xref: /freebsd/sys/x86/xen/pv.c (revision 5ca8e32633c4ffbbcd6762e5888b6a4ba0708c6c)
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 	halt();						\
153 } while (0)
154 
155 uint64_t
156 hammer_time_xen(vm_paddr_t start_info_paddr)
157 {
158 	struct hvm_modlist_entry *mod;
159 	uint64_t physfree;
160 	char *kenv;
161 
162 	start_info = (struct hvm_start_info *)(start_info_paddr + KERNBASE);
163 	if (start_info->magic != XEN_HVM_START_MAGIC_VALUE) {
164 		CRASH("Unknown magic value in start_info struct: %#x\n",
165 		    start_info->magic);
166 	}
167 
168 	/*
169 	 * Select the higher address to use as physfree: either after
170 	 * start_info, after the kernel, after the memory map or after any of
171 	 * the modules.  We assume enough memory to be available after the
172 	 * selected address for the needs of very early memory allocations.
173 	 */
174 	physfree = roundup2(start_info_paddr + sizeof(struct hvm_start_info),
175 	    PAGE_SIZE);
176 	physfree = MAX(roundup2((vm_paddr_t)_end - KERNBASE, PAGE_SIZE),
177 	    physfree);
178 
179 	if (start_info->memmap_paddr != 0)
180 		physfree = MAX(roundup2(start_info->memmap_paddr +
181 		    start_info->memmap_entries *
182 		    sizeof(struct hvm_memmap_table_entry), PAGE_SIZE),
183 		    physfree);
184 
185 	if (start_info->modlist_paddr != 0) {
186 		unsigned int i;
187 
188 		if (start_info->nr_modules == 0) {
189 			CRASH(
190 			    "ERROR: modlist_paddr != 0 but nr_modules == 0\n");
191 		}
192 		mod = (struct hvm_modlist_entry *)
193 		    (start_info->modlist_paddr + KERNBASE);
194 		for (i = 0; i < start_info->nr_modules; i++)
195 			physfree = MAX(roundup2(mod[i].paddr + mod[i].size,
196 			    PAGE_SIZE), physfree);
197 	}
198 
199 	/*
200 	 * Init a static kenv using a free page. The contents will be filled
201 	 * from the parse_preload_data hook.
202 	 */
203 	kenv = (void *)(physfree + KERNBASE);
204 	physfree += PAGE_SIZE;
205 	bzero_early(kenv, PAGE_SIZE);
206 	init_static_kenv(kenv, PAGE_SIZE);
207 
208 	/* Set the hooks for early functions that diverge from bare metal */
209 	init_ops = xen_pvh_init_ops;
210 	hvm_start_flags = start_info->flags;
211 
212 	/* Now we can jump into the native init function */
213 	return (hammer_time(0, physfree));
214 }
215 
216 /*-------------------------------- PV specific -------------------------------*/
217 
218 /*
219  * When booted as a PVH guest FreeBSD needs to avoid using the RSDP address
220  * hint provided by the loader because it points to the native set of ACPI
221  * tables instead of the ones crafted by Xen. The acpi.rsdp env variable is
222  * removed from kenv if present, and a new acpi.rsdp is added to kenv that
223  * points to the address of the Xen crafted RSDP.
224  */
225 static bool reject_option(const char *option)
226 {
227 	static const char *reject[] = {
228 		"acpi.rsdp",
229 	};
230 	unsigned int i;
231 
232 	for (i = 0; i < nitems(reject); i++)
233 		if (strncmp(option, reject[i], strlen(reject[i])) == 0)
234 			return (true);
235 
236 	return (false);
237 }
238 
239 static void
240 xen_pvh_set_env(char *env, bool (*filter)(const char *))
241 {
242 	char *option;
243 
244 	if (env == NULL)
245 		return;
246 
247 	option = env;
248 	while (*option != 0) {
249 		char *value;
250 
251 		if (filter != NULL && filter(option)) {
252 			option += strlen(option) + 1;
253 			continue;
254 		}
255 
256 		value = option;
257 		option = strsep(&value, "=");
258 		if (kern_setenv(option, value) != 0 && isxen())
259 			xc_printf("unable to add kenv %s=%s\n", option, value);
260 		option = value + strlen(value) + 1;
261 	}
262 }
263 
264 #ifdef DDB
265 /*
266  * The way Xen loads the symtab is different from the native boot loader,
267  * because it's tailored for NetBSD. So we have to adapt and use the same
268  * method as NetBSD. Portions of the code below have been picked from NetBSD:
269  * sys/kern/kern_ksyms.c CVS Revision 1.71.
