xref: /openbmc/qemu/hw/arm/boot.c (revision 7f709ce7)
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 "qapi/error.h"
12 #include <libfdt.h>
13 #include "hw/hw.h"
14 #include "hw/arm/arm.h"
15 #include "hw/arm/linux-boot-if.h"
16 #include "sysemu/kvm.h"
17 #include "sysemu/sysemu.h"
18 #include "sysemu/numa.h"
19 #include "hw/boards.h"
20 #include "hw/loader.h"
21 #include "elf.h"
22 #include "sysemu/device_tree.h"
23 #include "qemu/config-file.h"
24 #include "exec/address-spaces.h"
25 
26 /* Kernel boot protocol is specified in the kernel docs
27  * Documentation/arm/Booting and Documentation/arm64/booting.txt
28  * They have different preferred image load offsets from system RAM base.
29  */
30 #define KERNEL_ARGS_ADDR 0x100
31 #define KERNEL_LOAD_ADDR 0x00010000
32 #define KERNEL64_LOAD_ADDR 0x00080000
33 
34 #define ARM64_TEXT_OFFSET_OFFSET    8
35 #define ARM64_MAGIC_OFFSET          56
36 
37 typedef enum {
38     FIXUP_NONE = 0,     /* do nothing */
39     FIXUP_TERMINATOR,   /* end of insns */
40     FIXUP_BOARDID,      /* overwrite with board ID number */
41     FIXUP_BOARD_SETUP,  /* overwrite with board specific setup code address */
42     FIXUP_ARGPTR,       /* overwrite with pointer to kernel args */
43     FIXUP_ENTRYPOINT,   /* overwrite with kernel entry point */
44     FIXUP_GIC_CPU_IF,   /* overwrite with GIC CPU interface address */
45     FIXUP_BOOTREG,      /* overwrite with boot register address */
46     FIXUP_DSB,          /* overwrite with correct DSB insn for cpu */
47     FIXUP_MAX,
48 } FixupType;
49 
50 typedef struct ARMInsnFixup {
51     uint32_t insn;
52     FixupType fixup;
53 } ARMInsnFixup;
54 
55 static const ARMInsnFixup bootloader_aarch64[] = {
56     { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */
57     { 0xaa1f03e1 }, /* mov x1, xzr */
58     { 0xaa1f03e2 }, /* mov x2, xzr */
59     { 0xaa1f03e3 }, /* mov x3, xzr */
60     { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */
61     { 0xd61f0080 }, /* br x4      ; Jump to the kernel entry point */
62     { 0, FIXUP_ARGPTR }, /* arg: .word @DTB Lower 32-bits */
63     { 0 }, /* .word @DTB Higher 32-bits */
64     { 0, FIXUP_ENTRYPOINT }, /* entry: .word @Kernel Entry Lower 32-bits */
65     { 0 }, /* .word @Kernel Entry Higher 32-bits */
66     { 0, FIXUP_TERMINATOR }
67 };
68 
69 /* A very small bootloader: call the board-setup code (if needed),
70  * set r0-r2, then jump to the kernel.
71  * If we're not calling boot setup code then we don't copy across
72  * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array.
73  */
74 
75 static const ARMInsnFixup bootloader[] = {
76     { 0xe28fe004 }, /* add     lr, pc, #4 */
77     { 0xe51ff004 }, /* ldr     pc, [pc, #-4] */
78     { 0, FIXUP_BOARD_SETUP },
79 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3
80     { 0xe3a00000 }, /* mov     r0, #0 */
81     { 0xe59f1004 }, /* ldr     r1, [pc, #4] */
82     { 0xe59f2004 }, /* ldr     r2, [pc, #4] */
83     { 0xe59ff004 }, /* ldr     pc, [pc, #4] */
84     { 0, FIXUP_BOARDID },
85     { 0, FIXUP_ARGPTR },
86     { 0, FIXUP_ENTRYPOINT },
87     { 0, FIXUP_TERMINATOR }
88 };
89 
90 /* Handling for secondary CPU boot in a multicore system.
91  * Unlike the uniprocessor/primary CPU boot, this is platform
92  * dependent. The default code here is based on the secondary
93  * CPU boot protocol used on realview/vexpress boards, with
94  * some parameterisation to increase its flexibility.
95  * QEMU platform models for which this code is not appropriate
96  * should override write_secondary_boot and secondary_cpu_reset_hook
97  * instead.
98  *
99  * This code enables the interrupt controllers for the secondary
100  * CPUs and then puts all the secondary CPUs into a loop waiting
101  * for an interprocessor interrupt and polling a configurable
102  * location for the kernel secondary CPU entry point.
