2 * Copyright (c) 2003-2004 Fabrice Bellard
3 * Copyright (c) 2019, 2024 Red Hat, Inc.
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:
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
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
23 #include "qemu/osdep.h"
24 #include "qemu/error-report.h"
25 #include "qemu/cutils.h"
26 #include "qemu/units.h"
27 #include "qemu/datadir.h"
28 #include "qapi/error.h"
29 #include "sysemu/numa.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/xen.h"
34 #include "hw/i386/x86.h"
35 #include "target/i386/cpu.h"
36 #include "hw/rtc/mc146818rtc.h"
37 #include "target/i386/sev.h"
39 #include "hw/acpi/cpu_hotplug.h"
41 #include "hw/loader.h"
42 #include "multiboot.h"
44 #include "standard-headers/asm-x86/bootparam.h"
45 #include CONFIG_DEVICES
46 #include "kvm/kvm_i386.h"
49 #include "hw/xen/xen.h"
50 #include "hw/i386/kvm/xen_evtchn.h"
53 /* Physical Address of PVH entry point read from kernel ELF NOTE */
54 static size_t pvh_start_addr
;
56 static void x86_cpu_new(X86MachineState
*x86ms
, int64_t apic_id
, Error
**errp
)
58 Object
*cpu
= object_new(MACHINE(x86ms
)->cpu_type
);
60 if (!object_property_set_uint(cpu
, "apic-id", apic_id
, errp
)) {
63 qdev_realize(DEVICE(cpu
), NULL
, errp
);
69 void x86_cpus_init(X86MachineState
*x86ms
, int default_cpu_version
)
72 const CPUArchIdList
*possible_cpus
;
73 MachineState
*ms
= MACHINE(x86ms
);
74 MachineClass
*mc
= MACHINE_GET_CLASS(x86ms
);
76 x86_cpu_set_default_version(default_cpu_version
);
79 * Calculates the limit to CPU APIC ID values
81 * Limit for the APIC ID value, so that all
82 * CPU APIC IDs are < x86ms->apic_id_limit.
84 * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
86 x86ms
->apic_id_limit
= x86_cpu_apic_id_from_index(x86ms
,
87 ms
->smp
.max_cpus
- 1) + 1;
90 * Can we support APIC ID 255 or higher? With KVM, that requires
91 * both in-kernel lapic and X2APIC userspace API.
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)
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.");
105 kvm_set_max_apic_id(x86ms
->apic_id_limit
);
108 if (!kvm_irqchip_in_kernel()) {
109 apic_set_max_apic_id(x86ms
->apic_id_limit
);
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
);
118 void x86_rtc_set_cpus_count(ISADevice
*s
, uint16_t cpus_count
)
120 MC146818RtcState
*rtc
= MC146818_RTC(s
);
122 if (cpus_count
> 0xff) {
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.
128 mc146818rtc_set_cmos_data(rtc
, 0x5f, 0);
130 mc146818rtc_set_cmos_data(rtc
, 0x5f, cpus_count
- 1);
134 static int x86_apic_cmp(const void *a
, const void *b
)
136 CPUArchId
*apic_a
= (CPUArchId
*)a
;
137 CPUArchId
*apic_b
= (CPUArchId
*)b
;
139 return apic_a
->arch_id
- apic_b
->arch_id
;
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.
