xref: /openbmc/qemu/hw/arm/boot.c (revision b3e02049)
1 /*
2  * ARM kernel loader.
3  *
4  * Copyright (c) 2006-2007 CodeSourcery.
5  * Written by Paul Brook
6  *
7  * This code is licensed under the GPL.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "qemu/datadir.h"
12 #include "qemu/error-report.h"
13 #include "qapi/error.h"
14 #include <libfdt.h>
15 #include "hw/arm/boot.h"
16 #include "hw/arm/linux-boot-if.h"
17 #include "sysemu/kvm.h"
18 #include "sysemu/sysemu.h"
19 #include "sysemu/numa.h"
20 #include "hw/boards.h"
21 #include "sysemu/reset.h"
22 #include "hw/loader.h"
23 #include "elf.h"
24 #include "sysemu/device_tree.h"
25 #include "qemu/config-file.h"
26 #include "qemu/option.h"
27 #include "qemu/units.h"
28 
29 /* Kernel boot protocol is specified in the kernel docs
30  * Documentation/arm/Booting and Documentation/arm64/booting.txt
31  * They have different preferred image load offsets from system RAM base.
32  */
33 #define KERNEL_ARGS_ADDR   0x100
34 #define KERNEL_NOLOAD_ADDR 0x02000000
35 #define KERNEL_LOAD_ADDR   0x00010000
36 #define KERNEL64_LOAD_ADDR 0x00080000
37 
38 #define ARM64_TEXT_OFFSET_OFFSET    8
39 #define ARM64_MAGIC_OFFSET          56
40 
41 #define BOOTLOADER_MAX_SIZE         (4 * KiB)
42 
43 AddressSpace *arm_boot_address_space(ARMCPU *cpu,
44                                      const struct arm_boot_info *info)
45 {
46     /* Return the address space to use for bootloader reads and writes.
47      * We prefer the secure address space if the CPU has it and we're
48      * going to boot the guest into it.
49      */
50     int asidx;
51     CPUState *cs = CPU(cpu);
52 
53     if (arm_feature(&cpu->env, ARM_FEATURE_EL3) && info->secure_boot) {
54         asidx = ARMASIdx_S;
55     } else {
56         asidx = ARMASIdx_NS;
57     }
58 
59     return cpu_get_address_space(cs, asidx);
60 }
61 
62 typedef enum {
63     FIXUP_NONE = 0,     /* do nothing */
64     FIXUP_TERMINATOR,   /* end of insns */
65     FIXUP_BOARDID,      /* overwrite with board ID number */
66     FIXUP_BOARD_SETUP,  /* overwrite with board specific setup code address */
67     FIXUP_ARGPTR_LO,    /* overwrite with pointer to kernel args */
68     FIXUP_ARGPTR_HI,    /* overwrite with pointer to kernel args (high half) */
69     FIXUP_ENTRYPOINT_LO, /* overwrite with kernel entry point */
70     FIXUP_ENTRYPOINT_HI, /* overwrite with kernel entry point (high half) */
71     FIXUP_GIC_CPU_IF,   /* overwrite with GIC CPU interface address */
72     FIXUP_BOOTREG,      /* overwrite with boot register address */
73     FIXUP_DSB,          /* overwrite with correct DSB insn for cpu */
74     FIXUP_MAX,
75 } FixupType;
76 
77 typedef struct ARMInsnFixup {
78     uint32_t insn;
79     FixupType fixup;
80 } ARMInsnFixup;
81 
82 static const ARMInsnFixup bootloader_aarch64[] = {
83     { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */
84     { 0xaa1f03e1 }, /* mov x1, xzr */
85     { 0xaa1f03e2 }, /* mov x2, xzr */
86     { 0xaa1f03e3 }, /* mov x3, xzr */
87     { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */
88     { 0xd61f0080 }, /* br x4      ; Jump to the kernel entry point */
89     { 0, FIXUP_ARGPTR_LO }, /* arg: .word @DTB Lower 32-bits */
90     { 0, FIXUP_ARGPTR_HI}, /* .word @DTB Higher 32-bits */
91     { 0, FIXUP_ENTRYPOINT_LO }, /* entry: .word @Kernel Entry Lower 32-bits */
92     { 0, FIXUP_ENTRYPOINT_HI }, /* .word @Kernel Entry Higher 32-bits */
93     { 0, FIXUP_TERMINATOR }
94 };
95 
96 /* A very small bootloader: call the board-setup code (if needed),
97  * set r0-r2, then jump to the kernel.
98  * If we're not calling boot setup code then we don't copy across
99  * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array.
100  */
101 
102 static const ARMInsnFixup bootloader[] = {
103     { 0xe28fe004 }, /* add     lr, pc, #4 */
104     { 0xe51ff004 }, /* ldr     pc, [pc, #-4] */
105     { 0, FIXUP_BOARD_SETUP },
106 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3
107     { 0xe3a00000 }, /* mov     r0, #0 */
108     { 0xe59f1004 }, /* ldr     r1, [pc, #4] */
109     { 0xe59f2004 }, /* ldr     r2, [pc, #4] */
110     { 0xe59ff004 }, /* ldr     pc, [pc, #4] */
111     { 0, FIXUP_BOARDID },
112     { 0, FIXUP_ARGPTR_LO },
113     { 0, FIXUP_ENTRYPOINT_LO },
114     { 0, FIXUP_TERMINATOR }
115 };
116 
117 /* Handling for secondary CPU boot in a multicore system.
118  * Unlike the uniprocessor/primary CPU boot, this is platform
119  * dependent. The default code here is based on the secondary
120  * CPU boot protocol used on realview/vexpress boards, with
121  * some parameterisation to increase its flexibility.
122  * QEMU platform models for which this code is not appropriate
123  * should override write_secondary_boot and secondary_cpu_reset_hook
124  * instead.
125  *
126  * This code enables the interrupt controllers for the secondary
127  * CPUs and then puts all the secondary CPUs into a loop waiting
128  * for an interprocessor interrupt and polling a configurable
129  * location for the kernel secondary CPU entry point.
