xref: /openbmc/qemu/hw/arm/virt.c (revision 9277d81f5c2c6f4d0b5e47c8476eb7ee7e5c0beb)
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/vfio/vfio-calxeda-xgmac.h"
38 #include "hw/vfio/vfio-amd-xgbe.h"
39 #include "hw/display/ramfb.h"
40 #include "hw/devices.h"
41 #include "net/net.h"
42 #include "sysemu/device_tree.h"
43 #include "sysemu/numa.h"
44 #include "sysemu/sysemu.h"
45 #include "sysemu/kvm.h"
46 #include "hw/compat.h"
47 #include "hw/loader.h"
48 #include "exec/address-spaces.h"
49 #include "qemu/bitops.h"
50 #include "qemu/error-report.h"
51 #include "hw/pci-host/gpex.h"
52 #include "hw/arm/sysbus-fdt.h"
53 #include "hw/platform-bus.h"
54 #include "hw/arm/fdt.h"
55 #include "hw/intc/arm_gic.h"
56 #include "hw/intc/arm_gicv3_common.h"
57 #include "kvm_arm.h"
58 #include "hw/smbios/smbios.h"
59 #include "qapi/visitor.h"
60 #include "standard-headers/linux/input.h"
61 #include "hw/arm/smmuv3.h"
62 
63 #define DEFINE_VIRT_MACHINE_LATEST(major, minor, latest) \
64     static void virt_##major##_##minor##_class_init(ObjectClass *oc, \
65                                                     void *data) \
66     { \
67         MachineClass *mc = MACHINE_CLASS(oc); \
68         virt_machine_##major##_##minor##_options(mc); \
69         mc->desc = "QEMU " # major "." # minor " ARM Virtual Machine"; \
70         if (latest) { \
71             mc->alias = "virt"; \
72         } \
73     } \
74     static const TypeInfo machvirt_##major##_##minor##_info = { \
75         .name = MACHINE_TYPE_NAME("virt-" # major "." # minor), \
76         .parent = TYPE_VIRT_MACHINE, \
77         .instance_init = virt_##major##_##minor##_instance_init, \
78         .class_init = virt_##major##_##minor##_class_init, \
79     }; \
80     static void machvirt_machine_##major##_##minor##_init(void) \
81     { \
82         type_register_static(&machvirt_##major##_##minor##_info); \
83     } \
84     type_init(machvirt_machine_##major##_##minor##_init);
85 
86 #define DEFINE_VIRT_MACHINE_AS_LATEST(major, minor) \
87     DEFINE_VIRT_MACHINE_LATEST(major, minor, true)
88 #define DEFINE_VIRT_MACHINE(major, minor) \
89     DEFINE_VIRT_MACHINE_LATEST(major, minor, false)
90 
91 
92 /* Number of external interrupt lines to configure the GIC with */
93 #define NUM_IRQS 256
94 
95 #define PLATFORM_BUS_NUM_IRQS 64
96 
97 /* RAM limit in GB. Since VIRT_MEM starts at the 1GB mark, this means
98  * RAM can go up to the 256GB mark, leaving 256GB of the physical
99  * address space unallocated and free for future use between 256G and 512G.
100  * If we need to provide more RAM to VMs in the future then we need to:
101  *  * allocate a second bank of RAM starting at 2TB and working up
102  *  * fix the DT and ACPI table generation code in QEMU to correctly
103  *    report two split lumps of RAM to the guest
104  *  * fix KVM in the host kernel to allow guests with >40 bit address spaces
105  * (We don't want to fill all the way up to 512GB with RAM because
106  * we might want it for non-RAM purposes later. Conversely it seems
107  * reasonable to assume that anybody configuring a VM with a quarter
108  * of a terabyte of RAM will be doing it on a host with more than a
109  * terabyte of physical address space.)
110  */
111 #define RAMLIMIT_GB 255
112 #define RAMLIMIT_BYTES (RAMLIMIT_GB * 1024ULL * 1024 * 1024)
113 
114 /* Addresses and sizes of our components.
115  * 0..128MB is space for a flash device so we can run bootrom code such as UEFI.
116  * 128MB..256MB is used for miscellaneous device I/O.
117  * 256MB..1GB is reserved for possible future PCI support (ie where the
118  * PCI memory window will go if we add a PCI host controller).
119  * 1GB and up is RAM (which may happily spill over into the
120  * high memory region beyond 4GB).
121  * This represents a compromise between how much RAM can be given to
122  * a 32 bit VM and leaving space for expansion and in particular for PCI.
123  * Note that devices should generally be placed at multiples of 0x10000,
124  * to accommodate guests using 64K pages.
125  */
126 static const MemMapEntry a15memmap[] = {
127     /* Space up to 0x8000000 is reserved for a boot ROM */
128     [VIRT_FLASH] =              {          0, 0x08000000 },
129     [VIRT_CPUPERIPHS] =         { 0x08000000, 0x00020000 },
130     /* GIC distributor and CPU interfaces sit inside the CPU peripheral space */
131     [VIRT_GIC_DIST] =           { 0x08000000, 0x00010000 },
132     [VIRT_GIC_CPU] =            { 0x08010000, 0x00010000 },
133     [VIRT_GIC_V2M] =            { 0x08020000, 0x00001000 },
134     /* The space in between here is reserved for GICv3 CPU/vCPU/HYP */
135     [VIRT_GIC_ITS] =            { 0x08080000, 0x00020000 },
136     /* This redistributor space allows up to 2*64kB*123 CPUs */
137     [VIRT_GIC_REDIST] =         { 0x080A0000, 0x00F60000 },
138     [VIRT_UART] =               { 0x09000000, 0x00001000 },
139     [VIRT_RTC] =                { 0x09010000, 0x00001000 },
140     [VIRT_FW_CFG] =             { 0x09020000, 0x00000018 },
141     [VIRT_GPIO] =               { 0x09030000, 0x00001000 },
142     [VIRT_SECURE_UART] =        { 0x09040000, 0x00001000 },
143     [VIRT_SMMU] =               { 0x09050000, 0x00020000 },
144     [VIRT_MMIO] =               { 0x0a000000, 0x00000200 },
145     /* ...repeating for a total of NUM_VIRTIO_TRANSPORTS, each of that size */
146     [VIRT_PLATFORM_BUS] =       { 0x0c000000, 0x02000000 },
147     [VIRT_SECURE_MEM] =         { 0x0e000000, 0x01000000 },
148     [VIRT_PCIE_MMIO] =          { 0x10000000, 0x2eff0000 },
149     [VIRT_PCIE_PIO] =           { 0x3eff0000, 0x00010000 },
150     [VIRT_PCIE_ECAM] =          { 0x3f000000, 0x01000000 },
151     [VIRT_MEM] =                { 0x40000000, RAMLIMIT_BYTES },
152     /* Additional 64 MB redist region (can contain up to 512 redistributors) */
153     [VIRT_GIC_REDIST2] =        { 0x4000000000ULL, 0x4000000 },
154     [VIRT_PCIE_ECAM_HIGH] =     { 0x4010000000ULL, 0x10000000 },
155     /* Second PCIe window, 512GB wide at the 512GB boundary */
156     [VIRT_PCIE_MMIO_HIGH] =   { 0x8000000000ULL, 0x8000000000ULL },
157 };
158 
159 static const int a15irqmap[] = {
160     [VIRT_UART] = 1,
161     [VIRT_RTC] = 2,
162     [VIRT_PCIE] = 3, /* ... to 6 */
163     [VIRT_GPIO] = 7,
164     [VIRT_SECURE_UART] = 8,
165     [VIRT_MMIO] = 16, /* ...to 16 + NUM_VIRTIO_TRANSPORTS - 1 */
166     [VIRT_GIC_V2M] = 48, /* ...to 48 + NUM_GICV2M_SPIS - 1 */
167     [VIRT_SMMU] = 74,    /* ...to 74 + NUM_SMMU_IRQS - 1 */
168     [VIRT_PLATFORM_BUS] = 112, /* ...to 112 + PLATFORM_BUS_NUM_IRQS -1 */
169 };
170 
171 static const char *valid_cpus[] = {
172     ARM_CPU_TYPE_NAME("cortex-a15"),
173     ARM_CPU_TYPE_NAME("cortex-a53"),
174     ARM_CPU_TYPE_NAME("cortex-a57"),
175     ARM_CPU_TYPE_NAME("host"),
176     ARM_CPU_TYPE_NAME("max"),
177 };
178 
179 static bool cpu_type_valid(const char *cpu)
180 {
181     int i;
182 
183     for (i = 0; i < ARRAY_SIZE(valid_cpus); i++) {
184         if (strcmp(cpu, valid_cpus[i]) == 0) {
185             return true;
186         }
187     }
188     return false;
189 }
190 
191 static void create_fdt(VirtMachineState *vms)
192 {
193     void *fdt = create_device_tree(&vms->fdt_size);
194 
195     if (!fdt) {
196         error_report("create_device_tree() failed");
197         exit(1);
198     }
199 
200     vms->fdt = fdt;
201 
202     /* Header */
203     qemu_fdt_setprop_string(fdt, "/", "compatible", "linux,dummy-virt");
204     qemu_fdt_setprop_cell(fdt, "/", "#address-cells", 0x2);
205     qemu_fdt_setprop_cell(fdt, "/", "#size-cells", 0x2);
206 
207     /* /chosen must exist for load_dtb to fill in necessary properties later */
208     qemu_fdt_add_subnode(fdt, "/chosen");
209 
210     /* Clock node, for the benefit of the UART. The kernel device tree
211      * binding documentation claims the PL011 node clock properties are
212      * optional but in practice if you omit them the kernel refuses to
213      * probe for the device.
