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(unsigned long start_pfn,
99 * No need to check for zero size, should happen rarely and will only
100 * write a new entry regarded to be unused due to zero size.
102 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
103 /* Add new region. */
104 if (xen_extra_mem[i].n_pfns == 0) {
105 xen_extra_mem[i].start_pfn = start_pfn;
106 xen_extra_mem[i].n_pfns = n_pfns;
109 /* Append to existing region. */
110 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
112 xen_extra_mem[i].n_pfns += n_pfns;
116 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
117 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
119 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
122 static void __init xen_del_extra_mem(unsigned long start_pfn,
123 unsigned long n_pfns)
126 unsigned long start_r, size_r;
128 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
129 start_r = xen_extra_mem[i].start_pfn;
130 size_r = xen_extra_mem[i].n_pfns;
132 /* Start of region. */
133 if (start_r == start_pfn) {
134 BUG_ON(n_pfns > size_r);
135 xen_extra_mem[i].start_pfn += n_pfns;
136 xen_extra_mem[i].n_pfns -= n_pfns;
140 if (start_r + size_r == start_pfn + n_pfns) {
141 BUG_ON(n_pfns > size_r);
142 xen_extra_mem[i].n_pfns -= n_pfns;
146 if (start_pfn > start_r && start_pfn < start_r + size_r) {
147 BUG_ON(start_pfn + n_pfns > start_r + size_r);
148 xen_extra_mem[i].n_pfns = start_pfn - start_r;
149 /* Calling memblock_reserve() again is okay. */
150 xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
151 (start_pfn + n_pfns));
155 memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
159 * Called during boot before the p2m list can take entries beyond the
160 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
163 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
167 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
168 if (pfn >= xen_extra_mem[i].start_pfn &&
169 pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
170 return INVALID_P2M_ENTRY;
173 return IDENTITY_FRAME(pfn);
177 * Mark all pfns of extra mem as invalid in p2m list.
179 void __init xen_inv_extra_mem(void)
181 unsigned long pfn, pfn_s, pfn_e;
184 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
185 if (!xen_extra_mem[i].n_pfns)
187 pfn_s = xen_extra_mem[i].start_pfn;
188 pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
189 for (pfn = pfn_s; pfn < pfn_e; pfn++)
190 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
195 * Finds the next RAM pfn available in the E820 map after min_pfn.
196 * This function updates min_pfn with the pfn found and returns
197 * the size of that range or zero if not found.
199 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
201 const struct e820entry *entry = xen_e820_map;
203 unsigned long done = 0;
205 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
209 if (entry->type != E820_RAM)
212 e_pfn = PFN_DOWN(entry->addr + entry->size);
214 /* We only care about E820 after this */
215 if (e_pfn < *min_pfn)
218 s_pfn = PFN_UP(entry->addr);
220 /* If min_pfn falls within the E820 entry, we want to start
221 * at the min_pfn PFN.
223 if (s_pfn <= *min_pfn) {
224 done = e_pfn - *min_pfn;
226 done = e_pfn - s_pfn;
235 static int __init xen_free_mfn(unsigned long mfn)
237 struct xen_memory_reservation reservation = {
243 set_xen_guest_handle(reservation.extent_start, &mfn);
244 reservation.nr_extents = 1;
246 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
250 * This releases a chunk of memory and then does the identity map. It's used
251 * as a fallback if the remapping fails.
253 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
254 unsigned long end_pfn, unsigned long nr_pages)
256 unsigned long pfn, end;
259 WARN_ON(start_pfn > end_pfn);
261 /* Release pages first. */
262 end = min(end_pfn, nr_pages);
263 for (pfn = start_pfn; pfn < end; pfn++) {
264 unsigned long mfn = pfn_to_mfn(pfn);
266 /* Make sure pfn exists to start with */
267 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
270 ret = xen_free_mfn(mfn);
271 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
274 xen_released_pages++;
275 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
281 set_phys_range_identity(start_pfn, end_pfn);
285 * Helper function to update the p2m and m2p tables and kernel mapping.
