xref: /openbmc/qemu/hw/arm/boot.c (revision ab4c1361)
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/tcg.h"
19 #include "sysemu/sysemu.h"
20 #include "sysemu/numa.h"
21 #include "hw/boards.h"
22 #include "sysemu/reset.h"
23 #include "hw/loader.h"
24 #include "elf.h"
25 #include "sysemu/device_tree.h"
26 #include "qemu/config-file.h"
27 #include "qemu/option.h"
28 #include "qemu/units.h"
29 
30 /* Kernel boot protocol is specified in the kernel docs
31  * Documentation/arm/Booting and Documentation/arm64/booting.txt
32  * They have different preferred image load offsets from system RAM base.
33  */
34 #define KERNEL_ARGS_ADDR   0x100
35 #define KERNEL_NOLOAD_ADDR 0x02000000
36 #define KERNEL_LOAD_ADDR   0x00010000
37 #define KERNEL64_LOAD_ADDR 0x00080000
38 
39 #define ARM64_TEXT_OFFSET_OFFSET    8
40 #define ARM64_MAGIC_OFFSET          56
41 
42 #define BOOTLOADER_MAX_SIZE         (4 * KiB)
43 
44 AddressSpace *arm_boot_address_space(ARMCPU *cpu,
45                                      const struct arm_boot_info *info)
46 {
47     /* Return the address space to use for bootloader reads and writes.
48      * We prefer the secure address space if the CPU has it and we're
49      * going to boot the guest into it.
50      */
51     int asidx;
52     CPUState *cs = CPU(cpu);
53 
54     if (arm_feature(&cpu->env, ARM_FEATURE_EL3) && info->secure_boot) {
55         asidx = ARMASIdx_S;
56     } else {
57         asidx = ARMASIdx_NS;
58     }
59 
60     return cpu_get_address_space(cs, asidx);
61 }
62 
63 typedef enum {
64     FIXUP_NONE = 0,     /* do nothing */
65     FIXUP_TERMINATOR,   /* end of insns */
66     FIXUP_BOARDID,      /* overwrite with board ID number */
67     FIXUP_BOARD_SETUP,  /* overwrite with board specific setup code address */
68     FIXUP_ARGPTR_LO,    /* overwrite with pointer to kernel args */
69     FIXUP_ARGPTR_HI,    /* overwrite with pointer to kernel args (high half) */
70     FIXUP_ENTRYPOINT_LO, /* overwrite with kernel entry point */
71     FIXUP_ENTRYPOINT_HI, /* overwrite with kernel entry point (high half) */
72     FIXUP_GIC_CPU_IF,   /* overwrite with GIC CPU interface address */
73     FIXUP_BOOTREG,      /* overwrite with boot register address */
74     FIXUP_DSB,          /* overwrite with correct DSB insn for cpu */
75     FIXUP_MAX,
76 } FixupType;
77 
78 typedef struct ARMInsnFixup {
79     uint32_t insn;
80     FixupType fixup;
81 } ARMInsnFixup;
82 
83 static const ARMInsnFixup bootloader_aarch64[] = {
84     { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */
85     { 0xaa1f03e1 }, /* mov x1, xzr */
86     { 0xaa1f03e2 }, /* mov x2, xzr */
87     { 0xaa1f03e3 }, /* mov x3, xzr */
88     { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */
89     { 0xd61f0080 }, /* br x4      ; Jump to the kernel entry point */
90     { 0, FIXUP_ARGPTR_LO }, /* arg: .word @DTB Lower 32-bits */
91     { 0, FIXUP_ARGPTR_HI}, /* .word @DTB Higher 32-bits */
92     { 0, FIXUP_ENTRYPOINT_LO }, /* entry: .word @Kernel Entry Lower 32-bits */
93     { 0, FIXUP_ENTRYPOINT_HI }, /* .word @Kernel Entry Higher 32-bits */
94     { 0, FIXUP_TERMINATOR }
95 };
96 
97 /* A very small bootloader: call the board-setup code (if needed),
98  * set r0-r2, then jump to the kernel.
99  * If we're not calling boot setup code then we don't copy across
100  * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array.
101  */
102 
103 static const ARMInsnFixup bootloader[] = {
104     { 0xe28fe004 }, /* add     lr, pc, #4 */
105     { 0xe51ff004 }, /* ldr     pc, [pc, #-4] */
106     { 0, FIXUP_BOARD_SETUP },
107 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3
108     { 0xe3a00000 }, /* mov     r0, #0 */
109     { 0xe59f1004 }, /* ldr     r1, [pc, #4] */
110     { 0xe59f2004 }, /* ldr     r2, [pc, #4] */
111     { 0xe59ff004 }, /* ldr     pc, [pc, #4] */
112     { 0, FIXUP_BOARDID },
113     { 0, FIXUP_ARGPTR_LO },
114     { 0, FIXUP_ENTRYPOINT_LO },
115     { 0, FIXUP_TERMINATOR }
116 };
117 
118 /* Handling for secondary CPU boot in a multicore system.
119  * Unlike the uniprocessor/primary CPU boot, this is platform
120  * dependent. The default code here is based on the secondary
121  * CPU boot protocol used on realview/vexpress boards, with
122  * some parameterisation to increase its flexibility.
123  * QEMU platform models for which this code is not appropriate
124  * should override write_secondary_boot and secondary_cpu_reset_hook
125  * instead.
126  *
127  * This code enables the interrupt controllers for the secondary
128  * CPUs and then puts all the secondary CPUs into a loop waiting
129  * for an interprocessor interrupt and polling a configurable
130  * location for the kernel secondary CPU entry point.
131  */
132 #define DSB_INSN 0xf57ff04f
133 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
134 
135 static const ARMInsnFixup smpboot[] = {
136     { 0xe59f2028 }, /* ldr r2, gic_cpu_if */
137     { 0xe59f0028 }, /* ldr r0, bootreg_addr */
138     { 0xe3a01001 }, /* mov r1, #1 */
139     { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */
140     { 0xe3a010ff }, /* mov r1, #0xff */
141     { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */
142     { 0, FIXUP_DSB },   /* dsb */
143     { 0xe320f003 }, /* wfi */
144     { 0xe5901000 }, /* ldr     r1, [r0] */
145     { 0xe1110001 }, /* tst     r1, r1 */
146     { 0x0afffffb }, /* beq     <wfi> */
147     { 0xe12fff11 }, /* bx      r1 */
148     { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */
149     { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */
150     { 0, FIXUP_TERMINATOR }
151 };
152 
153 static void write_bootloader(const char *name, hwaddr addr,
154                              const ARMInsnFixup *insns, uint32_t *fixupcontext,
155                              AddressSpace *as)
156 {
157     /* Fix up the specified bootloader fragment and write it into
158      * guest memory using rom_add_blob_fixed(). fixupcontext is
159      * an array giving the values to write in for the fixup types
160      * which write a value into the code array.