130  */
131 #define DSB_INSN 0xf57ff04f
132 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
133 
134 static const ARMInsnFixup smpboot[] = {
135     { 0xe59f2028 }, /* ldr r2, gic_cpu_if */
136     { 0xe59f0028 }, /* ldr r0, bootreg_addr */
137     { 0xe3a01001 }, /* mov r1, #1 */
138     { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */
139     { 0xe3a010ff }, /* mov r1, #0xff */
140     { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */
141     { 0, FIXUP_DSB },   /* dsb */
142     { 0xe320f003 }, /* wfi */
143     { 0xe5901000 }, /* ldr     r1, [r0] */
144     { 0xe1110001 }, /* tst     r1, r1 */
145     { 0x0afffffb }, /* beq     <wfi> */
146     { 0xe12fff11 }, /* bx      r1 */
147     { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */
148     { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */
149     { 0, FIXUP_TERMINATOR }
150 };
151 
152 static void write_bootloader(const char *name, hwaddr addr,
153                              const ARMInsnFixup *insns, uint32_t *fixupcontext,
154                              AddressSpace *as)
155 {
156     /* Fix up the specified bootloader fragment and write it into
157      * guest memory using rom_add_blob_fixed(). fixupcontext is
158      * an array giving the values to write in for the fixup types
159      * which write a value into the code array.
160      */
161     int i, len;
162     uint32_t *code;
163 
164     len = 0;
165     while (insns[len].fixup != FIXUP_TERMINATOR) {
166         len++;
167     }
168 
169     code = g_new0(uint32_t, len);
170 
171     for (i = 0; i < len; i++) {
172         uint32_t insn = insns[i].insn;
173         FixupType fixup = insns[i].fixup;
174 
175         switch (fixup) {
176         case FIXUP_NONE:
177             break;
178         case FIXUP_BOARDID:
179         case FIXUP_BOARD_SETUP:
180         case FIXUP_ARGPTR_LO:
181         case FIXUP_ARGPTR_HI:
182         case FIXUP_ENTRYPOINT_LO:
183         case FIXUP_ENTRYPOINT_HI:
184         case FIXUP_GIC_CPU_IF:
185         case FIXUP_BOOTREG:
186         case FIXUP_DSB:
187             insn = fixupcontext[fixup];
188             break;
189         default:
190             abort();
191         }
192         code[i] = tswap32(insn);
193     }
194 
195     assert((len * sizeof(uint32_t)) < BOOTLOADER_MAX_SIZE);
196 
197     rom_add_blob_fixed_as(name, code, len * sizeof(uint32_t), addr, as);
198 
199     g_free(code);
200 }
201 
202 static void default_write_secondary(ARMCPU *cpu,
203                                     const struct arm_boot_info *info)
204 {
205     uint32_t fixupcontext[FIXUP_MAX];
206     AddressSpace *as = arm_boot_address_space(cpu, info);
207 
208     fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr;
209     fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr;
210     if (arm_feature(&cpu->env, ARM_FEATURE_V7)) {
211         fixupcontext[FIXUP_DSB] = DSB_INSN;
212     } else {
213         fixupcontext[FIXUP_DSB] = CP15_DSB_INSN;
214     }
215 
216     write_bootloader("smpboot", info->smp_loader_start,
217                      smpboot, fixupcontext, as);
218 }
219 
220 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
221                                             const struct arm_boot_info *info,
222                                             hwaddr mvbar_addr)
223 {
224     AddressSpace *as = arm_boot_address_space(cpu, info);
225     int n;
226     uint32_t mvbar_blob[] = {
227         /* mvbar_addr: secure monitor vectors
228          * Default unimplemented and unused vectors to spin. Makes it
229          * easier to debug (as opposed to the CPU running away).
230          */
231         0xeafffffe, /* (spin) */
232         0xeafffffe, /* (spin) */
233         0xe1b0f00e, /* movs pc, lr ;SMC exception return */
234         0xeafffffe, /* (spin) */
235         0xeafffffe, /* (spin) */
236         0xeafffffe, /* (spin) */
237         0xeafffffe, /* (spin) */
238         0xeafffffe, /* (spin) */
239     };
240     uint32_t board_setup_blob[] = {
241         /* board setup addr */
242         0xee110f51, /* mrc     p15, 0, r0, c1, c1, 2  ;read NSACR */
243         0xe3800b03, /* orr     r0, #0xc00             ;set CP11, CP10 */
244         0xee010f51, /* mcr     p15, 0, r0, c1, c1, 2  ;write NSACR */
245         0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */
246         0xee0c0f30, /* mcr     p15, 0, r0, c12, c0, 1 ;set MVBAR */
247         0xee110f11, /* mrc     p15, 0, r0, c1 , c1, 0 ;read SCR */
248         0xe3800031, /* orr     r0, #0x31              ;enable AW, FW, NS */
249         0xee010f11, /* mcr     p15, 0, r0, c1, c1, 0  ;write SCR */
250         0xe1a0100e, /* mov     r1, lr                 ;save LR across SMC */
251         0xe1600070, /* smc     #0                     ;call monitor to flush SCR */
252         0xe1a0f001, /* mov     pc, r1                 ;return */
253     };
254 
255     /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
256     assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
257 
258     /* check that these blobs don't overlap */
259     assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr)
260           || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr));
261 
262     for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
263         mvbar_blob[n] = tswap32(mvbar_blob[n]);
264     }
265     rom_add_blob_fixed_as("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
266                           mvbar_addr, as);
267 
268     for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
269         board_setup_blob[n] = tswap32(board_setup_blob[n]);
270     }
271     rom_add_blob_fixed_as("board-setup", board_setup_blob,
272                           sizeof(board_setup_blob), info->board_setup_addr, as);
273 }
274 
275 static void default_reset_secondary(ARMCPU *cpu,
276                                     const struct arm_boot_info *info)
277 {
278     AddressSpace *as = arm_boot_address_space(cpu, info);
279     CPUState *cs = CPU(cpu);
280 
281     address_space_stl_notdirty(as, info->smp_bootreg_addr,
282                                0, MEMTXATTRS_UNSPECIFIED, NULL);
283     cpu_set_pc(cs, info->smp_loader_start);
284 }
285 
286 static inline bool have_dtb(const struct arm_boot_info *info)
287 {
288     return info->dtb_filename || info->get_dtb;
289 }
290 
291 #define WRITE_WORD(p, value) do { \
292     address_space_stl_notdirty(as, p, value, \
293                                MEMTXATTRS_UNSPECIFIED, NULL);  \
294     p += 4;                       \
295 } while (0)
296 
297 static void set_kernel_args(const struct arm_boot_info *info, AddressSpace *as)
298 {
299     int initrd_size = info->initrd_size;
300     hwaddr base = info->loader_start;
301     hwaddr p;
302 
303     p = base + KERNEL_ARGS_ADDR;
304     /* ATAG_CORE */
305     WRITE_WORD(p, 5);
306     WRITE_WORD(p, 0x54410001);
307     WRITE_WORD(p, 1);
308     WRITE_WORD(p, 0x1000);
309     WRITE_WORD(p, 0);
310     /* ATAG_MEM */
311     /* TODO: handle multiple chips on one ATAG list */
