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>
30 #include <xen/hvc-console.h>
35 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
37 /* Amount of extra memory space we add to the e820 ranges */
38 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
40 /* Number of pages released from the initial allocation. */
41 unsigned long xen_released_pages;
43 /* E820 map used during setting up memory. */
44 static struct e820entry xen_e820_map[E820MAX] __initdata;
45 static u32 xen_e820_map_entries __initdata;
48 * Buffer used to remap identity mapped pages. We only need the virtual space.
49 * The physical page behind this address is remapped as needed to different
52 #define REMAP_SIZE (P2M_PER_PAGE - 3)
54 unsigned long next_area_mfn;
55 unsigned long target_pfn;
57 unsigned long mfns[REMAP_SIZE];
58 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
59 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
62 * The maximum amount of extra memory compared to the base size. The
63 * main scaling factor is the size of struct page. At extreme ratios
64 * of base:extra, all the base memory can be filled with page
65 * structures for the extra memory, leaving no space for anything
68 * 10x seems like a reasonable balance between scaling flexibility and
69 * leaving a practically usable system.
71 #define EXTRA_MEM_RATIO (10)
73 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
75 static void __init xen_parse_512gb(void)
80 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
84 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
87 else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
90 xen_512gb_limit = val;
93 static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
97 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
99 if (xen_extra_mem[i].size == 0) {
100 xen_extra_mem[i].start = start;
101 xen_extra_mem[i].size = size;
104 /* Append to existing region. */
105 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
106 xen_extra_mem[i].size += size;
110 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
111 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
113 memblock_reserve(start, size);
116 static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
119 phys_addr_t start_r, size_r;
121 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
122 start_r = xen_extra_mem[i].start;
123 size_r = xen_extra_mem[i].size;
125 /* Start of region. */
126 if (start_r == start) {
127 BUG_ON(size > size_r);
128 xen_extra_mem[i].start += size;
129 xen_extra_mem[i].size -= size;
133 if (start_r + size_r == start + size) {
134 BUG_ON(size > size_r);
135 xen_extra_mem[i].size -= size;
139 if (start > start_r && start < start_r + size_r) {
140 BUG_ON(start + size > start_r + size_r);
141 xen_extra_mem[i].size = start - start_r;
142 /* Calling memblock_reserve() again is okay. */
143 xen_add_extra_mem(start + size, start_r + size_r -
148 memblock_free(start, size);
152 * Called during boot before the p2m list can take entries beyond the
153 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
156 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
159 phys_addr_t addr = PFN_PHYS(pfn);
161 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
162 if (addr >= xen_extra_mem[i].start &&
163 addr < xen_extra_mem[i].start + xen_extra_mem[i].size)
164 return INVALID_P2M_ENTRY;
167 return IDENTITY_FRAME(pfn);
171 * Mark all pfns of extra mem as invalid in p2m list.
173 void __init xen_inv_extra_mem(void)
175 unsigned long pfn, pfn_s, pfn_e;
178 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
179 if (!xen_extra_mem[i].size)
181 pfn_s = PFN_DOWN(xen_extra_mem[i].start);
182 pfn_e = PFN_UP(xen_extra_mem[i].start + xen_extra_mem[i].size);
183 for (pfn = pfn_s; pfn < pfn_e; pfn++)
184 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
189 * Finds the next RAM pfn available in the E820 map after min_pfn.
190 * This function updates min_pfn with the pfn found and returns
191 * the size of that range or zero if not found.
193 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
195 const struct e820entry *entry = xen_e820_map;
197 unsigned long done = 0;
199 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
203 if (entry->type != E820_RAM)
206 e_pfn = PFN_DOWN(entry->addr + entry->size);
208 /* We only care about E820 after this */
209 if (e_pfn < *min_pfn)
212 s_pfn = PFN_UP(entry->addr);
214 /* If min_pfn falls within the E820 entry, we want to start
215 * at the min_pfn PFN.
