master
c 1,120 lines 36.8 KB
Raw
1 /*
2 * Copyright (c) 2003-2004 Fabrice Bellard
3 * Copyright (c) 2019, 2024 Red Hat, Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a copy
6 * of this software and associated documentation files (the "Software"), to deal
7 * in the Software without restriction, including without limitation the rights
8 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9 * copies of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
21 * THE SOFTWARE.
22 */
23 #include "qemu/osdep.h"
24 #include "qemu/error-report.h"
25 #include "qemu/cutils.h"
26 #include "qemu/units.h"
27 #include "qemu/datadir.h"
28 #include "qapi/error.h"
29 #include "system/numa.h"
30 #include "system/system.h"
31 #include "system/xen.h"
32 #include "trace.h"
33
34 #include "hw/i386/x86.h"
35 #include "target/i386/cpu.h"
36 #include "hw/rtc/mc146818rtc.h"
37 #include "target/i386/sev.h"
38
39 #include "hw/core/irq.h"
40 #include "hw/core/loader.h"
41 #include "multiboot.h"
42 #include "elf.h"
43 #include "standard-headers/asm-x86/bootparam.h"
44 #include CONFIG_DEVICES
45 #include "kvm/kvm_i386.h"
46 #include "kvm/tdx.h"
47
48 #ifdef CONFIG_XEN_EMU
49 #include "hw/xen/xen.h"
50 #include "hw/i386/kvm/xen_evtchn.h"
51 #endif
52
53 /* Physical Address of PVH entry point read from kernel ELF NOTE */
54 static size_t pvh_start_addr;
55
56 static void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp)
57 {
58 Object *cpu = object_new(MACHINE(x86ms)->cpu_type);
59
60 if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) {
61 goto out;
62 }
63 qdev_realize(DEVICE(cpu), NULL, errp);
64
65 out:
66 object_unref(cpu);
67 }
68
69 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version)
70 {
71 int i;
72 const CPUArchIdList *possible_cpus;
73 MachineState *ms = MACHINE(x86ms);
74 MachineClass *mc = MACHINE_GET_CLASS(x86ms);
75
76 x86_cpu_set_default_version(default_cpu_version);
77
78 /*
79 * Calculates the limit to CPU APIC ID values
80 *
81 * Limit for the APIC ID value, so that all
82 * CPU APIC IDs are < x86ms->apic_id_limit.
83 *
84 * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
85 */
86 x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms,
87 ms->smp.max_cpus - 1) + 1;
88
89 /*
90 * Can we support APIC ID 255 or higher? With KVM, that requires
91 * both in-kernel lapic and X2APIC userspace API.
92 *
93 * kvm_enabled() must go first to ensure that kvm_* references are
94 * not emitted for the linker to consume (kvm_enabled() is
95 * a literal `0` in configurations where kvm_* aren't defined)
96 */
97 if (kvm_enabled() && x86ms->apic_id_limit > 255 &&
98 kvm_irqchip_in_kernel() && !kvm_enable_x2apic()) {
99 error_report("current -smp configuration requires kernel "
100 "irqchip and X2APIC API support.");
101 exit(EXIT_FAILURE);
102 }
103
104 if (kvm_enabled()) {
105 kvm_set_max_apic_id(x86ms->apic_id_limit);
106 }
107
108 if (!kvm_irqchip_in_kernel()) {
109 apic_set_max_apic_id(x86ms->apic_id_limit);
110 }
111
112 possible_cpus = mc->possible_cpu_arch_ids(ms);
113 for (i = 0; i < ms->smp.cpus; i++) {
114 x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal);
115 }
116 }
117
118 void x86_rtc_set_cpus_count(ISADevice *s, uint16_t cpus_count)
119 {
120 MC146818RtcState *rtc = MC146818_RTC(s);
121
122 if (cpus_count > 0xff) {
123 /*
124 * If the number of CPUs can't be represented in 8 bits, the
125 * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just
126 * to make old BIOSes fail more predictably.
127 */
128 mc146818rtc_set_cmos_data(rtc, 0x5f, 0);
129 } else {
130 mc146818rtc_set_cmos_data(rtc, 0x5f, cpus_count - 1);
131 }
132 }
133
134 static int x86_apic_cmp(const void *a, const void *b)
135 {
136 CPUArchId *apic_a = (CPUArchId *)a;
137 CPUArchId *apic_b = (CPUArchId *)b;
138
139 return apic_a->arch_id - apic_b->arch_id;
140 }
141
142 /*
143 * returns pointer to CPUArchId descriptor that matches CPU's apic_id
144 * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no
145 * entry corresponding to CPU's apic_id returns NULL.
