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/acpi/cpu_hotplug.h"
40 #include "hw/irq.h"
41 #include "hw/loader.h"
42 #include "multiboot.h"
43 #include "elf.h"
44 #include "standard-headers/asm-x86/bootparam.h"
45 #include CONFIG_DEVICES
46 #include "kvm/kvm_i386.h"
47 #include "kvm/tdx.h"
48
49 #ifdef CONFIG_XEN_EMU
50 #include "hw/xen/xen.h"
51 #include "hw/i386/kvm/xen_evtchn.h"
52 #endif
53
54 /* Physical Address of PVH entry point read from kernel ELF NOTE */
55 static size_t pvh_start_addr;
56
x86_cpu_new(X86MachineState * x86ms,int64_t apic_id,Error ** errp)57 static void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp)
58 {
59 Object *cpu = object_new(MACHINE(x86ms)->cpu_type);
60
61 if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) {
62 goto out;
63 }
64 qdev_realize(DEVICE(cpu), NULL, errp);
65
66 out:
67 object_unref(cpu);
68 }
69
x86_cpus_init(X86MachineState * x86ms,int default_cpu_version)70 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version)
71 {
72 int i;
73 const CPUArchIdList *possible_cpus;
74 MachineState *ms = MACHINE(x86ms);
75 MachineClass *mc = MACHINE_GET_CLASS(x86ms);
76
77 x86_cpu_set_default_version(default_cpu_version);
78
79 /*
80 * Calculates the limit to CPU APIC ID values
81 *
82 * Limit for the APIC ID value, so that all
83 * CPU APIC IDs are < x86ms->apic_id_limit.
84 *
85 * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
86 */
87 x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms,
88 ms->smp.max_cpus - 1) + 1;
89
90 /*
91 * Can we support APIC ID 255 or higher? With KVM, that requires
92 * both in-kernel lapic and X2APIC userspace API.
93 *
94 * kvm_enabled() must go first to ensure that kvm_* references are
95 * not emitted for the linker to consume (kvm_enabled() is
96 * a literal `0` in configurations where kvm_* aren't defined)
97 */
98 if (kvm_enabled() && x86ms->apic_id_limit > 255 &&
99 kvm_irqchip_in_kernel() && !kvm_enable_x2apic()) {
100 error_report("current -smp configuration requires kernel "
101 "irqchip and X2APIC API support.");
102 exit(EXIT_FAILURE);
103 }
104
105 if (kvm_enabled()) {
106 kvm_set_max_apic_id(x86ms->apic_id_limit);
107 }
108
109 if (!kvm_irqchip_in_kernel()) {
110 apic_set_max_apic_id(x86ms->apic_id_limit);
111 }
112
113 possible_cpus = mc->possible_cpu_arch_ids(ms);
114 for (i = 0; i < ms->smp.cpus; i++) {
115 x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal);
116 }
117 }
118
x86_rtc_set_cpus_count(ISADevice * s,uint16_t cpus_count)119 void x86_rtc_set_cpus_count(ISADevice *s, uint16_t cpus_count)
120 {
121 MC146818RtcState *rtc = MC146818_RTC(s);
122
123 if (cpus_count > 0xff) {
124 /*
125 * If the number of CPUs can't be represented in 8 bits, the
126 * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just
127 * to make old BIOSes fail more predictably.
128 */
129 mc146818rtc_set_cmos_data(rtc, 0x5f, 0);
130 } else {
131 mc146818rtc_set_cmos_data(rtc, 0x5f, cpus_count - 1);
132 }
133 }
134
x86_apic_cmp(const void * a,const void * b)135 static int x86_apic_cmp(const void *a, const void *b)
136 {
137 CPUArchId *apic_a = (CPUArchId *)a;
138 CPUArchId *apic_b = (CPUArchId *)b;
139
140 return apic_a->arch_id - apic_b->arch_id;
141 }
142
143 /*
144 * returns pointer to CPUArchId descriptor that matches CPU's apic_id
145 * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no
146 * entry corresponding to CPU's apic_id returns NULL.
