xref: /openbmc/qemu/hw/arm/virt.c (revision 1fea334eeed2b747d1c91ee0099401595ba697f8)
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     vms->gic = qdev_new(gictype);
801     qdev_prop_set_uint32(vms->gic, "revision", revision);
802     qdev_prop_set_uint32(vms->gic, "num-cpu", smp_cpus);
803     /* Note that the num-irq property counts both internal and external
804      * interrupts; there are always 32 of the former (mandated by GIC spec).
805      */
806     qdev_prop_set_uint32(vms->gic, "num-irq", NUM_IRQS + 32);
807     if (!kvm_irqchip_in_kernel()) {
808         qdev_prop_set_bit(vms->gic, "has-security-extensions", vms->secure);
809     }
810 
811     if (vms->gic_version != VIRT_GIC_VERSION_2) {
812         QList *redist_region_count;
813         uint32_t redist0_capacity = virt_redist_capacity(vms, VIRT_GIC_REDIST);
814         uint32_t redist0_count = MIN(smp_cpus, redist0_capacity);
815 
816         nb_redist_regions = virt_gicv3_redist_region_count(vms);
817 
818         redist_region_count = qlist_new();
819         qlist_append_int(redist_region_count, redist0_count);
820         if (nb_redist_regions == 2) {
821             uint32_t redist1_capacity =
822                 virt_redist_capacity(vms, VIRT_HIGH_GIC_REDIST2);
823 
824             qlist_append_int(redist_region_count,
825                 MIN(smp_cpus - redist0_count, redist1_capacity));
826         }
827         qdev_prop_set_array(vms->gic, "redist-region-count",
828                             redist_region_count);
829 
830         if (!kvm_irqchip_in_kernel()) {
831             if (vms->tcg_its) {
832                 object_property_set_link(OBJECT(vms->gic), "sysmem",
833                                          OBJECT(mem), &error_fatal);
834                 qdev_prop_set_bit(vms->gic, "has-lpi", true);
835             }
836         }
837     } else {
838         if (!kvm_irqchip_in_kernel()) {
839             qdev_prop_set_bit(vms->gic, "has-virtualization-extensions",
840                               vms->virt);
841         }
842     }
843 
844     if (gicv3_nmi_present(vms)) {
845         qdev_prop_set_bit(vms->gic, "has-nmi", true);
846     }
847 
848     gicbusdev = SYS_BUS_DEVICE(vms->gic);
849     sysbus_realize_and_unref(gicbusdev, &error_fatal);
850     sysbus_mmio_map(gicbusdev, 0, vms->memmap[VIRT_GIC_DIST].base);
851     if (vms->gic_version != VIRT_GIC_VERSION_2) {
852         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_REDIST].base);
853         if (nb_redist_regions == 2) {
854             sysbus_mmio_map(gicbusdev, 2,
855                             vms->memmap[VIRT_HIGH_GIC_REDIST2].base);
856         }
857     } else {
858         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base);
859         if (vms->virt) {
860             sysbus_mmio_map(gicbusdev, 2, vms->memmap[VIRT_GIC_HYP].base);
861             sysbus_mmio_map(gicbusdev, 3, vms->memmap[VIRT_GIC_VCPU].base);
862         }
863     }
864 
865     /* Wire the outputs from each CPU's generic timer and the GICv3
866      * maintenance interrupt signal to the appropriate GIC PPI inputs,
867      * and the GIC's IRQ/FIQ/VIRQ/VFIQ/NMI/VINMI interrupt outputs to the
868      * CPU's inputs.
869      */
870     for (i = 0; i < smp_cpus; i++) {
871         DeviceState *cpudev = DEVICE(qemu_get_cpu(i));
872         int intidbase = NUM_IRQS + i * GIC_INTERNAL;
873         /* Mapping from the output timer irq lines from the CPU to the
874          * GIC PPI inputs we use for the virt board.
875          */
876         const int timer_irq[] = {
877             [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ,
878             [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ,
879             [GTIMER_HYP]  = ARCH_TIMER_NS_EL2_IRQ,
880             [GTIMER_SEC]  = ARCH_TIMER_S_EL1_IRQ,
881             [GTIMER_HYPVIRT] = ARCH_TIMER_NS_EL2_VIRT_IRQ,
882             [GTIMER_S_EL2_PHYS] = ARCH_TIMER_S_EL2_IRQ,
883             [GTIMER_S_EL2_VIRT] = ARCH_TIMER_S_EL2_VIRT_IRQ,
884         };
885 
886         for (unsigned irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) {
887             qdev_connect_gpio_out(cpudev, irq,
888                                   qdev_get_gpio_in(vms->gic,
889                                                    intidbase + timer_irq[irq]));
890         }
891 
892         if (vms->gic_version != VIRT_GIC_VERSION_2) {
893             qemu_irq irq = qdev_get_gpio_in(vms->gic,
894                                             intidbase + ARCH_GIC_MAINT_IRQ);
895             qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt",
896                                         0, irq);
897         } else if (vms->virt) {
898             qemu_irq irq = qdev_get_gpio_in(vms->gic,
899                                             intidbase + ARCH_GIC_MAINT_IRQ);
900             sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus, irq);
901         }
902 
903         qdev_connect_gpio_out_named(cpudev, "pmu-interrupt", 0,
904                                     qdev_get_gpio_in(vms->gic, intidbase
905                                                      + VIRTUAL_PMU_IRQ));
906 
907         sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ));
908         sysbus_connect_irq(gicbusdev, i + smp_cpus,
909                            qdev_get_gpio_in(cpudev, ARM_CPU_FIQ));
910         sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus,
911                            qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ));
912         sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus,
913                            qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ));
914 
915         if (vms->gic_version != VIRT_GIC_VERSION_2) {
916             sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus,
917                                qdev_get_gpio_in(cpudev, ARM_CPU_NMI));
918             sysbus_connect_irq(gicbusdev, i + 5 * smp_cpus,
919                                qdev_get_gpio_in(cpudev, ARM_CPU_VINMI));
920         }
921     }
922 
923     fdt_add_gic_node(vms);
924 
925     if (vms->gic_version != VIRT_GIC_VERSION_2 && vms->its) {
926         create_its(vms);
927     } else if (vms->gic_version == VIRT_GIC_VERSION_2) {
928         create_v2m(vms);
929     }
930 }
931 
932 static void create_uart(const VirtMachineState *vms, int uart,
933                         MemoryRegion *mem, Chardev *chr, bool secure)
934 {
935     char *nodename;
936     hwaddr base = vms->memmap[uart].base;
937     hwaddr size = vms->memmap[uart].size;
938     int irq = vms->irqmap[uart];
939     const char compat[] = "arm,pl011\0arm,primecell";
940     const char clocknames[] = "uartclk\0apb_pclk";
941     DeviceState *dev = qdev_new(TYPE_PL011);
942     SysBusDevice *s = SYS_BUS_DEVICE(dev);
943     MachineState *ms = MACHINE(vms);
944 
945     qdev_prop_set_chr(dev, "chardev", chr);
946     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
947     memory_region_add_subregion(mem, base,
948                                 sysbus_mmio_get_region(s, 0));
949     sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
950 
951     nodename = g_strdup_printf("/pl011@%" PRIx64, base);
952     qemu_fdt_add_subnode(ms->fdt, nodename);
953     /* Note that we can't use setprop_string because of the embedded NUL */
954     qemu_fdt_setprop(ms->fdt, nodename, "compatible",
955                          compat, sizeof(compat));
956     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
957                                      2, base, 2, size);
958     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
959                                GIC_FDT_IRQ_TYPE_SPI, irq,
960                                GIC_FDT_IRQ_FLAGS_LEVEL_HI);
961     qemu_fdt_setprop_cells(ms->fdt, nodename, "clocks",
962                                vms->clock_phandle, vms->clock_phandle);
963     qemu_fdt_setprop(ms->fdt, nodename, "clock-names",
964                          clocknames, sizeof(clocknames));
965 
966     if (uart == VIRT_UART0) {
967         qemu_fdt_setprop_string(ms->fdt, "/chosen", "stdout-path", nodename);
968         qemu_fdt_setprop_string(ms->fdt, "/aliases", "serial0", nodename);
969     } else {
970         qemu_fdt_setprop_string(ms->fdt, "/aliases", "serial1", nodename);
971     }
972     if (secure) {
973         /* Mark as not usable by the normal world */
974         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
975         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
976 
977         qemu_fdt_setprop_string(ms->fdt, "/secure-chosen", "stdout-path",
978                                 nodename);
979     }
980 
981     g_free(nodename);
982 }
983 
984 static void create_rtc(const VirtMachineState *vms)
985 {
986     char *nodename;
987     hwaddr base = vms->memmap[VIRT_RTC].base;
988     hwaddr size = vms->memmap[VIRT_RTC].size;
989     int irq = vms->irqmap[VIRT_RTC];
990     const char compat[] = "arm,pl031\0arm,primecell";
991     MachineState *ms = MACHINE(vms);
992 
993     sysbus_create_simple("pl031", base, qdev_get_gpio_in(vms->gic, irq));
994 
995     nodename = g_strdup_printf("/pl031@%" PRIx64, base);
996     qemu_fdt_add_subnode(ms->fdt, nodename);
997     qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
998     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
999                                  2, base, 2, size);
1000     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1001                            GIC_FDT_IRQ_TYPE_SPI, irq,
1002                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
1003     qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
1004     qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
1005     g_free(nodename);
1006 }
1007 
1008 static DeviceState *gpio_key_dev;
1009 static void virt_powerdown_req(Notifier *n, void *opaque)
1010 {
1011     VirtMachineState *s = container_of(n, VirtMachineState, powerdown_notifier);
1012 
1013     if (s->acpi_dev) {
1014         acpi_send_event(s->acpi_dev, ACPI_POWER_DOWN_STATUS);
1015     } else {
1016         /* use gpio Pin for power button event */
1017         qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1);
1018     }
1019 }
1020 
1021 static void create_gpio_keys(char *fdt, DeviceState *pl061_dev,
1022                              uint32_t phandle)
1023 {
1024     gpio_key_dev = sysbus_create_simple("gpio-key", -1,
1025                                         qdev_get_gpio_in(pl061_dev,
1026                                                          GPIO_PIN_POWER_BUTTON));
1027 
1028     qemu_fdt_add_subnode(fdt, "/gpio-keys");
1029     qemu_fdt_setprop_string(fdt, "/gpio-keys", "compatible", "gpio-keys");
1030 
1031     qemu_fdt_add_subnode(fdt, "/gpio-keys/poweroff");
1032     qemu_fdt_setprop_string(fdt, "/gpio-keys/poweroff",
1033                             "label", "GPIO Key Poweroff");
1034     qemu_fdt_setprop_cell(fdt, "/gpio-keys/poweroff", "linux,code",
1035                           KEY_POWER);
1036     qemu_fdt_setprop_cells(fdt, "/gpio-keys/poweroff",
1037                            "gpios", phandle, GPIO_PIN_POWER_BUTTON, 0);
1038 }
1039 
1040 #define SECURE_GPIO_POWEROFF 0
1041 #define SECURE_GPIO_RESET    1
1042 
1043 static void create_secure_gpio_pwr(char *fdt, DeviceState *pl061_dev,
1044                                    uint32_t phandle)
1045 {
1046     DeviceState *gpio_pwr_dev;
1047 
1048     /* gpio-pwr */
1049     gpio_pwr_dev = sysbus_create_simple("gpio-pwr", -1, NULL);
1050 
1051     /* connect secure pl061 to gpio-pwr */
1052     qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_RESET,
1053                           qdev_get_gpio_in_named(gpio_pwr_dev, "reset", 0));
1054     qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_POWEROFF,
1055                           qdev_get_gpio_in_named(gpio_pwr_dev, "shutdown", 0));
1056 
1057     qemu_fdt_add_subnode(fdt, "/gpio-poweroff");
1058     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "compatible",
1059                             "gpio-poweroff");
1060     qemu_fdt_setprop_cells(fdt, "/gpio-poweroff",
1061                            "gpios", phandle, SECURE_GPIO_POWEROFF, 0);
1062     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "status", "disabled");
1063     qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "secure-status",
1064                             "okay");
1065 
1066     qemu_fdt_add_subnode(fdt, "/gpio-restart");
1067     qemu_fdt_setprop_string(fdt, "/gpio-restart", "compatible",
1068                             "gpio-restart");
1069     qemu_fdt_setprop_cells(fdt, "/gpio-restart",
1070                            "gpios", phandle, SECURE_GPIO_RESET, 0);
1071     qemu_fdt_setprop_string(fdt, "/gpio-restart", "status", "disabled");
1072     qemu_fdt_setprop_string(fdt, "/gpio-restart", "secure-status",
1073                             "okay");
1074 }
1075 
1076 static void create_gpio_devices(const VirtMachineState *vms, int gpio,
1077                                 MemoryRegion *mem)
1078 {
1079     char *nodename;
1080     DeviceState *pl061_dev;
1081     hwaddr base = vms->memmap[gpio].base;
1082     hwaddr size = vms->memmap[gpio].size;
1083     int irq = vms->irqmap[gpio];
1084     const char compat[] = "arm,pl061\0arm,primecell";
1085     SysBusDevice *s;
1086     MachineState *ms = MACHINE(vms);
1087 
1088     pl061_dev = qdev_new("pl061");
1089     /* Pull lines down to 0 if not driven by the PL061 */
1090     qdev_prop_set_uint32(pl061_dev, "pullups", 0);
1091     qdev_prop_set_uint32(pl061_dev, "pulldowns", 0xff);
1092     s = SYS_BUS_DEVICE(pl061_dev);
1093     sysbus_realize_and_unref(s, &error_fatal);
1094     memory_region_add_subregion(mem, base, sysbus_mmio_get_region(s, 0));
1095     sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
1096 
1097     uint32_t phandle = qemu_fdt_alloc_phandle(ms->fdt);
1098     nodename = g_strdup_printf("/pl061@%" PRIx64, base);
1099     qemu_fdt_add_subnode(ms->fdt, nodename);
1100     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1101                                  2, base, 2, size);
1102     qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
1103     qemu_fdt_setprop_cell(ms->fdt, nodename, "#gpio-cells", 2);
1104     qemu_fdt_setprop(ms->fdt, nodename, "gpio-controller", NULL, 0);
1105     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1106                            GIC_FDT_IRQ_TYPE_SPI, irq,
1107                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
1108     qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
1109     qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
1110     qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", phandle);
1111 
1112     if (gpio != VIRT_GPIO) {
1113         /* Mark as not usable by the normal world */
1114         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1115         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1116     }
1117     g_free(nodename);
1118 
1119     /* Child gpio devices */
1120     if (gpio == VIRT_GPIO) {
1121         create_gpio_keys(ms->fdt, pl061_dev, phandle);
1122     } else {
1123         create_secure_gpio_pwr(ms->fdt, pl061_dev, phandle);
1124     }
1125 }
1126 
1127 static void create_virtio_devices(const VirtMachineState *vms)
1128 {
1129     int i;
1130     hwaddr size = vms->memmap[VIRT_MMIO].size;
1131     MachineState *ms = MACHINE(vms);
1132 
1133     /* We create the transports in forwards order. Since qbus_realize()
1134      * prepends (not appends) new child buses, the incrementing loop below will
1135      * create a list of virtio-mmio buses with decreasing base addresses.
