xref: /openbmc/qemu/hw/arm/boot.c (revision f31cd9e4e2172a4807f390194978c61e717791d2)
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 "exec/address-spaces.h"
26 
27 /* Kernel boot protocol is specified in the kernel docs
28  * Documentation/arm/Booting and Documentation/arm64/booting.txt
29  * They have different preferred image load offsets from system RAM base.
30  */
31 #define KERNEL_ARGS_ADDR 0x100
32 #define KERNEL_LOAD_ADDR 0x00010000
33 #define KERNEL64_LOAD_ADDR 0x00080000
34 
35 #define ARM64_TEXT_OFFSET_OFFSET    8
36 #define ARM64_MAGIC_OFFSET          56
37 
38 typedef enum {
39     FIXUP_NONE = 0,     /* do nothing */
40     FIXUP_TERMINATOR,   /* end of insns */
41     FIXUP_BOARDID,      /* overwrite with board ID number */
42     FIXUP_BOARD_SETUP,  /* overwrite with board specific setup code address */
43     FIXUP_ARGPTR,       /* overwrite with pointer to kernel args */
44     FIXUP_ENTRYPOINT,   /* overwrite with kernel entry point */
45     FIXUP_GIC_CPU_IF,   /* overwrite with GIC CPU interface address */
46     FIXUP_BOOTREG,      /* overwrite with boot register address */
47     FIXUP_DSB,          /* overwrite with correct DSB insn for cpu */
48     FIXUP_MAX,
49 } FixupType;
50 
51 typedef struct ARMInsnFixup {
52     uint32_t insn;
53     FixupType fixup;
54 } ARMInsnFixup;
55 
56 static const ARMInsnFixup bootloader_aarch64[] = {
57     { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */
58     { 0xaa1f03e1 }, /* mov x1, xzr */
59     { 0xaa1f03e2 }, /* mov x2, xzr */
60     { 0xaa1f03e3 }, /* mov x3, xzr */
61     { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */
62     { 0xd61f0080 }, /* br x4      ; Jump to the kernel entry point */
63     { 0, FIXUP_ARGPTR }, /* arg: .word @DTB Lower 32-bits */
64     { 0 }, /* .word @DTB Higher 32-bits */
65     { 0, FIXUP_ENTRYPOINT }, /* entry: .word @Kernel Entry Lower 32-bits */
66     { 0 }, /* .word @Kernel Entry Higher 32-bits */
67     { 0, FIXUP_TERMINATOR }
68 };
69 
70 /* A very small bootloader: call the board-setup code (if needed),
71  * set r0-r2, then jump to the kernel.
72  * If we're not calling boot setup code then we don't copy across
73  * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array.
74  */
75 
76 static const ARMInsnFixup bootloader[] = {
77     { 0xe28fe004 }, /* add     lr, pc, #4 */
78     { 0xe51ff004 }, /* ldr     pc, [pc, #-4] */
79     { 0, FIXUP_BOARD_SETUP },
80 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3
81     { 0xe3a00000 }, /* mov     r0, #0 */
82     { 0xe59f1004 }, /* ldr     r1, [pc, #4] */
83     { 0xe59f2004 }, /* ldr     r2, [pc, #4] */
84     { 0xe59ff004 }, /* ldr     pc, [pc, #4] */
85     { 0, FIXUP_BOARDID },
86     { 0, FIXUP_ARGPTR },
87     { 0, FIXUP_ENTRYPOINT },
88     { 0, FIXUP_TERMINATOR }
89 };
90 
91 /* Handling for secondary CPU boot in a multicore system.
92  * Unlike the uniprocessor/primary CPU boot, this is platform
93  * dependent. The default code here is based on the secondary
94  * CPU boot protocol used on realview/vexpress boards, with
95  * some parameterisation to increase its flexibility.
96  * QEMU platform models for which this code is not appropriate
97  * should override write_secondary_boot and secondary_cpu_reset_hook
98  * instead.
99  *
100  * This code enables the interrupt controllers for the secondary
101  * CPUs and then puts all the secondary CPUs into a loop waiting
102  * for an interprocessor interrupt and polling a configurable
103  * location for the kernel secondary CPU entry point.
104  */
105 #define DSB_INSN 0xf57ff04f
106 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
107 
108 static const ARMInsnFixup smpboot[] = {
109     { 0xe59f2028 }, /* ldr r2, gic_cpu_if */
110     { 0xe59f0028 }, /* ldr r0, bootreg_addr */
111     { 0xe3a01001 }, /* mov r1, #1 */
112     { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */
113     { 0xe3a010ff }, /* mov r1, #0xff */
114     { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */
115     { 0, FIXUP_DSB },   /* dsb */
116     { 0xe320f003 }, /* wfi */
117     { 0xe5901000 }, /* ldr     r1, [r0] */
118     { 0xe1110001 }, /* tst     r1, r1 */
119     { 0x0afffffb }, /* beq     <wfi> */
120     { 0xe12fff11 }, /* bx      r1 */
121     { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */
122     { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */
123     { 0, FIXUP_TERMINATOR }
124 };
125 
126 static void write_bootloader(const char *name, hwaddr addr,
127                              const ARMInsnFixup *insns, uint32_t *fixupcontext)
128 {
129     /* Fix up the specified bootloader fragment and write it into
130      * guest memory using rom_add_blob_fixed(). fixupcontext is
131      * an array giving the values to write in for the fixup types
132      * which write a value into the code array.
