xref: /qemu/hw/riscv/virt.c (revision 7778cddddaeea98920589bb7ba91effd78f0cf04)
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 "hw/pci/pci.h"
52 #include "hw/pci-host/gpex.h"
53 #include "hw/display/ramfb.h"
54 #include "hw/acpi/aml-build.h"
55 #include "qapi/qapi-visit-common.h"
56 #include "hw/virtio/virtio-iommu.h"
57 
58 /* KVM AIA only supports APLIC MSI. APLIC Wired is always emulated by QEMU. */
59 static bool virt_use_kvm_aia(RISCVVirtState *s)
60 {
61     return kvm_irqchip_in_kernel() && s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC;
62 }
63 
64 static const MemMapEntry virt_memmap[] = {
65     [VIRT_DEBUG] =        {        0x0,         0x100 },
66     [VIRT_MROM] =         {     0x1000,        0xf000 },
67     [VIRT_TEST] =         {   0x100000,        0x1000 },
68     [VIRT_RTC] =          {   0x101000,        0x1000 },
69     [VIRT_CLINT] =        {  0x2000000,       0x10000 },
70     [VIRT_ACLINT_SSWI] =  {  0x2F00000,        0x4000 },
71     [VIRT_PCIE_PIO] =     {  0x3000000,       0x10000 },
72     [VIRT_PLATFORM_BUS] = {  0x4000000,     0x2000000 },
73     [VIRT_PLIC] =         {  0xc000000, VIRT_PLIC_SIZE(VIRT_CPUS_MAX * 2) },
74     [VIRT_APLIC_M] =      {  0xc000000, APLIC_SIZE(VIRT_CPUS_MAX) },
75     [VIRT_APLIC_S] =      {  0xd000000, APLIC_SIZE(VIRT_CPUS_MAX) },
76     [VIRT_UART0] =        { 0x10000000,         0x100 },
77     [VIRT_VIRTIO] =       { 0x10001000,        0x1000 },
78     [VIRT_FW_CFG] =       { 0x10100000,          0x18 },
79     [VIRT_FLASH] =        { 0x20000000,     0x4000000 },
80     [VIRT_IMSIC_M] =      { 0x24000000, VIRT_IMSIC_MAX_SIZE },
81     [VIRT_IMSIC_S] =      { 0x28000000, VIRT_IMSIC_MAX_SIZE },
82     [VIRT_PCIE_ECAM] =    { 0x30000000,    0x10000000 },
83     [VIRT_PCIE_MMIO] =    { 0x40000000,    0x40000000 },
84     [VIRT_DRAM] =         { 0x80000000,           0x0 },
85 };
86 
87 /* PCIe high mmio is fixed for RV32 */
88 #define VIRT32_HIGH_PCIE_MMIO_BASE  0x300000000ULL
89 #define VIRT32_HIGH_PCIE_MMIO_SIZE  (4 * GiB)
90 
91 /* PCIe high mmio for RV64, size is fixed but base depends on top of RAM */
92 #define VIRT64_HIGH_PCIE_MMIO_SIZE  (16 * GiB)
93 
94 static MemMapEntry virt_high_pcie_memmap;
95 
96 #define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
97 
98 static PFlashCFI01 *virt_flash_create1(RISCVVirtState *s,
99                                        const char *name,
100                                        const char *alias_prop_name)
101 {
102     /*
103      * Create a single flash device.  We use the same parameters as
104      * the flash devices on the ARM virt board.
105      */
106     DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
107 
108     qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
109     qdev_prop_set_uint8(dev, "width", 4);
110     qdev_prop_set_uint8(dev, "device-width", 2);
111     qdev_prop_set_bit(dev, "big-endian", false);
112     qdev_prop_set_uint16(dev, "id0", 0x89);
113     qdev_prop_set_uint16(dev, "id1", 0x18);
114     qdev_prop_set_uint16(dev, "id2", 0x00);
115     qdev_prop_set_uint16(dev, "id3", 0x00);
116     qdev_prop_set_string(dev, "name", name);
117 
118     object_property_add_child(OBJECT(s), name, OBJECT(dev));
119     object_property_add_alias(OBJECT(s), alias_prop_name,
120                               OBJECT(dev), "drive");
121 
122     return PFLASH_CFI01(dev);
123 }
124 
125 static void virt_flash_create(RISCVVirtState *s)
126 {
127     s->flash[0] = virt_flash_create1(s, "virt.flash0", "pflash0");
128     s->flash[1] = virt_flash_create1(s, "virt.flash1", "pflash1");
129 }
130 
131 static void virt_flash_map1(PFlashCFI01 *flash,
132                             hwaddr base, hwaddr size,
133                             MemoryRegion *sysmem)
134 {
135     DeviceState *dev = DEVICE(flash);
136 
137     assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
138     assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
139     qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
140     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
141 
142     memory_region_add_subregion(sysmem, base,
143                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
144                                                        0));
145 }
146 
147 static void virt_flash_map(RISCVVirtState *s,
148                            MemoryRegion *sysmem)
149 {
150     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
151     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
152 
153     virt_flash_map1(s->flash[0], flashbase, flashsize,
154                     sysmem);
155     virt_flash_map1(s->flash[1], flashbase + flashsize, flashsize,
156                     sysmem);
157 }
158 
159 static void create_pcie_irq_map(RISCVVirtState *s, void *fdt, char *nodename,
160                                 uint32_t irqchip_phandle)
161 {
162     int pin, dev;
163     uint32_t irq_map_stride = 0;
164     uint32_t full_irq_map[GPEX_NUM_IRQS * GPEX_NUM_IRQS *
165                           FDT_MAX_INT_MAP_WIDTH] = {};
166     uint32_t *irq_map = full_irq_map;
167 
168     /* This code creates a standard swizzle of interrupts such that
169      * each device's first interrupt is based on it's PCI_SLOT number.
170      * (See pci_swizzle_map_irq_fn())
171      *
172      * We only need one entry per interrupt in the table (not one per
173      * possible slot) seeing the interrupt-map-mask will allow the table
174      * to wrap to any number of devices.
175      */
176     for (dev = 0; dev < GPEX_NUM_IRQS; dev++) {
177         int devfn = dev * 0x8;
178 
179         for (pin = 0; pin < GPEX_NUM_IRQS; pin++) {
180             int irq_nr = PCIE_IRQ + ((pin + PCI_SLOT(devfn)) % GPEX_NUM_IRQS);
181             int i = 0;
182 
183             /* Fill PCI address cells */
184             irq_map[i] = cpu_to_be32(devfn << 8);
185             i += FDT_PCI_ADDR_CELLS;
186 
187             /* Fill PCI Interrupt cells */
188             irq_map[i] = cpu_to_be32(pin + 1);
189             i += FDT_PCI_INT_CELLS;
190 
191             /* Fill interrupt controller phandle and cells */
192             irq_map[i++] = cpu_to_be32(irqchip_phandle);
193             irq_map[i++] = cpu_to_be32(irq_nr);
194             if (s->aia_type != VIRT_AIA_TYPE_NONE) {
195                 irq_map[i++] = cpu_to_be32(0x4);
196             }
197 
198             if (!irq_map_stride) {
199                 irq_map_stride = i;
200             }
201             irq_map += irq_map_stride;
202         }
203     }
204 
205     qemu_fdt_setprop(fdt, nodename, "interrupt-map", full_irq_map,
206                      GPEX_NUM_IRQS * GPEX_NUM_IRQS *
207                      irq_map_stride * sizeof(uint32_t));
208 
209     qemu_fdt_setprop_cells(fdt, nodename, "interrupt-map-mask",
210                            0x1800, 0, 0, 0x7);
211 }
212 
213 static void create_fdt_socket_cpus(RISCVVirtState *s, int socket,
214                                    char *clust_name, uint32_t *phandle,
215                                    uint32_t *intc_phandles)
216 {
217     int cpu;
218     uint32_t cpu_phandle;
219     MachineState *ms = MACHINE(s);
220     bool is_32_bit = riscv_is_32bit(&s->soc[0]);
221     uint8_t satp_mode_max;
222 
223     for (cpu = s->soc[socket].num_harts - 1; cpu >= 0; cpu--) {
224         RISCVCPU *cpu_ptr = &s->soc[socket].harts[cpu];
225         g_autofree char *cpu_name = NULL;
226         g_autofree char *core_name = NULL;
227         g_autofree char *intc_name = NULL;
228         g_autofree char *sv_name = NULL;
229 
230         cpu_phandle = (*phandle)++;
231 
232         cpu_name = g_strdup_printf("/cpus/cpu@%d",
233             s->soc[socket].hartid_base + cpu);
234         qemu_fdt_add_subnode(ms->fdt, cpu_name);
235 
236         if (cpu_ptr->cfg.satp_mode.supported != 0) {
237             satp_mode_max = satp_mode_max_from_map(cpu_ptr->cfg.satp_mode.map);
238             sv_name = g_strdup_printf("riscv,%s",
239                                       satp_mode_str(satp_mode_max, is_32_bit));
240             qemu_fdt_setprop_string(ms->fdt, cpu_name, "mmu-type", sv_name);
241         }
242 
243         riscv_isa_write_fdt(cpu_ptr, ms->fdt, cpu_name);
244 
245         if (cpu_ptr->cfg.ext_zicbom) {
246             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cbom-block-size",
247                                   cpu_ptr->cfg.cbom_blocksize);
248         }
249 
250         if (cpu_ptr->cfg.ext_zicboz) {
251             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cboz-block-size",
252                                   cpu_ptr->cfg.cboz_blocksize);
253         }
254 
255         if (cpu_ptr->cfg.ext_zicbop) {
256             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cbop-block-size",
257                                   cpu_ptr->cfg.cbop_blocksize);
258         }
259 
260         qemu_fdt_setprop_string(ms->fdt, cpu_name, "compatible", "riscv");
261         qemu_fdt_setprop_string(ms->fdt, cpu_name, "status", "okay");