147 */
x86_find_cpu_slot(MachineState * ms,uint32_t id,int * idx)148 static CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
149 {
150 CPUArchId apic_id, *found_cpu;
151
152 apic_id.arch_id = id;
153 found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus,
154 ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus),
155 x86_apic_cmp);
156 if (found_cpu && idx) {
157 *idx = found_cpu - ms->possible_cpus->cpus;
158 }
159 return found_cpu;
160 }
161
x86_cpu_plug(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)162 void x86_cpu_plug(HotplugHandler *hotplug_dev,
163 DeviceState *dev, Error **errp)
164 {
165 CPUArchId *found_cpu;
166 Error *local_err = NULL;
167 X86CPU *cpu = X86_CPU(dev);
168 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
169
170 if (x86ms->acpi_dev) {
171 hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err);
172 if (local_err) {
173 goto out;
174 }
175 }
176
177 /* increment the number of CPUs */
178 x86ms->boot_cpus++;
179 if (x86ms->rtc) {
180 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
181 }
182 if (x86ms->fw_cfg) {
183 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
184 }
185
186 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
187 found_cpu->cpu = CPU(dev);
188 out:
189 error_propagate(errp, local_err);
190 }
191
x86_cpu_unplug_request_cb(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)192 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev,
193 DeviceState *dev, Error **errp)
194 {
195 int idx = -1;
196 X86CPU *cpu = X86_CPU(dev);
197 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
198
199 if (!x86ms->acpi_dev) {
200 error_setg(errp, "CPU hot unplug not supported without ACPI");
201 return;
202 }
203
204 x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
205 assert(idx != -1);
206 if (idx == 0) {
207 error_setg(errp, "Boot CPU is unpluggable");
208 return;
209 }
210
211 hotplug_handler_unplug_request(x86ms->acpi_dev, dev,
212 errp);
213 }
214
x86_cpu_unplug_cb(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)215 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev,
216 DeviceState *dev, Error **errp)
217 {
218 CPUArchId *found_cpu;
219 Error *local_err = NULL;
220 X86CPU *cpu = X86_CPU(dev);
221 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
222
223 hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err);
224 if (local_err) {
225 goto out;
226 }
227
228 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
229 found_cpu->cpu = NULL;
230 qdev_unrealize(dev);
231
232 /* decrement the number of CPUs */
233 x86ms->boot_cpus--;
234 /* Update the number of CPUs in CMOS */
235 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
236 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
237 out:
238 error_propagate(errp, local_err);
239 }
240
x86_cpu_pre_plug(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)241 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev,
242 DeviceState *dev, Error **errp)
243 {
244 int idx;
245 CPUState *cs;
246 CPUArchId *cpu_slot;
247 X86CPUTopoIDs topo_ids;
248 X86CPU *cpu = X86_CPU(dev);
249 CPUX86State *env = &cpu->env;
250 MachineState *ms = MACHINE(hotplug_dev);
251 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
252 X86CPUTopoInfo *topo_info = &env->topo_info;
253
254 if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) {
255 error_setg(errp, "Invalid CPU type, expected cpu type: '%s'",
256 ms->cpu_type);
257 return;
258 }
259
260 if (x86ms->acpi_dev) {
261 Error *local_err = NULL;
262
263 hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev,
264 &local_err);
265 if (local_err) {
266 error_propagate(errp, local_err);
267 return;
268 }
269 }
270
271 init_topo_info(topo_info, x86ms);
272
273 if (ms->smp.modules > 1) {
274 set_bit(CPU_TOPOLOGY_LEVEL_MODULE, env->avail_cpu_topo);
275 }
276
277 if (ms->smp.dies > 1) {
278 set_bit(CPU_TOPOLOGY_LEVEL_DIE, env->avail_cpu_topo);
279 }
280
281 /*
282 * If APIC ID is not set,
283 * set it based on socket/die/module/core/thread properties.
284 */
285 if (cpu->apic_id == UNASSIGNED_APIC_ID) {
286 /*
287 * die-id was optional in QEMU 4.0 and older, so keep it optional
288 * if there's only one die per socket.
289 */
290 if (cpu->die_id < 0 && ms->smp.dies == 1) {
291 cpu->die_id = 0;
292 }
293
294 /*
295 * module-id was optional in QEMU 9.0 and older, so keep it optional
296 * if there's only one module per die.
