xref: /openbmc/qemu/hw/i386/x86.c (revision aa09b3d5)
1 /*
2  * Copyright (c) 2003-2004 Fabrice Bellard
3  * Copyright (c) 2019 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/option.h"
26 #include "qemu/cutils.h"
27 #include "qemu/units.h"
28 #include "qemu/datadir.h"
29 #include "qemu/guest-random.h"
30 #include "qapi/error.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qapi/qapi-visit-common.h"
33 #include "qapi/clone-visitor.h"
34 #include "qapi/qapi-visit-machine.h"
35 #include "qapi/visitor.h"
36 #include "sysemu/qtest.h"
37 #include "sysemu/whpx.h"
38 #include "sysemu/numa.h"
39 #include "sysemu/replay.h"
40 #include "sysemu/reset.h"
41 #include "sysemu/sysemu.h"
42 #include "sysemu/cpu-timers.h"
43 #include "sysemu/xen.h"
44 #include "trace.h"
45 
46 #include "hw/i386/x86.h"
47 #include "target/i386/cpu.h"
48 #include "hw/i386/topology.h"
49 #include "hw/i386/fw_cfg.h"
50 #include "hw/intc/i8259.h"
51 #include "hw/rtc/mc146818rtc.h"
52 #include "target/i386/sev.h"
53 #include "hw/i386/microvm.h"
54 
55 #include "hw/acpi/cpu_hotplug.h"
56 #include "hw/irq.h"
57 #include "hw/nmi.h"
58 #include "hw/loader.h"
59 #include "multiboot.h"
60 #include "elf.h"
61 #include "standard-headers/asm-x86/bootparam.h"
62 #include CONFIG_DEVICES
63 #include "kvm/kvm_i386.h"
64 
65 /* Physical Address of PVH entry point read from kernel ELF NOTE */
66 static size_t pvh_start_addr;
67 
68 inline void init_topo_info(X86CPUTopoInfo *topo_info,
69                            const X86MachineState *x86ms)
70 {
71     MachineState *ms = MACHINE(x86ms);
72 
73     topo_info->dies_per_pkg = ms->smp.dies;
74     topo_info->cores_per_die = ms->smp.cores;
75     topo_info->threads_per_core = ms->smp.threads;
76 }
77 
78 /*
79  * Calculates initial APIC ID for a specific CPU index
80  *
81  * Currently we need to be able to calculate the APIC ID from the CPU index
82  * alone (without requiring a CPU object), as the QEMU<->Seabios interfaces have
83  * no concept of "CPU index", and the NUMA tables on fw_cfg need the APIC ID of
84  * all CPUs up to max_cpus.
85  */
86 uint32_t x86_cpu_apic_id_from_index(X86MachineState *x86ms,
87                                     unsigned int cpu_index)
88 {
89     X86CPUTopoInfo topo_info;
90 
91     init_topo_info(&topo_info, x86ms);
92 
93     return x86_apicid_from_cpu_idx(&topo_info, cpu_index);
94 }
95 
96 
97 void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp)
98 {
99     Object *cpu = object_new(MACHINE(x86ms)->cpu_type);
100 
101     if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) {
102         goto out;
103     }
104     qdev_realize(DEVICE(cpu), NULL, errp);
105 
106 out:
107     object_unref(cpu);
108 }
109 
110 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version)
111 {
112     int i;
113     const CPUArchIdList *possible_cpus;
114     MachineState *ms = MACHINE(x86ms);
115     MachineClass *mc = MACHINE_GET_CLASS(x86ms);
116 
117     x86_cpu_set_default_version(default_cpu_version);
118 
119     /*
120      * Calculates the limit to CPU APIC ID values
121      *
122      * Limit for the APIC ID value, so that all
123      * CPU APIC IDs are < x86ms->apic_id_limit.
124      *
125      * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
126      */
127     x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms,
128                                                       ms->smp.max_cpus - 1) + 1;
129 
130     /*
131      * Can we support APIC ID 255 or higher?
132      *
133      * Under Xen: yes.
134      * With userspace emulated lapic: no
135      * With KVM's in-kernel lapic: only if X2APIC API is enabled.
136      */
137     if (x86ms->apic_id_limit > 255 && !xen_enabled() &&
138         (!kvm_irqchip_in_kernel() || !kvm_enable_x2apic())) {
139         error_report("current -smp configuration requires kernel "
140                      "irqchip and X2APIC API support.");
141         exit(EXIT_FAILURE);
142     }
143 
144     if (kvm_enabled()) {
145         kvm_set_max_apic_id(x86ms->apic_id_limit);
146     }
147 
148     possible_cpus = mc->possible_cpu_arch_ids(ms);
149     for (i = 0; i < ms->smp.cpus; i++) {
150         x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal);
151     }
152 }
153 
154 void x86_rtc_set_cpus_count(ISADevice *rtc, uint16_t cpus_count)
155 {
156     if (cpus_count > 0xff) {
157         /*
158          * If the number of CPUs can't be represented in 8 bits, the
159          * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just
160          * to make old BIOSes fail more predictably.
161          */
162         rtc_set_memory(rtc, 0x5f, 0);
163     } else {
164         rtc_set_memory(rtc, 0x5f, cpus_count - 1);
165     }
166 }
167 
168 static int x86_apic_cmp(const void *a, const void *b)
169 {
170    CPUArchId *apic_a = (CPUArchId *)a;
171    CPUArchId *apic_b = (CPUArchId *)b;
172 
173    return apic_a->arch_id - apic_b->arch_id;
174 }
175 
176 /*
177  * returns pointer to CPUArchId descriptor that matches CPU's apic_id
178  * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no
179  * entry corresponding to CPU's apic_id returns NULL.
