2 * Machine specific setup for xen
4 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
7 #include <linux/module.h>
8 #include <linux/sched.h>
11 #include <linux/memblock.h>
12 #include <linux/cpuidle.h>
13 #include <linux/cpufreq.h>
18 #include <asm/setup.h>
21 #include <asm/xen/hypervisor.h>
22 #include <asm/xen/hypercall.h>
26 #include <xen/interface/callback.h>
27 #include <xen/interface/memory.h>
28 #include <xen/interface/physdev.h>
29 #include <xen/features.h>
35 /* Amount of extra memory space we add to the e820 ranges */
36 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
38 /* Number of pages released from the initial allocation. */
39 unsigned long xen_released_pages;
41 /* E820 map used during setting up memory. */
42 static struct e820entry xen_e820_map[E820MAX] __initdata;
43 static u32 xen_e820_map_entries __initdata;
46 * Buffer used to remap identity mapped pages. We only need the virtual space.
47 * The physical page behind this address is remapped as needed to different
50 #define REMAP_SIZE (P2M_PER_PAGE - 3)
52 unsigned long next_area_mfn;
53 unsigned long target_pfn;
55 unsigned long mfns[REMAP_SIZE];
56 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
57 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
60 * The maximum amount of extra memory compared to the base size. The
61 * main scaling factor is the size of struct page. At extreme ratios
62 * of base:extra, all the base memory can be filled with page
63 * structures for the extra memory, leaving no space for anything
66 * 10x seems like a reasonable balance between scaling flexibility and
67 * leaving a practically usable system.
69 #define EXTRA_MEM_RATIO (10)
71 static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
75 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
77 if (xen_extra_mem[i].size == 0) {
78 xen_extra_mem[i].start = start;
79 xen_extra_mem[i].size = size;
82 /* Append to existing region. */
83 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
84 xen_extra_mem[i].size += size;
88 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
89 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
91 memblock_reserve(start, size);
94 static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
97 phys_addr_t start_r, size_r;
99 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
100 start_r = xen_extra_mem[i].start;
101 size_r = xen_extra_mem[i].size;
103 /* Start of region. */
104 if (start_r == start) {
105 BUG_ON(size > size_r);
106 xen_extra_mem[i].start += size;
107 xen_extra_mem[i].size -= size;
111 if (start_r + size_r == start + size) {
112 BUG_ON(size > size_r);
113 xen_extra_mem[i].size -= size;
117 if (start > start_r && start < start_r + size_r) {
118 BUG_ON(start + size > start_r + size_r);
119 xen_extra_mem[i].size = start - start_r;
120 /* Calling memblock_reserve() again is okay. */
121 xen_add_extra_mem(start + size, start_r + size_r -
126 memblock_free(start, size);
130 * Called during boot before the p2m list can take entries beyond the
131 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
134 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
137 phys_addr_t addr = PFN_PHYS(pfn);
139 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
140 if (addr >= xen_extra_mem[i].start &&
141 addr < xen_extra_mem[i].start + xen_extra_mem[i].size)
142 return INVALID_P2M_ENTRY;
145 return IDENTITY_FRAME(pfn);
149 * Mark all pfns of extra mem as invalid in p2m list.
151 void __init xen_inv_extra_mem(void)
153 unsigned long pfn, pfn_s, pfn_e;
156 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
157 if (!xen_extra_mem[i].size)
159 pfn_s = PFN_DOWN(xen_extra_mem[i].start);
160 pfn_e = PFN_UP(xen_extra_mem[i].start + xen_extra_mem[i].size);
161 for (pfn = pfn_s; pfn < pfn_e; pfn++)
162 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
167 * Finds the next RAM pfn available in the E820 map after min_pfn.
168 * This function updates min_pfn with the pfn found and returns
169 * the size of that range or zero if not found.
171 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
173 const struct e820entry *entry = xen_e820_map;
175 unsigned long done = 0;
177 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
181 if (entry->type != E820_RAM)
184 e_pfn = PFN_DOWN(entry->addr + entry->size);
186 /* We only care about E820 after this */
187 if (e_pfn < *min_pfn)
190 s_pfn = PFN_UP(entry->addr);
192 /* If min_pfn falls within the E820 entry, we want to start
193 * at the min_pfn PFN.
195 if (s_pfn <= *min_pfn) {
196 done = e_pfn - *min_pfn;
198 done = e_pfn - s_pfn;
207 static int __init xen_free_mfn(unsigned long mfn)
209 struct xen_memory_reservation reservation = {
215 set_xen_guest_handle(reservation.extent_start, &mfn);
216 reservation.nr_extents = 1;
218 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
222 * This releases a chunk of memory and then does the identity map. It's used
223 * as a fallback if the remapping fails.
