xref: /openbmc/qemu/hw/arm/virt.c (revision 88c725c7)
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 "qapi/error.h"
33 #include "hw/sysbus.h"
34 #include "hw/arm/arm.h"
35 #include "hw/arm/primecell.h"
36 #include "hw/arm/virt.h"
37 #include "hw/devices.h"
38 #include "net/net.h"
39 #include "sysemu/block-backend.h"
40 #include "sysemu/device_tree.h"
41 #include "sysemu/numa.h"
42 #include "sysemu/sysemu.h"
43 #include "sysemu/kvm.h"
44 #include "hw/compat.h"
45 #include "hw/loader.h"
46 #include "exec/address-spaces.h"
47 #include "qemu/bitops.h"
48 #include "qemu/error-report.h"
49 #include "hw/pci-host/gpex.h"
50 #include "hw/arm/sysbus-fdt.h"
51 #include "hw/platform-bus.h"
52 #include "hw/arm/fdt.h"
53 #include "hw/intc/arm_gic.h"
54 #include "hw/intc/arm_gicv3_common.h"
55 #include "kvm_arm.h"
56 #include "hw/smbios/smbios.h"
57 #include "qapi/visitor.h"
58 #include "standard-headers/linux/input.h"
59 
60 #define DEFINE_VIRT_MACHINE_LATEST(major, minor, latest) \
61     static void virt_##major##_##minor##_class_init(ObjectClass *oc, \
62                                                     void *data) \
63     { \
64         MachineClass *mc = MACHINE_CLASS(oc); \
65         virt_machine_##major##_##minor##_options(mc); \
66         mc->desc = "QEMU " # major "." # minor " ARM Virtual Machine"; \
67         if (latest) { \
68             mc->alias = "virt"; \
69         } \
70     } \
71     static const TypeInfo machvirt_##major##_##minor##_info = { \
72         .name = MACHINE_TYPE_NAME("virt-" # major "." # minor), \
73         .parent = TYPE_VIRT_MACHINE, \
74         .instance_init = virt_##major##_##minor##_instance_init, \
75         .class_init = virt_##major##_##minor##_class_init, \
76     }; \
77     static void machvirt_machine_##major##_##minor##_init(void) \
78     { \
79         type_register_static(&machvirt_##major##_##minor##_info); \
80     } \
81     type_init(machvirt_machine_##major##_##minor##_init);
82 
83 #define DEFINE_VIRT_MACHINE_AS_LATEST(major, minor) \
84     DEFINE_VIRT_MACHINE_LATEST(major, minor, true)
85 #define DEFINE_VIRT_MACHINE(major, minor) \
86     DEFINE_VIRT_MACHINE_LATEST(major, minor, false)
87 
88 
89 /* Number of external interrupt lines to configure the GIC with */
90 #define NUM_IRQS 256
91 
92 #define PLATFORM_BUS_NUM_IRQS 64
93 
94 static ARMPlatformBusSystemParams platform_bus_params;
95 
96 /* RAM limit in GB. Since VIRT_MEM starts at the 1GB mark, this means
97  * RAM can go up to the 256GB mark, leaving 256GB of the physical
98  * address space unallocated and free for future use between 256G and 512G.
99  * If we need to provide more RAM to VMs in the future then we need to:
100  *  * allocate a second bank of RAM starting at 2TB and working up
101  *  * fix the DT and ACPI table generation code in QEMU to correctly
102  *    report two split lumps of RAM to the guest
103  *  * fix KVM in the host kernel to allow guests with >40 bit address spaces
104  * (We don't want to fill all the way up to 512GB with RAM because
105  * we might want it for non-RAM purposes later. Conversely it seems
106  * reasonable to assume that anybody configuring a VM with a quarter
107  * of a terabyte of RAM will be doing it on a host with more than a
108  * terabyte of physical address space.)
109  */
110 #define RAMLIMIT_GB 255
111 #define RAMLIMIT_BYTES (RAMLIMIT_GB * 1024ULL * 1024 * 1024)
112 
113 /* Addresses and sizes of our components.
114  * 0..128MB is space for a flash device so we can run bootrom code such as UEFI.
115  * 128MB..256MB is used for miscellaneous device I/O.
116  * 256MB..1GB is reserved for possible future PCI support (ie where the
117  * PCI memory window will go if we add a PCI host controller).
118  * 1GB and up is RAM (which may happily spill over into the
119  * high memory region beyond 4GB).
120  * This represents a compromise between how much RAM can be given to
121  * a 32 bit VM and leaving space for expansion and in particular for PCI.
122  * Note that devices should generally be placed at multiples of 0x10000,
123  * to accommodate guests using 64K pages.
124  */
125 static const MemMapEntry a15memmap[] = {
126     /* Space up to 0x8000000 is reserved for a boot ROM */
127     [VIRT_FLASH] =              {          0, 0x08000000 },
128     [VIRT_CPUPERIPHS] =         { 0x08000000, 0x00020000 },
129     /* GIC distributor and CPU interfaces sit inside the CPU peripheral space */
130     [VIRT_GIC_DIST] =           { 0x08000000, 0x00010000 },
131     [VIRT_GIC_CPU] =            { 0x08010000, 0x00010000 },
132     [VIRT_GIC_V2M] =            { 0x08020000, 0x00001000 },
133     /* The space in between here is reserved for GICv3 CPU/vCPU/HYP */
134     [VIRT_GIC_ITS] =            { 0x08080000, 0x00020000 },
135     /* This redistributor space allows up to 2*64kB*123 CPUs */
136     [VIRT_GIC_REDIST] =         { 0x080A0000, 0x00F60000 },
137     [VIRT_UART] =               { 0x09000000, 0x00001000 },
138     [VIRT_RTC] =                { 0x09010000, 0x00001000 },
139     [VIRT_FW_CFG] =             { 0x09020000, 0x00000018 },
140     [VIRT_GPIO] =               { 0x09030000, 0x00001000 },
141     [VIRT_SECURE_UART] =        { 0x09040000, 0x00001000 },
142     [VIRT_MMIO] =               { 0x0a000000, 0x00000200 },
143     /* ...repeating for a total of NUM_VIRTIO_TRANSPORTS, each of that size */
144     [VIRT_PLATFORM_BUS] =       { 0x0c000000, 0x02000000 },
145     [VIRT_SECURE_MEM] =         { 0x0e000000, 0x01000000 },
146     [VIRT_PCIE_MMIO] =          { 0x10000000, 0x2eff0000 },
147     [VIRT_PCIE_PIO] =           { 0x3eff0000, 0x00010000 },
148     [VIRT_PCIE_ECAM] =          { 0x3f000000, 0x01000000 },
149     [VIRT_MEM] =                { 0x40000000, RAMLIMIT_BYTES },
150     /* Second PCIe window, 512GB wide at the 512GB boundary */
151     [VIRT_PCIE_MMIO_HIGH] =   { 0x8000000000ULL, 0x8000000000ULL },
152 };
153 
154 static const int a15irqmap[] = {
155     [VIRT_UART] = 1,
156     [VIRT_RTC] = 2,
157     [VIRT_PCIE] = 3, /* ... to 6 */
158     [VIRT_GPIO] = 7,
159     [VIRT_SECURE_UART] = 8,
160     [VIRT_MMIO] = 16, /* ...to 16 + NUM_VIRTIO_TRANSPORTS - 1 */
161     [VIRT_GIC_V2M] = 48, /* ...to 48 + NUM_GICV2M_SPIS - 1 */
162     [VIRT_PLATFORM_BUS] = 112, /* ...to 112 + PLATFORM_BUS_NUM_IRQS -1 */
163 };
164 
165 static const char *valid_cpus[] = {
166     "cortex-a15",
167     "cortex-a53",
168     "cortex-a57",
169     "host",
170 };
171 
172 static bool cpuname_valid(const char *cpu)
173 {
174     int i;
175 
176     for (i = 0; i < ARRAY_SIZE(valid_cpus); i++) {
177         if (strcmp(cpu, valid_cpus[i]) == 0) {
178             return true;
179         }
180     }
181     return false;
182 }
183 
184 static void create_fdt(VirtMachineState *vms)
185 {
186     void *fdt = create_device_tree(&vms->fdt_size);
187 
188     if (!fdt) {
189         error_report("create_device_tree() failed");
190         exit(1);
191     }
192 
193     vms->fdt = fdt;
194 
195     /* Header */
196     qemu_fdt_setprop_string(fdt, "/", "compatible", "linux,dummy-virt");
197     qemu_fdt_setprop_cell(fdt, "/", "#address-cells", 0x2);
198     qemu_fdt_setprop_cell(fdt, "/", "#size-cells", 0x2);
199 
200     /*
201      * /chosen and /memory nodes must exist for load_dtb
202      * to fill in necessary properties later
203      */
204     qemu_fdt_add_subnode(fdt, "/chosen");
205     qemu_fdt_add_subnode(fdt, "/memory");
206     qemu_fdt_setprop_string(fdt, "/memory", "device_type", "memory");
207 
208     /* Clock node, for the benefit of the UART. The kernel device tree
209      * binding documentation claims the PL011 node clock properties are
210      * optional but in practice if you omit them the kernel refuses to
211      * probe for the device.