180  */
181 CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
182 {
183     CPUArchId apic_id, *found_cpu;
184 
185     apic_id.arch_id = id;
186     found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus,
187         ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus),
188         x86_apic_cmp);
189     if (found_cpu && idx) {
190         *idx = found_cpu - ms->possible_cpus->cpus;
191     }
192     return found_cpu;
193 }
194 
195 void x86_cpu_plug(HotplugHandler *hotplug_dev,
196                   DeviceState *dev, Error **errp)
197 {
198     CPUArchId *found_cpu;
199     Error *local_err = NULL;
200     X86CPU *cpu = X86_CPU(dev);
201     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
202 
203     if (x86ms->acpi_dev) {
204         hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err);
205         if (local_err) {
206             goto out;
207         }
208     }
209 
210     /* increment the number of CPUs */
211     x86ms->boot_cpus++;
212     if (x86ms->rtc) {
213         x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
214     }
215     if (x86ms->fw_cfg) {
216         fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
217     }
218 
219     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
220     found_cpu->cpu = OBJECT(dev);
221 out:
222     error_propagate(errp, local_err);
223 }
224 
225 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev,
226                                DeviceState *dev, Error **errp)
227 {
228     int idx = -1;
229     X86CPU *cpu = X86_CPU(dev);
230     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
231 
232     if (!x86ms->acpi_dev) {
233         error_setg(errp, "CPU hot unplug not supported without ACPI");
234         return;
235     }
236 
237     x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
238     assert(idx != -1);
239     if (idx == 0) {
240         error_setg(errp, "Boot CPU is unpluggable");
241         return;
242     }
243 
244     hotplug_handler_unplug_request(x86ms->acpi_dev, dev,
245                                    errp);
246 }
247 
248 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev,
249                        DeviceState *dev, Error **errp)
250 {
251     CPUArchId *found_cpu;
252     Error *local_err = NULL;
253     X86CPU *cpu = X86_CPU(dev);
254     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
255 
256     hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err);
257     if (local_err) {
258         goto out;
259     }
260 
261     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
262     found_cpu->cpu = NULL;
263     qdev_unrealize(dev);
264 
265     /* decrement the number of CPUs */
266     x86ms->boot_cpus--;
267     /* Update the number of CPUs in CMOS */
268     x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
269     fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
270  out:
271     error_propagate(errp, local_err);
272 }
273 
274 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev,
275                       DeviceState *dev, Error **errp)
276 {
277     int idx;
278     CPUState *cs;
279     CPUArchId *cpu_slot;
280     X86CPUTopoIDs topo_ids;
281     X86CPU *cpu = X86_CPU(dev);
282     CPUX86State *env = &cpu->env;
283     MachineState *ms = MACHINE(hotplug_dev);
284     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
285     unsigned int smp_cores = ms->smp.cores;
286     unsigned int smp_threads = ms->smp.threads;
287     X86CPUTopoInfo topo_info;
288 
289     if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) {
290         error_setg(errp, "Invalid CPU type, expected cpu type: '%s'",
291                    ms->cpu_type);
292         return;
293     }
294 
295     if (x86ms->acpi_dev) {
296         Error *local_err = NULL;
297 
298         hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev,
299                                  &local_err);
300         if (local_err) {
301             error_propagate(errp, local_err);
302             return;
303         }
304     }
305 
306     init_topo_info(&topo_info, x86ms);
307 
308     env->nr_dies = ms->smp.dies;
309 
310     /*
311      * If APIC ID is not set,
312      * set it based on socket/die/core/thread properties.
313      */
314     if (cpu->apic_id == UNASSIGNED_APIC_ID) {
315         int max_socket = (ms->smp.max_cpus - 1) /
316                                 smp_threads / smp_cores / ms->smp.dies;
317 
318         /*
319          * die-id was optional in QEMU 4.0 and older, so keep it optional
320          * if there's only one die per socket.
321          */
322         if (cpu->die_id < 0 && ms->smp.dies == 1) {
323             cpu->die_id = 0;
324         }
325 
326         if (cpu->socket_id < 0) {
327             error_setg(errp, "CPU socket-id is not set");
328             return;
329         } else if (cpu->socket_id > max_socket) {
330             error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u",
331                        cpu->socket_id, max_socket);
332             return;
333         }
334         if (cpu->die_id < 0) {
335             error_setg(errp, "CPU die-id is not set");
336             return;
337         } else if (cpu->die_id > ms->smp.dies - 1) {
338             error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u",
339                        cpu->die_id, ms->smp.dies - 1);
340             return;
341         }
342         if (cpu->core_id < 0) {
343             error_setg(errp, "CPU core-id is not set");
344             return;
345         } else if (cpu->core_id > (smp_cores - 1)) {
346             error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u",
347                        cpu->core_id, smp_cores - 1);
348             return;
349         }
350         if (cpu->thread_id < 0) {
351             error_setg(errp, "CPU thread-id is not set");
352             return;
353         } else if (cpu->thread_id > (smp_threads - 1)) {
354             error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u",
355                        cpu->thread_id, smp_threads - 1);
356             return;
357         }
358 
359         topo_ids.pkg_id = cpu->socket_id;
360         topo_ids.die_id = cpu->die_id;
361         topo_ids.core_id = cpu->core_id;
362         topo_ids.smt_id = cpu->thread_id;
363         cpu->apic_id = x86_apicid_from_topo_ids(&topo_info, &topo_ids);
364     }
365 
366     cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
367     if (!cpu_slot) {
368         MachineState *ms = MACHINE(x86ms);
369 
370         x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
371         error_setg(errp,
372             "Invalid CPU [socket: %u, die: %u, core: %u, thread: %u] with"
373             " APIC ID %" PRIu32 ", valid index range 0:%d",
374             topo_ids.pkg_id, topo_ids.die_id, topo_ids.core_id, topo_ids.smt_id,
375             cpu->apic_id, ms->possible_cpus->len - 1);
376         return;
377     }
378 
379     if (cpu_slot->cpu) {
380         error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists",
381                    idx, cpu->apic_id);
382         return;
383     }
384 
385     /* if 'address' properties socket-id/core-id/thread-id are not set, set them
386      * so that machine_query_hotpluggable_cpus would show correct values
387      */
388     /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
389      * once -smp refactoring is complete and there will be CPU private
390      * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
391     x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
392     if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) {
393         error_setg(errp, "property socket-id: %u doesn't match set apic-id:"
394             " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id,
395             topo_ids.pkg_id);
396         return;
397     }
398     cpu->socket_id = topo_ids.pkg_id;
399 
400     if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) {
401         error_setg(errp, "property die-id: %u doesn't match set apic-id:"
402             " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id);
403         return;
404     }
405     cpu->die_id = topo_ids.die_id;
406 
407     if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) {
408         error_setg(errp, "property core-id: %u doesn't match set apic-id:"
409             " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id,
410             topo_ids.core_id);
411         return;
412     }
413     cpu->core_id = topo_ids.core_id;
414 
415     if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) {
416         error_setg(errp, "property thread-id: %u doesn't match set apic-id:"
417             " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id,
418             topo_ids.smt_id);
419         return;
420     }
421     cpu->thread_id = topo_ids.smt_id;
422 
423     if (hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) &&
424         !kvm_hv_vpindex_settable()) {
425         error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX");
426         return;
427     }
428 
429     cs = CPU(cpu);
430     cs->cpu_index = idx;
431 
432     numa_cpu_pre_plug(cpu_slot, dev, errp);
433 }
434 
435 CpuInstanceProperties
436 x86_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
437 {
438     MachineClass *mc = MACHINE_GET_CLASS(ms);
439     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
440 
441     assert(cpu_index < possible_cpus->len);
442     return possible_cpus->cpus[cpu_index].props;
443 }
444 
445 int64_t x86_get_default_cpu_node_id(const MachineState *ms, int idx)
446 {
447    X86CPUTopoIDs topo_ids;
448    X86MachineState *x86ms = X86_MACHINE(ms);
449    X86CPUTopoInfo topo_info;
450 
451    init_topo_info(&topo_info, x86ms);
452 
453    assert(idx < ms->possible_cpus->len);
454    x86_topo_ids_from_apicid(ms->possible_cpus->cpus[idx].arch_id,
455                             &topo_info, &topo_ids);
456    return topo_ids.pkg_id % ms->numa_state->num_nodes;
457 }
458 
459 const CPUArchIdList *x86_possible_cpu_arch_ids(MachineState *ms)
460 {
461     X86MachineState *x86ms = X86_MACHINE(ms);
462     unsigned int max_cpus = ms->smp.max_cpus;
463     X86CPUTopoInfo topo_info;
464     int i;
465 
466     if (ms->possible_cpus) {
467         /*
468          * make sure that max_cpus hasn't changed since the first use, i.e.