103  */
104 #define DSB_INSN 0xf57ff04f
105 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
106 
107 static const ARMInsnFixup smpboot[] = {
108     { 0xe59f2028 }, /* ldr r2, gic_cpu_if */
109     { 0xe59f0028 }, /* ldr r0, bootreg_addr */
110     { 0xe3a01001 }, /* mov r1, #1 */
111     { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */
112     { 0xe3a010ff }, /* mov r1, #0xff */
113     { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */
114     { 0, FIXUP_DSB },   /* dsb */
115     { 0xe320f003 }, /* wfi */
116     { 0xe5901000 }, /* ldr     r1, [r0] */
117     { 0xe1110001 }, /* tst     r1, r1 */
118     { 0x0afffffb }, /* beq     <wfi> */
119     { 0xe12fff11 }, /* bx      r1 */
120     { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */
121     { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */
122     { 0, FIXUP_TERMINATOR }
123 };
124 
125 static void write_bootloader(const char *name, hwaddr addr,
126                              const ARMInsnFixup *insns, uint32_t *fixupcontext)
127 {
128     /* Fix up the specified bootloader fragment and write it into
129      * guest memory using rom_add_blob_fixed(). fixupcontext is
130      * an array giving the values to write in for the fixup types
131      * which write a value into the code array.
132      */
133     int i, len;
134     uint32_t *code;
135 
136     len = 0;
137     while (insns[len].fixup != FIXUP_TERMINATOR) {
138         len++;
139     }
140 
141     code = g_new0(uint32_t, len);
142 
143     for (i = 0; i < len; i++) {
144         uint32_t insn = insns[i].insn;
145         FixupType fixup = insns[i].fixup;
146 
147         switch (fixup) {
148         case FIXUP_NONE:
149             break;
150         case FIXUP_BOARDID:
151         case FIXUP_BOARD_SETUP:
152         case FIXUP_ARGPTR:
153         case FIXUP_ENTRYPOINT:
154         case FIXUP_GIC_CPU_IF:
155         case FIXUP_BOOTREG:
156         case FIXUP_DSB:
157             insn = fixupcontext[fixup];
158             break;
159         default:
160             abort();
161         }
162         code[i] = tswap32(insn);
163     }
164 
165     rom_add_blob_fixed(name, code, len * sizeof(uint32_t), addr);
166 
167     g_free(code);
168 }
169 
170 static void default_write_secondary(ARMCPU *cpu,
171                                     const struct arm_boot_info *info)
172 {
173     uint32_t fixupcontext[FIXUP_MAX];
174 
175     fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr;
176     fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr;
177     if (arm_feature(&cpu->env, ARM_FEATURE_V7)) {
178         fixupcontext[FIXUP_DSB] = DSB_INSN;
179     } else {
180         fixupcontext[FIXUP_DSB] = CP15_DSB_INSN;
181     }
182 
183     write_bootloader("smpboot", info->smp_loader_start,
184                      smpboot, fixupcontext);
185 }
186 
187 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
188                                             const struct arm_boot_info *info,
189                                             hwaddr mvbar_addr)
190 {
191     int n;
192     uint32_t mvbar_blob[] = {
193         /* mvbar_addr: secure monitor vectors
194          * Default unimplemented and unused vectors to spin. Makes it
195          * easier to debug (as opposed to the CPU running away).
196          */
197         0xeafffffe, /* (spin) */
198         0xeafffffe, /* (spin) */
199         0xe1b0f00e, /* movs pc, lr ;SMC exception return */
200         0xeafffffe, /* (spin) */
201         0xeafffffe, /* (spin) */
202         0xeafffffe, /* (spin) */
203         0xeafffffe, /* (spin) */
204         0xeafffffe, /* (spin) */
205     };
206     uint32_t board_setup_blob[] = {
207         /* board setup addr */
208         0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */
209         0xee0c0f30, /* mcr     p15, 0, r0, c12, c0, 1 ;set MVBAR */
210         0xee110f11, /* mrc     p15, 0, r0, c1 , c1, 0 ;read SCR */
211         0xe3800031, /* orr     r0, #0x31              ;enable AW, FW, NS */
212         0xee010f11, /* mcr     p15, 0, r0, c1, c1, 0  ;write SCR */
213         0xe1a0100e, /* mov     r1, lr                 ;save LR across SMC */
214         0xe1600070, /* smc     #0                     ;call monitor to flush SCR */
215         0xe1a0f001, /* mov     pc, r1                 ;return */
216     };
217 
218     /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
219     assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
220 
221     /* check that these blobs don't overlap */
222     assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr)
223           || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr));
224 
225     for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
226         mvbar_blob[n] = tswap32(mvbar_blob[n]);
227     }
228     rom_add_blob_fixed("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
229                        mvbar_addr);
230 
231     for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
232         board_setup_blob[n] = tswap32(board_setup_blob[n]);
233     }
234     rom_add_blob_fixed("board-setup", board_setup_blob,
235                        sizeof(board_setup_blob), info->board_setup_addr);