212      */
213     vms->clock_phandle = qemu_fdt_alloc_phandle(fdt);
214     qemu_fdt_add_subnode(fdt, "/apb-pclk");
215     qemu_fdt_setprop_string(fdt, "/apb-pclk", "compatible", "fixed-clock");
216     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "#clock-cells", 0x0);
217     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "clock-frequency", 24000000);
218     qemu_fdt_setprop_string(fdt, "/apb-pclk", "clock-output-names",
219                                 "clk24mhz");
220     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "phandle", vms->clock_phandle);
221 
222     if (have_numa_distance) {
223         int size = nb_numa_nodes * nb_numa_nodes * 3 * sizeof(uint32_t);
224         uint32_t *matrix = g_malloc0(size);
225         int idx, i, j;
226 
227         for (i = 0; i < nb_numa_nodes; i++) {
228             for (j = 0; j < nb_numa_nodes; j++) {
229                 idx = (i * nb_numa_nodes + j) * 3;
230                 matrix[idx + 0] = cpu_to_be32(i);
231                 matrix[idx + 1] = cpu_to_be32(j);
232                 matrix[idx + 2] = cpu_to_be32(numa_info[i].distance[j]);
233             }
234         }
235 
236         qemu_fdt_add_subnode(fdt, "/distance-map");
237         qemu_fdt_setprop_string(fdt, "/distance-map", "compatible",
238                                 "numa-distance-map-v1");
239         qemu_fdt_setprop(fdt, "/distance-map", "distance-matrix",
240                          matrix, size);
241         g_free(matrix);
242     }
243 }
244 
245 static void fdt_add_psci_node(const VirtMachineState *vms)
246 {
247     uint32_t cpu_suspend_fn;
248     uint32_t cpu_off_fn;
249     uint32_t cpu_on_fn;
250     uint32_t migrate_fn;
251     void *fdt = vms->fdt;
252     ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
253     const char *psci_method;
254 
255     switch (vms->psci_conduit) {
256     case QEMU_PSCI_CONDUIT_DISABLED:
257         return;
258     case QEMU_PSCI_CONDUIT_HVC:
259         psci_method = "hvc";
260         break;
261     case QEMU_PSCI_CONDUIT_SMC:
262         psci_method = "smc";
263         break;
264     default:
265         g_assert_not_reached();
266     }
267 
268     qemu_fdt_add_subnode(fdt, "/psci");
269     if (armcpu->psci_version == 2) {
270         const char comp[] = "arm,psci-0.2\0arm,psci";
271         qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
272 
273         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
274         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
275             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
276             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
277             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
278         } else {
279             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
280             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
281             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
282         }
283     } else {
284         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
285 
286         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
287         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
288         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
289         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
290     }
291 
292     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
293      * to the instruction that should be used to invoke PSCI functions.
294      * However, the device tree binding uses 'method' instead, so that is
295      * what we should use here.
296      */
297     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
298 
299     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
300     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
301     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
302     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
303 }
304 
305 static void fdt_add_timer_nodes(const VirtMachineState *vms)
306 {
307     /* On real hardware these interrupts are level-triggered.
308      * On KVM they were edge-triggered before host kernel version 4.4,
309      * and level-triggered afterwards.
310      * On emulated QEMU they are level-triggered.
311      *
312      * Getting the DTB info about them wrong is awkward for some
313      * guest kernels:
314      *  pre-4.8 ignore the DT and leave the interrupt configured
315      *   with whatever the GIC reset value (or the bootloader) left it at
316      *  4.8 before rc6 honour the incorrect data by programming it back
317      *   into the GIC, causing problems
318      *  4.8rc6 and later ignore the DT and always write "level triggered"
319      *   into the GIC
320      *
321      * For backwards-compatibility, virt-2.8 and earlier will continue
322      * to say these are edge-triggered, but later machines will report
323      * the correct information.
324      */
325     ARMCPU *armcpu;
326     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
327     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
328 
329     if (vmc->claim_edge_triggered_timers) {
330         irqflags = GIC_FDT_IRQ_FLAGS_EDGE_LO_HI;
331     }
332 
333     if (vms->gic_version == 2) {
334         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
335                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
336                              (1 << vms->smp_cpus) - 1);
337     }
338 
339     qemu_fdt_add_subnode(vms->fdt, "/timer");
340 
341     armcpu = ARM_CPU(qemu_get_cpu(0));
342     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
343         const char compat[] = "arm,armv8-timer\0arm,armv7-timer";
344         qemu_fdt_setprop(vms->fdt, "/timer", "compatible",
345                          compat, sizeof(compat));
346     } else {
347         qemu_fdt_setprop_string(vms->fdt, "/timer", "compatible",
348                                 "arm,armv7-timer");
349     }
350     qemu_fdt_setprop(vms->fdt, "/timer", "always-on", NULL, 0);
351     qemu_fdt_setprop_cells(vms->fdt, "/timer", "interrupts",
352                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_S_EL1_IRQ, irqflags,
353                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL1_IRQ, irqflags,
354                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_VIRT_IRQ, irqflags,
355                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL2_IRQ, irqflags);
356 }
357 
358 static void fdt_add_cpu_nodes(const VirtMachineState *vms)
359 {
360     int cpu;
361     int addr_cells = 1;
362     const MachineState *ms = MACHINE(vms);
363 
364     /*
365      * From Documentation/devicetree/bindings/arm/cpus.txt
366      *  On ARM v8 64-bit systems value should be set to 2,
367      *  that corresponds to the MPIDR_EL1 register size.
368      *  If MPIDR_EL1[63:32] value is equal to 0 on all CPUs
369      *  in the system, #address-cells can be set to 1, since
370      *  MPIDR_EL1[63:32] bits are not used for CPUs
371      *  identification.
372      *
373      *  Here we actually don't know whether our system is 32- or 64-bit one.
374      *  The simplest way to go is to examine affinity IDs of all our CPUs. If
375      *  at least one of them has Aff3 populated, we set #address-cells to 2.
