ia64/xen-unstable

view xen/arch/x86/domain_build.c @ 17696:3ed3d0688fa8

x86: Remove 32-bit non-PAE code from dom0 builder, and avoid use of
get_page_type() on pagetable pages (pending support for preemptive
get_page_type(), as dom0 builder would be an awkward place to deal
with preemption).

Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Thu May 22 10:40:40 2008 +0100 (2008-05-22)
parents 8bd776540ab3
children 31a941632a5d
line source
1 /******************************************************************************
2 * domain_build.c
3 *
4 * Copyright (c) 2002-2005, K A Fraser
5 */
7 #include <xen/config.h>
8 #include <xen/init.h>
9 #include <xen/lib.h>
10 #include <xen/ctype.h>
11 #include <xen/sched.h>
12 #include <xen/smp.h>
13 #include <xen/delay.h>
14 #include <xen/event.h>
15 #include <xen/console.h>
16 #include <xen/kernel.h>
17 #include <xen/domain.h>
18 #include <xen/version.h>
19 #include <xen/iocap.h>
20 #include <xen/bitops.h>
21 #include <xen/compat.h>
22 #include <asm/regs.h>
23 #include <asm/system.h>
24 #include <asm/io.h>
25 #include <asm/processor.h>
26 #include <asm/desc.h>
27 #include <asm/i387.h>
28 #include <asm/paging.h>
29 #include <asm/e820.h>
31 #include <public/version.h>
32 #include <public/libelf.h>
34 extern unsigned long initial_images_nrpages(void);
35 extern void discard_initial_images(void);
37 static long __initdata dom0_nrpages;
38 static long __initdata dom0_min_nrpages;
39 static long __initdata dom0_max_nrpages = LONG_MAX;
41 /*
42 * dom0_mem=[min:<min_amt>,][max:<max_amt>,][<amt>]
43 *
44 * <min_amt>: The minimum amount of memory which should be allocated for dom0.
45 * <max_amt>: The maximum amount of memory which should be allocated for dom0.
46 * <amt>: The precise amount of memory to allocate for dom0.
47 *
48 * Notes:
49 * 1. <amt> is clamped from below by <min_amt> and from above by available
50 * memory and <max_amt>
51 * 2. <min_amt> is clamped from above by available memory and <max_amt>
52 * 3. <min_amt> is ignored if it is greater than <max_amt>
53 * 4. If <amt> is not specified, it is calculated as follows:
54 * "All of memory is allocated to domain 0, minus 1/16th which is reserved
55 * for uses such as DMA buffers (the reservation is clamped to 128MB)."
56 *
57 * Each value can be specified as positive or negative:
58 * If +ve: The specified amount is an absolute value.
59 * If -ve: The specified amount is subtracted from total available memory.
60 */
61 static long __init parse_amt(const char *s, const char **ps)
62 {
63 long pages = parse_size_and_unit((*s == '-') ? s+1 : s, ps) >> PAGE_SHIFT;
64 return (*s == '-') ? -pages : pages;
65 }
66 static void __init parse_dom0_mem(const char *s)
67 {
68 do {
69 if ( !strncmp(s, "min:", 4) )
70 dom0_min_nrpages = parse_amt(s+4, &s);
71 else if ( !strncmp(s, "max:", 4) )
72 dom0_max_nrpages = parse_amt(s+4, &s);
73 else
74 dom0_nrpages = parse_amt(s, &s);
75 if ( *s != ',' )
76 break;
77 } while ( *s++ == ',' );
78 }
79 custom_param("dom0_mem", parse_dom0_mem);
81 static unsigned int opt_dom0_max_vcpus;
82 integer_param("dom0_max_vcpus", opt_dom0_max_vcpus);
84 static unsigned int opt_dom0_shadow;
85 boolean_param("dom0_shadow", opt_dom0_shadow);
87 static char opt_dom0_ioports_disable[200] = "";
88 string_param("dom0_ioports_disable", opt_dom0_ioports_disable);
90 #if defined(__i386__)
91 /* No ring-3 access in initial leaf page tables. */
92 #define L1_PROT (_PAGE_PRESENT|_PAGE_RW|_PAGE_ACCESSED)
93 #define L2_PROT (_PAGE_PRESENT|_PAGE_RW|_PAGE_ACCESSED|_PAGE_DIRTY|_PAGE_USER)
94 #define L3_PROT (_PAGE_PRESENT)
95 #elif defined(__x86_64__)
96 /* Allow ring-3 access in long mode as guest cannot use ring 1 ... */
97 #define BASE_PROT (_PAGE_PRESENT|_PAGE_RW|_PAGE_ACCESSED|_PAGE_USER)
98 #define L1_PROT (BASE_PROT|_PAGE_GUEST_KERNEL)
99 /* ... except for compatibility mode guests. */
100 #define COMPAT_L1_PROT (_PAGE_PRESENT|_PAGE_RW|_PAGE_ACCESSED)
101 #define L2_PROT (BASE_PROT|_PAGE_DIRTY)
102 #define L3_PROT (BASE_PROT|_PAGE_DIRTY)
103 #define L4_PROT (BASE_PROT|_PAGE_DIRTY)
104 #endif
106 #define round_pgup(_p) (((_p)+(PAGE_SIZE-1))&PAGE_MASK)
107 #define round_pgdown(_p) ((_p)&PAGE_MASK)
109 static struct page_info * __init alloc_chunk(
110 struct domain *d, unsigned long max_pages)
111 {
112 struct page_info *page;
113 unsigned int order;
114 /*
115 * Allocate up to 2MB at a time: It prevents allocating very large chunks
116 * from DMA pools before the >4GB pool is fully depleted.
