!kvm_irqchip_in_kernel()) {
815         qdev_prop_set_bit(vms->gic, "has-security-extensions", vms->secure);
816     }
817 
818     if (vms->gic_version != VIRT_GIC_VERSION_2) {
819         QList *redist_region_count;
820         uint32_t redist0_capacity = virt_redist_capacity(vms, VIRT_GIC_REDIST);
821         uint32_t redist0_count = MIN(smp_cpus, redist0_capacity);
822 
823         nb_redist_regions = virt_gicv3_redist_region_count(vms);
824 
825         redist_region_count = qlist_new();
826         qlist_append_int(redist_region_count, redist0_count);
827         if (nb_redist_regions == 2) {
828             uint32_t redist1_capacity =
829                 virt_redist_capacity(vms, VIRT_HIGH_GIC_REDIST2);
830 
831             qlist_append_int(redist_region_count,
832                 MIN(smp_cpus - redist0_count, redist1_capacity));
833         }
834         qdev_prop_set_array(vms->gic, "redist-region-count",
835                             redist_region_count);
836 
837         if (!kvm_irqchip_in_kernel()) {
838             if (vms->tcg_its) {
839                 object_property_set_link(OBJECT(vms->gic), "sysmem",
840                                          OBJECT(mem), &error_fatal);
841                 qdev_prop_set_bit(vms->gic, "has-lpi", true);
842             }
843         } else if (vms->virt) {
844             qdev_prop_set_uint32(vms->gic, "maintenance-interrupt-id",
845                                  ARCH_GIC_MAINT_IRQ);
846         }
847     } else {
848         if (!kvm_irqchip_in_kernel()) {
849             qdev_prop_set_bit(vms->gic, "has-virtualization-extensions",
850                               vms->virt);
851         }
852     }
853 
854     if (gicv3_nmi_present(vms)) {
855         qdev_prop_set_bit(vms->gic, "has-nmi", true);
856     }
857 
858     gicbusdev = SYS_BUS_DEVICE(vms->gic);
859     sysbus_realize_and_unref(gicbusdev, &error_fatal);
860     sysbus_mmio_map(gicbusdev, 0, vms->memmap[VIRT_GIC_DIST].base);
861     if (vms->gic_version != VIRT_GIC_VERSION_2) {
862         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_REDIST].base);
863         if (nb_redist_regions == 2) {
864             sysbus_mmio_map(gicbusdev, 2,
865                             vms->memmap[VIRT_HIGH_GIC_REDIST2].base);
866         }
867     } else {
868         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base);
869         if (vms->virt) {
870             sysbus_mmio_map(gicbusdev, 2, vms->memmap[VIRT_GIC_HYP].base);
871             sysbus_mmio_map(gicbusdev, 3, vms->memmap[VIRT_GIC_VCPU].base);
872         }
873     }
874 
875     /* Wire the outputs from each CPU's generic timer and the GICv3
876      * maintenance interrupt signal to the appropriate GIC PPI inputs,
877      * and the GIC's IRQ/FIQ/VIRQ/VFIQ/NMI/VINMI interrupt outputs to the
878      * CPU's inputs.
879      */
880     for (i = 0; i < smp_cpus; i++) {
881         DeviceState *cpudev = DEVICE(qemu_get_cpu(i));
882         int intidbase = NUM_IRQS + i * GIC_INTERNAL;
883         /* Mapping from the output timer irq lines from the CPU to the
884          * GIC PPI inputs we use for the virt board.
885          */
886         const int timer_irq[] = {
887             [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ,
888             [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ,
889             [GTIMER_HYP]  = ARCH_TIMER_NS_EL2_IRQ,
890             [GTIMER_SEC]  = ARCH_TIMER_S_EL1_IRQ,
891             [GTIMER_HYPVIRT] = ARCH_TIMER_NS_EL2_VIRT_IRQ,
892             [GTIMER_S_EL2_PHYS] = ARCH_TIMER_S_EL2_IRQ,
893             [GTIMER_S_EL2_VIRT] = ARCH_TIMER_S_EL2_VIRT_IRQ,
894         };
895 
896         for (unsigned irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) {
897             qdev_connect_gpio_out(cpudev, irq,
898                                   qdev_get_gpio_in(vms->gic,
899                                                    intidbase + timer_irq[irq]));
900         }
901 
902         if (vms->gic_version != VIRT_GIC_VERSION_2) {
903             qemu_irq irq = qdev_get_gpio_in(vms->gic,
904                                             intidbase + ARCH_GIC_MAINT_IRQ);
905             qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt",
906                                         0, irq);
907         } else if (vms->virt) {
908             qemu_irq irq = qdev_get_gpio_in(vms->gic,
909                                             intidbase + ARCH_GIC_MAINT_IRQ);
910             sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus, irq);
911         }
912 
913         qdev_connect_gpio_out_named(cpudev, "pmu-interrupt", 0,
914                                     qdev_get_gpio_in(vms->gic, intidbase
915                                                      + VIRTUAL_PMU_IRQ));
916 
917         sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ));
918         sysbus_connect_irq(gicbusdev, i + smp_cpus,
919                            qdev_get_gpio_in(cpudev, ARM_CPU_FIQ));
920         sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus,
921                            qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ));
922         sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus,
923                            qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ));
924 
925         if (vms->gic_version != VIRT_GIC_VERSION_2) {
926             sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus,
927                                qdev_get_gpio_in(cpudev, ARM_CPU_NMI));
928             sysbus_connect_irq(gicbusdev, i + 5 * smp_cpus,
929                                qdev_get_gpio_in(cpudev, ARM_CPU_VINMI));
930         }
931     }
932 
933     fdt_add_gic_node(vms);
934 
935     if (vms->gic_version != VIRT_GIC_VERSION_2 && vms->its) {
936         create_its(vms);
937     } else if (vms->gic_version == VIRT_GIC_VERSION_2) {
938         create_v2m(vms);
939     }
940 }
941 
942 static void create_uart(const VirtMachineState *vms, int uart,
943                         MemoryRegion *mem, Chardev *chr, bool secure)
944 {
945     char *nodename;
946     hwaddr base = vms->memmap[uart].base;
947     hwaddr size = vms->memmap[uart].size;
948     int irq = vms->irqmap[uart];
949     const char compat[] = "arm,pl011\0arm,primecell";
950     const char clocknames[] = "uartclk\0apb_pclk";
951     DeviceState *dev = qdev_new(TYPE_PL011);
952     SysBusDevice *s = SYS_BUS_DEVICE(dev);
953     MachineState *ms = MACHINE(vms);
954 
955     qdev_prop_set_chr(dev, "chardev", chr);
956     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
957     memory_region_add_subregion(mem, base,
958                                 sysbus_mmio_get_region(s, 0));
959     sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
960 
961     nodename = g_strdup_printf("/pl011@%" PRIx64, base);
962     qemu_fdt_add_subnode(ms->fdt, nodename);
963     /* Note that we can't use setprop_string because of the embedded NUL */
964     qemu_fdt_setprop(ms->fdt, nodename, "compatible",
965                          compat, sizeof(compat));
966     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
967                                      2, base, 2, size);
968     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
969                                GIC_FDT_IRQ_TYPE_SPI, irq,
970                                GIC_FDT_IRQ_FLAGS_LEVEL_HI);
971     qemu_fdt_setprop_cells(ms->fdt, nodename, "clocks",
972                                vms->clock_phandle, vms->clock_phandle);
973     qemu_fdt_setprop(ms->fdt, nodename, "clock-names",
974                          clocknames, sizeof(clocknames));
975 
976     if (uart == VIRT_UART0) {
977         qemu_fdt_setprop_string(ms->fdt, "/chosen", "stdout-path", nodename);
978         qemu_fdt_setprop_string(ms->fdt, "/aliases", "serial0", nodename);
979     } else {
980         qemu_fdt_setprop_string(ms->fdt, "/aliases", "serial1", nodename);
981     }
982     if (secure) {
983         /* Mark as not usable by the normal world */
984         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
985         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
986 
987         qemu_fdt_setprop_string(ms->fdt, "/secure-chosen", "stdout-path",
988                                 nodename);
989     }
990 
991     g_free(nodename);
992 }
993 
994 static void create_rtc(const VirtMachineState *vms)
995 {
996     char *nodename;
997     hwaddr base = vms->memmap[VIRT_RTC].base;
998     hwaddr size = vms->memmap[VIRT_RTC].size;
999     int irq = vms->irqmap[VIRT_RTC];
1000     const char compat[] = "arm,pl031\0arm,primecell";
1001     MachineState *ms = MACHINE(vms);
1002 
1003     sysbus_create_simple("pl031", base, qdev_get_gpio_in(vms->gic, irq));
1004 
1005     nodename = g_strdup_printf("/pl031@%" PRIx64, base);
1006     qemu_fdt_add_subnode(ms->fdt, nodename);
1007     qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
1008     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1009                                  2, base, 2, size);
1010     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1011                            GIC_FDT_IRQ_TYPE_SPI, irq,
1012                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
1013     qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
1014     qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
1015     g_free(nodename);
1016 }
1017 
1018 static DeviceState *gpio_key_dev;
1019 static void virt_powerdown_req(Notifier *n, void *opaque)
1020 {
1021     VirtMachineState *s = container_of(n, VirtMachineState, powerdown_notifier);
1022 
1023     if (s->acpi_dev) {
1024         acpi_send_event(s->acpi_dev, ACPI_POWER_DOWN_STATUS);
1025     } else {
1026         /* use gpio Pin for power button event */
1027         qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1);
1028     }
1029 }
1030 
1031 static void create_gpio_keys(char *fdt, DeviceState *pl061_dev,
1032                              uint32_t phandle)
1033 {
1034     gpio_key_dev = sysbus_create_simple("gpio-key", -1,
1035                                         qdev_get_gpio_in(pl061_dev,
1036                                                          GPIO_PIN_POWER_BUTTON));
1037 
1038     qemu_fdt_add_subnode(fdt, "/gpio-keys");
1039     qemu_fdt_setprop_string(fdt, "/gpio-keys", "compatible", "gpio-keys");
1040 
1041     qemu_fdt_add_subnode(fdt, "/gpio-keys/poweroff");
1042     qemu_fdt_setprop_string(fdt, "/gpio-keys/poweroff",
1043                             "label", "GPIO Key Poweroff");
1044     qemu_fdt_setprop_cell(fdt, "/gpio-keys/poweroff", "linux,code",
1045                           KEY_POWER);
1046     qemu_fdt_setprop_cells(fdt, "/gpio-keys/poweroff",
1047                            "gpios", phandle, GPIO_PIN_POWER_BUTTON, 0);
1048 }
1049 
1050 #define SECURE_GPIO_POWEROFF 0
1051 #define SECURE_GPIO_RESET    1
1052 
1053 static void create_secure_gpio_pwr(char *fdt, DeviceState *pl061_dev,
1054                                    uint32_t phandle)
1055 {
1056     DeviceState *gpio_pwr_dev;
1057 
1058     /* gpio-pwr */
1059     gpio_pwr_dev = sysbus_create_simple("gpio-pwr", -1, NULL);
1060 
1061     /* connect secure pl061 to gpio-pwr */
1062     qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_RESET,
1063                           qdev_get_gpio_in_named(gpio_pwr_dev, "reset", 0));
1064     qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_POWEROFF,
1065                           qdev_get_gpio_in_named(gpio_pwr_dev, "shutdown", 0));
1066 
1067     qemu_fdt_add_subnode(fdt, "/gpio-poweroff");
1068     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "compatible",
1069                             "gpio-poweroff");
1070     qemu_fdt_setprop_cells(fdt, "/gpio-poweroff",
1071                            "gpios", phandle, SECURE_GPIO_POWEROFF, 0);
1072     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "status", "disabled");
1073     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "secure-status",
1074                             "okay");
1075 
1076     qemu_fdt_add_subnode(fdt, "/gpio-restart");
1077     qemu_fdt_setprop_string(fdt, "/gpio-restart", "compatible",
1078                             "gpio-restart");
1079     qemu_fdt_setprop_cells(fdt, "/gpio-restart",
1080                            "gpios", phandle, SECURE_GPIO_RESET, 0);
1081     qemu_fdt_setprop_string(fdt, "/gpio-restart", "status", "disabled");
1082     qemu_fdt_setprop_string(fdt, "/gpio-restart", "secure-status",
1083                             "okay");
1084 }
1085 
1086 static void create_gpio_devices(const VirtMachineState *vms, int gpio,
1087                                 MemoryRegion *mem)
1088 {
1089     char *nodename;
1090     DeviceState *pl061_dev;
1091     hwaddr base = vms->memmap[gpio].base;
1092     hwaddr size = vms->memmap[gpio].size;
1093     int irq = vms->irqmap[gpio];
1094     const char compat[] = "arm,pl061\0arm,primecell";
1095     SysBusDevice *s;
1096     MachineState *ms = MACHINE(vms);
1097 
1098     pl061_dev = qdev_new("pl061");
1099     /* Pull lines down to 0 if not driven by the PL061 */
1100     qdev_prop_set_uint32(pl061_dev, "pullups", 0);
1101     qdev_prop_set_uint32(pl061_dev, "pulldowns", 0xff);
1102     s = SYS_BUS_DEVICE(pl061_dev);
1103     sysbus_realize_and_unref(s, &error_fatal);
1104     memory_region_add_subregion(mem, base, sysbus_mmio_get_region(s, 0));
1105     sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
1106 
1107     uint32_t phandle = qemu_fdt_alloc_phandle(ms->fdt);
1108     nodename = g_strdup_printf("/pl061@%" PRIx64, base);
1109     qemu_fdt_add_subnode(ms->fdt, nodename);
1110     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1111                                  2, base, 2, size);
1112     qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
1113     qemu_fdt_setprop_cell(ms->fdt, nodename, "#gpio-cells", 2);
1114     qemu_fdt_setprop(ms->fdt, nodename, "gpio-controller", NULL, 0);
1115     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1116                            GIC_FDT_IRQ_TYPE_SPI, irq,
1117                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
1118     qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
1119     qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
1120     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", phandle);
1121 
1122     if (gpio != VIRT_GPIO) {
1123         /* Mark as not usable by the normal world */
1124         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1125         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1126     }
1127     g_free(nodename);
1128 
1129     /* Child gpio devices */
1130     if (gpio == VIRT_GPIO) {
1131         create_gpio_keys(ms->fdt, pl061_dev, phandle);
1132     } else {
1133         create_secure_gpio_pwr(ms->fdt, pl061_dev, phandle);
1134     }
1135 }
1136 
1137 static void create_virtio_devices(const VirtMachineState *vms)
1138 {
1139     int i;
1140     hwaddr size = vms->memmap[VIRT_MMIO].size;
1141     MachineState *ms = MACHINE(vms);
1142 
1143     /* We create the transports in forwards order. Since qbus_realize()
1144      * prepends (not appends) new child buses, the incrementing loop below will
1145      * create a list of virtio-mmio buses with decreasing base addresses.