376      */
377     for (cpu = 0; cpu < vms->smp_cpus; cpu++) {
378         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
379 
380         if (armcpu->mp_affinity & ARM_AFF3_MASK) {
381             addr_cells = 2;
382             break;
383         }
384     }
385 
386     qemu_fdt_add_subnode(vms->fdt, "/cpus");
387     qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#address-cells", addr_cells);
388     qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#size-cells", 0x0);
389 
390     for (cpu = vms->smp_cpus - 1; cpu >= 0; cpu--) {
391         char *nodename = g_strdup_printf("/cpus/cpu@%d", cpu);
392         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
393         CPUState *cs = CPU(armcpu);
394 
395         qemu_fdt_add_subnode(vms->fdt, nodename);
396         qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "cpu");
397         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible",
398                                     armcpu->dtb_compatible);
399 
400         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED
401             && vms->smp_cpus > 1) {
402             qemu_fdt_setprop_string(vms->fdt, nodename,
403                                         "enable-method", "psci");
404         }
405 
406         if (addr_cells == 2) {
407             qemu_fdt_setprop_u64(vms->fdt, nodename, "reg",
408                                  armcpu->mp_affinity);
409         } else {
410             qemu_fdt_setprop_cell(vms->fdt, nodename, "reg",
411                                   armcpu->mp_affinity);
412         }
413 
414         if (ms->possible_cpus->cpus[cs->cpu_index].props.has_node_id) {
415             qemu_fdt_setprop_cell(vms->fdt, nodename, "numa-node-id",
416                 ms->possible_cpus->cpus[cs->cpu_index].props.node_id);
417         }
418 
419         g_free(nodename);
420     }
421 }
422 
423 static void fdt_add_its_gic_node(VirtMachineState *vms)
424 {
425     vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt);
426     qemu_fdt_add_subnode(vms->fdt, "/intc/its");
427     qemu_fdt_setprop_string(vms->fdt, "/intc/its", "compatible",
428                             "arm,gic-v3-its");
429     qemu_fdt_setprop(vms->fdt, "/intc/its", "msi-controller", NULL, 0);
430     qemu_fdt_setprop_sized_cells(vms->fdt, "/intc/its", "reg",
431                                  2, vms->memmap[VIRT_GIC_ITS].base,
432                                  2, vms->memmap[VIRT_GIC_ITS].size);
433     qemu_fdt_setprop_cell(vms->fdt, "/intc/its", "phandle", vms->msi_phandle);
434 }
435 
436 static void fdt_add_v2m_gic_node(VirtMachineState *vms)
437 {
438     vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt);
439     qemu_fdt_add_subnode(vms->fdt, "/intc/v2m");
440     qemu_fdt_setprop_string(vms->fdt, "/intc/v2m", "compatible",
441                             "arm,gic-v2m-frame");
442     qemu_fdt_setprop(vms->fdt, "/intc/v2m", "msi-controller", NULL, 0);
443     qemu_fdt_setprop_sized_cells(vms->fdt, "/intc/v2m", "reg",
444                                  2, vms->memmap[VIRT_GIC_V2M].base,
445                                  2, vms->memmap[VIRT_GIC_V2M].size);
446     qemu_fdt_setprop_cell(vms->fdt, "/intc/v2m", "phandle", vms->msi_phandle);
447 }
448 
449 static void fdt_add_gic_node(VirtMachineState *vms)
450 {
451     vms->gic_phandle = qemu_fdt_alloc_phandle(vms->fdt);
452     qemu_fdt_setprop_cell(vms->fdt, "/", "interrupt-parent", vms->gic_phandle);
453 
454     qemu_fdt_add_subnode(vms->fdt, "/intc");
455     qemu_fdt_setprop_cell(vms->fdt, "/intc", "#interrupt-cells", 3);
456     qemu_fdt_setprop(vms->fdt, "/intc", "interrupt-controller", NULL, 0);
457     qemu_fdt_setprop_cell(vms->fdt, "/intc", "#address-cells", 0x2);
458     qemu_fdt_setprop_cell(vms->fdt, "/intc", "#size-cells", 0x2);
459     qemu_fdt_setprop(vms->fdt, "/intc", "ranges", NULL, 0);
460     if (vms->gic_version == 3) {
461         qemu_fdt_setprop_string(vms->fdt, "/intc", "compatible",
462                                 "arm,gic-v3");
463         qemu_fdt_setprop_sized_cells(vms->fdt, "/intc", "reg",
464                                      2, vms->memmap[VIRT_GIC_DIST].base,
465                                      2, vms->memmap[VIRT_GIC_DIST].size,
466                                      2, vms->memmap[VIRT_GIC_REDIST].base,
467                                      2, vms->memmap[VIRT_GIC_REDIST].size);
468         if (vms->virt) {
469             qemu_fdt_setprop_cells(vms->fdt, "/intc", "interrupts",
470                                    GIC_FDT_IRQ_TYPE_PPI, ARCH_GICV3_MAINT_IRQ,
471                                    GIC_FDT_IRQ_FLAGS_LEVEL_HI);
472         }
473     } else {
474         /* 'cortex-a15-gic' means 'GIC v2' */
475         qemu_fdt_setprop_string(vms->fdt, "/intc", "compatible",
476                                 "arm,cortex-a15-gic");
477         qemu_fdt_setprop_sized_cells(vms->fdt, "/intc", "reg",
478                                       2, vms->memmap[VIRT_GIC_DIST].base,
479                                       2, vms->memmap[VIRT_GIC_DIST].size,
480                                       2, vms->memmap[VIRT_GIC_CPU].base,
481                                       2, vms->memmap[VIRT_GIC_CPU].size);
482     }
483 
484     qemu_fdt_setprop_cell(vms->fdt, "/intc", "phandle", vms->gic_phandle);
485 }
486 
487 static void fdt_add_pmu_nodes(const VirtMachineState *vms)
488 {
489     CPUState *cpu;
490     ARMCPU *armcpu;
491     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
492 
493     CPU_FOREACH(cpu) {
494         armcpu = ARM_CPU(cpu);
495         if (!arm_feature(&armcpu->env, ARM_FEATURE_PMU) ||
496             (kvm_enabled() && !kvm_arm_pmu_create(cpu, PPI(VIRTUAL_PMU_IRQ)))) {
497             return;
498         }
499     }
500 
501     if (vms->gic_version == 2) {
502         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
503                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
504                              (1 << vms->smp_cpus) - 1);
505     }
506 
507     armcpu = ARM_CPU(qemu_get_cpu(0));
508     qemu_fdt_add_subnode(vms->fdt, "/pmu");
509     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
510         const char compat[] = "arm,armv8-pmuv3";
511         qemu_fdt_setprop(vms->fdt, "/pmu", "compatible",
512                          compat, sizeof(compat));
513         qemu_fdt_setprop_cells(vms->fdt, "/pmu", "interrupts",
514                                GIC_FDT_IRQ_TYPE_PPI, VIRTUAL_PMU_IRQ, irqflags);
515     }
516 }
517 
518 static void create_its(VirtMachineState *vms, DeviceState *gicdev)
519 {
520     const char *itsclass = its_class_name();
521     DeviceState *dev;
522 
523     if (!itsclass) {
524         /* Do nothing if not supported */
525         return;
526     }
527 
528     dev = qdev_create(NULL, itsclass);
529 
530     object_property_set_link(OBJECT(dev), OBJECT(gicdev), "parent-gicv3",
531                              &error_abort);
532     qdev_init_nofail(dev);
533     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_ITS].base);
534 
535     fdt_add_its_gic_node(vms);
536 }
537 
538 static void create_v2m(VirtMachineState *vms, qemu_irq *pic)
539 {
540     int i;
541     int irq = vms->irqmap[VIRT_GIC_V2M];
542     DeviceState *dev;
543 
544     dev = qdev_create(NULL, "arm-gicv2m");
545     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_V2M].base);
546     qdev_prop_set_uint32(dev, "base-spi", irq);
547     qdev_prop_set_uint32(dev, "num-spi", NUM_GICV2M_SPIS);
548     qdev_init_nofail(dev);
549 
550     for (i = 0; i < NUM_GICV2M_SPIS; i++) {
551         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]);
552     }
553 
554     fdt_add_v2m_gic_node(vms);
555 }
556 
557 static void create_gic(VirtMachineState *vms, qemu_irq *pic)
558 {
559     /* We create a standalone GIC */
560     DeviceState *gicdev;
561     SysBusDevice *gicbusdev;
562     const char *gictype;
563     int type = vms->gic_version, i;
564 
565     gictype = (type == 3) ? gicv3_class_name() : gic_class_name();
566 
567     gicdev = qdev_create(NULL, gictype);
568     qdev_prop_set_uint32(gicdev, "revision", type);
569     qdev_prop_set_uint32(gicdev, "num-cpu", smp_cpus);
570     /* Note that the num-irq property counts both internal and external
571      * interrupts; there are always 32 of the former (mandated by GIC spec).
572      */
573     qdev_prop_set_uint32(gicdev, "num-irq", NUM_IRQS + 32);
574     if (!kvm_irqchip_in_kernel()) {
575         qdev_prop_set_bit(gicdev, "has-security-extensions", vms->secure);
576     }
577     qdev_init_nofail(gicdev);
578     gicbusdev = SYS_BUS_DEVICE(gicdev);
579     sysbus_mmio_map(gicbusdev, 0, vms->memmap[VIRT_GIC_DIST].base);
580     if (type == 3) {
581         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_REDIST].base);
582     } else {
583         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base);
584     }
585 
586     /* Wire the outputs from each CPU's generic timer and the GICv3
587      * maintenance interrupt signal to the appropriate GIC PPI inputs,
588      * and the GIC's IRQ/FIQ/VIRQ/VFIQ interrupt outputs to the CPU's inputs.
589      */
590     for (i = 0; i < smp_cpus; i++) {
591         DeviceState *cpudev = DEVICE(qemu_get_cpu(i));
592         int ppibase = NUM_IRQS + i * GIC_INTERNAL + GIC_NR_SGIS;
593         int irq;
594         /* Mapping from the output timer irq lines from the CPU to the
595          * GIC PPI inputs we use for the virt board.