117 */
118 if ( max_pages > (2UL << (20 - PAGE_SHIFT)) )
119 max_pages = 2UL << (20 - PAGE_SHIFT);
120 order = get_order_from_pages(max_pages);
121 if ( (max_pages & (max_pages-1)) != 0 )
122 order--;
123 while ( (page = alloc_domheap_pages(d, order, 0)) == NULL )
124 if ( order-- == 0 )
125 break;
126 return page;
127 }
129 static unsigned long __init compute_dom0_nr_pages(void)
130 {
131 unsigned long avail = avail_domheap_pages() + initial_images_nrpages();
133 /*
134 * If domain 0 allocation isn't specified, reserve 1/16th of available
135 * memory for things like DMA buffers. This reservation is clamped to
136 * a maximum of 128MB.
137 */
138 if ( dom0_nrpages == 0 )
139 {
140 dom0_nrpages = avail;
141 dom0_nrpages = min(dom0_nrpages / 16, 128L << (20 - PAGE_SHIFT));
142 dom0_nrpages = -dom0_nrpages;
143 }
145 /* Negative memory specification means "all memory - specified amount". */
146 if ( dom0_nrpages < 0 ) dom0_nrpages += avail;
147 if ( dom0_min_nrpages < 0 ) dom0_min_nrpages += avail;
148 if ( dom0_max_nrpages < 0 ) dom0_max_nrpages += avail;
150 /* Clamp dom0 memory according to min/max limits and available memory. */
151 dom0_nrpages = max(dom0_nrpages, dom0_min_nrpages);
152 dom0_nrpages = min(dom0_nrpages, dom0_max_nrpages);
153 dom0_nrpages = min(dom0_nrpages, (long)avail);
155 return dom0_nrpages;
156 }
158 static void __init process_dom0_ioports_disable(void)
159 {
160 unsigned long io_from, io_to;
161 char *t, *s = opt_dom0_ioports_disable;
162 const char *u;
164 if ( *s == '\0' )
165 return;
167 while ( (t = strsep(&s, ",")) != NULL )
168 {
169 io_from = simple_strtoul(t, &u, 16);
170 if ( u == t )
171 {
172 parse_error:
173 printk("Invalid ioport range <%s> "
174 "in dom0_ioports_disable, skipping\n", t);
175 continue;
176 }
178 if ( *u == '\0' )
179 io_to = io_from;
180 else if ( *u == '-' )
181 io_to = simple_strtoul(u + 1, &u, 16);
182 else
183 goto parse_error;
185 if ( (*u != '\0') || (io_to < io_from) || (io_to >= 65536) )
186 goto parse_error;
188 printk("Disabling dom0 access to ioport range %04lx-%04lx\n",
189 io_from, io_to);
191 if ( ioports_deny_access(dom0, io_from, io_to) != 0 )
192 BUG();
193 }
194 }
196 int __init construct_dom0(
197 struct domain *d,
198 unsigned long _image_start, unsigned long image_len,
199 unsigned long _initrd_start, unsigned long initrd_len,
200 char *cmdline)
201 {
202 int i, rc, compatible, compat32, order, machine;
203 struct cpu_user_regs *regs;
204 unsigned long pfn, mfn;
205 unsigned long nr_pages;
206 unsigned long nr_pt_pages;
207 unsigned long alloc_spfn;
208 unsigned long alloc_epfn;
209 unsigned long count;
210 struct page_info *page = NULL;
211 start_info_t *si;
212 struct vcpu *v = d->vcpu[0];
213 unsigned long long value;
214 #if defined(__i386__)
215 char *image_start = (char *)_image_start; /* use lowmem mappings */
216 char *initrd_start = (char *)_initrd_start; /* use lowmem mappings */
217 #elif defined(__x86_64__)
218 char *image_start = __va(_image_start);
219 char *initrd_start = __va(_initrd_start);
220 #endif
221 #if CONFIG_PAGING_LEVELS >= 4
222 l4_pgentry_t *l4tab = NULL, *l4start = NULL;
223 #endif
224 l3_pgentry_t *l3tab = NULL, *l3start = NULL;
225 l2_pgentry_t *l2tab = NULL, *l2start = NULL;
226 l1_pgentry_t *l1tab = NULL, *l1start = NULL;
228 /*
229 * This fully describes the memory layout of the initial domain. All
230 * *_start address are page-aligned, except v_start (and v_end) which are
231 * superpage-aligned.