469          * -smp hasn't been parsed after it
470          */
471         assert(ms->possible_cpus->len == max_cpus);
472         return ms->possible_cpus;
473     }
474 
475     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
476                                   sizeof(CPUArchId) * max_cpus);
477     ms->possible_cpus->len = max_cpus;
478 
479     init_topo_info(&topo_info, x86ms);
480 
481     for (i = 0; i < ms->possible_cpus->len; i++) {
482         X86CPUTopoIDs topo_ids;
483 
484         ms->possible_cpus->cpus[i].type = ms->cpu_type;
485         ms->possible_cpus->cpus[i].vcpus_count = 1;
486         ms->possible_cpus->cpus[i].arch_id =
487             x86_cpu_apic_id_from_index(x86ms, i);
488         x86_topo_ids_from_apicid(ms->possible_cpus->cpus[i].arch_id,
489                                  &topo_info, &topo_ids);
490         ms->possible_cpus->cpus[i].props.has_socket_id = true;
491         ms->possible_cpus->cpus[i].props.socket_id = topo_ids.pkg_id;
492         if (ms->smp.dies > 1) {
493             ms->possible_cpus->cpus[i].props.has_die_id = true;
494             ms->possible_cpus->cpus[i].props.die_id = topo_ids.die_id;
495         }
496         ms->possible_cpus->cpus[i].props.has_core_id = true;
497         ms->possible_cpus->cpus[i].props.core_id = topo_ids.core_id;
498         ms->possible_cpus->cpus[i].props.has_thread_id = true;
499         ms->possible_cpus->cpus[i].props.thread_id = topo_ids.smt_id;
500     }
501     return ms->possible_cpus;
502 }
503 
504 static void x86_nmi(NMIState *n, int cpu_index, Error **errp)
505 {
506     /* cpu index isn't used */
507     CPUState *cs;
508 
509     CPU_FOREACH(cs) {
510         X86CPU *cpu = X86_CPU(cs);
511 
512         if (!cpu->apic_state) {
513             cpu_interrupt(cs, CPU_INTERRUPT_NMI);
514         } else {
515             apic_deliver_nmi(cpu->apic_state);
516         }
517     }
518 }
519 
520 static long get_file_size(FILE *f)
521 {
522     long where, size;
523 
524     /* XXX: on Unix systems, using fstat() probably makes more sense */
525 
526     where = ftell(f);
527     fseek(f, 0, SEEK_END);
528     size = ftell(f);
529     fseek(f, where, SEEK_SET);
530 
531     return size;
532 }
533 
534 /* TSC handling */
535 uint64_t cpu_get_tsc(CPUX86State *env)
536 {
537     return cpus_get_elapsed_ticks();
538 }
539 
540 /* IRQ handling */
541 static void pic_irq_request(void *opaque, int irq, int level)
542 {
543     CPUState *cs = first_cpu;
544     X86CPU *cpu = X86_CPU(cs);
545 
546     trace_x86_pic_interrupt(irq, level);
547     if (cpu->apic_state && !kvm_irqchip_in_kernel() &&
548         !whpx_apic_in_platform()) {
549         CPU_FOREACH(cs) {
550             cpu = X86_CPU(cs);
551             if (apic_accept_pic_intr(cpu->apic_state)) {
552                 apic_deliver_pic_intr(cpu->apic_state, level);
553             }
554         }
555     } else {
556         if (level) {
557             cpu_interrupt(cs, CPU_INTERRUPT_HARD);
558         } else {
559             cpu_reset_interrupt(cs, CPU_INTERRUPT_HARD);
560         }
561     }
562 }
563 
564 qemu_irq x86_allocate_cpu_irq(void)
565 {
566     return qemu_allocate_irq(pic_irq_request, NULL, 0);
567 }
568 
569 int cpu_get_pic_interrupt(CPUX86State *env)
570 {
571     X86CPU *cpu = env_archcpu(env);
572     int intno;
573 
574     if (!kvm_irqchip_in_kernel() && !whpx_apic_in_platform()) {
575         intno = apic_get_interrupt(cpu->apic_state);
576         if (intno >= 0) {
577             return intno;
578         }
579         /* read the irq from the PIC */
580         if (!apic_accept_pic_intr(cpu->apic_state)) {
581             return -1;
582         }
583     }
584 
585     intno = pic_read_irq(isa_pic);
586     return intno;
587 }
588 
589 DeviceState *cpu_get_current_apic(void)
590 {
591     if (current_cpu) {
592         X86CPU *cpu = X86_CPU(current_cpu);
593         return cpu->apic_state;
594     } else {
595         return NULL;
596     }
597 }
598 
599 void gsi_handler(void *opaque, int n, int level)
600 {
601     GSIState *s = opaque;
602 
603     trace_x86_gsi_interrupt(n, level);
604     switch (n) {
605     case 0 ... ISA_NUM_IRQS - 1:
606         if (s->i8259_irq[n]) {
607             /* Under KVM, Kernel will forward to both PIC and IOAPIC */
608             qemu_set_irq(s->i8259_irq[n], level);
609         }
610         /* fall through */
611     case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
612         qemu_set_irq(s->ioapic_irq[n], level);
613         break;
614     case IO_APIC_SECONDARY_IRQBASE
615         ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1:
616         qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level);
617         break;
618     }
619 }
620 
621 void ioapic_init_gsi(GSIState *gsi_state, const char *parent_name)
622 {
623     DeviceState *dev;
624     SysBusDevice *d;
625     unsigned int i;
626 
627     assert(parent_name);
628     if (kvm_ioapic_in_kernel()) {
629         dev = qdev_new(TYPE_KVM_IOAPIC);
630     } else {
631         dev = qdev_new(TYPE_IOAPIC);
632     }
633     object_property_add_child(object_resolve_path(parent_name, NULL),
634                               "ioapic", OBJECT(dev));
635     d = SYS_BUS_DEVICE(dev);
636     sysbus_realize_and_unref(d, &error_fatal);
637     sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS);
638 
639     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
640         gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i);
641     }
642 }
643 
644 DeviceState *ioapic_init_secondary(GSIState *gsi_state)
645 {
646     DeviceState *dev;
647     SysBusDevice *d;
648     unsigned int i;
649 
650     dev = qdev_new(TYPE_IOAPIC);
651     d = SYS_BUS_DEVICE(dev);
652     sysbus_realize_and_unref(d, &error_fatal);
653     sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS);
654 
655     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
656         gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i);
657     }
658     return dev;
659 }
660 
661 typedef struct SetupData {
662     uint64_t next;
663     uint32_t type;
664     uint32_t len;
665     uint8_t data[];
666 } __attribute__((packed)) SetupData;
667 
668 
669 /*
670  * The entry point into the kernel for PVH boot is different from
671  * the native entry point.  The PVH entry is defined by the x86/HVM
672  * direct boot ABI and is available in an ELFNOTE in the kernel binary.