146 */
147 static CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
148 {
149 CPUArchId apic_id, *found_cpu;
150
151 apic_id.arch_id = id;
152 found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus,
153 ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus),
154 x86_apic_cmp);
155 if (found_cpu && idx) {
156 *idx = found_cpu - ms->possible_cpus->cpus;
157 }
158 return found_cpu;
159 }
160
161 void x86_cpu_plug(HotplugHandler *hotplug_dev,
162 DeviceState *dev, Error **errp)
163 {
164 CPUArchId *found_cpu;
165 Error *local_err = NULL;
166 X86CPU *cpu = X86_CPU(dev);
167 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
168
169 if (x86ms->acpi_dev) {
170 hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err);
171 if (local_err) {
172 goto out;
173 }
174 }
175
176 /* increment the number of CPUs */
177 x86ms->boot_cpus++;
178 if (x86ms->rtc) {
179 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
180 }
181 if (x86ms->fw_cfg) {
182 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
183 }
184
185 /*
186 * Non-hotplugged CPUs get their SMM cpu address space initialized in
187 * machine init done notifier: register_smram_listener().
188 *
189 * We need initialize the SMM cpu address space for the hotplugged CPU
190 * specifically.
191 */
192 if (kvm_enabled() && dev->hotplugged && x86_machine_is_smm_enabled(x86ms)) {
193 kvm_smm_cpu_address_space_init(cpu);
194 }
195
196 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
197 found_cpu->cpu = CPU(dev);
198 out:
199 error_propagate(errp, local_err);
200 }
201
202 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev,
203 DeviceState *dev, Error **errp)
204 {
205 int idx = -1;
206 X86CPU *cpu = X86_CPU(dev);
207 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
208
209 if (!x86ms->acpi_dev) {
210 error_setg(errp, "CPU hot unplug not supported without ACPI");
211 return;
212 }
213
214 x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
215 assert(idx != -1);
216 if (idx == 0) {
217 error_setg(errp, "Boot CPU is unpluggable");
218 return;
219 }
220
221 hotplug_handler_unplug_request(x86ms->acpi_dev, dev,
222 errp);
223 }
224
225 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev,
226 DeviceState *dev, Error **errp)
227 {
228 CPUArchId *found_cpu;
229 Error *local_err = NULL;
230 X86CPU *cpu = X86_CPU(dev);
231 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
232
233 hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err);
234 if (local_err) {
235 goto out;
236 }
237
238 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
239 found_cpu->cpu = NULL;
240 qdev_unrealize(dev);
241
242 /* decrement the number of CPUs */
243 x86ms->boot_cpus--;
244 /* Update the number of CPUs in CMOS */
245 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
246 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
247 out:
248 error_propagate(errp, local_err);
249 }
250
251 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev,
252 DeviceState *dev, Error **errp)
253 {
254 int idx;
255 CPUState *cs;
256 CPUArchId *cpu_slot;
257 X86CPUTopoIDs topo_ids;
258 X86CPU *cpu = X86_CPU(dev);
259 CPUX86State *env = &cpu->env;
260 MachineState *ms = MACHINE(hotplug_dev);
261 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
262 X86CPUTopoInfo *topo_info = &env->topo_info;
263
264 if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) {
265 error_setg(errp, "Invalid CPU type, expected cpu type: '%s'",
266 ms->cpu_type);
267 return;
268 }
269
270 if (x86ms->acpi_dev) {
271 Error *local_err = NULL;
272
273 hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev,
274 &local_err);
275 if (local_err) {
276 error_propagate(errp, local_err);
277 return;
278 }
279 }
280
281 init_topo_info(topo_info, x86ms);
282
283 if (ms->smp.modules > 1) {
284 set_bit(CPU_TOPOLOGY_LEVEL_MODULE, env->avail_cpu_topo);
285 }
286
287 if (ms->smp.dies > 1) {
288 set_bit(CPU_TOPOLOGY_LEVEL_DIE, env->avail_cpu_topo);
289 }
290
291 /*
292 * If APIC ID is not set,
293 * set it based on socket/die/module/core/thread properties.
294 */
295 if (cpu->apic_id == UNASSIGNED_APIC_ID) {
296 /*
297 * die-id was optional in QEMU 4.0 and older, so keep it optional
298 * if there's only one die per socket.
299 */
300 if (cpu->die_id < 0 && ms->smp.dies == 1) {
301 cpu->die_id = 0;
302 }
303
304 /*
305 * module-id was optional in QEMU 9.0 and older, so keep it optional
306 * if there's only one module per die.
