xref: /openbmc/qemu/hw/arm/boot.c (revision 8f9abdf5)
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      * A pre-existing /psci node might specify function ID values
482      * that don't match QEMU's PSCI implementation. Delete the whole
483      * node and put our own in instead.
484      */
485     rc = fdt_path_offset(fdt, "/psci");
486     if (rc >= 0) {
487         qemu_fdt_nop_node(fdt, "/psci");
488     }
489 
490     qemu_fdt_add_subnode(fdt, "/psci");
491     if (armcpu->psci_version >= QEMU_PSCI_VERSION_0_2) {
492         if (armcpu->psci_version < QEMU_PSCI_VERSION_1_0) {
493             const char comp[] = "arm,psci-0.2\0arm,psci";
494             qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
495         } else {
496             const char comp[] = "arm,psci-1.0\0arm,psci-0.2\0arm,psci";
497             qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
498         }
499 
500         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
501         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
502             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
503             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
504             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
505         } else {
506             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
507             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
508             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
509         }
510     } else {
511         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
512 
513         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
514         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
515         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
516         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
517     }
518 
519     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
520      * to the instruction that should be used to invoke PSCI functions.
521      * However, the device tree binding uses 'method' instead, so that is
522      * what we should use here.
523      */
524     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
525 
526     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
527     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
528     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
529     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
530 }
531 
532 int arm_load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
533                  hwaddr addr_limit, AddressSpace *as, MachineState *ms)
534 {
535     void *fdt = NULL;
536     int size, rc, n = 0;
537     uint32_t acells, scells;
538     unsigned int i;
539     hwaddr mem_base, mem_len;
540     char **node_path;
541     Error *err = NULL;
542 
543     if (binfo->dtb_filename) {
544         char *filename;
545         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
546         if (!filename) {
547             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
548             goto fail;
549         }
550 
551         fdt = load_device_tree(filename, &size);
552         if (!fdt) {
553             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
554             g_free(filename);
555             goto fail;
556         }
557         g_free(filename);
558     } else {
559         fdt = binfo->get_dtb(binfo, &size);
560         if (!fdt) {
561             fprintf(stderr, "Board was unable to create a dtb blob\n");
562             goto fail;
563         }
564     }
565 
566     if (addr_limit > addr && size > (addr_limit - addr)) {
567         /* Installing the device tree blob at addr would exceed addr_limit.
568          * Whether this constitutes failure is up to the caller to decide,
569          * so just return 0 as size, i.e., no error.
570          */
571         g_free(fdt);
572         return 0;
573     }
574 
575     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
576                                    NULL, &error_fatal);
577     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
578                                    NULL, &error_fatal);
579     if (acells == 0 || scells == 0) {
580         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
581         goto fail;
582     }
583 
584     if (scells < 2 && binfo->ram_size >= 4 * GiB) {
585         /* This is user error so deserves a friendlier error message
586          * than the failure of setprop_sized_cells would provide
587          */
588         fprintf(stderr, "qemu: dtb file not compatible with "
589                 "RAM size > 4GB\n");
590         goto fail;
591     }
592 
593     /* nop all root nodes matching /memory or /memory@unit-address */
594     node_path = qemu_fdt_node_unit_path(fdt, "memory", &err);
595     if (err) {
596         error_report_err(err);
597         goto fail;
598     }
599     while (node_path[n]) {
600         if (g_str_has_prefix(node_path[n], "/memory")) {
601             qemu_fdt_nop_node(fdt, node_path[n]);
602         }
603         n++;
604     }
605     g_strfreev(node_path);
606 
607     /*
608      * We drop all the memory nodes which correspond to empty NUMA nodes
609      * from the device tree, because the Linux NUMA binding document
610      * states they should not be generated. Linux will get the NUMA node
611      * IDs of the empty NUMA nodes from the distance map if they are needed.
612      * This means QEMU users may be obliged to provide command lines which
613      * configure distance maps when the empty NUMA node IDs are needed and
614      * Linux's default distance map isn't sufficient.
