xref: /openbmc/qemu/hw/arm/boot.c (revision 4a9b31b8)
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/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"
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_LOAD_ADDR 0x00010000
35 #define KERNEL64_LOAD_ADDR 0x00080000
36 
37 #define ARM64_TEXT_OFFSET_OFFSET    8
38 #define ARM64_MAGIC_OFFSET          56
39 
40 #define BOOTLOADER_MAX_SIZE         (4 * KiB)
41 
42 AddressSpace *arm_boot_address_space(ARMCPU *cpu,
43                                      const struct arm_boot_info *info)
44 {
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.
48      */
49     int asidx;
50     CPUState *cs = CPU(cpu);
51 
52     if (arm_feature(&cpu->env, ARM_FEATURE_EL3) && info->secure_boot) {
53         asidx = ARMASIdx_S;
54     } else {
55         asidx = ARMASIdx_NS;
56     }
57 
58     return cpu_get_address_space(cs, asidx);
59 }
60 
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;
73 
74 typedef struct ARMInsnFixup {
75     uint32_t insn;
76     FixupType fixup;
77 } ARMInsnFixup;
78 
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 }
91 };
92 
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.
97  */
98 
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 }
112 };
113 
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.
122  *
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.
127  */
128 #define DSB_INSN 0xf57ff04f
129 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
130 
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 }
147 };
148 
149 static void write_bootloader(const char *name, hwaddr addr,
150                              const ARMInsnFixup *insns, uint32_t *fixupcontext,
151                              AddressSpace *as)
152 {
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.
157      */
158     int i, len;
159     uint32_t *code;
160 
161     len = 0;
162     while (insns[len].fixup != FIXUP_TERMINATOR) {
163         len++;
164     }
165 
166     code = g_new0(uint32_t, len);
167 
168     for (i = 0; i < len; i++) {
169         uint32_t insn = insns[i].insn;
170         FixupType fixup = insns[i].fixup;
171 
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();
186         }
187         code[i] = tswap32(insn);
188     }
189 
190     assert((len * sizeof(uint32_t)) < BOOTLOADER_MAX_SIZE);
191 
192     rom_add_blob_fixed_as(name, code, len * sizeof(uint32_t), addr, as);
193 
194     g_free(code);
195 }
196 
197 static void default_write_secondary(ARMCPU *cpu,
198                                     const struct arm_boot_info *info)
199 {
200     uint32_t fixupcontext[FIXUP_MAX];
201     AddressSpace *as = arm_boot_address_space(cpu, info);
202 
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;
209     }
210 
211     write_bootloader("smpboot", info->smp_loader_start,
212                      smpboot, fixupcontext, as);
213 }
214 
215 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
216                                             const struct arm_boot_info *info,
217                                             hwaddr mvbar_addr)
218 {
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).
225          */
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) */
234     };
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 */
245     };
246 
247     /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
248     assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
249 
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));
253 
254     for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
255         mvbar_blob[n] = tswap32(mvbar_blob[n]);
256     }
257     rom_add_blob_fixed_as("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
258                           mvbar_addr, as);
259 
260     for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
261         board_setup_blob[n] = tswap32(board_setup_blob[n]);
262     }
263     rom_add_blob_fixed_as("board-setup", board_setup_blob,
264                           sizeof(board_setup_blob), info->board_setup_addr, as);
265 }
266 
267 static void default_reset_secondary(ARMCPU *cpu,
268                                     const struct arm_boot_info *info)
269 {
270     AddressSpace *as = arm_boot_address_space(cpu, info);
271     CPUState *cs = CPU(cpu);
272 
273     address_space_stl_notdirty(as, info->smp_bootreg_addr,
274                                0, MEMTXATTRS_UNSPECIFIED, NULL);
275     cpu_set_pc(cs, info->smp_loader_start);
276 }
277 
278 static inline bool have_dtb(const struct arm_boot_info *info)
279 {
280     return info->dtb_filename || info->get_dtb;
281 }
282 
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)
288 
289 static void set_kernel_args(const struct arm_boot_info *info, AddressSpace *as)
290 {
291     int initrd_size = info->initrd_size;
292     hwaddr base = info->loader_start;
293     hwaddr p;
294 
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);
314     }
315     if (info->kernel_cmdline && *info->kernel_cmdline) {
316         /* ATAG_CMDLINE */
317         int cmdline_size;
318 
319         cmdline_size = strlen(info->kernel_cmdline);
320         address_space_write(as, p + 8, MEMTXATTRS_UNSPECIFIED,
321                             (const uint8_t *)info->kernel_cmdline,
322                             cmdline_size + 1);
323         cmdline_size = (cmdline_size >> 2) + 1;
