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