596          */
597         const int timer_irq[] = {
598             [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ,
599             [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ,
600             [GTIMER_HYP]  = ARCH_TIMER_NS_EL2_IRQ,
601             [GTIMER_SEC]  = ARCH_TIMER_S_EL1_IRQ,
602         };
603 
604         for (irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) {
605             qdev_connect_gpio_out(cpudev, irq,
606                                   qdev_get_gpio_in(gicdev,
607                                                    ppibase + timer_irq[irq]));
608         }
609 
610         qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt", 0,
611                                     qdev_get_gpio_in(gicdev, ppibase
612                                                      + ARCH_GICV3_MAINT_IRQ));
613 
614         sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ));
615         sysbus_connect_irq(gicbusdev, i + smp_cpus,
616                            qdev_get_gpio_in(cpudev, ARM_CPU_FIQ));
617         sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus,
618                            qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ));
619         sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus,
620                            qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ));
621     }
622 
623     for (i = 0; i < NUM_IRQS; i++) {
624         pic[i] = qdev_get_gpio_in(gicdev, i);
625     }
626 
627     fdt_add_gic_node(vms);
628 
629     if (type == 3 && vms->its) {
630         create_its(vms, gicdev);
631     } else if (type == 2) {
632         create_v2m(vms, pic);
633     }
634 }
635 
636 static void create_uart(const VirtMachineState *vms, qemu_irq *pic, int uart,
637                         MemoryRegion *mem, Chardev *chr)
638 {
639     char *nodename;
640     hwaddr base = vms->memmap[uart].base;
641     hwaddr size = vms->memmap[uart].size;
642     int irq = vms->irqmap[uart];
643     const char compat[] = "arm,pl011\0arm,primecell";
644     const char clocknames[] = "uartclk\0apb_pclk";
645     DeviceState *dev = qdev_create(NULL, "pl011");
646     SysBusDevice *s = SYS_BUS_DEVICE(dev);
647 
648     qdev_prop_set_chr(dev, "chardev", chr);
649     qdev_init_nofail(dev);
650     memory_region_add_subregion(mem, base,
651                                 sysbus_mmio_get_region(s, 0));
652     sysbus_connect_irq(s, 0, pic[irq]);
653 
654     nodename = g_strdup_printf("/pl011@%" PRIx64, base);
655     qemu_fdt_add_subnode(vms->fdt, nodename);
656     /* Note that we can't use setprop_string because of the embedded NUL */
657     qemu_fdt_setprop(vms->fdt, nodename, "compatible",
658                          compat, sizeof(compat));
659     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
660                                      2, base, 2, size);
661     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
662                                GIC_FDT_IRQ_TYPE_SPI, irq,
663                                GIC_FDT_IRQ_FLAGS_LEVEL_HI);
664     qemu_fdt_setprop_cells(vms->fdt, nodename, "clocks",
665                                vms->clock_phandle, vms->clock_phandle);
666     qemu_fdt_setprop(vms->fdt, nodename, "clock-names",
667                          clocknames, sizeof(clocknames));
668 
669     if (uart == VIRT_UART) {
670         qemu_fdt_setprop_string(vms->fdt, "/chosen", "stdout-path", nodename);
671     } else {
672         /* Mark as not usable by the normal world */
673         qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
674         qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
675     }
676 
677     g_free(nodename);
678 }
679 
680 static void create_rtc(const VirtMachineState *vms, qemu_irq *pic)
681 {
682     char *nodename;
683     hwaddr base = vms->memmap[VIRT_RTC].base;
684     hwaddr size = vms->memmap[VIRT_RTC].size;
685     int irq = vms->irqmap[VIRT_RTC];
686     const char compat[] = "arm,pl031\0arm,primecell";
687 
688     sysbus_create_simple("pl031", base, pic[irq]);
689 
690     nodename = g_strdup_printf("/pl031@%" PRIx64, base);
691     qemu_fdt_add_subnode(vms->fdt, nodename);
692     qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat));
693     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
694                                  2, base, 2, size);
695     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
696                            GIC_FDT_IRQ_TYPE_SPI, irq,
697                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
698     qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle);
699     qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk");
700     g_free(nodename);
701 }
702 
703 static DeviceState *gpio_key_dev;
704 static void virt_powerdown_req(Notifier *n, void *opaque)
705 {
706     /* use gpio Pin 3 for power button event */
707     qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1);
708 }
709 
710 static Notifier virt_system_powerdown_notifier = {
711     .notify = virt_powerdown_req
712 };
713 
714 static void create_gpio(const VirtMachineState *vms, qemu_irq *pic)
715 {
716     char *nodename;
717     DeviceState *pl061_dev;
718     hwaddr base = vms->memmap[VIRT_GPIO].base;
719     hwaddr size = vms->memmap[VIRT_GPIO].size;
720     int irq = vms->irqmap[VIRT_GPIO];
721     const char compat[] = "arm,pl061\0arm,primecell";
722 
723     pl061_dev = sysbus_create_simple("pl061", base, pic[irq]);
724 
725     uint32_t phandle = qemu_fdt_alloc_phandle(vms->fdt);
726     nodename = g_strdup_printf("/pl061@%" PRIx64, base);
727     qemu_fdt_add_subnode(vms->fdt, nodename);
728     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
729                                  2, base, 2, size);
730     qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat));
731     qemu_fdt_setprop_cell(vms->fdt, nodename, "#gpio-cells", 2);
732     qemu_fdt_setprop(vms->fdt, nodename, "gpio-controller", NULL, 0);
733     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
734                            GIC_FDT_IRQ_TYPE_SPI, irq,
735                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
736     qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle);
737     qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk");
738     qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", phandle);
739 
740     gpio_key_dev = sysbus_create_simple("gpio-key", -1,
741                                         qdev_get_gpio_in(pl061_dev, 3));
742     qemu_fdt_add_subnode(vms->fdt, "/gpio-keys");
743     qemu_fdt_setprop_string(vms->fdt, "/gpio-keys", "compatible", "gpio-keys");
744     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#size-cells", 0);
745     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#address-cells", 1);
746 
747     qemu_fdt_add_subnode(vms->fdt, "/gpio-keys/poweroff");
748     qemu_fdt_setprop_string(vms->fdt, "/gpio-keys/poweroff",
749                             "label", "GPIO Key Poweroff");
750     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys/poweroff", "linux,code",
751                           KEY_POWER);
752     qemu_fdt_setprop_cells(vms->fdt, "/gpio-keys/poweroff",
753                            "gpios", phandle, 3, 0);
754 
755     /* connect powerdown request */
756     qemu_register_powerdown_notifier(&virt_system_powerdown_notifier);
757 
758     g_free(nodename);
759 }
760 
761 static void create_virtio_devices(const VirtMachineState *vms, qemu_irq *pic)
762 {
763     int i;
764     hwaddr size = vms->memmap[VIRT_MMIO].size;
765 
766     /* We create the transports in forwards order. Since qbus_realize()
767      * prepends (not appends) new child buses, the incrementing loop below will
768      * create a list of virtio-mmio buses with decreasing base addresses.
769      *
770      * When a -device option is processed from the command line,
771      * qbus_find_recursive() picks the next free virtio-mmio bus in forwards
772      * order. The upshot is that -device options in increasing command line
773      * order are mapped to virtio-mmio buses with decreasing base addresses.
774      *
775      * When this code was originally written, that arrangement ensured that the
776      * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to
777      * the first -device on the command line. (The end-to-end order is a
778      * function of this loop, qbus_realize(), qbus_find_recursive(), and the
779      * guest kernel's name-to-address assignment strategy.)
780      *
781      * Meanwhile, the kernel's traversal seems to have been reversed; see eg.
782      * the message, if not necessarily the code, of commit 70161ff336.
783      * Therefore the loop now establishes the inverse of the original intent.
784      *
785      * Unfortunately, we can't counteract the kernel change by reversing the
786      * loop; it would break existing command lines.
787      *
788      * In any case, the kernel makes no guarantee about the stability of
789      * enumeration order of virtio devices (as demonstrated by it changing
790      * between kernel versions). For reliable and stable identification
791      * of disks users must use UUIDs or similar mechanisms.
792      */
793     for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) {
794         int irq = vms->irqmap[VIRT_MMIO] + i;
795         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
796 
797         sysbus_create_simple("virtio-mmio", base, pic[irq]);
798     }
799 
800     /* We add dtb nodes in reverse order so that they appear in the finished
801      * device tree lowest address first.
802      *
803      * Note that this mapping is independent of the loop above. The previous
804      * loop influences virtio device to virtio transport assignment, whereas
805      * this loop controls how virtio transports are laid out in the dtb.
806      */
807     for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) {
808         char *nodename;
809         int irq = vms->irqmap[VIRT_MMIO] + i;
810         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
811 
812         nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base);
813         qemu_fdt_add_subnode(vms->fdt, nodename);
814         qemu_fdt_setprop_string(vms->fdt, nodename,
815                                 "compatible", "virtio,mmio");
816         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
817                                      2, base, 2, size);
818         qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
819                                GIC_FDT_IRQ_TYPE_SPI, irq,
820                                GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
821         qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
822         g_free(nodename);
823     }
824 }
825 
826 static void create_one_flash(const char *name, hwaddr flashbase,
827                              hwaddr flashsize, const char *file,
828                              MemoryRegion *sysmem)
829 {
830     /* Create and map a single flash device. We use the same
831      * parameters as the flash devices on the Versatile Express board.