673  *
674  * This function is passed to load_elf() when it is called from
675  * load_elfboot() which then additionally checks for an ELF Note of
676  * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
677  * parse the PVH entry address from the ELF Note.
678  *
679  * Due to trickery in elf_opts.h, load_elf() is actually available as
680  * load_elf32() or load_elf64() and this routine needs to be able
681  * to deal with being called as 32 or 64 bit.
682  *
683  * The address of the PVH entry point is saved to the 'pvh_start_addr'
684  * global variable.  (although the entry point is 32-bit, the kernel
685  * binary can be either 32-bit or 64-bit).
686  */
687 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64)
688 {
689     size_t *elf_note_data_addr;
690 
691     /* Check if ELF Note header passed in is valid */
692     if (arg1 == NULL) {
693         return 0;
694     }
695 
696     if (is64) {
697         struct elf64_note *nhdr64 = (struct elf64_note *)arg1;
698         uint64_t nhdr_size64 = sizeof(struct elf64_note);
699         uint64_t phdr_align = *(uint64_t *)arg2;
700         uint64_t nhdr_namesz = nhdr64->n_namesz;
701 
702         elf_note_data_addr =
703             ((void *)nhdr64) + nhdr_size64 +
704             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
705 
706         pvh_start_addr = *elf_note_data_addr;
707     } else {
708         struct elf32_note *nhdr32 = (struct elf32_note *)arg1;
709         uint32_t nhdr_size32 = sizeof(struct elf32_note);
710         uint32_t phdr_align = *(uint32_t *)arg2;
711         uint32_t nhdr_namesz = nhdr32->n_namesz;
712 
713         elf_note_data_addr =
714             ((void *)nhdr32) + nhdr_size32 +
715             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
716 
717         pvh_start_addr = *(uint32_t *)elf_note_data_addr;
718     }
719 
720     return pvh_start_addr;
721 }
722 
723 static bool load_elfboot(const char *kernel_filename,
724                          int kernel_file_size,
725                          uint8_t *header,
726                          size_t pvh_xen_start_addr,
727                          FWCfgState *fw_cfg)
728 {
729     uint32_t flags = 0;
730     uint32_t mh_load_addr = 0;
731     uint32_t elf_kernel_size = 0;
732     uint64_t elf_entry;
733     uint64_t elf_low, elf_high;
734     int kernel_size;
735 
736     if (ldl_p(header) != 0x464c457f) {
737         return false; /* no elfboot */
738     }
739 
740     bool elf_is64 = header[EI_CLASS] == ELFCLASS64;
741     flags = elf_is64 ?
742         ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags;
743 
744     if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
745         error_report("elfboot unsupported flags = %x", flags);
746         exit(1);
747     }
748 
749     uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY;
750     kernel_size = load_elf(kernel_filename, read_pvh_start_addr,
751                            NULL, &elf_note_type, &elf_entry,
752                            &elf_low, &elf_high, NULL, 0, I386_ELF_MACHINE,
753                            0, 0);
754 
755     if (kernel_size < 0) {
756         error_report("Error while loading elf kernel");
757         exit(1);
758     }
759     mh_load_addr = elf_low;
760     elf_kernel_size = elf_high - elf_low;
761 
762     if (pvh_start_addr == 0) {
763         error_report("Error loading uncompressed kernel without PVH ELF Note");
764         exit(1);
765     }
766     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr);
767     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr);
768     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size);
769 
770     return true;
771 }
772 
773 typedef struct SetupDataFixup {
774     void *pos;
775     hwaddr orig_val, new_val;
776     uint32_t addr;
777 } SetupDataFixup;
778 
779 static void fixup_setup_data(void *opaque)
780 {
781     SetupDataFixup *fixup = opaque;
782     stq_p(fixup->pos, fixup->new_val);
783 }
784 
785 static void reset_setup_data(void *opaque)
786 {
787     SetupDataFixup *fixup = opaque;
788     stq_p(fixup->pos, fixup->orig_val);
789 }
790 
791 static void reset_rng_seed(void *opaque)
792 {
793     SetupData *setup_data = opaque;
794     qemu_guest_getrandom_nofail(setup_data->data, le32_to_cpu(setup_data->len));
795 }
796 
797 void x86_load_linux(X86MachineState *x86ms,
798                     FWCfgState *fw_cfg,
799                     int acpi_data_size,
800                     bool pvh_enabled,
801                     bool legacy_no_rng_seed)
802 {
803     bool linuxboot_dma_enabled = X86_MACHINE_GET_CLASS(x86ms)->fwcfg_dma_enabled;
804     uint16_t protocol;
805     int setup_size, kernel_size, cmdline_size;
806     int dtb_size, setup_data_offset;
807     uint32_t initrd_max;
808     uint8_t header[8192], *setup, *kernel;
809     hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0, first_setup_data = 0;
810     FILE *f;
811     char *vmode;
812     MachineState *machine = MACHINE(x86ms);
813     SetupData *setup_data;
814     const char *kernel_filename = machine->kernel_filename;
815     const char *initrd_filename = machine->initrd_filename;
816     const char *dtb_filename = machine->dtb;
817     char *kernel_cmdline;
818     SevKernelLoaderContext sev_load_ctx = {};
819     enum { RNG_SEED_LENGTH = 32 };
820 
821     /*
822      * Add the NUL terminator, some padding for the microvm cmdline fiddling
823      * hack, and then align to 16 bytes as a paranoia measure
824      */
825     cmdline_size = (strlen(machine->kernel_cmdline) + 1 +
826                     VIRTIO_CMDLINE_TOTAL_MAX_LEN + 16) & ~15;
827     /* Make a copy, since we might append arbitrary bytes to it later. */
828     kernel_cmdline = g_strndup(machine->kernel_cmdline, cmdline_size);
829 
830     /* load the kernel header */
831     f = fopen(kernel_filename, "rb");
832     if (!f) {
833         fprintf(stderr, "qemu: could not open kernel file '%s': %s\n",
834                 kernel_filename, strerror(errno));
835         exit(1);
836     }
837 
838     kernel_size = get_file_size(f);
839     if (!kernel_size ||
840         fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
841         MIN(ARRAY_SIZE(header), kernel_size)) {
842         fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
843                 kernel_filename, strerror(errno));
844         exit(1);
845     }
846 
847     /* kernel protocol version */
848     if (ldl_p(header + 0x202) == 0x53726448) {
849         protocol = lduw_p(header + 0x206);
850     } else {
851         /*
852          * This could be a multiboot kernel. If it is, let's stop treating it
853          * like a Linux kernel.