1146      *
1147      * When a -device option is processed from the command line,
1148      * qbus_find_recursive() picks the next free virtio-mmio bus in forwards
1149      * order. The upshot is that -device options in increasing command line
1150      * order are mapped to virtio-mmio buses with decreasing base addresses.
1151      *
1152      * When this code was originally written, that arrangement ensured that the
1153      * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to
1154      * the first -device on the command line. (The end-to-end order is a
1155      * function of this loop, qbus_realize(), qbus_find_recursive(), and the
1156      * guest kernel's name-to-address assignment strategy.)
1157      *
1158      * Meanwhile, the kernel's traversal seems to have been reversed; see eg.
1159      * the message, if not necessarily the code, of commit 70161ff336.
1160      * Therefore the loop now establishes the inverse of the original intent.
1161      *
1162      * Unfortunately, we can't counteract the kernel change by reversing the
1163      * loop; it would break existing command lines.
1164      *
1165      * In any case, the kernel makes no guarantee about the stability of
1166      * enumeration order of virtio devices (as demonstrated by it changing
1167      * between kernel versions). For reliable and stable identification
1168      * of disks users must use UUIDs or similar mechanisms.
1169      */
1170     for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) {
1171         int irq = vms->irqmap[VIRT_MMIO] + i;
1172         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1173 
1174         sysbus_create_simple("virtio-mmio", base,
1175                              qdev_get_gpio_in(vms->gic, irq));
1176     }
1177 
1178     /* We add dtb nodes in reverse order so that they appear in the finished
1179      * device tree lowest address first.
1180      *
1181      * Note that this mapping is independent of the loop above. The previous
1182      * loop influences virtio device to virtio transport assignment, whereas
1183      * this loop controls how virtio transports are laid out in the dtb.
1184      */
1185     for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) {
1186         char *nodename;
1187         int irq = vms->irqmap[VIRT_MMIO] + i;
1188         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1189 
1190         nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base);
1191         qemu_fdt_add_subnode(ms->fdt, nodename);
1192         qemu_fdt_setprop_string(ms->fdt, nodename,
1193                                 "compatible", "virtio,mmio");
1194         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1195                                      2, base, 2, size);
1196         qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1197                                GIC_FDT_IRQ_TYPE_SPI, irq,
1198                                GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1199         qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1200         g_free(nodename);
1201     }
1202 }
1203 
1204 #define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
1205 
1206 static PFlashCFI01 *virt_flash_create1(VirtMachineState *vms,
1207                                         const char *name,
1208                                         const char *alias_prop_name)
1209 {
1210     /*
1211      * Create a single flash device.  We use the same parameters as
1212      * the flash devices on the Versatile Express board.
1213      */
1214     DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
1215 
1216     qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
1217     qdev_prop_set_uint8(dev, "width", 4);
1218     qdev_prop_set_uint8(dev, "device-width", 2);
1219     qdev_prop_set_bit(dev, "big-endian", false);
1220     qdev_prop_set_uint16(dev, "id0", 0x89);
1221     qdev_prop_set_uint16(dev, "id1", 0x18);
1222     qdev_prop_set_uint16(dev, "id2", 0x00);
1223     qdev_prop_set_uint16(dev, "id3", 0x00);
1224     qdev_prop_set_string(dev, "name", name);
1225     object_property_add_child(OBJECT(vms), name, OBJECT(dev));
1226     object_property_add_alias(OBJECT(vms), alias_prop_name,
1227                               OBJECT(dev), "drive");
1228     return PFLASH_CFI01(dev);
1229 }
1230 
1231 static void virt_flash_create(VirtMachineState *vms)
1232 {
1233     vms->flash[0] = virt_flash_create1(vms, "virt.flash0", "pflash0");
1234     vms->flash[1] = virt_flash_create1(vms, "virt.flash1", "pflash1");
1235 }
1236 
1237 static void virt_flash_map1(PFlashCFI01 *flash,
1238                             hwaddr base, hwaddr size,
1239                             MemoryRegion *sysmem)
1240 {
1241     DeviceState *dev = DEVICE(flash);
1242 
1243     assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
1244     assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
1245     qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
1246     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1247 
1248     memory_region_add_subregion(sysmem, base,
1249                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
1250                                                        0));
1251 }
1252 
1253 static void virt_flash_map(VirtMachineState *vms,
1254                            MemoryRegion *sysmem,
1255                            MemoryRegion *secure_sysmem)
1256 {
1257     /*
1258      * Map two flash devices to fill the VIRT_FLASH space in the memmap.
1259      * sysmem is the system memory space. secure_sysmem is the secure view
1260      * of the system, and the first flash device should be made visible only
1261      * there. The second flash device is visible to both secure and nonsecure.
1262      * If sysmem == secure_sysmem this means there is no separate Secure
1263      * address space and both flash devices are generally visible.
1264      */
1265     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1266     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1267 
1268     virt_flash_map1(vms->flash[0], flashbase, flashsize,
1269                     secure_sysmem);
1270     virt_flash_map1(vms->flash[1], flashbase + flashsize, flashsize,
1271                     sysmem);
1272 }
1273 
1274 static void virt_flash_fdt(VirtMachineState *vms,
1275                            MemoryRegion *sysmem,
1276                            MemoryRegion *secure_sysmem)
1277 {
1278     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1279     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1280     MachineState *ms = MACHINE(vms);
1281     char *nodename;
1282 
1283     if (sysmem == secure_sysmem) {
1284         /* Report both flash devices as a single node in the DT */
1285         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
1286         qemu_fdt_add_subnode(ms->fdt, nodename);
1287         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1288         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1289                                      2, flashbase, 2, flashsize,
1290                                      2, flashbase + flashsize, 2, flashsize);
1291         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1292         g_free(nodename);
1293     } else {
1294         /*
1295          * Report the devices as separate nodes so we can mark one as
1296          * only visible to the secure world.
1297          */
1298         nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase);
1299         qemu_fdt_add_subnode(ms->fdt, nodename);
1300         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1301         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1302                                      2, flashbase, 2, flashsize);
1303         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1304         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1305         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1306         g_free(nodename);
1307 
1308         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase + flashsize);
1309         qemu_fdt_add_subnode(ms->fdt, nodename);
1310         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1311         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1312                                      2, flashbase + flashsize, 2, flashsize);
1313         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1314         g_free(nodename);
1315     }
1316 }
1317 
1318 static bool virt_firmware_init(VirtMachineState *vms,
1319                                MemoryRegion *sysmem,
1320                                MemoryRegion *secure_sysmem)
1321 {
1322     int i;
1323     const char *bios_name;
1324     BlockBackend *pflash_blk0;
1325 
1326     /* Map legacy -drive if=pflash to machine properties */
1327     for (i = 0; i < ARRAY_SIZE(vms->flash); i++) {
1328         pflash_cfi01_legacy_drive(vms->flash[i],
1329                                   drive_get(IF_PFLASH, 0, i));
1330     }
1331 
1332     virt_flash_map(vms, sysmem, secure_sysmem);
1333 
1334     pflash_blk0 = pflash_cfi01_get_blk(vms->flash[0]);
1335 
1336     bios_name = MACHINE(vms)->firmware;
1337     if (bios_name) {
1338         char *fname;
1339         MemoryRegion *mr;
1340         int image_size;
1341 
1342         if (pflash_blk0) {
1343             error_report("The contents of the first flash device may be "
1344                          "specified with -bios or with -drive if=pflash... "
1345                          "but you cannot use both options at once");
1346             exit(1);
1347         }
1348 
1349         /* Fall back to -bios */
1350 
1351         fname = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1352         if (!fname) {
1353             error_report("Could not find ROM image '%s'", bios_name);
1354             exit(1);
1355         }
1356         mr = sysbus_mmio_get_region(SYS_BUS_DEVICE(vms->flash[0]), 0);
1357         image_size = load_image_mr(fname, mr);
1358         g_free(fname);
1359         if (image_size < 0) {
1360             error_report("Could not load ROM image '%s'", bios_name);
1361             exit(1);
1362         }
1363     }
1364 
1365     return pflash_blk0 || bios_name;
1366 }
1367 
1368 static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as)
1369 {
1370     MachineState *ms = MACHINE(vms);
1371     hwaddr base = vms->memmap[VIRT_FW_CFG].base;
1372     hwaddr size = vms->memmap[VIRT_FW_CFG].size;
1373     FWCfgState *fw_cfg;
1374     char *nodename;
1375 
1376     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as);
1377     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)ms->smp.cpus);
1378 
1379     nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
1380     qemu_fdt_add_subnode(ms->fdt, nodename);
1381     qemu_fdt_setprop_string(ms->fdt, nodename,
1382                             "compatible", "qemu,fw-cfg-mmio");
1383     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1384                                  2, base, 2, size);
1385     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1386     g_free(nodename);
1387     return fw_cfg;
1388 }
1389 
1390 static void create_pcie_irq_map(const MachineState *ms,
1391                                 uint32_t gic_phandle,
1392                                 int first_irq, const char *nodename)
1393 {
1394     int devfn, pin;
1395     uint32_t full_irq_map[4 * 4 * 10] = { 0 };
1396     uint32_t *irq_map = full_irq_map;
1397 
1398     for (devfn = 0; devfn <= 0x18; devfn += 0x8) {
1399         for (pin = 0; pin < 4; pin++) {
1400             int irq_type = GIC_FDT_IRQ_TYPE_SPI;
1401             int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS);
1402             int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
1403             int i;
1404 
1405             uint32_t map[] = {
1406                 devfn << 8, 0, 0,                           /* devfn */
1407                 pin + 1,                                    /* PCI pin */
1408                 gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */
1409 
1410             /* Convert map to big endian */
1411             for (i = 0; i < 10; i++) {
1412                 irq_map[i] = cpu_to_be32(map[i]);
1413             }
1414             irq_map += 10;
1415         }
1416     }
1417 
1418     qemu_fdt_setprop(ms->fdt, nodename, "interrupt-map",
1419                      full_irq_map, sizeof(full_irq_map));
1420 
1421     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupt-map-mask",
1422                            cpu_to_be16(PCI_DEVFN(3, 0)), /* Slot 3 */
1423                            0, 0,
1424                            0x7           /* PCI irq */);
1425 }
1426 
1427 static void create_smmu(const VirtMachineState *vms,
1428                         PCIBus *bus)
1429 {
1430     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1431     char *node;
1432     const char compat[] = "arm,smmu-v3";
1433     int irq =  vms->irqmap[VIRT_SMMU];
1434     int i;
1435     hwaddr base = vms->memmap[VIRT_SMMU].base;
1436     hwaddr size = vms->memmap[VIRT_SMMU].size;
1437     const char irq_names[] = "eventq\0priq\0cmdq-sync\0gerror";
1438     DeviceState *dev;
1439     MachineState *ms = MACHINE(vms);
1440 
1441     if (vms->iommu != VIRT_IOMMU_SMMUV3 || !vms->iommu_phandle) {
1442         return;
1443     }
1444 
1445     dev = qdev_new(TYPE_ARM_SMMUV3);
1446 
1447     if (!vmc->no_nested_smmu) {
1448         object_property_set_str(OBJECT(dev), "stage", "nested", &error_fatal);
1449     }
1450     object_property_set_link(OBJECT(dev), "primary-bus", OBJECT(bus),
1451                              &error_abort);
1452     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1453     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, base);
1454     for (i = 0; i < NUM_SMMU_IRQS; i++) {
1455         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1456                            qdev_get_gpio_in(vms->gic, irq + i));
1457     }
1458 
1459     node = g_strdup_printf("/smmuv3@%" PRIx64, base);
1460     qemu_fdt_add_subnode(ms->fdt, node);
1461     qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1462     qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg", 2, base, 2, size);
1463 
1464     qemu_fdt_setprop_cells(ms->fdt, node, "interrupts",
1465             GIC_FDT_IRQ_TYPE_SPI, irq    , GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1466             GIC_FDT_IRQ_TYPE_SPI, irq + 1, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1467             GIC_FDT_IRQ_TYPE_SPI, irq + 2, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1468             GIC_FDT_IRQ_TYPE_SPI, irq + 3, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1469 
1470     qemu_fdt_setprop(ms->fdt, node, "interrupt-names", irq_names,
1471                      sizeof(irq_names));
1472 
1473     qemu_fdt_setprop(ms->fdt, node, "dma-coherent", NULL, 0);
1474 
1475     qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1476 
1477     qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1478     g_free(node);
1479 }
1480 
1481 static void create_virtio_iommu_dt_bindings(VirtMachineState *vms)
1482 {
1483     const char compat[] = "virtio,pci-iommu\0pci1af4,1057";
1484     uint16_t bdf = vms->virtio_iommu_bdf;
1485     MachineState *ms = MACHINE(vms);
1486     char *node;
1487 
1488     vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1489 
1490     node = g_strdup_printf("%s/virtio_iommu@%x,%x", vms->pciehb_nodename,
1491                            PCI_SLOT(bdf), PCI_FUNC(bdf));
1492     qemu_fdt_add_subnode(ms->fdt, node);
1493     qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1494     qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg",
1495                                  1, bdf << 8, 1, 0, 1, 0,
1496                                  1, 0, 1, 0);
1497 
1498     qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1499     qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1500     g_free(node);
1501 
1502     if (!vms->default_bus_bypass_iommu) {
1503         qemu_fdt_setprop_cells(ms->fdt, vms->pciehb_nodename, "iommu-map",
1504                                0x0, vms->iommu_phandle, 0x0, bdf,
1505                                bdf + 1, vms->iommu_phandle, bdf + 1,
1506                                0xffff - bdf);
1507     }
1508 }
1509 
1510 static void create_pcie(VirtMachineState *vms)
1511 {
1512     hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base;
1513     hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size;
1514     hwaddr base_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].base;
1515     hwaddr size_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].size;
1516     hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base;
1517     hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size;
1518     hwaddr base_ecam, size_ecam;
1519     hwaddr base = base_mmio;
1520     int nr_pcie_buses;
1521     int irq = vms->irqmap[VIRT_PCIE];
1522     MemoryRegion *mmio_alias;
1523     MemoryRegion *mmio_reg;
1524     MemoryRegion *ecam_alias;
1525     MemoryRegion *ecam_reg;
1526     DeviceState *dev;
1527     char *nodename;
1528     int i, ecam_id;
1529     PCIHostState *pci;
1530     MachineState *ms = MACHINE(vms);
1531     MachineClass *mc = MACHINE_GET_CLASS(ms);
1532 
1533     dev = qdev_new(TYPE_GPEX_HOST);
1534     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1535 
1536     ecam_id = VIRT_ECAM_ID(vms->highmem_ecam);
1537     base_ecam = vms->memmap[ecam_id].base;
1538     size_ecam = vms->memmap[ecam_id].size;
1539     nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN;
1540     /* Map only the first size_ecam bytes of ECAM space */
1541     ecam_alias = g_new0(MemoryRegion, 1);
1542     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1543     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1544                              ecam_reg, 0, size_ecam);
1545     memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias);
1546 
1547     /* Map the MMIO window into system address space so as to expose
1548      * the section of PCI MMIO space which starts at the same base address
1549      * (ie 1:1 mapping for that part of PCI MMIO space visible through
1550      * the window).
