xref: /openbmc/qemu/hw/i386/x86.c (revision c16de0d9)
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 "qapi/error.h"
30 #include "qapi/qapi-visit-common.h"
31 #include "qapi/clone-visitor.h"
32 #include "qapi/qapi-visit-machine.h"
33 #include "qapi/visitor.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/whpx.h"
36 #include "sysemu/numa.h"
37 #include "sysemu/replay.h"
38 #include "sysemu/sysemu.h"
39 #include "sysemu/cpu-timers.h"
40 #include "sysemu/xen.h"
41 #include "trace.h"
42 
43 #include "hw/i386/x86.h"
44 #include "target/i386/cpu.h"
45 #include "hw/i386/topology.h"
46 #include "hw/i386/fw_cfg.h"
47 #include "hw/intc/i8259.h"
48 #include "hw/rtc/mc146818rtc.h"
49 #include "target/i386/sev.h"
50 
51 #include "hw/acpi/cpu_hotplug.h"
52 #include "hw/irq.h"
53 #include "hw/nmi.h"
54 #include "hw/loader.h"
55 #include "multiboot.h"
56 #include "elf.h"
57 #include "standard-headers/asm-x86/bootparam.h"
58 #include CONFIG_DEVICES
59 #include "kvm/kvm_i386.h"
60 
61 #ifdef CONFIG_XEN_EMU
62 #include "hw/xen/xen.h"
63 #include "hw/i386/kvm/xen_evtchn.h"
64 #endif
65 
66 /* Physical Address of PVH entry point read from kernel ELF NOTE */
67 static size_t pvh_start_addr;
68 
69 static void init_topo_info(X86CPUTopoInfo *topo_info,
70                            const X86MachineState *x86ms)
71 {
72     MachineState *ms = MACHINE(x86ms);
73 
74     topo_info->dies_per_pkg = ms->smp.dies;
75     topo_info->cores_per_die = ms->smp.cores;
76     topo_info->threads_per_core = ms->smp.threads;
77 }
78 
79 /*
80  * Calculates initial APIC ID for a specific CPU index
81  *
82  * Currently we need to be able to calculate the APIC ID from the CPU index
83  * alone (without requiring a CPU object), as the QEMU<->Seabios interfaces have
84  * no concept of "CPU index", and the NUMA tables on fw_cfg need the APIC ID of
85  * all CPUs up to max_cpus.
86  */
87 uint32_t x86_cpu_apic_id_from_index(X86MachineState *x86ms,
88                                     unsigned int cpu_index)
89 {
90     X86CPUTopoInfo topo_info;
91 
92     init_topo_info(&topo_info, x86ms);
93 
94     return x86_apicid_from_cpu_idx(&topo_info, cpu_index);
95 }
96 
97 
98 void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp)
99 {
100     Object *cpu = object_new(MACHINE(x86ms)->cpu_type);
101 
102     if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) {
103         goto out;
104     }
105     qdev_realize(DEVICE(cpu), NULL, errp);
106 
107 out:
108     object_unref(cpu);
109 }
110 
111 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version)
112 {
113     int i;
114     const CPUArchIdList *possible_cpus;
115     MachineState *ms = MACHINE(x86ms);
116     MachineClass *mc = MACHINE_GET_CLASS(x86ms);
117 
118     x86_cpu_set_default_version(default_cpu_version);
119 
120     /*
121      * Calculates the limit to CPU APIC ID values
122      *
123      * Limit for the APIC ID value, so that all
124      * CPU APIC IDs are < x86ms->apic_id_limit.
125      *
126      * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
127      */
128     x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms,
129                                                       ms->smp.max_cpus - 1) + 1;
130 
131     /*
132      * Can we support APIC ID 255 or higher?  With KVM, that requires
133      * both in-kernel lapic and X2APIC userspace API.
134      *
135      * kvm_enabled() must go first to ensure that kvm_* references are
136      * not emitted for the linker to consume (kvm_enabled() is
137      * a literal `0` in configurations where kvm_* aren't defined)
138      */
139     if (kvm_enabled() && x86ms->apic_id_limit > 255 &&
140         (!kvm_irqchip_in_kernel() || !kvm_enable_x2apic())) {
141         error_report("current -smp configuration requires kernel "
142                      "irqchip and X2APIC API support.");
143         exit(EXIT_FAILURE);
144     }
145 
146     if (kvm_enabled()) {
147         kvm_set_max_apic_id(x86ms->apic_id_limit);
148     }
149 
150     possible_cpus = mc->possible_cpu_arch_ids(ms);
151     for (i = 0; i < ms->smp.cpus; i++) {
152         x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal);
153     }
154 }
155 
156 void x86_rtc_set_cpus_count(ISADevice *s, uint16_t cpus_count)
157 {
158     MC146818RtcState *rtc = MC146818_RTC(s);
159 
160     if (cpus_count > 0xff) {
161         /*
162          * If the number of CPUs can't be represented in 8 bits, the
163          * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just
164          * to make old BIOSes fail more predictably.
165          */
166         mc146818rtc_set_cmos_data(rtc, 0x5f, 0);
167     } else {
168         mc146818rtc_set_cmos_data(rtc, 0x5f, cpus_count - 1);
169     }
170 }
171 
172 static int x86_apic_cmp(const void *a, const void *b)
173 {
174    CPUArchId *apic_a = (CPUArchId *)a;
175    CPUArchId *apic_b = (CPUArchId *)b;
176 
177    return apic_a->arch_id - apic_b->arch_id;
178 }
179 
180 /*
181  * returns pointer to CPUArchId descriptor that matches CPU's apic_id
182  * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no
183  * entry corresponding to CPU's apic_id returns NULL.