1136      *
1137      * When a -device option is processed from the command line,
1138      * qbus_find_recursive() picks the next free virtio-mmio bus in forwards
1139      * order. The upshot is that -device options in increasing command line
1140      * order are mapped to virtio-mmio buses with decreasing base addresses.
1141      *
1142      * When this code was originally written, that arrangement ensured that the
1143      * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to
1144      * the first -device on the command line. (The end-to-end order is a
1145      * function of this loop, qbus_realize(), qbus_find_recursive(), and the
1146      * guest kernel's name-to-address assignment strategy.)
1147      *
1148      * Meanwhile, the kernel's traversal seems to have been reversed; see eg.
1149      * the message, if not necessarily the code, of commit 70161ff336.
1150      * Therefore the loop now establishes the inverse of the original intent.
1151      *
1152      * Unfortunately, we can't counteract the kernel change by reversing the
1153      * loop; it would break existing command lines.
1154      *
1155      * In any case, the kernel makes no guarantee about the stability of
1156      * enumeration order of virtio devices (as demonstrated by it changing
1157      * between kernel versions). For reliable and stable identification
1158      * of disks users must use UUIDs or similar mechanisms.
1159      */
1160     for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) {
1161         int irq = vms->irqmap[VIRT_MMIO] + i;
1162         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1163 
1164         sysbus_create_simple("virtio-mmio", base,
1165                              qdev_get_gpio_in(vms->gic, irq));
1166     }
1167 
1168     /* We add dtb nodes in reverse order so that they appear in the finished
1169      * device tree lowest address first.
1170      *
1171      * Note that this mapping is independent of the loop above. The previous
1172      * loop influences virtio device to virtio transport assignment, whereas
1173      * this loop controls how virtio transports are laid out in the dtb.
1174      */
1175     for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) {
1176         char *nodename;
1177         int irq = vms->irqmap[VIRT_MMIO] + i;
1178         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1179 
1180         nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base);
1181         qemu_fdt_add_subnode(ms->fdt, nodename);
1182         qemu_fdt_setprop_string(ms->fdt, nodename,
1183                                 "compatible", "virtio,mmio");
1184         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1185                                      2, base, 2, size);
1186         qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1187                                GIC_FDT_IRQ_TYPE_SPI, irq,
1188                                GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1189         qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1190         g_free(nodename);
1191     }
1192 }
1193 
1194 #define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
1195 
1196 static PFlashCFI01 *virt_flash_create1(VirtMachineState *vms,
1197                                         const char *name,
1198                                         const char *alias_prop_name)
1199 {
1200     /*
1201      * Create a single flash device.  We use the same parameters as
1202      * the flash devices on the Versatile Express board.
1203      */
1204     DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
1205 
1206     qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
1207     qdev_prop_set_uint8(dev, "width", 4);
1208     qdev_prop_set_uint8(dev, "device-width", 2);
1209     qdev_prop_set_bit(dev, "big-endian", false);
1210     qdev_prop_set_uint16(dev, "id0", 0x89);
1211     qdev_prop_set_uint16(dev, "id1", 0x18);
1212     qdev_prop_set_uint16(dev, "id2", 0x00);
1213     qdev_prop_set_uint16(dev, "id3", 0x00);
1214     qdev_prop_set_string(dev, "name", name);
1215     object_property_add_child(OBJECT(vms), name, OBJECT(dev));
1216     object_property_add_alias(OBJECT(vms), alias_prop_name,
1217                               OBJECT(dev), "drive");
1218     return PFLASH_CFI01(dev);
1219 }
1220 
1221 static void virt_flash_create(VirtMachineState *vms)
1222 {
1223     vms->flash[0] = virt_flash_create1(vms, "virt.flash0", "pflash0");
1224     vms->flash[1] = virt_flash_create1(vms, "virt.flash1", "pflash1");
1225 }
1226 
1227 static void virt_flash_map1(PFlashCFI01 *flash,
1228                             hwaddr base, hwaddr size,
1229                             MemoryRegion *sysmem)
1230 {
1231     DeviceState *dev = DEVICE(flash);
1232 
1233     assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
1234     assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
1235     qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
1236     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1237 
1238     memory_region_add_subregion(sysmem, base,
1239                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
1240                                                        0));
1241 }
1242 
1243 static void virt_flash_map(VirtMachineState *vms,
1244                            MemoryRegion *sysmem,
1245                            MemoryRegion *secure_sysmem)
1246 {
1247     /*
1248      * Map two flash devices to fill the VIRT_FLASH space in the memmap.
1249      * sysmem is the system memory space. secure_sysmem is the secure view
1250      * of the system, and the first flash device should be made visible only
1251      * there. The second flash device is visible to both secure and nonsecure.
1252      * If sysmem == secure_sysmem this means there is no separate Secure
1253      * address space and both flash devices are generally visible.
1254      */
1255     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1256     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1257 
1258     virt_flash_map1(vms->flash[0], flashbase, flashsize,
1259                     secure_sysmem);
1260     virt_flash_map1(vms->flash[1], flashbase + flashsize, flashsize,
1261                     sysmem);
1262 }
1263 
1264 static void virt_flash_fdt(VirtMachineState *vms,
1265                            MemoryRegion *sysmem,
1266                            MemoryRegion *secure_sysmem)
1267 {
1268     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1269     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1270     MachineState *ms = MACHINE(vms);
1271     char *nodename;
1272 
1273     if (sysmem == secure_sysmem) {
1274         /* Report both flash devices as a single node in the DT */
1275         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
1276         qemu_fdt_add_subnode(ms->fdt, nodename);
1277         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1278         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1279                                      2, flashbase, 2, flashsize,
1280                                      2, flashbase + flashsize, 2, flashsize);
1281         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1282         g_free(nodename);
1283     } else {
1284         /*
1285          * Report the devices as separate nodes so we can mark one as
1286          * only visible to the secure world.
1287          */
1288         nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase);
1289         qemu_fdt_add_subnode(ms->fdt, nodename);
1290         qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1291         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1292                                      2, flashbase, 2, flashsize);
1293         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1294         qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1295         qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1296         g_free(nodename);
1297 
1298         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase + flashsize);
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 + flashsize, 2, flashsize);
1303         qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1304         g_free(nodename);
1305     }
1306 }
1307 
1308 static bool virt_firmware_init(VirtMachineState *vms,
1309                                MemoryRegion *sysmem,
1310                                MemoryRegion *secure_sysmem)
1311 {
1312     int i;
1313     const char *bios_name;
1314     BlockBackend *pflash_blk0;
1315 
1316     /* Map legacy -drive if=pflash to machine properties */
1317     for (i = 0; i < ARRAY_SIZE(vms->flash); i++) {
1318         pflash_cfi01_legacy_drive(vms->flash[i],
1319                                   drive_get(IF_PFLASH, 0, i));
1320     }
1321 
1322     virt_flash_map(vms, sysmem, secure_sysmem);
1323 
1324     pflash_blk0 = pflash_cfi01_get_blk(vms->flash[0]);
1325 
1326     bios_name = MACHINE(vms)->firmware;
1327     if (bios_name) {
1328         char *fname;
1329         MemoryRegion *mr;
1330         int image_size;
1331 
1332         if (pflash_blk0) {
1333             error_report("The contents of the first flash device may be "
1334                          "specified with -bios or with -drive if=pflash... "
1335                          "but you cannot use both options at once");
1336             exit(1);
1337         }
1338 
1339         /* Fall back to -bios */
1340 
1341         fname = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1342         if (!fname) {
1343             error_report("Could not find ROM image '%s'", bios_name);
1344             exit(1);
1345         }
1346         mr = sysbus_mmio_get_region(SYS_BUS_DEVICE(vms->flash[0]), 0);
1347         image_size = load_image_mr(fname, mr);
1348         g_free(fname);
1349         if (image_size < 0) {
1350             error_report("Could not load ROM image '%s'", bios_name);
1351             exit(1);
1352         }
1353     }
1354 
1355     return pflash_blk0 || bios_name;
1356 }
1357 
1358 static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as)
1359 {
1360     MachineState *ms = MACHINE(vms);
1361     hwaddr base = vms->memmap[VIRT_FW_CFG].base;
1362     hwaddr size = vms->memmap[VIRT_FW_CFG].size;
1363     FWCfgState *fw_cfg;
1364     char *nodename;
1365 
1366     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as);
1367     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)ms->smp.cpus);
1368 
1369     nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
1370     qemu_fdt_add_subnode(ms->fdt, nodename);
1371     qemu_fdt_setprop_string(ms->fdt, nodename,
1372                             "compatible", "qemu,fw-cfg-mmio");
1373     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1374                                  2, base, 2, size);
1375     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1376     g_free(nodename);
1377     return fw_cfg;
1378 }
1379 
1380 static void create_pcie_irq_map(const MachineState *ms,
1381                                 uint32_t gic_phandle,
1382                                 int first_irq, const char *nodename)
1383 {
1384     int devfn, pin;
1385     uint32_t full_irq_map[4 * 4 * 10] = { 0 };
1386     uint32_t *irq_map = full_irq_map;
1387 
1388     for (devfn = 0; devfn <= 0x18; devfn += 0x8) {
1389         for (pin = 0; pin < 4; pin++) {
1390             int irq_type = GIC_FDT_IRQ_TYPE_SPI;
1391             int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS);
1392             int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
1393             int i;
1394 
1395             uint32_t map[] = {
1396                 devfn << 8, 0, 0,                           /* devfn */
1397                 pin + 1,                                    /* PCI pin */
1398                 gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */
1399 
1400             /* Convert map to big endian */
1401             for (i = 0; i < 10; i++) {
1402                 irq_map[i] = cpu_to_be32(map[i]);
1403             }
1404             irq_map += 10;
1405         }
1406     }
1407 
1408     qemu_fdt_setprop(ms->fdt, nodename, "interrupt-map",
1409                      full_irq_map, sizeof(full_irq_map));
1410 
1411     qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupt-map-mask",
1412                            cpu_to_be16(PCI_DEVFN(3, 0)), /* Slot 3 */
1413                            0, 0,
1414                            0x7           /* PCI irq */);
1415 }
1416 
1417 static void create_smmu(const VirtMachineState *vms,
1418                         PCIBus *bus)
1419 {
1420     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1421     char *node;
1422     const char compat[] = "arm,smmu-v3";
1423     int irq =  vms->irqmap[VIRT_SMMU];
1424     int i;
1425     hwaddr base = vms->memmap[VIRT_SMMU].base;
1426     hwaddr size = vms->memmap[VIRT_SMMU].size;
1427     const char irq_names[] = "eventq\0priq\0cmdq-sync\0gerror";
1428     DeviceState *dev;
1429     MachineState *ms = MACHINE(vms);
1430 
1431     if (vms->iommu != VIRT_IOMMU_SMMUV3 || !vms->iommu_phandle) {
1432         return;
1433     }
1434 
1435     dev = qdev_new(TYPE_ARM_SMMUV3);
1436 
1437     if (!vmc->no_nested_smmu) {
1438         object_property_set_str(OBJECT(dev), "stage", "nested", &error_fatal);
1439     }
1440     object_property_set_link(OBJECT(dev), "primary-bus", OBJECT(bus),
1441                              &error_abort);
1442     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1443     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, base);
1444     for (i = 0; i < NUM_SMMU_IRQS; i++) {
1445         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1446                            qdev_get_gpio_in(vms->gic, irq + i));
1447     }
1448 
1449     node = g_strdup_printf("/smmuv3@%" PRIx64, base);
1450     qemu_fdt_add_subnode(ms->fdt, node);
1451     qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1452     qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg", 2, base, 2, size);
1453 
1454     qemu_fdt_setprop_cells(ms->fdt, node, "interrupts",
1455             GIC_FDT_IRQ_TYPE_SPI, irq    , GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1456             GIC_FDT_IRQ_TYPE_SPI, irq + 1, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1457             GIC_FDT_IRQ_TYPE_SPI, irq + 2, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1458             GIC_FDT_IRQ_TYPE_SPI, irq + 3, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1459 
1460     qemu_fdt_setprop(ms->fdt, node, "interrupt-names", irq_names,
1461                      sizeof(irq_names));
1462 
1463     qemu_fdt_setprop(ms->fdt, node, "dma-coherent", NULL, 0);
1464 
1465     qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1466 
1467     qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1468     g_free(node);
1469 }
1470 
1471 static void create_virtio_iommu_dt_bindings(VirtMachineState *vms)
1472 {
1473     const char compat[] = "virtio,pci-iommu\0pci1af4,1057";
1474     uint16_t bdf = vms->virtio_iommu_bdf;
1475     MachineState *ms = MACHINE(vms);
1476     char *node;
1477 
1478     vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1479 
1480     node = g_strdup_printf("%s/virtio_iommu@%x,%x", vms->pciehb_nodename,
1481                            PCI_SLOT(bdf), PCI_FUNC(bdf));
1482     qemu_fdt_add_subnode(ms->fdt, node);
1483     qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1484     qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg",
1485                                  1, bdf << 8, 1, 0, 1, 0,
1486                                  1, 0, 1, 0);
1487 
1488     qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1489     qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1490     g_free(node);
1491 
1492     if (!vms->default_bus_bypass_iommu) {
1493         qemu_fdt_setprop_cells(ms->fdt, vms->pciehb_nodename, "iommu-map",
1494                                0x0, vms->iommu_phandle, 0x0, bdf,
1495                                bdf + 1, vms->iommu_phandle, bdf + 1,
1496                                0xffff - bdf);
1497     }
1498 }
1499 
1500 static void create_pcie(VirtMachineState *vms)
1501 {
1502     hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base;
1503     hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size;
1504     hwaddr base_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].base;
1505     hwaddr size_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].size;
1506     hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base;
1507     hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size;
1508     hwaddr base_ecam, size_ecam;
1509     hwaddr base = base_mmio;
1510     int nr_pcie_buses;
1511     int irq = vms->irqmap[VIRT_PCIE];
1512     MemoryRegion *mmio_alias;
1513     MemoryRegion *mmio_reg;
1514     MemoryRegion *ecam_alias;
1515     MemoryRegion *ecam_reg;
1516     DeviceState *dev;
1517     char *nodename;
1518     int i, ecam_id;
1519     PCIHostState *pci;
1520     MachineState *ms = MACHINE(vms);
1521     MachineClass *mc = MACHINE_GET_CLASS(ms);
1522 
1523     dev = qdev_new(TYPE_GPEX_HOST);
1524     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1525 
1526     ecam_id = VIRT_ECAM_ID(vms->highmem_ecam);
1527     base_ecam = vms->memmap[ecam_id].base;
1528     size_ecam = vms->memmap[ecam_id].size;
1529     nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN;
1530     /* Map only the first size_ecam bytes of ECAM space */
1531     ecam_alias = g_new0(MemoryRegion, 1);
1532     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1533     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1534                              ecam_reg, 0, size_ecam);
1535     memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias);
1536 
1537     /* Map the MMIO window into system address space so as to expose
1538      * the section of PCI MMIO space which starts at the same base address
1539      * (ie 1:1 mapping for that part of PCI MMIO space visible through
1540      * the window).