1551      */
1552     mmio_alias = g_new0(MemoryRegion, 1);
1553     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1554     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1555                              mmio_reg, base_mmio, size_mmio);
1556     memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias);
1557 
1558     if (vms->highmem_mmio) {
1559         /* Map high MMIO space */
1560         MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1);
1561 
1562         memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1563                                  mmio_reg, base_mmio_high, size_mmio_high);
1564         memory_region_add_subregion(get_system_memory(), base_mmio_high,
1565                                     high_mmio_alias);
1566     }
1567 
1568     /* Map IO port space */
1569     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio);
1570 
1571     for (i = 0; i < PCI_NUM_PINS; i++) {
1572         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1573                            qdev_get_gpio_in(vms->gic, irq + i));
1574         gpex_set_irq_num(GPEX_HOST(dev), i, irq + i);
1575     }
1576 
1577     pci = PCI_HOST_BRIDGE(dev);
1578     pci->bypass_iommu = vms->default_bus_bypass_iommu;
1579     vms->bus = pci->bus;
1580     if (vms->bus) {
1581         pci_init_nic_devices(pci->bus, mc->default_nic);
1582     }
1583 
1584     nodename = vms->pciehb_nodename = g_strdup_printf("/pcie@%" PRIx64, base);
1585     qemu_fdt_add_subnode(ms->fdt, nodename);
1586     qemu_fdt_setprop_string(ms->fdt, nodename,
1587                             "compatible", "pci-host-ecam-generic");
1588     qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "pci");
1589     qemu_fdt_setprop_cell(ms->fdt, nodename, "#address-cells", 3);
1590     qemu_fdt_setprop_cell(ms->fdt, nodename, "#size-cells", 2);
1591     qemu_fdt_setprop_cell(ms->fdt, nodename, "linux,pci-domain", 0);
1592     qemu_fdt_setprop_cells(ms->fdt, nodename, "bus-range", 0,
1593                            nr_pcie_buses - 1);
1594     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1595 
1596     if (vms->msi_phandle) {
1597         qemu_fdt_setprop_cells(ms->fdt, nodename, "msi-map",
1598                                0, vms->msi_phandle, 0, 0x10000);
1599     }
1600 
1601     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1602                                  2, base_ecam, 2, size_ecam);
1603 
1604     if (vms->highmem_mmio) {
1605         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1606                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1607                                      2, base_pio, 2, size_pio,
1608                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1609                                      2, base_mmio, 2, size_mmio,
1610                                      1, FDT_PCI_RANGE_MMIO_64BIT,
1611                                      2, base_mmio_high,
1612                                      2, base_mmio_high, 2, size_mmio_high);
1613     } else {
1614         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1615                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1616                                      2, base_pio, 2, size_pio,
1617                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1618                                      2, base_mmio, 2, size_mmio);
1619     }
1620 
1621     qemu_fdt_setprop_cell(ms->fdt, nodename, "#interrupt-cells", 1);
1622     create_pcie_irq_map(ms, vms->gic_phandle, irq, nodename);
1623 
1624     if (vms->iommu) {
1625         vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1626 
1627         switch (vms->iommu) {
1628         case VIRT_IOMMU_SMMUV3:
1629             create_smmu(vms, vms->bus);
1630             if (!vms->default_bus_bypass_iommu) {
1631                 qemu_fdt_setprop_cells(ms->fdt, nodename, "iommu-map",
1632                                        0x0, vms->iommu_phandle, 0x0, 0x10000);
1633             }
1634             break;
1635         default:
1636             g_assert_not_reached();
1637         }
1638     }
1639 }
1640 
1641 static void create_cxl_host_reg_region(VirtMachineState *vms)
1642 {
1643     MemoryRegion *sysmem = get_system_memory();
1644     MemoryRegion *mr = &vms->cxl_devices_state.host_mr;
1645 
1646     memory_region_init(mr, OBJECT(vms), "cxl_host_reg",
1647                        vms->memmap[VIRT_CXL_HOST].size);
1648     memory_region_add_subregion(sysmem, vms->memmap[VIRT_CXL_HOST].base, mr);
1649     vms->highmem_cxl = true;
1650 }
1651 
1652 static void create_platform_bus(VirtMachineState *vms)
1653 {
1654     DeviceState *dev;
1655     SysBusDevice *s;
1656     int i;
1657     MemoryRegion *sysmem = get_system_memory();
1658 
1659     dev = qdev_new(TYPE_PLATFORM_BUS_DEVICE);
1660     dev->id = g_strdup(TYPE_PLATFORM_BUS_DEVICE);
1661     qdev_prop_set_uint32(dev, "num_irqs", PLATFORM_BUS_NUM_IRQS);
1662     qdev_prop_set_uint32(dev, "mmio_size", vms->memmap[VIRT_PLATFORM_BUS].size);
1663     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1664     vms->platform_bus_dev = dev;
1665 
1666     s = SYS_BUS_DEVICE(dev);
1667     for (i = 0; i < PLATFORM_BUS_NUM_IRQS; i++) {
1668         int irq = vms->irqmap[VIRT_PLATFORM_BUS] + i;
1669         sysbus_connect_irq(s, i, qdev_get_gpio_in(vms->gic, irq));
1670     }
1671 
1672     memory_region_add_subregion(sysmem,
1673                                 vms->memmap[VIRT_PLATFORM_BUS].base,
1674                                 sysbus_mmio_get_region(s, 0));
1675 }
1676 
1677 static void create_tag_ram(MemoryRegion *tag_sysmem,
1678                            hwaddr base, hwaddr size,
1679                            const char *name)
1680 {
1681     MemoryRegion *tagram = g_new(MemoryRegion, 1);
1682 
1683     memory_region_init_ram(tagram, NULL, name, size / 32, &error_fatal);
1684     memory_region_add_subregion(tag_sysmem, base / 32, tagram);
1685 }
1686 
1687 static void create_secure_ram(VirtMachineState *vms,
1688                               MemoryRegion *secure_sysmem,
1689                               MemoryRegion *secure_tag_sysmem)
1690 {
1691     MemoryRegion *secram = g_new(MemoryRegion, 1);
1692     char *nodename;
1693     hwaddr base = vms->memmap[VIRT_SECURE_MEM].base;
1694     hwaddr size = vms->memmap[VIRT_SECURE_MEM].size;
1695     MachineState *ms = MACHINE(vms);
1696 
1697     memory_region_init_ram(secram, NULL, "virt.secure-ram", size,
1698                            &error_fatal);
1699     memory_region_add_subregion(secure_sysmem, base, secram);
1700 
1701     nodename = g_strdup_printf("/secram@%" PRIx64, base);
1702     qemu_fdt_add_subnode(ms->fdt, nodename);
1703     qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "memory");
1704     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg", 2, base, 2, size);
1705     qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1706     qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1707 
1708     if (secure_tag_sysmem) {
1709         create_tag_ram(secure_tag_sysmem, base, size, "mach-virt.secure-tag");
1710     }
1711 
1712     g_free(nodename);
1713 }
1714 
1715 static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size)
1716 {
1717     const VirtMachineState *board = container_of(binfo, VirtMachineState,
1718                                                  bootinfo);
1719     MachineState *ms = MACHINE(board);
1720 
1721 
1722     *fdt_size = board->fdt_size;
1723     return ms->fdt;
1724 }
1725 
1726 static void virt_build_smbios(VirtMachineState *vms)
1727 {
1728     MachineClass *mc = MACHINE_GET_CLASS(vms);
1729     MachineState *ms = MACHINE(vms);
1730     uint8_t *smbios_tables, *smbios_anchor;
1731     size_t smbios_tables_len, smbios_anchor_len;
1732     struct smbios_phys_mem_area mem_array;
1733     const char *product = "QEMU Virtual Machine";
1734 
1735     if (kvm_enabled()) {
1736         product = "KVM Virtual Machine";
1737     }
1738 
1739     smbios_set_defaults("QEMU", product, mc->name);
1740 
1741     /* build the array of physical mem area from base_memmap */
1742     mem_array.address = vms->memmap[VIRT_MEM].base;
1743     mem_array.length = ms->ram_size;
1744 
1745     smbios_get_tables(ms, SMBIOS_ENTRY_POINT_TYPE_64, &mem_array, 1,
1746                       &smbios_tables, &smbios_tables_len,
1747                       &smbios_anchor, &smbios_anchor_len,
1748                       &error_fatal);
1749 
1750     if (smbios_anchor) {
1751         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables",
1752                         smbios_tables, smbios_tables_len);
1753         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor",
1754                         smbios_anchor, smbios_anchor_len);
1755     }
1756 }
1757 
1758 static
1759 void virt_machine_done(Notifier *notifier, void *data)
1760 {
1761     VirtMachineState *vms = container_of(notifier, VirtMachineState,
1762                                          machine_done);
1763     MachineState *ms = MACHINE(vms);
1764     ARMCPU *cpu = ARM_CPU(first_cpu);
1765     struct arm_boot_info *info = &vms->bootinfo;
1766     AddressSpace *as = arm_boot_address_space(cpu, info);
1767 
1768     cxl_hook_up_pxb_registers(vms->bus, &vms->cxl_devices_state,
1769                               &error_fatal);
1770 
1771     if (vms->cxl_devices_state.is_enabled) {
1772         cxl_fmws_link_targets(&error_fatal);
1773     }
1774     /*
1775      * If the user provided a dtb, we assume the dynamic sysbus nodes
1776      * already are integrated there. This corresponds to a use case where
1777      * the dynamic sysbus nodes are complex and their generation is not yet
1778      * supported. In that case the user can take charge of the guest dt
1779      * while qemu takes charge of the qom stuff.
1780      */
1781     if (info->dtb_filename == NULL) {
1782         platform_bus_add_all_fdt_nodes(ms->fdt, "/intc",
1783                                        vms->memmap[VIRT_PLATFORM_BUS].base,
1784                                        vms->memmap[VIRT_PLATFORM_BUS].size,
1785                                        vms->irqmap[VIRT_PLATFORM_BUS]);
1786     }
1787     if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms, cpu) < 0) {
1788         exit(1);
1789     }
1790 
1791     pci_bus_add_fw_cfg_extra_pci_roots(vms->fw_cfg, vms->bus,
1792                                        &error_abort);
1793 
1794     virt_acpi_setup(vms);
1795     virt_build_smbios(vms);
1796 }
1797 
1798 static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx)
1799 {
1800     uint8_t clustersz;
1801 
1802     /*
1803      * Adjust MPIDR to make TCG consistent (with 64-bit KVM hosts)
1804      * and to improve SGI efficiency.
1805      */
1806     if (vms->gic_version == VIRT_GIC_VERSION_2) {
1807         clustersz = GIC_TARGETLIST_BITS;
1808     } else {
1809         clustersz = GICV3_TARGETLIST_BITS;
1810     }
1811 
1812     return arm_build_mp_affinity(idx, clustersz);
1813 }
1814 
1815 static inline bool *virt_get_high_memmap_enabled(VirtMachineState *vms,
1816                                                  int index)
1817 {
1818     bool *enabled_array[] = {
1819         &vms->highmem_redists,
1820         &vms->highmem_cxl,
1821         &vms->highmem_ecam,
1822         &vms->highmem_mmio,
1823     };
1824 
1825     assert(ARRAY_SIZE(extended_memmap) - VIRT_LOWMEMMAP_LAST ==
1826            ARRAY_SIZE(enabled_array));
1827     assert(index - VIRT_LOWMEMMAP_LAST < ARRAY_SIZE(enabled_array));
1828 
1829     return enabled_array[index - VIRT_LOWMEMMAP_LAST];
1830 }
1831 
1832 static void virt_set_high_memmap(VirtMachineState *vms,
1833                                  hwaddr base, int pa_bits)
1834 {
1835     hwaddr region_base, region_size;
1836     bool *region_enabled, fits;
1837     int i;
1838 
1839     for (i = VIRT_LOWMEMMAP_LAST; i < ARRAY_SIZE(extended_memmap); i++) {
1840         region_enabled = virt_get_high_memmap_enabled(vms, i);
1841         region_base = ROUND_UP(base, extended_memmap[i].size);
1842         region_size = extended_memmap[i].size;
1843 
1844         vms->memmap[i].base = region_base;
1845         vms->memmap[i].size = region_size;
1846 
1847         /*
1848          * Check each device to see if it fits in the PA space,
1849          * moving highest_gpa as we go. For compatibility, move
1850          * highest_gpa for disabled fitting devices as well, if
1851          * the compact layout has been disabled.