225 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
226 unsigned long end_pfn, unsigned long nr_pages)
228 unsigned long pfn, end;
231 WARN_ON(start_pfn > end_pfn);
233 /* Release pages first. */
234 end = min(end_pfn, nr_pages);
235 for (pfn = start_pfn; pfn < end; pfn++) {
236 unsigned long mfn = pfn_to_mfn(pfn);
238 /* Make sure pfn exists to start with */
239 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
242 ret = xen_free_mfn(mfn);
243 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
246 xen_released_pages++;
247 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
253 set_phys_range_identity(start_pfn, end_pfn);
257 * Helper function to update the p2m and m2p tables and kernel mapping.
259 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
261 struct mmu_update update = {
262 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
267 if (!set_phys_to_machine(pfn, mfn)) {
268 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
274 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
275 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
280 /* Update kernel mapping, but not for highmem. */
281 if (pfn >= PFN_UP(__pa(high_memory - 1)))
284 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
285 mfn_pte(mfn, PAGE_KERNEL), 0)) {
286 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
293 * This function updates the p2m and m2p tables with an identity map from
294 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
295 * original allocation at remap_pfn. The information needed for remapping is
296 * saved in the memory itself to avoid the need for allocating buffers. The
297 * complete remap information is contained in a list of MFNs each containing
298 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
299 * This enables us to preserve the original mfn sequence while doing the
300 * remapping at a time when the memory management is capable of allocating
301 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
304 static void __init xen_do_set_identity_and_remap_chunk(
305 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
307 unsigned long buf = (unsigned long)&xen_remap_buf;
308 unsigned long mfn_save, mfn;
309 unsigned long ident_pfn_iter, remap_pfn_iter;
310 unsigned long ident_end_pfn = start_pfn + size;
311 unsigned long left = size;
312 unsigned int i, chunk;
316 BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
318 mfn_save = virt_to_mfn(buf);
320 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
321 ident_pfn_iter < ident_end_pfn;
322 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
323 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
325 /* Map first pfn to xen_remap_buf */
326 mfn = pfn_to_mfn(ident_pfn_iter);
327 set_pte_mfn(buf, mfn, PAGE_KERNEL);
329 /* Save mapping information in page */
330 xen_remap_buf.next_area_mfn = xen_remap_mfn;
331 xen_remap_buf.target_pfn = remap_pfn_iter;
332 xen_remap_buf.size = chunk;
333 for (i = 0; i < chunk; i++)
334 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
336 /* Put remap buf into list. */
339 /* Set identity map */
340 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
345 /* Restore old xen_remap_buf mapping */
346 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
350 * This function takes a contiguous pfn range that needs to be identity mapped
353 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
354 * 2) Calls the do_ function to actually do the mapping/remapping work.
356 * The goal is to not allocate additional memory but to remap the existing
357 * pages. In the case of an error the underlying memory is simply released back
358 * to Xen and not remapped.
360 static unsigned long __init xen_set_identity_and_remap_chunk(
361 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
362 unsigned long remap_pfn)
366 unsigned long n = end_pfn - start_pfn;
369 unsigned long cur_pfn = start_pfn + i;
370 unsigned long left = n - i;
371 unsigned long size = left;
372 unsigned long remap_range_size;
374 /* Do not remap pages beyond the current allocation */
375 if (cur_pfn >= nr_pages) {
376 /* Identity map remaining pages */
377 set_phys_range_identity(cur_pfn, cur_pfn + size);
380 if (cur_pfn + size > nr_pages)
381 size = nr_pages - cur_pfn;
383 remap_range_size = xen_find_pfn_range(&remap_pfn);
384 if (!remap_range_size) {
385 pr_warning("Unable to find available pfn range, not remapping identity pages\n");
386 xen_set_identity_and_release_chunk(cur_pfn,
387 cur_pfn + left, nr_pages);
390 /* Adjust size to fit in current e820 RAM region */
391 if (size > remap_range_size)
392 size = remap_range_size;
394 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
396 /* Update variables to reflect new mappings. */
402 * If the PFNs are currently mapped, the VA mapping also needs
403 * to be updated to be 1:1.
