xref: /freebsd/sys/x86/xen/pv.c (revision b72ae900d4348118829fe04abdc11b620930c30f)
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/acpica_machdep.h>
64 #include <x86/apicvar.h>
65 #include <x86/init.h>
66 #include <machine/pc/bios.h>
67 #include <machine/smp.h>
68 #include <machine/intr_machdep.h>
69 #include <machine/md_var.h>
70 #include <machine/metadata.h>
71 #include <machine/cpu.h>
72 
73 #include <xen/xen-os.h>
74 #include <xen/hvm.h>
75 #include <xen/hypervisor.h>
76 #include <xen/xenstore/xenstorevar.h>
77 #include <xen/xen_pv.h>
78 
79 #include <contrib/xen/arch-x86/cpuid.h>
80 #include <contrib/xen/arch-x86/hvm/start_info.h>
81 #include <contrib/xen/vcpu.h>
82 
83 #include <dev/xen/timer/timer.h>
84 
85 #ifdef DDB
86 #include <ddb/ddb.h>
87 #endif
88 
89 /* Native initial function */
90 extern u_int64_t hammer_time(u_int64_t, u_int64_t);
91 /* Xen initial function */
92 uint64_t hammer_time_xen(vm_paddr_t);
93 
94 #define MAX_E820_ENTRIES	128
95 
96 /*--------------------------- Forward Declarations ---------------------------*/
97 static void xen_pvh_parse_preload_data(uint64_t);
98 static void pvh_parse_memmap(vm_paddr_t *, int *);
99 
100 /*---------------------------- Extern Declarations ---------------------------*/
101 /*
102  * Placed by the linker at the end of the bss section, which is the last
103  * section loaded by Xen before loading the symtab and strtab.
104  */
105 extern uint32_t end;
106 
107 /*-------------------------------- Global Data -------------------------------*/
108 struct init_ops xen_pvh_init_ops = {
109 	.parse_preload_data		= xen_pvh_parse_preload_data,
110 	.early_clock_source_init	= xen_clock_init,
111 	.early_delay			= xen_delay,
112 	.parse_memmap			= pvh_parse_memmap,
113 };
114 
115 static struct bios_smap xen_smap[MAX_E820_ENTRIES];
116 
117 static struct hvm_start_info *start_info;
118 
119 /*-------------------------------- Xen PV init -------------------------------*/
120 
121 static int
isxen(void)122 isxen(void)
123 {
124 	static int xen = -1;
125 	uint32_t base;
126 	u_int regs[4];
127 
128 	if (xen != -1)
129 		return (xen);
130 
131 	/*
132 	 * The full code for identifying which hypervisor we're running under
133 	 * is in sys/x86/x86/identcpu.c and runs later in the boot process;
134 	 * this is sufficient to distinguish Xen PVH booting from non-Xen PVH
135 	 * and skip some very early Xen-specific code in the non-Xen case.
136 	 */
137 	xen = 0;
138 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
139 		do_cpuid(base, regs);
140 		if (regs[1] == XEN_CPUID_SIGNATURE_EBX &&
141 		    regs[2] == XEN_CPUID_SIGNATURE_ECX &&
142 		    regs[3] == XEN_CPUID_SIGNATURE_EDX) {
143 			xen = 1;
144 			break;
145 		}
146 	}
147 	return (xen);
148 }
149 
150 #define CRASH(...) do {					\
151 	if (isxen())					\
152 		xc_printf(__VA_ARGS__);			\
153 	halt();						\
154 } while (0)
155 
156 uint64_t
hammer_time_xen(vm_paddr_t start_info_paddr)157 hammer_time_xen(vm_paddr_t start_info_paddr)
158 {
159 	struct hvm_modlist_entry *mod;
160 	uint64_t physfree;
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 	/* Set the hooks for early functions that diverge from bare metal */
200 	init_ops = xen_pvh_init_ops;
201 	hvm_start_flags = start_info->flags;
202 
203 	/* Now we can jump into the native init function */
204 	return (hammer_time(0, physfree));
205 }
206 
207 /*-------------------------------- PV specific -------------------------------*/
208 
209 #ifdef DDB
210 /*
211  * The way Xen loads the symtab is different from the native boot loader,
212  * because it's tailored for NetBSD. So we have to adapt and use the same
213  * method as NetBSD. Portions of the code below have been picked from NetBSD:
214  * sys/kern/kern_ksyms.c CVS Revision 1.71.