236 }
237 
238 static void default_reset_secondary(ARMCPU *cpu,
239                                     const struct arm_boot_info *info)
240 {
241     CPUState *cs = CPU(cpu);
242 
243     address_space_stl_notdirty(&address_space_memory, info->smp_bootreg_addr,
244                                0, MEMTXATTRS_UNSPECIFIED, NULL);
245     cpu_set_pc(cs, info->smp_loader_start);
246 }
247 
248 static inline bool have_dtb(const struct arm_boot_info *info)
249 {
250     return info->dtb_filename || info->get_dtb;
251 }
252 
253 #define WRITE_WORD(p, value) do { \
254     address_space_stl_notdirty(&address_space_memory, p, value, \
255                                MEMTXATTRS_UNSPECIFIED, NULL);  \
256     p += 4;                       \
257 } while (0)
258 
259 static void set_kernel_args(const struct arm_boot_info *info)
260 {
261     int initrd_size = info->initrd_size;
262     hwaddr base = info->loader_start;
263     hwaddr p;
264 
265     p = base + KERNEL_ARGS_ADDR;
266     /* ATAG_CORE */
267     WRITE_WORD(p, 5);
268     WRITE_WORD(p, 0x54410001);
269     WRITE_WORD(p, 1);
270     WRITE_WORD(p, 0x1000);
271     WRITE_WORD(p, 0);
272     /* ATAG_MEM */
273     /* TODO: handle multiple chips on one ATAG list */
274     WRITE_WORD(p, 4);
275     WRITE_WORD(p, 0x54410002);
276     WRITE_WORD(p, info->ram_size);
277     WRITE_WORD(p, info->loader_start);
278     if (initrd_size) {
279         /* ATAG_INITRD2 */
280         WRITE_WORD(p, 4);
281         WRITE_WORD(p, 0x54420005);
282         WRITE_WORD(p, info->initrd_start);
283         WRITE_WORD(p, initrd_size);
284     }
285     if (info->kernel_cmdline && *info->kernel_cmdline) {
286         /* ATAG_CMDLINE */
287         int cmdline_size;
288 
289         cmdline_size = strlen(info->kernel_cmdline);
290         cpu_physical_memory_write(p + 8, info->kernel_cmdline,
291                                   cmdline_size + 1);
292         cmdline_size = (cmdline_size >> 2) + 1;
293         WRITE_WORD(p, cmdline_size + 2);
294         WRITE_WORD(p, 0x54410009);
295         p += cmdline_size * 4;
296     }
297     if (info->atag_board) {
298         /* ATAG_BOARD */
299         int atag_board_len;
300         uint8_t atag_board_buf[0x1000];
301 
302         atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
303         WRITE_WORD(p, (atag_board_len + 8) >> 2);
304         WRITE_WORD(p, 0x414f4d50);
305         cpu_physical_memory_write(p, atag_board_buf, atag_board_len);
306         p += atag_board_len;
307     }
308     /* ATAG_END */
309     WRITE_WORD(p, 0);
310     WRITE_WORD(p, 0);
311 }
312 
313 static void set_kernel_args_old(const struct arm_boot_info *info)
314 {
315     hwaddr p;
316     const char *s;
317     int initrd_size = info->initrd_size;
318     hwaddr base = info->loader_start;
319 
320     /* see linux/include/asm-arm/setup.h */
321     p = base + KERNEL_ARGS_ADDR;
322     /* page_size */
323     WRITE_WORD(p, 4096);
324     /* nr_pages */
325     WRITE_WORD(p, info->ram_size / 4096);
326     /* ramdisk_size */
327     WRITE_WORD(p, 0);
328 #define FLAG_READONLY	1
329 #define FLAG_RDLOAD	4
330 #define FLAG_RDPROMPT	8
331     /* flags */
332     WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
333     /* rootdev */
334     WRITE_WORD(p, (31 << 8) | 0);	/* /dev/mtdblock0 */
335     /* video_num_cols */
336     WRITE_WORD(p, 0);
337     /* video_num_rows */
338     WRITE_WORD(p, 0);
339     /* video_x */
340     WRITE_WORD(p, 0);
341     /* video_y */
342     WRITE_WORD(p, 0);
343     /* memc_control_reg */
344     WRITE_WORD(p, 0);
345     /* unsigned char sounddefault */
346     /* unsigned char adfsdrives */
347     /* unsigned char bytes_per_char_h */
348     /* unsigned char bytes_per_char_v */
349     WRITE_WORD(p, 0);
350     /* pages_in_bank[4] */
351     WRITE_WORD(p, 0);
352     WRITE_WORD(p, 0);
353     WRITE_WORD(p, 0);
354     WRITE_WORD(p, 0);
355     /* pages_in_vram */
356     WRITE_WORD(p, 0);
357     /* initrd_start */
358     if (initrd_size) {
359         WRITE_WORD(p, info->initrd_start);
360     } else {
361         WRITE_WORD(p, 0);
362     }
363     /* initrd_size */
364     WRITE_WORD(p, initrd_size);
365     /* rd_start */
366     WRITE_WORD(p, 0);
367     /* system_rev */
368     WRITE_WORD(p, 0);
369     /* system_serial_low */
370     WRITE_WORD(p, 0);
371     /* system_serial_high */
372     WRITE_WORD(p, 0);
373     /* mem_fclk_21285 */
374     WRITE_WORD(p, 0);
375     /* zero unused fields */
376     while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
377         WRITE_WORD(p, 0);
378     }
379     s = info->kernel_cmdline;
380     if (s) {
381         cpu_physical_memory_write(p, s, strlen(s) + 1);
382     } else {
383         WRITE_WORD(p, 0);
384     }
385 }
386 
387 /**
388  * load_dtb() - load a device tree binary image into memory
389  * @addr:       the address to load the image at
390  * @binfo:      struct describing the boot environment
391  * @addr_limit: upper limit of the available memory area at @addr
392  *
393  * Load a device tree supplied by the machine or by the user  with the
394  * '-dtb' command line option, and put it at offset @addr in target
395  * memory.