133      */
134     int i, len;
135     uint32_t *code;
136 
137     len = 0;
138     while (insns[len].fixup != FIXUP_TERMINATOR) {
139         len++;
140     }
141 
142     code = g_new0(uint32_t, len);
143 
144     for (i = 0; i < len; i++) {
145         uint32_t insn = insns[i].insn;
146         FixupType fixup = insns[i].fixup;
147 
148         switch (fixup) {
149         case FIXUP_NONE:
150             break;
151         case FIXUP_BOARDID:
152         case FIXUP_BOARD_SETUP:
153         case FIXUP_ARGPTR:
154         case FIXUP_ENTRYPOINT:
155         case FIXUP_GIC_CPU_IF:
156         case FIXUP_BOOTREG:
157         case FIXUP_DSB:
158             insn = fixupcontext[fixup];
159             break;
160         default:
161             abort();
162         }
163         code[i] = tswap32(insn);
164     }
165 
166     rom_add_blob_fixed(name, code, len * sizeof(uint32_t), addr);
167 
168     g_free(code);
169 }
170 
171 static void default_write_secondary(ARMCPU *cpu,
172                                     const struct arm_boot_info *info)
173 {
174     uint32_t fixupcontext[FIXUP_MAX];
175 
176     fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr;
177     fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr;
178     if (arm_feature(&cpu->env, ARM_FEATURE_V7)) {
179         fixupcontext[FIXUP_DSB] = DSB_INSN;
180     } else {
181         fixupcontext[FIXUP_DSB] = CP15_DSB_INSN;
182     }
183 
184     write_bootloader("smpboot", info->smp_loader_start,
185                      smpboot, fixupcontext);
186 }
187 
188 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
189                                             const struct arm_boot_info *info,
190                                             hwaddr mvbar_addr)
191 {
192     int n;
193     uint32_t mvbar_blob[] = {
194         /* mvbar_addr: secure monitor vectors
195          * Default unimplemented and unused vectors to spin. Makes it
196          * easier to debug (as opposed to the CPU running away).
197          */
198         0xeafffffe, /* (spin) */
199         0xeafffffe, /* (spin) */
200         0xe1b0f00e, /* movs pc, lr ;SMC exception return */
201         0xeafffffe, /* (spin) */
202         0xeafffffe, /* (spin) */
203         0xeafffffe, /* (spin) */
204         0xeafffffe, /* (spin) */
205         0xeafffffe, /* (spin) */
206     };
207     uint32_t board_setup_blob[] = {
208         /* board setup addr */
209         0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */
210         0xee0c0f30, /* mcr     p15, 0, r0, c12, c0, 1 ;set MVBAR */
211         0xee110f11, /* mrc     p15, 0, r0, c1 , c1, 0 ;read SCR */
212         0xe3800031, /* orr     r0, #0x31              ;enable AW, FW, NS */
213         0xee010f11, /* mcr     p15, 0, r0, c1, c1, 0  ;write SCR */
214         0xe1a0100e, /* mov     r1, lr                 ;save LR across SMC */
215         0xe1600070, /* smc     #0                     ;call monitor to flush SCR */
216         0xe1a0f001, /* mov     pc, r1                 ;return */
217     };
218 
219     /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
220     assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
221 
222     /* check that these blobs don't overlap */
223     assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr)
224           || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr));
225 
226     for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
227         mvbar_blob[n] = tswap32(mvbar_blob[n]);
228     }
229     rom_add_blob_fixed("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
230                        mvbar_addr);
231 
232     for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
233         board_setup_blob[n] = tswap32(board_setup_blob[n]);
234     }
235     rom_add_blob_fixed("board-setup", board_setup_blob,
236                        sizeof(board_setup_blob), info->board_setup_addr);
237 }
238 
239 static void default_reset_secondary(ARMCPU *cpu,
240                                     const struct arm_boot_info *info)
241 {
242     CPUState *cs = CPU(cpu);
243 
244     address_space_stl_notdirty(&address_space_memory, info->smp_bootreg_addr,
245                                0, MEMTXATTRS_UNSPECIFIED, NULL);
246     cpu_set_pc(cs, info->smp_loader_start);
247 }
248 
249 static inline bool have_dtb(const struct arm_boot_info *info)
250 {
251     return info->dtb_filename || info->get_dtb;
252 }
253 
254 #define WRITE_WORD(p, value) do { \
255     address_space_stl_notdirty(&address_space_memory, p, value, \
256                                MEMTXATTRS_UNSPECIFIED, NULL);  \
257     p += 4;                       \
258 } while (0)
259 
260 static void set_kernel_args(const struct arm_boot_info *info)
261 {
262     int initrd_size = info->initrd_size;
263     hwaddr base = info->loader_start;
264     hwaddr p;
265 
266     p = base + KERNEL_ARGS_ADDR;
267     /* ATAG_CORE */
268     WRITE_WORD(p, 5);
269     WRITE_WORD(p, 0x54410001);
270     WRITE_WORD(p, 1);
271     WRITE_WORD(p, 0x1000);
272     WRITE_WORD(p, 0);
273     /* ATAG_MEM */
274     /* TODO: handle multiple chips on one ATAG list */
275     WRITE_WORD(p, 4);
276     WRITE_WORD(p, 0x54410002);
277     WRITE_WORD(p, info->ram_size);
278     WRITE_WORD(p, info->loader_start);
279     if (initrd_size) {
280         /* ATAG_INITRD2 */
281         WRITE_WORD(p, 4);
282         WRITE_WORD(p, 0x54420005);