147 static CPUArchId
*x86_find_cpu_slot(MachineState
*ms
, uint32_t id
, int *idx
)
149 CPUArchId apic_id
, *found_cpu
;
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
),
155 if (found_cpu
&& idx
) {
156 *idx
= found_cpu
- ms
->possible_cpus
->cpus
;
161 void x86_cpu_plug(HotplugHandler
*hotplug_dev
,
162 DeviceState
*dev
, Error
**errp
)
164 CPUArchId
*found_cpu
;
165 Error
*local_err
= NULL
;
166 X86CPU
*cpu
= X86_CPU(dev
);
167 X86MachineState
*x86ms
= X86_MACHINE(hotplug_dev
);
169 if (x86ms
->acpi_dev
) {
170 hotplug_handler_plug(x86ms
->acpi_dev
, dev
, &local_err
);
176 /* increment the number of CPUs */
179 x86_rtc_set_cpus_count(x86ms
->rtc
, x86ms
->boot_cpus
);
182 fw_cfg_modify_i16(x86ms
->fw_cfg
, FW_CFG_NB_CPUS
, x86ms
->boot_cpus
);
185 found_cpu
= x86_find_cpu_slot(MACHINE(x86ms
), cpu
->apic_id
, NULL
);
186 found_cpu
->cpu
= CPU(dev
);
188 error_propagate(errp
, local_err
);
191 void x86_cpu_unplug_request_cb(HotplugHandler
*hotplug_dev
,
192 DeviceState
*dev
, Error
**errp
)
195 X86CPU
*cpu
= X86_CPU(dev
);
196 X86MachineState
*x86ms
= X86_MACHINE(hotplug_dev
);
198 if (!x86ms
->acpi_dev
) {
199 error_setg(errp
, "CPU hot unplug not supported without ACPI");
203 x86_find_cpu_slot(MACHINE(x86ms
), cpu
->apic_id
, &idx
);
206 error_setg(errp
, "Boot CPU is unpluggable");
210 hotplug_handler_unplug_request(x86ms
->acpi_dev
, dev
,
214 void x86_cpu_unplug_cb(HotplugHandler
*hotplug_dev
,
215 DeviceState
*dev
, Error
**errp
)
217 CPUArchId
*found_cpu
;
218 Error
*local_err
= NULL
;
219 X86CPU
*cpu
= X86_CPU(dev
);
220 X86MachineState
*x86ms
= X86_MACHINE(hotplug_dev
);
222 hotplug_handler_unplug(x86ms
->acpi_dev
, dev
, &local_err
);
227 found_cpu
= x86_find_cpu_slot(MACHINE(x86ms
), cpu
->apic_id
, NULL
);
228 found_cpu
->cpu
= NULL
;
231 /* decrement the number of 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
);
237 error_propagate(errp
, local_err
);
240 void x86_cpu_pre_plug(HotplugHandler
*hotplug_dev
,
241 DeviceState
*dev
, Error
**errp
)
246 X86CPUTopoIDs topo_ids
;
247 X86CPU
*cpu
= X86_CPU(dev
);
248 CPUX86State
*env
= &cpu
->env
;
249 MachineState
*ms
= MACHINE(hotplug_dev
);
250 X86MachineState
*x86ms
= X86_MACHINE(hotplug_dev
);
251 unsigned int smp_cores
= ms
->smp
.cores
;
252 unsigned int smp_threads
= ms
->smp
.threads
;
253 X86CPUTopoInfo topo_info
;
255 if (!object_dynamic_cast(OBJECT(cpu
), ms
->cpu_type
)) {
256 error_setg(errp
, "Invalid CPU type, expected cpu type: '%s'",
261 if (x86ms
->acpi_dev
) {
262 Error
*local_err
= NULL
;
264 hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms
->acpi_dev
), dev
,
267 error_propagate(errp
, local_err
);
272 init_topo_info(&topo_info
, x86ms
);
274 if (ms
->smp
.modules
> 1) {
275 env
->nr_modules
= ms
->smp
.modules
;
276 set_bit(CPU_TOPO_LEVEL_MODULE
, env
->avail_cpu_topo
);
279 if (ms
->smp
.dies
> 1) {
280 env
->nr_dies
= ms
->smp
.dies
;
281 set_bit(CPU_TOPO_LEVEL_DIE
, env
->avail_cpu_topo
);
285 * If APIC ID is not set,
286 * set it based on socket/die/module/core/thread properties.
288 if (cpu
->apic_id
== UNASSIGNED_APIC_ID
) {
290 * die-id was optional in QEMU 4.0 and older, so keep it optional
291 * if there's only one die per socket.
293 if (cpu
->die_id
< 0 && ms
->smp
.dies
== 1) {
298 * module-id was optional in QEMU 9.0 and older, so keep it optional
299 * if there's only one module per die.
