xref: /openbmc/qemu/hw/arm/boot.c (revision 7d405b2f)
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;
494     uint32_t acells, scells;
495     char *nodename;
496     unsigned int i;
497     hwaddr mem_base, mem_len;
498 
499     if (binfo->dtb_filename) {
500         char *filename;
501         filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename);
502         if (!filename) {
503             fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename);
504             goto fail;
505         }
506 
507         fdt = load_device_tree(filename, &size);
508         if (!fdt) {
509             fprintf(stderr, "Couldn't open dtb file %s\n", filename);
510             g_free(filename);
511             goto fail;
512         }
513         g_free(filename);
514     } else {
515         fdt = binfo->get_dtb(binfo, &size);
516         if (!fdt) {
517             fprintf(stderr, "Board was unable to create a dtb blob\n");
518             goto fail;
519         }
520     }
521 
522     if (addr_limit > addr && size > (addr_limit - addr)) {
523         /* Installing the device tree blob at addr would exceed addr_limit.
524          * Whether this constitutes failure is up to the caller to decide,
525          * so just return 0 as size, i.e., no error.
526          */
527         g_free(fdt);
528         return 0;
529     }
530 
531     acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells",
532                                    NULL, &error_fatal);
533     scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells",
534                                    NULL, &error_fatal);
535     if (acells == 0 || scells == 0) {
536         fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n");
537         goto fail;
538     }
539 
540     if (scells < 2 && binfo->ram_size >= (1ULL << 32)) {
541         /* This is user error so deserves a friendlier error message
542          * than the failure of setprop_sized_cells would provide
543          */
544         fprintf(stderr, "qemu: dtb file not compatible with "
545                 "RAM size > 4GB\n");
546         goto fail;
547     }
548 
549     if (nb_numa_nodes > 0) {
550         /*
551          * Turn the /memory node created before into a NOP node, then create
552          * /memory@addr nodes for all numa nodes respectively.
553          */
554         qemu_fdt_nop_node(fdt, "/memory");
555         mem_base = binfo->loader_start;
556         for (i = 0; i < nb_numa_nodes; i++) {
557             mem_len = numa_info[i].node_mem;
558             nodename = g_strdup_printf("/memory@%" PRIx64, mem_base);
559             qemu_fdt_add_subnode(fdt, nodename);
560             qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory");
561             rc = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg",
562                                               acells, mem_base,
563                                               scells, mem_len);
564             if (rc < 0) {
565                 fprintf(stderr, "couldn't set %s/reg for node %d\n", nodename,
566                         i);
567                 goto fail;
568             }
569 
570             qemu_fdt_setprop_cell(fdt, nodename, "numa-node-id", i);
571             mem_base += mem_len;
572             g_free(nodename);
573         }
574     } else {
575         Error *err = NULL;
576 
577         rc = fdt_path_offset(fdt, "/memory");
578         if (rc < 0) {
579             qemu_fdt_add_subnode(fdt, "/memory");
580         }
581 
582         if (!qemu_fdt_getprop(fdt, "/memory", "device_type", NULL, &err)) {
583             qemu_fdt_setprop_string(fdt, "/memory", "device_type", "memory");
584         }
585 
586         rc = qemu_fdt_setprop_sized_cells(fdt, "/memory", "reg",
587                                           acells, binfo->loader_start,
588                                           scells, binfo->ram_size);
589         if (rc < 0) {
590             fprintf(stderr, "couldn't set /memory/reg\n");
591             goto fail;
592         }
593     }
594 
595     rc = fdt_path_offset(fdt, "/chosen");
596     if (rc < 0) {
597         qemu_fdt_add_subnode(fdt, "/chosen");
598     }
599 
600     if (binfo->kernel_cmdline && *binfo->kernel_cmdline) {
601         rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
602                                      binfo->kernel_cmdline);
603         if (rc < 0) {
604             fprintf(stderr, "couldn't set /chosen/bootargs\n");
605             goto fail;
606         }
607     }
608 
609     if (binfo->initrd_size) {
610         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start",
611                                    binfo->initrd_start);
612         if (rc < 0) {
613             fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n");
614             goto fail;
615         }
616 
617         rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end",
618                                    binfo->initrd_start + binfo->initrd_size);
619         if (rc < 0) {
620             fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n");
621             goto fail;
622         }
623     }
624 
625     fdt_add_psci_node(fdt);
626 
627     if (binfo->modify_dtb) {
628         binfo->modify_dtb(binfo, fdt);
629     }
630 
631     qemu_fdt_dumpdtb(fdt, size);
632 
633     /* Put the DTB into the memory map as a ROM image: this will ensure
634      * the DTB is copied again upon reset, even if addr points into RAM.
