Merge tag 'v3.1.0-rc0'
[qemu/ar7.git] / hw / arm / boot.c
blob269fe6dacc9f3eee7d66d665c3becf2140de2c6c
1 /*
2 * ARM kernel loader.
4 * Copyright (c) 2006-2007 CodeSourcery.
5 * Written by Paul Brook
7 * This code is licensed under the GPL.
8 */
10 #include "qemu/osdep.h"
11 #include "qemu/error-report.h"
12 #include "qapi/error.h"
13 #include <libfdt.h>
14 #include "hw/hw.h"
15 #include "hw/arm/arm.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 "hw/loader.h"
22 #include "elf.h"
23 #include "sysemu/device_tree.h"
24 #include "qemu/config-file.h"
25 #include "qemu/option.h"
26 #include "exec/address-spaces.h"
27 #include "qemu/units.h"
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.
33 #define KERNEL_ARGS_ADDR 0x100
34 #define KERNEL_LOAD_ADDR 0x00010000
35 #define KERNEL64_LOAD_ADDR 0x00080000
37 #define ARM64_TEXT_OFFSET_OFFSET 8
38 #define ARM64_MAGIC_OFFSET 56
40 #define BOOTLOADER_MAX_SIZE (4 * KiB)
42 AddressSpace *arm_boot_address_space(ARMCPU *cpu,
43 const struct arm_boot_info *info)
45 /* Return the address space to use for bootloader reads and writes.
46 * We prefer the secure address space if the CPU has it and we're
47 * going to boot the guest into it.
49 int asidx;
50 CPUState *cs = CPU(cpu);
52 if (arm_feature(&cpu->env, ARM_FEATURE_EL3) && info->secure_boot) {
53 asidx = ARMASIdx_S;
54 } else {
55 asidx = ARMASIdx_NS;
58 return cpu_get_address_space(cs, asidx);
61 typedef enum {
62 FIXUP_NONE = 0, /* do nothing */
63 FIXUP_TERMINATOR, /* end of insns */
64 FIXUP_BOARDID, /* overwrite with board ID number */
65 FIXUP_BOARD_SETUP, /* overwrite with board specific setup code address */
66 FIXUP_ARGPTR, /* overwrite with pointer to kernel args */
67 FIXUP_ENTRYPOINT, /* overwrite with kernel entry point */
68 FIXUP_GIC_CPU_IF, /* overwrite with GIC CPU interface address */
69 FIXUP_BOOTREG, /* overwrite with boot register address */
70 FIXUP_DSB, /* overwrite with correct DSB insn for cpu */
71 FIXUP_MAX,
72 } FixupType;
74 typedef struct ARMInsnFixup {
75 uint32_t insn;
76 FixupType fixup;
77 } ARMInsnFixup;
79 static const ARMInsnFixup bootloader_aarch64[] = {
80 { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */
81 { 0xaa1f03e1 }, /* mov x1, xzr */
82 { 0xaa1f03e2 }, /* mov x2, xzr */
83 { 0xaa1f03e3 }, /* mov x3, xzr */
84 { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */
85 { 0xd61f0080 }, /* br x4 ; Jump to the kernel entry point */
86 { 0, FIXUP_ARGPTR }, /* arg: .word @DTB Lower 32-bits */
87 { 0 }, /* .word @DTB Higher 32-bits */
88 { 0, FIXUP_ENTRYPOINT }, /* entry: .word @Kernel Entry Lower 32-bits */
89 { 0 }, /* .word @Kernel Entry Higher 32-bits */
90 { 0, FIXUP_TERMINATOR }
93 /* A very small bootloader: call the board-setup code (if needed),
94 * set r0-r2, then jump to the kernel.
95 * If we're not calling boot setup code then we don't copy across
96 * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array.
99 static const ARMInsnFixup bootloader[] = {
100 { 0xe28fe004 }, /* add lr, pc, #4 */
101 { 0xe51ff004 }, /* ldr pc, [pc, #-4] */
102 { 0, FIXUP_BOARD_SETUP },
103 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3
104 { 0xe3a00000 }, /* mov r0, #0 */
105 { 0xe59f1004 }, /* ldr r1, [pc, #4] */
106 { 0xe59f2004 }, /* ldr r2, [pc, #4] */
107 { 0xe59ff004 }, /* ldr pc, [pc, #4] */
108 { 0, FIXUP_BOARDID },
109 { 0, FIXUP_ARGPTR },
110 { 0, FIXUP_ENTRYPOINT },
111 { 0, FIXUP_TERMINATOR }
114 /* Handling for secondary CPU boot in a multicore system.
115 * Unlike the uniprocessor/primary CPU boot, this is platform
116 * dependent. The default code here is based on the secondary
117 * CPU boot protocol used on realview/vexpress boards, with
118 * some parameterisation to increase its flexibility.
119 * QEMU platform models for which this code is not appropriate
120 * should override write_secondary_boot and secondary_cpu_reset_hook
121 * instead.