232 */
233 struct elf_binary elf;
234 struct elf_dom_parms parms;
235 unsigned long vkern_start;
236 unsigned long vkern_end;
237 unsigned long vinitrd_start;
238 unsigned long vinitrd_end;
239 unsigned long vphysmap_start;
240 unsigned long vphysmap_end;
241 unsigned long vstartinfo_start;
242 unsigned long vstartinfo_end;
243 unsigned long vstack_start;
244 unsigned long vstack_end;
245 unsigned long vpt_start;
246 unsigned long vpt_end;
247 unsigned long v_start;
248 unsigned long v_end;
250 /* Machine address of next candidate page-table page. */
251 unsigned long mpt_alloc;
253 /* Sanity! */
254 BUG_ON(d->domain_id != 0);
255 BUG_ON(d->vcpu[0] == NULL);
256 BUG_ON(v->is_initialised);
258 printk("*** LOADING DOMAIN 0 ***\n");
260 d->max_pages = ~0U;
262 nr_pages = compute_dom0_nr_pages();
264 if ( (rc = elf_init(&elf, image_start, image_len)) != 0 )
265 return rc;
266 #ifdef VERBOSE
267 elf_set_verbose(&elf);
268 #endif
269 elf_parse_binary(&elf);
270 if ( (rc = elf_xen_parse(&elf, &parms)) != 0 )
271 return rc;
273 /* compatibility check */
274 compatible = 0;
275 compat32 = 0;
276 machine = elf_uval(&elf, elf.ehdr, e_machine);
277 switch (CONFIG_PAGING_LEVELS) {
278 case 3: /* x86_32p */
279 if (parms.pae == PAEKERN_bimodal)
280 parms.pae = PAEKERN_extended_cr3;
281 printk(" Xen kernel: 32-bit, PAE, lsb\n");
282 if (elf_32bit(&elf) && parms.pae && machine == EM_386)
283 compatible = 1;
284 break;
285 case 4: /* x86_64 */
286 printk(" Xen kernel: 64-bit, lsb, compat32\n");
287 if (elf_32bit(&elf) && parms.pae == PAEKERN_bimodal)
288 parms.pae = PAEKERN_extended_cr3;
289 if (elf_32bit(&elf) && parms.pae && machine == EM_386)
290 {
291 compat32 = 1;
292 compatible = 1;
293 }
294 if (elf_64bit(&elf) && machine == EM_X86_64)
295 compatible = 1;
296 break;
297 }
298 printk(" Dom0 kernel: %s%s, %s, paddr 0x%" PRIx64 " -> 0x%" PRIx64 "\n",
299 elf_64bit(&elf) ? "64-bit" : "32-bit",
300 parms.pae ? ", PAE" : "",
301 elf_msb(&elf) ? "msb" : "lsb",
302 elf.pstart, elf.pend);
303 if ( elf.bsd_symtab_pstart )
304 printk(" Dom0 symbol map 0x%" PRIx64 " -> 0x%" PRIx64 "\n",
305 elf.bsd_symtab_pstart, elf.bsd_symtab_pend);
307 if ( !compatible )
308 {
309 printk("Mismatch between Xen and DOM0 kernel\n");
310 return -EINVAL;
311 }
313 #if defined(__x86_64__)
314 if ( compat32 )
315 {
316 l1_pgentry_t gdt_l1e;
318 d->arch.is_32bit_pv = d->arch.has_32bit_shinfo = 1;
319 v->vcpu_info = (void *)&d->shared_info->compat.vcpu_info[0];
321 if ( nr_pages != (unsigned int)nr_pages )
322 nr_pages = UINT_MAX;
324 /*
325 * Map compatibility Xen segments into every VCPU's GDT. See
326 * arch_domain_create() for further comments.
327 */
328 gdt_l1e = l1e_from_page(virt_to_page(compat_gdt_table),
329 PAGE_HYPERVISOR);
330 for ( i = 0; i < MAX_VIRT_CPUS; i++ )
331 d->arch.mm_perdomain_pt[((i << GDT_LDT_VCPU_SHIFT) +
332 FIRST_RESERVED_GDT_PAGE)] = gdt_l1e;
333 flush_tlb_one_local(GDT_LDT_VIRT_START + FIRST_RESERVED_GDT_BYTE);
334 }
335 #endif
337 if ( parms.pae == PAEKERN_extended_cr3 )
338 set_bit(VMASST_TYPE_pae_extended_cr3, &d->vm_assist);
340 if ( (parms.virt_hv_start_low != UNSET_ADDR) && elf_32bit(&elf) )
341 {
342 unsigned long mask = (1UL << L2_PAGETABLE_SHIFT) - 1;
343 value = (parms.virt_hv_start_low + mask) & ~mask;
344 BUG_ON(!is_pv_32bit_domain(d));
345 #if defined(__i386__)
346 if ( value > HYPERVISOR_VIRT_START )
347 panic("Domain 0 expects too high a hypervisor start address.\n");
348 #else
349 if ( value > __HYPERVISOR_COMPAT_VIRT_START )
350 panic("Domain 0 expects too high a hypervisor start address.\n");
351 HYPERVISOR_COMPAT_VIRT_START(d) =
352 max_t(unsigned int, m2p_compat_vstart, value);
353 #endif
354 }
356 #if defined(__x86_64__)
357 if ( is_pv_32on64_domain(d) )
358 d->arch.physaddr_bitsize =
359 fls((1UL << 32) - HYPERVISOR_COMPAT_VIRT_START(d)) - 1
360 + (PAGE_SIZE - 2);
361 #endif
363 /*
364 * Why do we need this? The number of page-table frames depends on the
365 * size of the bootstrap address space. But the size of the address space
366 * depends on the number of page-table frames (since each one is mapped
367 * read-only). We have a pair of simultaneous equations in two unknowns,
368 * which we solve by exhaustive search.
369 */
370 v_start = parms.virt_base;
371 vkern_start = parms.virt_kstart;
372 vkern_end = parms.virt_kend;
373 vinitrd_start = round_pgup(vkern_end);
374 vinitrd_end = vinitrd_start + initrd_len;
375 vphysmap_start = round_pgup(vinitrd_end);
376 vphysmap_end = vphysmap_start + (nr_pages * (!is_pv_32on64_domain(d) ?