396  *
397  * If @addr_limit contains a meaningful value (i.e., it is strictly greater
398  * than @addr), the device tree is only loaded if its size does not exceed
399  * the limit.
400  *
401  * Returns: the size of the device tree image on success,
402  *          0 if the image size exceeds the limit,
403  *          -1 on errors.
404  *
405  * Note: Must not be called unless have_dtb(binfo) is true.
406  */
407 static int load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
408                     hwaddr addr_limit)
409 {
410     void *fdt = NULL;
411     int size, rc;
412     uint32_t acells, scells;
413     char *nodename;
414     unsigned int i;
415     hwaddr mem_base, mem_len;
416 
417     if (binfo->dtb_filename) {
418         char *filename;
419         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
420         if (!filename) {
421             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
422             goto fail;
423         }
424 
425         fdt = load_device_tree(filename, &size);
426         if (!fdt) {
427             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
428             g_free(filename);
429             goto fail;
430         }
431         g_free(filename);
432     } else {
433         fdt = binfo->get_dtb(binfo, &size);
434         if (!fdt) {
435             fprintf(stderr, "Board was unable to create a dtb blob\n");
436             goto fail;
437         }
438     }
439 
440     if (addr_limit > addr && size > (addr_limit - addr)) {
441         /* Installing the device tree blob at addr would exceed addr_limit.
442          * Whether this constitutes failure is up to the caller to decide,
443          * so just return 0 as size, i.e., no error.
444          */
445         g_free(fdt);
446         return 0;
447     }
448 
449     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
450                                    NULL, &error_fatal);
451     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
452                                    NULL, &error_fatal);
453     if (acells == 0 || scells == 0) {
454         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
455         goto fail;
456     }
457 
458     if (scells < 2 && binfo->ram_size >= (1ULL << 32)) {
459         /* This is user error so deserves a friendlier error message
460          * than the failure of setprop_sized_cells would provide
461          */
462         fprintf(stderr, "qemu: dtb file not compatible with "
463                 "RAM size > 4GB\n");
464         goto fail;
465     }
466 
467     if (nb_numa_nodes > 0) {
468         /*
469          * Turn the /memory node created before into a NOP node, then create
470          * /memory@addr nodes for all numa nodes respectively.
471          */
472         qemu_fdt_nop_node(fdt, "/memory");
473         mem_base = binfo->loader_start;
474         for (i = 0; i < nb_numa_nodes; i++) {
475             mem_len = numa_info[i].node_mem;
476             nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
477             qemu_fdt_add_subnode(fdt, nodename);
478             qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
479             rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
480                                               acells, mem_base,
481                                               scells, mem_len);
482             if (rc < 0) {
483                 fprintf(stderr, "couldn't set %s/reg for node %d\n", nodename,
484                         i);
485                 goto fail;
486             }
487 
488             qemu_fdt_setprop_cell(fdt, nodename, "numa-node-id", i);
489             mem_base += mem_len;
490             g_free(nodename);
491         }
492     } else {
493         Error *err = NULL;
494 
495         rc = fdt_path_offset(fdt, "/memory");
496         if (rc < 0) {
497             qemu_fdt_add_subnode(fdt, "/memory");
498         }
499 
500         if (!qemu_fdt_getprop(fdt, "/memory", "device_type", NULL, &err)) {
501             qemu_fdt_setprop_string(fdt, "/memory", "device_type", "memory");
502         }
503 
504         rc = qemu_fdt_setprop_sized_cells(fdt, "/memory", "reg",
505                                           acells, binfo->loader_start,
506                                           scells, binfo->ram_size);
507         if (rc < 0) {
508             fprintf(stderr, "couldn't set /memory/reg\n");
509             goto fail;
510         }
511     }
512 
513     rc = fdt_path_offset(fdt, "/chosen");
514     if (rc < 0) {
515         qemu_fdt_add_subnode(fdt, "/chosen");
516     }
517 
518     if (binfo->kernel_cmdline && *binfo->kernel_cmdline) {
519         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
520                                      binfo->kernel_cmdline);
521         if (rc < 0) {
522             fprintf(stderr, "couldn't set /chosen/bootargs\n");
523             goto fail;
524         }
525     }
526 
527     if (binfo->initrd_size) {
528         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
529                                    binfo->initrd_start);
530         if (rc < 0) {
531             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
532             goto fail;
533         }
534 
535         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
536                                    binfo->initrd_start + binfo->initrd_size);
537         if (rc < 0) {
538             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
539             goto fail;
540         }
541     }
542 
543     if (binfo->modify_dtb) {
544         binfo->modify_dtb(binfo, fdt);
545     }
546 
547     qemu_fdt_dumpdtb(fdt, size);
548 
549     /* Put the DTB into the memory map as a ROM image: this will ensure
550      * the DTB is copied again upon reset, even if addr points into RAM.