297 */
298 if (cpu->module_id < 0 && ms->smp.modules == 1) {
299 cpu->module_id = 0;
300 }
301
302 if (cpu->socket_id < 0) {
303 error_setg(errp, "CPU socket-id is not set");
304 return;
305 } else if (cpu->socket_id > ms->smp.sockets - 1) {
306 error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u",
307 cpu->socket_id, ms->smp.sockets - 1);
308 return;
309 }
310 if (cpu->die_id < 0) {
311 error_setg(errp, "CPU die-id is not set");
312 return;
313 } else if (cpu->die_id > ms->smp.dies - 1) {
314 error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u",
315 cpu->die_id, ms->smp.dies - 1);
316 return;
317 }
318 if (cpu->module_id < 0) {
319 error_setg(errp, "CPU module-id is not set");
320 return;
321 } else if (cpu->module_id > ms->smp.modules - 1) {
322 error_setg(errp, "Invalid CPU module-id: %u must be in range 0:%u",
323 cpu->module_id, ms->smp.modules - 1);
324 return;
325 }
326 if (cpu->core_id < 0) {
327 error_setg(errp, "CPU core-id is not set");
328 return;
329 } else if (cpu->core_id > (ms->smp.cores - 1)) {
330 error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u",
331 cpu->core_id, ms->smp.cores - 1);
332 return;
333 }
334 if (cpu->thread_id < 0) {
335 error_setg(errp, "CPU thread-id is not set");
336 return;
337 } else if (cpu->thread_id > (ms->smp.threads - 1)) {
338 error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u",
339 cpu->thread_id, ms->smp.threads - 1);
340 return;
341 }
342
343 topo_ids.pkg_id = cpu->socket_id;
344 topo_ids.die_id = cpu->die_id;
345 topo_ids.module_id = cpu->module_id;
346 topo_ids.core_id = cpu->core_id;
347 topo_ids.smt_id = cpu->thread_id;
348 cpu->apic_id = x86_apicid_from_topo_ids(topo_info, &topo_ids);
349 }
350
351 cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
352 if (!cpu_slot) {
353 x86_topo_ids_from_apicid(cpu->apic_id, topo_info, &topo_ids);
354
355 error_setg(errp,
356 "Invalid CPU [socket: %u, die: %u, module: %u, core: %u, thread: %u]"
357 " with APIC ID %" PRIu32 ", valid index range 0:%d",
358 topo_ids.pkg_id, topo_ids.die_id, topo_ids.module_id,
359 topo_ids.core_id, topo_ids.smt_id, cpu->apic_id,
360 ms->possible_cpus->len - 1);
361 return;
362 }
363
364 if (cpu_slot->cpu) {
365 error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists",
366 idx, cpu->apic_id);
367 return;
368 }
369
370 /* if 'address' properties socket-id/core-id/thread-id are not set, set them
371 * so that machine_query_hotpluggable_cpus would show correct values
372 */
373 /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
374 * once -smp refactoring is complete and there will be CPU private
375 * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
376 x86_topo_ids_from_apicid(cpu->apic_id, topo_info, &topo_ids);
377 if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) {
378 error_setg(errp, "property socket-id: %u doesn't match set apic-id:"
379 " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id,
380 topo_ids.pkg_id);
381 return;
382 }
383 cpu->socket_id = topo_ids.pkg_id;
384
385 if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) {
386 error_setg(errp, "property die-id: %u doesn't match set apic-id:"
387 " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id);
388 return;
389 }
390 cpu->die_id = topo_ids.die_id;
391
392 if (cpu->module_id != -1 && cpu->module_id != topo_ids.module_id) {
393 error_setg(errp, "property module-id: %u doesn't match set apic-id:"
394 " 0x%x (module-id: %u)", cpu->module_id, cpu->apic_id,
395 topo_ids.module_id);
396 return;
397 }
398 cpu->module_id = topo_ids.module_id;
399
400 if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) {
401 error_setg(errp, "property core-id: %u doesn't match set apic-id:"
402 " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id,
403 topo_ids.core_id);
404 return;
405 }
406 cpu->core_id = topo_ids.core_id;
407
408 if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) {
409 error_setg(errp, "property thread-id: %u doesn't match set apic-id:"
410 " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id,
411 topo_ids.smt_id);
412 return;
413 }
414 cpu->thread_id = topo_ids.smt_id;
415
416 /*
417 * kvm_enabled() must go first to ensure that kvm_* references are
418 * not emitted for the linker to consume (kvm_enabled() is
419 * a literal `0` in configurations where kvm_* aren't defined)
420 */
421 if (kvm_enabled() && hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) &&
422 !kvm_hv_vpindex_settable()) {
423 error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX");
424 return;
425 }
426
427 cs = CPU(cpu);
428 cs->cpu_index = idx;
429
430 numa_cpu_pre_plug(cpu_slot, dev, errp);
431 }
432
get_file_size(FILE * f)433 static long get_file_size(FILE *f)
434 {
435 long where, size;
436
437 /* XXX: on Unix systems, using fstat() probably makes more sense */
438
439 where = ftell(f);
440 fseek(f, 0, SEEK_END);
441 size = ftell(f);
442 fseek(f, where, SEEK_SET);
443
444 return size;
445 }
446
gsi_handler(void * opaque,int n,int level)447 void gsi_handler(void *opaque, int n, int level)
448 {
449 GSIState *s = opaque;
450 bool bypass_ioapic = false;
451
452 trace_x86_gsi_interrupt(n, level);
453
454 #ifdef CONFIG_XEN_EMU
455 /*
456 * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
457 * routing actually works properly under Xen). And then to
458 * *either* the PIRQ handling or the I/OAPIC depending on whether
459 * the former wants it.
