1 /* 2 * ARM mach-virt emulation 3 * 4 * Copyright (c) 2013 Linaro Limited 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms and conditions of the GNU General Public License, 8 * version 2 or later, as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 * more details. 14 * 15 * You should have received a copy of the GNU General Public License along with 16 * this program. If not, see <http://www.gnu.org/licenses/>. 17 * 18 * Emulate a virtual board which works by passing Linux all the information 19 * it needs about what devices are present via the device tree. 20 * There are some restrictions about what we can do here: 21 * + we can only present devices whose Linux drivers will work based 22 * purely on the device tree with no platform data at all 23 * + we want to present a very stripped-down minimalist platform, 24 * both because this reduces the security attack surface from the guest 25 * and also because it reduces our exposure to being broken when 26 * the kernel updates its device tree bindings and requires further 27 * information in a device binding that we aren't providing. 28 * This is essentially the same approach kvmtool uses. 29 */ 30 31 #include "qemu/osdep.h" 32 #include "qemu/units.h" 33 #include "qapi/error.h" 34 #include "hw/sysbus.h" 35 #include "hw/arm/arm.h" 36 #include "hw/arm/primecell.h" 37 #include "hw/arm/virt.h" 38 #include "hw/block/flash.h" 39 #include "hw/vfio/vfio-calxeda-xgmac.h" 40 #include "hw/vfio/vfio-amd-xgbe.h" 41 #include "hw/display/ramfb.h" 42 #include "net/net.h" 43 #include "sysemu/device_tree.h" 44 #include "sysemu/numa.h" 45 #include "sysemu/sysemu.h" 46 #include "sysemu/kvm.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/firmware/smbios.h" 59 #include "qapi/visitor.h" 60 #include "standard-headers/linux/input.h" 61 #include "hw/arm/smmuv3.h" 62 #include "hw/acpi/acpi.h" 63 #include "target/arm/internals.h" 64 65 #define DEFINE_VIRT_MACHINE_LATEST(major, minor, latest) \ 66 static void virt_##major##_##minor##_class_init(ObjectClass *oc, \ 67 void *data) \ 68 { \ 69 MachineClass *mc = MACHINE_CLASS(oc); \ 70 virt_machine_##major##_##minor##_options(mc); \ 71 mc->desc = "QEMU " # major "." # minor " ARM Virtual Machine"; \ 72 if (latest) { \ 73 mc->alias = "virt"; \ 74 } \ 75 } \ 76 static const TypeInfo machvirt_##major##_##minor##_info = { \ 77 .name = MACHINE_TYPE_NAME("virt-" # major "." # minor), \ 78 .parent = TYPE_VIRT_MACHINE, \ 79 .class_init = virt_##major##_##minor##_class_init, \ 80 }; \ 81 static void machvirt_machine_##major##_##minor##_init(void) \ 82 { \ 83 type_register_static(&machvirt_##major##_##minor##_info); \ 84 } \ 85 type_init(machvirt_machine_##major##_##minor##_init); 86 87 #define DEFINE_VIRT_MACHINE_AS_LATEST(major, minor) \ 88 DEFINE_VIRT_MACHINE_LATEST(major, minor, true) 89 #define DEFINE_VIRT_MACHINE(major, minor) \ 90 DEFINE_VIRT_MACHINE_LATEST(major, minor, false) 91 92 93 /* Number of external interrupt lines to configure the GIC with */ 94 #define NUM_IRQS 256 95 96 #define PLATFORM_BUS_NUM_IRQS 64 97 98 /* Legacy RAM limit in GB (< version 4.0) */ 99 #define LEGACY_RAMLIMIT_GB 255 100 #define LEGACY_RAMLIMIT_BYTES (LEGACY_RAMLIMIT_GB * GiB) 101 102 /* Addresses and sizes of our components. 103 * 0..128MB is space for a flash device so we can run bootrom code such as UEFI. 104 * 128MB..256MB is used for miscellaneous device I/O. 105 * 256MB..1GB is reserved for possible future PCI support (ie where the 106 * PCI memory window will go if we add a PCI host controller). 107 * 1GB and up is RAM (which may happily spill over into the 108 * high memory region beyond 4GB). 109 * This represents a compromise between how much RAM can be given to 110 * a 32 bit VM and leaving space for expansion and in particular for PCI. 111 * Note that devices should generally be placed at multiples of 0x10000, 112 * to accommodate guests using 64K pages. 113 */ 114 static const MemMapEntry base_memmap[] = { 115 /* Space up to 0x8000000 is reserved for a boot ROM */ 116 [VIRT_FLASH] = { 0, 0x08000000 }, 117 [VIRT_CPUPERIPHS] = { 0x08000000, 0x00020000 }, 118 /* GIC distributor and CPU interfaces sit inside the CPU peripheral space */ 119 [VIRT_GIC_DIST] = { 0x08000000, 0x00010000 }, 120 [VIRT_GIC_CPU] = { 0x08010000, 0x00010000 }, 121 [VIRT_GIC_V2M] = { 0x08020000, 0x00001000 }, 122 [VIRT_GIC_HYP] = { 0x08030000, 0x00010000 }, 123 [VIRT_GIC_VCPU] = { 0x08040000, 0x00010000 }, 124 /* The space in between here is reserved for GICv3 CPU/vCPU/HYP */ 125 [VIRT_GIC_ITS] = { 0x08080000, 0x00020000 }, 126 /* This redistributor space allows up to 2*64kB*123 CPUs */ 127 [VIRT_GIC_REDIST] = { 0x080A0000, 0x00F60000 }, 128 [VIRT_UART] = { 0x09000000, 0x00001000 }, 129 [VIRT_RTC] = { 0x09010000, 0x00001000 }, 130 [VIRT_FW_CFG] = { 0x09020000, 0x00000018 }, 131 [VIRT_GPIO] = { 0x09030000, 0x00001000 }, 132 [VIRT_SECURE_UART] = { 0x09040000, 0x00001000 }, 133 [VIRT_SMMU] = { 0x09050000, 0x00020000 }, 134 [VIRT_MMIO] = { 0x0a000000, 0x00000200 }, 135 /* ...repeating for a total of NUM_VIRTIO_TRANSPORTS, each of that size */ 136 [VIRT_PLATFORM_BUS] = { 0x0c000000, 0x02000000 }, 137 [VIRT_SECURE_MEM] = { 0x0e000000, 0x01000000 }, 138 [VIRT_PCIE_MMIO] = { 0x10000000, 0x2eff0000 }, 139 [VIRT_PCIE_PIO] = { 0x3eff0000, 0x00010000 }, 140 [VIRT_PCIE_ECAM] = { 0x3f000000, 0x01000000 }, 141 /* Actual RAM size depends on initial RAM and device memory settings */ 142 [VIRT_MEM] = { GiB, LEGACY_RAMLIMIT_BYTES }, 143 }; 144 145 /* 146 * Highmem IO Regions: This memory map is floating, located after the RAM. 147 * Each MemMapEntry base (GPA) will be dynamically computed, depending on the 148 * top of the RAM, so that its base get the same alignment as the size, 149 * ie. a 512GiB entry will be aligned on a 512GiB boundary. If there is 150 * less than 256GiB of RAM, the floating area starts at the 256GiB mark. 151 * Note the extended_memmap is sized so that it eventually also includes the 152 * base_memmap entries (VIRT_HIGH_GIC_REDIST2 index is greater than the last 153 * index of base_memmap). 154 */ 155 static MemMapEntry extended_memmap[] = { 156 /* Additional 64 MB redist region (can contain up to 512 redistributors) */ 157 [VIRT_HIGH_GIC_REDIST2] = { 0x0, 64 * MiB }, 158 [VIRT_HIGH_PCIE_ECAM] = { 0x0, 256 * MiB }, 159 /* Second PCIe window */ 160 [VIRT_HIGH_PCIE_MMIO] = { 0x0, 512 * GiB }, 161 }; 162 163 static const int a15irqmap[] = { 164 [VIRT_UART] = 1, 165 [VIRT_RTC] = 2, 166 [VIRT_PCIE] = 3, /* ... to 6 */ 167 [VIRT_GPIO] = 7, 168 [VIRT_SECURE_UART] = 8, 169 [VIRT_MMIO] = 16, /* ...to 16 + NUM_VIRTIO_TRANSPORTS - 1 */ 170 [VIRT_GIC_V2M] = 48, /* ...to 48 + NUM_GICV2M_SPIS - 1 */ 171 [VIRT_SMMU] = 74, /* ...to 74 + NUM_SMMU_IRQS - 1 */ 172 [VIRT_PLATFORM_BUS] = 112, /* ...to 112 + PLATFORM_BUS_NUM_IRQS -1 */ 173 }; 174 175 static const char *valid_cpus[] = { 176 ARM_CPU_TYPE_NAME("cortex-a15"), 177 ARM_CPU_TYPE_NAME("cortex-a53"), 178 ARM_CPU_TYPE_NAME("cortex-a57"), 179 ARM_CPU_TYPE_NAME("cortex-a72"), 180 ARM_CPU_TYPE_NAME("host"), 181 ARM_CPU_TYPE_NAME("max"), 182 }; 183 184 static bool cpu_type_valid(const char *cpu) 185 { 186 int i; 187 188 for (i = 0; i < ARRAY_SIZE(valid_cpus); i++) { 189 if (strcmp(cpu, valid_cpus[i]) == 0) { 190 return true; 191 } 192 } 193 return false; 194 } 195 196 static void create_fdt(VirtMachineState *vms) 197 { 198 void *fdt = create_device_tree(&vms->fdt_size); 199 200 if (!fdt) { 201 error_report("create_device_tree() failed"); 202 exit(1); 203 } 204 205 vms->fdt = fdt; 206 207 /* Header */ 208 qemu_fdt_setprop_string(fdt, "/", "compatible", "linux,dummy-virt"); 209 qemu_fdt_setprop_cell(fdt, "/", "#address-cells", 0x2); 210 qemu_fdt_setprop_cell(fdt, "/", "#size-cells", 0x2); 211 212 /* /chosen must exist for load_dtb to fill in necessary properties later */ 213 qemu_fdt_add_subnode(fdt, "/chosen"); 214 215 /* Clock node, for the benefit of the UART. The kernel device tree 216 * binding documentation claims the PL011 node clock properties are 217 * optional but in practice if you omit them the kernel refuses to 218 * probe for the device. 219 */ 220 vms->clock_phandle = qemu_fdt_alloc_phandle(fdt); 221 qemu_fdt_add_subnode(fdt, "/apb-pclk"); 222 qemu_fdt_setprop_string(fdt, "/apb-pclk", "compatible", "fixed-clock"); 223 qemu_fdt_setprop_cell(fdt, "/apb-pclk", "#clock-cells", 0x0); 224 qemu_fdt_setprop_cell(fdt, "/apb-pclk", "clock-frequency", 24000000); 225 qemu_fdt_setprop_string(fdt, "/apb-pclk", "clock-output-names", 226 "clk24mhz"); 227 qemu_fdt_setprop_cell(fdt, "/apb-pclk", "phandle", vms->clock_phandle); 228 229 if (have_numa_distance) { 230 int size = nb_numa_nodes * nb_numa_nodes * 3 * sizeof(uint32_t); 231 uint32_t *matrix = g_malloc0(size); 232 int idx, i, j; 233 234 for (i = 0; i < nb_numa_nodes; i++) { 235 for (j = 0; j < nb_numa_nodes; j++) { 236 idx = (i * nb_numa_nodes + j) * 3; 237 matrix[idx + 0] = cpu_to_be32(i); 238 matrix[idx + 1] = cpu_to_be32(j); 239 matrix[idx + 2] = cpu_to_be32(numa_info[i].distance[j]); 240 } 241 } 242 243 qemu_fdt_add_subnode(fdt, "/distance-map"); 244 qemu_fdt_setprop_string(fdt, "/distance-map", "compatible", 245 "numa-distance-map-v1"); 246 qemu_fdt_setprop(fdt, "/distance-map", "distance-matrix", 247 matrix, size); 248 g_free(matrix); 249 } 250 } 251 252 static void fdt_add_timer_nodes(const VirtMachineState *vms) 253 { 254 /* On real hardware these interrupts are level-triggered. 