615      */
616     if (ms->numa_state != NULL && ms->numa_state->num_nodes > 0) {
617         mem_base = binfo->loader_start;
618         for (i = 0; i < ms->numa_state->num_nodes; i++) {
619             mem_len = ms->numa_state->nodes[i].node_mem;
620             if (!mem_len) {
621                 continue;
622             }
623 
624             rc = fdt_add_memory_node(fdt, acells, mem_base,
625                                      scells, mem_len, i);
626             if (rc < 0) {
627                 fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n",
628                         mem_base);
629                 goto fail;
630             }
631 
632             mem_base += mem_len;
633         }
634     } else {
635         rc = fdt_add_memory_node(fdt, acells, binfo->loader_start,
636                                  scells, binfo->ram_size, -1);
637         if (rc < 0) {
638             fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n",
639                     binfo->loader_start);
640             goto fail;
641         }
642     }
643 
644     rc = fdt_path_offset(fdt, "/chosen");
645     if (rc < 0) {
646         qemu_fdt_add_subnode(fdt, "/chosen");
647     }
648 
649     if (ms->kernel_cmdline && *ms->kernel_cmdline) {
650         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
651                                      ms->kernel_cmdline);
652         if (rc < 0) {
653             fprintf(stderr, "couldn't set /chosen/bootargs\n");
654             goto fail;
655         }
656     }
657 
658     if (binfo->initrd_size) {
659         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
660                                    binfo->initrd_start);
661         if (rc < 0) {
662             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
663             goto fail;
664         }
665 
666         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
667                                    binfo->initrd_start + binfo->initrd_size);
668         if (rc < 0) {
669             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
670             goto fail;
671         }
672     }
673 
674     fdt_add_psci_node(fdt);
675 
676     if (binfo->modify_dtb) {
677         binfo->modify_dtb(binfo, fdt);
678     }
679 
680     qemu_fdt_dumpdtb(fdt, size);
681 
682     /* Put the DTB into the memory map as a ROM image: this will ensure
683      * the DTB is copied again upon reset, even if addr points into RAM.
684      */
685     rom_add_blob_fixed_as("dtb", fdt, size, addr, as);
686 
687     g_free(fdt);
688 
689     return size;
690 
691 fail:
692     g_free(fdt);
693     return -1;
694 }
695 
696 static void do_cpu_reset(void *opaque)
697 {
698     ARMCPU *cpu = opaque;
699     CPUState *cs = CPU(cpu);
700     CPUARMState *env = &cpu->env;
701     const struct arm_boot_info *info = env->boot_info;
702 
703     cpu_reset(cs);
704     if (info) {
705         if (!info->is_linux) {
706             int i;
707             /* Jump to the entry point.  */
708             uint64_t entry = info->entry;
709 
710             switch (info->endianness) {
711             case ARM_ENDIANNESS_LE:
712                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
713                 for (i = 1; i < 4; ++i) {
714                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
715                 }
716                 env->uncached_cpsr &= ~CPSR_E;
717                 break;
718             case ARM_ENDIANNESS_BE8:
719                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
720                 for (i = 1; i < 4; ++i) {
721                     env->cp15.sctlr_el[i] |= SCTLR_EE;
722                 }
723                 env->uncached_cpsr |= CPSR_E;
724                 break;
725             case ARM_ENDIANNESS_BE32:
726                 env->cp15.sctlr_el[1] |= SCTLR_B;
727                 break;
728             case ARM_ENDIANNESS_UNKNOWN:
729                 break; /* Board's decision */
730             default:
731                 g_assert_not_reached();
732             }
733 
734             cpu_set_pc(cs, entry);
735         } else {
736             /* If we are booting Linux then we need to check whether we are
737              * booting into secure or non-secure state and adjust the state
738              * accordingly.  Out of reset, ARM is defined to be in secure state
739              * (SCR.NS = 0), we change that here if non-secure boot has been
740              * requested.
741              */
742             if (arm_feature(env, ARM_FEATURE_EL3)) {
743                 /* AArch64 is defined to come out of reset into EL3 if enabled.
744                  * If we are booting Linux then we need to adjust our EL as
745                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
746                  * SVC, which Linux expects, so no privilege/exception level to
747                  * adjust.