1541      */
1542     mmio_alias = g_new0(MemoryRegion, 1);
1543     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1544     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1545                              mmio_reg, base_mmio, size_mmio);
1546     memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias);
1547 
1548     if (vms->highmem_mmio) {
1549         /* Map high MMIO space */
1550         MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1);
1551 
1552         memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1553                                  mmio_reg, base_mmio_high, size_mmio_high);
1554         memory_region_add_subregion(get_system_memory(), base_mmio_high,
1555                                     high_mmio_alias);
1556     }
1557 
1558     /* Map IO port space */
1559     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio);
1560 
1561     for (i = 0; i < PCI_NUM_PINS; i++) {
1562         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1563                            qdev_get_gpio_in(vms->gic, irq + i));
1564         gpex_set_irq_num(GPEX_HOST(dev), i, irq + i);
1565     }
1566 
1567     pci = PCI_HOST_BRIDGE(dev);
1568     pci->bypass_iommu = vms->default_bus_bypass_iommu;
1569     vms->bus = pci->bus;
1570     if (vms->bus) {
1571         pci_init_nic_devices(pci->bus, mc->default_nic);
1572     }
1573 
1574     nodename = vms->pciehb_nodename = g_strdup_printf("/pcie@%" PRIx64, base);
1575     qemu_fdt_add_subnode(ms->fdt, nodename);
1576     qemu_fdt_setprop_string(ms->fdt, nodename,
1577                             "compatible", "pci-host-ecam-generic");
1578     qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "pci");
1579     qemu_fdt_setprop_cell(ms->fdt, nodename, "#address-cells", 3);
1580     qemu_fdt_setprop_cell(ms->fdt, nodename, "#size-cells", 2);
1581     qemu_fdt_setprop_cell(ms->fdt, nodename, "linux,pci-domain", 0);
1582     qemu_fdt_setprop_cells(ms->fdt, nodename, "bus-range", 0,
1583                            nr_pcie_buses - 1);
1584     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1585 
1586     if (vms->msi_phandle) {
1587         qemu_fdt_setprop_cells(ms->fdt, nodename, "msi-map",
1588                                0, vms->msi_phandle, 0, 0x10000);
1589     }
1590 
1591     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1592                                  2, base_ecam, 2, size_ecam);
1593 
1594     if (vms->highmem_mmio) {
1595         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1596                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1597                                      2, base_pio, 2, size_pio,
1598                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1599                                      2, base_mmio, 2, size_mmio,
1600                                      1, FDT_PCI_RANGE_MMIO_64BIT,
1601                                      2, base_mmio_high,
1602                                      2, base_mmio_high, 2, size_mmio_high);
1603     } else {
1604         qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1605                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1606                                      2, base_pio, 2, size_pio,
1607                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1608                                      2, base_mmio, 2, size_mmio);
1609     }
1610 
1611     qemu_fdt_setprop_cell(ms->fdt, nodename, "#interrupt-cells", 1);
1612     create_pcie_irq_map(ms, vms->gic_phandle, irq, nodename);
1613 
1614     if (vms->iommu) {
1615         vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1616 
1617         switch (vms->iommu) {
1618         case VIRT_IOMMU_SMMUV3:
1619             create_smmu(vms, vms->bus);
1620             if (!vms->default_bus_bypass_iommu) {
1621                 qemu_fdt_setprop_cells(ms->fdt, nodename, "iommu-map",
1622                                        0x0, vms->iommu_phandle, 0x0, 0x10000);
1623             }
1624             break;
1625         default:
1626             g_assert_not_reached();
1627         }
1628     }
1629 }
1630 
1631 static void create_cxl_host_reg_region(VirtMachineState *vms)
1632 {
1633     MemoryRegion *sysmem = get_system_memory();
1634     MemoryRegion *mr = &vms->cxl_devices_state.host_mr;
1635 
1636     memory_region_init(mr, OBJECT(vms), "cxl_host_reg",
1637                        vms->memmap[VIRT_CXL_HOST].size);
1638     memory_region_add_subregion(sysmem, vms->memmap[VIRT_CXL_HOST].base, mr);
1639     vms->highmem_cxl = true;
1640 }
1641 
1642 static void create_platform_bus(VirtMachineState *vms)
1643 {
1644     DeviceState *dev;
1645     SysBusDevice *s;
1646     int i;
1647     MemoryRegion *sysmem = get_system_memory();
1648 
1649     dev = qdev_new(TYPE_PLATFORM_BUS_DEVICE);
1650     dev->id = g_strdup(TYPE_PLATFORM_BUS_DEVICE);
1651     qdev_prop_set_uint32(dev, "num_irqs", PLATFORM_BUS_NUM_IRQS);
1652     qdev_prop_set_uint32(dev, "mmio_size", vms->memmap[VIRT_PLATFORM_BUS].size);
1653     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1654     vms->platform_bus_dev = dev;
1655 
1656     s = SYS_BUS_DEVICE(dev);
1657     for (i = 0; i < PLATFORM_BUS_NUM_IRQS; i++) {
1658         int irq = vms->irqmap[VIRT_PLATFORM_BUS] + i;
1659         sysbus_connect_irq(s, i, qdev_get_gpio_in(vms->gic, irq));
1660     }
1661 
1662     memory_region_add_subregion(sysmem,
1663                                 vms->memmap[VIRT_PLATFORM_BUS].base,
1664                                 sysbus_mmio_get_region(s, 0));
1665 }
1666 
1667 static void create_tag_ram(MemoryRegion *tag_sysmem,
1668                            hwaddr base, hwaddr size,
1669                            const char *name)
1670 {
1671     MemoryRegion *tagram = g_new(MemoryRegion, 1);
1672 
1673     memory_region_init_ram(tagram, NULL, name, size / 32, &error_fatal);
1674     memory_region_add_subregion(tag_sysmem, base / 32, tagram);
1675 }
1676 
1677 static void create_secure_ram(VirtMachineState *vms,
1678                               MemoryRegion *secure_sysmem,
1679                               MemoryRegion *secure_tag_sysmem)
1680 {
1681     MemoryRegion *secram = g_new(MemoryRegion, 1);
1682     char *nodename;
1683     hwaddr base = vms->memmap[VIRT_SECURE_MEM].base;
1684     hwaddr size = vms->memmap[VIRT_SECURE_MEM].size;
1685     MachineState *ms = MACHINE(vms);
1686 
1687     memory_region_init_ram(secram, NULL, "virt.secure-ram", size,
1688                            &error_fatal);
1689     memory_region_add_subregion(secure_sysmem, base, secram);
1690 
1691     nodename = g_strdup_printf("/secram@%" PRIx64, base);
1692     qemu_fdt_add_subnode(ms->fdt, nodename);
1693     qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "memory");
1694     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg", 2, base, 2, size);
1695     qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1696     qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1697 
1698     if (secure_tag_sysmem) {
1699         create_tag_ram(secure_tag_sysmem, base, size, "mach-virt.secure-tag");
1700     }
1701 
1702     g_free(nodename);
1703 }
1704 
1705 static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size)
1706 {
1707     const VirtMachineState *board = container_of(binfo, VirtMachineState,
1708                                                  bootinfo);
1709     MachineState *ms = MACHINE(board);
1710 
1711 
1712     *fdt_size = board->fdt_size;
1713     return ms->fdt;
1714 }
1715 
1716 static void virt_build_smbios(VirtMachineState *vms)
1717 {
1718     MachineClass *mc = MACHINE_GET_CLASS(vms);
1719     MachineState *ms = MACHINE(vms);
1720     uint8_t *smbios_tables, *smbios_anchor;
1721     size_t smbios_tables_len, smbios_anchor_len;
1722     struct smbios_phys_mem_area mem_array;
1723     const char *product = "QEMU Virtual Machine";
1724 
1725     if (kvm_enabled()) {
1726         product = "KVM Virtual Machine";
1727     }
1728 
1729     smbios_set_defaults("QEMU", product, mc->name);
1730 
1731     /* build the array of physical mem area from base_memmap */
1732     mem_array.address = vms->memmap[VIRT_MEM].base;
1733     mem_array.length = ms->ram_size;
1734 
1735     smbios_get_tables(ms, SMBIOS_ENTRY_POINT_TYPE_64, &mem_array, 1,
1736                       &smbios_tables, &smbios_tables_len,
1737                       &smbios_anchor, &smbios_anchor_len,
1738                       &error_fatal);
1739 
1740     if (smbios_anchor) {
1741         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables",
1742                         smbios_tables, smbios_tables_len);
1743         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor",
1744                         smbios_anchor, smbios_anchor_len);
1745     }
1746 }
1747 
1748 static
1749 void virt_machine_done(Notifier *notifier, void *data)
1750 {
1751     VirtMachineState *vms = container_of(notifier, VirtMachineState,
1752                                          machine_done);
1753     MachineState *ms = MACHINE(vms);
1754     ARMCPU *cpu = ARM_CPU(first_cpu);
1755     struct arm_boot_info *info = &vms->bootinfo;
1756     AddressSpace *as = arm_boot_address_space(cpu, info);
1757 
1758     cxl_hook_up_pxb_registers(vms->bus, &vms->cxl_devices_state,
1759                               &error_fatal);
1760 
1761     if (vms->cxl_devices_state.is_enabled) {
1762         cxl_fmws_link_targets(&error_fatal);
1763     }
1764     /*
1765      * If the user provided a dtb, we assume the dynamic sysbus nodes
1766      * already are integrated there. This corresponds to a use case where
1767      * the dynamic sysbus nodes are complex and their generation is not yet
1768      * supported. In that case the user can take charge of the guest dt
1769      * while qemu takes charge of the qom stuff.
1770      */
1771     if (info->dtb_filename == NULL) {
1772         platform_bus_add_all_fdt_nodes(ms->fdt, "/intc",
1773                                        vms->memmap[VIRT_PLATFORM_BUS].base,
1774                                        vms->memmap[VIRT_PLATFORM_BUS].size,
1775                                        vms->irqmap[VIRT_PLATFORM_BUS]);
1776     }
1777     if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms, cpu) < 0) {
1778         exit(1);
1779     }
1780 
1781     pci_bus_add_fw_cfg_extra_pci_roots(vms->fw_cfg, vms->bus,
1782                                        &error_abort);
1783 
1784     virt_acpi_setup(vms);
1785     virt_build_smbios(vms);
1786 }
1787 
1788 static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx)
1789 {
1790     uint8_t clustersz;
1791 
1792     /*
1793      * Adjust MPIDR to make TCG consistent (with 64-bit KVM hosts)
1794      * and to improve SGI efficiency.
