xref: /qemu/hw/riscv/virt.c (revision 8fb0bb5e8a601c66af2f1b5261256451d3d23587)
1 /*
2  * QEMU RISC-V VirtIO Board
3  *
4  * Copyright (c) 2017 SiFive, Inc.
5  *
6  * RISC-V machine with 16550a UART and VirtIO MMIO
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms and conditions of the GNU General Public License,
10  * version 2 or later, as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * You should have received a copy of the GNU General Public License along with
18  * this program.  If not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #include "qemu/osdep.h"
22 #include "qemu/units.h"
23 #include "qemu/error-report.h"
24 #include "qemu/guest-random.h"
25 #include "qapi/error.h"
26 #include "hw/boards.h"
27 #include "hw/loader.h"
28 #include "hw/sysbus.h"
29 #include "hw/qdev-properties.h"
30 #include "hw/char/serial.h"
31 #include "target/riscv/cpu.h"
32 #include "hw/core/sysbus-fdt.h"
33 #include "target/riscv/pmu.h"
34 #include "hw/riscv/riscv_hart.h"
35 #include "hw/riscv/virt.h"
36 #include "hw/riscv/boot.h"
37 #include "hw/riscv/numa.h"
38 #include "kvm/kvm_riscv.h"
39 #include "hw/firmware/smbios.h"
40 #include "hw/intc/riscv_aclint.h"
41 #include "hw/intc/riscv_aplic.h"
42 #include "hw/intc/sifive_plic.h"
43 #include "hw/misc/sifive_test.h"
44 #include "hw/platform-bus.h"
45 #include "chardev/char.h"
46 #include "sysemu/device_tree.h"
47 #include "sysemu/sysemu.h"
48 #include "sysemu/tcg.h"
49 #include "sysemu/kvm.h"
50 #include "sysemu/tpm.h"
51 #include "sysemu/qtest.h"
52 #include "hw/pci/pci.h"
53 #include "hw/pci-host/gpex.h"
54 #include "hw/display/ramfb.h"
55 #include "hw/acpi/aml-build.h"
56 #include "qapi/qapi-visit-common.h"
57 #include "hw/virtio/virtio-iommu.h"
58 
59 /* KVM AIA only supports APLIC MSI. APLIC Wired is always emulated by QEMU. */
60 static bool virt_use_kvm_aia(RISCVVirtState *s)
61 {
62     return kvm_irqchip_in_kernel() && s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC;
63 }
64 
65 static bool virt_aclint_allowed(void)
66 {
67     return tcg_enabled() || qtest_enabled();
68 }
69 
70 static const MemMapEntry virt_memmap[] = {
71     [VIRT_DEBUG] =        {        0x0,         0x100 },
72     [VIRT_MROM] =         {     0x1000,        0xf000 },
73     [VIRT_TEST] =         {   0x100000,        0x1000 },
74     [VIRT_RTC] =          {   0x101000,        0x1000 },
75     [VIRT_CLINT] =        {  0x2000000,       0x10000 },
76     [VIRT_ACLINT_SSWI] =  {  0x2F00000,        0x4000 },
77     [VIRT_PCIE_PIO] =     {  0x3000000,       0x10000 },
78     [VIRT_PLATFORM_BUS] = {  0x4000000,     0x2000000 },
79     [VIRT_PLIC] =         {  0xc000000, VIRT_PLIC_SIZE(VIRT_CPUS_MAX * 2) },
80     [VIRT_APLIC_M] =      {  0xc000000, APLIC_SIZE(VIRT_CPUS_MAX) },
81     [VIRT_APLIC_S] =      {  0xd000000, APLIC_SIZE(VIRT_CPUS_MAX) },
82     [VIRT_UART0] =        { 0x10000000,         0x100 },
83     [VIRT_VIRTIO] =       { 0x10001000,        0x1000 },
84     [VIRT_FW_CFG] =       { 0x10100000,          0x18 },
85     [VIRT_FLASH] =        { 0x20000000,     0x4000000 },
86     [VIRT_IMSIC_M] =      { 0x24000000, VIRT_IMSIC_MAX_SIZE },
87     [VIRT_IMSIC_S] =      { 0x28000000, VIRT_IMSIC_MAX_SIZE },
88     [VIRT_PCIE_ECAM] =    { 0x30000000,    0x10000000 },
89     [VIRT_PCIE_MMIO] =    { 0x40000000,    0x40000000 },
90     [VIRT_DRAM] =         { 0x80000000,           0x0 },
91 };
92 
93 /* PCIe high mmio is fixed for RV32 */
94 #define VIRT32_HIGH_PCIE_MMIO_BASE  0x300000000ULL
95 #define VIRT32_HIGH_PCIE_MMIO_SIZE  (4 * GiB)
96 
97 /* PCIe high mmio for RV64, size is fixed but base depends on top of RAM */
98 #define VIRT64_HIGH_PCIE_MMIO_SIZE  (16 * GiB)
99 
100 static MemMapEntry virt_high_pcie_memmap;
101 
102 #define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
103 
104 static PFlashCFI01 *virt_flash_create1(RISCVVirtState *s,
105                                        const char *name,
106                                        const char *alias_prop_name)
107 {
108     /*
109      * Create a single flash device.  We use the same parameters as
110      * the flash devices on the ARM virt board.
111      */
112     DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
113 
114     qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
115     qdev_prop_set_uint8(dev, "width", 4);
116     qdev_prop_set_uint8(dev, "device-width", 2);
117     qdev_prop_set_bit(dev, "big-endian", false);
118     qdev_prop_set_uint16(dev, "id0", 0x89);
119     qdev_prop_set_uint16(dev, "id1", 0x18);
120     qdev_prop_set_uint16(dev, "id2", 0x00);
121     qdev_prop_set_uint16(dev, "id3", 0x00);
122     qdev_prop_set_string(dev, "name", name);
123 
124     object_property_add_child(OBJECT(s), name, OBJECT(dev));
125     object_property_add_alias(OBJECT(s), alias_prop_name,
126                               OBJECT(dev), "drive");
127 
128     return PFLASH_CFI01(dev);
129 }
130 
131 static void virt_flash_create(RISCVVirtState *s)
132 {
133     s->flash[0] = virt_flash_create1(s, "virt.flash0", "pflash0");
134     s->flash[1] = virt_flash_create1(s, "virt.flash1", "pflash1");
135 }
136 
137 static void virt_flash_map1(PFlashCFI01 *flash,
138                             hwaddr base, hwaddr size,
139                             MemoryRegion *sysmem)
140 {
141     DeviceState *dev = DEVICE(flash);
142 
143     assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
144     assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
145     qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
146     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
147 
148     memory_region_add_subregion(sysmem, base,
149                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
150                                                        0));
151 }
152 
153 static void virt_flash_map(RISCVVirtState *s,
154                            MemoryRegion *sysmem)
155 {
156     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
157     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
158 
159     virt_flash_map1(s->flash[0], flashbase, flashsize,
160                     sysmem);
161     virt_flash_map1(s->flash[1], flashbase + flashsize, flashsize,
162                     sysmem);
163 }
164 
165 static void create_pcie_irq_map(RISCVVirtState *s, void *fdt, char *nodename,
166                                 uint32_t irqchip_phandle)
167 {
168     int pin, dev;
169     uint32_t irq_map_stride = 0;
170     uint32_t full_irq_map[GPEX_NUM_IRQS * GPEX_NUM_IRQS *
171                           FDT_MAX_INT_MAP_WIDTH] = {};
172     uint32_t *irq_map = full_irq_map;
173 
174     /* This code creates a standard swizzle of interrupts such that
175      * each device's first interrupt is based on it's PCI_SLOT number.
176      * (See pci_swizzle_map_irq_fn())
177      *
178      * We only need one entry per interrupt in the table (not one per
179      * possible slot) seeing the interrupt-map-mask will allow the table
180      * to wrap to any number of devices.