748                  */
749                 if (env->aarch64) {
750                     env->cp15.scr_el3 |= SCR_RW;
751                     if (arm_feature(env, ARM_FEATURE_EL2)) {
752                         env->cp15.hcr_el2 |= HCR_RW;
753                         env->pstate = PSTATE_MODE_EL2h;
754                     } else {
755                         env->pstate = PSTATE_MODE_EL1h;
756                     }
757                     if (cpu_isar_feature(aa64_pauth, cpu)) {
758                         env->cp15.scr_el3 |= SCR_API | SCR_APK;
759                     }
760                     if (cpu_isar_feature(aa64_mte, cpu)) {
761                         env->cp15.scr_el3 |= SCR_ATA;
762                     }
763                     if (cpu_isar_feature(aa64_sve, cpu)) {
764                         env->cp15.cptr_el[3] |= R_CPTR_EL3_EZ_MASK;
765                     }
766                     /* AArch64 kernels never boot in secure mode */
767                     assert(!info->secure_boot);
768                     /* This hook is only supported for AArch32 currently:
769                      * bootloader_aarch64[] will not call the hook, and
770                      * the code above has already dropped us into EL2 or EL1.
771                      */
772                     assert(!info->secure_board_setup);
773                 }
774 
775                 if (arm_feature(env, ARM_FEATURE_EL2)) {
776                     /* If we have EL2 then Linux expects the HVC insn to work */
777                     env->cp15.scr_el3 |= SCR_HCE;
778                 }
779 
780                 /* Set to non-secure if not a secure boot */
781                 if (!info->secure_boot &&
782                     (cs != first_cpu || !info->secure_board_setup)) {
783                     /* Linux expects non-secure state */
784                     env->cp15.scr_el3 |= SCR_NS;
785                     /* Set NSACR.{CP11,CP10} so NS can access the FPU */
786                     env->cp15.nsacr |= 3 << 10;
787                 }
788             }
789 
790             if (!env->aarch64 && !info->secure_boot &&
791                 arm_feature(env, ARM_FEATURE_EL2)) {
792                 /*
793                  * This is an AArch32 boot not to Secure state, and
794                  * we have Hyp mode available, so boot the kernel into
795                  * Hyp mode. This is not how the CPU comes out of reset,
796                  * so we need to manually put it there.
797                  */
798                 cpsr_write(env, ARM_CPU_MODE_HYP, CPSR_M, CPSRWriteRaw);
799             }
800 
801             if (cs == first_cpu) {
802                 AddressSpace *as = arm_boot_address_space(cpu, info);
803 
804                 cpu_set_pc(cs, info->loader_start);
805 
806                 if (!have_dtb(info)) {
807                     if (old_param) {
808                         set_kernel_args_old(info, as);
809                     } else {
810                         set_kernel_args(info, as);
811                     }
812                 }
813             } else if (info->secondary_cpu_reset_hook) {
814                 info->secondary_cpu_reset_hook(cpu, info);
815             }
816         }
817         arm_rebuild_hflags(env);
818     }
819 }
820 
821 /**
822  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
823  *                          by key.
824  * @fw_cfg:         The firmware config instance to store the data in.
825  * @size_key:       The firmware config key to store the size of the loaded
826  *                  data under, with fw_cfg_add_i32().
827  * @data_key:       The firmware config key to store the loaded data under,
828  *                  with fw_cfg_add_bytes().
829  * @image_name:     The name of the image file to load. If it is NULL, the
830  *                  function returns without doing anything.
831  * @try_decompress: Whether the image should be decompressed (gunzipped) before
832  *                  adding it to fw_cfg. If decompression fails, the image is
833  *                  loaded as-is.
834  *
835  * In case of failure, the function prints an error message to stderr and the
836  * process exits with status 1.