307 */
308 if (cpu->module_id < 0 && ms->smp.modules == 1) {
309 cpu->module_id = 0;
310 }
311
312 if (cpu->socket_id < 0) {
313 error_setg(errp, "CPU socket-id is not set");
314 return;
315 } else if (cpu->socket_id > ms->smp.sockets - 1) {
316 error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u",
317 cpu->socket_id, ms->smp.sockets - 1);
318 return;
319 }
320 if (cpu->die_id < 0) {
321 error_setg(errp, "CPU die-id is not set");
322 return;
323 } else if (cpu->die_id > ms->smp.dies - 1) {
324 error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u",
325 cpu->die_id, ms->smp.dies - 1);
326 return;
327 }
328 if (cpu->module_id < 0) {
329 error_setg(errp, "CPU module-id is not set");
330 return;
331 } else if (cpu->module_id > ms->smp.modules - 1) {
332 error_setg(errp, "Invalid CPU module-id: %u must be in range 0:%u",
333 cpu->module_id, ms->smp.modules - 1);
334 return;
335 }
336 if (cpu->core_id < 0) {
337 error_setg(errp, "CPU core-id is not set");
338 return;
339 } else if (cpu->core_id > (ms->smp.cores - 1)) {
340 error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u",
341 cpu->core_id, ms->smp.cores - 1);
342 return;
343 }
344 if (cpu->thread_id < 0) {
345 error_setg(errp, "CPU thread-id is not set");
346 return;
347 } else if (cpu->thread_id > (ms->smp.threads - 1)) {
348 error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u",
349 cpu->thread_id, ms->smp.threads - 1);
350 return;
351 }
352
353 topo_ids.pkg_id = cpu->socket_id;
354 topo_ids.die_id = cpu->die_id;
355 topo_ids.module_id = cpu->module_id;
356 topo_ids.core_id = cpu->core_id;
357 topo_ids.smt_id = cpu->thread_id;
358 cpu->apic_id = x86_apicid_from_topo_ids(topo_info, &topo_ids);
359 }
360
361 cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
362 if (!cpu_slot) {
363 x86_topo_ids_from_apicid(cpu->apic_id, topo_info, &topo_ids);
364
365 error_setg(errp,
366 "Invalid CPU [socket: %u, die: %u, module: %u, core: %u, thread: %u]"
367 " with APIC ID %" PRIu32 ", valid index range 0:%d",
368 topo_ids.pkg_id, topo_ids.die_id, topo_ids.module_id,
369 topo_ids.core_id, topo_ids.smt_id, cpu->apic_id,
370 ms->possible_cpus->len - 1);
371 return;
372 }
373
374 if (cpu_slot->cpu) {
375 error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists",
376 idx, cpu->apic_id);
377 return;
378 }
379
380 /* if 'address' properties socket-id/core-id/thread-id are not set, set them
381 * so that machine_query_hotpluggable_cpus would show correct values
382 */
383 /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
384 * once -smp refactoring is complete and there will be CPU private
385 * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
386 x86_topo_ids_from_apicid(cpu->apic_id, topo_info, &topo_ids);
387 if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) {
388 error_setg(errp, "property socket-id: %u doesn't match set apic-id:"
389 " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id,
390 topo_ids.pkg_id);
391 return;
392 }
393 cpu->socket_id = topo_ids.pkg_id;
394
395 if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) {
396 error_setg(errp, "property die-id: %u doesn't match set apic-id:"
397 " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id);
398 return;
399 }
400 cpu->die_id = topo_ids.die_id;
401
402 if (cpu->module_id != -1 && cpu->module_id != topo_ids.module_id) {
403 error_setg(errp, "property module-id: %u doesn't match set apic-id:"
404 " 0x%x (module-id: %u)", cpu->module_id, cpu->apic_id,
405 topo_ids.module_id);
406 return;
407 }
408 cpu->module_id = topo_ids.module_id;
409
410 if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) {
411 error_setg(errp, "property core-id: %u doesn't match set apic-id:"
412 " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id,
413 topo_ids.core_id);
414 return;
415 }
416 cpu->core_id = topo_ids.core_id;
417
418 if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) {
419 error_setg(errp, "property thread-id: %u doesn't match set apic-id:"
420 " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id,
421 topo_ids.smt_id);
422 return;
423 }
424 cpu->thread_id = topo_ids.smt_id;
425
426 /*
427 * kvm_enabled() must go first to ensure that kvm_* references are
428 * not emitted for the linker to consume (kvm_enabled() is
429 * a literal `0` in configurations where kvm_* aren't defined)
430 */
431 if (kvm_enabled() && hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) &&
432 !kvm_hv_vpindex_settable()) {
433 error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX");
434 return;
435 }
436
437 cs = CPU(cpu);
438 cs->cpu_index = idx;
439
440 numa_cpu_pre_plug(cpu_slot, dev, errp);
441 }
442
443 static long get_file_size(FILE *f)
444 {
445 long where, size;
446
447 /* XXX: on Unix systems, using fstat() probably makes more sense */
448
449 where = ftell(f);
450 fseek(f, 0, SEEK_END);
451 size = ftell(f);
452 fseek(f, where, SEEK_SET);
453
454 return size;
455 }
456
457 void gsi_handler(void *opaque, int n, int level)
458 {
459 GSIState *s = opaque;
460 bool bypass_ioapic = false;
461
462 trace_x86_gsi_interrupt(n, level);
463
464 #ifdef CONFIG_XEN_EMU
465 /*
466 * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
467 * routing actually works properly under Xen). And then to
468 * *either* the PIRQ handling or the I/OAPIC depending on whether
469 * the former wants it.
