xref: /openbmc/qemu/hw/riscv/virt.c (revision 9746e583fe6ca67d9645448989535bc19adb6150)
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 "qapi/error.h"
25 #include "hw/boards.h"
26 #include "hw/loader.h"
27 #include "hw/sysbus.h"
28 #include "hw/qdev-properties.h"
29 #include "hw/char/serial.h"
30 #include "target/riscv/cpu.h"
31 #include "hw/riscv/riscv_hart.h"
32 #include "hw/riscv/virt.h"
33 #include "hw/riscv/boot.h"
34 #include "hw/riscv/numa.h"
35 #include "hw/intc/riscv_aclint.h"
36 #include "hw/intc/riscv_aplic.h"
37 #include "hw/intc/sifive_plic.h"
38 #include "hw/misc/sifive_test.h"
39 #include "chardev/char.h"
40 #include "sysemu/device_tree.h"
41 #include "sysemu/sysemu.h"
42 #include "sysemu/kvm.h"
43 #include "hw/pci/pci.h"
44 #include "hw/pci-host/gpex.h"
45 #include "hw/display/ramfb.h"
46 
47 static const MemMapEntry virt_memmap[] = {
48     [VIRT_DEBUG] =       {        0x0,         0x100 },
49     [VIRT_MROM] =        {     0x1000,        0xf000 },
50     [VIRT_TEST] =        {   0x100000,        0x1000 },
51     [VIRT_RTC] =         {   0x101000,        0x1000 },
52     [VIRT_CLINT] =       {  0x2000000,       0x10000 },
53     [VIRT_ACLINT_SSWI] = {  0x2F00000,        0x4000 },
54     [VIRT_PCIE_PIO] =    {  0x3000000,       0x10000 },
55     [VIRT_PLIC] =        {  0xc000000, VIRT_PLIC_SIZE(VIRT_CPUS_MAX * 2) },
56     [VIRT_APLIC_M] =     {  0xc000000, APLIC_SIZE(VIRT_CPUS_MAX) },
57     [VIRT_APLIC_S] =     {  0xd000000, APLIC_SIZE(VIRT_CPUS_MAX) },
58     [VIRT_UART0] =       { 0x10000000,         0x100 },
59     [VIRT_VIRTIO] =      { 0x10001000,        0x1000 },
60     [VIRT_FW_CFG] =      { 0x10100000,          0x18 },
61     [VIRT_FLASH] =       { 0x20000000,     0x4000000 },
62     [VIRT_PCIE_ECAM] =   { 0x30000000,    0x10000000 },
63     [VIRT_PCIE_MMIO] =   { 0x40000000,    0x40000000 },
64     [VIRT_DRAM] =        { 0x80000000,           0x0 },
65 };
66 
67 /* PCIe high mmio is fixed for RV32 */
68 #define VIRT32_HIGH_PCIE_MMIO_BASE  0x300000000ULL
69 #define VIRT32_HIGH_PCIE_MMIO_SIZE  (4 * GiB)
70 
71 /* PCIe high mmio for RV64, size is fixed but base depends on top of RAM */
72 #define VIRT64_HIGH_PCIE_MMIO_SIZE  (16 * GiB)
73 
74 static MemMapEntry virt_high_pcie_memmap;
75 
76 #define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
77 
78 static PFlashCFI01 *virt_flash_create1(RISCVVirtState *s,
79                                        const char *name,
80                                        const char *alias_prop_name)
81 {
82     /*
83      * Create a single flash device.  We use the same parameters as
84      * the flash devices on the ARM virt board.
85      */
86     DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
87 
88     qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
89     qdev_prop_set_uint8(dev, "width", 4);
90     qdev_prop_set_uint8(dev, "device-width", 2);
91     qdev_prop_set_bit(dev, "big-endian", false);
92     qdev_prop_set_uint16(dev, "id0", 0x89);
93     qdev_prop_set_uint16(dev, "id1", 0x18);
94     qdev_prop_set_uint16(dev, "id2", 0x00);
95     qdev_prop_set_uint16(dev, "id3", 0x00);
96     qdev_prop_set_string(dev, "name", name);
97 
98     object_property_add_child(OBJECT(s), name, OBJECT(dev));
99     object_property_add_alias(OBJECT(s), alias_prop_name,
100                               OBJECT(dev), "drive");
101 
102     return PFLASH_CFI01(dev);
103 }
104 
105 static void virt_flash_create(RISCVVirtState *s)
106 {
107     s->flash[0] = virt_flash_create1(s, "virt.flash0", "pflash0");
108     s->flash[1] = virt_flash_create1(s, "virt.flash1", "pflash1");
109 }
110 
111 static void virt_flash_map1(PFlashCFI01 *flash,
112                             hwaddr base, hwaddr size,
113                             MemoryRegion *sysmem)
114 {
115     DeviceState *dev = DEVICE(flash);
116 
117     assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
118     assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
119     qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
120     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
121 
122     memory_region_add_subregion(sysmem, base,
123                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
124                                                        0));
125 }
126 
127 static void virt_flash_map(RISCVVirtState *s,
128                            MemoryRegion *sysmem)
129 {
130     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
131     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
132 
133     virt_flash_map1(s->flash[0], flashbase, flashsize,
134                     sysmem);
135     virt_flash_map1(s->flash[1], flashbase + flashsize, flashsize,
136                     sysmem);
137 }
138 
139 static void create_pcie_irq_map(RISCVVirtState *s, void *fdt, char *nodename,
140                                 uint32_t irqchip_phandle)
141 {
142     int pin, dev;
143     uint32_t irq_map_stride = 0;
144     uint32_t full_irq_map[GPEX_NUM_IRQS * GPEX_NUM_IRQS *
145                           FDT_MAX_INT_MAP_WIDTH] = {};
146     uint32_t *irq_map = full_irq_map;
147 
148     /* This code creates a standard swizzle of interrupts such that
149      * each device's first interrupt is based on it's PCI_SLOT number.
150      * (See pci_swizzle_map_irq_fn())
151      *
152      * We only need one entry per interrupt in the table (not one per
153      * possible slot) seeing the interrupt-map-mask will allow the table
154      * to wrap to any number of devices.