1852          *
1853          * For each device that doesn't fit, disable it.
1854          */
1855         fits = (region_base + region_size) <= BIT_ULL(pa_bits);
1856         *region_enabled &= fits;
1857         if (vms->highmem_compact && !*region_enabled) {
1858             continue;
1859         }
1860 
1861         base = region_base + region_size;
1862         if (fits) {
1863             vms->highest_gpa = base - 1;
1864         }
1865     }
1866 }
1867 
1868 static void virt_set_memmap(VirtMachineState *vms, int pa_bits)
1869 {
1870     MachineState *ms = MACHINE(vms);
1871     hwaddr base, device_memory_base, device_memory_size, memtop;
1872     int i;
1873 
1874     vms->memmap = extended_memmap;
1875 
1876     for (i = 0; i < ARRAY_SIZE(base_memmap); i++) {
1877         vms->memmap[i] = base_memmap[i];
1878     }
1879 
1880     if (ms->ram_slots > ACPI_MAX_RAM_SLOTS) {
1881         error_report("unsupported number of memory slots: %"PRIu64,
1882                      ms->ram_slots);
1883         exit(EXIT_FAILURE);
1884     }
1885 
1886     /*
1887      * !highmem is exactly the same as limiting the PA space to 32bit,
1888      * irrespective of the underlying capabilities of the HW.
1889      */
1890     if (!vms->highmem) {
1891         pa_bits = 32;
1892     }
1893 
1894     /*
1895      * We compute the base of the high IO region depending on the
1896      * amount of initial and device memory. The device memory start/size
1897      * is aligned on 1GiB. We never put the high IO region below 256GiB
1898      * so that if maxram_size is < 255GiB we keep the legacy memory map.
1899      * The device region size assumes 1GiB page max alignment per slot.
1900      */
1901     device_memory_base =
1902         ROUND_UP(vms->memmap[VIRT_MEM].base + ms->ram_size, GiB);
1903     device_memory_size = ms->maxram_size - ms->ram_size + ms->ram_slots * GiB;
1904 
1905     /* Base address of the high IO region */
1906     memtop = base = device_memory_base + ROUND_UP(device_memory_size, GiB);
1907     if (memtop > BIT_ULL(pa_bits)) {
1908         error_report("Addressing limited to %d bits, but memory exceeds it by %llu bytes",
1909                      pa_bits, memtop - BIT_ULL(pa_bits));
1910         exit(EXIT_FAILURE);
1911     }
1912     if (base < device_memory_base) {
1913         error_report("maxmem/slots too huge");
1914         exit(EXIT_FAILURE);
1915     }
1916     if (base < vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES) {
1917         base = vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES;
1918     }
1919 
1920     /* We know for sure that at least the memory fits in the PA space */
1921     vms->highest_gpa = memtop - 1;
1922 
1923     virt_set_high_memmap(vms, base, pa_bits);
1924 
1925     if (device_memory_size > 0) {
1926         machine_memory_devices_init(ms, device_memory_base, device_memory_size);
1927     }
1928     vms->highest_gpa = cxl_fmws_set_memmap(ROUND_UP(vms->highest_gpa + 1,
1929                                                     256 * MiB),
1930                                            BIT_ULL(pa_bits)) - 1;
1931 }
1932 
1933 static VirtGICType finalize_gic_version_do(const char *accel_name,
1934                                            VirtGICType gic_version,
1935                                            int gics_supported,
1936                                            unsigned int max_cpus)
1937 {
1938     /* Convert host/max/nosel to GIC version number */
1939     switch (gic_version) {
1940     case VIRT_GIC_VERSION_HOST:
1941         if (!kvm_enabled()) {
1942             error_report("gic-version=host requires KVM");
1943             exit(1);
1944         }
1945 
1946         /* For KVM, gic-version=host means gic-version=max */
1947         return finalize_gic_version_do(accel_name, VIRT_GIC_VERSION_MAX,
1948                                        gics_supported, max_cpus);
1949     case VIRT_GIC_VERSION_MAX:
1950         if (gics_supported & VIRT_GIC_VERSION_4_MASK) {
1951             gic_version = VIRT_GIC_VERSION_4;
1952         } else if (gics_supported & VIRT_GIC_VERSION_3_MASK) {
1953             gic_version = VIRT_GIC_VERSION_3;
1954         } else {
1955             gic_version = VIRT_GIC_VERSION_2;
1956         }
1957         break;
1958     case VIRT_GIC_VERSION_NOSEL:
1959         if ((gics_supported & VIRT_GIC_VERSION_2_MASK) &&
1960             max_cpus <= GIC_NCPU) {
1961             gic_version = VIRT_GIC_VERSION_2;
1962         } else if (gics_supported & VIRT_GIC_VERSION_3_MASK) {
1963             /*
1964              * in case the host does not support v2 emulation or
1965              * the end-user requested more than 8 VCPUs we now default
1966              * to v3. In any case defaulting to v2 would be broken.
1967              */
1968             gic_version = VIRT_GIC_VERSION_3;
1969         } else if (max_cpus > GIC_NCPU) {
1970             error_report("%s only supports GICv2 emulation but more than 8 "
1971                          "vcpus are requested", accel_name);
1972             exit(1);
1973         }
1974         break;
1975     case VIRT_GIC_VERSION_2:
1976     case VIRT_GIC_VERSION_3:
1977     case VIRT_GIC_VERSION_4:
1978         break;
1979     }
1980 
1981     /* Check chosen version is effectively supported */
1982     switch (gic_version) {
1983     case VIRT_GIC_VERSION_2:
1984         if (!(gics_supported & VIRT_GIC_VERSION_2_MASK)) {
1985             error_report("%s does not support GICv2 emulation", accel_name);
1986             exit(1);
1987         }
1988         break;
1989     case VIRT_GIC_VERSION_3:
1990         if (!(gics_supported & VIRT_GIC_VERSION_3_MASK)) {
1991             error_report("%s does not support GICv3 emulation", accel_name);
1992             exit(1);
1993         }
1994         break;
1995     case VIRT_GIC_VERSION_4:
1996         if (!(gics_supported & VIRT_GIC_VERSION_4_MASK)) {
1997             error_report("%s does not support GICv4 emulation, is virtualization=on?",
1998                          accel_name);
1999             exit(1);
2000         }
2001         break;
2002     default:
2003         error_report("logic error in finalize_gic_version");
2004         exit(1);
2005         break;
2006     }
2007 
2008     return gic_version;
2009 }
2010 
2011 /*
2012  * finalize_gic_version - Determines the final gic_version
2013  * according to the gic-version property
2014  *
2015  * Default GIC type is v2
2016  */
2017 static void finalize_gic_version(VirtMachineState *vms)
2018 {
2019     const char *accel_name = current_accel_name();
2020     unsigned int max_cpus = MACHINE(vms)->smp.max_cpus;
2021     int gics_supported = 0;
2022 
2023     /* Determine which GIC versions the current environment supports */
2024     if (kvm_enabled() && kvm_irqchip_in_kernel()) {
2025         int probe_bitmap = kvm_arm_vgic_probe();
2026 
2027         if (!probe_bitmap) {
2028             error_report("Unable to determine GIC version supported by host");
2029             exit(1);
2030         }
2031 
2032         if (probe_bitmap & KVM_ARM_VGIC_V2) {
2033             gics_supported |= VIRT_GIC_VERSION_2_MASK;
2034         }
2035         if (probe_bitmap & KVM_ARM_VGIC_V3) {
2036             gics_supported |= VIRT_GIC_VERSION_3_MASK;
2037         }
2038     } else if (kvm_enabled() && !kvm_irqchip_in_kernel()) {
2039         /* KVM w/o kernel irqchip can only deal with GICv2 */
2040         gics_supported |= VIRT_GIC_VERSION_2_MASK;
2041         accel_name = "KVM with kernel-irqchip=off";
2042     } else if (tcg_enabled() || hvf_enabled() || qtest_enabled())  {
2043         gics_supported |= VIRT_GIC_VERSION_2_MASK;
2044         if (module_object_class_by_name("arm-gicv3")) {
2045             gics_supported |= VIRT_GIC_VERSION_3_MASK;
2046             if (vms->virt) {
2047                 /* GICv4 only makes sense if CPU has EL2 */
2048                 gics_supported |= VIRT_GIC_VERSION_4_MASK;
2049             }
2050         }
2051     } else {
2052         error_report("Unsupported accelerator, can not determine GIC support");
2053         exit(1);
2054     }
2055 
2056     /*
2057      * Then convert helpers like host/max to concrete GIC versions and ensure
2058      * the desired version is supported
2059      */
2060     vms->gic_version = finalize_gic_version_do(accel_name, vms->gic_version,
2061                                                gics_supported, max_cpus);
2062 }
2063 
2064 /*
2065  * virt_post_cpus_gic_realized() must be called after the CPUs and
2066  * the GIC have both been realized.
2067  */
2068 static void virt_post_cpus_gic_realized(VirtMachineState *vms,
2069                                         MemoryRegion *sysmem)
2070 {
2071     int max_cpus = MACHINE(vms)->smp.max_cpus;
2072     bool aarch64, pmu, steal_time;
2073     CPUState *cpu;
2074 
2075     aarch64 = object_property_get_bool(OBJECT(first_cpu), "aarch64", NULL);
2076     pmu = object_property_get_bool(OBJECT(first_cpu), "pmu", NULL);
2077     steal_time = object_property_get_bool(OBJECT(first_cpu),
2078                                           "kvm-steal-time", NULL);
2079 
2080     if (kvm_enabled()) {
2081         hwaddr pvtime_reg_base = vms->memmap[VIRT_PVTIME].base;
2082         hwaddr pvtime_reg_size = vms->memmap[VIRT_PVTIME].size;
2083 
2084         if (steal_time) {
2085             MemoryRegion *pvtime = g_new(MemoryRegion, 1);
2086             hwaddr pvtime_size = max_cpus * PVTIME_SIZE_PER_CPU;
2087 
2088             /* The memory region size must be a multiple of host page size. */
2089             pvtime_size = REAL_HOST_PAGE_ALIGN(pvtime_size);
2090 
2091             if (pvtime_size > pvtime_reg_size) {
2092                 error_report("pvtime requires a %" HWADDR_PRId
2093                              " byte memory region for %d CPUs,"
2094                              " but only %" HWADDR_PRId " has been reserved",
2095                              pvtime_size, max_cpus, pvtime_reg_size);
2096                 exit(1);
2097             }
2098 
2099             memory_region_init_ram(pvtime, NULL, "pvtime", pvtime_size, NULL);
2100             memory_region_add_subregion(sysmem, pvtime_reg_base, pvtime);
2101         }
2102         if (!aarch64 && vms->virt) {
2103             error_report("KVM does not support EL2 on an AArch32 vCPU");
2104             exit(1);
2105         }
2106 
2107         CPU_FOREACH(cpu) {
2108             if (pmu) {
2109                 assert(arm_feature(&ARM_CPU(cpu)->env, ARM_FEATURE_PMU));
2110                 if (kvm_irqchip_in_kernel()) {
2111                     kvm_arm_pmu_set_irq(ARM_CPU(cpu), VIRTUAL_PMU_IRQ);
2112                 }
2113                 kvm_arm_pmu_init(ARM_CPU(cpu));
2114             }
2115             if (steal_time) {
2116                 kvm_arm_pvtime_init(ARM_CPU(cpu), pvtime_reg_base
2117                                                   + cpu->cpu_index
2118                                                     * PVTIME_SIZE_PER_CPU);
2119             }
2120         }
2121     } else {
2122         if (aarch64 && vms->highmem) {
2123             int requested_pa_size = 64 - clz64(vms->highest_gpa);
2124             int pamax = arm_pamax(ARM_CPU(first_cpu));
2125 
2126             if (pamax < requested_pa_size) {
2127                 error_report("VCPU supports less PA bits (%d) than "
2128                              "requested by the memory map (%d)",
2129                              pamax, requested_pa_size);
2130                 exit(1);
2131             }
2132         }
2133     }
2134 }
2135 
2136 static void machvirt_init(MachineState *machine)
2137 {
2138     VirtMachineState *vms = VIRT_MACHINE(machine);
2139     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine);
2140     MachineClass *mc = MACHINE_GET_CLASS(machine);
2141     const CPUArchIdList *possible_cpus;
2142     MemoryRegion *sysmem = get_system_memory();
2143     MemoryRegion *secure_sysmem = NULL;
2144     MemoryRegion *tag_sysmem = NULL;
2145     MemoryRegion *secure_tag_sysmem = NULL;
2146     int n, virt_max_cpus;
2147     bool firmware_loaded;
2148     bool aarch64 = true;
2149     bool has_ged = !vmc->no_ged;
2150     unsigned int smp_cpus = machine->smp.cpus;
2151     unsigned int max_cpus = machine->smp.max_cpus;
2152 
2153     possible_cpus = mc->possible_cpu_arch_ids(machine);
2154 
2155     /*
2156      * In accelerated mode, the memory map is computed earlier in kvm_type()
2157      * for Linux, or hvf_get_physical_address_range() for macOS to create a
2158      * VM with the right number of IPA bits.