837  */
838 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
839                                  uint16_t data_key, const char *image_name,
840                                  bool try_decompress)
841 {
842     size_t size = -1;
843     uint8_t *data;
844 
845     if (image_name == NULL) {
846         return;
847     }
848 
849     if (try_decompress) {
850         size = load_image_gzipped_buffer(image_name,
851                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
852     }
853 
854     if (size == (size_t)-1) {
855         gchar *contents;
856         gsize length;
857 
858         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
859             error_report("failed to load \"%s\"", image_name);
860             exit(1);
861         }
862         size = length;
863         data = (uint8_t *)contents;
864     }
865 
866     fw_cfg_add_i32(fw_cfg, size_key, size);
867     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
868 }
869 
870 static int do_arm_linux_init(Object *obj, void *opaque)
871 {
872     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
873         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
874         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
875         struct arm_boot_info *info = opaque;
876 
877         if (albifc->arm_linux_init) {
878             albifc->arm_linux_init(albif, info->secure_boot);
879         }
880     }
881     return 0;
882 }
883 
884 static ssize_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
885                             uint64_t *lowaddr, uint64_t *highaddr,
886                             int elf_machine, AddressSpace *as)
887 {
888     bool elf_is64;
889     union {
890         Elf32_Ehdr h32;
891         Elf64_Ehdr h64;
892     } elf_header;
893     int data_swab = 0;
894     bool big_endian;
895     ssize_t ret = -1;
896     Error *err = NULL;
897 
898 
899     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
900     if (err) {
901         error_free(err);
902         return ret;
903     }
904 
905     if (elf_is64) {
906         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
907         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
908                                       : ARM_ENDIANNESS_LE;
909     } else {
910         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
911         if (big_endian) {
912             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
913                 info->endianness = ARM_ENDIANNESS_BE8;
914             } else {
915                 info->endianness = ARM_ENDIANNESS_BE32;
916                 /* In BE32, the CPU has a different view of the per-byte
917                  * address map than the rest of the system. BE32 ELF files
918                  * are organised such that they can be programmed through
919                  * the CPU's per-word byte-reversed view of the world. QEMU
920                  * however loads ELF files independently of the CPU. So
921                  * tell the ELF loader to byte reverse the data for us.
922                  */
923                 data_swab = 2;
924             }
925         } else {
926             info->endianness = ARM_ENDIANNESS_LE;
927         }
928     }
929 
930     ret = load_elf_as(info->kernel_filename, NULL, NULL, NULL,
931                       pentry, lowaddr, highaddr, NULL, big_endian, elf_machine,
932                       1, data_swab, as);
933     if (ret <= 0) {
934         /* The header loaded but the image didn't */
935         exit(1);
936     }
937 
938     return ret;
939 }
940 
941 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
942                                    hwaddr *entry, AddressSpace *as)
943 {
944     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
945     uint64_t kernel_size = 0;
946     uint8_t *buffer;
947     int size;
948 
949     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
950     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
951                                      &buffer);
952 
953     if (size < 0) {
954         gsize len;
955 
956         /* Load as raw file otherwise */
957         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
958             return -1;
959         }
960         size = len;
961     }
962 
963     /* check the arm64 magic header value -- very old kernels may not have it */
964     if (size > ARM64_MAGIC_OFFSET + 4 &&
965         memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
966         uint64_t hdrvals[2];
967 
968         /* The arm64 Image header has text_offset and image_size fields at 8 and
969          * 16 bytes into the Image header, respectively. The text_offset field
970          * is only valid if the image_size is non-zero.
971          */
972         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
973 
974         kernel_size = le64_to_cpu(hdrvals[1]);
975 
976         if (kernel_size != 0) {
977             kernel_load_offset = le64_to_cpu(hdrvals[0]);
978 
979             /*
980              * We write our startup "bootloader" at the very bottom of RAM,
981              * so that bit can't be used for the image. Luckily the Image
982              * format specification is that the image requests only an offset
983              * from a 2MB boundary, not an absolute load address. So if the
984              * image requests an offset that might mean it overlaps with the
985              * bootloader, we can just load it starting at 2MB+offset rather
986              * than 0MB + offset.
987              */
988             if (kernel_load_offset < BOOTLOADER_MAX_SIZE) {
989                 kernel_load_offset += 2 * MiB;
990             }
991         }
992     }
993 
994     /*
995      * Kernels before v3.17 don't populate the image_size field, and
996      * raw images have no header. For those our best guess at the size
997      * is the size of the Image file itself.