283         WRITE_WORD(p, info->initrd_start);
284         WRITE_WORD(p, initrd_size);
285     }
286     if (info->kernel_cmdline && *info->kernel_cmdline) {
287         /* ATAG_CMDLINE */
288         int cmdline_size;
289 
290         cmdline_size = strlen(info->kernel_cmdline);
291         cpu_physical_memory_write(p + 8, info->kernel_cmdline,
292                                   cmdline_size + 1);
293         cmdline_size = (cmdline_size >> 2) + 1;
294         WRITE_WORD(p, cmdline_size + 2);
295         WRITE_WORD(p, 0x54410009);
296         p += cmdline_size * 4;
297     }
298     if (info->atag_board) {
299         /* ATAG_BOARD */
300         int atag_board_len;
301         uint8_t atag_board_buf[0x1000];
302 
303         atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
304         WRITE_WORD(p, (atag_board_len + 8) >> 2);
305         WRITE_WORD(p, 0x414f4d50);
306         cpu_physical_memory_write(p, atag_board_buf, atag_board_len);
307         p += atag_board_len;
308     }
309     /* ATAG_END */
310     WRITE_WORD(p, 0);
311     WRITE_WORD(p, 0);
312 }
313 
314 static void set_kernel_args_old(const struct arm_boot_info *info)
315 {
316     hwaddr p;
317     const char *s;
318     int initrd_size = info->initrd_size;
319     hwaddr base = info->loader_start;
320 
321     /* see linux/include/asm-arm/setup.h */
322     p = base + KERNEL_ARGS_ADDR;
323     /* page_size */
324     WRITE_WORD(p, 4096);
325     /* nr_pages */
326     WRITE_WORD(p, info->ram_size / 4096);
327     /* ramdisk_size */
328     WRITE_WORD(p, 0);
329 #define FLAG_READONLY	1
330 #define FLAG_RDLOAD	4
331 #define FLAG_RDPROMPT	8
332     /* flags */
333     WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
334     /* rootdev */
335     WRITE_WORD(p, (31 << 8) | 0);	/* /dev/mtdblock0 */
336     /* video_num_cols */
337     WRITE_WORD(p, 0);
338     /* video_num_rows */
339     WRITE_WORD(p, 0);
340     /* video_x */
341     WRITE_WORD(p, 0);
342     /* video_y */
343     WRITE_WORD(p, 0);
344     /* memc_control_reg */
345     WRITE_WORD(p, 0);
346     /* unsigned char sounddefault */
347     /* unsigned char adfsdrives */
348     /* unsigned char bytes_per_char_h */
349     /* unsigned char bytes_per_char_v */
350     WRITE_WORD(p, 0);
351     /* pages_in_bank[4] */
352     WRITE_WORD(p, 0);
353     WRITE_WORD(p, 0);
354     WRITE_WORD(p, 0);
355     WRITE_WORD(p, 0);
356     /* pages_in_vram */
357     WRITE_WORD(p, 0);
358     /* initrd_start */
359     if (initrd_size) {
360         WRITE_WORD(p, info->initrd_start);
361     } else {
362         WRITE_WORD(p, 0);
363     }
364     /* initrd_size */
365     WRITE_WORD(p, initrd_size);
366     /* rd_start */
367     WRITE_WORD(p, 0);
368     /* system_rev */
369     WRITE_WORD(p, 0);
370     /* system_serial_low */
371     WRITE_WORD(p, 0);
372     /* system_serial_high */
373     WRITE_WORD(p, 0);
374     /* mem_fclk_21285 */
375     WRITE_WORD(p, 0);
376     /* zero unused fields */
377     while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
378         WRITE_WORD(p, 0);
379     }
380     s = info->kernel_cmdline;
381     if (s) {
382         cpu_physical_memory_write(p, s, strlen(s) + 1);
383     } else {
384         WRITE_WORD(p, 0);
385     }
386 }
387 
388 static void fdt_add_psci_node(void *fdt)
389 {
390     uint32_t cpu_suspend_fn;
391     uint32_t cpu_off_fn;
392     uint32_t cpu_on_fn;
393     uint32_t migrate_fn;
394     ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
395     const char *psci_method;
396     int64_t psci_conduit;
397 
398     psci_conduit = object_property_get_int(OBJECT(armcpu),
399                                            "psci-conduit",
400                                            &error_abort);
401     switch (psci_conduit) {
402     case QEMU_PSCI_CONDUIT_DISABLED:
403         return;
404     case QEMU_PSCI_CONDUIT_HVC:
405         psci_method = "hvc";
406         break;
407     case QEMU_PSCI_CONDUIT_SMC:
408         psci_method = "smc";
409         break;
410     default:
411         g_assert_not_reached();
412     }
413 
414     qemu_fdt_add_subnode(fdt, "/psci");
415     if (armcpu->psci_version == 2) {
416         const char comp[] = "arm,psci-0.2\0arm,psci";
417         qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
418 
419         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
420         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
421             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
422             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
423             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
424         } else {
425             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
426             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
427             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
428         }
429     } else {
430         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
431 
432         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
433         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
434         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
435         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
436     }
437 
438     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
439      * to the instruction that should be used to invoke PSCI functions.