181      */
182     for (dev = 0; dev < GPEX_NUM_IRQS; dev++) {
183         int devfn = dev * 0x8;
184 
185         for (pin = 0; pin < GPEX_NUM_IRQS; pin++) {
186             int irq_nr = PCIE_IRQ + ((pin + PCI_SLOT(devfn)) % GPEX_NUM_IRQS);
187             int i = 0;
188 
189             /* Fill PCI address cells */
190             irq_map[i] = cpu_to_be32(devfn << 8);
191             i += FDT_PCI_ADDR_CELLS;
192 
193             /* Fill PCI Interrupt cells */
194             irq_map[i] = cpu_to_be32(pin + 1);
195             i += FDT_PCI_INT_CELLS;
196 
197             /* Fill interrupt controller phandle and cells */
198             irq_map[i++] = cpu_to_be32(irqchip_phandle);
199             irq_map[i++] = cpu_to_be32(irq_nr);
200             if (s->aia_type != VIRT_AIA_TYPE_NONE) {
201                 irq_map[i++] = cpu_to_be32(0x4);
202             }
203 
204             if (!irq_map_stride) {
205                 irq_map_stride = i;
206             }
207             irq_map += irq_map_stride;
208         }
209     }
210 
211     qemu_fdt_setprop(fdt, nodename, "interrupt-map", full_irq_map,
212                      GPEX_NUM_IRQS * GPEX_NUM_IRQS *
213                      irq_map_stride * sizeof(uint32_t));
214 
215     qemu_fdt_setprop_cells(fdt, nodename, "interrupt-map-mask",
216                            0x1800, 0, 0, 0x7);
217 }
218 
219 static void create_fdt_socket_cpus(RISCVVirtState *s, int socket,
220                                    char *clust_name, uint32_t *phandle,
221                                    uint32_t *intc_phandles)
222 {
223     int cpu;
224     uint32_t cpu_phandle;
225     MachineState *ms = MACHINE(s);
226     bool is_32_bit = riscv_is_32bit(&s->soc[0]);
227     uint8_t satp_mode_max;
228 
229     for (cpu = s->soc[socket].num_harts - 1; cpu >= 0; cpu--) {
230         RISCVCPU *cpu_ptr = &s->soc[socket].harts[cpu];
231         g_autofree char *cpu_name = NULL;
232         g_autofree char *core_name = NULL;
233         g_autofree char *intc_name = NULL;
234         g_autofree char *sv_name = NULL;
235 
236         cpu_phandle = (*phandle)++;
237 
238         cpu_name = g_strdup_printf("/cpus/cpu@%d",
239             s->soc[socket].hartid_base + cpu);
240         qemu_fdt_add_subnode(ms->fdt, cpu_name);
241 
242         if (cpu_ptr->cfg.satp_mode.supported != 0) {
243             satp_mode_max = satp_mode_max_from_map(cpu_ptr->cfg.satp_mode.map);
244             sv_name = g_strdup_printf("riscv,%s",
245                                       satp_mode_str(satp_mode_max, is_32_bit));
246             qemu_fdt_setprop_string(ms->fdt, cpu_name, "mmu-type", sv_name);
247         }
248 
249         riscv_isa_write_fdt(cpu_ptr, ms->fdt, cpu_name);
250 
251         if (cpu_ptr->cfg.ext_zicbom) {
252             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cbom-block-size",
253                                   cpu_ptr->cfg.cbom_blocksize);
254         }
255 
256         if (cpu_ptr->cfg.ext_zicboz) {
257             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cboz-block-size",
258                                   cpu_ptr->cfg.cboz_blocksize);
259         }
260 
261         if (cpu_ptr->cfg.ext_zicbop) {
262             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cbop-block-size",
263                                   cpu_ptr->cfg.cbop_blocksize);
264         }
265 
266         qemu_fdt_setprop_string(ms->fdt, cpu_name, "compatible", "riscv");
267         qemu_fdt_setprop_string(ms->fdt, cpu_name, "status", "okay");
268         qemu_fdt_setprop_cell(ms->fdt, cpu_name, "reg",
269             s->soc[socket].hartid_base + cpu);
270         qemu_fdt_setprop_string(ms->fdt, cpu_name, "device_type", "cpu");
271         riscv_socket_fdt_write_id(ms, cpu_name, socket);
272         qemu_fdt_setprop_cell(ms->fdt, cpu_name, "phandle", cpu_phandle);
273 
274         intc_phandles[cpu] = (*phandle)++;
275 
276         intc_name = g_strdup_printf("%s/interrupt-controller", cpu_name);
277         qemu_fdt_add_subnode(ms->fdt, intc_name);
278         qemu_fdt_setprop_cell(ms->fdt, intc_name, "phandle",
279             intc_phandles[cpu]);
280         qemu_fdt_setprop_string(ms->fdt, intc_name, "compatible",
281             "riscv,cpu-intc");
282         qemu_fdt_setprop(ms->fdt, intc_name, "interrupt-controller", NULL, 0);
283         qemu_fdt_setprop_cell(ms->fdt, intc_name, "#interrupt-cells", 1);
284 
285         core_name = g_strdup_printf("%s/core%d", clust_name, cpu);
286         qemu_fdt_add_subnode(ms->fdt, core_name);
287         qemu_fdt_setprop_cell(ms->fdt, core_name, "cpu", cpu_phandle);
288     }
289 }
290 
291 static void create_fdt_socket_memory(RISCVVirtState *s,
292                                      const MemMapEntry *memmap, int socket)
293 {
294     g_autofree char *mem_name = NULL;
295     uint64_t addr, size;
296     MachineState *ms = MACHINE(s);
297 
298     addr = memmap[VIRT_DRAM].base + riscv_socket_mem_offset(ms, socket);
299     size = riscv_socket_mem_size(ms, socket);
300     mem_name = g_strdup_printf("/memory@%lx", (long)addr);
301     qemu_fdt_add_subnode(ms->fdt, mem_name);
302     qemu_fdt_setprop_cells(ms->fdt, mem_name, "reg",
303         addr >> 32, addr, size >> 32, size);
304     qemu_fdt_setprop_string(ms->fdt, mem_name, "device_type", "memory");
305     riscv_socket_fdt_write_id(ms, mem_name, socket);
306 }
307 
308 static void create_fdt_socket_clint(RISCVVirtState *s,
309                                     const MemMapEntry *memmap, int socket,
310                                     uint32_t *intc_phandles)
311 {
312     int cpu;
313     g_autofree char *clint_name = NULL;
314     g_autofree uint32_t *clint_cells = NULL;
315     unsigned long clint_addr;
316     MachineState *ms = MACHINE(s);
317     static const char * const clint_compat[2] = {
318         "sifive,clint0", "riscv,clint0"
319     };
320 
321     clint_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
322 
323     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
324         clint_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
325         clint_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_SOFT);
326         clint_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
327         clint_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_M_TIMER);
328     }
329 
330     clint_addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
331     clint_name = g_strdup_printf("/soc/clint@%lx", clint_addr);
332     qemu_fdt_add_subnode(ms->fdt, clint_name);
333     qemu_fdt_setprop_string_array(ms->fdt, clint_name, "compatible",
334                                   (char **)&clint_compat,
335                                   ARRAY_SIZE(clint_compat));
336     qemu_fdt_setprop_cells(ms->fdt, clint_name, "reg",
337         0x0, clint_addr, 0x0, memmap[VIRT_CLINT].size);
338     qemu_fdt_setprop(ms->fdt, clint_name, "interrupts-extended",
339         clint_cells, s->soc[socket].num_harts * sizeof(uint32_t) * 4);
340     riscv_socket_fdt_write_id(ms, clint_name, socket);
341 }
342 
343 static void create_fdt_socket_aclint(RISCVVirtState *s,
344                                      const MemMapEntry *memmap, int socket,
345                                      uint32_t *intc_phandles)
346 {
347     int cpu;
348     char *name;
349     unsigned long addr, size;
350     uint32_t aclint_cells_size;
351     g_autofree uint32_t *aclint_mswi_cells = NULL;
352     g_autofree uint32_t *aclint_sswi_cells = NULL;
353     g_autofree uint32_t *aclint_mtimer_cells = NULL;
354     MachineState *ms = MACHINE(s);
355 
356     aclint_mswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
357     aclint_mtimer_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
358     aclint_sswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
359 
360     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
361         aclint_mswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
362         aclint_mswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_SOFT);
363         aclint_mtimer_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
364         aclint_mtimer_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_TIMER);
365         aclint_sswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
366         aclint_sswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_SOFT);
367     }
368     aclint_cells_size = s->soc[socket].num_harts * sizeof(uint32_t) * 2;
369 
370     if (s->aia_type != VIRT_AIA_TYPE_APLIC_IMSIC) {
371         addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
372         name = g_strdup_printf("/soc/mswi@%lx", addr);
373         qemu_fdt_add_subnode(ms->fdt, name);
374         qemu_fdt_setprop_string(ms->fdt, name, "compatible",
375             "riscv,aclint-mswi");
376         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
377             0x0, addr, 0x0, RISCV_ACLINT_SWI_SIZE);
378         qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
379             aclint_mswi_cells, aclint_cells_size);
380         qemu_fdt_setprop(ms->fdt, name, "interrupt-controller", NULL, 0);
381         qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells", 0);
382         riscv_socket_fdt_write_id(ms, name, socket);
383         g_free(name);
384     }
385 
386     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
387         addr = memmap[VIRT_CLINT].base +
388                (RISCV_ACLINT_DEFAULT_MTIMER_SIZE * socket);
389         size = RISCV_ACLINT_DEFAULT_MTIMER_SIZE;
390     } else {
391         addr = memmap[VIRT_CLINT].base + RISCV_ACLINT_SWI_SIZE +
392             (memmap[VIRT_CLINT].size * socket);
393         size = memmap[VIRT_CLINT].size - RISCV_ACLINT_SWI_SIZE;
394     }
395     name = g_strdup_printf("/soc/mtimer@%lx", addr);
396     qemu_fdt_add_subnode(ms->fdt, name);
397     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
398         "riscv,aclint-mtimer");
399     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
400         0x0, addr + RISCV_ACLINT_DEFAULT_MTIME,
401         0x0, size - RISCV_ACLINT_DEFAULT_MTIME,
402         0x0, addr + RISCV_ACLINT_DEFAULT_MTIMECMP,
403         0x0, RISCV_ACLINT_DEFAULT_MTIME);
404     qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
405         aclint_mtimer_cells, aclint_cells_size);
406     riscv_socket_fdt_write_id(ms, name, socket);
407     g_free(name);
408 
409     if (s->aia_type != VIRT_AIA_TYPE_APLIC_IMSIC) {
410         addr = memmap[VIRT_ACLINT_SSWI].base +
411             (memmap[VIRT_ACLINT_SSWI].size * socket);
412         name = g_strdup_printf("/soc/sswi@%lx", addr);
413         qemu_fdt_add_subnode(ms->fdt, name);
414         qemu_fdt_setprop_string(ms->fdt, name, "compatible",
415             "riscv,aclint-sswi");
416         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
417             0x0, addr, 0x0, memmap[VIRT_ACLINT_SSWI].size);
418         qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
419             aclint_sswi_cells, aclint_cells_size);
420         qemu_fdt_setprop(ms->fdt, name, "interrupt-controller", NULL, 0);
421         qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells", 0);
422         riscv_socket_fdt_write_id(ms, name, socket);
423         g_free(name);
424     }
425 }
426 
427 static void create_fdt_socket_plic(RISCVVirtState *s,
428                                    const MemMapEntry *memmap, int socket,
429                                    uint32_t *phandle, uint32_t *intc_phandles,
430                                    uint32_t *plic_phandles)
431 {
432     int cpu;
433     g_autofree char *plic_name = NULL;
434     g_autofree uint32_t *plic_cells;
435     unsigned long plic_addr;
436     MachineState *ms = MACHINE(s);
437     static const char * const plic_compat[2] = {
438         "sifive,plic-1.0.0", "riscv,plic0"
439     };
440 
441     plic_phandles[socket] = (*phandle)++;
442     plic_addr = memmap[VIRT_PLIC].base + (memmap[VIRT_PLIC].size * socket);
443     plic_name = g_strdup_printf("/soc/plic@%lx", plic_addr);
444     qemu_fdt_add_subnode(ms->fdt, plic_name);
445     qemu_fdt_setprop_cell(ms->fdt, plic_name,
446         "#interrupt-cells", FDT_PLIC_INT_CELLS);
447     qemu_fdt_setprop_cell(ms->fdt, plic_name,
448         "#address-cells", FDT_PLIC_ADDR_CELLS);
449     qemu_fdt_setprop_string_array(ms->fdt, plic_name, "compatible",
450                                   (char **)&plic_compat,
451                                   ARRAY_SIZE(plic_compat));
452     qemu_fdt_setprop(ms->fdt, plic_name, "interrupt-controller", NULL, 0);
453 
454     if (kvm_enabled()) {