832      */
833     DriveInfo *dinfo = drive_get_next(IF_PFLASH);
834     DeviceState *dev = qdev_create(NULL, "cfi.pflash01");
835     SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
836     const uint64_t sectorlength = 256 * 1024;
837 
838     if (dinfo) {
839         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
840                             &error_abort);
841     }
842 
843     qdev_prop_set_uint32(dev, "num-blocks", flashsize / sectorlength);
844     qdev_prop_set_uint64(dev, "sector-length", sectorlength);
845     qdev_prop_set_uint8(dev, "width", 4);
846     qdev_prop_set_uint8(dev, "device-width", 2);
847     qdev_prop_set_bit(dev, "big-endian", false);
848     qdev_prop_set_uint16(dev, "id0", 0x89);
849     qdev_prop_set_uint16(dev, "id1", 0x18);
850     qdev_prop_set_uint16(dev, "id2", 0x00);
851     qdev_prop_set_uint16(dev, "id3", 0x00);
852     qdev_prop_set_string(dev, "name", name);
853     qdev_init_nofail(dev);
854 
855     memory_region_add_subregion(sysmem, flashbase,
856                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0));
857 
858     if (file) {
859         char *fn;
860         int image_size;
861 
862         if (drive_get(IF_PFLASH, 0, 0)) {
863             error_report("The contents of the first flash device may be "
864                          "specified with -bios or with -drive if=pflash... "
865                          "but you cannot use both options at once");
866             exit(1);
867         }
868         fn = qemu_find_file(QEMU_FILE_TYPE_BIOS, file);
869         if (!fn) {
870             error_report("Could not find ROM image '%s'", file);
871             exit(1);
872         }
873         image_size = load_image_mr(fn, sysbus_mmio_get_region(sbd, 0));
874         g_free(fn);
875         if (image_size < 0) {
876             error_report("Could not load ROM image '%s'", file);
877             exit(1);
878         }
879     }
880 }
881 
882 static void create_flash(const VirtMachineState *vms,
883                          MemoryRegion *sysmem,
884                          MemoryRegion *secure_sysmem)
885 {
886     /* Create two flash devices to fill the VIRT_FLASH space in the memmap.
887      * Any file passed via -bios goes in the first of these.
888      * sysmem is the system memory space. secure_sysmem is the secure view
889      * of the system, and the first flash device should be made visible only
890      * there. The second flash device is visible to both secure and nonsecure.
891      * If sysmem == secure_sysmem this means there is no separate Secure
892      * address space and both flash devices are generally visible.
893      */
894     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
895     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
896     char *nodename;
897 
898     create_one_flash("virt.flash0", flashbase, flashsize,
899                      bios_name, secure_sysmem);
900     create_one_flash("virt.flash1", flashbase + flashsize, flashsize,
901                      NULL, sysmem);
902 
903     if (sysmem == secure_sysmem) {
904         /* Report both flash devices as a single node in the DT */
905         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
906         qemu_fdt_add_subnode(vms->fdt, nodename);
907         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
908         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
909                                      2, flashbase, 2, flashsize,
910                                      2, flashbase + flashsize, 2, flashsize);
911         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
912         g_free(nodename);
913     } else {
914         /* Report the devices as separate nodes so we can mark one as
915          * only visible to the secure world.
916          */
917         nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase);
918         qemu_fdt_add_subnode(vms->fdt, nodename);
919         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
920         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
921                                      2, flashbase, 2, flashsize);
922         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
923         qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
924         qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
925         g_free(nodename);
926 
927         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
928         qemu_fdt_add_subnode(vms->fdt, nodename);
929         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
930         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
931                                      2, flashbase + flashsize, 2, flashsize);
932         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
933         g_free(nodename);
934     }
935 }
936 
937 static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as)
938 {
939     hwaddr base = vms->memmap[VIRT_FW_CFG].base;
940     hwaddr size = vms->memmap[VIRT_FW_CFG].size;
941     FWCfgState *fw_cfg;
942     char *nodename;
943 
944     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as);
945     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)smp_cpus);
946 
947     nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
948     qemu_fdt_add_subnode(vms->fdt, nodename);
949     qemu_fdt_setprop_string(vms->fdt, nodename,
950                             "compatible", "qemu,fw-cfg-mmio");
951     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
952                                  2, base, 2, size);
953     qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
954     g_free(nodename);
955     return fw_cfg;
956 }
957 
958 static void create_pcie_irq_map(const VirtMachineState *vms,
959                                 uint32_t gic_phandle,
960                                 int first_irq, const char *nodename)
961 {
962     int devfn, pin;
963     uint32_t full_irq_map[4 * 4 * 10] = { 0 };
964     uint32_t *irq_map = full_irq_map;
965 
966     for (devfn = 0; devfn <= 0x18; devfn += 0x8) {
967         for (pin = 0; pin < 4; pin++) {
968             int irq_type = GIC_FDT_IRQ_TYPE_SPI;
969             int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS);
970             int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
971             int i;
972 
973             uint32_t map[] = {
974                 devfn << 8, 0, 0,                           /* devfn */
975                 pin + 1,                                    /* PCI pin */
976                 gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */
977 
978             /* Convert map to big endian */
979             for (i = 0; i < 10; i++) {
980                 irq_map[i] = cpu_to_be32(map[i]);
981             }
982             irq_map += 10;
983         }
984     }
985 
986     qemu_fdt_setprop(vms->fdt, nodename, "interrupt-map",
987                      full_irq_map, sizeof(full_irq_map));
988 
989     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupt-map-mask",
990                            0x1800, 0, 0, /* devfn (PCI_SLOT(3)) */
991                            0x7           /* PCI irq */);
992 }
993 
994 static void create_pcie(const VirtMachineState *vms, qemu_irq *pic)
995 {
996     hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base;
997     hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size;
998     hwaddr base_mmio_high = vms->memmap[VIRT_PCIE_MMIO_HIGH].base;
999     hwaddr size_mmio_high = vms->memmap[VIRT_PCIE_MMIO_HIGH].size;
1000     hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base;
1001     hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size;
1002     hwaddr base_ecam = vms->memmap[VIRT_PCIE_ECAM].base;
1003     hwaddr size_ecam = vms->memmap[VIRT_PCIE_ECAM].size;
1004     hwaddr base = base_mmio;
1005     int nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN;
1006     int irq = vms->irqmap[VIRT_PCIE];
1007     MemoryRegion *mmio_alias;
1008     MemoryRegion *mmio_reg;
1009     MemoryRegion *ecam_alias;
1010     MemoryRegion *ecam_reg;
1011     DeviceState *dev;
1012     char *nodename;
1013     int i;
1014     PCIHostState *pci;
1015 
1016     dev = qdev_create(NULL, TYPE_GPEX_HOST);
1017     qdev_init_nofail(dev);
1018 
1019     /* Map only the first size_ecam bytes of ECAM space */
1020     ecam_alias = g_new0(MemoryRegion, 1);
1021     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1022     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1023                              ecam_reg, 0, size_ecam);
1024     memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias);
1025 
1026     /* Map the MMIO window into system address space so as to expose
1027      * the section of PCI MMIO space which starts at the same base address
1028      * (ie 1:1 mapping for that part of PCI MMIO space visible through
1029      * the window).
