xref: /openbmc/qemu/hw/arm/boot.c (revision 922a01a013d2270682a188258cbccacfecf8129c)
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 
28 /* Kernel boot protocol is specified in the kernel docs
29  * Documentation/arm/Booting and Documentation/arm64/booting.txt
30  * They have different preferred image load offsets from system RAM base.
31  */
32 #define KERNEL_ARGS_ADDR 0x100
33 #define KERNEL_LOAD_ADDR 0x00010000
34 #define KERNEL64_LOAD_ADDR 0x00080000
35 
36 #define ARM64_TEXT_OFFSET_OFFSET    8
37 #define ARM64_MAGIC_OFFSET          56
38 
39 typedef enum {
40     FIXUP_NONE = 0,     /* do nothing */
41     FIXUP_TERMINATOR,   /* end of insns */
42     FIXUP_BOARDID,      /* overwrite with board ID number */
43     FIXUP_BOARD_SETUP,  /* overwrite with board specific setup code address */
44     FIXUP_ARGPTR,       /* overwrite with pointer to kernel args */
45     FIXUP_ENTRYPOINT,   /* overwrite with kernel entry point */
46     FIXUP_GIC_CPU_IF,   /* overwrite with GIC CPU interface address */
47     FIXUP_BOOTREG,      /* overwrite with boot register address */
48     FIXUP_DSB,          /* overwrite with correct DSB insn for cpu */
49     FIXUP_MAX,
50 } FixupType;
51 
52 typedef struct ARMInsnFixup {
53     uint32_t insn;
54     FixupType fixup;
55 } ARMInsnFixup;
56 
57 static const ARMInsnFixup bootloader_aarch64[] = {
58     { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */
59     { 0xaa1f03e1 }, /* mov x1, xzr */
60     { 0xaa1f03e2 }, /* mov x2, xzr */
61     { 0xaa1f03e3 }, /* mov x3, xzr */
62     { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */
63     { 0xd61f0080 }, /* br x4      ; Jump to the kernel entry point */
64     { 0, FIXUP_ARGPTR }, /* arg: .word @DTB Lower 32-bits */
65     { 0 }, /* .word @DTB Higher 32-bits */
66     { 0, FIXUP_ENTRYPOINT }, /* entry: .word @Kernel Entry Lower 32-bits */
67     { 0 }, /* .word @Kernel Entry Higher 32-bits */
68     { 0, FIXUP_TERMINATOR }
69 };
70 
71 /* A very small bootloader: call the board-setup code (if needed),
72  * set r0-r2, then jump to the kernel.
73  * If we're not calling boot setup code then we don't copy across
74  * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array.
75  */
76 
77 static const ARMInsnFixup bootloader[] = {
78     { 0xe28fe004 }, /* add     lr, pc, #4 */
79     { 0xe51ff004 }, /* ldr     pc, [pc, #-4] */
80     { 0, FIXUP_BOARD_SETUP },
81 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3
82     { 0xe3a00000 }, /* mov     r0, #0 */
83     { 0xe59f1004 }, /* ldr     r1, [pc, #4] */
84     { 0xe59f2004 }, /* ldr     r2, [pc, #4] */
85     { 0xe59ff004 }, /* ldr     pc, [pc, #4] */
86     { 0, FIXUP_BOARDID },
87     { 0, FIXUP_ARGPTR },
88     { 0, FIXUP_ENTRYPOINT },
89     { 0, FIXUP_TERMINATOR }
90 };
91 
92 /* Handling for secondary CPU boot in a multicore system.
93  * Unlike the uniprocessor/primary CPU boot, this is platform
94  * dependent. The default code here is based on the secondary
95  * CPU boot protocol used on realview/vexpress boards, with
96  * some parameterisation to increase its flexibility.
97  * QEMU platform models for which this code is not appropriate
98  * should override write_secondary_boot and secondary_cpu_reset_hook
99  * instead.
100  *
101  * This code enables the interrupt controllers for the secondary
102  * CPUs and then puts all the secondary CPUs into a loop waiting
103  * for an interprocessor interrupt and polling a configurable
104  * location for the kernel secondary CPU entry point.