1795      */
1796     if (vms->gic_version == VIRT_GIC_VERSION_2) {
1797         clustersz = GIC_TARGETLIST_BITS;
1798     } else {
1799         clustersz = GICV3_TARGETLIST_BITS;
1800     }
1801 
1802     return arm_build_mp_affinity(idx, clustersz);
1803 }
1804 
1805 static inline bool *virt_get_high_memmap_enabled(VirtMachineState *vms,
1806                                                  int index)
1807 {
1808     bool *enabled_array[] = {
1809         &vms->highmem_redists,
1810         &vms->highmem_cxl,
1811         &vms->highmem_ecam,
1812         &vms->highmem_mmio,
1813     };
1814 
1815     assert(ARRAY_SIZE(extended_memmap) - VIRT_LOWMEMMAP_LAST ==
1816            ARRAY_SIZE(enabled_array));
1817     assert(index - VIRT_LOWMEMMAP_LAST < ARRAY_SIZE(enabled_array));
1818 
1819     return enabled_array[index - VIRT_LOWMEMMAP_LAST];
1820 }
1821 
1822 static void virt_set_high_memmap(VirtMachineState *vms,
1823                                  hwaddr base, int pa_bits)
1824 {
1825     hwaddr region_base, region_size;
1826     bool *region_enabled, fits;
1827     int i;
1828 
1829     for (i = VIRT_LOWMEMMAP_LAST; i < ARRAY_SIZE(extended_memmap); i++) {
1830         region_enabled = virt_get_high_memmap_enabled(vms, i);
1831         region_base = ROUND_UP(base, extended_memmap[i].size);
1832         region_size = extended_memmap[i].size;
1833 
1834         vms->memmap[i].base = region_base;
1835         vms->memmap[i].size = region_size;
1836 
1837         /*
1838          * Check each device to see if it fits in the PA space,
1839          * moving highest_gpa as we go. For compatibility, move
1840          * highest_gpa for disabled fitting devices as well, if
1841          * the compact layout has been disabled.
1842          *
1843          * For each device that doesn't fit, disable it.
1844          */
1845         fits = (region_base + region_size) <= BIT_ULL(pa_bits);
1846         *region_enabled &= fits;
1847         if (vms->highmem_compact && !*region_enabled) {
1848             continue;
1849         }
1850 
1851         base = region_base + region_size;
1852         if (fits) {
1853             vms->highest_gpa = base - 1;
1854         }
1855     }
1856 }
1857 
1858 static void virt_set_memmap(VirtMachineState *vms, int pa_bits)
1859 {
1860     MachineState *ms = MACHINE(vms);
1861     hwaddr base, device_memory_base, device_memory_size, memtop;
1862     int i;
1863 
1864     vms->memmap = extended_memmap;
1865 
1866     for (i = 0; i < ARRAY_SIZE(base_memmap); i++) {
1867         vms->memmap[i] = base_memmap[i];
1868     }
1869 
1870     if (ms->ram_slots > ACPI_MAX_RAM_SLOTS) {
1871         error_report("unsupported number of memory slots: %"PRIu64,
1872                      ms->ram_slots);
1873         exit(EXIT_FAILURE);
1874     }
1875 
1876     /*
1877      * !highmem is exactly the same as limiting the PA space to 32bit,
1878      * irrespective of the underlying capabilities of the HW.
1879      */
1880     if (!vms->highmem) {
1881         pa_bits = 32;
1882     }
1883 
1884     /*
1885      * We compute the base of the high IO region depending on the
1886      * amount of initial and device memory. The device memory start/size
1887      * is aligned on 1GiB. We never put the high IO region below 256GiB
1888      * so that if maxram_size is < 255GiB we keep the legacy memory map.
1889      * The device region size assumes 1GiB page max alignment per slot.
1890      */
1891     device_memory_base =
1892         ROUND_UP(vms->memmap[VIRT_MEM].base + ms->ram_size, GiB);
1893     device_memory_size = ms->maxram_size - ms->ram_size + ms->ram_slots * GiB;
1894 
1895     /* Base address of the high IO region */
1896     memtop = base = device_memory_base + ROUND_UP(device_memory_size, GiB);
1897     if (memtop > BIT_ULL(pa_bits)) {
1898         error_report("Addressing limited to %d bits, but memory exceeds it by %llu bytes",
1899                      pa_bits, memtop - BIT_ULL(pa_bits));
1900         exit(EXIT_FAILURE);
1901     }
1902     if (base < device_memory_base) {
1903         error_report("maxmem/slots too huge");
1904         exit(EXIT_FAILURE);
1905     }
1906     if (base < vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES) {
1907         base = vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES;
1908     }
1909 
1910     /* We know for sure that at least the memory fits in the PA space */
1911     vms->highest_gpa = memtop - 1;
1912 
1913     virt_set_high_memmap(vms, base, pa_bits);
1914 
1915     if (device_memory_size > 0) {
1916         machine_memory_devices_init(ms, device_memory_base, device_memory_size);
1917     }
1918     vms->highest_gpa = cxl_fmws_set_memmap(ROUND_UP(vms->highest_gpa + 1,
1919                                                     256 * MiB),
1920                                            BIT_ULL(pa_bits)) - 1;
1921 }
1922 
1923 static VirtGICType finalize_gic_version_do(const char *accel_name,
1924                                            VirtGICType gic_version,
1925                                            int gics_supported,
1926                                            unsigned int max_cpus)
1927 {
1928     /* Convert host/max/nosel to GIC version number */
1929     switch (gic_version) {
1930     case VIRT_GIC_VERSION_HOST:
1931         if (!kvm_enabled()) {
1932             error_report("gic-version=host requires KVM");
1933             exit(1);
1934         }
1935 
1936         /* For KVM, gic-version=host means gic-version=max */
1937         return finalize_gic_version_do(accel_name, VIRT_GIC_VERSION_MAX,
1938                                        gics_supported, max_cpus);
1939     case VIRT_GIC_VERSION_MAX:
1940         if (gics_supported & VIRT_GIC_VERSION_4_MASK) {
1941             gic_version = VIRT_GIC_VERSION_4;
1942         } else if (gics_supported & VIRT_GIC_VERSION_3_MASK) {
1943             gic_version = VIRT_GIC_VERSION_3;
1944         } else {
1945             gic_version = VIRT_GIC_VERSION_2;
1946         }
1947         break;
1948     case VIRT_GIC_VERSION_NOSEL:
1949         if ((gics_supported & VIRT_GIC_VERSION_2_MASK) &&
1950             max_cpus <= GIC_NCPU) {
1951             gic_version = VIRT_GIC_VERSION_2;
1952         } else if (gics_supported & VIRT_GIC_VERSION_3_MASK) {
1953             /*
1954              * in case the host does not support v2 emulation or
1955              * the end-user requested more than 8 VCPUs we now default
1956              * to v3. In any case defaulting to v2 would be broken.
1957              */
1958             gic_version = VIRT_GIC_VERSION_3;
1959         } else if (max_cpus > GIC_NCPU) {
1960             error_report("%s only supports GICv2 emulation but more than 8 "
1961                          "vcpus are requested", accel_name);
1962             exit(1);
1963         }
1964         break;
1965     case VIRT_GIC_VERSION_2:
1966     case VIRT_GIC_VERSION_3:
1967     case VIRT_GIC_VERSION_4:
1968         break;
1969     }
1970 
1971     /* Check chosen version is effectively supported */
1972     switch (gic_version) {
1973     case VIRT_GIC_VERSION_2:
1974         if (!(gics_supported & VIRT_GIC_VERSION_2_MASK)) {
1975             error_report("%s does not support GICv2 emulation", accel_name);
1976             exit(1);
1977         }
1978         break;
1979     case VIRT_GIC_VERSION_3:
1980         if (!(gics_supported & VIRT_GIC_VERSION_3_MASK)) {
1981             error_report("%s does not support GICv3 emulation", accel_name);
1982             exit(1);
1983         }
1984         break;
1985     case VIRT_GIC_VERSION_4:
1986         if (!(gics_supported & VIRT_GIC_VERSION_4_MASK)) {
1987             error_report("%s does not support GICv4 emulation, is virtualization=on?",
1988                          accel_name);
1989             exit(1);
1990         }
1991         break;
1992     default:
1993         error_report("logic error in finalize_gic_version");
1994         exit(1);
1995         break;
1996     }
1997 
1998     return gic_version;
1999 }
2000 
2001 /*
2002  * finalize_gic_version - Determines the final gic_version
2003  * according to the gic-version property
2004  *
2005  * Default GIC type is v2
2006  */
2007 static void finalize_gic_version(VirtMachineState *vms)
2008 {
2009     const char *accel_name = current_accel_name();
2010     unsigned int max_cpus = MACHINE(vms)->smp.max_cpus;
2011     int gics_supported = 0;
2012 
2013     /* Determine which GIC versions the current environment supports */
2014     if (kvm_enabled() && kvm_irqchip_in_kernel()) {
2015         int probe_bitmap = kvm_arm_vgic_probe();
2016 
2017         if (!probe_bitmap) {
2018             error_report("Unable to determine GIC version supported by host");
2019             exit(1);
2020         }
2021 
2022         if (probe_bitmap & KVM_ARM_VGIC_V2) {
2023             gics_supported |= VIRT_GIC_VERSION_2_MASK;
2024         }
2025         if (probe_bitmap & KVM_ARM_VGIC_V3) {
2026             gics_supported |= VIRT_GIC_VERSION_3_MASK;
2027         }
2028     } else if (kvm_enabled() && !kvm_irqchip_in_kernel()) {
2029         /* KVM w/o kernel irqchip can only deal with GICv2 */
2030         gics_supported |= VIRT_GIC_VERSION_2_MASK;
2031         accel_name = "KVM with kernel-irqchip=off";
2032     } else if (tcg_enabled() || hvf_enabled() || qtest_enabled())  {
2033         gics_supported |= VIRT_GIC_VERSION_2_MASK;
2034         if (module_object_class_by_name("arm-gicv3")) {
2035             gics_supported |= VIRT_GIC_VERSION_3_MASK;
2036             if (vms->virt) {
2037                 /* GICv4 only makes sense if CPU has EL2 */
2038                 gics_supported |= VIRT_GIC_VERSION_4_MASK;
2039             }
2040         }
2041     } else {
2042         error_report("Unsupported accelerator, can not determine GIC support");
2043         exit(1);
2044     }
2045 
2046     /*
2047      * Then convert helpers like host/max to concrete GIC versions and ensure
2048      * the desired version is supported
2049      */
2050     vms->gic_version = finalize_gic_version_do(accel_name, vms->gic_version,
2051                                                gics_supported, max_cpus);
2052 }
2053 
2054 /*
2055  * virt_post_cpus_gic_realized() must be called after the CPUs and
2056  * the GIC have both been realized.
2057  */
2058 static void virt_post_cpus_gic_realized(VirtMachineState *vms,
2059                                         MemoryRegion *sysmem)
2060 {
2061     int max_cpus = MACHINE(vms)->smp.max_cpus;
2062     bool aarch64, pmu, steal_time;
2063     CPUState *cpu;
2064 
2065     aarch64 = object_property_get_bool(OBJECT(first_cpu), "aarch64", NULL);
2066     pmu = object_property_get_bool(OBJECT(first_cpu), "pmu", NULL);
2067     steal_time = object_property_get_bool(OBJECT(first_cpu),
2068                                           "kvm-steal-time", NULL);
2069 
2070     if (kvm_enabled()) {
2071         hwaddr pvtime_reg_base = vms->memmap[VIRT_PVTIME].base;
2072         hwaddr pvtime_reg_size = vms->memmap[VIRT_PVTIME].size;
2073 
2074         if (steal_time) {
2075             MemoryRegion *pvtime = g_new(MemoryRegion, 1);
2076             hwaddr pvtime_size = max_cpus * PVTIME_SIZE_PER_CPU;
2077 
2078             /* The memory region size must be a multiple of host page size. */
2079             pvtime_size = REAL_HOST_PAGE_ALIGN(pvtime_size);
2080 
2081             if (pvtime_size > pvtime_reg_size) {
2082                 error_report("pvtime requires a %" HWADDR_PRId
2083                              " byte memory region for %d CPUs,"
2084                              " but only %" HWADDR_PRId " has been reserved",
2085                              pvtime_size, max_cpus, pvtime_reg_size);
2086                 exit(1);
2087             }
2088 
2089             memory_region_init_ram(pvtime, NULL, "pvtime", pvtime_size, NULL);
2090             memory_region_add_subregion(sysmem, pvtime_reg_base, pvtime);
2091         }
2092 
2093         CPU_FOREACH(cpu) {
2094             if (pmu) {
2095                 assert(arm_feature(&ARM_CPU(cpu)->env, ARM_FEATURE_PMU));
2096                 if (kvm_irqchip_in_kernel()) {
2097                     kvm_arm_pmu_set_irq(ARM_CPU(cpu), VIRTUAL_PMU_IRQ);
2098                 }
2099                 kvm_arm_pmu_init(ARM_CPU(cpu));
2100             }
2101             if (steal_time) {
2102                 kvm_arm_pvtime_init(ARM_CPU(cpu), pvtime_reg_base
2103                                                   + cpu->cpu_index
2104                                                     * PVTIME_SIZE_PER_CPU);
2105             }
2106         }
2107     } else {
2108         if (aarch64 && vms->highmem) {
2109             int requested_pa_size = 64 - clz64(vms->highest_gpa);
2110             int pamax = arm_pamax(ARM_CPU(first_cpu));
2111 
2112             if (pamax < requested_pa_size) {
2113                 error_report("VCPU supports less PA bits (%d) than "
2114                              "requested by the memory map (%d)",
2115                              pamax, requested_pa_size);
2116                 exit(1);
2117             }
2118         }
2119     }
2120 }
2121 
2122 static void machvirt_init(MachineState *machine)
2123 {
2124     VirtMachineState *vms = VIRT_MACHINE(machine);
2125     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine);
2126     MachineClass *mc = MACHINE_GET_CLASS(machine);
2127     const CPUArchIdList *possible_cpus;
2128     MemoryRegion *sysmem = get_system_memory();
2129     MemoryRegion *secure_sysmem = NULL;
2130     MemoryRegion *tag_sysmem = NULL;
2131     MemoryRegion *secure_tag_sysmem = NULL;
2132     int n, virt_max_cpus;
2133     bool firmware_loaded;
2134     bool aarch64 = true;
2135     bool has_ged = !vmc->no_ged;
2136     unsigned int smp_cpus = machine->smp.cpus;
2137     unsigned int max_cpus = machine->smp.max_cpus;
2138 
2139     possible_cpus = mc->possible_cpu_arch_ids(machine);
2140 
2141     /*
2142      * In accelerated mode, the memory map is computed earlier in kvm_type()
2143      * for Linux, or hvf_get_physical_address_range() for macOS to create a
2144      * VM with the right number of IPA bits.