184  */
185 CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
186 {
187     CPUArchId apic_id, *found_cpu;
188 
189     apic_id.arch_id = id;
190     found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus,
191         ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus),
192         x86_apic_cmp);
193     if (found_cpu && idx) {
194         *idx = found_cpu - ms->possible_cpus->cpus;
195     }
196     return found_cpu;
197 }
198 
199 void x86_cpu_plug(HotplugHandler *hotplug_dev,
200                   DeviceState *dev, Error **errp)
201 {
202     CPUArchId *found_cpu;
203     Error *local_err = NULL;
204     X86CPU *cpu = X86_CPU(dev);
205     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
206 
207     if (x86ms->acpi_dev) {
208         hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err);
209         if (local_err) {
210             goto out;
211         }
212     }
213 
214     /* increment the number of CPUs */
215     x86ms->boot_cpus++;
216     if (x86ms->rtc) {
217         x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
218     }
219     if (x86ms->fw_cfg) {
220         fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
221     }
222 
223     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
224     found_cpu->cpu = OBJECT(dev);
225 out:
226     error_propagate(errp, local_err);
227 }
228 
229 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev,
230                                DeviceState *dev, Error **errp)
231 {
232     int idx = -1;
233     X86CPU *cpu = X86_CPU(dev);
234     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
235 
236     if (!x86ms->acpi_dev) {
237         error_setg(errp, "CPU hot unplug not supported without ACPI");
238         return;
239     }
240 
241     x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
242     assert(idx != -1);
243     if (idx == 0) {
244         error_setg(errp, "Boot CPU is unpluggable");
245         return;
246     }
247 
248     hotplug_handler_unplug_request(x86ms->acpi_dev, dev,
249                                    errp);
250 }
251 
252 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev,
253                        DeviceState *dev, Error **errp)
254 {
255     CPUArchId *found_cpu;
256     Error *local_err = NULL;
257     X86CPU *cpu = X86_CPU(dev);
258     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
259 
260     hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err);
261     if (local_err) {
262         goto out;
263     }
264 
265     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
266     found_cpu->cpu = NULL;
267     qdev_unrealize(dev);
268 
269     /* decrement the number of CPUs */
270     x86ms->boot_cpus--;
271     /* Update the number of CPUs in CMOS */
272     x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
273     fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
274  out:
275     error_propagate(errp, local_err);
276 }
277 
278 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev,
279                       DeviceState *dev, Error **errp)
280 {
281     int idx;
282     CPUState *cs;
283     CPUArchId *cpu_slot;
284     X86CPUTopoIDs topo_ids;
285     X86CPU *cpu = X86_CPU(dev);
286     CPUX86State *env = &cpu->env;
287     MachineState *ms = MACHINE(hotplug_dev);
288     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
289     unsigned int smp_cores = ms->smp.cores;
290     unsigned int smp_threads = ms->smp.threads;
291     X86CPUTopoInfo topo_info;
292 
293     if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) {
294         error_setg(errp, "Invalid CPU type, expected cpu type: '%s'",
295                    ms->cpu_type);
296         return;
297     }
298 
299     if (x86ms->acpi_dev) {
300         Error *local_err = NULL;
301 
302         hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev,
303                                  &local_err);
304         if (local_err) {
305             error_propagate(errp, local_err);
306             return;
307         }
308     }
309 
310     init_topo_info(&topo_info, x86ms);
311 
312     env->nr_dies = ms->smp.dies;
313 
314     /*
315      * If APIC ID is not set,
316      * set it based on socket/die/core/thread properties.
317      */
318     if (cpu->apic_id == UNASSIGNED_APIC_ID) {
319         int max_socket = (ms->smp.max_cpus - 1) /
320                                 smp_threads / smp_cores / ms->smp.dies;
321 
322         /*
323          * die-id was optional in QEMU 4.0 and older, so keep it optional
324          * if there's only one die per socket.
325          */
326         if (cpu->die_id < 0 && ms->smp.dies == 1) {
327             cpu->die_id = 0;
328         }
329 
330         if (cpu->socket_id < 0) {
331             error_setg(errp, "CPU socket-id is not set");
332             return;
333         } else if (cpu->socket_id > max_socket) {
334             error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u",
335                        cpu->socket_id, max_socket);
336             return;
337         }
338         if (cpu->die_id < 0) {
339             error_setg(errp, "CPU die-id is not set");
340             return;
341         } else if (cpu->die_id > ms->smp.dies - 1) {
342             error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u",
343                        cpu->die_id, ms->smp.dies - 1);
344             return;
345         }
346         if (cpu->core_id < 0) {
347             error_setg(errp, "CPU core-id is not set");
348             return;
349         } else if (cpu->core_id > (smp_cores - 1)) {
350             error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u",
351                        cpu->core_id, smp_cores - 1);
352             return;
353         }
354         if (cpu->thread_id < 0) {
355             error_setg(errp, "CPU thread-id is not set");
356             return;
357         } else if (cpu->thread_id > (smp_threads - 1)) {
358             error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u",
359                        cpu->thread_id, smp_threads - 1);
360             return;
361         }
362 
363         topo_ids.pkg_id = cpu->socket_id;
364         topo_ids.die_id = cpu->die_id;
365         topo_ids.core_id = cpu->core_id;
366         topo_ids.smt_id = cpu->thread_id;
367         cpu->apic_id = x86_apicid_from_topo_ids(&topo_info, &topo_ids);
368     }
369 
370     cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
371     if (!cpu_slot) {
372         x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
373         error_setg(errp,
374             "Invalid CPU [socket: %u, die: %u, core: %u, thread: %u] with"
375             " APIC ID %" PRIu32 ", valid index range 0:%d",
376             topo_ids.pkg_id, topo_ids.die_id, topo_ids.core_id, topo_ids.smt_id,
377             cpu->apic_id, ms->possible_cpus->len - 1);
378         return;
379     }
380 
381     if (cpu_slot->cpu) {
382         error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists",
383                    idx, cpu->apic_id);
384         return;
385     }
386 
387     /* if 'address' properties socket-id/core-id/thread-id are not set, set them
388      * so that machine_query_hotpluggable_cpus would show correct values
389      */
390     /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
391      * once -smp refactoring is complete and there will be CPU private
392      * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
393     x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
394     if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) {
395         error_setg(errp, "property socket-id: %u doesn't match set apic-id:"
396             " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id,
397             topo_ids.pkg_id);
398         return;
399     }
400     cpu->socket_id = topo_ids.pkg_id;
401 
402     if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) {
403         error_setg(errp, "property die-id: %u doesn't match set apic-id:"
404             " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id);
405         return;
406     }
407     cpu->die_id = topo_ids.die_id;
408 
409     if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) {
410         error_setg(errp, "property core-id: %u doesn't match set apic-id:"
411             " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id,
412             topo_ids.core_id);
413         return;
414     }
415     cpu->core_id = topo_ids.core_id;
416 
417     if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) {
418         error_setg(errp, "property thread-id: %u doesn't match set apic-id:"
419             " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id,
420             topo_ids.smt_id);
421         return;
422     }
423     cpu->thread_id = topo_ids.smt_id;
424 
425     /*
426     * kvm_enabled() must go first to ensure that kvm_* references are
427     * not emitted for the linker to consume (kvm_enabled() is
428     * a literal `0` in configurations where kvm_* aren't defined)
429     */
430     if (kvm_enabled() && hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) &&
431         !kvm_hv_vpindex_settable()) {
432         error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX");
433         return;
434     }
435 
436     cs = CPU(cpu);
437     cs->cpu_index = idx;
438 
439     numa_cpu_pre_plug(cpu_slot, dev, errp);
440 }
441 
442 CpuInstanceProperties
443 x86_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
444 {
445     MachineClass *mc = MACHINE_GET_CLASS(ms);
446     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
447 
448     assert(cpu_index < possible_cpus->len);
449     return possible_cpus->cpus[cpu_index].props;
450 }
451 
452 int64_t x86_get_default_cpu_node_id(const MachineState *ms, int idx)
453 {
454    X86CPUTopoIDs topo_ids;
455    X86MachineState *x86ms = X86_MACHINE(ms);
456    X86CPUTopoInfo topo_info;
457 
458    init_topo_info(&topo_info, x86ms);
459 
460    assert(idx < ms->possible_cpus->len);
461    x86_topo_ids_from_apicid(ms->possible_cpus->cpus[idx].arch_id,
462                             &topo_info, &topo_ids);
463    return topo_ids.pkg_id % ms->numa_state->num_nodes;
464 }
465 
466 const CPUArchIdList *x86_possible_cpu_arch_ids(MachineState *ms)
467 {
468     X86MachineState *x86ms = X86_MACHINE(ms);
469     unsigned int max_cpus = ms->smp.max_cpus;
470     X86CPUTopoInfo topo_info;
471     int i;
472 
473     if (ms->possible_cpus) {
474         /*
475          * make sure that max_cpus hasn't changed since the first use, i.e.
