xref: /openbmc/qemu/hw/arm/boot.c (revision 966f2ec3)
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 AddressSpace *arm_boot_address_space(ARMCPU *cpu,
40                                      const struct arm_boot_info *info)
41 {
42     /* Return the address space to use for bootloader reads and writes.
43      * We prefer the secure address space if the CPU has it and we're
44      * going to boot the guest into it.
45      */
46     int asidx;
47     CPUState *cs = CPU(cpu);
48 
49     if (arm_feature(&cpu->env, ARM_FEATURE_EL3) && info->secure_boot) {
50         asidx = ARMASIdx_S;
51     } else {
52         asidx = ARMASIdx_NS;
53     }
54 
55     return cpu_get_address_space(cs, asidx);
56 }
57 
58 typedef enum {
59     FIXUP_NONE = 0,     /* do nothing */
60     FIXUP_TERMINATOR,   /* end of insns */
61     FIXUP_BOARDID,      /* overwrite with board ID number */
62     FIXUP_BOARD_SETUP,  /* overwrite with board specific setup code address */
63     FIXUP_ARGPTR,       /* overwrite with pointer to kernel args */
64     FIXUP_ENTRYPOINT,   /* overwrite with kernel entry point */
65     FIXUP_GIC_CPU_IF,   /* overwrite with GIC CPU interface address */
66     FIXUP_BOOTREG,      /* overwrite with boot register address */
67     FIXUP_DSB,          /* overwrite with correct DSB insn for cpu */
68     FIXUP_MAX,
69 } FixupType;
70 
71 typedef struct ARMInsnFixup {
72     uint32_t insn;
73     FixupType fixup;
74 } ARMInsnFixup;
75 
76 static const ARMInsnFixup bootloader_aarch64[] = {
77     { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */
78     { 0xaa1f03e1 }, /* mov x1, xzr */
79     { 0xaa1f03e2 }, /* mov x2, xzr */
80     { 0xaa1f03e3 }, /* mov x3, xzr */
81     { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */
82     { 0xd61f0080 }, /* br x4      ; Jump to the kernel entry point */
83     { 0, FIXUP_ARGPTR }, /* arg: .word @DTB Lower 32-bits */
84     { 0 }, /* .word @DTB Higher 32-bits */
85     { 0, FIXUP_ENTRYPOINT }, /* entry: .word @Kernel Entry Lower 32-bits */
86     { 0 }, /* .word @Kernel Entry Higher 32-bits */
87     { 0, FIXUP_TERMINATOR }
88 };
89 
90 /* A very small bootloader: call the board-setup code (if needed),
91  * set r0-r2, then jump to the kernel.
92  * If we're not calling boot setup code then we don't copy across
93  * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array.
94  */
95 
96 static const ARMInsnFixup bootloader[] = {
97     { 0xe28fe004 }, /* add     lr, pc, #4 */
98     { 0xe51ff004 }, /* ldr     pc, [pc, #-4] */
99     { 0, FIXUP_BOARD_SETUP },
100 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3
101     { 0xe3a00000 }, /* mov     r0, #0 */
102     { 0xe59f1004 }, /* ldr     r1, [pc, #4] */
103     { 0xe59f2004 }, /* ldr     r2, [pc, #4] */
104     { 0xe59ff004 }, /* ldr     pc, [pc, #4] */
105     { 0, FIXUP_BOARDID },
106     { 0, FIXUP_ARGPTR },
107     { 0, FIXUP_ENTRYPOINT },
108     { 0, FIXUP_TERMINATOR }
109 };
110 
111 /* Handling for secondary CPU boot in a multicore system.
112  * Unlike the uniprocessor/primary CPU boot, this is platform
113  * dependent. The default code here is based on the secondary
114  * CPU boot protocol used on realview/vexpress boards, with
115  * some parameterisation to increase its flexibility.
116  * QEMU platform models for which this code is not appropriate
117  * should override write_secondary_boot and secondary_cpu_reset_hook
118  * instead.
119  *
120  * This code enables the interrupt controllers for the secondary
121  * CPUs and then puts all the secondary CPUs into a loop waiting
122  * for an interprocessor interrupt and polling a configurable
123  * location for the kernel secondary CPU entry point.
124  */
125 #define DSB_INSN 0xf57ff04f
126 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */
127 
128 static const ARMInsnFixup smpboot[] = {
129     { 0xe59f2028 }, /* ldr r2, gic_cpu_if */
130     { 0xe59f0028 }, /* ldr r0, bootreg_addr */
131     { 0xe3a01001 }, /* mov r1, #1 */
132     { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */
133     { 0xe3a010ff }, /* mov r1, #0xff */
134     { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */
135     { 0, FIXUP_DSB },   /* dsb */
136     { 0xe320f003 }, /* wfi */
137     { 0xe5901000 }, /* ldr     r1, [r0] */
138     { 0xe1110001 }, /* tst     r1, r1 */
139     { 0x0afffffb }, /* beq     <wfi> */
140     { 0xe12fff11 }, /* bx      r1 */
141     { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */
142     { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */
143     { 0, FIXUP_TERMINATOR }
144 };
145 
146 static void write_bootloader(const char *name, hwaddr addr,
147                              const ARMInsnFixup *insns, uint32_t *fixupcontext,
148                              AddressSpace *as)
149 {
150     /* Fix up the specified bootloader fragment and write it into
151      * guest memory using rom_add_blob_fixed(). fixupcontext is
152      * an array giving the values to write in for the fixup types
153      * which write a value into the code array.