455         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
456 
457         for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
458             plic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
459             plic_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_EXT);
460         }
461 
462         qemu_fdt_setprop(ms->fdt, plic_name, "interrupts-extended",
463                          plic_cells,
464                          s->soc[socket].num_harts * sizeof(uint32_t) * 2);
465    } else {
466         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
467 
468         for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
469             plic_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
470             plic_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_EXT);
471             plic_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
472             plic_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_S_EXT);
473         }
474 
475         qemu_fdt_setprop(ms->fdt, plic_name, "interrupts-extended",
476                          plic_cells,
477                          s->soc[socket].num_harts * sizeof(uint32_t) * 4);
478     }
479 
480     qemu_fdt_setprop_cells(ms->fdt, plic_name, "reg",
481         0x0, plic_addr, 0x0, memmap[VIRT_PLIC].size);
482     qemu_fdt_setprop_cell(ms->fdt, plic_name, "riscv,ndev",
483                           VIRT_IRQCHIP_NUM_SOURCES - 1);
484     riscv_socket_fdt_write_id(ms, plic_name, socket);
485     qemu_fdt_setprop_cell(ms->fdt, plic_name, "phandle",
486         plic_phandles[socket]);
487 
488     if (!socket) {
489         platform_bus_add_all_fdt_nodes(ms->fdt, plic_name,
490                                        memmap[VIRT_PLATFORM_BUS].base,
491                                        memmap[VIRT_PLATFORM_BUS].size,
492                                        VIRT_PLATFORM_BUS_IRQ);
493     }
494 }
495 
496 uint32_t imsic_num_bits(uint32_t count)
497 {
498     uint32_t ret = 0;
499 
500     while (BIT(ret) < count) {
501         ret++;
502     }
503 
504     return ret;
505 }
506 
507 static void create_fdt_one_imsic(RISCVVirtState *s, hwaddr base_addr,
508                                  uint32_t *intc_phandles, uint32_t msi_phandle,
509                                  bool m_mode, uint32_t imsic_guest_bits)
510 {
511     int cpu, socket;
512     g_autofree char *imsic_name = NULL;
513     MachineState *ms = MACHINE(s);
514     int socket_count = riscv_socket_count(ms);
515     uint32_t imsic_max_hart_per_socket, imsic_addr, imsic_size;
516     g_autofree uint32_t *imsic_cells = NULL;
517     g_autofree uint32_t *imsic_regs = NULL;
518     static const char * const imsic_compat[2] = {
519         "qemu,imsics", "riscv,imsics"
520     };
521 
522     imsic_cells = g_new0(uint32_t, ms->smp.cpus * 2);
523     imsic_regs = g_new0(uint32_t, socket_count * 4);
524 
525     for (cpu = 0; cpu < ms->smp.cpus; cpu++) {
526         imsic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
527         imsic_cells[cpu * 2 + 1] = cpu_to_be32(m_mode ? IRQ_M_EXT : IRQ_S_EXT);
528     }
529 
530     imsic_max_hart_per_socket = 0;
531     for (socket = 0; socket < socket_count; socket++) {
532         imsic_addr = base_addr + socket * VIRT_IMSIC_GROUP_MAX_SIZE;
533         imsic_size = IMSIC_HART_SIZE(imsic_guest_bits) *
534                      s->soc[socket].num_harts;
535         imsic_regs[socket * 4 + 0] = 0;
536         imsic_regs[socket * 4 + 1] = cpu_to_be32(imsic_addr);
537         imsic_regs[socket * 4 + 2] = 0;
538         imsic_regs[socket * 4 + 3] = cpu_to_be32(imsic_size);
539         if (imsic_max_hart_per_socket < s->soc[socket].num_harts) {
540             imsic_max_hart_per_socket = s->soc[socket].num_harts;
541         }
542     }
543 
544     imsic_name = g_strdup_printf("/soc/interrupt-controller@%lx",
545                                  (unsigned long)base_addr);
546     qemu_fdt_add_subnode(ms->fdt, imsic_name);
547     qemu_fdt_setprop_string_array(ms->fdt, imsic_name, "compatible",
548                                   (char **)&imsic_compat,
549                                   ARRAY_SIZE(imsic_compat));
550 
551     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "#interrupt-cells",
552                           FDT_IMSIC_INT_CELLS);
553     qemu_fdt_setprop(ms->fdt, imsic_name, "interrupt-controller", NULL, 0);
554     qemu_fdt_setprop(ms->fdt, imsic_name, "msi-controller", NULL, 0);
555     qemu_fdt_setprop(ms->fdt, imsic_name, "interrupts-extended",
556                      imsic_cells, ms->smp.cpus * sizeof(uint32_t) * 2);
557     qemu_fdt_setprop(ms->fdt, imsic_name, "reg", imsic_regs,
558                      socket_count * sizeof(uint32_t) * 4);
559     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,num-ids",
560                      VIRT_IRQCHIP_NUM_MSIS);
561 
562     if (imsic_guest_bits) {
563         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,guest-index-bits",
564                               imsic_guest_bits);
565     }
566 
567     if (socket_count > 1) {
568         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,hart-index-bits",
569                               imsic_num_bits(imsic_max_hart_per_socket));
570         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,group-index-bits",
571                               imsic_num_bits(socket_count));
572         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,group-index-shift",
573                               IMSIC_MMIO_GROUP_MIN_SHIFT);
574     }
575     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "phandle", msi_phandle);
576 }
577 
578 static void create_fdt_imsic(RISCVVirtState *s, const MemMapEntry *memmap,
579                              uint32_t *phandle, uint32_t *intc_phandles,
580                              uint32_t *msi_m_phandle, uint32_t *msi_s_phandle)
581 {
582     *msi_m_phandle = (*phandle)++;
583     *msi_s_phandle = (*phandle)++;
584 
585     if (!kvm_enabled()) {
586         /* M-level IMSIC node */
587         create_fdt_one_imsic(s, memmap[VIRT_IMSIC_M].base, intc_phandles,
588                              *msi_m_phandle, true, 0);
589     }
590 
591     /* S-level IMSIC node */
592     create_fdt_one_imsic(s, memmap[VIRT_IMSIC_S].base, intc_phandles,
593                          *msi_s_phandle, false,
594                          imsic_num_bits(s->aia_guests + 1));
595 
596 }
597 
598 /* Caller must free string after use */
599 static char *fdt_get_aplic_nodename(unsigned long aplic_addr)
600 {
601     return g_strdup_printf("/soc/interrupt-controller@%lx", aplic_addr);
602 }
603 
604 static void create_fdt_one_aplic(RISCVVirtState *s, int socket,
605                                  unsigned long aplic_addr, uint32_t aplic_size,
606                                  uint32_t msi_phandle,
607                                  uint32_t *intc_phandles,
608                                  uint32_t aplic_phandle,
609                                  uint32_t aplic_child_phandle,
610                                  bool m_mode, int num_harts)
611 {
612     int cpu;
613     g_autofree char *aplic_name = fdt_get_aplic_nodename(aplic_addr);
614     g_autofree uint32_t *aplic_cells = g_new0(uint32_t, num_harts * 2);
615     MachineState *ms = MACHINE(s);
616     static const char * const aplic_compat[2] = {
617         "qemu,aplic", "riscv,aplic"
618     };
619 
620     for (cpu = 0; cpu < num_harts; cpu++) {
621         aplic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
622         aplic_cells[cpu * 2 + 1] = cpu_to_be32(m_mode ? IRQ_M_EXT : IRQ_S_EXT);
623     }
624 
625     qemu_fdt_add_subnode(ms->fdt, aplic_name);
626     qemu_fdt_setprop_string_array(ms->fdt, aplic_name, "compatible",
627                                   (char **)&aplic_compat,
628                                   ARRAY_SIZE(aplic_compat));
629     qemu_fdt_setprop_cell(ms->fdt, aplic_name, "#address-cells",
630                           FDT_APLIC_ADDR_CELLS);
631     qemu_fdt_setprop_cell(ms->fdt, aplic_name,
632                           "#interrupt-cells", FDT_APLIC_INT_CELLS);
633     qemu_fdt_setprop(ms->fdt, aplic_name, "interrupt-controller", NULL, 0);
634 
635     if (s->aia_type == VIRT_AIA_TYPE_APLIC) {
636         qemu_fdt_setprop(ms->fdt, aplic_name, "interrupts-extended",
637                          aplic_cells, num_harts * sizeof(uint32_t) * 2);
638     } else {
639         qemu_fdt_setprop_cell(ms->fdt, aplic_name, "msi-parent", msi_phandle);
640     }
641 
642     qemu_fdt_setprop_cells(ms->fdt, aplic_name, "reg",
643                            0x0, aplic_addr, 0x0, aplic_size);
644     qemu_fdt_setprop_cell(ms->fdt, aplic_name, "riscv,num-sources",
645                           VIRT_IRQCHIP_NUM_SOURCES);
646 
647     if (aplic_child_phandle) {
648         qemu_fdt_setprop_cell(ms->fdt, aplic_name, "riscv,children",
649                               aplic_child_phandle);
650         qemu_fdt_setprop_cells(ms->fdt, aplic_name, "riscv,delegation",
651                                aplic_child_phandle, 0x1,
652                                VIRT_IRQCHIP_NUM_SOURCES);
653     }
654 
655     riscv_socket_fdt_write_id(ms, aplic_name, socket);
656     qemu_fdt_setprop_cell(ms->fdt, aplic_name, "phandle", aplic_phandle);
657 }
658 
659 static void create_fdt_socket_aplic(RISCVVirtState *s,
660                                     const MemMapEntry *memmap, int socket,
661                                     uint32_t msi_m_phandle,
662                                     uint32_t msi_s_phandle,
663                                     uint32_t *phandle,
664                                     uint32_t *intc_phandles,
665                                     uint32_t *aplic_phandles,
666                                     int num_harts)
667 {
668     unsigned long aplic_addr;
669     MachineState *ms = MACHINE(s);
670     uint32_t aplic_m_phandle, aplic_s_phandle;
671 
672     aplic_m_phandle = (*phandle)++;
673     aplic_s_phandle = (*phandle)++;
674 
675     if (!kvm_enabled()) {
676         /* M-level APLIC node */
677         aplic_addr = memmap[VIRT_APLIC_M].base +
678                      (memmap[VIRT_APLIC_M].size * socket);
679         create_fdt_one_aplic(s, socket, aplic_addr, memmap[VIRT_APLIC_M].size,
680                              msi_m_phandle, intc_phandles,
681                              aplic_m_phandle, aplic_s_phandle,
682                              true, num_harts);
683     }
684 
685     /* S-level APLIC node */
686     aplic_addr = memmap[VIRT_APLIC_S].base +
687                  (memmap[VIRT_APLIC_S].size * socket);
688     create_fdt_one_aplic(s, socket, aplic_addr, memmap[VIRT_APLIC_S].size,
689                          msi_s_phandle, intc_phandles,
690                          aplic_s_phandle, 0,
691                          false, num_harts);
692 
693     if (!socket) {
694         g_autofree char *aplic_name = fdt_get_aplic_nodename(aplic_addr);
695         platform_bus_add_all_fdt_nodes(ms->fdt, aplic_name,
696                                        memmap[VIRT_PLATFORM_BUS].base,
697                                        memmap[VIRT_PLATFORM_BUS].size,
698                                        VIRT_PLATFORM_BUS_IRQ);
699     }
700 
701     aplic_phandles[socket] = aplic_s_phandle;
702 }
703 
704 static void create_fdt_pmu(RISCVVirtState *s)
705 {
706     g_autofree char *pmu_name = g_strdup_printf("/pmu");
707     MachineState *ms = MACHINE(s);
708     RISCVCPU hart = s->soc[0].harts[0];
709 
710     qemu_fdt_add_subnode(ms->fdt, pmu_name);
711     qemu_fdt_setprop_string(ms->fdt, pmu_name, "compatible", "riscv,pmu");
712     riscv_pmu_generate_fdt_node(ms->fdt, hart.pmu_avail_ctrs, pmu_name);
713 }
714 
715 static void create_fdt_sockets(RISCVVirtState *s, const MemMapEntry *memmap,
716                                uint32_t *phandle,
717                                uint32_t *irq_mmio_phandle,
718                                uint32_t *irq_pcie_phandle,
719                                uint32_t *irq_virtio_phandle,
720                                uint32_t *msi_pcie_phandle)
721 {
722     int socket, phandle_pos;
723     MachineState *ms = MACHINE(s);
724     uint32_t msi_m_phandle = 0, msi_s_phandle = 0;
725     uint32_t xplic_phandles[MAX_NODES];
726     g_autofree uint32_t *intc_phandles = NULL;
727     int socket_count = riscv_socket_count(ms);
728 
729     qemu_fdt_add_subnode(ms->fdt, "/cpus");
730     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "timebase-frequency",
731                           kvm_enabled() ?
