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