635      */
636     rom_add_blob_fixed_as("dtb", fdt, size, addr, as);
637 
638     g_free(fdt);
639 
640     return size;
641 
642 fail:
643     g_free(fdt);
644     return -1;
645 }
646 
647 static void do_cpu_reset(void *opaque)
648 {
649     ARMCPU *cpu = opaque;
650     CPUState *cs = CPU(cpu);
651     CPUARMState *env = &cpu->env;
652     const struct arm_boot_info *info = env->boot_info;
653 
654     cpu_reset(cs);
655     if (info) {
656         if (!info->is_linux) {
657             int i;
658             /* Jump to the entry point.  */
659             uint64_t entry = info->entry;
660 
661             switch (info->endianness) {
662             case ARM_ENDIANNESS_LE:
663                 env->cp15.sctlr_el[1] &= ~SCTLR_E0E;
664                 for (i = 1; i < 4; ++i) {
665                     env->cp15.sctlr_el[i] &= ~SCTLR_EE;
666                 }
667                 env->uncached_cpsr &= ~CPSR_E;
668                 break;
669             case ARM_ENDIANNESS_BE8:
670                 env->cp15.sctlr_el[1] |= SCTLR_E0E;
671                 for (i = 1; i < 4; ++i) {
672                     env->cp15.sctlr_el[i] |= SCTLR_EE;
673                 }
674                 env->uncached_cpsr |= CPSR_E;
675                 break;
676             case ARM_ENDIANNESS_BE32:
677                 env->cp15.sctlr_el[1] |= SCTLR_B;
678                 break;
679             case ARM_ENDIANNESS_UNKNOWN:
680                 break; /* Board's decision */
681             default:
682                 g_assert_not_reached();
683             }
684 
685             if (!env->aarch64) {
686                 env->thumb = info->entry & 1;
687                 entry &= 0xfffffffe;
688             }
689             cpu_set_pc(cs, entry);
690         } else {
691             /* If we are booting Linux then we need to check whether we are
692              * booting into secure or non-secure state and adjust the state
693              * accordingly.  Out of reset, ARM is defined to be in secure state
694              * (SCR.NS = 0), we change that here if non-secure boot has been
695              * requested.
696              */
697             if (arm_feature(env, ARM_FEATURE_EL3)) {
698                 /* AArch64 is defined to come out of reset into EL3 if enabled.
699                  * If we are booting Linux then we need to adjust our EL as
700                  * Linux expects us to be in EL2 or EL1.  AArch32 resets into
701                  * SVC, which Linux expects, so no privilege/exception level to
702                  * adjust.
703                  */
704                 if (env->aarch64) {
705                     env->cp15.scr_el3 |= SCR_RW;
706                     if (arm_feature(env, ARM_FEATURE_EL2)) {
707                         env->cp15.hcr_el2 |= HCR_RW;
708                         env->pstate = PSTATE_MODE_EL2h;
709                     } else {
710                         env->pstate = PSTATE_MODE_EL1h;
711                     }
712                     /* AArch64 kernels never boot in secure mode */
713                     assert(!info->secure_boot);
714                     /* This hook is only supported for AArch32 currently:
715                      * bootloader_aarch64[] will not call the hook, and
716                      * the code above has already dropped us into EL2 or EL1.