155      */
156     for (dev = 0; dev < GPEX_NUM_IRQS; dev++) {
157         int devfn = dev * 0x8;
158 
159         for (pin = 0; pin < GPEX_NUM_IRQS; pin++) {
160             int irq_nr = PCIE_IRQ + ((pin + PCI_SLOT(devfn)) % GPEX_NUM_IRQS);
161             int i = 0;
162 
163             /* Fill PCI address cells */
164             irq_map[i] = cpu_to_be32(devfn << 8);
165             i += FDT_PCI_ADDR_CELLS;
166 
167             /* Fill PCI Interrupt cells */
168             irq_map[i] = cpu_to_be32(pin + 1);
169             i += FDT_PCI_INT_CELLS;
170 
171             /* Fill interrupt controller phandle and cells */
172             irq_map[i++] = cpu_to_be32(irqchip_phandle);
173             irq_map[i++] = cpu_to_be32(irq_nr);
174             if (s->aia_type != VIRT_AIA_TYPE_NONE) {
175                 irq_map[i++] = cpu_to_be32(0x4);
176             }
177 
178             if (!irq_map_stride) {
179                 irq_map_stride = i;
180             }
181             irq_map += irq_map_stride;
182         }
183     }
184 
185     qemu_fdt_setprop(fdt, nodename, "interrupt-map", full_irq_map,
186                      GPEX_NUM_IRQS * GPEX_NUM_IRQS *
187                      irq_map_stride * sizeof(uint32_t));
188 
189     qemu_fdt_setprop_cells(fdt, nodename, "interrupt-map-mask",
190                            0x1800, 0, 0, 0x7);
191 }
192 
193 static void create_fdt_socket_cpus(RISCVVirtState *s, int socket,
194                                    char *clust_name, uint32_t *phandle,
195                                    bool is_32_bit, uint32_t *intc_phandles)
196 {
197     int cpu;
198     uint32_t cpu_phandle;
199     MachineState *mc = MACHINE(s);
200     char *name, *cpu_name, *core_name, *intc_name;
201 
202     for (cpu = s->soc[socket].num_harts - 1; cpu >= 0; cpu--) {
203         cpu_phandle = (*phandle)++;
204 
205         cpu_name = g_strdup_printf("/cpus/cpu@%d",
206             s->soc[socket].hartid_base + cpu);
207         qemu_fdt_add_subnode(mc->fdt, cpu_name);
208         qemu_fdt_setprop_string(mc->fdt, cpu_name, "mmu-type",
209             (is_32_bit) ? "riscv,sv32" : "riscv,sv48");
210         name = riscv_isa_string(&s->soc[socket].harts[cpu]);
211         qemu_fdt_setprop_string(mc->fdt, cpu_name, "riscv,isa", name);
212         g_free(name);
213         qemu_fdt_setprop_string(mc->fdt, cpu_name, "compatible", "riscv");
214         qemu_fdt_setprop_string(mc->fdt, cpu_name, "status", "okay");
215         qemu_fdt_setprop_cell(mc->fdt, cpu_name, "reg",
216             s->soc[socket].hartid_base + cpu);
217         qemu_fdt_setprop_string(mc->fdt, cpu_name, "device_type", "cpu");
218         riscv_socket_fdt_write_id(mc, mc->fdt, cpu_name, socket);
219         qemu_fdt_setprop_cell(mc->fdt, cpu_name, "phandle", cpu_phandle);
220 
221         intc_phandles[cpu] = (*phandle)++;
222 
223         intc_name = g_strdup_printf("%s/interrupt-controller", cpu_name);
224         qemu_fdt_add_subnode(mc->fdt, intc_name);
225         qemu_fdt_setprop_cell(mc->fdt, intc_name, "phandle",
226             intc_phandles[cpu]);
227         if (riscv_feature(&s->soc[socket].harts[cpu].env,
228                           RISCV_FEATURE_AIA)) {
229             static const char * const compat[2] = {
230                 "riscv,cpu-intc-aia", "riscv,cpu-intc"
231             };
232             qemu_fdt_setprop_string_array(mc->fdt, intc_name, "compatible",
233                                       (char **)&compat, ARRAY_SIZE(compat));
234         } else {
235             qemu_fdt_setprop_string(mc->fdt, intc_name, "compatible",
236                 "riscv,cpu-intc");
237         }
238         qemu_fdt_setprop(mc->fdt, intc_name, "interrupt-controller", NULL, 0);
239         qemu_fdt_setprop_cell(mc->fdt, intc_name, "#interrupt-cells", 1);
240 
241         core_name = g_strdup_printf("%s/core%d", clust_name, cpu);
242         qemu_fdt_add_subnode(mc->fdt, core_name);
243         qemu_fdt_setprop_cell(mc->fdt, core_name, "cpu", cpu_phandle);
244 
245         g_free(core_name);
246         g_free(intc_name);
247         g_free(cpu_name);
248     }
249 }
250 
251 static void create_fdt_socket_memory(RISCVVirtState *s,
252                                      const MemMapEntry *memmap, int socket)
253 {
254     char *mem_name;
255     uint64_t addr, size;
256     MachineState *mc = MACHINE(s);
257 
258     addr = memmap[VIRT_DRAM].base + riscv_socket_mem_offset(mc, socket);
259     size = riscv_socket_mem_size(mc, socket);
260     mem_name = g_strdup_printf("/memory@%lx", (long)addr);
261     qemu_fdt_add_subnode(mc->fdt, mem_name);
262     qemu_fdt_setprop_cells(mc->fdt, mem_name, "reg",
263         addr >> 32, addr, size >> 32, size);
264     qemu_fdt_setprop_string(mc->fdt, mem_name, "device_type", "memory");
265     riscv_socket_fdt_write_id(mc, mc->fdt, mem_name, socket);
266     g_free(mem_name);
267 }
268 
269 static void create_fdt_socket_clint(RISCVVirtState *s,
270                                     const MemMapEntry *memmap, int socket,
271                                     uint32_t *intc_phandles)
272 {
273     int cpu;
274     char *clint_name;
275     uint32_t *clint_cells;
276     unsigned long clint_addr;
277     MachineState *mc = MACHINE(s);
278     static const char * const clint_compat[2] = {
279         "sifive,clint0", "riscv,clint0"
280     };
281 
282     clint_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
283 
284     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
285         clint_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
286         clint_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_SOFT);
287         clint_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
288         clint_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_M_TIMER);
289     }
290 
291     clint_addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
292     clint_name = g_strdup_printf("/soc/clint@%lx", clint_addr);
293     qemu_fdt_add_subnode(mc->fdt, clint_name);
294     qemu_fdt_setprop_string_array(mc->fdt, clint_name, "compatible",
295                                   (char **)&clint_compat,
296                                   ARRAY_SIZE(clint_compat));
297     qemu_fdt_setprop_cells(mc->fdt, clint_name, "reg",
298         0x0, clint_addr, 0x0, memmap[VIRT_CLINT].size);
299     qemu_fdt_setprop(mc->fdt, clint_name, "interrupts-extended",
300         clint_cells, s->soc[socket].num_harts * sizeof(uint32_t) * 4);
301     riscv_socket_fdt_write_id(mc, mc->fdt, clint_name, socket);
302     g_free(clint_name);
303 
304     g_free(clint_cells);
305 }
306 
307 static void create_fdt_socket_aclint(RISCVVirtState *s,
308                                      const MemMapEntry *memmap, int socket,
309                                      uint32_t *intc_phandles)
310 {
311     int cpu;
312     char *name;
313     unsigned long addr;