440      * However, the device tree binding uses 'method' instead, so that is
441      * what we should use here.
442      */
443     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
444 
445     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
446     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
447     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
448     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
449 }
450 
451 /**
452  * load_dtb() - load a device tree binary image into memory
453  * @addr:       the address to load the image at
454  * @binfo:      struct describing the boot environment
455  * @addr_limit: upper limit of the available memory area at @addr
456  *
457  * Load a device tree supplied by the machine or by the user  with the
458  * '-dtb' command line option, and put it at offset @addr in target
459  * memory.
460  *
461  * If @addr_limit contains a meaningful value (i.e., it is strictly greater
462  * than @addr), the device tree is only loaded if its size does not exceed
463  * the limit.
464  *
465  * Returns: the size of the device tree image on success,
466  *          0 if the image size exceeds the limit,
467  *          -1 on errors.
468  *
469  * Note: Must not be called unless have_dtb(binfo) is true.
470  */
471 static int load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
472                     hwaddr addr_limit)
473 {
474     void *fdt = NULL;
475     int size, rc;
476     uint32_t acells, scells;
477     char *nodename;
478     unsigned int i;
479     hwaddr mem_base, mem_len;
480 
481     if (binfo->dtb_filename) {
482         char *filename;
483         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
484         if (!filename) {
485             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
486             goto fail;
487         }
488 
489         fdt = load_device_tree(filename, &size);
490         if (!fdt) {
491             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
492             g_free(filename);
493             goto fail;
494         }
495         g_free(filename);
496     } else {
497         fdt = binfo->get_dtb(binfo, &size);
498         if (!fdt) {
499             fprintf(stderr, "Board was unable to create a dtb blob\n");
500             goto fail;
501         }
502     }
503 
504     if (addr_limit > addr && size > (addr_limit - addr)) {
505         /* Installing the device tree blob at addr would exceed addr_limit.
506          * Whether this constitutes failure is up to the caller to decide,
507          * so just return 0 as size, i.e., no error.
508          */
509         g_free(fdt);
510         return 0;
511     }
512 
513     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
514                                    NULL, &error_fatal);
515     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
516                                    NULL, &error_fatal);
517     if (acells == 0 || scells == 0) {
518         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
519         goto fail;
520     }
521 
522     if (scells < 2 && binfo->ram_size >= (1ULL << 32)) {
523         /* This is user error so deserves a friendlier error message
524          * than the failure of setprop_sized_cells would provide
525          */
526         fprintf(stderr, "qemu: dtb file not compatible with "
527                 "RAM size > 4GB\n");
528         goto fail;
529     }
530 
531     if (nb_numa_nodes > 0) {
532         /*
533          * Turn the /memory node created before into a NOP node, then create
534          * /memory@addr nodes for all numa nodes respectively.