214      */
215     vms->clock_phandle = qemu_fdt_alloc_phandle(fdt);
216     qemu_fdt_add_subnode(fdt, "/apb-pclk");
217     qemu_fdt_setprop_string(fdt, "/apb-pclk", "compatible", "fixed-clock");
218     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "#clock-cells", 0x0);
219     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "clock-frequency", 24000000);
220     qemu_fdt_setprop_string(fdt, "/apb-pclk", "clock-output-names",
221                                 "clk24mhz");
222     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "phandle", vms->clock_phandle);
223 
224     if (have_numa_distance) {
225         int size = nb_numa_nodes * nb_numa_nodes * 3 * sizeof(uint32_t);
226         uint32_t *matrix = g_malloc0(size);
227         int idx, i, j;
228 
229         for (i = 0; i < nb_numa_nodes; i++) {
230             for (j = 0; j < nb_numa_nodes; j++) {
231                 idx = (i * nb_numa_nodes + j) * 3;
232                 matrix[idx + 0] = cpu_to_be32(i);
233                 matrix[idx + 1] = cpu_to_be32(j);
234                 matrix[idx + 2] = cpu_to_be32(numa_info[i].distance[j]);
235             }
236         }
237 
238         qemu_fdt_add_subnode(fdt, "/distance-map");
239         qemu_fdt_setprop_string(fdt, "/distance-map", "compatible",
240                                 "numa-distance-map-v1");
241         qemu_fdt_setprop(fdt, "/distance-map", "distance-matrix",
242                          matrix, size);
243         g_free(matrix);
244     }
245 }
246 
247 static void fdt_add_timer_nodes(const VirtMachineState *vms)
248 {
249     /* On real hardware these interrupts are level-triggered.
250      * On KVM they were edge-triggered before host kernel version 4.4,
251      * and level-triggered afterwards.
252      * On emulated QEMU they are level-triggered.
253      *
254      * Getting the DTB info about them wrong is awkward for some
255      * guest kernels:
256      *  pre-4.8 ignore the DT and leave the interrupt configured
257      *   with whatever the GIC reset value (or the bootloader) left it at
258      *  4.8 before rc6 honour the incorrect data by programming it back
259      *   into the GIC, causing problems
260      *  4.8rc6 and later ignore the DT and always write "level triggered"
261      *   into the GIC
262      *
263      * For backwards-compatibility, virt-2.8 and earlier will continue
264      * to say these are edge-triggered, but later machines will report
265      * the correct information.
266      */
267     ARMCPU *armcpu;
268     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
269     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
270 
271     if (vmc->claim_edge_triggered_timers) {
272         irqflags = GIC_FDT_IRQ_FLAGS_EDGE_LO_HI;
273     }
274 
275     if (vms->gic_version == 2) {
276         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
277                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
278                              (1 << vms->smp_cpus) - 1);
279     }
280 
281     qemu_fdt_add_subnode(vms->fdt, "/timer");
282 
283     armcpu = ARM_CPU(qemu_get_cpu(0));
284     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
285         const char compat[] = "arm,armv8-timer\0arm,armv7-timer";
286         qemu_fdt_setprop(vms->fdt, "/timer", "compatible",
287                          compat, sizeof(compat));
288     } else {
289         qemu_fdt_setprop_string(vms->fdt, "/timer", "compatible",
290                                 "arm,armv7-timer");
291     }
292     qemu_fdt_setprop(vms->fdt, "/timer", "always-on", NULL, 0);
293     qemu_fdt_setprop_cells(vms->fdt, "/timer", "interrupts",
294                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_S_EL1_IRQ, irqflags,
295                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL1_IRQ, irqflags,
296                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_VIRT_IRQ, irqflags,
297                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL2_IRQ, irqflags);
298 }
299 
300 static void fdt_add_cpu_nodes(const VirtMachineState *vms)
301 {
302     int cpu;
303     int addr_cells = 1;
304     const MachineState *ms = MACHINE(vms);
305 
306     /*
307      * From Documentation/devicetree/bindings/arm/cpus.txt
308      *  On ARM v8 64-bit systems value should be set to 2,
309      *  that corresponds to the MPIDR_EL1 register size.
310      *  If MPIDR_EL1[63:32] value is equal to 0 on all CPUs
311      *  in the system, #address-cells can be set to 1, since
312      *  MPIDR_EL1[63:32] bits are not used for CPUs
313      *  identification.
314      *
315      *  Here we actually don't know whether our system is 32- or 64-bit one.
316      *  The simplest way to go is to examine affinity IDs of all our CPUs. If
317      *  at least one of them has Aff3 populated, we set #address-cells to 2.
318      */
319     for (cpu = 0; cpu < vms->smp_cpus; cpu++) {
320         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
321 
322         if (armcpu->mp_affinity & ARM_AFF3_MASK) {
323             addr_cells = 2;
324             break;
325         }
326     }
327 
328     qemu_fdt_add_subnode(vms->fdt, "/cpus");
329     qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#address-cells", addr_cells);
330     qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#size-cells", 0x0);
331 
332     for (cpu = vms->smp_cpus - 1; cpu >= 0; cpu--) {
333         char *nodename = g_strdup_printf("/cpus/cpu@%d", cpu);
334         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
335         CPUState *cs = CPU(armcpu);
336 
337         qemu_fdt_add_subnode(vms->fdt, nodename);
338         qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "cpu");
339         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible",
340                                     armcpu->dtb_compatible);
341 
342         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED
343             && vms->smp_cpus > 1) {
344             qemu_fdt_setprop_string(vms->fdt, nodename,
345                                         "enable-method", "psci");
346         }
347 
348         if (addr_cells == 2) {
349             qemu_fdt_setprop_u64(vms->fdt, nodename, "reg",
350                                  armcpu->mp_affinity);
351         } else {
352             qemu_fdt_setprop_cell(vms->fdt, nodename, "reg",
353                                   armcpu->mp_affinity);
354         }
355 
356         if (ms->possible_cpus->cpus[cs->cpu_index].props.has_node_id) {
357             qemu_fdt_setprop_cell(vms->fdt, nodename, "numa-node-id",
358                 ms->possible_cpus->cpus[cs->cpu_index].props.node_id);
359         }
360 
361         g_free(nodename);
362     }
363 }
364 
365 static void fdt_add_its_gic_node(VirtMachineState *vms)
366 {
367     char *nodename;
368 
369     vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt);
370     nodename = g_strdup_printf("/intc/its@%" PRIx64,
371                                vms->memmap[VIRT_GIC_ITS].base);
372     qemu_fdt_add_subnode(vms->fdt, nodename);
373     qemu_fdt_setprop_string(vms->fdt, nodename, "compatible",
374                             "arm,gic-v3-its");
375     qemu_fdt_setprop(vms->fdt, nodename, "msi-controller", NULL, 0);
376     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
377                                  2, vms->memmap[VIRT_GIC_ITS].base,
378                                  2, vms->memmap[VIRT_GIC_ITS].size);
379     qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", vms->msi_phandle);
380     g_free(nodename);
381 }
382 
383 static void fdt_add_v2m_gic_node(VirtMachineState *vms)
384 {
385     char *nodename;
386 
387     nodename = g_strdup_printf("/intc/v2m@%" PRIx64,
388                                vms->memmap[VIRT_GIC_V2M].base);
389     vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt);
390     qemu_fdt_add_subnode(vms->fdt, nodename);
391     qemu_fdt_setprop_string(vms->fdt, nodename, "compatible",
392                             "arm,gic-v2m-frame");
393     qemu_fdt_setprop(vms->fdt, nodename, "msi-controller", NULL, 0);
394     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
395                                  2, vms->memmap[VIRT_GIC_V2M].base,
396                                  2, vms->memmap[VIRT_GIC_V2M].size);
397     qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", vms->msi_phandle);
398     g_free(nodename);
399 }
400 
401 static void fdt_add_gic_node(VirtMachineState *vms)
402 {
403     char *nodename;
404 
405     vms->gic_phandle = qemu_fdt_alloc_phandle(vms->fdt);
406     qemu_fdt_setprop_cell(vms->fdt, "/", "interrupt-parent", vms->gic_phandle);
407 
408     nodename = g_strdup_printf("/intc@%" PRIx64,
409                                vms->memmap[VIRT_GIC_DIST].base);
410     qemu_fdt_add_subnode(vms->fdt, nodename);
411     qemu_fdt_setprop_cell(vms->fdt, nodename, "#interrupt-cells", 3);
412     qemu_fdt_setprop(vms->fdt, nodename, "interrupt-controller", NULL, 0);
413     qemu_fdt_setprop_cell(vms->fdt, nodename, "#address-cells", 0x2);
414     qemu_fdt_setprop_cell(vms->fdt, nodename, "#size-cells", 0x2);
415     qemu_fdt_setprop(vms->fdt, nodename, "ranges", NULL, 0);
416     if (vms->gic_version == 3) {
417         int nb_redist_regions = virt_gicv3_redist_region_count(vms);
418 
419         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible",
420                                 "arm,gic-v3");
421 
422         qemu_fdt_setprop_cell(vms->fdt, nodename,
423                               "#redistributor-regions", nb_redist_regions);
424 
425         if (nb_redist_regions == 1) {
426             qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
427                                          2, vms->memmap[VIRT_GIC_DIST].base,
428                                          2, vms->memmap[VIRT_GIC_DIST].size,
429                                          2, vms->memmap[VIRT_GIC_REDIST].base,
430                                          2, vms->memmap[VIRT_GIC_REDIST].size);
431         } else {
432             qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
433                                          2, vms->memmap[VIRT_GIC_DIST].base,
434                                          2, vms->memmap[VIRT_GIC_DIST].size,
435                                          2, vms->memmap[VIRT_GIC_REDIST].base,
436                                          2, vms->memmap[VIRT_GIC_REDIST].size,
437                                          2, vms->memmap[VIRT_GIC_REDIST2].base,
438                                          2, vms->memmap[VIRT_GIC_REDIST2].size);