470 *
471 * Additionally, this hook allows the Xen event channel GSI to
472 * work around QEMU's lack of support for shared level interrupts,
473 * by keeping track of the externally driven state of the pin and
474 * implementing a logical OR with the state of the evtchn GSI.
475 */
476 if (xen_mode == XEN_EMULATE) {
477 bypass_ioapic = xen_evtchn_set_gsi(n, &level);
478 }
479 #endif
480
481 switch (n) {
482 case 0 ... ISA_NUM_IRQS - 1:
483 if (s->i8259_irq[n]) {
484 /* Under KVM, Kernel will forward to both PIC and IOAPIC */
485 qemu_set_irq(s->i8259_irq[n], level);
486 }
487 /* fall through */
488 case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
489 if (!bypass_ioapic) {
490 qemu_set_irq(s->ioapic_irq[n], level);
491 }
492 break;
493 case IO_APIC_SECONDARY_IRQBASE
494 ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1:
495 qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level);
496 break;
497 }
498 }
499
500 void ioapic_init_gsi(GSIState *gsi_state, Object *parent)
501 {
502 DeviceState *dev;
503 SysBusDevice *d;
504 unsigned int i;
505
506 assert(parent);
507 if (kvm_ioapic_in_kernel()) {
508 dev = qdev_new(TYPE_KVM_IOAPIC);
509 } else {
510 dev = qdev_new(TYPE_IOAPIC);
511 }
512 object_property_add_child(parent, "ioapic", OBJECT(dev));
513 d = SYS_BUS_DEVICE(dev);
514 sysbus_realize_and_unref(d, &error_fatal);
515 sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS);
516
517 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
518 gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i);
519 }
520 }
521
522 DeviceState *ioapic_init_secondary(GSIState *gsi_state)
523 {
524 DeviceState *dev;
525 SysBusDevice *d;
526 unsigned int i;
527
528 dev = qdev_new(TYPE_IOAPIC);
529 d = SYS_BUS_DEVICE(dev);
530 sysbus_realize_and_unref(d, &error_fatal);
531 sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS);
532
533 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
534 gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i);
535 }
536 return dev;
537 }
538
539 /*
540 * The entry point into the kernel for PVH boot is different from
541 * the native entry point. The PVH entry is defined by the x86/HVM
542 * direct boot ABI and is available in an ELFNOTE in the kernel binary.
543 *
544 * This function is passed to load_elf() when it is called from
545 * load_elfboot() which then additionally checks for an ELF Note of
546 * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
547 * parse the PVH entry address from the ELF Note.
548 *
549 * Due to trickery in elf_opts.h, load_elf() is actually available as
550 * load_elf32() or load_elf64() and this routine needs to be able
551 * to deal with being called as 32 or 64 bit.
552 *
553 * The address of the PVH entry point is saved to the 'pvh_start_addr'
554 * global variable. (although the entry point is 32-bit, the kernel
555 * binary can be either 32-bit or 64-bit).
556 */
557 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64)
558 {
559 size_t *elf_note_data_addr;
560
561 /* Check if ELF Note header passed in is valid */
562 if (arg1 == NULL) {
563 return 0;
564 }
565
566 if (is64) {
567 struct elf64_note *nhdr64 = (struct elf64_note *)arg1;
568 uint64_t nhdr_size64 = sizeof(struct elf64_note);
569 uint64_t phdr_align = *(uint64_t *)arg2;
570 uint64_t nhdr_namesz = nhdr64->n_namesz;
571
572 elf_note_data_addr =
573 ((void *)nhdr64) + nhdr_size64 +
574 QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
575
576 pvh_start_addr = *elf_note_data_addr;
577 } else {
578 struct elf32_note *nhdr32 = (struct elf32_note *)arg1;
579 uint32_t nhdr_size32 = sizeof(struct elf32_note);
580 uint32_t phdr_align = *(uint32_t *)arg2;
581 uint32_t nhdr_namesz = nhdr32->n_namesz;
582
583 elf_note_data_addr =
584 ((void *)nhdr32) + nhdr_size32 +
585 QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
586
587 pvh_start_addr = *(uint32_t *)elf_note_data_addr;
588 }
589
590 return pvh_start_addr;
591 }
592
593 static bool load_elfboot(const char *kernel_filename,
594 int kernel_file_size,
595 uint8_t *header,
596 size_t pvh_xen_start_addr,
597 FWCfgState *fw_cfg)
598 {
599 uint32_t flags = 0;
600 uint32_t mh_load_addr = 0;
601 uint32_t elf_kernel_size = 0;
602 uint64_t elf_entry;
603 uint64_t elf_low, elf_high;
604 int kernel_size;
605
606 if (ldl_le_p(header) != 0x464c457f) {
607 return false; /* no elfboot */
608 }
609
610 bool elf_is64 = header[EI_CLASS] == ELFCLASS64;
611 flags = elf_is64 ?