998      */
999     if (kernel_size == 0) {
1000         kernel_size = size;
1001     }
1002 
1003     *entry = mem_base + kernel_load_offset;
1004     rom_add_blob_fixed_as(filename, buffer, size, *entry, as);
1005 
1006     g_free(buffer);
1007 
1008     return kernel_size;
1009 }
1010 
1011 static void arm_setup_direct_kernel_boot(ARMCPU *cpu,
1012                                          struct arm_boot_info *info)
1013 {
1014     /* Set up for a direct boot of a kernel image file. */
1015     CPUState *cs;
1016     AddressSpace *as = arm_boot_address_space(cpu, info);
1017     ssize_t kernel_size;
1018     int initrd_size;
1019     int is_linux = 0;
1020     uint64_t elf_entry;
1021     /* Addresses of first byte used and first byte not used by the image */
1022     uint64_t image_low_addr = 0, image_high_addr = 0;
1023     int elf_machine;
1024     hwaddr entry;
1025     static const ARMInsnFixup *primary_loader;
1026     uint64_t ram_end = info->loader_start + info->ram_size;
1027 
1028     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
1029         primary_loader = bootloader_aarch64;
1030         elf_machine = EM_AARCH64;
1031     } else {
1032         primary_loader = bootloader;
1033         if (!info->write_board_setup) {
1034             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
1035         }
1036         elf_machine = EM_ARM;
1037     }
1038 
1039     /* Assume that raw images are linux kernels, and ELF images are not.  */
1040     kernel_size = arm_load_elf(info, &elf_entry, &image_low_addr,
1041                                &image_high_addr, elf_machine, as);
1042     if (kernel_size > 0 && have_dtb(info)) {
1043         /*
1044          * If there is still some room left at the base of RAM, try and put
1045          * the DTB there like we do for images loaded with -bios or -pflash.
1046          */
1047         if (image_low_addr > info->loader_start
1048             || image_high_addr < info->loader_start) {
1049             /*
1050              * Set image_low_addr as address limit for arm_load_dtb if it may be
1051              * pointing into RAM, otherwise pass '0' (no limit)
1052              */
1053             if (image_low_addr < info->loader_start) {
1054                 image_low_addr = 0;
1055             }
1056             info->dtb_start = info->loader_start;
1057             info->dtb_limit = image_low_addr;
1058         }
1059     }
1060     entry = elf_entry;
1061     if (kernel_size < 0) {
1062         uint64_t loadaddr = info->loader_start + KERNEL_NOLOAD_ADDR;
1063         kernel_size = load_uimage_as(info->kernel_filename, &entry, &loadaddr,
1064                                      &is_linux, NULL, NULL, as);
1065         if (kernel_size >= 0) {
1066             image_low_addr = loadaddr;
1067             image_high_addr = image_low_addr + kernel_size;
1068         }
1069     }
1070     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1071         kernel_size = load_aarch64_image(info->kernel_filename,
1072                                          info->loader_start, &entry, as);
1073         is_linux = 1;
1074         if (kernel_size >= 0) {
1075             image_low_addr = entry;
1076             image_high_addr = image_low_addr + kernel_size;
1077         }
1078     } else if (kernel_size < 0) {
1079         /* 32-bit ARM */
1080         entry = info->loader_start + KERNEL_LOAD_ADDR;
1081         kernel_size = load_image_targphys_as(info->kernel_filename, entry,
1082                                              ram_end - KERNEL_LOAD_ADDR, as);
1083         is_linux = 1;
1084         if (kernel_size >= 0) {
1085             image_low_addr = entry;
1086             image_high_addr = image_low_addr + kernel_size;
1087         }
1088     }
1089     if (kernel_size < 0) {
1090         error_report("could not load kernel '%s'", info->kernel_filename);
1091         exit(1);
1092     }
1093 
1094     if (kernel_size > info->ram_size) {
1095         error_report("kernel '%s' is too large to fit in RAM "
1096                      "(kernel size %zd, RAM size %" PRId64 ")",
1097                      info->kernel_filename, kernel_size, info->ram_size);
1098         exit(1);
1099     }
1100 
1101     info->entry = entry;
1102 
1103     /*
1104      * We want to put the initrd far enough into RAM that when the
1105      * kernel is uncompressed it will not clobber the initrd. However
1106      * on boards without much RAM we must ensure that we still leave
1107      * enough room for a decent sized initrd, and on boards with large
1108      * amounts of RAM we must avoid the initrd being so far up in RAM
1109      * that it is outside lowmem and inaccessible to the kernel.