301 if (cpu
->module_id
< 0 && ms
->smp
.modules
== 1) {
305 if (cpu
->socket_id
< 0) {
306 error_setg(errp
, "CPU socket-id is not set");
308 } else if (cpu
->socket_id
> ms
->smp
.sockets
- 1) {
309 error_setg(errp
, "Invalid CPU socket-id: %u must be in range 0:%u",
310 cpu
->socket_id
, ms
->smp
.sockets
- 1);
313 if (cpu
->die_id
< 0) {
314 error_setg(errp
, "CPU die-id is not set");
316 } else if (cpu
->die_id
> ms
->smp
.dies
- 1) {
317 error_setg(errp
, "Invalid CPU die-id: %u must be in range 0:%u",
318 cpu
->die_id
, ms
->smp
.dies
- 1);
321 if (cpu
->module_id
< 0) {
322 error_setg(errp
, "CPU module-id is not set");
324 } else if (cpu
->module_id
> ms
->smp
.modules
- 1) {
325 error_setg(errp
, "Invalid CPU module-id: %u must be in range 0:%u",
326 cpu
->module_id
, ms
->smp
.modules
- 1);
329 if (cpu
->core_id
< 0) {
330 error_setg(errp
, "CPU core-id is not set");
332 } else if (cpu
->core_id
> (smp_cores
- 1)) {
333 error_setg(errp
, "Invalid CPU core-id: %u must be in range 0:%u",
334 cpu
->core_id
, smp_cores
- 1);
337 if (cpu
->thread_id
< 0) {
338 error_setg(errp
, "CPU thread-id is not set");
340 } else if (cpu
->thread_id
> (smp_threads
- 1)) {
341 error_setg(errp
, "Invalid CPU thread-id: %u must be in range 0:%u",
342 cpu
->thread_id
, smp_threads
- 1);
346 topo_ids
.pkg_id
= cpu
->socket_id
;
347 topo_ids
.die_id
= cpu
->die_id
;
348 topo_ids
.module_id
= cpu
->module_id
;
349 topo_ids
.core_id
= cpu
->core_id
;
350 topo_ids
.smt_id
= cpu
->thread_id
;
351 cpu
->apic_id
= x86_apicid_from_topo_ids(&topo_info
, &topo_ids
);
354 cpu_slot
= x86_find_cpu_slot(MACHINE(x86ms
), cpu
->apic_id
, &idx
);
356 x86_topo_ids_from_apicid(cpu
->apic_id
, &topo_info
, &topo_ids
);
359 "Invalid CPU [socket: %u, die: %u, module: %u, core: %u, thread: %u]"
360 " with APIC ID %" PRIu32
", valid index range 0:%d",
361 topo_ids
.pkg_id
, topo_ids
.die_id
, topo_ids
.module_id
,
362 topo_ids
.core_id
, topo_ids
.smt_id
, cpu
->apic_id
,
363 ms
->possible_cpus
->len
- 1);
368 error_setg(errp
, "CPU[%d] with APIC ID %" PRIu32
" exists",
373 /* if 'address' properties socket-id/core-id/thread-id are not set, set them
374 * so that machine_query_hotpluggable_cpus would show correct values
376 /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
377 * once -smp refactoring is complete and there will be CPU private
378 * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
379 x86_topo_ids_from_apicid(cpu
->apic_id
, &topo_info
, &topo_ids
);
380 if (cpu
->socket_id
!= -1 && cpu
->socket_id
!= topo_ids
.pkg_id
) {
381 error_setg(errp
, "property socket-id: %u doesn't match set apic-id:"
382 " 0x%x (socket-id: %u)", cpu
->socket_id
, cpu
->apic_id
,
386 cpu
->socket_id
= topo_ids
.pkg_id
;
388 if (cpu
->die_id
!= -1 && cpu
->die_id
!= topo_ids
.die_id
) {
389 error_setg(errp
, "property die-id: %u doesn't match set apic-id:"
390 " 0x%x (die-id: %u)", cpu
->die_id
, cpu
->apic_id
, topo_ids
.die_id
);
393 cpu
->die_id
= topo_ids
.die_id
;
395 if (cpu
->module_id
!= -1 && cpu
->module_id
!= topo_ids
.module_id
) {
396 error_setg(errp
, "property module-id: %u doesn't match set apic-id:"
397 " 0x%x (module-id: %u)", cpu
->module_id
, cpu
->apic_id
,
401 cpu
->module_id
= topo_ids
.module_id
;
403 if (cpu
->core_id
!= -1 && cpu
->core_id
!= topo_ids
.core_id
) {
404 error_setg(errp
, "property core-id: %u doesn't match set apic-id:"
405 " 0x%x (core-id: %u)", cpu
->core_id
, cpu
->apic_id
,
409 cpu
->core_id
= topo_ids
.core_id
;
411 if (cpu
->thread_id
!= -1 && cpu
->thread_id
!= topo_ids
.smt_id
) {
412 error_setg(errp
, "property thread-id: %u doesn't match set apic-id:"
413 " 0x%x (thread-id: %u)", cpu
->thread_id
, cpu
->apic_id
,
417 cpu
->thread_id
= topo_ids
.smt_id
;
420 * kvm_enabled() must go first to ensure that kvm_* references are
421 * not emitted for the linker to consume (kvm_enabled() is
422 * a literal `0` in configurations where kvm_* aren't defined)
424 if (kvm_enabled() && hyperv_feat_enabled(cpu
, HYPERV_FEAT_VPINDEX
) &&
425 !kvm_hv_vpindex_settable()) {
426 error_setg(errp
, "kernel doesn't allow setting HyperV VP_INDEX");
433 numa_cpu_pre_plug(cpu_slot
, dev
, errp
);
436 static long get_file_size(FILE *f
)
440 /* XXX: on Unix systems, using fstat() probably makes more sense */
443 fseek(f
, 0, SEEK_END
);
445 fseek(f
, where
, SEEK_SET
);
450 void gsi_handler(void *opaque
, int n
, int level
)
452 GSIState
*s
= opaque
;
454 trace_x86_gsi_interrupt(n
, level
);
456 case 0 ... ISA_NUM_IRQS
- 1:
457 if (s
->i8259_irq
[n
]) {
458 /* Under KVM, Kernel will forward to both PIC and IOAPIC */
459 qemu_set_irq(s
->i8259_irq
[n
], level
);
462 case ISA_NUM_IRQS
... IOAPIC_NUM_PINS
- 1:
463 #ifdef CONFIG_XEN_EMU
465 * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
466 * routing actually works properly under Xen). And then to
467 * *either* the PIRQ handling or the I/OAPIC depending on
468 * whether the former wants it.