161      */
162     int i, len;
163     uint32_t *code;
164 
165     len = 0;
166     while (insns[len].fixup != FIXUP_TERMINATOR) {
167         len++;
168     }
169 
170     code = g_new0(uint32_t, len);
171 
172     for (i = 0; i < len; i++) {
173         uint32_t insn = insns[i].insn;
174         FixupType fixup = insns[i].fixup;
175 
176         switch (fixup) {
177         case FIXUP_NONE:
178             break;
179         case FIXUP_BOARDID:
180         case FIXUP_BOARD_SETUP:
181         case FIXUP_ARGPTR_LO:
182         case FIXUP_ARGPTR_HI:
183         case FIXUP_ENTRYPOINT_LO:
184         case FIXUP_ENTRYPOINT_HI:
185         case FIXUP_GIC_CPU_IF:
186         case FIXUP_BOOTREG:
187         case FIXUP_DSB:
188             insn = fixupcontext[fixup];
189             break;
190         default:
191             abort();
192         }
193         code[i] = tswap32(insn);
194     }
195 
196     assert((len * sizeof(uint32_t)) < BOOTLOADER_MAX_SIZE);
197 
198     rom_add_blob_fixed_as(name, code, len * sizeof(uint32_t), addr, as);
199 
200     g_free(code);
201 }
202 
203 static void default_write_secondary(ARMCPU *cpu,
204                                     const struct arm_boot_info *info)
205 {
206     uint32_t fixupcontext[FIXUP_MAX];
207     AddressSpace *as = arm_boot_address_space(cpu, info);
208 
209     fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr;
210     fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr;
211     if (arm_feature(&cpu->env, ARM_FEATURE_V7)) {
212         fixupcontext[FIXUP_DSB] = DSB_INSN;
213     } else {
214         fixupcontext[FIXUP_DSB] = CP15_DSB_INSN;
215     }
216 
217     write_bootloader("smpboot", info->smp_loader_start,
218                      smpboot, fixupcontext, as);
219 }
220 
221 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
222                                             const struct arm_boot_info *info,
223                                             hwaddr mvbar_addr)
224 {
225     AddressSpace *as = arm_boot_address_space(cpu, info);
226     int n;
227     uint32_t mvbar_blob[] = {
228         /* mvbar_addr: secure monitor vectors
229          * Default unimplemented and unused vectors to spin. Makes it
230          * easier to debug (as opposed to the CPU running away).
231          */
232         0xeafffffe, /* (spin) */
233         0xeafffffe, /* (spin) */
234         0xe1b0f00e, /* movs pc, lr ;SMC exception return */
235         0xeafffffe, /* (spin) */
236         0xeafffffe, /* (spin) */
237         0xeafffffe, /* (spin) */
238         0xeafffffe, /* (spin) */
239         0xeafffffe, /* (spin) */
240     };
241     uint32_t board_setup_blob[] = {
242         /* board setup addr */
243         0xee110f51, /* mrc     p15, 0, r0, c1, c1, 2  ;read NSACR */
244         0xe3800b03, /* orr     r0, #0xc00             ;set CP11, CP10 */
245         0xee010f51, /* mcr     p15, 0, r0, c1, c1, 2  ;write NSACR */
246         0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */
247         0xee0c0f30, /* mcr     p15, 0, r0, c12, c0, 1 ;set MVBAR */
248         0xee110f11, /* mrc     p15, 0, r0, c1 , c1, 0 ;read SCR */
249         0xe3800031, /* orr     r0, #0x31              ;enable AW, FW, NS */
250         0xee010f11, /* mcr     p15, 0, r0, c1, c1, 0  ;write SCR */
251         0xe1a0100e, /* mov     r1, lr                 ;save LR across SMC */
252         0xe1600070, /* smc     #0                     ;call monitor to flush SCR */
253         0xe1a0f001, /* mov     pc, r1                 ;return */
254     };
255 
256     /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
257     assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
258 
259     /* check that these blobs don't overlap */
260     assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr)
261           || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr));
262 
263     for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
264         mvbar_blob[n] = tswap32(mvbar_blob[n]);
265     }
266     rom_add_blob_fixed_as("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
267                           mvbar_addr, as);
268 
269     for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
270         board_setup_blob[n] = tswap32(board_setup_blob[n]);
271     }
272     rom_add_blob_fixed_as("board-setup", board_setup_blob,
273                           sizeof(board_setup_blob), info->board_setup_addr, as);
274 }
275 
276 static void default_reset_secondary(ARMCPU *cpu,
277                                     const struct arm_boot_info *info)
278 {
279     AddressSpace *as = arm_boot_address_space(cpu, info);
280     CPUState *cs = CPU(cpu);
281 
282     address_space_stl_notdirty(as, info->smp_bootreg_addr,
283                                0, MEMTXATTRS_UNSPECIFIED, NULL);
284     cpu_set_pc(cs, info->smp_loader_start);
285 }
286 
287 static inline bool have_dtb(const struct arm_boot_info *info)
288 {
289     return info->dtb_filename || info->get_dtb;
290 }
291 
292 #define WRITE_WORD(p, value) do { \
293     address_space_stl_notdirty(as, p, value, \
294                                MEMTXATTRS_UNSPECIFIED, NULL);  \
295     p += 4;                       \
296 } while (0)
297 
298 static void set_kernel_args(const struct arm_boot_info *info, AddressSpace *as)
299 {
300     int initrd_size = info->initrd_size;
301     hwaddr base = info->loader_start;
302     hwaddr p;
303 
304     p = base + KERNEL_ARGS_ADDR;
305     /* ATAG_CORE */
306     WRITE_WORD(p, 5);
307     WRITE_WORD(p, 0x54410001);
308     WRITE_WORD(p, 1);
309     WRITE_WORD(p, 0x1000);
310     WRITE_WORD(p, 0);
311     /* ATAG_MEM */