439         }
440 
441         if (vms->virt) {
442             qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
443                                    GIC_FDT_IRQ_TYPE_PPI, ARCH_GICV3_MAINT_IRQ,
444                                    GIC_FDT_IRQ_FLAGS_LEVEL_HI);
445         }
446     } else {
447         /* 'cortex-a15-gic' means 'GIC v2' */
448         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible",
449                                 "arm,cortex-a15-gic");
450         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
451                                       2, vms->memmap[VIRT_GIC_DIST].base,
452                                       2, vms->memmap[VIRT_GIC_DIST].size,
453                                       2, vms->memmap[VIRT_GIC_CPU].base,
454                                       2, vms->memmap[VIRT_GIC_CPU].size);
455     }
456 
457     qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", vms->gic_phandle);
458     g_free(nodename);
459 }
460 
461 static void fdt_add_pmu_nodes(const VirtMachineState *vms)
462 {
463     CPUState *cpu;
464     ARMCPU *armcpu;
465     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
466 
467     CPU_FOREACH(cpu) {
468         armcpu = ARM_CPU(cpu);
469         if (!arm_feature(&armcpu->env, ARM_FEATURE_PMU)) {
470             return;
471         }
472         if (kvm_enabled()) {
473             if (kvm_irqchip_in_kernel()) {
474                 kvm_arm_pmu_set_irq(cpu, PPI(VIRTUAL_PMU_IRQ));
475             }
476             kvm_arm_pmu_init(cpu);
477         }
478     }
479 
480     if (vms->gic_version == 2) {
481         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
482                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
483                              (1 << vms->smp_cpus) - 1);
484     }
485 
486     armcpu = ARM_CPU(qemu_get_cpu(0));
487     qemu_fdt_add_subnode(vms->fdt, "/pmu");
488     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
489         const char compat[] = "arm,armv8-pmuv3";
490         qemu_fdt_setprop(vms->fdt, "/pmu", "compatible",
491                          compat, sizeof(compat));
492         qemu_fdt_setprop_cells(vms->fdt, "/pmu", "interrupts",
493                                GIC_FDT_IRQ_TYPE_PPI, VIRTUAL_PMU_IRQ, irqflags);
494     }
495 }
496 
497 static void create_its(VirtMachineState *vms, DeviceState *gicdev)
498 {
499     const char *itsclass = its_class_name();
500     DeviceState *dev;
501 
502     if (!itsclass) {
503         /* Do nothing if not supported */
504         return;
505     }
506 
507     dev = qdev_create(NULL, itsclass);
508 
509     object_property_set_link(OBJECT(dev), OBJECT(gicdev), "parent-gicv3",
510                              &error_abort);
511     qdev_init_nofail(dev);
512     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_ITS].base);
513 
514     fdt_add_its_gic_node(vms);
515 }
516 
517 static void create_v2m(VirtMachineState *vms, qemu_irq *pic)
518 {
519     int i;
520     int irq = vms->irqmap[VIRT_GIC_V2M];
521     DeviceState *dev;
522 
523     dev = qdev_create(NULL, "arm-gicv2m");
524     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_V2M].base);
525     qdev_prop_set_uint32(dev, "base-spi", irq);
526     qdev_prop_set_uint32(dev, "num-spi", NUM_GICV2M_SPIS);
527     qdev_init_nofail(dev);
528 
529     for (i = 0; i < NUM_GICV2M_SPIS; i++) {
530         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]);
531     }
532 
533     fdt_add_v2m_gic_node(vms);
534 }
535 
536 static void create_gic(VirtMachineState *vms, qemu_irq *pic)
537 {
538     /* We create a standalone GIC */
539     DeviceState *gicdev;
540     SysBusDevice *gicbusdev;
541     const char *gictype;
542     int type = vms->gic_version, i;
543     uint32_t nb_redist_regions = 0;
544 
545     gictype = (type == 3) ? gicv3_class_name() : gic_class_name();
546 
547     gicdev = qdev_create(NULL, gictype);
548     qdev_prop_set_uint32(gicdev, "revision", type);
549     qdev_prop_set_uint32(gicdev, "num-cpu", smp_cpus);
550     /* Note that the num-irq property counts both internal and external
551      * interrupts; there are always 32 of the former (mandated by GIC spec).
552      */
553     qdev_prop_set_uint32(gicdev, "num-irq", NUM_IRQS + 32);
554     if (!kvm_irqchip_in_kernel()) {
555         qdev_prop_set_bit(gicdev, "has-security-extensions", vms->secure);
556     }
557 
558     if (type == 3) {
559         uint32_t redist0_capacity =
560                     vms->memmap[VIRT_GIC_REDIST].size / GICV3_REDIST_SIZE;
561         uint32_t redist0_count = MIN(smp_cpus, redist0_capacity);
562 
563         nb_redist_regions = virt_gicv3_redist_region_count(vms);
564 
565         qdev_prop_set_uint32(gicdev, "len-redist-region-count",
566                              nb_redist_regions);
567         qdev_prop_set_uint32(gicdev, "redist-region-count[0]", redist0_count);
568 
569         if (nb_redist_regions == 2) {
570             uint32_t redist1_capacity =
571                         vms->memmap[VIRT_GIC_REDIST2].size / GICV3_REDIST_SIZE;
572 
573             qdev_prop_set_uint32(gicdev, "redist-region-count[1]",
574                 MIN(smp_cpus - redist0_count, redist1_capacity));
575         }
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         if (nb_redist_regions == 2) {
583             sysbus_mmio_map(gicbusdev, 2, vms->memmap[VIRT_GIC_REDIST2].base);
584         }
585     } else {
586         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base);
587     }
588 
589     /* Wire the outputs from each CPU's generic timer and the GICv3
590      * maintenance interrupt signal to the appropriate GIC PPI inputs,
591      * and the GIC's IRQ/FIQ/VIRQ/VFIQ interrupt outputs to the CPU's inputs.
592      */
593     for (i = 0; i < smp_cpus; i++) {
594         DeviceState *cpudev = DEVICE(qemu_get_cpu(i));
595         int ppibase = NUM_IRQS + i * GIC_INTERNAL + GIC_NR_SGIS;
596         int irq;
597         /* Mapping from the output timer irq lines from the CPU to the
598          * GIC PPI inputs we use for the virt board.
599          */
600         const int timer_irq[] = {
601             [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ,
602             [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ,
603             [GTIMER_HYP]  = ARCH_TIMER_NS_EL2_IRQ,
604             [GTIMER_SEC]  = ARCH_TIMER_S_EL1_IRQ,
605         };
606 
607         for (irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) {
608             qdev_connect_gpio_out(cpudev, irq,
609                                   qdev_get_gpio_in(gicdev,
610                                                    ppibase + timer_irq[irq]));
611         }
612 
613         qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt", 0,
614                                     qdev_get_gpio_in(gicdev, ppibase
615                                                      + ARCH_GICV3_MAINT_IRQ));
616         qdev_connect_gpio_out_named(cpudev, "pmu-interrupt", 0,
617                                     qdev_get_gpio_in(gicdev, ppibase
618                                                      + VIRTUAL_PMU_IRQ));
619 
620         sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ));
621         sysbus_connect_irq(gicbusdev, i + smp_cpus,
622                            qdev_get_gpio_in(cpudev, ARM_CPU_FIQ));
623         sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus,
624                            qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ));
625         sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus,
626                            qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ));
627     }
628 
629     for (i = 0; i < NUM_IRQS; i++) {
630         pic[i] = qdev_get_gpio_in(gicdev, i);
631     }
632 
633     fdt_add_gic_node(vms);
634 
635     if (type == 3 && vms->its) {
636         create_its(vms, gicdev);
637     } else if (type == 2) {
638         create_v2m(vms, pic);
639     }
640 }
641 
642 static void create_uart(const VirtMachineState *vms, qemu_irq *pic, int uart,
643                         MemoryRegion *mem, Chardev *chr)
644 {
645     char *nodename;
646     hwaddr base = vms->memmap[uart].base;
647     hwaddr size = vms->memmap[uart].size;
648     int irq = vms->irqmap[uart];
649     const char compat[] = "arm,pl011\0arm,primecell";
650     const char clocknames[] = "uartclk\0apb_pclk";
651     DeviceState *dev = qdev_create(NULL, "pl011");
652     SysBusDevice *s = SYS_BUS_DEVICE(dev);
653 
654     qdev_prop_set_chr(dev, "chardev", chr);
655     qdev_init_nofail(dev);
656     memory_region_add_subregion(mem, base,
657                                 sysbus_mmio_get_region(s, 0));
658     sysbus_connect_irq(s, 0, pic[irq]);
659 
660     nodename = g_strdup_printf("/pl011@%" PRIx64, base);
661     qemu_fdt_add_subnode(vms->fdt, nodename);
662     /* Note that we can't use setprop_string because of the embedded NUL */
663     qemu_fdt_setprop(vms->fdt, nodename, "compatible",
664                          compat, sizeof(compat));
665     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
666                                      2, base, 2, size);
667     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
668                                GIC_FDT_IRQ_TYPE_SPI, irq,
669                                GIC_FDT_IRQ_FLAGS_LEVEL_HI);
670     qemu_fdt_setprop_cells(vms->fdt, nodename, "clocks",
671                                vms->clock_phandle, vms->clock_phandle);
672     qemu_fdt_setprop(vms->fdt, nodename, "clock-names",
673                          clocknames, sizeof(clocknames));
674 
675     if (uart == VIRT_UART) {
676         qemu_fdt_setprop_string(vms->fdt, "/chosen", "stdout-path", nodename);
677     } else {
678         /* Mark as not usable by the normal world */
679         qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
680         qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
681     }
682 
683     g_free(nodename);
684 }
685 
686 static void create_rtc(const VirtMachineState *vms, qemu_irq *pic)
687 {
688     char *nodename;
689     hwaddr base = vms->memmap[VIRT_RTC].base;
690     hwaddr size = vms->memmap[VIRT_RTC].size;
691     int irq = vms->irqmap[VIRT_RTC];
692     const char compat[] = "arm,pl031\0arm,primecell";
693 
694     sysbus_create_simple("pl031", base, pic[irq]);
695 
696     nodename = g_strdup_printf("/pl031@%" PRIx64, base);
697     qemu_fdt_add_subnode(vms->fdt, nodename);
698     qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat));
699     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