105  */
106 #define DSB_INSN 0xf57ff04f
107 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
108 
109 static const ARMInsnFixup smpboot[] = {
110     { 0xe59f2028 }, /* ldr r2, gic_cpu_if */
111     { 0xe59f0028 }, /* ldr r0, bootreg_addr */
112     { 0xe3a01001 }, /* mov r1, #1 */
113     { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */
114     { 0xe3a010ff }, /* mov r1, #0xff */
115     { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */
116     { 0, FIXUP_DSB },   /* dsb */
117     { 0xe320f003 }, /* wfi */
118     { 0xe5901000 }, /* ldr     r1, [r0] */
119     { 0xe1110001 }, /* tst     r1, r1 */
120     { 0x0afffffb }, /* beq     <wfi> */
121     { 0xe12fff11 }, /* bx      r1 */
122     { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */
123     { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */
124     { 0, FIXUP_TERMINATOR }
125 };
126 
127 static void write_bootloader(const char *name, hwaddr addr,
128                              const ARMInsnFixup *insns, uint32_t *fixupcontext)
129 {
130     /* Fix up the specified bootloader fragment and write it into
131      * guest memory using rom_add_blob_fixed(). fixupcontext is
132      * an array giving the values to write in for the fixup types
133      * which write a value into the code array.
134      */
135     int i, len;
136     uint32_t *code;
137 
138     len = 0;
139     while (insns[len].fixup != FIXUP_TERMINATOR) {
140         len++;
141     }
142 
143     code = g_new0(uint32_t, len);
144 
145     for (i = 0; i < len; i++) {
146         uint32_t insn = insns[i].insn;
147         FixupType fixup = insns[i].fixup;
148 
149         switch (fixup) {
150         case FIXUP_NONE:
151             break;
152         case FIXUP_BOARDID:
153         case FIXUP_BOARD_SETUP:
154         case FIXUP_ARGPTR:
155         case FIXUP_ENTRYPOINT:
156         case FIXUP_GIC_CPU_IF:
157         case FIXUP_BOOTREG:
158         case FIXUP_DSB:
159             insn = fixupcontext[fixup];
160             break;
161         default:
162             abort();
163         }
164         code[i] = tswap32(insn);
165     }
166 
167     rom_add_blob_fixed(name, code, len * sizeof(uint32_t), addr);
168 
169     g_free(code);
170 }
171 
172 static void default_write_secondary(ARMCPU *cpu,
173                                     const struct arm_boot_info *info)
174 {
175     uint32_t fixupcontext[FIXUP_MAX];
176 
177     fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr;
178     fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr;
179     if (arm_feature(&cpu->env, ARM_FEATURE_V7)) {
180         fixupcontext[FIXUP_DSB] = DSB_INSN;
181     } else {
182         fixupcontext[FIXUP_DSB] = CP15_DSB_INSN;
183     }
184 
185     write_bootloader("smpboot", info->smp_loader_start,
186                      smpboot, fixupcontext);
187 }
188 
189 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
190                                             const struct arm_boot_info *info,
191                                             hwaddr mvbar_addr)
192 {
193     int n;
194     uint32_t mvbar_blob[] = {
195         /* mvbar_addr: secure monitor vectors
196          * Default unimplemented and unused vectors to spin. Makes it
197          * easier to debug (as opposed to the CPU running away).
198          */
199         0xeafffffe, /* (spin) */
200         0xeafffffe, /* (spin) */
201         0xe1b0f00e, /* movs pc, lr ;SMC exception return */
202         0xeafffffe, /* (spin) */
203         0xeafffffe, /* (spin) */
204         0xeafffffe, /* (spin) */
205         0xeafffffe, /* (spin) */
206         0xeafffffe, /* (spin) */
207     };
208     uint32_t board_setup_blob[] = {
209         /* board setup addr */
210         0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */
211         0xee0c0f30, /* mcr     p15, 0, r0, c12, c0, 1 ;set MVBAR */
212         0xee110f11, /* mrc     p15, 0, r0, c1 , c1, 0 ;read SCR */
213         0xe3800031, /* orr     r0, #0x31              ;enable AW, FW, NS */
214         0xee010f11, /* mcr     p15, 0, r0, c1, c1, 0  ;write SCR */
215         0xe1a0100e, /* mov     r1, lr                 ;save LR across SMC */
216         0xe1600070, /* smc     #0                     ;call monitor to flush SCR */
217         0xe1a0f001, /* mov     pc, r1                 ;return */
218     };
219 
220     /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
221     assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
222 
223     /* check that these blobs don't overlap */
224     assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr)
225           || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr));
226 
227     for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
228         mvbar_blob[n] = tswap32(mvbar_blob[n]);
229     }
230     rom_add_blob_fixed("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
231                        mvbar_addr);
232 
233     for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
234         board_setup_blob[n] = tswap32(board_setup_blob[n]);
235     }
236     rom_add_blob_fixed("board-setup", board_setup_blob,
237                        sizeof(board_setup_blob), info->board_setup_addr);