262         qemu_fdt_setprop_cell(ms->fdt, cpu_name, "reg",
263             s->soc[socket].hartid_base + cpu);
264         qemu_fdt_setprop_string(ms->fdt, cpu_name, "device_type", "cpu");
265         riscv_socket_fdt_write_id(ms, cpu_name, socket);
266         qemu_fdt_setprop_cell(ms->fdt, cpu_name, "phandle", cpu_phandle);
267 
268         intc_phandles[cpu] = (*phandle)++;
269 
270         intc_name = g_strdup_printf("%s/interrupt-controller", cpu_name);
271         qemu_fdt_add_subnode(ms->fdt, intc_name);
272         qemu_fdt_setprop_cell(ms->fdt, intc_name, "phandle",
273             intc_phandles[cpu]);
274         qemu_fdt_setprop_string(ms->fdt, intc_name, "compatible",
275             "riscv,cpu-intc");
276         qemu_fdt_setprop(ms->fdt, intc_name, "interrupt-controller", NULL, 0);
277         qemu_fdt_setprop_cell(ms->fdt, intc_name, "#interrupt-cells", 1);
278 
279         core_name = g_strdup_printf("%s/core%d", clust_name, cpu);
280         qemu_fdt_add_subnode(ms->fdt, core_name);
281         qemu_fdt_setprop_cell(ms->fdt, core_name, "cpu", cpu_phandle);
282     }
283 }
284 
285 static void create_fdt_socket_memory(RISCVVirtState *s,
286                                      const MemMapEntry *memmap, int socket)
287 {
288     g_autofree char *mem_name = NULL;
289     uint64_t addr, size;
290     MachineState *ms = MACHINE(s);
291 
292     addr = memmap[VIRT_DRAM].base + riscv_socket_mem_offset(ms, socket);
293     size = riscv_socket_mem_size(ms, socket);
294     mem_name = g_strdup_printf("/memory@%lx", (long)addr);
295     qemu_fdt_add_subnode(ms->fdt, mem_name);
296     qemu_fdt_setprop_cells(ms->fdt, mem_name, "reg",
297         addr >> 32, addr, size >> 32, size);
298     qemu_fdt_setprop_string(ms->fdt, mem_name, "device_type", "memory");
299     riscv_socket_fdt_write_id(ms, mem_name, socket);
300 }
301 
302 static void create_fdt_socket_clint(RISCVVirtState *s,
303                                     const MemMapEntry *memmap, int socket,
304                                     uint32_t *intc_phandles)
305 {
306     int cpu;
307     g_autofree char *clint_name = NULL;
308     g_autofree uint32_t *clint_cells = NULL;
309     unsigned long clint_addr;
310     MachineState *ms = MACHINE(s);
311     static const char * const clint_compat[2] = {
312         "sifive,clint0", "riscv,clint0"
313     };
314 
315     clint_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
316 
317     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
318         clint_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
319         clint_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_SOFT);
320         clint_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
321         clint_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_M_TIMER);
322     }
323 
324     clint_addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
325     clint_name = g_strdup_printf("/soc/clint@%lx", clint_addr);
326     qemu_fdt_add_subnode(ms->fdt, clint_name);
327     qemu_fdt_setprop_string_array(ms->fdt, clint_name, "compatible",
328                                   (char **)&clint_compat,
329                                   ARRAY_SIZE(clint_compat));
330     qemu_fdt_setprop_cells(ms->fdt, clint_name, "reg",
331         0x0, clint_addr, 0x0, memmap[VIRT_CLINT].size);
332     qemu_fdt_setprop(ms->fdt, clint_name, "interrupts-extended",
333         clint_cells, s->soc[socket].num_harts * sizeof(uint32_t) * 4);
334     riscv_socket_fdt_write_id(ms, clint_name, socket);
335 }
336 
337 static void create_fdt_socket_aclint(RISCVVirtState *s,
338                                      const MemMapEntry *memmap, int socket,
339                                      uint32_t *intc_phandles)
340 {
341     int cpu;
342     char *name;
343     unsigned long addr, size;
344     uint32_t aclint_cells_size;
345     g_autofree uint32_t *aclint_mswi_cells = NULL;
346     g_autofree uint32_t *aclint_sswi_cells = NULL;
347     g_autofree uint32_t *aclint_mtimer_cells = NULL;
348     MachineState *ms = MACHINE(s);
349 
350     aclint_mswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
351     aclint_mtimer_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
352     aclint_sswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
353 
354     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
355         aclint_mswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
356         aclint_mswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_SOFT);
357         aclint_mtimer_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
358         aclint_mtimer_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_TIMER);
359         aclint_sswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
360         aclint_sswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_SOFT);
361     }
362     aclint_cells_size = s->soc[socket].num_harts * sizeof(uint32_t) * 2;
363 
364     if (s->aia_type != VIRT_AIA_TYPE_APLIC_IMSIC) {
365         addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
366         name = g_strdup_printf("/soc/mswi@%lx", addr);
367         qemu_fdt_add_subnode(ms->fdt, name);
368         qemu_fdt_setprop_string(ms->fdt, name, "compatible",
369             "riscv,aclint-mswi");
370         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
371             0x0, addr, 0x0, RISCV_ACLINT_SWI_SIZE);
372         qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
373             aclint_mswi_cells, aclint_cells_size);
374         qemu_fdt_setprop(ms->fdt, name, "interrupt-controller", NULL, 0);
375         qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells", 0);
376         riscv_socket_fdt_write_id(ms, name, socket);
377         g_free(name);
378     }
379 
380     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
381         addr = memmap[VIRT_CLINT].base +
382                (RISCV_ACLINT_DEFAULT_MTIMER_SIZE * socket);
383         size = RISCV_ACLINT_DEFAULT_MTIMER_SIZE;
384     } else {
385         addr = memmap[VIRT_CLINT].base + RISCV_ACLINT_SWI_SIZE +
386             (memmap[VIRT_CLINT].size * socket);
387         size = memmap[VIRT_CLINT].size - RISCV_ACLINT_SWI_SIZE;
388     }
389     name = g_strdup_printf("/soc/mtimer@%lx", addr);
390     qemu_fdt_add_subnode(ms->fdt, name);
391     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
392         "riscv,aclint-mtimer");
393     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
394         0x0, addr + RISCV_ACLINT_DEFAULT_MTIME,
395         0x0, size - RISCV_ACLINT_DEFAULT_MTIME,
396         0x0, addr + RISCV_ACLINT_DEFAULT_MTIMECMP,
397         0x0, RISCV_ACLINT_DEFAULT_MTIME);
398     qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
399         aclint_mtimer_cells, aclint_cells_size);
400     riscv_socket_fdt_write_id(ms, name, socket);
401     g_free(name);
402 
403     if (s->aia_type != VIRT_AIA_TYPE_APLIC_IMSIC) {
404         addr = memmap[VIRT_ACLINT_SSWI].base +
405             (memmap[VIRT_ACLINT_SSWI].size * socket);
406         name = g_strdup_printf("/soc/sswi@%lx", addr);
407         qemu_fdt_add_subnode(ms->fdt, name);
408         qemu_fdt_setprop_string(ms->fdt, name, "compatible",
409             "riscv,aclint-sswi");
410         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
411             0x0, addr, 0x0, memmap[VIRT_ACLINT_SSWI].size);
412         qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
413             aclint_sswi_cells, aclint_cells_size);
414         qemu_fdt_setprop(ms->fdt, name, "interrupt-controller", NULL, 0);
415         qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells", 0);
416         riscv_socket_fdt_write_id(ms, name, socket);
417         g_free(name);
418     }
419 }
420 
421 static void create_fdt_socket_plic(RISCVVirtState *s,
422                                    const MemMapEntry *memmap, int socket,
423                                    uint32_t *phandle, uint32_t *intc_phandles,
424                                    uint32_t *plic_phandles)
425 {
426     int cpu;
427     g_autofree char *plic_name = NULL;
428     g_autofree uint32_t *plic_cells;
429     unsigned long plic_addr;
430     MachineState *ms = MACHINE(s);
431     static const char * const plic_compat[2] = {
432         "sifive,plic-1.0.0", "riscv,plic0"
433     };
434 
435     plic_phandles[socket] = (*phandle)++;
436     plic_addr = memmap[VIRT_PLIC].base + (memmap[VIRT_PLIC].size * socket);
437     plic_name = g_strdup_printf("/soc/plic@%lx", plic_addr);
438     qemu_fdt_add_subnode(ms->fdt, plic_name);
439     qemu_fdt_setprop_cell(ms->fdt, plic_name,
440         "#interrupt-cells", FDT_PLIC_INT_CELLS);
441     qemu_fdt_setprop_cell(ms->fdt, plic_name,
442         "#address-cells", FDT_PLIC_ADDR_CELLS);
443     qemu_fdt_setprop_string_array(ms->fdt, plic_name, "compatible",
444                                   (char **)&plic_compat,
445                                   ARRAY_SIZE(plic_compat));
