xref: /qemu/hw/i386/x86-common.c (revision 81bd60625fc23cb8d4d0e682dcc4223d5e1ead84)
1 /*
2  * Copyright (c) 2003-2004 Fabrice Bellard
3  * Copyright (c) 2019, 2024 Red Hat, Inc.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a copy
6  * of this software and associated documentation files (the "Software"), to deal
7  * in the Software without restriction, including without limitation the rights
8  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9  * copies of the Software, and to permit persons to whom the Software is
10  * furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be included in
13  * all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
21  * THE SOFTWARE.
22  */
23 #include "qemu/osdep.h"
24 #include "qemu/error-report.h"
25 #include "qemu/cutils.h"
26 #include "qemu/units.h"
27 #include "qemu/datadir.h"
28 #include "qapi/error.h"
29 #include "system/numa.h"
30 #include "system/system.h"
31 #include "system/xen.h"
32 #include "trace.h"
33 
34 #include "hw/i386/x86.h"
35 #include "target/i386/cpu.h"
36 #include "hw/rtc/mc146818rtc.h"
37 #include "target/i386/sev.h"
38 
39 #include "hw/acpi/cpu_hotplug.h"
40 #include "hw/irq.h"
41 #include "hw/loader.h"
42 #include "multiboot.h"
43 #include "elf.h"
44 #include "standard-headers/asm-x86/bootparam.h"
45 #include CONFIG_DEVICES
46 #include "kvm/kvm_i386.h"
47 
48 #ifdef CONFIG_XEN_EMU
49 #include "hw/xen/xen.h"
50 #include "hw/i386/kvm/xen_evtchn.h"
51 #endif
52 
53 /* Physical Address of PVH entry point read from kernel ELF NOTE */
54 static size_t pvh_start_addr;
55 
56 static void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp)
57 {
58     Object *cpu = object_new(MACHINE(x86ms)->cpu_type);
59 
60     if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) {
61         goto out;
62     }
63     qdev_realize(DEVICE(cpu), NULL, errp);
64 
65 out:
66     object_unref(cpu);
67 }
68 
69 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version)
70 {
71     int i;
72     const CPUArchIdList *possible_cpus;
73     MachineState *ms = MACHINE(x86ms);
74     MachineClass *mc = MACHINE_GET_CLASS(x86ms);
75 
76     x86_cpu_set_default_version(default_cpu_version);
77 
78     /*
79      * Calculates the limit to CPU APIC ID values
80      *
81      * Limit for the APIC ID value, so that all
82      * CPU APIC IDs are < x86ms->apic_id_limit.
83      *
84      * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
85      */
86     x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms,
87                                                       ms->smp.max_cpus - 1) + 1;
88 
89     /*
90      * Can we support APIC ID 255 or higher?  With KVM, that requires
91      * both in-kernel lapic and X2APIC userspace API.
92      *
93      * kvm_enabled() must go first to ensure that kvm_* references are
94      * not emitted for the linker to consume (kvm_enabled() is
95      * a literal `0` in configurations where kvm_* aren't defined)
96      */
97     if (kvm_enabled() && x86ms->apic_id_limit > 255 &&
98         kvm_irqchip_in_kernel() && !kvm_enable_x2apic()) {
99         error_report("current -smp configuration requires kernel "
100                      "irqchip and X2APIC API support.");
101         exit(EXIT_FAILURE);
102     }
103 
104     if (kvm_enabled()) {
105         kvm_set_max_apic_id(x86ms->apic_id_limit);
106     }
107 
108     if (!kvm_irqchip_in_kernel()) {
109         apic_set_max_apic_id(x86ms->apic_id_limit);
110     }
111 
112     possible_cpus = mc->possible_cpu_arch_ids(ms);
113     for (i = 0; i < ms->smp.cpus; i++) {
114         x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal);
115     }
116 }
117 
118 void x86_rtc_set_cpus_count(ISADevice *s, uint16_t cpus_count)
119 {
120     MC146818RtcState *rtc = MC146818_RTC(s);
121 
122     if (cpus_count > 0xff) {
123         /*
124          * If the number of CPUs can't be represented in 8 bits, the
125          * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just
126          * to make old BIOSes fail more predictably.
127          */
128         mc146818rtc_set_cmos_data(rtc, 0x5f, 0);
129     } else {
130         mc146818rtc_set_cmos_data(rtc, 0x5f, cpus_count - 1);
131     }
132 }
133 
134 static int x86_apic_cmp(const void *a, const void *b)
135 {
136    CPUArchId *apic_a = (CPUArchId *)a;
137    CPUArchId *apic_b = (CPUArchId *)b;
138 
139    return apic_a->arch_id - apic_b->arch_id;
140 }
141 
142 /*
143  * returns pointer to CPUArchId descriptor that matches CPU's apic_id
144  * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no
145  * entry corresponding to CPU's apic_id returns NULL.