2159      */
2160     if (!vms->memmap) {
2161         Object *cpuobj;
2162         ARMCPU *armcpu;
2163         int pa_bits;
2164 
2165         /*
2166          * Instantiate a temporary CPU object to find out about what
2167          * we are about to deal with. Once this is done, get rid of
2168          * the object.
2169          */
2170         cpuobj = object_new(possible_cpus->cpus[0].type);
2171         armcpu = ARM_CPU(cpuobj);
2172 
2173         pa_bits = arm_pamax(armcpu);
2174 
2175         object_unref(cpuobj);
2176 
2177         virt_set_memmap(vms, pa_bits);
2178     }
2179 
2180     /* We can probe only here because during property set
2181      * KVM is not available yet
2182      */
2183     finalize_gic_version(vms);
2184 
2185     if (vms->secure) {
2186         /*
2187          * The Secure view of the world is the same as the NonSecure,
2188          * but with a few extra devices. Create it as a container region
2189          * containing the system memory at low priority; any secure-only
2190          * devices go in at higher priority and take precedence.
2191          */
2192         secure_sysmem = g_new(MemoryRegion, 1);
2193         memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory",
2194                            UINT64_MAX);
2195         memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1);
2196     }
2197 
2198     firmware_loaded = virt_firmware_init(vms, sysmem,
2199                                          secure_sysmem ?: sysmem);
2200 
2201     /* If we have an EL3 boot ROM then the assumption is that it will
2202      * implement PSCI itself, so disable QEMU's internal implementation
2203      * so it doesn't get in the way. Instead of starting secondary
2204      * CPUs in PSCI powerdown state we will start them all running and
2205      * let the boot ROM sort them out.
2206      * The usual case is that we do use QEMU's PSCI implementation;
2207      * if the guest has EL2 then we will use SMC as the conduit,
2208      * and otherwise we will use HVC (for backwards compatibility and
2209      * because if we're using KVM then we must use HVC).
2210      */
2211     if (vms->secure && firmware_loaded) {
2212         vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
2213     } else if (vms->virt) {
2214         vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC;
2215     } else {
2216         vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC;
2217     }
2218 
2219     /*
2220      * The maximum number of CPUs depends on the GIC version, or on how
2221      * many redistributors we can fit into the memory map (which in turn
2222      * depends on whether this is a GICv3 or v4).
2223      */
2224     if (vms->gic_version == VIRT_GIC_VERSION_2) {
2225         virt_max_cpus = GIC_NCPU;
2226     } else {
2227         virt_max_cpus = virt_redist_capacity(vms, VIRT_GIC_REDIST);
2228         if (vms->highmem_redists) {
2229             virt_max_cpus += virt_redist_capacity(vms, VIRT_HIGH_GIC_REDIST2);
2230         }
2231     }
2232 
2233     if (max_cpus > virt_max_cpus) {
2234         error_report("Number of SMP CPUs requested (%d) exceeds max CPUs "
2235                      "supported by machine 'mach-virt' (%d)",
2236                      max_cpus, virt_max_cpus);
2237         if (vms->gic_version != VIRT_GIC_VERSION_2 && !vms->highmem_redists) {
2238             error_printf("Try 'highmem-redists=on' for more CPUs\n");
2239         }
2240 
2241         exit(1);
2242     }
2243 
2244     if (vms->secure && !tcg_enabled() && !qtest_enabled()) {
2245         error_report("mach-virt: %s does not support providing "
2246                      "Security extensions (TrustZone) to the guest CPU",
2247                      current_accel_name());
2248         exit(1);
2249     }
2250 
2251     if (vms->virt && kvm_enabled() && !kvm_arm_el2_supported()) {
2252         error_report("mach-virt: host kernel KVM does not support providing "
2253                      "Virtualization extensions to the guest CPU");
2254         exit(1);
2255     }
2256 
2257     if (vms->virt && !kvm_enabled() && !tcg_enabled() && !qtest_enabled()) {
2258         error_report("mach-virt: %s does not support providing "
2259                      "Virtualization extensions to the guest CPU",
2260                      current_accel_name());
2261         exit(1);
2262     }
2263 
2264     if (vms->mte && hvf_enabled()) {
2265         error_report("mach-virt: %s does not support providing "
2266                      "MTE to the guest CPU",
2267                      current_accel_name());
2268         exit(1);
2269     }
2270 
2271     create_fdt(vms);
2272 
2273     assert(possible_cpus->len == max_cpus);
2274     for (n = 0; n < possible_cpus->len; n++) {
2275         Object *cpuobj;
2276         CPUState *cs;
2277 
2278         if (n >= smp_cpus) {
2279             break;
2280         }
2281 
2282         cpuobj = object_new(possible_cpus->cpus[n].type);
2283         object_property_set_int(cpuobj, "mp-affinity",
2284                                 possible_cpus->cpus[n].arch_id, NULL);
2285 
2286         cs = CPU(cpuobj);
2287         cs->cpu_index = n;
2288 
2289         numa_cpu_pre_plug(&possible_cpus->cpus[cs->cpu_index], DEVICE(cpuobj),
2290                           &error_fatal);
2291 
2292         aarch64 &= object_property_get_bool(cpuobj, "aarch64", NULL);
2293 
2294         if (!vms->secure) {
2295             object_property_set_bool(cpuobj, "has_el3", false, NULL);
2296         }
2297 
2298         if (!vms->virt && object_property_find(cpuobj, "has_el2")) {
2299             object_property_set_bool(cpuobj, "has_el2", false, NULL);
2300         }
2301 
2302         if (vmc->kvm_no_adjvtime &&
2303             object_property_find(cpuobj, "kvm-no-adjvtime")) {
2304             object_property_set_bool(cpuobj, "kvm-no-adjvtime", true, NULL);
2305         }
2306 
2307         if (vmc->no_kvm_steal_time &&
2308             object_property_find(cpuobj, "kvm-steal-time")) {
2309             object_property_set_bool(cpuobj, "kvm-steal-time", false, NULL);
2310         }
2311 
2312         if (vmc->no_tcg_lpa2 && object_property_find(cpuobj, "lpa2")) {
2313             object_property_set_bool(cpuobj, "lpa2", false, NULL);
2314         }
2315 
2316         if (object_property_find(cpuobj, "reset-cbar")) {
2317             object_property_set_int(cpuobj, "reset-cbar",
2318                                     vms->memmap[VIRT_CPUPERIPHS].base,
2319                                     &error_abort);
2320         }
2321 
2322         object_property_set_link(cpuobj, "memory", OBJECT(sysmem),
2323                                  &error_abort);
2324         if (vms->secure) {
2325             object_property_set_link(cpuobj, "secure-memory",
2326                                      OBJECT(secure_sysmem), &error_abort);
2327         }
2328 
2329         if (vms->mte) {
2330             if (tcg_enabled()) {
2331                 /* Create the memory region only once, but link to all cpus. */
2332                 if (!tag_sysmem) {
2333                     /*
2334                      * The property exists only if MemTag is supported.
2335                      * If it is, we must allocate the ram to back that up.
2336                      */
2337                     if (!object_property_find(cpuobj, "tag-memory")) {
2338                         error_report("MTE requested, but not supported "
2339                                      "by the guest CPU");
2340                         exit(1);
2341                     }
2342 
2343                     tag_sysmem = g_new(MemoryRegion, 1);
2344                     memory_region_init(tag_sysmem, OBJECT(machine),
2345                                        "tag-memory", UINT64_MAX / 32);
2346 
2347                     if (vms->secure) {
2348                         secure_tag_sysmem = g_new(MemoryRegion, 1);
2349                         memory_region_init(secure_tag_sysmem, OBJECT(machine),
2350                                            "secure-tag-memory",
2351                                            UINT64_MAX / 32);
2352 
2353                         /* As with ram, secure-tag takes precedence over tag. */
2354                         memory_region_add_subregion_overlap(secure_tag_sysmem,
2355                                                             0, tag_sysmem, -1);
2356                     }
2357                 }
2358 
2359                 object_property_set_link(cpuobj, "tag-memory",
2360                                          OBJECT(tag_sysmem), &error_abort);
2361                 if (vms->secure) {
2362                     object_property_set_link(cpuobj, "secure-tag-memory",
2363                                              OBJECT(secure_tag_sysmem),
2364                                              &error_abort);
2365                 }
2366             } else if (kvm_enabled()) {
2367                 if (!kvm_arm_mte_supported()) {
2368                     error_report("MTE requested, but not supported by KVM");
2369                     exit(1);
2370                 }
2371                 kvm_arm_enable_mte(cpuobj, &error_abort);
2372             } else {
2373                     error_report("MTE requested, but not supported ");
2374                     exit(1);
2375             }
2376         }
2377 
2378         qdev_realize(DEVICE(cpuobj), NULL, &error_fatal);
2379         object_unref(cpuobj);
2380     }
2381 
2382     /* Now we've created the CPUs we can see if they have the hypvirt timer */
2383     vms->ns_el2_virt_timer_irq = ns_el2_virt_timer_present() &&
2384         !vmc->no_ns_el2_virt_timer_irq;
2385 
2386     fdt_add_timer_nodes(vms);
2387     fdt_add_cpu_nodes(vms);
2388 
2389     memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base,
2390                                 machine->ram);
2391 
2392     cxl_fmws_update_mmio();
2393 
2394     virt_flash_fdt(vms, sysmem, secure_sysmem ?: sysmem);
2395 
2396     create_gic(vms, sysmem);
2397 
2398     virt_post_cpus_gic_realized(vms, sysmem);
2399 
2400     fdt_add_pmu_nodes(vms);
2401 
2402     /*
2403      * The first UART always exists. If the security extensions are
2404      * enabled, the second UART also always exists. Otherwise, it only exists
2405      * if a backend is configured explicitly via '-serial <backend>'.
2406      * This avoids potentially breaking existing user setups that expect
2407      * only one NonSecure UART to be present (for instance, older EDK2
2408      * binaries).
2409      *
2410      * The nodes end up in the DTB in reverse order of creation, so we must
2411      * create UART0 last to ensure it appears as the first node in the DTB,
2412      * for compatibility with guest software that just iterates through the
2413      * DTB to find the first UART, as older versions of EDK2 do.
2414      * DTB readers that follow the spec, as Linux does, should honour the
2415      * aliases node information and /chosen/stdout-path regardless of
2416      * the order that nodes appear in the DTB.
2417      *
2418      * For similar back-compatibility reasons, if UART1 is the secure UART
2419      * we create it second (and so it appears first in the DTB), because
2420      * that's what QEMU has always done.
2421      */
2422     if (!vms->secure) {
2423         Chardev *serial1 = serial_hd(1);
2424 
2425         if (serial1) {
2426             vms->second_ns_uart_present = true;
2427             create_uart(vms, VIRT_UART1, sysmem, serial1, false);
2428         }
2429     }
2430     create_uart(vms, VIRT_UART0, sysmem, serial_hd(0), false);
2431     if (vms->secure) {
2432         create_uart(vms, VIRT_UART1, secure_sysmem, serial_hd(1), true);
2433     }
2434 
2435     if (vms->secure) {
2436         create_secure_ram(vms, secure_sysmem, secure_tag_sysmem);
2437     }
2438 
2439     if (tag_sysmem) {
2440         create_tag_ram(tag_sysmem, vms->memmap[VIRT_MEM].base,
2441                        machine->ram_size, "mach-virt.tag");
2442     }
2443 
2444     vms->highmem_ecam &= (!firmware_loaded || aarch64);
2445 
2446     create_rtc(vms);
2447 
2448     create_pcie(vms);
2449     create_cxl_host_reg_region(vms);
2450 
2451     if (has_ged && aarch64 && firmware_loaded && virt_is_acpi_enabled(vms)) {
2452         vms->acpi_dev = create_acpi_ged(vms);
2453     } else {
2454         create_gpio_devices(vms, VIRT_GPIO, sysmem);
2455     }
2456 
2457     if (vms->secure && !vmc->no_secure_gpio) {
2458         create_gpio_devices(vms, VIRT_SECURE_GPIO, secure_sysmem);
2459     }
2460 
2461      /* connect powerdown request */
2462      vms->powerdown_notifier.notify = virt_powerdown_req;
2463      qemu_register_powerdown_notifier(&vms->powerdown_notifier);
2464 
2465     /* Create mmio transports, so the user can create virtio backends
2466      * (which will be automatically plugged in to the transports). If
2467      * no backend is created the transport will just sit harmlessly idle.