154      */
155     int i, len;
156     uint32_t *code;
157 
158     len = 0;
159     while (insns[len].fixup != FIXUP_TERMINATOR) {
160         len++;
161     }
162 
163     code = g_new0(uint32_t, len);
164 
165     for (i = 0; i < len; i++) {
166         uint32_t insn = insns[i].insn;
167         FixupType fixup = insns[i].fixup;
168 
169         switch (fixup) {
170         case FIXUP_NONE:
171             break;
172         case FIXUP_BOARDID:
173         case FIXUP_BOARD_SETUP:
174         case FIXUP_ARGPTR:
175         case FIXUP_ENTRYPOINT:
176         case FIXUP_GIC_CPU_IF:
177         case FIXUP_BOOTREG:
178         case FIXUP_DSB:
179             insn = fixupcontext[fixup];
180             break;
181         default:
182             abort();
183         }
184         code[i] = tswap32(insn);
185     }
186 
187     rom_add_blob_fixed_as(name, code, len * sizeof(uint32_t), addr, as);
188 
189     g_free(code);
190 }
191 
192 static void default_write_secondary(ARMCPU *cpu,
193                                     const struct arm_boot_info *info)
194 {
195     uint32_t fixupcontext[FIXUP_MAX];
196     AddressSpace *as = arm_boot_address_space(cpu, info);
197 
198     fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr;
199     fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr;
200     if (arm_feature(&cpu->env, ARM_FEATURE_V7)) {
201         fixupcontext[FIXUP_DSB] = DSB_INSN;
202     } else {
203         fixupcontext[FIXUP_DSB] = CP15_DSB_INSN;
204     }
205 
206     write_bootloader("smpboot", info->smp_loader_start,
207                      smpboot, fixupcontext, as);
208 }
209 
210 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu,
211                                             const struct arm_boot_info *info,
212                                             hwaddr mvbar_addr)
213 {
214     AddressSpace *as = arm_boot_address_space(cpu, info);
215     int n;
216     uint32_t mvbar_blob[] = {
217         /* mvbar_addr: secure monitor vectors
218          * Default unimplemented and unused vectors to spin. Makes it
219          * easier to debug (as opposed to the CPU running away).
220          */
221         0xeafffffe, /* (spin) */
222         0xeafffffe, /* (spin) */
223         0xe1b0f00e, /* movs pc, lr ;SMC exception return */
224         0xeafffffe, /* (spin) */
225         0xeafffffe, /* (spin) */
226         0xeafffffe, /* (spin) */
227         0xeafffffe, /* (spin) */
228         0xeafffffe, /* (spin) */
229     };
230     uint32_t board_setup_blob[] = {
231         /* board setup addr */
232         0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */
233         0xee0c0f30, /* mcr     p15, 0, r0, c12, c0, 1 ;set MVBAR */
234         0xee110f11, /* mrc     p15, 0, r0, c1 , c1, 0 ;read SCR */
235         0xe3800031, /* orr     r0, #0x31              ;enable AW, FW, NS */
236         0xee010f11, /* mcr     p15, 0, r0, c1, c1, 0  ;write SCR */
237         0xe1a0100e, /* mov     r1, lr                 ;save LR across SMC */
238         0xe1600070, /* smc     #0                     ;call monitor to flush SCR */
239         0xe1a0f001, /* mov     pc, r1                 ;return */
240     };
241 
242     /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */
243     assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100);
244 
245     /* check that these blobs don't overlap */
246     assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr)
247           || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr));
248 
249     for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) {
250         mvbar_blob[n] = tswap32(mvbar_blob[n]);
251     }
252     rom_add_blob_fixed_as("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob),
253                           mvbar_addr, as);
254 
255     for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) {
256         board_setup_blob[n] = tswap32(board_setup_blob[n]);
257     }
258     rom_add_blob_fixed_as("board-setup", board_setup_blob,
259                           sizeof(board_setup_blob), info->board_setup_addr, as);
260 }
261 
262 static void default_reset_secondary(ARMCPU *cpu,
263                                     const struct arm_boot_info *info)
264 {
265     AddressSpace *as = arm_boot_address_space(cpu, info);
266     CPUState *cs = CPU(cpu);
267 
268     address_space_stl_notdirty(as, info->smp_bootreg_addr,
269                                0, MEMTXATTRS_UNSPECIFIED, NULL);
270     cpu_set_pc(cs, info->smp_loader_start);
271 }
272 
273 static inline bool have_dtb(const struct arm_boot_info *info)
274 {
275     return info->dtb_filename || info->get_dtb;
276 }
277 
278 #define WRITE_WORD(p, value) do { \
279     address_space_stl_notdirty(as, p, value, \
280                                MEMTXATTRS_UNSPECIFIED, NULL);  \
281     p += 4;                       \
282 } while (0)
283 
284 static void set_kernel_args(const struct arm_boot_info *info, AddressSpace *as)
285 {
286     int initrd_size = info->initrd_size;
287     hwaddr base = info->loader_start;
288     hwaddr p;
289 
290     p = base + KERNEL_ARGS_ADDR;
291     /* ATAG_CORE */
292     WRITE_WORD(p, 5);
293     WRITE_WORD(p, 0x54410001);
294     WRITE_WORD(p, 1);
295     WRITE_WORD(p, 0x1000);
296     WRITE_WORD(p, 0);
297     /* ATAG_MEM */
298     /* TODO: handle multiple chips on one ATAG list */
299     WRITE_WORD(p, 4);
300     WRITE_WORD(p, 0x54410002);
301     WRITE_WORD(p, info->ram_size);
302     WRITE_WORD(p, info->loader_start);
303     if (initrd_size) {
304         /* ATAG_INITRD2 */
305         WRITE_WORD(p, 4);
306         WRITE_WORD(p, 0x54420005);
307         WRITE_WORD(p, info->initrd_start);
308         WRITE_WORD(p, initrd_size);
309     }
310     if (info->kernel_cmdline && *info->kernel_cmdline) {
311         /* ATAG_CMDLINE */
312         int cmdline_size;
313 