312     WRITE_WORD(p, 4);
313     WRITE_WORD(p, 0x54410002);
314     WRITE_WORD(p, info->ram_size);
315     WRITE_WORD(p, info->loader_start);
316     if (initrd_size) {
317         /* ATAG_INITRD2 */
318         WRITE_WORD(p, 4);
319         WRITE_WORD(p, 0x54420005);
320         WRITE_WORD(p, info->initrd_start);
321         WRITE_WORD(p, initrd_size);
322     }
323     if (info->kernel_cmdline && *info->kernel_cmdline) {
324         /* ATAG_CMDLINE */
325         int cmdline_size;
326 
327         cmdline_size = strlen(info->kernel_cmdline);
328         address_space_write(as, p + 8, MEMTXATTRS_UNSPECIFIED,
329                             info->kernel_cmdline, cmdline_size + 1);
330         cmdline_size = (cmdline_size >> 2) + 1;
331         WRITE_WORD(p, cmdline_size + 2);
332         WRITE_WORD(p, 0x54410009);
333         p += cmdline_size * 4;
334     }
335     if (info->atag_board) {
336         /* ATAG_BOARD */
337         int atag_board_len;
338         uint8_t atag_board_buf[0x1000];
339 
340         atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
341         WRITE_WORD(p, (atag_board_len + 8) >> 2);
342         WRITE_WORD(p, 0x414f4d50);
343         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
344                             atag_board_buf, atag_board_len);
345         p += atag_board_len;
346     }
347     /* ATAG_END */
348     WRITE_WORD(p, 0);
349     WRITE_WORD(p, 0);
350 }
351 
352 static void set_kernel_args_old(const struct arm_boot_info *info,
353                                 AddressSpace *as)
354 {
355     hwaddr p;
356     const char *s;
357     int initrd_size = info->initrd_size;
358     hwaddr base = info->loader_start;
359 
360     /* see linux/include/asm-arm/setup.h */
361     p = base + KERNEL_ARGS_ADDR;
362     /* page_size */
363     WRITE_WORD(p, 4096);
364     /* nr_pages */
365     WRITE_WORD(p, info->ram_size / 4096);
366     /* ramdisk_size */
367     WRITE_WORD(p, 0);
368 #define FLAG_READONLY	1
369 #define FLAG_RDLOAD	4
370 #define FLAG_RDPROMPT	8
371     /* flags */
372     WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
373     /* rootdev */
374     WRITE_WORD(p, (31 << 8) | 0);	/* /dev/mtdblock0 */
375     /* video_num_cols */
376     WRITE_WORD(p, 0);
377     /* video_num_rows */
378     WRITE_WORD(p, 0);
379     /* video_x */
380     WRITE_WORD(p, 0);
381     /* video_y */
382     WRITE_WORD(p, 0);
383     /* memc_control_reg */
384     WRITE_WORD(p, 0);
385     /* unsigned char sounddefault */
386     /* unsigned char adfsdrives */
387     /* unsigned char bytes_per_char_h */
388     /* unsigned char bytes_per_char_v */
389     WRITE_WORD(p, 0);
390     /* pages_in_bank[4] */
391     WRITE_WORD(p, 0);
392     WRITE_WORD(p, 0);
393     WRITE_WORD(p, 0);
394     WRITE_WORD(p, 0);
395     /* pages_in_vram */
396     WRITE_WORD(p, 0);
397     /* initrd_start */
398     if (initrd_size) {
399         WRITE_WORD(p, info->initrd_start);
400     } else {
401         WRITE_WORD(p, 0);
402     }
403     /* initrd_size */
404     WRITE_WORD(p, initrd_size);
405     /* rd_start */
406     WRITE_WORD(p, 0);
407     /* system_rev */
408     WRITE_WORD(p, 0);
409     /* system_serial_low */
410     WRITE_WORD(p, 0);
411     /* system_serial_high */
412     WRITE_WORD(p, 0);
413     /* mem_fclk_21285 */
414     WRITE_WORD(p, 0);
415     /* zero unused fields */
416     while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
417         WRITE_WORD(p, 0);
418     }
419     s = info->kernel_cmdline;
420     if (s) {
421         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED, s, strlen(s) + 1);
422     } else {
423         WRITE_WORD(p, 0);
424     }
425 }
426 
427 static int fdt_add_memory_node(void *fdt, uint32_t acells, hwaddr mem_base,
428                                uint32_t scells, hwaddr mem_len,
429                                int numa_node_id)
430 {
431     char *nodename;
432     int ret;
433 
434     nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
435     qemu_fdt_add_subnode(fdt, nodename);
436     qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
437     ret = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg", acells, mem_base,
438                                        scells, mem_len);
439     if (ret < 0) {
440         goto out;
441     }
442 
443     /* only set the NUMA ID if it is specified */
444     if (numa_node_id >= 0) {
445         ret = qemu_fdt_setprop_cell(fdt, nodename,
446                                     "numa-node-id", numa_node_id);
447     }
448 out:
449     g_free(nodename);
450     return ret;
451 }
452 
453 static void fdt_add_psci_node(void *fdt)
454 {
455     uint32_t cpu_suspend_fn;
456     uint32_t cpu_off_fn;
457     uint32_t cpu_on_fn;
458     uint32_t migrate_fn;
459     ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
460     const char *psci_method;
461     int64_t psci_conduit;
462     int rc;
463 
464     psci_conduit = object_property_get_int(OBJECT(armcpu),
465                                            "psci-conduit",
466                                            &error_abort);
467     switch (psci_conduit) {
468     case QEMU_PSCI_CONDUIT_DISABLED:
469         return;
470     case QEMU_PSCI_CONDUIT_HVC:
471         psci_method = "hvc";
472         break;
473     case QEMU_PSCI_CONDUIT_SMC:
474         psci_method = "smc";
475         break;
476     default:
477         g_assert_not_reached();
478     }
479 
480     /*
481      * If /psci node is present in provided DTB, assume that no fixup
482      * is necessary and all PSCI configuration should be taken as-is
483      */
484     rc = fdt_path_offset(fdt, "/psci");
485     if (rc >= 0) {
486         return;
487     }
488 
489     qemu_fdt_add_subnode(fdt, "/psci");
490     if (armcpu->psci_version == 2) {
491         const char comp[] = "arm,psci-0.2\0arm,psci";
492         qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
493 
494         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
495         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
496             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
497             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
498             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
499         } else {
500             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
501             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
502             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
503         }
504     } else {
505         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
506 
507         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
508         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
509         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
510         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
511     }
512 
513     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
514      * to the instruction that should be used to invoke PSCI functions.