287 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
289 struct mmu_update update = {
290 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
295 if (!set_phys_to_machine(pfn, mfn)) {
296 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
302 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
303 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
308 /* Update kernel mapping, but not for highmem. */
309 if (pfn >= PFN_UP(__pa(high_memory - 1)))
312 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
313 mfn_pte(mfn, PAGE_KERNEL), 0)) {
314 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
321 * This function updates the p2m and m2p tables with an identity map from
322 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
323 * original allocation at remap_pfn. The information needed for remapping is
324 * saved in the memory itself to avoid the need for allocating buffers. The
325 * complete remap information is contained in a list of MFNs each containing
326 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
327 * This enables us to preserve the original mfn sequence while doing the
328 * remapping at a time when the memory management is capable of allocating
329 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
332 static void __init xen_do_set_identity_and_remap_chunk(
333 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
335 unsigned long buf = (unsigned long)&xen_remap_buf;
336 unsigned long mfn_save, mfn;
337 unsigned long ident_pfn_iter, remap_pfn_iter;
338 unsigned long ident_end_pfn = start_pfn + size;
339 unsigned long left = size;
340 unsigned int i, chunk;
344 BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
346 mfn_save = virt_to_mfn(buf);
348 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
349 ident_pfn_iter < ident_end_pfn;
350 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
351 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
353 /* Map first pfn to xen_remap_buf */
354 mfn = pfn_to_mfn(ident_pfn_iter);
355 set_pte_mfn(buf, mfn, PAGE_KERNEL);
357 /* Save mapping information in page */
358 xen_remap_buf.next_area_mfn = xen_remap_mfn;
359 xen_remap_buf.target_pfn = remap_pfn_iter;
360 xen_remap_buf.size = chunk;
361 for (i = 0; i < chunk; i++)
362 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
364 /* Put remap buf into list. */
367 /* Set identity map */
368 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
373 /* Restore old xen_remap_buf mapping */
374 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
378 * This function takes a contiguous pfn range that needs to be identity mapped
381 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
382 * 2) Calls the do_ function to actually do the mapping/remapping work.
384 * The goal is to not allocate additional memory but to remap the existing
385 * pages. In the case of an error the underlying memory is simply released back
386 * to Xen and not remapped.
388 static unsigned long __init xen_set_identity_and_remap_chunk(
389 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
390 unsigned long remap_pfn)
394 unsigned long n = end_pfn - start_pfn;
397 unsigned long cur_pfn = start_pfn + i;
398 unsigned long left = n - i;
399 unsigned long size = left;
400 unsigned long remap_range_size;
402 /* Do not remap pages beyond the current allocation */
403 if (cur_pfn >= nr_pages) {
404 /* Identity map remaining pages */
405 set_phys_range_identity(cur_pfn, cur_pfn + size);
408 if (cur_pfn + size > nr_pages)
409 size = nr_pages - cur_pfn;
411 remap_range_size = xen_find_pfn_range(&remap_pfn);
412 if (!remap_range_size) {
413 pr_warning("Unable to find available pfn range, not remapping identity pages\n");
414 xen_set_identity_and_release_chunk(cur_pfn,
415 cur_pfn + left, nr_pages);
418 /* Adjust size to fit in current e820 RAM region */
419 if (size > remap_range_size)
420 size = remap_range_size;
422 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
424 /* Update variables to reflect new mappings. */
430 * If the PFNs are currently mapped, the VA mapping also needs
431 * to be updated to be 1:1.
433 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
434 (void)HYPERVISOR_update_va_mapping(
435 (unsigned long)__va(pfn << PAGE_SHIFT),
436 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
441 static void __init xen_set_identity_and_remap(unsigned long nr_pages)
443 phys_addr_t start = 0;
444 unsigned long last_pfn = nr_pages;
445 const struct e820entry *entry = xen_e820_map;
449 * Combine non-RAM regions and gaps until a RAM region (or the
450 * end of the map) is reached, then set the 1:1 map and
451 * remap the memory in those non-RAM regions.
453 * The combined non-RAM regions are rounded to a whole number
454 * of pages so any partial pages are accessible via the 1:1
455 * mapping. This is needed for some BIOSes that put (for
456 * example) the DMI tables in a reserved region that begins on
457 * a non-page boundary.
459 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
460 phys_addr_t end = entry->addr + entry->size;
461 if (entry->type == E820_RAM || i == xen_e820_map_entries - 1) {
462 unsigned long start_pfn = PFN_DOWN(start);
463 unsigned long end_pfn = PFN_UP(end);
465 if (entry->type == E820_RAM)
466 end_pfn = PFN_UP(entry->addr);
468 if (start_pfn < end_pfn)
469 last_pfn = xen_set_identity_and_remap_chunk(
470 start_pfn, end_pfn, nr_pages,
476 pr_info("Released %ld page(s)\n", xen_released_pages);
480 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
481 * The remap information (which mfn remap to which pfn) is contained in the
482 * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
483 * This scheme allows to remap the different chunks in arbitrary order while
484 * the resulting mapping will be independant from the order.
