FreeBSD/Linux Kernel Cross Reference
sys/x86/xen/pv.c
1 /*
2 * Copyright (c) 2004 Christian Limpach.
3 * Copyright (c) 2004-2006,2008 Kip Macy
4 * Copyright (c) 2008 The NetBSD Foundation, Inc.
5 * Copyright (c) 2013 Roger Pau Monné <roger.pau@citrix.com>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32
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
50 #include <vm/vm.h>
51 #include <vm/vm_extern.h>
52 #include <vm/vm_kern.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_map.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_pager.h>
57 #include <vm/vm_param.h>
58
59 #include <machine/intr_machdep.h>
60 #include <x86/apicvar.h>
61 #include <x86/init.h>
62 #include <machine/pc/bios.h>
63 #include <machine/smp.h>
64 #include <machine/intr_machdep.h>
65 #include <machine/metadata.h>
66
67 #include <xen/xen-os.h>
68 #include <xen/hypervisor.h>
69 #include <xen/xenstore/xenstorevar.h>
70 #include <xen/xen_pv.h>
71 #include <xen/xen_msi.h>
72
73 #include <xen/interface/vcpu.h>
74
75 #include <dev/xen/timer/timer.h>
76
77 #ifdef DDB
78 #include <ddb/ddb.h>
79 #endif
80
81 /* Native initial function */
82 extern u_int64_t hammer_time(u_int64_t, u_int64_t);
83 /* Xen initial function */
84 uint64_t hammer_time_xen(start_info_t *, uint64_t);
85
86 #define MAX_E820_ENTRIES 128
87
88 /*--------------------------- Forward Declarations ---------------------------*/
89 static caddr_t xen_pv_parse_preload_data(u_int64_t);
90 static void xen_pv_parse_memmap(caddr_t, vm_paddr_t *, int *);
91
92 #ifdef SMP
93 static int xen_pv_start_all_aps(void);
94 #endif
95
96 /*---------------------------- Extern Declarations ---------------------------*/
97 #ifdef SMP
98 /* Variables used by amd64 mp_machdep to start APs */
99 extern char *doublefault_stack;
100 extern char *mce_stack;
101 extern char *nmi_stack;
102 #endif
103
104 /*
105 * Placed by the linker at the end of the bss section, which is the last
106 * section loaded by Xen before loading the symtab and strtab.
107 */
108 extern uint32_t end;
109
110 /*-------------------------------- Global Data -------------------------------*/
111 /* Xen init_ops implementation. */
112 struct init_ops xen_init_ops = {
113 .parse_preload_data = xen_pv_parse_preload_data,
114 .early_clock_source_init = xen_clock_init,
115 .early_delay = xen_delay,
116 .parse_memmap = xen_pv_parse_memmap,
117 #ifdef SMP
118 .start_all_aps = xen_pv_start_all_aps,
119 #endif
120 .msi_init = xen_msi_init,
121 };
122
123 static struct bios_smap xen_smap[MAX_E820_ENTRIES];
124
125 /*-------------------------------- Xen PV init -------------------------------*/
126 /*
127 * First function called by the Xen PVH boot sequence.
128 *
129 * Set some Xen global variables and prepare the environment so it is
130 * as similar as possible to what native FreeBSD init function expects.
131 */
132 uint64_t
133 hammer_time_xen(start_info_t *si, uint64_t xenstack)
134 {
135 uint64_t physfree;
136 uint64_t *PT4 = (u_int64_t *)xenstack;
137 uint64_t *PT3 = (u_int64_t *)(xenstack + PAGE_SIZE);
138 uint64_t *PT2 = (u_int64_t *)(xenstack + 2 * PAGE_SIZE);
139 int i;
140
141 xen_domain_type = XEN_PV_DOMAIN;
142 vm_guest = VM_GUEST_XEN;
143
144 if ((si == NULL) || (xenstack == 0)) {
145 xc_printf("ERROR: invalid start_info or xen stack, halting\n");
146 HYPERVISOR_shutdown(SHUTDOWN_crash);
147 }
148
149 xc_printf("FreeBSD PVH running on %s\n", si->magic);
150
151 /* We use 3 pages of xen stack for the boot pagetables */
152 physfree = xenstack + 3 * PAGE_SIZE - KERNBASE;
153
154 /* Setup Xen global variables */
155 HYPERVISOR_start_info = si;
156 HYPERVISOR_shared_info =
157 (shared_info_t *)(si->shared_info + KERNBASE);
158
159 /*
160 * Setup some misc global variables for Xen devices
161 *
162 * XXX: Devices that need these specific variables should
163 * be rewritten to fetch this info by themselves from the
164 * start_info page.