515      * However, the device tree binding uses 'method' instead, so that is
516      * what we should use here.
517      */
518     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
519 
520     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
521     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
522     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
523     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
524 }
525 
526 int arm_load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
527                  hwaddr addr_limit, AddressSpace *as, MachineState *ms)
528 {
529     void *fdt = NULL;
530     int size, rc, n = 0;
531     uint32_t acells, scells;
532     unsigned int i;
533     hwaddr mem_base, mem_len;
534     char **node_path;
535     Error *err = NULL;
536 
537     if (binfo->dtb_filename) {
538         char *filename;
539         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
540         if (!filename) {
541             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
542             goto fail;
543         }
544 
545         fdt = load_device_tree(filename, &size);
546         if (!fdt) {
547             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
548             g_free(filename);
549             goto fail;
550         }
551         g_free(filename);
552     } else {
553         fdt = binfo->get_dtb(binfo, &size);
554         if (!fdt) {
555             fprintf(stderr, "Board was unable to create a dtb blob\n");
556             goto fail;
557         }
558     }
559 
560     if (addr_limit > addr && size > (addr_limit - addr)) {
561         /* Installing the device tree blob at addr would exceed addr_limit.
562          * Whether this constitutes failure is up to the caller to decide,
563          * so just return 0 as size, i.e., no error.
564          */
565         g_free(fdt);
566         return 0;
567     }
568 
569     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
570                                    NULL, &error_fatal);
571     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
572                                    NULL, &error_fatal);
573     if (acells == 0 || scells == 0) {
574         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
575         goto fail;
576     }
577 
578     if (scells < 2 && binfo->ram_size >= 4 * GiB) {
579         /* This is user error so deserves a friendlier error message
580          * than the failure of setprop_sized_cells would provide
581          */
582         fprintf(stderr, "qemu: dtb file not compatible with "
583                 "RAM size > 4GB\n");
584         goto fail;
585     }
586 
587     /* nop all root nodes matching /memory or /memory@unit-address */
588     node_path = qemu_fdt_node_unit_path(fdt, "memory", &err);
589     if (err) {
590         error_report_err(err);
591         goto fail;
592     }
593     while (node_path[n]) {
594         if (g_str_has_prefix(node_path[n], "/memory")) {
595             qemu_fdt_nop_node(fdt, node_path[n]);
596         }
597         n++;
598     }
599     g_strfreev(node_path);
600 
601     /*
602      * We drop all the memory nodes which correspond to empty NUMA nodes
603      * from the device tree, because the Linux NUMA binding document
604      * states they should not be generated. Linux will get the NUMA node
605      * IDs of the empty NUMA nodes from the distance map if they are needed.
606      * This means QEMU users may be obliged to provide command lines which
607      * configure distance maps when the empty NUMA node IDs are needed and
608      * Linux's default distance map isn't sufficient.
609      */
610     if (ms->numa_state != NULL && ms->numa_state->num_nodes > 0) {
611         mem_base = binfo->loader_start;
612         for (i = 0; i < ms->numa_state->num_nodes; i++) {
613             mem_len = ms->numa_state->nodes[i].node_mem;
614             if (!mem_len) {
615                 continue;
616             }
617 
618             rc = fdt_add_memory_node(fdt, acells, mem_base,
619                                      scells, mem_len, i);
620             if (rc < 0) {
621                 fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n",
622                         mem_base);
623                 goto fail;
624             }
625 
626             mem_base += mem_len;
627         }
628     } else {
629         rc = fdt_add_memory_node(fdt, acells, binfo->loader_start,
630                                  scells, binfo->ram_size, -1);
631         if (rc < 0) {
632             fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n",
633                     binfo->loader_start);
634             goto fail;
635         }
636     }
637 
638     rc = fdt_path_offset(fdt, "/chosen");
639     if (rc < 0) {
640         qemu_fdt_add_subnode(fdt, "/chosen");
641     }
642 
643     if (ms->kernel_cmdline && *ms->kernel_cmdline) {
644         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
645                                      ms->kernel_cmdline);
646         if (rc < 0) {
647             fprintf(stderr, "couldn't set /chosen/bootargs\n");
648             goto fail;
649         }
650     }
651 
652     if (binfo->initrd_size) {
653         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
654                                    binfo->initrd_start);
655         if (rc < 0) {
656             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
657             goto fail;
658         }
659 
660         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
661                                    binfo->initrd_start + binfo->initrd_size);
662         if (rc < 0) {
663             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
664             goto fail;
665         }
666     }
667 
668     fdt_add_psci_node(fdt);
669 
670     if (binfo->modify_dtb) {
671         binfo->modify_dtb(binfo, fdt);
672     }
673 
674     qemu_fdt_dumpdtb(fdt, size);
675 
676     /* Put the DTB into the memory map as a ROM image: this will ensure
677      * the DTB is copied again upon reset, even if addr points into RAM.