324         WRITE_WORD(p, cmdline_size + 2);
325         WRITE_WORD(p, 0x54410009);
326         p += cmdline_size * 4;
327     }
328     if (info->atag_board) {
329         /* ATAG_BOARD */
330         int atag_board_len;
331         uint8_t atag_board_buf[0x1000];
332 
333         atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
334         WRITE_WORD(p, (atag_board_len + 8) >> 2);
335         WRITE_WORD(p, 0x414f4d50);
336         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
337                             atag_board_buf, atag_board_len);
338         p += atag_board_len;
339     }
340     /* ATAG_END */
341     WRITE_WORD(p, 0);
342     WRITE_WORD(p, 0);
343 }
344 
345 static void set_kernel_args_old(const struct arm_boot_info *info,
346                                 AddressSpace *as)
347 {
348     hwaddr p;
349     const char *s;
350     int initrd_size = info->initrd_size;
351     hwaddr base = info->loader_start;
352 
353     /* see linux/include/asm-arm/setup.h */
354     p = base + KERNEL_ARGS_ADDR;
355     /* page_size */
356     WRITE_WORD(p, 4096);
357     /* nr_pages */
358     WRITE_WORD(p, info->ram_size / 4096);
359     /* ramdisk_size */
360     WRITE_WORD(p, 0);
361 #define FLAG_READONLY	1
362 #define FLAG_RDLOAD	4
363 #define FLAG_RDPROMPT	8
364     /* flags */
365     WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
366     /* rootdev */
367     WRITE_WORD(p, (31 << 8) | 0);	/* /dev/mtdblock0 */
368     /* video_num_cols */
369     WRITE_WORD(p, 0);
370     /* video_num_rows */
371     WRITE_WORD(p, 0);
372     /* video_x */
373     WRITE_WORD(p, 0);
374     /* video_y */
375     WRITE_WORD(p, 0);
376     /* memc_control_reg */
377     WRITE_WORD(p, 0);
378     /* unsigned char sounddefault */
379     /* unsigned char adfsdrives */
380     /* unsigned char bytes_per_char_h */
381     /* unsigned char bytes_per_char_v */
382     WRITE_WORD(p, 0);
383     /* pages_in_bank[4] */
384     WRITE_WORD(p, 0);
385     WRITE_WORD(p, 0);
386     WRITE_WORD(p, 0);
387     WRITE_WORD(p, 0);
388     /* pages_in_vram */
389     WRITE_WORD(p, 0);
390     /* initrd_start */
391     if (initrd_size) {
392         WRITE_WORD(p, info->initrd_start);
393     } else {
394         WRITE_WORD(p, 0);
395     }
396     /* initrd_size */
397     WRITE_WORD(p, initrd_size);
398     /* rd_start */
399     WRITE_WORD(p, 0);
400     /* system_rev */
401     WRITE_WORD(p, 0);
402     /* system_serial_low */
403     WRITE_WORD(p, 0);
404     /* system_serial_high */
405     WRITE_WORD(p, 0);
406     /* mem_fclk_21285 */
407     WRITE_WORD(p, 0);
408     /* zero unused fields */
409     while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
410         WRITE_WORD(p, 0);
411     }
412     s = info->kernel_cmdline;
413     if (s) {
414         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
415                             (const uint8_t *)s, strlen(s) + 1);
416     } else {
417         WRITE_WORD(p, 0);
418     }
419 }
420 
421 static void fdt_add_psci_node(void *fdt)
422 {
423     uint32_t cpu_suspend_fn;
424     uint32_t cpu_off_fn;
425     uint32_t cpu_on_fn;
426     uint32_t migrate_fn;
427     ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
428     const char *psci_method;
429     int64_t psci_conduit;
430     int rc;
431 
432     psci_conduit = object_property_get_int(OBJECT(armcpu),
433                                            "psci-conduit",
434                                            &error_abort);
435     switch (psci_conduit) {
436     case QEMU_PSCI_CONDUIT_DISABLED:
437         return;
438     case QEMU_PSCI_CONDUIT_HVC:
439         psci_method = "hvc";
440         break;
441     case QEMU_PSCI_CONDUIT_SMC:
442         psci_method = "smc";
443         break;
444     default:
445         g_assert_not_reached();
446     }
447 
448     /*
449      * If /psci node is present in provided DTB, assume that no fixup
450      * is necessary and all PSCI configuration should be taken as-is
451      */
452     rc = fdt_path_offset(fdt, "/psci");
453     if (rc >= 0) {
454         return;
455     }
456 
457     qemu_fdt_add_subnode(fdt, "/psci");
458     if (armcpu->psci_version == 2) {
459         const char comp[] = "arm,psci-0.2\0arm,psci";
460         qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
461 
462         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
463         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
464             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
465             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
466             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
467         } else {
468             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
469             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
470             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
471         }
472     } else {
473         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
474 
475         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
476         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
477         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
478         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
479     }
480 
481     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
482      * to the instruction that should be used to invoke PSCI functions.