146  */
147 static CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
148 {
149     CPUArchId apic_id, *found_cpu;
150 
151     apic_id.arch_id = id;
152     found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus,
153         ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus),
154         x86_apic_cmp);
155     if (found_cpu && idx) {
156         *idx = found_cpu - ms->possible_cpus->cpus;
157     }
158     return found_cpu;
159 }
160 
161 void x86_cpu_plug(HotplugHandler *hotplug_dev,
162                   DeviceState *dev, Error **errp)
163 {
164     CPUArchId *found_cpu;
165     Error *local_err = NULL;
166     X86CPU *cpu = X86_CPU(dev);
167     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
168 
169     if (x86ms->acpi_dev) {
170         hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err);
171         if (local_err) {
172             goto out;
173         }
174     }
175 
176     /* increment the number of CPUs */
177     x86ms->boot_cpus++;
178     if (x86ms->rtc) {
179         x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
180     }
181     if (x86ms->fw_cfg) {
182         fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
183     }
184 
185     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
186     found_cpu->cpu = CPU(dev);
187 out:
188     error_propagate(errp, local_err);
189 }
190 
191 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev,
192                                DeviceState *dev, Error **errp)
193 {
194     int idx = -1;
195     X86CPU *cpu = X86_CPU(dev);
196     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
197 
198     if (!x86ms->acpi_dev) {
199         error_setg(errp, "CPU hot unplug not supported without ACPI");
200         return;
201     }
202 
203     x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
204     assert(idx != -1);
205     if (idx == 0) {
206         error_setg(errp, "Boot CPU is unpluggable");
207         return;
208     }
209 
210     hotplug_handler_unplug_request(x86ms->acpi_dev, dev,
211                                    errp);
212 }
213 
214 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev,
215                        DeviceState *dev, Error **errp)
216 {
217     CPUArchId *found_cpu;
218     Error *local_err = NULL;
219     X86CPU *cpu = X86_CPU(dev);
220     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
221 
222     hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err);
223     if (local_err) {
224         goto out;
225     }
226 
227     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
228     found_cpu->cpu = NULL;
229     qdev_unrealize(dev);
230 
231     /* decrement the number of CPUs */
232     x86ms->boot_cpus--;
233     /* Update the number of CPUs in CMOS */
234     x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
235     fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
236  out:
237     error_propagate(errp, local_err);
238 }
239 
240 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev,
241                       DeviceState *dev, Error **errp)
242 {
243     int idx;
244     CPUState *cs;
245     CPUArchId *cpu_slot;
246     X86CPUTopoIDs topo_ids;
247     X86CPU *cpu = X86_CPU(dev);
248     CPUX86State *env = &cpu->env;
249     MachineState *ms = MACHINE(hotplug_dev);
250     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
251     X86CPUTopoInfo topo_info;
252 
253     if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) {
254         error_setg(errp, "Invalid CPU type, expected cpu type: '%s'",
255                    ms->cpu_type);
256         return;
257     }
258 
259     if (x86ms->acpi_dev) {
260         Error *local_err = NULL;
261 
262         hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev,
263                                  &local_err);
264         if (local_err) {
265             error_propagate(errp, local_err);
266             return;
267         }
268     }
269 
270     init_topo_info(&topo_info, x86ms);
271 
272     if (ms->smp.modules > 1) {
273         env->nr_modules = ms->smp.modules;
274         set_bit(CPU_TOPOLOGY_LEVEL_MODULE, env->avail_cpu_topo);
275     }
276 
277     if (ms->smp.dies > 1) {
278         env->nr_dies = ms->smp.dies;
279         set_bit(CPU_TOPOLOGY_LEVEL_DIE, env->avail_cpu_topo);
280     }
281 
282     /*
283      * If APIC ID is not set,
284      * set it based on socket/die/module/core/thread properties.
285      */
286     if (cpu->apic_id == UNASSIGNED_APIC_ID) {
287         /*
288          * die-id was optional in QEMU 4.0 and older, so keep it optional
289          * if there's only one die per socket.
290          */
291         if (cpu->die_id < 0 && ms->smp.dies == 1) {
292             cpu->die_id = 0;
293         }
294 
295         /*
296          * module-id was optional in QEMU 9.0 and older, so keep it optional
297          * if there's only one module per die.