612 ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags;
613
614 if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
615 error_report("elfboot unsupported flags = %x", flags);
616 exit(1);
617 }
618
619 uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY;
620 kernel_size = load_elf(kernel_filename, read_pvh_start_addr,
621 NULL, &elf_note_type, &elf_entry,
622 &elf_low, &elf_high, NULL,
623 ELFDATA2LSB, I386_ELF_MACHINE, 0, 0);
624
625 if (kernel_size < 0) {
626 error_report("Error while loading elf kernel");
627 exit(1);
628 }
629 mh_load_addr = elf_low;
630 elf_kernel_size = elf_high - elf_low;
631
632 if (pvh_start_addr == 0) {
633 error_report("Error loading uncompressed kernel without PVH ELF Note");
634 exit(1);
635 }
636 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr);
637 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr);
638 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size);
639
640 return true;
641 }
642
643 void x86_load_linux(X86MachineState *x86ms,
644 FWCfgState *fw_cfg,
645 int acpi_data_size)
646 {
647 uint16_t protocol;
648 int setup_size, kernel_size, cmdline_size;
649 int dtb_size, setup_data_offset;
650 uint32_t initrd_max;
651 uint8_t header[8192], *setup, *kernel;
652 hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0;
653 FILE *f;
654 const char *vmode;
655 MachineState *machine = MACHINE(x86ms);
656 struct setup_data *setup_data;
657 const char *kernel_filename = machine->kernel_filename;
658 const char *initrd_filename = machine->initrd_filename;
659 const char *dtb_filename = machine->dtb;
660 const char *kernel_cmdline = machine->kernel_cmdline;
661 SevKernelLoaderContext sev_load_ctx = {};
662
663 /* Align to 16 bytes as a paranoia measure */
664 cmdline_size = (strlen(kernel_cmdline) + 16) & ~15;
665
666 /* load the kernel header */
667 f = fopen(kernel_filename, "rb");
668 if (!f) {
669 fprintf(stderr, "qemu: could not open kernel file '%s': %s\n",
670 kernel_filename, strerror(errno));
671 exit(1);
672 }
673
674 kernel_size = get_file_size(f);
675 if (!kernel_size ||
676 fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
677 MIN(ARRAY_SIZE(header), kernel_size)) {
678 fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
679 kernel_filename, strerror(errno));
680 exit(1);
681 }
682
683 /*
684 * kernel protocol version.
685 * Please see https://www.kernel.org/doc/Documentation/x86/boot.txt
686 */
687 if (ldl_le_p(header + 0x202) == 0x53726448) /* Magic signature "HdrS" */ {
688 protocol = lduw_le_p(header + 0x206);
689 } else {
690 /*
691 * This could be a multiboot kernel. If it is, let's stop treating it
692 * like a Linux kernel.
693 * Note: some multiboot images could be in the ELF format (the same of
694 * PVH), so we try multiboot first since we check the multiboot magic
695 * header before to load it.
696 */
697 if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename,
698 kernel_cmdline, kernel_size, header)) {
699 return;
700 }
701 /*
702 * Check if the file is an uncompressed kernel file (ELF) and load it,
703 * saving the PVH entry point used by the x86/HVM direct boot ABI.
704 * If load_elfboot() is successful, populate the fw_cfg info.