314     uint32_t aclint_cells_size;
315     uint32_t *aclint_mswi_cells;
316     uint32_t *aclint_sswi_cells;
317     uint32_t *aclint_mtimer_cells;
318     MachineState *mc = MACHINE(s);
319 
320     aclint_mswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
321     aclint_mtimer_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
322     aclint_sswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
323 
324     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
325         aclint_mswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
326         aclint_mswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_SOFT);
327         aclint_mtimer_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
328         aclint_mtimer_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_TIMER);
329         aclint_sswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
330         aclint_sswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_SOFT);
331     }
332     aclint_cells_size = s->soc[socket].num_harts * sizeof(uint32_t) * 2;
333 
334     addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
335     name = g_strdup_printf("/soc/mswi@%lx", addr);
336     qemu_fdt_add_subnode(mc->fdt, name);
337     qemu_fdt_setprop_string(mc->fdt, name, "compatible", "riscv,aclint-mswi");
338     qemu_fdt_setprop_cells(mc->fdt, name, "reg",
339         0x0, addr, 0x0, RISCV_ACLINT_SWI_SIZE);
340     qemu_fdt_setprop(mc->fdt, name, "interrupts-extended",
341         aclint_mswi_cells, aclint_cells_size);
342     qemu_fdt_setprop(mc->fdt, name, "interrupt-controller", NULL, 0);
343     qemu_fdt_setprop_cell(mc->fdt, name, "#interrupt-cells", 0);
344     riscv_socket_fdt_write_id(mc, mc->fdt, name, socket);
345     g_free(name);
346 
347     addr = memmap[VIRT_CLINT].base + RISCV_ACLINT_SWI_SIZE +
348         (memmap[VIRT_CLINT].size * socket);
349     name = g_strdup_printf("/soc/mtimer@%lx", addr);
350     qemu_fdt_add_subnode(mc->fdt, name);
351     qemu_fdt_setprop_string(mc->fdt, name, "compatible",
352         "riscv,aclint-mtimer");
353     qemu_fdt_setprop_cells(mc->fdt, name, "reg",
354         0x0, addr + RISCV_ACLINT_DEFAULT_MTIME,
355         0x0, memmap[VIRT_CLINT].size - RISCV_ACLINT_SWI_SIZE -
356              RISCV_ACLINT_DEFAULT_MTIME,
357         0x0, addr + RISCV_ACLINT_DEFAULT_MTIMECMP,
358         0x0, RISCV_ACLINT_DEFAULT_MTIME);
359     qemu_fdt_setprop(mc->fdt, name, "interrupts-extended",
360         aclint_mtimer_cells, aclint_cells_size);
361     riscv_socket_fdt_write_id(mc, mc->fdt, name, socket);
362     g_free(name);
363 
364     addr = memmap[VIRT_ACLINT_SSWI].base +
365         (memmap[VIRT_ACLINT_SSWI].size * socket);
366     name = g_strdup_printf("/soc/sswi@%lx", addr);
367     qemu_fdt_add_subnode(mc->fdt, name);
368     qemu_fdt_setprop_string(mc->fdt, name, "compatible", "riscv,aclint-sswi");
369     qemu_fdt_setprop_cells(mc->fdt, name, "reg",
370         0x0, addr, 0x0, memmap[VIRT_ACLINT_SSWI].size);
371     qemu_fdt_setprop(mc->fdt, name, "interrupts-extended",
372         aclint_sswi_cells, aclint_cells_size);
373     qemu_fdt_setprop(mc->fdt, name, "interrupt-controller", NULL, 0);
374     qemu_fdt_setprop_cell(mc->fdt, name, "#interrupt-cells", 0);
375     riscv_socket_fdt_write_id(mc, mc->fdt, name, socket);
376     g_free(name);
377 
378     g_free(aclint_mswi_cells);
379     g_free(aclint_mtimer_cells);
380     g_free(aclint_sswi_cells);
381 }
382 
383 static void create_fdt_socket_plic(RISCVVirtState *s,
384                                    const MemMapEntry *memmap, int socket,
385                                    uint32_t *phandle, uint32_t *intc_phandles,
386                                    uint32_t *plic_phandles)
387 {
388     int cpu;
389     char *plic_name;
390     uint32_t *plic_cells;
391     unsigned long plic_addr;
392     MachineState *mc = MACHINE(s);
393     static const char * const plic_compat[2] = {
394         "sifive,plic-1.0.0", "riscv,plic0"
395     };
396 
397     if (kvm_enabled()) {
398         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
399     } else {
400         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
401     }
402 
403     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
404         if (kvm_enabled()) {
405             plic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
406             plic_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_EXT);
407         } else {
408             plic_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
409             plic_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_EXT);
410             plic_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
411             plic_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_S_EXT);
412         }
413     }
414 
415     plic_phandles[socket] = (*phandle)++;
416     plic_addr = memmap[VIRT_PLIC].base + (memmap[VIRT_PLIC].size * socket);
417     plic_name = g_strdup_printf("/soc/plic@%lx", plic_addr);
418     qemu_fdt_add_subnode(mc->fdt, plic_name);
419     qemu_fdt_setprop_cell(mc->fdt, plic_name,
420         "#interrupt-cells", FDT_PLIC_INT_CELLS);
421     qemu_fdt_setprop_string_array(mc->fdt, plic_name, "compatible",
422                                   (char **)&plic_compat,
423                                   ARRAY_SIZE(plic_compat));
424     qemu_fdt_setprop(mc->fdt, plic_name, "interrupt-controller", NULL, 0);
425     qemu_fdt_setprop(mc->fdt, plic_name, "interrupts-extended",
426         plic_cells, s->soc[socket].num_harts * sizeof(uint32_t) * 4);
427     qemu_fdt_setprop_cells(mc->fdt, plic_name, "reg",
428         0x0, plic_addr, 0x0, memmap[VIRT_PLIC].size);
429     qemu_fdt_setprop_cell(mc->fdt, plic_name, "riscv,ndev", VIRTIO_NDEV);
430     riscv_socket_fdt_write_id(mc, mc->fdt, plic_name, socket);
431     qemu_fdt_setprop_cell(mc->fdt, plic_name, "phandle",
432         plic_phandles[socket]);
433     g_free(plic_name);
434 
435     g_free(plic_cells);
436 }
437 
438 static void create_fdt_socket_aia(RISCVVirtState *s,
439                                   const MemMapEntry *memmap, int socket,
440                                   uint32_t *phandle, uint32_t *intc_phandles,
441                                   uint32_t *aplic_phandles)
442 {
443     int cpu;
444     char *aplic_name;
445     uint32_t *aplic_cells;
446     unsigned long aplic_addr;
447     MachineState *mc = MACHINE(s);
448     uint32_t aplic_m_phandle, aplic_s_phandle;
449 
450     aplic_m_phandle = (*phandle)++;
451     aplic_s_phandle = (*phandle)++;
452     aplic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
453 
454     /* M-level APLIC node */
455     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
456         aplic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
457         aplic_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_EXT);
458     }
