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>
33 /* These are code, but not functions. Defined in entry.S */
34 extern const char xen_hypervisor_callback[];
35 extern const char xen_failsafe_callback[];
37 extern asmlinkage void nmi(void);
39 extern void xen_sysenter_target(void);
40 extern void xen_syscall_target(void);
41 extern void xen_syscall32_target(void);
43 /* Amount of extra memory space we add to the e820 ranges */
44 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
46 /* Number of pages released from the initial allocation. */
47 unsigned long xen_released_pages;
50 * The maximum amount of extra memory compared to the base size. The
51 * main scaling factor is the size of struct page. At extreme ratios
52 * of base:extra, all the base memory can be filled with page
53 * structures for the extra memory, leaving no space for anything
56 * 10x seems like a reasonable balance between scaling flexibility and
57 * leaving a practically usable system.
59 #define EXTRA_MEM_RATIO (10)
61 static void __init xen_add_extra_mem(u64 start, u64 size)
66 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
68 if (xen_extra_mem[i].size == 0) {
69 xen_extra_mem[i].start = start;
70 xen_extra_mem[i].size = size;
73 /* Append to existing region. */
74 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
75 xen_extra_mem[i].size += size;
79 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
80 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
82 memblock_reserve(start, size);
84 xen_max_p2m_pfn = PFN_DOWN(start + size);
85 for (pfn = PFN_DOWN(start); pfn < xen_max_p2m_pfn; pfn++) {
86 unsigned long mfn = pfn_to_mfn(pfn);
88 if (WARN_ONCE(mfn == pfn, "Trying to over-write 1-1 mapping (pfn: %lx)\n", pfn))
90 WARN_ONCE(mfn != INVALID_P2M_ENTRY, "Trying to remove %lx which has %lx mfn!\n",
93 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
97 static unsigned long __init xen_do_chunk(unsigned long start,
98 unsigned long end, bool release)
100 struct xen_memory_reservation reservation = {
105 unsigned long len = 0;
109 for (pfn = start; pfn < end; pfn++) {
111 unsigned long mfn = pfn_to_mfn(pfn);
114 /* Make sure pfn exists to start with */
115 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
119 if (mfn != INVALID_P2M_ENTRY)
123 set_xen_guest_handle(reservation.extent_start, &frame);
124 reservation.nr_extents = 1;
126 ret = HYPERVISOR_memory_op(release ? XENMEM_decrease_reservation : XENMEM_populate_physmap,
128 WARN(ret != 1, "Failed to %s pfn %lx err=%d\n",
129 release ? "release" : "populate", pfn, ret);
132 if (!early_set_phys_to_machine(pfn, release ? INVALID_P2M_ENTRY : frame)) {
135 set_xen_guest_handle(reservation.extent_start, &frame);
136 reservation.nr_extents = 1;
137 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
146 printk(KERN_INFO "%s %lx-%lx pfn range: %lu pages %s\n",
147 release ? "Freeing" : "Populating",
149 release ? "freed" : "added");
154 static unsigned long __init xen_release_chunk(unsigned long start,
157 return xen_do_chunk(start, end, true);
160 static unsigned long __init xen_populate_chunk(
161 const struct e820entry *list, size_t map_size,
162 unsigned long max_pfn, unsigned long *last_pfn,
163 unsigned long credits_left)
165 const struct e820entry *entry;
167 unsigned long done = 0;
168 unsigned long dest_pfn;
170 for (i = 0, entry = list; i < map_size; i++, entry++) {
176 if (credits_left <= 0)
179 if (entry->type != E820_RAM)
182 e_pfn = PFN_DOWN(entry->addr + entry->size);
184 /* We only care about E820 after the xen_start_info->nr_pages */
185 if (e_pfn <= max_pfn)
188 s_pfn = PFN_UP(entry->addr);
189 /* If the E820 falls within the nr_pages, we want to start
190 * at the nr_pages PFN.
191 * If that would mean going past the E820 entry, skip it
193 if (s_pfn <= max_pfn) {
194 capacity = e_pfn - max_pfn;
197 capacity = e_pfn - s_pfn;
201 if (credits_left < capacity)
202 capacity = credits_left;
204 pfns = xen_do_chunk(dest_pfn, dest_pfn + capacity, false);
206 *last_pfn = (dest_pfn + pfns);
209 credits_left -= pfns;
214 static void __init xen_set_identity_and_release_chunk(
215 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
216 unsigned long *released, unsigned long *identity)
221 * If the PFNs are currently mapped, clear the mappings
222 * (except for the ISA region which must be 1:1 mapped) to
223 * release the refcounts (in Xen) on the original frames.