2145      */
2146     if (!vms->memmap) {
2147         Object *cpuobj;
2148         ARMCPU *armcpu;
2149         int pa_bits;
2150 
2151         /*
2152          * Instantiate a temporary CPU object to find out about what
2153          * we are about to deal with. Once this is done, get rid of
2154          * the object.
2155          */
2156         cpuobj = object_new(possible_cpus->cpus[0].type);
2157         armcpu = ARM_CPU(cpuobj);
2158 
2159         pa_bits = arm_pamax(armcpu);
2160 
2161         object_unref(cpuobj);
2162 
2163         virt_set_memmap(vms, pa_bits);
2164     }
2165 
2166     /* We can probe only here because during property set
2167      * KVM is not available yet
2168      */
2169     finalize_gic_version(vms);
2170 
2171     if (vms->secure) {
2172         /*
2173          * The Secure view of the world is the same as the NonSecure,
2174          * but with a few extra devices. Create it as a container region
2175          * containing the system memory at low priority; any secure-only
2176          * devices go in at higher priority and take precedence.
2177          */
2178         secure_sysmem = g_new(MemoryRegion, 1);
2179         memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory",
2180                            UINT64_MAX);
2181         memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1);
2182     }
2183 
2184     firmware_loaded = virt_firmware_init(vms, sysmem,
2185                                          secure_sysmem ?: sysmem);
2186 
2187     /* If we have an EL3 boot ROM then the assumption is that it will
2188      * implement PSCI itself, so disable QEMU's internal implementation
2189      * so it doesn't get in the way. Instead of starting secondary
2190      * CPUs in PSCI powerdown state we will start them all running and
2191      * let the boot ROM sort them out.
2192      * The usual case is that we do use QEMU's PSCI implementation;
2193      * if the guest has EL2 then we will use SMC as the conduit,
2194      * and otherwise we will use HVC (for backwards compatibility and
2195      * because if we're using KVM then we must use HVC).
2196      */
2197     if (vms->secure && firmware_loaded) {
2198         vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
2199     } else if (vms->virt) {
2200         vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC;
2201     } else {
2202         vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC;
2203     }
2204 
2205     /*
2206      * The maximum number of CPUs depends on the GIC version, or on how
2207      * many redistributors we can fit into the memory map (which in turn
2208      * depends on whether this is a GICv3 or v4).
2209      */
2210     if (vms->gic_version == VIRT_GIC_VERSION_2) {
2211         virt_max_cpus = GIC_NCPU;
2212     } else {
2213         virt_max_cpus = virt_redist_capacity(vms, VIRT_GIC_REDIST);
2214         if (vms->highmem_redists) {
2215             virt_max_cpus += virt_redist_capacity(vms, VIRT_HIGH_GIC_REDIST2);
2216         }
2217     }
2218 
2219     if (max_cpus > virt_max_cpus) {
2220         error_report("Number of SMP CPUs requested (%d) exceeds max CPUs "
2221                      "supported by machine 'mach-virt' (%d)",
2222                      max_cpus, virt_max_cpus);
2223         if (vms->gic_version != VIRT_GIC_VERSION_2 && !vms->highmem_redists) {
2224             error_printf("Try 'highmem-redists=on' for more CPUs\n");
2225         }
2226 
2227         exit(1);
2228     }
2229 
2230     if (vms->secure && !tcg_enabled() && !qtest_enabled()) {
2231         error_report("mach-virt: %s does not support providing "
2232                      "Security extensions (TrustZone) to the guest CPU",
2233                      current_accel_name());
2234         exit(1);
2235     }
2236 
2237     if (vms->virt && !tcg_enabled() && !qtest_enabled()) {
2238         error_report("mach-virt: %s does not support providing "
2239                      "Virtualization extensions to the guest CPU",
2240                      current_accel_name());
2241         exit(1);
2242     }
2243 
2244     if (vms->mte && hvf_enabled()) {
2245         error_report("mach-virt: %s does not support providing "
2246                      "MTE to the guest CPU",
2247                      current_accel_name());
2248         exit(1);
2249     }
2250 
2251     create_fdt(vms);
2252 
2253     assert(possible_cpus->len == max_cpus);
2254     for (n = 0; n < possible_cpus->len; n++) {
2255         Object *cpuobj;
2256         CPUState *cs;
2257 
2258         if (n >= smp_cpus) {
2259             break;
2260         }
2261 
2262         cpuobj = object_new(possible_cpus->cpus[n].type);
2263         object_property_set_int(cpuobj, "mp-affinity",
2264                                 possible_cpus->cpus[n].arch_id, NULL);
2265 
2266         cs = CPU(cpuobj);
2267         cs->cpu_index = n;
2268 
2269         numa_cpu_pre_plug(&possible_cpus->cpus[cs->cpu_index], DEVICE(cpuobj),
2270                           &error_fatal);
2271 
2272         aarch64 &= object_property_get_bool(cpuobj, "aarch64", NULL);
2273 
2274         if (!vms->secure) {
2275             object_property_set_bool(cpuobj, "has_el3", false, NULL);
2276         }
2277 
2278         if (!vms->virt && object_property_find(cpuobj, "has_el2")) {
2279             object_property_set_bool(cpuobj, "has_el2", false, NULL);
2280         }
2281 
2282         if (vmc->kvm_no_adjvtime &&
2283             object_property_find(cpuobj, "kvm-no-adjvtime")) {
2284             object_property_set_bool(cpuobj, "kvm-no-adjvtime", true, NULL);
2285         }
2286 
2287         if (vmc->no_kvm_steal_time &&
2288             object_property_find(cpuobj, "kvm-steal-time")) {
2289             object_property_set_bool(cpuobj, "kvm-steal-time", false, NULL);
2290         }
2291 
2292         if (vmc->no_tcg_lpa2 && object_property_find(cpuobj, "lpa2")) {
2293             object_property_set_bool(cpuobj, "lpa2", false, NULL);
2294         }
2295 
2296         if (object_property_find(cpuobj, "reset-cbar")) {
2297             object_property_set_int(cpuobj, "reset-cbar",
2298                                     vms->memmap[VIRT_CPUPERIPHS].base,
2299                                     &error_abort);
2300         }
2301 
2302         object_property_set_link(cpuobj, "memory", OBJECT(sysmem),
2303                                  &error_abort);
2304         if (vms->secure) {
2305             object_property_set_link(cpuobj, "secure-memory",
2306                                      OBJECT(secure_sysmem), &error_abort);
2307         }
2308 
2309         if (vms->mte) {
2310             if (tcg_enabled()) {
2311                 /* Create the memory region only once, but link to all cpus. */
2312                 if (!tag_sysmem) {
2313                     /*
2314                      * The property exists only if MemTag is supported.
2315                      * If it is, we must allocate the ram to back that up.
2316                      */
2317                     if (!object_property_find(cpuobj, "tag-memory")) {
2318                         error_report("MTE requested, but not supported "
2319                                      "by the guest CPU");
2320                         exit(1);
2321                     }
2322 
2323                     tag_sysmem = g_new(MemoryRegion, 1);
2324                     memory_region_init(tag_sysmem, OBJECT(machine),
2325                                        "tag-memory", UINT64_MAX / 32);
2326 
2327                     if (vms->secure) {
2328                         secure_tag_sysmem = g_new(MemoryRegion, 1);
2329                         memory_region_init(secure_tag_sysmem, OBJECT(machine),
2330                                            "secure-tag-memory",
2331                                            UINT64_MAX / 32);
2332 
2333                         /* As with ram, secure-tag takes precedence over tag. */
2334                         memory_region_add_subregion_overlap(secure_tag_sysmem,
2335                                                             0, tag_sysmem, -1);
2336                     }
2337                 }
2338 
2339                 object_property_set_link(cpuobj, "tag-memory",
2340                                          OBJECT(tag_sysmem), &error_abort);
2341                 if (vms->secure) {
2342                     object_property_set_link(cpuobj, "secure-tag-memory",
2343                                              OBJECT(secure_tag_sysmem),
2344                                              &error_abort);
2345                 }
2346             } else if (kvm_enabled()) {
2347                 if (!kvm_arm_mte_supported()) {
2348                     error_report("MTE requested, but not supported by KVM");
2349                     exit(1);
2350                 }
2351                 kvm_arm_enable_mte(cpuobj, &error_abort);
2352             } else {
2353                     error_report("MTE requested, but not supported ");
2354                     exit(1);
2355             }
2356         }
2357 
2358         qdev_realize(DEVICE(cpuobj), NULL, &error_fatal);
2359         object_unref(cpuobj);
2360     }
2361 
2362     /* Now we've created the CPUs we can see if they have the hypvirt timer */
2363     vms->ns_el2_virt_timer_irq = ns_el2_virt_timer_present() &&
2364         !vmc->no_ns_el2_virt_timer_irq;
2365 
2366     fdt_add_timer_nodes(vms);
2367     fdt_add_cpu_nodes(vms);
2368 
2369     memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base,
2370                                 machine->ram);
2371 
2372     cxl_fmws_update_mmio();
2373 
2374     virt_flash_fdt(vms, sysmem, secure_sysmem ?: sysmem);
2375 
2376     create_gic(vms, sysmem);
2377 
2378     virt_post_cpus_gic_realized(vms, sysmem);
2379 
2380     fdt_add_pmu_nodes(vms);
2381 
2382     /*
2383      * The first UART always exists. If the security extensions are
2384      * enabled, the second UART also always exists. Otherwise, it only exists
2385      * if a backend is configured explicitly via '-serial <backend>'.
2386      * This avoids potentially breaking existing user setups that expect
2387      * only one NonSecure UART to be present (for instance, older EDK2
2388      * binaries).
2389      *
2390      * The nodes end up in the DTB in reverse order of creation, so we must
2391      * create UART0 last to ensure it appears as the first node in the DTB,
2392      * for compatibility with guest software that just iterates through the
2393      * DTB to find the first UART, as older versions of EDK2 do.
2394      * DTB readers that follow the spec, as Linux does, should honour the
2395      * aliases node information and /chosen/stdout-path regardless of
2396      * the order that nodes appear in the DTB.
2397      *
2398      * For similar back-compatibility reasons, if UART1 is the secure UART
2399      * we create it second (and so it appears first in the DTB), because
2400      * that's what QEMU has always done.
2401      */
2402     if (!vms->secure) {
2403         Chardev *serial1 = serial_hd(1);
2404 
2405         if (serial1) {
2406             vms->second_ns_uart_present = true;
2407             create_uart(vms, VIRT_UART1, sysmem, serial1, false);
2408         }
2409     }
2410     create_uart(vms, VIRT_UART0, sysmem, serial_hd(0), false);
2411     if (vms->secure) {
2412         create_uart(vms, VIRT_UART1, secure_sysmem, serial_hd(1), true);
2413     }
2414 
2415     if (vms->secure) {
2416         create_secure_ram(vms, secure_sysmem, secure_tag_sysmem);
2417     }
2418 
2419     if (tag_sysmem) {
2420         create_tag_ram(tag_sysmem, vms->memmap[VIRT_MEM].base,
2421                        machine->ram_size, "mach-virt.tag");
2422     }
2423 
2424     vms->highmem_ecam &= (!firmware_loaded || aarch64);
2425 
2426     create_rtc(vms);
2427 
2428     create_pcie(vms);
2429     create_cxl_host_reg_region(vms);
2430 
2431     if (has_ged && aarch64 && firmware_loaded && virt_is_acpi_enabled(vms)) {
2432         vms->acpi_dev = create_acpi_ged(vms);
2433     } else {
2434         create_gpio_devices(vms, VIRT_GPIO, sysmem);
2435     }
2436 
2437     if (vms->secure && !vmc->no_secure_gpio) {
2438         create_gpio_devices(vms, VIRT_SECURE_GPIO, secure_sysmem);
2439     }
2440 
2441      /* connect powerdown request */
2442      vms->powerdown_notifier.notify = virt_powerdown_req;
2443      qemu_register_powerdown_notifier(&vms->powerdown_notifier);
2444 
2445     /* Create mmio transports, so the user can create virtio backends
2446      * (which will be automatically plugged in to the transports). If
2447      * no backend is created the transport will just sit harmlessly idle.