470 if (xen_mode
== XEN_EMULATE
&& xen_evtchn_set_gsi(n
, level
)) {
474 qemu_set_irq(s
->ioapic_irq
[n
], level
);
476 case IO_APIC_SECONDARY_IRQBASE
477 ... IO_APIC_SECONDARY_IRQBASE
+ IOAPIC_NUM_PINS
- 1:
478 qemu_set_irq(s
->ioapic2_irq
[n
- IO_APIC_SECONDARY_IRQBASE
], level
);
483 void ioapic_init_gsi(GSIState
*gsi_state
, Object
*parent
)
490 if (kvm_ioapic_in_kernel()) {
491 dev
= qdev_new(TYPE_KVM_IOAPIC
);
493 dev
= qdev_new(TYPE_IOAPIC
);
495 object_property_add_child(parent
, "ioapic", OBJECT(dev
));
496 d
= SYS_BUS_DEVICE(dev
);
497 sysbus_realize_and_unref(d
, &error_fatal
);
498 sysbus_mmio_map(d
, 0, IO_APIC_DEFAULT_ADDRESS
);
500 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++) {
501 gsi_state
->ioapic_irq
[i
] = qdev_get_gpio_in(dev
, i
);
505 DeviceState
*ioapic_init_secondary(GSIState
*gsi_state
)
511 dev
= qdev_new(TYPE_IOAPIC
);
512 d
= SYS_BUS_DEVICE(dev
);
513 sysbus_realize_and_unref(d
, &error_fatal
);
514 sysbus_mmio_map(d
, 0, IO_APIC_SECONDARY_ADDRESS
);
516 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++) {
517 gsi_state
->ioapic2_irq
[i
] = qdev_get_gpio_in(dev
, i
);
523 * The entry point into the kernel for PVH boot is different from
524 * the native entry point. The PVH entry is defined by the x86/HVM
525 * direct boot ABI and is available in an ELFNOTE in the kernel binary.
527 * This function is passed to load_elf() when it is called from
528 * load_elfboot() which then additionally checks for an ELF Note of
529 * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
530 * parse the PVH entry address from the ELF Note.
532 * Due to trickery in elf_opts.h, load_elf() is actually available as
533 * load_elf32() or load_elf64() and this routine needs to be able
534 * to deal with being called as 32 or 64 bit.
536 * The address of the PVH entry point is saved to the 'pvh_start_addr'
537 * global variable. (although the entry point is 32-bit, the kernel
538 * binary can be either 32-bit or 64-bit).