476          * -smp hasn't been parsed after it
477          */
478         assert(ms->possible_cpus->len == max_cpus);
479         return ms->possible_cpus;
480     }
481 
482     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
483                                   sizeof(CPUArchId) * max_cpus);
484     ms->possible_cpus->len = max_cpus;
485 
486     init_topo_info(&topo_info, x86ms);
487 
488     for (i = 0; i < ms->possible_cpus->len; i++) {
489         X86CPUTopoIDs topo_ids;
490 
491         ms->possible_cpus->cpus[i].type = ms->cpu_type;
492         ms->possible_cpus->cpus[i].vcpus_count = 1;
493         ms->possible_cpus->cpus[i].arch_id =
494             x86_cpu_apic_id_from_index(x86ms, i);
495         x86_topo_ids_from_apicid(ms->possible_cpus->cpus[i].arch_id,
496                                  &topo_info, &topo_ids);
497         ms->possible_cpus->cpus[i].props.has_socket_id = true;
498         ms->possible_cpus->cpus[i].props.socket_id = topo_ids.pkg_id;
499         if (ms->smp.dies > 1) {
500             ms->possible_cpus->cpus[i].props.has_die_id = true;
501             ms->possible_cpus->cpus[i].props.die_id = topo_ids.die_id;
502         }
503         ms->possible_cpus->cpus[i].props.has_core_id = true;
504         ms->possible_cpus->cpus[i].props.core_id = topo_ids.core_id;
505         ms->possible_cpus->cpus[i].props.has_thread_id = true;
506         ms->possible_cpus->cpus[i].props.thread_id = topo_ids.smt_id;
507     }
508     return ms->possible_cpus;
509 }
510 
511 static void x86_nmi(NMIState *n, int cpu_index, Error **errp)
512 {
513     /* cpu index isn't used */
514     CPUState *cs;
515 
516     CPU_FOREACH(cs) {
517         X86CPU *cpu = X86_CPU(cs);
518 
519         if (!cpu->apic_state) {
520             cpu_interrupt(cs, CPU_INTERRUPT_NMI);
521         } else {
522             apic_deliver_nmi(cpu->apic_state);
523         }
524     }
525 }
526 
527 static long get_file_size(FILE *f)
528 {
529     long where, size;
530 
531     /* XXX: on Unix systems, using fstat() probably makes more sense */
532 
533     where = ftell(f);
534     fseek(f, 0, SEEK_END);
535     size = ftell(f);
536     fseek(f, where, SEEK_SET);
537 
538     return size;
539 }
540 
541 /* TSC handling */
542 uint64_t cpu_get_tsc(CPUX86State *env)
543 {
544     return cpus_get_elapsed_ticks();
545 }
546 
547 /* IRQ handling */
548 static void pic_irq_request(void *opaque, int irq, int level)
549 {
550     CPUState *cs = first_cpu;
551     X86CPU *cpu = X86_CPU(cs);
552 
553     trace_x86_pic_interrupt(irq, level);
554     if (cpu->apic_state && !kvm_irqchip_in_kernel() &&
555         !whpx_apic_in_platform()) {
556         CPU_FOREACH(cs) {
557             cpu = X86_CPU(cs);
558             if (apic_accept_pic_intr(cpu->apic_state)) {
559                 apic_deliver_pic_intr(cpu->apic_state, level);
560             }
561         }
562     } else {
563         if (level) {
564             cpu_interrupt(cs, CPU_INTERRUPT_HARD);
565         } else {
566             cpu_reset_interrupt(cs, CPU_INTERRUPT_HARD);
567         }
568     }
569 }
570 
571 qemu_irq x86_allocate_cpu_irq(void)
572 {
573     return qemu_allocate_irq(pic_irq_request, NULL, 0);
574 }
575 
576 int cpu_get_pic_interrupt(CPUX86State *env)
577 {
578     X86CPU *cpu = env_archcpu(env);
579     int intno;
580 
581     if (!kvm_irqchip_in_kernel() && !whpx_apic_in_platform()) {
582         intno = apic_get_interrupt(cpu->apic_state);
583         if (intno >= 0) {
584             return intno;
585         }
586         /* read the irq from the PIC */
587         if (!apic_accept_pic_intr(cpu->apic_state)) {
588             return -1;
589         }
590     }
591 
592     intno = pic_read_irq(isa_pic);
593     return intno;
594 }
595 
596 DeviceState *cpu_get_current_apic(void)
597 {
598     if (current_cpu) {
599         X86CPU *cpu = X86_CPU(current_cpu);
600         return cpu->apic_state;
601     } else {
602         return NULL;
603     }
604 }
605 
606 void gsi_handler(void *opaque, int n, int level)
607 {
608     GSIState *s = opaque;
609 
610     trace_x86_gsi_interrupt(n, level);
611     switch (n) {
612     case 0 ... ISA_NUM_IRQS - 1:
613         if (s->i8259_irq[n]) {
614             /* Under KVM, Kernel will forward to both PIC and IOAPIC */
615             qemu_set_irq(s->i8259_irq[n], level);
616         }
617         /* fall through */
618     case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
619 #ifdef CONFIG_XEN_EMU
620         /*
621          * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
622          * routing actually works properly under Xen). And then to
623          * *either* the PIRQ handling or the I/OAPIC depending on
624          * whether the former wants it.