700                                  2, base, 2, size);
701     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
702                            GIC_FDT_IRQ_TYPE_SPI, irq,
703                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
704     qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle);
705     qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk");
706     g_free(nodename);
707 }
708 
709 static DeviceState *gpio_key_dev;
710 static void virt_powerdown_req(Notifier *n, void *opaque)
711 {
712     /* use gpio Pin 3 for power button event */
713     qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1);
714 }
715 
716 static Notifier virt_system_powerdown_notifier = {
717     .notify = virt_powerdown_req
718 };
719 
720 static void create_gpio(const VirtMachineState *vms, qemu_irq *pic)
721 {
722     char *nodename;
723     DeviceState *pl061_dev;
724     hwaddr base = vms->memmap[VIRT_GPIO].base;
725     hwaddr size = vms->memmap[VIRT_GPIO].size;
726     int irq = vms->irqmap[VIRT_GPIO];
727     const char compat[] = "arm,pl061\0arm,primecell";
728 
729     pl061_dev = sysbus_create_simple("pl061", base, pic[irq]);
730 
731     uint32_t phandle = qemu_fdt_alloc_phandle(vms->fdt);
732     nodename = g_strdup_printf("/pl061@%" PRIx64, base);
733     qemu_fdt_add_subnode(vms->fdt, nodename);
734     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
735                                  2, base, 2, size);
736     qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat));
737     qemu_fdt_setprop_cell(vms->fdt, nodename, "#gpio-cells", 2);
738     qemu_fdt_setprop(vms->fdt, nodename, "gpio-controller", NULL, 0);
739     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
740                            GIC_FDT_IRQ_TYPE_SPI, irq,
741                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
742     qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle);
743     qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk");
744     qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", phandle);
745 
746     gpio_key_dev = sysbus_create_simple("gpio-key", -1,
747                                         qdev_get_gpio_in(pl061_dev, 3));
748     qemu_fdt_add_subnode(vms->fdt, "/gpio-keys");
749     qemu_fdt_setprop_string(vms->fdt, "/gpio-keys", "compatible", "gpio-keys");
750     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#size-cells", 0);
751     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#address-cells", 1);
752 
753     qemu_fdt_add_subnode(vms->fdt, "/gpio-keys/poweroff");
754     qemu_fdt_setprop_string(vms->fdt, "/gpio-keys/poweroff",
755                             "label", "GPIO Key Poweroff");
756     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys/poweroff", "linux,code",
757                           KEY_POWER);
758     qemu_fdt_setprop_cells(vms->fdt, "/gpio-keys/poweroff",
759                            "gpios", phandle, 3, 0);
760 
761     /* connect powerdown request */
762     qemu_register_powerdown_notifier(&virt_system_powerdown_notifier);
763 
764     g_free(nodename);
765 }
766 
767 static void create_virtio_devices(const VirtMachineState *vms, qemu_irq *pic)
768 {
769     int i;
770     hwaddr size = vms->memmap[VIRT_MMIO].size;
771 
772     /* We create the transports in forwards order. Since qbus_realize()
773      * prepends (not appends) new child buses, the incrementing loop below will
774      * create a list of virtio-mmio buses with decreasing base addresses.
775      *
776      * When a -device option is processed from the command line,
777      * qbus_find_recursive() picks the next free virtio-mmio bus in forwards
778      * order. The upshot is that -device options in increasing command line
779      * order are mapped to virtio-mmio buses with decreasing base addresses.
780      *
781      * When this code was originally written, that arrangement ensured that the
782      * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to
783      * the first -device on the command line. (The end-to-end order is a
784      * function of this loop, qbus_realize(), qbus_find_recursive(), and the
785      * guest kernel's name-to-address assignment strategy.)
786      *
787      * Meanwhile, the kernel's traversal seems to have been reversed; see eg.
788      * the message, if not necessarily the code, of commit 70161ff336.
789      * Therefore the loop now establishes the inverse of the original intent.
790      *
791      * Unfortunately, we can't counteract the kernel change by reversing the
792      * loop; it would break existing command lines.
793      *
794      * In any case, the kernel makes no guarantee about the stability of
795      * enumeration order of virtio devices (as demonstrated by it changing
796      * between kernel versions). For reliable and stable identification
797      * of disks users must use UUIDs or similar mechanisms.
798      */
799     for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) {
800         int irq = vms->irqmap[VIRT_MMIO] + i;
801         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
802 
803         sysbus_create_simple("virtio-mmio", base, pic[irq]);
804     }
805 
806     /* We add dtb nodes in reverse order so that they appear in the finished
807      * device tree lowest address first.
808      *
809      * Note that this mapping is independent of the loop above. The previous
810      * loop influences virtio device to virtio transport assignment, whereas
811      * this loop controls how virtio transports are laid out in the dtb.
812      */
813     for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) {
814         char *nodename;
815         int irq = vms->irqmap[VIRT_MMIO] + i;
816         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
817 
818         nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base);
819         qemu_fdt_add_subnode(vms->fdt, nodename);
820         qemu_fdt_setprop_string(vms->fdt, nodename,
821                                 "compatible", "virtio,mmio");
822         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
823                                      2, base, 2, size);
824         qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
825                                GIC_FDT_IRQ_TYPE_SPI, irq,
826                                GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
827         qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
828         g_free(nodename);
829     }
830 }
831 
832 static void create_one_flash(const char *name, hwaddr flashbase,
833                              hwaddr flashsize, const char *file,
834                              MemoryRegion *sysmem)
835 {
836     /* Create and map a single flash device. We use the same
837      * parameters as the flash devices on the Versatile Express board.
838      */
839     DriveInfo *dinfo = drive_get_next(IF_PFLASH);
840     DeviceState *dev = qdev_create(NULL, "cfi.pflash01");
841     SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
842     const uint64_t sectorlength = 256 * 1024;
843 
844     if (dinfo) {
845         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
846                             &error_abort);
847     }
848 
849     qdev_prop_set_uint32(dev, "num-blocks", flashsize / sectorlength);
850     qdev_prop_set_uint64(dev, "sector-length", sectorlength);
851     qdev_prop_set_uint8(dev, "width", 4);
852     qdev_prop_set_uint8(dev, "device-width", 2);
853     qdev_prop_set_bit(dev, "big-endian", false);
854     qdev_prop_set_uint16(dev, "id0", 0x89);
855     qdev_prop_set_uint16(dev, "id1", 0x18);
856     qdev_prop_set_uint16(dev, "id2", 0x00);
857     qdev_prop_set_uint16(dev, "id3", 0x00);
858     qdev_prop_set_string(dev, "name", name);
859     qdev_init_nofail(dev);
860 
861     memory_region_add_subregion(sysmem, flashbase,
862                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0));
863 
864     if (file) {
865         char *fn;
866         int image_size;
867 
868         if (drive_get(IF_PFLASH, 0, 0)) {
869             error_report("The contents of the first flash device may be "
870                          "specified with -bios or with -drive if=pflash... "
871                          "but you cannot use both options at once");
872             exit(1);
873         }
874         fn = qemu_find_file(QEMU_FILE_TYPE_BIOS, file);
875         if (!fn) {
876             error_report("Could not find ROM image '%s'", file);
877             exit(1);
878         }
879         image_size = load_image_mr(fn, sysbus_mmio_get_region(sbd, 0));
880         g_free(fn);
881         if (image_size < 0) {
882             error_report("Could not load ROM image '%s'", file);
883             exit(1);
884         }
885     }
886 }
887 
888 static void create_flash(const VirtMachineState *vms,
889                          MemoryRegion *sysmem,
890                          MemoryRegion *secure_sysmem)
891 {
892     /* Create two flash devices to fill the VIRT_FLASH space in the memmap.
893      * Any file passed via -bios goes in the first of these.
894      * sysmem is the system memory space. secure_sysmem is the secure view
895      * of the system, and the first flash device should be made visible only
896      * there. The second flash device is visible to both secure and nonsecure.
897      * If sysmem == secure_sysmem this means there is no separate Secure
898      * address space and both flash devices are generally visible.
899      */
900     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
901     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
902     char *nodename;
903 
904     create_one_flash("virt.flash0", flashbase, flashsize,
905                      bios_name, secure_sysmem);
906     create_one_flash("virt.flash1", flashbase + flashsize, flashsize,
907                      NULL, sysmem);
908 
909     if (sysmem == secure_sysmem) {
910         /* Report both flash devices as a single node in the DT */
911         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
912         qemu_fdt_add_subnode(vms->fdt, nodename);
913         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
914         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
915                                      2, flashbase, 2, flashsize,
916                                      2, flashbase + flashsize, 2, flashsize);
917         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
918         g_free(nodename);
919     } else {
920         /* Report the devices as separate nodes so we can mark one as
921          * only visible to the secure world.