483      * However, the device tree binding uses 'method' instead, so that is
484      * what we should use here.
485      */
486     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
487 
488     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
489     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
490     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
491     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
492 }
493 
494 int arm_load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
495                  hwaddr addr_limit, AddressSpace *as)
496 {
497     void *fdt = NULL;
498     int size, rc, n = 0;
499     uint32_t acells, scells;
500     char *nodename;
501     unsigned int i;
502     hwaddr mem_base, mem_len;
503     char **node_path;
504     Error *err = NULL;
505 
506     if (binfo->dtb_filename) {
507         char *filename;
508         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
509         if (!filename) {
510             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
511             goto fail;
512         }
513 
514         fdt = load_device_tree(filename, &size);
515         if (!fdt) {
516             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
517             g_free(filename);
518             goto fail;
519         }
520         g_free(filename);
521     } else {
522         fdt = binfo->get_dtb(binfo, &size);
523         if (!fdt) {
524             fprintf(stderr, "Board was unable to create a dtb blob\n");
525             goto fail;
526         }
527     }
528 
529     if (addr_limit > addr && size > (addr_limit - addr)) {
530         /* Installing the device tree blob at addr would exceed addr_limit.
531          * Whether this constitutes failure is up to the caller to decide,
532          * so just return 0 as size, i.e., no error.
533          */
534         g_free(fdt);
535         return 0;
536     }
537 
538     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
539                                    NULL, &error_fatal);
540     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
541                                    NULL, &error_fatal);
542     if (acells == 0 || scells == 0) {
543         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
544         goto fail;
545     }
546 
547     if (scells < 2 && binfo->ram_size >= (1ULL << 32)) {
548         /* This is user error so deserves a friendlier error message
549          * than the failure of setprop_sized_cells would provide
550          */
551         fprintf(stderr, "qemu: dtb file not compatible with "
552                 "RAM size > 4GB\n");
553         goto fail;
554     }
555 
556     /* nop all root nodes matching /memory or /memory@unit-address */
557     node_path = qemu_fdt_node_unit_path(fdt, "memory", &err);
558     if (err) {
559         error_report_err(err);
560         goto fail;
561     }
562     while (node_path[n]) {
563         if (g_str_has_prefix(node_path[n], "/memory")) {
564             qemu_fdt_nop_node(fdt, node_path[n]);
565         }
566         n++;
567     }
568     g_strfreev(node_path);
569 
570     if (nb_numa_nodes > 0) {
571         mem_base = binfo->loader_start;
572         for (i = 0; i < nb_numa_nodes; i++) {
573             mem_len = numa_info[i].node_mem;
574             nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
575             qemu_fdt_add_subnode(fdt, nodename);
576             qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
577             rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
578                                               acells, mem_base,
579                                               scells, mem_len);
580             if (rc < 0) {
581                 fprintf(stderr, "couldn't set %s/reg for node %d\n", nodename,
582                         i);
583                 goto fail;
584             }
585 
586             qemu_fdt_setprop_cell(fdt, nodename, "numa-node-id", i);
587             mem_base += mem_len;
588             g_free(nodename);
589         }
590     } else {
591         nodename = g_strdup_printf("/memory@%" PRIx64, binfo->loader_start);
592         qemu_fdt_add_subnode(fdt, nodename);
593         qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
594 
595         rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