314         cmdline_size = strlen(info->kernel_cmdline);
315         address_space_write(as, p + 8, MEMTXATTRS_UNSPECIFIED,
316                             (const uint8_t *)info->kernel_cmdline,
317                             cmdline_size + 1);
318         cmdline_size = (cmdline_size >> 2) + 1;
319         WRITE_WORD(p, cmdline_size + 2);
320         WRITE_WORD(p, 0x54410009);
321         p += cmdline_size * 4;
322     }
323     if (info->atag_board) {
324         /* ATAG_BOARD */
325         int atag_board_len;
326         uint8_t atag_board_buf[0x1000];
327 
328         atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3;
329         WRITE_WORD(p, (atag_board_len + 8) >> 2);
330         WRITE_WORD(p, 0x414f4d50);
331         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
332                             atag_board_buf, atag_board_len);
333         p += atag_board_len;
334     }
335     /* ATAG_END */
336     WRITE_WORD(p, 0);
337     WRITE_WORD(p, 0);
338 }
339 
340 static void set_kernel_args_old(const struct arm_boot_info *info,
341                                 AddressSpace *as)
342 {
343     hwaddr p;
344     const char *s;
345     int initrd_size = info->initrd_size;
346     hwaddr base = info->loader_start;
347 
348     /* see linux/include/asm-arm/setup.h */
349     p = base + KERNEL_ARGS_ADDR;
350     /* page_size */
351     WRITE_WORD(p, 4096);
352     /* nr_pages */
353     WRITE_WORD(p, info->ram_size / 4096);
354     /* ramdisk_size */
355     WRITE_WORD(p, 0);
356 #define FLAG_READONLY	1
357 #define FLAG_RDLOAD	4
358 #define FLAG_RDPROMPT	8
359     /* flags */
360     WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT);
361     /* rootdev */
362     WRITE_WORD(p, (31 << 8) | 0);	/* /dev/mtdblock0 */
363     /* video_num_cols */
364     WRITE_WORD(p, 0);
365     /* video_num_rows */
366     WRITE_WORD(p, 0);
367     /* video_x */
368     WRITE_WORD(p, 0);
369     /* video_y */
370     WRITE_WORD(p, 0);
371     /* memc_control_reg */
372     WRITE_WORD(p, 0);
373     /* unsigned char sounddefault */
374     /* unsigned char adfsdrives */
375     /* unsigned char bytes_per_char_h */
376     /* unsigned char bytes_per_char_v */
377     WRITE_WORD(p, 0);
378     /* pages_in_bank[4] */
379     WRITE_WORD(p, 0);
380     WRITE_WORD(p, 0);
381     WRITE_WORD(p, 0);
382     WRITE_WORD(p, 0);
383     /* pages_in_vram */
384     WRITE_WORD(p, 0);
385     /* initrd_start */
386     if (initrd_size) {
387         WRITE_WORD(p, info->initrd_start);
388     } else {
389         WRITE_WORD(p, 0);
390     }
391     /* initrd_size */
392     WRITE_WORD(p, initrd_size);
393     /* rd_start */
394     WRITE_WORD(p, 0);
395     /* system_rev */
396     WRITE_WORD(p, 0);
397     /* system_serial_low */
398     WRITE_WORD(p, 0);
399     /* system_serial_high */
400     WRITE_WORD(p, 0);
401     /* mem_fclk_21285 */
402     WRITE_WORD(p, 0);
403     /* zero unused fields */
404     while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) {
405         WRITE_WORD(p, 0);
406     }
407     s = info->kernel_cmdline;
408     if (s) {
409         address_space_write(as, p, MEMTXATTRS_UNSPECIFIED,
410                             (const uint8_t *)s, strlen(s) + 1);
411     } else {
412         WRITE_WORD(p, 0);
413     }
414 }
415 
416 static void fdt_add_psci_node(void *fdt)
417 {
418     uint32_t cpu_suspend_fn;
419     uint32_t cpu_off_fn;
420     uint32_t cpu_on_fn;
421     uint32_t migrate_fn;
422     ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
423     const char *psci_method;
424     int64_t psci_conduit;
425     int rc;
426 
427     psci_conduit = object_property_get_int(OBJECT(armcpu),
428                                            "psci-conduit",
429                                            &error_abort);
430     switch (psci_conduit) {
431     case QEMU_PSCI_CONDUIT_DISABLED:
432         return;
433     case QEMU_PSCI_CONDUIT_HVC:
434         psci_method = "hvc";
435         break;
436     case QEMU_PSCI_CONDUIT_SMC:
437         psci_method = "smc";
438         break;
439     default:
440         g_assert_not_reached();
441     }
442 
443     /*
444      * If /psci node is present in provided DTB, assume that no fixup
445      * is necessary and all PSCI configuration should be taken as-is
446      */
447     rc = fdt_path_offset(fdt, "/psci");
448     if (rc >= 0) {
449         return;
450     }
451 
452     qemu_fdt_add_subnode(fdt, "/psci");
453     if (armcpu->psci_version == 2) {
454         const char comp[] = "arm,psci-0.2\0arm,psci";
455         qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
456 
457         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
458         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
459             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
460             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
461             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
462         } else {
463             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
464             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
465             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
466         }
467     } else {
468         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
469 
470         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
471         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
472         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
473         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
474     }
475 
476     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
477      * to the instruction that should be used to invoke PSCI functions.
478      * However, the device tree binding uses 'method' instead, so that is
479      * what we should use here.
480      */
481     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