312     /* TODO: handle multiple chips on one ATAG list */
313     WRITE_WORD(p, 4);
314     WRITE_WORD(p, 0x54410002);
315     WRITE_WORD(p, info->ram_size);
316     WRITE_WORD(p, info->loader_start);
317     if (initrd_size) {
318         /* ATAG_INITRD2 */
319         WRITE_WORD(p, 4);
320         WRITE_WORD(p, 0x54420005);
321         WRITE_WORD(p, info->initrd_start);
322         WRITE_WORD(p, initrd_size);
323     }
324     if (info->kernel_cmdline && *info->kernel_cmdline) {
325         /* ATAG_CMDLINE */
326         int cmdline_size;
327 
328         cmdline_size = strlen(info->kernel_cmdline);
329         address_space_write(as, p + 8, MEMTXATTRS_UNSPECIFIED,
330                             info->kernel_cmdline, cmdline_size + 1);
331         cmdline_size = (cmdline_size >> 2) + 1;
332         WRITE_WORD(p, cmdline_size + 2);
333         WRITE_WORD(p, 0x54410009);
334         p += cmdline_size * 4;
335     }
336     if (info->atag_board) {
337         /* ATAG_BOARD */
338         int atag_board_len;
339         uint8_t atag_board_buf[0x1000];
340 
341         atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
342         WRITE_WORD(p, (atag_board_len + 8) >> 2);
343         WRITE_WORD(p, 0x414f4d50);
344         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
345                             atag_board_buf, atag_board_len);
346         p += atag_board_len;
347     }
348     /* ATAG_END */
349     WRITE_WORD(p, 0);
350     WRITE_WORD(p, 0);
351 }
352 
353 static void set_kernel_args_old(const struct arm_boot_info *info,
354                                 AddressSpace *as)
355 {
356     hwaddr p;
357     const char *s;
358     int initrd_size = info->initrd_size;
359     hwaddr base = info->loader_start;
360 
361     /* see linux/include/asm-arm/setup.h */
362     p = base + KERNEL_ARGS_ADDR;
363     /* page_size */
364     WRITE_WORD(p, 4096);
365     /* nr_pages */
366     WRITE_WORD(p, info->ram_size / 4096);
367     /* ramdisk_size */
368     WRITE_WORD(p, 0);
369 #define FLAG_READONLY	1
370 #define FLAG_RDLOAD	4
371 #define FLAG_RDPROMPT	8
372     /* flags */
373     WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
374     /* rootdev */
375     WRITE_WORD(p, (31 << 8) | 0);	/* /dev/mtdblock0 */
376     /* video_num_cols */
377     WRITE_WORD(p, 0);
378     /* video_num_rows */
379     WRITE_WORD(p, 0);
380     /* video_x */
381     WRITE_WORD(p, 0);
382     /* video_y */
383     WRITE_WORD(p, 0);
384     /* memc_control_reg */
385     WRITE_WORD(p, 0);
386     /* unsigned char sounddefault */
387     /* unsigned char adfsdrives */
388     /* unsigned char bytes_per_char_h */
389     /* unsigned char bytes_per_char_v */
390     WRITE_WORD(p, 0);
391     /* pages_in_bank[4] */
392     WRITE_WORD(p, 0);
393     WRITE_WORD(p, 0);
394     WRITE_WORD(p, 0);
395     WRITE_WORD(p, 0);
396     /* pages_in_vram */
397     WRITE_WORD(p, 0);
398     /* initrd_start */
399     if (initrd_size) {
400         WRITE_WORD(p, info->initrd_start);
401     } else {
402         WRITE_WORD(p, 0);
403     }
404     /* initrd_size */
405     WRITE_WORD(p, initrd_size);
406     /* rd_start */
407     WRITE_WORD(p, 0);
408     /* system_rev */
409     WRITE_WORD(p, 0);
410     /* system_serial_low */
411     WRITE_WORD(p, 0);
412     /* system_serial_high */
413     WRITE_WORD(p, 0);
414     /* mem_fclk_21285 */
415     WRITE_WORD(p, 0);
416     /* zero unused fields */
417     while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
418         WRITE_WORD(p, 0);
419     }
420     s = info->kernel_cmdline;
421     if (s) {
422         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED, s, strlen(s) + 1);
423     } else {
424         WRITE_WORD(p, 0);
425     }
426 }
427 
428 static int fdt_add_memory_node(void *fdt, uint32_t acells, hwaddr mem_base,
429                                uint32_t scells, hwaddr mem_len,
430                                int numa_node_id)
431 {
432     char *nodename;
433     int ret;
434 
435     nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
436     qemu_fdt_add_subnode(fdt, nodename);
437     qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
438     ret = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg", acells, mem_base,
439                                        scells, mem_len);
440     if (ret < 0) {
441         goto out;
442     }
443 
444     /* only set the NUMA ID if it is specified */
445     if (numa_node_id >= 0) {
446         ret = qemu_fdt_setprop_cell(fdt, nodename,
447                                     "numa-node-id", numa_node_id);
448     }
449 out:
450     g_free(nodename);
451     return ret;
452 }
453 
454 static void fdt_add_psci_node(void *fdt)
455 {
456     uint32_t cpu_suspend_fn;
457     uint32_t cpu_off_fn;
458     uint32_t cpu_on_fn;
459     uint32_t migrate_fn;
460     ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
461     const char *psci_method;
462     int64_t psci_conduit;
463     int rc;
464 
465     psci_conduit = object_property_get_int(OBJECT(armcpu),
466                                            "psci-conduit",
467                                            &error_abort);
468     switch (psci_conduit) {
469     case QEMU_PSCI_CONDUIT_DISABLED:
470         return;
471     case QEMU_PSCI_CONDUIT_HVC:
472         psci_method = "hvc";
473         break;
474     case QEMU_PSCI_CONDUIT_SMC:
475         psci_method = "smc";
476         break;
477     default:
478         g_assert_not_reached();
479     }
480 
481     /*
482      * A pre-existing /psci node might specify function ID values
483      * that don't match QEMU's PSCI implementation. Delete the whole
484      * node and put our own in instead.