298          */
299         if (cpu->module_id < 0 && ms->smp.modules == 1) {
300             cpu->module_id = 0;
301         }
302 
303         if (cpu->socket_id < 0) {
304             error_setg(errp, "CPU socket-id is not set");
305             return;
306         } else if (cpu->socket_id > ms->smp.sockets - 1) {
307             error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u",
308                        cpu->socket_id, ms->smp.sockets - 1);
309             return;
310         }
311         if (cpu->die_id < 0) {
312             error_setg(errp, "CPU die-id is not set");
313             return;
314         } else if (cpu->die_id > ms->smp.dies - 1) {
315             error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u",
316                        cpu->die_id, ms->smp.dies - 1);
317             return;
318         }
319         if (cpu->module_id < 0) {
320             error_setg(errp, "CPU module-id is not set");
321             return;
322         } else if (cpu->module_id > ms->smp.modules - 1) {
323             error_setg(errp, "Invalid CPU module-id: %u must be in range 0:%u",
324                        cpu->module_id, ms->smp.modules - 1);
325             return;
326         }
327         if (cpu->core_id < 0) {
328             error_setg(errp, "CPU core-id is not set");
329             return;
330         } else if (cpu->core_id > (ms->smp.cores - 1)) {
331             error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u",
332                        cpu->core_id, ms->smp.cores - 1);
333             return;
334         }
335         if (cpu->thread_id < 0) {
336             error_setg(errp, "CPU thread-id is not set");
337             return;
338         } else if (cpu->thread_id > (ms->smp.threads - 1)) {
339             error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u",
340                        cpu->thread_id, ms->smp.threads - 1);
341             return;
342         }
343 
344         topo_ids.pkg_id = cpu->socket_id;
345         topo_ids.die_id = cpu->die_id;
346         topo_ids.module_id = cpu->module_id;
347         topo_ids.core_id = cpu->core_id;
348         topo_ids.smt_id = cpu->thread_id;
349         cpu->apic_id = x86_apicid_from_topo_ids(&topo_info, &topo_ids);
350     }
351 
352     cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
353     if (!cpu_slot) {
354         x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
355 
356         error_setg(errp,
357             "Invalid CPU [socket: %u, die: %u, module: %u, core: %u, thread: %u]"
358             " with APIC ID %" PRIu32 ", valid index range 0:%d",
359             topo_ids.pkg_id, topo_ids.die_id, topo_ids.module_id,
360             topo_ids.core_id, topo_ids.smt_id, cpu->apic_id,
361             ms->possible_cpus->len - 1);
362         return;
363     }
364 
365     if (cpu_slot->cpu) {
366         error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists",
367                    idx, cpu->apic_id);
368         return;
369     }
370 
371     /* if 'address' properties socket-id/core-id/thread-id are not set, set them
372      * so that machine_query_hotpluggable_cpus would show correct values
373      */
374     /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
375      * once -smp refactoring is complete and there will be CPU private
376      * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
377     x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
378     if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) {
379         error_setg(errp, "property socket-id: %u doesn't match set apic-id:"
380             " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id,
381             topo_ids.pkg_id);
382         return;
383     }
384     cpu->socket_id = topo_ids.pkg_id;
385 
386     if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) {
387         error_setg(errp, "property die-id: %u doesn't match set apic-id:"
388             " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id);
389         return;
390     }
391     cpu->die_id = topo_ids.die_id;
392 
393     if (cpu->module_id != -1 && cpu->module_id != topo_ids.module_id) {
394         error_setg(errp, "property module-id: %u doesn't match set apic-id:"
395             " 0x%x (module-id: %u)", cpu->module_id, cpu->apic_id,
396             topo_ids.module_id);
397         return;
398     }
399     cpu->module_id = topo_ids.module_id;
400 
401     if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) {
402         error_setg(errp, "property core-id: %u doesn't match set apic-id:"
403             " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id,
404             topo_ids.core_id);
405         return;
406     }
407     cpu->core_id = topo_ids.core_id;
408 
409     if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) {
410         error_setg(errp, "property thread-id: %u doesn't match set apic-id:"
411             " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id,
412             topo_ids.smt_id);
413         return;
414     }
415     cpu->thread_id = topo_ids.smt_id;
416 
417     /*
418     * kvm_enabled() must go first to ensure that kvm_* references are
419     * not emitted for the linker to consume (kvm_enabled() is
420     * a literal `0` in configurations where kvm_* aren't defined)
421     */
422     if (kvm_enabled() && hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) &&
423         !kvm_hv_vpindex_settable()) {
424         error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX");
425         return;
426     }
427 
428     cs = CPU(cpu);
429     cs->cpu_index = idx;
430 
431     numa_cpu_pre_plug(cpu_slot, dev, errp);
432 }
433 
434 static long get_file_size(FILE *f)
435 {
436     long where, size;
437 
438     /* XXX: on Unix systems, using fstat() probably makes more sense */
439 
440     where = ftell(f);
441     fseek(f, 0, SEEK_END);
442     size = ftell(f);
443     fseek(f, where, SEEK_SET);
444 
445     return size;
446 }
447 
448 void gsi_handler(void *opaque, int n, int level)
449 {
450     GSIState *s = opaque;
451 
452     trace_x86_gsi_interrupt(n, level);
453     switch (n) {
454     case 0 ... ISA_NUM_IRQS - 1:
455         if (s->i8259_irq[n]) {
456             /* Under KVM, Kernel will forward to both PIC and IOAPIC */
457             qemu_set_irq(s->i8259_irq[n], level);
458         }
459         /* fall through */
460     case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
461 #ifdef CONFIG_XEN_EMU
462         /*
463          * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
464          * routing actually works properly under Xen). And then to
465          * *either* the PIRQ handling or the I/OAPIC depending on
466          * whether the former wants it.