732                           kvm_riscv_get_timebase_frequency(first_cpu) :
733                           RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ);
734     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#size-cells", 0x0);
735     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#address-cells", 0x1);
736     qemu_fdt_add_subnode(ms->fdt, "/cpus/cpu-map");
737 
738     intc_phandles = g_new0(uint32_t, ms->smp.cpus);
739 
740     phandle_pos = ms->smp.cpus;
741     for (socket = (socket_count - 1); socket >= 0; socket--) {
742         g_autofree char *clust_name = NULL;
743         phandle_pos -= s->soc[socket].num_harts;
744 
745         clust_name = g_strdup_printf("/cpus/cpu-map/cluster%d", socket);
746         qemu_fdt_add_subnode(ms->fdt, clust_name);
747 
748         create_fdt_socket_cpus(s, socket, clust_name, phandle,
749                                &intc_phandles[phandle_pos]);
750 
751         create_fdt_socket_memory(s, memmap, socket);
752 
753         if (virt_aclint_allowed() && s->have_aclint) {
754             create_fdt_socket_aclint(s, memmap, socket,
755                                      &intc_phandles[phandle_pos]);
756         } else if (tcg_enabled()) {
757             create_fdt_socket_clint(s, memmap, socket,
758                                     &intc_phandles[phandle_pos]);
759         }
760     }
761 
762     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
763         create_fdt_imsic(s, memmap, phandle, intc_phandles,
764             &msi_m_phandle, &msi_s_phandle);
765         *msi_pcie_phandle = msi_s_phandle;
766     }
767 
768     /* KVM AIA only has one APLIC instance */
769     if (kvm_enabled() && virt_use_kvm_aia(s)) {
770         create_fdt_socket_aplic(s, memmap, 0,
771                                 msi_m_phandle, msi_s_phandle, phandle,
772                                 &intc_phandles[0], xplic_phandles,
773                                 ms->smp.cpus);
774     } else {
775         phandle_pos = ms->smp.cpus;
776         for (socket = (socket_count - 1); socket >= 0; socket--) {
777             phandle_pos -= s->soc[socket].num_harts;
778 
779             if (s->aia_type == VIRT_AIA_TYPE_NONE) {
780                 create_fdt_socket_plic(s, memmap, socket, phandle,
781                                        &intc_phandles[phandle_pos],
782                                        xplic_phandles);
783             } else {
784                 create_fdt_socket_aplic(s, memmap, socket,
785                                         msi_m_phandle, msi_s_phandle, phandle,
786                                         &intc_phandles[phandle_pos],
787                                         xplic_phandles,
788                                         s->soc[socket].num_harts);
789             }
790         }
791     }
792 
793     if (kvm_enabled() && virt_use_kvm_aia(s)) {
794         *irq_mmio_phandle = xplic_phandles[0];
795         *irq_virtio_phandle = xplic_phandles[0];
796         *irq_pcie_phandle = xplic_phandles[0];
797     } else {
798         for (socket = 0; socket < socket_count; socket++) {
799             if (socket == 0) {
800                 *irq_mmio_phandle = xplic_phandles[socket];
801                 *irq_virtio_phandle = xplic_phandles[socket];
802                 *irq_pcie_phandle = xplic_phandles[socket];
803             }
804             if (socket == 1) {
805                 *irq_virtio_phandle = xplic_phandles[socket];
806                 *irq_pcie_phandle = xplic_phandles[socket];
807             }
808             if (socket == 2) {
809                 *irq_pcie_phandle = xplic_phandles[socket];
810             }
811         }
812     }
813 
814     riscv_socket_fdt_write_distance_matrix(ms);
815 }
816 
817 static void create_fdt_virtio(RISCVVirtState *s, const MemMapEntry *memmap,
818                               uint32_t irq_virtio_phandle)
819 {
820     int i;
821     MachineState *ms = MACHINE(s);
822 
823     for (i = 0; i < VIRTIO_COUNT; i++) {
824         g_autofree char *name =  g_strdup_printf("/soc/virtio_mmio@%lx",
825             (long)(memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size));
826 
827         qemu_fdt_add_subnode(ms->fdt, name);
828         qemu_fdt_setprop_string(ms->fdt, name, "compatible", "virtio,mmio");
829         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
830             0x0, memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
831             0x0, memmap[VIRT_VIRTIO].size);
832         qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent",
833             irq_virtio_phandle);
834         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
835             qemu_fdt_setprop_cell(ms->fdt, name, "interrupts",
836                                   VIRTIO_IRQ + i);
837         } else {
838             qemu_fdt_setprop_cells(ms->fdt, name, "interrupts",
839                                    VIRTIO_IRQ + i, 0x4);
840         }
841     }
842 }
843 
844 static void create_fdt_pcie(RISCVVirtState *s, const MemMapEntry *memmap,
845                             uint32_t irq_pcie_phandle,
846                             uint32_t msi_pcie_phandle)
847 {
848     g_autofree char *name = NULL;
849     MachineState *ms = MACHINE(s);
850 
851     name = g_strdup_printf("/soc/pci@%lx",
852         (long) memmap[VIRT_PCIE_ECAM].base);
853     qemu_fdt_setprop_cell(ms->fdt, name, "#address-cells",
854         FDT_PCI_ADDR_CELLS);
855     qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells",
856         FDT_PCI_INT_CELLS);
857     qemu_fdt_setprop_cell(ms->fdt, name, "#size-cells", 0x2);
858     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
859         "pci-host-ecam-generic");
860     qemu_fdt_setprop_string(ms->fdt, name, "device_type", "pci");
861     qemu_fdt_setprop_cell(ms->fdt, name, "linux,pci-domain", 0);
862     qemu_fdt_setprop_cells(ms->fdt, name, "bus-range", 0,
863         memmap[VIRT_PCIE_ECAM].size / PCIE_MMCFG_SIZE_MIN - 1);
864     qemu_fdt_setprop(ms->fdt, name, "dma-coherent", NULL, 0);
865     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
866         qemu_fdt_setprop_cell(ms->fdt, name, "msi-parent", msi_pcie_phandle);
867     }
868     qemu_fdt_setprop_cells(ms->fdt, name, "reg", 0,
869         memmap[VIRT_PCIE_ECAM].base, 0, memmap[VIRT_PCIE_ECAM].size);
870     qemu_fdt_setprop_sized_cells(ms->fdt, name, "ranges",
871         1, FDT_PCI_RANGE_IOPORT, 2, 0,
872         2, memmap[VIRT_PCIE_PIO].base, 2, memmap[VIRT_PCIE_PIO].size,
873         1, FDT_PCI_RANGE_MMIO,
874         2, memmap[VIRT_PCIE_MMIO].base,
875         2, memmap[VIRT_PCIE_MMIO].base, 2, memmap[VIRT_PCIE_MMIO].size,
876         1, FDT_PCI_RANGE_MMIO_64BIT,
877         2, virt_high_pcie_memmap.base,
878         2, virt_high_pcie_memmap.base, 2, virt_high_pcie_memmap.size);
879 
880     create_pcie_irq_map(s, ms->fdt, name, irq_pcie_phandle);
881 }
882 
883 static void create_fdt_reset(RISCVVirtState *s, const MemMapEntry *memmap,
884                              uint32_t *phandle)
885 {
886     char *name;
887     uint32_t test_phandle;
888     MachineState *ms = MACHINE(s);
889 
890     test_phandle = (*phandle)++;
891     name = g_strdup_printf("/soc/test@%lx",
892         (long)memmap[VIRT_TEST].base);
893     qemu_fdt_add_subnode(ms->fdt, name);
894     {
895         static const char * const compat[3] = {
896             "sifive,test1", "sifive,test0", "syscon"
897         };
898         qemu_fdt_setprop_string_array(ms->fdt, name, "compatible",
899                                       (char **)&compat, ARRAY_SIZE(compat));
900     }