217 if (s_pfn <= *min_pfn) {
218 done = e_pfn - *min_pfn;
220 done = e_pfn - s_pfn;
229 static int __init xen_free_mfn(unsigned long mfn)
231 struct xen_memory_reservation reservation = {
237 set_xen_guest_handle(reservation.extent_start, &mfn);
238 reservation.nr_extents = 1;
240 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
244 * This releases a chunk of memory and then does the identity map. It's used
245 * as a fallback if the remapping fails.
247 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
248 unsigned long end_pfn, unsigned long nr_pages)
250 unsigned long pfn, end;
253 WARN_ON(start_pfn > end_pfn);
255 /* Release pages first. */
256 end = min(end_pfn, nr_pages);
257 for (pfn = start_pfn; pfn < end; pfn++) {
258 unsigned long mfn = pfn_to_mfn(pfn);
260 /* Make sure pfn exists to start with */
261 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
264 ret = xen_free_mfn(mfn);
265 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
268 xen_released_pages++;
269 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
275 set_phys_range_identity(start_pfn, end_pfn);
279 * Helper function to update the p2m and m2p tables and kernel mapping.
281 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
283 struct mmu_update update = {
284 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
289 if (!set_phys_to_machine(pfn, mfn)) {
290 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
296 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
297 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
302 /* Update kernel mapping, but not for highmem. */
303 if (pfn >= PFN_UP(__pa(high_memory - 1)))
306 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
307 mfn_pte(mfn, PAGE_KERNEL), 0)) {
308 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
315 * This function updates the p2m and m2p tables with an identity map from
316 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
317 * original allocation at remap_pfn. The information needed for remapping is
318 * saved in the memory itself to avoid the need for allocating buffers. The
319 * complete remap information is contained in a list of MFNs each containing
320 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
321 * This enables us to preserve the original mfn sequence while doing the
322 * remapping at a time when the memory management is capable of allocating
323 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
326 static void __init xen_do_set_identity_and_remap_chunk(
327 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
329 unsigned long buf = (unsigned long)&xen_remap_buf;
330 unsigned long mfn_save, mfn;
331 unsigned long ident_pfn_iter, remap_pfn_iter;
332 unsigned long ident_end_pfn = start_pfn + size;
333 unsigned long left = size;
334 unsigned int i, chunk;
338 BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
340 mfn_save = virt_to_mfn(buf);
342 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
343 ident_pfn_iter < ident_end_pfn;
344 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
345 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
347 /* Map first pfn to xen_remap_buf */
348 mfn = pfn_to_mfn(ident_pfn_iter);
349 set_pte_mfn(buf, mfn, PAGE_KERNEL);
351 /* Save mapping information in page */
352 xen_remap_buf.next_area_mfn = xen_remap_mfn;
353 xen_remap_buf.target_pfn = remap_pfn_iter;
354 xen_remap_buf.size = chunk;
355 for (i = 0; i < chunk; i++)
356 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
358 /* Put remap buf into list. */
361 /* Set identity map */
362 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
367 /* Restore old xen_remap_buf mapping */
368 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
372 * This function takes a contiguous pfn range that needs to be identity mapped
375 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
376 * 2) Calls the do_ function to actually do the mapping/remapping work.
378 * The goal is to not allocate additional memory but to remap the existing
379 * pages. In the case of an error the underlying memory is simply released back
380 * to Xen and not remapped.
382 static unsigned long __init xen_set_identity_and_remap_chunk(
383 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
384 unsigned long remap_pfn)
388 unsigned long n = end_pfn - start_pfn;
391 unsigned long cur_pfn = start_pfn + i;
392 unsigned long left = n - i;
393 unsigned long size = left;
394 unsigned long remap_range_size;
396 /* Do not remap pages beyond the current allocation */
397 if (cur_pfn >= nr_pages) {
398 /* Identity map remaining pages */
399 set_phys_range_identity(cur_pfn, cur_pfn + size);
402 if (cur_pfn + size > nr_pages)
403 size = nr_pages - cur_pfn;
405 remap_range_size = xen_find_pfn_range(&remap_pfn);
406 if (!remap_range_size) {
407 pr_warning("Unable to find available pfn range, not remapping identity pages\n");
408 xen_set_identity_and_release_chunk(cur_pfn,
409 cur_pfn + left, nr_pages);
412 /* Adjust size to fit in current e820 RAM region */
413 if (size > remap_range_size)
414 size = remap_range_size;
416 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
418 /* Update variables to reflect new mappings. */
424 * If the PFNs are currently mapped, the VA mapping also needs
425 * to be updated to be 1:1.