405 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
406 (void)HYPERVISOR_update_va_mapping(
407 (unsigned long)__va(pfn << PAGE_SHIFT),
408 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
413 static void __init xen_set_identity_and_remap(unsigned long nr_pages)
415 phys_addr_t start = 0;
416 unsigned long last_pfn = nr_pages;
417 const struct e820entry *entry = xen_e820_map;
421 * Combine non-RAM regions and gaps until a RAM region (or the
422 * end of the map) is reached, then set the 1:1 map and
423 * remap the memory in those non-RAM regions.
425 * The combined non-RAM regions are rounded to a whole number
426 * of pages so any partial pages are accessible via the 1:1
427 * mapping. This is needed for some BIOSes that put (for
428 * example) the DMI tables in a reserved region that begins on
429 * a non-page boundary.
431 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
432 phys_addr_t end = entry->addr + entry->size;
433 if (entry->type == E820_RAM || i == xen_e820_map_entries - 1) {
434 unsigned long start_pfn = PFN_DOWN(start);
435 unsigned long end_pfn = PFN_UP(end);
437 if (entry->type == E820_RAM)
438 end_pfn = PFN_UP(entry->addr);
440 if (start_pfn < end_pfn)
441 last_pfn = xen_set_identity_and_remap_chunk(
442 start_pfn, end_pfn, nr_pages,
448 pr_info("Released %ld page(s)\n", xen_released_pages);
452 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
453 * The remap information (which mfn remap to which pfn) is contained in the
454 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
455 * This scheme allows to remap the different chunks in arbitrary order while
456 * the resulting mapping will be independant from the order.
458 void __init xen_remap_memory(void)
460 unsigned long buf = (unsigned long)&xen_remap_buf;
461 unsigned long mfn_save, mfn, pfn;
462 unsigned long remapped = 0;
464 unsigned long pfn_s = ~0UL;
465 unsigned long len = 0;
467 mfn_save = virt_to_mfn(buf);
469 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
470 /* Map the remap information */
471 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
473 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
475 pfn = xen_remap_buf.target_pfn;
476 for (i = 0; i < xen_remap_buf.size; i++) {
477 mfn = xen_remap_buf.mfns[i];
478 xen_update_mem_tables(pfn, mfn);
482 if (pfn_s == ~0UL || pfn == pfn_s) {
483 pfn_s = xen_remap_buf.target_pfn;
484 len += xen_remap_buf.size;
485 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
486 len += xen_remap_buf.size;
488 xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
489 pfn_s = xen_remap_buf.target_pfn;
490 len = xen_remap_buf.size;
494 xen_remap_mfn = xen_remap_buf.next_area_mfn;
497 if (pfn_s != ~0UL && len)
498 xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
500 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
502 pr_info("Remapped %ld page(s)\n", remapped);
505 static unsigned long __init xen_get_max_pages(void)
507 unsigned long max_pages = MAX_DOMAIN_PAGES;
508 domid_t domid = DOMID_SELF;
512 * For the initial domain we use the maximum reservation as
515 * For guest domains the current maximum reservation reflects
516 * the current maximum rather than the static maximum. In this
517 * case the e820 map provided to us will cover the static
520 if (xen_initial_domain()) {
521 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
526 return min(max_pages, MAX_DOMAIN_PAGES);
529 static void __init xen_align_and_add_e820_region(phys_addr_t start,
530 phys_addr_t size, int type)
532 phys_addr_t end = start + size;
534 /* Align RAM regions to page boundaries. */
535 if (type == E820_RAM) {
536 start = PAGE_ALIGN(start);
537 end &= ~((phys_addr_t)PAGE_SIZE - 1);
540 e820_add_region(start, end - start, type);
543 static void __init xen_ignore_unusable(void)
545 struct e820entry *entry = xen_e820_map;
548 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
549 if (entry->type == E820_UNUSABLE)
550 entry->type = E820_RAM;
554 static unsigned long __init xen_count_remap_pages(unsigned long max_pfn)
556 unsigned long extra = 0;
557 const struct e820entry *entry = xen_e820_map;
560 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
561 unsigned long start_pfn = PFN_DOWN(entry->addr);
562 unsigned long end_pfn = PFN_UP(entry->addr + entry->size);
564 if (start_pfn >= max_pfn)
566 if (entry->type == E820_RAM)
568 if (end_pfn >= max_pfn)
570 extra += end_pfn - start_pfn;
576 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
578 struct e820entry *entry;
586 entry = xen_e820_map;
588 for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++) {
589 if (entry->type == E820_RAM && entry->addr <= start &&
590 (entry->addr + entry->size) >= end)
600 * Find a free area in physical memory not yet reserved and compliant with
602 * Used to relocate pre-allocated areas like initrd or p2m list which are in
603 * conflict with the to be used E820 map.