467          */
468         if (xen_mode == XEN_EMULATE && xen_evtchn_set_gsi(n, level)) {
469             break;
470         }
471 #endif
472         qemu_set_irq(s->ioapic_irq[n], level);
473         break;
474     case IO_APIC_SECONDARY_IRQBASE
475         ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1:
476         qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level);
477         break;
478     }
479 }
480 
481 void ioapic_init_gsi(GSIState *gsi_state, Object *parent)
482 {
483     DeviceState *dev;
484     SysBusDevice *d;
485     unsigned int i;
486 
487     assert(parent);
488     if (kvm_ioapic_in_kernel()) {
489         dev = qdev_new(TYPE_KVM_IOAPIC);
490     } else {
491         dev = qdev_new(TYPE_IOAPIC);
492     }
493     object_property_add_child(parent, "ioapic", OBJECT(dev));
494     d = SYS_BUS_DEVICE(dev);
495     sysbus_realize_and_unref(d, &error_fatal);
496     sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS);
497 
498     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
499         gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i);
500     }
501 }
502 
503 DeviceState *ioapic_init_secondary(GSIState *gsi_state)
504 {
505     DeviceState *dev;
506     SysBusDevice *d;
507     unsigned int i;
508 
509     dev = qdev_new(TYPE_IOAPIC);
510     d = SYS_BUS_DEVICE(dev);
511     sysbus_realize_and_unref(d, &error_fatal);
512     sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS);
513 
514     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
515         gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i);
516     }
517     return dev;
518 }
519 
520 /*
521  * The entry point into the kernel for PVH boot is different from
522  * the native entry point.  The PVH entry is defined by the x86/HVM
523  * direct boot ABI and is available in an ELFNOTE in the kernel binary.
524  *
525  * This function is passed to load_elf() when it is called from
526  * load_elfboot() which then additionally checks for an ELF Note of
527  * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
528  * parse the PVH entry address from the ELF Note.
529  *
530  * Due to trickery in elf_opts.h, load_elf() is actually available as
531  * load_elf32() or load_elf64() and this routine needs to be able
532  * to deal with being called as 32 or 64 bit.
533  *
534  * The address of the PVH entry point is saved to the 'pvh_start_addr'
535  * global variable.  (although the entry point is 32-bit, the kernel
536  * binary can be either 32-bit or 64-bit).
537  */
538 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64)
539 {
540     size_t *elf_note_data_addr;
541 
542     /* Check if ELF Note header passed in is valid */
543     if (arg1 == NULL) {
544         return 0;
545     }
546 
547     if (is64) {
548         struct elf64_note *nhdr64 = (struct elf64_note *)arg1;
549         uint64_t nhdr_size64 = sizeof(struct elf64_note);
550         uint64_t phdr_align = *(uint64_t *)arg2;
551         uint64_t nhdr_namesz = nhdr64->n_namesz;
552 
553         elf_note_data_addr =
554             ((void *)nhdr64) + nhdr_size64 +
555             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
556 
557         pvh_start_addr = *elf_note_data_addr;
558     } else {
559         struct elf32_note *nhdr32 = (struct elf32_note *)arg1;
560         uint32_t nhdr_size32 = sizeof(struct elf32_note);
561         uint32_t phdr_align = *(uint32_t *)arg2;
562         uint32_t nhdr_namesz = nhdr32->n_namesz;
563 
564         elf_note_data_addr =
565             ((void *)nhdr32) + nhdr_size32 +
566             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
567 
568         pvh_start_addr = *(uint32_t *)elf_note_data_addr;
569     }
570 
571     return pvh_start_addr;
572 }
573 
574 static bool load_elfboot(const char *kernel_filename,
575                          int kernel_file_size,
576                          uint8_t *header,
577                          size_t pvh_xen_start_addr,
578                          FWCfgState *fw_cfg)
579 {
580     uint32_t flags = 0;
581     uint32_t mh_load_addr = 0;
582     uint32_t elf_kernel_size = 0;
583     uint64_t elf_entry;
584     uint64_t elf_low, elf_high;
585     int kernel_size;
586 
587     if (ldl_le_p(header) != 0x464c457f) {
588         return false; /* no elfboot */
589     }
590 
591     bool elf_is64 = header[EI_CLASS] == ELFCLASS64;
592     flags = elf_is64 ?