551      */
552     rom_add_blob_fixed("dtb", fdt, size, addr);
553 
554     g_free(fdt);
555 
556     return size;
557 
558 fail:
559     g_free(fdt);
560     return -1;
561 }
562 
563 static void do_cpu_reset(void *opaque)
564 {
565     ARMCPU *cpu = opaque;
566     CPUState *cs = CPU(cpu);
567     CPUARMState *env = &cpu->env;
568     const struct arm_boot_info *info = env->boot_info;
569 
570     cpu_reset(cs);
571     if (info) {
572         if (!info->is_linux) {
573             int i;
574             /* Jump to the entry point.  */
575             uint64_t entry = info->entry;
576 
577             switch (info->endianness) {
578             case ARM_ENDIANNESS_LE:
579                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
580                 for (i = 1; i < 4; ++i) {
581                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
582                 }
583                 env->uncached_cpsr &= ~CPSR_E;
584                 break;
585             case ARM_ENDIANNESS_BE8:
586                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
587                 for (i = 1; i < 4; ++i) {
588                     env->cp15.sctlr_el[i] |= SCTLR_EE;
589                 }
590                 env->uncached_cpsr |= CPSR_E;
591                 break;
592             case ARM_ENDIANNESS_BE32:
593                 env->cp15.sctlr_el[1] |= SCTLR_B;
594                 break;
595             case ARM_ENDIANNESS_UNKNOWN:
596                 break; /* Board's decision */
597             default:
598                 g_assert_not_reached();
599             }
600 
601             if (!env->aarch64) {
602                 env->thumb = info->entry & 1;
603                 entry &= 0xfffffffe;
604             }
605             cpu_set_pc(cs, entry);
606         } else {
607             /* If we are booting Linux then we need to check whether we are
608              * booting into secure or non-secure state and adjust the state
609              * accordingly.  Out of reset, ARM is defined to be in secure state
610              * (SCR.NS = 0), we change that here if non-secure boot has been
611              * requested.
612              */
613             if (arm_feature(env, ARM_FEATURE_EL3)) {
614                 /* AArch64 is defined to come out of reset into EL3 if enabled.
615                  * If we are booting Linux then we need to adjust our EL as
616                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
617                  * SVC, which Linux expects, so no privilege/exception level to
618                  * adjust.
619                  */
620                 if (env->aarch64) {
621                     env->cp15.scr_el3 |= SCR_RW;
622                     if (arm_feature(env, ARM_FEATURE_EL2)) {
623                         env->cp15.hcr_el2 |= HCR_RW;
624                         env->pstate = PSTATE_MODE_EL2h;
625                     } else {
626                         env->pstate = PSTATE_MODE_EL1h;
627                     }
628                 }
629 
630                 /* Set to non-secure if not a secure boot */
631                 if (!info->secure_boot &&
632                     (cs != first_cpu || !info->secure_board_setup)) {
633                     /* Linux expects non-secure state */
634                     env->cp15.scr_el3 |= SCR_NS;
635                 }
636             }
637 
638             if (cs == first_cpu) {
639                 cpu_set_pc(cs, info->loader_start);
640 
641                 if (!have_dtb(info)) {
642                     if (old_param) {
643                         set_kernel_args_old(info);
644                     } else {
645                         set_kernel_args(info);
646                     }
647                 }
648             } else {
649                 info->secondary_cpu_reset_hook(cpu, info);
650             }
651         }
652     }
653 }
654 
655 /**
656  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
657  *                          by key.
658  * @fw_cfg:         The firmware config instance to store the data in.