854          * Note: some multiboot images could be in the ELF format (the same of
855          * PVH), so we try multiboot first since we check the multiboot magic
856          * header before to load it.
857          */
858         if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename,
859                            kernel_cmdline, kernel_size, header)) {
860             return;
861         }
862         /*
863          * Check if the file is an uncompressed kernel file (ELF) and load it,
864          * saving the PVH entry point used by the x86/HVM direct boot ABI.
865          * If load_elfboot() is successful, populate the fw_cfg info.
866          */
867         if (pvh_enabled &&
868             load_elfboot(kernel_filename, kernel_size,
869                          header, pvh_start_addr, fw_cfg)) {
870             fclose(f);
871 
872             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
873                 strlen(kernel_cmdline) + 1);
874             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
875 
876             fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header));
877             fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA,
878                              header, sizeof(header));
879 
880             /* load initrd */
881             if (initrd_filename) {
882                 GMappedFile *mapped_file;
883                 gsize initrd_size;
884                 gchar *initrd_data;
885                 GError *gerr = NULL;
886 
887                 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
888                 if (!mapped_file) {
889                     fprintf(stderr, "qemu: error reading initrd %s: %s\n",
890                             initrd_filename, gerr->message);
891                     exit(1);
892                 }
893                 x86ms->initrd_mapped_file = mapped_file;
894 
895                 initrd_data = g_mapped_file_get_contents(mapped_file);
896                 initrd_size = g_mapped_file_get_length(mapped_file);
897                 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
898                 if (initrd_size >= initrd_max) {
899                     fprintf(stderr, "qemu: initrd is too large, cannot support."
900                             "(max: %"PRIu32", need %"PRId64")\n",
901                             initrd_max, (uint64_t)initrd_size);
902                     exit(1);
903                 }
904 
905                 initrd_addr = (initrd_max - initrd_size) & ~4095;
906 
907                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
908                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
909                 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data,
910                                  initrd_size);
911             }
912 
913             option_rom[nb_option_roms].bootindex = 0;
914             option_rom[nb_option_roms].name = "pvh.bin";
915             nb_option_roms++;
916 
917             return;
918         }
919         protocol = 0;
920     }
921 
922     if (protocol < 0x200 || !(header[0x211] & 0x01)) {
923         /* Low kernel */
924         real_addr    = 0x90000;
925         cmdline_addr = 0x9a000 - cmdline_size;
926         prot_addr    = 0x10000;
927     } else if (protocol < 0x202) {
928         /* High but ancient kernel */
929         real_addr    = 0x90000;
930         cmdline_addr = 0x9a000 - cmdline_size;
931         prot_addr    = 0x100000;
932     } else {
933         /* High and recent kernel */
934         real_addr    = 0x10000;
935         cmdline_addr = 0x20000;
936         prot_addr    = 0x100000;
937     }
938 
939     /* highest address for loading the initrd */
940     if (protocol >= 0x20c &&
941         lduw_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) {
942         /*
943          * Linux has supported initrd up to 4 GB for a very long time (2007,
944          * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
945          * though it only sets initrd_max to 2 GB to "work around bootloader
946          * bugs". Luckily, QEMU firmware(which does something like bootloader)
947          * has supported this.
948          *
949          * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
950          * be loaded into any address.
951          *
952          * In addition, initrd_max is uint32_t simply because QEMU doesn't
953          * support the 64-bit boot protocol (specifically the ext_ramdisk_image
954          * field).
955          *
956          * Therefore here just limit initrd_max to UINT32_MAX simply as well.
957          */
958         initrd_max = UINT32_MAX;
959     } else if (protocol >= 0x203) {
960         initrd_max = ldl_p(header + 0x22c);
961     } else {
962         initrd_max = 0x37ffffff;
963     }
964 
965     if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) {
966         initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
967     }
968 
969     if (protocol >= 0x202) {
970         stl_p(header + 0x228, cmdline_addr);
971     } else {
972         stw_p(header + 0x20, 0xA33F);
973         stw_p(header + 0x22, cmdline_addr - real_addr);
974     }
975 
976     /* handle vga= parameter */
977     vmode = strstr(kernel_cmdline, "vga=");
978     if (vmode) {
979         unsigned int video_mode;
980         const char *end;
981         int ret;
982         /* skip "vga=" */
983         vmode += 4;
984         if (!strncmp(vmode, "normal", 6)) {
985             video_mode = 0xffff;
986         } else if (!strncmp(vmode, "ext", 3)) {
987             video_mode = 0xfffe;
988         } else if (!strncmp(vmode, "ask", 3)) {
989             video_mode = 0xfffd;
990         } else {
991             ret = qemu_strtoui(vmode, &end, 0, &video_mode);
992             if (ret != 0 || (*end && *end != ' ')) {
993                 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n");
994                 exit(1);
995             }
996         }
997         stw_p(header + 0x1fa, video_mode);
998     }
999 
1000     /* loader type */
1001     /*
1002      * High nybble = B reserved for QEMU; low nybble is revision number.
1003      * If this code is substantially changed, you may want to consider
1004      * incrementing the revision.
1005      */
1006     if (protocol >= 0x200) {
1007         header[0x210] = 0xB0;
1008     }
1009     /* heap */
1010     if (protocol >= 0x201) {
1011         header[0x211] |= 0x80; /* CAN_USE_HEAP */
1012         stw_p(header + 0x224, cmdline_addr - real_addr - 0x200);
1013     }
1014 
1015     /* load initrd */
1016     if (initrd_filename) {
1017         GMappedFile *mapped_file;
1018         gsize initrd_size;
1019         gchar *initrd_data;
1020         GError *gerr = NULL;
1021 
1022         if (protocol < 0x200) {
1023             fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
1024             exit(1);
1025         }
1026 
1027         mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
1028         if (!mapped_file) {
1029             fprintf(stderr, "qemu: error reading initrd %s: %s\n",
1030                     initrd_filename, gerr->message);
1031             exit(1);
1032         }
1033         x86ms->initrd_mapped_file = mapped_file;
1034 
1035         initrd_data = g_mapped_file_get_contents(mapped_file);
1036         initrd_size = g_mapped_file_get_length(mapped_file);
1037         if (initrd_size >= initrd_max) {
1038             fprintf(stderr, "qemu: initrd is too large, cannot support."