165 */
166 xen_store = (struct xenstore_domain_interface *)
167 (ptoa(si->store_mfn) + KERNBASE);
168 console_page = (char *)(ptoa(si->console.domU.mfn) + KERNBASE);
169
170 /*
171 * Use the stack Xen gives us to build the page tables
172 * as native FreeBSD expects to find them (created
173 * by the boot trampoline).
174 */
175 for (i = 0; i < (PAGE_SIZE / sizeof(uint64_t)); i++) {
176 /*
177 * Each slot of the level 4 pages points
178 * to the same level 3 page
179 */
180 PT4[i] = ((uint64_t)&PT3[0]) - KERNBASE;
181 PT4[i] |= PG_V | PG_RW | PG_U;
182
183 /*
184 * Each slot of the level 3 pages points
185 * to the same level 2 page
186 */
187 PT3[i] = ((uint64_t)&PT2[0]) - KERNBASE;
188 PT3[i] |= PG_V | PG_RW | PG_U;
189
190 /*
191 * The level 2 page slots are mapped with
192 * 2MB pages for 1GB.
193 */
194 PT2[i] = i * (2 * 1024 * 1024);
195 PT2[i] |= PG_V | PG_RW | PG_PS | PG_U;
196 }
197 load_cr3(((uint64_t)&PT4[0]) - KERNBASE);
198
199 /* Set the hooks for early functions that diverge from bare metal */
200 init_ops = xen_init_ops;
201 apic_ops = xen_apic_ops;
202
203 /* Now we can jump into the native init function */
204 return (hammer_time(0, physfree));
205 }
206
207 /*-------------------------------- PV specific -------------------------------*/
208 #ifdef SMP
209 static bool
210 start_xen_ap(int cpu)
211 {
212 struct vcpu_guest_context *ctxt;
213 int ms, cpus = mp_naps;
214 const size_t stacksize = kstack_pages * PAGE_SIZE;
215
216 /* allocate and set up an idle stack data page */
217 bootstacks[cpu] =
218 (void *)kmem_malloc(kernel_arena, stacksize, M_WAITOK | M_ZERO);
219 doublefault_stack =
220 (char *)kmem_malloc(kernel_arena, PAGE_SIZE, M_WAITOK | M_ZERO);
221 mce_stack =
222 (char *)kmem_malloc(kernel_arena, PAGE_SIZE, M_WAITOK | M_ZERO);
223 nmi_stack =
224 (char *)kmem_malloc(kernel_arena, PAGE_SIZE, M_WAITOK | M_ZERO);
225 dpcpu =
226 (void *)kmem_malloc(kernel_arena, DPCPU_SIZE, M_WAITOK | M_ZERO);
227
228 bootSTK = (char *)bootstacks[cpu] + kstack_pages * PAGE_SIZE - 8;
229 bootAP = cpu;
230
231 ctxt = malloc(sizeof(*ctxt), M_TEMP, M_WAITOK | M_ZERO);
232
233 ctxt->flags = VGCF_IN_KERNEL;
234 ctxt->user_regs.rip = (unsigned long) init_secondary;
235 ctxt->user_regs.rsp = (unsigned long) bootSTK;
236
237 /* Set the AP to use the same page tables */
238 ctxt->ctrlreg[3] = KPML4phys;
239
240 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
241 panic("unable to initialize AP#%d", cpu);
242
243 free(ctxt, M_TEMP);
244
245 /* Launch the vCPU */
246 if (HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
247 panic("unable to start AP#%d", cpu);
248
249 /* Wait up to 5 seconds for it to start. */
250 for (ms = 0; ms < 5000; ms++) {
251 if (mp_naps > cpus)
252 return (true);
253 DELAY(1000);
254 }
255
256 return (false);
257 }
258
259 static int
260 xen_pv_start_all_aps(void)
261 {
262 int cpu;
263
264 mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
265
266 for (cpu = 1; cpu < mp_ncpus; cpu++) {
267
268 /* attempt to start the Application Processor */
269 if (!start_xen_ap(cpu))
270 panic("AP #%d failed to start!", cpu);
271
272 CPU_SET(cpu, &all_cpus); /* record AP in CPU map */
273 }
274
275 return (mp_naps);
276 }
277 #endif /* SMP */
278
279 /*
280 * Functions to convert the "extra" parameters passed by Xen
281 * into FreeBSD boot options.