255 * On KVM they were edge-triggered before host kernel version 4.4, 256 * and level-triggered afterwards. 257 * On emulated QEMU they are level-triggered. 258 * 259 * Getting the DTB info about them wrong is awkward for some 260 * guest kernels: 261 * pre-4.8 ignore the DT and leave the interrupt configured 262 * with whatever the GIC reset value (or the bootloader) left it at 263 * 4.8 before rc6 honour the incorrect data by programming it back 264 * into the GIC, causing problems 265 * 4.8rc6 and later ignore the DT and always write "level triggered" 266 * into the GIC 267 * 268 * For backwards-compatibility, virt-2.8 and earlier will continue 269 * to say these are edge-triggered, but later machines will report 270 * the correct information. 271 */ 272 ARMCPU *armcpu; 273 VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms); 274 uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI; 275 276 if (vmc->claim_edge_triggered_timers) { 277 irqflags = GIC_FDT_IRQ_FLAGS_EDGE_LO_HI; 278 } 279 280 if (vms->gic_version == 2) { 281 irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START, 282 GIC_FDT_IRQ_PPI_CPU_WIDTH, 283 (1 << vms->smp_cpus) - 1); 284 } 285 286 qemu_fdt_add_subnode(vms->fdt, "/timer"); 287 288 armcpu = ARM_CPU(qemu_get_cpu(0)); 289 if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) { 290 const char compat[] = "arm,armv8-timer\0arm,armv7-timer"; 291 qemu_fdt_setprop(vms->fdt, "/timer", "compatible", 292 compat, sizeof(compat)); 293 } else { 294 qemu_fdt_setprop_string(vms->fdt, "/timer", "compatible", 295 "arm,armv7-timer"); 296 } 297 qemu_fdt_setprop(vms->fdt, "/timer", "always-on", NULL, 0); 298 qemu_fdt_setprop_cells(vms->fdt, "/timer", "interrupts", 299 GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_S_EL1_IRQ, irqflags, 300 GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL1_IRQ, irqflags, 301 GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_VIRT_IRQ, irqflags, 302 GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL2_IRQ, irqflags); 303 } 304 305 static void fdt_add_cpu_nodes(const VirtMachineState *vms) 306 { 307 int cpu; 308 int addr_cells = 1; 309 const MachineState *ms = MACHINE(vms); 310 311 /* 312 * From Documentation/devicetree/bindings/arm/cpus.txt 313 * On ARM v8 64-bit systems value should be set to 2, 314 * that corresponds to the MPIDR_EL1 register size. 315 * If MPIDR_EL1[63:32] value is equal to 0 on all CPUs 316 * in the system, #address-cells can be set to 1, since 317 * MPIDR_EL1[63:32] bits are not used for CPUs 318 * identification. 319 * 320 * Here we actually don't know whether our system is 32- or 64-bit one. 321 * The simplest way to go is to examine affinity IDs of all our CPUs. If 322 * at least one of them has Aff3 populated, we set #address-cells to 2. 323 */ 324 for (cpu = 0; cpu < vms->smp_cpus; cpu++) { 325 ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu)); 326 327 if (armcpu->mp_affinity & ARM_AFF3_MASK) { 328 addr_cells = 2; 329 break; 330 } 331 } 332 333 qemu_fdt_add_subnode(vms->fdt, "/cpus"); 334 qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#address-cells", addr_cells); 335 qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#size-cells", 0x0); 336 337 for (cpu = vms->smp_cpus - 1; cpu >= 0; cpu--) { 338 char *nodename = g_strdup_printf("/cpus/cpu@%d", cpu); 339 ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu)); 340 CPUState *cs = CPU(armcpu); 341 342 qemu_fdt_add_subnode(vms->fdt, nodename); 343 qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "cpu"); 344 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", 345 armcpu->dtb_compatible); 346 347 if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED 348 && vms->smp_cpus > 1) { 349 qemu_fdt_setprop_string(vms->fdt, nodename, 350 "enable-method", "psci"); 351 } 352 353 if (addr_cells == 2) { 354 qemu_fdt_setprop_u64(vms->fdt, nodename, "reg", 355 armcpu->mp_affinity); 356 } else { 357 qemu_fdt_setprop_cell(vms->fdt, nodename, "reg", 358 armcpu->mp_affinity); 359 } 360 361 if (ms->possible_cpus->cpus[cs->cpu_index].props.has_node_id) { 362 qemu_fdt_setprop_cell(vms->fdt, nodename, "numa-node-id", 363 ms->possible_cpus->cpus[cs->cpu_index].props.node_id); 364 } 365 366 g_free(nodename); 367 } 368 } 369 370 static void fdt_add_its_gic_node(VirtMachineState *vms) 371 { 372 char *nodename; 373 374 vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt); 375 nodename = g_strdup_printf("/intc/its@%" PRIx64, 376 vms->memmap[VIRT_GIC_ITS].base); 377 qemu_fdt_add_subnode(vms->fdt, nodename); 378 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", 379 "arm,gic-v3-its"); 380 qemu_fdt_setprop(vms->fdt, nodename, "msi-controller", NULL, 0); 381 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 382 2, vms->memmap[VIRT_GIC_ITS].base, 383 2, vms->memmap[VIRT_GIC_ITS].size); 384 qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", vms->msi_phandle); 385 g_free(nodename); 386 } 387 388 static void fdt_add_v2m_gic_node(VirtMachineState *vms) 389 { 390 char *nodename; 391 392 nodename = g_strdup_printf("/intc/v2m@%" PRIx64, 393 vms->memmap[VIRT_GIC_V2M].base); 394 vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt); 395 qemu_fdt_add_subnode(vms->fdt, nodename); 396 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", 397 "arm,gic-v2m-frame"); 398 qemu_fdt_setprop(vms->fdt, nodename, "msi-controller", NULL, 0); 399 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 400 2, vms->memmap[VIRT_GIC_V2M].base, 401 2, vms->memmap[VIRT_GIC_V2M].size); 402 qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", vms->msi_phandle); 403 g_free(nodename); 404 } 405 406 static void fdt_add_gic_node(VirtMachineState *vms) 407 { 408 char *nodename; 409 410 vms->gic_phandle = qemu_fdt_alloc_phandle(vms->fdt); 411 qemu_fdt_setprop_cell(vms->fdt, "/", "interrupt-parent", vms->gic_phandle); 412 413 nodename = g_strdup_printf("/intc@%" PRIx64, 414 vms->memmap[VIRT_GIC_DIST].base); 415 qemu_fdt_add_subnode(vms->fdt, nodename); 416 qemu_fdt_setprop_cell(vms->fdt, nodename, "#interrupt-cells", 3); 417 qemu_fdt_setprop(vms->fdt, nodename, "interrupt-controller", NULL, 0); 418 qemu_fdt_setprop_cell(vms->fdt, nodename, "#address-cells", 0x2); 419 qemu_fdt_setprop_cell(vms->fdt, nodename, "#size-cells", 0x2); 420 qemu_fdt_setprop(vms->fdt, nodename, "ranges", NULL, 0); 421 if (vms->gic_version == 3) { 422 int nb_redist_regions = virt_gicv3_redist_region_count(vms); 423 424 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", 425 "arm,gic-v3"); 426 427 qemu_fdt_setprop_cell(vms->fdt, nodename, 428 "#redistributor-regions", nb_redist_regions); 429 430 if (nb_redist_regions == 1) { 431 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 432 2, vms->memmap[VIRT_GIC_DIST].base, 433 2, vms->memmap[VIRT_GIC_DIST].size, 434 2, vms->memmap[VIRT_GIC_REDIST].base, 435 2, vms->memmap[VIRT_GIC_REDIST].size); 436 } else { 437 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 438 2, vms->memmap[VIRT_GIC_DIST].base, 439 2, vms->memmap[VIRT_GIC_DIST].size, 440 2, vms->memmap[VIRT_GIC_REDIST].base, 441 2, vms->memmap[VIRT_GIC_REDIST].size, 442 2, vms->memmap[VIRT_HIGH_GIC_REDIST2].base, 443 2, vms->memmap[VIRT_HIGH_GIC_REDIST2].size); 444 } 445 446 if (vms->virt) { 447 qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts", 448 GIC_FDT_IRQ_TYPE_PPI, ARCH_GIC_MAINT_IRQ, 449 GIC_FDT_IRQ_FLAGS_LEVEL_HI); 450 } 451 } else { 452 /* 'cortex-a15-gic' means 'GIC v2' */ 453 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", 454 "arm,cortex-a15-gic"); 455 if (!vms->virt) { 456 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 457 2, vms->memmap[VIRT_GIC_DIST].base, 458 2, vms->memmap[VIRT_GIC_DIST].size, 459 2, vms->memmap[VIRT_GIC_CPU].base, 460 2, vms->memmap[VIRT_GIC_CPU].size); 461 } else { 462 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 463 2, vms->memmap[VIRT_GIC_DIST].base, 464 2, vms->memmap[VIRT_GIC_DIST].size, 465 2, vms->memmap[VIRT_GIC_CPU].base, 466 2, vms->memmap[VIRT_GIC_CPU].size, 467 2, vms->memmap[VIRT_GIC_HYP].base, 468 2, vms->memmap[VIRT_GIC_HYP].size, 469 2, vms->memmap[VIRT_GIC_VCPU].base, 470 2, vms->memmap[VIRT_GIC_VCPU].size); 471 qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts", 472 GIC_FDT_IRQ_TYPE_PPI, ARCH_GIC_MAINT_IRQ, 473 GIC_FDT_IRQ_FLAGS_LEVEL_HI); 474 } 475 } 476 477 qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", vms->gic_phandle); 478 g_free(nodename); 479 } 480 481 static void fdt_add_pmu_nodes(const VirtMachineState *vms) 482 { 483 CPUState *cpu; 484 ARMCPU *armcpu; 485 uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI; 486 487 CPU_FOREACH(cpu) { 488 armcpu = ARM_CPU(cpu); 489 if (!arm_feature(&armcpu->env, ARM_FEATURE_PMU)) { 490 return; 491 } 492 if (kvm_enabled()) { 493 if (kvm_irqchip_in_kernel()) { 