659  * @size_key:       The firmware config key to store the size of the loaded
660  *                  data under, with fw_cfg_add_i32().
661  * @data_key:       The firmware config key to store the loaded data under,
662  *                  with fw_cfg_add_bytes().
663  * @image_name:     The name of the image file to load. If it is NULL, the
664  *                  function returns without doing anything.
665  * @try_decompress: Whether the image should be decompressed (gunzipped) before
666  *                  adding it to fw_cfg. If decompression fails, the image is
667  *                  loaded as-is.
668  *
669  * In case of failure, the function prints an error message to stderr and the
670  * process exits with status 1.
671  */
672 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
673                                  uint16_t data_key, const char *image_name,
674                                  bool try_decompress)
675 {
676     size_t size = -1;
677     uint8_t *data;
678 
679     if (image_name == NULL) {
680         return;
681     }
682 
683     if (try_decompress) {
684         size = load_image_gzipped_buffer(image_name,
685                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
686     }
687 
688     if (size == (size_t)-1) {
689         gchar *contents;
690         gsize length;
691 
692         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
693             fprintf(stderr, "failed to load \"%s\"\n", image_name);
694             exit(1);
695         }
696         size = length;
697         data = (uint8_t *)contents;
698     }
699 
700     fw_cfg_add_i32(fw_cfg, size_key, size);
701     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
702 }
703 
704 static int do_arm_linux_init(Object *obj, void *opaque)
705 {
706     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
707         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
708         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
709         struct arm_boot_info *info = opaque;
710 
711         if (albifc->arm_linux_init) {
712             albifc->arm_linux_init(albif, info->secure_boot);
713         }
714     }
715     return 0;
716 }
717 
718 static uint64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
719                              uint64_t *lowaddr, uint64_t *highaddr,
720                              int elf_machine)
721 {
722     bool elf_is64;
723     union {
724         Elf32_Ehdr h32;
725         Elf64_Ehdr h64;
726     } elf_header;
727     int data_swab = 0;
728     bool big_endian;
729     uint64_t ret = -1;
730     Error *err = NULL;
731 
732 
733     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
734     if (err) {
735         return ret;
736     }
737 
738     if (elf_is64) {
739         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
740         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
741                                       : ARM_ENDIANNESS_LE;
742     } else {
743         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
744         if (big_endian) {
745             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
746                 info->endianness = ARM_ENDIANNESS_BE8;
747             } else {
748                 info->endianness = ARM_ENDIANNESS_BE32;
749                 /* In BE32, the CPU has a different view of the per-byte
750                  * address map than the rest of the system. BE32 ELF files
751                  * are organised such that they can be programmed through
752                  * the CPU's per-word byte-reversed view of the world. QEMU
753                  * however loads ELF files independently of the CPU. So
754                  * tell the ELF loader to byte reverse the data for us.
755                  */
756                 data_swab = 2;
757             }
758         } else {
759             info->endianness = ARM_ENDIANNESS_LE;
760         }
761     }
762 
763     ret = load_elf(info->kernel_filename, NULL, NULL,
764                    pentry, lowaddr, highaddr, big_endian, elf_machine,
765                    1, data_swab);
766     if (ret <= 0) {
767         /* The header loaded but the image didn't */
768         exit(1);
769     }
770 
771     return ret;
772 }
773 
774 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
775                                    hwaddr *entry)
776 {
777     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
778     uint8_t *buffer;
779     int size;
780 
781     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
782     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
783                                      &buffer);
784 
785     if (size < 0) {
786         gsize len;
787 
788         /* Load as raw file otherwise */
789         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
790             return -1;
791         }
792         size = len;
793     }
794 
795     /* check the arm64 magic header value -- very old kernels may not have it */
796     if (memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
797         uint64_t hdrvals[2];
798 
799         /* The arm64 Image header has text_offset and image_size fields at 8 and
800          * 16 bytes into the Image header, respectively. The text_offset field
801          * is only valid if the image_size is non-zero.
802          */
803         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
804         if (hdrvals[1] != 0) {
805             kernel_load_offset = le64_to_cpu(hdrvals[0]);
806         }
807     }
808 
809     *entry = mem_base + kernel_load_offset;
810     rom_add_blob_fixed(filename, buffer, size, *entry);
811 
812     g_free(buffer);
813 
814     return size;
815 }
816 
817 static void arm_load_kernel_notify(Notifier *notifier, void *data)
818 {
819     CPUState *cs;
820     int kernel_size;
821     int initrd_size;
822     int is_linux = 0;
823     uint64_t elf_entry, elf_low_addr, elf_high_addr;
824     int elf_machine;
825     hwaddr entry;
826     static const ARMInsnFixup *primary_loader;
827     ArmLoadKernelNotifier *n = DO_UPCAST(ArmLoadKernelNotifier,
828                                          notifier, notifier);
829     ARMCPU *cpu = n->cpu;
830     struct arm_boot_info *info =
831         container_of(n, struct arm_boot_info, load_kernel_notifier);
832 
833     /* The board code is not supposed to set secure_board_setup unless
834      * running its code in secure mode is actually possible, and KVM
835      * doesn't support secure.