123 * This code enables the interrupt controllers for the secondary
124 * CPUs and then puts all the secondary CPUs into a loop waiting
125 * for an interprocessor interrupt and polling a configurable
126 * location for the kernel secondary CPU entry point.
128 #define DSB_INSN 0xf57ff04f
129 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
131 static const ARMInsnFixup smpboot[] = {
132 { 0xe59f2028 }, /* ldr r2, gic_cpu_if */
133 { 0xe59f0028 }, /* ldr r0, bootreg_addr */
134 { 0xe3a01001 }, /* mov r1, #1 */
135 { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */
136 { 0xe3a010ff }, /* mov r1, #0xff */
137 { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */
138 { 0, FIXUP_DSB }, /* dsb */
139 { 0xe320f003 }, /* wfi */
140 { 0xe5901000 }, /* ldr r1, [r0] */
141 { 0xe1110001 }, /* tst r1, r1 */
142 { 0x0afffffb }, /* beq <wfi> */
143 { 0xe12fff11 }, /* bx r1 */
144 { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */
145 { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */
146 { 0, FIXUP_TERMINATOR }
149 static void write_bootloader(const char *name, hwaddr addr,
150 const ARMInsnFixup *insns, uint32_t *fixupcontext,
151 AddressSpace *as)
153 /* Fix up the specified bootloader fragment and write it into
154 * guest memory using rom_add_blob_fixed(). fixupcontext is
155 * an array giving the values to write in for the fixup types
156 * which write a value into the code array.
158 int i, len;
159 uint32_t *code;
161 len = 0;
162 while (insns[len].fixup != FIXUP_TERMINATOR) {
163 len++;
166 code = g_new0(uint32_t, len);
168 for (i = 0; i < len; i++) {
169 uint32_t insn = insns[i].insn;
170 FixupType fixup = insns[i].fixup;
172 switch (fixup) {
173 case FIXUP_NONE:
174 break;
175 case FIXUP_BOARDID:
176 case FIXUP_BOARD_SETUP:
177 case FIXUP_ARGPTR:
178 case FIXUP_ENTRYPOINT:
179 case FIXUP_GIC_CPU_IF:
180 case FIXUP_BOOTREG:
181 case FIXUP_DSB:
182 insn = fixupcontext[fixup];
183 break;
184 default:
185 abort();
187 code[i] = tswap32(insn);
190 assert((len * sizeof(uint32_t)) < BOOTLOADER_MAX_SIZE);
192 rom_add_blob_fixed_as(name, code, len * sizeof(uint32_t), addr, as);
194 g_free(code);
197 static void default_write_secondary(ARMCPU *cpu,
198 const struct arm_boot_info *info)
200 uint32_t fixupcontext[FIXUP_MAX];
201 AddressSpace *as = arm_boot_address_space(cpu, info);
203 fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr;
204 fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr;
205 if (arm_feature(&cpu->env, ARM_FEATURE_V7)) {
206 fixupcontext[FIXUP_DSB] = DSB_INSN;
207 } else {
208 fixupcontext[FIXUP_DSB] = CP15_DSB_INSN;
211 write_bootloader("smpboot", info->smp_loader_start,
212 smpboot, fixupcontext, as);
215 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
216 const struct arm_boot_info *info,
217 hwaddr mvbar_addr)
219 AddressSpace *as = arm_boot_address_space(cpu, info);
220 int n;
221 uint32_t mvbar_blob[] = {
222 /* mvbar_addr: secure monitor vectors
223 * Default unimplemented and unused vectors to spin. Makes it
224 * easier to debug (as opposed to the CPU running away).