482 
483     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
484     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
485     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
486     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
487 }
488 
489 int arm_load_dtb(hwaddr addr, const struct arm_boot_info *binfo,
490                  hwaddr addr_limit, AddressSpace *as)
491 {
492     void *fdt = NULL;
493     int size, rc, n = 0;
494     uint32_t acells, scells;
495     char *nodename;
496     unsigned int i;
497     hwaddr mem_base, mem_len;
498     char **node_path;
499     Error *err = NULL;
500 
501     if (binfo->dtb_filename) {
502         char *filename;
503         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
504         if (!filename) {
505             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
506             goto fail;
507         }
508 
509         fdt = load_device_tree(filename, &size);
510         if (!fdt) {
511             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
512             g_free(filename);
513             goto fail;
514         }
515         g_free(filename);
516     } else {
517         fdt = binfo->get_dtb(binfo, &size);
518         if (!fdt) {
519             fprintf(stderr, "Board was unable to create a dtb blob\n");
520             goto fail;
521         }
522     }
523 
524     if (addr_limit > addr && size > (addr_limit - addr)) {
525         /* Installing the device tree blob at addr would exceed addr_limit.
526          * Whether this constitutes failure is up to the caller to decide,
527          * so just return 0 as size, i.e., no error.
528          */
529         g_free(fdt);
530         return 0;
531     }
532 
533     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
534                                    NULL, &error_fatal);
535     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
536                                    NULL, &error_fatal);
537     if (acells == 0 || scells == 0) {
538         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
539         goto fail;
540     }
541 
542     if (scells < 2 && binfo->ram_size >= (1ULL << 32)) {
543         /* This is user error so deserves a friendlier error message
544          * than the failure of setprop_sized_cells would provide
545          */
546         fprintf(stderr, "qemu: dtb file not compatible with "
547                 "RAM size > 4GB\n");
548         goto fail;
549     }
550 
551     /* nop all root nodes matching /memory or /memory@unit-address */
552     node_path = qemu_fdt_node_unit_path(fdt, "memory", &err);
553     if (err) {
554         error_report_err(err);
555         goto fail;
556     }
557     while (node_path[n]) {
558         if (g_str_has_prefix(node_path[n], "/memory")) {
559             qemu_fdt_nop_node(fdt, node_path[n]);
560         }
561         n++;
562     }
563     g_strfreev(node_path);
564 
565     if (nb_numa_nodes > 0) {
566         mem_base = binfo->loader_start;
567         for (i = 0; i < nb_numa_nodes; i++) {
568             mem_len = numa_info[i].node_mem;
569             nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
570             qemu_fdt_add_subnode(fdt, nodename);
571             qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
572             rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
573                                               acells, mem_base,
574                                               scells, mem_len);
575             if (rc < 0) {
576                 fprintf(stderr, "couldn't set %s/reg for node %d\n", nodename,
577                         i);
578                 goto fail;
579             }
580 
581             qemu_fdt_setprop_cell(fdt, nodename, "numa-node-id", i);
582             mem_base += mem_len;
583             g_free(nodename);
584         }
585     } else {
586         nodename = g_strdup_printf("/memory@%" PRIx64, binfo->loader_start);
587         qemu_fdt_add_subnode(fdt, nodename);
588         qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
589 
590         rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
591                                           acells, binfo->loader_start,
592                                           scells, binfo->ram_size);
593         if (rc < 0) {
594             fprintf(stderr, "couldn't set %s reg\n", nodename);
595             goto fail;
596         }
597         g_free(nodename);
598     }
599 
600     rc = fdt_path_offset(fdt, "/chosen");
601     if (rc < 0) {
602         qemu_fdt_add_subnode(fdt, "/chosen");
603     }
604 
605     if (binfo->kernel_cmdline && *binfo->kernel_cmdline) {
606         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
607                                      binfo->kernel_cmdline);
608         if (rc < 0) {
609             fprintf(stderr, "couldn't set /chosen/bootargs\n");
610             goto fail;
611         }
612     }
613 
614     if (binfo->initrd_size) {
615         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
616                                    binfo->initrd_start);
617         if (rc < 0) {
618             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
619             goto fail;
620         }
621 
622         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
623                                    binfo->initrd_start + binfo->initrd_size);
624         if (rc < 0) {
625             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
626             goto fail;
627         }
628     }
629 
630     fdt_add_psci_node(fdt);
631 
632     if (binfo->modify_dtb) {
633         binfo->modify_dtb(binfo, fdt);
634     }
635 
636     qemu_fdt_dumpdtb(fdt, size);
637 
638     /* Put the DTB into the memory map as a ROM image: this will ensure
639      * the DTB is copied again upon reset, even if addr points into RAM.