836      */
837     assert(!(info->secure_board_setup && kvm_enabled()));
838 
839     info->dtb_filename = qemu_opt_get(qemu_get_machine_opts(), "dtb");
840 
841     /* Load the kernel.  */
842     if (!info->kernel_filename || info->firmware_loaded) {
843 
844         if (have_dtb(info)) {
845             /* If we have a device tree blob, but no kernel to supply it to (or
846              * the kernel is supposed to be loaded by the bootloader), copy the
847              * DTB to the base of RAM for the bootloader to pick up.
848              */
849             if (load_dtb(info->loader_start, info, 0) < 0) {
850                 exit(1);
851             }
852         }
853 
854         if (info->kernel_filename) {
855             FWCfgState *fw_cfg;
856             bool try_decompressing_kernel;
857 
858             fw_cfg = fw_cfg_find();
859             try_decompressing_kernel = arm_feature(&cpu->env,
860                                                    ARM_FEATURE_AARCH64);
861 
862             /* Expose the kernel, the command line, and the initrd in fw_cfg.
863              * We don't process them here at all, it's all left to the
864              * firmware.
865              */
866             load_image_to_fw_cfg(fw_cfg,
867                                  FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
868                                  info->kernel_filename,
869                                  try_decompressing_kernel);
870             load_image_to_fw_cfg(fw_cfg,
871                                  FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
872                                  info->initrd_filename, false);
873 
874             if (info->kernel_cmdline) {
875                 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
876                                strlen(info->kernel_cmdline) + 1);
877                 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
878                                   info->kernel_cmdline);
879             }
880         }
881 
882         /* We will start from address 0 (typically a boot ROM image) in the
883          * same way as hardware.
884          */
885         return;
886     }
887 
888     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
889         primary_loader = bootloader_aarch64;
890         elf_machine = EM_AARCH64;
891     } else {
892         primary_loader = bootloader;
893         if (!info->write_board_setup) {
894             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
895         }
896         elf_machine = EM_ARM;
897     }
898 
899     if (!info->secondary_cpu_reset_hook) {
900         info->secondary_cpu_reset_hook = default_reset_secondary;
901     }
902     if (!info->write_secondary_boot) {
903         info->write_secondary_boot = default_write_secondary;
904     }
905 
906     if (info->nb_cpus == 0)
907         info->nb_cpus = 1;
908 
909     /* We want to put the initrd far enough into RAM that when the
910      * kernel is uncompressed it will not clobber the initrd. However
911      * on boards without much RAM we must ensure that we still leave
912      * enough room for a decent sized initrd, and on boards with large
913      * amounts of RAM we must avoid the initrd being so far up in RAM
914      * that it is outside lowmem and inaccessible to the kernel.
915      * So for boards with less  than 256MB of RAM we put the initrd
916      * halfway into RAM, and for boards with 256MB of RAM or more we put
917      * the initrd at 128MB.
918      */
919     info->initrd_start = info->loader_start +
920         MIN(info->ram_size / 2, 128 * 1024 * 1024);
921 
922     /* Assume that raw images are linux kernels, and ELF images are not.  */
923     kernel_size = arm_load_elf(info, &elf_entry, &elf_low_addr,
924                                &elf_high_addr, elf_machine);
925     if (kernel_size > 0 && have_dtb(info)) {
926         /* If there is still some room left at the base of RAM, try and put
927          * the DTB there like we do for images loaded with -bios or -pflash.