226 0xeafffffe, /* (spin) */
227 0xeafffffe, /* (spin) */
228 0xe1b0f00e, /* movs pc, lr ;SMC exception return */
229 0xeafffffe, /* (spin) */
230 0xeafffffe, /* (spin) */
231 0xeafffffe, /* (spin) */
232 0xeafffffe, /* (spin) */
233 0xeafffffe, /* (spin) */
235 uint32_t board_setup_blob[] = {
236 /* board setup addr */
237 0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */
238 0xee0c0f30, /* mcr p15, 0, r0, c12, c0, 1 ;set MVBAR */
239 0xee110f11, /* mrc p15, 0, r0, c1 , c1, 0 ;read SCR */
240 0xe3800031, /* orr r0, #0x31 ;enable AW, FW, NS */
241 0xee010f11, /* mcr p15, 0, r0, c1, c1, 0 ;write SCR */
242 0xe1a0100e, /* mov r1, lr ;save LR across SMC */
243 0xe1600070, /* smc #0 ;call monitor to flush SCR */
244 0xe1a0f001, /* mov pc, r1 ;return */
247 /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
248 assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
250 /* check that these blobs don't overlap */
251 assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr)
252 || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr));
254 for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
255 mvbar_blob[n] = tswap32(mvbar_blob[n]);
257 rom_add_blob_fixed_as("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
258 mvbar_addr, as);
260 for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
261 board_setup_blob[n] = tswap32(board_setup_blob[n]);
263 rom_add_blob_fixed_as("board-setup", board_setup_blob,
264 sizeof(board_setup_blob), info->board_setup_addr, as);
267 static void default_reset_secondary(ARMCPU *cpu,
268 const struct arm_boot_info *info)
270 AddressSpace *as = arm_boot_address_space(cpu, info);
271 CPUState *cs = CPU(cpu);
273 address_space_stl_notdirty(as, info->smp_bootreg_addr,
274 0, MEMTXATTRS_UNSPECIFIED, NULL);
275 cpu_set_pc(cs, info->smp_loader_start);
278 static inline bool have_dtb(const struct arm_boot_info *info)
280 return info->dtb_filename || info->get_dtb;
283 #define WRITE_WORD(p, value) do { \
284 address_space_stl_notdirty(as, p, value, \
285 MEMTXATTRS_UNSPECIFIED, NULL); \
286 p += 4; \
287 } while (0)
289 static void set_kernel_args(const struct arm_boot_info *info, AddressSpace *as)
291 int initrd_size = info->initrd_size;
292 hwaddr base = info->loader_start;
293 hwaddr p;
295 p = base + KERNEL_ARGS_ADDR;
296 /* ATAG_CORE */
297 WRITE_WORD(p, 5);
298 WRITE_WORD(p, 0x54410001);
299 WRITE_WORD(p, 1);
300 WRITE_WORD(p, 0x1000);
301 WRITE_WORD(p, 0);
302 /* ATAG_MEM */
303 /* TODO: handle multiple chips on one ATAG list */
304 WRITE_WORD(p, 4);
305 WRITE_WORD(p, 0x54410002);
306 WRITE_WORD(p, info->ram_size);
307 WRITE_WORD(p, info->loader_start);
308 if (initrd_size) {
309 /* ATAG_INITRD2 */
310 WRITE_WORD(p, 4);
311 WRITE_WORD(p, 0x54420005);
312 WRITE_WORD(p, info->initrd_start);
313 WRITE_WORD(p, initrd_size);
315 if (info->atag_revision) {
316 /* ATAG REVISION. */
317 WRITE_WORD(p, 3);
318 WRITE_WORD(p, 0x54410007);
319 WRITE_WORD(p, info->atag_revision);
321 if (info->kernel_cmdline && *info->kernel_cmdline) {
322 /* ATAG_CMDLINE */
323 int cmdline_size;
325 cmdline_size = strlen(info->kernel_cmdline);
326 address_space_write(as, p + 8, MEMTXATTRS_UNSPECIFIED,
327 (const uint8_t *)info->kernel_cmdline,
328 cmdline_size + 1);
329 cmdline_size = (cmdline_size >> 2) + 1;
330 WRITE_WORD(p, cmdline_size + 2);
331 WRITE_WORD(p, 0x54410009);
332 p += cmdline_size * 4;
334 if (info->atag_board) {
335 /* ATAG_BOARD */
336 int atag_board_len;
337 uint8_t atag_board_buf[0x1000];
339 atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
340 WRITE_WORD(p, (atag_board_len + 8) >> 2);
341 WRITE_WORD(p, 0x414f4d50);
342 address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
343 atag_board_buf, atag_board_len);
344 p += atag_board_len;
346 /* ATAG_END */
347 WRITE_WORD(p, 0);
348 WRITE_WORD(p, 0);
351 static void set_kernel_args_old(const struct arm_boot_info *info,
352 AddressSpace *as)
354 hwaddr p;
355 const char *s;
356 int initrd_size = info->initrd_size;
357 hwaddr base = info->loader_start;
359 /* see linux/include/asm-arm/setup.h */
360 p = base + KERNEL_ARGS_ADDR;
361 /* page_size */
362 WRITE_WORD(p, 4096);
363 /* nr_pages */
364 WRITE_WORD(p, info->ram_size / 4096);
365 /* ramdisk_size */
366 WRITE_WORD(p, 0);
367 #define FLAG_READONLY 1
368 #define FLAG_RDLOAD 4
369 #define FLAG_RDPROMPT 8
370 /* flags */
371 WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
372 /* rootdev */
373 WRITE_WORD(p, (31 << 8) | 0); /* /dev/mtdblock0 */
374 /* video_num_cols */
375 WRITE_WORD(p, 0);
376 /* video_num_rows */
377 WRITE_WORD(p, 0);
378 /* video_x */