901     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
902         0x0, memmap[VIRT_TEST].base, 0x0, memmap[VIRT_TEST].size);
903     qemu_fdt_setprop_cell(ms->fdt, name, "phandle", test_phandle);
904     test_phandle = qemu_fdt_get_phandle(ms->fdt, name);
905     g_free(name);
906 
907     name = g_strdup_printf("/reboot");
908     qemu_fdt_add_subnode(ms->fdt, name);
909     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "syscon-reboot");
910     qemu_fdt_setprop_cell(ms->fdt, name, "regmap", test_phandle);
911     qemu_fdt_setprop_cell(ms->fdt, name, "offset", 0x0);
912     qemu_fdt_setprop_cell(ms->fdt, name, "value", FINISHER_RESET);
913     g_free(name);
914 
915     name = g_strdup_printf("/poweroff");
916     qemu_fdt_add_subnode(ms->fdt, name);
917     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "syscon-poweroff");
918     qemu_fdt_setprop_cell(ms->fdt, name, "regmap", test_phandle);
919     qemu_fdt_setprop_cell(ms->fdt, name, "offset", 0x0);
920     qemu_fdt_setprop_cell(ms->fdt, name, "value", FINISHER_PASS);
921     g_free(name);
922 }
923 
924 static void create_fdt_uart(RISCVVirtState *s, const MemMapEntry *memmap,
925                             uint32_t irq_mmio_phandle)
926 {
927     g_autofree char *name = NULL;
928     MachineState *ms = MACHINE(s);
929 
930     name = g_strdup_printf("/soc/serial@%lx", (long)memmap[VIRT_UART0].base);
931     qemu_fdt_add_subnode(ms->fdt, name);
932     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "ns16550a");
933     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
934         0x0, memmap[VIRT_UART0].base,
935         0x0, memmap[VIRT_UART0].size);
936     qemu_fdt_setprop_cell(ms->fdt, name, "clock-frequency", 3686400);
937     qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent", irq_mmio_phandle);
938     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
939         qemu_fdt_setprop_cell(ms->fdt, name, "interrupts", UART0_IRQ);
940     } else {
941         qemu_fdt_setprop_cells(ms->fdt, name, "interrupts", UART0_IRQ, 0x4);
942     }
943 
944     qemu_fdt_setprop_string(ms->fdt, "/chosen", "stdout-path", name);
945 }
946 
947 static void create_fdt_rtc(RISCVVirtState *s, const MemMapEntry *memmap,
948                            uint32_t irq_mmio_phandle)
949 {
950     g_autofree char *name = NULL;
951     MachineState *ms = MACHINE(s);
952 
953     name = g_strdup_printf("/soc/rtc@%lx", (long)memmap[VIRT_RTC].base);
954     qemu_fdt_add_subnode(ms->fdt, name);
955     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
956         "google,goldfish-rtc");
957     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
958         0x0, memmap[VIRT_RTC].base, 0x0, memmap[VIRT_RTC].size);
959     qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent",
960         irq_mmio_phandle);
961     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
962         qemu_fdt_setprop_cell(ms->fdt, name, "interrupts", RTC_IRQ);
963     } else {
964         qemu_fdt_setprop_cells(ms->fdt, name, "interrupts", RTC_IRQ, 0x4);
965     }
966 }
967 
968 static void create_fdt_flash(RISCVVirtState *s, const MemMapEntry *memmap)
969 {
970     MachineState *ms = MACHINE(s);
971     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
972     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
973     g_autofree char *name = g_strdup_printf("/flash@%" PRIx64, flashbase);
974 
975     qemu_fdt_add_subnode(ms->fdt, name);
976     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "cfi-flash");
977     qemu_fdt_setprop_sized_cells(ms->fdt, name, "reg",
978                                  2, flashbase, 2, flashsize,
979                                  2, flashbase + flashsize, 2, flashsize);
980     qemu_fdt_setprop_cell(ms->fdt, name, "bank-width", 4);
981 }
982 
983 static void create_fdt_fw_cfg(RISCVVirtState *s, const MemMapEntry *memmap)
984 {
985     MachineState *ms = MACHINE(s);
986     hwaddr base = memmap[VIRT_FW_CFG].base;
987     hwaddr size = memmap[VIRT_FW_CFG].size;
988     g_autofree char *nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
989 
990     qemu_fdt_add_subnode(ms->fdt, nodename);
991     qemu_fdt_setprop_string(ms->fdt, nodename,
992                             "compatible", "qemu,fw-cfg-mmio");
993     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
994                                  2, base, 2, size);
995     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
996 }
997 
998 static void create_fdt_virtio_iommu(RISCVVirtState *s, uint16_t bdf)
999 {
1000     const char compat[] = "virtio,pci-iommu\0pci1af4,1057";
1001     void *fdt = MACHINE(s)->fdt;
1002     uint32_t iommu_phandle;
1003     g_autofree char *iommu_node = NULL;
1004     g_autofree char *pci_node = NULL;
1005 
1006     pci_node = g_strdup_printf("/soc/pci@%lx",
1007                                (long) virt_memmap[VIRT_PCIE_ECAM].base);
1008     iommu_node = g_strdup_printf("%s/virtio_iommu@%x,%x", pci_node,
1009                                  PCI_SLOT(bdf), PCI_FUNC(bdf));
1010     iommu_phandle = qemu_fdt_alloc_phandle(fdt);
1011 
1012     qemu_fdt_add_subnode(fdt, iommu_node);
1013 
1014     qemu_fdt_setprop(fdt, iommu_node, "compatible", compat, sizeof(compat));
1015     qemu_fdt_setprop_sized_cells(fdt, iommu_node, "reg",
1016                                  1, bdf << 8, 1, 0, 1, 0,
1017                                  1, 0, 1, 0);
1018     qemu_fdt_setprop_cell(fdt, iommu_node, "#iommu-cells", 1);
1019     qemu_fdt_setprop_cell(fdt, iommu_node, "phandle", iommu_phandle);
1020 
1021     qemu_fdt_setprop_cells(fdt, pci_node, "iommu-map",
1022                            0, iommu_phandle, 0, bdf,
1023                            bdf + 1, iommu_phandle, bdf + 1, 0xffff - bdf);
1024 }
1025 
1026 static void finalize_fdt(RISCVVirtState *s)
1027 {
1028     uint32_t phandle = 1, irq_mmio_phandle = 1, msi_pcie_phandle = 1;
1029     uint32_t irq_pcie_phandle = 1, irq_virtio_phandle = 1;
1030 
1031     create_fdt_sockets(s, virt_memmap, &phandle, &irq_mmio_phandle,
1032                        &irq_pcie_phandle, &irq_virtio_phandle,
1033                        &msi_pcie_phandle);
1034 
1035     create_fdt_virtio(s, virt_memmap, irq_virtio_phandle);
1036 
1037     create_fdt_pcie(s, virt_memmap, irq_pcie_phandle, msi_pcie_phandle);
1038 
1039     create_fdt_reset(s, virt_memmap, &phandle);
1040 
1041     create_fdt_uart(s, virt_memmap, irq_mmio_phandle);
1042 
1043     create_fdt_rtc(s, virt_memmap, irq_mmio_phandle);
1044 }
1045 
1046 static void create_fdt(RISCVVirtState *s, const MemMapEntry *memmap)
1047 {
1048     MachineState *ms = MACHINE(s);
1049     uint8_t rng_seed[32];
1050     g_autofree char *name = NULL;
1051 
1052     ms->fdt = create_device_tree(&s->fdt_size);
1053     if (!ms->fdt) {
1054         error_report("create_device_tree() failed");
1055         exit(1);
1056     }
1057 
1058     qemu_fdt_setprop_string(ms->fdt, "/", "model", "riscv-virtio,qemu");
1059     qemu_fdt_setprop_string(ms->fdt, "/", "compatible", "riscv-virtio");
1060     qemu_fdt_setprop_cell(ms->fdt, "/", "#size-cells", 0x2);
1061     qemu_fdt_setprop_cell(ms->fdt, "/", "#address-cells", 0x2);
1062 
1063     qemu_fdt_add_subnode(ms->fdt, "/soc");
1064     qemu_fdt_setprop(ms->fdt, "/soc", "ranges", NULL, 0);
1065     qemu_fdt_setprop_string(ms->fdt, "/soc", "compatible", "simple-bus");
1066     qemu_fdt_setprop_cell(ms->fdt, "/soc", "#size-cells", 0x2);
1067     qemu_fdt_setprop_cell(ms->fdt, "/soc", "#address-cells", 0x2);
1068 
1069     /*
1070      * The "/soc/pci@..." node is needed for PCIE hotplugs
1071      * that might happen before finalize_fdt().