270  */
271 static void
272 xen_pvh_parse_symtab(void)
273 {
274 	Elf_Ehdr *ehdr;
275 	Elf_Shdr *shdr;
276 	int i, j;
277 
278 	ehdr = (Elf_Ehdr *)(&end + 1);
279 	if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) ||
280 	    ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
281 	    ehdr->e_version > 1) {
282 		if (isxen())
283 			xc_printf("Unable to load ELF symtab: invalid symbol table\n");
284 		return;
285 	}
286 
287 	shdr = (Elf_Shdr *)((uint8_t *)ehdr + ehdr->e_shoff);
288 	/* Find the symbol table and the corresponding string table. */
289 	for (i = 1; i < ehdr->e_shnum; i++) {
290 		if (shdr[i].sh_type != SHT_SYMTAB)
291 			continue;
292 		if (shdr[i].sh_offset == 0)
293 			continue;
294 		ksymtab = (uintptr_t)((uint8_t *)ehdr + shdr[i].sh_offset);
295 		ksymtab_size = shdr[i].sh_size;
296 		j = shdr[i].sh_link;
297 		if (shdr[j].sh_offset == 0)
298 			continue; /* Can this happen? */
299 		kstrtab = (uintptr_t)((uint8_t *)ehdr + shdr[j].sh_offset);
300 		break;
301 	}
302 
303 	if ((ksymtab == 0 || kstrtab == 0) && isxen())
304 		xc_printf(
305     "Unable to load ELF symtab: could not find symtab or strtab\n");
306 }
307 #endif
308 
309 static caddr_t
310 xen_pvh_parse_preload_data(uint64_t modulep)
311 {
312 	caddr_t kmdp;
313 	vm_ooffset_t off;
314 	vm_paddr_t metadata;
315 	char *envp;
316 	char acpi_rsdp[19];
317 
318 	TSENTER();
319 	if (start_info->modlist_paddr != 0) {
320 		struct hvm_modlist_entry *mod;
321 		const char *cmdline;
322 
323 		mod = (struct hvm_modlist_entry *)
324 		    (start_info->modlist_paddr + KERNBASE);
325 		cmdline = mod[0].cmdline_paddr ?
326 		    (const char *)(mod[0].cmdline_paddr + KERNBASE) : NULL;
327 
328 		if (strcmp(cmdline, "header") == 0) {
329 			struct xen_header *header;
330 
331 			header = (struct xen_header *)(mod[0].paddr + KERNBASE);
332 
333 			if ((header->flags & XENHEADER_HAS_MODULEP_OFFSET) !=
334 			    XENHEADER_HAS_MODULEP_OFFSET) {
335 				xc_printf("Unable to load module metadata\n");
336 				HYPERVISOR_shutdown(SHUTDOWN_crash);
337 			}
338 
339 			preload_metadata = (caddr_t)(mod[0].paddr +
340 			    header->modulep_offset + KERNBASE);
341 
342 			kmdp = preload_search_by_type("elf kernel");
343 			if (kmdp == NULL)
344 				kmdp = preload_search_by_type("elf64 kernel");
345 			if (kmdp == NULL) {
346 				xc_printf("Unable to find kernel\n");
347 				HYPERVISOR_shutdown(SHUTDOWN_crash);
348 			}
349 
350 			/*
351 			 * Xen has relocated the metadata and the modules, so
352 			 * we need to recalculate it's position. This is done
353 			 * by saving the original modulep address and then
354 			 * calculating the offset from the real modulep
355 			 * position.
