xref: /freebsd/sys/x86/xen/pv.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
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 		xen_domain_type = XEN_HVM_DOMAIN;
169 		vm_guest = VM_GUEST_XEN;
170 		rc = xen_hvm_init_hypercall_stubs(XEN_HVM_INIT_EARLY);
171 		if (rc) {
172 			xc_printf("ERROR: failed to initialize hypercall page: %d\n",
173 			    rc);
174 			HYPERVISOR_shutdown(SHUTDOWN_crash);
175 		}
176 	}
177 
178 	start_info = (struct hvm_start_info *)(start_info_paddr + KERNBASE);
179 	if (start_info->magic != XEN_HVM_START_MAGIC_VALUE) {
180 		CRASH("Unknown magic value in start_info struct: %#x\n",
181 		    start_info->magic);
182 	}
183 
184 	/*
185 	 * Select the higher address to use as physfree: either after
186 	 * start_info, after the kernel, after the memory map or after any of
187 	 * the modules.  We assume enough memory to be available after the
188 	 * selected address for the needs of very early memory allocations.
189 	 */
190 	physfree = roundup2(start_info_paddr + sizeof(struct hvm_start_info),
191 	    PAGE_SIZE);
192 	physfree = MAX(roundup2((vm_paddr_t)_end - KERNBASE, PAGE_SIZE),
193 	    physfree);
194 
195 	if (start_info->memmap_paddr != 0)
196 		physfree = MAX(roundup2(start_info->memmap_paddr +
197 		    start_info->memmap_entries *
198 		    sizeof(struct hvm_memmap_table_entry), PAGE_SIZE),
199 		    physfree);
200 
201 	if (start_info->modlist_paddr != 0) {
202 		unsigned int i;
203 
204 		if (start_info->nr_modules == 0) {
205 			CRASH(
206 			    "ERROR: modlist_paddr != 0 but nr_modules == 0\n");
207 		}
208 		mod = (struct hvm_modlist_entry *)
209 		    (start_info->modlist_paddr + KERNBASE);
210 		for (i = 0; i < start_info->nr_modules; i++)
211 			physfree = MAX(roundup2(mod[i].paddr + mod[i].size,
212 			    PAGE_SIZE), physfree);
213 	}
214 
215 	if (isxen()) {
216 		xatp.domid = DOMID_SELF;
217 		xatp.idx = 0;
218 		xatp.space = XENMAPSPACE_shared_info;
219 		xatp.gpfn = atop(physfree);
220 		if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) {
221 			xc_printf("ERROR: failed to setup shared_info page\n");
222 			HYPERVISOR_shutdown(SHUTDOWN_crash);
223 		}
224 		HYPERVISOR_shared_info = (shared_info_t *)(physfree + KERNBASE);
225 		physfree += PAGE_SIZE;
226 	}
227 
228 	/*
229 	 * Init a static kenv using a free page. The contents will be filled
230 	 * from the parse_preload_data hook.
231 	 */
232 	kenv = (void *)(physfree + KERNBASE);
233 	physfree += PAGE_SIZE;
234 	bzero_early(kenv, PAGE_SIZE);
235 	init_static_kenv(kenv, PAGE_SIZE);
236 
237 	/* Set the hooks for early functions that diverge from bare metal */
238 	init_ops = xen_pvh_init_ops;
239 	hvm_start_flags = start_info->flags;
240 
241 	/* Now we can jump into the native init function */
242 	return (hammer_time(0, physfree));
243 }
244 
245 /*-------------------------------- PV specific -------------------------------*/
246 
247 /*
248  * When booted as a PVH guest FreeBSD needs to avoid using the RSDP address
249  * hint provided by the loader because it points to the native set of ACPI
250  * tables instead of the ones crafted by Xen. The acpi.rsdp env variable is
251  * removed from kenv if present, and a new acpi.rsdp is added to kenv that
252  * points to the address of the Xen crafted RSDP.