625          */
626         if (xen_mode == XEN_EMULATE && xen_evtchn_set_gsi(n, level)) {
627             break;
628         }
629 #endif
630         qemu_set_irq(s->ioapic_irq[n], level);
631         break;
632     case IO_APIC_SECONDARY_IRQBASE
633         ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1:
634         qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level);
635         break;
636     }
637 }
638 
639 void ioapic_init_gsi(GSIState *gsi_state, const char *parent_name)
640 {
641     DeviceState *dev;
642     SysBusDevice *d;
643     unsigned int i;
644 
645     assert(parent_name);
646     if (kvm_ioapic_in_kernel()) {
647         dev = qdev_new(TYPE_KVM_IOAPIC);
648     } else {
649         dev = qdev_new(TYPE_IOAPIC);
650     }
651     object_property_add_child(object_resolve_path(parent_name, NULL),
652                               "ioapic", OBJECT(dev));
653     d = SYS_BUS_DEVICE(dev);
654     sysbus_realize_and_unref(d, &error_fatal);
655     sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS);
656 
657     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
658         gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i);
659     }
660 }
661 
662 DeviceState *ioapic_init_secondary(GSIState *gsi_state)
663 {
664     DeviceState *dev;
665     SysBusDevice *d;
666     unsigned int i;
667 
668     dev = qdev_new(TYPE_IOAPIC);
669     d = SYS_BUS_DEVICE(dev);
670     sysbus_realize_and_unref(d, &error_fatal);
671     sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS);
672 
673     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
674         gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i);
675     }
676     return dev;
677 }
678 
679 struct setup_data {
680     uint64_t next;
681     uint32_t type;
682     uint32_t len;
683     uint8_t data[];
684 } __attribute__((packed));
685 
686 
687 /*
688  * The entry point into the kernel for PVH boot is different from
689  * the native entry point.  The PVH entry is defined by the x86/HVM
690  * direct boot ABI and is available in an ELFNOTE in the kernel binary.
691  *
692  * This function is passed to load_elf() when it is called from
693  * load_elfboot() which then additionally checks for an ELF Note of
694  * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
695  * parse the PVH entry address from the ELF Note.
696  *
697  * Due to trickery in elf_opts.h, load_elf() is actually available as
698  * load_elf32() or load_elf64() and this routine needs to be able
699  * to deal with being called as 32 or 64 bit.
700  *
701  * The address of the PVH entry point is saved to the 'pvh_start_addr'
702  * global variable.  (although the entry point is 32-bit, the kernel
703  * binary can be either 32-bit or 64-bit).
704  */
705 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64)
706 {
707     size_t *elf_note_data_addr;
708 
709     /* Check if ELF Note header passed in is valid */
710     if (arg1 == NULL) {
711         return 0;
712     }
713 
714     if (is64) {
715         struct elf64_note *nhdr64 = (struct elf64_note *)arg1;
716         uint64_t nhdr_size64 = sizeof(struct elf64_note);
717         uint64_t phdr_align = *(uint64_t *)arg2;
718         uint64_t nhdr_namesz = nhdr64->n_namesz;
719 
720         elf_note_data_addr =
721             ((void *)nhdr64) + nhdr_size64 +
722             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
723 
724         pvh_start_addr = *elf_note_data_addr;
725     } else {
726         struct elf32_note *nhdr32 = (struct elf32_note *)arg1;
727         uint32_t nhdr_size32 = sizeof(struct elf32_note);
728         uint32_t phdr_align = *(uint32_t *)arg2;
729         uint32_t nhdr_namesz = nhdr32->n_namesz;
730 
731         elf_note_data_addr =
732             ((void *)nhdr32) + nhdr_size32 +
733             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
734 
735         pvh_start_addr = *(uint32_t *)elf_note_data_addr;
736     }
737 
738     return pvh_start_addr;
739 }
740 
741 static bool load_elfboot(const char *kernel_filename,
742                          int kernel_file_size,
743                          uint8_t *header,
744                          size_t pvh_xen_start_addr,
745                          FWCfgState *fw_cfg)
746 {
747     uint32_t flags = 0;
748     uint32_t mh_load_addr = 0;
749     uint32_t elf_kernel_size = 0;
750     uint64_t elf_entry;
751     uint64_t elf_low, elf_high;
752     int kernel_size;
753 
754     if (ldl_p(header) != 0x464c457f) {
755         return false; /* no elfboot */
756     }
757 
758     bool elf_is64 = header[EI_CLASS] == ELFCLASS64;
759     flags = elf_is64 ?
760         ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags;
761 
762     if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
763         error_report("elfboot unsupported flags = %x", flags);
764         exit(1);
765     }
766 
767     uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY;
768     kernel_size = load_elf(kernel_filename, read_pvh_start_addr,
769                            NULL, &elf_note_type, &elf_entry,
770                            &elf_low, &elf_high, NULL, 0, I386_ELF_MACHINE,
771                            0, 0);
772 
773     if (kernel_size < 0) {
774         error_report("Error while loading elf kernel");
775         exit(1);
776     }
777     mh_load_addr = elf_low;
778     elf_kernel_size = elf_high - elf_low;
779 
780     if (pvh_start_addr == 0) {
781         error_report("Error loading uncompressed kernel without PVH ELF Note");
782         exit(1);
783     }
784     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr);
785     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr);
786     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size);
787 
788     return true;
789 }
790 
791 void x86_load_linux(X86MachineState *x86ms,
792                     FWCfgState *fw_cfg,
793                     int acpi_data_size,
794                     bool pvh_enabled)
795 {
796     bool linuxboot_dma_enabled = X86_MACHINE_GET_CLASS(x86ms)->fwcfg_dma_enabled;
797     uint16_t protocol;
798     int setup_size, kernel_size, cmdline_size;
799     int dtb_size, setup_data_offset;
800     uint32_t initrd_max;
801     uint8_t header[8192], *setup, *kernel;
802     hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0;
803     FILE *f;
804     char *vmode;
805     MachineState *machine = MACHINE(x86ms);
806     struct setup_data *setup_data;
807     const char *kernel_filename = machine->kernel_filename;
808     const char *initrd_filename = machine->initrd_filename;
809     const char *dtb_filename = machine->dtb;
810     const char *kernel_cmdline = machine->kernel_cmdline;
811     SevKernelLoaderContext sev_load_ctx = {};
812 
813     /* Align to 16 bytes as a paranoia measure */
814     cmdline_size = (strlen(kernel_cmdline) + 16) & ~15;
815 
816     /* load the kernel header */
817     f = fopen(kernel_filename, "rb");
818     if (!f) {
819         fprintf(stderr, "qemu: could not open kernel file '%s': %s\n",
820                 kernel_filename, strerror(errno));
821         exit(1);
822     }
823 
824     kernel_size = get_file_size(f);
825     if (!kernel_size ||
826         fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
827         MIN(ARRAY_SIZE(header), kernel_size)) {
828         fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
829                 kernel_filename, strerror(errno));
830         exit(1);
831     }
832 
833     /* kernel protocol version */
834     if (ldl_p(header + 0x202) == 0x53726448) {
835         protocol = lduw_p(header + 0x206);
836     } else {
837         /*
838          * This could be a multiboot kernel. If it is, let's stop treating it
839          * like a Linux kernel.