427 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
428 (void)HYPERVISOR_update_va_mapping(
429 (unsigned long)__va(pfn << PAGE_SHIFT),
430 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
435 static void __init xen_set_identity_and_remap(unsigned long nr_pages)
437 phys_addr_t start = 0;
438 unsigned long last_pfn = nr_pages;
439 const struct e820entry *entry = xen_e820_map;
443 * Combine non-RAM regions and gaps until a RAM region (or the
444 * end of the map) is reached, then set the 1:1 map and
445 * remap the memory in those non-RAM regions.
447 * The combined non-RAM regions are rounded to a whole number
448 * of pages so any partial pages are accessible via the 1:1
449 * mapping. This is needed for some BIOSes that put (for
450 * example) the DMI tables in a reserved region that begins on
451 * a non-page boundary.
453 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
454 phys_addr_t end = entry->addr + entry->size;
455 if (entry->type == E820_RAM || i == xen_e820_map_entries - 1) {
456 unsigned long start_pfn = PFN_DOWN(start);
457 unsigned long end_pfn = PFN_UP(end);
459 if (entry->type == E820_RAM)
460 end_pfn = PFN_UP(entry->addr);
462 if (start_pfn < end_pfn)
463 last_pfn = xen_set_identity_and_remap_chunk(
464 start_pfn, end_pfn, nr_pages,
470 pr_info("Released %ld page(s)\n", xen_released_pages);
474 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
475 * The remap information (which mfn remap to which pfn) is contained in the
476 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
477 * This scheme allows to remap the different chunks in arbitrary order while
478 * the resulting mapping will be independant from the order.
480 void __init xen_remap_memory(void)
482 unsigned long buf = (unsigned long)&xen_remap_buf;
483 unsigned long mfn_save, mfn, pfn;
484 unsigned long remapped = 0;
486 unsigned long pfn_s = ~0UL;
487 unsigned long len = 0;
489 mfn_save = virt_to_mfn(buf);
491 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
492 /* Map the remap information */
493 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
495 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
497 pfn = xen_remap_buf.target_pfn;
498 for (i = 0; i < xen_remap_buf.size; i++) {
499 mfn = xen_remap_buf.mfns[i];
500 xen_update_mem_tables(pfn, mfn);
504 if (pfn_s == ~0UL || pfn == pfn_s) {
505 pfn_s = xen_remap_buf.target_pfn;
506 len += xen_remap_buf.size;
507 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
508 len += xen_remap_buf.size;
510 xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
511 pfn_s = xen_remap_buf.target_pfn;
512 len = xen_remap_buf.size;
516 xen_remap_mfn = xen_remap_buf.next_area_mfn;
519 if (pfn_s != ~0UL && len)
520 xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
522 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
524 pr_info("Remapped %ld page(s)\n", remapped);
527 static unsigned long __init xen_get_pages_limit(void)
532 limit = GB(64) / PAGE_SIZE;
534 limit = MAXMEM / PAGE_SIZE;
535 if (!xen_initial_domain() && xen_512gb_limit)
536 limit = GB(512) / PAGE_SIZE;
541 static unsigned long __init xen_get_max_pages(void)
543 unsigned long max_pages, limit;
544 domid_t domid = DOMID_SELF;
547 limit = xen_get_pages_limit();
551 * For the initial domain we use the maximum reservation as
554 * For guest domains the current maximum reservation reflects
555 * the current maximum rather than the static maximum. In this
556 * case the e820 map provided to us will cover the static
559 if (xen_initial_domain()) {
560 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
565 return min(max_pages, limit);
568 static void __init xen_align_and_add_e820_region(phys_addr_t start,
569 phys_addr_t size, int type)