485      */
486     rc = fdt_path_offset(fdt, "/psci");
487     if (rc >= 0) {
488         qemu_fdt_nop_node(fdt, "/psci");
489     }
490 
491     qemu_fdt_add_subnode(fdt, "/psci");
492     if (armcpu->psci_version >= QEMU_PSCI_VERSION_0_2) {
493         if (armcpu->psci_version < QEMU_PSCI_VERSION_1_0) {
494             const char comp[] = "arm,psci-0.2\0arm,psci";
495             qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
496         } else {
497             const char comp[] = "arm,psci-1.0\0arm,psci-0.2\0arm,psci";
498             qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
499         }
500 
501         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
502         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
503             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
504             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
505             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
506         } else {
507             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
508             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
509             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
510         }
511     } else {
512         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
513 
514         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
515         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
516         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
517         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
518     }
519 
520     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
521      * to the instruction that should be used to invoke PSCI functions.
522      * However, the device tree binding uses 'method' instead, so that is
523      * what we should use here.
524      */
525     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
526 
527     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
528     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
529     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
530     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
531 }
532 
533 int arm_load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
534                  hwaddr addr_limit, AddressSpace *as, MachineState *ms)
535 {
536     void *fdt = NULL;
537     int size, rc, n = 0;
538     uint32_t acells, scells;
539     unsigned int i;
540     hwaddr mem_base, mem_len;
541     char **node_path;
542     Error *err = NULL;
543 
544     if (binfo->dtb_filename) {
545         char *filename;
546         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
547         if (!filename) {
548             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
549             goto fail;
550         }
551 
552         fdt = load_device_tree(filename, &size);
553         if (!fdt) {
554             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
555             g_free(filename);
556             goto fail;
557         }
558         g_free(filename);
559     } else {
560         fdt = binfo->get_dtb(binfo, &size);
561         if (!fdt) {
562             fprintf(stderr, "Board was unable to create a dtb blob\n");
563             goto fail;
564         }
565     }
566 
567     if (addr_limit > addr && size > (addr_limit - addr)) {
568         /* Installing the device tree blob at addr would exceed addr_limit.
569          * Whether this constitutes failure is up to the caller to decide,
570          * so just return 0 as size, i.e., no error.
571          */
572         g_free(fdt);
573         return 0;
574     }
575 
576     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
577                                    NULL, &error_fatal);
578     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
579                                    NULL, &error_fatal);
580     if (acells == 0 || scells == 0) {
581         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
582         goto fail;
583     }
584 
585     if (scells < 2 && binfo->ram_size >= 4 * GiB) {
586         /* This is user error so deserves a friendlier error message
587          * than the failure of setprop_sized_cells would provide
588          */
589         fprintf(stderr, "qemu: dtb file not compatible with "
590                 "RAM size > 4GB\n");
591         goto fail;
592     }
593 
594     /* nop all root nodes matching /memory or /memory@unit-address */
595     node_path = qemu_fdt_node_unit_path(fdt, "memory", &err);
596     if (err) {
597         error_report_err(err);
598         goto fail;
599     }
600     while (node_path[n]) {
601         if (g_str_has_prefix(node_path[n], "/memory")) {
602             qemu_fdt_nop_node(fdt, node_path[n]);
603         }
604         n++;
605     }
606     g_strfreev(node_path);
607 
608     /*
609      * We drop all the memory nodes which correspond to empty NUMA nodes
610      * from the device tree, because the Linux NUMA binding document
611      * states they should not be generated. Linux will get the NUMA node
612      * IDs of the empty NUMA nodes from the distance map if they are needed.
613      * This means QEMU users may be obliged to provide command lines which
614      * configure distance maps when the empty NUMA node IDs are needed and
615      * Linux's default distance map isn't sufficient.
616      */
617     if (ms->numa_state != NULL && ms->numa_state->num_nodes > 0) {
618         mem_base = binfo->loader_start;
619         for (i = 0; i < ms->numa_state->num_nodes; i++) {
620             mem_len = ms->numa_state->nodes[i].node_mem;
621             if (!mem_len) {
622                 continue;
623             }
624 
625             rc = fdt_add_memory_node(fdt, acells, mem_base,
626                                      scells, mem_len, i);
627             if (rc < 0) {
628                 fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n",
629                         mem_base);
630                 goto fail;
631             }
632 
633             mem_base += mem_len;
634         }
635     } else {
636         rc = fdt_add_memory_node(fdt, acells, binfo->loader_start,
637                                  scells, binfo->ram_size, -1);
638         if (rc < 0) {
639             fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n",
640                     binfo->loader_start);
641             goto fail;
642         }
643     }
644 
645     rc = fdt_path_offset(fdt, "/chosen");
646     if (rc < 0) {
647         qemu_fdt_add_subnode(fdt, "/chosen");
648     }
649 
650     if (ms->kernel_cmdline && *ms->kernel_cmdline) {
651         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
652                                      ms->kernel_cmdline);
653         if (rc < 0) {
654             fprintf(stderr, "couldn't set /chosen/bootargs\n");
655             goto fail;
656         }
657     }
658 
659     if (binfo->initrd_size) {
660         rc = qemu_fdt_setprop_sized_cells(fdt, "/chosen", "linux,initrd-start",
661                                           acells, binfo->initrd_start);
662         if (rc < 0) {
663             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
664             goto fail;
665         }
666 
667         rc = qemu_fdt_setprop_sized_cells(fdt, "/chosen", "linux,initrd-end",
668                                           acells,
669                                           binfo->initrd_start +
670                                           binfo->initrd_size);
671         if (rc < 0) {
672             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
673             goto fail;
674         }
675     }
676 
677     fdt_add_psci_node(fdt);
678 
679     if (binfo->modify_dtb) {
680         binfo->modify_dtb(binfo, fdt);
681     }
682 
683     qemu_fdt_dumpdtb(fdt, size);
684 
685     /* Put the DTB into the memory map as a ROM image: this will ensure
686      * the DTB is copied again upon reset, even if addr points into RAM.