1072      */
1073     name = g_strdup_printf("/soc/pci@%lx", (long) memmap[VIRT_PCIE_ECAM].base);
1074     qemu_fdt_add_subnode(ms->fdt, name);
1075 
1076     qemu_fdt_add_subnode(ms->fdt, "/chosen");
1077 
1078     /* Pass seed to RNG */
1079     qemu_guest_getrandom_nofail(rng_seed, sizeof(rng_seed));
1080     qemu_fdt_setprop(ms->fdt, "/chosen", "rng-seed",
1081                      rng_seed, sizeof(rng_seed));
1082 
1083     create_fdt_flash(s, memmap);
1084     create_fdt_fw_cfg(s, memmap);
1085     create_fdt_pmu(s);
1086 }
1087 
1088 static inline DeviceState *gpex_pcie_init(MemoryRegion *sys_mem,
1089                                           DeviceState *irqchip,
1090                                           RISCVVirtState *s)
1091 {
1092     DeviceState *dev;
1093     MemoryRegion *ecam_alias, *ecam_reg;
1094     MemoryRegion *mmio_alias, *high_mmio_alias, *mmio_reg;
1095     hwaddr ecam_base = s->memmap[VIRT_PCIE_ECAM].base;
1096     hwaddr ecam_size = s->memmap[VIRT_PCIE_ECAM].size;
1097     hwaddr mmio_base = s->memmap[VIRT_PCIE_MMIO].base;
1098     hwaddr mmio_size = s->memmap[VIRT_PCIE_MMIO].size;
1099     hwaddr high_mmio_base = virt_high_pcie_memmap.base;
1100     hwaddr high_mmio_size = virt_high_pcie_memmap.size;
1101     hwaddr pio_base = s->memmap[VIRT_PCIE_PIO].base;
1102     hwaddr pio_size = s->memmap[VIRT_PCIE_PIO].size;
1103     qemu_irq irq;
1104     int i;
1105 
1106     dev = qdev_new(TYPE_GPEX_HOST);
1107 
1108     /* Set GPEX object properties for the virt machine */
1109     object_property_set_uint(OBJECT(GPEX_HOST(dev)), PCI_HOST_ECAM_BASE,
1110                             ecam_base, NULL);
1111     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_ECAM_SIZE,
1112                             ecam_size, NULL);
1113     object_property_set_uint(OBJECT(GPEX_HOST(dev)),
1114                              PCI_HOST_BELOW_4G_MMIO_BASE,
1115                              mmio_base, NULL);
1116     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_BELOW_4G_MMIO_SIZE,
1117                             mmio_size, NULL);
1118     object_property_set_uint(OBJECT(GPEX_HOST(dev)),
1119                              PCI_HOST_ABOVE_4G_MMIO_BASE,
1120                              high_mmio_base, NULL);
1121     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_ABOVE_4G_MMIO_SIZE,
1122                             high_mmio_size, NULL);
1123     object_property_set_uint(OBJECT(GPEX_HOST(dev)), PCI_HOST_PIO_BASE,
1124                             pio_base, NULL);
1125     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_PIO_SIZE,
1126                             pio_size, NULL);
1127 
1128     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1129 
1130     ecam_alias = g_new0(MemoryRegion, 1);
1131     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1132     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1133                              ecam_reg, 0, ecam_size);
1134     memory_region_add_subregion(get_system_memory(), ecam_base, ecam_alias);
1135 
1136     mmio_alias = g_new0(MemoryRegion, 1);
1137     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1138     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1139                              mmio_reg, mmio_base, mmio_size);
1140     memory_region_add_subregion(get_system_memory(), mmio_base, mmio_alias);
1141 
1142     /* Map high MMIO space */
1143     high_mmio_alias = g_new0(MemoryRegion, 1);
1144     memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1145                              mmio_reg, high_mmio_base, high_mmio_size);
1146     memory_region_add_subregion(get_system_memory(), high_mmio_base,
1147                                 high_mmio_alias);
1148 
1149     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, pio_base);
1150 
1151     for (i = 0; i < GPEX_NUM_IRQS; i++) {
1152         irq = qdev_get_gpio_in(irqchip, PCIE_IRQ + i);
1153 
1154         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, irq);
1155         gpex_set_irq_num(GPEX_HOST(dev), i, PCIE_IRQ + i);
1156     }
1157 
1158     GPEX_HOST(dev)->gpex_cfg.bus = PCI_HOST_BRIDGE(GPEX_HOST(dev))->bus;
1159     return dev;
1160 }
1161 
1162 static FWCfgState *create_fw_cfg(const MachineState *ms)
1163 {
1164     hwaddr base = virt_memmap[VIRT_FW_CFG].base;
1165     FWCfgState *fw_cfg;
1166 
1167     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16,
1168                                   &address_space_memory);
1169     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)ms->smp.cpus);
1170 
1171     return fw_cfg;
1172 }
1173 
1174 static DeviceState *virt_create_plic(const MemMapEntry *memmap, int socket,
1175                                      int base_hartid, int hart_count)
1176 {
1177     DeviceState *ret;
1178     g_autofree char *plic_hart_config = NULL;
1179 
1180     /* Per-socket PLIC hart topology configuration string */
1181     plic_hart_config = riscv_plic_hart_config_string(hart_count);
1182 
1183     /* Per-socket PLIC */
1184     ret = sifive_plic_create(
1185             memmap[VIRT_PLIC].base + socket * memmap[VIRT_PLIC].size,
1186             plic_hart_config, hart_count, base_hartid,
1187             VIRT_IRQCHIP_NUM_SOURCES,
1188             ((1U << VIRT_IRQCHIP_NUM_PRIO_BITS) - 1),
1189             VIRT_PLIC_PRIORITY_BASE,
1190             VIRT_PLIC_PENDING_BASE,
1191             VIRT_PLIC_ENABLE_BASE,
1192             VIRT_PLIC_ENABLE_STRIDE,
1193             VIRT_PLIC_CONTEXT_BASE,
1194             VIRT_PLIC_CONTEXT_STRIDE,
1195             memmap[VIRT_PLIC].size);
1196 
1197     return ret;
1198 }
1199 
1200 static DeviceState *virt_create_aia(RISCVVirtAIAType aia_type, int aia_guests,
1201                                     const MemMapEntry *memmap, int socket,
1202                                     int base_hartid, int hart_count)
1203 {
1204     int i;
1205     hwaddr addr;
1206     uint32_t guest_bits;
1207     DeviceState *aplic_s = NULL;
1208     DeviceState *aplic_m = NULL;
1209     bool msimode = aia_type == VIRT_AIA_TYPE_APLIC_IMSIC;
1210 
1211     if (msimode) {
1212         if (!kvm_enabled()) {
1213             /* Per-socket M-level IMSICs */
1214             addr = memmap[VIRT_IMSIC_M].base +
1215                    socket * VIRT_IMSIC_GROUP_MAX_SIZE;
1216             for (i = 0; i < hart_count; i++) {
1217                 riscv_imsic_create(addr + i * IMSIC_HART_SIZE(0),
1218                                    base_hartid + i, true, 1,
1219                                    VIRT_IRQCHIP_NUM_MSIS);
1220             }
1221         }
1222 
1223         /* Per-socket S-level IMSICs */
1224         guest_bits = imsic_num_bits(aia_guests + 1);
1225         addr = memmap[VIRT_IMSIC_S].base + socket * VIRT_IMSIC_GROUP_MAX_SIZE;
1226         for (i = 0; i < hart_count; i++) {
1227             riscv_imsic_create(addr + i * IMSIC_HART_SIZE(guest_bits),
1228                                base_hartid + i, false, 1 + aia_guests,
1229                                VIRT_IRQCHIP_NUM_MSIS);
1230         }
1231     }
1232 
1233     if (!kvm_enabled()) {
1234         /* Per-socket M-level APLIC */
1235         aplic_m = riscv_aplic_create(memmap[VIRT_APLIC_M].base +
1236                                      socket * memmap[VIRT_APLIC_M].size,
1237                                      memmap[VIRT_APLIC_M].size,
1238                                      (msimode) ? 0 : base_hartid,
1239                                      (msimode) ? 0 : hart_count,
1240                                      VIRT_IRQCHIP_NUM_SOURCES,
1241                                      VIRT_IRQCHIP_NUM_PRIO_BITS,
1242                                      msimode, true, NULL);
1243     }
1244 
1245     /* Per-socket S-level APLIC */
1246     aplic_s = riscv_aplic_create(memmap[VIRT_APLIC_S].base +
1247                                  socket * memmap[VIRT_APLIC_S].size,
1248                                  memmap[VIRT_APLIC_S].size,
1249                                  (msimode) ? 0 : base_hartid,
1250                                  (msimode) ? 0 : hart_count,
1251                                  VIRT_IRQCHIP_NUM_SOURCES,
1252                                  VIRT_IRQCHIP_NUM_PRIO_BITS,
1253                                  msimode, false, aplic_m);
1254 
1255     return kvm_enabled() ? aplic_s : aplic_m;
1256 }
1257 
1258 static void create_platform_bus(RISCVVirtState *s, DeviceState *irqchip)
1259 {
1260     DeviceState *dev;
1261     SysBusDevice *sysbus;
1262     const MemMapEntry *memmap = virt_memmap;
1263     int i;
1264     MemoryRegion *sysmem = get_system_memory();
1265 
1266     dev = qdev_new(TYPE_PLATFORM_BUS_DEVICE);
1267     dev->id = g_strdup(TYPE_PLATFORM_BUS_DEVICE);
1268     qdev_prop_set_uint32(dev, "num_irqs", VIRT_PLATFORM_BUS_NUM_IRQS);
1269     qdev_prop_set_uint32(dev, "mmio_size", memmap[VIRT_PLATFORM_BUS].size);
1270     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1271     s->platform_bus_dev = dev;
1272 
1273     sysbus = SYS_BUS_DEVICE(dev);
1274     for (i = 0; i < VIRT_PLATFORM_BUS_NUM_IRQS; i++) {
1275         int irq = VIRT_PLATFORM_BUS_IRQ + i;
1276         sysbus_connect_irq(sysbus, i, qdev_get_gpio_in(irqchip, irq));
1277     }
1278 
1279     memory_region_add_subregion(sysmem,
1280                                 memmap[VIRT_PLATFORM_BUS].base,
1281                                 sysbus_mmio_get_region(sysbus, 0));
1282 }
1283 
1284 static void virt_build_smbios(RISCVVirtState *s)
1285 {
1286     MachineClass *mc = MACHINE_GET_CLASS(s);
1287     MachineState *ms = MACHINE(s);
1288     uint8_t *smbios_tables, *smbios_anchor;
1289     size_t smbios_tables_len, smbios_anchor_len;
1290     struct smbios_phys_mem_area mem_array;
1291     const char *product = "QEMU Virtual Machine";
1292 
1293     if (kvm_enabled()) {
1294         product = "KVM Virtual Machine";
1295     }
1296 
1297     smbios_set_defaults("QEMU", product, mc->name);
1298 
1299     if (riscv_is_32bit(&s->soc[0])) {
1300         smbios_set_default_processor_family(0x200);
1301     } else {
1302         smbios_set_default_processor_family(0x201);
1303     }
1304 
1305     /* build the array of physical mem area from base_memmap */
1306     mem_array.address = s->memmap[VIRT_DRAM].base;
1307     mem_array.length = ms->ram_size;
1308 
1309     smbios_get_tables(ms, SMBIOS_ENTRY_POINT_TYPE_64,
1310                       &mem_array, 1,
1311                       &smbios_tables, &smbios_tables_len,
1312                       &smbios_anchor, &smbios_anchor_len,
1313                       &error_fatal);
1314 
1315     if (smbios_anchor) {
1316         fw_cfg_add_file(s->fw_cfg, "etc/smbios/smbios-tables",
1317                         smbios_tables, smbios_tables_len);
1318         fw_cfg_add_file(s->fw_cfg, "etc/smbios/smbios-anchor",
1319                         smbios_anchor, smbios_anchor_len);
1320     }
1321 }
1322 
1323 static void virt_machine_done(Notifier *notifier, void *data)
1324 {
1325     RISCVVirtState *s = container_of(notifier, RISCVVirtState,
1326                                      machine_done);
1327     const MemMapEntry *memmap = virt_memmap;
1328     MachineState *machine = MACHINE(s);
1329     target_ulong start_addr = memmap[VIRT_DRAM].base;
1330     target_ulong firmware_end_addr, kernel_start_addr;
1331     const char *firmware_name = riscv_default_firmware_name(&s->soc[0]);
1332     uint64_t fdt_load_addr;
1333     uint64_t kernel_entry = 0;
1334     BlockBackend *pflash_blk0;
1335 
1336     /*
1337      * An user provided dtb must include everything, including
1338      * dynamic sysbus devices. Our FDT needs to be finalized.