377 sizeof(unsigned long) :
378 sizeof(unsigned int)));
379 vstartinfo_start = round_pgup(vphysmap_end);
380 vstartinfo_end = (vstartinfo_start +
381 sizeof(struct start_info) +
382 sizeof(struct dom0_vga_console_info));
383 vpt_start = round_pgup(vstartinfo_end);
384 for ( nr_pt_pages = 2; ; nr_pt_pages++ )
385 {
386 vpt_end = vpt_start + (nr_pt_pages * PAGE_SIZE);
387 vstack_start = vpt_end;
388 vstack_end = vstack_start + PAGE_SIZE;
389 v_end = (vstack_end + (1UL<<22)-1) & ~((1UL<<22)-1);
390 if ( (v_end - vstack_end) < (512UL << 10) )
391 v_end += 1UL << 22; /* Add extra 4MB to get >= 512kB padding. */
392 #if defined(__i386__)
393 /* 5 pages: 1x 3rd + 4x 2nd level */
394 if ( (((v_end - v_start + ((1UL<<L2_PAGETABLE_SHIFT)-1)) >>
395 L2_PAGETABLE_SHIFT) + 5) <= nr_pt_pages )
396 break;
397 #elif defined(__x86_64__)
398 #define NR(_l,_h,_s) \
399 (((((_h) + ((1UL<<(_s))-1)) & ~((1UL<<(_s))-1)) - \
400 ((_l) & ~((1UL<<(_s))-1))) >> (_s))
401 if ( (1 + /* # L4 */
402 NR(v_start, v_end, L4_PAGETABLE_SHIFT) + /* # L3 */
403 (!is_pv_32on64_domain(d) ?
404 NR(v_start, v_end, L3_PAGETABLE_SHIFT) : /* # L2 */
405 4) + /* # compat L2 */
406 NR(v_start, v_end, L2_PAGETABLE_SHIFT)) /* # L1 */
407 <= nr_pt_pages )
408 break;
409 #endif
410 }
412 order = get_order_from_bytes(v_end - v_start);
413 if ( (1UL << order) > nr_pages )
414 panic("Domain 0 allocation is too small for kernel image.\n");
416 #ifdef __i386__
417 /* Ensure that our low-memory 1:1 mapping covers the allocation. */
418 page = alloc_domheap_pages(d, order, MEMF_bits(30));
419 #else
420 page = alloc_domheap_pages(d, order, 0);
421 #endif
422 if ( page == NULL )
423 panic("Not enough RAM for domain 0 allocation.\n");
424 alloc_spfn = page_to_mfn(page);
425 alloc_epfn = alloc_spfn + d->tot_pages;
427 printk("PHYSICAL MEMORY ARRANGEMENT:\n"
428 " Dom0 alloc.: %"PRIpaddr"->%"PRIpaddr,
429 pfn_to_paddr(alloc_spfn), pfn_to_paddr(alloc_epfn));
430 if ( d->tot_pages < nr_pages )
431 printk(" (%lu pages to be allocated)",
432 nr_pages - d->tot_pages);
433 printk("\nVIRTUAL MEMORY ARRANGEMENT:\n"
434 " Loaded kernel: %p->%p\n"
435 " Init. ramdisk: %p->%p\n"
436 " Phys-Mach map: %p->%p\n"
437 " Start info: %p->%p\n"
438 " Page tables: %p->%p\n"
439 " Boot stack: %p->%p\n"
440 " TOTAL: %p->%p\n",
441 _p(vkern_start), _p(vkern_end),
442 _p(vinitrd_start), _p(vinitrd_end),
443 _p(vphysmap_start), _p(vphysmap_end),
444 _p(vstartinfo_start), _p(vstartinfo_end),
445 _p(vpt_start), _p(vpt_end),
446 _p(vstack_start), _p(vstack_end),
447 _p(v_start), _p(v_end));
448 printk(" ENTRY ADDRESS: %p\n", _p(parms.virt_entry));
450 if ( ((v_end - v_start)>>PAGE_SHIFT) > nr_pages )
451 {
452 printk("Initial guest OS requires too much space\n"
453 "(%luMB is greater than %luMB limit)\n",
454 (v_end-v_start)>>20, nr_pages>>(20-PAGE_SHIFT));
455 return -ENOMEM;
456 }
458 mpt_alloc = (vpt_start - v_start) +
459 (unsigned long)pfn_to_paddr(alloc_spfn);
461 #if defined(__i386__)
462 /*
463 * Protect the lowest 1GB of memory. We use a temporary mapping there
464 * from which we copy the kernel and ramdisk images.