1110      * So for boards with less  than 256MB of RAM we put the initrd
1111      * halfway into RAM, and for boards with 256MB of RAM or more we put
1112      * the initrd at 128MB.
1113      * We also refuse to put the initrd somewhere that will definitely
1114      * overlay the kernel we just loaded, though for kernel formats which
1115      * don't tell us their exact size (eg self-decompressing 32-bit kernels)
1116      * we might still make a bad choice here.
1117      */
1118     info->initrd_start = info->loader_start +
1119         MIN(info->ram_size / 2, 128 * MiB);
1120     if (image_high_addr) {
1121         info->initrd_start = MAX(info->initrd_start, image_high_addr);
1122     }
1123     info->initrd_start = TARGET_PAGE_ALIGN(info->initrd_start);
1124 
1125     if (is_linux) {
1126         uint32_t fixupcontext[FIXUP_MAX];
1127 
1128         if (info->initrd_filename) {
1129 
1130             if (info->initrd_start >= ram_end) {
1131                 error_report("not enough space after kernel to load initrd");
1132                 exit(1);
1133             }
1134 
1135             initrd_size = load_ramdisk_as(info->initrd_filename,
1136                                           info->initrd_start,
1137                                           ram_end - info->initrd_start, as);
1138             if (initrd_size < 0) {
1139                 initrd_size = load_image_targphys_as(info->initrd_filename,
1140                                                      info->initrd_start,
1141                                                      ram_end -
1142                                                      info->initrd_start,
1143                                                      as);
1144             }
1145             if (initrd_size < 0) {
1146                 error_report("could not load initrd '%s'",
1147                              info->initrd_filename);
1148                 exit(1);
1149             }
1150             if (info->initrd_start + initrd_size > ram_end) {
1151                 error_report("could not load initrd '%s': "
1152                              "too big to fit into RAM after the kernel",
1153                              info->initrd_filename);
1154                 exit(1);
1155             }
1156         } else {
1157             initrd_size = 0;
1158         }
1159         info->initrd_size = initrd_size;
1160 
1161         fixupcontext[FIXUP_BOARDID] = info->board_id;
1162         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1163 
1164         /*
1165          * for device tree boot, we pass the DTB directly in r2. Otherwise
1166          * we point to the kernel args.
1167          */
1168         if (have_dtb(info)) {
1169             hwaddr align;
1170 
1171             if (elf_machine == EM_AARCH64) {
1172                 /*
1173                  * Some AArch64 kernels on early bootup map the fdt region as
1174                  *
1175                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1176                  *
1177                  * Let's play safe and prealign it to 2MB to give us some space.
1178                  */
1179                 align = 2 * MiB;
1180             } else {
1181                 /*
1182                  * Some 32bit kernels will trash anything in the 4K page the
1183                  * initrd ends in, so make sure the DTB isn't caught up in that.
1184                  */
1185                 align = 4 * KiB;
1186             }
1187 
1188             /* Place the DTB after the initrd in memory with alignment. */
1189             info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size,
1190                                            align);
1191             if (info->dtb_start >= ram_end) {
1192                 error_report("Not enough space for DTB after kernel/initrd");
1193                 exit(1);
1194             }
1195             fixupcontext[FIXUP_ARGPTR_LO] = info->dtb_start;
1196             fixupcontext[FIXUP_ARGPTR_HI] = info->dtb_start >> 32;
1197         } else {
1198             fixupcontext[FIXUP_ARGPTR_LO] =
1199                 info->loader_start + KERNEL_ARGS_ADDR;
1200             fixupcontext[FIXUP_ARGPTR_HI] =
1201                 (info->loader_start + KERNEL_ARGS_ADDR) >> 32;
1202             if (info->ram_size >= 4 * GiB) {
1203                 error_report("RAM size must be less than 4GB to boot"
1204                              " Linux kernel using ATAGS (try passing a device tree"
1205                              " using -dtb)");
1206                 exit(1);
1207             }
1208         }
1209         fixupcontext[FIXUP_ENTRYPOINT_LO] = entry;
1210         fixupcontext[FIXUP_ENTRYPOINT_HI] = entry >> 32;
1211 
1212         write_bootloader("bootloader", info->loader_start,
1213                          primary_loader, fixupcontext, as);
1214 
1215         if (info->write_board_setup) {
1216             info->write_board_setup(cpu, info);
1217         }
1218 
1219         /*
1220          * Notify devices which need to fake up firmware initialization
1221          * that we're doing a direct kernel boot.