604 * In case no area is found, return 0. Otherwise return the physical address
605 * of the area which is already reserved for convenience.
607 phys_addr_t __init xen_find_free_area(phys_addr_t size)
610 phys_addr_t addr, start;
611 struct e820entry *entry = xen_e820_map;
613 for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++, entry++) {
614 if (entry->type != E820_RAM || entry->size < size)
617 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
618 if (!memblock_is_reserved(addr))
620 start = addr + PAGE_SIZE;
621 if (start + size > entry->addr + entry->size)
624 if (addr >= start + size) {
625 memblock_reserve(start, size);
634 * Reserve Xen mfn_list.
635 * See comment above "struct start_info" in <xen/interface/xen.h>
636 * We tried to make the the memblock_reserve more selective so
637 * that it would be clear what region is reserved. Sadly we ran
638 * in the problem wherein on a 64-bit hypervisor with a 32-bit
639 * initial domain, the pt_base has the cr3 value which is not
640 * neccessarily where the pagetable starts! As Jan put it: "
641 * Actually, the adjustment turns out to be correct: The page
642 * tables for a 32-on-64 dom0 get allocated in the order "first L1",
643 * "first L2", "first L3", so the offset to the page table base is
644 * indeed 2. When reading xen/include/public/xen.h's comment
645 * very strictly, this is not a violation (since there nothing is said
646 * that the first thing in the page table space is pointed to by
647 * pt_base; I admit that this seems to be implied though, namely
648 * do I think that it is implied that the page table space is the
649 * range [pt_base, pt_base + nt_pt_frames), whereas that
650 * range here indeed is [pt_base - 2, pt_base - 2 + nt_pt_frames),
651 * which - without a priori knowledge - the kernel would have
652 * difficulty to figure out)." - so lets just fall back to the
653 * easy way and reserve the whole region.
655 static void __init xen_reserve_xen_mfnlist(void)
657 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
658 memblock_reserve(__pa(xen_start_info->mfn_list),
659 xen_start_info->pt_base -
660 xen_start_info->mfn_list);
664 memblock_reserve(PFN_PHYS(xen_start_info->first_p2m_pfn),
665 PFN_PHYS(xen_start_info->nr_p2m_frames));
669 * machine_specific_memory_setup - Hook for machine specific memory setup.
671 char * __init xen_memory_setup(void)
673 unsigned long max_pfn = xen_start_info->nr_pages;
674 phys_addr_t mem_end, addr, size, chunk_size;
677 struct xen_memory_map memmap;
678 unsigned long max_pages;
679 unsigned long extra_pages = 0;
683 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
684 mem_end = PFN_PHYS(max_pfn);
686 memmap.nr_entries = E820MAX;
687 set_xen_guest_handle(memmap.buffer, xen_e820_map);
689 op = xen_initial_domain() ?
690 XENMEM_machine_memory_map :
692 rc = HYPERVISOR_memory_op(op, &memmap);
694 BUG_ON(xen_initial_domain());
695 memmap.nr_entries = 1;
696 xen_e820_map[0].addr = 0ULL;
697 xen_e820_map[0].size = mem_end;
698 /* 8MB slack (to balance backend allocations). */
699 xen_e820_map[0].size += 8ULL << 20;
700 xen_e820_map[0].type = E820_RAM;
704 BUG_ON(memmap.nr_entries == 0);
705 xen_e820_map_entries = memmap.nr_entries;
708 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
709 * regions, so if we're using the machine memory map leave the
710 * region as RAM as it is in the pseudo-physical map.
712 * UNUSABLE regions in domUs are not handled and will need
713 * a patch in the future.
715 if (xen_initial_domain())
716 xen_ignore_unusable();
718 /* Make sure the Xen-supplied memory map is well-ordered. */
719 sanitize_e820_map(xen_e820_map, xen_e820_map_entries,
720 &xen_e820_map_entries);
722 max_pages = xen_get_max_pages();
723 if (max_pages > max_pfn)
724 extra_pages += max_pages - max_pfn;
726 /* How many extra pages do we need due to remapping? */
727 extra_pages += xen_count_remap_pages(max_pfn);
730 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
731 * factor the base size. On non-highmem systems, the base
732 * size is the full initial memory allocation; on highmem it
733 * is limited to the max size of lowmem, so that it doesn't
734 * get completely filled.