356 			 */
357 			metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP,
358 			    vm_paddr_t);
359 			off = mod[0].paddr + header->modulep_offset - metadata +
360 			    KERNBASE;
361 		} else {
362 			preload_metadata = (caddr_t)(mod[0].paddr + KERNBASE);
363 
364 			kmdp = preload_search_by_type("elf kernel");
365 			if (kmdp == NULL)
366 				kmdp = preload_search_by_type("elf64 kernel");
367 			if (kmdp == NULL) {
368 				xc_printf("Unable to find kernel\n");
369 				HYPERVISOR_shutdown(SHUTDOWN_crash);
370 			}
371 
372 			metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t);
373 			off = mod[0].paddr + KERNBASE - metadata;
374 		}
375 
376 		preload_bootstrap_relocate(off);
377 
378 		boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
379 		envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *);
380 		if (envp != NULL)
381 			envp += off;
382 		xen_pvh_set_env(envp, reject_option);
383 
384 		if (MD_FETCH(kmdp, MODINFOMD_EFI_MAP, void *) != NULL)
385 		    strlcpy(bootmethod, "UEFI", sizeof(bootmethod));
386 		else
387 		    strlcpy(bootmethod, "BIOS", sizeof(bootmethod));
388 	} else {
389 		/* Parse the extra boot information given by Xen */
390 		if (start_info->cmdline_paddr != 0)
391 			boot_parse_cmdline_delim(
392 			    (char *)(start_info->cmdline_paddr + KERNBASE),
393 			    ", \t\n");
394 		kmdp = NULL;
395 		strlcpy(bootmethod, "PVH", sizeof(bootmethod));
396 	}
397 
398 	boothowto |= boot_env_to_howto();
399 
400 	snprintf(acpi_rsdp, sizeof(acpi_rsdp), "%#" PRIx64,
401 	    start_info->rsdp_paddr);
402 	kern_setenv("acpi.rsdp", acpi_rsdp);
403 
404 #ifdef DDB
405 	xen_pvh_parse_symtab();
406 #endif
407 	TSEXIT();
408 	return (kmdp);
409 }
410 
411 static void
412 pvh_parse_memmap_start_info(caddr_t kmdp, vm_paddr_t *physmap,
413     int *physmap_idx)
414 {
415 	const struct hvm_memmap_table_entry * entries;
416 	size_t nentries;
417 	size_t i;
418 
419 	/* Extract from HVM start_info. */
420 	entries = (struct hvm_memmap_table_entry *)(start_info->memmap_paddr + KERNBASE);
421 	nentries = start_info->memmap_entries;
422 
423 	/* Convert into E820 format and handle one by one. */
424 	for (i = 0; i < nentries; i++) {
425 		struct bios_smap entry;
426 
427 		entry.base = entries[i].addr;
428 		entry.length = entries[i].size;
429 
430 		/*
431 		 * Luckily for us, the XEN_HVM_MEMMAP_TYPE_* values exactly
432 		 * match the SMAP_TYPE_* values so we don't need to translate
433 		 * anything here.
434 		 */
435 		entry.type = entries[i].type;
436 
437 		bios_add_smap_entries(&entry, 1, physmap, physmap_idx);
438 	}
439 }
440 
441 static void
442 xen_pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx)
443 {
444 	struct xen_memory_map memmap;
445 	u_int32_t size;
446 	int rc;
447 
448 	/* We should only reach here if we're running under Xen. */
449 	KASSERT(isxen(), ("xen_pvh_parse_memmap reached when !Xen"));
450 
451 	/* Fetch the E820 map from Xen */
452 	memmap.nr_entries = MAX_E820_ENTRIES;
453 	set_xen_guest_handle(memmap.buffer, xen_smap);
454 	rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
455 	if (rc) {
456 		xc_printf("ERROR: unable to fetch Xen E820 memory map: %d\n",
457 		    rc);
458 		HYPERVISOR_shutdown(SHUTDOWN_crash);
459 	}
460 
461 	size = memmap.nr_entries * sizeof(xen_smap[0]);
462 
463 	bios_add_smap_entries(xen_smap, size, physmap, physmap_idx);
464 }
465 
466 static void
467 pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx)
468 {
469 
470 	/*
471 	 * If version >= 1 and memmap_paddr != 0, use the memory map provided
472 	 * in the start_info structure; if not, we're running under legacy
473 	 * Xen and need to use the Xen hypercall.
474 	 */
475 	if ((start_info->version >= 1) && (start_info->memmap_paddr != 0))
476 		pvh_parse_memmap_start_info(kmdp, physmap, physmap_idx);
477 	else
478 		xen_pvh_parse_memmap(kmdp, physmap, physmap_idx);
479 }
480