540 static uint64_t read_pvh_start_addr(void *arg1
, void *arg2
, bool is64
)
542 size_t *elf_note_data_addr
;
544 /* Check if ELF Note header passed in is valid */
550 struct elf64_note
*nhdr64
= (struct elf64_note
*)arg1
;
551 uint64_t nhdr_size64
= sizeof(struct elf64_note
);
552 uint64_t phdr_align
= *(uint64_t *)arg2
;
553 uint64_t nhdr_namesz
= nhdr64
->n_namesz
;
556 ((void *)nhdr64
) + nhdr_size64
+
557 QEMU_ALIGN_UP(nhdr_namesz
, phdr_align
);
559 pvh_start_addr
= *elf_note_data_addr
;
561 struct elf32_note
*nhdr32
= (struct elf32_note
*)arg1
;
562 uint32_t nhdr_size32
= sizeof(struct elf32_note
);
563 uint32_t phdr_align
= *(uint32_t *)arg2
;
564 uint32_t nhdr_namesz
= nhdr32
->n_namesz
;
567 ((void *)nhdr32
) + nhdr_size32
+
568 QEMU_ALIGN_UP(nhdr_namesz
, phdr_align
);
570 pvh_start_addr
= *(uint32_t *)elf_note_data_addr
;
573 return pvh_start_addr
;
576 static bool load_elfboot(const char *kernel_filename
,
577 int kernel_file_size
,
579 size_t pvh_xen_start_addr
,
583 uint32_t mh_load_addr
= 0;
584 uint32_t elf_kernel_size
= 0;
586 uint64_t elf_low
, elf_high
;
589 if (ldl_p(header
) != 0x464c457f) {
590 return false; /* no elfboot */
593 bool elf_is64
= header
[EI_CLASS
] == ELFCLASS64
;
595 ((Elf64_Ehdr
*)header
)->e_flags
: ((Elf32_Ehdr
*)header
)->e_flags
;
597 if (flags
& 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
598 error_report("elfboot unsupported flags = %x", flags
);
602 uint64_t elf_note_type
= XEN_ELFNOTE_PHYS32_ENTRY
;
603 kernel_size
= load_elf(kernel_filename
, read_pvh_start_addr
,
604 NULL
, &elf_note_type
, &elf_entry
,
605 &elf_low
, &elf_high
, NULL
, 0, I386_ELF_MACHINE
,
608 if (kernel_size
< 0) {
609 error_report("Error while loading elf kernel");
612 mh_load_addr
= elf_low
;
613 elf_kernel_size
= elf_high
- elf_low
;
615 if (pvh_start_addr
== 0) {
616 error_report("Error loading uncompressed kernel without PVH ELF Note");
619 fw_cfg_add_i32(fw_cfg
, FW_CFG_KERNEL_ENTRY
, pvh_start_addr
);
620 fw_cfg_add_i32(fw_cfg
, FW_CFG_KERNEL_ADDR
, mh_load_addr
);
621 fw_cfg_add_i32(fw_cfg
, FW_CFG_KERNEL_SIZE
, elf_kernel_size
);
626 void x86_load_linux(X86MachineState
*x86ms
,
631 bool linuxboot_dma_enabled
= X86_MACHINE_GET_CLASS(x86ms
)->fwcfg_dma_enabled
;
633 int setup_size
, kernel_size
, cmdline_size
;
634 int dtb_size
, setup_data_offset
;
636 uint8_t header
[8192], *setup
, *kernel
;
637 hwaddr real_addr
, prot_addr
, cmdline_addr
, initrd_addr
= 0;
640 MachineState
*machine
= MACHINE(x86ms
);
641 struct setup_data
*setup_data
;
642 const char *kernel_filename
= machine
->kernel_filename
;
643 const char *initrd_filename
= machine
->initrd_filename
;
644 const char *dtb_filename
= machine
->dtb
;
645 const char *kernel_cmdline
= machine
->kernel_cmdline
;
646 SevKernelLoaderContext sev_load_ctx
= {};
648 /* Align to 16 bytes as a paranoia measure */
649 cmdline_size
= (strlen(kernel_cmdline
) + 16) & ~15;
651 /* load the kernel header */
652 f
= fopen(kernel_filename
, "rb");
654 fprintf(stderr
, "qemu: could not open kernel file '%s': %s\n",
655 kernel_filename
, strerror(errno
));
659 kernel_size
= get_file_size(f
);
661 fread(header
, 1, MIN(ARRAY_SIZE(header
), kernel_size
), f
) !=
662 MIN(ARRAY_SIZE(header
), kernel_size
)) {
663 fprintf(stderr
, "qemu: could not load kernel '%s': %s\n",
664 kernel_filename
, strerror(errno
));
668 /* kernel protocol version */
669 if (ldl_p(header
+ 0x202) == 0x53726448) {
670 protocol
= lduw_p(header
+ 0x206);
673 * This could be a multiboot kernel. If it is, let's stop treating it
674 * like a Linux kernel.
675 * Note: some multiboot images could be in the ELF format (the same of
676 * PVH), so we try multiboot first since we check the multiboot magic
677 * header before to load it.
679 if (load_multiboot(x86ms
, fw_cfg
, f
, kernel_filename
, initrd_filename
,
680 kernel_cmdline
, kernel_size
, header
)) {
684 * Check if the file is an uncompressed kernel file (ELF) and load it,
685 * saving the PVH entry point used by the x86/HVM direct boot ABI.
686 * If load_elfboot() is successful, populate the fw_cfg info.