2468      */
2469     create_virtio_devices(vms);
2470 
2471     vms->fw_cfg = create_fw_cfg(vms, &address_space_memory);
2472     rom_set_fw(vms->fw_cfg);
2473 
2474     create_platform_bus(vms);
2475 
2476     if (machine->nvdimms_state->is_enabled) {
2477         const struct AcpiGenericAddress arm_virt_nvdimm_acpi_dsmio = {
2478             .space_id = AML_AS_SYSTEM_MEMORY,
2479             .address = vms->memmap[VIRT_NVDIMM_ACPI].base,
2480             .bit_width = NVDIMM_ACPI_IO_LEN << 3
2481         };
2482 
2483         nvdimm_init_acpi_state(machine->nvdimms_state, sysmem,
2484                                arm_virt_nvdimm_acpi_dsmio,
2485                                vms->fw_cfg, OBJECT(vms));
2486     }
2487 
2488     vms->bootinfo.ram_size = machine->ram_size;
2489     vms->bootinfo.board_id = -1;
2490     vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base;
2491     vms->bootinfo.get_dtb = machvirt_dtb;
2492     vms->bootinfo.skip_dtb_autoload = true;
2493     vms->bootinfo.firmware_loaded = firmware_loaded;
2494     vms->bootinfo.psci_conduit = vms->psci_conduit;
2495     arm_load_kernel(ARM_CPU(first_cpu), machine, &vms->bootinfo);
2496 
2497     vms->machine_done.notify = virt_machine_done;
2498     qemu_add_machine_init_done_notifier(&vms->machine_done);
2499 }
2500 
2501 static bool virt_get_secure(Object *obj, Error **errp)
2502 {
2503     VirtMachineState *vms = VIRT_MACHINE(obj);
2504 
2505     return vms->secure;
2506 }
2507 
2508 static void virt_set_secure(Object *obj, bool value, Error **errp)
2509 {
2510     VirtMachineState *vms = VIRT_MACHINE(obj);
2511 
2512     vms->secure = value;
2513 }
2514 
2515 static bool virt_get_virt(Object *obj, Error **errp)
2516 {
2517     VirtMachineState *vms = VIRT_MACHINE(obj);
2518 
2519     return vms->virt;
2520 }
2521 
2522 static void virt_set_virt(Object *obj, bool value, Error **errp)
2523 {
2524     VirtMachineState *vms = VIRT_MACHINE(obj);
2525 
2526     vms->virt = value;
2527 }
2528 
2529 static bool virt_get_highmem(Object *obj, Error **errp)
2530 {
2531     VirtMachineState *vms = VIRT_MACHINE(obj);
2532 
2533     return vms->highmem;
2534 }
2535 
2536 static void virt_set_highmem(Object *obj, bool value, Error **errp)
2537 {
2538     VirtMachineState *vms = VIRT_MACHINE(obj);
2539 
2540     vms->highmem = value;
2541 }
2542 
2543 static bool virt_get_compact_highmem(Object *obj, Error **errp)
2544 {
2545     VirtMachineState *vms = VIRT_MACHINE(obj);
2546 
2547     return vms->highmem_compact;
2548 }
2549 
2550 static void virt_set_compact_highmem(Object *obj, bool value, Error **errp)
2551 {
2552     VirtMachineState *vms = VIRT_MACHINE(obj);
2553 
2554     vms->highmem_compact = value;
2555 }
2556 
2557 static bool virt_get_highmem_redists(Object *obj, Error **errp)
2558 {
2559     VirtMachineState *vms = VIRT_MACHINE(obj);
2560 
2561     return vms->highmem_redists;
2562 }
2563 
2564 static void virt_set_highmem_redists(Object *obj, bool value, Error **errp)
2565 {
2566     VirtMachineState *vms = VIRT_MACHINE(obj);
2567 
2568     vms->highmem_redists = value;
2569 }
2570 
2571 static bool virt_get_highmem_ecam(Object *obj, Error **errp)
2572 {
2573     VirtMachineState *vms = VIRT_MACHINE(obj);
2574 
2575     return vms->highmem_ecam;
2576 }
2577 
2578 static void virt_set_highmem_ecam(Object *obj, bool value, Error **errp)
2579 {
2580     VirtMachineState *vms = VIRT_MACHINE(obj);
2581 
2582     vms->highmem_ecam = value;
2583 }
2584 
2585 static bool virt_get_highmem_mmio(Object *obj, Error **errp)
2586 {
2587     VirtMachineState *vms = VIRT_MACHINE(obj);
2588 
2589     return vms->highmem_mmio;
2590 }
2591 
2592 static void virt_set_highmem_mmio(Object *obj, bool value, Error **errp)
2593 {
2594     VirtMachineState *vms = VIRT_MACHINE(obj);
2595 
2596     vms->highmem_mmio = value;
2597 }
2598 
2599 static void virt_get_highmem_mmio_size(Object *obj, Visitor *v,
2600                                        const char *name, void *opaque,
2601                                        Error **errp)
2602 {
2603     uint64_t size = extended_memmap[VIRT_HIGH_PCIE_MMIO].size;
2604 
2605     visit_type_size(v, name, &size, errp);
2606 }
2607 
2608 static void virt_set_highmem_mmio_size(Object *obj, Visitor *v,
2609                                        const char *name, void *opaque,
2610                                        Error **errp)
2611 {
2612     uint64_t size;
2613 
2614     if (!visit_type_size(v, name, &size, errp)) {
2615         return;
2616     }
2617 
2618     if (!is_power_of_2(size)) {
2619         error_setg(errp, "highmem-mmio-size is not a power of 2");
2620         return;
2621     }
2622 
2623     if (size < DEFAULT_HIGH_PCIE_MMIO_SIZE) {
2624         char *sz = size_to_str(DEFAULT_HIGH_PCIE_MMIO_SIZE);
2625         error_setg(errp, "highmem-mmio-size cannot be set to a lower value "
2626                          "than the default (%s)", sz);
2627         g_free(sz);
2628         return;
2629     }
2630 
2631     extended_memmap[VIRT_HIGH_PCIE_MMIO].size = size;
2632 }
2633 
2634 static bool virt_get_its(Object *obj, Error **errp)
2635 {
2636     VirtMachineState *vms = VIRT_MACHINE(obj);
2637 
2638     return vms->its;
2639 }
2640 
2641 static void virt_set_its(Object *obj, bool value, Error **errp)
2642 {
2643     VirtMachineState *vms = VIRT_MACHINE(obj);
2644 
2645     vms->its = value;
2646 }
2647 
2648 static bool virt_get_dtb_randomness(Object *obj, Error **errp)
2649 {
2650     VirtMachineState *vms = VIRT_MACHINE(obj);
2651 
2652     return vms->dtb_randomness;
2653 }
2654 
2655 static void virt_set_dtb_randomness(Object *obj, bool value, Error **errp)
2656 {
2657     VirtMachineState *vms = VIRT_MACHINE(obj);
2658 
2659     vms->dtb_randomness = value;
2660 }
2661 
2662 static char *virt_get_oem_id(Object *obj, Error **errp)
2663 {
2664     VirtMachineState *vms = VIRT_MACHINE(obj);
2665 
2666     return g_strdup(vms->oem_id);
2667 }
2668 
2669 static void virt_set_oem_id(Object *obj, const char *value, Error **errp)
2670 {
2671     VirtMachineState *vms = VIRT_MACHINE(obj);
2672     size_t len = strlen(value);
2673 
2674     if (len > 6) {
2675         error_setg(errp,
2676                    "User specified oem-id value is bigger than 6 bytes in size");
2677         return;
2678     }
2679 
2680     strncpy(vms->oem_id, value, 6);
2681 }
2682 
2683 static char *virt_get_oem_table_id(Object *obj, Error **errp)
2684 {
2685     VirtMachineState *vms = VIRT_MACHINE(obj);
2686 
2687     return g_strdup(vms->oem_table_id);
2688 }
2689 
2690 static void virt_set_oem_table_id(Object *obj, const char *value,
2691                                   Error **errp)
2692 {
2693     VirtMachineState *vms = VIRT_MACHINE(obj);
2694     size_t len = strlen(value);
2695 
2696     if (len > 8) {
2697         error_setg(errp,
2698                    "User specified oem-table-id value is bigger than 8 bytes in size");
2699         return;
2700     }
2701     strncpy(vms->oem_table_id, value, 8);
2702 }
2703 
2704 
2705 bool virt_is_acpi_enabled(VirtMachineState *vms)
2706 {
2707     if (vms->acpi == ON_OFF_AUTO_OFF) {
2708         return false;
2709     }
2710     return true;
2711 }
2712 
2713 static void virt_get_acpi(Object *obj, Visitor *v, const char *name,
2714                           void *opaque, Error **errp)
2715 {
2716     VirtMachineState *vms = VIRT_MACHINE(obj);
2717     OnOffAuto acpi = vms->acpi;
2718 
2719     visit_type_OnOffAuto(v, name, &acpi, errp);
2720 }
2721 
2722 static void virt_set_acpi(Object *obj, Visitor *v, const char *name,
2723                           void *opaque, Error **errp)
2724 {
2725     VirtMachineState *vms = VIRT_MACHINE(obj);
2726 
2727     visit_type_OnOffAuto(v, name, &vms->acpi, errp);
2728 }
2729 
2730 static bool virt_get_ras(Object *obj, Error **errp)
2731 {
2732     VirtMachineState *vms = VIRT_MACHINE(obj);
2733 
2734     return vms->ras;
2735 }
2736 
2737 static void virt_set_ras(Object *obj, bool value, Error **errp)
2738 {
2739     VirtMachineState *vms = VIRT_MACHINE(obj);
2740 
2741     vms->ras = value;
2742 }
2743 
2744 static bool virt_get_mte(Object *obj, Error **errp)
2745 {
2746     VirtMachineState *vms = VIRT_MACHINE(obj);
2747 
2748     return vms->mte;
2749 }
2750 
2751 static void virt_set_mte(Object *obj, bool value, Error **errp)
2752 {
2753     VirtMachineState *vms = VIRT_MACHINE(obj);
2754 
2755     vms->mte = value;
2756 }
2757 
2758 static char *virt_get_gic_version(Object *obj, Error **errp)
2759 {
2760     VirtMachineState *vms = VIRT_MACHINE(obj);
2761     const char *val;
2762 
2763     switch (vms->gic_version) {
2764     case VIRT_GIC_VERSION_4:
2765         val = "4";
2766         break;
2767     case VIRT_GIC_VERSION_3:
2768         val = "3";
2769         break;
2770     default:
2771         val = "2";
2772         break;
2773     }
2774     return g_strdup(val);
2775 }
2776 
2777 static void virt_set_gic_version(Object *obj, const char *value, Error **errp)
2778 {
2779     VirtMachineState *vms = VIRT_MACHINE(obj);
2780 
2781     if (!strcmp(value, "4")) {
2782         vms->gic_version = VIRT_GIC_VERSION_4;
2783     } else if (!strcmp(value, "3")) {
2784         vms->gic_version = VIRT_GIC_VERSION_3;
2785     } else if (!strcmp(value, "2")) {
2786         vms->gic_version = VIRT_GIC_VERSION_2;
2787     } else if (!strcmp(value, "host")) {
2788         vms->gic_version = VIRT_GIC_VERSION_HOST; /* Will probe later */
2789     } else if (!strcmp(value, "max")) {
2790         vms->gic_version = VIRT_GIC_VERSION_MAX; /* Will probe later */
2791     } else {
2792         error_setg(errp, "Invalid gic-version value");
2793         error_append_hint(errp, "Valid values are 3, 2, host, max.\n");
2794     }
2795 }
2796 
2797 static char *virt_get_iommu(Object *obj, Error **errp)
2798 {
2799     VirtMachineState *vms = VIRT_MACHINE(obj);
2800 
2801     switch (vms->iommu) {
2802     case VIRT_IOMMU_NONE:
2803         return g_strdup("none");
2804     case VIRT_IOMMU_SMMUV3:
2805         return g_strdup("smmuv3");
2806     default:
2807         g_assert_not_reached();
2808     }
2809 }
2810 
2811 static void virt_set_iommu(Object *obj, const char *value, Error **errp)
2812 {
2813     VirtMachineState *vms = VIRT_MACHINE(obj);
2814 
2815     if (!strcmp(value, "smmuv3")) {
2816         vms->iommu = VIRT_IOMMU_SMMUV3;
2817     } else if (!strcmp(value, "none")) {
2818         vms->iommu = VIRT_IOMMU_NONE;
2819     } else {
2820         error_setg(errp, "Invalid iommu value");
2821         error_append_hint(errp, "Valid values are none, smmuv3.\n");
2822     }
2823 }
2824 
2825 static bool virt_get_default_bus_bypass_iommu(Object *obj, Error **errp)
2826 {
2827     VirtMachineState *vms = VIRT_MACHINE(obj);
2828 
2829     return vms->default_bus_bypass_iommu;
2830 }
2831 
2832 static void virt_set_default_bus_bypass_iommu(Object *obj, bool value,
2833                                               Error **errp)
2834 {
2835     VirtMachineState *vms = VIRT_MACHINE(obj);
2836 
2837     vms->default_bus_bypass_iommu = value;
2838 }
2839 
2840 static CpuInstanceProperties
2841 virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
2842 {
2843     MachineClass *mc = MACHINE_GET_CLASS(ms);
2844     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
2845 
2846     assert(cpu_index < possible_cpus->len);
2847     return possible_cpus->cpus[cpu_index].props;
2848 }
2849 
2850 static int64_t virt_get_default_cpu_node_id(const MachineState *ms, int idx)
2851 {
2852     int64_t socket_id = ms->possible_cpus->cpus[idx].props.socket_id;
2853 
2854     return socket_id % ms->numa_state->num_nodes;
2855 }
2856 
2857 static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms)
2858 {
2859     int n;
2860     unsigned int max_cpus = ms->smp.max_cpus;
2861     VirtMachineState *vms = VIRT_MACHINE(ms);
2862     MachineClass *mc = MACHINE_GET_CLASS(vms);
2863 
2864     if (ms->possible_cpus) {
2865         assert(ms->possible_cpus->len == max_cpus);
2866         return ms->possible_cpus;
2867     }
2868 
2869     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2870                                   sizeof(CPUArchId) * max_cpus);
2871     ms->possible_cpus->len = max_cpus;
2872     for (n = 0; n < ms->possible_cpus->len; n++) {
2873         ms->possible_cpus->cpus[n].type = ms->cpu_type;
2874         ms->possible_cpus->cpus[n].arch_id =
2875             virt_cpu_mp_affinity(vms, n);
2876 
2877         assert(!mc->smp_props.dies_supported);
2878         ms->possible_cpus->cpus[n].props.has_socket_id = true;
2879         ms->possible_cpus->cpus[n].props.socket_id =