840          * Note: some multiboot images could be in the ELF format (the same of
841          * PVH), so we try multiboot first since we check the multiboot magic
842          * header before to load it.
843          */
844         if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename,
845                            kernel_cmdline, kernel_size, header)) {
846             return;
847         }
848         /*
849          * Check if the file is an uncompressed kernel file (ELF) and load it,
850          * saving the PVH entry point used by the x86/HVM direct boot ABI.
851          * If load_elfboot() is successful, populate the fw_cfg info.
852          */
853         if (pvh_enabled &&
854             load_elfboot(kernel_filename, kernel_size,
855                          header, pvh_start_addr, fw_cfg)) {
856             fclose(f);
857 
858             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
859                 strlen(kernel_cmdline) + 1);
860             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
861 
862             fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header));
863             fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA,
864                              header, sizeof(header));
865 
866             /* load initrd */
867             if (initrd_filename) {
868                 GMappedFile *mapped_file;
869                 gsize initrd_size;
870                 gchar *initrd_data;
871                 GError *gerr = NULL;
872 
873                 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
874                 if (!mapped_file) {
875                     fprintf(stderr, "qemu: error reading initrd %s: %s\n",
876                             initrd_filename, gerr->message);
877                     exit(1);
878                 }
879                 x86ms->initrd_mapped_file = mapped_file;
880 
881                 initrd_data = g_mapped_file_get_contents(mapped_file);
882                 initrd_size = g_mapped_file_get_length(mapped_file);
883                 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
884                 if (initrd_size >= initrd_max) {
885                     fprintf(stderr, "qemu: initrd is too large, cannot support."
886                             "(max: %"PRIu32", need %"PRId64")\n",
887                             initrd_max, (uint64_t)initrd_size);
888                     exit(1);
889                 }
890 
891                 initrd_addr = (initrd_max - initrd_size) & ~4095;
892 
893                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
894                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
895                 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data,
896                                  initrd_size);
897             }
898 
899             option_rom[nb_option_roms].bootindex = 0;
900             option_rom[nb_option_roms].name = "pvh.bin";
901             nb_option_roms++;
902 
903             return;
904         }
905         protocol = 0;
906     }
907 
908     if (protocol < 0x200 || !(header[0x211] & 0x01)) {
909         /* Low kernel */
910         real_addr    = 0x90000;
911         cmdline_addr = 0x9a000 - cmdline_size;
912         prot_addr    = 0x10000;
913     } else if (protocol < 0x202) {
914         /* High but ancient kernel */
915         real_addr    = 0x90000;
916         cmdline_addr = 0x9a000 - cmdline_size;
917         prot_addr    = 0x100000;
918     } else {
919         /* High and recent kernel */
920         real_addr    = 0x10000;
921         cmdline_addr = 0x20000;
922         prot_addr    = 0x100000;
923     }
924 
925     /* highest address for loading the initrd */
926     if (protocol >= 0x20c &&
927         lduw_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) {
928         /*
929          * Linux has supported initrd up to 4 GB for a very long time (2007,
930          * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
931          * though it only sets initrd_max to 2 GB to "work around bootloader
932          * bugs". Luckily, QEMU firmware(which does something like bootloader)
933          * has supported this.
934          *
935          * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
936          * be loaded into any address.
937          *
938          * In addition, initrd_max is uint32_t simply because QEMU doesn't
939          * support the 64-bit boot protocol (specifically the ext_ramdisk_image
940          * field).
941          *
942          * Therefore here just limit initrd_max to UINT32_MAX simply as well.
943          */
944         initrd_max = UINT32_MAX;
945     } else if (protocol >= 0x203) {
946         initrd_max = ldl_p(header + 0x22c);
947     } else {
948         initrd_max = 0x37ffffff;
949     }
950 
951     if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) {
952         initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
953     }
954 
955     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
956     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline) + 1);
957     fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
958     sev_load_ctx.cmdline_data = (char *)kernel_cmdline;
959     sev_load_ctx.cmdline_size = strlen(kernel_cmdline) + 1;
960 
961     if (protocol >= 0x202) {
962         stl_p(header + 0x228, cmdline_addr);
963     } else {
964         stw_p(header + 0x20, 0xA33F);
965         stw_p(header + 0x22, cmdline_addr - real_addr);
966     }
967 
968     /* handle vga= parameter */
969     vmode = strstr(kernel_cmdline, "vga=");
970     if (vmode) {
971         unsigned int video_mode;
972         const char *end;
973         int ret;
974         /* skip "vga=" */
975         vmode += 4;
976         if (!strncmp(vmode, "normal", 6)) {
977             video_mode = 0xffff;
978         } else if (!strncmp(vmode, "ext", 3)) {
979             video_mode = 0xfffe;
980         } else if (!strncmp(vmode, "ask", 3)) {
981             video_mode = 0xfffd;
982         } else {
983             ret = qemu_strtoui(vmode, &end, 0, &video_mode);
984             if (ret != 0 || (*end && *end != ' ')) {
985                 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n");
986                 exit(1);
987             }
988         }
989         stw_p(header + 0x1fa, video_mode);
990     }
991 
992     /* loader type */
993     /*
994      * High nybble = B reserved for QEMU; low nybble is revision number.
995      * If this code is substantially changed, you may want to consider
996      * incrementing the revision.