717                      */
718                     assert(!info->secure_board_setup);
719                 }
720 
721                 if (arm_feature(env, ARM_FEATURE_EL2)) {
722                     /* If we have EL2 then Linux expects the HVC insn to work */
723                     env->cp15.scr_el3 |= SCR_HCE;
724                 }
725 
726                 /* Set to non-secure if not a secure boot */
727                 if (!info->secure_boot &&
728                     (cs != first_cpu || !info->secure_board_setup)) {
729                     /* Linux expects non-secure state */
730                     env->cp15.scr_el3 |= SCR_NS;
731                 }
732             }
733 
734             if (cs == first_cpu) {
735                 AddressSpace *as = arm_boot_address_space(cpu, info);
736 
737                 cpu_set_pc(cs, info->loader_start);
738 
739                 if (!have_dtb(info)) {
740                     if (old_param) {
741                         set_kernel_args_old(info, as);
742                     } else {
743                         set_kernel_args(info, as);
744                     }
745                 }
746             } else {
747                 info->secondary_cpu_reset_hook(cpu, info);
748             }
749         }
750     }
751 }
752 
753 /**
754  * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified
755  *                          by key.
756  * @fw_cfg:         The firmware config instance to store the data in.
757  * @size_key:       The firmware config key to store the size of the loaded
758  *                  data under, with fw_cfg_add_i32().
759  * @data_key:       The firmware config key to store the loaded data under,
760  *                  with fw_cfg_add_bytes().
761  * @image_name:     The name of the image file to load. If it is NULL, the
762  *                  function returns without doing anything.
763  * @try_decompress: Whether the image should be decompressed (gunzipped) before
764  *                  adding it to fw_cfg. If decompression fails, the image is
765  *                  loaded as-is.
766  *
767  * In case of failure, the function prints an error message to stderr and the
768  * process exits with status 1.
769  */
770 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key,
771                                  uint16_t data_key, const char *image_name,
772                                  bool try_decompress)
773 {
774     size_t size = -1;
775     uint8_t *data;
776 
777     if (image_name == NULL) {
778         return;
779     }
780 
781     if (try_decompress) {
782         size = load_image_gzipped_buffer(image_name,
783                                          LOAD_IMAGE_MAX_GUNZIP_BYTES, &data);
784     }
785 
786     if (size == (size_t)-1) {
787         gchar *contents;
788         gsize length;
789 
790         if (!g_file_get_contents(image_name, &contents, &length, NULL)) {
791             error_report("failed to load \"%s\"", image_name);
792             exit(1);
793         }
794         size = length;
795         data = (uint8_t *)contents;
796     }
797 
798     fw_cfg_add_i32(fw_cfg, size_key, size);
799     fw_cfg_add_bytes(fw_cfg, data_key, data, size);
800 }
801 
802 static int do_arm_linux_init(Object *obj, void *opaque)
803 {
804     if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) {
805         ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj);
806         ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj);
807         struct arm_boot_info *info = opaque;
808 
809         if (albifc->arm_linux_init) {
810             albifc->arm_linux_init(albif, info->secure_boot);
811         }
812     }
813     return 0;
814 }
815 
816 static uint64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry,
817                              uint64_t *lowaddr, uint64_t *highaddr,
818                              int elf_machine, AddressSpace *as)
819 {
820     bool elf_is64;
821     union {
822         Elf32_Ehdr h32;
823         Elf64_Ehdr h64;
824     } elf_header;
825     int data_swab = 0;
826     bool big_endian;
827     uint64_t ret = -1;
828     Error *err = NULL;
829 
830 
831     load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err);
832     if (err) {
833         error_free(err);
834         return ret;
835     }
836 
837     if (elf_is64) {
838         big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB;
839         info->endianness = big_endian ? ARM_ENDIANNESS_BE8
840                                       : ARM_ENDIANNESS_LE;
841     } else {
842         big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB;
843         if (big_endian) {
844             if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) {
845                 info->endianness = ARM_ENDIANNESS_BE8;
846             } else {
847                 info->endianness = ARM_ENDIANNESS_BE32;
848                 /* In BE32, the CPU has a different view of the per-byte
849                  * address map than the rest of the system. BE32 ELF files
850                  * are organised such that they can be programmed through
851                  * the CPU's per-word byte-reversed view of the world. QEMU
852                  * however loads ELF files independently of the CPU. So
853                  * tell the ELF loader to byte reverse the data for us.