535          */
536         qemu_fdt_nop_node(fdt, "/memory");
537         mem_base = binfo->loader_start;
538         for (i = 0; i < nb_numa_nodes; i++) {
539             mem_len = numa_info[i].node_mem;
540             nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
541             qemu_fdt_add_subnode(fdt, nodename);
542             qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
543             rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
544                                               acells, mem_base,
545                                               scells, mem_len);
546             if (rc < 0) {
547                 fprintf(stderr, "couldn't set %s/reg for node %d\n", nodename,
548                         i);
549                 goto fail;
550             }
551 
552             qemu_fdt_setprop_cell(fdt, nodename, "numa-node-id", i);
553             mem_base += mem_len;
554             g_free(nodename);
555         }
556     } else {
557         Error *err = NULL;
558 
559         rc = fdt_path_offset(fdt, "/memory");
560         if (rc < 0) {
561             qemu_fdt_add_subnode(fdt, "/memory");
562         }
563 
564         if (!qemu_fdt_getprop(fdt, "/memory", "device_type", NULL, &err)) {
565             qemu_fdt_setprop_string(fdt, "/memory", "device_type", "memory");
566         }
567 
568         rc = qemu_fdt_setprop_sized_cells(fdt, "/memory", "reg",
569                                           acells, binfo->loader_start,
570                                           scells, binfo->ram_size);
571         if (rc < 0) {
572             fprintf(stderr, "couldn't set /memory/reg\n");
573             goto fail;
574         }
575     }
576 
577     rc = fdt_path_offset(fdt, "/chosen");
578     if (rc < 0) {
579         qemu_fdt_add_subnode(fdt, "/chosen");
580     }
581 
582     if (binfo->kernel_cmdline && *binfo->kernel_cmdline) {
583         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
584                                      binfo->kernel_cmdline);
585         if (rc < 0) {
586             fprintf(stderr, "couldn't set /chosen/bootargs\n");
587             goto fail;
588         }
589     }
590 
591     if (binfo->initrd_size) {
592         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
593                                    binfo->initrd_start);
594         if (rc < 0) {
595             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
596             goto fail;
597         }
598 
599         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
600                                    binfo->initrd_start + binfo->initrd_size);
601         if (rc < 0) {
602             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
603             goto fail;
604         }
605     }
606 
607     fdt_add_psci_node(fdt);
608 
609     if (binfo->modify_dtb) {
610         binfo->modify_dtb(binfo, fdt);
611     }
612 
613     qemu_fdt_dumpdtb(fdt, size);
614 
615     /* Put the DTB into the memory map as a ROM image: this will ensure
616      * the DTB is copied again upon reset, even if addr points into RAM.
617      */
618     rom_add_blob_fixed("dtb", fdt, size, addr);
619 
620     g_free(fdt);
621 
622     return size;
623 
624 fail:
625     g_free(fdt);
626     return -1;
627 }
628 
629 static void do_cpu_reset(void *opaque)
630 {
631     ARMCPU *cpu = opaque;
632     CPUState *cs = CPU(cpu);
633     CPUARMState *env = &cpu->env;
634     const struct arm_boot_info *info = env->boot_info;
635 
636     cpu_reset(cs);
637     if (info) {
638         if (!info->is_linux) {
639             int i;
640             /* Jump to the entry point.  */
641             uint64_t entry = info->entry;
642 
643             switch (info->endianness) {
644             case ARM_ENDIANNESS_LE:
645                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
646                 for (i = 1; i < 4; ++i) {
647                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
648                 }
649                 env->uncached_cpsr &= ~CPSR_E;
650                 break;
651             case ARM_ENDIANNESS_BE8:
652                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
653                 for (i = 1; i < 4; ++i) {
654                     env->cp15.sctlr_el[i] |= SCTLR_EE;
655                 }
656                 env->uncached_cpsr |= CPSR_E;
657                 break;
658             case ARM_ENDIANNESS_BE32:
659                 env->cp15.sctlr_el[1] |= SCTLR_B;
660                 break;
661             case ARM_ENDIANNESS_UNKNOWN:
662                 break; /* Board's decision */
663             default:
664                 g_assert_not_reached();
665             }
666 
667             if (!env->aarch64) {
668                 env->thumb = info->entry & 1;
669                 entry &= 0xfffffffe;
670             }
671             cpu_set_pc(cs, entry);
672         } else {
673             /* If we are booting Linux then we need to check whether we are
674              * booting into secure or non-secure state and adjust the state
675              * accordingly.  Out of reset, ARM is defined to be in secure state
676              * (SCR.NS = 0), we change that here if non-secure boot has been
677              * requested.
678              */
679             if (arm_feature(env, ARM_FEATURE_EL3)) {
680                 /* AArch64 is defined to come out of reset into EL3 if enabled.
681                  * If we are booting Linux then we need to adjust our EL as
682                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
683                  * SVC, which Linux expects, so no privilege/exception level to
684                  * adjust.
685                  */
686                 if (env->aarch64) {
687                     env->cp15.scr_el3 |= SCR_RW;
688                     if (arm_feature(env, ARM_FEATURE_EL2)) {
689                         env->cp15.hcr_el2 |= HCR_RW;
690                         env->pstate = PSTATE_MODE_EL2h;
691                     } else {
692                         env->pstate = PSTATE_MODE_EL1h;
693                     }
694                 }
695 
696                 /* Set to non-secure if not a secure boot */
697                 if (!info->secure_boot &&
698                     (cs != first_cpu || !info->secure_board_setup)) {
699                     /* Linux expects non-secure state */
700                     env->cp15.scr_el3 |= SCR_NS;
701                 }
702             }
703 
704             if (cs == first_cpu) {
705                 cpu_set_pc(cs, info->loader_start);
706 
707                 if (!have_dtb(info)) {
708                     if (old_param) {
709                         set_kernel_args_old(info);
710                     } else {
711                         set_kernel_args(info);
712                     }
713                 }
714             } else {
715                 info->secondary_cpu_reset_hook(cpu, info);
716             }
717         }
718     }
719 }
720 
721 /**
722  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
723  *                          by key.
