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