446     qemu_fdt_setprop(ms->fdt, plic_name, "interrupt-controller", NULL, 0);
447 
448     if (kvm_enabled()) {
449         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
450 
451         for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
452             plic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
453             plic_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_EXT);
454         }
455 
456         qemu_fdt_setprop(ms->fdt, plic_name, "interrupts-extended",
457                          plic_cells,
458                          s->soc[socket].num_harts * sizeof(uint32_t) * 2);
459    } else {
460         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
461 
462         for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
463             plic_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
464             plic_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_EXT);
465             plic_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
466             plic_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_S_EXT);
467         }
468 
469         qemu_fdt_setprop(ms->fdt, plic_name, "interrupts-extended",
470                          plic_cells,
471                          s->soc[socket].num_harts * sizeof(uint32_t) * 4);
472     }
473 
474     qemu_fdt_setprop_cells(ms->fdt, plic_name, "reg",
475         0x0, plic_addr, 0x0, memmap[VIRT_PLIC].size);
476     qemu_fdt_setprop_cell(ms->fdt, plic_name, "riscv,ndev",
477                           VIRT_IRQCHIP_NUM_SOURCES - 1);
478     riscv_socket_fdt_write_id(ms, plic_name, socket);
479     qemu_fdt_setprop_cell(ms->fdt, plic_name, "phandle",
480         plic_phandles[socket]);
481 
482     if (!socket) {
483         platform_bus_add_all_fdt_nodes(ms->fdt, plic_name,
484                                        memmap[VIRT_PLATFORM_BUS].base,
485                                        memmap[VIRT_PLATFORM_BUS].size,
486                                        VIRT_PLATFORM_BUS_IRQ);
487     }
488 }
489 
490 uint32_t imsic_num_bits(uint32_t count)
491 {
492     uint32_t ret = 0;
493 
494     while (BIT(ret) < count) {
495         ret++;
496     }
497 
498     return ret;
499 }
500 
501 static void create_fdt_one_imsic(RISCVVirtState *s, hwaddr base_addr,
502                                  uint32_t *intc_phandles, uint32_t msi_phandle,
503                                  bool m_mode, uint32_t imsic_guest_bits)
504 {
505     int cpu, socket;
506     g_autofree char *imsic_name = NULL;
507     MachineState *ms = MACHINE(s);
508     int socket_count = riscv_socket_count(ms);
509     uint32_t imsic_max_hart_per_socket, imsic_addr, imsic_size;
510     g_autofree uint32_t *imsic_cells = NULL;
511     g_autofree uint32_t *imsic_regs = NULL;
512 
513     imsic_cells = g_new0(uint32_t, ms->smp.cpus * 2);
514     imsic_regs = g_new0(uint32_t, socket_count * 4);
515 
516     for (cpu = 0; cpu < ms->smp.cpus; cpu++) {
517         imsic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
518         imsic_cells[cpu * 2 + 1] = cpu_to_be32(m_mode ? IRQ_M_EXT : IRQ_S_EXT);
519     }
520 
521     imsic_max_hart_per_socket = 0;
522     for (socket = 0; socket < socket_count; socket++) {
523         imsic_addr = base_addr + socket * VIRT_IMSIC_GROUP_MAX_SIZE;
524         imsic_size = IMSIC_HART_SIZE(imsic_guest_bits) *
525                      s->soc[socket].num_harts;
526         imsic_regs[socket * 4 + 0] = 0;
527         imsic_regs[socket * 4 + 1] = cpu_to_be32(imsic_addr);
528         imsic_regs[socket * 4 + 2] = 0;
529         imsic_regs[socket * 4 + 3] = cpu_to_be32(imsic_size);
530         if (imsic_max_hart_per_socket < s->soc[socket].num_harts) {
531             imsic_max_hart_per_socket = s->soc[socket].num_harts;
532         }
533     }
534 
535     imsic_name = g_strdup_printf("/soc/imsics@%lx", (unsigned long)base_addr);
536     qemu_fdt_add_subnode(ms->fdt, imsic_name);
537     qemu_fdt_setprop_string(ms->fdt, imsic_name, "compatible", "riscv,imsics");
538     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "#interrupt-cells",
539                           FDT_IMSIC_INT_CELLS);
540     qemu_fdt_setprop(ms->fdt, imsic_name, "interrupt-controller", NULL, 0);
541     qemu_fdt_setprop(ms->fdt, imsic_name, "msi-controller", NULL, 0);
542     qemu_fdt_setprop(ms->fdt, imsic_name, "interrupts-extended",
543                      imsic_cells, ms->smp.cpus * sizeof(uint32_t) * 2);
544     qemu_fdt_setprop(ms->fdt, imsic_name, "reg", imsic_regs,
545                      socket_count * sizeof(uint32_t) * 4);
546     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,num-ids",
547                      VIRT_IRQCHIP_NUM_MSIS);
548 
549     if (imsic_guest_bits) {
550         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,guest-index-bits",
551                               imsic_guest_bits);
552     }
553 
554     if (socket_count > 1) {
555         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,hart-index-bits",
556                               imsic_num_bits(imsic_max_hart_per_socket));
557         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,group-index-bits",
558                               imsic_num_bits(socket_count));
559         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,group-index-shift",
560                               IMSIC_MMIO_GROUP_MIN_SHIFT);
561     }
562     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "phandle", msi_phandle);
563 }
564 
565 static void create_fdt_imsic(RISCVVirtState *s, const MemMapEntry *memmap,
566                              uint32_t *phandle, uint32_t *intc_phandles,
567                              uint32_t *msi_m_phandle, uint32_t *msi_s_phandle)
568 {
569     *msi_m_phandle = (*phandle)++;
570     *msi_s_phandle = (*phandle)++;
571 
572     if (!kvm_enabled()) {
573         /* M-level IMSIC node */
574         create_fdt_one_imsic(s, memmap[VIRT_IMSIC_M].base, intc_phandles,
575                              *msi_m_phandle, true, 0);
576     }
577 
578     /* S-level IMSIC node */
579     create_fdt_one_imsic(s, memmap[VIRT_IMSIC_S].base, intc_phandles,
580                          *msi_s_phandle, false,
581                          imsic_num_bits(s->aia_guests + 1));
582 
583 }
584 
585 static void create_fdt_one_aplic(RISCVVirtState *s, int socket,
586                                  unsigned long aplic_addr, uint32_t aplic_size,
587                                  uint32_t msi_phandle,
588                                  uint32_t *intc_phandles,
589                                  uint32_t aplic_phandle,
590                                  uint32_t aplic_child_phandle,
591                                  bool m_mode, int num_harts)
592 {
593     int cpu;
594     g_autofree char *aplic_name = NULL;
595     g_autofree uint32_t *aplic_cells = g_new0(uint32_t, num_harts * 2);
596     MachineState *ms = MACHINE(s);
597 
598     for (cpu = 0; cpu < num_harts; cpu++) {
599         aplic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
600         aplic_cells[cpu * 2 + 1] = cpu_to_be32(m_mode ? IRQ_M_EXT : IRQ_S_EXT);
601     }
602 
603     aplic_name = g_strdup_printf("/soc/aplic@%lx", aplic_addr);
604     qemu_fdt_add_subnode(ms->fdt, aplic_name);
605     qemu_fdt_setprop_string(ms->fdt, aplic_name, "compatible", "riscv,aplic");
606     qemu_fdt_setprop_cell(ms->fdt, aplic_name,
607                           "#interrupt-cells", FDT_APLIC_INT_CELLS);
608     qemu_fdt_setprop(ms->fdt, aplic_name, "interrupt-controller", NULL, 0);
609 
610     if (s->aia_type == VIRT_AIA_TYPE_APLIC) {
611         qemu_fdt_setprop(ms->fdt, aplic_name, "interrupts-extended",
612                          aplic_cells, num_harts * sizeof(uint32_t) * 2);
613     } else {
614         qemu_fdt_setprop_cell(ms->fdt, aplic_name, "msi-parent", msi_phandle);
615     }
616 
617     qemu_fdt_setprop_cells(ms->fdt, aplic_name, "reg",
618                            0x0, aplic_addr, 0x0, aplic_size);
619     qemu_fdt_setprop_cell(ms->fdt, aplic_name, "riscv,num-sources",
620                           VIRT_IRQCHIP_NUM_SOURCES);
621 
622     if (aplic_child_phandle) {
623         qemu_fdt_setprop_cell(ms->fdt, aplic_name, "riscv,children",
624                               aplic_child_phandle);
625         qemu_fdt_setprop_cells(ms->fdt, aplic_name, "riscv,delegate",
626                                aplic_child_phandle, 0x1,
627                                VIRT_IRQCHIP_NUM_SOURCES);
628     }
629 
630     riscv_socket_fdt_write_id(ms, aplic_name, socket);
631     qemu_fdt_setprop_cell(ms->fdt, aplic_name, "phandle", aplic_phandle);
632 }
633 
634 static void create_fdt_socket_aplic(RISCVVirtState *s,
635                                     const MemMapEntry *memmap, int socket,
636                                     uint32_t msi_m_phandle,
637                                     uint32_t msi_s_phandle,
638                                     uint32_t *phandle,