705 */
706 if (load_elfboot(kernel_filename, kernel_size,
707 header, pvh_start_addr, fw_cfg)) {
708 fclose(f);
709
710 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
711 strlen(kernel_cmdline) + 1);
712 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
713
714 setup = g_memdup2(header, sizeof(header));
715
716 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header));
717 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA,
718 setup, sizeof(header));
719
720 /* load initrd */
721 if (initrd_filename) {
722 GMappedFile *mapped_file;
723 gsize initrd_size;
724 gchar *initrd_data;
725 GError *gerr = NULL;
726
727 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
728 if (!mapped_file) {
729 fprintf(stderr, "qemu: error reading initrd %s: %s\n",
730 initrd_filename, gerr->message);
731 exit(1);
732 }
733 x86ms->initrd_mapped_file = mapped_file;
734
735 initrd_data = g_mapped_file_get_contents(mapped_file);
736 initrd_size = g_mapped_file_get_length(mapped_file);
737 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
738 if (initrd_size >= initrd_max) {
739 fprintf(stderr, "qemu: initrd is too large, cannot support."
740 "(max: %"PRIu32", need %"PRId64")\n",
741 initrd_max, (uint64_t)initrd_size);
742 exit(1);
743 }
744
745 initrd_addr = (initrd_max - initrd_size) & ~4095;
746
747 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
748 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
749 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data,
750 initrd_size);
751 }
752
753 option_rom[nb_option_roms].bootindex = 0;
754 option_rom[nb_option_roms].name = "pvh.bin";
755 nb_option_roms++;
756
757 return;
758 }
759 protocol = 0;
760 }
761
762 if (protocol < 0x200 || !(header[0x211] & 0x01)) {
763 /* Low kernel */
764 real_addr = 0x90000;
765 cmdline_addr = 0x9a000 - cmdline_size;
766 prot_addr = 0x10000;
767 } else if (protocol < 0x202) {
768 /* High but ancient kernel */
769 real_addr = 0x90000;
770 cmdline_addr = 0x9a000 - cmdline_size;
771 prot_addr = 0x100000;
772 } else {
773 /* High and recent kernel */
774 real_addr = 0x10000;
775 cmdline_addr = 0x20000;
776 prot_addr = 0x100000;
777 }
778
779 /* highest address for loading the initrd */
780 if (protocol >= 0x20c &&
781 lduw_le_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) {
782 /*
783 * Linux has supported initrd up to 4 GB for a very long time (2007,
784 * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
785 * though it only sets initrd_max to 2 GB to "work around bootloader
786 * bugs". Luckily, QEMU firmware(which does something like bootloader)
787 * has supported this.
788 *
789 * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
790 * be loaded into any address.
791 *
792 * In addition, initrd_max is uint32_t simply because QEMU doesn't
793 * support the 64-bit boot protocol (specifically the ext_ramdisk_image
794 * field).
795 *
796 * Therefore here just limit initrd_max to UINT32_MAX simply as well.
797 */
798 initrd_max = UINT32_MAX;
799 } else if (protocol >= 0x203) {
800 initrd_max = ldl_le_p(header + 0x22c);
801 } else {
802 initrd_max = 0x37ffffff;
803 }
804
805 if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) {
806 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
807 }
808
809 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
810 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline) + 1);
811 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
812 sev_load_ctx.cmdline_data = (char *)kernel_cmdline;
813 sev_load_ctx.cmdline_size = strlen(kernel_cmdline) + 1;
814
815 if (protocol >= 0x202) {
816 stl_le_p(header + 0x228, cmdline_addr);
817 } else {
818 stw_le_p(header + 0x20, 0xA33F);
819 stw_le_p(header + 0x22, cmdline_addr - real_addr);
820 }
821
822 /* handle vga= parameter */
823 vmode = strstr(kernel_cmdline, "vga=");
824 if (vmode) {
825 unsigned int video_mode;
826 const char *end;
827 int ret;
828 /* skip "vga=" */
829 vmode += 4;
830 if (!strncmp(vmode, "normal", 6)) {
831 video_mode = 0xffff;
832 } else if (!strncmp(vmode, "ext", 3)) {
833 video_mode = 0xfffe;
834 } else if (!strncmp(vmode, "ask", 3)) {
835 video_mode = 0xfffd;
836 } else {
837 ret = qemu_strtoui(vmode, &end, 0, &video_mode);
838 if (ret != 0 || (*end && *end != ' ')) {
839 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n");
840 exit(1);
841 }
842 }
843 stw_le_p(header + 0x1fa, video_mode);
844 }
845
846 /* loader type */
847 /*
848 * High nybble = B reserved for QEMU; low nybble is revision number.
849 * If this code is substantially changed, you may want to consider
850 * incrementing the revision.