593         ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags;
594 
595     if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
596         error_report("elfboot unsupported flags = %x", flags);
597         exit(1);
598     }
599 
600     uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY;
601     kernel_size = load_elf(kernel_filename, read_pvh_start_addr,
602                            NULL, &elf_note_type, &elf_entry,
603                            &elf_low, &elf_high, NULL, 0, I386_ELF_MACHINE,
604                            0, 0);
605 
606     if (kernel_size < 0) {
607         error_report("Error while loading elf kernel");
608         exit(1);
609     }
610     mh_load_addr = elf_low;
611     elf_kernel_size = elf_high - elf_low;
612 
613     if (pvh_start_addr == 0) {
614         error_report("Error loading uncompressed kernel without PVH ELF Note");
615         exit(1);
616     }
617     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr);
618     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr);
619     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size);
620 
621     return true;
622 }
623 
624 void x86_load_linux(X86MachineState *x86ms,
625                     FWCfgState *fw_cfg,
626                     int acpi_data_size,
627                     bool pvh_enabled)
628 {
629     bool linuxboot_dma_enabled = X86_MACHINE_GET_CLASS(x86ms)->fwcfg_dma_enabled;
630     uint16_t protocol;
631     int setup_size, kernel_size, cmdline_size;
632     int dtb_size, setup_data_offset;
633     uint32_t initrd_max;
634     uint8_t header[8192], *setup, *kernel;
635     hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0;
636     FILE *f;
637     char *vmode;
638     MachineState *machine = MACHINE(x86ms);
639     struct setup_data *setup_data;
640     const char *kernel_filename = machine->kernel_filename;
641     const char *initrd_filename = machine->initrd_filename;
642     const char *dtb_filename = machine->dtb;
643     const char *kernel_cmdline = machine->kernel_cmdline;
644     SevKernelLoaderContext sev_load_ctx = {};
645 
646     /* Align to 16 bytes as a paranoia measure */
647     cmdline_size = (strlen(kernel_cmdline) + 16) & ~15;
648 
649     /* load the kernel header */
650     f = fopen(kernel_filename, "rb");
651     if (!f) {
652         fprintf(stderr, "qemu: could not open kernel file '%s': %s\n",
653                 kernel_filename, strerror(errno));
654         exit(1);
655     }
656 
657     kernel_size = get_file_size(f);
658     if (!kernel_size ||
659         fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
660         MIN(ARRAY_SIZE(header), kernel_size)) {
661         fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
662                 kernel_filename, strerror(errno));
663         exit(1);
664     }
665 
666     /*
667      * kernel protocol version.
668      * Please see https://www.kernel.org/doc/Documentation/x86/boot.txt
669      */
670     if (ldl_le_p(header + 0x202) == 0x53726448) /* Magic signature "HdrS" */ {
671         protocol = lduw_le_p(header + 0x206);
672     } else {
673         /*
674          * This could be a multiboot kernel. If it is, let's stop treating it
675          * like a Linux kernel.
676          * Note: some multiboot images could be in the ELF format (the same of
677          * PVH), so we try multiboot first since we check the multiboot magic
678          * header before to load it.
679          */
680         if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename,
681                            kernel_cmdline, kernel_size, header)) {
682             return;
683         }
684         /*
685          * Check if the file is an uncompressed kernel file (ELF) and load it,
686          * saving the PVH entry point used by the x86/HVM direct boot ABI.
687          * If load_elfboot() is successful, populate the fw_cfg info.