571 phys_addr_t end = start + size;
573 /* Align RAM regions to page boundaries. */
574 if (type == E820_RAM) {
575 start = PAGE_ALIGN(start);
576 end &= ~((phys_addr_t)PAGE_SIZE - 1);
579 e820_add_region(start, end - start, type);
582 static void __init xen_ignore_unusable(void)
584 struct e820entry *entry = xen_e820_map;
587 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
588 if (entry->type == E820_UNUSABLE)
589 entry->type = E820_RAM;
593 static unsigned long __init xen_count_remap_pages(unsigned long max_pfn)
595 unsigned long extra = 0;
596 unsigned long start_pfn, end_pfn;
597 const struct e820entry *entry = xen_e820_map;
601 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
602 start_pfn = PFN_DOWN(entry->addr);
603 /* Adjacent regions on non-page boundaries handling! */
604 end_pfn = min(end_pfn, start_pfn);
606 if (start_pfn >= max_pfn)
607 return extra + max_pfn - end_pfn;
609 /* Add any holes in map to result. */
610 extra += start_pfn - end_pfn;
612 end_pfn = PFN_UP(entry->addr + entry->size);
613 end_pfn = min(end_pfn, max_pfn);
615 if (entry->type != E820_RAM)
616 extra += end_pfn - start_pfn;
622 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
624 struct e820entry *entry;
632 entry = xen_e820_map;
634 for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++) {
635 if (entry->type == E820_RAM && entry->addr <= start &&
636 (entry->addr + entry->size) >= end)
646 * Find a free area in physical memory not yet reserved and compliant with
648 * Used to relocate pre-allocated areas like initrd or p2m list which are in
649 * conflict with the to be used E820 map.
650 * In case no area is found, return 0. Otherwise return the physical address
651 * of the area which is already reserved for convenience.
653 phys_addr_t __init xen_find_free_area(phys_addr_t size)
656 phys_addr_t addr, start;
657 struct e820entry *entry = xen_e820_map;
659 for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++, entry++) {
660 if (entry->type != E820_RAM || entry->size < size)
663 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
664 if (!memblock_is_reserved(addr))
666 start = addr + PAGE_SIZE;
667 if (start + size > entry->addr + entry->size)
670 if (addr >= start + size) {
671 memblock_reserve(start, size);
680 * Like memcpy, but with physical addresses for dest and src.
682 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
685 phys_addr_t dest_off, src_off, dest_len, src_len, len;
689 dest_off = dest & ~PAGE_MASK;
690 src_off = src & ~PAGE_MASK;
692 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
693 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
695 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
696 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
697 len = min(dest_len, src_len);
698 to = early_memremap(dest - dest_off, dest_len + dest_off);
699 from = early_memremap(src - src_off, src_len + src_off);
700 memcpy(to, from, len);
701 early_memunmap(to, dest_len + dest_off);
702 early_memunmap(from, src_len + src_off);
710 * Reserve Xen mfn_list.
712 static void __init xen_reserve_xen_mfnlist(void)
714 phys_addr_t start, size;
716 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
717 start = __pa(xen_start_info->mfn_list);
718 size = PFN_ALIGN(xen_start_info->nr_pages *
719 sizeof(unsigned long));
721 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
722 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
725 if (!xen_is_e820_reserved(start, size)) {
726 memblock_reserve(start, size);
732 * Relocating the p2m on 32 bit system to an arbitrary virtual address
733 * is not supported, so just give up.
735 xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n");
743 * machine_specific_memory_setup - Hook for machine specific memory setup.