238 }
239 
240 static void default_reset_secondary(ARMCPU *cpu,
241                                     const struct arm_boot_info *info)
242 {
243     CPUState *cs = CPU(cpu);
244 
245     address_space_stl_notdirty(&address_space_memory, info->smp_bootreg_addr,
246                                0, MEMTXATTRS_UNSPECIFIED, NULL);
247     cpu_set_pc(cs, info->smp_loader_start);
248 }
249 
250 static inline bool have_dtb(const struct arm_boot_info *info)
251 {
252     return info->dtb_filename || info->get_dtb;
253 }
254 
255 #define WRITE_WORD(p, value) do { \
256     address_space_stl_notdirty(&address_space_memory, p, value, \
257                                MEMTXATTRS_UNSPECIFIED, NULL);  \
258     p += 4;                       \
259 } while (0)
260 
261 static void set_kernel_args(const struct arm_boot_info *info)
262 {
263     int initrd_size = info->initrd_size;
264     hwaddr base = info->loader_start;
265     hwaddr p;
266 
267     p = base + KERNEL_ARGS_ADDR;
268     /* ATAG_CORE */
269     WRITE_WORD(p, 5);
270     WRITE_WORD(p, 0x54410001);
271     WRITE_WORD(p, 1);
272     WRITE_WORD(p, 0x1000);
273     WRITE_WORD(p, 0);
274     /* ATAG_MEM */
275     /* TODO: handle multiple chips on one ATAG list */
276     WRITE_WORD(p, 4);
277     WRITE_WORD(p, 0x54410002);
278     WRITE_WORD(p, info->ram_size);
279     WRITE_WORD(p, info->loader_start);
280     if (initrd_size) {
281         /* ATAG_INITRD2 */
282         WRITE_WORD(p, 4);
283         WRITE_WORD(p, 0x54420005);
284         WRITE_WORD(p, info->initrd_start);
285         WRITE_WORD(p, initrd_size);
286     }
287     if (info->kernel_cmdline && *info->kernel_cmdline) {
288         /* ATAG_CMDLINE */
289         int cmdline_size;
290 
291         cmdline_size = strlen(info->kernel_cmdline);
292         cpu_physical_memory_write(p + 8, info->kernel_cmdline,
293                                   cmdline_size + 1);
294         cmdline_size = (cmdline_size >> 2) + 1;
295         WRITE_WORD(p, cmdline_size + 2);
296         WRITE_WORD(p, 0x54410009);
297         p += cmdline_size * 4;
298     }
299     if (info->atag_board) {
300         /* ATAG_BOARD */
301         int atag_board_len;
302         uint8_t atag_board_buf[0x1000];
303 
304         atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
305         WRITE_WORD(p, (atag_board_len + 8) >> 2);
306         WRITE_WORD(p, 0x414f4d50);
307         cpu_physical_memory_write(p, atag_board_buf, atag_board_len);
308         p += atag_board_len;
309     }
310     /* ATAG_END */
311     WRITE_WORD(p, 0);
312     WRITE_WORD(p, 0);
313 }
314 
315 static void set_kernel_args_old(const struct arm_boot_info *info)
316 {
317     hwaddr p;
318     const char *s;
319     int initrd_size = info->initrd_size;
320     hwaddr base = info->loader_start;
321 
322     /* see linux/include/asm-arm/setup.h */
323     p = base + KERNEL_ARGS_ADDR;
324     /* page_size */
325     WRITE_WORD(p, 4096);
326     /* nr_pages */
327     WRITE_WORD(p, info->ram_size / 4096);
328     /* ramdisk_size */
329     WRITE_WORD(p, 0);
330 #define FLAG_READONLY	1
331 #define FLAG_RDLOAD	4
332 #define FLAG_RDPROMPT	8
333     /* flags */
334     WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
335     /* rootdev */
336     WRITE_WORD(p, (31 << 8) | 0);	/* /dev/mtdblock0 */
337     /* video_num_cols */
338     WRITE_WORD(p, 0);
339     /* video_num_rows */
340     WRITE_WORD(p, 0);
341     /* video_x */
342     WRITE_WORD(p, 0);
343     /* video_y */
344     WRITE_WORD(p, 0);
345     /* memc_control_reg */
346     WRITE_WORD(p, 0);
347     /* unsigned char sounddefault */
348     /* unsigned char adfsdrives */
349     /* unsigned char bytes_per_char_h */
350     /* unsigned char bytes_per_char_v */
351     WRITE_WORD(p, 0);
352     /* pages_in_bank[4] */
353     WRITE_WORD(p, 0);
354     WRITE_WORD(p, 0);
355     WRITE_WORD(p, 0);
356     WRITE_WORD(p, 0);
357     /* pages_in_vram */
358     WRITE_WORD(p, 0);
359     /* initrd_start */
360     if (initrd_size) {
361         WRITE_WORD(p, info->initrd_start);
362     } else {
363         WRITE_WORD(p, 0);
364     }
365     /* initrd_size */
366     WRITE_WORD(p, initrd_size);
367     /* rd_start */
368     WRITE_WORD(p, 0);
369     /* system_rev */
370     WRITE_WORD(p, 0);
371     /* system_serial_low */
372     WRITE_WORD(p, 0);
373     /* system_serial_high */
374     WRITE_WORD(p, 0);
375     /* mem_fclk_21285 */
376     WRITE_WORD(p, 0);
377     /* zero unused fields */
378     while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
379         WRITE_WORD(p, 0);
380     }
381     s = info->kernel_cmdline;
382     if (s) {
383         cpu_physical_memory_write(p, s, strlen(s) + 1);
384     } else {
385         WRITE_WORD(p, 0);
386     }
387 }
388 
389 /**
390  * load_dtb() - load a device tree binary image into memory
391  * @addr:       the address to load the image at
392  * @binfo:      struct describing the boot environment
393  * @addr_limit: upper limit of the available memory area at @addr
394  *
395  * Load a device tree supplied by the machine or by the user  with the
396  * '-dtb' command line option, and put it at offset @addr in target
397  * memory.