2880             n / (ms->smp.clusters * ms->smp.cores * ms->smp.threads);
2881         ms->possible_cpus->cpus[n].props.has_cluster_id = true;
2882         ms->possible_cpus->cpus[n].props.cluster_id =
2883             (n / (ms->smp.cores * ms->smp.threads)) % ms->smp.clusters;
2884         ms->possible_cpus->cpus[n].props.has_core_id = true;
2885         ms->possible_cpus->cpus[n].props.core_id =
2886             (n / ms->smp.threads) % ms->smp.cores;
2887         ms->possible_cpus->cpus[n].props.has_thread_id = true;
2888         ms->possible_cpus->cpus[n].props.thread_id =
2889             n % ms->smp.threads;
2890     }
2891     return ms->possible_cpus;
2892 }
2893 
2894 static void virt_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2895                                  Error **errp)
2896 {
2897     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2898     const MachineState *ms = MACHINE(hotplug_dev);
2899     const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2900 
2901     if (!vms->acpi_dev) {
2902         error_setg(errp,
2903                    "memory hotplug is not enabled: missing acpi-ged device");
2904         return;
2905     }
2906 
2907     if (vms->mte) {
2908         error_setg(errp, "memory hotplug is not enabled: MTE is enabled");
2909         return;
2910     }
2911 
2912     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
2913         error_setg(errp, "nvdimm is not enabled: add 'nvdimm=on' to '-M'");
2914         return;
2915     }
2916 
2917     pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev), errp);
2918 }
2919 
2920 static void virt_memory_plug(HotplugHandler *hotplug_dev,
2921                              DeviceState *dev, Error **errp)
2922 {
2923     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2924     MachineState *ms = MACHINE(hotplug_dev);
2925     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2926 
2927     pc_dimm_plug(PC_DIMM(dev), MACHINE(vms));
2928 
2929     if (is_nvdimm) {
2930         nvdimm_plug(ms->nvdimms_state);
2931     }
2932 
2933     hotplug_handler_plug(HOTPLUG_HANDLER(vms->acpi_dev),
2934                          dev, &error_abort);
2935 }
2936 
2937 static void virt_machine_device_pre_plug_cb(HotplugHandler *hotplug_dev,
2938                                             DeviceState *dev, Error **errp)
2939 {
2940     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2941 
2942     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2943         virt_memory_pre_plug(hotplug_dev, dev, errp);
2944     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
2945         virtio_md_pci_pre_plug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
2946     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
2947         hwaddr db_start = 0, db_end = 0;
2948         QList *reserved_regions;
2949         char *resv_prop_str;
2950 
2951         if (vms->iommu != VIRT_IOMMU_NONE) {
2952             error_setg(errp, "virt machine does not support multiple IOMMUs");
2953             return;
2954         }
2955 
2956         switch (vms->msi_controller) {
2957         case VIRT_MSI_CTRL_NONE:
2958             return;
2959         case VIRT_MSI_CTRL_ITS:
2960             /* GITS_TRANSLATER page */
2961             db_start = base_memmap[VIRT_GIC_ITS].base + 0x10000;
2962             db_end = base_memmap[VIRT_GIC_ITS].base +
2963                      base_memmap[VIRT_GIC_ITS].size - 1;
2964             break;
2965         case VIRT_MSI_CTRL_GICV2M:
2966             /* MSI_SETSPI_NS page */
2967             db_start = base_memmap[VIRT_GIC_V2M].base;
2968             db_end = db_start + base_memmap[VIRT_GIC_V2M].size - 1;
2969             break;
2970         }
2971         resv_prop_str = g_strdup_printf("0x%"PRIx64":0x%"PRIx64":%u",
2972                                         db_start, db_end,
2973                                         VIRTIO_IOMMU_RESV_MEM_T_MSI);
2974 
2975         reserved_regions = qlist_new();
2976         qlist_append_str(reserved_regions, resv_prop_str);
2977         qdev_prop_set_array(dev, "reserved-regions", reserved_regions);
2978         g_free(resv_prop_str);
2979     }
2980 }
2981 
2982 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
2983                                         DeviceState *dev, Error **errp)
2984 {
2985     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2986 
2987     if (vms->platform_bus_dev) {
2988         MachineClass *mc = MACHINE_GET_CLASS(vms);
2989 
2990         if (device_is_dynamic_sysbus(mc, dev)) {
2991             platform_bus_link_device(PLATFORM_BUS_DEVICE(vms->platform_bus_dev),
2992                                      SYS_BUS_DEVICE(dev));
2993         }
2994     }
2995 
2996     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2997         virt_memory_plug(hotplug_dev, dev, errp);
2998     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
2999         virtio_md_pci_plug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
3000     }
3001 
3002     if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
3003         PCIDevice *pdev = PCI_DEVICE(dev);
3004 
3005         vms->iommu = VIRT_IOMMU_VIRTIO;
3006         vms->virtio_iommu_bdf = pci_get_bdf(pdev);
3007         create_virtio_iommu_dt_bindings(vms);
3008     }
3009 }
3010 
3011 static void virt_dimm_unplug_request(HotplugHandler *hotplug_dev,
3012                                      DeviceState *dev, Error **errp)
3013 {
3014     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
3015 
3016     if (!vms->acpi_dev) {
3017         error_setg(errp,
3018                    "memory hotplug is not enabled: missing acpi-ged device");
3019         return;
3020     }
3021 
3022     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3023         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3024         return;
3025     }
3026 
3027     hotplug_handler_unplug_request(HOTPLUG_HANDLER(vms->acpi_dev), dev,
3028                                    errp);
3029 }
3030 
3031 static void virt_dimm_unplug(HotplugHandler *hotplug_dev,
3032                              DeviceState *dev, Error **errp)
3033 {
3034     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
3035     Error *local_err = NULL;
3036 
3037     hotplug_handler_unplug(HOTPLUG_HANDLER(vms->acpi_dev), dev, &local_err);
3038     if (local_err) {
3039         goto out;
3040     }
3041 
3042     pc_dimm_unplug(PC_DIMM(dev), MACHINE(vms));
3043     qdev_unrealize(dev);
3044 
3045 out:
3046     error_propagate(errp, local_err);
3047 }
3048 
3049 static void virt_machine_device_unplug_request_cb(HotplugHandler *hotplug_dev,
3050                                           DeviceState *dev, Error **errp)
3051 {
3052     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3053         virt_dimm_unplug_request(hotplug_dev, dev, errp);
3054     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
3055         virtio_md_pci_unplug_request(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev),
3056                                      errp);
3057     } else {
3058         error_setg(errp, "device unplug request for unsupported device"
3059                    " type: %s", object_get_typename(OBJECT(dev)));
3060     }
3061 }
3062 
3063 static void virt_machine_device_unplug_cb(HotplugHandler *hotplug_dev,
3064                                           DeviceState *dev, Error **errp)
3065 {
3066     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3067         virt_dimm_unplug(hotplug_dev, dev, errp);
3068     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
3069         virtio_md_pci_unplug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
3070     } else {
3071         error_setg(errp, "virt: device unplug for unsupported device"
3072                    " type: %s", object_get_typename(OBJECT(dev)));
3073     }
3074 }
3075 
3076 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
3077                                                         DeviceState *dev)
3078 {
3079     MachineClass *mc = MACHINE_GET_CLASS(machine);
3080 
3081     if (device_is_dynamic_sysbus(mc, dev) ||
3082         object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3083         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI) ||
3084         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
3085         return HOTPLUG_HANDLER(machine);
3086     }
3087     return NULL;
3088 }
3089 
3090 /*
3091  * for arm64 kvm_type [7-0] encodes the requested number of bits
3092  * in the IPA address space
3093  */
3094 static int virt_kvm_type(MachineState *ms, const char *type_str)
3095 {
3096     VirtMachineState *vms = VIRT_MACHINE(ms);
3097     int max_vm_pa_size, requested_pa_size;
3098     bool fixed_ipa;
3099 
3100     max_vm_pa_size = kvm_arm_get_max_vm_ipa_size(ms, &fixed_ipa);
3101 
3102     /* we freeze the memory map to compute the highest gpa */
3103     virt_set_memmap(vms, max_vm_pa_size);
3104 
3105     requested_pa_size = 64 - clz64(vms->highest_gpa);
3106 
3107     /*
3108      * KVM requires the IPA size to be at least 32 bits.
3109      */
3110     if (requested_pa_size < 32) {
3111         requested_pa_size = 32;
3112     }
3113 
3114     if (requested_pa_size > max_vm_pa_size) {
3115         error_report("-m and ,maxmem option values "
3116                      "require an IPA range (%d bits) larger than "
3117                      "the one supported by the host (%d bits)",
3118                      requested_pa_size, max_vm_pa_size);
3119         return -1;
3120     }
3121     /*
3122      * We return the requested PA log size, unless KVM only supports
3123      * the implicit legacy 40b IPA setting, in which case the kvm_type
3124      * must be 0.
3125      */
3126     return fixed_ipa ? 0 : requested_pa_size;
3127 }
3128 
3129 static int virt_hvf_get_physical_address_range(MachineState *ms)
3130 {
3131     VirtMachineState *vms = VIRT_MACHINE(ms);
3132 
3133     int default_ipa_size = hvf_arm_get_default_ipa_bit_size();
3134     int max_ipa_size = hvf_arm_get_max_ipa_bit_size();
3135 
3136     /* We freeze the memory map to compute the highest gpa */
3137     virt_set_memmap(vms, max_ipa_size);
3138 
3139     int requested_ipa_size = 64 - clz64(vms->highest_gpa);
3140 
3141     /*
3142      * If we're <= the default IPA size just use the default.
3143      * If we're above the default but below the maximum, round up to
3144      * the maximum. hvf_arm_get_max_ipa_bit_size() conveniently only
3145      * returns values that are valid ARM PARange values.
3146      */
3147     if (requested_ipa_size <= default_ipa_size) {
3148         requested_ipa_size = default_ipa_size;
3149     } else if (requested_ipa_size <= max_ipa_size) {
3150         requested_ipa_size = max_ipa_size;
3151     } else {
3152         error_report("-m and ,maxmem option values "
3153                      "require an IPA range (%d bits) larger than "
3154                      "the one supported by the host (%d bits)",
3155                      requested_ipa_size, max_ipa_size);
3156         return -1;
3157     }
3158 
3159     return requested_ipa_size;
3160 }
3161 
3162 static void virt_machine_class_init(ObjectClass *oc, const void *data)
3163 {
3164     MachineClass *mc = MACHINE_CLASS(oc);
3165     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3166     static const char * const valid_cpu_types[] = {
3167 #ifdef CONFIG_TCG
3168         ARM_CPU_TYPE_NAME("cortex-a7"),
3169         ARM_CPU_TYPE_NAME("cortex-a15"),
3170 #ifdef TARGET_AARCH64
3171         ARM_CPU_TYPE_NAME("cortex-a35"),
3172         ARM_CPU_TYPE_NAME("cortex-a55"),
3173         ARM_CPU_TYPE_NAME("cortex-a72"),
3174         ARM_CPU_TYPE_NAME("cortex-a76"),
3175         ARM_CPU_TYPE_NAME("cortex-a710"),
3176         ARM_CPU_TYPE_NAME("a64fx"),
3177         ARM_CPU_TYPE_NAME("neoverse-n1"),
3178         ARM_CPU_TYPE_NAME("neoverse-v1"),
3179         ARM_CPU_TYPE_NAME("neoverse-n2"),
3180 #endif /* TARGET_AARCH64 */
3181 #endif /* CONFIG_TCG */
3182 #ifdef TARGET_AARCH64
3183         ARM_CPU_TYPE_NAME("cortex-a53"),
3184         ARM_CPU_TYPE_NAME("cortex-a57"),
3185 #if defined(CONFIG_KVM) || defined(CONFIG_HVF)
3186         ARM_CPU_TYPE_NAME("host"),
3187 #endif /* CONFIG_KVM || CONFIG_HVF */
3188 #endif /* TARGET_AARCH64 */
3189         ARM_CPU_TYPE_NAME("max"),
3190         NULL
3191     };
3192 
3193     mc->init = machvirt_init;
3194     /* Start with max_cpus set to 512, which is the maximum supported by KVM.
3195      * The value may be reduced later when we have more information about the
3196      * configuration of the particular instance.