460 *
461 * Additionally, this hook allows the Xen event channel GSI to
462 * work around QEMU's lack of support for shared level interrupts,
463 * by keeping track of the externally driven state of the pin and
464 * implementing a logical OR with the state of the evtchn GSI.
465 */
466 if (xen_mode == XEN_EMULATE) {
467 bypass_ioapic = xen_evtchn_set_gsi(n, &level);
468 }
469 #endif
470
471 switch (n) {
472 case 0 ... ISA_NUM_IRQS - 1:
473 if (s->i8259_irq[n]) {
474 /* Under KVM, Kernel will forward to both PIC and IOAPIC */
475 qemu_set_irq(s->i8259_irq[n], level);
476 }
477 /* fall through */
478 case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
479 if (!bypass_ioapic) {
480 qemu_set_irq(s->ioapic_irq[n], level);
481 }
482 break;
483 case IO_APIC_SECONDARY_IRQBASE
484 ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1:
485 qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level);
486 break;
487 }
488 }
489
ioapic_init_gsi(GSIState * gsi_state,Object * parent)490 void ioapic_init_gsi(GSIState *gsi_state, Object *parent)
491 {
492 DeviceState *dev;
493 SysBusDevice *d;
494 unsigned int i;
495
496 assert(parent);
497 if (kvm_ioapic_in_kernel()) {
498 dev = qdev_new(TYPE_KVM_IOAPIC);
499 } else {
500 dev = qdev_new(TYPE_IOAPIC);
501 }
502 object_property_add_child(parent, "ioapic", OBJECT(dev));
503 d = SYS_BUS_DEVICE(dev);
504 sysbus_realize_and_unref(d, &error_fatal);
505 sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS);
506
507 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
508 gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i);
509 }
510 }
511
ioapic_init_secondary(GSIState * gsi_state)512 DeviceState *ioapic_init_secondary(GSIState *gsi_state)
513 {
514 DeviceState *dev;
515 SysBusDevice *d;
516 unsigned int i;
517
518 dev = qdev_new(TYPE_IOAPIC);
519 d = SYS_BUS_DEVICE(dev);
520 sysbus_realize_and_unref(d, &error_fatal);
521 sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS);
522
523 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
524 gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i);
525 }
526 return dev;
527 }
528
529 /*
530 * The entry point into the kernel for PVH boot is different from
531 * the native entry point. The PVH entry is defined by the x86/HVM
532 * direct boot ABI and is available in an ELFNOTE in the kernel binary.
533 *
534 * This function is passed to load_elf() when it is called from
535 * load_elfboot() which then additionally checks for an ELF Note of
536 * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
537 * parse the PVH entry address from the ELF Note.
538 *
539 * Due to trickery in elf_opts.h, load_elf() is actually available as
540 * load_elf32() or load_elf64() and this routine needs to be able
541 * to deal with being called as 32 or 64 bit.
542 *
543 * The address of the PVH entry point is saved to the 'pvh_start_addr'
544 * global variable. (although the entry point is 32-bit, the kernel
545 * binary can be either 32-bit or 64-bit).
546 */
read_pvh_start_addr(void * arg1,void * arg2,bool is64)547 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64)
548 {
549 size_t *elf_note_data_addr;
550
551 /* Check if ELF Note header passed in is valid */
552 if (arg1 == NULL) {
553 return 0;
554 }
555
556 if (is64) {
557 struct elf64_note *nhdr64 = (struct elf64_note *)arg1;
558 uint64_t nhdr_size64 = sizeof(struct elf64_note);
559 uint64_t phdr_align = *(uint64_t *)arg2;
560 uint64_t nhdr_namesz = nhdr64->n_namesz;
561
562 elf_note_data_addr =
563 ((void *)nhdr64) + nhdr_size64 +
564 QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
565
566 pvh_start_addr = *elf_note_data_addr;
567 } else {
568 struct elf32_note *nhdr32 = (struct elf32_note *)arg1;
569 uint32_t nhdr_size32 = sizeof(struct elf32_note);
570 uint32_t phdr_align = *(uint32_t *)arg2;
571 uint32_t nhdr_namesz = nhdr32->n_namesz;
572
573 elf_note_data_addr =
574 ((void *)nhdr32) + nhdr_size32 +
575 QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
576
577 pvh_start_addr = *(uint32_t *)elf_note_data_addr;
578 }
579
580 return pvh_start_addr;
581 }
582
load_elfboot(const char * kernel_filename,int kernel_file_size,uint8_t * header,size_t pvh_xen_start_addr,FWCfgState * fw_cfg)583 static bool load_elfboot(const char *kernel_filename,
584 int kernel_file_size,
585 uint8_t *header,
586 size_t pvh_xen_start_addr,
587 FWCfgState *fw_cfg)
588 {
589 uint32_t flags = 0;
590 uint32_t mh_load_addr = 0;
591 uint32_t elf_kernel_size = 0;
592 uint64_t elf_entry;
593 uint64_t elf_low, elf_high;
594 int kernel_size;
595
596 if (ldl_le_p(header) != 0x464c457f) {
597 return false; /* no elfboot */
598 }
599
600 bool elf_is64 = header[EI_CLASS] == ELFCLASS64;
601 flags = elf_is64 ?