398  *
399  * If @addr_limit contains a meaningful value (i.e., it is strictly greater
400  * than @addr), the device tree is only loaded if its size does not exceed
401  * the limit.
402  *
403  * Returns: the size of the device tree image on success,
404  *          0 if the image size exceeds the limit,
405  *          -1 on errors.
406  *
407  * Note: Must not be called unless have_dtb(binfo) is true.
408  */
409 static int load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
410                     hwaddr addr_limit)
411 {
412     void *fdt = NULL;
413     int size, rc;
414     uint32_t acells, scells;
415     char *nodename;
416     unsigned int i;
417     hwaddr mem_base, mem_len;
418 
419     if (binfo->dtb_filename) {
420         char *filename;
421         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
422         if (!filename) {
423             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
424             goto fail;
425         }
426 
427         fdt = load_device_tree(filename, &size);
428         if (!fdt) {
429             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
430             g_free(filename);
431             goto fail;
432         }
433         g_free(filename);
434     } else {
435         fdt = binfo->get_dtb(binfo, &size);
436         if (!fdt) {
437             fprintf(stderr, "Board was unable to create a dtb blob\n");
438             goto fail;
439         }
440     }
441 
442     if (addr_limit > addr && size > (addr_limit - addr)) {
443         /* Installing the device tree blob at addr would exceed addr_limit.
444          * Whether this constitutes failure is up to the caller to decide,
445          * so just return 0 as size, i.e., no error.
446          */
447         g_free(fdt);
448         return 0;
449     }
450 
451     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
452                                    NULL, &error_fatal);
453     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
454                                    NULL, &error_fatal);
455     if (acells == 0 || scells == 0) {
456         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
457         goto fail;
458     }
459 
460     if (scells < 2 && binfo->ram_size >= (1ULL << 32)) {
461         /* This is user error so deserves a friendlier error message
462          * than the failure of setprop_sized_cells would provide
463          */
464         fprintf(stderr, "qemu: dtb file not compatible with "
465                 "RAM size > 4GB\n");
466         goto fail;
467     }
468 
469     if (nb_numa_nodes > 0) {
470         /*
471          * Turn the /memory node created before into a NOP node, then create
472          * /memory@addr nodes for all numa nodes respectively.
473          */
474         qemu_fdt_nop_node(fdt, "/memory");
475         mem_base = binfo->loader_start;
476         for (i = 0; i < nb_numa_nodes; i++) {
477             mem_len = numa_info[i].node_mem;
478             nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
479             qemu_fdt_add_subnode(fdt, nodename);
480             qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
481             rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
482                                               acells, mem_base,
483                                               scells, mem_len);
484             if (rc < 0) {
485                 fprintf(stderr, "couldn't set %s/reg for node %d\n", nodename,
486                         i);
487                 goto fail;
488             }
489 
490             qemu_fdt_setprop_cell(fdt, nodename, "numa-node-id", i);
491             mem_base += mem_len;
492             g_free(nodename);
493         }
494     } else {
495         Error *err = NULL;
496 
497         rc = fdt_path_offset(fdt, "/memory");
498         if (rc < 0) {
499             qemu_fdt_add_subnode(fdt, "/memory");
500         }
501 
502         if (!qemu_fdt_getprop(fdt, "/memory", "device_type", NULL, &err)) {
503             qemu_fdt_setprop_string(fdt, "/memory", "device_type", "memory");
504         }
505 
506         rc = qemu_fdt_setprop_sized_cells(fdt, "/memory", "reg",
507                                           acells, binfo->loader_start,
508                                           scells, binfo->ram_size);
509         if (rc < 0) {
510             fprintf(stderr, "couldn't set /memory/reg\n");
511             goto fail;
512         }
513     }
514 
515     rc = fdt_path_offset(fdt, "/chosen");
516     if (rc < 0) {
517         qemu_fdt_add_subnode(fdt, "/chosen");
518     }
519 
520     if (binfo->kernel_cmdline && *binfo->kernel_cmdline) {
521         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
522                                      binfo->kernel_cmdline);
523         if (rc < 0) {
524             fprintf(stderr, "couldn't set /chosen/bootargs\n");
525             goto fail;
526         }
527     }
528 
529     if (binfo->initrd_size) {
530         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
531                                    binfo->initrd_start);
532         if (rc < 0) {
533             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
534             goto fail;
535         }
536 
537         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
538                                    binfo->initrd_start + binfo->initrd_size);
539         if (rc < 0) {
540             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
541             goto fail;
542         }
543     }
544 
545     if (binfo->modify_dtb) {
546         binfo->modify_dtb(binfo, fdt);
547     }
548 
549     qemu_fdt_dumpdtb(fdt, size);
550 
551     /* Put the DTB into the memory map as a ROM image: this will ensure
552      * the DTB is copied again upon reset, even if addr points into RAM.