459     aplic_addr = memmap[VIRT_APLIC_M].base +
460                  (memmap[VIRT_APLIC_M].size * socket);
461     aplic_name = g_strdup_printf("/soc/aplic@%lx", aplic_addr);
462     qemu_fdt_add_subnode(mc->fdt, aplic_name);
463     qemu_fdt_setprop_string(mc->fdt, aplic_name, "compatible", "riscv,aplic");
464     qemu_fdt_setprop_cell(mc->fdt, aplic_name,
465         "#interrupt-cells", FDT_APLIC_INT_CELLS);
466     qemu_fdt_setprop(mc->fdt, aplic_name, "interrupt-controller", NULL, 0);
467     qemu_fdt_setprop(mc->fdt, aplic_name, "interrupts-extended",
468         aplic_cells, s->soc[socket].num_harts * sizeof(uint32_t) * 2);
469     qemu_fdt_setprop_cells(mc->fdt, aplic_name, "reg",
470         0x0, aplic_addr, 0x0, memmap[VIRT_APLIC_M].size);
471     qemu_fdt_setprop_cell(mc->fdt, aplic_name, "riscv,num-sources",
472         VIRT_IRQCHIP_NUM_SOURCES);
473     qemu_fdt_setprop_cell(mc->fdt, aplic_name, "riscv,children",
474         aplic_s_phandle);
475     qemu_fdt_setprop_cells(mc->fdt, aplic_name, "riscv,delegate",
476         aplic_s_phandle, 0x1, VIRT_IRQCHIP_NUM_SOURCES);
477     riscv_socket_fdt_write_id(mc, mc->fdt, aplic_name, socket);
478     qemu_fdt_setprop_cell(mc->fdt, aplic_name, "phandle", aplic_m_phandle);
479     g_free(aplic_name);
480 
481     /* S-level APLIC node */
482     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
483         aplic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
484         aplic_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_EXT);
485     }
486     aplic_addr = memmap[VIRT_APLIC_S].base +
487                  (memmap[VIRT_APLIC_S].size * socket);
488     aplic_name = g_strdup_printf("/soc/aplic@%lx", aplic_addr);
489     qemu_fdt_add_subnode(mc->fdt, aplic_name);
490     qemu_fdt_setprop_string(mc->fdt, aplic_name, "compatible", "riscv,aplic");
491     qemu_fdt_setprop_cell(mc->fdt, aplic_name,
492         "#interrupt-cells", FDT_APLIC_INT_CELLS);
493     qemu_fdt_setprop(mc->fdt, aplic_name, "interrupt-controller", NULL, 0);
494     qemu_fdt_setprop(mc->fdt, aplic_name, "interrupts-extended",
495         aplic_cells, s->soc[socket].num_harts * sizeof(uint32_t) * 2);
496     qemu_fdt_setprop_cells(mc->fdt, aplic_name, "reg",
497         0x0, aplic_addr, 0x0, memmap[VIRT_APLIC_S].size);
498     qemu_fdt_setprop_cell(mc->fdt, aplic_name, "riscv,num-sources",
499         VIRT_IRQCHIP_NUM_SOURCES);
500     riscv_socket_fdt_write_id(mc, mc->fdt, aplic_name, socket);
501     qemu_fdt_setprop_cell(mc->fdt, aplic_name, "phandle", aplic_s_phandle);
502     g_free(aplic_name);
503 
504     g_free(aplic_cells);
505     aplic_phandles[socket] = aplic_s_phandle;
506 }
507 
508 static void create_fdt_sockets(RISCVVirtState *s, const MemMapEntry *memmap,
509                                bool is_32_bit, uint32_t *phandle,
510                                uint32_t *irq_mmio_phandle,
511                                uint32_t *irq_pcie_phandle,
512                                uint32_t *irq_virtio_phandle)
513 {
514     int socket;
515     char *clust_name;
516     uint32_t *intc_phandles;
517     MachineState *mc = MACHINE(s);
518     uint32_t xplic_phandles[MAX_NODES];
519 
520     qemu_fdt_add_subnode(mc->fdt, "/cpus");
521     qemu_fdt_setprop_cell(mc->fdt, "/cpus", "timebase-frequency",
522                           RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ);
523     qemu_fdt_setprop_cell(mc->fdt, "/cpus", "#size-cells", 0x0);
524     qemu_fdt_setprop_cell(mc->fdt, "/cpus", "#address-cells", 0x1);
525     qemu_fdt_add_subnode(mc->fdt, "/cpus/cpu-map");
526 
527     for (socket = (riscv_socket_count(mc) - 1); socket >= 0; socket--) {
528         clust_name = g_strdup_printf("/cpus/cpu-map/cluster%d", socket);
529         qemu_fdt_add_subnode(mc->fdt, clust_name);
530 
531         intc_phandles = g_new0(uint32_t, s->soc[socket].num_harts);
532 
533         create_fdt_socket_cpus(s, socket, clust_name, phandle,
534             is_32_bit, intc_phandles);
535 
536         create_fdt_socket_memory(s, memmap, socket);
537 
538         if (!kvm_enabled()) {
539             if (s->have_aclint) {
540                 create_fdt_socket_aclint(s, memmap, socket, intc_phandles);
541             } else {
542                 create_fdt_socket_clint(s, memmap, socket, intc_phandles);
543             }
544         }
545 
546         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
547             create_fdt_socket_plic(s, memmap, socket, phandle,
548                 intc_phandles, xplic_phandles);
549         } else {
550             create_fdt_socket_aia(s, memmap, socket, phandle,
551                 intc_phandles, xplic_phandles);
552         }
553 
554         g_free(intc_phandles);
555         g_free(clust_name);
556     }
557 
558     for (socket = 0; socket < riscv_socket_count(mc); socket++) {
559         if (socket == 0) {
560             *irq_mmio_phandle = xplic_phandles[socket];
561             *irq_virtio_phandle = xplic_phandles[socket];
562             *irq_pcie_phandle = xplic_phandles[socket];
563         }
564         if (socket == 1) {
565             *irq_virtio_phandle = xplic_phandles[socket];
566             *irq_pcie_phandle = xplic_phandles[socket];
567         }
568         if (socket == 2) {
569             *irq_pcie_phandle = xplic_phandles[socket];
570         }
571     }
572 
573     riscv_socket_fdt_write_distance_matrix(mc, mc->fdt);
574 }
575 
576 static void create_fdt_virtio(RISCVVirtState *s, const MemMapEntry *memmap,
577                               uint32_t irq_virtio_phandle)
578 {
579     int i;
580     char *name;
581     MachineState *mc = MACHINE(s);
582 
583     for (i = 0; i < VIRTIO_COUNT; i++) {
584         name = g_strdup_printf("/soc/virtio_mmio@%lx",
585             (long)(memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size));
586         qemu_fdt_add_subnode(mc->fdt, name);
587         qemu_fdt_setprop_string(mc->fdt, name, "compatible", "virtio,mmio");
588         qemu_fdt_setprop_cells(mc->fdt, name, "reg",
589             0x0, memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
590             0x0, memmap[VIRT_VIRTIO].size);
591         qemu_fdt_setprop_cell(mc->fdt, name, "interrupt-parent",
592             irq_virtio_phandle);
593         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
594             qemu_fdt_setprop_cell(mc->fdt, name, "interrupts",
595                                   VIRTIO_IRQ + i);
596         } else {
597             qemu_fdt_setprop_cells(mc->fdt, name, "interrupts",
598                                    VIRTIO_IRQ + i, 0x4);
599         }
600         g_free(name);
601     }
602 }
603 
604 static void create_fdt_pcie(RISCVVirtState *s, const MemMapEntry *memmap,
605                             uint32_t irq_pcie_phandle)
606 {
607     char *name;
608     MachineState *mc = MACHINE(s);
609 