854                  */
855                 data_swab = 2;
856             }
857         } else {
858             info->endianness = ARM_ENDIANNESS_LE;
859         }
860     }
861 
862     ret = load_elf_as(info->kernel_filename, NULL, NULL,
863                       pentry, lowaddr, highaddr, big_endian, elf_machine,
864                       1, data_swab, as);
865     if (ret <= 0) {
866         /* The header loaded but the image didn't */
867         exit(1);
868     }
869 
870     return ret;
871 }
872 
873 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base,
874                                    hwaddr *entry, AddressSpace *as)
875 {
876     hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR;
877     uint8_t *buffer;
878     int size;
879 
880     /* On aarch64, it's the bootloader's job to uncompress the kernel. */
881     size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES,
882                                      &buffer);
883 
884     if (size < 0) {
885         gsize len;
886 
887         /* Load as raw file otherwise */
888         if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) {
889             return -1;
890         }
891         size = len;
892     }
893 
894     /* check the arm64 magic header value -- very old kernels may not have it */
895     if (size > ARM64_MAGIC_OFFSET + 4 &&
896         memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) {
897         uint64_t hdrvals[2];
898 
899         /* The arm64 Image header has text_offset and image_size fields at 8 and
900          * 16 bytes into the Image header, respectively. The text_offset field
901          * is only valid if the image_size is non-zero.
902          */
903         memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals));
904         if (hdrvals[1] != 0) {
905             kernel_load_offset = le64_to_cpu(hdrvals[0]);
906         }
907     }
908 
909     *entry = mem_base + kernel_load_offset;
910     rom_add_blob_fixed_as(filename, buffer, size, *entry, as);
911 
912     g_free(buffer);
913 
914     return size;
915 }
916 
917 void arm_load_kernel(ARMCPU *cpu, struct arm_boot_info *info)
918 {
919     CPUState *cs;
920     int kernel_size;
921     int initrd_size;
922     int is_linux = 0;
923     uint64_t elf_entry, elf_low_addr, elf_high_addr;
924     int elf_machine;
925     hwaddr entry;
926     static const ARMInsnFixup *primary_loader;
927     AddressSpace *as = arm_boot_address_space(cpu, info);
928 
929     /* The board code is not supposed to set secure_board_setup unless
930      * running its code in secure mode is actually possible, and KVM
931      * doesn't support secure.
932      */
933     assert(!(info->secure_board_setup && kvm_enabled()));
934 
935     info->dtb_filename = qemu_opt_get(qemu_get_machine_opts(), "dtb");
936     info->dtb_limit = 0;
937 
938     /* Load the kernel.  */
939     if (!info->kernel_filename || info->firmware_loaded) {
940 
941         if (have_dtb(info)) {
942             /* If we have a device tree blob, but no kernel to supply it to (or
943              * the kernel is supposed to be loaded by the bootloader), copy the
944              * DTB to the base of RAM for the bootloader to pick up.
945              */
946             info->dtb_start = info->loader_start;
947         }
948 
949         if (info->kernel_filename) {
950             FWCfgState *fw_cfg;
951             bool try_decompressing_kernel;
952 
953             fw_cfg = fw_cfg_find();
954             try_decompressing_kernel = arm_feature(&cpu->env,
955                                                    ARM_FEATURE_AARCH64);
956 
957             /* Expose the kernel, the command line, and the initrd in fw_cfg.
958              * We don't process them here at all, it's all left to the
959              * firmware.