596                                           acells, binfo->loader_start,
597                                           scells, binfo->ram_size);
598         if (rc < 0) {
599             fprintf(stderr, "couldn't set %s reg\n", nodename);
600             goto fail;
601         }
602         g_free(nodename);
603     }
604 
605     rc = fdt_path_offset(fdt, "/chosen");
606     if (rc < 0) {
607         qemu_fdt_add_subnode(fdt, "/chosen");
608     }
609 
610     if (binfo->kernel_cmdline && *binfo->kernel_cmdline) {
611         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
612                                      binfo->kernel_cmdline);
613         if (rc < 0) {
614             fprintf(stderr, "couldn't set /chosen/bootargs\n");
615             goto fail;
616         }
617     }
618 
619     if (binfo->initrd_size) {
620         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
621                                    binfo->initrd_start);
622         if (rc < 0) {
623             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
624             goto fail;
625         }
626 
627         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
628                                    binfo->initrd_start + binfo->initrd_size);
629         if (rc < 0) {
630             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
631             goto fail;
632         }
633     }
634 
635     fdt_add_psci_node(fdt);
636 
637     if (binfo->modify_dtb) {
638         binfo->modify_dtb(binfo, fdt);
639     }
640 
641     qemu_fdt_dumpdtb(fdt, size);
642 
643     /* Put the DTB into the memory map as a ROM image: this will ensure
644      * the DTB is copied again upon reset, even if addr points into RAM.
645      */
646     rom_add_blob_fixed_as("dtb", fdt, size, addr, as);
647 
648     g_free(fdt);
649 
650     return size;
651 
652 fail:
653     g_free(fdt);
654     return -1;
655 }
656 
657 static void do_cpu_reset(void *opaque)
658 {
659     ARMCPU *cpu = opaque;
660     CPUState *cs = CPU(cpu);
661     CPUARMState *env = &cpu->env;
662     const struct arm_boot_info *info = env->boot_info;
663 
664     cpu_reset(cs);
665     if (info) {
666         if (!info->is_linux) {
667             int i;
668             /* Jump to the entry point.  */
669             uint64_t entry = info->entry;
670 
671             switch (info->endianness) {
672             case ARM_ENDIANNESS_LE:
673                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
674                 for (i = 1; i < 4; ++i) {
675                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
676                 }
677                 env->uncached_cpsr &= ~CPSR_E;
678                 break;
679             case ARM_ENDIANNESS_BE8:
680                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
681                 for (i = 1; i < 4; ++i) {
682                     env->cp15.sctlr_el[i] |= SCTLR_EE;
683                 }
684                 env->uncached_cpsr |= CPSR_E;
685                 break;
686             case ARM_ENDIANNESS_BE32:
687                 env->cp15.sctlr_el[1] |= SCTLR_B;
688                 break;
689             case ARM_ENDIANNESS_UNKNOWN:
690                 break; /* Board's decision */
691             default:
692                 g_assert_not_reached();
693             }
694 
695             if (!env->aarch64) {
696                 env->thumb = info->entry & 1;
697                 entry &= 0xfffffffe;
698             }
699             cpu_set_pc(cs, entry);
700         } else {
701             /* If we are booting Linux then we need to check whether we are
702              * booting into secure or non-secure state and adjust the state
703              * accordingly.  Out of reset, ARM is defined to be in secure state
704              * (SCR.NS = 0), we change that here if non-secure boot has been
705              * requested.
706              */
707             if (arm_feature(env, ARM_FEATURE_EL3)) {
708                 /* AArch64 is defined to come out of reset into EL3 if enabled.
709                  * If we are booting Linux then we need to adjust our EL as
710                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
711                  * SVC, which Linux expects, so no privilege/exception level to
712                  * adjust.