610     name = g_strdup_printf("/soc/pci@%lx",
611         (long) memmap[VIRT_PCIE_ECAM].base);
612     qemu_fdt_add_subnode(mc->fdt, name);
613     qemu_fdt_setprop_cell(mc->fdt, name, "#address-cells",
614         FDT_PCI_ADDR_CELLS);
615     qemu_fdt_setprop_cell(mc->fdt, name, "#interrupt-cells",
616         FDT_PCI_INT_CELLS);
617     qemu_fdt_setprop_cell(mc->fdt, name, "#size-cells", 0x2);
618     qemu_fdt_setprop_string(mc->fdt, name, "compatible",
619         "pci-host-ecam-generic");
620     qemu_fdt_setprop_string(mc->fdt, name, "device_type", "pci");
621     qemu_fdt_setprop_cell(mc->fdt, name, "linux,pci-domain", 0);
622     qemu_fdt_setprop_cells(mc->fdt, name, "bus-range", 0,
623         memmap[VIRT_PCIE_ECAM].size / PCIE_MMCFG_SIZE_MIN - 1);
624     qemu_fdt_setprop(mc->fdt, name, "dma-coherent", NULL, 0);
625     qemu_fdt_setprop_cells(mc->fdt, name, "reg", 0,
626         memmap[VIRT_PCIE_ECAM].base, 0, memmap[VIRT_PCIE_ECAM].size);
627     qemu_fdt_setprop_sized_cells(mc->fdt, name, "ranges",
628         1, FDT_PCI_RANGE_IOPORT, 2, 0,
629         2, memmap[VIRT_PCIE_PIO].base, 2, memmap[VIRT_PCIE_PIO].size,
630         1, FDT_PCI_RANGE_MMIO,
631         2, memmap[VIRT_PCIE_MMIO].base,
632         2, memmap[VIRT_PCIE_MMIO].base, 2, memmap[VIRT_PCIE_MMIO].size,
633         1, FDT_PCI_RANGE_MMIO_64BIT,
634         2, virt_high_pcie_memmap.base,
635         2, virt_high_pcie_memmap.base, 2, virt_high_pcie_memmap.size);
636 
637     create_pcie_irq_map(s, mc->fdt, name, irq_pcie_phandle);
638     g_free(name);
639 }
640 
641 static void create_fdt_reset(RISCVVirtState *s, const MemMapEntry *memmap,
642                              uint32_t *phandle)
643 {
644     char *name;
645     uint32_t test_phandle;
646     MachineState *mc = MACHINE(s);
647 
648     test_phandle = (*phandle)++;
649     name = g_strdup_printf("/soc/test@%lx",
650         (long)memmap[VIRT_TEST].base);
651     qemu_fdt_add_subnode(mc->fdt, name);
652     {
653         static const char * const compat[3] = {
654             "sifive,test1", "sifive,test0", "syscon"
655         };
656         qemu_fdt_setprop_string_array(mc->fdt, name, "compatible",
657                                       (char **)&compat, ARRAY_SIZE(compat));
658     }
659     qemu_fdt_setprop_cells(mc->fdt, name, "reg",
660         0x0, memmap[VIRT_TEST].base, 0x0, memmap[VIRT_TEST].size);
661     qemu_fdt_setprop_cell(mc->fdt, name, "phandle", test_phandle);
662     test_phandle = qemu_fdt_get_phandle(mc->fdt, name);
663     g_free(name);
664 
665     name = g_strdup_printf("/soc/reboot");
666     qemu_fdt_add_subnode(mc->fdt, name);
667     qemu_fdt_setprop_string(mc->fdt, name, "compatible", "syscon-reboot");
668     qemu_fdt_setprop_cell(mc->fdt, name, "regmap", test_phandle);
669     qemu_fdt_setprop_cell(mc->fdt, name, "offset", 0x0);
670     qemu_fdt_setprop_cell(mc->fdt, name, "value", FINISHER_RESET);
671     g_free(name);
672 
673     name = g_strdup_printf("/soc/poweroff");
674     qemu_fdt_add_subnode(mc->fdt, name);
675     qemu_fdt_setprop_string(mc->fdt, name, "compatible", "syscon-poweroff");
676     qemu_fdt_setprop_cell(mc->fdt, name, "regmap", test_phandle);
677     qemu_fdt_setprop_cell(mc->fdt, name, "offset", 0x0);
678     qemu_fdt_setprop_cell(mc->fdt, name, "value", FINISHER_PASS);
679     g_free(name);
680 }
681 
682 static void create_fdt_uart(RISCVVirtState *s, const MemMapEntry *memmap,
683                             uint32_t irq_mmio_phandle)
684 {
685     char *name;
686     MachineState *mc = MACHINE(s);
687 
688     name = g_strdup_printf("/soc/uart@%lx", (long)memmap[VIRT_UART0].base);
689     qemu_fdt_add_subnode(mc->fdt, name);
690     qemu_fdt_setprop_string(mc->fdt, name, "compatible", "ns16550a");
691     qemu_fdt_setprop_cells(mc->fdt, name, "reg",
692         0x0, memmap[VIRT_UART0].base,
693         0x0, memmap[VIRT_UART0].size);
694     qemu_fdt_setprop_cell(mc->fdt, name, "clock-frequency", 3686400);
695     qemu_fdt_setprop_cell(mc->fdt, name, "interrupt-parent", irq_mmio_phandle);
696     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
697         qemu_fdt_setprop_cell(mc->fdt, name, "interrupts", UART0_IRQ);
698     } else {
699         qemu_fdt_setprop_cells(mc->fdt, name, "interrupts", UART0_IRQ, 0x4);
700     }
701 
702     qemu_fdt_add_subnode(mc->fdt, "/chosen");
703     qemu_fdt_setprop_string(mc->fdt, "/chosen", "stdout-path", name);
704     g_free(name);
705 }
706 
707 static void create_fdt_rtc(RISCVVirtState *s, const MemMapEntry *memmap,
708                            uint32_t irq_mmio_phandle)
709 {
710     char *name;
711     MachineState *mc = MACHINE(s);
712 
713     name = g_strdup_printf("/soc/rtc@%lx", (long)memmap[VIRT_RTC].base);
714     qemu_fdt_add_subnode(mc->fdt, name);
715     qemu_fdt_setprop_string(mc->fdt, name, "compatible",
716         "google,goldfish-rtc");
717     qemu_fdt_setprop_cells(mc->fdt, name, "reg",
718         0x0, memmap[VIRT_RTC].base, 0x0, memmap[VIRT_RTC].size);
719     qemu_fdt_setprop_cell(mc->fdt, name, "interrupt-parent",
720         irq_mmio_phandle);
721     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
722         qemu_fdt_setprop_cell(mc->fdt, name, "interrupts", RTC_IRQ);
723     } else {
724         qemu_fdt_setprop_cells(mc->fdt, name, "interrupts", RTC_IRQ, 0x4);
725     }
726     g_free(name);
727 }
728 
729 static void create_fdt_flash(RISCVVirtState *s, const MemMapEntry *memmap)
730 {
731     char *name;
732     MachineState *mc = MACHINE(s);
733     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
734     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
735 
736     name = g_strdup_printf("/flash@%" PRIx64, flashbase);
737     qemu_fdt_add_subnode(mc->fdt, name);
738     qemu_fdt_setprop_string(mc->fdt, name, "compatible", "cfi-flash");
739     qemu_fdt_setprop_sized_cells(mc->fdt, name, "reg",
740                                  2, flashbase, 2, flashsize,
741                                  2, flashbase + flashsize, 2, flashsize);
742     qemu_fdt_setprop_cell(mc->fdt, name, "bank-width", 4);
743     g_free(name);
744 }
745 
746 static void create_fdt(RISCVVirtState *s, const MemMapEntry *memmap,
747                        uint64_t mem_size, const char *cmdline, bool is_32_bit)
748 {
749     MachineState *mc = MACHINE(s);
750     uint32_t phandle = 1, irq_mmio_phandle = 1;
751     uint32_t irq_pcie_phandle = 1, irq_virtio_phandle = 1;
752 
753     if (mc->dtb) {
754         mc->fdt = load_device_tree(mc->dtb, &s->fdt_size);
755         if (!mc->fdt) {
756             error_report("load_device_tree() failed");
757             exit(1);
758         }
759         goto update_bootargs;
760     } else {
761         mc->fdt = create_device_tree(&s->fdt_size);
762         if (!mc->fdt) {
763             error_report("create_device_tree() failed");
764             exit(1);
765         }
766     }
767 