851 */
852 if (protocol >= 0x200) {
853 header[0x210] = 0xB0;
854 }
855 /* heap */
856 if (protocol >= 0x201) {
857 header[0x211] |= 0x80; /* CAN_USE_HEAP */
858 stw_le_p(header + 0x224, cmdline_addr - real_addr - 0x200);
859 }
860
861 /* load initrd */
862 if (initrd_filename) {
863 GMappedFile *mapped_file;
864 gsize initrd_size;
865 gchar *initrd_data;
866 GError *gerr = NULL;
867
868 if (protocol < 0x200) {
869 fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
870 exit(1);
871 }
872
873 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
874 if (!mapped_file) {
875 fprintf(stderr, "qemu: error reading initrd %s: %s\n",
876 initrd_filename, gerr->message);
877 exit(1);
878 }
879 x86ms->initrd_mapped_file = mapped_file;
880
881 initrd_data = g_mapped_file_get_contents(mapped_file);
882 initrd_size = g_mapped_file_get_length(mapped_file);
883 if (initrd_size >= initrd_max) {
884 fprintf(stderr, "qemu: initrd is too large, cannot support."
885 "(max: %"PRIu32", need %"PRId64")\n",
886 initrd_max, (uint64_t)initrd_size);
887 exit(1);
888 }
889
890 initrd_addr = (initrd_max - initrd_size) & ~4095;
891
892 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
893 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
894 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
895 sev_load_ctx.initrd_data = initrd_data;
896 sev_load_ctx.initrd_size = initrd_size;
897
898 stl_le_p(header + 0x218, initrd_addr);
899 stl_le_p(header + 0x21c, initrd_size);
900 }
901
902 /* load kernel and setup */
903 setup_size = header[0x1f1];
904 if (setup_size == 0) {
905 setup_size = 4;
906 }
907 setup_size = (setup_size + 1) * 512;
908 if (setup_size > kernel_size) {
909 fprintf(stderr, "qemu: invalid kernel header\n");
910 exit(1);
911 }
912
913 setup = g_malloc(setup_size);
914 kernel = g_malloc(kernel_size);
915 fseek(f, 0, SEEK_SET);
916 if (fread(setup, 1, setup_size, f) != setup_size) {
917 fprintf(stderr, "fread() failed\n");
918 exit(1);
919 }
920 fseek(f, 0, SEEK_SET);
921 if (fread(kernel, 1, kernel_size, f) != kernel_size) {
922 fprintf(stderr, "fread() failed\n");
923 exit(1);
924 }
925 fclose(f);
926
927 /* append dtb to kernel */
928 if (dtb_filename) {
929 if (protocol < 0x209) {
930 fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n");
931 exit(1);
932 }
933
934 dtb_size = get_image_size(dtb_filename, NULL);
935 if (dtb_size <= 0) {
936 fprintf(stderr, "qemu: error reading dtb %s: %s\n",
937 dtb_filename, strerror(errno));
938 exit(1);
939 }
940
941 setup_data_offset = QEMU_ALIGN_UP(kernel_size, 16);
942 kernel_size = setup_data_offset + sizeof(struct setup_data) + dtb_size;
943 kernel = g_realloc(kernel, kernel_size);
944
945 stq_le_p(header + 0x250, prot_addr + setup_data_offset);
946
947 setup_data = (struct setup_data *)(kernel + setup_data_offset);
948 setup_data->next = 0;
949 setup_data->type = cpu_to_le32(SETUP_DTB);
950 setup_data->len = cpu_to_le32(dtb_size);
951
952 load_image_size(dtb_filename, setup_data->data, dtb_size);
953 }
954
955 /*
956 * If we're starting an encrypted VM, it will be OVMF based, which uses the
957 * efi stub for booting and doesn't require any values to be placed in the
958 * kernel header. We therefore don't update the header so the hash of the
959 * kernel on the other side of the fw_cfg interface matches the hash of the
960 * file the user passed in.