602 ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags;
603
604 if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
605 error_report("elfboot unsupported flags = %x", flags);
606 exit(1);
607 }
608
609 uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY;
610 kernel_size = load_elf(kernel_filename, read_pvh_start_addr,
611 NULL, &elf_note_type, &elf_entry,
612 &elf_low, &elf_high, NULL,
613 ELFDATA2LSB, I386_ELF_MACHINE, 0, 0);
614
615 if (kernel_size < 0) {
616 error_report("Error while loading elf kernel");
617 exit(1);
618 }
619 mh_load_addr = elf_low;
620 elf_kernel_size = elf_high - elf_low;
621
622 if (pvh_start_addr == 0) {
623 error_report("Error loading uncompressed kernel without PVH ELF Note");
624 exit(1);
625 }
626 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr);
627 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr);
628 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size);
629
630 return true;
631 }
632
x86_load_linux(X86MachineState * x86ms,FWCfgState * fw_cfg,int acpi_data_size,bool pvh_enabled)633 void x86_load_linux(X86MachineState *x86ms,
634 FWCfgState *fw_cfg,
635 int acpi_data_size,
636 bool pvh_enabled)
637 {
638 bool linuxboot_dma_enabled = X86_MACHINE_GET_CLASS(x86ms)->fwcfg_dma_enabled;
639 uint16_t protocol;
640 int setup_size, kernel_size, cmdline_size;
641 int dtb_size, setup_data_offset;
642 uint32_t initrd_max;
643 uint8_t header[8192], *setup, *kernel;
644 hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0;
645 FILE *f;
646 char *vmode;
647 MachineState *machine = MACHINE(x86ms);
648 struct setup_data *setup_data;
649 const char *kernel_filename = machine->kernel_filename;
650 const char *initrd_filename = machine->initrd_filename;
651 const char *dtb_filename = machine->dtb;
652 const char *kernel_cmdline = machine->kernel_cmdline;
653 SevKernelLoaderContext sev_load_ctx = {};
654
655 /* Align to 16 bytes as a paranoia measure */
656 cmdline_size = (strlen(kernel_cmdline) + 16) & ~15;
657
658 /* load the kernel header */
659 f = fopen(kernel_filename, "rb");
660 if (!f) {
661 fprintf(stderr, "qemu: could not open kernel file '%s': %s\n",
662 kernel_filename, strerror(errno));
663 exit(1);
664 }
665
666 kernel_size = get_file_size(f);
667 if (!kernel_size ||
668 fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
669 MIN(ARRAY_SIZE(header), kernel_size)) {
670 fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
671 kernel_filename, strerror(errno));
672 exit(1);
673 }
674
675 /*
676 * kernel protocol version.
677 * Please see https://www.kernel.org/doc/Documentation/x86/boot.txt
678 */
679 if (ldl_le_p(header + 0x202) == 0x53726448) /* Magic signature "HdrS" */ {
680 protocol = lduw_le_p(header + 0x206);
681 } else {
682 /*
683 * This could be a multiboot kernel. If it is, let's stop treating it
684 * like a Linux kernel.
685 * Note: some multiboot images could be in the ELF format (the same of
686 * PVH), so we try multiboot first since we check the multiboot magic
687 * header before to load it.
688 */
689 if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename,
690 kernel_cmdline, kernel_size, header)) {
691 return;
692 }
693 /*
694 * Check if the file is an uncompressed kernel file (ELF) and load it,
695 * saving the PVH entry point used by the x86/HVM direct boot ABI.
696 * If load_elfboot() is successful, populate the fw_cfg info.