688          */
689         if (pvh_enabled &&
690             load_elfboot(kernel_filename, kernel_size,
691                          header, pvh_start_addr, fw_cfg)) {
692             fclose(f);
693 
694             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
695                 strlen(kernel_cmdline) + 1);
696             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
697 
698             setup = g_memdup2(header, sizeof(header));
699 
700             fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header));
701             fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA,
702                              setup, sizeof(header));
703 
704             /* load initrd */
705             if (initrd_filename) {
706                 GMappedFile *mapped_file;
707                 gsize initrd_size;
708                 gchar *initrd_data;
709                 GError *gerr = NULL;
710 
711                 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
712                 if (!mapped_file) {
713                     fprintf(stderr, "qemu: error reading initrd %s: %s\n",
714                             initrd_filename, gerr->message);
715                     exit(1);
716                 }
717                 x86ms->initrd_mapped_file = mapped_file;
718 
719                 initrd_data = g_mapped_file_get_contents(mapped_file);
720                 initrd_size = g_mapped_file_get_length(mapped_file);
721                 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
722                 if (initrd_size >= initrd_max) {
723                     fprintf(stderr, "qemu: initrd is too large, cannot support."
724                             "(max: %"PRIu32", need %"PRId64")\n",
725                             initrd_max, (uint64_t)initrd_size);
726                     exit(1);
727                 }
728 
729                 initrd_addr = (initrd_max - initrd_size) & ~4095;
730 
731                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
732                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
733                 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data,
734                                  initrd_size);
735             }
736 
737             option_rom[nb_option_roms].bootindex = 0;
738             option_rom[nb_option_roms].name = "pvh.bin";
739             nb_option_roms++;
740 
741             return;
742         }
743         protocol = 0;
744     }
745 
746     if (protocol < 0x200 || !(header[0x211] & 0x01)) {
747         /* Low kernel */
748         real_addr    = 0x90000;
749         cmdline_addr = 0x9a000 - cmdline_size;
750         prot_addr    = 0x10000;
751     } else if (protocol < 0x202) {
752         /* High but ancient kernel */
753         real_addr    = 0x90000;
754         cmdline_addr = 0x9a000 - cmdline_size;
755         prot_addr    = 0x100000;
756     } else {
757         /* High and recent kernel */
758         real_addr    = 0x10000;
759         cmdline_addr = 0x20000;
760         prot_addr    = 0x100000;
761     }
762 
763     /* highest address for loading the initrd */
764     if (protocol >= 0x20c &&
765         lduw_le_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) {
766         /*
767          * Linux has supported initrd up to 4 GB for a very long time (2007,
768          * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
769          * though it only sets initrd_max to 2 GB to "work around bootloader
770          * bugs". Luckily, QEMU firmware(which does something like bootloader)
771          * has supported this.
772          *
773          * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
774          * be loaded into any address.
775          *
776          * In addition, initrd_max is uint32_t simply because QEMU doesn't
777          * support the 64-bit boot protocol (specifically the ext_ramdisk_image
778          * field).
779          *
780          * Therefore here just limit initrd_max to UINT32_MAX simply as well.
781          */
782         initrd_max = UINT32_MAX;
783     } else if (protocol >= 0x203) {
784         initrd_max = ldl_le_p(header + 0x22c);
785     } else {
786         initrd_max = 0x37ffffff;
787     }
788 
789     if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) {
790         initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
791     }
792 
793     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
794     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline) + 1);
795     fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
796     sev_load_ctx.cmdline_data = (char *)kernel_cmdline;
797     sev_load_ctx.cmdline_size = strlen(kernel_cmdline) + 1;
798 
799     if (protocol >= 0x202) {
800         stl_le_p(header + 0x228, cmdline_addr);
801     } else {
802         stw_le_p(header + 0x20, 0xA33F);
803         stw_le_p(header + 0x22, cmdline_addr - real_addr);
804     }
805 
806     /* handle vga= parameter */
807     vmode = strstr(kernel_cmdline, "vga=");
808     if (vmode) {
809         unsigned int video_mode;
810         const char *end;
811         int ret;
812         /* skip "vga=" */
813         vmode += 4;
814         if (!strncmp(vmode, "normal", 6)) {
815             video_mode = 0xffff;
816         } else if (!strncmp(vmode, "ext", 3)) {
817             video_mode = 0xfffe;
818         } else if (!strncmp(vmode, "ask", 3)) {
819             video_mode = 0xfffd;
820         } else {
821             ret = qemu_strtoui(vmode, &end, 0, &video_mode);
822             if (ret != 0 || (*end && *end != ' ')) {
823                 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n");
824                 exit(1);
825             }
826         }
827         stw_le_p(header + 0x1fa, video_mode);
828     }
829 
830     /* loader type */
831     /*
832      * High nybble = B reserved for QEMU; low nybble is revision number.