639                                     uint32_t *intc_phandles,
640                                     uint32_t *aplic_phandles,
641                                     int num_harts)
642 {
643     g_autofree char *aplic_name = NULL;
644     unsigned long aplic_addr;
645     MachineState *ms = MACHINE(s);
646     uint32_t aplic_m_phandle, aplic_s_phandle;
647 
648     aplic_m_phandle = (*phandle)++;
649     aplic_s_phandle = (*phandle)++;
650 
651     if (!kvm_enabled()) {
652         /* M-level APLIC node */
653         aplic_addr = memmap[VIRT_APLIC_M].base +
654                      (memmap[VIRT_APLIC_M].size * socket);
655         create_fdt_one_aplic(s, socket, aplic_addr, memmap[VIRT_APLIC_M].size,
656                              msi_m_phandle, intc_phandles,
657                              aplic_m_phandle, aplic_s_phandle,
658                              true, num_harts);
659     }
660 
661     /* S-level APLIC node */
662     aplic_addr = memmap[VIRT_APLIC_S].base +
663                  (memmap[VIRT_APLIC_S].size * socket);
664     create_fdt_one_aplic(s, socket, aplic_addr, memmap[VIRT_APLIC_S].size,
665                          msi_s_phandle, intc_phandles,
666                          aplic_s_phandle, 0,
667                          false, num_harts);
668 
669     aplic_name = g_strdup_printf("/soc/aplic@%lx", aplic_addr);
670 
671     if (!socket) {
672         platform_bus_add_all_fdt_nodes(ms->fdt, aplic_name,
673                                        memmap[VIRT_PLATFORM_BUS].base,
674                                        memmap[VIRT_PLATFORM_BUS].size,
675                                        VIRT_PLATFORM_BUS_IRQ);
676     }
677 
678     aplic_phandles[socket] = aplic_s_phandle;
679 }
680 
681 static void create_fdt_pmu(RISCVVirtState *s)
682 {
683     g_autofree char *pmu_name = g_strdup_printf("/pmu");
684     MachineState *ms = MACHINE(s);
685     RISCVCPU hart = s->soc[0].harts[0];
686 
687     qemu_fdt_add_subnode(ms->fdt, pmu_name);
688     qemu_fdt_setprop_string(ms->fdt, pmu_name, "compatible", "riscv,pmu");
689     riscv_pmu_generate_fdt_node(ms->fdt, hart.pmu_avail_ctrs, pmu_name);
690 }
691 
692 static void create_fdt_sockets(RISCVVirtState *s, const MemMapEntry *memmap,
693                                uint32_t *phandle,
694                                uint32_t *irq_mmio_phandle,
695                                uint32_t *irq_pcie_phandle,
696                                uint32_t *irq_virtio_phandle,
697                                uint32_t *msi_pcie_phandle)
698 {
699     int socket, phandle_pos;
700     MachineState *ms = MACHINE(s);
701     uint32_t msi_m_phandle = 0, msi_s_phandle = 0;
702     uint32_t xplic_phandles[MAX_NODES];
703     g_autofree uint32_t *intc_phandles = NULL;
704     int socket_count = riscv_socket_count(ms);
705 
706     qemu_fdt_add_subnode(ms->fdt, "/cpus");
707     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "timebase-frequency",
708                           RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ);
709     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#size-cells", 0x0);
710     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#address-cells", 0x1);
711     qemu_fdt_add_subnode(ms->fdt, "/cpus/cpu-map");
712 
713     intc_phandles = g_new0(uint32_t, ms->smp.cpus);
714 
715     phandle_pos = ms->smp.cpus;
716     for (socket = (socket_count - 1); socket >= 0; socket--) {
717         g_autofree char *clust_name = NULL;
718         phandle_pos -= s->soc[socket].num_harts;
719 
720         clust_name = g_strdup_printf("/cpus/cpu-map/cluster%d", socket);
721         qemu_fdt_add_subnode(ms->fdt, clust_name);
722 
723         create_fdt_socket_cpus(s, socket, clust_name, phandle,
724                                &intc_phandles[phandle_pos]);
725 
726         create_fdt_socket_memory(s, memmap, socket);
727 
728         if (tcg_enabled()) {
729             if (s->have_aclint) {
730                 create_fdt_socket_aclint(s, memmap, socket,
731                     &intc_phandles[phandle_pos]);
732             } else {
733                 create_fdt_socket_clint(s, memmap, socket,
734                     &intc_phandles[phandle_pos]);
735             }
736         }
737     }
738 
739     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
740         create_fdt_imsic(s, memmap, phandle, intc_phandles,
741             &msi_m_phandle, &msi_s_phandle);
742         *msi_pcie_phandle = msi_s_phandle;
743     }
744 
745     /* KVM AIA only has one APLIC instance */
746     if (kvm_enabled() && virt_use_kvm_aia(s)) {
747         create_fdt_socket_aplic(s, memmap, 0,
748                                 msi_m_phandle, msi_s_phandle, phandle,
749                                 &intc_phandles[0], xplic_phandles,
750                                 ms->smp.cpus);
751     } else {
752         phandle_pos = ms->smp.cpus;
753         for (socket = (socket_count - 1); socket >= 0; socket--) {
754             phandle_pos -= s->soc[socket].num_harts;
755 
756             if (s->aia_type == VIRT_AIA_TYPE_NONE) {
757                 create_fdt_socket_plic(s, memmap, socket, phandle,
758                                        &intc_phandles[phandle_pos],
759                                        xplic_phandles);
760             } else {
761                 create_fdt_socket_aplic(s, memmap, socket,
762                                         msi_m_phandle, msi_s_phandle, phandle,
763                                         &intc_phandles[phandle_pos],
764                                         xplic_phandles,
765                                         s->soc[socket].num_harts);
766             }
767         }
768     }
769 
770     if (kvm_enabled() && virt_use_kvm_aia(s)) {
771         *irq_mmio_phandle = xplic_phandles[0];
772         *irq_virtio_phandle = xplic_phandles[0];
773         *irq_pcie_phandle = xplic_phandles[0];
774     } else {
775         for (socket = 0; socket < socket_count; socket++) {
776             if (socket == 0) {
777                 *irq_mmio_phandle = xplic_phandles[socket];
778                 *irq_virtio_phandle = xplic_phandles[socket];
779                 *irq_pcie_phandle = xplic_phandles[socket];
780             }
781             if (socket == 1) {
782                 *irq_virtio_phandle = xplic_phandles[socket];
783                 *irq_pcie_phandle = xplic_phandles[socket];
784             }
785             if (socket == 2) {
786                 *irq_pcie_phandle = xplic_phandles[socket];
787             }
788         }
789     }
790 
791     riscv_socket_fdt_write_distance_matrix(ms);
792 }
793 
794 static void create_fdt_virtio(RISCVVirtState *s, const MemMapEntry *memmap,
795                               uint32_t irq_virtio_phandle)
796 {
797     int i;
798     MachineState *ms = MACHINE(s);
799 
800     for (i = 0; i < VIRTIO_COUNT; i++) {
801         g_autofree char *name =  g_strdup_printf("/soc/virtio_mmio@%lx",
802             (long)(memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size));
803 
804         qemu_fdt_add_subnode(ms->fdt, name);
805         qemu_fdt_setprop_string(ms->fdt, name, "compatible", "virtio,mmio");
806         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
807             0x0, memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
808             0x0, memmap[VIRT_VIRTIO].size);
809         qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent",
810             irq_virtio_phandle);
811         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
812             qemu_fdt_setprop_cell(ms->fdt, name, "interrupts",
813                                   VIRTIO_IRQ + i);
814         } else {
815             qemu_fdt_setprop_cells(ms->fdt, name, "interrupts",
816                                    VIRTIO_IRQ + i, 0x4);
817         }
818     }
819 }
820 
821 static void create_fdt_pcie(RISCVVirtState *s, const MemMapEntry *memmap,
822                             uint32_t irq_pcie_phandle,
823                             uint32_t msi_pcie_phandle)
824 {
825     g_autofree char *name = NULL;
826     MachineState *ms = MACHINE(s);
827 
828     name = g_strdup_printf("/soc/pci@%lx",
829         (long) memmap[VIRT_PCIE_ECAM].base);
830     qemu_fdt_setprop_cell(ms->fdt, name, "#address-cells",
831         FDT_PCI_ADDR_CELLS);
832     qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells",
833         FDT_PCI_INT_CELLS);
834     qemu_fdt_setprop_cell(ms->fdt, name, "#size-cells", 0x2);
835     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
836         "pci-host-ecam-generic");
837     qemu_fdt_setprop_string(ms->fdt, name, "device_type", "pci");
838     qemu_fdt_setprop_cell(ms->fdt, name, "linux,pci-domain", 0);
839     qemu_fdt_setprop_cells(ms->fdt, name, "bus-range", 0,
840         memmap[VIRT_PCIE_ECAM].size / PCIE_MMCFG_SIZE_MIN - 1);
841     qemu_fdt_setprop(ms->fdt, name, "dma-coherent", NULL, 0);
842     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