697 */
698 if (pvh_enabled &&
699 load_elfboot(kernel_filename, kernel_size,
700 header, pvh_start_addr, fw_cfg)) {
701 fclose(f);
702
703 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
704 strlen(kernel_cmdline) + 1);
705 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
706
707 setup = g_memdup2(header, sizeof(header));
708
709 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header));
710 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA,
711 setup, sizeof(header));
712
713 /* load initrd */
714 if (initrd_filename) {
715 GMappedFile *mapped_file;
716 gsize initrd_size;
717 gchar *initrd_data;
718 GError *gerr = NULL;
719
720 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
721 if (!mapped_file) {
722 fprintf(stderr, "qemu: error reading initrd %s: %s\n",
723 initrd_filename, gerr->message);
724 exit(1);
725 }
726 x86ms->initrd_mapped_file = mapped_file;
727
728 initrd_data = g_mapped_file_get_contents(mapped_file);
729 initrd_size = g_mapped_file_get_length(mapped_file);
730 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
731 if (initrd_size >= initrd_max) {
732 fprintf(stderr, "qemu: initrd is too large, cannot support."
733 "(max: %"PRIu32", need %"PRId64")\n",
734 initrd_max, (uint64_t)initrd_size);
735 exit(1);
736 }
737
738 initrd_addr = (initrd_max - initrd_size) & ~4095;
739
740 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
741 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
742 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data,
743 initrd_size);
744 }
745
746 option_rom[nb_option_roms].bootindex = 0;
747 option_rom[nb_option_roms].name = "pvh.bin";
748 nb_option_roms++;
749
750 return;
751 }
752 protocol = 0;
753 }
754
755 if (protocol < 0x200 || !(header[0x211] & 0x01)) {
756 /* Low kernel */
757 real_addr = 0x90000;
758 cmdline_addr = 0x9a000 - cmdline_size;
759 prot_addr = 0x10000;
760 } else if (protocol < 0x202) {
761 /* High but ancient kernel */
762 real_addr = 0x90000;
763 cmdline_addr = 0x9a000 - cmdline_size;
764 prot_addr = 0x100000;
765 } else {
766 /* High and recent kernel */
767 real_addr = 0x10000;
768 cmdline_addr = 0x20000;
769 prot_addr = 0x100000;
770 }
771
772 /* highest address for loading the initrd */
773 if (protocol >= 0x20c &&
774 lduw_le_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) {
775 /*
776 * Linux has supported initrd up to 4 GB for a very long time (2007,
777 * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
778 * though it only sets initrd_max to 2 GB to "work around bootloader
779 * bugs". Luckily, QEMU firmware(which does something like bootloader)
780 * has supported this.
781 *
782 * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
783 * be loaded into any address.
784 *
785 * In addition, initrd_max is uint32_t simply because QEMU doesn't
786 * support the 64-bit boot protocol (specifically the ext_ramdisk_image
787 * field).
788 *
789 * Therefore here just limit initrd_max to UINT32_MAX simply as well.
790 */
791 initrd_max = UINT32_MAX;
792 } else if (protocol >= 0x203) {
793 initrd_max = ldl_le_p(header + 0x22c);
794 } else {
795 initrd_max = 0x37ffffff;
796 }
797
798 if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) {
799 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
800 }
801
802 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
803 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline) + 1);
804 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
805 sev_load_ctx.cmdline_data = (char *)kernel_cmdline;
806 sev_load_ctx.cmdline_size = strlen(kernel_cmdline) + 1;
807
808 if (protocol >= 0x202) {
809 stl_le_p(header + 0x228, cmdline_addr);
810 } else {
811 stw_le_p(header + 0x20, 0xA33F);
812 stw_le_p(header + 0x22, cmdline_addr - real_addr);
813 }
814
815 /* handle vga= parameter */
816 vmode = strstr(kernel_cmdline, "vga=");
817 if (vmode) {
818 unsigned int video_mode;
819 const char *end;
820 int ret;
821 /* skip "vga=" */
822 vmode += 4;
823 if (!strncmp(vmode, "normal", 6)) {
824 video_mode = 0xffff;
825 } else if (!strncmp(vmode, "ext", 3)) {
826 video_mode = 0xfffe;
827 } else if (!strncmp(vmode, "ask", 3)) {
828 video_mode = 0xfffd;
829 } else {
830 ret = qemu_strtoui(vmode, &end, 0, &video_mode);
831 if (ret != 0 || (*end && *end != ' ')) {
832 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n");
833 exit(1);
834 }
835 }
836 stw_le_p(header + 0x1fa, video_mode);
837 }
838
839 /* loader type */
840 /*
841 * High nybble = B reserved for QEMU; low nybble is revision number.
842 * If this code is substantially changed, you may want to consider
843 * incrementing the revision.