1030      */
1031     mmio_alias = g_new0(MemoryRegion, 1);
1032     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1033     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1034                              mmio_reg, base_mmio, size_mmio);
1035     memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias);
1036 
1037     if (vms->highmem) {
1038         /* Map high MMIO space */
1039         MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1);
1040 
1041         memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1042                                  mmio_reg, base_mmio_high, size_mmio_high);
1043         memory_region_add_subregion(get_system_memory(), base_mmio_high,
1044                                     high_mmio_alias);
1045     }
1046 
1047     /* Map IO port space */
1048     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio);
1049 
1050     for (i = 0; i < GPEX_NUM_IRQS; i++) {
1051         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]);
1052     }
1053 
1054     pci = PCI_HOST_BRIDGE(dev);
1055     if (pci->bus) {
1056         for (i = 0; i < nb_nics; i++) {
1057             NICInfo *nd = &nd_table[i];
1058 
1059             if (!nd->model) {
1060                 nd->model = g_strdup("virtio");
1061             }
1062 
1063             pci_nic_init_nofail(nd, pci->bus, nd->model, NULL);
1064         }
1065     }
1066 
1067     nodename = g_strdup_printf("/pcie@%" PRIx64, base);
1068     qemu_fdt_add_subnode(vms->fdt, nodename);
1069     qemu_fdt_setprop_string(vms->fdt, nodename,
1070                             "compatible", "pci-host-ecam-generic");
1071     qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "pci");
1072     qemu_fdt_setprop_cell(vms->fdt, nodename, "#address-cells", 3);
1073     qemu_fdt_setprop_cell(vms->fdt, nodename, "#size-cells", 2);
1074     qemu_fdt_setprop_cells(vms->fdt, nodename, "bus-range", 0,
1075                            nr_pcie_buses - 1);
1076     qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
1077 
1078     if (vms->msi_phandle) {
1079         qemu_fdt_setprop_cells(vms->fdt, nodename, "msi-parent",
1080                                vms->msi_phandle);
1081     }
1082 
1083     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
1084                                  2, base_ecam, 2, size_ecam);
1085 
1086     if (vms->highmem) {
1087         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges",
1088                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1089                                      2, base_pio, 2, size_pio,
1090                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1091                                      2, base_mmio, 2, size_mmio,
1092                                      1, FDT_PCI_RANGE_MMIO_64BIT,
1093                                      2, base_mmio_high,
1094                                      2, base_mmio_high, 2, size_mmio_high);
1095     } else {
1096         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges",
1097                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1098                                      2, base_pio, 2, size_pio,
1099                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1100                                      2, base_mmio, 2, size_mmio);
1101     }
1102 
1103     qemu_fdt_setprop_cell(vms->fdt, nodename, "#interrupt-cells", 1);
1104     create_pcie_irq_map(vms, vms->gic_phandle, irq, nodename);
1105 
1106     g_free(nodename);
1107 }
1108 
1109 static void create_platform_bus(VirtMachineState *vms, qemu_irq *pic)
1110 {
1111     DeviceState *dev;
1112     SysBusDevice *s;
1113     int i;
1114     ARMPlatformBusFDTParams *fdt_params = g_new(ARMPlatformBusFDTParams, 1);
1115     MemoryRegion *sysmem = get_system_memory();
1116 
1117     platform_bus_params.platform_bus_base = vms->memmap[VIRT_PLATFORM_BUS].base;
1118     platform_bus_params.platform_bus_size = vms->memmap[VIRT_PLATFORM_BUS].size;
1119     platform_bus_params.platform_bus_first_irq = vms->irqmap[VIRT_PLATFORM_BUS];
1120     platform_bus_params.platform_bus_num_irqs = PLATFORM_BUS_NUM_IRQS;
1121 
1122     fdt_params->system_params = &platform_bus_params;
1123     fdt_params->binfo = &vms->bootinfo;
1124     fdt_params->intc = "/intc";
1125     /*
1126      * register a machine init done notifier that creates the device tree
1127      * nodes of the platform bus and its children dynamic sysbus devices
1128      */
1129     arm_register_platform_bus_fdt_creator(fdt_params);
1130 
1131     dev = qdev_create(NULL, TYPE_PLATFORM_BUS_DEVICE);
1132     dev->id = TYPE_PLATFORM_BUS_DEVICE;
1133     qdev_prop_set_uint32(dev, "num_irqs",
1134         platform_bus_params.platform_bus_num_irqs);
1135     qdev_prop_set_uint32(dev, "mmio_size",
1136         platform_bus_params.platform_bus_size);
1137     qdev_init_nofail(dev);
1138     s = SYS_BUS_DEVICE(dev);
1139 
1140     for (i = 0; i < platform_bus_params.platform_bus_num_irqs; i++) {
1141         int irqn = platform_bus_params.platform_bus_first_irq + i;
1142         sysbus_connect_irq(s, i, pic[irqn]);
1143     }
1144 
1145     memory_region_add_subregion(sysmem,
1146                                 platform_bus_params.platform_bus_base,
1147                                 sysbus_mmio_get_region(s, 0));
1148 }
1149 
1150 static void create_secure_ram(VirtMachineState *vms,
1151                               MemoryRegion *secure_sysmem)
1152 {
1153     MemoryRegion *secram = g_new(MemoryRegion, 1);
1154     char *nodename;
1155     hwaddr base = vms->memmap[VIRT_SECURE_MEM].base;
1156     hwaddr size = vms->memmap[VIRT_SECURE_MEM].size;
1157 
1158     memory_region_init_ram(secram, NULL, "virt.secure-ram", size,
1159                            &error_fatal);
1160     memory_region_add_subregion(secure_sysmem, base, secram);
1161 
1162     nodename = g_strdup_printf("/secram@%" PRIx64, base);
1163     qemu_fdt_add_subnode(vms->fdt, nodename);
1164     qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "memory");
1165     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 2, base, 2, size);
1166     qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
1167     qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
1168 
1169     g_free(nodename);
1170 }
1171 
1172 static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size)
1173 {
1174     const VirtMachineState *board = container_of(binfo, VirtMachineState,
1175                                                  bootinfo);
1176 
1177     *fdt_size = board->fdt_size;
1178     return board->fdt;
1179 }
1180 
1181 static void virt_build_smbios(VirtMachineState *vms)
1182 {
1183     uint8_t *smbios_tables, *smbios_anchor;
1184     size_t smbios_tables_len, smbios_anchor_len;
1185     const char *product = "QEMU Virtual Machine";
1186 
1187     if (!vms->fw_cfg) {
1188         return;
1189     }
1190 
1191     if (kvm_enabled()) {
1192         product = "KVM Virtual Machine";
1193     }
1194 
1195     smbios_set_defaults("QEMU", product,
1196                         "1.0", false, true, SMBIOS_ENTRY_POINT_30);
1197 
1198     smbios_get_tables(NULL, 0, &smbios_tables, &smbios_tables_len,
1199                       &smbios_anchor, &smbios_anchor_len);
1200 
1201     if (smbios_anchor) {
1202         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables",
1203                         smbios_tables, smbios_tables_len);
1204         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor",
1205                         smbios_anchor, smbios_anchor_len);
1206     }
1207 }
1208 
1209 static
1210 void virt_machine_done(Notifier *notifier, void *data)
1211 {
1212     VirtMachineState *vms = container_of(notifier, VirtMachineState,
1213                                          machine_done);
1214 
1215     virt_acpi_setup(vms);
1216     virt_build_smbios(vms);
1217 }
1218 
1219 static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx)
1220 {
1221     uint8_t clustersz = ARM_DEFAULT_CPUS_PER_CLUSTER;
1222     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1223 
1224     if (!vmc->disallow_affinity_adjustment) {
1225         /* Adjust MPIDR like 64-bit KVM hosts, which incorporate the
1226          * GIC's target-list limitations. 32-bit KVM hosts currently
1227          * always create clusters of 4 CPUs, but that is expected to
1228          * change when they gain support for gicv3. When KVM is enabled
1229          * it will override the changes we make here, therefore our
1230          * purposes are to make TCG consistent (with 64-bit KVM hosts)
1231          * and to improve SGI efficiency.
1232          */
1233         if (vms->gic_version == 3) {
1234             clustersz = GICV3_TARGETLIST_BITS;
1235         } else {
1236             clustersz = GIC_TARGETLIST_BITS;
1237         }
1238     }
1239     return arm_cpu_mp_affinity(idx, clustersz);
1240 }
1241 
1242 static void machvirt_init(MachineState *machine)
1243 {
1244     VirtMachineState *vms = VIRT_MACHINE(machine);
1245     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine);
1246     MachineClass *mc = MACHINE_GET_CLASS(machine);
1247     const CPUArchIdList *possible_cpus;
1248     qemu_irq pic[NUM_IRQS];
1249     MemoryRegion *sysmem = get_system_memory();
1250     MemoryRegion *secure_sysmem = NULL;
1251     int n, virt_max_cpus;
1252     MemoryRegion *ram = g_new(MemoryRegion, 1);
1253     const char *cpu_model = machine->cpu_model;
1254     char **cpustr;
1255     ObjectClass *oc;
1256     const char *typename;
1257     CPUClass *cc;
1258     Error *err = NULL;
1259     bool firmware_loaded = bios_name || drive_get(IF_PFLASH, 0, 0);
1260 
1261     if (!cpu_model) {
1262         cpu_model = "cortex-a15";
1263     }
1264 
1265     /* We can probe only here because during property set
1266      * KVM is not available yet
1267      */
1268     if (!vms->gic_version) {
1269         if (!kvm_enabled()) {
1270             error_report("gic-version=host requires KVM");
1271             exit(1);
1272         }
1273 
1274         vms->gic_version = kvm_arm_vgic_probe();
1275         if (!vms->gic_version) {
1276             error_report("Unable to determine GIC version supported by host");
1277             exit(1);
1278         }
1279     }
1280 
1281     /* Separate the actual CPU model name from any appended features */
1282     cpustr = g_strsplit(cpu_model, ",", 2);
1283 
1284     if (!cpuname_valid(cpustr[0])) {
1285         error_report("mach-virt: CPU %s not supported", cpustr[0]);
1286         exit(1);
1287     }
1288 
1289     /* If we have an EL3 boot ROM then the assumption is that it will
1290      * implement PSCI itself, so disable QEMU's internal implementation
1291      * so it doesn't get in the way. Instead of starting secondary
1292      * CPUs in PSCI powerdown state we will start them all running and
1293      * let the boot ROM sort them out.
1294      * The usual case is that we do use QEMU's PSCI implementation;
1295      * if the guest has EL2 then we will use SMC as the conduit,
1296      * and otherwise we will use HVC (for backwards compatibility and
1297      * because if we're using KVM then we must use HVC).
1298      */
1299     if (vms->secure && firmware_loaded) {
1300         vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
1301     } else if (vms->virt) {
1302         vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC;
1303     } else {
1304         vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC;
1305     }
1306 
1307     /* The maximum number of CPUs depends on the GIC version, or on how
1308      * many redistributors we can fit into the memory map.