3197      */
3198     mc->max_cpus = 512;
3199     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_CALXEDA_XGMAC);
3200     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_AMD_XGBE);
3201     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
3202     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_PLATFORM);
3203     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_UEFI_VARS_SYSBUS);
3204 #ifdef CONFIG_TPM
3205     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_TPM_TIS_SYSBUS);
3206 #endif
3207     mc->block_default_type = IF_VIRTIO;
3208     mc->no_cdrom = 1;
3209     mc->pci_allow_0_address = true;
3210     /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */
3211     mc->minimum_page_bits = 12;
3212     mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids;
3213     mc->cpu_index_to_instance_props = virt_cpu_index_to_props;
3214 #ifdef CONFIG_TCG
3215     mc->default_cpu_type = ARM_CPU_TYPE_NAME("cortex-a15");
3216 #else
3217     mc->default_cpu_type = ARM_CPU_TYPE_NAME("max");
3218 #endif
3219     mc->valid_cpu_types = valid_cpu_types;
3220     mc->get_default_cpu_node_id = virt_get_default_cpu_node_id;
3221     mc->kvm_type = virt_kvm_type;
3222     mc->hvf_get_physical_address_range = virt_hvf_get_physical_address_range;
3223     assert(!mc->get_hotplug_handler);
3224     mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
3225     hc->pre_plug = virt_machine_device_pre_plug_cb;
3226     hc->plug = virt_machine_device_plug_cb;
3227     hc->unplug_request = virt_machine_device_unplug_request_cb;
3228     hc->unplug = virt_machine_device_unplug_cb;
3229     mc->nvdimm_supported = true;
3230     mc->smp_props.clusters_supported = true;
3231     mc->auto_enable_numa_with_memhp = true;
3232     mc->auto_enable_numa_with_memdev = true;
3233     /* platform instead of architectural choice */
3234     mc->cpu_cluster_has_numa_boundary = true;
3235     mc->default_ram_id = "mach-virt.ram";
3236     mc->default_nic = "virtio-net-pci";
3237 
3238     object_class_property_add(oc, "acpi", "OnOffAuto",
3239         virt_get_acpi, virt_set_acpi,
3240         NULL, NULL);
3241     object_class_property_set_description(oc, "acpi",
3242         "Enable ACPI");
3243     object_class_property_add_bool(oc, "secure", virt_get_secure,
3244                                    virt_set_secure);
3245     object_class_property_set_description(oc, "secure",
3246                                                 "Set on/off to enable/disable the ARM "
3247                                                 "Security Extensions (TrustZone)");
3248 
3249     object_class_property_add_bool(oc, "virtualization", virt_get_virt,
3250                                    virt_set_virt);
3251     object_class_property_set_description(oc, "virtualization",
3252                                           "Set on/off to enable/disable emulating a "
3253                                           "guest CPU which implements the ARM "
3254                                           "Virtualization Extensions");
3255 
3256     object_class_property_add_bool(oc, "highmem", virt_get_highmem,
3257                                    virt_set_highmem);
3258     object_class_property_set_description(oc, "highmem",
3259                                           "Set on/off to enable/disable using "
3260                                           "physical address space above 32 bits");
3261 
3262     object_class_property_add_bool(oc, "compact-highmem",
3263                                    virt_get_compact_highmem,
3264                                    virt_set_compact_highmem);
3265     object_class_property_set_description(oc, "compact-highmem",
3266                                           "Set on/off to enable/disable compact "
3267                                           "layout for high memory regions");
3268 
3269     object_class_property_add_bool(oc, "highmem-redists",
3270                                    virt_get_highmem_redists,
3271                                    virt_set_highmem_redists);
3272     object_class_property_set_description(oc, "highmem-redists",
3273                                           "Set on/off to enable/disable high "
3274                                           "memory region for GICv3 or GICv4 "
3275                                           "redistributor");
3276 
3277     object_class_property_add_bool(oc, "highmem-ecam",
3278                                    virt_get_highmem_ecam,
3279                                    virt_set_highmem_ecam);
3280     object_class_property_set_description(oc, "highmem-ecam",
3281                                           "Set on/off to enable/disable high "
3282                                           "memory region for PCI ECAM");
3283 
3284     object_class_property_add_bool(oc, "highmem-mmio",
3285                                    virt_get_highmem_mmio,
3286                                    virt_set_highmem_mmio);
3287     object_class_property_set_description(oc, "highmem-mmio",
3288                                           "Set on/off to enable/disable high "
3289                                           "memory region for PCI MMIO");
3290 
3291     object_class_property_add(oc, "highmem-mmio-size", "size",
3292                                    virt_get_highmem_mmio_size,
3293                                    virt_set_highmem_mmio_size,
3294                                    NULL, NULL);
3295     object_class_property_set_description(oc, "highmem-mmio-size",
3296                                           "Set the high memory region size "
3297                                           "for PCI MMIO");
3298 
3299     object_class_property_add_str(oc, "gic-version", virt_get_gic_version,
3300                                   virt_set_gic_version);
3301     object_class_property_set_description(oc, "gic-version",
3302                                           "Set GIC version. "
3303                                           "Valid values are 2, 3, 4, host and max");
3304 
3305     object_class_property_add_str(oc, "iommu", virt_get_iommu, virt_set_iommu);
3306     object_class_property_set_description(oc, "iommu",
3307                                           "Set the IOMMU type. "
3308                                           "Valid values are none and smmuv3");
3309 
3310     object_class_property_add_bool(oc, "default-bus-bypass-iommu",
3311                                    virt_get_default_bus_bypass_iommu,
3312                                    virt_set_default_bus_bypass_iommu);
3313     object_class_property_set_description(oc, "default-bus-bypass-iommu",
3314                                           "Set on/off to enable/disable "
3315                                           "bypass_iommu for default root bus");
3316 
3317     object_class_property_add_bool(oc, "ras", virt_get_ras,
3318                                    virt_set_ras);
3319     object_class_property_set_description(oc, "ras",
3320                                           "Set on/off to enable/disable reporting host memory errors "
3321                                           "to a KVM guest using ACPI and guest external abort exceptions");
3322 
3323     object_class_property_add_bool(oc, "mte", virt_get_mte, virt_set_mte);
3324     object_class_property_set_description(oc, "mte",
3325                                           "Set on/off to enable/disable emulating a "
3326                                           "guest CPU which implements the ARM "
3327                                           "Memory Tagging Extension");
3328 
3329     object_class_property_add_bool(oc, "its", virt_get_its,
3330                                    virt_set_its);
3331     object_class_property_set_description(oc, "its",
3332                                           "Set on/off to enable/disable "
3333                                           "ITS instantiation");
3334 
3335     object_class_property_add_bool(oc, "dtb-randomness",
3336                                    virt_get_dtb_randomness,
3337                                    virt_set_dtb_randomness);
3338     object_class_property_set_description(oc, "dtb-randomness",
3339                                           "Set off to disable passing random or "
3340                                           "non-deterministic dtb nodes to guest");
3341 
3342     object_class_property_add_bool(oc, "dtb-kaslr-seed",
3343                                    virt_get_dtb_randomness,
3344                                    virt_set_dtb_randomness);
3345     object_class_property_set_description(oc, "dtb-kaslr-seed",
3346                                           "Deprecated synonym of dtb-randomness");
3347 
3348     object_class_property_add_str(oc, "x-oem-id",
3349                                   virt_get_oem_id,
3350                                   virt_set_oem_id);
3351     object_class_property_set_description(oc, "x-oem-id",
3352                                           "Override the default value of field OEMID "
3353                                           "in ACPI table header."
3354                                           "The string may be up to 6 bytes in size");
3355 
3356 
3357     object_class_property_add_str(oc, "x-oem-table-id",
3358                                   virt_get_oem_table_id,
3359                                   virt_set_oem_table_id);
3360     object_class_property_set_description(oc, "x-oem-table-id",
3361                                           "Override the default value of field OEM Table ID "
3362                                           "in ACPI table header."
3363                                           "The string may be up to 8 bytes in size");
3364 
3365 }
3366 
3367 static void virt_instance_init(Object *obj)
3368 {
3369     VirtMachineState *vms = VIRT_MACHINE(obj);
3370     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
3371 
3372     /* EL3 is disabled by default on virt: this makes us consistent
3373      * between KVM and TCG for this board, and it also allows us to
3374      * boot UEFI blobs which assume no TrustZone support.
3375      */
3376     vms->secure = false;
3377 
3378     /* EL2 is also disabled by default, for similar reasons */
3379     vms->virt = false;
3380 
3381     /* High memory is enabled by default */
3382     vms->highmem = true;
3383     vms->highmem_compact = !vmc->no_highmem_compact;
3384     vms->gic_version = VIRT_GIC_VERSION_NOSEL;
3385 
3386     vms->highmem_ecam = true;
3387     vms->highmem_mmio = true;
3388     vms->highmem_redists = true;
3389 
3390     /* Default allows ITS instantiation */
3391     vms->its = true;
3392     /* Allow ITS emulation if the machine version supports it */
3393     vms->tcg_its = !vmc->no_tcg_its;
3394 
3395     /* Default disallows iommu instantiation */
3396     vms->iommu = VIRT_IOMMU_NONE;
3397 
3398     /* The default root bus is attached to iommu by default */
3399     vms->default_bus_bypass_iommu = false;
3400 
3401     /* Default disallows RAS instantiation */
3402     vms->ras = false;
3403 
3404     /* MTE is disabled by default.  */
3405     vms->mte = false;
3406 
3407     /* Supply kaslr-seed and rng-seed by default */
3408     vms->dtb_randomness = true;
3409 
3410     vms->irqmap = a15irqmap;
3411 
3412     virt_flash_create(vms);
3413 
3414     vms->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
3415     vms->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
3416     cxl_machine_init(obj, &vms->cxl_devices_state);
3417 }
3418 
3419 static const TypeInfo virt_machine_info = {
3420     .name          = TYPE_VIRT_MACHINE,
3421     .parent        = TYPE_MACHINE,
3422     .abstract      = true,
3423     .instance_size = sizeof(VirtMachineState),
3424     .class_size    = sizeof(VirtMachineClass),
3425     .class_init    = virt_machine_class_init,
3426     .instance_init = virt_instance_init,
3427     .interfaces = (const InterfaceInfo[]) {
3428          { TYPE_HOTPLUG_HANDLER },
3429          { }
3430     },
3431 };
3432 
3433 static void machvirt_machine_init(void)
3434 {
3435     type_register_static(&virt_machine_info);
3436 }
3437 type_init(machvirt_machine_init);
3438 
3439 static void virt_machine_10_1_options(MachineClass *mc)
3440 {
3441 }
3442 DEFINE_VIRT_MACHINE_AS_LATEST(10, 1)
3443 
3444 static void virt_machine_10_0_options(MachineClass *mc)
3445 {
3446     virt_machine_10_1_options(mc);
3447     compat_props_add(mc->compat_props, hw_compat_10_0, hw_compat_10_0_len);
3448 }
3449 DEFINE_VIRT_MACHINE(10, 0)
3450 
3451 static void virt_machine_9_2_options(MachineClass *mc)
3452 {
3453     virt_machine_10_0_options(mc);
3454     compat_props_add(mc->compat_props, hw_compat_9_2, hw_compat_9_2_len);
3455 }
3456 DEFINE_VIRT_MACHINE(9, 2)
3457 
3458 static void virt_machine_9_1_options(MachineClass *mc)
3459 {
3460     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3461 
3462     virt_machine_9_2_options(mc);
3463     compat_props_add(mc->compat_props, hw_compat_9_1, hw_compat_9_1_len);
3464     /* 9.1 and earlier have only a stage-1 SMMU, not a nested s1+2 one */
3465     vmc->no_nested_smmu = true;
3466 }
3467 DEFINE_VIRT_MACHINE(9, 1)
3468 
3469 static void virt_machine_9_0_options(MachineClass *mc)
3470 {
3471     virt_machine_9_1_options(mc);
3472     mc->smbios_memory_device_size = 16 * GiB;
3473     compat_props_add(mc->compat_props, hw_compat_9_0, hw_compat_9_0_len);
3474 }
3475 DEFINE_VIRT_MACHINE(9, 0)
3476 
3477 static void virt_machine_8_2_options(MachineClass *mc)
3478 {
3479     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3480 
3481     virt_machine_9_0_options(mc);
3482     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
3483     /*
3484      * Don't expose NS_EL2_VIRT timer IRQ in DTB on ACPI on 8.2 and
3485      * earlier machines. (Exposing it tickles a bug in older EDK2
3486      * guest BIOS binaries.)
3487      */
3488     vmc->no_ns_el2_virt_timer_irq = true;
3489 }
3490 DEFINE_VIRT_MACHINE(8, 2)
3491 
3492 static void virt_machine_8_1_options(MachineClass *mc)
3493 {
3494     virt_machine_8_2_options(mc);
3495     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
3496 }
3497 DEFINE_VIRT_MACHINE(8, 1)
3498 
3499 static void virt_machine_8_0_options(MachineClass *mc)
3500 {
3501     virt_machine_8_1_options(mc);
3502     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
3503 }
3504 DEFINE_VIRT_MACHINE(8, 0)
3505 
3506 static void virt_machine_7_2_options(MachineClass *mc)
3507 {
3508     virt_machine_8_0_options(mc);
3509     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
3510 }
3511 DEFINE_VIRT_MACHINE(7, 2)
3512 
3513 static void virt_machine_7_1_options(MachineClass *mc)
3514 {
3515     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3516 
3517     virt_machine_7_2_options(mc);
3518     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
3519     /* Compact layout for high memory regions was introduced with 7.2 */
3520     vmc->no_highmem_compact = true;
3521 }
3522 DEFINE_VIRT_MACHINE(7, 1)
3523 
3524 static void virt_machine_7_0_options(MachineClass *mc)
3525 {
3526     virt_machine_7_1_options(mc);
3527     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
3528 }
3529 DEFINE_VIRT_MACHINE(7, 0)
3530 
3531 static void virt_machine_6_2_options(MachineClass *mc)
3532 {
3533     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3534 
3535     virt_machine_7_0_options(mc);
3536     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
3537     vmc->no_tcg_lpa2 = true;
3538 }
3539 DEFINE_VIRT_MACHINE(6, 2)
3540 
3541 static void virt_machine_6_1_options(MachineClass *mc)
3542 {
3543     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3544 
3545     virt_machine_6_2_options(mc);
3546     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
3547     mc->smp_props.prefer_sockets = true;
3548     vmc->no_cpu_topology = true;
3549 
3550     /* qemu ITS was introduced with 6.2 */
3551     vmc->no_tcg_its = true;
3552 }
3553 DEFINE_VIRT_MACHINE(6, 1)
3554 
3555 static void virt_machine_6_0_options(MachineClass *mc)
3556 {
3557     virt_machine_6_1_options(mc);
3558     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
3559 }
3560 DEFINE_VIRT_MACHINE(6, 0)
3561 
3562 static void virt_machine_5_2_options(MachineClass *mc)
3563 {
3564     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3565 
3566     virt_machine_6_0_options(mc);
3567     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
3568     vmc->no_secure_gpio = true;
3569 }
3570 DEFINE_VIRT_MACHINE(5, 2)
3571 
3572 static void virt_machine_5_1_options(MachineClass *mc)
3573 {
3574     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3575 
3576     virt_machine_5_2_options(mc);
3577     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
3578     vmc->no_kvm_steal_time = true;
3579 }
3580 DEFINE_VIRT_MACHINE(5, 1)
3581 
3582 static void virt_machine_5_0_options(MachineClass *mc)
3583 {
3584     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3585 
3586     virt_machine_5_1_options(mc);
3587     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
3588     mc->numa_mem_supported = true;
3589     vmc->acpi_expose_flash = true;
3590     mc->auto_enable_numa_with_memdev = false;
3591 }
3592 DEFINE_VIRT_MACHINE(5, 0)
3593 
3594 static void virt_machine_4_2_options(MachineClass *mc)
3595 {
3596     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3597 
3598     virt_machine_5_0_options(mc);
3599     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
3600     vmc->kvm_no_adjvtime = true;
3601 }
3602 DEFINE_VIRT_MACHINE(4, 2)
3603 
3604 static void virt_machine_4_1_options(MachineClass *mc)
3605 {
3606     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3607 
3608     virt_machine_4_2_options(mc);
3609     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
3610     vmc->no_ged = true;
3611     mc->auto_enable_numa_with_memhp = false;
3612 }
3613 DEFINE_VIRT_MACHINE(4, 1)
3614