724  * @fw_cfg:         The firmware config instance to store the data in.
725  * @size_key:       The firmware config key to store the size of the loaded
726  *                  data under, with fw_cfg_add_i32().
727  * @data_key:       The firmware config key to store the loaded data under,
728  *                  with fw_cfg_add_bytes().
729  * @image_name:     The name of the image file to load. If it is NULL, the
730  *                  function returns without doing anything.
731  * @try_decompress: Whether the image should be decompressed (gunzipped) before
732  *                  adding it to fw_cfg. If decompression fails, the image is
733  *                  loaded as-is.
734  *
735  * In case of failure, the function prints an error message to stderr and the
736  * process exits with status 1.
737  */
738 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
739                                  uint16_t data_key, const char *image_name,
740                                  bool try_decompress)
741 {
742     size_t size = -1;
743     uint8_t *data;
744 
745     if (image_name == NULL) {
746         return;
747     }
748 
749     if (try_decompress) {
750         size = load_image_gzipped_buffer(image_name,
751                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
752     }
753 
754     if (size == (size_t)-1) {
755         gchar *contents;
756         gsize length;
757 
758         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
759             error_report("failed to load \"%s\"", image_name);
760             exit(1);
761         }
762         size = length;
763         data = (uint8_t *)contents;
764     }
765 
766     fw_cfg_add_i32(fw_cfg, size_key, size);
767     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
768 }
769 
770 static int do_arm_linux_init(Object *obj, void *opaque)
771 {
772     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
773         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
774         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
775         struct arm_boot_info *info = opaque;
776 
777         if (albifc->arm_linux_init) {
778             albifc->arm_linux_init(albif, info->secure_boot);
779         }
780     }
781     return 0;
782 }
783 
784 static uint64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
785                              uint64_t *lowaddr, uint64_t *highaddr,
786                              int elf_machine)
787 {
788     bool elf_is64;
789     union {
790         Elf32_Ehdr h32;
791         Elf64_Ehdr h64;
792     } elf_header;
793     int data_swab = 0;
794     bool big_endian;
795     uint64_t ret = -1;
796     Error *err = NULL;
797 
798 
799     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
800     if (err) {
801         return ret;
802     }
803 
804     if (elf_is64) {
805         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
806         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
807                                       : ARM_ENDIANNESS_LE;
808     } else {
809         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
810         if (big_endian) {
811             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
812                 info->endianness = ARM_ENDIANNESS_BE8;
813             } else {
814                 info->endianness = ARM_ENDIANNESS_BE32;
815                 /* In BE32, the CPU has a different view of the per-byte
816                  * address map than the rest of the system. BE32 ELF files
817                  * are organised such that they can be programmed through
818                  * the CPU's per-word byte-reversed view of the world. QEMU
819                  * however loads ELF files independently of the CPU. So
820                  * tell the ELF loader to byte reverse the data for us.
821                  */
822                 data_swab = 2;
823             }
824         } else {
825             info->endianness = ARM_ENDIANNESS_LE;
826         }
827     }
828 
829     ret = load_elf(info->kernel_filename, NULL, NULL,
830                    pentry, lowaddr, highaddr, big_endian, elf_machine,
831                    1, data_swab);
832     if (ret <= 0) {
833         /* The header loaded but the image didn't */
834         exit(1);
835     }
836 
837     return ret;
838 }
839 
840 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
841                                    hwaddr *entry)
842 {
843     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
844     uint8_t *buffer;
845     int size;
846 
847     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
848     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
849                                      &buffer);
850 
851     if (size < 0) {
852         gsize len;
853 
854         /* Load as raw file otherwise */
855         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
856             return -1;
857         }
858         size = len;
859     }
860 
861     /* check the arm64 magic header value -- very old kernels may not have it */
862     if (memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
863         uint64_t hdrvals[2];
864 
865         /* The arm64 Image header has text_offset and image_size fields at 8 and
866          * 16 bytes into the Image header, respectively. The text_offset field
867          * is only valid if the image_size is non-zero.
868          */
869         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
870         if (hdrvals[1] != 0) {
871             kernel_load_offset = le64_to_cpu(hdrvals[0]);
872         }
873     }
874 
875     *entry = mem_base + kernel_load_offset;
876     rom_add_blob_fixed(filename, buffer, size, *entry);
877 
878     g_free(buffer);
879 
880     return size;
881 }
882 
883 static void arm_load_kernel_notify(Notifier *notifier, void *data)
884 {
885     CPUState *cs;
886     int kernel_size;
887     int initrd_size;
888     int is_linux = 0;
889     uint64_t elf_entry, elf_low_addr, elf_high_addr;
890     int elf_machine;
891     hwaddr entry;
892     static const ARMInsnFixup *primary_loader;
893     ArmLoadKernelNotifier *n = DO_UPCAST(ArmLoadKernelNotifier,
894                                          notifier, notifier);
895     ARMCPU *cpu = n->cpu;
896     struct arm_boot_info *info =
897         container_of(n, struct arm_boot_info, load_kernel_notifier);
898 
899     /* The board code is not supposed to set secure_board_setup unless
900      * running its code in secure mode is actually possible, and KVM
901      * doesn't support secure.