465 */
466 if ( v_start < (1UL<<30) )
467 {
468 printk("Initial loading isn't allowed to lowest 1GB of memory.\n");
469 return -EINVAL;
470 }
472 /* WARNING: The new domain must have its 'processor' field filled in! */
473 l3start = l3tab = (l3_pgentry_t *)mpt_alloc; mpt_alloc += PAGE_SIZE;
474 l2start = l2tab = (l2_pgentry_t *)mpt_alloc; mpt_alloc += 4*PAGE_SIZE;
475 memcpy(l2tab, idle_pg_table_l2, 4*PAGE_SIZE);
476 for (i = 0; i < 4; i++) {
477 l3tab[i] = l3e_from_paddr((u32)l2tab + i*PAGE_SIZE, L3_PROT);
478 l2tab[(LINEAR_PT_VIRT_START >> L2_PAGETABLE_SHIFT)+i] =
479 l2e_from_paddr((u32)l2tab + i*PAGE_SIZE, __PAGE_HYPERVISOR);
480 }
481 v->arch.guest_table = pagetable_from_paddr((unsigned long)l3start);
483 for ( i = 0; i < PDPT_L2_ENTRIES; i++ )
484 l2tab[l2_linear_offset(PERDOMAIN_VIRT_START) + i] =
485 l2e_from_page(virt_to_page(d->arch.mm_perdomain_pt) + i,
486 __PAGE_HYPERVISOR);
488 l2tab += l2_linear_offset(v_start);
489 mfn = alloc_spfn;
490 for ( count = 0; count < ((v_end-v_start)>>PAGE_SHIFT); count++ )
491 {
492 if ( !((unsigned long)l1tab & (PAGE_SIZE-1)) )
493 {
494 l1start = l1tab = (l1_pgentry_t *)mpt_alloc;
495 mpt_alloc += PAGE_SIZE;
496 *l2tab = l2e_from_paddr((unsigned long)l1start, L2_PROT);
497 l2tab++;
498 clear_page(l1tab);
499 if ( count == 0 )
500 l1tab += l1_table_offset(v_start);
501 }
502 *l1tab = l1e_from_pfn(mfn, L1_PROT);
503 l1tab++;
505 page = mfn_to_page(mfn);
506 if ( !get_page_and_type(page, d, PGT_writable_page) )
507 BUG();
509 mfn++;
510 }
512 /* Pages that are part of page tables must be read only. */
513 l2tab = l2start + l2_linear_offset(vpt_start);
514 l1start = l1tab = (l1_pgentry_t *)(u32)l2e_get_paddr(*l2tab);
515 l1tab += l1_table_offset(vpt_start);
516 for ( count = 0; count < nr_pt_pages; count++ )
517 {
518 page = mfn_to_page(l1e_get_pfn(*l1tab));
519 if ( !opt_dom0_shadow )
520 l1e_remove_flags(*l1tab, _PAGE_RW);
521 else
522 if ( !get_page_type(page, PGT_writable_page) )
523 BUG();
525 switch ( count )
526 {
527 case 0:
528 page->u.inuse.type_info &= ~PGT_type_mask;
529 page->u.inuse.type_info |= PGT_l3_page_table;
530 get_page(page, d); /* an extra ref because of readable mapping */
532 /* Get another ref to L3 page so that it can be pinned. */
533 page->u.inuse.type_info++;
534 page->count_info++;
535 set_bit(_PGT_pinned, &page->u.inuse.type_info);
536 break;
537 case 1 ... 4:
538 page->u.inuse.type_info &= ~PGT_type_mask;
539 page->u.inuse.type_info |= PGT_l2_page_table;
540 if ( count == 4 )
541 page->u.inuse.type_info |= PGT_pae_xen_l2;
542 get_page(page, d); /* an extra ref because of readable mapping */
543 break;
544 default:
545 page->u.inuse.type_info &= ~PGT_type_mask;
546 page->u.inuse.type_info |= PGT_l1_page_table;
547 get_page(page, d); /* an extra ref because of readable mapping */
548 break;
549 }
550 if ( !((unsigned long)++l1tab & (PAGE_SIZE - 1)) )
551 l1start = l1tab = (l1_pgentry_t *)(u32)l2e_get_paddr(*++l2tab);
552 }
554 #elif defined(__x86_64__)
556 /* Overlap with Xen protected area? */
557 if ( !is_pv_32on64_domain(d) ?
558 ((v_start < HYPERVISOR_VIRT_END) &&
559 (v_end > HYPERVISOR_VIRT_START)) :
560 (v_end > HYPERVISOR_COMPAT_VIRT_START(d)) )
561 {
562 printk("DOM0 image overlaps with Xen private area.\n");
563 return -EINVAL;
564 }
566 if ( is_pv_32on64_domain(d) )
567 {
568 v->arch.guest_context.failsafe_callback_cs = FLAT_COMPAT_KERNEL_CS;
569 v->arch.guest_context.event_callback_cs = FLAT_COMPAT_KERNEL_CS;
570 }
572 /* WARNING: The new domain must have its 'processor' field filled in! */
573 if ( !is_pv_32on64_domain(d) )
574 {
575 maddr_to_page(mpt_alloc)->u.inuse.type_info = PGT_l4_page_table;
576 l4start = l4tab = __va(mpt_alloc); mpt_alloc += PAGE_SIZE;
577 }
578 else
579 {
