xref: /freebsd/sys/x86/xen/pv.c (revision 06c3fb2749bda94cb5201f81ffdb8fa6c3161b2e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2004 Christian Limpach.
5  * Copyright (c) 2004-2006,2008 Kip Macy
6  * Copyright (c) 2008 The NetBSD Foundation, Inc.
7  * Copyright (c) 2013 Roger Pau Monné <roger.pau@citrix.com>
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 #include "opt_ddb.h"
34 #include "opt_kstack_pages.h"
35 
36 #include <sys/param.h>
37 #include <sys/bus.h>
38 #include <sys/kernel.h>
39 #include <sys/reboot.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/linker.h>
43 #include <sys/lock.h>
44 #include <sys/rwlock.h>
45 #include <sys/boot.h>
46 #include <sys/ctype.h>
47 #include <sys/mutex.h>
48 #include <sys/smp.h>
49 #include <sys/efi.h>
50 #include <sys/tslog.h>
51 
52 #include <vm/vm.h>
53 #include <vm/vm_extern.h>
54 #include <vm/vm_kern.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_pager.h>
59 #include <vm/vm_param.h>
60 
61 #include <machine/_inttypes.h>
62 #include <machine/intr_machdep.h>
63 #include <x86/apicvar.h>
64 #include <x86/init.h>
65 #include <machine/pc/bios.h>
66 #include <machine/smp.h>
67 #include <machine/intr_machdep.h>
68 #include <machine/md_var.h>
69 #include <machine/metadata.h>
70 #include <machine/cpu.h>
71 
72 #include <xen/xen-os.h>
73 #include <xen/hvm.h>
74 #include <xen/hypervisor.h>
75 #include <xen/xenstore/xenstorevar.h>
76 #include <xen/xen_pv.h>
77 
78 #include <contrib/xen/arch-x86/cpuid.h>
79 #include <contrib/xen/arch-x86/hvm/start_info.h>
80 #include <contrib/xen/vcpu.h>
81 
82 #include <dev/xen/timer/timer.h>
83 
84 #ifdef DDB
85 #include <ddb/ddb.h>
86 #endif
87 
88 /* Native initial function */
89 extern u_int64_t hammer_time(u_int64_t, u_int64_t);
90 /* Xen initial function */
91 uint64_t hammer_time_xen(vm_paddr_t);
92 
93 #define MAX_E820_ENTRIES	128
94 
95 /*--------------------------- Forward Declarations ---------------------------*/
96 static caddr_t xen_pvh_parse_preload_data(uint64_t);
97 static void pvh_parse_memmap(caddr_t, vm_paddr_t *, int *);
98 
99 /*---------------------------- Extern Declarations ---------------------------*/
100 /*
101  * Placed by the linker at the end of the bss section, which is the last
102  * section loaded by Xen before loading the symtab and strtab.
103  */
104 extern uint32_t end;
105 
106 /*-------------------------------- Global Data -------------------------------*/
107 struct init_ops xen_pvh_init_ops = {
108 	.parse_preload_data		= xen_pvh_parse_preload_data,
109 	.early_clock_source_init	= xen_clock_init,
110 	.early_delay			= xen_delay,
111 	.parse_memmap			= pvh_parse_memmap,
112 };
113 
114 static struct bios_smap xen_smap[MAX_E820_ENTRIES];
115 
116 static struct hvm_start_info *start_info;
117 
118 /*-------------------------------- Xen PV init -------------------------------*/
119 
120 static int
121 isxen(void)
122 {
123 	static int xen = -1;
124 	uint32_t base;
125 	u_int regs[4];
126 
127 	if (xen != -1)
128 		return (xen);
129 
130 	/*
131 	 * The full code for identifying which hypervisor we're running under
132 	 * is in sys/x86/x86/identcpu.c and runs later in the boot process;
133 	 * this is sufficient to distinguish Xen PVH booting from non-Xen PVH
134 	 * and skip some very early Xen-specific code in the non-Xen case.