902      */
903     assert(!(info->secure_board_setup && kvm_enabled()));
904 
905     info->dtb_filename = qemu_opt_get(qemu_get_machine_opts(), "dtb");
906 
907     /* Load the kernel.  */
908     if (!info->kernel_filename || info->firmware_loaded) {
909 
910         if (have_dtb(info)) {
911             /* If we have a device tree blob, but no kernel to supply it to (or
912              * the kernel is supposed to be loaded by the bootloader), copy the
913              * DTB to the base of RAM for the bootloader to pick up.
914              */
915             if (load_dtb(info->loader_start, info, 0) < 0) {
916                 exit(1);
917             }
918         }
919 
920         if (info->kernel_filename) {
921             FWCfgState *fw_cfg;
922             bool try_decompressing_kernel;
923 
924             fw_cfg = fw_cfg_find();
925             try_decompressing_kernel = arm_feature(&cpu->env,
926                                                    ARM_FEATURE_AARCH64);
927 
928             /* Expose the kernel, the command line, and the initrd in fw_cfg.
929              * We don't process them here at all, it's all left to the
930              * firmware.
931              */
932             load_image_to_fw_cfg(fw_cfg,
933                                  FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
934                                  info->kernel_filename,
935                                  try_decompressing_kernel);
936             load_image_to_fw_cfg(fw_cfg,
937                                  FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
938                                  info->initrd_filename, false);
939 
940             if (info->kernel_cmdline) {
941                 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
942                                strlen(info->kernel_cmdline) + 1);
943                 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
944                                   info->kernel_cmdline);
945             }
946         }
947 
948         /* We will start from address 0 (typically a boot ROM image) in the
949          * same way as hardware.
950          */
951         return;
952     }
953 
954     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
955         primary_loader = bootloader_aarch64;
956         elf_machine = EM_AARCH64;
957     } else {
958         primary_loader = bootloader;
959         if (!info->write_board_setup) {
960             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
961         }
962         elf_machine = EM_ARM;
963     }
964 
965     if (!info->secondary_cpu_reset_hook) {
966         info->secondary_cpu_reset_hook = default_reset_secondary;
967     }
968     if (!info->write_secondary_boot) {
969         info->write_secondary_boot = default_write_secondary;
970     }
971 
972     if (info->nb_cpus == 0)
973         info->nb_cpus = 1;
974 
975     /* We want to put the initrd far enough into RAM that when the
976      * kernel is uncompressed it will not clobber the initrd. However
977      * on boards without much RAM we must ensure that we still leave
978      * enough room for a decent sized initrd, and on boards with large
979      * amounts of RAM we must avoid the initrd being so far up in RAM
980      * that it is outside lowmem and inaccessible to the kernel.
981      * So for boards with less  than 256MB of RAM we put the initrd
982      * halfway into RAM, and for boards with 256MB of RAM or more we put
983      * the initrd at 128MB.
984      */
985     info->initrd_start = info->loader_start +
986         MIN(info->ram_size / 2, 128 * 1024 * 1024);
987 
988     /* Assume that raw images are linux kernels, and ELF images are not.  */
989     kernel_size = arm_load_elf(info, &elf_entry, &elf_low_addr,
990                                &elf_high_addr, elf_machine);
991     if (kernel_size > 0 && have_dtb(info)) {
992         /* If there is still some room left at the base of RAM, try and put
993          * the DTB there like we do for images loaded with -bios or -pflash.