253  */
254 static bool reject_option(const char *option)
255 {
256 	static const char *reject[] = {
257 		"acpi.rsdp",
258 	};
259 	unsigned int i;
260 
261 	for (i = 0; i < nitems(reject); i++)
262 		if (strncmp(option, reject[i], strlen(reject[i])) == 0)
263 			return (true);
264 
265 	return (false);
266 }
267 
268 static void
269 xen_pvh_set_env(char *env, bool (*filter)(const char *))
270 {
271 	char *option;
272 
273 	if (env == NULL)
274 		return;
275 
276 	option = env;
277 	while (*option != 0) {
278 		char *value;
279 
280 		if (filter != NULL && filter(option)) {
281 			option += strlen(option) + 1;
282 			continue;
283 		}
284 
285 		value = option;
286 		option = strsep(&value, "=");
287 		if (kern_setenv(option, value) != 0 && isxen())
288 			xc_printf("unable to add kenv %s=%s\n", option, value);
289 		option = value + strlen(value) + 1;
290 	}
291 }
292 
293 #ifdef DDB
294 /*
295  * The way Xen loads the symtab is different from the native boot loader,
296  * because it's tailored for NetBSD. So we have to adapt and use the same
297  * method as NetBSD. Portions of the code below have been picked from NetBSD:
298  * sys/kern/kern_ksyms.c CVS Revision 1.71.
299  */
300 static void
301 xen_pvh_parse_symtab(void)
302 {
303 	Elf_Ehdr *ehdr;
304 	Elf_Shdr *shdr;
305 	int i, j;
306 
307 	ehdr = (Elf_Ehdr *)(&end + 1);
308 	if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) ||
309 	    ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
310 	    ehdr->e_version > 1) {
311 		if (isxen())
312 			xc_printf("Unable to load ELF symtab: invalid symbol table\n");
313 		return;
314 	}
315 
316 	shdr = (Elf_Shdr *)((uint8_t *)ehdr + ehdr->e_shoff);
317 	/* Find the symbol table and the corresponding string table. */
318 	for (i = 1; i < ehdr->e_shnum; i++) {
319 		if (shdr[i].sh_type != SHT_SYMTAB)
320 			continue;
321 		if (shdr[i].sh_offset == 0)
322 			continue;
323 		ksymtab = (uintptr_t)((uint8_t *)ehdr + shdr[i].sh_offset);
324 		ksymtab_size = shdr[i].sh_size;
325 		j = shdr[i].sh_link;
326 		if (shdr[j].sh_offset == 0)
327 			continue; /* Can this happen? */
328 		kstrtab = (uintptr_t)((uint8_t *)ehdr + shdr[j].sh_offset);
329 		break;
330 	}
331 
332 	if ((ksymtab == 0 || kstrtab == 0) && isxen())
333 		xc_printf(
334     "Unable to load ELF symtab: could not find symtab or strtab\n");
335 }
336 #endif
337 
338 static void
339 fixup_console(caddr_t kmdp)
340 {
341 	struct xen_platform_op op = {
342 		.cmd = XENPF_get_dom0_console,
343 	};
344 	xenpf_dom0_console_t *console = &op.u.dom0_console;
345 	union {
346 		struct efi_fb efi;
347 		struct vbe_fb vbe;
348 	} *fb = NULL;
349 	int size;
350 
351 	size = HYPERVISOR_platform_op(&op);
352 	if (size < 0) {
353 		xc_printf("Failed to get dom0 video console info: %d\n", size);
354 		return;
355 	}
356 
357 	switch (console->video_type) {
358 	case XEN_VGATYPE_VESA_LFB:
359 		fb = (__typeof__ (fb))preload_search_info(kmdp,
360 		    MODINFO_METADATA | MODINFOMD_VBE_FB);
361 
362 		if (fb == NULL) {
363 			xc_printf("No VBE FB in kernel metadata\n");
364 			return;