928          */
929         if (elf_low_addr > info->loader_start
930             || elf_high_addr < info->loader_start) {
931             /* Pass elf_low_addr as address limit to load_dtb if it may be
932              * pointing into RAM, otherwise pass '0' (no limit)
933              */
934             if (elf_low_addr < info->loader_start) {
935                 elf_low_addr = 0;
936             }
937             if (load_dtb(info->loader_start, info, elf_low_addr) < 0) {
938                 exit(1);
939             }
940         }
941     }
942     entry = elf_entry;
943     if (kernel_size < 0) {
944         kernel_size = load_uimage(info->kernel_filename, &entry, NULL,
945                                   &is_linux, NULL, NULL);
946     }
947     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
948         kernel_size = load_aarch64_image(info->kernel_filename,
949                                          info->loader_start, &entry);
950         is_linux = 1;
951     } else if (kernel_size < 0) {
952         /* 32-bit ARM */
953         entry = info->loader_start + KERNEL_LOAD_ADDR;
954         kernel_size = load_image_targphys(info->kernel_filename, entry,
955                                           info->ram_size - KERNEL_LOAD_ADDR);
956         is_linux = 1;
957     }
958     if (kernel_size < 0) {
959         fprintf(stderr, "qemu: could not load kernel '%s'\n",
960                 info->kernel_filename);
961         exit(1);
962     }
963     info->entry = entry;
964     if (is_linux) {
965         uint32_t fixupcontext[FIXUP_MAX];
966 
967         if (info->initrd_filename) {
968             initrd_size = load_ramdisk(info->initrd_filename,
969                                        info->initrd_start,
970                                        info->ram_size -
971                                        info->initrd_start);
972             if (initrd_size < 0) {
973                 initrd_size = load_image_targphys(info->initrd_filename,
974                                                   info->initrd_start,
975                                                   info->ram_size -
976                                                   info->initrd_start);
977             }
978             if (initrd_size < 0) {
979                 fprintf(stderr, "qemu: could not load initrd '%s'\n",
980                         info->initrd_filename);
981                 exit(1);
982             }
983         } else {
984             initrd_size = 0;
985         }
986         info->initrd_size = initrd_size;
987 
988         fixupcontext[FIXUP_BOARDID] = info->board_id;
989         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
990 
991         /* for device tree boot, we pass the DTB directly in r2. Otherwise
992          * we point to the kernel args.
993          */
994         if (have_dtb(info)) {
995             hwaddr align;
996             hwaddr dtb_start;
997 
998             if (elf_machine == EM_AARCH64) {
999                 /*
1000                  * Some AArch64 kernels on early bootup map the fdt region as
1001                  *
1002                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1003                  *
1004                  * Let's play safe and prealign it to 2MB to give us some space.
1005                  */
1006                 align = 2 * 1024 * 1024;
1007             } else {
1008                 /*
1009                  * Some 32bit kernels will trash anything in the 4K page the
1010                  * initrd ends in, so make sure the DTB isn't caught up in that.
1011                  */
1012                 align = 4096;
1013             }
1014 
1015             /* Place the DTB after the initrd in memory with alignment. */
1016             dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size, align);
1017             if (load_dtb(dtb_start, info, 0) < 0) {
1018                 exit(1);
1019             }
1020             fixupcontext[FIXUP_ARGPTR] = dtb_start;
1021         } else {
1022             fixupcontext[FIXUP_ARGPTR] = info->loader_start + KERNEL_ARGS_ADDR;
1023             if (info->ram_size >= (1ULL << 32)) {
1024                 fprintf(stderr, "qemu: RAM size must be less than 4GB to boot"
1025                         " Linux kernel using ATAGS (try passing a device tree"
1026                         " using -dtb)\n");
1027                 exit(1);
1028             }
1029         }
1030         fixupcontext[FIXUP_ENTRYPOINT] = entry;
1031 
1032         write_bootloader("bootloader", info->loader_start,
1033                          primary_loader, fixupcontext);
1034 
1035         if (info->nb_cpus > 1) {
1036             info->write_secondary_boot(cpu, info);
1037         }
1038         if (info->write_board_setup) {
1039             info->write_board_setup(cpu, info);
1040         }
1041 
1042         /* Notify devices which need to fake up firmware initialization
1043          * that we're doing a direct kernel boot.
1044          */
1045         object_child_foreach_recursive(object_get_root(),
1046                                        do_arm_linux_init, info);
1047     }
1048     info->is_linux = is_linux;
1049 
1050     for (cs = CPU(cpu); cs; cs = CPU_NEXT(cs)) {
1051         ARM_CPU(cs)->env.boot_info = info;
1052     }
1053 }
1054 
1055 void arm_load_kernel(ARMCPU *cpu, struct arm_boot_info *info)
1056 {
1057     CPUState *cs;
1058 
1059     info->load_kernel_notifier.cpu = cpu;
1060     info->load_kernel_notifier.notifier.notify = arm_load_kernel_notify;
1061     qemu_add_machine_init_done_notifier(&info->load_kernel_notifier.notifier);
1062 
1063     /* CPU objects (unlike devices) are not automatically reset on system
1064      * reset, so we must always register a handler to do so. If we're
1065      * actually loading a kernel, the handler is also responsible for
1066      * arranging that we start it correctly.
1067      */
1068     for (cs = CPU(cpu); cs; cs = CPU_NEXT(cs)) {
1069         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
1070     }
1071 }
1072 
1073 static const TypeInfo arm_linux_boot_if_info = {
1074     .name = TYPE_ARM_LINUX_BOOT_IF,
1075     .parent = TYPE_INTERFACE,
1076     .class_size = sizeof(ARMLinuxBootIfClass),
1077 };
1078 
1079 static void arm_linux_boot_register_types(void)
1080 {
1081     type_register_static(&arm_linux_boot_if_info);
1082 }
1083 
1084 type_init(arm_linux_boot_register_types)
1085