843         qemu_fdt_setprop_cell(ms->fdt, name, "msi-parent", msi_pcie_phandle);
844     }
845     qemu_fdt_setprop_cells(ms->fdt, name, "reg", 0,
846         memmap[VIRT_PCIE_ECAM].base, 0, memmap[VIRT_PCIE_ECAM].size);
847     qemu_fdt_setprop_sized_cells(ms->fdt, name, "ranges",
848         1, FDT_PCI_RANGE_IOPORT, 2, 0,
849         2, memmap[VIRT_PCIE_PIO].base, 2, memmap[VIRT_PCIE_PIO].size,
850         1, FDT_PCI_RANGE_MMIO,
851         2, memmap[VIRT_PCIE_MMIO].base,
852         2, memmap[VIRT_PCIE_MMIO].base, 2, memmap[VIRT_PCIE_MMIO].size,
853         1, FDT_PCI_RANGE_MMIO_64BIT,
854         2, virt_high_pcie_memmap.base,
855         2, virt_high_pcie_memmap.base, 2, virt_high_pcie_memmap.size);
856 
857     create_pcie_irq_map(s, ms->fdt, name, irq_pcie_phandle);
858 }
859 
860 static void create_fdt_reset(RISCVVirtState *s, const MemMapEntry *memmap,
861                              uint32_t *phandle)
862 {
863     char *name;
864     uint32_t test_phandle;
865     MachineState *ms = MACHINE(s);
866 
867     test_phandle = (*phandle)++;
868     name = g_strdup_printf("/soc/test@%lx",
869         (long)memmap[VIRT_TEST].base);
870     qemu_fdt_add_subnode(ms->fdt, name);
871     {
872         static const char * const compat[3] = {
873             "sifive,test1", "sifive,test0", "syscon"
874         };
875         qemu_fdt_setprop_string_array(ms->fdt, name, "compatible",
876                                       (char **)&compat, ARRAY_SIZE(compat));
877     }
878     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
879         0x0, memmap[VIRT_TEST].base, 0x0, memmap[VIRT_TEST].size);
880     qemu_fdt_setprop_cell(ms->fdt, name, "phandle", test_phandle);
881     test_phandle = qemu_fdt_get_phandle(ms->fdt, name);
882     g_free(name);
883 
884     name = g_strdup_printf("/reboot");
885     qemu_fdt_add_subnode(ms->fdt, name);
886     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "syscon-reboot");
887     qemu_fdt_setprop_cell(ms->fdt, name, "regmap", test_phandle);
888     qemu_fdt_setprop_cell(ms->fdt, name, "offset", 0x0);
889     qemu_fdt_setprop_cell(ms->fdt, name, "value", FINISHER_RESET);
890     g_free(name);
891 
892     name = g_strdup_printf("/poweroff");
893     qemu_fdt_add_subnode(ms->fdt, name);
894     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "syscon-poweroff");
895     qemu_fdt_setprop_cell(ms->fdt, name, "regmap", test_phandle);
896     qemu_fdt_setprop_cell(ms->fdt, name, "offset", 0x0);
897     qemu_fdt_setprop_cell(ms->fdt, name, "value", FINISHER_PASS);
898     g_free(name);
899 }
900 
901 static void create_fdt_uart(RISCVVirtState *s, const MemMapEntry *memmap,
902                             uint32_t irq_mmio_phandle)
903 {
904     g_autofree char *name = NULL;
905     MachineState *ms = MACHINE(s);
906 
907     name = g_strdup_printf("/soc/serial@%lx", (long)memmap[VIRT_UART0].base);
908     qemu_fdt_add_subnode(ms->fdt, name);
909     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "ns16550a");
910     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
911         0x0, memmap[VIRT_UART0].base,
912         0x0, memmap[VIRT_UART0].size);
913     qemu_fdt_setprop_cell(ms->fdt, name, "clock-frequency", 3686400);
914     qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent", irq_mmio_phandle);
915     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
916         qemu_fdt_setprop_cell(ms->fdt, name, "interrupts", UART0_IRQ);
917     } else {
918         qemu_fdt_setprop_cells(ms->fdt, name, "interrupts", UART0_IRQ, 0x4);
919     }
920 
921     qemu_fdt_setprop_string(ms->fdt, "/chosen", "stdout-path", name);
922 }
923 
924 static void create_fdt_rtc(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/rtc@%lx", (long)memmap[VIRT_RTC].base);
931     qemu_fdt_add_subnode(ms->fdt, name);
932     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
933         "google,goldfish-rtc");
934     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
935         0x0, memmap[VIRT_RTC].base, 0x0, memmap[VIRT_RTC].size);
936     qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent",
937         irq_mmio_phandle);
938     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
939         qemu_fdt_setprop_cell(ms->fdt, name, "interrupts", RTC_IRQ);
940     } else {
941         qemu_fdt_setprop_cells(ms->fdt, name, "interrupts", RTC_IRQ, 0x4);
942     }
943 }
944 
945 static void create_fdt_flash(RISCVVirtState *s, const MemMapEntry *memmap)
946 {
947     MachineState *ms = MACHINE(s);
948     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
949     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
950     g_autofree char *name = g_strdup_printf("/flash@%" PRIx64, flashbase);
951 
952     qemu_fdt_add_subnode(ms->fdt, name);
953     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "cfi-flash");
954     qemu_fdt_setprop_sized_cells(ms->fdt, name, "reg",
955                                  2, flashbase, 2, flashsize,
956                                  2, flashbase + flashsize, 2, flashsize);
957     qemu_fdt_setprop_cell(ms->fdt, name, "bank-width", 4);
958 }
959 
960 static void create_fdt_fw_cfg(RISCVVirtState *s, const MemMapEntry *memmap)
961 {
962     MachineState *ms = MACHINE(s);
963     hwaddr base = memmap[VIRT_FW_CFG].base;
964     hwaddr size = memmap[VIRT_FW_CFG].size;
965     g_autofree char *nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
966 
967     qemu_fdt_add_subnode(ms->fdt, nodename);
968     qemu_fdt_setprop_string(ms->fdt, nodename,
969                             "compatible", "qemu,fw-cfg-mmio");
970     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
971                                  2, base, 2, size);
972     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
973 }
974 
975 static void create_fdt_virtio_iommu(RISCVVirtState *s, uint16_t bdf)
976 {
977     const char compat[] = "virtio,pci-iommu\0pci1af4,1057";
978     void *fdt = MACHINE(s)->fdt;
979     uint32_t iommu_phandle;
980     g_autofree char *iommu_node = NULL;
981     g_autofree char *pci_node = NULL;
982 
983     pci_node = g_strdup_printf("/soc/pci@%lx",
984                                (long) virt_memmap[VIRT_PCIE_ECAM].base);
985     iommu_node = g_strdup_printf("%s/virtio_iommu@%x,%x", pci_node,
986                                  PCI_SLOT(bdf), PCI_FUNC(bdf));
987     iommu_phandle = qemu_fdt_alloc_phandle(fdt);
988 
989     qemu_fdt_add_subnode(fdt, iommu_node);
990 
991     qemu_fdt_setprop(fdt, iommu_node, "compatible", compat, sizeof(compat));
992     qemu_fdt_setprop_sized_cells(fdt, iommu_node, "reg",
993                                  1, bdf << 8, 1, 0, 1, 0,
994                                  1, 0, 1, 0);
995     qemu_fdt_setprop_cell(fdt, iommu_node, "#iommu-cells", 1);
996     qemu_fdt_setprop_cell(fdt, iommu_node, "phandle", iommu_phandle);
997 
998     qemu_fdt_setprop_cells(fdt, pci_node, "iommu-map",
999                            0, iommu_phandle, 0, bdf,
1000                            bdf + 1, iommu_phandle, bdf + 1, 0xffff - bdf);
1001 }
1002 
1003 static void finalize_fdt(RISCVVirtState *s)
1004 {
1005     uint32_t phandle = 1, irq_mmio_phandle = 1, msi_pcie_phandle = 1;
1006     uint32_t irq_pcie_phandle = 1, irq_virtio_phandle = 1;
1007 
1008     create_fdt_sockets(s, virt_memmap, &phandle, &irq_mmio_phandle,
1009                        &irq_pcie_phandle, &irq_virtio_phandle,
1010                        &msi_pcie_phandle);
1011 
1012     create_fdt_virtio(s, virt_memmap, irq_virtio_phandle);
1013 
1014     create_fdt_pcie(s, virt_memmap, irq_pcie_phandle, msi_pcie_phandle);
1015 
1016     create_fdt_reset(s, virt_memmap, &phandle);
1017 
1018     create_fdt_uart(s, virt_memmap, irq_mmio_phandle);
1019 
1020     create_fdt_rtc(s, virt_memmap, irq_mmio_phandle);
1021 }
1022 
1023 static void create_fdt(RISCVVirtState *s, const MemMapEntry *memmap)
1024 {
1025     MachineState *ms = MACHINE(s);
1026     uint8_t rng_seed[32];
1027     g_autofree char *name = NULL;
1028 
1029     ms->fdt = create_device_tree(&s->fdt_size);
1030     if (!ms->fdt) {
1031         error_report("create_device_tree() failed");
1032         exit(1);
1033     }
1034 
1035     qemu_fdt_setprop_string(ms->fdt, "/", "model", "riscv-virtio,qemu");
1036     qemu_fdt_setprop_string(ms->fdt, "/", "compatible", "riscv-virtio");
1037     qemu_fdt_setprop_cell(ms->fdt, "/", "#size-cells", 0x2);
1038     qemu_fdt_setprop_cell(ms->fdt, "/", "#address-cells", 0x2);
1039 
1040     qemu_fdt_add_subnode(ms->fdt, "/soc");
1041     qemu_fdt_setprop(ms->fdt, "/soc", "ranges", NULL, 0);
1042     qemu_fdt_setprop_string(ms->fdt, "/soc", "compatible", "simple-bus");
1043     qemu_fdt_setprop_cell(ms->fdt, "/soc", "#size-cells", 0x2);
1044     qemu_fdt_setprop_cell(ms->fdt, "/soc", "#address-cells", 0x2);
1045 
1046     /*
1047      * The "/soc/pci@..." node is needed for PCIE hotplugs
1048      * that might happen before finalize_fdt().