922          */
923         nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase);
924         qemu_fdt_add_subnode(vms->fdt, nodename);
925         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
926         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
927                                      2, flashbase, 2, flashsize);
928         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
929         qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
930         qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
931         g_free(nodename);
932 
933         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
934         qemu_fdt_add_subnode(vms->fdt, nodename);
935         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
936         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
937                                      2, flashbase + flashsize, 2, flashsize);
938         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
939         g_free(nodename);
940     }
941 }
942 
943 static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as)
944 {
945     hwaddr base = vms->memmap[VIRT_FW_CFG].base;
946     hwaddr size = vms->memmap[VIRT_FW_CFG].size;
947     FWCfgState *fw_cfg;
948     char *nodename;
949 
950     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as);
951     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)smp_cpus);
952 
953     nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
954     qemu_fdt_add_subnode(vms->fdt, nodename);
955     qemu_fdt_setprop_string(vms->fdt, nodename,
956                             "compatible", "qemu,fw-cfg-mmio");
957     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
958                                  2, base, 2, size);
959     qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
960     g_free(nodename);
961     return fw_cfg;
962 }
963 
964 static void create_pcie_irq_map(const VirtMachineState *vms,
965                                 uint32_t gic_phandle,
966                                 int first_irq, const char *nodename)
967 {
968     int devfn, pin;
969     uint32_t full_irq_map[4 * 4 * 10] = { 0 };
970     uint32_t *irq_map = full_irq_map;
971 
972     for (devfn = 0; devfn <= 0x18; devfn += 0x8) {
973         for (pin = 0; pin < 4; pin++) {
974             int irq_type = GIC_FDT_IRQ_TYPE_SPI;
975             int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS);
976             int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
977             int i;
978 
979             uint32_t map[] = {
980                 devfn << 8, 0, 0,                           /* devfn */
981                 pin + 1,                                    /* PCI pin */
982                 gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */
983 
984             /* Convert map to big endian */
985             for (i = 0; i < 10; i++) {
986                 irq_map[i] = cpu_to_be32(map[i]);
987             }
988             irq_map += 10;
989         }
990     }
991 
992     qemu_fdt_setprop(vms->fdt, nodename, "interrupt-map",
993                      full_irq_map, sizeof(full_irq_map));
994 
995     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupt-map-mask",
996                            0x1800, 0, 0, /* devfn (PCI_SLOT(3)) */
997                            0x7           /* PCI irq */);
998 }
999 
1000 static void create_smmu(const VirtMachineState *vms, qemu_irq *pic,
1001                         PCIBus *bus)
1002 {
1003     char *node;
1004     const char compat[] = "arm,smmu-v3";
1005     int irq =  vms->irqmap[VIRT_SMMU];
1006     int i;
1007     hwaddr base = vms->memmap[VIRT_SMMU].base;
1008     hwaddr size = vms->memmap[VIRT_SMMU].size;
1009     const char irq_names[] = "eventq\0priq\0cmdq-sync\0gerror";
1010     DeviceState *dev;
1011 
1012     if (vms->iommu != VIRT_IOMMU_SMMUV3 || !vms->iommu_phandle) {
1013         return;
1014     }
1015 
1016     dev = qdev_create(NULL, "arm-smmuv3");
1017 
1018     object_property_set_link(OBJECT(dev), OBJECT(bus), "primary-bus",
1019                              &error_abort);
1020     qdev_init_nofail(dev);
1021     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, base);
1022     for (i = 0; i < NUM_SMMU_IRQS; i++) {
1023         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]);
1024     }
1025 
1026     node = g_strdup_printf("/smmuv3@%" PRIx64, base);
1027     qemu_fdt_add_subnode(vms->fdt, node);
1028     qemu_fdt_setprop(vms->fdt, node, "compatible", compat, sizeof(compat));
1029     qemu_fdt_setprop_sized_cells(vms->fdt, node, "reg", 2, base, 2, size);
1030 
1031     qemu_fdt_setprop_cells(vms->fdt, node, "interrupts",
1032             GIC_FDT_IRQ_TYPE_SPI, irq    , GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1033             GIC_FDT_IRQ_TYPE_SPI, irq + 1, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1034             GIC_FDT_IRQ_TYPE_SPI, irq + 2, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1035             GIC_FDT_IRQ_TYPE_SPI, irq + 3, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1036 
1037     qemu_fdt_setprop(vms->fdt, node, "interrupt-names", irq_names,
1038                      sizeof(irq_names));
1039 
1040     qemu_fdt_setprop_cell(vms->fdt, node, "clocks", vms->clock_phandle);
1041     qemu_fdt_setprop_string(vms->fdt, node, "clock-names", "apb_pclk");
1042     qemu_fdt_setprop(vms->fdt, node, "dma-coherent", NULL, 0);
1043 
1044     qemu_fdt_setprop_cell(vms->fdt, node, "#iommu-cells", 1);
1045 
1046     qemu_fdt_setprop_cell(vms->fdt, node, "phandle", vms->iommu_phandle);
1047     g_free(node);
1048 }
1049 
1050 static void create_pcie(VirtMachineState *vms, qemu_irq *pic)
1051 {
1052     hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base;
1053     hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size;
1054     hwaddr base_mmio_high = vms->memmap[VIRT_PCIE_MMIO_HIGH].base;
1055     hwaddr size_mmio_high = vms->memmap[VIRT_PCIE_MMIO_HIGH].size;
1056     hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base;
1057     hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size;
1058     hwaddr base_ecam, size_ecam;
1059     hwaddr base = base_mmio;
1060     int nr_pcie_buses;
1061     int irq = vms->irqmap[VIRT_PCIE];
1062     MemoryRegion *mmio_alias;
1063     MemoryRegion *mmio_reg;
1064     MemoryRegion *ecam_alias;
1065     MemoryRegion *ecam_reg;
1066     DeviceState *dev;
1067     char *nodename;
1068     int i, ecam_id;
1069     PCIHostState *pci;
1070 
1071     dev = qdev_create(NULL, TYPE_GPEX_HOST);
1072     qdev_init_nofail(dev);
1073 
1074     ecam_id = VIRT_ECAM_ID(vms->highmem_ecam);
1075     base_ecam = vms->memmap[ecam_id].base;
1076     size_ecam = vms->memmap[ecam_id].size;
1077     nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN;
1078     /* Map only the first size_ecam bytes of ECAM space */
1079     ecam_alias = g_new0(MemoryRegion, 1);
1080     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1081     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1082                              ecam_reg, 0, size_ecam);
1083     memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias);
1084 
1085     /* Map the MMIO window into system address space so as to expose
1086      * the section of PCI MMIO space which starts at the same base address
1087      * (ie 1:1 mapping for that part of PCI MMIO space visible through
1088      * the window).
1089      */
1090     mmio_alias = g_new0(MemoryRegion, 1);
1091     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1092     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1093                              mmio_reg, base_mmio, size_mmio);
1094     memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias);
1095 
1096     if (vms->highmem) {
1097         /* Map high MMIO space */
1098         MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1);
1099 
1100         memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1101                                  mmio_reg, base_mmio_high, size_mmio_high);
1102         memory_region_add_subregion(get_system_memory(), base_mmio_high,
1103                                     high_mmio_alias);
1104     }
1105 
1106     /* Map IO port space */
1107     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio);
1108 
1109     for (i = 0; i < GPEX_NUM_IRQS; i++) {
1110         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]);
1111         gpex_set_irq_num(GPEX_HOST(dev), i, irq + i);
1112     }
1113 
1114     pci = PCI_HOST_BRIDGE(dev);
1115     if (pci->bus) {
1116         for (i = 0; i < nb_nics; i++) {
1117             NICInfo *nd = &nd_table[i];
1118 
1119             if (!nd->model) {
1120                 nd->model = g_strdup("virtio");
1121             }
1122 
1123             pci_nic_init_nofail(nd, pci->bus, nd->model, NULL);
1124         }
1125     }
1126 
1127     nodename = g_strdup_printf("/pcie@%" PRIx64, base);
1128     qemu_fdt_add_subnode(vms->fdt, nodename);
1129     qemu_fdt_setprop_string(vms->fdt, nodename,
1130                             "compatible", "pci-host-ecam-generic");
1131     qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "pci");
1132     qemu_fdt_setprop_cell(vms->fdt, nodename, "#address-cells", 3);
1133     qemu_fdt_setprop_cell(vms->fdt, nodename, "#size-cells", 2);
1134     qemu_fdt_setprop_cell(vms->fdt, nodename, "linux,pci-domain", 0);
1135     qemu_fdt_setprop_cells(vms->fdt, nodename, "bus-range", 0,
1136                            nr_pcie_buses - 1);
1137     qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
1138 
1139     if (vms->msi_phandle) {
1140         qemu_fdt_setprop_cells(vms->fdt, nodename, "msi-parent",
1141                                vms->msi_phandle);
1142     }
1143 
1144     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
1145                                  2, base_ecam, 2, size_ecam);
1146 
1147     if (vms->highmem) {
1148         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges",
1149                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1150                                      2, base_pio, 2, size_pio,
1151                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1152                                      2, base_mmio, 2, size_mmio,
1153                                      1, FDT_PCI_RANGE_MMIO_64BIT,
1154                                      2, base_mmio_high,
1155                                      2, base_mmio_high, 2, size_mmio_high);
1156     } else {
1157         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges",
1158                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1159                                      2, base_pio, 2, size_pio,