1309      */
1310     if (vms->gic_version == 3) {
1311         virt_max_cpus = vms->memmap[VIRT_GIC_REDIST].size / 0x20000;
1312     } else {
1313         virt_max_cpus = GIC_NCPU;
1314     }
1315 
1316     if (max_cpus > virt_max_cpus) {
1317         error_report("Number of SMP CPUs requested (%d) exceeds max CPUs "
1318                      "supported by machine 'mach-virt' (%d)",
1319                      max_cpus, virt_max_cpus);
1320         exit(1);
1321     }
1322 
1323     vms->smp_cpus = smp_cpus;
1324 
1325     if (machine->ram_size > vms->memmap[VIRT_MEM].size) {
1326         error_report("mach-virt: cannot model more than %dGB RAM", RAMLIMIT_GB);
1327         exit(1);
1328     }
1329 
1330     if (vms->virt && kvm_enabled()) {
1331         error_report("mach-virt: KVM does not support providing "
1332                      "Virtualization extensions to the guest CPU");
1333         exit(1);
1334     }
1335 
1336     if (vms->secure) {
1337         if (kvm_enabled()) {
1338             error_report("mach-virt: KVM does not support Security extensions");
1339             exit(1);
1340         }
1341 
1342         /* The Secure view of the world is the same as the NonSecure,
1343          * but with a few extra devices. Create it as a container region
1344          * containing the system memory at low priority; any secure-only
1345          * devices go in at higher priority and take precedence.
1346          */
1347         secure_sysmem = g_new(MemoryRegion, 1);
1348         memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory",
1349                            UINT64_MAX);
1350         memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1);
1351     }
1352 
1353     create_fdt(vms);
1354 
1355     oc = cpu_class_by_name(TYPE_ARM_CPU, cpustr[0]);
1356     if (!oc) {
1357         error_report("Unable to find CPU definition");
1358         exit(1);
1359     }
1360     typename = object_class_get_name(oc);
1361 
1362     /* convert -smp CPU options specified by the user into global props */
1363     cc = CPU_CLASS(oc);
1364     cc->parse_features(typename, cpustr[1], &err);
1365     g_strfreev(cpustr);
1366     if (err) {
1367         error_report_err(err);
1368         exit(1);
1369     }
1370 
1371     possible_cpus = mc->possible_cpu_arch_ids(machine);
1372     for (n = 0; n < possible_cpus->len; n++) {
1373         Object *cpuobj;
1374         CPUState *cs;
1375 
1376         if (n >= smp_cpus) {
1377             break;
1378         }
1379 
1380         cpuobj = object_new(typename);
1381         object_property_set_int(cpuobj, possible_cpus->cpus[n].arch_id,
1382                                 "mp-affinity", NULL);
1383 
1384         cs = CPU(cpuobj);
1385         cs->cpu_index = n;
1386 
1387         numa_cpu_pre_plug(&possible_cpus->cpus[cs->cpu_index], DEVICE(cpuobj),
1388                           &error_fatal);
1389 
1390         if (!vms->secure) {
1391             object_property_set_bool(cpuobj, false, "has_el3", NULL);
1392         }
1393 
1394         if (!vms->virt && object_property_find(cpuobj, "has_el2", NULL)) {
1395             object_property_set_bool(cpuobj, false, "has_el2", NULL);
1396         }
1397 
1398         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) {
1399             object_property_set_int(cpuobj, vms->psci_conduit,
1400                                     "psci-conduit", NULL);
1401 
1402             /* Secondary CPUs start in PSCI powered-down state */
1403             if (n > 0) {
1404                 object_property_set_bool(cpuobj, true,
1405                                          "start-powered-off", NULL);
1406             }
1407         }
1408 
1409         if (vmc->no_pmu && object_property_find(cpuobj, "pmu", NULL)) {
1410             object_property_set_bool(cpuobj, false, "pmu", NULL);
1411         }
1412 
1413         if (object_property_find(cpuobj, "reset-cbar", NULL)) {
1414             object_property_set_int(cpuobj, vms->memmap[VIRT_CPUPERIPHS].base,
1415                                     "reset-cbar", &error_abort);
1416         }
1417 
1418         object_property_set_link(cpuobj, OBJECT(sysmem), "memory",
1419                                  &error_abort);
1420         if (vms->secure) {
1421             object_property_set_link(cpuobj, OBJECT(secure_sysmem),
1422                                      "secure-memory", &error_abort);
1423         }
1424 
1425         object_property_set_bool(cpuobj, true, "realized", NULL);
1426         object_unref(cpuobj);
1427     }
1428     fdt_add_timer_nodes(vms);
1429     fdt_add_cpu_nodes(vms);
1430     fdt_add_psci_node(vms);
1431 
1432     memory_region_allocate_system_memory(ram, NULL, "mach-virt.ram",
1433                                          machine->ram_size);
1434     memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base, ram);
1435 
1436     create_flash(vms, sysmem, secure_sysmem ? secure_sysmem : sysmem);
1437 
1438     create_gic(vms, pic);
1439 
1440     fdt_add_pmu_nodes(vms);
1441 
1442     create_uart(vms, pic, VIRT_UART, sysmem, serial_hds[0]);
1443 
1444     if (vms->secure) {
1445         create_secure_ram(vms, secure_sysmem);
1446         create_uart(vms, pic, VIRT_SECURE_UART, secure_sysmem, serial_hds[1]);
1447     }
1448 
1449     create_rtc(vms, pic);
1450 
1451     create_pcie(vms, pic);
1452 
1453     create_gpio(vms, pic);
1454 
1455     /* Create mmio transports, so the user can create virtio backends
1456      * (which will be automatically plugged in to the transports). If
1457      * no backend is created the transport will just sit harmlessly idle.
1458      */
1459     create_virtio_devices(vms, pic);
1460 
1461     vms->fw_cfg = create_fw_cfg(vms, &address_space_memory);
1462     rom_set_fw(vms->fw_cfg);
1463 
1464     vms->machine_done.notify = virt_machine_done;
1465     qemu_add_machine_init_done_notifier(&vms->machine_done);
1466 
1467     vms->bootinfo.ram_size = machine->ram_size;
1468     vms->bootinfo.kernel_filename = machine->kernel_filename;
1469     vms->bootinfo.kernel_cmdline = machine->kernel_cmdline;
1470     vms->bootinfo.initrd_filename = machine->initrd_filename;
1471     vms->bootinfo.nb_cpus = smp_cpus;
1472     vms->bootinfo.board_id = -1;
1473     vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base;
1474     vms->bootinfo.get_dtb = machvirt_dtb;
1475     vms->bootinfo.firmware_loaded = firmware_loaded;
1476     arm_load_kernel(ARM_CPU(first_cpu), &vms->bootinfo);
1477 
1478     /*
1479      * arm_load_kernel machine init done notifier registration must
1480      * happen before the platform_bus_create call. In this latter,
1481      * another notifier is registered which adds platform bus nodes.
1482      * Notifiers are executed in registration reverse order.