379 WRITE_WORD(p, 0);
380 /* video_y */
381 WRITE_WORD(p, 0);
382 /* memc_control_reg */
383 WRITE_WORD(p, 0);
384 /* unsigned char sounddefault */
385 /* unsigned char adfsdrives */
386 /* unsigned char bytes_per_char_h */
387 /* unsigned char bytes_per_char_v */
388 WRITE_WORD(p, 0);
389 /* pages_in_bank[4] */
390 WRITE_WORD(p, 0);
391 WRITE_WORD(p, 0);
392 WRITE_WORD(p, 0);
393 WRITE_WORD(p, 0);
394 /* pages_in_vram */
395 WRITE_WORD(p, 0);
396 /* initrd_start */
397 if (initrd_size) {
398 WRITE_WORD(p, info->initrd_start);
399 } else {
400 WRITE_WORD(p, 0);
402 /* initrd_size */
403 WRITE_WORD(p, initrd_size);
404 /* rd_start */
405 WRITE_WORD(p, 0);
406 /* system_rev */
407 WRITE_WORD(p, 0);
408 /* system_serial_low */
409 WRITE_WORD(p, 0);
410 /* system_serial_high */
411 WRITE_WORD(p, 0);
412 /* mem_fclk_21285 */
413 WRITE_WORD(p, 0);
414 /* zero unused fields */
415 while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
416 WRITE_WORD(p, 0);
418 s = info->kernel_cmdline;
419 if (s) {
420 address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
421 (const uint8_t *)s, strlen(s) + 1);
422 } else {
423 WRITE_WORD(p, 0);
427 static void fdt_add_psci_node(void *fdt)
429 uint32_t cpu_suspend_fn;
430 uint32_t cpu_off_fn;
431 uint32_t cpu_on_fn;
432 uint32_t migrate_fn;
433 ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
434 const char *psci_method;
435 int64_t psci_conduit;
436 int rc;
438 psci_conduit = object_property_get_int(OBJECT(armcpu),
439 "psci-conduit",
440 &error_abort);
441 switch (psci_conduit) {
442 case QEMU_PSCI_CONDUIT_DISABLED:
443 return;
444 case QEMU_PSCI_CONDUIT_HVC:
445 psci_method = "hvc";
446 break;
447 case QEMU_PSCI_CONDUIT_SMC:
448 psci_method = "smc";
449 break;
450 default:
451 g_assert_not_reached();
455 * If /psci node is present in provided DTB, assume that no fixup
456 * is necessary and all PSCI configuration should be taken as-is
458 rc = fdt_path_offset(fdt, "/psci");
459 if (rc >= 0) {
460 return;
463 qemu_fdt_add_subnode(fdt, "/psci");
464 if (armcpu->psci_version == 2) {
465 const char comp[] = "arm,psci-0.2\0arm,psci";
466 qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
468 cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
469 if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
470 cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
471 cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
472 migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
473 } else {
474 cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
475 cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
476 migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
478 } else {
479 qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
481 cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
482 cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
483 cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
484 migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
487 /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
488 * to the instruction that should be used to invoke PSCI functions.
489 * However, the device tree binding uses 'method' instead, so that is
490 * what we should use here.
492 qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
494 qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
495 qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
496 qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
497 qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
500 int arm_load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
501 hwaddr addr_limit, AddressSpace *as)
503 void *fdt = NULL;
504 int size, rc, n = 0;
505 uint32_t acells, scells;
506 char *nodename;
507 unsigned int i;
508 hwaddr mem_base, mem_len;
509 char **node_path;
510 Error *err = NULL;
512 if (binfo->dtb_filename) {
513 char *filename;
514 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
515 if (!filename) {
516 fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
517 goto fail;
520 fdt = load_device_tree(filename, &size);
521 if (!fdt) {
522 fprintf(stderr, "Couldn't open dtb file %s\n", filename);
523 g_free(filename);
524 goto fail;
526 g_free(filename);
527 } else {
528 fdt = binfo->get_dtb(binfo, &size);
529 if (!fdt) {
530 fprintf(stderr, "Board was unable to create a dtb blob\n");
531 goto fail;
535 if (addr_limit > addr && size > (addr_limit - addr)) {
536 /* Installing the device tree blob at addr would exceed addr_limit.
537 * Whether this constitutes failure is up to the caller to decide,
538 * so just return 0 as size, i.e., no error.