678      */
679     rom_add_blob_fixed_as("dtb", fdt, size, addr, as);
680 
681     g_free(fdt);
682 
683     return size;
684 
685 fail:
686     g_free(fdt);
687     return -1;
688 }
689 
690 static void do_cpu_reset(void *opaque)
691 {
692     ARMCPU *cpu = opaque;
693     CPUState *cs = CPU(cpu);
694     CPUARMState *env = &cpu->env;
695     const struct arm_boot_info *info = env->boot_info;
696 
697     cpu_reset(cs);
698     if (info) {
699         if (!info->is_linux) {
700             int i;
701             /* Jump to the entry point.  */
702             uint64_t entry = info->entry;
703 
704             switch (info->endianness) {
705             case ARM_ENDIANNESS_LE:
706                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
707                 for (i = 1; i < 4; ++i) {
708                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
709                 }
710                 env->uncached_cpsr &= ~CPSR_E;
711                 break;
712             case ARM_ENDIANNESS_BE8:
713                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
714                 for (i = 1; i < 4; ++i) {
715                     env->cp15.sctlr_el[i] |= SCTLR_EE;
716                 }
717                 env->uncached_cpsr |= CPSR_E;
718                 break;
719             case ARM_ENDIANNESS_BE32:
720                 env->cp15.sctlr_el[1] |= SCTLR_B;
721                 break;
722             case ARM_ENDIANNESS_UNKNOWN:
723                 break; /* Board's decision */
724             default:
725                 g_assert_not_reached();
726             }
727 
728             cpu_set_pc(cs, entry);
729         } else {
730             /* If we are booting Linux then we need to check whether we are
731              * booting into secure or non-secure state and adjust the state
732              * accordingly.  Out of reset, ARM is defined to be in secure state
733              * (SCR.NS = 0), we change that here if non-secure boot has been
734              * requested.
735              */
736             if (arm_feature(env, ARM_FEATURE_EL3)) {
737                 /* AArch64 is defined to come out of reset into EL3 if enabled.
738                  * If we are booting Linux then we need to adjust our EL as
739                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
740                  * SVC, which Linux expects, so no privilege/exception level to
741                  * adjust.
742                  */
743                 if (env->aarch64) {
744                     env->cp15.scr_el3 |= SCR_RW;
745                     if (arm_feature(env, ARM_FEATURE_EL2)) {
746                         env->cp15.hcr_el2 |= HCR_RW;
747                         env->pstate = PSTATE_MODE_EL2h;
748                     } else {
749                         env->pstate = PSTATE_MODE_EL1h;
750                     }
751                     if (cpu_isar_feature(aa64_pauth, cpu)) {
752                         env->cp15.scr_el3 |= SCR_API | SCR_APK;
753                     }
754                     if (cpu_isar_feature(aa64_mte, cpu)) {
755                         env->cp15.scr_el3 |= SCR_ATA;
756                     }
757                     if (cpu_isar_feature(aa64_sve, cpu)) {
758                         env->cp15.cptr_el[3] |= CPTR_EZ;
759                     }
760                     /* AArch64 kernels never boot in secure mode */
761                     assert(!info->secure_boot);
762                     /* This hook is only supported for AArch32 currently:
763                      * bootloader_aarch64[] will not call the hook, and
764                      * the code above has already dropped us into EL2 or EL1.
765                      */
766                     assert(!info->secure_board_setup);
767                 }
768 
769                 if (arm_feature(env, ARM_FEATURE_EL2)) {
770                     /* If we have EL2 then Linux expects the HVC insn to work */
771                     env->cp15.scr_el3 |= SCR_HCE;
772                 }
773 
774                 /* Set to non-secure if not a secure boot */
775                 if (!info->secure_boot &&
776                     (cs != first_cpu || !info->secure_board_setup)) {
777                     /* Linux expects non-secure state */
778                     env->cp15.scr_el3 |= SCR_NS;
779                     /* Set NSACR.{CP11,CP10} so NS can access the FPU */
780                     env->cp15.nsacr |= 3 << 10;
781                 }
782             }
783 
784             if (!env->aarch64 && !info->secure_boot &&
785                 arm_feature(env, ARM_FEATURE_EL2)) {
786                 /*
787                  * This is an AArch32 boot not to Secure state, and
788                  * we have Hyp mode available, so boot the kernel into
789                  * Hyp mode. This is not how the CPU comes out of reset,
790                  * so we need to manually put it there.
791                  */
792                 cpsr_write(env, ARM_CPU_MODE_HYP, CPSR_M, CPSRWriteRaw);
793             }
794 
795             if (cs == first_cpu) {
796                 AddressSpace *as = arm_boot_address_space(cpu, info);
797 
798                 cpu_set_pc(cs, info->loader_start);
799 
800                 if (!have_dtb(info)) {
801                     if (old_param) {
802                         set_kernel_args_old(info, as);
803                     } else {
804                         set_kernel_args(info, as);
805                     }
806                 }
807             } else {
808                 info->secondary_cpu_reset_hook(cpu, info);
809             }
810         }
811         arm_rebuild_hflags(env);
812     }
813 }
814 
815 /**
816  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
817  *                          by key.
818  * @fw_cfg:         The firmware config instance to store the data in.
819  * @size_key:       The firmware config key to store the size of the loaded
820  *                  data under, with fw_cfg_add_i32().
821  * @data_key:       The firmware config key to store the loaded data under,
822  *                  with fw_cfg_add_bytes().
823  * @image_name:     The name of the image file to load. If it is NULL, the
824  *                  function returns without doing anything.
825  * @try_decompress: Whether the image should be decompressed (gunzipped) before
826  *                  adding it to fw_cfg. If decompression fails, the image is
827  *                  loaded as-is.