1039                     "(max: %"PRIu32", need %"PRId64")\n",
1040                     initrd_max, (uint64_t)initrd_size);
1041             exit(1);
1042         }
1043 
1044         initrd_addr = (initrd_max - initrd_size) & ~4095;
1045 
1046         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
1047         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
1048         fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
1049         sev_load_ctx.initrd_data = initrd_data;
1050         sev_load_ctx.initrd_size = initrd_size;
1051 
1052         stl_p(header + 0x218, initrd_addr);
1053         stl_p(header + 0x21c, initrd_size);
1054     }
1055 
1056     /* load kernel and setup */
1057     setup_size = header[0x1f1];
1058     if (setup_size == 0) {
1059         setup_size = 4;
1060     }
1061     setup_size = (setup_size + 1) * 512;
1062     if (setup_size > kernel_size) {
1063         fprintf(stderr, "qemu: invalid kernel header\n");
1064         exit(1);
1065     }
1066     kernel_size -= setup_size;
1067 
1068     setup  = g_malloc(setup_size);
1069     kernel = g_malloc(kernel_size);
1070     fseek(f, 0, SEEK_SET);
1071     if (fread(setup, 1, setup_size, f) != setup_size) {
1072         fprintf(stderr, "fread() failed\n");
1073         exit(1);
1074     }
1075     if (fread(kernel, 1, kernel_size, f) != kernel_size) {
1076         fprintf(stderr, "fread() failed\n");
1077         exit(1);
1078     }
1079     fclose(f);
1080 
1081     /* append dtb to kernel */
1082     if (dtb_filename) {
1083         if (protocol < 0x209) {
1084             fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n");
1085             exit(1);
1086         }
1087 
1088         dtb_size = get_image_size(dtb_filename);
1089         if (dtb_size <= 0) {
1090             fprintf(stderr, "qemu: error reading dtb %s: %s\n",
1091                     dtb_filename, strerror(errno));
1092             exit(1);
1093         }
1094 
1095         setup_data_offset = cmdline_size;
1096         cmdline_size += sizeof(SetupData) + dtb_size;
1097         kernel_cmdline = g_realloc(kernel_cmdline, cmdline_size);
1098         setup_data = (void *)kernel_cmdline + setup_data_offset;
1099         setup_data->next = cpu_to_le64(first_setup_data);
1100         first_setup_data = cmdline_addr + setup_data_offset;
1101         setup_data->type = cpu_to_le32(SETUP_DTB);
1102         setup_data->len = cpu_to_le32(dtb_size);
1103         load_image_size(dtb_filename, setup_data->data, dtb_size);
1104     }
1105 
1106     if (!legacy_no_rng_seed && protocol >= 0x209) {
1107         setup_data_offset = cmdline_size;
1108         cmdline_size += sizeof(SetupData) + RNG_SEED_LENGTH;
1109         kernel_cmdline = g_realloc(kernel_cmdline, cmdline_size);
1110         setup_data = (void *)kernel_cmdline + setup_data_offset;
1111         setup_data->next = cpu_to_le64(first_setup_data);
1112         first_setup_data = cmdline_addr + setup_data_offset;
1113         setup_data->type = cpu_to_le32(SETUP_RNG_SEED);
1114         setup_data->len = cpu_to_le32(RNG_SEED_LENGTH);
1115         qemu_guest_getrandom_nofail(setup_data->data, RNG_SEED_LENGTH);
1116         qemu_register_reset_nosnapshotload(reset_rng_seed, setup_data);
1117         fw_cfg_add_bytes_callback(fw_cfg, FW_CFG_KERNEL_DATA, reset_rng_seed, NULL,
1118                                   setup_data, kernel, kernel_size, true);
1119     } else {
1120         fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA, kernel, kernel_size);
1121     }
1122 
1123     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
1124     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, cmdline_size);
1125     fw_cfg_add_bytes(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline, cmdline_size);
1126     sev_load_ctx.cmdline_data = (char *)kernel_cmdline;
1127     sev_load_ctx.cmdline_size = cmdline_size;
1128 
1129     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
1130     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size);
1131     sev_load_ctx.kernel_data = (char *)kernel;
1132     sev_load_ctx.kernel_size = kernel_size;
1133 
1134     /*
1135      * If we're starting an encrypted VM, it will be OVMF based, which uses the
1136      * efi stub for booting and doesn't require any values to be placed in the
1137      * kernel header.  We therefore don't update the header so the hash of the
1138      * kernel on the other side of the fw_cfg interface matches the hash of the
1139      * file the user passed in.
1140      */
1141     if (!sev_enabled() && first_setup_data) {
1142         SetupDataFixup *fixup = g_malloc(sizeof(*fixup));
1143 
1144         memcpy(setup, header, MIN(sizeof(header), setup_size));
1145         /* Offset 0x250 is a pointer to the first setup_data link. */
1146         fixup->pos = setup + 0x250;
1147         fixup->orig_val = ldq_p(fixup->pos);
1148         fixup->new_val = first_setup_data;
1149         fixup->addr = cpu_to_le32(real_addr);
1150         fw_cfg_add_bytes_callback(fw_cfg, FW_CFG_SETUP_ADDR, fixup_setup_data, NULL,
1151                                   fixup, &fixup->addr, sizeof(fixup->addr), true);
1152         qemu_register_reset(reset_setup_data, fixup);
1153     } else {
1154         fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
1155     }
1156     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
1157     fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
1158     sev_load_ctx.setup_data = (char *)setup;
1159     sev_load_ctx.setup_size = setup_size;
1160 
1161     if (sev_enabled()) {
1162         sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal);
1163     }
1164 
1165     option_rom[nb_option_roms].bootindex = 0;
1166     option_rom[nb_option_roms].name = "linuxboot.bin";
1167     if (linuxboot_dma_enabled && fw_cfg_dma_enabled(fw_cfg)) {
1168         option_rom[nb_option_roms].name = "linuxboot_dma.bin";
1169     }
1170     nb_option_roms++;
1171 }
1172 
1173 void x86_bios_rom_init(MachineState *ms, const char *default_firmware,
1174                        MemoryRegion *rom_memory, bool isapc_ram_fw)
1175 {
1176     const char *bios_name;
1177     char *filename;
1178     MemoryRegion *bios, *isa_bios;
1179     int bios_size, isa_bios_size;
1180     ssize_t ret;
1181 
1182     /* BIOS load */
1183     bios_name = ms->firmware ?: default_firmware;
1184     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1185     if (filename) {
1186         bios_size = get_image_size(filename);
1187     } else {
1188         bios_size = -1;
1189     }
1190     if (bios_size <= 0 ||
1191         (bios_size % 65536) != 0) {
1192         goto bios_error;
1193     }
1194     bios = g_malloc(sizeof(*bios));
1195     memory_region_init_ram(bios, NULL, "pc.bios", bios_size, &error_fatal);
1196     if (sev_enabled()) {
1197         /*
1198          * The concept of a "reset" simply doesn't exist for
1199          * confidential computing guests, we have to destroy and
1200          * re-launch them instead.  So there is no need to register
1201          * the firmware as rom to properly re-initialize on reset.