540 g_free(fdt);
541 return 0;
544 acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
545 NULL, &error_fatal);
546 scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
547 NULL, &error_fatal);
548 if (acells == 0 || scells == 0) {
549 fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
550 goto fail;
553 if (scells < 2 && binfo->ram_size >= (1ULL << 32)) {
554 /* This is user error so deserves a friendlier error message
555 * than the failure of setprop_sized_cells would provide
557 fprintf(stderr, "qemu: dtb file not compatible with "
558 "RAM size > 4GB\n");
559 goto fail;
562 /* nop all root nodes matching /memory or /memory@unit-address */
563 node_path = qemu_fdt_node_unit_path(fdt, "memory", &err);
564 if (err) {
565 error_report_err(err);
566 goto fail;
568 while (node_path[n]) {
569 if (g_str_has_prefix(node_path[n], "/memory")) {
570 qemu_fdt_nop_node(fdt, node_path[n]);
572 n++;
574 g_strfreev(node_path);
576 if (nb_numa_nodes > 0) {
577 mem_base = binfo->loader_start;
578 for (i = 0; i < nb_numa_nodes; i++) {
579 mem_len = numa_info[i].node_mem;
580 nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
581 qemu_fdt_add_subnode(fdt, nodename);
582 qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
583 rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
584 acells, mem_base,
585 scells, mem_len);
586 if (rc < 0) {
587 fprintf(stderr, "couldn't set %s/reg for node %d\n", nodename,
589 goto fail;
592 qemu_fdt_setprop_cell(fdt, nodename, "numa-node-id", i);
593 mem_base += mem_len;
594 g_free(nodename);
596 } else {
597 nodename = g_strdup_printf("/memory@%" PRIx64, binfo->loader_start);
598 qemu_fdt_add_subnode(fdt, nodename);
599 qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
601 rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
602 acells, binfo->loader_start,
603 scells, binfo->ram_size);
604 if (rc < 0) {
605 fprintf(stderr, "couldn't set %s reg\n", nodename);
606 goto fail;
608 g_free(nodename);
611 rc = fdt_path_offset(fdt, "/chosen");
612 if (rc < 0) {
613 qemu_fdt_add_subnode(fdt, "/chosen");
616 if (binfo->kernel_cmdline && *binfo->kernel_cmdline) {
617 rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
618 binfo->kernel_cmdline);
619 if (rc < 0) {
620 fprintf(stderr, "couldn't set /chosen/bootargs\n");
621 goto fail;
625 if (binfo->initrd_size) {
626 rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
627 binfo->initrd_start);
628 if (rc < 0) {
629 fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
630 goto fail;
633 rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
634 binfo->initrd_start + binfo->initrd_size);
635 if (rc < 0) {
636 fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
637 goto fail;
641 fdt_add_psci_node(fdt);
643 if (binfo->modify_dtb) {
644 binfo->modify_dtb(binfo, fdt);
647 qemu_fdt_dumpdtb(fdt, size);
649 /* Put the DTB into the memory map as a ROM image: this will ensure
650 * the DTB is copied again upon reset, even if addr points into RAM.
652 rom_add_blob_fixed_as("dtb", fdt, size, addr, as);
654 g_free(fdt);
656 return size;
658 fail:
659 g_free(fdt);
660 return -1;
663 static void do_cpu_reset(void *opaque)
665 ARMCPU *cpu = opaque;
666 CPUState *cs = CPU(cpu);
667 CPUARMState *env = &cpu->env;
668 const struct arm_boot_info *info = env->boot_info;
670 cpu_reset(cs);
671 if (info) {
672 if (!info->is_linux) {
673 int i;
674 /* Jump to the entry point. */
675 uint64_t entry = info->entry;
677 switch (info->endianness) {
678 case ARM_ENDIANNESS_LE:
679 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
680 for (i = 1; i < 4; ++i) {
681 env->cp15.sctlr_el[i] &= ~SCTLR_EE;
683 env->uncached_cpsr &= ~CPSR_E;
684 break;
685 case ARM_ENDIANNESS_BE8:
686 env->cp15.sctlr_el[1] |= SCTLR_E0E;
687 for (i = 1; i < 4; ++i) {
688 env->cp15.sctlr_el[i] |= SCTLR_EE;
690 env->uncached_cpsr |= CPSR_E;
691 break;
692 case ARM_ENDIANNESS_BE32:
693 env->cp15.sctlr_el[1] |= SCTLR_B;
694 break;
695 case ARM_ENDIANNESS_UNKNOWN:
696 break; /* Board's decision */
697 default:
698 g_assert_not_reached();
701 if (!env->aarch64) {
702 env->thumb = info->entry & 1;
703 entry &= 0xfffffffe;
705 cpu_set_pc(cs, entry);
706 } else {
707 /* If we are booting Linux then we need to check whether we are
708 * booting into secure or non-secure state and adjust the state
709 * accordingly. Out of reset, ARM is defined to be in secure state
710 * (SCR.NS = 0), we change that here if non-secure boot has been
711 * requested.
713 if (arm_feature(env, ARM_FEATURE_EL3)) {
714 /* AArch64 is defined to come out of reset into EL3 if enabled.
715 * If we are booting Linux then we need to adjust our EL as
716 * Linux expects us to be in EL2 or EL1. AArch32 resets into
717 * SVC, which Linux expects, so no privilege/exception level to
718 * adjust.
720 if (env->aarch64) {
721 env->cp15.scr_el3 |= SCR_RW;
722 if (arm_feature(env, ARM_FEATURE_EL2)) {
723 env->cp15.hcr_el2 |= HCR_RW;
724 env->pstate = PSTATE_MODE_EL2h;
725 } else {
726 env->pstate = PSTATE_MODE_EL1h;
728 /* AArch64 kernels never boot in secure mode */
729 assert(!info->secure_boot);
730 /* This hook is only supported for AArch32 currently:
731 * bootloader_aarch64[] will not call the hook, and
732 * the code above has already dropped us into EL2 or EL1.