1160                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1161                                      2, base_mmio, 2, size_mmio);
1162     }
1163 
1164     qemu_fdt_setprop_cell(vms->fdt, nodename, "#interrupt-cells", 1);
1165     create_pcie_irq_map(vms, vms->gic_phandle, irq, nodename);
1166 
1167     if (vms->iommu) {
1168         vms->iommu_phandle = qemu_fdt_alloc_phandle(vms->fdt);
1169 
1170         create_smmu(vms, pic, pci->bus);
1171 
1172         qemu_fdt_setprop_cells(vms->fdt, nodename, "iommu-map",
1173                                0x0, vms->iommu_phandle, 0x0, 0x10000);
1174     }
1175 
1176     g_free(nodename);
1177 }
1178 
1179 static void create_platform_bus(VirtMachineState *vms, qemu_irq *pic)
1180 {
1181     DeviceState *dev;
1182     SysBusDevice *s;
1183     int i;
1184     MemoryRegion *sysmem = get_system_memory();
1185 
1186     dev = qdev_create(NULL, TYPE_PLATFORM_BUS_DEVICE);
1187     dev->id = TYPE_PLATFORM_BUS_DEVICE;
1188     qdev_prop_set_uint32(dev, "num_irqs", PLATFORM_BUS_NUM_IRQS);
1189     qdev_prop_set_uint32(dev, "mmio_size", vms->memmap[VIRT_PLATFORM_BUS].size);
1190     qdev_init_nofail(dev);
1191     vms->platform_bus_dev = dev;
1192 
1193     s = SYS_BUS_DEVICE(dev);
1194     for (i = 0; i < PLATFORM_BUS_NUM_IRQS; i++) {
1195         int irqn = vms->irqmap[VIRT_PLATFORM_BUS] + i;
1196         sysbus_connect_irq(s, i, pic[irqn]);
1197     }
1198 
1199     memory_region_add_subregion(sysmem,
1200                                 vms->memmap[VIRT_PLATFORM_BUS].base,
1201                                 sysbus_mmio_get_region(s, 0));
1202 }
1203 
1204 static void create_secure_ram(VirtMachineState *vms,
1205                               MemoryRegion *secure_sysmem)
1206 {
1207     MemoryRegion *secram = g_new(MemoryRegion, 1);
1208     char *nodename;
1209     hwaddr base = vms->memmap[VIRT_SECURE_MEM].base;
1210     hwaddr size = vms->memmap[VIRT_SECURE_MEM].size;
1211 
1212     memory_region_init_ram(secram, NULL, "virt.secure-ram", size,
1213                            &error_fatal);
1214     memory_region_add_subregion(secure_sysmem, base, secram);
1215 
1216     nodename = g_strdup_printf("/secram@%" PRIx64, base);
1217     qemu_fdt_add_subnode(vms->fdt, nodename);
1218     qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "memory");
1219     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 2, base, 2, size);
1220     qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
1221     qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
1222 
1223     g_free(nodename);
1224 }
1225 
1226 static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size)
1227 {
1228     const VirtMachineState *board = container_of(binfo, VirtMachineState,
1229                                                  bootinfo);
1230 
1231     *fdt_size = board->fdt_size;
1232     return board->fdt;
1233 }
1234 
1235 static void virt_build_smbios(VirtMachineState *vms)
1236 {
1237     MachineClass *mc = MACHINE_GET_CLASS(vms);
1238     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1239     uint8_t *smbios_tables, *smbios_anchor;
1240     size_t smbios_tables_len, smbios_anchor_len;
1241     const char *product = "QEMU Virtual Machine";
1242 
1243     if (!vms->fw_cfg) {
1244         return;
1245     }
1246 
1247     if (kvm_enabled()) {
1248         product = "KVM Virtual Machine";
1249     }
1250 
1251     smbios_set_defaults("QEMU", product,
1252                         vmc->smbios_old_sys_ver ? "1.0" : mc->name, false,
1253                         true, SMBIOS_ENTRY_POINT_30);
1254 
1255     smbios_get_tables(NULL, 0, &smbios_tables, &smbios_tables_len,
1256                       &smbios_anchor, &smbios_anchor_len);
1257 
1258     if (smbios_anchor) {
1259         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables",
1260                         smbios_tables, smbios_tables_len);
1261         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor",
1262                         smbios_anchor, smbios_anchor_len);
1263     }
1264 }
1265 
1266 static
1267 void virt_machine_done(Notifier *notifier, void *data)
1268 {
1269     VirtMachineState *vms = container_of(notifier, VirtMachineState,
1270                                          machine_done);
1271     ARMCPU *cpu = ARM_CPU(first_cpu);
1272     struct arm_boot_info *info = &vms->bootinfo;
1273     AddressSpace *as = arm_boot_address_space(cpu, info);
1274 
1275     /*
1276      * If the user provided a dtb, we assume the dynamic sysbus nodes
1277      * already are integrated there. This corresponds to a use case where
1278      * the dynamic sysbus nodes are complex and their generation is not yet
1279      * supported. In that case the user can take charge of the guest dt
1280      * while qemu takes charge of the qom stuff.
1281      */
1282     if (info->dtb_filename == NULL) {
1283         platform_bus_add_all_fdt_nodes(vms->fdt, "/intc",
1284                                        vms->memmap[VIRT_PLATFORM_BUS].base,
1285                                        vms->memmap[VIRT_PLATFORM_BUS].size,
1286                                        vms->irqmap[VIRT_PLATFORM_BUS]);
1287     }
1288     if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as) < 0) {
1289         exit(1);
1290     }
1291 
1292     virt_acpi_setup(vms);
1293     virt_build_smbios(vms);
1294 }
1295 
1296 static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx)
1297 {
1298     uint8_t clustersz = ARM_DEFAULT_CPUS_PER_CLUSTER;
1299     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1300 
1301     if (!vmc->disallow_affinity_adjustment) {
1302         /* Adjust MPIDR like 64-bit KVM hosts, which incorporate the
1303          * GIC's target-list limitations. 32-bit KVM hosts currently
1304          * always create clusters of 4 CPUs, but that is expected to
1305          * change when they gain support for gicv3. When KVM is enabled
1306          * it will override the changes we make here, therefore our
1307          * purposes are to make TCG consistent (with 64-bit KVM hosts)
1308          * and to improve SGI efficiency.
1309          */
1310         if (vms->gic_version == 3) {
1311             clustersz = GICV3_TARGETLIST_BITS;
1312         } else {
1313             clustersz = GIC_TARGETLIST_BITS;
1314         }
1315     }
1316     return arm_cpu_mp_affinity(idx, clustersz);
1317 }
1318 
1319 static void machvirt_init(MachineState *machine)
1320 {
1321     VirtMachineState *vms = VIRT_MACHINE(machine);
1322     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine);
1323     MachineClass *mc = MACHINE_GET_CLASS(machine);
1324     const CPUArchIdList *possible_cpus;
1325     qemu_irq pic[NUM_IRQS];
1326     MemoryRegion *sysmem = get_system_memory();
1327     MemoryRegion *secure_sysmem = NULL;
1328     int n, virt_max_cpus;
1329     MemoryRegion *ram = g_new(MemoryRegion, 1);
1330     bool firmware_loaded = bios_name || drive_get(IF_PFLASH, 0, 0);
1331     bool aarch64 = true;
1332 
1333     /* We can probe only here because during property set
1334      * KVM is not available yet
1335      */
1336     if (vms->gic_version <= 0) {
1337         /* "host" or "max" */
1338         if (!kvm_enabled()) {
1339             if (vms->gic_version == 0) {
1340                 error_report("gic-version=host requires KVM");
1341                 exit(1);
1342             } else {
1343                 /* "max": currently means 3 for TCG */
1344                 vms->gic_version = 3;
1345             }
1346         } else {
1347             vms->gic_version = kvm_arm_vgic_probe();
1348             if (!vms->gic_version) {
1349                 error_report(
1350                     "Unable to determine GIC version supported by host");
1351                 exit(1);
1352             }
1353         }
1354     }
1355 
1356     if (!cpu_type_valid(machine->cpu_type)) {
1357         error_report("mach-virt: CPU type %s not supported", machine->cpu_type);
1358         exit(1);
1359     }
1360 
1361     /* If we have an EL3 boot ROM then the assumption is that it will
1362      * implement PSCI itself, so disable QEMU's internal implementation
1363      * so it doesn't get in the way. Instead of starting secondary
1364      * CPUs in PSCI powerdown state we will start them all running and
1365      * let the boot ROM sort them out.
1366      * The usual case is that we do use QEMU's PSCI implementation;
1367      * if the guest has EL2 then we will use SMC as the conduit,
1368      * and otherwise we will use HVC (for backwards compatibility and
1369      * because if we're using KVM then we must use HVC).
1370      */
1371     if (vms->secure && firmware_loaded) {
1372         vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
1373     } else if (vms->virt) {
1374         vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC;
1375     } else {
1376         vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC;
1377     }
1378 
1379     /* The maximum number of CPUs depends on the GIC version, or on how
1380      * many redistributors we can fit into the memory map.
1381      */
1382     if (vms->gic_version == 3) {
1383         virt_max_cpus = vms->memmap[VIRT_GIC_REDIST].size / GICV3_REDIST_SIZE;
1384         virt_max_cpus += vms->memmap[VIRT_GIC_REDIST2].size / GICV3_REDIST_SIZE;
1385     } else {
1386         virt_max_cpus = GIC_NCPU;
1387     }
1388 
1389     if (max_cpus > virt_max_cpus) {
1390         error_report("Number of SMP CPUs requested (%d) exceeds max CPUs "
1391                      "supported by machine 'mach-virt' (%d)",
1392                      max_cpus, virt_max_cpus);
1393         exit(1);
1394     }
1395 
1396     vms->smp_cpus = smp_cpus;
1397 
1398     if (machine->ram_size > vms->memmap[VIRT_MEM].size) {
1399         error_report("mach-virt: cannot model more than %dGB RAM", RAMLIMIT_GB);
1400         exit(1);
1401     }
1402 
1403     if (vms->virt && kvm_enabled()) {
1404         error_report("mach-virt: KVM does not support providing "
1405                      "Virtualization extensions to the guest CPU");
1406         exit(1);
1407     }
1408 
1409     if (vms->secure) {
1410         if (kvm_enabled()) {
1411             error_report("mach-virt: KVM does not support Security extensions");
1412             exit(1);
1413         }
1414 
1415         /* The Secure view of the world is the same as the NonSecure,
1416          * but with a few extra devices. Create it as a container region
1417          * containing the system memory at low priority; any secure-only
1418          * devices go in at higher priority and take precedence.