689 load_elfboot(kernel_filename
, kernel_size
,
690 header
, pvh_start_addr
, fw_cfg
)) {
693 fw_cfg_add_i32(fw_cfg
, FW_CFG_CMDLINE_SIZE
,
694 strlen(kernel_cmdline
) + 1);
695 fw_cfg_add_string(fw_cfg
, FW_CFG_CMDLINE_DATA
, kernel_cmdline
);
697 fw_cfg_add_i32(fw_cfg
, FW_CFG_SETUP_SIZE
, sizeof(header
));
698 fw_cfg_add_bytes(fw_cfg
, FW_CFG_SETUP_DATA
,
699 header
, sizeof(header
));
702 if (initrd_filename
) {
703 GMappedFile
*mapped_file
;
708 mapped_file
= g_mapped_file_new(initrd_filename
, false, &gerr
);
710 fprintf(stderr
, "qemu: error reading initrd %s: %s\n",
711 initrd_filename
, gerr
->message
);
714 x86ms
->initrd_mapped_file
= mapped_file
;
716 initrd_data
= g_mapped_file_get_contents(mapped_file
);
717 initrd_size
= g_mapped_file_get_length(mapped_file
);
718 initrd_max
= x86ms
->below_4g_mem_size
- acpi_data_size
- 1;
719 if (initrd_size
>= initrd_max
) {
720 fprintf(stderr
, "qemu: initrd is too large, cannot support."
721 "(max: %"PRIu32
", need %"PRId64
")\n",
722 initrd_max
, (uint64_t)initrd_size
);
726 initrd_addr
= (initrd_max
- initrd_size
) & ~4095;
728 fw_cfg_add_i32(fw_cfg
, FW_CFG_INITRD_ADDR
, initrd_addr
);
729 fw_cfg_add_i32(fw_cfg
, FW_CFG_INITRD_SIZE
, initrd_size
);
730 fw_cfg_add_bytes(fw_cfg
, FW_CFG_INITRD_DATA
, initrd_data
,
734 option_rom
[nb_option_roms
].bootindex
= 0;
735 option_rom
[nb_option_roms
].name
= "pvh.bin";
743 if (protocol
< 0x200 || !(header
[0x211] & 0x01)) {
746 cmdline_addr
= 0x9a000 - cmdline_size
;
748 } else if (protocol
< 0x202) {
749 /* High but ancient kernel */
751 cmdline_addr
= 0x9a000 - cmdline_size
;
752 prot_addr
= 0x100000;
754 /* High and recent kernel */
756 cmdline_addr
= 0x20000;
757 prot_addr
= 0x100000;
760 /* highest address for loading the initrd */
761 if (protocol
>= 0x20c &&
762 lduw_p(header
+ 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G
) {
764 * Linux has supported initrd up to 4 GB for a very long time (2007,
765 * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
766 * though it only sets initrd_max to 2 GB to "work around bootloader
767 * bugs". Luckily, QEMU firmware(which does something like bootloader)
768 * has supported this.
770 * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
771 * be loaded into any address.
773 * In addition, initrd_max is uint32_t simply because QEMU doesn't
774 * support the 64-bit boot protocol (specifically the ext_ramdisk_image
777 * Therefore here just limit initrd_max to UINT32_MAX simply as well.
779 initrd_max
= UINT32_MAX
;
780 } else if (protocol
>= 0x203) {
781 initrd_max
= ldl_p(header
+ 0x22c);
783 initrd_max
= 0x37ffffff;
786 if (initrd_max
>= x86ms
->below_4g_mem_size
- acpi_data_size
) {
787 initrd_max
= x86ms
->below_4g_mem_size
- acpi_data_size
- 1;
790 fw_cfg_add_i32(fw_cfg
, FW_CFG_CMDLINE_ADDR
, cmdline_addr
);
791 fw_cfg_add_i32(fw_cfg
, FW_CFG_CMDLINE_SIZE
, strlen(kernel_cmdline
) + 1);
792 fw_cfg_add_string(fw_cfg
, FW_CFG_CMDLINE_DATA
, kernel_cmdline
);
793 sev_load_ctx
.cmdline_data
= (char *)kernel_cmdline
;
794 sev_load_ctx
.cmdline_size
= strlen(kernel_cmdline
) + 1;
796 if (protocol
>= 0x202) {
797 stl_p(header
+ 0x228, cmdline_addr
);
799 stw_p(header
+ 0x20, 0xA33F);
800 stw_p(header
+ 0x22, cmdline_addr
- real_addr
);
803 /* handle vga= parameter */
804 vmode
= strstr(kernel_cmdline
, "vga=");
806 unsigned int video_mode
;
811 if (!strncmp(vmode
, "normal", 6)) {
813 } else if (!strncmp(vmode
, "ext", 3)) {
815 } else if (!strncmp(vmode
, "ask", 3)) {
818 ret
= qemu_strtoui(vmode
, &end
, 0, &video_mode
);
819 if (ret
!= 0 || (*end
&& *end
!= ' ')) {
820 fprintf(stderr
, "qemu: invalid 'vga=' kernel parameter.\n");
824 stw_p(header
+ 0x1fa, video_mode
);
829 * High nybble = B reserved for QEMU; low nybble is revision number.