997      */
998     if (protocol >= 0x200) {
999         header[0x210] = 0xB0;
1000     }
1001     /* heap */
1002     if (protocol >= 0x201) {
1003         header[0x211] |= 0x80; /* CAN_USE_HEAP */
1004         stw_p(header + 0x224, cmdline_addr - real_addr - 0x200);
1005     }
1006 
1007     /* load initrd */
1008     if (initrd_filename) {
1009         GMappedFile *mapped_file;
1010         gsize initrd_size;
1011         gchar *initrd_data;
1012         GError *gerr = NULL;
1013 
1014         if (protocol < 0x200) {
1015             fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
1016             exit(1);
1017         }
1018 
1019         mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
1020         if (!mapped_file) {
1021             fprintf(stderr, "qemu: error reading initrd %s: %s\n",
1022                     initrd_filename, gerr->message);
1023             exit(1);
1024         }
1025         x86ms->initrd_mapped_file = mapped_file;
1026 
1027         initrd_data = g_mapped_file_get_contents(mapped_file);
1028         initrd_size = g_mapped_file_get_length(mapped_file);
1029         if (initrd_size >= initrd_max) {
1030             fprintf(stderr, "qemu: initrd is too large, cannot support."
1031                     "(max: %"PRIu32", need %"PRId64")\n",
1032                     initrd_max, (uint64_t)initrd_size);
1033             exit(1);
1034         }
1035 
1036         initrd_addr = (initrd_max - initrd_size) & ~4095;
1037 
1038         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
1039         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
1040         fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
1041         sev_load_ctx.initrd_data = initrd_data;
1042         sev_load_ctx.initrd_size = initrd_size;
1043 
1044         stl_p(header + 0x218, initrd_addr);
1045         stl_p(header + 0x21c, initrd_size);
1046     }
1047 
1048     /* load kernel and setup */
1049     setup_size = header[0x1f1];
1050     if (setup_size == 0) {
1051         setup_size = 4;
1052     }
1053     setup_size = (setup_size + 1) * 512;
1054     if (setup_size > kernel_size) {
1055         fprintf(stderr, "qemu: invalid kernel header\n");
1056         exit(1);
1057     }
1058     kernel_size -= setup_size;
1059 
1060     setup  = g_malloc(setup_size);
1061     kernel = g_malloc(kernel_size);
1062     fseek(f, 0, SEEK_SET);
1063     if (fread(setup, 1, setup_size, f) != setup_size) {
1064         fprintf(stderr, "fread() failed\n");
1065         exit(1);
1066     }
1067     if (fread(kernel, 1, kernel_size, f) != kernel_size) {
1068         fprintf(stderr, "fread() failed\n");
1069         exit(1);
1070     }
1071     fclose(f);
1072 
1073     /* append dtb to kernel */
1074     if (dtb_filename) {
1075         if (protocol < 0x209) {
1076             fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n");
1077             exit(1);
1078         }
1079 
1080         dtb_size = get_image_size(dtb_filename);
1081         if (dtb_size <= 0) {
1082             fprintf(stderr, "qemu: error reading dtb %s: %s\n",
1083                     dtb_filename, strerror(errno));
1084             exit(1);
1085         }
1086 
1087         setup_data_offset = QEMU_ALIGN_UP(kernel_size, 16);
1088         kernel_size = setup_data_offset + sizeof(struct setup_data) + dtb_size;
1089         kernel = g_realloc(kernel, kernel_size);
1090 
1091         stq_p(header + 0x250, prot_addr + setup_data_offset);
1092 
1093         setup_data = (struct setup_data *)(kernel + setup_data_offset);
1094         setup_data->next = 0;
1095         setup_data->type = cpu_to_le32(SETUP_DTB);
1096         setup_data->len = cpu_to_le32(dtb_size);
1097 
1098         load_image_size(dtb_filename, setup_data->data, dtb_size);
1099     }
1100 
1101     /*
1102      * If we're starting an encrypted VM, it will be OVMF based, which uses the
1103      * efi stub for booting and doesn't require any values to be placed in the
1104      * kernel header.  We therefore don't update the header so the hash of the
1105      * kernel on the other side of the fw_cfg interface matches the hash of the
1106      * file the user passed in.
1107      */
1108     if (!sev_enabled()) {
1109         memcpy(setup, header, MIN(sizeof(header), setup_size));
1110     }
1111 
1112     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
1113     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size);
1114     fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA, kernel, kernel_size);
1115     sev_load_ctx.kernel_data = (char *)kernel;
1116     sev_load_ctx.kernel_size = kernel_size;
1117 
1118     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
1119     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
1120     fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
1121     sev_load_ctx.setup_data = (char *)setup;
1122     sev_load_ctx.setup_size = setup_size;
1123 
1124     if (sev_enabled()) {
1125         sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal);
1126     }
1127 
1128     option_rom[nb_option_roms].bootindex = 0;
1129     option_rom[nb_option_roms].name = "linuxboot.bin";
1130     if (linuxboot_dma_enabled && fw_cfg_dma_enabled(fw_cfg)) {
1131         option_rom[nb_option_roms].name = "linuxboot_dma.bin";
1132     }
1133     nb_option_roms++;
1134 }
1135 
1136 void x86_bios_rom_init(MachineState *ms, const char *default_firmware,
1137                        MemoryRegion *rom_memory, bool isapc_ram_fw)
1138 {
1139     const char *bios_name;
1140     char *filename;
1141     MemoryRegion *bios, *isa_bios;
1142     int bios_size, isa_bios_size;
1143     ssize_t ret;
1144 
1145     /* BIOS load */
1146     bios_name = ms->firmware ?: default_firmware;
1147     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1148     if (filename) {
1149         bios_size = get_image_size(filename);
1150     } else {
1151         bios_size = -1;
1152     }
1153     if (bios_size <= 0 ||
1154         (bios_size % 65536) != 0) {
1155         goto bios_error;
1156     }
1157     bios = g_malloc(sizeof(*bios));
1158     memory_region_init_ram(bios, NULL, "pc.bios", bios_size, &error_fatal);
1159     if (sev_enabled()) {
1160         /*
1161          * The concept of a "reset" simply doesn't exist for
1162          * confidential computing guests, we have to destroy and
1163          * re-launch them instead.  So there is no need to register
1164          * the firmware as rom to properly re-initialize on reset.
1165          * Just go for a straight file load instead.