960              */
961             load_image_to_fw_cfg(fw_cfg,
962                                  FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
963                                  info->kernel_filename,
964                                  try_decompressing_kernel);
965             load_image_to_fw_cfg(fw_cfg,
966                                  FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
967                                  info->initrd_filename, false);
968 
969             if (info->kernel_cmdline) {
970                 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
971                                strlen(info->kernel_cmdline) + 1);
972                 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
973                                   info->kernel_cmdline);
974             }
975         }
976 
977         /* We will start from address 0 (typically a boot ROM image) in the
978          * same way as hardware.
979          */
980         return;
981     }
982 
983     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
984         primary_loader = bootloader_aarch64;
985         elf_machine = EM_AARCH64;
986     } else {
987         primary_loader = bootloader;
988         if (!info->write_board_setup) {
989             primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET;
990         }
991         elf_machine = EM_ARM;
992     }
993 
994     if (!info->secondary_cpu_reset_hook) {
995         info->secondary_cpu_reset_hook = default_reset_secondary;
996     }
997     if (!info->write_secondary_boot) {
998         info->write_secondary_boot = default_write_secondary;
999     }
1000 
1001     if (info->nb_cpus == 0)
1002         info->nb_cpus = 1;
1003 
1004     /* We want to put the initrd far enough into RAM that when the
1005      * kernel is uncompressed it will not clobber the initrd. However
1006      * on boards without much RAM we must ensure that we still leave
1007      * enough room for a decent sized initrd, and on boards with large
1008      * amounts of RAM we must avoid the initrd being so far up in RAM
1009      * that it is outside lowmem and inaccessible to the kernel.
1010      * So for boards with less  than 256MB of RAM we put the initrd
1011      * halfway into RAM, and for boards with 256MB of RAM or more we put
1012      * the initrd at 128MB.
1013      */
1014     info->initrd_start = info->loader_start +
1015         MIN(info->ram_size / 2, 128 * 1024 * 1024);
1016 
1017     /* Assume that raw images are linux kernels, and ELF images are not.  */
1018     kernel_size = arm_load_elf(info, &elf_entry, &elf_low_addr,
1019                                &elf_high_addr, elf_machine, as);
1020     if (kernel_size > 0 && have_dtb(info)) {
1021         /* If there is still some room left at the base of RAM, try and put
1022          * the DTB there like we do for images loaded with -bios or -pflash.
1023          */
1024         if (elf_low_addr > info->loader_start
1025             || elf_high_addr < info->loader_start) {
1026             /* Set elf_low_addr as address limit for arm_load_dtb if it may be
1027              * pointing into RAM, otherwise pass '0' (no limit)
1028              */
1029             if (elf_low_addr < info->loader_start) {
1030                 elf_low_addr = 0;
1031             }
1032             info->dtb_start = info->loader_start;
1033             info->dtb_limit = elf_low_addr;
1034         }
1035     }
1036     entry = elf_entry;
1037     if (kernel_size < 0) {
1038         kernel_size = load_uimage_as(info->kernel_filename, &entry, NULL,
1039                                      &is_linux, NULL, NULL, as);
1040     }
1041     if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) {
1042         kernel_size = load_aarch64_image(info->kernel_filename,
1043                                          info->loader_start, &entry, as);
1044         is_linux = 1;
1045     } else if (kernel_size < 0) {
1046         /* 32-bit ARM */
1047         entry = info->loader_start + KERNEL_LOAD_ADDR;
1048         kernel_size = load_image_targphys_as(info->kernel_filename, entry,
1049                                              info->ram_size - KERNEL_LOAD_ADDR,
1050                                              as);
1051         is_linux = 1;
1052     }
1053     if (kernel_size < 0) {
1054         error_report("could not load kernel '%s'", info->kernel_filename);
1055         exit(1);
1056     }
1057     info->entry = entry;
1058     if (is_linux) {
1059         uint32_t fixupcontext[FIXUP_MAX];
1060 
1061         if (info->initrd_filename) {
1062             initrd_size = load_ramdisk_as(info->initrd_filename,
1063                                           info->initrd_start,
1064                                           info->ram_size - info->initrd_start,
1065                                           as);
1066             if (initrd_size < 0) {
1067                 initrd_size = load_image_targphys_as(info->initrd_filename,
1068                                                      info->initrd_start,
1069                                                      info->ram_size -
1070                                                      info->initrd_start,
1071                                                      as);
1072             }
1073             if (initrd_size < 0) {
1074                 error_report("could not load initrd '%s'",
1075                              info->initrd_filename);
1076                 exit(1);
1077             }
1078         } else {
1079             initrd_size = 0;
1080         }
1081         info->initrd_size = initrd_size;
1082 
1083         fixupcontext[FIXUP_BOARDID] = info->board_id;
1084         fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr;
1085 
1086         /* for device tree boot, we pass the DTB directly in r2. Otherwise
1087          * we point to the kernel args.