1222          */
1223         object_child_foreach_recursive(object_get_root(),
1224                                        do_arm_linux_init, info);
1225     }
1226     info->is_linux = is_linux;
1227 
1228     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1229         ARM_CPU(cs)->env.boot_info = info;
1230     }
1231 }
1232 
1233 static void arm_setup_firmware_boot(ARMCPU *cpu, struct arm_boot_info *info)
1234 {
1235     /* Set up for booting firmware (which might load a kernel via fw_cfg) */
1236 
1237     if (have_dtb(info)) {
1238         /*
1239          * If we have a device tree blob, but no kernel to supply it to (or
1240          * the kernel is supposed to be loaded by the bootloader), copy the
1241          * DTB to the base of RAM for the bootloader to pick up.
1242          */
1243         info->dtb_start = info->loader_start;
1244     }
1245 
1246     if (info->kernel_filename) {
1247         FWCfgState *fw_cfg;
1248         bool try_decompressing_kernel;
1249 
1250         fw_cfg = fw_cfg_find();
1251 
1252         if (!fw_cfg) {
1253             error_report("This machine type does not support loading both "
1254                          "a guest firmware/BIOS image and a guest kernel at "
1255                          "the same time. You should change your QEMU command "
1256                          "line to specify one or the other, but not both.");
1257             exit(1);
1258         }
1259 
1260         try_decompressing_kernel = arm_feature(&cpu->env,
1261                                                ARM_FEATURE_AARCH64);
1262 
1263         /*
1264          * Expose the kernel, the command line, and the initrd in fw_cfg.
1265          * We don't process them here at all, it's all left to the
1266          * firmware.
1267          */
1268         load_image_to_fw_cfg(fw_cfg,
1269                              FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
1270                              info->kernel_filename,
1271                              try_decompressing_kernel);
1272         load_image_to_fw_cfg(fw_cfg,
1273                              FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
1274                              info->initrd_filename, false);
1275 
1276         if (info->kernel_cmdline) {
1277             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
1278                            strlen(info->kernel_cmdline) + 1);
1279             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
1280                               info->kernel_cmdline);
1281         }
1282     }
1283 
1284     /*
1285      * We will start from address 0 (typically a boot ROM image) in the
1286      * same way as hardware. Leave env->boot_info NULL, so that
1287      * do_cpu_reset() knows it does not need to alter the PC on reset.
1288      */
1289 }
1290 
1291 void arm_load_kernel(ARMCPU *cpu, MachineState *ms, struct arm_boot_info *info)
1292 {
1293     CPUState *cs;
1294     AddressSpace *as = arm_boot_address_space(cpu, info);
1295     int boot_el;
1296     CPUARMState *env = &cpu->env;
1297     int nb_cpus = 0;
1298 
1299     /*
1300      * CPU objects (unlike devices) are not automatically reset on system
1301      * reset, so we must always register a handler to do so. If we're
1302      * actually loading a kernel, the handler is also responsible for
1303      * arranging that we start it correctly.
1304      */
1305     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1306         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
1307         nb_cpus++;
1308     }
1309 
1310     /*
1311      * The board code is not supposed to set secure_board_setup unless
1312      * running its code in secure mode is actually possible, and KVM
1313      * doesn't support secure.
1314      */
1315     assert(!(info->secure_board_setup && kvm_enabled()));
1316     info->kernel_filename = ms->kernel_filename;
1317     info->kernel_cmdline = ms->kernel_cmdline;
1318     info->initrd_filename = ms->initrd_filename;
1319     info->dtb_filename = ms->dtb;
1320     info->dtb_limit = 0;
1321 
1322     /* Load the kernel.  */
1323     if (!info->kernel_filename || info->firmware_loaded) {
1324         arm_setup_firmware_boot(cpu, info);
1325     } else {
1326         arm_setup_direct_kernel_boot(cpu, info);
1327     }
1328 
1329     /*
1330      * Disable the PSCI conduit if it is set up to target the same
1331      * or a lower EL than the one we're going to start the guest code in.