486 void __init xen_remap_memory(void)
488 unsigned long buf = (unsigned long)&xen_remap_buf;
489 unsigned long mfn_save, mfn, pfn;
490 unsigned long remapped = 0;
492 unsigned long pfn_s = ~0UL;
493 unsigned long len = 0;
495 mfn_save = virt_to_mfn(buf);
497 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
498 /* Map the remap information */
499 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
501 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
503 pfn = xen_remap_buf.target_pfn;
504 for (i = 0; i < xen_remap_buf.size; i++) {
505 mfn = xen_remap_buf.mfns[i];
506 xen_update_mem_tables(pfn, mfn);
510 if (pfn_s == ~0UL || pfn == pfn_s) {
511 pfn_s = xen_remap_buf.target_pfn;
512 len += xen_remap_buf.size;
513 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
514 len += xen_remap_buf.size;
516 xen_del_extra_mem(pfn_s, len);
517 pfn_s = xen_remap_buf.target_pfn;
518 len = xen_remap_buf.size;
522 xen_remap_mfn = xen_remap_buf.next_area_mfn;
525 if (pfn_s != ~0UL && len)
526 xen_del_extra_mem(pfn_s, len);
528 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
530 pr_info("Remapped %ld page(s)\n", remapped);
533 static unsigned long __init xen_get_pages_limit(void)
538 limit = GB(64) / PAGE_SIZE;
540 limit = MAXMEM / PAGE_SIZE;
541 if (!xen_initial_domain() && xen_512gb_limit)
542 limit = GB(512) / PAGE_SIZE;
547 static unsigned long __init xen_get_max_pages(void)
549 unsigned long max_pages, limit;
550 domid_t domid = DOMID_SELF;
553 limit = xen_get_pages_limit();
557 * For the initial domain we use the maximum reservation as
560 * For guest domains the current maximum reservation reflects
561 * the current maximum rather than the static maximum. In this
562 * case the e820 map provided to us will cover the static
565 if (xen_initial_domain()) {
566 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
571 return min(max_pages, limit);
574 static void __init xen_align_and_add_e820_region(phys_addr_t start,
575 phys_addr_t size, int type)
577 phys_addr_t end = start + size;
579 /* Align RAM regions to page boundaries. */
580 if (type == E820_RAM) {
581 start = PAGE_ALIGN(start);
582 end &= ~((phys_addr_t)PAGE_SIZE - 1);
585 e820_add_region(start, end - start, type);
588 static void __init xen_ignore_unusable(void)
590 struct e820entry *entry = xen_e820_map;
593 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
594 if (entry->type == E820_UNUSABLE)
595 entry->type = E820_RAM;
599 static unsigned long __init xen_count_remap_pages(unsigned long max_pfn)
601 unsigned long extra = 0;
602 unsigned long start_pfn, end_pfn;
603 const struct e820entry *entry = xen_e820_map;
607 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
608 start_pfn = PFN_DOWN(entry->addr);
609 /* Adjacent regions on non-page boundaries handling! */
610 end_pfn = min(end_pfn, start_pfn);
612 if (start_pfn >= max_pfn)
613 return extra + max_pfn - end_pfn;
615 /* Add any holes in map to result. */
616 extra += start_pfn - end_pfn;
618 end_pfn = PFN_UP(entry->addr + entry->size);
619 end_pfn = min(end_pfn, max_pfn);
621 if (entry->type != E820_RAM)
622 extra += end_pfn - start_pfn;
628 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
630 struct e820entry *entry;
638 entry = xen_e820_map;
640 for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++) {
641 if (entry->type == E820_RAM && entry->addr <= start &&
642 (entry->addr + entry->size) >= end)
652 * Find a free area in physical memory not yet reserved and compliant with
654 * Used to relocate pre-allocated areas like initrd or p2m list which are in
655 * conflict with the to be used E820 map.
656 * In case no area is found, return 0. Otherwise return the physical address
657 * of the area which is already reserved for convenience.
659 phys_addr_t __init xen_find_free_area(phys_addr_t size)
662 phys_addr_t addr, start;
663 struct e820entry *entry = xen_e820_map;
665 for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++, entry++) {
666 if (entry->type != E820_RAM || entry->size < size)
669 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
670 if (!memblock_is_reserved(addr))
672 start = addr + PAGE_SIZE;
673 if (start + size > entry->addr + entry->size)
676 if (addr >= start + size) {
677 memblock_reserve(start, size);
686 * Like memcpy, but with physical addresses for dest and src.