736 * In principle there could be a problem in lowmem systems if
737 * the initial memory is also very large with respect to
738 * lowmem, but we won't try to deal with that here.
740 extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
743 addr = xen_e820_map[0].addr;
744 size = xen_e820_map[0].size;
745 while (i < xen_e820_map_entries) {
747 type = xen_e820_map[i].type;
749 if (type == E820_RAM) {
750 if (addr < mem_end) {
751 chunk_size = min(size, mem_end - addr);
752 } else if (extra_pages) {
753 chunk_size = min(size, PFN_PHYS(extra_pages));
754 extra_pages -= PFN_DOWN(chunk_size);
755 xen_add_extra_mem(addr, chunk_size);
756 xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
758 type = E820_UNUSABLE;
761 xen_align_and_add_e820_region(addr, chunk_size, type);
767 if (i < xen_e820_map_entries) {
768 addr = xen_e820_map[i].addr;
769 size = xen_e820_map[i].size;
775 * Set the rest as identity mapped, in case PCI BARs are
778 * PFNs above MAX_P2M_PFN are considered identity mapped as
781 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
784 * In domU, the ISA region is normal, usable memory, but we
785 * reserve ISA memory anyway because too many things poke
788 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
791 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
793 xen_reserve_xen_mfnlist();
796 * Set identity map on non-RAM pages and prepare remapping the
799 xen_set_identity_and_remap(max_pfn);
805 * Machine specific memory setup for auto-translated guests.
807 char * __init xen_auto_xlated_memory_setup(void)
809 struct xen_memory_map memmap;
813 memmap.nr_entries = E820MAX;
814 set_xen_guest_handle(memmap.buffer, xen_e820_map);
816 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
818 panic("No memory map (%d)\n", rc);
820 xen_e820_map_entries = memmap.nr_entries;
822 sanitize_e820_map(xen_e820_map, ARRAY_SIZE(xen_e820_map),
823 &xen_e820_map_entries);
825 for (i = 0; i < xen_e820_map_entries; i++)
826 e820_add_region(xen_e820_map[i].addr, xen_e820_map[i].size,
827 xen_e820_map[i].type);
829 xen_reserve_xen_mfnlist();
835 * Set the bit indicating "nosegneg" library variants should be used.
836 * We only need to bother in pure 32-bit mode; compat 32-bit processes
837 * can have un-truncated segments, so wrapping around is allowed.
839 static void __init fiddle_vdso(void)
843 * This could be called before selected_vdso32 is initialized, so
844 * just fiddle with both possible images. vdso_image_32_syscall
845 * can't be selected, since it only exists on 64-bit systems.
848 mask = vdso_image_32_int80.data +
849 vdso_image_32_int80.sym_VDSO32_NOTE_MASK;
850 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
851 mask = vdso_image_32_sysenter.data +
852 vdso_image_32_sysenter.sym_VDSO32_NOTE_MASK;
853 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
857 static int register_callback(unsigned type, const void *func)
859 struct callback_register callback = {
861 .address = XEN_CALLBACK(__KERNEL_CS, func),
862 .flags = CALLBACKF_mask_events,
865 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
868 void xen_enable_sysenter(void)
871 unsigned sysenter_feature;
874 sysenter_feature = X86_FEATURE_SEP;
876 sysenter_feature = X86_FEATURE_SYSENTER32;
879 if (!boot_cpu_has(sysenter_feature))
882 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
884 setup_clear_cpu_cap(sysenter_feature);
887 void xen_enable_syscall(void)
892 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
894 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
895 /* Pretty fatal; 64-bit userspace has no other
896 mechanism for syscalls. */
899 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
900 ret = register_callback(CALLBACKTYPE_syscall32,
901 xen_syscall32_target);
903 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
905 #endif /* CONFIG_X86_64 */
908 void __init xen_pvmmu_arch_setup(void)
910 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
911 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
913 HYPERVISOR_vm_assist(VMASST_CMD_enable,
914 VMASST_TYPE_pae_extended_cr3);
916 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
917 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
920 xen_enable_sysenter();
921 xen_enable_syscall();
924 /* This function is not called for HVM domains */
925 void __init xen_arch_setup(void)
927 xen_panic_handler_init();
928 if (!xen_feature(XENFEAT_auto_translated_physmap))
929 xen_pvmmu_arch_setup();
932 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
933 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
938 memcpy(boot_command_line, xen_start_info->cmd_line,
939 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
940 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
942 /* Set up idle, making sure it calls safe_halt() pvop */
945 WARN_ON(xen_set_default_idle());