734 assert(!info->secure_board_setup);
737 if (arm_feature(env, ARM_FEATURE_EL2)) {
738 /* If we have EL2 then Linux expects the HVC insn to work */
739 env->cp15.scr_el3 |= SCR_HCE;
742 /* Set to non-secure if not a secure boot */
743 if (!info->secure_boot &&
744 (cs != first_cpu || !info->secure_board_setup)) {
745 /* Linux expects non-secure state */
746 env->cp15.scr_el3 |= SCR_NS;
750 if (!env->aarch64 && !info->secure_boot &&
751 arm_feature(env, ARM_FEATURE_EL2)) {
753 * This is an AArch32 boot not to Secure state, and
754 * we have Hyp mode available, so boot the kernel into
755 * Hyp mode. This is not how the CPU comes out of reset,
756 * so we need to manually put it there.
758 cpsr_write(env, ARM_CPU_MODE_HYP, CPSR_M, CPSRWriteRaw);
761 if (cs == first_cpu) {
762 AddressSpace *as = arm_boot_address_space(cpu, info);
764 cpu_set_pc(cs, info->loader_start);
766 if (!have_dtb(info)) {
767 if (old_param) {
768 set_kernel_args_old(info, as);
769 } else {
770 set_kernel_args(info, as);
773 } else {
774 info->secondary_cpu_reset_hook(cpu, info);
781 * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
782 * by key.
783 * @fw_cfg: The firmware config instance to store the data in.
784 * @size_key: The firmware config key to store the size of the loaded
785 * data under, with fw_cfg_add_i32().
786 * @data_key: The firmware config key to store the loaded data under,
787 * with fw_cfg_add_bytes().
788 * @image_name: The name of the image file to load. If it is NULL, the
789 * function returns without doing anything.
790 * @try_decompress: Whether the image should be decompressed (gunzipped) before
791 * adding it to fw_cfg. If decompression fails, the image is
792 * loaded as-is.
794 * In case of failure, the function prints an error message to stderr and the
795 * process exits with status 1.
797 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
798 uint16_t data_key, const char *image_name,
799 bool try_decompress)
801 size_t size = -1;
802 uint8_t *data;
804 if (image_name == NULL) {
805 return;
808 if (try_decompress) {
809 size = load_image_gzipped_buffer(image_name,
810 LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
813 if (size == (size_t)-1) {
814 gchar *contents;
815 gsize length;
817 if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
818 error_report("failed to load \"%s\"", image_name);
819 exit(1);
821 size = length;
822 data = (uint8_t *)contents;
825 fw_cfg_add_i32(fw_cfg, size_key, size);
826 fw_cfg_add_bytes(fw_cfg, data_key, data, size);
829 static int do_arm_linux_init(Object *obj, void *opaque)
831 if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
832 ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
833 ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
834 struct arm_boot_info *info = opaque;
836 if (albifc->arm_linux_init) {
837 albifc->arm_linux_init(albif, info->secure_boot);
840 return 0;
843 static int64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
844 uint64_t *lowaddr, uint64_t *highaddr,
845 int elf_machine, AddressSpace *as)
847 bool elf_is64;
848 union {
849 Elf32_Ehdr h32;
850 Elf64_Ehdr h64;
851 } elf_header;
852 int data_swab = 0;
853 bool big_endian;
854 int64_t ret = -1;
855 Error *err = NULL;
858 load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
859 if (err) {
860 error_free(err);
861 return ret;
864 if (elf_is64) {
865 big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
866 info->endianness = big_endian ? ARM_ENDIANNESS_BE8
867 : ARM_ENDIANNESS_LE;
868 } else {
869 big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
870 if (big_endian) {
871 if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
872 info->endianness = ARM_ENDIANNESS_BE8;
873 } else {
874 info->endianness = ARM_ENDIANNESS_BE32;
875 /* In BE32, the CPU has a different view of the per-byte
876 * address map than the rest of the system. BE32 ELF files
877 * are organised such that they can be programmed through
878 * the CPU's per-word byte-reversed view of the world. QEMU
879 * however loads ELF files independently of the CPU. So
880 * tell the ELF loader to byte reverse the data for us.
882 data_swab = 2;
884 } else {
885 info->endianness = ARM_ENDIANNESS_LE;
889 ret = load_elf_as(info->kernel_filename, NULL, NULL,
890 pentry, lowaddr, highaddr, big_endian, elf_machine,
891 1, data_swab, as);
892 if (ret <= 0) {
893 /* The header loaded but the image didn't */
894 exit(1);
897 return ret;
900 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
901 hwaddr *entry, AddressSpace *as)
903 hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
904 uint8_t *buffer;
905 int size;
907 /* On aarch64, it's the bootloader's job to uncompress the kernel. */
908 size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
909 &buffer);
911 if (size < 0) {
912 gsize len;
914 /* Load as raw file otherwise */
915 if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
916 return -1;
918 size = len;
921 /* check the arm64 magic header value -- very old kernels may not have it */
922 if (size > ARM64_MAGIC_OFFSET + 4 &&
923 memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
924 uint64_t hdrvals[2];
926 /* The arm64 Image header has text_offset and image_size fields at 8 and
927 * 16 bytes into the Image header, respectively. The text_offset field
928 * is only valid if the image_size is non-zero.