494 kvm_arm_pmu_set_irq(cpu, PPI(VIRTUAL_PMU_IRQ)); 495 } 496 kvm_arm_pmu_init(cpu); 497 } 498 } 499 500 if (vms->gic_version == 2) { 501 irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START, 502 GIC_FDT_IRQ_PPI_CPU_WIDTH, 503 (1 << vms->smp_cpus) - 1); 504 } 505 506 armcpu = ARM_CPU(qemu_get_cpu(0)); 507 qemu_fdt_add_subnode(vms->fdt, "/pmu"); 508 if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) { 509 const char compat[] = "arm,armv8-pmuv3"; 510 qemu_fdt_setprop(vms->fdt, "/pmu", "compatible", 511 compat, sizeof(compat)); 512 qemu_fdt_setprop_cells(vms->fdt, "/pmu", "interrupts", 513 GIC_FDT_IRQ_TYPE_PPI, VIRTUAL_PMU_IRQ, irqflags); 514 } 515 } 516 517 static void create_its(VirtMachineState *vms, DeviceState *gicdev) 518 { 519 const char *itsclass = its_class_name(); 520 DeviceState *dev; 521 522 if (!itsclass) { 523 /* Do nothing if not supported */ 524 return; 525 } 526 527 dev = qdev_create(NULL, itsclass); 528 529 object_property_set_link(OBJECT(dev), OBJECT(gicdev), "parent-gicv3", 530 &error_abort); 531 qdev_init_nofail(dev); 532 sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_ITS].base); 533 534 fdt_add_its_gic_node(vms); 535 } 536 537 static void create_v2m(VirtMachineState *vms, qemu_irq *pic) 538 { 539 int i; 540 int irq = vms->irqmap[VIRT_GIC_V2M]; 541 DeviceState *dev; 542 543 dev = qdev_create(NULL, "arm-gicv2m"); 544 sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_V2M].base); 545 qdev_prop_set_uint32(dev, "base-spi", irq); 546 qdev_prop_set_uint32(dev, "num-spi", NUM_GICV2M_SPIS); 547 qdev_init_nofail(dev); 548 549 for (i = 0; i < NUM_GICV2M_SPIS; i++) { 550 sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]); 551 } 552 553 fdt_add_v2m_gic_node(vms); 554 } 555 556 static void create_gic(VirtMachineState *vms, qemu_irq *pic) 557 { 558 /* We create a standalone GIC */ 559 DeviceState *gicdev; 560 SysBusDevice *gicbusdev; 561 const char *gictype; 562 int type = vms->gic_version, i; 563 uint32_t nb_redist_regions = 0; 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 } 577 578 if (type == 3) { 579 uint32_t redist0_capacity = 580 vms->memmap[VIRT_GIC_REDIST].size / GICV3_REDIST_SIZE; 581 uint32_t redist0_count = MIN(smp_cpus, redist0_capacity); 582 583 nb_redist_regions = virt_gicv3_redist_region_count(vms); 584 585 qdev_prop_set_uint32(gicdev, "len-redist-region-count", 586 nb_redist_regions); 587 qdev_prop_set_uint32(gicdev, "redist-region-count[0]", redist0_count); 588 589 if (nb_redist_regions == 2) { 590 uint32_t redist1_capacity = 591 vms->memmap[VIRT_HIGH_GIC_REDIST2].size / GICV3_REDIST_SIZE; 592 593 qdev_prop_set_uint32(gicdev, "redist-region-count[1]", 594 MIN(smp_cpus - redist0_count, redist1_capacity)); 595 } 596 } else { 597 if (!kvm_irqchip_in_kernel()) { 598 qdev_prop_set_bit(gicdev, "has-virtualization-extensions", 599 vms->virt); 600 } 601 } 602 qdev_init_nofail(gicdev); 603 gicbusdev = SYS_BUS_DEVICE(gicdev); 604 sysbus_mmio_map(gicbusdev, 0, vms->memmap[VIRT_GIC_DIST].base); 605 if (type == 3) { 606 sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_REDIST].base); 607 if (nb_redist_regions == 2) { 608 sysbus_mmio_map(gicbusdev, 2, 609 vms->memmap[VIRT_HIGH_GIC_REDIST2].base); 610 } 611 } else { 612 sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base); 613 if (vms->virt) { 614 sysbus_mmio_map(gicbusdev, 2, vms->memmap[VIRT_GIC_HYP].base); 615 sysbus_mmio_map(gicbusdev, 3, vms->memmap[VIRT_GIC_VCPU].base); 616 } 617 } 618 619 /* Wire the outputs from each CPU's generic timer and the GICv3 620 * maintenance interrupt signal to the appropriate GIC PPI inputs, 621 * and the GIC's IRQ/FIQ/VIRQ/VFIQ interrupt outputs to the CPU's inputs. 622 */ 623 for (i = 0; i < smp_cpus; i++) { 624 DeviceState *cpudev = DEVICE(qemu_get_cpu(i)); 625 int ppibase = NUM_IRQS + i * GIC_INTERNAL + GIC_NR_SGIS; 626 int irq; 627 /* Mapping from the output timer irq lines from the CPU to the 628 * GIC PPI inputs we use for the virt board. 629 */ 630 const int timer_irq[] = { 631 [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ, 632 [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ, 633 [GTIMER_HYP] = ARCH_TIMER_NS_EL2_IRQ, 634 [GTIMER_SEC] = ARCH_TIMER_S_EL1_IRQ, 635 }; 636 637 for (irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) { 638 qdev_connect_gpio_out(cpudev, irq, 639 qdev_get_gpio_in(gicdev, 640 ppibase + timer_irq[irq])); 641 } 642 643 if (type == 3) { 644 qemu_irq irq = qdev_get_gpio_in(gicdev, 645 ppibase + ARCH_GIC_MAINT_IRQ); 646 qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt", 647 0, irq); 648 } else if (vms->virt) { 649 qemu_irq irq = qdev_get_gpio_in(gicdev, 650 ppibase + ARCH_GIC_MAINT_IRQ); 651 sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus, irq); 652 } 653 654 qdev_connect_gpio_out_named(cpudev, "pmu-interrupt", 0, 655 qdev_get_gpio_in(gicdev, ppibase 656 + VIRTUAL_PMU_IRQ)); 657 658 sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ)); 659 sysbus_connect_irq(gicbusdev, i + smp_cpus, 660 qdev_get_gpio_in(cpudev, ARM_CPU_FIQ)); 661 sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus, 662 qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ)); 663 sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus, 664 qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ)); 665 } 666 667 for (i = 0; i < NUM_IRQS; i++) { 668 pic[i] = qdev_get_gpio_in(gicdev, i); 669 } 670 671 fdt_add_gic_node(vms); 672 673 if (type == 3 && vms->its) { 674 create_its(vms, gicdev); 675 } else if (type == 2) { 676 create_v2m(vms, pic); 677 } 678 } 679 680 static void create_uart(const VirtMachineState *vms, qemu_irq *pic, int uart, 681 MemoryRegion *mem, Chardev *chr) 682 { 683 char *nodename; 684 hwaddr base = vms->memmap[uart].base; 685 hwaddr size = vms->memmap[uart].size; 686 int irq = vms->irqmap[uart]; 687 const char compat[] = "arm,pl011\0arm,primecell"; 688 const char clocknames[] = "uartclk\0apb_pclk"; 689 DeviceState *dev = qdev_create(NULL, "pl011"); 690 SysBusDevice *s = SYS_BUS_DEVICE(dev); 691 692 qdev_prop_set_chr(dev, "chardev", chr); 693 qdev_init_nofail(dev); 694 memory_region_add_subregion(mem, base, 695 sysbus_mmio_get_region(s, 0)); 696 sysbus_connect_irq(s, 0, pic[irq]); 697 698 nodename = g_strdup_printf("/pl011@%" PRIx64, base); 699 qemu_fdt_add_subnode(vms->fdt, nodename); 700 /* Note that we can't use setprop_string because of the embedded NUL */ 701 qemu_fdt_setprop(vms->fdt, nodename, "compatible", 702 compat, sizeof(compat)); 703 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 704 2, base, 2, size); 705 qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts", 706 GIC_FDT_IRQ_TYPE_SPI, irq, 707 GIC_FDT_IRQ_FLAGS_LEVEL_HI); 708 qemu_fdt_setprop_cells(vms->fdt, nodename, "clocks", 709 vms->clock_phandle, vms->clock_phandle); 710 qemu_fdt_setprop(vms->fdt, nodename, "clock-names", 711 clocknames, sizeof(clocknames)); 712 713 if (uart == VIRT_UART) { 714 qemu_fdt_setprop_string(vms->fdt, "/chosen", "stdout-path", nodename); 715 } else { 716 /* Mark as not usable by the normal world */ 717 qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled"); 718 qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay"); 719 720 qemu_fdt_add_subnode(vms->fdt, "/secure-chosen"); 721 qemu_fdt_setprop_string(vms->fdt, "/secure-chosen", "stdout-path", 722 nodename); 723 } 724 725 g_free(nodename); 726 } 727 728 static void create_rtc(const VirtMachineState *vms, qemu_irq *pic) 729 { 730 char *nodename; 731 hwaddr base = vms->memmap[VIRT_RTC].base; 732 hwaddr size = vms->memmap[VIRT_RTC].size; 733 int irq = vms->irqmap[VIRT_RTC]; 734 const char compat[] = "arm,pl031\0arm,primecell"; 735 736 sysbus_create_simple("pl031", base, pic[irq]); 737 738 nodename = g_strdup_printf("/pl031@%" PRIx64, base); 739 qemu_fdt_add_subnode(vms->fdt, nodename); 740 qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat)); 741 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 742 2, base, 2, size); 743 qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts", 744 GIC_FDT_IRQ_TYPE_SPI, irq, 745 GIC_FDT_IRQ_FLAGS_LEVEL_HI); 746 qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle); 747 qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk"); 748 g_free(nodename); 749 } 750 751 static DeviceState *gpio_key_dev; 752 static void virt_powerdown_req(Notifier *n, void *opaque) 753 { 754 /* use gpio Pin 3 for power button event */ 755 qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1); 756 } 757 758 static Notifier virt_system_powerdown_notifier = { 759 .notify = virt_powerdown_req 760 }; 761 762 static void create_gpio(const VirtMachineState *vms, qemu_irq *pic) 763 { 764 char *nodename; 765 DeviceState *pl061_dev; 766 hwaddr base = vms->memmap[VIRT_GPIO].base; 767 hwaddr size = vms->memmap[VIRT_GPIO].size; 768 int irq = vms->irqmap[VIRT_GPIO]; 769 const char compat[] = "arm,pl061\0arm,primecell"; 770 771 pl061_dev = sysbus_create_simple("pl061", base, pic[irq]); 772 773 uint32_t phandle = qemu_fdt_alloc_phandle(vms->fdt); 774 nodename = g_strdup_printf("/pl061@%" PRIx64, base); 775 qemu_fdt_add_subnode(vms->fdt, nodename); 