553      */
554     rom_add_blob_fixed("dtb", fdt, size, addr);
555 
556     g_free(fdt);
557 
558     return size;
559 
560 fail:
561     g_free(fdt);
562     return -1;
563 }
564 
565 static void do_cpu_reset(void *opaque)
566 {
567     ARMCPU *cpu = opaque;
568     CPUState *cs = CPU(cpu);
569     CPUARMState *env = &cpu->env;
570     const struct arm_boot_info *info = env->boot_info;
571 
572     cpu_reset(cs);
573     if (info) {
574         if (!info->is_linux) {
575             int i;
576             /* Jump to the entry point.  */
577             uint64_t entry = info->entry;
578 
579             switch (info->endianness) {
580             case ARM_ENDIANNESS_LE:
581                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
582                 for (i = 1; i < 4; ++i) {
583                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
584                 }
585                 env->uncached_cpsr &= ~CPSR_E;
586                 break;
587             case ARM_ENDIANNESS_BE8:
588                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
589                 for (i = 1; i < 4; ++i) {
590                     env->cp15.sctlr_el[i] |= SCTLR_EE;
591                 }
592                 env->uncached_cpsr |= CPSR_E;
593                 break;
594             case ARM_ENDIANNESS_BE32:
595                 env->cp15.sctlr_el[1] |= SCTLR_B;
596                 break;
597             case ARM_ENDIANNESS_UNKNOWN:
598                 break; /* Board's decision */
599             default:
600                 g_assert_not_reached();
601             }
602 
603             if (!env->aarch64) {
604                 env->thumb = info->entry & 1;
605                 entry &= 0xfffffffe;
606             }
607             cpu_set_pc(cs, entry);
608         } else {
609             /* If we are booting Linux then we need to check whether we are
610              * booting into secure or non-secure state and adjust the state
611              * accordingly.  Out of reset, ARM is defined to be in secure state
612              * (SCR.NS = 0), we change that here if non-secure boot has been
613              * requested.
614              */
615             if (arm_feature(env, ARM_FEATURE_EL3)) {
616                 /* AArch64 is defined to come out of reset into EL3 if enabled.
617                  * If we are booting Linux then we need to adjust our EL as
618                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
619                  * SVC, which Linux expects, so no privilege/exception level to
620                  * adjust.
621                  */
622                 if (env->aarch64) {
623                     env->cp15.scr_el3 |= SCR_RW;
624                     if (arm_feature(env, ARM_FEATURE_EL2)) {
625                         env->cp15.hcr_el2 |= HCR_RW;
626                         env->pstate = PSTATE_MODE_EL2h;
627                     } else {
628                         env->pstate = PSTATE_MODE_EL1h;
629                     }
630                 }
631 
632                 /* Set to non-secure if not a secure boot */
633                 if (!info->secure_boot &&
634                     (cs != first_cpu || !info->secure_board_setup)) {
635                     /* Linux expects non-secure state */
636                     env->cp15.scr_el3 |= SCR_NS;
637                 }
638             }
639 
640             if (cs == first_cpu) {
641                 cpu_set_pc(cs, info->loader_start);
642 
643                 if (!have_dtb(info)) {
644                     if (old_param) {
645                         set_kernel_args_old(info);
646                     } else {
647                         set_kernel_args(info);
648                     }
649                 }
650             } else {
651                 info->secondary_cpu_reset_hook(cpu, info);
652             }
653         }
654     }
655 }
656 
657 /**
658  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
659  *                          by key.
660  * @fw_cfg:         The firmware config instance to store the data in.
661  * @size_key:       The firmware config key to store the size of the loaded
662  *                  data under, with fw_cfg_add_i32().