930 memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
931 if (hdrvals[1] != 0) {
932 kernel_load_offset = le64_to_cpu(hdrvals[0]);
935 * We write our startup "bootloader" at the very bottom of RAM,
936 * so that bit can't be used for the image. Luckily the Image
937 * format specification is that the image requests only an offset
938 * from a 2MB boundary, not an absolute load address. So if the
939 * image requests an offset that might mean it overlaps with the
940 * bootloader, we can just load it starting at 2MB+offset rather
941 * than 0MB + offset.
943 if (kernel_load_offset < BOOTLOADER_MAX_SIZE) {
944 kernel_load_offset += 2 * MiB;
949 *entry = mem_base + kernel_load_offset;
950 rom_add_blob_fixed_as(filename, buffer, size, *entry, as);
952 g_free(buffer);
954 return size;
957 void arm_load_kernel(ARMCPU *cpu, struct arm_boot_info *info)
959 CPUState *cs;
960 int kernel_size;
961 int initrd_size;
962 int is_linux = 0;
963 uint64_t elf_entry, elf_low_addr, elf_high_addr;
964 int elf_machine;
965 hwaddr entry;
966 static const ARMInsnFixup *primary_loader;
967 AddressSpace *as = arm_boot_address_space(cpu, info);
969 /* CPU objects (unlike devices) are not automatically reset on system
970 * reset, so we must always register a handler to do so. If we're
971 * actually loading a kernel, the handler is also responsible for
972 * arranging that we start it correctly.
974 for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
975 qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
978 /* The board code is not supposed to set secure_board_setup unless
979 * running its code in secure mode is actually possible, and KVM
980 * doesn't support secure.
982 assert(!(info->secure_board_setup && kvm_enabled()));
984 info->dtb_filename = qemu_opt_get(qemu_get_machine_opts(), "dtb");
985 info->dtb_limit = 0;
987 /* Load the kernel. */
988 if (!info->kernel_filename || info->firmware_loaded) {
990 if (have_dtb(info)) {
991 /* If we have a device tree blob, but no kernel to supply it to (or
992 * the kernel is supposed to be loaded by the bootloader), copy the
993 * DTB to the base of RAM for the bootloader to pick up.
995 info->dtb_start = info->loader_start;
998 if (info->kernel_filename) {
999 FWCfgState *fw_cfg;
1000 bool try_decompressing_kernel;
1002 fw_cfg = fw_cfg_find();
1003 try_decompressing_kernel = arm_feature(&cpu->env,
1004 ARM_FEATURE_AARCH64);
1006 /* Expose the kernel, the command line, and the initrd in fw_cfg.
1007 * We don't process them here at all, it's all left to the
1008 * firmware.
1010 load_image_to_fw_cfg(fw_cfg,
1011 FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
1012 info->kernel_filename,
1013 try_decompressing_kernel);
1014 load_image_to_fw_cfg(fw_cfg,
1015 FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
1016 info->initrd_filename, false);
1018 if (info->kernel_cmdline) {
1019 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
1020 strlen(info->kernel_cmdline) + 1);
1021 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
1022 info->kernel_cmdline);
1026 /* We will start from address 0 (typically a boot ROM image) in the
1027 * same way as hardware.
1029 return;
1032 if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
1033 primary_loader = bootloader_aarch64;
1034 elf_machine = EM_AARCH64;
1035 } else {
1036 primary_loader = bootloader;
1037 if (!info->write_board_setup) {
1038 primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
1040 elf_machine = EM_ARM;
1043 if (!info->secondary_cpu_reset_hook) {
1044 info->secondary_cpu_reset_hook = default_reset_secondary;
1046 if (!info->write_secondary_boot) {
1047 info->write_secondary_boot = default_write_secondary;
1050 if (info->nb_cpus == 0)
1051 info->nb_cpus = 1;
1053 /* We want to put the initrd far enough into RAM that when the
1054 * kernel is uncompressed it will not clobber the initrd. However
1055 * on boards without much RAM we must ensure that we still leave
1056 * enough room for a decent sized initrd, and on boards with large
1057 * amounts of RAM we must avoid the initrd being so far up in RAM
1058 * that it is outside lowmem and inaccessible to the kernel.
1059 * So for boards with less than 256MB of RAM we put the initrd
1060 * halfway into RAM, and for boards with 256MB of RAM or more we put
1061 * the initrd at 128MB.
1063 info->initrd_start = info->loader_start +
1064 MIN(info->ram_size / 2, 128 * 1024 * 1024);
1066 /* Assume that raw images are linux kernels, and ELF images are not. */
1067 /* If the filename contains 'vmlinux', assume ELF images are linux, too. */
1068 is_linux = (strstr(info->kernel_filename, "vmlinux") != NULL);
1069 kernel_size = arm_load_elf(info, &elf_entry, &elf_low_addr,
1070 &elf_high_addr, elf_machine, as);
1071 if (kernel_size > 0 && have_dtb(info)) {
1072 /* If there is still some room left at the base of RAM, try and put
1073 * the DTB there like we do for images loaded with -bios or -pflash.