776 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 777 2, base, 2, size); 778 qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat)); 779 qemu_fdt_setprop_cell(vms->fdt, nodename, "#gpio-cells", 2); 780 qemu_fdt_setprop(vms->fdt, nodename, "gpio-controller", NULL, 0); 781 qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts", 782 GIC_FDT_IRQ_TYPE_SPI, irq, 783 GIC_FDT_IRQ_FLAGS_LEVEL_HI); 784 qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle); 785 qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk"); 786 qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", phandle); 787 788 gpio_key_dev = sysbus_create_simple("gpio-key", -1, 789 qdev_get_gpio_in(pl061_dev, 3)); 790 qemu_fdt_add_subnode(vms->fdt, "/gpio-keys"); 791 qemu_fdt_setprop_string(vms->fdt, "/gpio-keys", "compatible", "gpio-keys"); 792 qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#size-cells", 0); 793 qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#address-cells", 1); 794 795 qemu_fdt_add_subnode(vms->fdt, "/gpio-keys/poweroff"); 796 qemu_fdt_setprop_string(vms->fdt, "/gpio-keys/poweroff", 797 "label", "GPIO Key Poweroff"); 798 qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys/poweroff", "linux,code", 799 KEY_POWER); 800 qemu_fdt_setprop_cells(vms->fdt, "/gpio-keys/poweroff", 801 "gpios", phandle, 3, 0); 802 803 /* connect powerdown request */ 804 qemu_register_powerdown_notifier(&virt_system_powerdown_notifier); 805 806 g_free(nodename); 807 } 808 809 static void create_virtio_devices(const VirtMachineState *vms, qemu_irq *pic) 810 { 811 int i; 812 hwaddr size = vms->memmap[VIRT_MMIO].size; 813 814 /* We create the transports in forwards order. Since qbus_realize() 815 * prepends (not appends) new child buses, the incrementing loop below will 816 * create a list of virtio-mmio buses with decreasing base addresses. 817 * 818 * When a -device option is processed from the command line, 819 * qbus_find_recursive() picks the next free virtio-mmio bus in forwards 820 * order. The upshot is that -device options in increasing command line 821 * order are mapped to virtio-mmio buses with decreasing base addresses. 822 * 823 * When this code was originally written, that arrangement ensured that the 824 * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to 825 * the first -device on the command line. (The end-to-end order is a 826 * function of this loop, qbus_realize(), qbus_find_recursive(), and the 827 * guest kernel's name-to-address assignment strategy.) 828 * 829 * Meanwhile, the kernel's traversal seems to have been reversed; see eg. 830 * the message, if not necessarily the code, of commit 70161ff336. 831 * Therefore the loop now establishes the inverse of the original intent. 832 * 833 * Unfortunately, we can't counteract the kernel change by reversing the 834 * loop; it would break existing command lines. 835 * 836 * In any case, the kernel makes no guarantee about the stability of 837 * enumeration order of virtio devices (as demonstrated by it changing 838 * between kernel versions). For reliable and stable identification 839 * of disks users must use UUIDs or similar mechanisms. 840 */ 841 for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) { 842 int irq = vms->irqmap[VIRT_MMIO] + i; 843 hwaddr base = vms->memmap[VIRT_MMIO].base + i * size; 844 845 sysbus_create_simple("virtio-mmio", base, pic[irq]); 846 } 847 848 /* We add dtb nodes in reverse order so that they appear in the finished 849 * device tree lowest address first. 850 * 851 * Note that this mapping is independent of the loop above. The previous 852 * loop influences virtio device to virtio transport assignment, whereas 853 * this loop controls how virtio transports are laid out in the dtb. 854 */ 855 for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) { 856 char *nodename; 857 int irq = vms->irqmap[VIRT_MMIO] + i; 858 hwaddr base = vms->memmap[VIRT_MMIO].base + i * size; 859 860 nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base); 861 qemu_fdt_add_subnode(vms->fdt, nodename); 862 qemu_fdt_setprop_string(vms->fdt, nodename, 863 "compatible", "virtio,mmio"); 864 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 865 2, base, 2, size); 866 qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts", 867 GIC_FDT_IRQ_TYPE_SPI, irq, 868 GIC_FDT_IRQ_FLAGS_EDGE_LO_HI); 869 qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0); 870 g_free(nodename); 871 } 872 } 873 874 static void create_one_flash(const char *name, hwaddr flashbase, 875 hwaddr flashsize, const char *file, 876 MemoryRegion *sysmem) 877 { 878 /* Create and map a single flash device. We use the same 879 * parameters as the flash devices on the Versatile Express board. 880 */ 881 DriveInfo *dinfo = drive_get_next(IF_PFLASH); 882 DeviceState *dev = qdev_create(NULL, TYPE_PFLASH_CFI01); 883 SysBusDevice *sbd = SYS_BUS_DEVICE(dev); 884 const uint64_t sectorlength = 256 * 1024; 885 886 if (dinfo) { 887 qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo), 888 &error_abort); 889 } 890 891 qdev_prop_set_uint32(dev, "num-blocks", flashsize / sectorlength); 892 qdev_prop_set_uint64(dev, "sector-length", sectorlength); 893 qdev_prop_set_uint8(dev, "width", 4); 894 qdev_prop_set_uint8(dev, "device-width", 2); 895 qdev_prop_set_bit(dev, "big-endian", false); 896 qdev_prop_set_uint16(dev, "id0", 0x89); 897 qdev_prop_set_uint16(dev, "id1", 0x18); 898 qdev_prop_set_uint16(dev, "id2", 0x00); 899 qdev_prop_set_uint16(dev, "id3", 0x00); 900 qdev_prop_set_string(dev, "name", name); 901 qdev_init_nofail(dev); 902 903 memory_region_add_subregion(sysmem, flashbase, 904 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0)); 905 906 if (file) { 907 char *fn; 908 int image_size; 909 910 if (drive_get(IF_PFLASH, 0, 0)) { 911 error_report("The contents of the first flash device may be " 912 "specified with -bios or with -drive if=pflash... " 913 "but you cannot use both options at once"); 914 exit(1); 915 } 916 fn = qemu_find_file(QEMU_FILE_TYPE_BIOS, file); 917 if (!fn) { 918 error_report("Could not find ROM image '%s'", file); 919 exit(1); 920 } 921 image_size = load_image_mr(fn, sysbus_mmio_get_region(sbd, 0)); 922 g_free(fn); 923 if (image_size < 0) { 924 error_report("Could not load ROM image '%s'", file); 925 exit(1); 926 } 927 } 928 } 929 930 static void create_flash(const VirtMachineState *vms, 931 MemoryRegion *sysmem, 932 MemoryRegion *secure_sysmem) 933 { 934 /* Create two flash devices to fill the VIRT_FLASH space in the memmap. 935 * Any file passed via -bios goes in the first of these. 936 * sysmem is the system memory space. secure_sysmem is the secure view 937 * of the system, and the first flash device should be made visible only 938 * there. The second flash device is visible to both secure and nonsecure. 939 * If sysmem == secure_sysmem this means there is no separate Secure 940 * address space and both flash devices are generally visible. 941 */ 942 hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2; 943 hwaddr flashbase = vms->memmap[VIRT_FLASH].base; 944 char *nodename; 945 946 create_one_flash("virt.flash0", flashbase, flashsize, 947 bios_name, secure_sysmem); 948 create_one_flash("virt.flash1", flashbase + flashsize, flashsize, 949 NULL, sysmem); 950 951 if (sysmem == secure_sysmem) { 952 /* Report both flash devices as a single node in the DT */ 953 nodename = g_strdup_printf("/flash@%" PRIx64, flashbase); 954 qemu_fdt_add_subnode(vms->fdt, nodename); 955 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash"); 956 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 957 2, flashbase, 2, flashsize, 958 2, flashbase + flashsize, 2, flashsize); 959 qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4); 960 g_free(nodename); 961 } else { 962 /* Report the devices as separate nodes so we can mark one as 963 * only visible to the secure world. 