580 page = alloc_domheap_page(NULL, 0);
581 if ( !page )
582 panic("Not enough RAM for domain 0 PML4.\n");
583 l4start = l4tab = page_to_virt(page);
584 }
585 copy_page(l4tab, idle_pg_table);
586 l4tab[0] = l4e_empty(); /* zap trampoline mapping */
587 l4tab[l4_table_offset(LINEAR_PT_VIRT_START)] =
588 l4e_from_paddr(__pa(l4start), __PAGE_HYPERVISOR);
589 l4tab[l4_table_offset(PERDOMAIN_VIRT_START)] =
590 l4e_from_paddr(__pa(d->arch.mm_perdomain_l3), __PAGE_HYPERVISOR);
591 v->arch.guest_table = pagetable_from_paddr(__pa(l4start));
592 if ( is_pv_32on64_domain(d) )
593 {
594 v->arch.guest_table_user = v->arch.guest_table;
595 if ( setup_arg_xlat_area(v, l4start) < 0 )
596 panic("Not enough RAM for domain 0 hypercall argument translation.\n");
597 }
599 l4tab += l4_table_offset(v_start);
600 mfn = alloc_spfn;
601 for ( count = 0; count < ((v_end-v_start)>>PAGE_SHIFT); count++ )
602 {
603 if ( !((unsigned long)l1tab & (PAGE_SIZE-1)) )
604 {
605 maddr_to_page(mpt_alloc)->u.inuse.type_info = PGT_l1_page_table;
606 l1start = l1tab = __va(mpt_alloc); mpt_alloc += PAGE_SIZE;
607 clear_page(l1tab);
608 if ( count == 0 )
609 l1tab += l1_table_offset(v_start);
610 if ( !((unsigned long)l2tab & (PAGE_SIZE-1)) )
611 {
612 maddr_to_page(mpt_alloc)->u.inuse.type_info = PGT_l2_page_table;
613 l2start = l2tab = __va(mpt_alloc); mpt_alloc += PAGE_SIZE;
614 clear_page(l2tab);
615 if ( count == 0 )
616 l2tab += l2_table_offset(v_start);
617 if ( !((unsigned long)l3tab & (PAGE_SIZE-1)) )
618 {
619 maddr_to_page(mpt_alloc)->u.inuse.type_info =
620 PGT_l3_page_table;
621 l3start = l3tab = __va(mpt_alloc); mpt_alloc += PAGE_SIZE;
622 clear_page(l3tab);
623 if ( count == 0 )
624 l3tab += l3_table_offset(v_start);
625 *l4tab = l4e_from_paddr(__pa(l3start), L4_PROT);
626 l4tab++;
627 }
628 *l3tab = l3e_from_paddr(__pa(l2start), L3_PROT);
629 l3tab++;
630 }
631 *l2tab = l2e_from_paddr(__pa(l1start), L2_PROT);
632 l2tab++;
633 }
634 *l1tab = l1e_from_pfn(mfn, (!is_pv_32on64_domain(d) ?
635 L1_PROT : COMPAT_L1_PROT));
636 l1tab++;
638 page = mfn_to_page(mfn);
639 if ( (page->u.inuse.type_info == 0) &&
640 !get_page_and_type(page, d, PGT_writable_page) )
641 BUG();
643 mfn++;
644 }
646 if ( is_pv_32on64_domain(d) )
647 {
648 /* Ensure the first four L3 entries are all populated. */
649 for ( i = 0, l3tab = l3start; i < 4; ++i, ++l3tab )
650 {
651 if ( !l3e_get_intpte(*l3tab) )
652 {
653 maddr_to_page(mpt_alloc)->u.inuse.type_info = PGT_l2_page_table;
654 l2tab = __va(mpt_alloc); mpt_alloc += PAGE_SIZE;
655 clear_page(l2tab);
656 *l3tab = l3e_from_paddr(__pa(l2tab), L3_PROT);
657 }
658 if ( i == 3 )
659 l3e_get_page(*l3tab)->u.inuse.type_info |= PGT_pae_xen_l2;
660 }
661 /* Install read-only guest visible MPT mapping. */
662 l2tab = l3e_to_l2e(l3start[3]);
663 memcpy(&l2tab[COMPAT_L2_PAGETABLE_FIRST_XEN_SLOT(d)],
664 &compat_idle_pg_table_l2[l2_table_offset(HIRO_COMPAT_MPT_VIRT_START)],
665 COMPAT_L2_PAGETABLE_XEN_SLOTS(d) * sizeof(*l2tab));
666 }
668 /* Pages that are part of page tables must be read only. */
669 l4tab = l4start + l4_table_offset(vpt_start);
670 l3start = l3tab = l4e_to_l3e(*l4tab);
671 l3tab += l3_table_offset(vpt_start);
672 l2start = l2tab = l3e_to_l2e(*l3tab);
673 l2tab += l2_table_offset(vpt_start);
674 l1start = l1tab = l2e_to_l1e(*l2tab);
675 l1tab += l1_table_offset(vpt_start);
676 for ( count = 0; count < nr_pt_pages; count++ )
677 {
678 l1e_remove_flags(*l1tab, _PAGE_RW);
679 page = mfn_to_page(l1e_get_pfn(*l1tab));
681 /* Read-only mapping + PGC_allocated + page-table page. */
682 page->count_info = PGC_allocated | 3;
683 page->u.inuse.type_info |= PGT_validated | 1;
685 /* Top-level p.t. is pinned. */
686 if ( (page->u.inuse.type_info & PGT_type_mask) ==
687 (!is_pv_32on64_domain(d) ?