994          */
995         if (elf_low_addr > info->loader_start
996             || elf_high_addr < info->loader_start) {
997             /* Pass elf_low_addr as address limit to load_dtb if it may be
998              * pointing into RAM, otherwise pass '0' (no limit)
999              */
1000             if (elf_low_addr < info->loader_start) {
1001                 elf_low_addr = 0;
1002             }
1003             if (load_dtb(info->loader_start, info, elf_low_addr) < 0) {
1004                 exit(1);
1005             }
1006         }
1007     }
1008     entry = elf_entry;
1009     if (kernel_size < 0) {
1010         kernel_size = load_uimage(info->kernel_filename, &entry, NULL,
1011                                   &is_linux, NULL, NULL);
1012     }
1013     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1014         kernel_size = load_aarch64_image(info->kernel_filename,
1015                                          info->loader_start, &entry);
1016         is_linux = 1;
1017     } else if (kernel_size < 0) {
1018         /* 32-bit ARM */
1019         entry = info->loader_start + KERNEL_LOAD_ADDR;
1020         kernel_size = load_image_targphys(info->kernel_filename, entry,
1021                                           info->ram_size - KERNEL_LOAD_ADDR);
1022         is_linux = 1;
1023     }
1024     if (kernel_size < 0) {
1025         error_report("could not load kernel '%s'", info->kernel_filename);
1026         exit(1);
1027     }
1028     info->entry = entry;
1029     if (is_linux) {
1030         uint32_t fixupcontext[FIXUP_MAX];
1031 
1032         if (info->initrd_filename) {
1033             initrd_size = load_ramdisk(info->initrd_filename,
1034                                        info->initrd_start,
1035                                        info->ram_size -
1036                                        info->initrd_start);
1037             if (initrd_size < 0) {
1038                 initrd_size = load_image_targphys(info->initrd_filename,
1039                                                   info->initrd_start,
1040                                                   info->ram_size -
1041                                                   info->initrd_start);
1042             }
1043             if (initrd_size < 0) {
1044                 error_report("could not load initrd '%s'",
1045                              info->initrd_filename);
1046                 exit(1);
1047             }
1048         } else {
1049             initrd_size = 0;
1050         }
1051         info->initrd_size = initrd_size;
1052 
1053         fixupcontext[FIXUP_BOARDID] = info->board_id;
1054         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1055 
1056         /* for device tree boot, we pass the DTB directly in r2. Otherwise
1057          * we point to the kernel args.
1058          */
1059         if (have_dtb(info)) {
1060             hwaddr align;
1061             hwaddr dtb_start;
1062 
1063             if (elf_machine == EM_AARCH64) {
1064                 /*
1065                  * Some AArch64 kernels on early bootup map the fdt region as
1066                  *
1067                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1068                  *
1069                  * Let's play safe and prealign it to 2MB to give us some space.
1070                  */
1071                 align = 2 * 1024 * 1024;
1072             } else {
1073                 /*
1074                  * Some 32bit kernels will trash anything in the 4K page the
1075                  * initrd ends in, so make sure the DTB isn't caught up in that.
1076                  */
1077                 align = 4096;
1078             }
1079 
1080             /* Place the DTB after the initrd in memory with alignment. */
1081             dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size, align);
1082             if (load_dtb(dtb_start, info, 0) < 0) {
1083                 exit(1);
1084             }
1085             fixupcontext[FIXUP_ARGPTR] = dtb_start;
1086         } else {
1087             fixupcontext[FIXUP_ARGPTR] = info->loader_start + KERNEL_ARGS_ADDR;
1088             if (info->ram_size >= (1ULL << 32)) {
1089                 error_report("RAM size must be less than 4GB to boot"
1090                              " Linux kernel using ATAGS (try passing a device tree"
1091                              " using -dtb)");
1092                 exit(1);
1093             }
1094         }
1095         fixupcontext[FIXUP_ENTRYPOINT] = entry;
1096 
1097         write_bootloader("bootloader", info->loader_start,
1098                          primary_loader, fixupcontext);
1099 
1100         if (info->nb_cpus > 1) {
1101             info->write_secondary_boot(cpu, info);
1102         }
1103         if (info->write_board_setup) {
1104             info->write_board_setup(cpu, info);
1105         }
1106 
1107         /* Notify devices which need to fake up firmware initialization
1108          * that we're doing a direct kernel boot.
1109          */
1110         object_child_foreach_recursive(object_get_root(),
1111                                        do_arm_linux_init, info);
1112     }
1113     info->is_linux = is_linux;
1114 
1115     for (cs = CPU(cpu); cs; cs = CPU_NEXT(cs)) {
1116         ARM_CPU(cs)->env.boot_info = info;
1117     }
1118 }
1119 
1120 void arm_load_kernel(ARMCPU *cpu, struct arm_boot_info *info)
1121 {
1122     CPUState *cs;
1123 
1124     info->load_kernel_notifier.cpu = cpu;
1125     info->load_kernel_notifier.notifier.notify = arm_load_kernel_notify;
1126     qemu_add_machine_init_done_notifier(&info->load_kernel_notifier.notifier);
1127 
1128     /* CPU objects (unlike devices) are not automatically reset on system
1129      * reset, so we must always register a handler to do so. If we're
1130      * actually loading a kernel, the handler is also responsible for
1131      * arranging that we start it correctly.
1132      */
1133     for (cs = CPU(cpu); cs; cs = CPU_NEXT(cs)) {
1134         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
1135     }
1136 }
1137 
1138 static const TypeInfo arm_linux_boot_if_info = {
1139     .name = TYPE_ARM_LINUX_BOOT_IF,
1140     .parent = TYPE_INTERFACE,
1141     .class_size = sizeof(ARMLinuxBootIfClass),
1142 };
1143 
1144 static void arm_linux_boot_register_types(void)
1145 {
1146     type_register_static(&arm_linux_boot_if_info);
1147 }
1148 
1149 type_init(arm_linux_boot_register_types)
1150