713                  */
714                 if (env->aarch64) {
715                     env->cp15.scr_el3 |= SCR_RW;
716                     if (arm_feature(env, ARM_FEATURE_EL2)) {
717                         env->cp15.hcr_el2 |= HCR_RW;
718                         env->pstate = PSTATE_MODE_EL2h;
719                     } else {
720                         env->pstate = PSTATE_MODE_EL1h;
721                     }
722                     /* AArch64 kernels never boot in secure mode */
723                     assert(!info->secure_boot);
724                     /* This hook is only supported for AArch32 currently:
725                      * bootloader_aarch64[] will not call the hook, and
726                      * the code above has already dropped us into EL2 or EL1.
727                      */
728                     assert(!info->secure_board_setup);
729                 }
730 
731                 if (arm_feature(env, ARM_FEATURE_EL2)) {
732                     /* If we have EL2 then Linux expects the HVC insn to work */
733                     env->cp15.scr_el3 |= SCR_HCE;
734                 }
735 
736                 /* Set to non-secure if not a secure boot */
737                 if (!info->secure_boot &&
738                     (cs != first_cpu || !info->secure_board_setup)) {
739                     /* Linux expects non-secure state */
740                     env->cp15.scr_el3 |= SCR_NS;
741                 }
742             }
743 
744             if (!env->aarch64 && !info->secure_boot &&
745                 arm_feature(env, ARM_FEATURE_EL2)) {
746                 /*
747                  * This is an AArch32 boot not to Secure state, and
748                  * we have Hyp mode available, so boot the kernel into
749                  * Hyp mode. This is not how the CPU comes out of reset,
750                  * so we need to manually put it there.
751                  */
752                 cpsr_write(env, ARM_CPU_MODE_HYP, CPSR_M, CPSRWriteRaw);
753             }
754 
755             if (cs == first_cpu) {
756                 AddressSpace *as = arm_boot_address_space(cpu, info);
757 
758                 cpu_set_pc(cs, info->loader_start);
759 
760                 if (!have_dtb(info)) {
761                     if (old_param) {
762                         set_kernel_args_old(info, as);
763                     } else {
764                         set_kernel_args(info, as);
765                     }
766                 }
767             } else {
768                 info->secondary_cpu_reset_hook(cpu, info);
769             }
770         }
771     }
772 }
773 
774 /**
775  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
776  *                          by key.
777  * @fw_cfg:         The firmware config instance to store the data in.
778  * @size_key:       The firmware config key to store the size of the loaded
779  *                  data under, with fw_cfg_add_i32().
780  * @data_key:       The firmware config key to store the loaded data under,
781  *                  with fw_cfg_add_bytes().
782  * @image_name:     The name of the image file to load. If it is NULL, the
783  *                  function returns without doing anything.
784  * @try_decompress: Whether the image should be decompressed (gunzipped) before
785  *                  adding it to fw_cfg. If decompression fails, the image is
786  *                  loaded as-is.
787  *
788  * In case of failure, the function prints an error message to stderr and the
789  * process exits with status 1.
790  */
791 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
792                                  uint16_t data_key, const char *image_name,
793                                  bool try_decompress)
794 {
795     size_t size = -1;
796     uint8_t *data;
797 
798     if (image_name == NULL) {
799         return;
800     }
801 
802     if (try_decompress) {
803         size = load_image_gzipped_buffer(image_name,
804                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
805     }
806 
807     if (size == (size_t)-1) {
808         gchar *contents;
809         gsize length;
810 
811         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
812             error_report("failed to load \"%s\"", image_name);
813             exit(1);
814         }
815         size = length;
816         data = (uint8_t *)contents;
817     }
818 
819     fw_cfg_add_i32(fw_cfg, size_key, size);
820     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
821 }
822 
823 static int do_arm_linux_init(Object *obj, void *opaque)
824 {
825     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
826         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
827         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
828         struct arm_boot_info *info = opaque;
829 
830         if (albifc->arm_linux_init) {
831             albifc->arm_linux_init(albif, info->secure_boot);
832         }
833     }
834     return 0;
835 }
836 
837 static int64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
838                             uint64_t *lowaddr, uint64_t *highaddr,
839                             int elf_machine, AddressSpace *as)
840 {
841     bool elf_is64;
842     union {
843         Elf32_Ehdr h32;
844         Elf64_Ehdr h64;
845     } elf_header;
846     int data_swab = 0;
847     bool big_endian;
848     int64_t ret = -1;
849     Error *err = NULL;
850 
851 
852     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
853     if (err) {
854         error_free(err);
855         return ret;
856     }
857 
858     if (elf_is64) {
859         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
860         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
861                                       : ARM_ENDIANNESS_LE;
862     } else {
863         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
864         if (big_endian) {
865             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
866                 info->endianness = ARM_ENDIANNESS_BE8;
867             } else {
868                 info->endianness = ARM_ENDIANNESS_BE32;
869                 /* In BE32, the CPU has a different view of the per-byte
870                  * address map than the rest of the system. BE32 ELF files
871                  * are organised such that they can be programmed through
872                  * the CPU's per-word byte-reversed view of the world. QEMU
873                  * however loads ELF files independently of the CPU. So
874                  * tell the ELF loader to byte reverse the data for us.