663  * @data_key:       The firmware config key to store the loaded data under,
664  *                  with fw_cfg_add_bytes().
665  * @image_name:     The name of the image file to load. If it is NULL, the
666  *                  function returns without doing anything.
667  * @try_decompress: Whether the image should be decompressed (gunzipped) before
668  *                  adding it to fw_cfg. If decompression fails, the image is
669  *                  loaded as-is.
670  *
671  * In case of failure, the function prints an error message to stderr and the
672  * process exits with status 1.
673  */
674 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
675                                  uint16_t data_key, const char *image_name,
676                                  bool try_decompress)
677 {
678     size_t size = -1;
679     uint8_t *data;
680 
681     if (image_name == NULL) {
682         return;
683     }
684 
685     if (try_decompress) {
686         size = load_image_gzipped_buffer(image_name,
687                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
688     }
689 
690     if (size == (size_t)-1) {
691         gchar *contents;
692         gsize length;
693 
694         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
695             error_report("failed to load \"%s\"", image_name);
696             exit(1);
697         }
698         size = length;
699         data = (uint8_t *)contents;
700     }
701 
702     fw_cfg_add_i32(fw_cfg, size_key, size);
703     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
704 }
705 
706 static int do_arm_linux_init(Object *obj, void *opaque)
707 {
708     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
709         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
710         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
711         struct arm_boot_info *info = opaque;
712 
713         if (albifc->arm_linux_init) {
714             albifc->arm_linux_init(albif, info->secure_boot);
715         }
716     }
717     return 0;
718 }
719 
720 static uint64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
721                              uint64_t *lowaddr, uint64_t *highaddr,
722                              int elf_machine)
723 {
724     bool elf_is64;
725     union {
726         Elf32_Ehdr h32;
727         Elf64_Ehdr h64;
728     } elf_header;
729     int data_swab = 0;
730     bool big_endian;
731     uint64_t ret = -1;
732     Error *err = NULL;
733 
734 
735     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
736     if (err) {
737         return ret;
738     }
739 
740     if (elf_is64) {
741         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
742         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
743                                       : ARM_ENDIANNESS_LE;
744     } else {
745         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
746         if (big_endian) {
747             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
748                 info->endianness = ARM_ENDIANNESS_BE8;
749             } else {
750                 info->endianness = ARM_ENDIANNESS_BE32;
751                 /* In BE32, the CPU has a different view of the per-byte
752                  * address map than the rest of the system. BE32 ELF files
753                  * are organised such that they can be programmed through
754                  * the CPU's per-word byte-reversed view of the world. QEMU
755                  * however loads ELF files independently of the CPU. So
756                  * tell the ELF loader to byte reverse the data for us.
757                  */
758                 data_swab = 2;
759             }
760         } else {
761             info->endianness = ARM_ENDIANNESS_LE;
762         }
763     }
764 
765     ret = load_elf(info->kernel_filename, NULL, NULL,
766                    pentry, lowaddr, highaddr, big_endian, elf_machine,
767                    1, data_swab);
768     if (ret <= 0) {
769         /* The header loaded but the image didn't */
770         exit(1);
771     }
772 
773     return ret;
774 }
775 
776 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
777                                    hwaddr *entry)
778 {
779     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
780     uint8_t *buffer;
781     int size;
782 
783     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
784     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
785                                      &buffer);
786 
787     if (size < 0) {
788         gsize len;
789 
790         /* Load as raw file otherwise */
791         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
792             return -1;
793         }
794         size = len;
795     }
796 
797     /* check the arm64 magic header value -- very old kernels may not have it */
798     if (memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
799         uint64_t hdrvals[2];
800 
801         /* The arm64 Image header has text_offset and image_size fields at 8 and
802          * 16 bytes into the Image header, respectively. The text_offset field
803          * is only valid if the image_size is non-zero.
804          */
805         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
806         if (hdrvals[1] != 0) {
807             kernel_load_offset = le64_to_cpu(hdrvals[0]);
808         }
809     }
810 
811     *entry = mem_base + kernel_load_offset;
812     rom_add_blob_fixed(filename, buffer, size, *entry);
813 
814     g_free(buffer);
815 
816     return size;
817 }
818 
819 static void arm_load_kernel_notify(Notifier *notifier, void *data)
820 {
821     CPUState *cs;
822     int kernel_size;
823     int initrd_size;
824     int is_linux = 0;
825     uint64_t elf_entry, elf_low_addr, elf_high_addr;
826     int elf_machine;
827     hwaddr entry;
828     static const ARMInsnFixup *primary_loader;
829     ArmLoadKernelNotifier *n = DO_UPCAST(ArmLoadKernelNotifier,
830                                          notifier, notifier);
831     ARMCPU *cpu = n->cpu;
832     struct arm_boot_info *info =
833         container_of(n, struct arm_boot_info, load_kernel_notifier);
834 
835     /* The board code is not supposed to set secure_board_setup unless
836      * running its code in secure mode is actually possible, and KVM
837      * doesn't support secure.