844 */
845 if (protocol >= 0x200) {
846 header[0x210] = 0xB0;
847 }
848 /* heap */
849 if (protocol >= 0x201) {
850 header[0x211] |= 0x80; /* CAN_USE_HEAP */
851 stw_le_p(header + 0x224, cmdline_addr - real_addr - 0x200);
852 }
853
854 /* load initrd */
855 if (initrd_filename) {
856 GMappedFile *mapped_file;
857 gsize initrd_size;
858 gchar *initrd_data;
859 GError *gerr = NULL;
860
861 if (protocol < 0x200) {
862 fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
863 exit(1);
864 }
865
866 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
867 if (!mapped_file) {
868 fprintf(stderr, "qemu: error reading initrd %s: %s\n",
869 initrd_filename, gerr->message);
870 exit(1);
871 }
872 x86ms->initrd_mapped_file = mapped_file;
873
874 initrd_data = g_mapped_file_get_contents(mapped_file);
875 initrd_size = g_mapped_file_get_length(mapped_file);
876 if (initrd_size >= initrd_max) {
877 fprintf(stderr, "qemu: initrd is too large, cannot support."
878 "(max: %"PRIu32", need %"PRId64")\n",
879 initrd_max, (uint64_t)initrd_size);
880 exit(1);
881 }
882
883 initrd_addr = (initrd_max - initrd_size) & ~4095;
884
885 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
886 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
887 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
888 sev_load_ctx.initrd_data = initrd_data;
889 sev_load_ctx.initrd_size = initrd_size;
890
891 stl_le_p(header + 0x218, initrd_addr);
892 stl_le_p(header + 0x21c, initrd_size);
893 }
894
895 /* load kernel and setup */
896 setup_size = header[0x1f1];
897 if (setup_size == 0) {
898 setup_size = 4;
899 }
900 setup_size = (setup_size + 1) * 512;
901 if (setup_size > kernel_size) {
902 fprintf(stderr, "qemu: invalid kernel header\n");
903 exit(1);
904 }
905
906 setup = g_malloc(setup_size);
907 kernel = g_malloc(kernel_size);
908 fseek(f, 0, SEEK_SET);
909 if (fread(setup, 1, setup_size, f) != setup_size) {
910 fprintf(stderr, "fread() failed\n");
911 exit(1);
912 }
913 fseek(f, 0, SEEK_SET);
914 if (fread(kernel, 1, kernel_size, f) != kernel_size) {
915 fprintf(stderr, "fread() failed\n");
916 exit(1);
917 }
918 fclose(f);
919
920 /* append dtb to kernel */
921 if (dtb_filename) {
922 if (protocol < 0x209) {
923 fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n");
924 exit(1);
925 }
926
927 dtb_size = get_image_size(dtb_filename);
928 if (dtb_size <= 0) {
929 fprintf(stderr, "qemu: error reading dtb %s: %s\n",
930 dtb_filename, strerror(errno));
931 exit(1);
932 }
933
934 setup_data_offset = QEMU_ALIGN_UP(kernel_size, 16);
935 kernel_size = setup_data_offset + sizeof(struct setup_data) + dtb_size;
936 kernel = g_realloc(kernel, kernel_size);
937
938 stq_le_p(header + 0x250, prot_addr + setup_data_offset);
939
940 setup_data = (struct setup_data *)(kernel + setup_data_offset);
941 setup_data->next = 0;
942 setup_data->type = cpu_to_le32(SETUP_DTB);
943 setup_data->len = cpu_to_le32(dtb_size);
944
945 load_image_size(dtb_filename, setup_data->data, dtb_size);
946 }
947
948 /*
949 * If we're starting an encrypted VM, it will be OVMF based, which uses the
950 * efi stub for booting and doesn't require any values to be placed in the
951 * kernel header. We therefore don't update the header so the hash of the
952 * kernel on the other side of the fw_cfg interface matches the hash of the
953 * file the user passed in.