215  */
216 static void
xen_pvh_parse_symtab(void)217 xen_pvh_parse_symtab(void)
218 {
219 	Elf_Ehdr *ehdr;
220 	Elf_Shdr *shdr;
221 	int i, j;
222 
223 	ehdr = (Elf_Ehdr *)(&end + 1);
224 	if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) ||
225 	    ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
226 	    ehdr->e_version > 1) {
227 		if (isxen())
228 			xc_printf("Unable to load ELF symtab: invalid symbol table\n");
229 		return;
230 	}
231 
232 	shdr = (Elf_Shdr *)((uint8_t *)ehdr + ehdr->e_shoff);
233 	/* Find the symbol table and the corresponding string table. */
234 	for (i = 1; i < ehdr->e_shnum; i++) {
235 		if (shdr[i].sh_type != SHT_SYMTAB)
236 			continue;
237 		if (shdr[i].sh_offset == 0)
238 			continue;
239 		ksymtab = (uintptr_t)((uint8_t *)ehdr + shdr[i].sh_offset);
240 		ksymtab_size = shdr[i].sh_size;
241 		j = shdr[i].sh_link;
242 		if (shdr[j].sh_offset == 0)
243 			continue; /* Can this happen? */
244 		kstrtab = (uintptr_t)((uint8_t *)ehdr + shdr[j].sh_offset);
245 		break;
246 	}
247 
248 	if ((ksymtab == 0 || kstrtab == 0) && isxen())
249 		xc_printf(
250     "Unable to load ELF symtab: could not find symtab or strtab\n");
251 }
252 #endif
253 
254 static void
xen_pvh_parse_preload_data(uint64_t modulep)255 xen_pvh_parse_preload_data(uint64_t modulep)
256 {
257 	vm_ooffset_t off;
258 	vm_paddr_t metadata;
259 	char *envp;
260 
261 	TSENTER();
262 	if (start_info->modlist_paddr != 0) {
263 		struct hvm_modlist_entry *mod;
264 		const char *cmdline;
265 
266 		mod = (struct hvm_modlist_entry *)
267 		    (start_info->modlist_paddr + KERNBASE);
268 		cmdline = mod[0].cmdline_paddr ?
269 		    (const char *)(mod[0].cmdline_paddr + KERNBASE) : NULL;
270 
271 		if (strcmp(cmdline, "header") == 0) {
272 			struct xen_header *header;
273 
274 			header = (struct xen_header *)(mod[0].paddr + KERNBASE);
275 
276 			if ((header->flags & XENHEADER_HAS_MODULEP_OFFSET) !=
277 			    XENHEADER_HAS_MODULEP_OFFSET) {
278 				xc_printf("Unable to load module metadata\n");
279 				HYPERVISOR_shutdown(SHUTDOWN_crash);
280 			}
281 
282 			preload_metadata = (caddr_t)(mod[0].paddr +
283 			    header->modulep_offset + KERNBASE);
284 
285 			/*
286 			 * Xen has relocated the metadata and the modules, so
287 			 * we need to recalculate it's position. This is done
288 			 * by saving the original modulep address and then
289 			 * calculating the offset from the real modulep
290 			 * position.