1339      */
1340     if (machine->dtb == NULL) {
1341         finalize_fdt(s);
1342     }
1343 
1344     /*
1345      * Only direct boot kernel is currently supported for KVM VM,
1346      * so the "-bios" parameter is not supported when KVM is enabled.
1347      */
1348     if (kvm_enabled()) {
1349         if (machine->firmware) {
1350             if (strcmp(machine->firmware, "none")) {
1351                 error_report("Machine mode firmware is not supported in "
1352                              "combination with KVM.");
1353                 exit(1);
1354             }
1355         } else {
1356             machine->firmware = g_strdup("none");
1357         }
1358     }
1359 
1360     firmware_end_addr = riscv_find_and_load_firmware(machine, firmware_name,
1361                                                      start_addr, NULL);
1362 
1363     pflash_blk0 = pflash_cfi01_get_blk(s->flash[0]);
1364     if (pflash_blk0) {
1365         if (machine->firmware && !strcmp(machine->firmware, "none") &&
1366             !kvm_enabled()) {
1367             /*
1368              * Pflash was supplied but bios is none and not KVM guest,
1369              * let's overwrite the address we jump to after reset to
1370              * the base of the flash.
1371              */
1372             start_addr = virt_memmap[VIRT_FLASH].base;
1373         } else {
1374             /*
1375              * Pflash was supplied but either KVM guest or bios is not none.
1376              * In this case, base of the flash would contain S-mode payload.
1377              */
1378             riscv_setup_firmware_boot(machine);
1379             kernel_entry = virt_memmap[VIRT_FLASH].base;
1380         }
1381     }
1382 
1383     if (machine->kernel_filename && !kernel_entry) {
1384         kernel_start_addr = riscv_calc_kernel_start_addr(&s->soc[0],
1385                                                          firmware_end_addr);
1386 
1387         kernel_entry = riscv_load_kernel(machine, &s->soc[0],
1388                                          kernel_start_addr, true, NULL);
1389     }
1390 
1391     fdt_load_addr = riscv_compute_fdt_addr(memmap[VIRT_DRAM].base,
1392                                            memmap[VIRT_DRAM].size,
1393                                            machine);
1394     riscv_load_fdt(fdt_load_addr, machine->fdt);
1395 
1396     /* load the reset vector */
1397     riscv_setup_rom_reset_vec(machine, &s->soc[0], start_addr,
1398                               virt_memmap[VIRT_MROM].base,
1399                               virt_memmap[VIRT_MROM].size, kernel_entry,
1400                               fdt_load_addr);
1401 
1402     /*
1403      * Only direct boot kernel is currently supported for KVM VM,
1404      * So here setup kernel start address and fdt address.
1405      * TODO:Support firmware loading and integrate to TCG start
1406      */
1407     if (kvm_enabled()) {
1408         riscv_setup_direct_kernel(kernel_entry, fdt_load_addr);
1409     }
1410 
1411     virt_build_smbios(s);
1412 
1413     if (virt_is_acpi_enabled(s)) {
1414         virt_acpi_setup(s);
1415     }
1416 }
1417 
1418 static void virt_machine_init(MachineState *machine)
1419 {
1420     const MemMapEntry *memmap = virt_memmap;
1421     RISCVVirtState *s = RISCV_VIRT_MACHINE(machine);
1422     MemoryRegion *system_memory = get_system_memory();
1423     MemoryRegion *mask_rom = g_new(MemoryRegion, 1);
1424     DeviceState *mmio_irqchip, *virtio_irqchip, *pcie_irqchip;
1425     int i, base_hartid, hart_count;
1426     int socket_count = riscv_socket_count(machine);
1427 
1428     /* Check socket count limit */
1429     if (VIRT_SOCKETS_MAX < socket_count) {
1430         error_report("number of sockets/nodes should be less than %d",
1431             VIRT_SOCKETS_MAX);
1432         exit(1);
1433     }
1434 
1435     if (!virt_aclint_allowed() && s->have_aclint) {
1436         error_report("'aclint' is only available with TCG acceleration");
1437         exit(1);
1438     }
1439 
1440     /* Initialize sockets */
1441     mmio_irqchip = virtio_irqchip = pcie_irqchip = NULL;
1442     for (i = 0; i < socket_count; i++) {
1443         g_autofree char *soc_name = g_strdup_printf("soc%d", i);
1444 
1445         if (!riscv_socket_check_hartids(machine, i)) {
1446             error_report("discontinuous hartids in socket%d", i);
1447             exit(1);
1448         }
1449 
1450         base_hartid = riscv_socket_first_hartid(machine, i);
1451         if (base_hartid < 0) {
1452             error_report("can't find hartid base for socket%d", i);
1453             exit(1);
1454         }
1455 
1456         hart_count = riscv_socket_hart_count(machine, i);
1457         if (hart_count < 0) {
1458             error_report("can't find hart count for socket%d", i);
1459             exit(1);
1460         }
1461 
1462         object_initialize_child(OBJECT(machine), soc_name, &s->soc[i],
1463                                 TYPE_RISCV_HART_ARRAY);
1464         object_property_set_str(OBJECT(&s->soc[i]), "cpu-type",
1465                                 machine->cpu_type, &error_abort);
1466         object_property_set_int(OBJECT(&s->soc[i]), "hartid-base",
1467                                 base_hartid, &error_abort);
1468         object_property_set_int(OBJECT(&s->soc[i]), "num-harts",
1469                                 hart_count, &error_abort);
1470         sysbus_realize(SYS_BUS_DEVICE(&s->soc[i]), &error_fatal);
1471 
1472         if (virt_aclint_allowed() && s->have_aclint) {
1473             if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
1474                 /* Per-socket ACLINT MTIMER */
1475                 riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1476                             i * RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1477                         RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1478                         base_hartid, hart_count,
1479                         RISCV_ACLINT_DEFAULT_MTIMECMP,
1480                         RISCV_ACLINT_DEFAULT_MTIME,
1481                         RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1482             } else {
1483                 /* Per-socket ACLINT MSWI, MTIMER, and SSWI */
1484                 riscv_aclint_swi_create(memmap[VIRT_CLINT].base +
1485                             i * memmap[VIRT_CLINT].size,
1486                         base_hartid, hart_count, false);
1487                 riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1488                             i * memmap[VIRT_CLINT].size +
1489                             RISCV_ACLINT_SWI_SIZE,
1490                         RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1491                         base_hartid, hart_count,
1492                         RISCV_ACLINT_DEFAULT_MTIMECMP,
1493                         RISCV_ACLINT_DEFAULT_MTIME,
1494                         RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1495                 riscv_aclint_swi_create(memmap[VIRT_ACLINT_SSWI].base +
1496                             i * memmap[VIRT_ACLINT_SSWI].size,
1497                         base_hartid, hart_count, true);
1498             }
1499         } else if (tcg_enabled()) {
1500             /* Per-socket SiFive CLINT */
1501             riscv_aclint_swi_create(
1502                     memmap[VIRT_CLINT].base + i * memmap[VIRT_CLINT].size,
1503                     base_hartid, hart_count, false);
1504             riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1505                         i * memmap[VIRT_CLINT].size + RISCV_ACLINT_SWI_SIZE,
1506                     RISCV_ACLINT_DEFAULT_MTIMER_SIZE, base_hartid, hart_count,
1507                     RISCV_ACLINT_DEFAULT_MTIMECMP, RISCV_ACLINT_DEFAULT_MTIME,
1508                     RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1509         }
1510 
1511         /* Per-socket interrupt controller */
1512         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
1513             s->irqchip[i] = virt_create_plic(memmap, i,
1514                                              base_hartid, hart_count);
1515         } else {
1516             s->irqchip[i] = virt_create_aia(s->aia_type, s->aia_guests,
1517                                             memmap, i, base_hartid,
1518                                             hart_count);
1519         }
1520 
1521         /* Try to use different IRQCHIP instance based device type */
1522         if (i == 0) {
1523             mmio_irqchip = s->irqchip[i];
1524             virtio_irqchip = s->irqchip[i];
1525             pcie_irqchip = s->irqchip[i];
1526         }
1527         if (i == 1) {
1528             virtio_irqchip = s->irqchip[i];
1529             pcie_irqchip = s->irqchip[i];
1530         }
1531         if (i == 2) {
1532             pcie_irqchip = s->irqchip[i];
1533         }
1534     }
1535 
1536     if (kvm_enabled() && virt_use_kvm_aia(s)) {
1537         kvm_riscv_aia_create(machine, IMSIC_MMIO_GROUP_MIN_SHIFT,
1538                              VIRT_IRQCHIP_NUM_SOURCES, VIRT_IRQCHIP_NUM_MSIS,
1539                              memmap[VIRT_APLIC_S].base,
1540                              memmap[VIRT_IMSIC_S].base,
1541                              s->aia_guests);
1542     }
1543 
1544     if (riscv_is_32bit(&s->soc[0])) {
1545 #if HOST_LONG_BITS == 64
1546         /* limit RAM size in a 32-bit system */
1547         if (machine->ram_size > 10 * GiB) {
1548             machine->ram_size = 10 * GiB;
1549             error_report("Limiting RAM size to 10 GiB");
1550         }
1551 #endif
1552         virt_high_pcie_memmap.base = VIRT32_HIGH_PCIE_MMIO_BASE;
1553         virt_high_pcie_memmap.size = VIRT32_HIGH_PCIE_MMIO_SIZE;
1554     } else {
1555         virt_high_pcie_memmap.size = VIRT64_HIGH_PCIE_MMIO_SIZE;
1556         virt_high_pcie_memmap.base = memmap[VIRT_DRAM].base + machine->ram_size;
1557         virt_high_pcie_memmap.base =
1558             ROUND_UP(virt_high_pcie_memmap.base, virt_high_pcie_memmap.size);
1559     }
1560 
1561     s->memmap = virt_memmap;
1562 
1563     /* register system main memory (actual RAM) */
1564     memory_region_add_subregion(system_memory, memmap[VIRT_DRAM].base,
1565         machine->ram);
1566 
1567     /* boot rom */
1568     memory_region_init_rom(mask_rom, NULL, "riscv_virt_board.mrom",
1569                            memmap[VIRT_MROM].size, &error_fatal);
1570     memory_region_add_subregion(system_memory, memmap[VIRT_MROM].base,
1571                                 mask_rom);
1572 
1573     /*
1574      * Init fw_cfg. Must be done before riscv_load_fdt, otherwise the
1575      * device tree cannot be altered and we get FDT_ERR_NOSPACE.