875                  */
876                 data_swab = 2;
877             }
878         } else {
879             info->endianness = ARM_ENDIANNESS_LE;
880         }
881     }
882 
883     ret = load_elf_as(info->kernel_filename, NULL, NULL,
884                       pentry, lowaddr, highaddr, big_endian, elf_machine,
885                       1, data_swab, as);
886     if (ret <= 0) {
887         /* The header loaded but the image didn't */
888         exit(1);
889     }
890 
891     return ret;
892 }
893 
894 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
895                                    hwaddr *entry, AddressSpace *as)
896 {
897     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
898     uint8_t *buffer;
899     int size;
900 
901     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
902     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
903                                      &buffer);
904 
905     if (size < 0) {
906         gsize len;
907 
908         /* Load as raw file otherwise */
909         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
910             return -1;
911         }
912         size = len;
913     }
914 
915     /* check the arm64 magic header value -- very old kernels may not have it */
916     if (size > ARM64_MAGIC_OFFSET + 4 &&
917         memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
918         uint64_t hdrvals[2];
919 
920         /* The arm64 Image header has text_offset and image_size fields at 8 and
921          * 16 bytes into the Image header, respectively. The text_offset field
922          * is only valid if the image_size is non-zero.
923          */
924         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
925         if (hdrvals[1] != 0) {
926             kernel_load_offset = le64_to_cpu(hdrvals[0]);
927 
928             /*
929              * We write our startup "bootloader" at the very bottom of RAM,
930              * so that bit can't be used for the image. Luckily the Image
931              * format specification is that the image requests only an offset
932              * from a 2MB boundary, not an absolute load address. So if the
933              * image requests an offset that might mean it overlaps with the
934              * bootloader, we can just load it starting at 2MB+offset rather
935              * than 0MB + offset.
936              */
937             if (kernel_load_offset < BOOTLOADER_MAX_SIZE) {
938                 kernel_load_offset += 2 * MiB;
939             }
940         }
941     }
942 
943     *entry = mem_base + kernel_load_offset;
944     rom_add_blob_fixed_as(filename, buffer, size, *entry, as);
945 
946     g_free(buffer);
947 
948     return size;
949 }
950 
951 void arm_load_kernel(ARMCPU *cpu, struct arm_boot_info *info)
952 {
953     CPUState *cs;
954     int kernel_size;
955     int initrd_size;
956     int is_linux = 0;
957     uint64_t elf_entry, elf_low_addr, elf_high_addr;
958     int elf_machine;
959     hwaddr entry;
960     static const ARMInsnFixup *primary_loader;
961     AddressSpace *as = arm_boot_address_space(cpu, info);
962 
963     /* CPU objects (unlike devices) are not automatically reset on system
964      * reset, so we must always register a handler to do so. If we're
965      * actually loading a kernel, the handler is also responsible for
966      * arranging that we start it correctly.
967      */
968     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
969         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
970     }
971 
972     /* The board code is not supposed to set secure_board_setup unless
973      * running its code in secure mode is actually possible, and KVM
974      * doesn't support secure.
975      */
976     assert(!(info->secure_board_setup && kvm_enabled()));
977 
978     info->dtb_filename = qemu_opt_get(qemu_get_machine_opts(), "dtb");
979     info->dtb_limit = 0;
980 
981     /* Load the kernel.  */
982     if (!info->kernel_filename || info->firmware_loaded) {
983 
984         if (have_dtb(info)) {
985             /* If we have a device tree blob, but no kernel to supply it to (or
986              * the kernel is supposed to be loaded by the bootloader), copy the
987              * DTB to the base of RAM for the bootloader to pick up.