2448      */
2449     create_virtio_devices(vms);
2450 
2451     vms->fw_cfg = create_fw_cfg(vms, &address_space_memory);
2452     rom_set_fw(vms->fw_cfg);
2453 
2454     create_platform_bus(vms);
2455 
2456     if (machine->nvdimms_state->is_enabled) {
2457         const struct AcpiGenericAddress arm_virt_nvdimm_acpi_dsmio = {
2458             .space_id = AML_AS_SYSTEM_MEMORY,
2459             .address = vms->memmap[VIRT_NVDIMM_ACPI].base,
2460             .bit_width = NVDIMM_ACPI_IO_LEN << 3
2461         };
2462 
2463         nvdimm_init_acpi_state(machine->nvdimms_state, sysmem,
2464                                arm_virt_nvdimm_acpi_dsmio,
2465                                vms->fw_cfg, OBJECT(vms));
2466     }
2467 
2468     vms->bootinfo.ram_size = machine->ram_size;
2469     vms->bootinfo.board_id = -1;
2470     vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base;
2471     vms->bootinfo.get_dtb = machvirt_dtb;
2472     vms->bootinfo.skip_dtb_autoload = true;
2473     vms->bootinfo.firmware_loaded = firmware_loaded;
2474     vms->bootinfo.psci_conduit = vms->psci_conduit;
2475     arm_load_kernel(ARM_CPU(first_cpu), machine, &vms->bootinfo);
2476 
2477     vms->machine_done.notify = virt_machine_done;
2478     qemu_add_machine_init_done_notifier(&vms->machine_done);
2479 }
2480 
2481 static bool virt_get_secure(Object *obj, Error **errp)
2482 {
2483     VirtMachineState *vms = VIRT_MACHINE(obj);
2484 
2485     return vms->secure;
2486 }
2487 
2488 static void virt_set_secure(Object *obj, bool value, Error **errp)
2489 {
2490     VirtMachineState *vms = VIRT_MACHINE(obj);
2491 
2492     vms->secure = value;
2493 }
2494 
2495 static bool virt_get_virt(Object *obj, Error **errp)
2496 {
2497     VirtMachineState *vms = VIRT_MACHINE(obj);
2498 
2499     return vms->virt;
2500 }
2501 
2502 static void virt_set_virt(Object *obj, bool value, Error **errp)
2503 {
2504     VirtMachineState *vms = VIRT_MACHINE(obj);
2505 
2506     vms->virt = value;
2507 }
2508 
2509 static bool virt_get_highmem(Object *obj, Error **errp)
2510 {
2511     VirtMachineState *vms = VIRT_MACHINE(obj);
2512 
2513     return vms->highmem;
2514 }
2515 
2516 static void virt_set_highmem(Object *obj, bool value, Error **errp)
2517 {
2518     VirtMachineState *vms = VIRT_MACHINE(obj);
2519 
2520     vms->highmem = value;
2521 }
2522 
2523 static bool virt_get_compact_highmem(Object *obj, Error **errp)
2524 {
2525     VirtMachineState *vms = VIRT_MACHINE(obj);
2526 
2527     return vms->highmem_compact;
2528 }
2529 
2530 static void virt_set_compact_highmem(Object *obj, bool value, Error **errp)
2531 {
2532     VirtMachineState *vms = VIRT_MACHINE(obj);
2533 
2534     vms->highmem_compact = value;
2535 }
2536 
2537 static bool virt_get_highmem_redists(Object *obj, Error **errp)
2538 {
2539     VirtMachineState *vms = VIRT_MACHINE(obj);
2540 
2541     return vms->highmem_redists;
2542 }
2543 
2544 static void virt_set_highmem_redists(Object *obj, bool value, Error **errp)
2545 {
2546     VirtMachineState *vms = VIRT_MACHINE(obj);
2547 
2548     vms->highmem_redists = value;
2549 }
2550 
2551 static bool virt_get_highmem_ecam(Object *obj, Error **errp)
2552 {
2553     VirtMachineState *vms = VIRT_MACHINE(obj);
2554 
2555     return vms->highmem_ecam;
2556 }
2557 
2558 static void virt_set_highmem_ecam(Object *obj, bool value, Error **errp)
2559 {
2560     VirtMachineState *vms = VIRT_MACHINE(obj);
2561 
2562     vms->highmem_ecam = value;
2563 }
2564 
2565 static bool virt_get_highmem_mmio(Object *obj, Error **errp)
2566 {
2567     VirtMachineState *vms = VIRT_MACHINE(obj);
2568 
2569     return vms->highmem_mmio;
2570 }
2571 
2572 static void virt_set_highmem_mmio(Object *obj, bool value, Error **errp)
2573 {
2574     VirtMachineState *vms = VIRT_MACHINE(obj);
2575 
2576     vms->highmem_mmio = value;
2577 }
2578 
2579 static void virt_get_highmem_mmio_size(Object *obj, Visitor *v,
2580                                        const char *name, void *opaque,
2581                                        Error **errp)
2582 {
2583     uint64_t size = extended_memmap[VIRT_HIGH_PCIE_MMIO].size;
2584 
2585     visit_type_size(v, name, &size, errp);
2586 }
2587 
2588 static void virt_set_highmem_mmio_size(Object *obj, Visitor *v,
2589                                        const char *name, void *opaque,
2590                                        Error **errp)
2591 {
2592     uint64_t size;
2593 
2594     if (!visit_type_size(v, name, &size, errp)) {
2595         return;
2596     }
2597 
2598     if (!is_power_of_2(size)) {
2599         error_setg(errp, "highmem-mmio-size is not a power of 2");
2600         return;
2601     }
2602 
2603     if (size < DEFAULT_HIGH_PCIE_MMIO_SIZE) {
2604         char *sz = size_to_str(DEFAULT_HIGH_PCIE_MMIO_SIZE);
2605         error_setg(errp, "highmem-mmio-size cannot be set to a lower value "
2606                          "than the default (%s)", sz);
2607         g_free(sz);
2608         return;
2609     }
2610 
2611     extended_memmap[VIRT_HIGH_PCIE_MMIO].size = size;
2612 }
2613 
2614 static bool virt_get_its(Object *obj, Error **errp)
2615 {
2616     VirtMachineState *vms = VIRT_MACHINE(obj);
2617 
2618     return vms->its;
2619 }
2620 
2621 static void virt_set_its(Object *obj, bool value, Error **errp)
2622 {
2623     VirtMachineState *vms = VIRT_MACHINE(obj);
2624 
2625     vms->its = value;
2626 }
2627 
2628 static bool virt_get_dtb_randomness(Object *obj, Error **errp)
2629 {
2630     VirtMachineState *vms = VIRT_MACHINE(obj);
2631 
2632     return vms->dtb_randomness;
2633 }
2634 
2635 static void virt_set_dtb_randomness(Object *obj, bool value, Error **errp)
2636 {
2637     VirtMachineState *vms = VIRT_MACHINE(obj);
2638 
2639     vms->dtb_randomness = value;
2640 }
2641 
2642 static char *virt_get_oem_id(Object *obj, Error **errp)
2643 {
2644     VirtMachineState *vms = VIRT_MACHINE(obj);
2645 
2646     return g_strdup(vms->oem_id);
2647 }
2648 
2649 static void virt_set_oem_id(Object *obj, const char *value, Error **errp)
2650 {
2651     VirtMachineState *vms = VIRT_MACHINE(obj);
2652     size_t len = strlen(value);
2653 
2654     if (len > 6) {
2655         error_setg(errp,
2656                    "User specified oem-id value is bigger than 6 bytes in size");
2657         return;
2658     }
2659 
2660     strncpy(vms->oem_id, value, 6);
2661 }
2662 
2663 static char *virt_get_oem_table_id(Object *obj, Error **errp)
2664 {
2665     VirtMachineState *vms = VIRT_MACHINE(obj);
2666 
2667     return g_strdup(vms->oem_table_id);
2668 }
2669 
2670 static void virt_set_oem_table_id(Object *obj, const char *value,
2671                                   Error **errp)
2672 {
2673     VirtMachineState *vms = VIRT_MACHINE(obj);
2674     size_t len = strlen(value);
2675 
2676     if (len > 8) {
2677         error_setg(errp,
2678                    "User specified oem-table-id value is bigger than 8 bytes in size");
2679         return;
2680     }
2681     strncpy(vms->oem_table_id, value, 8);
2682 }
2683 
2684 
2685 bool virt_is_acpi_enabled(VirtMachineState *vms)
2686 {
2687     if (vms->acpi == ON_OFF_AUTO_OFF) {
2688         return false;
2689     }
2690     return true;
2691 }
2692 
2693 static void virt_get_acpi(Object *obj, Visitor *v, const char *name,
2694                           void *opaque, Error **errp)
2695 {
2696     VirtMachineState *vms = VIRT_MACHINE(obj);
2697     OnOffAuto acpi = vms->acpi;
2698 
2699     visit_type_OnOffAuto(v, name, &acpi, errp);
2700 }
2701 
2702 static void virt_set_acpi(Object *obj, Visitor *v, const char *name,
2703                           void *opaque, Error **errp)
2704 {
2705     VirtMachineState *vms = VIRT_MACHINE(obj);
2706 
2707     visit_type_OnOffAuto(v, name, &vms->acpi, errp);
2708 }
2709 
2710 static bool virt_get_ras(Object *obj, Error **errp)
2711 {
2712     VirtMachineState *vms = VIRT_MACHINE(obj);
2713 
2714     return vms->ras;
2715 }
2716 
2717 static void virt_set_ras(Object *obj, bool value, Error **errp)
2718 {
2719     VirtMachineState *vms = VIRT_MACHINE(obj);
2720 
2721     vms->ras = value;
2722 }
2723 
2724 static bool virt_get_mte(Object *obj, Error **errp)
2725 {
2726     VirtMachineState *vms = VIRT_MACHINE(obj);
2727 
2728     return vms->mte;
2729 }
2730 
2731 static void virt_set_mte(Object *obj, bool value, Error **errp)
2732 {
2733     VirtMachineState *vms = VIRT_MACHINE(obj);
2734 
2735     vms->mte = value;
2736 }
2737 
2738 static char *virt_get_gic_version(Object *obj, Error **errp)
2739 {
2740     VirtMachineState *vms = VIRT_MACHINE(obj);
2741     const char *val;
2742 
2743     switch (vms->gic_version) {
2744     case VIRT_GIC_VERSION_4:
2745         val = "4";
2746         break;
2747     case VIRT_GIC_VERSION_3:
2748         val = "3";
2749         break;
2750     default:
2751         val = "2";
2752         break;
2753     }
2754     return g_strdup(val);
2755 }
2756 
2757 static void virt_set_gic_version(Object *obj, const char *value, Error **errp)
2758 {
2759     VirtMachineState *vms = VIRT_MACHINE(obj);
2760 
2761     if (!strcmp(value, "4")) {
2762         vms->gic_version = VIRT_GIC_VERSION_4;
2763     } else if (!strcmp(value, "3")) {
2764         vms->gic_version = VIRT_GIC_VERSION_3;
2765     } else if (!strcmp(value, "2")) {
2766         vms->gic_version = VIRT_GIC_VERSION_2;
2767     } else if (!strcmp(value, "host")) {
2768         vms->gic_version = VIRT_GIC_VERSION_HOST; /* Will probe later */
2769     } else if (!strcmp(value, "max")) {
2770         vms->gic_version = VIRT_GIC_VERSION_MAX; /* Will probe later */
2771     } else {
2772         error_setg(errp, "Invalid gic-version value");
2773         error_append_hint(errp, "Valid values are 3, 2, host, max.\n");
2774     }
2775 }
2776 
2777 static char *virt_get_iommu(Object *obj, Error **errp)
2778 {
2779     VirtMachineState *vms = VIRT_MACHINE(obj);
2780 
2781     switch (vms->iommu) {
2782     case VIRT_IOMMU_NONE:
2783         return g_strdup("none");
2784     case VIRT_IOMMU_SMMUV3:
2785         return g_strdup("smmuv3");
2786     default:
2787         g_assert_not_reached();
2788     }
2789 }
2790 
2791 static void virt_set_iommu(Object *obj, const char *value, Error **errp)
2792 {
2793     VirtMachineState *vms = VIRT_MACHINE(obj);
2794 
2795     if (!strcmp(value, "smmuv3")) {
2796         vms->iommu = VIRT_IOMMU_SMMUV3;
2797     } else if (!strcmp(value, "none")) {
2798         vms->iommu = VIRT_IOMMU_NONE;
2799     } else {
2800         error_setg(errp, "Invalid iommu value");
2801         error_append_hint(errp, "Valid values are none, smmuv3.\n");
2802     }
2803 }
2804 
2805 static bool virt_get_default_bus_bypass_iommu(Object *obj, Error **errp)
2806 {
2807     VirtMachineState *vms = VIRT_MACHINE(obj);
2808 
2809     return vms->default_bus_bypass_iommu;
2810 }
2811 
2812 static void virt_set_default_bus_bypass_iommu(Object *obj, bool value,
2813                                               Error **errp)
2814 {
2815     VirtMachineState *vms = VIRT_MACHINE(obj);
2816 
2817     vms->default_bus_bypass_iommu = value;
2818 }
2819 
2820 static CpuInstanceProperties
2821 virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
2822 {
2823     MachineClass *mc = MACHINE_GET_CLASS(ms);
2824     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
2825 
2826     assert(cpu_index < possible_cpus->len);
2827     return possible_cpus->cpus[cpu_index].props;
2828 }
2829 
2830 static int64_t virt_get_default_cpu_node_id(const MachineState *ms, int idx)
2831 {
2832     int64_t socket_id = ms->possible_cpus->cpus[idx].props.socket_id;
2833 
2834     return socket_id % ms->numa_state->num_nodes;
2835 }
2836 
2837 static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms)
2838 {
2839     int n;
2840     unsigned int max_cpus = ms->smp.max_cpus;
2841     VirtMachineState *vms = VIRT_MACHINE(ms);
2842     MachineClass *mc = MACHINE_GET_CLASS(vms);
2843 
2844     if (ms->possible_cpus) {
2845         assert(ms->possible_cpus->len == max_cpus);
2846         return ms->possible_cpus;
2847     }
2848 
2849     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2850                                   sizeof(CPUArchId) * max_cpus);
2851     ms->possible_cpus->len = max_cpus;
2852     for (n = 0; n < ms->possible_cpus->len; n++) {
2853         ms->possible_cpus->cpus[n].type = ms->cpu_type;
2854         ms->possible_cpus->cpus[n].arch_id =
2855             virt_cpu_mp_affinity(vms, n);
2856 
2857         assert(!mc->smp_props.dies_supported);
2858         ms->possible_cpus->cpus[n].props.has_socket_id = true;
2859         ms->possible_cpus->cpus[n].props.socket_id =