1088          */
1089         if (have_dtb(info)) {
1090             hwaddr align;
1091 
1092             if (elf_machine == EM_AARCH64) {
1093                 /*
1094                  * Some AArch64 kernels on early bootup map the fdt region as
1095                  *
1096                  *   [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ]
1097                  *
1098                  * Let's play safe and prealign it to 2MB to give us some space.
1099                  */
1100                 align = 2 * 1024 * 1024;
1101             } else {
1102                 /*
1103                  * Some 32bit kernels will trash anything in the 4K page the
1104                  * initrd ends in, so make sure the DTB isn't caught up in that.
1105                  */
1106                 align = 4096;
1107             }
1108 
1109             /* Place the DTB after the initrd in memory with alignment. */
1110             info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size,
1111                                            align);
1112             fixupcontext[FIXUP_ARGPTR] = info->dtb_start;
1113         } else {
1114             fixupcontext[FIXUP_ARGPTR] = info->loader_start + KERNEL_ARGS_ADDR;
1115             if (info->ram_size >= (1ULL << 32)) {
1116                 error_report("RAM size must be less than 4GB to boot"
1117                              " Linux kernel using ATAGS (try passing a device tree"
1118                              " using -dtb)");
1119                 exit(1);
1120             }
1121         }
1122         fixupcontext[FIXUP_ENTRYPOINT] = entry;
1123 
1124         write_bootloader("bootloader", info->loader_start,
1125                          primary_loader, fixupcontext, as);
1126 
1127         if (info->nb_cpus > 1) {
1128             info->write_secondary_boot(cpu, info);
1129         }
1130         if (info->write_board_setup) {
1131             info->write_board_setup(cpu, info);
1132         }
1133 
1134         /* Notify devices which need to fake up firmware initialization
1135          * that we're doing a direct kernel boot.
1136          */
1137         object_child_foreach_recursive(object_get_root(),
1138                                        do_arm_linux_init, info);
1139     }
1140     info->is_linux = is_linux;
1141 
1142     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1143         ARM_CPU(cs)->env.boot_info = info;
1144     }
1145 
1146     /* CPU objects (unlike devices) are not automatically reset on system
1147      * reset, so we must always register a handler to do so. If we're
1148      * actually loading a kernel, the handler is also responsible for
1149      * arranging that we start it correctly.
1150      */
1151     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
1152         qemu_register_reset(do_cpu_reset, ARM_CPU(cs));
1153     }
1154 
1155     if (!info->skip_dtb_autoload && have_dtb(info)) {
1156         if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as) < 0) {
1157             exit(1);
1158         }
1159     }
1160 }
1161 
1162 static const TypeInfo arm_linux_boot_if_info = {
1163     .name = TYPE_ARM_LINUX_BOOT_IF,
1164     .parent = TYPE_INTERFACE,
1165     .class_size = sizeof(ARMLinuxBootIfClass),
1166 };
1167 
1168 static void arm_linux_boot_register_types(void)
1169 {
1170     type_register_static(&arm_linux_boot_if_info);
1171 }
1172 
1173 type_init(arm_linux_boot_register_types)
1174