988              */
989             info->dtb_start = info->loader_start;
990         }
991 
992         if (info->kernel_filename) {
993             FWCfgState *fw_cfg;
994             bool try_decompressing_kernel;
995 
996             fw_cfg = fw_cfg_find();
997             try_decompressing_kernel = arm_feature(&cpu->env,
998                                                    ARM_FEATURE_AARCH64);
999 
1000             /* Expose the kernel, the command line, and the initrd in fw_cfg.
1001              * We don't process them here at all, it's all left to the
1002              * firmware.
1003              */
1004             load_image_to_fw_cfg(fw_cfg,
1005                                  FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
1006                                  info->kernel_filename,
1007                                  try_decompressing_kernel);
1008             load_image_to_fw_cfg(fw_cfg,
1009                                  FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
1010                                  info->initrd_filename, false);
1011 
1012             if (info->kernel_cmdline) {
1013                 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
1014                                strlen(info->kernel_cmdline) + 1);
1015                 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
1016                                   info->kernel_cmdline);
1017             }
1018         }
1019 
1020         /* We will start from address 0 (typically a boot ROM image) in the
1021          * same way as hardware.
1022          */
1023         return;
1024     }
1025 
1026     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
1027         primary_loader = bootloader_aarch64;
1028         elf_machine = EM_AARCH64;
1029     } else {
1030         primary_loader = bootloader;
1031         if (!info->write_board_setup) {
1032             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
1033         }
1034         elf_machine = EM_ARM;
1035     }
1036 
1037     if (!info->secondary_cpu_reset_hook) {
1038         info->secondary_cpu_reset_hook = default_reset_secondary;
1039     }
1040     if (!info->write_secondary_boot) {
1041         info->write_secondary_boot = default_write_secondary;
1042     }
1043 
1044     if (info->nb_cpus == 0)
1045         info->nb_cpus = 1;
1046 
1047     /* We want to put the initrd far enough into RAM that when the
1048      * kernel is uncompressed it will not clobber the initrd. However
1049      * on boards without much RAM we must ensure that we still leave
1050      * enough room for a decent sized initrd, and on boards with large
1051      * amounts of RAM we must avoid the initrd being so far up in RAM
1052      * that it is outside lowmem and inaccessible to the kernel.
1053      * So for boards with less  than 256MB of RAM we put the initrd
1054      * halfway into RAM, and for boards with 256MB of RAM or more we put
1055      * the initrd at 128MB.
1056      */
1057     info->initrd_start = info->loader_start +
1058         MIN(info->ram_size / 2, 128 * 1024 * 1024);
1059 
1060     /* Assume that raw images are linux kernels, and ELF images are not.  */
1061     kernel_size = arm_load_elf(info, &elf_entry, &elf_low_addr,
1062                                &elf_high_addr, elf_machine, as);
1063     if (kernel_size > 0 && have_dtb(info)) {
1064         /* If there is still some room left at the base of RAM, try and put
1065          * the DTB there like we do for images loaded with -bios or -pflash.