1483      */
1484     create_platform_bus(vms, pic);
1485 }
1486 
1487 static bool virt_get_secure(Object *obj, Error **errp)
1488 {
1489     VirtMachineState *vms = VIRT_MACHINE(obj);
1490 
1491     return vms->secure;
1492 }
1493 
1494 static void virt_set_secure(Object *obj, bool value, Error **errp)
1495 {
1496     VirtMachineState *vms = VIRT_MACHINE(obj);
1497 
1498     vms->secure = value;
1499 }
1500 
1501 static bool virt_get_virt(Object *obj, Error **errp)
1502 {
1503     VirtMachineState *vms = VIRT_MACHINE(obj);
1504 
1505     return vms->virt;
1506 }
1507 
1508 static void virt_set_virt(Object *obj, bool value, Error **errp)
1509 {
1510     VirtMachineState *vms = VIRT_MACHINE(obj);
1511 
1512     vms->virt = value;
1513 }
1514 
1515 static bool virt_get_highmem(Object *obj, Error **errp)
1516 {
1517     VirtMachineState *vms = VIRT_MACHINE(obj);
1518 
1519     return vms->highmem;
1520 }
1521 
1522 static void virt_set_highmem(Object *obj, bool value, Error **errp)
1523 {
1524     VirtMachineState *vms = VIRT_MACHINE(obj);
1525 
1526     vms->highmem = value;
1527 }
1528 
1529 static bool virt_get_its(Object *obj, Error **errp)
1530 {
1531     VirtMachineState *vms = VIRT_MACHINE(obj);
1532 
1533     return vms->its;
1534 }
1535 
1536 static void virt_set_its(Object *obj, bool value, Error **errp)
1537 {
1538     VirtMachineState *vms = VIRT_MACHINE(obj);
1539 
1540     vms->its = value;
1541 }
1542 
1543 static char *virt_get_gic_version(Object *obj, Error **errp)
1544 {
1545     VirtMachineState *vms = VIRT_MACHINE(obj);
1546     const char *val = vms->gic_version == 3 ? "3" : "2";
1547 
1548     return g_strdup(val);
1549 }
1550 
1551 static void virt_set_gic_version(Object *obj, const char *value, Error **errp)
1552 {
1553     VirtMachineState *vms = VIRT_MACHINE(obj);
1554 
1555     if (!strcmp(value, "3")) {
1556         vms->gic_version = 3;
1557     } else if (!strcmp(value, "2")) {
1558         vms->gic_version = 2;
1559     } else if (!strcmp(value, "host")) {
1560         vms->gic_version = 0; /* Will probe later */
1561     } else {
1562         error_setg(errp, "Invalid gic-version value");
1563         error_append_hint(errp, "Valid values are 3, 2, host.\n");
1564     }
1565 }
1566 
1567 static CpuInstanceProperties
1568 virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
1569 {
1570     MachineClass *mc = MACHINE_GET_CLASS(ms);
1571     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
1572 
1573     assert(cpu_index < possible_cpus->len);
1574     return possible_cpus->cpus[cpu_index].props;
1575 }
1576 
1577 static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms)
1578 {
1579     int n;
1580     VirtMachineState *vms = VIRT_MACHINE(ms);
1581 
1582     if (ms->possible_cpus) {
1583         assert(ms->possible_cpus->len == max_cpus);
1584         return ms->possible_cpus;
1585     }
1586 
1587     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
1588                                   sizeof(CPUArchId) * max_cpus);
1589     ms->possible_cpus->len = max_cpus;
1590     for (n = 0; n < ms->possible_cpus->len; n++) {
1591         ms->possible_cpus->cpus[n].arch_id =
1592             virt_cpu_mp_affinity(vms, n);
1593         ms->possible_cpus->cpus[n].props.has_thread_id = true;
1594         ms->possible_cpus->cpus[n].props.thread_id = n;
1595 
1596         /* default distribution of CPUs over NUMA nodes */
1597         if (nb_numa_nodes) {
1598             /* preset values but do not enable them i.e. 'has_node_id = false',
1599              * numa init code will enable them later if manual mapping wasn't
1600              * present on CLI */
1601             ms->possible_cpus->cpus[n].props.node_id = n % nb_numa_nodes;
1602         }
1603     }
1604     return ms->possible_cpus;
1605 }
1606 
1607 static void virt_machine_class_init(ObjectClass *oc, void *data)
1608 {
1609     MachineClass *mc = MACHINE_CLASS(oc);
1610 
1611     mc->init = machvirt_init;
1612     /* Start max_cpus at the maximum QEMU supports. We'll further restrict
1613      * it later in machvirt_init, where we have more information about the
1614      * configuration of the particular instance.
1615      */
1616     mc->max_cpus = 255;
1617     mc->has_dynamic_sysbus = true;
1618     mc->block_default_type = IF_VIRTIO;
1619     mc->no_cdrom = 1;
1620     mc->pci_allow_0_address = true;
1621     /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */
1622     mc->minimum_page_bits = 12;
1623     mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids;
1624     mc->cpu_index_to_instance_props = virt_cpu_index_to_props;
1625 }
1626 
1627 static const TypeInfo virt_machine_info = {
1628     .name          = TYPE_VIRT_MACHINE,
1629     .parent        = TYPE_MACHINE,
1630     .abstract      = true,
1631     .instance_size = sizeof(VirtMachineState),
1632     .class_size    = sizeof(VirtMachineClass),
1633     .class_init    = virt_machine_class_init,
1634 };
1635 
1636 static void machvirt_machine_init(void)
1637 {
1638     type_register_static(&virt_machine_info);
1639 }
1640 type_init(machvirt_machine_init);
1641 
1642 static void virt_2_10_instance_init(Object *obj)
1643 {
1644     VirtMachineState *vms = VIRT_MACHINE(obj);
1645     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1646 
1647     /* EL3 is disabled by default on virt: this makes us consistent
1648      * between KVM and TCG for this board, and it also allows us to
1649      * boot UEFI blobs which assume no TrustZone support.
1650      */
1651     vms->secure = false;
1652     object_property_add_bool(obj, "secure", virt_get_secure,
1653                              virt_set_secure, NULL);
1654     object_property_set_description(obj, "secure",
1655                                     "Set on/off to enable/disable the ARM "
1656                                     "Security Extensions (TrustZone)",
1657                                     NULL);
1658 
1659     /* EL2 is also disabled by default, for similar reasons */
1660     vms->virt = false;
1661     object_property_add_bool(obj, "virtualization", virt_get_virt,
1662                              virt_set_virt, NULL);
1663     object_property_set_description(obj, "virtualization",
1664                                     "Set on/off to enable/disable emulating a "
1665                                     "guest CPU which implements the ARM "
1666                                     "Virtualization Extensions",
1667                                     NULL);
1668 
1669     /* High memory is enabled by default */
1670     vms->highmem = true;
1671     object_property_add_bool(obj, "highmem", virt_get_highmem,
1672                              virt_set_highmem, NULL);
1673     object_property_set_description(obj, "highmem",
1674                                     "Set on/off to enable/disable using "
1675                                     "physical address space above 32 bits",
1676                                     NULL);
1677     /* Default GIC type is v2 */
1678     vms->gic_version = 2;
1679     object_property_add_str(obj, "gic-version", virt_get_gic_version,
1680                         virt_set_gic_version, NULL);
1681     object_property_set_description(obj, "gic-version",
1682                                     "Set GIC version. "
1683                                     "Valid values are 2, 3 and host", NULL);
1684 
1685     if (vmc->no_its) {
1686         vms->its = false;
1687     } else {
1688         /* Default allows ITS instantiation */
1689         vms->its = true;
1690         object_property_add_bool(obj, "its", virt_get_its,
1691                                  virt_set_its, NULL);
1692         object_property_set_description(obj, "its",
1693                                         "Set on/off to enable/disable "
1694                                         "ITS instantiation",
1695                                         NULL);
1696     }
1697 
1698     vms->memmap = a15memmap;
1699     vms->irqmap = a15irqmap;
1700 }
1701 
1702 static void virt_machine_2_10_options(MachineClass *mc)
1703 {
1704 }
1705 DEFINE_VIRT_MACHINE_AS_LATEST(2, 10)
1706 
1707 #define VIRT_COMPAT_2_9 \
1708     HW_COMPAT_2_9
1709 
1710 static void virt_2_9_instance_init(Object *obj)
1711 {
1712     virt_2_10_instance_init(obj);
1713 }
1714 
1715 static void virt_machine_2_9_options(MachineClass *mc)
1716 {
1717     virt_machine_2_10_options(mc);
1718     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_9);
1719 }
1720 DEFINE_VIRT_MACHINE(2, 9)
1721 
1722 #define VIRT_COMPAT_2_8 \
1723     HW_COMPAT_2_8
1724 
1725 static void virt_2_8_instance_init(Object *obj)
1726 {
1727     virt_2_9_instance_init(obj);
1728 }
1729 
1730 static void virt_machine_2_8_options(MachineClass *mc)
1731 {
1732     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1733 
1734     virt_machine_2_9_options(mc);
1735     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_8);
1736     /* For 2.8 and earlier we falsely claimed in the DT that
1737      * our timers were edge-triggered, not level-triggered.
1738      */
1739     vmc->claim_edge_triggered_timers = true;
1740 }
1741 DEFINE_VIRT_MACHINE(2, 8)
1742 
1743 #define VIRT_COMPAT_2_7 \
1744     HW_COMPAT_2_7
1745 
1746 static void virt_2_7_instance_init(Object *obj)
1747 {
1748     virt_2_8_instance_init(obj);
1749 }
1750 
1751 static void virt_machine_2_7_options(MachineClass *mc)
1752 {
1753     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1754 
1755     virt_machine_2_8_options(mc);
1756     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_7);
1757     /* ITS was introduced with 2.8 */
1758     vmc->no_its = true;
1759     /* Stick with 1K pages for migration compatibility */
1760     mc->minimum_page_bits = 0;
1761 }
1762 DEFINE_VIRT_MACHINE(2, 7)
1763 
1764 #define VIRT_COMPAT_2_6 \
1765     HW_COMPAT_2_6
1766 
1767 static void virt_2_6_instance_init(Object *obj)
1768 {
1769     virt_2_7_instance_init(obj);
1770 }
1771 
1772 static void virt_machine_2_6_options(MachineClass *mc)
1773 {
1774     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1775 
1776     virt_machine_2_7_options(mc);
1777     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_6);
1778     vmc->disallow_affinity_adjustment = true;
1779     /* Disable PMU for 2.6 as PMU support was first introduced in 2.7 */
1780     vmc->no_pmu = true;
1781 }
1782 DEFINE_VIRT_MACHINE(2, 6)
1783