640      */
641     rom_add_blob_fixed_as("dtb", fdt, size, addr, as);
642 
643     g_free(fdt);
644 
645     return size;
646 
647 fail:
648     g_free(fdt);
649     return -1;
650 }
651 
652 static void do_cpu_reset(void *opaque)
653 {
654     ARMCPU *cpu = opaque;
655     CPUState *cs = CPU(cpu);
656     CPUARMState *env = &cpu->env;
657     const struct arm_boot_info *info = env->boot_info;
658 
659     cpu_reset(cs);
660     if (info) {
661         if (!info->is_linux) {
662             int i;
663             /* Jump to the entry point.  */
664             uint64_t entry = info->entry;
665 
666             switch (info->endianness) {
667             case ARM_ENDIANNESS_LE:
668                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
669                 for (i = 1; i < 4; ++i) {
670                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
671                 }
672                 env->uncached_cpsr &= ~CPSR_E;
673                 break;
674             case ARM_ENDIANNESS_BE8:
675                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
676                 for (i = 1; i < 4; ++i) {
677                     env->cp15.sctlr_el[i] |= SCTLR_EE;
678                 }
679                 env->uncached_cpsr |= CPSR_E;
680                 break;
681             case ARM_ENDIANNESS_BE32:
682                 env->cp15.sctlr_el[1] |= SCTLR_B;
683                 break;
684             case ARM_ENDIANNESS_UNKNOWN:
685                 break; /* Board's decision */
686             default:
687                 g_assert_not_reached();
688             }
689 
690             if (!env->aarch64) {
691                 env->thumb = info->entry & 1;
692                 entry &= 0xfffffffe;
693             }
694             cpu_set_pc(cs, entry);
695         } else {
696             /* If we are booting Linux then we need to check whether we are
697              * booting into secure or non-secure state and adjust the state
698              * accordingly.  Out of reset, ARM is defined to be in secure state
699              * (SCR.NS = 0), we change that here if non-secure boot has been
700              * requested.
701              */
702             if (arm_feature(env, ARM_FEATURE_EL3)) {
703                 /* AArch64 is defined to come out of reset into EL3 if enabled.
704                  * If we are booting Linux then we need to adjust our EL as
705                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
706                  * SVC, which Linux expects, so no privilege/exception level to
707                  * adjust.
708                  */
709                 if (env->aarch64) {
710                     env->cp15.scr_el3 |= SCR_RW;
711                     if (arm_feature(env, ARM_FEATURE_EL2)) {
712                         env->cp15.hcr_el2 |= HCR_RW;
713                         env->pstate = PSTATE_MODE_EL2h;
714                     } else {
715                         env->pstate = PSTATE_MODE_EL1h;
716                     }
717                     /* AArch64 kernels never boot in secure mode */
718                     assert(!info->secure_boot);
719                     /* This hook is only supported for AArch32 currently:
720                      * bootloader_aarch64[] will not call the hook, and
721                      * the code above has already dropped us into EL2 or EL1.
722                      */
723                     assert(!info->secure_board_setup);
724                 }
725 
726                 if (arm_feature(env, ARM_FEATURE_EL2)) {
727                     /* If we have EL2 then Linux expects the HVC insn to work */
728                     env->cp15.scr_el3 |= SCR_HCE;
729                 }
730 
731                 /* Set to non-secure if not a secure boot */
732                 if (!info->secure_boot &&
733                     (cs != first_cpu || !info->secure_board_setup)) {
734                     /* Linux expects non-secure state */
735                     env->cp15.scr_el3 |= SCR_NS;
736                 }
737             }
738 
739             if (!env->aarch64 && !info->secure_boot &&
740                 arm_feature(env, ARM_FEATURE_EL2)) {
741                 /*
742                  * This is an AArch32 boot not to Secure state, and
743                  * we have Hyp mode available, so boot the kernel into
744                  * Hyp mode. This is not how the CPU comes out of reset,
745                  * so we need to manually put it there.