954 */
955 if (!sev_enabled() && protocol > 0) {
956 memcpy(setup, header, MIN(sizeof(header), setup_size));
957 }
958
959 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
960 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size - setup_size);
961 fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA,
962 kernel + setup_size, kernel_size - setup_size);
963 sev_load_ctx.kernel_data = (char *)kernel + setup_size;
964 sev_load_ctx.kernel_size = kernel_size - setup_size;
965
966 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
967 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
968 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
969 sev_load_ctx.setup_data = (char *)setup;
970 sev_load_ctx.setup_size = setup_size;
971
972 /* kernel without setup header patches */
973 fw_cfg_add_file(fw_cfg, "etc/boot/kernel", kernel, kernel_size);
974
975 if (machine->shim_filename) {
976 GMappedFile *mapped_file;
977 GError *gerr = NULL;
978
979 mapped_file = g_mapped_file_new(machine->shim_filename, false, &gerr);
980 if (!mapped_file) {
981 fprintf(stderr, "qemu: error reading shim %s: %s\n",
982 machine->shim_filename, gerr->message);
983 exit(1);
984 }
985
986 fw_cfg_add_file(fw_cfg, "etc/boot/shim",
987 g_mapped_file_get_contents(mapped_file),
988 g_mapped_file_get_length(mapped_file));
989 }
990
991 if (sev_enabled()) {
992 sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal);
993 }
994
995 option_rom[nb_option_roms].bootindex = 0;
996 option_rom[nb_option_roms].name = "linuxboot.bin";
997 if (linuxboot_dma_enabled && fw_cfg_dma_enabled(fw_cfg)) {
998 option_rom[nb_option_roms].name = "linuxboot_dma.bin";
999 }
1000 nb_option_roms++;
1001 }
1002
x86_isa_bios_init(MemoryRegion * isa_bios,MemoryRegion * isa_memory,MemoryRegion * bios,bool read_only)1003 void x86_isa_bios_init(MemoryRegion *isa_bios, MemoryRegion *isa_memory,
1004 MemoryRegion *bios, bool read_only)
1005 {
1006 uint64_t bios_size = memory_region_size(bios);
1007 uint64_t isa_bios_size = MIN(bios_size, 128 * KiB);
1008
1009 memory_region_init_alias(isa_bios, NULL, "isa-bios", bios,
1010 bios_size - isa_bios_size, isa_bios_size);
1011 memory_region_add_subregion_overlap(isa_memory, 1 * MiB - isa_bios_size,
1012 isa_bios, 1);
1013 memory_region_set_readonly(isa_bios, read_only);
1014 }
1015
x86_bios_rom_init(X86MachineState * x86ms,const char * default_firmware,MemoryRegion * rom_memory,bool isapc_ram_fw)1016 void x86_bios_rom_init(X86MachineState *x86ms, const char *default_firmware,
1017 MemoryRegion *rom_memory, bool isapc_ram_fw)
1018 {
1019 const char *bios_name;
1020 char *filename;
1021 int bios_size;
1022 ssize_t ret;
1023
1024 /* BIOS load */
1025 bios_name = MACHINE(x86ms)->firmware ?: default_firmware;
1026 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1027 if (filename) {
1028 bios_size = get_image_size(filename);
1029 } else {
1030 bios_size = -1;
1031 }
1032 if (bios_size <= 0 ||
1033 (bios_size % 65536) != 0) {
1034 goto bios_error;
1035 }
1036 if (machine_require_guest_memfd(MACHINE(x86ms))) {
1037 memory_region_init_ram_guest_memfd(&x86ms->bios, NULL, "pc.bios",
1038 bios_size, &error_fatal);
1039 if (is_tdx_vm()) {
1040 tdx_set_tdvf_region(&x86ms->bios);
1041 }
1042 } else {
1043 memory_region_init_ram(&x86ms->bios, NULL, "pc.bios",
1044 bios_size, &error_fatal);
1045 }
1046 if (sev_enabled() || is_tdx_vm()) {
1047 /*
1048 * The concept of a "reset" simply doesn't exist for
1049 * confidential computing guests, we have to destroy and
1050 * re-launch them instead. So there is no need to register
1051 * the firmware as rom to properly re-initialize on reset.
1052 * Just go for a straight file load instead.
1053 */
1054 void *ptr = memory_region_get_ram_ptr(&x86ms->bios);
1055 load_image_size(filename, ptr, bios_size);
1056 x86_firmware_configure(0x100000000ULL - bios_size, ptr, bios_size);
1057 } else {
1058 memory_region_set_readonly(&x86ms->bios, !isapc_ram_fw);
1059 ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1);
1060 if (ret != 0) {
1061 goto bios_error;
1062 }
1063 }
1064 g_free(filename);
1065
1066 if (!machine_require_guest_memfd(MACHINE(x86ms))) {
1067 /* map the last 128KB of the BIOS in ISA space */
1068 x86_isa_bios_init(&x86ms->isa_bios, rom_memory, &x86ms->bios,
1069 !isapc_ram_fw);
1070 }
1071
1072 /* map all the bios at the top of memory */
1073 memory_region_add_subregion(rom_memory,
1074 (uint32_t)(-bios_size),
1075 &x86ms->bios);
1076 return;
1077
1078 bios_error:
1079 fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1080 exit(1);
1081 }
1082