135 	 */
136 	xen = 0;
137 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
138 		do_cpuid(base, regs);
139 		if (regs[1] == XEN_CPUID_SIGNATURE_EBX &&
140 		    regs[2] == XEN_CPUID_SIGNATURE_ECX &&
141 		    regs[3] == XEN_CPUID_SIGNATURE_EDX) {
142 			xen = 1;
143 			break;
144 		}
145 	}
146 	return (xen);
147 }
148 
149 #define CRASH(...) do {					\
150 	if (isxen()) {					\
151 		xc_printf(__VA_ARGS__);			\
152 		HYPERVISOR_shutdown(SHUTDOWN_crash);	\
153 	} else {					\
154 		halt();					\
155 	}						\
156 } while (0)
157 
158 uint64_t
159 hammer_time_xen(vm_paddr_t start_info_paddr)
160 {
161 	struct hvm_modlist_entry *mod;
162 	struct xen_add_to_physmap xatp;
163 	uint64_t physfree;
164 	char *kenv;
165 	int rc;
166 
167 	if (isxen()) {
168 		vm_guest = VM_GUEST_XEN;
169 		rc = xen_hvm_init_hypercall_stubs(XEN_HVM_INIT_EARLY);
170 		if (rc) {
171 			xc_printf("ERROR: failed to initialize hypercall page: %d\n",
172 			    rc);
173 			HYPERVISOR_shutdown(SHUTDOWN_crash);
174 		}
175 	}
176 
177 	start_info = (struct hvm_start_info *)(start_info_paddr + KERNBASE);
178 	if (start_info->magic != XEN_HVM_START_MAGIC_VALUE) {
179 		CRASH("Unknown magic value in start_info struct: %#x\n",
180 		    start_info->magic);
181 	}
182 
183 	/*
184 	 * Select the higher address to use as physfree: either after
185 	 * start_info, after the kernel, after the memory map or after any of
186 	 * the modules.  We assume enough memory to be available after the
187 	 * selected address for the needs of very early memory allocations.
188 	 */
189 	physfree = roundup2(start_info_paddr + sizeof(struct hvm_start_info),
190 	    PAGE_SIZE);
191 	physfree = MAX(roundup2((vm_paddr_t)_end - KERNBASE, PAGE_SIZE),
192 	    physfree);
193 
194 	if (start_info->memmap_paddr != 0)
195 		physfree = MAX(roundup2(start_info->memmap_paddr +
196 		    start_info->memmap_entries *
197 		    sizeof(struct hvm_memmap_table_entry), PAGE_SIZE),
198 		    physfree);
199 
200 	if (start_info->modlist_paddr != 0) {
201 		unsigned int i;
202 
203 		if (start_info->nr_modules == 0) {
204 			CRASH(
205 			    "ERROR: modlist_paddr != 0 but nr_modules == 0\n");
206 		}
207 		mod = (struct hvm_modlist_entry *)
208 		    (start_info->modlist_paddr + KERNBASE);
209 		for (i = 0; i < start_info->nr_modules; i++)
210 			physfree = MAX(roundup2(mod[i].paddr + mod[i].size,
211 			    PAGE_SIZE), physfree);
212 	}
213 
214 	if (isxen()) {
215 		xatp.domid = DOMID_SELF;
216 		xatp.idx = 0;
217 		xatp.space = XENMAPSPACE_shared_info;
218 		xatp.gpfn = atop(physfree);
219 		if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) {
220 			xc_printf("ERROR: failed to setup shared_info page\n");
221 			HYPERVISOR_shutdown(SHUTDOWN_crash);
222 		}
223 		HYPERVISOR_shared_info = (shared_info_t *)(physfree + KERNBASE);
224 		physfree += PAGE_SIZE;
225 	}
226 
227 	/*
228 	 * Init a static kenv using a free page. The contents will be filled
229 	 * from the parse_preload_data hook.
230 	 */
231 	kenv = (void *)(physfree + KERNBASE);
232 	physfree += PAGE_SIZE;
233 	bzero_early(kenv, PAGE_SIZE);
234 	init_static_kenv(kenv, PAGE_SIZE);
235 
236 	/* Set the hooks for early functions that diverge from bare metal */
237 	init_ops = xen_pvh_init_ops;
238 	hvm_start_flags = start_info->flags;
239 
240 	/* Now we can jump into the native init function */
241 	return (hammer_time(0, physfree));
242 }
243 
244 /*-------------------------------- PV specific -------------------------------*/
245 
246 /*
247  * When booted as a PVH guest FreeBSD needs to avoid using the RSDP address
248  * hint provided by the loader because it points to the native set of ACPI
249  * tables instead of the ones crafted by Xen. The acpi.rsdp env variable is
250  * removed from kenv if present, and a new acpi.rsdp is added to kenv that
251  * points to the address of the Xen crafted RSDP.