828  *
829  * In case of failure, the function prints an error message to stderr and the
830  * process exits with status 1.
831  */
832 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
833                                  uint16_t data_key, const char *image_name,
834                                  bool try_decompress)
835 {
836     size_t size = -1;
837     uint8_t *data;
838 
839     if (image_name == NULL) {
840         return;
841     }
842 
843     if (try_decompress) {
844         size = load_image_gzipped_buffer(image_name,
845                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
846     }
847 
848     if (size == (size_t)-1) {
849         gchar *contents;
850         gsize length;
851 
852         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
853             error_report("failed to load \"%s\"", image_name);
854             exit(1);
855         }
856         size = length;
857         data = (uint8_t *)contents;
858     }
859 
860     fw_cfg_add_i32(fw_cfg, size_key, size);
861     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
862 }
863 
864 static int do_arm_linux_init(Object *obj, void *opaque)
865 {
866     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
867         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
868         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
869         struct arm_boot_info *info = opaque;
870 
871         if (albifc->arm_linux_init) {
872             albifc->arm_linux_init(albif, info->secure_boot);
873         }
874     }
875     return 0;
876 }
877 
878 static int64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
879                             uint64_t *lowaddr, uint64_t *highaddr,
880                             int elf_machine, AddressSpace *as)
881 {
882     bool elf_is64;
883     union {
884         Elf32_Ehdr h32;
885         Elf64_Ehdr h64;
886     } elf_header;
887     int data_swab = 0;
888     bool big_endian;
889     int64_t ret = -1;
890     Error *err = NULL;
891 
892 
893     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
894     if (err) {
895         error_free(err);
896         return ret;
897     }
898 
899     if (elf_is64) {
900         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
901         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
902                                       : ARM_ENDIANNESS_LE;
903     } else {
904         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
905         if (big_endian) {
906             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
907                 info->endianness = ARM_ENDIANNESS_BE8;
908             } else {
909                 info->endianness = ARM_ENDIANNESS_BE32;
910                 /* In BE32, the CPU has a different view of the per-byte
911                  * address map than the rest of the system. BE32 ELF files
912                  * are organised such that they can be programmed through
913                  * the CPU's per-word byte-reversed view of the world. QEMU
914                  * however loads ELF files independently of the CPU. So
915                  * tell the ELF loader to byte reverse the data for us.
916                  */
917                 data_swab = 2;
918             }
919         } else {
920             info->endianness = ARM_ENDIANNESS_LE;
921         }
922     }
923 
924     ret = load_elf_as(info->kernel_filename, NULL, NULL, NULL,
925                       pentry, lowaddr, highaddr, NULL, big_endian, elf_machine,
926                       1, data_swab, as);
927     if (ret <= 0) {
928         /* The header loaded but the image didn't */
929         exit(1);
930     }
931 
932     return ret;
933 }
934 
935 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
936                                    hwaddr *entry, AddressSpace *as)
937 {
938     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
939     uint64_t kernel_size = 0;
940     uint8_t *buffer;
941     int size;
942 
943     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
944     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
945                                      &buffer);
946 
947     if (size < 0) {
948         gsize len;
949 
950         /* Load as raw file otherwise */
951         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
952             return -1;
953         }
954         size = len;
955     }
956 
957     /* check the arm64 magic header value -- very old kernels may not have it */
958     if (size > ARM64_MAGIC_OFFSET + 4 &&
959         memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
960         uint64_t hdrvals[2];
961 
962         /* The arm64 Image header has text_offset and image_size fields at 8 and
963          * 16 bytes into the Image header, respectively. The text_offset field
964          * is only valid if the image_size is non-zero.
965          */
966         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
967 
968         kernel_size = le64_to_cpu(hdrvals[1]);
969 
970         if (kernel_size != 0) {
971             kernel_load_offset = le64_to_cpu(hdrvals[0]);
972 
973             /*
974              * We write our startup "bootloader" at the very bottom of RAM,
975              * so that bit can't be used for the image. Luckily the Image
976              * format specification is that the image requests only an offset
977              * from a 2MB boundary, not an absolute load address. So if the
978              * image requests an offset that might mean it overlaps with the
979              * bootloader, we can just load it starting at 2MB+offset rather
980              * than 0MB + offset.
981              */
982             if (kernel_load_offset < BOOTLOADER_MAX_SIZE) {
983                 kernel_load_offset += 2 * MiB;
984             }
985         }
986     }
987 
988     /*
989      * Kernels before v3.17 don't populate the image_size field, and
990      * raw images have no header. For those our best guess at the size
991      * is the size of the Image file itself.
992      */
993     if (kernel_size == 0) {
994         kernel_size = size;
995     }
996 
997     *entry = mem_base + kernel_load_offset;
998     rom_add_blob_fixed_as(filename, buffer, size, *entry, as);
999 
1000     g_free(buffer);
1001 
1002     return kernel_size;
1003 }
1004 
1005 static void arm_setup_direct_kernel_boot(ARMCPU *cpu,
1006                                          struct arm_boot_info *info)
1007 {
1008     /* Set up for a direct boot of a kernel image file. */
1009     CPUState *cs;
1010     AddressSpace *as = arm_boot_address_space(cpu, info);
1011     int kernel_size;
1012     int initrd_size;
1013     int is_linux = 0;
1014     uint64_t elf_entry;
1015     /* Addresses of first byte used and first byte not used by the image */
1016     uint64_t image_low_addr = 0, image_high_addr = 0;
1017     int elf_machine;
1018     hwaddr entry;
1019     static const ARMInsnFixup *primary_loader;
1020     uint64_t ram_end = info->loader_start + info->ram_size;
1021 
1022     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
1023         primary_loader = bootloader_aarch64;
1024         elf_machine = EM_AARCH64;
1025     } else {
1026         primary_loader = bootloader;
1027         if (!info->write_board_setup) {
1028             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
1029         }
1030         elf_machine = EM_ARM;
1031     }
1032 
1033     if (!info->secondary_cpu_reset_hook) {
1034         info->secondary_cpu_reset_hook = default_reset_secondary;
1035     }
1036     if (!info->write_secondary_boot) {
1037         info->write_secondary_boot = default_write_secondary;
1038     }
1039 
1040     if (info->nb_cpus == 0)
1041         info->nb_cpus = 1;
1042 
1043     /* Assume that raw images are linux kernels, and ELF images are not.  */
1044     kernel_size = arm_load_elf(info, &elf_entry, &image_low_addr,
1045                                &image_high_addr, elf_machine, as);
1046     if (kernel_size > 0 && have_dtb(info)) {
1047         /*
1048          * If there is still some room left at the base of RAM, try and put
1049          * the DTB there like we do for images loaded with -bios or -pflash.