1075 if (elf_low_addr > info->loader_start
1076 || elf_high_addr < info->loader_start) {
1077 /* Set elf_low_addr as address limit for arm_load_dtb if it may be
1078 * pointing into RAM, otherwise pass '0' (no limit)
1080 if (elf_low_addr < info->loader_start) {
1081 elf_low_addr = 0;
1083 info->dtb_start = info->loader_start;
1084 info->dtb_limit = elf_low_addr;
1087 entry = elf_entry;
1088 if (kernel_size < 0) {
1089 kernel_size = load_uimage_as(info->kernel_filename, &entry, NULL,
1090 &is_linux, NULL, NULL, as);
1092 if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1093 kernel_size = load_aarch64_image(info->kernel_filename,
1094 info->loader_start, &entry, as);
1095 is_linux = 1;
1096 } else if (entry == info->loader_start) {
1097 /* Don't map bootloader memory if it conflicts with the kernel image. */
1098 /* TODO */
1099 } else if (kernel_size < 0) {
1100 /* 32-bit ARM */
1101 entry = info->loader_start + KERNEL_LOAD_ADDR;
1102 kernel_size = load_image_targphys_as(info->kernel_filename, entry,
1103 info->ram_size - KERNEL_LOAD_ADDR,
1104 as);
1105 is_linux = 1;
1107 if (kernel_size < 0) {
1108 error_report("could not load kernel '%s'", info->kernel_filename);
1109 exit(1);
1111 info->entry = entry;
1112 if (is_linux) {
1113 uint32_t fixupcontext[FIXUP_MAX];
1115 if (info->initrd_filename) {
1116 initrd_size = load_ramdisk_as(info->initrd_filename,
1117 info->initrd_start,
1118 info->ram_size - info->initrd_start,
1119 as);
1120 if (initrd_size < 0) {
1121 initrd_size = load_image_targphys_as(info->initrd_filename,
1122 info->initrd_start,
1123 info->ram_size -
1124 info->initrd_start,
1125 as);
1127 if (initrd_size < 0) {
1128 error_report("could not load initrd '%s'",
1129 info->initrd_filename);
1130 exit(1);
1132 } else {
1133 initrd_size = 0;
1135 info->initrd_size = initrd_size;
1137 fixupcontext[FIXUP_BOARDID] = info->board_id;
1138 fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1140 /* for device tree boot, we pass the DTB directly in r2. Otherwise
1141 * we point to the kernel args.
1143 if (have_dtb(info)) {
1144 hwaddr align;
1146 if (elf_machine == EM_AARCH64) {
1148 * Some AArch64 kernels on early bootup map the fdt region as
1150 * [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1152 * Let's play safe and prealign it to 2MB to give us some space.
1154 align = 2 * 1024 * 1024;
1155 } else {
1157 * Some 32bit kernels will trash anything in the 4K page the
1158 * initrd ends in, so make sure the DTB isn't caught up in that.
1160 align = 4096;
1163 /* Place the DTB after the initrd in memory with alignment. */
1164 info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size,
1165 align);
1166 fixupcontext[FIXUP_ARGPTR] = info->dtb_start;
1167 } else {
1168 fixupcontext[FIXUP_ARGPTR] = info->loader_start + KERNEL_ARGS_ADDR;
1169 if (info->ram_size >= (1ULL << 32)) {
1170 error_report("RAM size must be less than 4GB to boot"
1171 " Linux kernel using ATAGS (try passing a device tree"
1172 " using -dtb)");
1173 exit(1);
1176 fixupcontext[FIXUP_ENTRYPOINT] = entry;
1178 write_bootloader("bootloader", info->loader_start,
1179 primary_loader, fixupcontext, as);
1181 if (info->nb_cpus > 1) {
1182 info->write_secondary_boot(cpu, info);
1184 if (info->write_board_setup) {
1185 info->write_board_setup(cpu, info);
1188 /* Notify devices which need to fake up firmware initialization
1189 * that we're doing a direct kernel boot.
1191 object_child_foreach_recursive(object_get_root(),
1192 do_arm_linux_init, info);
1194 info->is_linux = is_linux;
1196 for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1197 ARM_CPU(cs)->env.boot_info = info;
1200 if (!info->skip_dtb_autoload && have_dtb(info)) {
1201 if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as) < 0) {
1202 exit(1);
1207 static const TypeInfo arm_linux_boot_if_info = {
1208 .name = TYPE_ARM_LINUX_BOOT_IF,
1209 .parent = TYPE_INTERFACE,
1210 .class_size = sizeof(ARMLinuxBootIfClass),
1213 static void arm_linux_boot_register_types(void)
1215 type_register_static(&arm_linux_boot_if_info);
1218 type_init(arm_linux_boot_register_types)