687      */
688     rom_add_blob_fixed_as("dtb", fdt, size, addr, as);
689     qemu_register_reset_nosnapshotload(qemu_fdt_randomize_seeds,
690                                        rom_ptr_for_as(as, addr, size));
691 
692     g_free(fdt);
693 
694     return size;
695 
696 fail:
697     g_free(fdt);
698     return -1;
699 }
700 
701 static void do_cpu_reset(void *opaque)
702 {
703     ARMCPU *cpu = opaque;
704     CPUState *cs = CPU(cpu);
705     CPUARMState *env = &cpu->env;
706     const struct arm_boot_info *info = env->boot_info;
707 
708     cpu_reset(cs);
709     if (info) {
710         if (!info->is_linux) {
711             int i;
712             /* Jump to the entry point.  */
713             uint64_t entry = info->entry;
714 
715             switch (info->endianness) {
716             case ARM_ENDIANNESS_LE:
717                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
718                 for (i = 1; i < 4; ++i) {
719                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
720                 }
721                 env->uncached_cpsr &= ~CPSR_E;
722                 break;
723             case ARM_ENDIANNESS_BE8:
724                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
725                 for (i = 1; i < 4; ++i) {
726                     env->cp15.sctlr_el[i] |= SCTLR_EE;
727                 }
728                 env->uncached_cpsr |= CPSR_E;
729                 break;
730             case ARM_ENDIANNESS_BE32:
731                 env->cp15.sctlr_el[1] |= SCTLR_B;
732                 break;
733             case ARM_ENDIANNESS_UNKNOWN:
734                 break; /* Board's decision */
735             default:
736                 g_assert_not_reached();
737             }
738 
739             cpu_set_pc(cs, entry);
740         } else {
741             /* If we are booting Linux then we need to check whether we are
742              * booting into secure or non-secure state and adjust the state
743              * accordingly.  Out of reset, ARM is defined to be in secure state
744              * (SCR.NS = 0), we change that here if non-secure boot has been
745              * requested.
746              */
747             if (arm_feature(env, ARM_FEATURE_EL3)) {
748                 /* AArch64 is defined to come out of reset into EL3 if enabled.
749                  * If we are booting Linux then we need to adjust our EL as
750                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
751                  * SVC, which Linux expects, so no privilege/exception level to
752                  * adjust.
753                  */
754                 if (env->aarch64) {
755                     env->cp15.scr_el3 |= SCR_RW;
756                     if (arm_feature(env, ARM_FEATURE_EL2)) {
757                         env->cp15.hcr_el2 |= HCR_RW;
758                         env->pstate = PSTATE_MODE_EL2h;
759                     } else {
760                         env->pstate = PSTATE_MODE_EL1h;
761                     }
762                     if (cpu_isar_feature(aa64_pauth, cpu)) {
763                         env->cp15.scr_el3 |= SCR_API | SCR_APK;
764                     }
765                     if (cpu_isar_feature(aa64_mte, cpu)) {
766                         env->cp15.scr_el3 |= SCR_ATA;
767                     }
768                     if (cpu_isar_feature(aa64_sve, cpu)) {
769                         env->cp15.cptr_el[3] |= R_CPTR_EL3_EZ_MASK;
770                         env->vfp.zcr_el[3] = 0xf;
771                     }
772                     if (cpu_isar_feature(aa64_sme, cpu)) {
773                         env->cp15.cptr_el[3] |= R_CPTR_EL3_ESM_MASK;
774                         env->cp15.scr_el3 |= SCR_ENTP2;
775                         env->vfp.smcr_el[3] = 0xf;
776                     }
777                     if (cpu_isar_feature(aa64_hcx, cpu)) {
778                         env->cp15.scr_el3 |= SCR_HXEN;
779                     }
780                     /* AArch64 kernels never boot in secure mode */
781                     assert(!info->secure_boot);
782                     /* This hook is only supported for AArch32 currently:
783                      * bootloader_aarch64[] will not call the hook, and
784                      * the code above has already dropped us into EL2 or EL1.
785                      */
786                     assert(!info->secure_board_setup);
787                 }
788 
789                 if (arm_feature(env, ARM_FEATURE_EL2)) {
790                     /* If we have EL2 then Linux expects the HVC insn to work */
791                     env->cp15.scr_el3 |= SCR_HCE;
792                 }
793 
794                 /* Set to non-secure if not a secure boot */
795                 if (!info->secure_boot &&
796                     (cs != first_cpu || !info->secure_board_setup)) {
797                     /* Linux expects non-secure state */
798                     env->cp15.scr_el3 |= SCR_NS;
799                     /* Set NSACR.{CP11,CP10} so NS can access the FPU */
800                     env->cp15.nsacr |= 3 << 10;
801                 }
802             }
803 
804             if (!env->aarch64 && !info->secure_boot &&
805                 arm_feature(env, ARM_FEATURE_EL2)) {
806                 /*
807                  * This is an AArch32 boot not to Secure state, and
808                  * we have Hyp mode available, so boot the kernel into
809                  * Hyp mode. This is not how the CPU comes out of reset,
810                  * so we need to manually put it there.
811                  */
812                 cpsr_write(env, ARM_CPU_MODE_HYP, CPSR_M, CPSRWriteRaw);
813             }
814 
815             if (cs == first_cpu) {
816                 AddressSpace *as = arm_boot_address_space(cpu, info);
817 
818                 cpu_set_pc(cs, info->loader_start);
819 
820                 if (!have_dtb(info)) {
821                     if (old_param) {
822                         set_kernel_args_old(info, as);
823                     } else {
824                         set_kernel_args(info, as);
825                     }
826                 }
827             } else if (info->secondary_cpu_reset_hook) {
828                 info->secondary_cpu_reset_hook(cpu, info);
829             }
830         }
831 
832         if (tcg_enabled()) {
833             arm_rebuild_hflags(env);
834         }
835     }
836 }
837 
838 static int do_arm_linux_init(Object *obj, void *opaque)
839 {
840     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
841         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
842         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
843         struct arm_boot_info *info = opaque;
844 
845         if (albifc->arm_linux_init) {
846             albifc->arm_linux_init(albif, info->secure_boot);
847         }
848     }
849     return 0;
850 }
851 
852 static ssize_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
853                             uint64_t *lowaddr, uint64_t *highaddr,
854                             int elf_machine, AddressSpace *as)
855 {
856     bool elf_is64;
857     union {
858         Elf32_Ehdr h32;
859         Elf64_Ehdr h64;
860     } elf_header;
861     int data_swab = 0;
862     bool big_endian;
863     ssize_t ret = -1;
864     Error *err = NULL;
865 
866 
867     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
868     if (err) {
869         error_free(err);
870         return ret;
871     }
872 
873     if (elf_is64) {
874         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
875         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
876                                       : ARM_ENDIANNESS_LE;
877     } else {
878         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
879         if (big_endian) {
880             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
881                 info->endianness = ARM_ENDIANNESS_BE8;
882             } else {
883                 info->endianness = ARM_ENDIANNESS_BE32;
884                 /* In BE32, the CPU has a different view of the per-byte
885                  * address map than the rest of the system. BE32 ELF files
886                  * are organised such that they can be programmed through
887                  * the CPU's per-word byte-reversed view of the world. QEMU
888                  * however loads ELF files independently of the CPU. So
889                  * tell the ELF loader to byte reverse the data for us.