688 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
691 phys_addr_t dest_off, src_off, dest_len, src_len, len;
695 dest_off = dest & ~PAGE_MASK;
696 src_off = src & ~PAGE_MASK;
698 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
699 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
701 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
702 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
703 len = min(dest_len, src_len);
704 to = early_memremap(dest - dest_off, dest_len + dest_off);
705 from = early_memremap(src - src_off, src_len + src_off);
706 memcpy(to, from, len);
707 early_memunmap(to, dest_len + dest_off);
708 early_memunmap(from, src_len + src_off);
716 * Reserve Xen mfn_list.
718 static void __init xen_reserve_xen_mfnlist(void)
720 phys_addr_t start, size;
722 if (xen_start_info->mfn_list >= __START_KERNEL_map) {
723 start = __pa(xen_start_info->mfn_list);
724 size = PFN_ALIGN(xen_start_info->nr_pages *
725 sizeof(unsigned long));
727 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
728 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
731 if (!xen_is_e820_reserved(start, size)) {
732 memblock_reserve(start, size);
738 * Relocating the p2m on 32 bit system to an arbitrary virtual address
739 * is not supported, so just give up.
741 xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n");
749 * machine_specific_memory_setup - Hook for machine specific memory setup.
751 char * __init xen_memory_setup(void)
753 unsigned long max_pfn, pfn_s, n_pfns;
754 phys_addr_t mem_end, addr, size, chunk_size;
757 struct xen_memory_map memmap;
758 unsigned long max_pages;
759 unsigned long extra_pages = 0;
764 max_pfn = xen_get_pages_limit();
765 max_pfn = min(max_pfn, xen_start_info->nr_pages);
766 mem_end = PFN_PHYS(max_pfn);
768 memmap.nr_entries = E820MAX;
769 set_xen_guest_handle(memmap.buffer, xen_e820_map);
771 op = xen_initial_domain() ?
772 XENMEM_machine_memory_map :
774 rc = HYPERVISOR_memory_op(op, &memmap);
776 BUG_ON(xen_initial_domain());
777 memmap.nr_entries = 1;
778 xen_e820_map[0].addr = 0ULL;
779 xen_e820_map[0].size = mem_end;
780 /* 8MB slack (to balance backend allocations). */
781 xen_e820_map[0].size += 8ULL << 20;
782 xen_e820_map[0].type = E820_RAM;
786 BUG_ON(memmap.nr_entries == 0);
787 xen_e820_map_entries = memmap.nr_entries;
790 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
791 * regions, so if we're using the machine memory map leave the
792 * region as RAM as it is in the pseudo-physical map.
794 * UNUSABLE regions in domUs are not handled and will need
795 * a patch in the future.
797 if (xen_initial_domain())
798 xen_ignore_unusable();
800 /* Make sure the Xen-supplied memory map is well-ordered. */
801 sanitize_e820_map(xen_e820_map, ARRAY_SIZE(xen_e820_map),
802 &xen_e820_map_entries);
804 max_pages = xen_get_max_pages();
806 /* How many extra pages do we need due to remapping? */
807 max_pages += xen_count_remap_pages(max_pfn);
809 if (max_pages > max_pfn)
810 extra_pages += max_pages - max_pfn;
813 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
814 * factor the base size. On non-highmem systems, the base
815 * size is the full initial memory allocation; on highmem it
816 * is limited to the max size of lowmem, so that it doesn't
817 * get completely filled.
819 * Make sure we have no memory above max_pages, as this area
820 * isn't handled by the p2m management.
822 * In principle there could be a problem in lowmem systems if
823 * the initial memory is also very large with respect to
824 * lowmem, but we won't try to deal with that here.
826 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
827 extra_pages, max_pages - max_pfn);
829 addr = xen_e820_map[0].addr;
830 size = xen_e820_map[0].size;
831 while (i < xen_e820_map_entries) {
833 type = xen_e820_map[i].type;
835 if (type == E820_RAM) {
836 if (addr < mem_end) {
837 chunk_size = min(size, mem_end - addr);
838 } else if (extra_pages) {
839 chunk_size = min(size, PFN_PHYS(extra_pages));
840 pfn_s = PFN_UP(addr);
841 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
842 extra_pages -= n_pfns;
843 xen_add_extra_mem(pfn_s, n_pfns);
844 xen_max_p2m_pfn = pfn_s + n_pfns;
846 type = E820_UNUSABLE;
849 xen_align_and_add_e820_region(addr, chunk_size, type);
855 if (i < xen_e820_map_entries) {
856 addr = xen_e820_map[i].addr;
857 size = xen_e820_map[i].size;
863 * Set the rest as identity mapped, in case PCI BARs are
866 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
869 * In domU, the ISA region is normal, usable memory, but we
870 * reserve ISA memory anyway because too many things poke
873 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
876 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
879 * Check whether the kernel itself conflicts with the target E820 map.