1202          * Just go for a straight file load instead.
1203          */
1204         void *ptr = memory_region_get_ram_ptr(bios);
1205         load_image_size(filename, ptr, bios_size);
1206         x86_firmware_configure(ptr, bios_size);
1207     } else {
1208         if (!isapc_ram_fw) {
1209             memory_region_set_readonly(bios, true);
1210         }
1211         ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1);
1212         if (ret != 0) {
1213             goto bios_error;
1214         }
1215     }
1216     g_free(filename);
1217 
1218     /* map the last 128KB of the BIOS in ISA space */
1219     isa_bios_size = MIN(bios_size, 128 * KiB);
1220     isa_bios = g_malloc(sizeof(*isa_bios));
1221     memory_region_init_alias(isa_bios, NULL, "isa-bios", bios,
1222                              bios_size - isa_bios_size, isa_bios_size);
1223     memory_region_add_subregion_overlap(rom_memory,
1224                                         0x100000 - isa_bios_size,
1225                                         isa_bios,
1226                                         1);
1227     if (!isapc_ram_fw) {
1228         memory_region_set_readonly(isa_bios, true);
1229     }
1230 
1231     /* map all the bios at the top of memory */
1232     memory_region_add_subregion(rom_memory,
1233                                 (uint32_t)(-bios_size),
1234                                 bios);
1235     return;
1236 
1237 bios_error:
1238     fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1239     exit(1);
1240 }
1241 
1242 bool x86_machine_is_smm_enabled(const X86MachineState *x86ms)
1243 {
1244     bool smm_available = false;
1245 
1246     if (x86ms->smm == ON_OFF_AUTO_OFF) {
1247         return false;
1248     }
1249 
1250     if (tcg_enabled() || qtest_enabled()) {
1251         smm_available = true;
1252     } else if (kvm_enabled()) {
1253         smm_available = kvm_has_smm();
1254     }
1255 
1256     if (smm_available) {
1257         return true;
1258     }
1259 
1260     if (x86ms->smm == ON_OFF_AUTO_ON) {
1261         error_report("System Management Mode not supported by this hypervisor.");
1262         exit(1);
1263     }
1264     return false;
1265 }
1266 
1267 static void x86_machine_get_smm(Object *obj, Visitor *v, const char *name,
1268                                void *opaque, Error **errp)
1269 {
1270     X86MachineState *x86ms = X86_MACHINE(obj);
1271     OnOffAuto smm = x86ms->smm;
1272 
1273     visit_type_OnOffAuto(v, name, &smm, errp);
1274 }
1275 
1276 static void x86_machine_set_smm(Object *obj, Visitor *v, const char *name,
1277                                void *opaque, Error **errp)
1278 {
1279     X86MachineState *x86ms = X86_MACHINE(obj);
1280 
1281     visit_type_OnOffAuto(v, name, &x86ms->smm, errp);
1282 }
1283 
1284 bool x86_machine_is_acpi_enabled(const X86MachineState *x86ms)
1285 {
1286     if (x86ms->acpi == ON_OFF_AUTO_OFF) {
1287         return false;
1288     }
1289     return true;
1290 }
1291 
1292 static void x86_machine_get_acpi(Object *obj, Visitor *v, const char *name,
1293                                  void *opaque, Error **errp)
1294 {
1295     X86MachineState *x86ms = X86_MACHINE(obj);
1296     OnOffAuto acpi = x86ms->acpi;
1297 
1298     visit_type_OnOffAuto(v, name, &acpi, errp);
1299 }
1300 
1301 static void x86_machine_set_acpi(Object *obj, Visitor *v, const char *name,
1302                                  void *opaque, Error **errp)
1303 {
1304     X86MachineState *x86ms = X86_MACHINE(obj);
1305 
1306     visit_type_OnOffAuto(v, name, &x86ms->acpi, errp);
1307 }
1308 
1309 static void x86_machine_get_pit(Object *obj, Visitor *v, const char *name,
1310                                     void *opaque, Error **errp)
1311 {
1312     X86MachineState *x86ms = X86_MACHINE(obj);
1313     OnOffAuto pit = x86ms->pit;
1314 
1315     visit_type_OnOffAuto(v, name, &pit, errp);
1316 }
1317 
1318 static void x86_machine_set_pit(Object *obj, Visitor *v, const char *name,
1319                                     void *opaque, Error **errp)
1320 {
1321     X86MachineState *x86ms = X86_MACHINE(obj);;
1322 
1323     visit_type_OnOffAuto(v, name, &x86ms->pit, errp);
1324 }
1325 
1326 static void x86_machine_get_pic(Object *obj, Visitor *v, const char *name,
1327                                 void *opaque, Error **errp)
1328 {
1329     X86MachineState *x86ms = X86_MACHINE(obj);
1330     OnOffAuto pic = x86ms->pic;
1331 
1332     visit_type_OnOffAuto(v, name, &pic, errp);
1333 }
1334 
1335 static void x86_machine_set_pic(Object *obj, Visitor *v, const char *name,
1336                                 void *opaque, Error **errp)
1337 {
1338     X86MachineState *x86ms = X86_MACHINE(obj);
1339 
1340     visit_type_OnOffAuto(v, name, &x86ms->pic, errp);
1341 }
1342 
1343 static char *x86_machine_get_oem_id(Object *obj, Error **errp)
1344 {
1345     X86MachineState *x86ms = X86_MACHINE(obj);
1346 
1347     return g_strdup(x86ms->oem_id);
1348 }
1349 
1350 static void x86_machine_set_oem_id(Object *obj, const char *value, Error **errp)
1351 {
1352     X86MachineState *x86ms = X86_MACHINE(obj);
1353     size_t len = strlen(value);
1354 
1355     if (len > 6) {
1356         error_setg(errp,
1357                    "User specified "X86_MACHINE_OEM_ID" value is bigger than "
1358                    "6 bytes in size");
1359         return;
1360     }
1361 
1362     strncpy(x86ms->oem_id, value, 6);
1363 }
1364 
1365 static char *x86_machine_get_oem_table_id(Object *obj, Error **errp)
1366 {
1367     X86MachineState *x86ms = X86_MACHINE(obj);
1368 
1369     return g_strdup(x86ms->oem_table_id);
1370 }
1371 
1372 static void x86_machine_set_oem_table_id(Object *obj, const char *value,
1373                                          Error **errp)
1374 {
1375     X86MachineState *x86ms = X86_MACHINE(obj);
1376     size_t len = strlen(value);
1377 
1378     if (len > 8) {
1379         error_setg(errp,
1380                    "User specified "X86_MACHINE_OEM_TABLE_ID
1381                    " value is bigger than "
1382                    "8 bytes in size");
1383         return;