833      * If this code is substantially changed, you may want to consider
834      * incrementing the revision.
835      */
836     if (protocol >= 0x200) {
837         header[0x210] = 0xB0;
838     }
839     /* heap */
840     if (protocol >= 0x201) {
841         header[0x211] |= 0x80; /* CAN_USE_HEAP */
842         stw_le_p(header + 0x224, cmdline_addr - real_addr - 0x200);
843     }
844 
845     /* load initrd */
846     if (initrd_filename) {
847         GMappedFile *mapped_file;
848         gsize initrd_size;
849         gchar *initrd_data;
850         GError *gerr = NULL;
851 
852         if (protocol < 0x200) {
853             fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
854             exit(1);
855         }
856 
857         mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
858         if (!mapped_file) {
859             fprintf(stderr, "qemu: error reading initrd %s: %s\n",
860                     initrd_filename, gerr->message);
861             exit(1);
862         }
863         x86ms->initrd_mapped_file = mapped_file;
864 
865         initrd_data = g_mapped_file_get_contents(mapped_file);
866         initrd_size = g_mapped_file_get_length(mapped_file);
867         if (initrd_size >= initrd_max) {
868             fprintf(stderr, "qemu: initrd is too large, cannot support."
869                     "(max: %"PRIu32", need %"PRId64")\n",
870                     initrd_max, (uint64_t)initrd_size);
871             exit(1);
872         }
873 
874         initrd_addr = (initrd_max - initrd_size) & ~4095;
875 
876         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
877         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
878         fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
879         sev_load_ctx.initrd_data = initrd_data;
880         sev_load_ctx.initrd_size = initrd_size;
881 
882         stl_le_p(header + 0x218, initrd_addr);
883         stl_le_p(header + 0x21c, initrd_size);
884     }
885 
886     /* load kernel and setup */
887     setup_size = header[0x1f1];
888     if (setup_size == 0) {
889         setup_size = 4;
890     }
891     setup_size = (setup_size + 1) * 512;
892     if (setup_size > kernel_size) {
893         fprintf(stderr, "qemu: invalid kernel header\n");
894         exit(1);
895     }
896 
897     setup  = g_malloc(setup_size);
898     kernel = g_malloc(kernel_size);
899     fseek(f, 0, SEEK_SET);
900     if (fread(setup, 1, setup_size, f) != setup_size) {
901         fprintf(stderr, "fread() failed\n");
902         exit(1);
903     }
904     fseek(f, 0, SEEK_SET);
905     if (fread(kernel, 1, kernel_size, f) != kernel_size) {
906         fprintf(stderr, "fread() failed\n");
907         exit(1);
908     }
909     fclose(f);
910 
911     /* append dtb to kernel */
912     if (dtb_filename) {
913         if (protocol < 0x209) {
914             fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n");
915             exit(1);
916         }
917 
918         dtb_size = get_image_size(dtb_filename);
919         if (dtb_size <= 0) {
920             fprintf(stderr, "qemu: error reading dtb %s: %s\n",
921                     dtb_filename, strerror(errno));
922             exit(1);
923         }
924 
925         setup_data_offset = QEMU_ALIGN_UP(kernel_size, 16);
926         kernel_size = setup_data_offset + sizeof(struct setup_data) + dtb_size;
927         kernel = g_realloc(kernel, kernel_size);
928 
929         stq_le_p(header + 0x250, prot_addr + setup_data_offset);
930 
931         setup_data = (struct setup_data *)(kernel + setup_data_offset);
932         setup_data->next = 0;
933         setup_data->type = cpu_to_le32(SETUP_DTB);
934         setup_data->len = cpu_to_le32(dtb_size);
935 
936         load_image_size(dtb_filename, setup_data->data, dtb_size);
937     }
938 
939     /*
940      * If we're starting an encrypted VM, it will be OVMF based, which uses the
941      * efi stub for booting and doesn't require any values to be placed in the
942      * kernel header.  We therefore don't update the header so the hash of the
943      * kernel on the other side of the fw_cfg interface matches the hash of the
944      * file the user passed in.