964 */ 965 nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase); 966 qemu_fdt_add_subnode(vms->fdt, nodename); 967 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash"); 968 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 969 2, flashbase, 2, flashsize); 970 qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4); 971 qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled"); 972 qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay"); 973 g_free(nodename); 974 975 nodename = g_strdup_printf("/flash@%" PRIx64, flashbase); 976 qemu_fdt_add_subnode(vms->fdt, nodename); 977 qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash"); 978 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 979 2, flashbase + flashsize, 2, flashsize); 980 qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4); 981 g_free(nodename); 982 } 983 } 984 985 static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as) 986 { 987 hwaddr base = vms->memmap[VIRT_FW_CFG].base; 988 hwaddr size = vms->memmap[VIRT_FW_CFG].size; 989 FWCfgState *fw_cfg; 990 char *nodename; 991 992 fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as); 993 fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)smp_cpus); 994 995 nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base); 996 qemu_fdt_add_subnode(vms->fdt, nodename); 997 qemu_fdt_setprop_string(vms->fdt, nodename, 998 "compatible", "qemu,fw-cfg-mmio"); 999 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 1000 2, base, 2, size); 1001 qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0); 1002 g_free(nodename); 1003 return fw_cfg; 1004 } 1005 1006 static void create_pcie_irq_map(const VirtMachineState *vms, 1007 uint32_t gic_phandle, 1008 int first_irq, const char *nodename) 1009 { 1010 int devfn, pin; 1011 uint32_t full_irq_map[4 * 4 * 10] = { 0 }; 1012 uint32_t *irq_map = full_irq_map; 1013 1014 for (devfn = 0; devfn <= 0x18; devfn += 0x8) { 1015 for (pin = 0; pin < 4; pin++) { 1016 int irq_type = GIC_FDT_IRQ_TYPE_SPI; 1017 int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS); 1018 int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI; 1019 int i; 1020 1021 uint32_t map[] = { 1022 devfn << 8, 0, 0, /* devfn */ 1023 pin + 1, /* PCI pin */ 1024 gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */ 1025 1026 /* Convert map to big endian */ 1027 for (i = 0; i < 10; i++) { 1028 irq_map[i] = cpu_to_be32(map[i]); 1029 } 1030 irq_map += 10; 1031 } 1032 } 1033 1034 qemu_fdt_setprop(vms->fdt, nodename, "interrupt-map", 1035 full_irq_map, sizeof(full_irq_map)); 1036 1037 qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupt-map-mask", 1038 0x1800, 0, 0, /* devfn (PCI_SLOT(3)) */ 1039 0x7 /* PCI irq */); 1040 } 1041 1042 static void create_smmu(const VirtMachineState *vms, qemu_irq *pic, 1043 PCIBus *bus) 1044 { 1045 char *node; 1046 const char compat[] = "arm,smmu-v3"; 1047 int irq = vms->irqmap[VIRT_SMMU]; 1048 int i; 1049 hwaddr base = vms->memmap[VIRT_SMMU].base; 1050 hwaddr size = vms->memmap[VIRT_SMMU].size; 1051 const char irq_names[] = "eventq\0priq\0cmdq-sync\0gerror"; 1052 DeviceState *dev; 1053 1054 if (vms->iommu != VIRT_IOMMU_SMMUV3 || !vms->iommu_phandle) { 1055 return; 1056 } 1057 1058 dev = qdev_create(NULL, "arm-smmuv3"); 1059 1060 object_property_set_link(OBJECT(dev), OBJECT(bus), "primary-bus", 1061 &error_abort); 1062 qdev_init_nofail(dev); 1063 sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, base); 1064 for (i = 0; i < NUM_SMMU_IRQS; i++) { 1065 sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]); 1066 } 1067 1068 node = g_strdup_printf("/smmuv3@%" PRIx64, base); 1069 qemu_fdt_add_subnode(vms->fdt, node); 1070 qemu_fdt_setprop(vms->fdt, node, "compatible", compat, sizeof(compat)); 1071 qemu_fdt_setprop_sized_cells(vms->fdt, node, "reg", 2, base, 2, size); 1072 1073 qemu_fdt_setprop_cells(vms->fdt, node, "interrupts", 1074 GIC_FDT_IRQ_TYPE_SPI, irq , GIC_FDT_IRQ_FLAGS_EDGE_LO_HI, 1075 GIC_FDT_IRQ_TYPE_SPI, irq + 1, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI, 1076 GIC_FDT_IRQ_TYPE_SPI, irq + 2, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI, 1077 GIC_FDT_IRQ_TYPE_SPI, irq + 3, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI); 1078 1079 qemu_fdt_setprop(vms->fdt, node, "interrupt-names", irq_names, 1080 sizeof(irq_names)); 1081 1082 qemu_fdt_setprop_cell(vms->fdt, node, "clocks", vms->clock_phandle); 1083 qemu_fdt_setprop_string(vms->fdt, node, "clock-names", "apb_pclk"); 1084 qemu_fdt_setprop(vms->fdt, node, "dma-coherent", NULL, 0); 1085 1086 qemu_fdt_setprop_cell(vms->fdt, node, "#iommu-cells", 1); 1087 1088 qemu_fdt_setprop_cell(vms->fdt, node, "phandle", vms->iommu_phandle); 1089 g_free(node); 1090 } 1091 1092 static void create_pcie(VirtMachineState *vms, qemu_irq *pic) 1093 { 1094 hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base; 1095 hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size; 1096 hwaddr base_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].base; 1097 hwaddr size_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].size; 1098 hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base; 1099 hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size; 1100 hwaddr base_ecam, size_ecam; 1101 hwaddr base = base_mmio; 1102 int nr_pcie_buses; 1103 int irq = vms->irqmap[VIRT_PCIE]; 1104 MemoryRegion *mmio_alias; 1105 MemoryRegion *mmio_reg; 1106 MemoryRegion *ecam_alias; 1107 MemoryRegion *ecam_reg; 1108 DeviceState *dev; 1109 char *nodename; 1110 int i, ecam_id; 1111 PCIHostState *pci; 1112 1113 dev = qdev_create(NULL, TYPE_GPEX_HOST); 1114 qdev_init_nofail(dev); 1115 1116 ecam_id = VIRT_ECAM_ID(vms->highmem_ecam); 1117 base_ecam = vms->memmap[ecam_id].base; 1118 size_ecam = vms->memmap[ecam_id].size; 1119 nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN; 1120 /* Map only the first size_ecam bytes of ECAM space */ 1121 ecam_alias = g_new0(MemoryRegion, 1); 1122 ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0); 1123 memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam", 1124 ecam_reg, 0, size_ecam); 1125 memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias); 1126 1127 /* Map the MMIO window into system address space so as to expose 1128 * the section of PCI MMIO space which starts at the same base address 1129 * (ie 1:1 mapping for that part of PCI MMIO space visible through 1130 * the window). 1131 */ 1132 mmio_alias = g_new0(MemoryRegion, 1); 1133 mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1); 1134 memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio", 1135 mmio_reg, base_mmio, size_mmio); 1136 memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias); 1137 1138 if (vms->highmem) { 1139 /* Map high MMIO space */ 1140 MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1); 1141 1142 memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high", 1143 mmio_reg, base_mmio_high, size_mmio_high); 1144 memory_region_add_subregion(get_system_memory(), base_mmio_high, 1145 high_mmio_alias); 1146 } 1147 1148 /* Map IO port space */ 1149 sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio); 1150 1151 for (i = 0; i < GPEX_NUM_IRQS; i++) { 1152 sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]); 1153 gpex_set_irq_num(GPEX_HOST(dev), i, irq + i); 1154 } 1155 1156 pci = PCI_HOST_BRIDGE(dev); 1157 if (pci->bus) { 1158 for (i = 0; i < nb_nics; i++) { 1159 NICInfo *nd = &nd_table[i]; 1160 1161 if (!nd->model) { 1162 nd->model = g_strdup("virtio"); 1163 } 1164 1165 pci_nic_init_nofail(nd, pci->bus, nd->model, NULL); 1166 } 1167 } 1168 1169 nodename = g_strdup_printf("/pcie@%" PRIx64, base); 1170 qemu_fdt_add_subnode(vms->fdt, nodename); 1171 qemu_fdt_setprop_string(vms->fdt, nodename, 1172 "compatible", "pci-host-ecam-generic"); 1173 qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "pci"); 1174 qemu_fdt_setprop_cell(vms->fdt, nodename, "#address-cells", 3); 1175 qemu_fdt_setprop_cell(vms->fdt, nodename, "#size-cells", 2); 1176 qemu_fdt_setprop_cell(vms->fdt, nodename, "linux,pci-domain", 0); 1177 qemu_fdt_setprop_cells(vms->fdt, nodename, "bus-range", 0, 1178 nr_pcie_buses - 1); 1179 qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0); 1180 1181 if (vms->msi_phandle) { 1182 qemu_fdt_setprop_cells(vms->fdt, nodename, "msi-parent", 1183 vms->msi_phandle); 1184 } 1185 1186 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 1187 2, base_ecam, 2, size_ecam); 1188 1189 if (vms->highmem) { 1190 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges", 1191 1, FDT_PCI_RANGE_IOPORT, 2, 0, 1192 2, base_pio, 2, size_pio, 1193 1, FDT_PCI_RANGE_MMIO, 2, base_mmio, 1194 2, base_mmio, 2, size_mmio, 1195 1, FDT_PCI_RANGE_MMIO_64BIT, 1196 2, base_mmio_high, 1197 2, base_mmio_high, 2, size_mmio_high); 1198 } else { 1199 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges", 1200 1, FDT_PCI_RANGE_IOPORT, 2, 0, 1201 2, base_pio, 2, size_pio, 1202 1, FDT_PCI_RANGE_MMIO, 2, base_mmio, 1203 2, base_mmio, 2, size_mmio); 1204 } 1205 1206 qemu_fdt_setprop_cell(vms->fdt, nodename, "#interrupt-cells", 1); 1207 create_pcie_irq_map(vms, vms->gic_phandle, irq, nodename); 1208 1209 if (vms->iommu) { 1210 vms->iommu_phandle = qemu_fdt_alloc_phandle(vms->fdt); 1211 1212 create_smmu(vms, pic, pci->bus); 1213 1214 qemu_fdt_setprop_cells(vms->fdt, nodename, "iommu-map", 1215 0x0, vms->iommu_phandle, 0x0, 0x10000); 1216 } 1217 1218 g_free(nodename); 1219 } 1220 1221 static void create_platform_bus(VirtMachineState *vms, qemu_irq *pic) 1222 { 1223 DeviceState *dev; 1224 SysBusDevice *s; 1225 int i; 1226 MemoryRegion *sysmem = get_system_memory(); 1227 1228 dev = qdev_create(NULL, TYPE_PLATFORM_BUS_DEVICE); 1229 dev->id = TYPE_PLATFORM_BUS_DEVICE; 1230 qdev_prop_set_uint32(dev, "num_irqs", PLATFORM_BUS_NUM_IRQS); 1231 qdev_prop_set_uint32(dev, "mmio_size", vms->memmap[VIRT_PLATFORM_BUS].size); 1232 qdev_init_nofail(dev); 1233 vms->platform_bus_dev = dev; 1234 1235 s = SYS_BUS_DEVICE(dev); 1236 for (i = 0; i < PLATFORM_BUS_NUM_IRQS; i++) { 1237 int irqn = vms->irqmap[VIRT_PLATFORM_BUS] + i; 1238 sysbus_connect_irq(s, i, pic[irqn]); 1239 } 1240 1241 memory_region_add_subregion(sysmem, 1242 vms->memmap[VIRT_PLATFORM_BUS].base, 1243 sysbus_mmio_get_region(s, 0)); 1244 } 1245 1246 static void create_secure_ram(VirtMachineState *vms, 1247 