890                  */
891                 data_swab = 2;
892             }
893         } else {
894             info->endianness = ARM_ENDIANNESS_LE;
895         }
896     }
897 
898     ret = load_elf_as(info->kernel_filename, NULL, NULL, NULL,
899                       pentry, lowaddr, highaddr, NULL, big_endian, elf_machine,
900                       1, data_swab, as);
901     if (ret <= 0) {
902         /* The header loaded but the image didn't */
903         exit(1);
904     }
905 
906     return ret;
907 }
908 
909 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
910                                    hwaddr *entry, AddressSpace *as)
911 {
912     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
913     uint64_t kernel_size = 0;
914     uint8_t *buffer;
915     int size;
916 
917     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
918     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
919                                      &buffer);
920 
921     if (size < 0) {
922         gsize len;
923 
924         /* Load as raw file otherwise */
925         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
926             return -1;
927         }
928         size = len;
929     }
930 
931     /* check the arm64 magic header value -- very old kernels may not have it */
932     if (size > ARM64_MAGIC_OFFSET + 4 &&
933         memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
934         uint64_t hdrvals[2];
935 
936         /* The arm64 Image header has text_offset and image_size fields at 8 and
937          * 16 bytes into the Image header, respectively. The text_offset field
938          * is only valid if the image_size is non-zero.
939          */
940         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
941 
942         kernel_size = le64_to_cpu(hdrvals[1]);
943 
944         if (kernel_size != 0) {
945             kernel_load_offset = le64_to_cpu(hdrvals[0]);
946 
947             /*
948              * We write our startup "bootloader" at the very bottom of RAM,
949              * so that bit can't be used for the image. Luckily the Image
950              * format specification is that the image requests only an offset
951              * from a 2MB boundary, not an absolute load address. So if the
952              * image requests an offset that might mean it overlaps with the
953              * bootloader, we can just load it starting at 2MB+offset rather
954              * than 0MB + offset.
955              */
956             if (kernel_load_offset < BOOTLOADER_MAX_SIZE) {
957                 kernel_load_offset += 2 * MiB;
958             }
959         }
960     }
961 
962     /*
963      * Kernels before v3.17 don't populate the image_size field, and
964      * raw images have no header. For those our best guess at the size
965      * is the size of the Image file itself.
966      */
967     if (kernel_size == 0) {
968         kernel_size = size;
969     }
970 
971     *entry = mem_base + kernel_load_offset;
972     rom_add_blob_fixed_as(filename, buffer, size, *entry, as);
973 
974     g_free(buffer);
975 
976     return kernel_size;
977 }
978 
979 static void arm_setup_direct_kernel_boot(ARMCPU *cpu,
980                                          struct arm_boot_info *info)
981 {
982     /* Set up for a direct boot of a kernel image file. */
983     CPUState *cs;
984     AddressSpace *as = arm_boot_address_space(cpu, info);
985     ssize_t kernel_size;
986     int initrd_size;
987     int is_linux = 0;
988     uint64_t elf_entry;
989     /* Addresses of first byte used and first byte not used by the image */
990     uint64_t image_low_addr = 0, image_high_addr = 0;
991     int elf_machine;
992     hwaddr entry;
993     static const ARMInsnFixup *primary_loader;
994     uint64_t ram_end = info->loader_start + info->ram_size;
995 
996     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
997         primary_loader = bootloader_aarch64;
998         elf_machine = EM_AARCH64;
999     } else {
1000         primary_loader = bootloader;
1001         if (!info->write_board_setup) {
1002             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
1003         }
1004         elf_machine = EM_ARM;
1005     }
1006 
1007     /* Assume that raw images are linux kernels, and ELF images are not.  */
1008     kernel_size = arm_load_elf(info, &elf_entry, &image_low_addr,
1009                                &image_high_addr, elf_machine, as);
1010     if (kernel_size > 0 && have_dtb(info)) {
1011         /*
1012          * If there is still some room left at the base of RAM, try and put
1013          * the DTB there like we do for images loaded with -bios or -pflash.
1014          */
1015         if (image_low_addr > info->loader_start
1016             || image_high_addr < info->loader_start) {
1017             /*
1018              * Set image_low_addr as address limit for arm_load_dtb if it may be
1019              * pointing into RAM, otherwise pass '0' (no limit)
1020              */
1021             if (image_low_addr < info->loader_start) {
1022                 image_low_addr = 0;
1023             }
1024             info->dtb_start = info->loader_start;
1025             info->dtb_limit = image_low_addr;
1026         }
1027     }
1028     entry = elf_entry;
1029     if (kernel_size < 0) {
1030         uint64_t loadaddr = info->loader_start + KERNEL_NOLOAD_ADDR;
1031         kernel_size = load_uimage_as(info->kernel_filename, &entry, &loadaddr,
1032                                      &is_linux, NULL, NULL, as);
1033         if (kernel_size >= 0) {
1034             image_low_addr = loadaddr;
1035             image_high_addr = image_low_addr + kernel_size;
1036         }
1037     }
1038     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1039         kernel_size = load_aarch64_image(info->kernel_filename,
1040                                          info->loader_start, &entry, as);
1041         is_linux = 1;
1042         if (kernel_size >= 0) {
1043             image_low_addr = entry;
1044             image_high_addr = image_low_addr + kernel_size;
1045         }
1046     } else if (kernel_size < 0) {
1047         /* 32-bit ARM */
1048         entry = info->loader_start + KERNEL_LOAD_ADDR;
1049         kernel_size = load_image_targphys_as(info->kernel_filename, entry,
1050                                              ram_end - KERNEL_LOAD_ADDR, as);
1051         is_linux = 1;
1052         if (kernel_size >= 0) {
1053             image_low_addr = entry;
1054             image_high_addr = image_low_addr + kernel_size;
1055         }
1056     }
1057     if (kernel_size < 0) {
1058         error_report("could not load kernel '%s'", info->kernel_filename);
1059         exit(1);
1060     }
1061 
1062     if (kernel_size > info->ram_size) {
1063         error_report("kernel '%s' is too large to fit in RAM "
1064                      "(kernel size %zd, RAM size %" PRId64 ")",
1065                      info->kernel_filename, kernel_size, info->ram_size);
1066         exit(1);
1067     }
1068 
1069     info->entry = entry;
1070 
1071     /*
1072      * We want to put the initrd far enough into RAM that when the
1073      * kernel is uncompressed it will not clobber the initrd. However
1074      * on boards without much RAM we must ensure that we still leave
1075      * enough room for a decent sized initrd, and on boards with large
1076      * amounts of RAM we must avoid the initrd being so far up in RAM
1077      * that it is outside lowmem and inaccessible to the kernel.