2860             n / (ms->smp.clusters * ms->smp.cores * ms->smp.threads);
2861         ms->possible_cpus->cpus[n].props.has_cluster_id = true;
2862         ms->possible_cpus->cpus[n].props.cluster_id =
2863             (n / (ms->smp.cores * ms->smp.threads)) % ms->smp.clusters;
2864         ms->possible_cpus->cpus[n].props.has_core_id = true;
2865         ms->possible_cpus->cpus[n].props.core_id =
2866             (n / ms->smp.threads) % ms->smp.cores;
2867         ms->possible_cpus->cpus[n].props.has_thread_id = true;
2868         ms->possible_cpus->cpus[n].props.thread_id =
2869             n % ms->smp.threads;
2870     }
2871     return ms->possible_cpus;
2872 }
2873 
2874 static void virt_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2875                                  Error **errp)
2876 {
2877     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2878     const MachineState *ms = MACHINE(hotplug_dev);
2879     const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2880 
2881     if (!vms->acpi_dev) {
2882         error_setg(errp,
2883                    "memory hotplug is not enabled: missing acpi-ged device");
2884         return;
2885     }
2886 
2887     if (vms->mte) {
2888         error_setg(errp, "memory hotplug is not enabled: MTE is enabled");
2889         return;
2890     }
2891 
2892     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
2893         error_setg(errp, "nvdimm is not enabled: add 'nvdimm=on' to '-M'");
2894         return;
2895     }
2896 
2897     pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev), errp);
2898 }
2899 
2900 static void virt_memory_plug(HotplugHandler *hotplug_dev,
2901                              DeviceState *dev, Error **errp)
2902 {
2903     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2904     MachineState *ms = MACHINE(hotplug_dev);
2905     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2906 
2907     pc_dimm_plug(PC_DIMM(dev), MACHINE(vms));
2908 
2909     if (is_nvdimm) {
2910         nvdimm_plug(ms->nvdimms_state);
2911     }
2912 
2913     hotplug_handler_plug(HOTPLUG_HANDLER(vms->acpi_dev),
2914                          dev, &error_abort);
2915 }
2916 
2917 static void virt_machine_device_pre_plug_cb(HotplugHandler *hotplug_dev,
2918                                             DeviceState *dev, Error **errp)
2919 {
2920     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2921 
2922     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2923         virt_memory_pre_plug(hotplug_dev, dev, errp);
2924     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
2925         virtio_md_pci_pre_plug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
2926     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
2927         hwaddr db_start = 0, db_end = 0;
2928         QList *reserved_regions;
2929         char *resv_prop_str;
2930 
2931         if (vms->iommu != VIRT_IOMMU_NONE) {
2932             error_setg(errp, "virt machine does not support multiple IOMMUs");
2933             return;
2934         }
2935 
2936         switch (vms->msi_controller) {
2937         case VIRT_MSI_CTRL_NONE:
2938             return;
2939         case VIRT_MSI_CTRL_ITS:
2940             /* GITS_TRANSLATER page */
2941             db_start = base_memmap[VIRT_GIC_ITS].base + 0x10000;
2942             db_end = base_memmap[VIRT_GIC_ITS].base +
2943                      base_memmap[VIRT_GIC_ITS].size - 1;
2944             break;
2945         case VIRT_MSI_CTRL_GICV2M:
2946             /* MSI_SETSPI_NS page */
2947             db_start = base_memmap[VIRT_GIC_V2M].base;
2948             db_end = db_start + base_memmap[VIRT_GIC_V2M].size - 1;
2949             break;
2950         }
2951         resv_prop_str = g_strdup_printf("0x%"PRIx64":0x%"PRIx64":%u",
2952                                         db_start, db_end,
2953                                         VIRTIO_IOMMU_RESV_MEM_T_MSI);
2954 
2955         reserved_regions = qlist_new();
2956         qlist_append_str(reserved_regions, resv_prop_str);
2957         qdev_prop_set_array(dev, "reserved-regions", reserved_regions);
2958         g_free(resv_prop_str);
2959     }
2960 }
2961 
2962 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
2963                                         DeviceState *dev, Error **errp)
2964 {
2965     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2966 
2967     if (vms->platform_bus_dev) {
2968         MachineClass *mc = MACHINE_GET_CLASS(vms);
2969 
2970         if (device_is_dynamic_sysbus(mc, dev)) {
2971             platform_bus_link_device(PLATFORM_BUS_DEVICE(vms->platform_bus_dev),
2972                                      SYS_BUS_DEVICE(dev));
2973         }
2974     }
2975 
2976     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2977         virt_memory_plug(hotplug_dev, dev, errp);
2978     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
2979         virtio_md_pci_plug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
2980     }
2981 
2982     if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
2983         PCIDevice *pdev = PCI_DEVICE(dev);
2984 
2985         vms->iommu = VIRT_IOMMU_VIRTIO;
2986         vms->virtio_iommu_bdf = pci_get_bdf(pdev);
2987         create_virtio_iommu_dt_bindings(vms);
2988     }
2989 }
2990 
2991 static void virt_dimm_unplug_request(HotplugHandler *hotplug_dev,
2992                                      DeviceState *dev, Error **errp)
2993 {
2994     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2995 
2996     if (!vms->acpi_dev) {
2997         error_setg(errp,
2998                    "memory hotplug is not enabled: missing acpi-ged device");
2999         return;
3000     }
3001 
3002     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3003         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3004         return;
3005     }
3006 
3007     hotplug_handler_unplug_request(HOTPLUG_HANDLER(vms->acpi_dev), dev,
3008                                    errp);
3009 }
3010 
3011 static void virt_dimm_unplug(HotplugHandler *hotplug_dev,
3012                              DeviceState *dev, Error **errp)
3013 {
3014     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
3015     Error *local_err = NULL;
3016 
3017     hotplug_handler_unplug(HOTPLUG_HANDLER(vms->acpi_dev), dev, &local_err);
3018     if (local_err) {
3019         goto out;
3020     }
3021 
3022     pc_dimm_unplug(PC_DIMM(dev), MACHINE(vms));
3023     qdev_unrealize(dev);
3024 
3025 out:
3026     error_propagate(errp, local_err);
3027 }
3028 
3029 static void virt_machine_device_unplug_request_cb(HotplugHandler *hotplug_dev,
3030                                           DeviceState *dev, Error **errp)
3031 {
3032     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3033         virt_dimm_unplug_request(hotplug_dev, dev, errp);
3034     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
3035         virtio_md_pci_unplug_request(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev),
3036                                      errp);
3037     } else {
3038         error_setg(errp, "device unplug request for unsupported device"
3039                    " type: %s", object_get_typename(OBJECT(dev)));
3040     }
3041 }
3042 
3043 static void virt_machine_device_unplug_cb(HotplugHandler *hotplug_dev,
3044                                           DeviceState *dev, Error **errp)
3045 {
3046     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3047         virt_dimm_unplug(hotplug_dev, dev, errp);
3048     } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI)) {
3049         virtio_md_pci_unplug(VIRTIO_MD_PCI(dev), MACHINE(hotplug_dev), errp);
3050     } else {
3051         error_setg(errp, "virt: device unplug for unsupported device"
3052                    " type: %s", object_get_typename(OBJECT(dev)));
3053     }
3054 }
3055 
3056 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
3057                                                         DeviceState *dev)
3058 {
3059     MachineClass *mc = MACHINE_GET_CLASS(machine);
3060 
3061     if (device_is_dynamic_sysbus(mc, dev) ||
3062         object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3063         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_MD_PCI) ||
3064         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
3065         return HOTPLUG_HANDLER(machine);
3066     }
3067     return NULL;
3068 }
3069 
3070 /*
3071  * for arm64 kvm_type [7-0] encodes the requested number of bits
3072  * in the IPA address space
3073  */
3074 static int virt_kvm_type(MachineState *ms, const char *type_str)
3075 {
3076     VirtMachineState *vms = VIRT_MACHINE(ms);
3077     int max_vm_pa_size, requested_pa_size;
3078     bool fixed_ipa;
3079 
3080     max_vm_pa_size = kvm_arm_get_max_vm_ipa_size(ms, &fixed_ipa);
3081 
3082     /* we freeze the memory map to compute the highest gpa */
3083     virt_set_memmap(vms, max_vm_pa_size);
3084 
3085     requested_pa_size = 64 - clz64(vms->highest_gpa);
3086 
3087     /*
3088      * KVM requires the IPA size to be at least 32 bits.
3089      */
3090     if (requested_pa_size < 32) {
3091         requested_pa_size = 32;
3092     }
3093 
3094     if (requested_pa_size > max_vm_pa_size) {
3095         error_report("-m and ,maxmem option values "
3096                      "require an IPA range (%d bits) larger than "
3097                      "the one supported by the host (%d bits)",
3098                      requested_pa_size, max_vm_pa_size);
3099         return -1;
3100     }
3101     /*
3102      * We return the requested PA log size, unless KVM only supports
3103      * the implicit legacy 40b IPA setting, in which case the kvm_type
3104      * must be 0.
3105      */
3106     return fixed_ipa ? 0 : requested_pa_size;
3107 }
3108 
3109 static int virt_hvf_get_physical_address_range(MachineState *ms)
3110 {
3111     VirtMachineState *vms = VIRT_MACHINE(ms);
3112 
3113     int default_ipa_size = hvf_arm_get_default_ipa_bit_size();
3114     int max_ipa_size = hvf_arm_get_max_ipa_bit_size();
3115 
3116     /* We freeze the memory map to compute the highest gpa */
3117     virt_set_memmap(vms, max_ipa_size);
3118 
3119     int requested_ipa_size = 64 - clz64(vms->highest_gpa);
3120 
3121     /*
3122      * If we're <= the default IPA size just use the default.
3123      * If we're above the default but below the maximum, round up to
3124      * the maximum. hvf_arm_get_max_ipa_bit_size() conveniently only
3125      * returns values that are valid ARM PARange values.
3126      */
3127     if (requested_ipa_size <= default_ipa_size) {
3128         requested_ipa_size = default_ipa_size;
3129     } else if (requested_ipa_size <= max_ipa_size) {
3130         requested_ipa_size = max_ipa_size;
3131     } else {
3132         error_report("-m and ,maxmem option values "
3133                      "require an IPA range (%d bits) larger than "
3134                      "the one supported by the host (%d bits)",
3135                      requested_ipa_size, max_ipa_size);
3136         return -1;
3137     }
3138 
3139     return requested_ipa_size;
3140 }
3141 
3142 static void virt_machine_class_init(ObjectClass *oc, const void *data)
3143 {
3144     MachineClass *mc = MACHINE_CLASS(oc);
3145     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3146     static const char * const valid_cpu_types[] = {
3147 #ifdef CONFIG_TCG
3148         ARM_CPU_TYPE_NAME("cortex-a7"),
3149         ARM_CPU_TYPE_NAME("cortex-a15"),
3150 #ifdef TARGET_AARCH64
3151         ARM_CPU_TYPE_NAME("cortex-a35"),
3152         ARM_CPU_TYPE_NAME("cortex-a55"),
3153         ARM_CPU_TYPE_NAME("cortex-a72"),
3154         ARM_CPU_TYPE_NAME("cortex-a76"),
3155         ARM_CPU_TYPE_NAME("cortex-a710"),
3156         ARM_CPU_TYPE_NAME("a64fx"),
3157         ARM_CPU_TYPE_NAME("neoverse-n1"),
3158         ARM_CPU_TYPE_NAME("neoverse-v1"),
3159         ARM_CPU_TYPE_NAME("neoverse-n2"),
3160 #endif /* TARGET_AARCH64 */
3161 #endif /* CONFIG_TCG */
3162 #ifdef TARGET_AARCH64
3163         ARM_CPU_TYPE_NAME("cortex-a53"),
3164         ARM_CPU_TYPE_NAME("cortex-a57"),
3165 #if defined(CONFIG_KVM) || defined(CONFIG_HVF)
3166         ARM_CPU_TYPE_NAME("host"),
3167 #endif /* CONFIG_KVM || CONFIG_HVF */
3168 #endif /* TARGET_AARCH64 */
3169         ARM_CPU_TYPE_NAME("max"),
3170         NULL
3171     };
3172 
3173     mc->init = machvirt_init;
3174     /* Start with max_cpus set to 512, which is the maximum supported by KVM.
3175      * The value may be reduced later when we have more information about the
3176      * configuration of the particular instance.