MemoryRegion *secure_sysmem) 1248 { 1249 MemoryRegion *secram = g_new(MemoryRegion, 1); 1250 char *nodename; 1251 hwaddr base = vms->memmap[VIRT_SECURE_MEM].base; 1252 hwaddr size = vms->memmap[VIRT_SECURE_MEM].size; 1253 1254 memory_region_init_ram(secram, NULL, "virt.secure-ram", size, 1255 &error_fatal); 1256 memory_region_add_subregion(secure_sysmem, base, secram); 1257 1258 nodename = g_strdup_printf("/secram@%" PRIx64, base); 1259 qemu_fdt_add_subnode(vms->fdt, nodename); 1260 qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "memory"); 1261 qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 2, base, 2, size); 1262 qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled"); 1263 qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay"); 1264 1265 g_free(nodename); 1266 } 1267 1268 static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size) 1269 { 1270 const VirtMachineState *board = container_of(binfo, VirtMachineState, 1271 bootinfo); 1272 1273 *fdt_size = board->fdt_size; 1274 return board->fdt; 1275 } 1276 1277 static void virt_build_smbios(VirtMachineState *vms) 1278 { 1279 MachineClass *mc = MACHINE_GET_CLASS(vms); 1280 VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms); 1281 uint8_t *smbios_tables, *smbios_anchor; 1282 size_t smbios_tables_len, smbios_anchor_len; 1283 const char *product = "QEMU Virtual Machine"; 1284 1285 if (kvm_enabled()) { 1286 product = "KVM Virtual Machine"; 1287 } 1288 1289 smbios_set_defaults("QEMU", product, 1290 vmc->smbios_old_sys_ver ? "1.0" : mc->name, false, 1291 true, SMBIOS_ENTRY_POINT_30); 1292 1293 smbios_get_tables(NULL, 0, &smbios_tables, &smbios_tables_len, 1294 &smbios_anchor, &smbios_anchor_len); 1295 1296 if (smbios_anchor) { 1297 fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables", 1298 smbios_tables, smbios_tables_len); 1299 fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor", 1300 smbios_anchor, smbios_anchor_len); 1301 } 1302 } 1303 1304 static 1305 void virt_machine_done(Notifier *notifier, void *data) 1306 { 1307 VirtMachineState *vms = container_of(notifier, VirtMachineState, 1308 machine_done); 1309 ARMCPU *cpu = ARM_CPU(first_cpu); 1310 struct arm_boot_info *info = &vms->bootinfo; 1311 AddressSpace *as = arm_boot_address_space(cpu, info); 1312 1313 /* 1314 * If the user provided a dtb, we assume the dynamic sysbus nodes 1315 * already are integrated there. This corresponds to a use case where 1316 * the dynamic sysbus nodes are complex and their generation is not yet 1317 * supported. In that case the user can take charge of the guest dt 1318 * while qemu takes charge of the qom stuff. 1319 */ 1320 if (info->dtb_filename == NULL) { 1321 platform_bus_add_all_fdt_nodes(vms->fdt, "/intc", 1322 vms->memmap[VIRT_PLATFORM_BUS].base, 1323 vms->memmap[VIRT_PLATFORM_BUS].size, 1324 vms->irqmap[VIRT_PLATFORM_BUS]); 1325 } 1326 if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as) < 0) { 1327 exit(1); 1328 } 1329 1330 virt_acpi_setup(vms); 1331 virt_build_smbios(vms); 1332 } 1333 1334 static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx) 1335 { 1336 uint8_t clustersz = ARM_DEFAULT_CPUS_PER_CLUSTER; 1337 VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms); 1338 1339 if (!vmc->disallow_affinity_adjustment) { 1340 /* Adjust MPIDR like 64-bit KVM hosts, which incorporate the 1341 * GIC's target-list limitations. 32-bit KVM hosts currently 1342 * always create clusters of 4 CPUs, but that is expected to 1343 * change when they gain support for gicv3. When KVM is enabled 1344 * it will override the changes we make here, therefore our 1345 * purposes are to make TCG consistent (with 64-bit KVM hosts) 1346 * and to improve SGI efficiency. 1347 */ 1348 if (vms->gic_version == 3) { 1349 clustersz = GICV3_TARGETLIST_BITS; 1350 } else { 1351 clustersz = GIC_TARGETLIST_BITS; 1352 } 1353 } 1354 return arm_cpu_mp_affinity(idx, clustersz); 1355 } 1356 1357 static void virt_set_memmap(VirtMachineState *vms) 1358 { 1359 MachineState *ms = MACHINE(vms); 1360 hwaddr base, device_memory_base, device_memory_size; 1361 int i; 1362 1363 vms->memmap = extended_memmap; 1364 1365 for (i = 0; i < ARRAY_SIZE(base_memmap); i++) { 1366 vms->memmap[i] = base_memmap[i]; 1367 } 1368 1369 if (ms->ram_slots > ACPI_MAX_RAM_SLOTS) { 1370 error_report("unsupported number of memory slots: %"PRIu64, 1371 ms->ram_slots); 1372 exit(EXIT_FAILURE); 1373 } 1374 1375 /* 1376 * We compute the base of the high IO region depending on the 1377 * amount of initial and device memory. The device memory start/size 1378 * is aligned on 1GiB. We never put the high IO region below 256GiB 1379 * so that if maxram_size is < 255GiB we keep the legacy memory map. 1380 * The device region size assumes 1GiB page max alignment per slot. 1381 */ 1382 device_memory_base = 1383 ROUND_UP(vms->memmap[VIRT_MEM].base + ms->ram_size, GiB); 1384 device_memory_size = ms->maxram_size - ms->ram_size + ms->ram_slots * GiB; 1385 1386 /* Base address of the high IO region */ 1387 base = device_memory_base + ROUND_UP(device_memory_size, GiB); 1388 if (base < device_memory_base) { 1389 error_report("maxmem/slots too huge"); 1390 exit(EXIT_FAILURE); 1391 } 1392 if (base < vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES) { 1393 base = vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES; 1394 } 1395 1396 for (i = VIRT_LOWMEMMAP_LAST; i < ARRAY_SIZE(extended_memmap); i++) { 1397 hwaddr size = extended_memmap[i].size; 1398 1399 base = ROUND_UP(base, size); 1400 vms->memmap[i].base = base; 1401 vms->memmap[i].size = size; 1402 base += size; 1403 } 1404 vms->highest_gpa = base - 1; 1405 if (device_memory_size > 0) { 1406 ms->device_memory = g_malloc0(sizeof(*ms->device_memory)); 1407 ms->device_memory->base = device_memory_base; 1408 memory_region_init(&ms->device_memory->mr, OBJECT(vms), 1409 "device-memory", device_memory_size); 1410 } 1411 } 1412 1413 static void machvirt_init(MachineState *machine) 1414 { 1415 VirtMachineState *vms = VIRT_MACHINE(machine); 1416 VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine); 1417 MachineClass *mc = MACHINE_GET_CLASS(machine); 1418 const CPUArchIdList *possible_cpus; 1419 qemu_irq pic[NUM_IRQS]; 1420 MemoryRegion *sysmem = get_system_memory(); 1421 MemoryRegion *secure_sysmem = NULL; 1422 int n, virt_max_cpus; 1423 MemoryRegion *ram = g_new(MemoryRegion, 1); 1424 bool firmware_loaded = bios_name || drive_get(IF_PFLASH, 0, 0); 1425 bool aarch64 = true; 1426 1427 /* 1428 * In accelerated mode, the memory map is computed earlier in kvm_type() 1429 * to create a VM with the right number of IPA bits. 1430 */ 1431 if (!vms->memmap) { 1432 virt_set_memmap(vms); 1433 } 1434 1435 /* We can probe only here because during property set 1436 * KVM is not available yet 1437 */ 1438 if (vms->gic_version <= 0) { 1439 /* "host" or "max" */ 1440 if (!kvm_enabled()) { 1441 if (vms->gic_version == 0) { 1442 error_report("gic-version=host requires KVM"); 1443 exit(1); 1444 } else { 1445 /* "max": currently means 3 for TCG */ 1446 vms->gic_version = 3; 1447 } 1448 } else { 1449 vms->gic_version = kvm_arm_vgic_probe(); 1450 if (!vms->gic_version) { 1451 error_report( 1452 "Unable to determine GIC version supported by host"); 1453 exit(1); 1454 } 1455 } 1456 } 1457 1458 if (!cpu_type_valid(machine->cpu_type)) { 1459 error_report("mach-virt: CPU type %s not supported", machine->cpu_type); 1460 exit(1); 1461 } 1462 1463 /* If we have an EL3 boot ROM then the assumption is that it will 1464 * implement PSCI itself, so disable QEMU's internal implementation 1465 * so it doesn't get in the way. Instead of starting secondary 1466 * CPUs in PSCI powerdown state we will start them all running and 1467 * let the boot ROM sort them out. 1468 * The usual case is that we do use QEMU's PSCI implementation; 1469 * if the guest has EL2 then we will use SMC as the conduit, 1470 * and otherwise we will use HVC (for backwards compatibility and 1471 * because if we're using KVM then we must use HVC). 1472 */ 1473 if (vms->secure && firmware_loaded) { 1474 vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED; 1475 } else if (vms->virt) { 1476 vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC; 1477 } else { 1478 vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC; 1479 } 1480 1481 /* The maximum number of CPUs depends on the GIC version, or on how 1482 * many redistributors we can fit into the memory map. 1483 */ 1484 if (vms->gic_version == 3) { 1485 virt_max_cpus = 1486 vms->memmap[VIRT_GIC_REDIST].size / GICV3_REDIST_SIZE; 1487 virt_max_cpus += 1488 vms->memmap[VIRT_HIGH_GIC_REDIST2].size / GICV3_REDIST_SIZE; 1489 } else { 1490 virt_max_cpus = GIC_NCPU; 1491 } 1492 1493 if (max_cpus > virt_max_cpus) { 1494 error_report("Number of SMP CPUs requested (%d) exceeds max CPUs " 1495 "supported by machine 'mach-virt' (%d)", 1496 max_cpus, virt_max_cpus); 1497 exit(1); 1498 } 1499 1500 vms->smp_cpus = smp_cpus; 1501 1502 if (vms->virt && kvm_enabled()) { 1503 error_report("mach-virt: KVM does not support providing " 1504 "Virtualization extensions to the guest CPU"); 1505 exit(1); 1506 } 1507 1508 if (vms->secure) { 1509 if (kvm_enabled()) { 1510 error_report("mach-virt: KVM does not support Security extensions"); 1511 exit(1); 1512 } 1513 1514 /* The Secure view of the world is the same as the NonSecure, 1515 * but with a few extra devices. Create it as a container region 1516 * containing the system memory at low priority; any secure-only 1517 * devices go in at higher priority and take precedence. 