1049      */
1050     name = g_strdup_printf("/soc/pci@%lx", (long) memmap[VIRT_PCIE_ECAM].base);
1051     qemu_fdt_add_subnode(ms->fdt, name);
1052 
1053     qemu_fdt_add_subnode(ms->fdt, "/chosen");
1054 
1055     /* Pass seed to RNG */
1056     qemu_guest_getrandom_nofail(rng_seed, sizeof(rng_seed));
1057     qemu_fdt_setprop(ms->fdt, "/chosen", "rng-seed",
1058                      rng_seed, sizeof(rng_seed));
1059 
1060     create_fdt_flash(s, memmap);
1061     create_fdt_fw_cfg(s, memmap);
1062     create_fdt_pmu(s);
1063 }
1064 
1065 static inline DeviceState *gpex_pcie_init(MemoryRegion *sys_mem,
1066                                           DeviceState *irqchip,
1067                                           RISCVVirtState *s)
1068 {
1069     DeviceState *dev;
1070     MemoryRegion *ecam_alias, *ecam_reg;
1071     MemoryRegion *mmio_alias, *high_mmio_alias, *mmio_reg;
1072     hwaddr ecam_base = s->memmap[VIRT_PCIE_ECAM].base;
1073     hwaddr ecam_size = s->memmap[VIRT_PCIE_ECAM].size;
1074     hwaddr mmio_base = s->memmap[VIRT_PCIE_MMIO].base;
1075     hwaddr mmio_size = s->memmap[VIRT_PCIE_MMIO].size;
1076     hwaddr high_mmio_base = virt_high_pcie_memmap.base;
1077     hwaddr high_mmio_size = virt_high_pcie_memmap.size;
1078     hwaddr pio_base = s->memmap[VIRT_PCIE_PIO].base;
1079     hwaddr pio_size = s->memmap[VIRT_PCIE_PIO].size;
1080     qemu_irq irq;
1081     int i;
1082 
1083     dev = qdev_new(TYPE_GPEX_HOST);
1084 
1085     /* Set GPEX object properties for the virt machine */
1086     object_property_set_uint(OBJECT(GPEX_HOST(dev)), PCI_HOST_ECAM_BASE,
1087                             ecam_base, NULL);
1088     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_ECAM_SIZE,
1089                             ecam_size, NULL);
1090     object_property_set_uint(OBJECT(GPEX_HOST(dev)),
1091                              PCI_HOST_BELOW_4G_MMIO_BASE,
1092                              mmio_base, NULL);
1093     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_BELOW_4G_MMIO_SIZE,
1094                             mmio_size, NULL);
1095     object_property_set_uint(OBJECT(GPEX_HOST(dev)),
1096                              PCI_HOST_ABOVE_4G_MMIO_BASE,
1097                              high_mmio_base, NULL);
1098     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_ABOVE_4G_MMIO_SIZE,
1099                             high_mmio_size, NULL);
1100     object_property_set_uint(OBJECT(GPEX_HOST(dev)), PCI_HOST_PIO_BASE,
1101                             pio_base, NULL);
1102     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_PIO_SIZE,
1103                             pio_size, NULL);
1104 
1105     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1106 
1107     ecam_alias = g_new0(MemoryRegion, 1);
1108     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1109     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1110                              ecam_reg, 0, ecam_size);
1111     memory_region_add_subregion(get_system_memory(), ecam_base, ecam_alias);
1112 
1113     mmio_alias = g_new0(MemoryRegion, 1);
1114     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1115     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1116                              mmio_reg, mmio_base, mmio_size);
1117     memory_region_add_subregion(get_system_memory(), mmio_base, mmio_alias);
1118 
1119     /* Map high MMIO space */
1120     high_mmio_alias = g_new0(MemoryRegion, 1);
1121     memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1122                              mmio_reg, high_mmio_base, high_mmio_size);
1123     memory_region_add_subregion(get_system_memory(), high_mmio_base,
1124                                 high_mmio_alias);
1125 
1126     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, pio_base);
1127 
1128     for (i = 0; i < GPEX_NUM_IRQS; i++) {
1129         irq = qdev_get_gpio_in(irqchip, PCIE_IRQ + i);
1130 
1131         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, irq);
1132         gpex_set_irq_num(GPEX_HOST(dev), i, PCIE_IRQ + i);
1133     }
1134 
1135     GPEX_HOST(dev)->gpex_cfg.bus = PCI_HOST_BRIDGE(GPEX_HOST(dev))->bus;
1136     return dev;
1137 }
1138 
1139 static FWCfgState *create_fw_cfg(const MachineState *ms)
1140 {
1141     hwaddr base = virt_memmap[VIRT_FW_CFG].base;
1142     FWCfgState *fw_cfg;
1143 
1144     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16,
1145                                   &address_space_memory);
1146     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)ms->smp.cpus);
1147 
1148     return fw_cfg;
1149 }
1150 
1151 static DeviceState *virt_create_plic(const MemMapEntry *memmap, int socket,
1152                                      int base_hartid, int hart_count)
1153 {
1154     DeviceState *ret;
1155     g_autofree char *plic_hart_config = NULL;
1156 
1157     /* Per-socket PLIC hart topology configuration string */
1158     plic_hart_config = riscv_plic_hart_config_string(hart_count);
1159 
1160     /* Per-socket PLIC */
1161     ret = sifive_plic_create(
1162             memmap[VIRT_PLIC].base + socket * memmap[VIRT_PLIC].size,
1163             plic_hart_config, hart_count, base_hartid,
1164             VIRT_IRQCHIP_NUM_SOURCES,
1165             ((1U << VIRT_IRQCHIP_NUM_PRIO_BITS) - 1),
1166             VIRT_PLIC_PRIORITY_BASE,
1167             VIRT_PLIC_PENDING_BASE,
1168             VIRT_PLIC_ENABLE_BASE,
1169             VIRT_PLIC_ENABLE_STRIDE,
1170             VIRT_PLIC_CONTEXT_BASE,
1171             VIRT_PLIC_CONTEXT_STRIDE,
1172             memmap[VIRT_PLIC].size);
1173 
1174     return ret;
1175 }
1176 
1177 static DeviceState *virt_create_aia(RISCVVirtAIAType aia_type, int aia_guests,
1178                                     const MemMapEntry *memmap, int socket,
1179                                     int base_hartid, int hart_count)
1180 {
1181     int i;
1182     hwaddr addr;
1183     uint32_t guest_bits;
1184     DeviceState *aplic_s = NULL;
1185     DeviceState *aplic_m = NULL;
1186     bool msimode = aia_type == VIRT_AIA_TYPE_APLIC_IMSIC;
1187 
1188     if (msimode) {
1189         if (!kvm_enabled()) {
1190             /* Per-socket M-level IMSICs */
1191             addr = memmap[VIRT_IMSIC_M].base +
1192                    socket * VIRT_IMSIC_GROUP_MAX_SIZE;
1193             for (i = 0; i < hart_count; i++) {
1194                 riscv_imsic_create(addr + i * IMSIC_HART_SIZE(0),
1195                                    base_hartid + i, true, 1,
1196                                    VIRT_IRQCHIP_NUM_MSIS);
1197             }
1198         }
1199 
1200         /* Per-socket S-level IMSICs */
1201         guest_bits = imsic_num_bits(aia_guests + 1);
1202         addr = memmap[VIRT_IMSIC_S].base + socket * VIRT_IMSIC_GROUP_MAX_SIZE;
1203         for (i = 0; i < hart_count; i++) {
1204             riscv_imsic_create(addr + i * IMSIC_HART_SIZE(guest_bits),
1205                                base_hartid + i, false, 1 + aia_guests,
1206                                VIRT_IRQCHIP_NUM_MSIS);
1207         }
1208     }
1209 
1210     if (!kvm_enabled()) {
1211         /* Per-socket M-level APLIC */
1212         aplic_m = riscv_aplic_create(memmap[VIRT_APLIC_M].base +
1213                                      socket * memmap[VIRT_APLIC_M].size,
1214                                      memmap[VIRT_APLIC_M].size,
1215                                      (msimode) ? 0 : base_hartid,
1216                                      (msimode) ? 0 : hart_count,
1217                                      VIRT_IRQCHIP_NUM_SOURCES,
1218                                      VIRT_IRQCHIP_NUM_PRIO_BITS,
1219                                      msimode, true, NULL);
1220     }
1221 
1222     /* Per-socket S-level APLIC */
1223     aplic_s = riscv_aplic_create(memmap[VIRT_APLIC_S].base +
1224                                  socket * memmap[VIRT_APLIC_S].size,
1225                                  memmap[VIRT_APLIC_S].size,
1226                                  (msimode) ? 0 : base_hartid,
1227                                  (msimode) ? 0 : hart_count,
1228                                  VIRT_IRQCHIP_NUM_SOURCES,
1229                                  VIRT_IRQCHIP_NUM_PRIO_BITS,
1230                                  msimode, false, aplic_m);
1231 
1232     return kvm_enabled() ? aplic_s : aplic_m;
1233 }
1234 
1235 static void create_platform_bus(RISCVVirtState *s, DeviceState *irqchip)
1236 {
1237     DeviceState *dev;
1238     SysBusDevice *sysbus;
1239     const MemMapEntry *memmap = virt_memmap;
1240     int i;
1241     MemoryRegion *sysmem = get_system_memory();
1242 
1243     dev = qdev_new(TYPE_PLATFORM_BUS_DEVICE);
1244     dev->id = g_strdup(TYPE_PLATFORM_BUS_DEVICE);
1245     qdev_prop_set_uint32(dev, "num_irqs", VIRT_PLATFORM_BUS_NUM_IRQS);
1246     qdev_prop_set_uint32(dev, "mmio_size", memmap[VIRT_PLATFORM_BUS].size);
1247     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1248     s->platform_bus_dev = dev;
1249 
1250     sysbus = SYS_BUS_DEVICE(dev);
1251     for (i = 0; i < VIRT_PLATFORM_BUS_NUM_IRQS; i++) {
1252         int irq = VIRT_PLATFORM_BUS_IRQ + i;
1253         sysbus_connect_irq(sysbus, i, qdev_get_gpio_in(irqchip, irq));
1254     }
1255 
1256     memory_region_add_subregion(sysmem,
1257                                 memmap[VIRT_PLATFORM_BUS].base,
1258                                 sysbus_mmio_get_region(sysbus, 0));
1259 }
1260 
1261 static void virt_build_smbios(RISCVVirtState *s)
1262 {
1263     MachineClass *mc = MACHINE_GET_CLASS(s);
1264     MachineState *ms = MACHINE(s);
1265     uint8_t *smbios_tables, *smbios_anchor;
1266     size_t smbios_tables_len, smbios_anchor_len;
1267     struct smbios_phys_mem_area mem_array;
1268     const char *product = "QEMU Virtual Machine";
1269 
1270     if (kvm_enabled()) {
1271         product = "KVM Virtual Machine";
1272     }
1273 
1274     smbios_set_defaults("QEMU", product, mc->name, false,
1275                         true, SMBIOS_ENTRY_POINT_TYPE_64);
1276 
1277     if (riscv_is_32bit(&s->soc[0])) {
1278         smbios_set_default_processor_family(0x200);
1279     } else {
1280         smbios_set_default_processor_family(0x201);
1281     }
1282 
1283     /* build the array of physical mem area from base_memmap */
1284     mem_array.address = s->memmap[VIRT_DRAM].base;
1285     mem_array.length = ms->ram_size;
1286 
1287     smbios_get_tables(ms, &mem_array, 1,
1288                       &smbios_tables, &smbios_tables_len,
1289                       &smbios_anchor, &smbios_anchor_len,
1290                       &error_fatal);
1291 
1292     if (smbios_anchor) {
1293         fw_cfg_add_file(s->fw_cfg, "etc/smbios/smbios-tables",
1294                         smbios_tables, smbios_tables_len);
1295         fw_cfg_add_file(s->fw_cfg, "etc/smbios/smbios-anchor",
1296                         smbios_anchor, smbios_anchor_len);
1297     }
1298 }
1299 
1300 static void virt_machine_done(Notifier *notifier, void *data)
1301 {
1302     RISCVVirtState *s = container_of(notifier, RISCVVirtState,
1303                                      machine_done);
1304     const MemMapEntry *memmap = virt_memmap;
1305     MachineState *machine = MACHINE(s);
1306     target_ulong start_addr = memmap[VIRT_DRAM].base;
1307     target_ulong firmware_end_addr, kernel_start_addr;
1308     const char *firmware_name = riscv_default_firmware_name(&s->soc[0]);
1309     uint64_t fdt_load_addr;
1310     uint64_t kernel_entry = 0;
1311     BlockBackend *pflash_blk0;
1312 
1313     /*
1314      * An user provided dtb must include everything, including
1315      * dynamic sysbus devices. Our FDT needs to be finalized.