1419          */
1420         secure_sysmem = g_new(MemoryRegion, 1);
1421         memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory",
1422                            UINT64_MAX);
1423         memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1);
1424     }
1425 
1426     create_fdt(vms);
1427 
1428     possible_cpus = mc->possible_cpu_arch_ids(machine);
1429     for (n = 0; n < possible_cpus->len; n++) {
1430         Object *cpuobj;
1431         CPUState *cs;
1432 
1433         if (n >= smp_cpus) {
1434             break;
1435         }
1436 
1437         cpuobj = object_new(possible_cpus->cpus[n].type);
1438         object_property_set_int(cpuobj, possible_cpus->cpus[n].arch_id,
1439                                 "mp-affinity", NULL);
1440 
1441         cs = CPU(cpuobj);
1442         cs->cpu_index = n;
1443 
1444         numa_cpu_pre_plug(&possible_cpus->cpus[cs->cpu_index], DEVICE(cpuobj),
1445                           &error_fatal);
1446 
1447         aarch64 &= object_property_get_bool(cpuobj, "aarch64", NULL);
1448 
1449         if (!vms->secure) {
1450             object_property_set_bool(cpuobj, false, "has_el3", NULL);
1451         }
1452 
1453         if (!vms->virt && object_property_find(cpuobj, "has_el2", NULL)) {
1454             object_property_set_bool(cpuobj, false, "has_el2", NULL);
1455         }
1456 
1457         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) {
1458             object_property_set_int(cpuobj, vms->psci_conduit,
1459                                     "psci-conduit", NULL);
1460 
1461             /* Secondary CPUs start in PSCI powered-down state */
1462             if (n > 0) {
1463                 object_property_set_bool(cpuobj, true,
1464                                          "start-powered-off", NULL);
1465             }
1466         }
1467 
1468         if (vmc->no_pmu && object_property_find(cpuobj, "pmu", NULL)) {
1469             object_property_set_bool(cpuobj, false, "pmu", NULL);
1470         }
1471 
1472         if (object_property_find(cpuobj, "reset-cbar", NULL)) {
1473             object_property_set_int(cpuobj, vms->memmap[VIRT_CPUPERIPHS].base,
1474                                     "reset-cbar", &error_abort);
1475         }
1476 
1477         object_property_set_link(cpuobj, OBJECT(sysmem), "memory",
1478                                  &error_abort);
1479         if (vms->secure) {
1480             object_property_set_link(cpuobj, OBJECT(secure_sysmem),
1481                                      "secure-memory", &error_abort);
1482         }
1483 
1484         object_property_set_bool(cpuobj, true, "realized", &error_fatal);
1485         object_unref(cpuobj);
1486     }
1487     fdt_add_timer_nodes(vms);
1488     fdt_add_cpu_nodes(vms);
1489 
1490     memory_region_allocate_system_memory(ram, NULL, "mach-virt.ram",
1491                                          machine->ram_size);
1492     memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base, ram);
1493 
1494     create_flash(vms, sysmem, secure_sysmem ? secure_sysmem : sysmem);
1495 
1496     create_gic(vms, pic);
1497 
1498     fdt_add_pmu_nodes(vms);
1499 
1500     create_uart(vms, pic, VIRT_UART, sysmem, serial_hd(0));
1501 
1502     if (vms->secure) {
1503         create_secure_ram(vms, secure_sysmem);
1504         create_uart(vms, pic, VIRT_SECURE_UART, secure_sysmem, serial_hd(1));
1505     }
1506 
1507     vms->highmem_ecam &= vms->highmem && (!firmware_loaded || aarch64);
1508 
1509     create_rtc(vms, pic);
1510 
1511     create_pcie(vms, pic);
1512 
1513     create_gpio(vms, pic);
1514 
1515     /* Create mmio transports, so the user can create virtio backends
1516      * (which will be automatically plugged in to the transports). If
1517      * no backend is created the transport will just sit harmlessly idle.
1518      */
1519     create_virtio_devices(vms, pic);
1520 
1521     vms->fw_cfg = create_fw_cfg(vms, &address_space_memory);
1522     rom_set_fw(vms->fw_cfg);
1523 
1524     create_platform_bus(vms, pic);
1525 
1526     vms->bootinfo.ram_size = machine->ram_size;
1527     vms->bootinfo.kernel_filename = machine->kernel_filename;
1528     vms->bootinfo.kernel_cmdline = machine->kernel_cmdline;
1529     vms->bootinfo.initrd_filename = machine->initrd_filename;
1530     vms->bootinfo.nb_cpus = smp_cpus;
1531     vms->bootinfo.board_id = -1;
1532     vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base;
1533     vms->bootinfo.get_dtb = machvirt_dtb;
1534     vms->bootinfo.skip_dtb_autoload = true;
1535     vms->bootinfo.firmware_loaded = firmware_loaded;
1536     arm_load_kernel(ARM_CPU(first_cpu), &vms->bootinfo);
1537 
1538     vms->machine_done.notify = virt_machine_done;
1539     qemu_add_machine_init_done_notifier(&vms->machine_done);
1540 }
1541 
1542 static bool virt_get_secure(Object *obj, Error **errp)
1543 {
1544     VirtMachineState *vms = VIRT_MACHINE(obj);
1545 
1546     return vms->secure;
1547 }
1548 
1549 static void virt_set_secure(Object *obj, bool value, Error **errp)
1550 {
1551     VirtMachineState *vms = VIRT_MACHINE(obj);
1552 
1553     vms->secure = value;
1554 }
1555 
1556 static bool virt_get_virt(Object *obj, Error **errp)
1557 {
1558     VirtMachineState *vms = VIRT_MACHINE(obj);
1559 
1560     return vms->virt;
1561 }
1562 
1563 static void virt_set_virt(Object *obj, bool value, Error **errp)
1564 {
1565     VirtMachineState *vms = VIRT_MACHINE(obj);
1566 
1567     vms->virt = value;
1568 }
1569 
1570 static bool virt_get_highmem(Object *obj, Error **errp)
1571 {
1572     VirtMachineState *vms = VIRT_MACHINE(obj);
1573 
1574     return vms->highmem;
1575 }
1576 
1577 static void virt_set_highmem(Object *obj, bool value, Error **errp)
1578 {
1579     VirtMachineState *vms = VIRT_MACHINE(obj);
1580 
1581     vms->highmem = value;
1582 }
1583 
1584 static bool virt_get_its(Object *obj, Error **errp)
1585 {
1586     VirtMachineState *vms = VIRT_MACHINE(obj);
1587 
1588     return vms->its;
1589 }
1590 
1591 static void virt_set_its(Object *obj, bool value, Error **errp)
1592 {
1593     VirtMachineState *vms = VIRT_MACHINE(obj);
1594 
1595     vms->its = value;
1596 }
1597 
1598 static char *virt_get_gic_version(Object *obj, Error **errp)
1599 {
1600     VirtMachineState *vms = VIRT_MACHINE(obj);
1601     const char *val = vms->gic_version == 3 ? "3" : "2";
1602 
1603     return g_strdup(val);
1604 }
1605 
1606 static void virt_set_gic_version(Object *obj, const char *value, Error **errp)
1607 {
1608     VirtMachineState *vms = VIRT_MACHINE(obj);
1609 
1610     if (!strcmp(value, "3")) {
1611         vms->gic_version = 3;
1612     } else if (!strcmp(value, "2")) {
1613         vms->gic_version = 2;
1614     } else if (!strcmp(value, "host")) {
1615         vms->gic_version = 0; /* Will probe later */
1616     } else if (!strcmp(value, "max")) {
1617         vms->gic_version = -1; /* Will probe later */
1618     } else {
1619         error_setg(errp, "Invalid gic-version value");
1620         error_append_hint(errp, "Valid values are 3, 2, host, max.\n");
1621     }
1622 }
1623 
1624 static char *virt_get_iommu(Object *obj, Error **errp)
1625 {
1626     VirtMachineState *vms = VIRT_MACHINE(obj);
1627 
1628     switch (vms->iommu) {
1629     case VIRT_IOMMU_NONE:
1630         return g_strdup("none");
1631     case VIRT_IOMMU_SMMUV3:
1632         return g_strdup("smmuv3");
1633     default:
1634         g_assert_not_reached();
1635     }
1636 }
1637 
1638 static void virt_set_iommu(Object *obj, const char *value, Error **errp)
1639 {
1640     VirtMachineState *vms = VIRT_MACHINE(obj);
1641 
1642     if (!strcmp(value, "smmuv3")) {
1643         vms->iommu = VIRT_IOMMU_SMMUV3;
1644     } else if (!strcmp(value, "none")) {
1645         vms->iommu = VIRT_IOMMU_NONE;
1646     } else {
1647         error_setg(errp, "Invalid iommu value");
1648         error_append_hint(errp, "Valid values are none, smmuv3.\n");
1649     }
1650 }
1651 
1652 static CpuInstanceProperties
1653 virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
1654 {
1655     MachineClass *mc = MACHINE_GET_CLASS(ms);
1656     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
1657 
1658     assert(cpu_index < possible_cpus->len);
1659     return possible_cpus->cpus[cpu_index].props;
1660 }
1661 
1662 static int64_t virt_get_default_cpu_node_id(const MachineState *ms, int idx)
1663 {
1664     return idx % nb_numa_nodes;
1665 }
1666 
1667 static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms)
1668 {
1669     int n;
1670     VirtMachineState *vms = VIRT_MACHINE(ms);
1671 
1672     if (ms->possible_cpus) {
1673         assert(ms->possible_cpus->len == max_cpus);
1674         return ms->possible_cpus;
1675     }
1676 
1677     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
1678                                   sizeof(CPUArchId) * max_cpus);
1679     ms->possible_cpus->len = max_cpus;
1680     for (n = 0; n < ms->possible_cpus->len; n++) {
1681         ms->possible_cpus->cpus[n].type = ms->cpu_type;
1682         ms->possible_cpus->cpus[n].arch_id =
1683             virt_cpu_mp_affinity(vms, n);
1684         ms->possible_cpus->cpus[n].props.has_thread_id = true;
1685         ms->possible_cpus->cpus[n].props.thread_id = n;
1686     }
1687     return ms->possible_cpus;
1688 }
1689 
1690 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
1691                                         DeviceState *dev, Error **errp)
1692 {
1693     VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
1694 
1695     if (vms->platform_bus_dev) {
1696         if (object_dynamic_cast(OBJECT(dev), TYPE_SYS_BUS_DEVICE)) {
1697             platform_bus_link_device(PLATFORM_BUS_DEVICE(vms->platform_bus_dev),
1698                                      SYS_BUS_DEVICE(dev));
1699         }
1700     }
1701 }
1702 
1703 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
1704                                                         DeviceState *dev)
1705 {
1706     if (object_dynamic_cast(OBJECT(dev), TYPE_SYS_BUS_DEVICE)) {
1707         return HOTPLUG_HANDLER(machine);
1708     }
1709 
1710     return NULL;
1711 }
1712 
1713 static void virt_machine_class_init(ObjectClass *oc, void *data)
1714 {
1715     MachineClass *mc = MACHINE_CLASS(oc);
1716     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
1717 
1718     mc->init = machvirt_init;
1719     /* Start with max_cpus set to 512, which is the maximum supported by KVM.