830 * If this code is substantially changed, you may want to consider
831 * incrementing the revision.
833 if (protocol
>= 0x200) {
834 header
[0x210] = 0xB0;
837 if (protocol
>= 0x201) {
838 header
[0x211] |= 0x80; /* CAN_USE_HEAP */
839 stw_p(header
+ 0x224, cmdline_addr
- real_addr
- 0x200);
843 if (initrd_filename
) {
844 GMappedFile
*mapped_file
;
849 if (protocol
< 0x200) {
850 fprintf(stderr
, "qemu: linux kernel too old to load a ram disk\n");
854 mapped_file
= g_mapped_file_new(initrd_filename
, false, &gerr
);
856 fprintf(stderr
, "qemu: error reading initrd %s: %s\n",
857 initrd_filename
, gerr
->message
);
860 x86ms
->initrd_mapped_file
= mapped_file
;
862 initrd_data
= g_mapped_file_get_contents(mapped_file
);
863 initrd_size
= g_mapped_file_get_length(mapped_file
);
864 if (initrd_size
>= initrd_max
) {
865 fprintf(stderr
, "qemu: initrd is too large, cannot support."
866 "(max: %"PRIu32
", need %"PRId64
")\n",
867 initrd_max
, (uint64_t)initrd_size
);
871 initrd_addr
= (initrd_max
- initrd_size
) & ~4095;
873 fw_cfg_add_i32(fw_cfg
, FW_CFG_INITRD_ADDR
, initrd_addr
);
874 fw_cfg_add_i32(fw_cfg
, FW_CFG_INITRD_SIZE
, initrd_size
);
875 fw_cfg_add_bytes(fw_cfg
, FW_CFG_INITRD_DATA
, initrd_data
, initrd_size
);
876 sev_load_ctx
.initrd_data
= initrd_data
;
877 sev_load_ctx
.initrd_size
= initrd_size
;
879 stl_p(header
+ 0x218, initrd_addr
);
880 stl_p(header
+ 0x21c, initrd_size
);
883 /* load kernel and setup */
884 setup_size
= header
[0x1f1];
885 if (setup_size
== 0) {
888 setup_size
= (setup_size
+ 1) * 512;
889 if (setup_size
> kernel_size
) {
890 fprintf(stderr
, "qemu: invalid kernel header\n");
893 kernel_size
-= setup_size
;
895 setup
= g_malloc(setup_size
);
896 kernel
= g_malloc(kernel_size
);
897 fseek(f
, 0, SEEK_SET
);
898 if (fread(setup
, 1, setup_size
, f
) != setup_size
) {
899 fprintf(stderr
, "fread() failed\n");
902 if (fread(kernel
, 1, kernel_size
, f
) != kernel_size
) {
903 fprintf(stderr
, "fread() failed\n");
908 /* append dtb to kernel */
910 if (protocol
< 0x209) {
911 fprintf(stderr
, "qemu: Linux kernel too old to load a dtb\n");
915 dtb_size
= get_image_size(dtb_filename
);
917 fprintf(stderr
, "qemu: error reading dtb %s: %s\n",
918 dtb_filename
, strerror(errno
));
922 setup_data_offset
= QEMU_ALIGN_UP(kernel_size
, 16);
923 kernel_size
= setup_data_offset
+ sizeof(struct setup_data
) + dtb_size
;
924 kernel
= g_realloc(kernel
, kernel_size
);
926 stq_p(header
+ 0x250, prot_addr
+ setup_data_offset
);
928 setup_data
= (struct setup_data
*)(kernel
+ setup_data_offset
);
929 setup_data
->next
= 0;
930 setup_data
->type
= cpu_to_le32(SETUP_DTB
);
931 setup_data
->len
= cpu_to_le32(dtb_size
);
933 load_image_size(dtb_filename
, setup_data
->data
, dtb_size
);
937 * If we're starting an encrypted VM, it will be OVMF based, which uses the
938 * efi stub for booting and doesn't require any values to be placed in the
939 * kernel header. We therefore don't update the header so the hash of the
940 * kernel on the other side of the fw_cfg interface matches the hash of the
941 * file the user passed in.