365 		}
366 
367 		_Static_assert(offsetof(struct vbe_fb, fb_bpp) ==
368 		    offsetof(struct efi_fb, fb_mask_reserved) +
369 		    sizeof(fb->efi.fb_mask_reserved),
370 		    "Bad structure overlay\n");
371 		fb->vbe.fb_bpp = console->u.vesa_lfb.bits_per_pixel;
372 		/* FALLTHROUGH */
373 	case XEN_VGATYPE_EFI_LFB:
374 		if (fb == NULL) {
375 			fb = (__typeof__ (fb))preload_search_info(kmdp,
376 			    MODINFO_METADATA | MODINFOMD_EFI_FB);
377 			if (fb == NULL) {
378 				xc_printf("No EFI FB in kernel metadata\n");
379 				return;
380 			}
381 		}
382 
383 		fb->efi.fb_addr = console->u.vesa_lfb.lfb_base;
384 		if (size >
385 		    offsetof(xenpf_dom0_console_t, u.vesa_lfb.ext_lfb_base))
386 			fb->efi.fb_addr |=
387 			    (uint64_t)console->u.vesa_lfb.ext_lfb_base << 32;
388 		fb->efi.fb_size = console->u.vesa_lfb.lfb_size << 16;
389 		fb->efi.fb_height = console->u.vesa_lfb.height;
390 		fb->efi.fb_width = console->u.vesa_lfb.width;
391 		fb->efi.fb_stride = (console->u.vesa_lfb.bytes_per_line << 3) /
392 		    console->u.vesa_lfb.bits_per_pixel;
393 #define FBMASK(c) \
394     ((~0u << console->u.vesa_lfb.c ## _pos) & \
395     (~0u >> (32 - console->u.vesa_lfb.c ## _pos - \
396     console->u.vesa_lfb.c ## _size)))
397 		fb->efi.fb_mask_red = FBMASK(red);
398 		fb->efi.fb_mask_green = FBMASK(green);
399 		fb->efi.fb_mask_blue = FBMASK(blue);
400 		fb->efi.fb_mask_reserved = FBMASK(rsvd);
401 #undef FBMASK
402 		break;
403 
404 	default:
405 		xc_printf("Video console type unsupported\n");
406 		return;
407 	}
408 }
409 
410 static caddr_t
411 xen_pvh_parse_preload_data(uint64_t modulep)
412 {
413 	caddr_t kmdp;
414 	vm_ooffset_t off;
415 	vm_paddr_t metadata;
416 	char *envp;
417 	char acpi_rsdp[19];
418 
419 	TSENTER();
420 	if (start_info->modlist_paddr != 0) {
421 		struct hvm_modlist_entry *mod;
422 		const char *cmdline;
423 
424 		mod = (struct hvm_modlist_entry *)
425 		    (start_info->modlist_paddr + KERNBASE);
426 		cmdline = mod[0].cmdline_paddr ?
427 		    (const char *)(mod[0].cmdline_paddr + KERNBASE) : NULL;
428 
429 		if (strcmp(cmdline, "header") == 0) {
430 			struct xen_header *header;
431 
432 			header = (struct xen_header *)(mod[0].paddr + KERNBASE);
433 
434 			if ((header->flags & XENHEADER_HAS_MODULEP_OFFSET) !=
435 			    XENHEADER_HAS_MODULEP_OFFSET) {
436 				xc_printf("Unable to load module metadata\n");
437 				HYPERVISOR_shutdown(SHUTDOWN_crash);
438 			}
439 
440 			preload_metadata = (caddr_t)(mod[0].paddr +
441 			    header->modulep_offset + KERNBASE);
442 
443 			kmdp = preload_search_by_type("elf kernel");
444 			if (kmdp == NULL)
445 				kmdp = preload_search_by_type("elf64 kernel");
446 			if (kmdp == NULL) {
447 				xc_printf("Unable to find kernel\n");
448 				HYPERVISOR_shutdown(SHUTDOWN_crash);
449 			}
450 
451 			/*
452 			 * Xen has relocated the metadata and the modules, so
453 			 * we need to recalculate it's position. This is done
454 			 * by saving the original modulep address and then
455 			 * calculating the offset from the real modulep
456 			 * position.