688 PGT_l4_page_table : PGT_l3_page_table) )
689 {
690 page->count_info += 1;
691 page->u.inuse.type_info += 1 | PGT_pinned;
692 }
694 /* Iterate. */
695 if ( !((unsigned long)++l1tab & (PAGE_SIZE - 1)) )
696 {
697 if ( !((unsigned long)++l2tab & (PAGE_SIZE - 1)) )
698 {
699 if ( !((unsigned long)++l3tab & (PAGE_SIZE - 1)) )
700 l3start = l3tab = l4e_to_l3e(*++l4tab);
701 l2start = l2tab = l3e_to_l2e(*l3tab);
702 }
703 l1start = l1tab = l2e_to_l1e(*l2tab);
704 }
705 }
707 #endif /* __x86_64__ */
709 /* Mask all upcalls... */
710 for ( i = 0; i < MAX_VIRT_CPUS; i++ )
711 shared_info(d, vcpu_info[i].evtchn_upcall_mask) = 1;
713 if ( opt_dom0_max_vcpus == 0 )
714 opt_dom0_max_vcpus = num_online_cpus();
715 if ( opt_dom0_max_vcpus > num_online_cpus() )
716 opt_dom0_max_vcpus = num_online_cpus();
717 if ( opt_dom0_max_vcpus > MAX_VIRT_CPUS )
718 opt_dom0_max_vcpus = MAX_VIRT_CPUS;
719 if ( opt_dom0_max_vcpus > BITS_PER_GUEST_LONG(d) )
720 opt_dom0_max_vcpus = BITS_PER_GUEST_LONG(d);
721 printk("Dom0 has maximum %u VCPUs\n", opt_dom0_max_vcpus);
723 for ( i = 1; i < opt_dom0_max_vcpus; i++ )
724 (void)alloc_vcpu(d, i, i);
726 /* Set up CR3 value for write_ptbase */
727 if ( paging_mode_enabled(v->domain) )
728 paging_update_paging_modes(v);
729 else
730 update_cr3(v);
732 /* Install the new page tables. */
733 local_irq_disable();
734 write_ptbase(v);
736 /* Copy the OS image and free temporary buffer. */
737 elf.dest = (void*)vkern_start;
738 elf_load_binary(&elf);
740 if ( UNSET_ADDR != parms.virt_hypercall )
741 {
742 if ( (parms.virt_hypercall < v_start) ||
743 (parms.virt_hypercall >= v_end) )
744 {
745 write_ptbase(current);
746 local_irq_enable();
747 printk("Invalid HYPERCALL_PAGE field in ELF notes.\n");
748 return -1;
749 }
750 hypercall_page_initialise(d, (void *)(unsigned long)parms.virt_hypercall);
751 }
753 /* Copy the initial ramdisk. */
754 if ( initrd_len != 0 )
755 memcpy((void *)vinitrd_start, initrd_start, initrd_len);
757 /* Free temporary buffers. */
758 discard_initial_images();
760 /* Set up start info area. */
761 si = (start_info_t *)vstartinfo_start;
762 clear_page(si);
763 si->nr_pages = nr_pages;
765 si->shared_info = virt_to_maddr(d->shared_info);
767 si->flags = SIF_PRIVILEGED | SIF_INITDOMAIN;
768 si->pt_base = vpt_start + 2 * PAGE_SIZE * !!is_pv_32on64_domain(d);
769 si->nr_pt_frames = nr_pt_pages;
770 si->mfn_list = vphysmap_start;
771 snprintf(si->magic, sizeof(si->magic), "xen-3.0-x86_%d%s",
772 elf_64bit(&elf) ? 64 : 32, parms.pae ? "p" : "");
774 /* Write the phys->machine and machine->phys table entries. */
775 for ( pfn = 0; pfn < d->tot_pages; pfn++ )
776 {
777 mfn = pfn + alloc_spfn;
778 #ifndef NDEBUG
779 #define REVERSE_START ((v_end - v_start) >> PAGE_SHIFT)
780 if ( pfn > REVERSE_START )
781 mfn = alloc_epfn - (pfn - REVERSE_START);
782 #endif
783 if ( !is_pv_32on64_domain(d) )
784 ((unsigned long *)vphysmap_start)[pfn] = mfn;
785 else
786 ((unsigned int *)vphysmap_start)[pfn] = mfn;
787 set_gpfn_from_mfn(mfn, pfn);
788 }
789 while ( pfn < nr_pages )
790 {
791 if ( (page = alloc_chunk(d, nr_pages - d->tot_pages)) == NULL )
792 panic("Not enough RAM for DOM0 reservation.\n");
793 while ( pfn < d->tot_pages )
794 {
795 mfn = page_to_mfn(page);
796 #ifndef NDEBUG
797 #define pfn (nr_pages - 1 - (pfn - (alloc_epfn - alloc_spfn)))
798 #endif
799 if ( !is_pv_32on64_domain(d) )
800 ((unsigned long *)vphysmap_start)[pfn] = mfn;
801 else
802 ((unsigned int *)vphysmap_start)[pfn] = mfn;
803 set_gpfn_from_mfn(mfn, pfn);
804 #undef pfn
805 page++; pfn++;
806 }
807 }
809 if ( initrd_len != 0 )
810 {
811 si->mod_start = vinitrd_start;
812 si->mod_len = initrd_len;
813 printk("Initrd len 0x%lx, start at 0x%lx\n",
814 si->mod_len, si->mod_start);
815 }
817 memset(si->cmd_line, 0, sizeof(si->cmd_line));
818 if ( cmdline != NULL )
819 strlcpy((char *)si->cmd_line, cmdline, sizeof(si->cmd_line));
821 if ( fill_console_start_info((void *)(si + 1)) )
822 {
823 si->console.dom0.info_off = sizeof(struct start_info);
824 si->console.dom0.info_size = sizeof(struct dom0_vga_console_info);
825 }
827 #if defined(__x86_64__)
828 if ( is_pv_32on64_domain(d) )
829 xlat_start_info(si, XLAT_start_info_console_dom0);
830 #endif
832 /* Reinstate the caller's page tables. */
833 write_ptbase(current);
834 local_irq_enable();
836 #if defined(__i386__)
837 /* Destroy low mappings - they were only for our convenience. */
838 zap_low_mappings(l2start);
839 #endif
841 update_domain_wallclock_time(d);
843 v->is_initialised = 1;
844 clear_bit(_VPF_down, &v->pause_flags);
846 /*
847 * Initial register values:
848 * DS,ES,FS,GS = FLAT_KERNEL_DS
849 * CS:EIP = FLAT_KERNEL_CS:start_pc
850 * SS:ESP = FLAT_KERNEL_SS:start_stack
851 * ESI = start_info
852 * [EAX,EBX,ECX,EDX,EDI,EBP are zero]
853 */
854 regs = &v->arch.guest_context.user_regs;
855 regs->ds = regs->es = regs->fs = regs->gs =
856 !is_pv_32on64_domain(d) ? FLAT_KERNEL_DS : FLAT_COMPAT_KERNEL_DS;
857 regs->ss = (!is_pv_32on64_domain(d) ?