961 */
962 if (!MACHINE(x86ms)->cgs && protocol > 0) {
963 memcpy(setup, header, MIN(sizeof(header), setup_size));
964 }
965
966 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
967 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size - setup_size);
968 fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA,
969 kernel + setup_size, kernel_size - setup_size);
970 sev_load_ctx.kernel_data = (char *)kernel + setup_size;
971 sev_load_ctx.kernel_size = kernel_size - setup_size;
972
973 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
974 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
975 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
976 sev_load_ctx.setup_data = (char *)setup;
977 sev_load_ctx.setup_size = setup_size;
978
979 /* kernel without setup header patches */
980 fw_cfg_add_file(fw_cfg, "etc/boot/kernel", kernel, kernel_size);
981
982 if (machine->shim_filename) {
983 load_image_to_fw_cfg_file(fw_cfg, "etc/boot/shim",
984 machine->shim_filename);
985 }
986
987 if (sev_enabled()) {
988 sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal);
989 }
990
991 option_rom[nb_option_roms].bootindex = 0;
992 assert(fw_cfg_dma_enabled(fw_cfg));
993 option_rom[nb_option_roms].name = "linuxboot_dma.bin";
994 nb_option_roms++;
995 }
996
997 void x86_isa_bios_init(MemoryRegion *isa_bios, MemoryRegion *isa_memory,
998 MemoryRegion *bios, bool read_only)
999 {
1000 uint64_t bios_size = memory_region_size(bios);
1001 uint64_t isa_bios_size = MIN(bios_size, 128 * KiB);
1002
1003 memory_region_init_alias(isa_bios, NULL, "isa-bios", bios,
1004 bios_size - isa_bios_size, isa_bios_size);
1005 memory_region_add_subregion_overlap(isa_memory, 1 * MiB - isa_bios_size,
1006 isa_bios, 1);
1007 memory_region_set_readonly(isa_bios, read_only);
1008 }
1009
1010 static int get_bios_size(X86MachineState *x86ms,
1011 const char *bios_name, char *filename)
1012 {
1013 int bios_size;
1014
1015 if (filename) {
1016 bios_size = get_image_size(filename, NULL);
1017 } else {
1018 bios_size = -1;
1019 }
1020 if (bios_size <= 0 ||
1021 (bios_size % 65536) != 0) {
1022 goto bios_error;
1023 }
1024
1025 return bios_size;
1026
1027 bios_error:
1028 fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1029 exit(1);
1030 }
1031
1032 static void load_bios_from_file(X86MachineState *x86ms, const char *bios_name,
1033 char *filename, int bios_size,
1034 bool isapc_ram_fw)
1035 {
1036 ssize_t ret;
1037
1038 /* BIOS load */
1039 if (machine_require_guest_memfd(MACHINE(x86ms))) {
1040 memory_region_init_ram_guest_memfd(&x86ms->bios, NULL, "pc.bios",
1041 bios_size, &error_fatal);
1042 if (is_tdx_vm()) {
1043 tdx_set_tdvf_region(&x86ms->bios);
1044 }
1045 } else {
1046 memory_region_init_ram(&x86ms->bios, NULL, "pc.bios",
1047 bios_size, &error_fatal);
1048 }
1049 if (sev_enabled() || is_tdx_vm()) {
1050 /*
1051 * The concept of a "reset" simply doesn't exist for
1052 * confidential computing guests, we have to destroy and
1053 * re-launch them instead. So there is no need to register
1054 * the firmware as rom to properly re-initialize on reset.
1055 * Just go for a straight file load instead.
1056 */
1057 void *ptr = memory_region_get_ram_ptr(&x86ms->bios);
1058 load_image_size(filename, ptr, bios_size);
1059 x86_firmware_configure(0x100000000ULL - bios_size, ptr, bios_size);
1060 } else {
1061 memory_region_set_readonly(&x86ms->bios, !isapc_ram_fw);
1062 ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1);
1063 if (ret != 0) {
1064 goto bios_error;
1065 }
1066 }
1067
1068 return;
1069
1070 bios_error:
1071 fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1072 exit(1);
1073 }
1074
1075 void x86_bios_rom_reload(X86MachineState *x86ms)
1076 {
1077 int bios_size;
1078 const char *bios_name;
1079 char *filename;
1080
1081 if (memory_region_size(&x86ms->bios) == 0) {
1082 /* if -bios is not used */
1083 return;
1084 }
1085
1086 bios_name = MACHINE(x86ms)->firmware ?: "bios.bin";
1087 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1088
1089 bios_size = get_bios_size(x86ms, bios_name, filename);
1090
1091 void *ptr = memory_region_get_ram_ptr(&x86ms->bios);
1092 load_image_size(filename, ptr, bios_size);
1093 x86_firmware_configure(0x100000000ULL - bios_size, ptr, bios_size);
1094 }
1095
1096 void x86_bios_rom_init(X86MachineState *x86ms, const char *default_firmware,
1097 MemoryRegion *rom_memory, bool isapc_ram_fw)
1098 {
1099 int bios_size;
1100 const char *bios_name;
1101 g_autofree char *filename;
1102
1103 bios_name = MACHINE(x86ms)->firmware ?: default_firmware;
1104 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1105
1106 bios_size = get_bios_size(x86ms, bios_name, filename);
1107 load_bios_from_file(x86ms, bios_name, filename, bios_size, isapc_ram_fw);
1108
1109 if (!machine_require_guest_memfd(MACHINE(x86ms))) {
1110 /* map the last 128KB of the BIOS in ISA space */
1111 x86_isa_bios_init(&x86ms->isa_bios, rom_memory, &x86ms->bios,
1112 !isapc_ram_fw);
1113 }
1114
1115 /* map all the bios at the top of memory */
1116 memory_region_add_subregion(rom_memory,
1117 (uint32_t)(-bios_size),
1118 &x86ms->bios);
1119 return;
1120 }