1518 */ 1519 secure_sysmem = g_new(MemoryRegion, 1); 1520 memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory", 1521 UINT64_MAX); 1522 memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1); 1523 } 1524 1525 create_fdt(vms); 1526 1527 possible_cpus = mc->possible_cpu_arch_ids(machine); 1528 for (n = 0; n < possible_cpus->len; n++) { 1529 Object *cpuobj; 1530 CPUState *cs; 1531 1532 if (n >= smp_cpus) { 1533 break; 1534 } 1535 1536 cpuobj = object_new(possible_cpus->cpus[n].type); 1537 object_property_set_int(cpuobj, possible_cpus->cpus[n].arch_id, 1538 "mp-affinity", NULL); 1539 1540 cs = CPU(cpuobj); 1541 cs->cpu_index = n; 1542 1543 numa_cpu_pre_plug(&possible_cpus->cpus[cs->cpu_index], DEVICE(cpuobj), 1544 &error_fatal); 1545 1546 aarch64 &= object_property_get_bool(cpuobj, "aarch64", NULL); 1547 1548 if (!vms->secure) { 1549 object_property_set_bool(cpuobj, false, "has_el3", NULL); 1550 } 1551 1552 if (!vms->virt && object_property_find(cpuobj, "has_el2", NULL)) { 1553 object_property_set_bool(cpuobj, false, "has_el2", NULL); 1554 } 1555 1556 if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) { 1557 object_property_set_int(cpuobj, vms->psci_conduit, 1558 "psci-conduit", NULL); 1559 1560 /* Secondary CPUs start in PSCI powered-down state */ 1561 if (n > 0) { 1562 object_property_set_bool(cpuobj, true, 1563 "start-powered-off", NULL); 1564 } 1565 } 1566 1567 if (vmc->no_pmu && object_property_find(cpuobj, "pmu", NULL)) { 1568 object_property_set_bool(cpuobj, false, "pmu", NULL); 1569 } 1570 1571 if (object_property_find(cpuobj, "reset-cbar", NULL)) { 1572 object_property_set_int(cpuobj, vms->memmap[VIRT_CPUPERIPHS].base, 1573 "reset-cbar", &error_abort); 1574 } 1575 1576 object_property_set_link(cpuobj, OBJECT(sysmem), "memory", 1577 &error_abort); 1578 if (vms->secure) { 1579 object_property_set_link(cpuobj, OBJECT(secure_sysmem), 1580 "secure-memory", &error_abort); 1581 } 1582 1583 object_property_set_bool(cpuobj, true, "realized", &error_fatal); 1584 object_unref(cpuobj); 1585 } 1586 fdt_add_timer_nodes(vms); 1587 fdt_add_cpu_nodes(vms); 1588 1589 if (!kvm_enabled()) { 1590 ARMCPU *cpu = ARM_CPU(first_cpu); 1591 bool aarch64 = object_property_get_bool(OBJECT(cpu), "aarch64", NULL); 1592 1593 if (aarch64 && vms->highmem) { 1594 int requested_pa_size, pamax = arm_pamax(cpu); 1595 1596 requested_pa_size = 64 - clz64(vms->highest_gpa); 1597 if (pamax < requested_pa_size) { 1598 error_report("VCPU supports less PA bits (%d) than requested " 1599 "by the memory map (%d)", pamax, requested_pa_size); 1600 exit(1); 1601 } 1602 } 1603 } 1604 1605 memory_region_allocate_system_memory(ram, NULL, "mach-virt.ram", 1606 machine->ram_size); 1607 memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base, ram); 1608 if (machine->device_memory) { 1609 memory_region_add_subregion(sysmem, machine->device_memory->base, 1610 &machine->device_memory->mr); 1611 } 1612 1613 create_flash(vms, sysmem, secure_sysmem ? secure_sysmem : sysmem); 1614 1615 create_gic(vms, pic); 1616 1617 fdt_add_pmu_nodes(vms); 1618 1619 create_uart(vms, pic, VIRT_UART, sysmem, serial_hd(0)); 1620 1621 if (vms->secure) { 1622 create_secure_ram(vms, secure_sysmem); 1623 create_uart(vms, pic, VIRT_SECURE_UART, secure_sysmem, serial_hd(1)); 1624 } 1625 1626 vms->highmem_ecam &= vms->highmem && (!firmware_loaded || aarch64); 1627 1628 create_rtc(vms, pic); 1629 1630 create_pcie(vms, pic); 1631 1632 create_gpio(vms, pic); 1633 1634 /* Create mmio transports, so the user can create virtio backends 1635 * (which will be automatically plugged in to the transports). If 1636 * no backend is created the transport will just sit harmlessly idle. 1637 */ 1638 create_virtio_devices(vms, pic); 1639 1640 vms->fw_cfg = create_fw_cfg(vms, &address_space_memory); 1641 rom_set_fw(vms->fw_cfg); 1642 1643 create_platform_bus(vms, pic); 1644 1645 vms->bootinfo.ram_size = machine->ram_size; 1646 vms->bootinfo.kernel_filename = machine->kernel_filename; 1647 vms->bootinfo.kernel_cmdline = machine->kernel_cmdline; 1648 vms->bootinfo.initrd_filename = machine->initrd_filename; 1649 vms->bootinfo.nb_cpus = smp_cpus; 1650 vms->bootinfo.board_id = -1; 1651 vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base; 1652 vms->bootinfo.get_dtb = machvirt_dtb; 1653 vms->bootinfo.skip_dtb_autoload = true; 1654 vms->bootinfo.firmware_loaded = firmware_loaded; 1655 arm_load_kernel(ARM_CPU(first_cpu), &vms->bootinfo); 1656 1657 vms->machine_done.notify = virt_machine_done; 1658 qemu_add_machine_init_done_notifier(&vms->machine_done); 1659 } 1660 1661 static bool virt_get_secure(Object *obj, Error **errp) 1662 { 1663 VirtMachineState *vms = VIRT_MACHINE(obj); 1664 1665 return vms->secure; 1666 } 1667 1668 static void virt_set_secure(Object *obj, bool value, Error **errp) 1669 { 1670 VirtMachineState *vms = VIRT_MACHINE(obj); 1671 1672 vms->secure = value; 1673 } 1674 1675 static bool virt_get_virt(Object *obj, Error **errp) 1676 { 1677 VirtMachineState *vms = VIRT_MACHINE(obj); 1678 1679 return vms->virt; 1680 } 1681 1682 static void virt_set_virt(Object *obj, bool value, Error **errp) 1683 { 1684 VirtMachineState *vms = VIRT_MACHINE(obj); 1685 1686 vms->virt = value; 1687 } 1688 1689 static bool virt_get_highmem(Object *obj, Error **errp) 1690 { 1691 VirtMachineState *vms = VIRT_MACHINE(obj); 1692 1693 return vms->highmem; 1694 } 1695 1696 static void virt_set_highmem(Object *obj, bool value, Error **errp) 1697 { 1698 VirtMachineState *vms = VIRT_MACHINE(obj); 1699 1700 vms->highmem = value; 1701 } 1702 1703 static bool virt_get_its(Object *obj, Error **errp) 1704 { 1705 VirtMachineState *vms = VIRT_MACHINE(obj); 1706 1707 return vms->its; 1708 } 1709 1710 static void virt_set_its(Object *obj, bool value, Error **errp) 1711 { 1712 VirtMachineState *vms = VIRT_MACHINE(obj); 1713 1714 vms->its = value; 1715 } 1716 1717 static char *virt_get_gic_version(Object *obj, Error **errp) 1718 { 1719 VirtMachineState *vms = VIRT_MACHINE(obj); 1720 const char *val = vms->gic_version == 3 ? "3" : "2"; 1721 1722 return g_strdup(val); 1723 } 1724 1725 static void virt_set_gic_version(Object *obj, const char *value, Error **errp) 1726 { 1727 VirtMachineState *vms = VIRT_MACHINE(obj); 1728 1729 if (!strcmp(value, "3")) { 1730 vms->gic_version = 3; 1731 } else if (!strcmp(value, "2")) { 1732 vms->gic_version = 2; 1733 } else if (!strcmp(value, "host")) { 1734 vms->gic_version = 0; /* Will probe later */ 1735 } else if (!strcmp(value, "max")) { 1736 vms->gic_version = -1; /* Will probe later */ 1737 } else { 1738 error_setg(errp, "Invalid gic-version value"); 1739 error_append_hint(errp, "Valid values are 3, 2, host, max.\n"); 1740 } 1741 } 1742 1743 static char *virt_get_iommu(Object *obj, Error **errp) 1744 { 1745 VirtMachineState *vms = VIRT_MACHINE(obj); 1746 1747 switch (vms->iommu) { 1748 case VIRT_IOMMU_NONE: 1749 return g_strdup("none"); 1750 case VIRT_IOMMU_SMMUV3: 1751 return g_strdup("smmuv3"); 1752 default: 1753 g_assert_not_reached(); 1754 } 1755 } 1756 1757 static void virt_set_iommu(Object *obj, const char *value, Error **errp) 1758 { 1759 VirtMachineState *vms = VIRT_MACHINE(obj); 1760 1761 if (!strcmp(value, "smmuv3")) { 1762 vms->iommu = VIRT_IOMMU_SMMUV3; 1763 } else if (!strcmp(value, "none")) { 1764 vms->iommu = VIRT_IOMMU_NONE; 1765 } else { 1766 error_setg(errp, "Invalid iommu value"); 1767 error_append_hint(errp, "Valid values are none, smmuv3.