1166          */
1167         void *ptr = memory_region_get_ram_ptr(bios);
1168         load_image_size(filename, ptr, bios_size);
1169         x86_firmware_configure(ptr, bios_size);
1170     } else {
1171         if (!isapc_ram_fw) {
1172             memory_region_set_readonly(bios, true);
1173         }
1174         ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1);
1175         if (ret != 0) {
1176             goto bios_error;
1177         }
1178     }
1179     g_free(filename);
1180 
1181     /* map the last 128KB of the BIOS in ISA space */
1182     isa_bios_size = MIN(bios_size, 128 * KiB);
1183     isa_bios = g_malloc(sizeof(*isa_bios));
1184     memory_region_init_alias(isa_bios, NULL, "isa-bios", bios,
1185                              bios_size - isa_bios_size, isa_bios_size);
1186     memory_region_add_subregion_overlap(rom_memory,
1187                                         0x100000 - isa_bios_size,
1188                                         isa_bios,
1189                                         1);
1190     if (!isapc_ram_fw) {
1191         memory_region_set_readonly(isa_bios, true);
1192     }
1193 
1194     /* map all the bios at the top of memory */
1195     memory_region_add_subregion(rom_memory,
1196                                 (uint32_t)(-bios_size),
1197                                 bios);
1198     return;
1199 
1200 bios_error:
1201     fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1202     exit(1);
1203 }
1204 
1205 bool x86_machine_is_smm_enabled(const X86MachineState *x86ms)
1206 {
1207     bool smm_available = false;
1208 
1209     if (x86ms->smm == ON_OFF_AUTO_OFF) {
1210         return false;
1211     }
1212 
1213     if (tcg_enabled() || qtest_enabled()) {
1214         smm_available = true;
1215     } else if (kvm_enabled()) {
1216         smm_available = kvm_has_smm();
1217     }
1218 
1219     if (smm_available) {
1220         return true;
1221     }
1222 
1223     if (x86ms->smm == ON_OFF_AUTO_ON) {
1224         error_report("System Management Mode not supported by this hypervisor.");
1225         exit(1);
1226     }
1227     return false;
1228 }
1229 
1230 static void x86_machine_get_smm(Object *obj, Visitor *v, const char *name,
1231                                void *opaque, Error **errp)
1232 {
1233     X86MachineState *x86ms = X86_MACHINE(obj);
1234     OnOffAuto smm = x86ms->smm;
1235 
1236     visit_type_OnOffAuto(v, name, &smm, errp);
1237 }
1238 
1239 static void x86_machine_set_smm(Object *obj, Visitor *v, const char *name,
1240                                void *opaque, Error **errp)
1241 {
1242     X86MachineState *x86ms = X86_MACHINE(obj);
1243 
1244     visit_type_OnOffAuto(v, name, &x86ms->smm, errp);
1245 }
1246 
1247 bool x86_machine_is_acpi_enabled(const X86MachineState *x86ms)
1248 {
1249     if (x86ms->acpi == ON_OFF_AUTO_OFF) {
1250         return false;
1251     }
1252     return true;
1253 }
1254 
1255 static void x86_machine_get_acpi(Object *obj, Visitor *v, const char *name,
1256                                  void *opaque, Error **errp)
1257 {
1258     X86MachineState *x86ms = X86_MACHINE(obj);
1259     OnOffAuto acpi = x86ms->acpi;
1260 
1261     visit_type_OnOffAuto(v, name, &acpi, errp);
1262 }
1263 
1264 static void x86_machine_set_acpi(Object *obj, Visitor *v, const char *name,
1265                                  void *opaque, Error **errp)
1266 {
1267     X86MachineState *x86ms = X86_MACHINE(obj);
1268 
1269     visit_type_OnOffAuto(v, name, &x86ms->acpi, errp);
1270 }
1271 
1272 static void x86_machine_get_pit(Object *obj, Visitor *v, const char *name,
1273                                     void *opaque, Error **errp)
1274 {
1275     X86MachineState *x86ms = X86_MACHINE(obj);
1276     OnOffAuto pit = x86ms->pit;
1277 
1278     visit_type_OnOffAuto(v, name, &pit, errp);
1279 }
1280 
1281 static void x86_machine_set_pit(Object *obj, Visitor *v, const char *name,
1282                                     void *opaque, Error **errp)
1283 {
1284     X86MachineState *x86ms = X86_MACHINE(obj);;
1285 
1286     visit_type_OnOffAuto(v, name, &x86ms->pit, errp);
1287 }
1288 
1289 static void x86_machine_get_pic(Object *obj, Visitor *v, const char *name,
1290                                 void *opaque, Error **errp)
1291 {
1292     X86MachineState *x86ms = X86_MACHINE(obj);
1293     OnOffAuto pic = x86ms->pic;
1294 
1295     visit_type_OnOffAuto(v, name, &pic, errp);
1296 }
1297 
1298 static void x86_machine_set_pic(Object *obj, Visitor *v, const char *name,
1299                                 void *opaque, Error **errp)
1300 {
1301     X86MachineState *x86ms = X86_MACHINE(obj);
1302 
1303     visit_type_OnOffAuto(v, name, &x86ms->pic, errp);
1304 }
1305 
1306 static char *x86_machine_get_oem_id(Object *obj, Error **errp)
1307 {
1308     X86MachineState *x86ms = X86_MACHINE(obj);
1309 
1310     return g_strdup(x86ms->oem_id);
1311 }
1312 
1313 static void x86_machine_set_oem_id(Object *obj, const char *value, Error **errp)
1314 {
1315     X86MachineState *x86ms = X86_MACHINE(obj);
1316     size_t len = strlen(value);
1317 
1318     if (len > 6) {
1319         error_setg(errp,
1320                    "User specified "X86_MACHINE_OEM_ID" value is bigger than "
1321                    "6 bytes in size");
1322         return;
1323     }
1324 
1325     strncpy(x86ms->oem_id, value, 6);
1326 }
1327 
1328 static char *x86_machine_get_oem_table_id(Object *obj, Error **errp)
1329 {
1330     X86MachineState *x86ms = X86_MACHINE(obj);
1331 
1332     return g_strdup(x86ms->oem_table_id);
1333 }
1334 
1335 static void x86_machine_set_oem_table_id(Object *obj, const char *value,
1336                                          Error **errp)
1337 {
1338     X86MachineState *x86ms = X86_MACHINE(obj);
1339     size_t len = strlen(value);
1340 
1341     if (len > 8) {
1342         error_setg(errp,
1343                    "User specified "X86_MACHINE_OEM_TABLE_ID
1344                    " value is bigger than "
1345                    "8 bytes in size");
1346         return;
1347     }