1316      */
1317     if (machine->dtb == NULL) {
1318         finalize_fdt(s);
1319     }
1320 
1321     /*
1322      * Only direct boot kernel is currently supported for KVM VM,
1323      * so the "-bios" parameter is not supported when KVM is enabled.
1324      */
1325     if (kvm_enabled()) {
1326         if (machine->firmware) {
1327             if (strcmp(machine->firmware, "none")) {
1328                 error_report("Machine mode firmware is not supported in "
1329                              "combination with KVM.");
1330                 exit(1);
1331             }
1332         } else {
1333             machine->firmware = g_strdup("none");
1334         }
1335     }
1336 
1337     firmware_end_addr = riscv_find_and_load_firmware(machine, firmware_name,
1338                                                      start_addr, NULL);
1339 
1340     pflash_blk0 = pflash_cfi01_get_blk(s->flash[0]);
1341     if (pflash_blk0) {
1342         if (machine->firmware && !strcmp(machine->firmware, "none") &&
1343             !kvm_enabled()) {
1344             /*
1345              * Pflash was supplied but bios is none and not KVM guest,
1346              * let's overwrite the address we jump to after reset to
1347              * the base of the flash.
1348              */
1349             start_addr = virt_memmap[VIRT_FLASH].base;
1350         } else {
1351             /*
1352              * Pflash was supplied but either KVM guest or bios is not none.
1353              * In this case, base of the flash would contain S-mode payload.
1354              */
1355             riscv_setup_firmware_boot(machine);
1356             kernel_entry = virt_memmap[VIRT_FLASH].base;
1357         }
1358     }
1359 
1360     if (machine->kernel_filename && !kernel_entry) {
1361         kernel_start_addr = riscv_calc_kernel_start_addr(&s->soc[0],
1362                                                          firmware_end_addr);
1363 
1364         kernel_entry = riscv_load_kernel(machine, &s->soc[0],
1365                                          kernel_start_addr, true, NULL);
1366     }
1367 
1368     fdt_load_addr = riscv_compute_fdt_addr(memmap[VIRT_DRAM].base,
1369                                            memmap[VIRT_DRAM].size,
1370                                            machine);
1371     riscv_load_fdt(fdt_load_addr, machine->fdt);
1372 
1373     /* load the reset vector */
1374     riscv_setup_rom_reset_vec(machine, &s->soc[0], start_addr,
1375                               virt_memmap[VIRT_MROM].base,
1376                               virt_memmap[VIRT_MROM].size, kernel_entry,
1377                               fdt_load_addr);
1378 
1379     /*
1380      * Only direct boot kernel is currently supported for KVM VM,
1381      * So here setup kernel start address and fdt address.
1382      * TODO:Support firmware loading and integrate to TCG start
1383      */
1384     if (kvm_enabled()) {
1385         riscv_setup_direct_kernel(kernel_entry, fdt_load_addr);
1386     }
1387 
1388     virt_build_smbios(s);
1389 
1390     if (virt_is_acpi_enabled(s)) {
1391         virt_acpi_setup(s);
1392     }
1393 }
1394 
1395 static void virt_machine_init(MachineState *machine)
1396 {
1397     const MemMapEntry *memmap = virt_memmap;
1398     RISCVVirtState *s = RISCV_VIRT_MACHINE(machine);
1399     MemoryRegion *system_memory = get_system_memory();
1400     MemoryRegion *mask_rom = g_new(MemoryRegion, 1);
1401     DeviceState *mmio_irqchip, *virtio_irqchip, *pcie_irqchip;
1402     int i, base_hartid, hart_count;
1403     int socket_count = riscv_socket_count(machine);
1404 
1405     /* Check socket count limit */
1406     if (VIRT_SOCKETS_MAX < socket_count) {
1407         error_report("number of sockets/nodes should be less than %d",
1408             VIRT_SOCKETS_MAX);
1409         exit(1);
1410     }
1411 
1412     if (!tcg_enabled() && s->have_aclint) {
1413         error_report("'aclint' is only available with TCG acceleration");
1414         exit(1);
1415     }
1416 
1417     /* Initialize sockets */
1418     mmio_irqchip = virtio_irqchip = pcie_irqchip = NULL;
1419     for (i = 0; i < socket_count; i++) {
1420         g_autofree char *soc_name = g_strdup_printf("soc%d", i);
1421 
1422         if (!riscv_socket_check_hartids(machine, i)) {
1423             error_report("discontinuous hartids in socket%d", i);
1424             exit(1);
1425         }
1426 
1427         base_hartid = riscv_socket_first_hartid(machine, i);
1428         if (base_hartid < 0) {
1429             error_report("can't find hartid base for socket%d", i);
1430             exit(1);
1431         }
1432 
1433         hart_count = riscv_socket_hart_count(machine, i);
1434         if (hart_count < 0) {
1435             error_report("can't find hart count for socket%d", i);
1436             exit(1);
1437         }
1438 
1439         object_initialize_child(OBJECT(machine), soc_name, &s->soc[i],
1440                                 TYPE_RISCV_HART_ARRAY);
1441         object_property_set_str(OBJECT(&s->soc[i]), "cpu-type",
1442                                 machine->cpu_type, &error_abort);
1443         object_property_set_int(OBJECT(&s->soc[i]), "hartid-base",
1444                                 base_hartid, &error_abort);
1445         object_property_set_int(OBJECT(&s->soc[i]), "num-harts",
1446                                 hart_count, &error_abort);
1447         sysbus_realize(SYS_BUS_DEVICE(&s->soc[i]), &error_fatal);
1448 
1449         if (tcg_enabled()) {
1450             if (s->have_aclint) {
1451                 if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
1452                     /* Per-socket ACLINT MTIMER */
1453                     riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1454                             i * RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1455                         RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1456                         base_hartid, hart_count,
1457                         RISCV_ACLINT_DEFAULT_MTIMECMP,
1458                         RISCV_ACLINT_DEFAULT_MTIME,
1459                         RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1460                 } else {
1461                     /* Per-socket ACLINT MSWI, MTIMER, and SSWI */
1462                     riscv_aclint_swi_create(memmap[VIRT_CLINT].base +
1463                             i * memmap[VIRT_CLINT].size,
1464                         base_hartid, hart_count, false);
1465                     riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1466                             i * memmap[VIRT_CLINT].size +
1467                             RISCV_ACLINT_SWI_SIZE,
1468                         RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1469                         base_hartid, hart_count,
1470                         RISCV_ACLINT_DEFAULT_MTIMECMP,
1471                         RISCV_ACLINT_DEFAULT_MTIME,
1472                         RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1473                     riscv_aclint_swi_create(memmap[VIRT_ACLINT_SSWI].base +
1474                             i * memmap[VIRT_ACLINT_SSWI].size,
1475                         base_hartid, hart_count, true);
1476                 }
1477             } else {
1478                 /* Per-socket SiFive CLINT */
1479                 riscv_aclint_swi_create(
1480                     memmap[VIRT_CLINT].base + i * memmap[VIRT_CLINT].size,
1481                     base_hartid, hart_count, false);
1482                 riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1483                         i * memmap[VIRT_CLINT].size + RISCV_ACLINT_SWI_SIZE,
1484                     RISCV_ACLINT_DEFAULT_MTIMER_SIZE, base_hartid, hart_count,
1485                     RISCV_ACLINT_DEFAULT_MTIMECMP, RISCV_ACLINT_DEFAULT_MTIME,
1486                     RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1487             }
1488         }
1489 
1490         /* Per-socket interrupt controller */
1491         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
1492             s->irqchip[i] = virt_create_plic(memmap, i,
1493                                              base_hartid, hart_count);
1494         } else {
1495             s->irqchip[i] = virt_create_aia(s->aia_type, s->aia_guests,
1496                                             memmap, i, base_hartid,
1497                                             hart_count);
1498         }
1499 
1500         /* Try to use different IRQCHIP instance based device type */
1501         if (i == 0) {
1502             mmio_irqchip = s->irqchip[i];
1503             virtio_irqchip = s->irqchip[i];
1504             pcie_irqchip = s->irqchip[i];
1505         }
1506         if (i == 1) {
1507             virtio_irqchip = s->irqchip[i];
1508             pcie_irqchip = s->irqchip[i];
1509         }
1510         if (i == 2) {
1511             pcie_irqchip = s->irqchip[i];
1512         }
1513     }
1514 
1515     if (kvm_enabled() && virt_use_kvm_aia(s)) {
1516         kvm_riscv_aia_create(machine, IMSIC_MMIO_GROUP_MIN_SHIFT,
1517                              VIRT_IRQCHIP_NUM_SOURCES, VIRT_IRQCHIP_NUM_MSIS,
1518                              memmap[VIRT_APLIC_S].base,
1519                              memmap[VIRT_IMSIC_S].base,
1520                              s->aia_guests);
1521     }
1522 
1523     if (riscv_is_32bit(&s->soc[0])) {
1524 #if HOST_LONG_BITS == 64
1525         /* limit RAM size in a 32-bit system */
1526         if (machine->ram_size > 10 * GiB) {
1527             machine->ram_size = 10 * GiB;
1528             error_report("Limiting RAM size to 10 GiB");
1529         }
1530 #endif
1531         virt_high_pcie_memmap.base = VIRT32_HIGH_PCIE_MMIO_BASE;
1532         virt_high_pcie_memmap.size = VIRT32_HIGH_PCIE_MMIO_SIZE;
1533     } else {
1534         virt_high_pcie_memmap.size = VIRT64_HIGH_PCIE_MMIO_SIZE;
1535         virt_high_pcie_memmap.base = memmap[VIRT_DRAM].base + machine->ram_size;
1536         virt_high_pcie_memmap.base =
1537             ROUND_UP(virt_high_pcie_memmap.base, virt_high_pcie_memmap.size);
1538     }
1539 
1540     s->memmap = virt_memmap;
1541 
1542     /* register system main memory (actual RAM) */
1543     memory_region_add_subregion(system_memory, memmap[VIRT_DRAM].base,
1544         machine->ram);
1545 
1546     /* boot rom */
1547     memory_region_init_rom(mask_rom, NULL, "riscv_virt_board.mrom",
1548                            memmap[VIRT_MROM].size, &error_fatal);
1549     memory_region_add_subregion(system_memory, memmap[VIRT_MROM].base,
1550                                 mask_rom);
1551 
1552     /*
1553      * Init fw_cfg. Must be done before riscv_load_fdt, otherwise the
1554      * device tree cannot be altered and we get FDT_ERR_NOSPACE.