746                  */
747                 cpsr_write(env, ARM_CPU_MODE_HYP, CPSR_M, CPSRWriteRaw);
748             }
749 
750             if (cs == first_cpu) {
751                 AddressSpace *as = arm_boot_address_space(cpu, info);
752 
753                 cpu_set_pc(cs, info->loader_start);
754 
755                 if (!have_dtb(info)) {
756                     if (old_param) {
757                         set_kernel_args_old(info, as);
758                     } else {
759                         set_kernel_args(info, as);
760                     }
761                 }
762             } else {
763                 info->secondary_cpu_reset_hook(cpu, info);
764             }
765         }
766     }
767 }
768 
769 /**
770  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
771  *                          by key.
772  * @fw_cfg:         The firmware config instance to store the data in.
773  * @size_key:       The firmware config key to store the size of the loaded
774  *                  data under, with fw_cfg_add_i32().
775  * @data_key:       The firmware config key to store the loaded data under,
776  *                  with fw_cfg_add_bytes().
777  * @image_name:     The name of the image file to load. If it is NULL, the
778  *                  function returns without doing anything.
779  * @try_decompress: Whether the image should be decompressed (gunzipped) before
780  *                  adding it to fw_cfg. If decompression fails, the image is
781  *                  loaded as-is.
782  *
783  * In case of failure, the function prints an error message to stderr and the
784  * process exits with status 1.
785  */
786 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
787                                  uint16_t data_key, const char *image_name,
788                                  bool try_decompress)
789 {
790     size_t size = -1;
791     uint8_t *data;
792 
793     if (image_name == NULL) {
794         return;
795     }
796 
797     if (try_decompress) {
798         size = load_image_gzipped_buffer(image_name,
799                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
800     }
801 
802     if (size == (size_t)-1) {
803         gchar *contents;
804         gsize length;
805 
806         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
807             error_report("failed to load \"%s\"", image_name);
808             exit(1);
809         }
810         size = length;
811         data = (uint8_t *)contents;
812     }
813 
814     fw_cfg_add_i32(fw_cfg, size_key, size);
815     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
816 }
817 
818 static int do_arm_linux_init(Object *obj, void *opaque)
819 {
820     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
821         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
822         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
823         struct arm_boot_info *info = opaque;
824 
825         if (albifc->arm_linux_init) {
826             albifc->arm_linux_init(albif, info->secure_boot);
827         }
828     }
829     return 0;
830 }
831 
832 static int64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
833                             uint64_t *lowaddr, uint64_t *highaddr,
834                             int elf_machine, AddressSpace *as)
835 {
836     bool elf_is64;
837     union {
838         Elf32_Ehdr h32;
839         Elf64_Ehdr h64;
840     } elf_header;
841     int data_swab = 0;
842     bool big_endian;
843     int64_t ret = -1;
844     Error *err = NULL;
845 
846 
847     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
848     if (err) {
849         error_free(err);
850         return ret;
851     }
852 
853     if (elf_is64) {
854         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
855         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
856                                       : ARM_ENDIANNESS_LE;
857     } else {
858         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
859         if (big_endian) {
860             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
861                 info->endianness = ARM_ENDIANNESS_BE8;
862             } else {
863                 info->endianness = ARM_ENDIANNESS_BE32;
864                 /* In BE32, the CPU has a different view of the per-byte
865                  * address map than the rest of the system. BE32 ELF files
866                  * are organised such that they can be programmed through
867                  * the CPU's per-word byte-reversed view of the world. QEMU
868                  * however loads ELF files independently of the CPU. So
869                  * tell the ELF loader to byte reverse the data for us.
870                  */
871                 data_swab = 2;
872             }
873         } else {
874             info->endianness = ARM_ENDIANNESS_LE;
875         }
876     }
877 
878     ret = load_elf_as(info->kernel_filename, NULL, NULL,
879                       pentry, lowaddr, highaddr, big_endian, elf_machine,
880                       1, data_swab, as);
881     if (ret <= 0) {
882         /* The header loaded but the image didn't */
883         exit(1);
884     }
885 
886     return ret;
887 }
888 
889 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
890                                    hwaddr *entry, AddressSpace *as)
891 {
892     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
893     uint8_t *buffer;
894     int size;
895 
896     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
897     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
898                                      &buffer);
899 
900     if (size < 0) {
901         gsize len;
902 
903         /* Load as raw file otherwise */
904         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
905             return -1;
906         }
907         size = len;
908     }
909 
910     /* check the arm64 magic header value -- very old kernels may not have it */
911     if (size > ARM64_MAGIC_OFFSET + 4 &&
912         memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
913         uint64_t hdrvals[2];
914 
915         /* The arm64 Image header has text_offset and image_size fields at 8 and
916          * 16 bytes into the Image header, respectively. The text_offset field
917          * is only valid if the image_size is non-zero.