282 */
283 static void
284 xen_pv_set_env(void)
285 {
286 char *cmd_line_next, *cmd_line;
287 size_t env_size;
288
289 cmd_line = HYPERVISOR_start_info->cmd_line;
290 env_size = sizeof(HYPERVISOR_start_info->cmd_line);
291
292 /* Skip leading spaces */
293 for (; isspace(*cmd_line) && (env_size != 0); cmd_line++)
294 env_size--;
295
296 /* Replace ',' with '\0' */
297 for (cmd_line_next = cmd_line; strsep(&cmd_line_next, ",") != NULL;)
298 ;
299
300 init_static_kenv(cmd_line, 0);
301 }
302
303 #ifdef DDB
304 /*
305 * The way Xen loads the symtab is different from the native boot loader,
306 * because it's tailored for NetBSD. So we have to adapt and use the same
307 * method as NetBSD. Portions of the code below have been picked from NetBSD:
308 * sys/kern/kern_ksyms.c CVS Revision 1.71.
309 */
310 static void
311 xen_pv_parse_symtab(void)
312 {
313 Elf_Ehdr *ehdr;
314 Elf_Shdr *shdr;
315 vm_offset_t sym_end;
316 uint32_t size;
317 int i, j;
318
319 size = end;
320 sym_end = HYPERVISOR_start_info->mod_start != 0 ?
321 HYPERVISOR_start_info->mod_start :
322 HYPERVISOR_start_info->mfn_list;
323
324 /*
325 * Make sure the size is right headed, sym_end is just a
326 * high boundary, but at least allows us to fail earlier.
327 */
328 if ((vm_offset_t)&end + size > sym_end) {
329 xc_printf("Unable to load ELF symtab: size mismatch\n");
330 return;
331 }
332
333 ehdr = (Elf_Ehdr *)(&end + 1);
334 if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) ||
335 ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
336 ehdr->e_version > 1) {
337 xc_printf("Unable to load ELF symtab: invalid symbol table\n");
338 return;
339 }
340
341 shdr = (Elf_Shdr *)((uint8_t *)ehdr + ehdr->e_shoff);
342 /* Find the symbol table and the corresponding string table. */
343 for (i = 1; i < ehdr->e_shnum; i++) {
344 if (shdr[i].sh_type != SHT_SYMTAB)
345 continue;
346 if (shdr[i].sh_offset == 0)
347 continue;
348 ksymtab = (uintptr_t)((uint8_t *)ehdr + shdr[i].sh_offset);
349 ksymtab_size = shdr[i].sh_size;
350 j = shdr[i].sh_link;
351 if (shdr[j].sh_offset == 0)
352 continue; /* Can this happen? */
353 kstrtab = (uintptr_t)((uint8_t *)ehdr + shdr[j].sh_offset);
354 break;
355 }
356
357 if (ksymtab == 0 || kstrtab == 0) {
358 xc_printf(
359 "Unable to load ELF symtab: could not find symtab or strtab\n");
360 return;
361 }
362 }
363 #endif
364
365 static caddr_t
366 xen_pv_parse_preload_data(u_int64_t modulep)
367 {
368 caddr_t kmdp;
369 vm_ooffset_t off;
370 vm_paddr_t metadata;
371 char *envp;
372
373 if (HYPERVISOR_start_info->mod_start != 0) {
374 preload_metadata = (caddr_t)(HYPERVISOR_start_info->mod_start);
375
376 kmdp = preload_search_by_type("elf kernel");
377 if (kmdp == NULL)
378 kmdp = preload_search_by_type("elf64 kernel");
379 KASSERT(kmdp != NULL, ("unable to find kernel"));
380
381 /*
382 * Xen has relocated the metadata and the modules,
383 * so we need to recalculate it's position. This is
384 * done by saving the original modulep address and
385 * then calculating the offset with mod_start,
386 * which contains the relocated modulep address.
387 */
388 metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t);
389 off = HYPERVISOR_start_info->mod_start - metadata;
390
391 preload_bootstrap_relocate(off);
392
393 boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
394 envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *);
395 if (envp != NULL)
396 envp += off;
397 init_static_kenv(envp, 0);
398 } else {
399 /* Parse the extra boot information given by Xen */
400 xen_pv_set_env();
401 boothowto |= boot_env_to_howto();
402 kmdp = NULL;
403 }
404
405 #ifdef DDB
406 xen_pv_parse_symtab();
407 #endif
408 return (kmdp);
409 }
410
411 static void
412 xen_pv_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx)
413 {
414 struct xen_memory_map memmap;
415 u_int32_t size;
416 int rc;
417
418 /* Fetch the E820 map from Xen */
419 memmap.nr_entries = MAX_E820_ENTRIES;
420 set_xen_guest_handle(memmap.buffer, xen_smap);
421 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
422 if (rc)
423 panic("unable to fetch Xen E820 memory map");
424 size = memmap.nr_entries * sizeof(xen_smap[0]);
425
426 bios_add_smap_entries(xen_smap, size, physmap, physmap_idx);
427 }
Cache object: 87e17ca06c291aa7e457767ebf1737e1
|