252  */
253 static bool reject_option(const char *option)
254 {
255 	static const char *reject[] = {
256 		"acpi.rsdp",
257 	};
258 	unsigned int i;
259 
260 	for (i = 0; i < nitems(reject); i++)
261 		if (strncmp(option, reject[i], strlen(reject[i])) == 0)
262 			return (true);
263 
264 	return (false);
265 }
266 
267 static void
268 xen_pvh_set_env(char *env, bool (*filter)(const char *))
269 {
270 	char *option;
271 
272 	if (env == NULL)
273 		return;
274 
275 	option = env;
276 	while (*option != 0) {
277 		char *value;
278 
279 		if (filter != NULL && filter(option)) {
280 			option += strlen(option) + 1;
281 			continue;
282 		}
283 
284 		value = option;
285 		option = strsep(&value, "=");
286 		if (kern_setenv(option, value) != 0 && isxen())
287 			xc_printf("unable to add kenv %s=%s\n", option, value);
288 		option = value + strlen(value) + 1;
289 	}
290 }
291 
292 #ifdef DDB
293 /*
294  * The way Xen loads the symtab is different from the native boot loader,
295  * because it's tailored for NetBSD. So we have to adapt and use the same
296  * method as NetBSD. Portions of the code below have been picked from NetBSD:
297  * sys/kern/kern_ksyms.c CVS Revision 1.71.
298  */
299 static void
300 xen_pvh_parse_symtab(void)
301 {
302 	Elf_Ehdr *ehdr;
303 	Elf_Shdr *shdr;
304 	int i, j;
305 
306 	ehdr = (Elf_Ehdr *)(&end + 1);
307 	if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) ||
308 	    ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
309 	    ehdr->e_version > 1) {
310 		if (isxen())
311 			xc_printf("Unable to load ELF symtab: invalid symbol table\n");
312 		return;
313 	}
314 
315 	shdr = (Elf_Shdr *)((uint8_t *)ehdr + ehdr->e_shoff);
316 	/* Find the symbol table and the corresponding string table. */
317 	for (i = 1; i < ehdr->e_shnum; i++) {
318 		if (shdr[i].sh_type != SHT_SYMTAB)
319 			continue;
320 		if (shdr[i].sh_offset == 0)
321 			continue;
322 		ksymtab = (uintptr_t)((uint8_t *)ehdr + shdr[i].sh_offset);
323 		ksymtab_size = shdr[i].sh_size;
324 		j = shdr[i].sh_link;
325 		if (shdr[j].sh_offset == 0)
326 			continue; /* Can this happen? */
327 		kstrtab = (uintptr_t)((uint8_t *)ehdr + shdr[j].sh_offset);
328 		break;
329 	}
330 
331 	if ((ksymtab == 0 || kstrtab == 0) && isxen())
332 		xc_printf(
333     "Unable to load ELF symtab: could not find symtab or strtab\n");
334 }
335 #endif
336 
337 static void
338 fixup_console(caddr_t kmdp)
339 {
340 	struct xen_platform_op op = {
341 		.cmd = XENPF_get_dom0_console,
342 	};
343 	xenpf_dom0_console_t *console = &op.u.dom0_console;
344 	union {
345 		struct efi_fb efi;
346 		struct vbe_fb vbe;
347 	} *fb = NULL;
348 	int size;
349 
350 	size = HYPERVISOR_platform_op(&op);
351 	if (size < 0) {
352 		xc_printf("Failed to get dom0 video console info: %d\n", size);
353 		return;
354 	}
355 
356 	switch (console->video_type) {
357 	case XEN_VGATYPE_VESA_LFB:
358 		fb = (__typeof__ (fb))preload_search_info(kmdp,
359 		    MODINFO_METADATA | MODINFOMD_VBE_FB);
360 
361 		if (fb == NULL) {
362 			xc_printf("No VBE FB in kernel metadata\n");
363 			return;