1066          */
1067         if (elf_low_addr > info->loader_start
1068             || elf_high_addr < info->loader_start) {
1069             /* Set elf_low_addr as address limit for arm_load_dtb if it may be
1070              * pointing into RAM, otherwise pass '0' (no limit)
1071              */
1072             if (elf_low_addr < info->loader_start) {
1073                 elf_low_addr = 0;
1074             }
1075             info->dtb_start = info->loader_start;
1076             info->dtb_limit = elf_low_addr;
1077         }
1078     }
1079     entry = elf_entry;
1080     if (kernel_size < 0) {
1081         kernel_size = load_uimage_as(info->kernel_filename, &entry, NULL,
1082                                      &is_linux, NULL, NULL, as);
1083     }
1084     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1085         kernel_size = load_aarch64_image(info->kernel_filename,
1086                                          info->loader_start, &entry, as);
1087         is_linux = 1;
1088     } else if (kernel_size < 0) {
1089         /* 32-bit ARM */
1090         entry = info->loader_start + KERNEL_LOAD_ADDR;
1091         kernel_size = load_image_targphys_as(info->kernel_filename, entry,
1092                                              info->ram_size - KERNEL_LOAD_ADDR,
1093                                              as);
1094         is_linux = 1;
1095     }
1096     if (kernel_size < 0) {
1097         error_report("could not load kernel '%s'", info->kernel_filename);
1098         exit(1);
1099     }
1100     info->entry = entry;
1101     if (is_linux) {
1102         uint32_t fixupcontext[FIXUP_MAX];
1103 
1104         if (info->initrd_filename) {
1105             initrd_size = load_ramdisk_as(info->initrd_filename,
1106                                           info->initrd_start,
1107                                           info->ram_size - info->initrd_start,
1108                                           as);
1109             if (initrd_size < 0) {
1110                 initrd_size = load_image_targphys_as(info->initrd_filename,
1111                                                      info->initrd_start,
1112                                                      info->ram_size -
1113                                                      info->initrd_start,
1114                                                      as);
1115             }
1116             if (initrd_size < 0) {
1117                 error_report("could not load initrd '%s'",
1118                              info->initrd_filename);
1119                 exit(1);
1120             }
1121         } else {
1122             initrd_size = 0;
1123         }
1124         info->initrd_size = initrd_size;
1125 
1126         fixupcontext[FIXUP_BOARDID] = info->board_id;
1127         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1128 
1129         /* for device tree boot, we pass the DTB directly in r2. Otherwise
1130          * we point to the kernel args.
1131          */
1132         if (have_dtb(info)) {
1133             hwaddr align;
1134 
1135             if (elf_machine == EM_AARCH64) {
1136                 /*
1137                  * Some AArch64 kernels on early bootup map the fdt region as
1138                  *
1139                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1140                  *
1141                  * Let's play safe and prealign it to 2MB to give us some space.
1142                  */
1143                 align = 2 * 1024 * 1024;
1144             } else {
1145                 /*
1146                  * Some 32bit kernels will trash anything in the 4K page the
1147                  * initrd ends in, so make sure the DTB isn't caught up in that.
1148                  */
1149                 align = 4096;
1150             }
1151 
1152             /* Place the DTB after the initrd in memory with alignment. */
1153             info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size,
1154                                            align);
1155             fixupcontext[FIXUP_ARGPTR] = info->dtb_start;
1156         } else {
1157             fixupcontext[FIXUP_ARGPTR] = info->loader_start + KERNEL_ARGS_ADDR;
1158             if (info->ram_size >= (1ULL << 32)) {
1159                 error_report("RAM size must be less than 4GB to boot"
1160                              " Linux kernel using ATAGS (try passing a device tree"
1161                              " using -dtb)");
1162                 exit(1);
1163             }
1164         }
1165         fixupcontext[FIXUP_ENTRYPOINT] = entry;
1166 
1167         write_bootloader("bootloader", info->loader_start,
1168                          primary_loader, fixupcontext, as);
1169 
1170         if (info->nb_cpus > 1) {
1171             info->write_secondary_boot(cpu, info);
1172         }
1173         if (info->write_board_setup) {
1174             info->write_board_setup(cpu, info);
1175         }
1176 
1177         /* Notify devices which need to fake up firmware initialization
1178          * that we're doing a direct kernel boot.
1179          */
1180         object_child_foreach_recursive(object_get_root(),
1181                                        do_arm_linux_init, info);
1182     }
1183     info->is_linux = is_linux;
1184 
1185     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1186         ARM_CPU(cs)->env.boot_info = info;
1187     }
1188 
1189     if (!info->skip_dtb_autoload && have_dtb(info)) {
1190         if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as) < 0) {
1191             exit(1);
1192         }
1193     }
1194 }
1195 
1196 static const TypeInfo arm_linux_boot_if_info = {
1197     .name = TYPE_ARM_LINUX_BOOT_IF,
1198     .parent = TYPE_INTERFACE,
1199     .class_size = sizeof(ARMLinuxBootIfClass),
1200 };
1201 
1202 static void arm_linux_boot_register_types(void)
1203 {
1204     type_register_static(&arm_linux_boot_if_info);
1205 }
1206 
1207 type_init(arm_linux_boot_register_types)
1208