1348     strncpy(x86ms->oem_table_id, value, 8);
1349 }
1350 
1351 static void x86_machine_get_bus_lock_ratelimit(Object *obj, Visitor *v,
1352                                 const char *name, void *opaque, Error **errp)
1353 {
1354     X86MachineState *x86ms = X86_MACHINE(obj);
1355     uint64_t bus_lock_ratelimit = x86ms->bus_lock_ratelimit;
1356 
1357     visit_type_uint64(v, name, &bus_lock_ratelimit, errp);
1358 }
1359 
1360 static void x86_machine_set_bus_lock_ratelimit(Object *obj, Visitor *v,
1361                                const char *name, void *opaque, Error **errp)
1362 {
1363     X86MachineState *x86ms = X86_MACHINE(obj);
1364 
1365     visit_type_uint64(v, name, &x86ms->bus_lock_ratelimit, errp);
1366 }
1367 
1368 static void machine_get_sgx_epc(Object *obj, Visitor *v, const char *name,
1369                                 void *opaque, Error **errp)
1370 {
1371     X86MachineState *x86ms = X86_MACHINE(obj);
1372     SgxEPCList *list = x86ms->sgx_epc_list;
1373 
1374     visit_type_SgxEPCList(v, name, &list, errp);
1375 }
1376 
1377 static void machine_set_sgx_epc(Object *obj, Visitor *v, const char *name,
1378                                 void *opaque, Error **errp)
1379 {
1380     X86MachineState *x86ms = X86_MACHINE(obj);
1381     SgxEPCList *list;
1382 
1383     list = x86ms->sgx_epc_list;
1384     visit_type_SgxEPCList(v, name, &x86ms->sgx_epc_list, errp);
1385 
1386     qapi_free_SgxEPCList(list);
1387 }
1388 
1389 static void x86_machine_initfn(Object *obj)
1390 {
1391     X86MachineState *x86ms = X86_MACHINE(obj);
1392 
1393     x86ms->smm = ON_OFF_AUTO_AUTO;
1394     x86ms->acpi = ON_OFF_AUTO_AUTO;
1395     x86ms->pit = ON_OFF_AUTO_AUTO;
1396     x86ms->pic = ON_OFF_AUTO_AUTO;
1397     x86ms->pci_irq_mask = ACPI_BUILD_PCI_IRQS;
1398     x86ms->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
1399     x86ms->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
1400     x86ms->bus_lock_ratelimit = 0;
1401     x86ms->above_4g_mem_start = 4 * GiB;
1402 }
1403 
1404 static void x86_machine_class_init(ObjectClass *oc, void *data)
1405 {
1406     MachineClass *mc = MACHINE_CLASS(oc);
1407     X86MachineClass *x86mc = X86_MACHINE_CLASS(oc);
1408     NMIClass *nc = NMI_CLASS(oc);
1409 
1410     mc->cpu_index_to_instance_props = x86_cpu_index_to_props;
1411     mc->get_default_cpu_node_id = x86_get_default_cpu_node_id;
1412     mc->possible_cpu_arch_ids = x86_possible_cpu_arch_ids;
1413     x86mc->save_tsc_khz = true;
1414     x86mc->fwcfg_dma_enabled = true;
1415     nc->nmi_monitor_handler = x86_nmi;
1416 
1417     object_class_property_add(oc, X86_MACHINE_SMM, "OnOffAuto",
1418         x86_machine_get_smm, x86_machine_set_smm,
1419         NULL, NULL);
1420     object_class_property_set_description(oc, X86_MACHINE_SMM,
1421         "Enable SMM");
1422 
1423     object_class_property_add(oc, X86_MACHINE_ACPI, "OnOffAuto",
1424         x86_machine_get_acpi, x86_machine_set_acpi,
1425         NULL, NULL);
1426     object_class_property_set_description(oc, X86_MACHINE_ACPI,
1427         "Enable ACPI");
1428 
1429     object_class_property_add(oc, X86_MACHINE_PIT, "OnOffAuto",
1430                               x86_machine_get_pit,
1431                               x86_machine_set_pit,
1432                               NULL, NULL);
1433     object_class_property_set_description(oc, X86_MACHINE_PIT,
1434         "Enable i8254 PIT");
1435 
1436     object_class_property_add(oc, X86_MACHINE_PIC, "OnOffAuto",
1437                               x86_machine_get_pic,
1438                               x86_machine_set_pic,
1439                               NULL, NULL);
1440     object_class_property_set_description(oc, X86_MACHINE_PIC,
1441         "Enable i8259 PIC");
1442 
1443     object_class_property_add_str(oc, X86_MACHINE_OEM_ID,
1444                                   x86_machine_get_oem_id,
1445                                   x86_machine_set_oem_id);
1446     object_class_property_set_description(oc, X86_MACHINE_OEM_ID,
1447                                           "Override the default value of field OEMID "
1448                                           "in ACPI table header."
1449                                           "The string may be up to 6 bytes in size");
1450 
1451 
1452     object_class_property_add_str(oc, X86_MACHINE_OEM_TABLE_ID,
1453                                   x86_machine_get_oem_table_id,
1454                                   x86_machine_set_oem_table_id);
1455     object_class_property_set_description(oc, X86_MACHINE_OEM_TABLE_ID,
1456                                           "Override the default value of field OEM Table ID "
1457                                           "in ACPI table header."
1458                                           "The string may be up to 8 bytes in size");
1459 
1460     object_class_property_add(oc, X86_MACHINE_BUS_LOCK_RATELIMIT, "uint64_t",
1461                                 x86_machine_get_bus_lock_ratelimit,
1462                                 x86_machine_set_bus_lock_ratelimit, NULL, NULL);
1463     object_class_property_set_description(oc, X86_MACHINE_BUS_LOCK_RATELIMIT,
1464             "Set the ratelimit for the bus locks acquired in VMs");
1465 
1466     object_class_property_add(oc, "sgx-epc", "SgxEPC",
1467         machine_get_sgx_epc, machine_set_sgx_epc,
1468         NULL, NULL);
1469     object_class_property_set_description(oc, "sgx-epc",
1470         "SGX EPC device");
1471 }
1472 
1473 static const TypeInfo x86_machine_info = {
1474     .name = TYPE_X86_MACHINE,
1475     .parent = TYPE_MACHINE,
1476     .abstract = true,
1477     .instance_size = sizeof(X86MachineState),
1478     .instance_init = x86_machine_initfn,
1479     .class_size = sizeof(X86MachineClass),
1480     .class_init = x86_machine_class_init,
1481     .interfaces = (InterfaceInfo[]) {
1482          { TYPE_NMI },
1483          { }
1484     },
1485 };
1486 
1487 static void x86_machine_register_types(void)
1488 {
1489     type_register_static(&x86_machine_info);
1490 }
1491 
1492 type_init(x86_machine_register_types)
1493