1050          */
1051         if (image_low_addr > info->loader_start
1052             || image_high_addr < info->loader_start) {
1053             /*
1054              * Set image_low_addr as address limit for arm_load_dtb if it may be
1055              * pointing into RAM, otherwise pass '0' (no limit)
1056              */
1057             if (image_low_addr < info->loader_start) {
1058                 image_low_addr = 0;
1059             }
1060             info->dtb_start = info->loader_start;
1061             info->dtb_limit = image_low_addr;
1062         }
1063     }
1064     entry = elf_entry;
1065     if (kernel_size < 0) {
1066         uint64_t loadaddr = info->loader_start + KERNEL_NOLOAD_ADDR;
1067         kernel_size = load_uimage_as(info->kernel_filename, &entry, &loadaddr,
1068                                      &is_linux, NULL, NULL, as);
1069         if (kernel_size >= 0) {
1070             image_low_addr = loadaddr;
1071             image_high_addr = image_low_addr + kernel_size;
1072         }
1073     }
1074     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1075         kernel_size = load_aarch64_image(info->kernel_filename,
1076                                          info->loader_start, &entry, as);
1077         is_linux = 1;
1078         if (kernel_size >= 0) {
1079             image_low_addr = entry;
1080             image_high_addr = image_low_addr + kernel_size;
1081         }
1082     } else if (kernel_size < 0) {
1083         /* 32-bit ARM */
1084         entry = info->loader_start + KERNEL_LOAD_ADDR;
1085         kernel_size = load_image_targphys_as(info->kernel_filename, entry,
1086                                              ram_end - KERNEL_LOAD_ADDR, as);
1087         is_linux = 1;
1088         if (kernel_size >= 0) {
1089             image_low_addr = entry;
1090             image_high_addr = image_low_addr + kernel_size;
1091         }
1092     }
1093     if (kernel_size < 0) {
1094         error_report("could not load kernel '%s'", info->kernel_filename);
1095         exit(1);
1096     }
1097 
1098     if (kernel_size > info->ram_size) {
1099         error_report("kernel '%s' is too large to fit in RAM "
1100                      "(kernel size %d, RAM size %" PRId64 ")",
1101                      info->kernel_filename, kernel_size, info->ram_size);
1102         exit(1);
1103     }
1104 
1105     info->entry = entry;
1106 
1107     /*
1108      * We want to put the initrd far enough into RAM that when the
1109      * kernel is uncompressed it will not clobber the initrd. However
1110      * on boards without much RAM we must ensure that we still leave
1111      * enough room for a decent sized initrd, and on boards with large
1112      * amounts of RAM we must avoid the initrd being so far up in RAM
1113      * that it is outside lowmem and inaccessible to the kernel.
1114      * So for boards with less  than 256MB of RAM we put the initrd
1115      * halfway into RAM, and for boards with 256MB of RAM or more we put
1116      * the initrd at 128MB.
1117      * We also refuse to put the initrd somewhere that will definitely
1118      * overlay the kernel we just loaded, though for kernel formats which
1119      * don't tell us their exact size (eg self-decompressing 32-bit kernels)
1120      * we might still make a bad choice here.
1121      */
1122     info->initrd_start = info->loader_start +
1123         MIN(info->ram_size / 2, 128 * MiB);
1124     if (image_high_addr) {
1125         info->initrd_start = MAX(info->initrd_start, image_high_addr);
1126     }
1127     info->initrd_start = TARGET_PAGE_ALIGN(info->initrd_start);
1128 
1129     if (is_linux) {
1130         uint32_t fixupcontext[FIXUP_MAX];
1131 
1132         if (info->initrd_filename) {
1133 
1134             if (info->initrd_start >= ram_end) {
1135                 error_report("not enough space after kernel to load initrd");
1136                 exit(1);
1137             }
1138 
1139             initrd_size = load_ramdisk_as(info->initrd_filename,
1140                                           info->initrd_start,
1141                                           ram_end - info->initrd_start, as);
1142             if (initrd_size < 0) {
1143                 initrd_size = load_image_targphys_as(info->initrd_filename,
1144                                                      info->initrd_start,
1145                                                      ram_end -
1146                                                      info->initrd_start,
1147                                                      as);
1148             }
1149             if (initrd_size < 0) {
1150                 error_report("could not load initrd '%s'",
1151                              info->initrd_filename);
1152                 exit(1);
1153             }
1154             if (info->initrd_start + initrd_size > ram_end) {
1155                 error_report("could not load initrd '%s': "
1156                              "too big to fit into RAM after the kernel",
1157                              info->initrd_filename);
1158                 exit(1);
1159             }
1160         } else {
1161             initrd_size = 0;
1162         }
1163         info->initrd_size = initrd_size;
1164 
1165         fixupcontext[FIXUP_BOARDID] = info->board_id;
1166         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1167 
1168         /*
1169          * for device tree boot, we pass the DTB directly in r2. Otherwise
1170          * we point to the kernel args.
1171          */
1172         if (have_dtb(info)) {
1173             hwaddr align;
1174 
1175             if (elf_machine == EM_AARCH64) {
1176                 /*
1177                  * Some AArch64 kernels on early bootup map the fdt region as
1178                  *
1179                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1180                  *
1181                  * Let's play safe and prealign it to 2MB to give us some space.
1182                  */
1183                 align = 2 * MiB;
1184             } else {
1185                 /*
1186                  * Some 32bit kernels will trash anything in the 4K page the
1187                  * initrd ends in, so make sure the DTB isn't caught up in that.