768     qemu_fdt_setprop_string(mc->fdt, "/", "model", "riscv-virtio,qemu");
769     qemu_fdt_setprop_string(mc->fdt, "/", "compatible", "riscv-virtio");
770     qemu_fdt_setprop_cell(mc->fdt, "/", "#size-cells", 0x2);
771     qemu_fdt_setprop_cell(mc->fdt, "/", "#address-cells", 0x2);
772 
773     qemu_fdt_add_subnode(mc->fdt, "/soc");
774     qemu_fdt_setprop(mc->fdt, "/soc", "ranges", NULL, 0);
775     qemu_fdt_setprop_string(mc->fdt, "/soc", "compatible", "simple-bus");
776     qemu_fdt_setprop_cell(mc->fdt, "/soc", "#size-cells", 0x2);
777     qemu_fdt_setprop_cell(mc->fdt, "/soc", "#address-cells", 0x2);
778 
779     create_fdt_sockets(s, memmap, is_32_bit, &phandle,
780         &irq_mmio_phandle, &irq_pcie_phandle, &irq_virtio_phandle);
781 
782     create_fdt_virtio(s, memmap, irq_virtio_phandle);
783 
784     create_fdt_pcie(s, memmap, irq_pcie_phandle);
785 
786     create_fdt_reset(s, memmap, &phandle);
787 
788     create_fdt_uart(s, memmap, irq_mmio_phandle);
789 
790     create_fdt_rtc(s, memmap, irq_mmio_phandle);
791 
792     create_fdt_flash(s, memmap);
793 
794 update_bootargs:
795     if (cmdline) {
796         qemu_fdt_setprop_string(mc->fdt, "/chosen", "bootargs", cmdline);
797     }
798 }
799 
800 static inline DeviceState *gpex_pcie_init(MemoryRegion *sys_mem,
801                                           hwaddr ecam_base, hwaddr ecam_size,
802                                           hwaddr mmio_base, hwaddr mmio_size,
803                                           hwaddr high_mmio_base,
804                                           hwaddr high_mmio_size,
805                                           hwaddr pio_base,
806                                           DeviceState *irqchip)
807 {
808     DeviceState *dev;
809     MemoryRegion *ecam_alias, *ecam_reg;
810     MemoryRegion *mmio_alias, *high_mmio_alias, *mmio_reg;
811     qemu_irq irq;
812     int i;
813 
814     dev = qdev_new(TYPE_GPEX_HOST);
815 
816     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
817 
818     ecam_alias = g_new0(MemoryRegion, 1);
819     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
820     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
821                              ecam_reg, 0, ecam_size);
822     memory_region_add_subregion(get_system_memory(), ecam_base, ecam_alias);
823 
824     mmio_alias = g_new0(MemoryRegion, 1);
825     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
826     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
827                              mmio_reg, mmio_base, mmio_size);
828     memory_region_add_subregion(get_system_memory(), mmio_base, mmio_alias);
829 
830     /* Map high MMIO space */
831     high_mmio_alias = g_new0(MemoryRegion, 1);
832     memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
833                              mmio_reg, high_mmio_base, high_mmio_size);
834     memory_region_add_subregion(get_system_memory(), high_mmio_base,
835                                 high_mmio_alias);
836 
837     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, pio_base);
838 
839     for (i = 0; i < GPEX_NUM_IRQS; i++) {
840         irq = qdev_get_gpio_in(irqchip, PCIE_IRQ + i);
841 
842         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, irq);
843         gpex_set_irq_num(GPEX_HOST(dev), i, PCIE_IRQ + i);
844     }
845 
846     return dev;
847 }
848 
849 static FWCfgState *create_fw_cfg(const MachineState *mc)
850 {
851     hwaddr base = virt_memmap[VIRT_FW_CFG].base;
852     hwaddr size = virt_memmap[VIRT_FW_CFG].size;
853     FWCfgState *fw_cfg;
854     char *nodename;
855 
856     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16,
857                                   &address_space_memory);
858     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)mc->smp.cpus);
859 
860     nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
861     qemu_fdt_add_subnode(mc->fdt, nodename);
862     qemu_fdt_setprop_string(mc->fdt, nodename,
863                             "compatible", "qemu,fw-cfg-mmio");
864     qemu_fdt_setprop_sized_cells(mc->fdt, nodename, "reg",
865                                  2, base, 2, size);
866     qemu_fdt_setprop(mc->fdt, nodename, "dma-coherent", NULL, 0);
867     g_free(nodename);
868     return fw_cfg;
869 }
870 
871 static DeviceState *virt_create_plic(const MemMapEntry *memmap, int socket,
872                                      int base_hartid, int hart_count)
873 {
874     DeviceState *ret;
875     char *plic_hart_config;
876 
877     /* Per-socket PLIC hart topology configuration string */
878     plic_hart_config = riscv_plic_hart_config_string(hart_count);
879 
880     /* Per-socket PLIC */
881     ret = sifive_plic_create(
882             memmap[VIRT_PLIC].base + socket * memmap[VIRT_PLIC].size,
883             plic_hart_config, hart_count, base_hartid,
884             VIRT_IRQCHIP_NUM_SOURCES,
885             ((1U << VIRT_IRQCHIP_NUM_PRIO_BITS) - 1),
886             VIRT_PLIC_PRIORITY_BASE,
887             VIRT_PLIC_PENDING_BASE,
888             VIRT_PLIC_ENABLE_BASE,
889             VIRT_PLIC_ENABLE_STRIDE,
890             VIRT_PLIC_CONTEXT_BASE,
891             VIRT_PLIC_CONTEXT_STRIDE,
892             memmap[VIRT_PLIC].size);
893 
894     g_free(plic_hart_config);
895 
896     return ret;
897 }
898 
899 static DeviceState *virt_create_aia(RISCVVirtAIAType aia_type,
900                                     const MemMapEntry *memmap, int socket,
901                                     int base_hartid, int hart_count)
902 {
903     DeviceState *aplic_m;
904 
905     /* Per-socket M-level APLIC */
906     aplic_m = riscv_aplic_create(
907         memmap[VIRT_APLIC_M].base + socket * memmap[VIRT_APLIC_M].size,
908         memmap[VIRT_APLIC_M].size,
909         base_hartid, hart_count,
910         VIRT_IRQCHIP_NUM_SOURCES,
911         VIRT_IRQCHIP_NUM_PRIO_BITS,
912         false, true, NULL);
913 
914     if (aplic_m) {
915         /* Per-socket S-level APLIC */
916         riscv_aplic_create(
917             memmap[VIRT_APLIC_S].base + socket * memmap[VIRT_APLIC_S].size,
918             memmap[VIRT_APLIC_S].size,
919             base_hartid, hart_count,
920             VIRT_IRQCHIP_NUM_SOURCES,
921             VIRT_IRQCHIP_NUM_PRIO_BITS,
922             false, false, aplic_m);
923     }
924 
925     return aplic_m;
926 }
927 
928 static void virt_machine_init(MachineState *machine)
929 {
930     const MemMapEntry *memmap = virt_memmap;
931     RISCVVirtState *s = RISCV_VIRT_MACHINE(machine);
932     MemoryRegion *system_memory = get_system_memory();
933     MemoryRegion *mask_rom = g_new(MemoryRegion, 1);
934     char *soc_name;
935     target_ulong start_addr = memmap[VIRT_DRAM].base;
936     target_ulong firmware_end_addr, kernel_start_addr;
937     uint32_t fdt_load_addr;
938     uint64_t kernel_entry;
939     DeviceState *mmio_irqchip, *virtio_irqchip, *pcie_irqchip;