\n"); 1768 } 1769 } 1770 1771 static CpuInstanceProperties 1772 virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index) 1773 { 1774 MachineClass *mc = MACHINE_GET_CLASS(ms); 1775 const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms); 1776 1777 assert(cpu_index < possible_cpus->len); 1778 return possible_cpus->cpus[cpu_index].props; 1779 } 1780 1781 static int64_t virt_get_default_cpu_node_id(const MachineState *ms, int idx) 1782 { 1783 return idx % nb_numa_nodes; 1784 } 1785 1786 static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms) 1787 { 1788 int n; 1789 VirtMachineState *vms = VIRT_MACHINE(ms); 1790 1791 if (ms->possible_cpus) { 1792 assert(ms->possible_cpus->len == max_cpus); 1793 return ms->possible_cpus; 1794 } 1795 1796 ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) + 1797 sizeof(CPUArchId) * max_cpus); 1798 ms->possible_cpus->len = max_cpus; 1799 for (n = 0; n < ms->possible_cpus->len; n++) { 1800 ms->possible_cpus->cpus[n].type = ms->cpu_type; 1801 ms->possible_cpus->cpus[n].arch_id = 1802 virt_cpu_mp_affinity(vms, n); 1803 ms->possible_cpus->cpus[n].props.has_thread_id = true; 1804 ms->possible_cpus->cpus[n].props.thread_id = n; 1805 } 1806 return ms->possible_cpus; 1807 } 1808 1809 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev, 1810 DeviceState *dev, Error **errp) 1811 { 1812 VirtMachineState *vms = VIRT_MACHINE(hotplug_dev); 1813 1814 if (vms->platform_bus_dev) { 1815 if (object_dynamic_cast(OBJECT(dev), TYPE_SYS_BUS_DEVICE)) { 1816 platform_bus_link_device(PLATFORM_BUS_DEVICE(vms->platform_bus_dev), 1817 SYS_BUS_DEVICE(dev)); 1818 } 1819 } 1820 } 1821 1822 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine, 1823 DeviceState *dev) 1824 { 1825 if (object_dynamic_cast(OBJECT(dev), TYPE_SYS_BUS_DEVICE)) { 1826 return HOTPLUG_HANDLER(machine); 1827 } 1828 1829 return NULL; 1830 } 1831 1832 /* 1833 * for arm64 kvm_type [7-0] encodes the requested number of bits 1834 * in the IPA address space 1835 */ 1836 static int virt_kvm_type(MachineState *ms, const char *type_str) 1837 { 1838 VirtMachineState *vms = VIRT_MACHINE(ms); 1839 int max_vm_pa_size = kvm_arm_get_max_vm_ipa_size(ms); 1840 int requested_pa_size; 1841 1842 /* we freeze the memory map to compute the highest gpa */ 1843 virt_set_memmap(vms); 1844 1845 requested_pa_size = 64 - clz64(vms->highest_gpa); 1846 1847 if (requested_pa_size > max_vm_pa_size) { 1848 error_report("-m and ,maxmem option values " 1849 "require an IPA range (%d bits) larger than " 1850 "the one supported by the host (%d bits)", 1851 requested_pa_size, max_vm_pa_size); 1852 exit(1); 1853 } 1854 /* 1855 * By default we return 0 which corresponds to an implicit legacy 1856 * 40b IPA setting. Otherwise we return the actual requested PA 1857 * logsize 1858 */ 1859 return requested_pa_size > 40 ? requested_pa_size : 0; 1860 } 1861 1862 static void virt_machine_class_init(ObjectClass *oc, void *data) 1863 { 1864 MachineClass *mc = MACHINE_CLASS(oc); 1865 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc); 1866 1867 mc->init = machvirt_init; 1868 /* Start with max_cpus set to 512, which is the maximum supported by KVM. 1869 * The value may be reduced later when we have more information about the 1870 * configuration of the particular instance. 1871 */ 1872 mc->max_cpus = 512; 1873 machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_CALXEDA_XGMAC); 1874 machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_AMD_XGBE); 1875 machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE); 1876 machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_PLATFORM); 1877 mc->block_default_type = IF_VIRTIO; 1878 mc->no_cdrom = 1; 1879 mc->pci_allow_0_address = true; 1880 /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */ 1881 mc->minimum_page_bits = 12; 1882 mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids; 1883 mc->cpu_index_to_instance_props = virt_cpu_index_to_props; 1884 mc->default_cpu_type = ARM_CPU_TYPE_NAME("cortex-a15"); 1885 mc->get_default_cpu_node_id = virt_get_default_cpu_node_id; 1886 mc->kvm_type = virt_kvm_type; 1887 assert(!mc->get_hotplug_handler); 1888 mc->get_hotplug_handler = virt_machine_get_hotplug_handler; 1889 hc->plug = virt_machine_device_plug_cb; 1890 } 1891 1892 static void virt_instance_init(Object *obj) 1893 { 1894 VirtMachineState *vms = VIRT_MACHINE(obj); 1895 VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms); 1896 1897 /* EL3 is disabled by default on virt: this makes us consistent 1898 * between KVM and TCG for this board, and it also allows us to 1899 * boot UEFI blobs which assume no TrustZone support. 1900 */ 1901 vms->secure = false; 1902 object_property_add_bool(obj, "secure", virt_get_secure, 1903 virt_set_secure, NULL); 1904 object_property_set_description(obj, "secure", 1905 "Set on/off to enable/disable the ARM " 1906 "Security Extensions (TrustZone)", 1907 NULL); 1908 1909 /* EL2 is also disabled by default, for similar reasons */ 1910 vms->virt = false; 1911 object_property_add_bool(obj, "virtualization", virt_get_virt, 1912 virt_set_virt, NULL); 1913 object_property_set_description(obj, "virtualization", 1914 "Set on/off to enable/disable emulating a " 1915 "guest CPU which implements the ARM " 1916 "Virtualization Extensions", 1917 NULL); 1918 1919 /* High memory is enabled by default */ 1920 vms->highmem = true; 1921 object_property_add_bool(obj, "highmem", virt_get_highmem, 1922 virt_set_highmem, NULL); 1923 object_property_set_description(obj, "highmem", 1924 "Set on/off to enable/disable using " 1925 "physical address space above 32 bits", 1926 NULL); 1927 /* Default GIC type is v2 */ 1928 vms->gic_version = 2; 1929 object_property_add_str(obj, "gic-version", virt_get_gic_version, 1930 virt_set_gic_version, NULL); 1931 object_property_set_description(obj, "gic-version", 1932 "Set GIC version. " 1933 "Valid values are 2, 3 and host", NULL); 1934 1935 vms->highmem_ecam = !vmc->no_highmem_ecam; 1936 1937 if (vmc->no_its) { 1938 vms->its = false; 1939 } else { 1940 /* Default allows ITS instantiation */ 1941 vms->its = true; 1942 object_property_add_bool(obj, "its", virt_get_its, 1943 virt_set_its, NULL); 1944 object_property_set_description(obj, "its", 1945 "Set on/off to enable/disable " 1946 "ITS instantiation", 1947 NULL); 1948 } 1949 1950 /* Default disallows iommu instantiation */ 1951 vms->iommu = VIRT_IOMMU_NONE; 1952 object_property_add_str(obj, "iommu", virt_get_iommu, virt_set_iommu, NULL); 1953 object_property_set_description(obj, "iommu", 1954 "Set the IOMMU type. " 1955 "Valid values are none and smmuv3", 1956 NULL); 1957 1958 vms->irqmap = a15irqmap; 1959 } 1960 1961 static const TypeInfo virt_machine_info = { 1962 .name = TYPE_VIRT_MACHINE, 1963 .parent = TYPE_MACHINE, 1964 .abstract = true, 1965 .instance_size = sizeof(VirtMachineState), 1966 .class_size = sizeof(VirtMachineClass), 1967 .class_init = virt_machine_class_init, 1968 .instance_init = virt_instance_init, 1969 .interfaces = (InterfaceInfo[]) { 1970 { TYPE_HOTPLUG_HANDLER }, 1971 { } 1972 }, 1973 }; 1974 1975 static void machvirt_machine_init(void) 1976 { 1977 type_register_static(&virt_machine_info); 1978 } 1979 type_init(machvirt_machine_init); 1980 1981 static void virt_machine_4_0_options(MachineClass *mc) 1982 { 1983 } 1984 DEFINE_VIRT_MACHINE_AS_LATEST(4, 0) 1985 1986 static void virt_machine_3_1_options(MachineClass *mc) 1987 { 1988 virt_machine_4_0_options(mc); 1989 compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len); 1990 } 1991 DEFINE_VIRT_MACHINE(3, 1) 1992 1993 static void virt_machine_3_0_options(MachineClass *mc) 1994 { 1995 virt_machine_3_1_options(mc); 1996 compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len); 1997 } 1998 DEFINE_VIRT_MACHINE(3, 0) 1999 2000 static void virt_machine_2_12_options(MachineClass *mc) 2001 { 2002 VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc)); 2003 2004 virt_machine_3_0_options(mc); 2005 compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len); 2006 vmc->no_highmem_ecam = true; 2007 mc->max_cpus = 255; 2008 } 2009 DEFINE_VIRT_MACHINE(2, 12) 2010 2011 static void virt_machine_2_11_options(MachineClass *mc) 2012 { 2013 VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc)); 2014 2015 virt_machine_2_12_options(mc); 2016 compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len); 2017 vmc->smbios_old_sys_ver = true; 2018 } 2019 DEFINE_VIRT_MACHINE(2, 11) 2020 2021 static void virt_machine_2_10_options(MachineClass *mc) 2022 { 2023 virt_machine_2_11_options(mc); 2024 compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len); 2025 /* before 2.11 we never faulted accesses to bad addresses */ 2026 mc->ignore_memory_transaction_failures = true; 2027 } 2028 DEFINE_VIRT_MACHINE(2, 10) 2029 2030 static void virt_machine_2_9_options(MachineClass *mc) 2031 { 2032 virt_machine_2_10_options(mc); 2033 compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len); 2034 } 2035 DEFINE_VIRT_MACHINE(2, 9) 2036 2037 static void virt_machine_2_8_options(MachineClass *mc) 2038 { 2039 VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc)); 2040 2041 virt_machine_2_9_options(mc); 2042 compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len); 2043 /* For 2.8 and earlier we falsely claimed in the DT that 2044 * our timers were edge-triggered, not level-triggered. 2045 */ 2046 vmc->claim_edge_triggered_timers = true; 2047 } 2048 DEFINE_VIRT_MACHINE(2, 8) 2049 2050 static void virt_machine_2_7_options(MachineClass *mc) 2051 { 2052 VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc)); 2053 2054 virt_machine_2_8_options(mc); 2055 compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len); 2056 /* ITS was introduced with 2.8 */ 2057 vmc->no_its = true; 2058 /* Stick with 1K pages for migration compatibility */ 2059 mc->minimum_page_bits = 0; 2060 } 2061 DEFINE_VIRT_MACHINE(2, 7) 2062 2063 static void virt_machine_2_6_options(MachineClass *mc) 2064 { 2065 VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc)); 2066 2067 virt_machine_2_7_options(mc); 2068 compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len); 2069 vmc->disallow_affinity_adjustment = true; 2070 /* Disable PMU for 2.6 as PMU support was first introduced in 2.7 */ 2071 vmc->no_pmu = true; 2072 } 2073 DEFINE_VIRT_MACHINE(2, 6) 2074