1555      */
1556     s->fw_cfg = create_fw_cfg(machine);
1557     rom_set_fw(s->fw_cfg);
1558 
1559     /* SiFive Test MMIO device */
1560     sifive_test_create(memmap[VIRT_TEST].base);
1561 
1562     /* VirtIO MMIO devices */
1563     for (i = 0; i < VIRTIO_COUNT; i++) {
1564         sysbus_create_simple("virtio-mmio",
1565             memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
1566             qdev_get_gpio_in(virtio_irqchip, VIRTIO_IRQ + i));
1567     }
1568 
1569     gpex_pcie_init(system_memory, pcie_irqchip, s);
1570 
1571     create_platform_bus(s, mmio_irqchip);
1572 
1573     serial_mm_init(system_memory, memmap[VIRT_UART0].base,
1574         0, qdev_get_gpio_in(mmio_irqchip, UART0_IRQ), 399193,
1575         serial_hd(0), DEVICE_LITTLE_ENDIAN);
1576 
1577     sysbus_create_simple("goldfish_rtc", memmap[VIRT_RTC].base,
1578         qdev_get_gpio_in(mmio_irqchip, RTC_IRQ));
1579 
1580     for (i = 0; i < ARRAY_SIZE(s->flash); i++) {
1581         /* Map legacy -drive if=pflash to machine properties */
1582         pflash_cfi01_legacy_drive(s->flash[i],
1583                                   drive_get(IF_PFLASH, 0, i));
1584     }
1585     virt_flash_map(s, system_memory);
1586 
1587     /* load/create device tree */
1588     if (machine->dtb) {
1589         machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
1590         if (!machine->fdt) {
1591             error_report("load_device_tree() failed");
1592             exit(1);
1593         }
1594     } else {
1595         create_fdt(s, memmap);
1596     }
1597 
1598     s->machine_done.notify = virt_machine_done;
1599     qemu_add_machine_init_done_notifier(&s->machine_done);
1600 }
1601 
1602 static void virt_machine_instance_init(Object *obj)
1603 {
1604     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1605 
1606     virt_flash_create(s);
1607 
1608     s->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
1609     s->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
1610     s->acpi = ON_OFF_AUTO_AUTO;
1611 }
1612 
1613 static char *virt_get_aia_guests(Object *obj, Error **errp)
1614 {
1615     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1616     char val[32];
1617 
1618     sprintf(val, "%d", s->aia_guests);
1619     return g_strdup(val);
1620 }
1621 
1622 static void virt_set_aia_guests(Object *obj, const char *val, Error **errp)
1623 {
1624     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1625 
1626     s->aia_guests = atoi(val);
1627     if (s->aia_guests < 0 || s->aia_guests > VIRT_IRQCHIP_MAX_GUESTS) {
1628         error_setg(errp, "Invalid number of AIA IMSIC guests");
1629         error_append_hint(errp, "Valid values be between 0 and %d.\n",
1630                           VIRT_IRQCHIP_MAX_GUESTS);
1631     }
1632 }
1633 
1634 static char *virt_get_aia(Object *obj, Error **errp)
1635 {
1636     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1637     const char *val;
1638 
1639     switch (s->aia_type) {
1640     case VIRT_AIA_TYPE_APLIC:
1641         val = "aplic";
1642         break;
1643     case VIRT_AIA_TYPE_APLIC_IMSIC:
1644         val = "aplic-imsic";
1645         break;
1646     default:
1647         val = "none";
1648         break;
1649     };
1650 
1651     return g_strdup(val);
1652 }
1653 
1654 static void virt_set_aia(Object *obj, const char *val, Error **errp)
1655 {
1656     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1657 
1658     if (!strcmp(val, "none")) {
1659         s->aia_type = VIRT_AIA_TYPE_NONE;
1660     } else if (!strcmp(val, "aplic")) {
1661         s->aia_type = VIRT_AIA_TYPE_APLIC;
1662     } else if (!strcmp(val, "aplic-imsic")) {
1663         s->aia_type = VIRT_AIA_TYPE_APLIC_IMSIC;
1664     } else {
1665         error_setg(errp, "Invalid AIA interrupt controller type");
1666         error_append_hint(errp, "Valid values are none, aplic, and "
1667                           "aplic-imsic.\n");
1668     }
1669 }
1670 
1671 static bool virt_get_aclint(Object *obj, Error **errp)
1672 {
1673     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1674 
1675     return s->have_aclint;
1676 }
1677 
1678 static void virt_set_aclint(Object *obj, bool value, Error **errp)
1679 {
1680     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1681 
1682     s->have_aclint = value;
1683 }
1684 
1685 bool virt_is_acpi_enabled(RISCVVirtState *s)
1686 {
1687     return s->acpi != ON_OFF_AUTO_OFF;
1688 }
1689 
1690 static void virt_get_acpi(Object *obj, Visitor *v, const char *name,
1691                           void *opaque, Error **errp)
1692 {
1693     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1694     OnOffAuto acpi = s->acpi;
1695 
1696     visit_type_OnOffAuto(v, name, &acpi, errp);
1697 }
1698 
1699 static void virt_set_acpi(Object *obj, Visitor *v, const char *name,
1700                           void *opaque, Error **errp)
1701 {
1702     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1703 
1704     visit_type_OnOffAuto(v, name, &s->acpi, errp);
1705 }
1706 
1707 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
1708                                                         DeviceState *dev)
1709 {
1710     MachineClass *mc = MACHINE_GET_CLASS(machine);
1711 
1712     if (device_is_dynamic_sysbus(mc, dev) ||
1713         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
1714         return HOTPLUG_HANDLER(machine);
1715     }
1716     return NULL;
1717 }
1718 
1719 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
1720                                         DeviceState *dev, Error **errp)
1721 {
1722     RISCVVirtState *s = RISCV_VIRT_MACHINE(hotplug_dev);
1723 
1724     if (s->platform_bus_dev) {
1725         MachineClass *mc = MACHINE_GET_CLASS(s);
1726 
1727         if (device_is_dynamic_sysbus(mc, dev)) {
1728             platform_bus_link_device(PLATFORM_BUS_DEVICE(s->platform_bus_dev),
1729                                      SYS_BUS_DEVICE(dev));
1730         }
1731     }
1732 
1733     if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
1734         create_fdt_virtio_iommu(s, pci_get_bdf(PCI_DEVICE(dev)));
1735     }
1736 }
1737 
1738 static void virt_machine_class_init(ObjectClass *oc, void *data)
1739 {
1740     char str[128];
1741     MachineClass *mc = MACHINE_CLASS(oc);
1742     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
1743 
1744     mc->desc = "RISC-V VirtIO board";
1745     mc->init = virt_machine_init;
1746     mc->max_cpus = VIRT_CPUS_MAX;
1747     mc->default_cpu_type = TYPE_RISCV_CPU_BASE;
1748     mc->pci_allow_0_address = true;
1749     mc->possible_cpu_arch_ids = riscv_numa_possible_cpu_arch_ids;
1750     mc->cpu_index_to_instance_props = riscv_numa_cpu_index_to_props;
1751     mc->get_default_cpu_node_id = riscv_numa_get_default_cpu_node_id;
1752     mc->numa_mem_supported = true;
1753     /* platform instead of architectural choice */
1754     mc->cpu_cluster_has_numa_boundary = true;
1755     mc->default_ram_id = "riscv_virt_board.ram";
1756     assert(!mc->get_hotplug_handler);
1757     mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
1758 
1759     hc->plug = virt_machine_device_plug_cb;
1760 
1761     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
1762 #ifdef CONFIG_TPM
1763     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_TPM_TIS_SYSBUS);
1764 #endif
1765 
1766 
1767     object_class_property_add_bool(oc, "aclint", virt_get_aclint,
1768                                    virt_set_aclint);
1769     object_class_property_set_description(oc, "aclint",
1770                                           "(TCG only) Set on/off to "
1771                                           "enable/disable emulating "
1772                                           "ACLINT devices");
1773 
1774     object_class_property_add_str(oc, "aia", virt_get_aia,
1775                                   virt_set_aia);
1776     object_class_property_set_description(oc, "aia",
1777                                           "Set type of AIA interrupt "
1778                                           "controller. Valid values are "
1779                                           "none, aplic, and aplic-imsic.");
1780 
1781     object_class_property_add_str(oc, "aia-guests",
1782                                   virt_get_aia_guests,
1783                                   virt_set_aia_guests);
1784     sprintf(str, "Set number of guest MMIO pages for AIA IMSIC. Valid value "
1785                  "should be between 0 and %d.", VIRT_IRQCHIP_MAX_GUESTS);
1786     object_class_property_set_description(oc, "aia-guests", str);
1787     object_class_property_add(oc, "acpi", "OnOffAuto",
1788                               virt_get_acpi, virt_set_acpi,
1789                               NULL, NULL);
1790     object_class_property_set_description(oc, "acpi",
1791                                           "Enable ACPI");
1792 }
1793 
1794 static const TypeInfo virt_machine_typeinfo = {
1795     .name       = MACHINE_TYPE_NAME("virt"),
1796     .parent     = TYPE_MACHINE,
1797     .class_init = virt_machine_class_init,
1798     .instance_init = virt_machine_instance_init,
1799     .instance_size = sizeof(RISCVVirtState),
1800     .interfaces = (InterfaceInfo[]) {
1801          { TYPE_HOTPLUG_HANDLER },
1802          { }
1803     },
1804 };
1805 
1806 static void virt_machine_init_register_types(void)
1807 {
1808     type_register_static(&virt_machine_typeinfo);
1809 }
1810 
1811 type_init(virt_machine_init_register_types)
1812