940     int i, base_hartid, hart_count;
941 
942     /* Check socket count limit */
943     if (VIRT_SOCKETS_MAX < riscv_socket_count(machine)) {
944         error_report("number of sockets/nodes should be less than %d",
945             VIRT_SOCKETS_MAX);
946         exit(1);
947     }
948 
949     /* Initialize sockets */
950     mmio_irqchip = virtio_irqchip = pcie_irqchip = NULL;
951     for (i = 0; i < riscv_socket_count(machine); i++) {
952         if (!riscv_socket_check_hartids(machine, i)) {
953             error_report("discontinuous hartids in socket%d", i);
954             exit(1);
955         }
956 
957         base_hartid = riscv_socket_first_hartid(machine, i);
958         if (base_hartid < 0) {
959             error_report("can't find hartid base for socket%d", i);
960             exit(1);
961         }
962 
963         hart_count = riscv_socket_hart_count(machine, i);
964         if (hart_count < 0) {
965             error_report("can't find hart count for socket%d", i);
966             exit(1);
967         }
968 
969         soc_name = g_strdup_printf("soc%d", i);
970         object_initialize_child(OBJECT(machine), soc_name, &s->soc[i],
971                                 TYPE_RISCV_HART_ARRAY);
972         g_free(soc_name);
973         object_property_set_str(OBJECT(&s->soc[i]), "cpu-type",
974                                 machine->cpu_type, &error_abort);
975         object_property_set_int(OBJECT(&s->soc[i]), "hartid-base",
976                                 base_hartid, &error_abort);
977         object_property_set_int(OBJECT(&s->soc[i]), "num-harts",
978                                 hart_count, &error_abort);
979         sysbus_realize(SYS_BUS_DEVICE(&s->soc[i]), &error_abort);
980 
981         if (!kvm_enabled()) {
982             /* Per-socket CLINT */
983             riscv_aclint_swi_create(
984                 memmap[VIRT_CLINT].base + i * memmap[VIRT_CLINT].size,
985                 base_hartid, hart_count, false);
986             riscv_aclint_mtimer_create(
987                 memmap[VIRT_CLINT].base + i * memmap[VIRT_CLINT].size +
988                     RISCV_ACLINT_SWI_SIZE,
989                 RISCV_ACLINT_DEFAULT_MTIMER_SIZE, base_hartid, hart_count,
990                 RISCV_ACLINT_DEFAULT_MTIMECMP, RISCV_ACLINT_DEFAULT_MTIME,
991                 RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
992 
993             /* Per-socket ACLINT SSWI */
994             if (s->have_aclint) {
995                 riscv_aclint_swi_create(
996                     memmap[VIRT_ACLINT_SSWI].base +
997                         i * memmap[VIRT_ACLINT_SSWI].size,
998                     base_hartid, hart_count, true);
999             }
1000         }
1001 
1002         /* Per-socket interrupt controller */
1003         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
1004             s->irqchip[i] = virt_create_plic(memmap, i,
1005                                              base_hartid, hart_count);
1006         } else {
1007             s->irqchip[i] = virt_create_aia(s->aia_type, memmap, i,
1008                                             base_hartid, hart_count);
1009         }
1010 
1011         /* Try to use different IRQCHIP instance based device type */
1012         if (i == 0) {
1013             mmio_irqchip = s->irqchip[i];
1014             virtio_irqchip = s->irqchip[i];
1015             pcie_irqchip = s->irqchip[i];
1016         }
1017         if (i == 1) {
1018             virtio_irqchip = s->irqchip[i];
1019             pcie_irqchip = s->irqchip[i];
1020         }
1021         if (i == 2) {
1022             pcie_irqchip = s->irqchip[i];
1023         }
1024     }
1025 
1026     if (riscv_is_32bit(&s->soc[0])) {
1027 #if HOST_LONG_BITS == 64
1028         /* limit RAM size in a 32-bit system */
1029         if (machine->ram_size > 10 * GiB) {
1030             machine->ram_size = 10 * GiB;
1031             error_report("Limiting RAM size to 10 GiB");
1032         }
1033 #endif
1034         virt_high_pcie_memmap.base = VIRT32_HIGH_PCIE_MMIO_BASE;
1035         virt_high_pcie_memmap.size = VIRT32_HIGH_PCIE_MMIO_SIZE;
1036     } else {
1037         virt_high_pcie_memmap.size = VIRT64_HIGH_PCIE_MMIO_SIZE;
1038         virt_high_pcie_memmap.base = memmap[VIRT_DRAM].base + machine->ram_size;
1039         virt_high_pcie_memmap.base =
1040             ROUND_UP(virt_high_pcie_memmap.base, virt_high_pcie_memmap.size);
1041     }
1042 
1043     /* register system main memory (actual RAM) */
1044     memory_region_add_subregion(system_memory, memmap[VIRT_DRAM].base,
1045         machine->ram);
1046 
1047     /* create device tree */
1048     create_fdt(s, memmap, machine->ram_size, machine->kernel_cmdline,
1049                riscv_is_32bit(&s->soc[0]));
1050 
1051     /* boot rom */
1052     memory_region_init_rom(mask_rom, NULL, "riscv_virt_board.mrom",
1053                            memmap[VIRT_MROM].size, &error_fatal);
1054     memory_region_add_subregion(system_memory, memmap[VIRT_MROM].base,
1055                                 mask_rom);
1056 
1057     /*
1058      * Only direct boot kernel is currently supported for KVM VM,
1059      * so the "-bios" parameter is ignored and treated like "-bios none"
1060      * when KVM is enabled.
1061      */
1062     if (kvm_enabled()) {
1063         g_free(machine->firmware);
1064         machine->firmware = g_strdup("none");
1065     }
1066 
1067     if (riscv_is_32bit(&s->soc[0])) {
1068         firmware_end_addr = riscv_find_and_load_firmware(machine,
1069                                     RISCV32_BIOS_BIN, start_addr, NULL);
1070     } else {
1071         firmware_end_addr = riscv_find_and_load_firmware(machine,
1072                                     RISCV64_BIOS_BIN, start_addr, NULL);
1073     }
1074 
1075     if (machine->kernel_filename) {
1076         kernel_start_addr = riscv_calc_kernel_start_addr(&s->soc[0],
1077                                                          firmware_end_addr);
1078 
1079         kernel_entry = riscv_load_kernel(machine->kernel_filename,
1080                                          kernel_start_addr, NULL);
1081 
1082         if (machine->initrd_filename) {
1083             hwaddr start;
1084             hwaddr end = riscv_load_initrd(machine->initrd_filename,
1085                                            machine->ram_size, kernel_entry,
1086                                            &start);
1087             qemu_fdt_setprop_cell(machine->fdt, "/chosen",
1088                                   "linux,initrd-start", start);
1089             qemu_fdt_setprop_cell(machine->fdt, "/chosen", "linux,initrd-end",
1090                                   end);
1091         }
1092     } else {
1093        /*
1094         * If dynamic firmware is used, it doesn't know where is the next mode
1095         * if kernel argument is not set.