364 		}
365 
366 		_Static_assert(offsetof(struct vbe_fb, fb_bpp) ==
367 		    offsetof(struct efi_fb, fb_mask_reserved) +
368 		    sizeof(fb->efi.fb_mask_reserved),
369 		    "Bad structure overlay\n");
370 		fb->vbe.fb_bpp = console->u.vesa_lfb.bits_per_pixel;
371 		/* FALLTHROUGH */
372 	case XEN_VGATYPE_EFI_LFB:
373 		if (fb == NULL) {
374 			fb = (__typeof__ (fb))preload_search_info(kmdp,
375 			    MODINFO_METADATA | MODINFOMD_EFI_FB);
376 			if (fb == NULL) {
377 				xc_printf("No EFI FB in kernel metadata\n");
378 				return;
379 			}
380 		}
381 
382 		fb->efi.fb_addr = console->u.vesa_lfb.lfb_base;
383 		if (size >
384 		    offsetof(xenpf_dom0_console_t, u.vesa_lfb.ext_lfb_base))
385 			fb->efi.fb_addr |=
386 			    (uint64_t)console->u.vesa_lfb.ext_lfb_base << 32;
387 		fb->efi.fb_size = console->u.vesa_lfb.lfb_size << 16;
388 		fb->efi.fb_height = console->u.vesa_lfb.height;
389 		fb->efi.fb_width = console->u.vesa_lfb.width;
390 		fb->efi.fb_stride = (console->u.vesa_lfb.bytes_per_line << 3) /
391 		    console->u.vesa_lfb.bits_per_pixel;
392 #define FBMASK(c) \
393     ((~0u << console->u.vesa_lfb.c ## _pos) & \
394     (~0u >> (32 - console->u.vesa_lfb.c ## _pos - \
395     console->u.vesa_lfb.c ## _size)))
396 		fb->efi.fb_mask_red = FBMASK(red);
397 		fb->efi.fb_mask_green = FBMASK(green);
398 		fb->efi.fb_mask_blue = FBMASK(blue);
399 		fb->efi.fb_mask_reserved = FBMASK(rsvd);
400 #undef FBMASK
401 		break;
402 
403 	default:
404 		xc_printf("Video console type unsupported\n");
405 		return;
406 	}
407 }
408 
409 static caddr_t
410 xen_pvh_parse_preload_data(uint64_t modulep)
411 {
412 	caddr_t kmdp;
413 	vm_ooffset_t off;
414 	vm_paddr_t metadata;
415 	char *envp;
416 	char acpi_rsdp[19];
417 
418 	TSENTER();
419 	if (start_info->modlist_paddr != 0) {
420 		struct hvm_modlist_entry *mod;
421 		const char *cmdline;
422 
423 		mod = (struct hvm_modlist_entry *)
424 		    (start_info->modlist_paddr + KERNBASE);
425 		cmdline = mod[0].cmdline_paddr ?
426 		    (const char *)(mod[0].cmdline_paddr + KERNBASE) : NULL;
427 
428 		if (strcmp(cmdline, "header") == 0) {
429 			struct xen_header *header;
430 
431 			header = (struct xen_header *)(mod[0].paddr + KERNBASE);
432 
433 			if ((header->flags & XENHEADER_HAS_MODULEP_OFFSET) !=
434 			    XENHEADER_HAS_MODULEP_OFFSET) {
435 				xc_printf("Unable to load module metadata\n");
436 				HYPERVISOR_shutdown(SHUTDOWN_crash);
437 			}
438 
439 			preload_metadata = (caddr_t)(mod[0].paddr +
440 			    header->modulep_offset + KERNBASE);
441 
442 			kmdp = preload_search_by_type("elf kernel");
443 			if (kmdp == NULL)
444 				kmdp = preload_search_by_type("elf64 kernel");
445 			if (kmdp == NULL) {
446 				xc_printf("Unable to find kernel\n");
447 				HYPERVISOR_shutdown(SHUTDOWN_crash);
448 			}
449 
450 			/*
451 			 * Xen has relocated the metadata and the modules, so
452 			 * we need to recalculate it's position. This is done
453 			 * by saving the original modulep address and then
454 			 * calculating the offset from the real modulep
455 			 * position.