1576      */
1577     s->fw_cfg = create_fw_cfg(machine);
1578     rom_set_fw(s->fw_cfg);
1579 
1580     /* SiFive Test MMIO device */
1581     sifive_test_create(memmap[VIRT_TEST].base);
1582 
1583     /* VirtIO MMIO devices */
1584     for (i = 0; i < VIRTIO_COUNT; i++) {
1585         sysbus_create_simple("virtio-mmio",
1586             memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
1587             qdev_get_gpio_in(virtio_irqchip, VIRTIO_IRQ + i));
1588     }
1589 
1590     gpex_pcie_init(system_memory, pcie_irqchip, s);
1591 
1592     create_platform_bus(s, mmio_irqchip);
1593 
1594     serial_mm_init(system_memory, memmap[VIRT_UART0].base,
1595         0, qdev_get_gpio_in(mmio_irqchip, UART0_IRQ), 399193,
1596         serial_hd(0), DEVICE_LITTLE_ENDIAN);
1597 
1598     sysbus_create_simple("goldfish_rtc", memmap[VIRT_RTC].base,
1599         qdev_get_gpio_in(mmio_irqchip, RTC_IRQ));
1600 
1601     for (i = 0; i < ARRAY_SIZE(s->flash); i++) {
1602         /* Map legacy -drive if=pflash to machine properties */
1603         pflash_cfi01_legacy_drive(s->flash[i],
1604                                   drive_get(IF_PFLASH, 0, i));
1605     }
1606     virt_flash_map(s, system_memory);
1607 
1608     /* load/create device tree */
1609     if (machine->dtb) {
1610         machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
1611         if (!machine->fdt) {
1612             error_report("load_device_tree() failed");
1613             exit(1);
1614         }
1615     } else {
1616         create_fdt(s, memmap);
1617     }
1618 
1619     s->machine_done.notify = virt_machine_done;
1620     qemu_add_machine_init_done_notifier(&s->machine_done);
1621 }
1622 
1623 static void virt_machine_instance_init(Object *obj)
1624 {
1625     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1626 
1627     virt_flash_create(s);
1628 
1629     s->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
1630     s->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
1631     s->acpi = ON_OFF_AUTO_AUTO;
1632 }
1633 
1634 static char *virt_get_aia_guests(Object *obj, Error **errp)
1635 {
1636     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1637 
1638     return g_strdup_printf("%d", s->aia_guests);
1639 }
1640 
1641 static void virt_set_aia_guests(Object *obj, const char *val, Error **errp)
1642 {
1643     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1644 
1645     s->aia_guests = atoi(val);
1646     if (s->aia_guests < 0 || s->aia_guests > VIRT_IRQCHIP_MAX_GUESTS) {
1647         error_setg(errp, "Invalid number of AIA IMSIC guests");
1648         error_append_hint(errp, "Valid values be between 0 and %d.\n",
1649                           VIRT_IRQCHIP_MAX_GUESTS);
1650     }
1651 }
1652 
1653 static char *virt_get_aia(Object *obj, Error **errp)
1654 {
1655     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1656     const char *val;
1657 
1658     switch (s->aia_type) {
1659     case VIRT_AIA_TYPE_APLIC:
1660         val = "aplic";
1661         break;
1662     case VIRT_AIA_TYPE_APLIC_IMSIC:
1663         val = "aplic-imsic";
1664         break;
1665     default:
1666         val = "none";
1667         break;
1668     };
1669 
1670     return g_strdup(val);
1671 }
1672 
1673 static void virt_set_aia(Object *obj, const char *val, Error **errp)
1674 {
1675     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1676 
1677     if (!strcmp(val, "none")) {
1678         s->aia_type = VIRT_AIA_TYPE_NONE;
1679     } else if (!strcmp(val, "aplic")) {
1680         s->aia_type = VIRT_AIA_TYPE_APLIC;
1681     } else if (!strcmp(val, "aplic-imsic")) {
1682         s->aia_type = VIRT_AIA_TYPE_APLIC_IMSIC;
1683     } else {
1684         error_setg(errp, "Invalid AIA interrupt controller type");
1685         error_append_hint(errp, "Valid values are none, aplic, and "
1686                           "aplic-imsic.\n");
1687     }
1688 }
1689 
1690 static bool virt_get_aclint(Object *obj, Error **errp)
1691 {
1692     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1693 
1694     return s->have_aclint;
1695 }
1696 
1697 static void virt_set_aclint(Object *obj, bool value, Error **errp)
1698 {
1699     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1700 
1701     s->have_aclint = value;
1702 }
1703 
1704 bool virt_is_acpi_enabled(RISCVVirtState *s)
1705 {
1706     return s->acpi != ON_OFF_AUTO_OFF;
1707 }
1708 
1709 static void virt_get_acpi(Object *obj, Visitor *v, const char *name,
1710                           void *opaque, Error **errp)
1711 {
1712     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1713     OnOffAuto acpi = s->acpi;
1714 
1715     visit_type_OnOffAuto(v, name, &acpi, errp);
1716 }
1717 
1718 static void virt_set_acpi(Object *obj, Visitor *v, const char *name,
1719                           void *opaque, Error **errp)
1720 {
1721     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1722 
1723     visit_type_OnOffAuto(v, name, &s->acpi, errp);
1724 }
1725 
1726 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
1727                                                         DeviceState *dev)
1728 {
1729     MachineClass *mc = MACHINE_GET_CLASS(machine);
1730 
1731     if (device_is_dynamic_sysbus(mc, dev) ||
1732         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
1733         return HOTPLUG_HANDLER(machine);
1734     }
1735     return NULL;
1736 }
1737 
1738 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
1739                                         DeviceState *dev, Error **errp)
1740 {
1741     RISCVVirtState *s = RISCV_VIRT_MACHINE(hotplug_dev);
1742 
1743     if (s->platform_bus_dev) {
1744         MachineClass *mc = MACHINE_GET_CLASS(s);
1745 
1746         if (device_is_dynamic_sysbus(mc, dev)) {
1747             platform_bus_link_device(PLATFORM_BUS_DEVICE(s->platform_bus_dev),
1748                                      SYS_BUS_DEVICE(dev));
1749         }
1750     }
1751 
1752     if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
1753         create_fdt_virtio_iommu(s, pci_get_bdf(PCI_DEVICE(dev)));
1754     }
1755 }
1756 
1757 static void virt_machine_class_init(ObjectClass *oc, void *data)
1758 {
1759     MachineClass *mc = MACHINE_CLASS(oc);
1760     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
1761 
1762     mc->desc = "RISC-V VirtIO board";
1763     mc->init = virt_machine_init;
1764     mc->max_cpus = VIRT_CPUS_MAX;
1765     mc->default_cpu_type = TYPE_RISCV_CPU_BASE;
1766     mc->pci_allow_0_address = true;
1767     mc->possible_cpu_arch_ids = riscv_numa_possible_cpu_arch_ids;
1768     mc->cpu_index_to_instance_props = riscv_numa_cpu_index_to_props;
1769     mc->get_default_cpu_node_id = riscv_numa_get_default_cpu_node_id;
1770     mc->numa_mem_supported = true;
1771     /* platform instead of architectural choice */
1772     mc->cpu_cluster_has_numa_boundary = true;
1773     mc->default_ram_id = "riscv_virt_board.ram";
1774     assert(!mc->get_hotplug_handler);
1775     mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
1776 
1777     hc->plug = virt_machine_device_plug_cb;
1778 
1779     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
1780 #ifdef CONFIG_TPM
1781     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_TPM_TIS_SYSBUS);
1782 #endif
1783 
1784     object_class_property_add_bool(oc, "aclint", virt_get_aclint,
1785                                    virt_set_aclint);
1786     object_class_property_set_description(oc, "aclint",
1787                                           "(TCG only) Set on/off to "
1788                                           "enable/disable emulating "
1789                                           "ACLINT devices");
1790 
1791     object_class_property_add_str(oc, "aia", virt_get_aia,
1792                                   virt_set_aia);
1793     object_class_property_set_description(oc, "aia",
1794                                           "Set type of AIA interrupt "
1795                                           "controller. Valid values are "
1796                                           "none, aplic, and aplic-imsic.");
1797 
1798     object_class_property_add_str(oc, "aia-guests",
1799                                   virt_get_aia_guests,
1800                                   virt_set_aia_guests);
1801     {
1802         g_autofree char *str =
1803             g_strdup_printf("Set number of guest MMIO pages for AIA IMSIC. "
1804                             "Valid value should be between 0 and %d.",
1805                             VIRT_IRQCHIP_MAX_GUESTS);
1806         object_class_property_set_description(oc, "aia-guests", str);
1807     }
1808 
1809     object_class_property_add(oc, "acpi", "OnOffAuto",
1810                               virt_get_acpi, virt_set_acpi,
1811                               NULL, NULL);
1812     object_class_property_set_description(oc, "acpi",
1813                                           "Enable ACPI");
1814 }
1815 
1816 static const TypeInfo virt_machine_typeinfo = {
1817     .name       = MACHINE_TYPE_NAME("virt"),
1818     .parent     = TYPE_MACHINE,
1819     .class_init = virt_machine_class_init,
1820     .instance_init = virt_machine_instance_init,
1821     .instance_size = sizeof(RISCVVirtState),
1822     .interfaces = (InterfaceInfo[]) {
1823          { TYPE_HOTPLUG_HANDLER },
1824          { }
1825     },
1826 };
1827 
1828 static void virt_machine_init_register_types(void)
1829 {
1830     type_register_static(&virt_machine_typeinfo);
1831 }
1832 
1833 type_init(virt_machine_init_register_types)
1834