745 char * __init xen_memory_setup(void)
747 unsigned long max_pfn;
748 phys_addr_t mem_end, addr, size, chunk_size;
751 struct xen_memory_map memmap;
752 unsigned long max_pages;
753 unsigned long extra_pages = 0;
758 max_pfn = xen_get_pages_limit();
759 max_pfn = min(max_pfn, xen_start_info->nr_pages);
760 mem_end = PFN_PHYS(max_pfn);
762 memmap.nr_entries = E820MAX;
763 set_xen_guest_handle(memmap.buffer, xen_e820_map);
765 op = xen_initial_domain() ?
766 XENMEM_machine_memory_map :
768 rc = HYPERVISOR_memory_op(op, &memmap);
770 BUG_ON(xen_initial_domain());
771 memmap.nr_entries = 1;
772 xen_e820_map[0].addr = 0ULL;
773 xen_e820_map[0].size = mem_end;
774 /* 8MB slack (to balance backend allocations). */
775 xen_e820_map[0].size += 8ULL << 20;
776 xen_e820_map[0].type = E820_RAM;
780 BUG_ON(memmap.nr_entries == 0);
781 xen_e820_map_entries = memmap.nr_entries;
784 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
785 * regions, so if we're using the machine memory map leave the
786 * region as RAM as it is in the pseudo-physical map.
788 * UNUSABLE regions in domUs are not handled and will need
789 * a patch in the future.
791 if (xen_initial_domain())
792 xen_ignore_unusable();
794 /* Make sure the Xen-supplied memory map is well-ordered. */
795 sanitize_e820_map(xen_e820_map, xen_e820_map_entries,
796 &xen_e820_map_entries);
798 max_pages = xen_get_max_pages();
800 /* How many extra pages do we need due to remapping? */
801 max_pages += xen_count_remap_pages(max_pfn);
803 if (max_pages > max_pfn)
804 extra_pages += max_pages - max_pfn;
807 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
808 * factor the base size. On non-highmem systems, the base
809 * size is the full initial memory allocation; on highmem it
810 * is limited to the max size of lowmem, so that it doesn't
811 * get completely filled.
813 * Make sure we have no memory above max_pages, as this area
814 * isn't handled by the p2m management.
816 * In principle there could be a problem in lowmem systems if
817 * the initial memory is also very large with respect to
818 * lowmem, but we won't try to deal with that here.
820 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
821 extra_pages, max_pages - max_pfn);
823 addr = xen_e820_map[0].addr;
824 size = xen_e820_map[0].size;
825 while (i < xen_e820_map_entries) {
827 type = xen_e820_map[i].type;
829 if (type == E820_RAM) {
830 if (addr < mem_end) {
831 chunk_size = min(size, mem_end - addr);
832 } else if (extra_pages) {
833 chunk_size = min(size, PFN_PHYS(extra_pages));
834 extra_pages -= PFN_DOWN(chunk_size);
835 xen_add_extra_mem(addr, chunk_size);
836 xen_max_p2m_pfn = PFN_DOWN(addr + chunk_size);
838 type = E820_UNUSABLE;
841 xen_align_and_add_e820_region(addr, chunk_size, type);
847 if (i < xen_e820_map_entries) {
848 addr = xen_e820_map[i].addr;
849 size = xen_e820_map[i].size;
855 * Set the rest as identity mapped, in case PCI BARs are
858 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
861 * In domU, the ISA region is normal, usable memory, but we
862 * reserve ISA memory anyway because too many things poke
865 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
868 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
871 * Check whether the kernel itself conflicts with the target E820 map.
872 * Failing now is better than running into weird problems later due
873 * to relocating (and even reusing) pages with kernel text or data.
875 if (xen_is_e820_reserved(__pa_symbol(_text),
876 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
877 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
882 * Check for a conflict of the hypervisor supplied page tables with
883 * the target E820 map.