457 			 */
458 			metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP,
459 			    vm_paddr_t);
460 			off = mod[0].paddr + header->modulep_offset - metadata +
461 			    KERNBASE;
462 		} else {
463 			preload_metadata = (caddr_t)(mod[0].paddr + KERNBASE);
464 
465 			kmdp = preload_search_by_type("elf kernel");
466 			if (kmdp == NULL)
467 				kmdp = preload_search_by_type("elf64 kernel");
468 			if (kmdp == NULL) {
469 				xc_printf("Unable to find kernel\n");
470 				HYPERVISOR_shutdown(SHUTDOWN_crash);
471 			}
472 
473 			metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t);
474 			off = mod[0].paddr + KERNBASE - metadata;
475 		}
476 
477 		preload_bootstrap_relocate(off);
478 
479 		boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
480 		envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *);
481 		if (envp != NULL)
482 			envp += off;
483 		xen_pvh_set_env(envp, reject_option);
484 
485 		if (MD_FETCH(kmdp, MODINFOMD_EFI_MAP, void *) != NULL)
486 		    strlcpy(bootmethod, "UEFI", sizeof(bootmethod));
487 		else
488 		    strlcpy(bootmethod, "BIOS", sizeof(bootmethod));
489 
490 		fixup_console(kmdp);
491 	} else {
492 		/* Parse the extra boot information given by Xen */
493 		if (start_info->cmdline_paddr != 0)
494 			boot_parse_cmdline_delim(
495 			    (char *)(start_info->cmdline_paddr + KERNBASE),
496 			    ", \t\n");
497 		kmdp = NULL;
498 		strlcpy(bootmethod, "PVH", sizeof(bootmethod));
499 	}
500 
501 	boothowto |= boot_env_to_howto();
502 
503 	snprintf(acpi_rsdp, sizeof(acpi_rsdp), "%#" PRIx64,
504 	    start_info->rsdp_paddr);
505 	kern_setenv("acpi.rsdp", acpi_rsdp);
506 
507 #ifdef DDB
508 	xen_pvh_parse_symtab();
509 #endif
510 	TSEXIT();
511 	return (kmdp);
512 }
513 
514 static void
515 pvh_parse_memmap_start_info(caddr_t kmdp, vm_paddr_t *physmap,
516     int *physmap_idx)
517 {
518 	const struct hvm_memmap_table_entry * entries;
519 	size_t nentries;
520 	size_t i;
521 
522 	/* Extract from HVM start_info. */
523 	entries = (struct hvm_memmap_table_entry *)(start_info->memmap_paddr + KERNBASE);
524 	nentries = start_info->memmap_entries;
525 
526 	/* Convert into E820 format and handle one by one. */
527 	for (i = 0; i < nentries; i++) {
528 		struct bios_smap entry;
529 
530 		entry.base = entries[i].addr;
531 		entry.length = entries[i].size;
532 
533 		/*
534 		 * Luckily for us, the XEN_HVM_MEMMAP_TYPE_* values exactly
535 		 * match the SMAP_TYPE_* values so we don't need to translate
536 		 * anything here.
537 		 */
538 		entry.type = entries[i].type;
539 
540 		bios_add_smap_entries(&entry, 1, physmap, physmap_idx);
541 	}
542 }
543 
544 static void
545 xen_pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx)
546 {
547 	struct xen_memory_map memmap;
548 	u_int32_t size;
549 	int rc;
550 
551 	/* We should only reach here if we're running under Xen. */
552 	KASSERT(isxen(), ("xen_pvh_parse_memmap reached when !Xen"));
553 
554 	/* Fetch the E820 map from Xen */
555 	memmap.nr_entries = MAX_E820_ENTRIES;
556 	set_xen_guest_handle(memmap.buffer, xen_smap);
557 	rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
558 	if (rc) {
559 		xc_printf("ERROR: unable to fetch Xen E820 memory map: %d\n",
560 		    rc);
561 		HYPERVISOR_shutdown(SHUTDOWN_crash);
562 	}
563 
564 	size = memmap.nr_entries * sizeof(xen_smap[0]);
565 
566 	bios_add_smap_entries(xen_smap, size, physmap, physmap_idx);
567 }
568 
569 static void
570 pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx)
571 {
572 
573 	/*
574 	 * If version >= 1 and memmap_paddr != 0, use the memory map provided
575 	 * in the start_info structure; if not, we're running under legacy
576 	 * Xen and need to use the Xen hypercall.
577 	 */
578 	if ((start_info->version >= 1) && (start_info->memmap_paddr != 0))
579 		pvh_parse_memmap_start_info(kmdp, physmap, physmap_idx);
580 	else
581 		xen_pvh_parse_memmap(kmdp, physmap, physmap_idx);
582 }
583