1188                  */
1189                 align = 4 * KiB;
1190             }
1191 
1192             /* Place the DTB after the initrd in memory with alignment. */
1193             info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size,
1194                                            align);
1195             if (info->dtb_start >= ram_end) {
1196                 error_report("Not enough space for DTB after kernel/initrd");
1197                 exit(1);
1198             }
1199             fixupcontext[FIXUP_ARGPTR_LO] = info->dtb_start;
1200             fixupcontext[FIXUP_ARGPTR_HI] = info->dtb_start >> 32;
1201         } else {
1202             fixupcontext[FIXUP_ARGPTR_LO] =
1203                 info->loader_start + KERNEL_ARGS_ADDR;
1204             fixupcontext[FIXUP_ARGPTR_HI] =
1205                 (info->loader_start + KERNEL_ARGS_ADDR) >> 32;
1206             if (info->ram_size >= 4 * GiB) {
1207                 error_report("RAM size must be less than 4GB to boot"
1208                              " Linux kernel using ATAGS (try passing a device tree"
1209                              " using -dtb)");
1210                 exit(1);
1211             }
1212         }
1213         fixupcontext[FIXUP_ENTRYPOINT_LO] = entry;
1214         fixupcontext[FIXUP_ENTRYPOINT_HI] = entry >> 32;
1215 
1216         write_bootloader("bootloader", info->loader_start,
1217                          primary_loader, fixupcontext, as);
1218 
1219         if (info->nb_cpus > 1) {
1220             info->write_secondary_boot(cpu, info);
1221         }
1222         if (info->write_board_setup) {
1223             info->write_board_setup(cpu, info);
1224         }
1225 
1226         /*
1227          * Notify devices which need to fake up firmware initialization
1228          * that we're doing a direct kernel boot.
1229          */
1230         object_child_foreach_recursive(object_get_root(),
1231                                        do_arm_linux_init, info);
1232     }
1233     info->is_linux = is_linux;
1234 
1235     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1236         ARM_CPU(cs)->env.boot_info = info;
1237     }
1238 }
1239 
1240 static void arm_setup_firmware_boot(ARMCPU *cpu, struct arm_boot_info *info)
1241 {
1242     /* Set up for booting firmware (which might load a kernel via fw_cfg) */
1243 
1244     if (have_dtb(info)) {
1245         /*
1246          * If we have a device tree blob, but no kernel to supply it to (or
1247          * the kernel is supposed to be loaded by the bootloader), copy the
1248          * DTB to the base of RAM for the bootloader to pick up.
1249          */
1250         info->dtb_start = info->loader_start;
1251     }
1252 
1253     if (info->kernel_filename) {
1254         FWCfgState *fw_cfg;
1255         bool try_decompressing_kernel;
1256 
1257         fw_cfg = fw_cfg_find();
1258 
1259         if (!fw_cfg) {
1260             error_report("This machine type does not support loading both "
1261                          "a guest firmware/BIOS image and a guest kernel at "
1262                          "the same time. You should change your QEMU command "
1263                          "line to specify one or the other, but not both.");
1264             exit(1);
1265         }
1266 
1267         try_decompressing_kernel = arm_feature(&cpu->env,
1268                                                ARM_FEATURE_AARCH64);
1269 
1270         /*
1271          * Expose the kernel, the command line, and the initrd in fw_cfg.
1272          * We don't process them here at all, it's all left to the
1273          * firmware.
1274          */
1275         load_image_to_fw_cfg(fw_cfg,
1276                              FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
1277                              info->kernel_filename,
1278                              try_decompressing_kernel);
1279         load_image_to_fw_cfg(fw_cfg,
1280                              FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
1281                              info->initrd_filename, false);
1282 
1283         if (info->kernel_cmdline) {
1284             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
1285                            strlen(info->kernel_cmdline) + 1);
1286             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
1287                               info->kernel_cmdline);
1288         }
1289     }
1290 
1291     /*
1292      * We will start from address 0 (typically a boot ROM image) in the
1293      * same way as hardware. Leave env->boot_info NULL, so that
1294      * do_cpu_reset() knows it does not need to alter the PC on reset.
1295      */
1296 }
1297 
1298 void arm_load_kernel(ARMCPU *cpu, MachineState *ms, struct arm_boot_info *info)
1299 {
1300     CPUState *cs;
1301     AddressSpace *as = arm_boot_address_space(cpu, info);
1302 
1303     /*
1304      * CPU objects (unlike devices) are not automatically reset on system
1305      * reset, so we must always register a handler to do so. If we're
1306      * actually loading a kernel, the handler is also responsible for
1307      * arranging that we start it correctly.
1308      */
1309     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1310         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
1311     }
1312 
1313     /*
1314      * The board code is not supposed to set secure_board_setup unless
1315      * running its code in secure mode is actually possible, and KVM
1316      * doesn't support secure.
1317      */
1318     assert(!(info->secure_board_setup && kvm_enabled()));
1319     info->kernel_filename = ms->kernel_filename;
1320     info->kernel_cmdline = ms->kernel_cmdline;
1321     info->initrd_filename = ms->initrd_filename;
1322     info->dtb_filename = ms->dtb;
1323     info->dtb_limit = 0;
1324 
1325     /* Load the kernel.  */
1326     if (!info->kernel_filename || info->firmware_loaded) {
1327         arm_setup_firmware_boot(cpu, info);
1328     } else {
1329         arm_setup_direct_kernel_boot(cpu, info);
1330     }
1331 
1332     if (!info->skip_dtb_autoload && have_dtb(info)) {
1333         if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms) < 0) {
1334             exit(1);
1335         }
1336     }
1337 }
1338 
1339 static const TypeInfo arm_linux_boot_if_info = {
1340     .name = TYPE_ARM_LINUX_BOOT_IF,
1341     .parent = TYPE_INTERFACE,
1342     .class_size = sizeof(ARMLinuxBootIfClass),
1343 };
1344 
1345 static void arm_linux_boot_register_types(void)
1346 {
1347     type_register_static(&arm_linux_boot_if_info);
1348 }
1349 
1350 type_init(arm_linux_boot_register_types)
1351