456 			 */
457 			metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP,
458 			    vm_paddr_t);
459 			off = mod[0].paddr + header->modulep_offset - metadata +
460 			    KERNBASE;
461 		} else {
462 			preload_metadata = (caddr_t)(mod[0].paddr + KERNBASE);
463 
464 			kmdp = preload_search_by_type("elf kernel");
465 			if (kmdp == NULL)
466 				kmdp = preload_search_by_type("elf64 kernel");
467 			if (kmdp == NULL) {
468 				xc_printf("Unable to find kernel\n");
469 				HYPERVISOR_shutdown(SHUTDOWN_crash);
470 			}
471 
472 			metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t);
473 			off = mod[0].paddr + KERNBASE - metadata;
474 		}
475 
476 		preload_bootstrap_relocate(off);
477 
478 		boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
479 		envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *);
480 		if (envp != NULL)
481 			envp += off;
482 		xen_pvh_set_env(envp, reject_option);
483 
484 		if (MD_FETCH(kmdp, MODINFOMD_EFI_MAP, void *) != NULL)
485 		    strlcpy(bootmethod, "UEFI", sizeof(bootmethod));
486 		else
487 		    strlcpy(bootmethod, "BIOS", sizeof(bootmethod));
488 
489 		fixup_console(kmdp);
490 	} else {
491 		/* Parse the extra boot information given by Xen */
492 		if (start_info->cmdline_paddr != 0)
493 			boot_parse_cmdline_delim(
494 			    (char *)(start_info->cmdline_paddr + KERNBASE),
495 			    ", \t\n");
496 		kmdp = NULL;
497 		strlcpy(bootmethod, "PVH", sizeof(bootmethod));
498 	}
499 
500 	boothowto |= boot_env_to_howto();
501 
502 	snprintf(acpi_rsdp, sizeof(acpi_rsdp), "%#" PRIx64,
503 	    start_info->rsdp_paddr);
504 	kern_setenv("acpi.rsdp", acpi_rsdp);
505 
506 #ifdef DDB
507 	xen_pvh_parse_symtab();
508 #endif
509 	TSEXIT();
510 	return (kmdp);
511 }
512 
513 static void
514 pvh_parse_memmap_start_info(caddr_t kmdp, vm_paddr_t *physmap,
515     int *physmap_idx)
516 {
517 	const struct hvm_memmap_table_entry * entries;
518 	size_t nentries;
519 	size_t i;
520 
521 	/* Extract from HVM start_info. */
522 	entries = (struct hvm_memmap_table_entry *)(start_info->memmap_paddr + KERNBASE);
523 	nentries = start_info->memmap_entries;
524 
525 	/* Convert into E820 format and handle one by one. */
526 	for (i = 0; i < nentries; i++) {
527 		struct bios_smap entry;
528 
529 		entry.base = entries[i].addr;
530 		entry.length = entries[i].size;
531 
532 		/*
533 		 * Luckily for us, the XEN_HVM_MEMMAP_TYPE_* values exactly
534 		 * match the SMAP_TYPE_* values so we don't need to translate
535 		 * anything here.
536 		 */
537 		entry.type = entries[i].type;
538 
539 		bios_add_smap_entries(&entry, 1, physmap, physmap_idx);
540 	}
541 }
542 
543 static void
544 xen_pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx)
545 {
546 	struct xen_memory_map memmap;
547 	u_int32_t size;
548 	int rc;
549 
550 	/* We should only reach here if we're running under Xen. */
551 	KASSERT(isxen(), ("xen_pvh_parse_memmap reached when !Xen"));
552 
553 	/* Fetch the E820 map from Xen */
554 	memmap.nr_entries = MAX_E820_ENTRIES;
555 	set_xen_guest_handle(memmap.buffer, xen_smap);
556 	rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
557 	if (rc) {
558 		xc_printf("ERROR: unable to fetch Xen E820 memory map: %d\n",
559 		    rc);
560 		HYPERVISOR_shutdown(SHUTDOWN_crash);
561 	}
562 
563 	size = memmap.nr_entries * sizeof(xen_smap[0]);
564 
565 	bios_add_smap_entries(xen_smap, size, physmap, physmap_idx);
566 }
567 
568 static void
569 pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx)
570 {
571 
572 	/*
573 	 * If version >= 1 and memmap_paddr != 0, use the memory map provided
574 	 * in the start_info structure; if not, we're running under legacy
575 	 * Xen and need to use the Xen hypercall.
576 	 */
577 	if ((start_info->version >= 1) && (start_info->memmap_paddr != 0))
578 		pvh_parse_memmap_start_info(kmdp, physmap, physmap_idx);
579 	else
580 		xen_pvh_parse_memmap(kmdp, physmap, physmap_idx);
581 }
582