858 FLAT_KERNEL_SS : FLAT_COMPAT_KERNEL_SS);
859 regs->cs = (!is_pv_32on64_domain(d) ?
860 FLAT_KERNEL_CS : FLAT_COMPAT_KERNEL_CS);
861 regs->eip = parms.virt_entry;
862 regs->esp = vstack_end;
863 regs->esi = vstartinfo_start;
864 regs->eflags = X86_EFLAGS_IF;
866 if ( opt_dom0_shadow )
867 if ( paging_enable(d, PG_SH_enable) == 0 )
868 paging_update_paging_modes(v);
870 if ( supervisor_mode_kernel )
871 {
872 v->arch.guest_context.kernel_ss &= ~3;
873 v->arch.guest_context.user_regs.ss &= ~3;
874 v->arch.guest_context.user_regs.es &= ~3;
875 v->arch.guest_context.user_regs.ds &= ~3;
876 v->arch.guest_context.user_regs.fs &= ~3;
877 v->arch.guest_context.user_regs.gs &= ~3;
878 printk("Dom0 runs in ring 0 (supervisor mode)\n");
879 if ( !test_bit(XENFEAT_supervisor_mode_kernel,
880 parms.f_supported) )
881 panic("Dom0 does not support supervisor-mode execution\n");
882 }
883 else
884 {
885 if ( test_bit(XENFEAT_supervisor_mode_kernel, parms.f_required) )
886 panic("Dom0 requires supervisor-mode execution\n");
887 }
889 rc = 0;
891 /* DOM0 is permitted full I/O capabilities. */
892 rc |= ioports_permit_access(dom0, 0, 0xFFFF);
893 rc |= iomem_permit_access(dom0, 0UL, ~0UL);
894 rc |= irqs_permit_access(dom0, 0, NR_IRQS-1);
896 /*
897 * Modify I/O port access permissions.
898 */
899 /* Master Interrupt Controller (PIC). */
900 rc |= ioports_deny_access(dom0, 0x20, 0x21);
901 /* Slave Interrupt Controller (PIC). */
902 rc |= ioports_deny_access(dom0, 0xA0, 0xA1);
903 /* Interval Timer (PIT). */
904 rc |= ioports_deny_access(dom0, 0x40, 0x43);
905 /* PIT Channel 2 / PC Speaker Control. */
906 rc |= ioports_deny_access(dom0, 0x61, 0x61);
907 /* PCI configuration space (NB. 0xcf8 has special treatment). */
908 rc |= ioports_deny_access(dom0, 0xcfc, 0xcff);
909 /* Command-line I/O ranges. */
910 process_dom0_ioports_disable();
912 /*
913 * Modify I/O memory access permissions.
914 */
915 /* Local APIC. */
916 if ( mp_lapic_addr != 0 )
917 {
918 mfn = paddr_to_pfn(mp_lapic_addr);
919 rc |= iomem_deny_access(dom0, mfn, mfn);
920 }
921 /* I/O APICs. */
922 for ( i = 0; i < nr_ioapics; i++ )
923 {
924 mfn = paddr_to_pfn(mp_ioapics[i].mpc_apicaddr);
925 if ( smp_found_config )
926 rc |= iomem_deny_access(dom0, mfn, mfn);
927 }
929 /* Remove access to E820_UNUSABLE I/O regions above 1MB. */
930 for ( i = 0; i < e820.nr_map; i++ )
931 {
932 unsigned long sfn, efn;
933 sfn = max_t(unsigned long, paddr_to_pfn(e820.map[i].addr), 0x100ul);
934 efn = paddr_to_pfn(e820.map[i].addr + e820.map[i].size - 1);
935 if ( (e820.map[i].type == E820_UNUSABLE) &&
936 (e820.map[i].size != 0) &&
937 (sfn <= efn) )
938 rc |= iomem_deny_access(dom0, sfn, efn);
939 }
941 BUG_ON(rc != 0);
943 return 0;
944 }
946 /*
947 * Local variables:
948 * mode: C
949 * c-set-style: "BSD"
950 * c-basic-offset: 4
951 * tab-width: 4
952 * indent-tabs-mode: nil
953 * End:
954 */