880 * Failing now is better than running into weird problems later due
881 * to relocating (and even reusing) pages with kernel text or data.
883 if (xen_is_e820_reserved(__pa_symbol(_text),
884 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
885 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
890 * Check for a conflict of the hypervisor supplied page tables with
891 * the target E820 map.
895 xen_reserve_xen_mfnlist();
897 /* Check for a conflict of the initrd with the target E820 map. */
898 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
899 boot_params.hdr.ramdisk_size)) {
900 phys_addr_t new_area, start, size;
902 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
904 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
908 start = boot_params.hdr.ramdisk_image;
909 size = boot_params.hdr.ramdisk_size;
910 xen_phys_memcpy(new_area, start, size);
911 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
912 start, start + size, new_area, new_area + size);
913 memblock_free(start, size);
914 boot_params.hdr.ramdisk_image = new_area;
915 boot_params.ext_ramdisk_image = new_area >> 32;
919 * Set identity map on non-RAM pages and prepare remapping the
922 xen_set_identity_and_remap(max_pfn);
928 * Machine specific memory setup for auto-translated guests.
930 char * __init xen_auto_xlated_memory_setup(void)
932 struct xen_memory_map memmap;
936 memmap.nr_entries = E820MAX;
937 set_xen_guest_handle(memmap.buffer, xen_e820_map);
939 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
941 panic("No memory map (%d)\n", rc);
943 xen_e820_map_entries = memmap.nr_entries;
945 sanitize_e820_map(xen_e820_map, ARRAY_SIZE(xen_e820_map),
946 &xen_e820_map_entries);
948 for (i = 0; i < xen_e820_map_entries; i++)
949 e820_add_region(xen_e820_map[i].addr, xen_e820_map[i].size,
950 xen_e820_map[i].type);
952 /* Remove p2m info, it is not needed. */
953 xen_start_info->mfn_list = 0;
954 xen_start_info->first_p2m_pfn = 0;
955 xen_start_info->nr_p2m_frames = 0;
961 * Set the bit indicating "nosegneg" library variants should be used.
962 * We only need to bother in pure 32-bit mode; compat 32-bit processes
963 * can have un-truncated segments, so wrapping around is allowed.
965 static void __init fiddle_vdso(void)
969 * This could be called before selected_vdso32 is initialized, so
970 * just fiddle with both possible images. vdso_image_32_syscall
971 * can't be selected, since it only exists on 64-bit systems.
974 mask = vdso_image_32_int80.data +
975 vdso_image_32_int80.sym_VDSO32_NOTE_MASK;
976 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
977 mask = vdso_image_32_sysenter.data +
978 vdso_image_32_sysenter.sym_VDSO32_NOTE_MASK;
979 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
983 static int register_callback(unsigned type, const void *func)
985 struct callback_register callback = {
987 .address = XEN_CALLBACK(__KERNEL_CS, func),
988 .flags = CALLBACKF_mask_events,
991 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
994 void xen_enable_sysenter(void)
997 unsigned sysenter_feature;
1000 sysenter_feature = X86_FEATURE_SEP;
1002 sysenter_feature = X86_FEATURE_SYSENTER32;
1005 if (!boot_cpu_has(sysenter_feature))
1008 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
1010 setup_clear_cpu_cap(sysenter_feature);
1013 void xen_enable_syscall(void)
1015 #ifdef CONFIG_X86_64
1018 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
1020 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
1021 /* Pretty fatal; 64-bit userspace has no other
1022 mechanism for syscalls. */
1025 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
1026 ret = register_callback(CALLBACKTYPE_syscall32,
1027 xen_syscall32_target);
1029 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
1031 #endif /* CONFIG_X86_64 */
1034 void __init xen_pvmmu_arch_setup(void)
1036 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
1037 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
1039 HYPERVISOR_vm_assist(VMASST_CMD_enable,
1040 VMASST_TYPE_pae_extended_cr3);
1042 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
1043 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1046 xen_enable_sysenter();
1047 xen_enable_syscall();
1050 /* This function is not called for HVM domains */
1051 void __init xen_arch_setup(void)
1053 xen_panic_handler_init();
1054 if (!xen_feature(XENFEAT_auto_translated_physmap))
1055 xen_pvmmu_arch_setup();
1058 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1059 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1064 memcpy(boot_command_line, xen_start_info->cmd_line,
1065 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1066 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1068 /* Set up idle, making sure it calls safe_halt() pvop */
1071 WARN_ON(xen_set_default_idle());