1332      * This logic needs to agree with the code in do_cpu_reset() which
1333      * decides whether we're going to boot the guest in the highest
1334      * supported exception level or in a lower one.
1335      */
1336 
1337     /*
1338      * If PSCI is enabled, then SMC calls all go to the PSCI handler and
1339      * are never emulated to trap into guest code. It therefore does not
1340      * make sense for the board to have a setup code fragment that runs
1341      * in Secure, because this will probably need to itself issue an SMC of some
1342      * kind as part of its operation.
1343      */
1344     assert(info->psci_conduit == QEMU_PSCI_CONDUIT_DISABLED ||
1345            !info->secure_board_setup);
1346 
1347     /* Boot into highest supported EL ... */
1348     if (arm_feature(env, ARM_FEATURE_EL3)) {
1349         boot_el = 3;
1350     } else if (arm_feature(env, ARM_FEATURE_EL2)) {
1351         boot_el = 2;
1352     } else {
1353         boot_el = 1;
1354     }
1355     /* ...except that if we're booting Linux we adjust the EL we boot into */
1356     if (info->is_linux && !info->secure_boot) {
1357         boot_el = arm_feature(env, ARM_FEATURE_EL2) ? 2 : 1;
1358     }
1359 
1360     if ((info->psci_conduit == QEMU_PSCI_CONDUIT_HVC && boot_el >= 2) ||
1361         (info->psci_conduit == QEMU_PSCI_CONDUIT_SMC && boot_el == 3)) {
1362         info->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
1363     }
1364 
1365     if (info->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) {
1366         for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1367             Object *cpuobj = OBJECT(cs);
1368 
1369             object_property_set_int(cpuobj, "psci-conduit", info->psci_conduit,
1370                                     &error_abort);
1371             /*
1372              * Secondary CPUs start in PSCI powered-down state. Like the
1373              * code in do_cpu_reset(), we assume first_cpu is the primary
1374              * CPU.
1375              */
1376             if (cs != first_cpu) {
1377                 object_property_set_bool(cpuobj, "start-powered-off", true,
1378                                          &error_abort);
1379             }
1380         }
1381     }
1382 
1383     if (info->psci_conduit == QEMU_PSCI_CONDUIT_DISABLED &&
1384         info->is_linux && nb_cpus > 1) {
1385         /*
1386          * We're booting Linux but not using PSCI, so for SMP we need
1387          * to write a custom secondary CPU boot loader stub, and arrange
1388          * for the secondary CPU reset to make the accompanying initialization.
1389          */
1390         if (!info->secondary_cpu_reset_hook) {
1391             info->secondary_cpu_reset_hook = default_reset_secondary;
1392         }
1393         if (!info->write_secondary_boot) {
1394             info->write_secondary_boot = default_write_secondary;
1395         }
1396         info->write_secondary_boot(cpu, info);
1397     } else {
1398         /*
1399          * No secondary boot stub; don't use the reset hook that would
1400          * have set the CPU up to call it
1401          */
1402         info->write_secondary_boot = NULL;
1403         info->secondary_cpu_reset_hook = NULL;
1404     }
1405 
1406     /*
1407      * arm_load_dtb() may add a PSCI node so it must be called after we have
1408      * decided whether to enable PSCI and set the psci-conduit CPU properties.
1409      */
1410     if (!info->skip_dtb_autoload && have_dtb(info)) {
1411         if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms) < 0) {
1412             exit(1);
1413         }
1414     }
1415 }
1416 
1417 static const TypeInfo arm_linux_boot_if_info = {
1418     .name = TYPE_ARM_LINUX_BOOT_IF,
1419     .parent = TYPE_INTERFACE,
1420     .class_size = sizeof(ARMLinuxBootIfClass),
1421 };
1422 
1423 static void arm_linux_boot_register_types(void)
1424 {
1425     type_register_static(&arm_linux_boot_if_info);
1426 }
1427 
1428 type_init(arm_linux_boot_register_types)
1429