1078      * So for boards with less  than 256MB of RAM we put the initrd
1079      * halfway into RAM, and for boards with 256MB of RAM or more we put
1080      * the initrd at 128MB.
1081      * We also refuse to put the initrd somewhere that will definitely
1082      * overlay the kernel we just loaded, though for kernel formats which
1083      * don't tell us their exact size (eg self-decompressing 32-bit kernels)
1084      * we might still make a bad choice here.
1085      */
1086     info->initrd_start = info->loader_start +
1087         MIN(info->ram_size / 2, 128 * MiB);
1088     if (image_high_addr) {
1089         info->initrd_start = MAX(info->initrd_start, image_high_addr);
1090     }
1091     info->initrd_start = TARGET_PAGE_ALIGN(info->initrd_start);
1092 
1093     if (is_linux) {
1094         uint32_t fixupcontext[FIXUP_MAX];
1095 
1096         if (info->initrd_filename) {
1097 
1098             if (info->initrd_start >= ram_end) {
1099                 error_report("not enough space after kernel to load initrd");
1100                 exit(1);
1101             }
1102 
1103             initrd_size = load_ramdisk_as(info->initrd_filename,
1104                                           info->initrd_start,
1105                                           ram_end - info->initrd_start, as);
1106             if (initrd_size < 0) {
1107                 initrd_size = load_image_targphys_as(info->initrd_filename,
1108                                                      info->initrd_start,
1109                                                      ram_end -
1110                                                      info->initrd_start,
1111                                                      as);
1112             }
1113             if (initrd_size < 0) {
1114                 error_report("could not load initrd '%s'",
1115                              info->initrd_filename);
1116                 exit(1);
1117             }
1118             if (info->initrd_start + initrd_size > ram_end) {
1119                 error_report("could not load initrd '%s': "
1120                              "too big to fit into RAM after the kernel",
1121                              info->initrd_filename);
1122                 exit(1);
1123             }
1124         } else {
1125             initrd_size = 0;
1126         }
1127         info->initrd_size = initrd_size;
1128 
1129         fixupcontext[FIXUP_BOARDID] = info->board_id;
1130         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1131 
1132         /*
1133          * for device tree boot, we pass the DTB directly in r2. Otherwise
1134          * we point to the kernel args.
1135          */
1136         if (have_dtb(info)) {
1137             hwaddr align;
1138 
1139             if (elf_machine == EM_AARCH64) {
1140                 /*
1141                  * Some AArch64 kernels on early bootup map the fdt region as
1142                  *
1143                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1144                  *
1145                  * Let's play safe and prealign it to 2MB to give us some space.
1146                  */
1147                 align = 2 * MiB;
1148             } else {
1149                 /*
1150                  * Some 32bit kernels will trash anything in the 4K page the
1151                  * initrd ends in, so make sure the DTB isn't caught up in that.
1152                  */
1153                 align = 4 * KiB;
1154             }
1155 
1156             /* Place the DTB after the initrd in memory with alignment. */
1157             info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size,
1158                                            align);
1159             if (info->dtb_start >= ram_end) {
1160                 error_report("Not enough space for DTB after kernel/initrd");
1161                 exit(1);
1162             }
1163             fixupcontext[FIXUP_ARGPTR_LO] = info->dtb_start;
1164             fixupcontext[FIXUP_ARGPTR_HI] = info->dtb_start >> 32;
1165         } else {
1166             fixupcontext[FIXUP_ARGPTR_LO] =
1167                 info->loader_start + KERNEL_ARGS_ADDR;
1168             fixupcontext[FIXUP_ARGPTR_HI] =
1169                 (info->loader_start + KERNEL_ARGS_ADDR) >> 32;
1170             if (info->ram_size >= 4 * GiB) {
1171                 error_report("RAM size must be less than 4GB to boot"
1172                              " Linux kernel using ATAGS (try passing a device tree"
1173                              " using -dtb)");
1174                 exit(1);
1175             }
1176         }
1177         fixupcontext[FIXUP_ENTRYPOINT_LO] = entry;
1178         fixupcontext[FIXUP_ENTRYPOINT_HI] = entry >> 32;
1179 
1180         write_bootloader("bootloader", info->loader_start,
1181                          primary_loader, fixupcontext, as);
1182 
1183         if (info->write_board_setup) {
1184             info->write_board_setup(cpu, info);
1185         }
1186 
1187         /*
1188          * Notify devices which need to fake up firmware initialization
1189          * that we're doing a direct kernel boot.
1190          */
1191         object_child_foreach_recursive(object_get_root(),
1192                                        do_arm_linux_init, info);
1193     }
1194     info->is_linux = is_linux;
1195 
1196     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1197         ARM_CPU(cs)->env.boot_info = info;
1198     }
1199 }
1200 
1201 static void arm_setup_firmware_boot(ARMCPU *cpu, struct arm_boot_info *info)
1202 {
1203     /* Set up for booting firmware (which might load a kernel via fw_cfg) */
1204 
1205     if (have_dtb(info)) {
1206         /*
1207          * If we have a device tree blob, but no kernel to supply it to (or
1208          * the kernel is supposed to be loaded by the bootloader), copy the
1209          * DTB to the base of RAM for the bootloader to pick up.
1210          */
1211         info->dtb_start = info->loader_start;
1212     }
1213 
1214     if (info->kernel_filename) {
1215         FWCfgState *fw_cfg;
1216         bool try_decompressing_kernel;
1217 
1218         fw_cfg = fw_cfg_find();
1219 
1220         if (!fw_cfg) {
1221             error_report("This machine type does not support loading both "
1222                          "a guest firmware/BIOS image and a guest kernel at "
1223                          "the same time. You should change your QEMU command "
1224                          "line to specify one or the other, but not both.");
1225             exit(1);
1226         }
1227 
1228         try_decompressing_kernel = arm_feature(&cpu->env,
1229                                                ARM_FEATURE_AARCH64);
1230 
1231         /*
1232          * Expose the kernel, the command line, and the initrd in fw_cfg.