887 xen_reserve_xen_mfnlist();
889 /* Check for a conflict of the initrd with the target E820 map. */
890 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
891 boot_params.hdr.ramdisk_size)) {
892 phys_addr_t new_area, start, size;
894 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
896 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
900 start = boot_params.hdr.ramdisk_image;
901 size = boot_params.hdr.ramdisk_size;
902 xen_phys_memcpy(new_area, start, size);
903 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
904 start, start + size, new_area, new_area + size);
905 memblock_free(start, size);
906 boot_params.hdr.ramdisk_image = new_area;
907 boot_params.ext_ramdisk_image = new_area >> 32;
911 * Set identity map on non-RAM pages and prepare remapping the
914 xen_set_identity_and_remap(max_pfn);
920 * Machine specific memory setup for auto-translated guests.
922 char * __init xen_auto_xlated_memory_setup(void)
924 struct xen_memory_map memmap;
928 memmap.nr_entries = E820MAX;
929 set_xen_guest_handle(memmap.buffer, xen_e820_map);
931 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
933 panic("No memory map (%d)\n", rc);
935 xen_e820_map_entries = memmap.nr_entries;
937 sanitize_e820_map(xen_e820_map, ARRAY_SIZE(xen_e820_map),
938 &xen_e820_map_entries);
940 for (i = 0; i < xen_e820_map_entries; i++)
941 e820_add_region(xen_e820_map[i].addr, xen_e820_map[i].size,
942 xen_e820_map[i].type);
944 /* Remove p2m info, it is not needed. */
945 xen_start_info->mfn_list = 0;
946 xen_start_info->first_p2m_pfn = 0;
947 xen_start_info->nr_p2m_frames = 0;
953 * Set the bit indicating "nosegneg" library variants should be used.
954 * We only need to bother in pure 32-bit mode; compat 32-bit processes
955 * can have un-truncated segments, so wrapping around is allowed.
957 static void __init fiddle_vdso(void)
961 * This could be called before selected_vdso32 is initialized, so
962 * just fiddle with both possible images. vdso_image_32_syscall
963 * can't be selected, since it only exists on 64-bit systems.
966 mask = vdso_image_32_int80.data +
967 vdso_image_32_int80.sym_VDSO32_NOTE_MASK;
968 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
969 mask = vdso_image_32_sysenter.data +
970 vdso_image_32_sysenter.sym_VDSO32_NOTE_MASK;
971 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
975 static int register_callback(unsigned type, const void *func)
977 struct callback_register callback = {
979 .address = XEN_CALLBACK(__KERNEL_CS, func),
980 .flags = CALLBACKF_mask_events,
983 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
986 void xen_enable_sysenter(void)
989 unsigned sysenter_feature;
992 sysenter_feature = X86_FEATURE_SEP;
994 sysenter_feature = X86_FEATURE_SYSENTER32;
997 if (!boot_cpu_has(sysenter_feature))
1000 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
1002 setup_clear_cpu_cap(sysenter_feature);
1005 void xen_enable_syscall(void)
1007 #ifdef CONFIG_X86_64
1010 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
1012 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
1013 /* Pretty fatal; 64-bit userspace has no other
1014 mechanism for syscalls. */
1017 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
1018 ret = register_callback(CALLBACKTYPE_syscall32,
1019 xen_syscall32_target);
1021 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
1023 #endif /* CONFIG_X86_64 */
1026 void __init xen_pvmmu_arch_setup(void)
1028 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
1029 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
1031 HYPERVISOR_vm_assist(VMASST_CMD_enable,
1032 VMASST_TYPE_pae_extended_cr3);
1034 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
1035 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1038 xen_enable_sysenter();
1039 xen_enable_syscall();
1042 /* This function is not called for HVM domains */
1043 void __init xen_arch_setup(void)
1045 xen_panic_handler_init();
1046 if (!xen_feature(XENFEAT_auto_translated_physmap))
1047 xen_pvmmu_arch_setup();
1050 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1051 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1056 memcpy(boot_command_line, xen_start_info->cmd_line,
1057 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1058 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1060 /* Set up idle, making sure it calls safe_halt() pvop */
1063 WARN_ON(xen_set_default_idle());