918          */
919         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
920         if (hdrvals[1] != 0) {
921             kernel_load_offset = le64_to_cpu(hdrvals[0]);
922         }
923     }
924 
925     *entry = mem_base + kernel_load_offset;
926     rom_add_blob_fixed_as(filename, buffer, size, *entry, as);
927 
928     g_free(buffer);
929 
930     return size;
931 }
932 
933 void arm_load_kernel(ARMCPU *cpu, struct arm_boot_info *info)
934 {
935     CPUState *cs;
936     int kernel_size;
937     int initrd_size;
938     int is_linux = 0;
939     uint64_t elf_entry, elf_low_addr, elf_high_addr;
940     int elf_machine;
941     hwaddr entry;
942     static const ARMInsnFixup *primary_loader;
943     AddressSpace *as = arm_boot_address_space(cpu, info);
944 
945     /* CPU objects (unlike devices) are not automatically reset on system
946      * reset, so we must always register a handler to do so. If we're
947      * actually loading a kernel, the handler is also responsible for
948      * arranging that we start it correctly.
949      */
950     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
951         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
952     }
953 
954     /* The board code is not supposed to set secure_board_setup unless
955      * running its code in secure mode is actually possible, and KVM
956      * doesn't support secure.
957      */
958     assert(!(info->secure_board_setup && kvm_enabled()));
959 
960     info->dtb_filename = qemu_opt_get(qemu_get_machine_opts(), "dtb");
961     info->dtb_limit = 0;
962 
963     /* Load the kernel.  */
964     if (!info->kernel_filename || info->firmware_loaded) {
965 
966         if (have_dtb(info)) {
967             /* If we have a device tree blob, but no kernel to supply it to (or
968              * the kernel is supposed to be loaded by the bootloader), copy the
969              * DTB to the base of RAM for the bootloader to pick up.
970              */
971             info->dtb_start = info->loader_start;
972         }
973 
974         if (info->kernel_filename) {
975             FWCfgState *fw_cfg;
976             bool try_decompressing_kernel;
977 
978             fw_cfg = fw_cfg_find();
979             try_decompressing_kernel = arm_feature(&cpu->env,
980                                                    ARM_FEATURE_AARCH64);
981 
982             /* Expose the kernel, the command line, and the initrd in fw_cfg.
983              * We don't process them here at all, it's all left to the
984              * firmware.
985              */
986             load_image_to_fw_cfg(fw_cfg,
987                                  FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
988                                  info->kernel_filename,
989                                  try_decompressing_kernel);
990             load_image_to_fw_cfg(fw_cfg,
991                                  FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
992                                  info->initrd_filename, false);
993 
994             if (info->kernel_cmdline) {
995                 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
996                                strlen(info->kernel_cmdline) + 1);
997                 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
998                                   info->kernel_cmdline);
999             }
1000         }
1001 
1002         /* We will start from address 0 (typically a boot ROM image) in the
1003          * same way as hardware.
1004          */
1005         return;
1006     }
1007 
1008     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
1009         primary_loader = bootloader_aarch64;
1010         elf_machine = EM_AARCH64;
1011     } else {
1012         primary_loader = bootloader;
1013         if (!info->write_board_setup) {
1014             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
1015         }
1016         elf_machine = EM_ARM;
1017     }
1018 
1019     if (!info->secondary_cpu_reset_hook) {
1020         info->secondary_cpu_reset_hook = default_reset_secondary;
1021     }
1022     if (!info->write_secondary_boot) {
1023         info->write_secondary_boot = default_write_secondary;
1024     }
1025 
1026     if (info->nb_cpus == 0)
1027         info->nb_cpus = 1;
1028 
1029     /* We want to put the initrd far enough into RAM that when the
1030      * kernel is uncompressed it will not clobber the initrd. However
1031      * on boards without much RAM we must ensure that we still leave
1032      * enough room for a decent sized initrd, and on boards with large
1033      * amounts of RAM we must avoid the initrd being so far up in RAM
1034      * that it is outside lowmem and inaccessible to the kernel.
1035      * So for boards with less  than 256MB of RAM we put the initrd
1036      * halfway into RAM, and for boards with 256MB of RAM or more we put
1037      * the initrd at 128MB.
1038      */
1039     info->initrd_start = info->loader_start +
1040         MIN(info->ram_size / 2, 128 * 1024 * 1024);
1041 
1042     /* Assume that raw images are linux kernels, and ELF images are not.  */
1043     kernel_size = arm_load_elf(info, &elf_entry, &elf_low_addr,
1044                                &elf_high_addr, elf_machine, as);
1045     if (kernel_size > 0 && have_dtb(info)) {
1046         /* If there is still some room left at the base of RAM, try and put
1047          * the DTB there like we do for images loaded with -bios or -pflash.