945      */
946     if (!sev_enabled() && protocol > 0) {
947         memcpy(setup, header, MIN(sizeof(header), setup_size));
948     }
949 
950     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
951     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size - setup_size);
952     fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA,
953                      kernel + setup_size, kernel_size - setup_size);
954     sev_load_ctx.kernel_data = (char *)kernel + setup_size;
955     sev_load_ctx.kernel_size = kernel_size - setup_size;
956 
957     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
958     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
959     fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
960     sev_load_ctx.setup_data = (char *)setup;
961     sev_load_ctx.setup_size = setup_size;
962 
963     /* kernel without setup header patches */
964     fw_cfg_add_file(fw_cfg, "etc/boot/kernel", kernel, kernel_size);
965 
966     if (machine->shim_filename) {
967         GMappedFile *mapped_file;
968         GError *gerr = NULL;
969 
970         mapped_file = g_mapped_file_new(machine->shim_filename, false, &gerr);
971         if (!mapped_file) {
972             fprintf(stderr, "qemu: error reading shim %s: %s\n",
973                     machine->shim_filename, gerr->message);
974             exit(1);
975         }
976 
977         fw_cfg_add_file(fw_cfg, "etc/boot/shim",
978                         g_mapped_file_get_contents(mapped_file),
979                         g_mapped_file_get_length(mapped_file));
980     }
981 
982     if (sev_enabled()) {
983         sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal);
984     }
985 
986     option_rom[nb_option_roms].bootindex = 0;
987     option_rom[nb_option_roms].name = "linuxboot.bin";
988     if (linuxboot_dma_enabled && fw_cfg_dma_enabled(fw_cfg)) {
989         option_rom[nb_option_roms].name = "linuxboot_dma.bin";
990     }
991     nb_option_roms++;
992 }
993 
994 void x86_isa_bios_init(MemoryRegion *isa_bios, MemoryRegion *isa_memory,
995                        MemoryRegion *bios, bool read_only)
996 {
997     uint64_t bios_size = memory_region_size(bios);
998     uint64_t isa_bios_size = MIN(bios_size, 128 * KiB);
999 
1000     memory_region_init_alias(isa_bios, NULL, "isa-bios", bios,
1001                              bios_size - isa_bios_size, isa_bios_size);
1002     memory_region_add_subregion_overlap(isa_memory, 1 * MiB - isa_bios_size,
1003                                         isa_bios, 1);
1004     memory_region_set_readonly(isa_bios, read_only);
1005 }
1006 
1007 void x86_bios_rom_init(X86MachineState *x86ms, const char *default_firmware,
1008                        MemoryRegion *rom_memory, bool isapc_ram_fw)
1009 {
1010     const char *bios_name;
1011     char *filename;
1012     int bios_size;
1013     ssize_t ret;
1014 
1015     /* BIOS load */
1016     bios_name = MACHINE(x86ms)->firmware ?: default_firmware;
1017     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1018     if (filename) {
1019         bios_size = get_image_size(filename);
1020     } else {
1021         bios_size = -1;
1022     }
1023     if (bios_size <= 0 ||
1024         (bios_size % 65536) != 0) {
1025         goto bios_error;
1026     }
1027     if (machine_require_guest_memfd(MACHINE(x86ms))) {
1028         memory_region_init_ram_guest_memfd(&x86ms->bios, NULL, "pc.bios",
1029                                            bios_size, &error_fatal);
1030     } else {
1031         memory_region_init_ram(&x86ms->bios, NULL, "pc.bios",
1032                                bios_size, &error_fatal);
1033     }
1034     if (sev_enabled()) {
1035         /*
1036          * The concept of a "reset" simply doesn't exist for
1037          * confidential computing guests, we have to destroy and
1038          * re-launch them instead.  So there is no need to register
1039          * the firmware as rom to properly re-initialize on reset.
1040          * Just go for a straight file load instead.
1041          */
1042         void *ptr = memory_region_get_ram_ptr(&x86ms->bios);
1043         load_image_size(filename, ptr, bios_size);
1044         x86_firmware_configure(0x100000000ULL - bios_size, ptr, bios_size);
1045     } else {
1046         memory_region_set_readonly(&x86ms->bios, !isapc_ram_fw);
1047         ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1);
1048         if (ret != 0) {
1049             goto bios_error;
1050         }
1051     }
1052     g_free(filename);
1053 
1054     if (!machine_require_guest_memfd(MACHINE(x86ms))) {
1055         /* map the last 128KB of the BIOS in ISA space */
1056         x86_isa_bios_init(&x86ms->isa_bios, rom_memory, &x86ms->bios,
1057                           !isapc_ram_fw);
1058     }
1059 
1060     /* map all the bios at the top of memory */
1061     memory_region_add_subregion(rom_memory,
1062                                 (uint32_t)(-bios_size),
1063                                 &x86ms->bios);
1064     return;
1065 
1066 bios_error:
1067     fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1068     exit(1);
1069 }
1070