1384     }
1385     strncpy(x86ms->oem_table_id, value, 8);
1386 }
1387 
1388 static void x86_machine_get_bus_lock_ratelimit(Object *obj, Visitor *v,
1389                                 const char *name, void *opaque, Error **errp)
1390 {
1391     X86MachineState *x86ms = X86_MACHINE(obj);
1392     uint64_t bus_lock_ratelimit = x86ms->bus_lock_ratelimit;
1393 
1394     visit_type_uint64(v, name, &bus_lock_ratelimit, errp);
1395 }
1396 
1397 static void x86_machine_set_bus_lock_ratelimit(Object *obj, Visitor *v,
1398                                const char *name, void *opaque, Error **errp)
1399 {
1400     X86MachineState *x86ms = X86_MACHINE(obj);
1401 
1402     visit_type_uint64(v, name, &x86ms->bus_lock_ratelimit, errp);
1403 }
1404 
1405 static void machine_get_sgx_epc(Object *obj, Visitor *v, const char *name,
1406                                 void *opaque, Error **errp)
1407 {
1408     X86MachineState *x86ms = X86_MACHINE(obj);
1409     SgxEPCList *list = x86ms->sgx_epc_list;
1410 
1411     visit_type_SgxEPCList(v, name, &list, errp);
1412 }
1413 
1414 static void machine_set_sgx_epc(Object *obj, Visitor *v, const char *name,
1415                                 void *opaque, Error **errp)
1416 {
1417     X86MachineState *x86ms = X86_MACHINE(obj);
1418     SgxEPCList *list;
1419 
1420     list = x86ms->sgx_epc_list;
1421     visit_type_SgxEPCList(v, name, &x86ms->sgx_epc_list, errp);
1422 
1423     qapi_free_SgxEPCList(list);
1424 }
1425 
1426 static void x86_machine_initfn(Object *obj)
1427 {
1428     X86MachineState *x86ms = X86_MACHINE(obj);
1429 
1430     x86ms->smm = ON_OFF_AUTO_AUTO;
1431     x86ms->acpi = ON_OFF_AUTO_AUTO;
1432     x86ms->pit = ON_OFF_AUTO_AUTO;
1433     x86ms->pic = ON_OFF_AUTO_AUTO;
1434     x86ms->pci_irq_mask = ACPI_BUILD_PCI_IRQS;
1435     x86ms->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
1436     x86ms->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
1437     x86ms->bus_lock_ratelimit = 0;
1438     x86ms->above_4g_mem_start = 4 * GiB;
1439 }
1440 
1441 static void x86_machine_class_init(ObjectClass *oc, void *data)
1442 {
1443     MachineClass *mc = MACHINE_CLASS(oc);
1444     X86MachineClass *x86mc = X86_MACHINE_CLASS(oc);
1445     NMIClass *nc = NMI_CLASS(oc);
1446 
1447     mc->cpu_index_to_instance_props = x86_cpu_index_to_props;
1448     mc->get_default_cpu_node_id = x86_get_default_cpu_node_id;
1449     mc->possible_cpu_arch_ids = x86_possible_cpu_arch_ids;
1450     x86mc->save_tsc_khz = true;
1451     x86mc->fwcfg_dma_enabled = true;
1452     nc->nmi_monitor_handler = x86_nmi;
1453 
1454     object_class_property_add(oc, X86_MACHINE_SMM, "OnOffAuto",
1455         x86_machine_get_smm, x86_machine_set_smm,
1456         NULL, NULL);
1457     object_class_property_set_description(oc, X86_MACHINE_SMM,
1458         "Enable SMM");
1459 
1460     object_class_property_add(oc, X86_MACHINE_ACPI, "OnOffAuto",
1461         x86_machine_get_acpi, x86_machine_set_acpi,
1462         NULL, NULL);
1463     object_class_property_set_description(oc, X86_MACHINE_ACPI,
1464         "Enable ACPI");
1465 
1466     object_class_property_add(oc, X86_MACHINE_PIT, "OnOffAuto",
1467                               x86_machine_get_pit,
1468                               x86_machine_set_pit,
1469                               NULL, NULL);
1470     object_class_property_set_description(oc, X86_MACHINE_PIT,
1471         "Enable i8254 PIT");
1472 
1473     object_class_property_add(oc, X86_MACHINE_PIC, "OnOffAuto",
1474                               x86_machine_get_pic,
1475                               x86_machine_set_pic,
1476                               NULL, NULL);
1477     object_class_property_set_description(oc, X86_MACHINE_PIC,
1478         "Enable i8259 PIC");
1479 
1480     object_class_property_add_str(oc, X86_MACHINE_OEM_ID,
1481                                   x86_machine_get_oem_id,
1482                                   x86_machine_set_oem_id);
1483     object_class_property_set_description(oc, X86_MACHINE_OEM_ID,
1484                                           "Override the default value of field OEMID "
1485                                           "in ACPI table header."
1486                                           "The string may be up to 6 bytes in size");
1487 
1488 
1489     object_class_property_add_str(oc, X86_MACHINE_OEM_TABLE_ID,
1490                                   x86_machine_get_oem_table_id,
1491                                   x86_machine_set_oem_table_id);
1492     object_class_property_set_description(oc, X86_MACHINE_OEM_TABLE_ID,
1493                                           "Override the default value of field OEM Table ID "
1494                                           "in ACPI table header."
1495                                           "The string may be up to 8 bytes in size");
1496 
1497     object_class_property_add(oc, X86_MACHINE_BUS_LOCK_RATELIMIT, "uint64_t",
1498                                 x86_machine_get_bus_lock_ratelimit,
1499                                 x86_machine_set_bus_lock_ratelimit, NULL, NULL);
1500     object_class_property_set_description(oc, X86_MACHINE_BUS_LOCK_RATELIMIT,
1501             "Set the ratelimit for the bus locks acquired in VMs");
1502 
1503     object_class_property_add(oc, "sgx-epc", "SgxEPC",
1504         machine_get_sgx_epc, machine_set_sgx_epc,
1505         NULL, NULL);
1506     object_class_property_set_description(oc, "sgx-epc",
1507         "SGX EPC device");
1508 }
1509 
1510 static const TypeInfo x86_machine_info = {
1511     .name = TYPE_X86_MACHINE,
1512     .parent = TYPE_MACHINE,
1513     .abstract = true,
1514     .instance_size = sizeof(X86MachineState),
1515     .instance_init = x86_machine_initfn,
1516     .class_size = sizeof(X86MachineClass),
1517     .class_init = x86_machine_class_init,
1518     .interfaces = (InterfaceInfo[]) {
1519          { TYPE_NMI },
1520          { }
1521     },
1522 };
1523 
1524 static void x86_machine_register_types(void)
1525 {
1526     type_register_static(&x86_machine_info);
1527 }
1528 
1529 type_init(x86_machine_register_types)
1530