3177      */
3178     mc->max_cpus = 512;
3179     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_CALXEDA_XGMAC);
3180     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_AMD_XGBE);
3181     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
3182     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_PLATFORM);
3183     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_UEFI_VARS_SYSBUS);
3184 #ifdef CONFIG_TPM
3185     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_TPM_TIS_SYSBUS);
3186 #endif
3187     mc->block_default_type = IF_VIRTIO;
3188     mc->no_cdrom = 1;
3189     mc->pci_allow_0_address = true;
3190     /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */
3191     mc->minimum_page_bits = 12;
3192     mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids;
3193     mc->cpu_index_to_instance_props = virt_cpu_index_to_props;
3194 #ifdef CONFIG_TCG
3195     mc->default_cpu_type = ARM_CPU_TYPE_NAME("cortex-a15");
3196 #else
3197     mc->default_cpu_type = ARM_CPU_TYPE_NAME("max");
3198 #endif
3199     mc->valid_cpu_types = valid_cpu_types;
3200     mc->get_default_cpu_node_id = virt_get_default_cpu_node_id;
3201     mc->kvm_type = virt_kvm_type;
3202     mc->hvf_get_physical_address_range = virt_hvf_get_physical_address_range;
3203     assert(!mc->get_hotplug_handler);
3204     mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
3205     hc->pre_plug = virt_machine_device_pre_plug_cb;
3206     hc->plug = virt_machine_device_plug_cb;
3207     hc->unplug_request = virt_machine_device_unplug_request_cb;
3208     hc->unplug = virt_machine_device_unplug_cb;
3209     mc->nvdimm_supported = true;
3210     mc->smp_props.clusters_supported = true;
3211     mc->auto_enable_numa_with_memhp = true;
3212     mc->auto_enable_numa_with_memdev = true;
3213     /* platform instead of architectural choice */
3214     mc->cpu_cluster_has_numa_boundary = true;
3215     mc->default_ram_id = "mach-virt.ram";
3216     mc->default_nic = "virtio-net-pci";
3217 
3218     object_class_property_add(oc, "acpi", "OnOffAuto",
3219         virt_get_acpi, virt_set_acpi,
3220         NULL, NULL);
3221     object_class_property_set_description(oc, "acpi",
3222         "Enable ACPI");
3223     object_class_property_add_bool(oc, "secure", virt_get_secure,
3224                                    virt_set_secure);
3225     object_class_property_set_description(oc, "secure",
3226                                                 "Set on/off to enable/disable the ARM "
3227                                                 "Security Extensions (TrustZone)");
3228 
3229     object_class_property_add_bool(oc, "virtualization", virt_get_virt,
3230                                    virt_set_virt);
3231     object_class_property_set_description(oc, "virtualization",
3232                                           "Set on/off to enable/disable emulating a "
3233                                           "guest CPU which implements the ARM "
3234                                           "Virtualization Extensions");
3235 
3236     object_class_property_add_bool(oc, "highmem", virt_get_highmem,
3237                                    virt_set_highmem);
3238     object_class_property_set_description(oc, "highmem",
3239                                           "Set on/off to enable/disable using "
3240                                           "physical address space above 32 bits");
3241 
3242     object_class_property_add_bool(oc, "compact-highmem",
3243                                    virt_get_compact_highmem,
3244                                    virt_set_compact_highmem);
3245     object_class_property_set_description(oc, "compact-highmem",
3246                                           "Set on/off to enable/disable compact "
3247                                           "layout for high memory regions");
3248 
3249     object_class_property_add_bool(oc, "highmem-redists",
3250                                    virt_get_highmem_redists,
3251                                    virt_set_highmem_redists);
3252     object_class_property_set_description(oc, "highmem-redists",
3253                                           "Set on/off to enable/disable high "
3254                                           "memory region for GICv3 or GICv4 "
3255                                           "redistributor");
3256 
3257     object_class_property_add_bool(oc, "highmem-ecam",
3258                                    virt_get_highmem_ecam,
3259                                    virt_set_highmem_ecam);
3260     object_class_property_set_description(oc, "highmem-ecam",
3261                                           "Set on/off to enable/disable high "
3262                                           "memory region for PCI ECAM");
3263 
3264     object_class_property_add_bool(oc, "highmem-mmio",
3265                                    virt_get_highmem_mmio,
3266                                    virt_set_highmem_mmio);
3267     object_class_property_set_description(oc, "highmem-mmio",
3268                                           "Set on/off to enable/disable high "
3269                                           "memory region for PCI MMIO");
3270 
3271     object_class_property_add(oc, "highmem-mmio-size", "size",
3272                                    virt_get_highmem_mmio_size,
3273                                    virt_set_highmem_mmio_size,
3274                                    NULL, NULL);
3275     object_class_property_set_description(oc, "highmem-mmio-size",
3276                                           "Set the high memory region size "
3277                                           "for PCI MMIO");
3278 
3279     object_class_property_add_str(oc, "gic-version", virt_get_gic_version,
3280                                   virt_set_gic_version);
3281     object_class_property_set_description(oc, "gic-version",
3282                                           "Set GIC version. "
3283                                           "Valid values are 2, 3, 4, host and max");
3284 
3285     object_class_property_add_str(oc, "iommu", virt_get_iommu, virt_set_iommu);
3286     object_class_property_set_description(oc, "iommu",
3287                                           "Set the IOMMU type. "
3288                                           "Valid values are none and smmuv3");
3289 
3290     object_class_property_add_bool(oc, "default-bus-bypass-iommu",
3291                                    virt_get_default_bus_bypass_iommu,
3292                                    virt_set_default_bus_bypass_iommu);
3293     object_class_property_set_description(oc, "default-bus-bypass-iommu",
3294                                           "Set on/off to enable/disable "
3295                                           "bypass_iommu for default root bus");
3296 
3297     object_class_property_add_bool(oc, "ras", virt_get_ras,
3298                                    virt_set_ras);
3299     object_class_property_set_description(oc, "ras",
3300                                           "Set on/off to enable/disable reporting host memory errors "
3301                                           "to a KVM guest using ACPI and guest external abort exceptions");
3302 
3303     object_class_property_add_bool(oc, "mte", virt_get_mte, virt_set_mte);
3304     object_class_property_set_description(oc, "mte",
3305                                           "Set on/off to enable/disable emulating a "
3306                                           "guest CPU which implements the ARM "
3307                                           "Memory Tagging Extension");
3308 
3309     object_class_property_add_bool(oc, "its", virt_get_its,
3310                                    virt_set_its);
3311     object_class_property_set_description(oc, "its",
3312                                           "Set on/off to enable/disable "
3313                                           "ITS instantiation");
3314 
3315     object_class_property_add_bool(oc, "dtb-randomness",
3316                                    virt_get_dtb_randomness,
3317                                    virt_set_dtb_randomness);
3318     object_class_property_set_description(oc, "dtb-randomness",
3319                                           "Set off to disable passing random or "
3320                                           "non-deterministic dtb nodes to guest");
3321 
3322     object_class_property_add_bool(oc, "dtb-kaslr-seed",
3323                                    virt_get_dtb_randomness,
3324                                    virt_set_dtb_randomness);
3325     object_class_property_set_description(oc, "dtb-kaslr-seed",
3326                                           "Deprecated synonym of dtb-randomness");
3327 
3328     object_class_property_add_str(oc, "x-oem-id",
3329                                   virt_get_oem_id,
3330                                   virt_set_oem_id);
3331     object_class_property_set_description(oc, "x-oem-id",
3332                                           "Override the default value of field OEMID "
3333                                           "in ACPI table header."
3334                                           "The string may be up to 6 bytes in size");
3335 
3336 
3337     object_class_property_add_str(oc, "x-oem-table-id",
3338                                   virt_get_oem_table_id,
3339                                   virt_set_oem_table_id);
3340     object_class_property_set_description(oc, "x-oem-table-id",
3341                                           "Override the default value of field OEM Table ID "
3342                                           "in ACPI table header."
3343                                           "The string may be up to 8 bytes in size");
3344 
3345 }
3346 
3347 static void virt_instance_init(Object *obj)
3348 {
3349     VirtMachineState *vms = VIRT_MACHINE(obj);
3350     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
3351 
3352     /* EL3 is disabled by default on virt: this makes us consistent
3353      * between KVM and TCG for this board, and it also allows us to
3354      * boot UEFI blobs which assume no TrustZone support.
3355      */
3356     vms->secure = false;
3357 
3358     /* EL2 is also disabled by default, for similar reasons */
3359     vms->virt = false;
3360 
3361     /* High memory is enabled by default */
3362     vms->highmem = true;
3363     vms->highmem_compact = !vmc->no_highmem_compact;
3364     vms->gic_version = VIRT_GIC_VERSION_NOSEL;
3365 
3366     vms->highmem_ecam = true;
3367     vms->highmem_mmio = true;
3368     vms->highmem_redists = true;
3369 
3370     /* Default allows ITS instantiation */
3371     vms->its = true;
3372     /* Allow ITS emulation if the machine version supports it */
3373     vms->tcg_its = !vmc->no_tcg_its;
3374 
3375     /* Default disallows iommu instantiation */
3376     vms->iommu = VIRT_IOMMU_NONE;
3377 
3378     /* The default root bus is attached to iommu by default */
3379     vms->default_bus_bypass_iommu = false;
3380 
3381     /* Default disallows RAS instantiation */
3382     vms->ras = false;
3383 
3384     /* MTE is disabled by default.  */
3385     vms->mte = false;
3386 
3387     /* Supply kaslr-seed and rng-seed by default */
3388     vms->dtb_randomness = true;
3389 
3390     vms->irqmap = a15irqmap;
3391 
3392     virt_flash_create(vms);
3393 
3394     vms->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
3395     vms->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
3396     cxl_machine_init(obj, &vms->cxl_devices_state);
3397 }
3398 
3399 static const TypeInfo virt_machine_info = {
3400     .name          = TYPE_VIRT_MACHINE,
3401     .parent        = TYPE_MACHINE,
3402     .abstract      = true,
3403     .instance_size = sizeof(VirtMachineState),
3404     .class_size    = sizeof(VirtMachineClass),
3405     .class_init    = virt_machine_class_init,
3406     .instance_init = virt_instance_init,
3407     .interfaces = (const InterfaceInfo[]) {
3408          { TYPE_HOTPLUG_HANDLER },
3409          { }
3410     },
3411 };
3412 
3413 static void machvirt_machine_init(void)
3414 {
3415     type_register_static(&virt_machine_info);
3416 }
3417 type_init(machvirt_machine_init);
3418 
3419 static void virt_machine_10_1_options(MachineClass *mc)
3420 {
3421 }
3422 DEFINE_VIRT_MACHINE_AS_LATEST(10, 1)
3423 
3424 static void virt_machine_10_0_options(MachineClass *mc)
3425 {
3426     virt_machine_10_1_options(mc);
3427     compat_props_add(mc->compat_props, hw_compat_10_0, hw_compat_10_0_len);
3428 }
3429 DEFINE_VIRT_MACHINE(10, 0)
3430 
3431 static void virt_machine_9_2_options(MachineClass *mc)
3432 {
3433     virt_machine_10_0_options(mc);
3434     compat_props_add(mc->compat_props, hw_compat_9_2, hw_compat_9_2_len);
3435 }
3436 DEFINE_VIRT_MACHINE(9, 2)
3437 
3438 static void virt_machine_9_1_options(MachineClass *mc)
3439 {
3440     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3441 
3442     virt_machine_9_2_options(mc);
3443     compat_props_add(mc->compat_props, hw_compat_9_1, hw_compat_9_1_len);
3444     /* 9.1 and earlier have only a stage-1 SMMU, not a nested s1+2 one */
3445     vmc->no_nested_smmu = true;
3446 }
3447 DEFINE_VIRT_MACHINE(9, 1)
3448 
3449 static void virt_machine_9_0_options(MachineClass *mc)
3450 {
3451     virt_machine_9_1_options(mc);
3452     mc->smbios_memory_device_size = 16 * GiB;
3453     compat_props_add(mc->compat_props, hw_compat_9_0, hw_compat_9_0_len);
3454 }
3455 DEFINE_VIRT_MACHINE(9, 0)
3456 
3457 static void virt_machine_8_2_options(MachineClass *mc)
3458 {
3459     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3460 
3461     virt_machine_9_0_options(mc);
3462     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
3463     /*
3464      * Don't expose NS_EL2_VIRT timer IRQ in DTB on ACPI on 8.2 and
3465      * earlier machines. (Exposing it tickles a bug in older EDK2
3466      * guest BIOS binaries.)
3467      */
3468     vmc->no_ns_el2_virt_timer_irq = true;
3469 }
3470 DEFINE_VIRT_MACHINE(8, 2)
3471 
3472 static void virt_machine_8_1_options(MachineClass *mc)
3473 {
3474     virt_machine_8_2_options(mc);
3475     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
3476 }
3477 DEFINE_VIRT_MACHINE(8, 1)
3478 
3479 static void virt_machine_8_0_options(MachineClass *mc)
3480 {
3481     virt_machine_8_1_options(mc);
3482     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
3483 }
3484 DEFINE_VIRT_MACHINE(8, 0)
3485 
3486 static void virt_machine_7_2_options(MachineClass *mc)
3487 {
3488     virt_machine_8_0_options(mc);
3489     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
3490 }
3491 DEFINE_VIRT_MACHINE(7, 2)
3492 
3493 static void virt_machine_7_1_options(MachineClass *mc)
3494 {
3495     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3496 
3497     virt_machine_7_2_options(mc);
3498     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
3499     /* Compact layout for high memory regions was introduced with 7.2 */
3500     vmc->no_highmem_compact = true;
3501 }
3502 DEFINE_VIRT_MACHINE(7, 1)
3503 
3504 static void virt_machine_7_0_options(MachineClass *mc)
3505 {
3506     virt_machine_7_1_options(mc);
3507     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
3508 }
3509 DEFINE_VIRT_MACHINE(7, 0)
3510 
3511 static void virt_machine_6_2_options(MachineClass *mc)
3512 {
3513     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3514 
3515     virt_machine_7_0_options(mc);
3516     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
3517     vmc->no_tcg_lpa2 = true;
3518 }
3519 DEFINE_VIRT_MACHINE(6, 2)
3520 
3521 static void virt_machine_6_1_options(MachineClass *mc)
3522 {
3523     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3524 
3525     virt_machine_6_2_options(mc);
3526     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
3527     mc->smp_props.prefer_sockets = true;
3528     vmc->no_cpu_topology = true;
3529 
3530     /* qemu ITS was introduced with 6.2 */
3531     vmc->no_tcg_its = true;
3532 }
3533 DEFINE_VIRT_MACHINE(6, 1)
3534 
3535 static void virt_machine_6_0_options(MachineClass *mc)
3536 {
3537     virt_machine_6_1_options(mc);
3538     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
3539 }
3540 DEFINE_VIRT_MACHINE(6, 0)
3541 
3542 static void virt_machine_5_2_options(MachineClass *mc)
3543 {
3544     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3545 
3546     virt_machine_6_0_options(mc);
3547     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
3548     vmc->no_secure_gpio = true;
3549 }
3550 DEFINE_VIRT_MACHINE(5, 2)
3551 
3552 static void virt_machine_5_1_options(MachineClass *mc)
3553 {
3554     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3555 
3556     virt_machine_5_2_options(mc);
3557     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
3558     vmc->no_kvm_steal_time = true;
3559 }
3560 DEFINE_VIRT_MACHINE(5, 1)
3561 
3562 static void virt_machine_5_0_options(MachineClass *mc)
3563 {
3564     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3565 
3566     virt_machine_5_1_options(mc);
3567     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
3568     mc->numa_mem_supported = true;
3569     vmc->acpi_expose_flash = true;
3570     mc->auto_enable_numa_with_memdev = false;
3571 }
3572 DEFINE_VIRT_MACHINE(5, 0)
3573 
3574 static void virt_machine_4_2_options(MachineClass *mc)
3575 {
3576     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3577 
3578     virt_machine_5_0_options(mc);
3579     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
3580     vmc->kvm_no_adjvtime = true;
3581 }
3582 DEFINE_VIRT_MACHINE(4, 2)
3583 
3584 static void virt_machine_4_1_options(MachineClass *mc)
3585 {
3586     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3587 
3588     virt_machine_4_2_options(mc);
3589     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
3590     vmc->no_ged = true;
3591     mc->auto_enable_numa_with_memhp = false;
3592 }
3593 DEFINE_VIRT_MACHINE(4, 1)
3594