838      */
839     assert(!(info->secure_board_setup && kvm_enabled()));
840 
841     info->dtb_filename = qemu_opt_get(qemu_get_machine_opts(), "dtb");
842 
843     /* Load the kernel.  */
844     if (!info->kernel_filename || info->firmware_loaded) {
845 
846         if (have_dtb(info)) {
847             /* If we have a device tree blob, but no kernel to supply it to (or
848              * the kernel is supposed to be loaded by the bootloader), copy the
849              * DTB to the base of RAM for the bootloader to pick up.
850              */
851             if (load_dtb(info->loader_start, info, 0) < 0) {
852                 exit(1);
853             }
854         }
855 
856         if (info->kernel_filename) {
857             FWCfgState *fw_cfg;
858             bool try_decompressing_kernel;
859 
860             fw_cfg = fw_cfg_find();
861             try_decompressing_kernel = arm_feature(&cpu->env,
862                                                    ARM_FEATURE_AARCH64);
863 
864             /* Expose the kernel, the command line, and the initrd in fw_cfg.
865              * We don't process them here at all, it's all left to the
866              * firmware.
867              */
868             load_image_to_fw_cfg(fw_cfg,
869                                  FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
870                                  info->kernel_filename,
871                                  try_decompressing_kernel);
872             load_image_to_fw_cfg(fw_cfg,
873                                  FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
874                                  info->initrd_filename, false);
875 
876             if (info->kernel_cmdline) {
877                 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
878                                strlen(info->kernel_cmdline) + 1);
879                 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
880                                   info->kernel_cmdline);
881             }
882         }
883 
884         /* We will start from address 0 (typically a boot ROM image) in the
885          * same way as hardware.
886          */
887         return;
888     }
889 
890     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
891         primary_loader = bootloader_aarch64;
892         elf_machine = EM_AARCH64;
893     } else {
894         primary_loader = bootloader;
895         if (!info->write_board_setup) {
896             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
897         }
898         elf_machine = EM_ARM;
899     }
900 
901     if (!info->secondary_cpu_reset_hook) {
902         info->secondary_cpu_reset_hook = default_reset_secondary;
903     }
904     if (!info->write_secondary_boot) {
905         info->write_secondary_boot = default_write_secondary;
906     }
907 
908     if (info->nb_cpus == 0)
909         info->nb_cpus = 1;
910 
911     /* We want to put the initrd far enough into RAM that when the
912      * kernel is uncompressed it will not clobber the initrd. However
913      * on boards without much RAM we must ensure that we still leave
914      * enough room for a decent sized initrd, and on boards with large
915      * amounts of RAM we must avoid the initrd being so far up in RAM
916      * that it is outside lowmem and inaccessible to the kernel.
917      * So for boards with less  than 256MB of RAM we put the initrd
918      * halfway into RAM, and for boards with 256MB of RAM or more we put
919      * the initrd at 128MB.
920      */
921     info->initrd_start = info->loader_start +
922         MIN(info->ram_size / 2, 128 * 1024 * 1024);
923 
924     /* Assume that raw images are linux kernels, and ELF images are not.  */
925     kernel_size = arm_load_elf(info, &elf_entry, &elf_low_addr,
926                                &elf_high_addr, elf_machine);
927     if (kernel_size > 0 && have_dtb(info)) {
928         /* If there is still some room left at the base of RAM, try and put
929          * the DTB there like we do for images loaded with -bios or -pflash.