291 			 */
292 			off = header->modulep_offset;
293 		} else {
294 			preload_metadata = (caddr_t)(mod[0].paddr + KERNBASE);
295 			off = 0;
296 		}
297 
298 		/* Initialize preload_kmdp */
299 		preload_initkmdp(false);
300 		if (preload_kmdp == NULL) {
301 			xc_printf("Unable to find kernel metadata\n");
302 			HYPERVISOR_shutdown(SHUTDOWN_crash);
303 		}
304 
305 		metadata = MD_FETCH(preload_kmdp, MODINFOMD_MODULEP,
306 		    vm_paddr_t);
307 		off += mod[0].paddr + KERNBASE - metadata;
308 
309 		preload_bootstrap_relocate(off);
310 
311 		boothowto = MD_FETCH(preload_kmdp, MODINFOMD_HOWTO, int);
312 		envp = MD_FETCH(preload_kmdp, MODINFOMD_ENVP, char *);
313 		if (envp != NULL)
314 			envp += off;
315 		init_static_kenv(envp, 0);
316 
317 		if (MD_FETCH(preload_kmdp, MODINFOMD_EFI_MAP, void *) != NULL)
318 		    strlcpy(bootmethod, "UEFI", sizeof(bootmethod));
319 		else
320 		    strlcpy(bootmethod, "BIOS", sizeof(bootmethod));
321 	} else {
322 		static char kenv_buffer[PAGE_SIZE];
323 
324 		/* Provide a static kenv so the command line can be parsed. */
325 		init_static_kenv(kenv_buffer, sizeof(kenv_buffer));
326 
327 		/* Parse the extra boot information given by Xen */
328 		if (start_info->cmdline_paddr != 0)
329 			boot_parse_cmdline_delim(
330 			    (char *)(start_info->cmdline_paddr + KERNBASE),
331 			    ", \t\n");
332 		strlcpy(bootmethod, "PVH", sizeof(bootmethod));
333 	}
334 
335 	boothowto |= boot_env_to_howto();
336 
337 	/*
338 	 * When booted as a PVH guest FreeBSD must not use the RSDP address
339 	 * hint provided by the loader because it points to the native set of
340 	 * ACPI tables instead of the ones crafted by Xen.
341 	 */
342 	acpi_set_root(start_info->rsdp_paddr);
343 
344 #ifdef DDB
345 	xen_pvh_parse_symtab();
346 #endif
347 	TSEXIT();
348 }
349 
350 static void
pvh_parse_memmap_start_info(vm_paddr_t * physmap,int * physmap_idx)351 pvh_parse_memmap_start_info(vm_paddr_t *physmap,
352     int *physmap_idx)
353 {
354 	const struct hvm_memmap_table_entry * entries;
355 	size_t nentries;
356 	size_t i;
357 
358 	/* Extract from HVM start_info. */
359 	entries = (struct hvm_memmap_table_entry *)(start_info->memmap_paddr + KERNBASE);
360 	nentries = start_info->memmap_entries;
361 
362 	/* Convert into E820 format and handle one by one. */
363 	for (i = 0; i < nentries; i++) {
364 		struct bios_smap entry;
365 
366 		entry.base = entries[i].addr;
367 		entry.length = entries[i].size;
368 
369 		/*
370 		 * Luckily for us, the XEN_HVM_MEMMAP_TYPE_* values exactly
371 		 * match the SMAP_TYPE_* values so we don't need to translate
372 		 * anything here.
373 		 */
374 		entry.type = entries[i].type;
375 
376 		bios_add_smap_entries(&entry, 1, physmap, physmap_idx);
377 	}
378 }
379 
380 static void
xen_pvh_parse_memmap(vm_paddr_t * physmap,int * physmap_idx)381 xen_pvh_parse_memmap(vm_paddr_t *physmap, int *physmap_idx)
382 {
383 	struct xen_memory_map memmap;
384 	u_int32_t size;
385 	int rc;
386 
387 	/* We should only reach here if we're running under Xen. */
388 	KASSERT(isxen(), ("xen_pvh_parse_memmap reached when !Xen"));
389 
390 	/* Fetch the E820 map from Xen */
391 	memmap.nr_entries = MAX_E820_ENTRIES;
392 	set_xen_guest_handle(memmap.buffer, xen_smap);
393 	rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
394 	if (rc) {
395 		xc_printf("ERROR: unable to fetch Xen E820 memory map: %d\n",
396 		    rc);
397 		HYPERVISOR_shutdown(SHUTDOWN_crash);
398 	}
399 
400 	size = memmap.nr_entries * sizeof(xen_smap[0]);
401 
402 	bios_add_smap_entries(xen_smap, size, physmap, physmap_idx);
403 }
404 
405 static void
pvh_parse_memmap(vm_paddr_t * physmap,int * physmap_idx)406 pvh_parse_memmap(vm_paddr_t *physmap, int *physmap_idx)
407 {
408 
409 	/*
410 	 * If version >= 1 and memmap_paddr != 0, use the memory map provided
411 	 * in the start_info structure; if not, we're running under legacy
412 	 * Xen and need to use the Xen hypercall.
413 	 */
414 	if ((start_info->version >= 1) && (start_info->memmap_paddr != 0))
415 		pvh_parse_memmap_start_info(physmap, physmap_idx);
416 	else
417 		xen_pvh_parse_memmap(physmap, physmap_idx);
418 }
419