943 if (!sev_enabled()) {
944 memcpy(setup
, header
, MIN(sizeof(header
), setup_size
));
947 fw_cfg_add_i32(fw_cfg
, FW_CFG_KERNEL_ADDR
, prot_addr
);
948 fw_cfg_add_i32(fw_cfg
, FW_CFG_KERNEL_SIZE
, kernel_size
);
949 fw_cfg_add_bytes(fw_cfg
, FW_CFG_KERNEL_DATA
, kernel
, kernel_size
);
950 sev_load_ctx
.kernel_data
= (char *)kernel
;
951 sev_load_ctx
.kernel_size
= kernel_size
;
953 fw_cfg_add_i32(fw_cfg
, FW_CFG_SETUP_ADDR
, real_addr
);
954 fw_cfg_add_i32(fw_cfg
, FW_CFG_SETUP_SIZE
, setup_size
);
955 fw_cfg_add_bytes(fw_cfg
, FW_CFG_SETUP_DATA
, setup
, setup_size
);
956 sev_load_ctx
.setup_data
= (char *)setup
;
957 sev_load_ctx
.setup_size
= setup_size
;
960 sev_add_kernel_loader_hashes(&sev_load_ctx
, &error_fatal
);
963 option_rom
[nb_option_roms
].bootindex
= 0;
964 option_rom
[nb_option_roms
].name
= "linuxboot.bin";
965 if (linuxboot_dma_enabled
&& fw_cfg_dma_enabled(fw_cfg
)) {
966 option_rom
[nb_option_roms
].name
= "linuxboot_dma.bin";
971 void x86_isa_bios_init(MemoryRegion
*isa_bios
, MemoryRegion
*isa_memory
,
972 MemoryRegion
*bios
, bool read_only
)
974 uint64_t bios_size
= memory_region_size(bios
);
975 uint64_t isa_bios_size
= MIN(bios_size
, 128 * KiB
);
977 memory_region_init_alias(isa_bios
, NULL
, "isa-bios", bios
,
978 bios_size
- isa_bios_size
, isa_bios_size
);
979 memory_region_add_subregion_overlap(isa_memory
, 1 * MiB
- isa_bios_size
,
981 memory_region_set_readonly(isa_bios
, read_only
);
984 void x86_bios_rom_init(X86MachineState
*x86ms
, const char *default_firmware
,
985 MemoryRegion
*rom_memory
, bool isapc_ram_fw
)
987 const char *bios_name
;
993 bios_name
= MACHINE(x86ms
)->firmware
?: default_firmware
;
994 filename
= qemu_find_file(QEMU_FILE_TYPE_BIOS
, bios_name
);
996 bios_size
= get_image_size(filename
);
1000 if (bios_size
<= 0 ||
1001 (bios_size
% 65536) != 0) {
1004 if (machine_require_guest_memfd(MACHINE(x86ms
))) {
1005 memory_region_init_ram_guest_memfd(&x86ms
->bios
, NULL
, "pc.bios",
1006 bios_size
, &error_fatal
);
1008 memory_region_init_ram(&x86ms
->bios
, NULL
, "pc.bios",
1009 bios_size
, &error_fatal
);
1011 if (sev_enabled()) {
1013 * The concept of a "reset" simply doesn't exist for
1014 * confidential computing guests, we have to destroy and
1015 * re-launch them instead. So there is no need to register
1016 * the firmware as rom to properly re-initialize on reset.
1017 * Just go for a straight file load instead.
1019 void *ptr
= memory_region_get_ram_ptr(&x86ms
->bios
);
1020 load_image_size(filename
, ptr
, bios_size
);
1021 x86_firmware_configure(0x100000000ULL
- bios_size
, ptr
, bios_size
);
1023 memory_region_set_readonly(&x86ms
->bios
, !isapc_ram_fw
);
1024 ret
= rom_add_file_fixed(bios_name
, (uint32_t)(-bios_size
), -1);
1031 if (!machine_require_guest_memfd(MACHINE(x86ms
))) {
1032 /* map the last 128KB of the BIOS in ISA space */
1033 x86_isa_bios_init(&x86ms
->isa_bios
, rom_memory
, &x86ms
->bios
,
1037 /* map all the bios at the top of memory */
1038 memory_region_add_subregion(rom_memory
,
1039 (uint32_t)(-bios_size
),
1044 fprintf(stderr
, "qemu: could not load PC BIOS '%s'\n", bios_name
);