930          */
931         if (elf_low_addr > info->loader_start
932             || elf_high_addr < info->loader_start) {
933             /* Pass elf_low_addr as address limit to load_dtb if it may be
934              * pointing into RAM, otherwise pass '0' (no limit)
935              */
936             if (elf_low_addr < info->loader_start) {
937                 elf_low_addr = 0;
938             }
939             if (load_dtb(info->loader_start, info, elf_low_addr) < 0) {
940                 exit(1);
941             }
942         }
943     }
944     entry = elf_entry;
945     if (kernel_size < 0) {
946         kernel_size = load_uimage(info->kernel_filename, &entry, NULL,
947                                   &is_linux, NULL, NULL);
948     }
949     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
950         kernel_size = load_aarch64_image(info->kernel_filename,
951                                          info->loader_start, &entry);
952         is_linux = 1;
953     } else if (kernel_size < 0) {
954         /* 32-bit ARM */
955         entry = info->loader_start + KERNEL_LOAD_ADDR;
956         kernel_size = load_image_targphys(info->kernel_filename, entry,
957                                           info->ram_size - KERNEL_LOAD_ADDR);
958         is_linux = 1;
959     }
960     if (kernel_size < 0) {
961         error_report("could not load kernel '%s'", info->kernel_filename);
962         exit(1);
963     }
964     info->entry = entry;
965     if (is_linux) {
966         uint32_t fixupcontext[FIXUP_MAX];
967 
968         if (info->initrd_filename) {
969             initrd_size = load_ramdisk(info->initrd_filename,
970                                        info->initrd_start,
971                                        info->ram_size -
972                                        info->initrd_start);
973             if (initrd_size < 0) {
974                 initrd_size = load_image_targphys(info->initrd_filename,
975                                                   info->initrd_start,
976                                                   info->ram_size -
977                                                   info->initrd_start);
978             }
979             if (initrd_size < 0) {
980                 error_report("could not load initrd '%s'",
981                              info->initrd_filename);
982                 exit(1);
983             }
984         } else {
985             initrd_size = 0;
986         }
987         info->initrd_size = initrd_size;
988 
989         fixupcontext[FIXUP_BOARDID] = info->board_id;
990         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
991 
992         /* for device tree boot, we pass the DTB directly in r2. Otherwise
993          * we point to the kernel args.
994          */
995         if (have_dtb(info)) {
996             hwaddr align;
997             hwaddr dtb_start;
998 
999             if (elf_machine == EM_AARCH64) {
1000                 /*
1001                  * Some AArch64 kernels on early bootup map the fdt region as
1002                  *
1003                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1004                  *
1005                  * Let's play safe and prealign it to 2MB to give us some space.
1006                  */
1007                 align = 2 * 1024 * 1024;
1008             } else {
1009                 /*
1010                  * Some 32bit kernels will trash anything in the 4K page the
1011                  * initrd ends in, so make sure the DTB isn't caught up in that.
1012                  */
1013                 align = 4096;
1014             }
1015 
1016             /* Place the DTB after the initrd in memory with alignment. */
1017             dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size, align);
1018             if (load_dtb(dtb_start, info, 0) < 0) {
1019                 exit(1);
1020             }
1021             fixupcontext[FIXUP_ARGPTR] = dtb_start;
1022         } else {
1023             fixupcontext[FIXUP_ARGPTR] = info->loader_start + KERNEL_ARGS_ADDR;
1024             if (info->ram_size >= (1ULL << 32)) {
1025                 error_report("RAM size must be less than 4GB to boot"
1026                              " Linux kernel using ATAGS (try passing a device tree"
1027                              " using -dtb)");
1028                 exit(1);
1029             }
1030         }
1031         fixupcontext[FIXUP_ENTRYPOINT] = entry;
1032 
1033         write_bootloader("bootloader", info->loader_start,
1034                          primary_loader, fixupcontext);
1035 
1036         if (info->nb_cpus > 1) {
1037             info->write_secondary_boot(cpu, info);
1038         }
1039         if (info->write_board_setup) {
1040             info->write_board_setup(cpu, info);
1041         }
1042 
1043         /* Notify devices which need to fake up firmware initialization
1044          * that we're doing a direct kernel boot.
1045          */
1046         object_child_foreach_recursive(object_get_root(),
1047                                        do_arm_linux_init, info);
1048     }
1049     info->is_linux = is_linux;
1050 
1051     for (cs = CPU(cpu); cs; cs = CPU_NEXT(cs)) {
1052         ARM_CPU(cs)->env.boot_info = info;
1053     }
1054 }
1055 
1056 void arm_load_kernel(ARMCPU *cpu, struct arm_boot_info *info)
1057 {
1058     CPUState *cs;
1059 
1060     info->load_kernel_notifier.cpu = cpu;
1061     info->load_kernel_notifier.notifier.notify = arm_load_kernel_notify;
1062     qemu_add_machine_init_done_notifier(&info->load_kernel_notifier.notifier);
1063 
1064     /* CPU objects (unlike devices) are not automatically reset on system
1065      * reset, so we must always register a handler to do so. If we're
1066      * actually loading a kernel, the handler is also responsible for
1067      * arranging that we start it correctly.
1068      */
1069     for (cs = CPU(cpu); cs; cs = CPU_NEXT(cs)) {
1070         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
1071     }
1072 }
1073 
1074 static const TypeInfo arm_linux_boot_if_info = {
1075     .name = TYPE_ARM_LINUX_BOOT_IF,
1076     .parent = TYPE_INTERFACE,
1077     .class_size = sizeof(ARMLinuxBootIfClass),
1078 };
1079 
1080 static void arm_linux_boot_register_types(void)
1081 {
1082     type_register_static(&arm_linux_boot_if_info);
1083 }
1084 
1085 type_init(arm_linux_boot_register_types)
1086