1233          * We don't process them here at all, it's all left to the
1234          * firmware.
1235          */
1236         load_image_to_fw_cfg(fw_cfg,
1237                              FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
1238                              info->kernel_filename,
1239                              try_decompressing_kernel);
1240         load_image_to_fw_cfg(fw_cfg,
1241                              FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
1242                              info->initrd_filename, false);
1243 
1244         if (info->kernel_cmdline) {
1245             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
1246                            strlen(info->kernel_cmdline) + 1);
1247             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
1248                               info->kernel_cmdline);
1249         }
1250     }
1251 
1252     /*
1253      * We will start from address 0 (typically a boot ROM image) in the
1254      * same way as hardware. Leave env->boot_info NULL, so that
1255      * do_cpu_reset() knows it does not need to alter the PC on reset.
1256      */
1257 }
1258 
1259 void arm_load_kernel(ARMCPU *cpu, MachineState *ms, struct arm_boot_info *info)
1260 {
1261     CPUState *cs;
1262     AddressSpace *as = arm_boot_address_space(cpu, info);
1263     int boot_el;
1264     CPUARMState *env = &cpu->env;
1265     int nb_cpus = 0;
1266 
1267     /*
1268      * CPU objects (unlike devices) are not automatically reset on system
1269      * reset, so we must always register a handler to do so. If we're
1270      * actually loading a kernel, the handler is also responsible for
1271      * arranging that we start it correctly.
1272      */
1273     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1274         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
1275         nb_cpus++;
1276     }
1277 
1278     /*
1279      * The board code is not supposed to set secure_board_setup unless
1280      * running its code in secure mode is actually possible, and KVM
1281      * doesn't support secure.
1282      */
1283     assert(!(info->secure_board_setup && kvm_enabled()));
1284     info->kernel_filename = ms->kernel_filename;
1285     info->kernel_cmdline = ms->kernel_cmdline;
1286     info->initrd_filename = ms->initrd_filename;
1287     info->dtb_filename = ms->dtb;
1288     info->dtb_limit = 0;
1289 
1290     /* Load the kernel.  */
1291     if (!info->kernel_filename || info->firmware_loaded) {
1292         arm_setup_firmware_boot(cpu, info);
1293     } else {
1294         arm_setup_direct_kernel_boot(cpu, info);
1295     }
1296 
1297     /*
1298      * Disable the PSCI conduit if it is set up to target the same
1299      * or a lower EL than the one we're going to start the guest code in.
1300      * This logic needs to agree with the code in do_cpu_reset() which
1301      * decides whether we're going to boot the guest in the highest
1302      * supported exception level or in a lower one.
1303      */
1304 
1305     /*
1306      * If PSCI is enabled, then SMC calls all go to the PSCI handler and
1307      * are never emulated to trap into guest code. It therefore does not
1308      * make sense for the board to have a setup code fragment that runs
1309      * in Secure, because this will probably need to itself issue an SMC of some
1310      * kind as part of its operation.
1311      */
1312     assert(info->psci_conduit == QEMU_PSCI_CONDUIT_DISABLED ||
1313            !info->secure_board_setup);
1314 
1315     /* Boot into highest supported EL ... */
1316     if (arm_feature(env, ARM_FEATURE_EL3)) {
1317         boot_el = 3;
1318     } else if (arm_feature(env, ARM_FEATURE_EL2)) {
1319         boot_el = 2;
1320     } else {
1321         boot_el = 1;
1322     }
1323     /* ...except that if we're booting Linux we adjust the EL we boot into */
1324     if (info->is_linux && !info->secure_boot) {
1325         boot_el = arm_feature(env, ARM_FEATURE_EL2) ? 2 : 1;
1326     }
1327 
1328     if ((info->psci_conduit == QEMU_PSCI_CONDUIT_HVC && boot_el >= 2) ||
1329         (info->psci_conduit == QEMU_PSCI_CONDUIT_SMC && boot_el == 3)) {
1330         info->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
1331     }
1332 
1333     if (info->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) {
1334         for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1335             Object *cpuobj = OBJECT(cs);
1336 
1337             object_property_set_int(cpuobj, "psci-conduit", info->psci_conduit,
1338                                     &error_abort);
1339             /*
1340              * Secondary CPUs start in PSCI powered-down state. Like the
1341              * code in do_cpu_reset(), we assume first_cpu is the primary
1342              * CPU.
1343              */
1344             if (cs != first_cpu) {
1345                 object_property_set_bool(cpuobj, "start-powered-off", true,
1346                                          &error_abort);
1347             }
1348         }
1349     }
1350 
1351     if (info->psci_conduit == QEMU_PSCI_CONDUIT_DISABLED &&
1352         info->is_linux && nb_cpus > 1) {
1353         /*
1354          * We're booting Linux but not using PSCI, so for SMP we need
1355          * to write a custom secondary CPU boot loader stub, and arrange
1356          * for the secondary CPU reset to make the accompanying initialization.
1357          */
1358         if (!info->secondary_cpu_reset_hook) {
1359             info->secondary_cpu_reset_hook = default_reset_secondary;
1360         }
1361         if (!info->write_secondary_boot) {
1362             info->write_secondary_boot = default_write_secondary;
1363         }
1364         info->write_secondary_boot(cpu, info);
1365     } else {
1366         /*
1367          * No secondary boot stub; don't use the reset hook that would
1368          * have set the CPU up to call it
1369          */
1370         info->write_secondary_boot = NULL;
1371         info->secondary_cpu_reset_hook = NULL;
1372     }
1373 
1374     /*
1375      * arm_load_dtb() may add a PSCI node so it must be called after we have
1376      * decided whether to enable PSCI and set the psci-conduit CPU properties.
1377      */
1378     if (!info->skip_dtb_autoload && have_dtb(info)) {
1379         if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms) < 0) {
1380             exit(1);
1381         }
1382     }
1383 }
1384 
1385 static const TypeInfo arm_linux_boot_if_info = {
1386     .name = TYPE_ARM_LINUX_BOOT_IF,
1387     .parent = TYPE_INTERFACE,
1388     .class_size = sizeof(ARMLinuxBootIfClass),
1389 };
1390 
1391 static void arm_linux_boot_register_types(void)
1392 {
1393     type_register_static(&arm_linux_boot_if_info);
1394 }
1395 
1396 type_init(arm_linux_boot_register_types)
1397