1720      * The value may be reduced later when we have more information about the
1721      * configuration of the particular instance.
1722      */
1723     mc->max_cpus = 512;
1724     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_CALXEDA_XGMAC);
1725     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_AMD_XGBE);
1726     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
1727     mc->block_default_type = IF_VIRTIO;
1728     mc->no_cdrom = 1;
1729     mc->pci_allow_0_address = true;
1730     /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */
1731     mc->minimum_page_bits = 12;
1732     mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids;
1733     mc->cpu_index_to_instance_props = virt_cpu_index_to_props;
1734     mc->default_cpu_type = ARM_CPU_TYPE_NAME("cortex-a15");
1735     mc->get_default_cpu_node_id = virt_get_default_cpu_node_id;
1736     assert(!mc->get_hotplug_handler);
1737     mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
1738     hc->plug = virt_machine_device_plug_cb;
1739 }
1740 
1741 static const TypeInfo virt_machine_info = {
1742     .name          = TYPE_VIRT_MACHINE,
1743     .parent        = TYPE_MACHINE,
1744     .abstract      = true,
1745     .instance_size = sizeof(VirtMachineState),
1746     .class_size    = sizeof(VirtMachineClass),
1747     .class_init    = virt_machine_class_init,
1748     .interfaces = (InterfaceInfo[]) {
1749          { TYPE_HOTPLUG_HANDLER },
1750          { }
1751     },
1752 };
1753 
1754 static void machvirt_machine_init(void)
1755 {
1756     type_register_static(&virt_machine_info);
1757 }
1758 type_init(machvirt_machine_init);
1759 
1760 #define VIRT_COMPAT_2_12 \
1761     HW_COMPAT_2_12
1762 
1763 static void virt_3_0_instance_init(Object *obj)
1764 {
1765     VirtMachineState *vms = VIRT_MACHINE(obj);
1766     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1767 
1768     /* EL3 is disabled by default on virt: this makes us consistent
1769      * between KVM and TCG for this board, and it also allows us to
1770      * boot UEFI blobs which assume no TrustZone support.
1771      */
1772     vms->secure = false;
1773     object_property_add_bool(obj, "secure", virt_get_secure,
1774                              virt_set_secure, NULL);
1775     object_property_set_description(obj, "secure",
1776                                     "Set on/off to enable/disable the ARM "
1777                                     "Security Extensions (TrustZone)",
1778                                     NULL);
1779 
1780     /* EL2 is also disabled by default, for similar reasons */
1781     vms->virt = false;
1782     object_property_add_bool(obj, "virtualization", virt_get_virt,
1783                              virt_set_virt, NULL);
1784     object_property_set_description(obj, "virtualization",
1785                                     "Set on/off to enable/disable emulating a "
1786                                     "guest CPU which implements the ARM "
1787                                     "Virtualization Extensions",
1788                                     NULL);
1789 
1790     /* High memory is enabled by default */
1791     vms->highmem = true;
1792     object_property_add_bool(obj, "highmem", virt_get_highmem,
1793                              virt_set_highmem, NULL);
1794     object_property_set_description(obj, "highmem",
1795                                     "Set on/off to enable/disable using "
1796                                     "physical address space above 32 bits",
1797                                     NULL);
1798     /* Default GIC type is v2 */
1799     vms->gic_version = 2;
1800     object_property_add_str(obj, "gic-version", virt_get_gic_version,
1801                         virt_set_gic_version, NULL);
1802     object_property_set_description(obj, "gic-version",
1803                                     "Set GIC version. "
1804                                     "Valid values are 2, 3 and host", NULL);
1805 
1806     vms->highmem_ecam = !vmc->no_highmem_ecam;
1807 
1808     if (vmc->no_its) {
1809         vms->its = false;
1810     } else {
1811         /* Default allows ITS instantiation */
1812         vms->its = true;
1813         object_property_add_bool(obj, "its", virt_get_its,
1814                                  virt_set_its, NULL);
1815         object_property_set_description(obj, "its",
1816                                         "Set on/off to enable/disable "
1817                                         "ITS instantiation",
1818                                         NULL);
1819     }
1820 
1821     /* Default disallows iommu instantiation */
1822     vms->iommu = VIRT_IOMMU_NONE;
1823     object_property_add_str(obj, "iommu", virt_get_iommu, virt_set_iommu, NULL);
1824     object_property_set_description(obj, "iommu",
1825                                     "Set the IOMMU type. "
1826                                     "Valid values are none and smmuv3",
1827                                     NULL);
1828 
1829     vms->memmap = a15memmap;
1830     vms->irqmap = a15irqmap;
1831 }
1832 
1833 static void virt_machine_3_0_options(MachineClass *mc)
1834 {
1835 }
1836 DEFINE_VIRT_MACHINE_AS_LATEST(3, 0)
1837 
1838 static void virt_2_12_instance_init(Object *obj)
1839 {
1840     virt_3_0_instance_init(obj);
1841 }
1842 
1843 static void virt_machine_2_12_options(MachineClass *mc)
1844 {
1845     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1846 
1847     virt_machine_3_0_options(mc);
1848     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_12);
1849     vmc->no_highmem_ecam = true;
1850     mc->max_cpus = 255;
1851 }
1852 DEFINE_VIRT_MACHINE(2, 12)
1853 
1854 #define VIRT_COMPAT_2_11 \
1855     HW_COMPAT_2_11
1856 
1857 static void virt_2_11_instance_init(Object *obj)
1858 {
1859     virt_2_12_instance_init(obj);
1860 }
1861 
1862 static void virt_machine_2_11_options(MachineClass *mc)
1863 {
1864     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1865 
1866     virt_machine_2_12_options(mc);
1867     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_11);
1868     vmc->smbios_old_sys_ver = true;
1869 }
1870 DEFINE_VIRT_MACHINE(2, 11)
1871 
1872 #define VIRT_COMPAT_2_10 \
1873     HW_COMPAT_2_10
1874 
1875 static void virt_2_10_instance_init(Object *obj)
1876 {
1877     virt_2_11_instance_init(obj);
1878 }
1879 
1880 static void virt_machine_2_10_options(MachineClass *mc)
1881 {
1882     virt_machine_2_11_options(mc);
1883     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_10);
1884 }
1885 DEFINE_VIRT_MACHINE(2, 10)
1886 
1887 #define VIRT_COMPAT_2_9 \
1888     HW_COMPAT_2_9
1889 
1890 static void virt_2_9_instance_init(Object *obj)
1891 {
1892     virt_2_10_instance_init(obj);
1893 }
1894 
1895 static void virt_machine_2_9_options(MachineClass *mc)
1896 {
1897     virt_machine_2_10_options(mc);
1898     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_9);
1899 }
1900 DEFINE_VIRT_MACHINE(2, 9)
1901 
1902 #define VIRT_COMPAT_2_8 \
1903     HW_COMPAT_2_8
1904 
1905 static void virt_2_8_instance_init(Object *obj)
1906 {
1907     virt_2_9_instance_init(obj);
1908 }
1909 
1910 static void virt_machine_2_8_options(MachineClass *mc)
1911 {
1912     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1913 
1914     virt_machine_2_9_options(mc);
1915     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_8);
1916     /* For 2.8 and earlier we falsely claimed in the DT that
1917      * our timers were edge-triggered, not level-triggered.
1918      */
1919     vmc->claim_edge_triggered_timers = true;
1920 }
1921 DEFINE_VIRT_MACHINE(2, 8)
1922 
1923 #define VIRT_COMPAT_2_7 \
1924     HW_COMPAT_2_7
1925 
1926 static void virt_2_7_instance_init(Object *obj)
1927 {
1928     virt_2_8_instance_init(obj);
1929 }
1930 
1931 static void virt_machine_2_7_options(MachineClass *mc)
1932 {
1933     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1934 
1935     virt_machine_2_8_options(mc);
1936     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_7);
1937     /* ITS was introduced with 2.8 */
1938     vmc->no_its = true;
1939     /* Stick with 1K pages for migration compatibility */
1940     mc->minimum_page_bits = 0;
1941 }
1942 DEFINE_VIRT_MACHINE(2, 7)
1943 
1944 #define VIRT_COMPAT_2_6 \
1945     HW_COMPAT_2_6
1946 
1947 static void virt_2_6_instance_init(Object *obj)
1948 {
1949     virt_2_7_instance_init(obj);
1950 }
1951 
1952 static void virt_machine_2_6_options(MachineClass *mc)
1953 {
1954     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1955 
1956     virt_machine_2_7_options(mc);
1957     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_6);
1958     vmc->disallow_affinity_adjustment = true;
1959     /* Disable PMU for 2.6 as PMU support was first introduced in 2.7 */
1960     vmc->no_pmu = true;
1961 }
1962 DEFINE_VIRT_MACHINE(2, 6)
1963