1096         */
1097         kernel_entry = 0;
1098     }
1099 
1100     if (drive_get(IF_PFLASH, 0, 0)) {
1101         /*
1102          * Pflash was supplied, let's overwrite the address we jump to after
1103          * reset to the base of the flash.
1104          */
1105         start_addr = virt_memmap[VIRT_FLASH].base;
1106     }
1107 
1108     /*
1109      * Init fw_cfg.  Must be done before riscv_load_fdt, otherwise the device
1110      * tree cannot be altered and we get FDT_ERR_NOSPACE.
1111      */
1112     s->fw_cfg = create_fw_cfg(machine);
1113     rom_set_fw(s->fw_cfg);
1114 
1115     /* Compute the fdt load address in dram */
1116     fdt_load_addr = riscv_load_fdt(memmap[VIRT_DRAM].base,
1117                                    machine->ram_size, machine->fdt);
1118     /* load the reset vector */
1119     riscv_setup_rom_reset_vec(machine, &s->soc[0], start_addr,
1120                               virt_memmap[VIRT_MROM].base,
1121                               virt_memmap[VIRT_MROM].size, kernel_entry,
1122                               fdt_load_addr, machine->fdt);
1123 
1124     /*
1125      * Only direct boot kernel is currently supported for KVM VM,
1126      * So here setup kernel start address and fdt address.
1127      * TODO:Support firmware loading and integrate to TCG start
1128      */
1129     if (kvm_enabled()) {
1130         riscv_setup_direct_kernel(kernel_entry, fdt_load_addr);
1131     }
1132 
1133     /* SiFive Test MMIO device */
1134     sifive_test_create(memmap[VIRT_TEST].base);
1135 
1136     /* VirtIO MMIO devices */
1137     for (i = 0; i < VIRTIO_COUNT; i++) {
1138         sysbus_create_simple("virtio-mmio",
1139             memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
1140             qdev_get_gpio_in(DEVICE(virtio_irqchip), VIRTIO_IRQ + i));
1141     }
1142 
1143     gpex_pcie_init(system_memory,
1144                    memmap[VIRT_PCIE_ECAM].base,
1145                    memmap[VIRT_PCIE_ECAM].size,
1146                    memmap[VIRT_PCIE_MMIO].base,
1147                    memmap[VIRT_PCIE_MMIO].size,
1148                    virt_high_pcie_memmap.base,
1149                    virt_high_pcie_memmap.size,
1150                    memmap[VIRT_PCIE_PIO].base,
1151                    DEVICE(pcie_irqchip));
1152 
1153     serial_mm_init(system_memory, memmap[VIRT_UART0].base,
1154         0, qdev_get_gpio_in(DEVICE(mmio_irqchip), UART0_IRQ), 399193,
1155         serial_hd(0), DEVICE_LITTLE_ENDIAN);
1156 
1157     sysbus_create_simple("goldfish_rtc", memmap[VIRT_RTC].base,
1158         qdev_get_gpio_in(DEVICE(mmio_irqchip), RTC_IRQ));
1159 
1160     virt_flash_create(s);
1161 
1162     for (i = 0; i < ARRAY_SIZE(s->flash); i++) {
1163         /* Map legacy -drive if=pflash to machine properties */
1164         pflash_cfi01_legacy_drive(s->flash[i],
1165                                   drive_get(IF_PFLASH, 0, i));
1166     }
1167     virt_flash_map(s, system_memory);
1168 }
1169 
1170 static void virt_machine_instance_init(Object *obj)
1171 {
1172 }
1173 
1174 static char *virt_get_aia(Object *obj, Error **errp)
1175 {
1176     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1177     const char *val;
1178 
1179     switch (s->aia_type) {
1180     case VIRT_AIA_TYPE_APLIC:
1181         val = "aplic";
1182         break;
1183     default:
1184         val = "none";
1185         break;
1186     };
1187 
1188     return g_strdup(val);
1189 }
1190 
1191 static void virt_set_aia(Object *obj, const char *val, Error **errp)
1192 {
1193     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1194 
1195     if (!strcmp(val, "none")) {
1196         s->aia_type = VIRT_AIA_TYPE_NONE;
1197     } else if (!strcmp(val, "aplic")) {
1198         s->aia_type = VIRT_AIA_TYPE_APLIC;
1199     } else {
1200         error_setg(errp, "Invalid AIA interrupt controller type");
1201         error_append_hint(errp, "Valid values are none, and aplic.\n");
1202     }
1203 }
1204 
1205 static bool virt_get_aclint(Object *obj, Error **errp)
1206 {
1207     MachineState *ms = MACHINE(obj);
1208     RISCVVirtState *s = RISCV_VIRT_MACHINE(ms);
1209 
1210     return s->have_aclint;
1211 }
1212 
1213 static void virt_set_aclint(Object *obj, bool value, Error **errp)
1214 {
1215     MachineState *ms = MACHINE(obj);
1216     RISCVVirtState *s = RISCV_VIRT_MACHINE(ms);
1217 
1218     s->have_aclint = value;
1219 }
1220 
1221 static void virt_machine_class_init(ObjectClass *oc, void *data)
1222 {
1223     MachineClass *mc = MACHINE_CLASS(oc);
1224 
1225     mc->desc = "RISC-V VirtIO board";
1226     mc->init = virt_machine_init;
1227     mc->max_cpus = VIRT_CPUS_MAX;
1228     mc->default_cpu_type = TYPE_RISCV_CPU_BASE;
1229     mc->pci_allow_0_address = true;
1230     mc->possible_cpu_arch_ids = riscv_numa_possible_cpu_arch_ids;
1231     mc->cpu_index_to_instance_props = riscv_numa_cpu_index_to_props;
1232     mc->get_default_cpu_node_id = riscv_numa_get_default_cpu_node_id;
1233     mc->numa_mem_supported = true;
1234     mc->default_ram_id = "riscv_virt_board.ram";
1235 
1236     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
1237 
1238     object_class_property_add_bool(oc, "aclint", virt_get_aclint,
1239                                    virt_set_aclint);
1240     object_class_property_set_description(oc, "aclint",
1241                                           "Set on/off to enable/disable "
1242                                           "emulating ACLINT devices");
1243 
1244     object_class_property_add_str(oc, "aia", virt_get_aia,
1245                                   virt_set_aia);
1246     object_class_property_set_description(oc, "aia",
1247                                           "Set type of AIA interrupt "
1248                                           "conttoller. Valid values are "
1249                                           "none, and aplic.");
1250 }
1251 
1252 static const TypeInfo virt_machine_typeinfo = {
1253     .name       = MACHINE_TYPE_NAME("virt"),
1254     .parent     = TYPE_MACHINE,
1255     .class_init = virt_machine_class_init,
1256     .instance_init = virt_machine_instance_init,
1257     .instance_size = sizeof(RISCVVirtState),
1258 };
1259 
1260 static void virt_machine_init_register_types(void)
1261 {
1262     type_register_static(&virt_machine_typeinfo);
1263 }
1264 
1265 type_init(virt_machine_init_register_types)
1266