1048          */
1049         if (elf_low_addr > info->loader_start
1050             || elf_high_addr < info->loader_start) {
1051             /* Set elf_low_addr as address limit for arm_load_dtb if it may be
1052              * pointing into RAM, otherwise pass '0' (no limit)
1053              */
1054             if (elf_low_addr < info->loader_start) {
1055                 elf_low_addr = 0;
1056             }
1057             info->dtb_start = info->loader_start;
1058             info->dtb_limit = elf_low_addr;
1059         }
1060     }
1061     entry = elf_entry;
1062     if (kernel_size < 0) {
1063         kernel_size = load_uimage_as(info->kernel_filename, &entry, NULL,
1064                                      &is_linux, NULL, NULL, as);
1065     }
1066     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1067         kernel_size = load_aarch64_image(info->kernel_filename,
1068                                          info->loader_start, &entry, as);
1069         is_linux = 1;
1070     } else if (kernel_size < 0) {
1071         /* 32-bit ARM */
1072         entry = info->loader_start + KERNEL_LOAD_ADDR;
1073         kernel_size = load_image_targphys_as(info->kernel_filename, entry,
1074                                              info->ram_size - KERNEL_LOAD_ADDR,
1075                                              as);
1076         is_linux = 1;
1077     }
1078     if (kernel_size < 0) {
1079         error_report("could not load kernel '%s'", info->kernel_filename);
1080         exit(1);
1081     }
1082     info->entry = entry;
1083     if (is_linux) {
1084         uint32_t fixupcontext[FIXUP_MAX];
1085 
1086         if (info->initrd_filename) {
1087             initrd_size = load_ramdisk_as(info->initrd_filename,
1088                                           info->initrd_start,
1089                                           info->ram_size - info->initrd_start,
1090                                           as);
1091             if (initrd_size < 0) {
1092                 initrd_size = load_image_targphys_as(info->initrd_filename,
1093                                                      info->initrd_start,
1094                                                      info->ram_size -
1095                                                      info->initrd_start,
1096                                                      as);
1097             }
1098             if (initrd_size < 0) {
1099                 error_report("could not load initrd '%s'",
1100                              info->initrd_filename);
1101                 exit(1);
1102             }
1103         } else {
1104             initrd_size = 0;
1105         }
1106         info->initrd_size = initrd_size;
1107 
1108         fixupcontext[FIXUP_BOARDID] = info->board_id;
1109         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1110 
1111         /* for device tree boot, we pass the DTB directly in r2. Otherwise
1112          * we point to the kernel args.
1113          */
1114         if (have_dtb(info)) {
1115             hwaddr align;
1116 
1117             if (elf_machine == EM_AARCH64) {
1118                 /*
1119                  * Some AArch64 kernels on early bootup map the fdt region as
1120                  *
1121                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1122                  *
1123                  * Let's play safe and prealign it to 2MB to give us some space.
1124                  */
1125                 align = 2 * 1024 * 1024;
1126             } else {
1127                 /*
1128                  * Some 32bit kernels will trash anything in the 4K page the
1129                  * initrd ends in, so make sure the DTB isn't caught up in that.
1130                  */
1131                 align = 4096;
1132             }
1133 
1134             /* Place the DTB after the initrd in memory with alignment. */
1135             info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size,
1136                                            align);
1137             fixupcontext[FIXUP_ARGPTR] = info->dtb_start;
1138         } else {
1139             fixupcontext[FIXUP_ARGPTR] = info->loader_start + KERNEL_ARGS_ADDR;
1140             if (info->ram_size >= (1ULL << 32)) {
1141                 error_report("RAM size must be less than 4GB to boot"
1142                              " Linux kernel using ATAGS (try passing a device tree"
1143                              " using -dtb)");
1144                 exit(1);
1145             }
1146         }
1147         fixupcontext[FIXUP_ENTRYPOINT] = entry;
1148 
1149         write_bootloader("bootloader", info->loader_start,
1150                          primary_loader, fixupcontext, as);
1151 
1152         if (info->nb_cpus > 1) {
1153             info->write_secondary_boot(cpu, info);
1154         }
1155         if (info->write_board_setup) {
1156             info->write_board_setup(cpu, info);
1157         }
1158 
1159         /* Notify devices which need to fake up firmware initialization
1160          * that we're doing a direct kernel boot.
1161          */
1162         object_child_foreach_recursive(object_get_root(),
1163                                        do_arm_linux_init, info);
1164     }
1165     info->is_linux = is_linux;
1166 
1167     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1168         ARM_CPU(cs)->env.boot_info = info;
1169     }
1170 
1171     if (!info->skip_dtb_autoload && have_dtb(info)) {
1172         if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as) < 0) {
1173             exit(1);
1174         }
1175     }
1176 }
1177 
1178 static const TypeInfo arm_linux_boot_if_info = {
1179     .name = TYPE_ARM_LINUX_BOOT_IF,
1180     .parent = TYPE_INTERFACE,
1181     .class_size = sizeof(ARMLinuxBootIfClass),
1182 };
1183 
1184 static void arm_linux_boot_register_types(void)
1185 {
1186     type_register_static(&arm_linux_boot_if_info);
1187 }
1188 
1189 type_init(arm_linux_boot_register_types)
1190