225 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++) {
226 pte_t pte = __pte_ma(0);
228 if (pfn < PFN_UP(ISA_END_ADDRESS))
229 pte = mfn_pte(pfn, PAGE_KERNEL_IO);
231 (void)HYPERVISOR_update_va_mapping(
232 (unsigned long)__va(pfn << PAGE_SHIFT), pte, 0);
235 if (start_pfn < nr_pages)
236 *released += xen_release_chunk(
237 start_pfn, min(end_pfn, nr_pages));
239 *identity += set_phys_range_identity(start_pfn, end_pfn);
242 static unsigned long __init xen_set_identity_and_release(
243 const struct e820entry *list, size_t map_size, unsigned long nr_pages)
245 phys_addr_t start = 0;
246 unsigned long released = 0;
247 unsigned long identity = 0;
248 const struct e820entry *entry;
252 * Combine non-RAM regions and gaps until a RAM region (or the
253 * end of the map) is reached, then set the 1:1 map and
254 * release the pages (if available) in those non-RAM regions.
256 * The combined non-RAM regions are rounded to a whole number
257 * of pages so any partial pages are accessible via the 1:1
258 * mapping. This is needed for some BIOSes that put (for
259 * example) the DMI tables in a reserved region that begins on
260 * a non-page boundary.
262 for (i = 0, entry = list; i < map_size; i++, entry++) {
263 phys_addr_t end = entry->addr + entry->size;
264 if (entry->type == E820_RAM || i == map_size - 1) {
265 unsigned long start_pfn = PFN_DOWN(start);
266 unsigned long end_pfn = PFN_UP(end);
268 if (entry->type == E820_RAM)
269 end_pfn = PFN_UP(entry->addr);
271 if (start_pfn < end_pfn)
272 xen_set_identity_and_release_chunk(
273 start_pfn, end_pfn, nr_pages,
274 &released, &identity);
281 printk(KERN_INFO "Released %lu pages of unused memory\n", released);
283 printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
288 static unsigned long __init xen_get_max_pages(void)
290 unsigned long max_pages = MAX_DOMAIN_PAGES;
291 domid_t domid = DOMID_SELF;
295 * For the initial domain we use the maximum reservation as
298 * For guest domains the current maximum reservation reflects
299 * the current maximum rather than the static maximum. In this
300 * case the e820 map provided to us will cover the static
303 if (xen_initial_domain()) {
304 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
309 return min(max_pages, MAX_DOMAIN_PAGES);
312 static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
314 u64 end = start + size;
316 /* Align RAM regions to page boundaries. */
317 if (type == E820_RAM) {
318 start = PAGE_ALIGN(start);
319 end &= ~((u64)PAGE_SIZE - 1);
322 e820_add_region(start, end - start, type);
325 void xen_ignore_unusable(struct e820entry *list, size_t map_size)
327 struct e820entry *entry;
330 for (i = 0, entry = list; i < map_size; i++, entry++) {
331 if (entry->type == E820_UNUSABLE)
332 entry->type = E820_RAM;
337 * machine_specific_memory_setup - Hook for machine specific memory setup.
339 char * __init xen_memory_setup(void)
341 static struct e820entry map[E820MAX] __initdata;
343 unsigned long max_pfn = xen_start_info->nr_pages;
344 unsigned long long mem_end;
346 struct xen_memory_map memmap;
347 unsigned long max_pages;
348 unsigned long last_pfn = 0;
349 unsigned long extra_pages = 0;
350 unsigned long populated;
354 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
355 mem_end = PFN_PHYS(max_pfn);
357 memmap.nr_entries = E820MAX;
358 set_xen_guest_handle(memmap.buffer, map);
360 op = xen_initial_domain() ?
361 XENMEM_machine_memory_map :
363 rc = HYPERVISOR_memory_op(op, &memmap);
365 BUG_ON(xen_initial_domain());
366 memmap.nr_entries = 1;
368 map[0].size = mem_end;
369 /* 8MB slack (to balance backend allocations). */
370 map[0].size += 8ULL << 20;
371 map[0].type = E820_RAM;
377 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
378 * regions, so if we're using the machine memory map leave the
379 * region as RAM as it is in the pseudo-physical map.
381 * UNUSABLE regions in domUs are not handled and will need
382 * a patch in the future.
384 if (xen_initial_domain())
385 xen_ignore_unusable(map, memmap.nr_entries);
387 /* Make sure the Xen-supplied memory map is well-ordered. */
388 sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
390 max_pages = xen_get_max_pages();
391 if (max_pages > max_pfn)
392 extra_pages += max_pages - max_pfn;
395 * Set P2M for all non-RAM pages and E820 gaps to be identity
396 * type PFNs. Any RAM pages that would be made inaccesible by
397 * this are first released.
399 xen_released_pages = xen_set_identity_and_release(
400 map, memmap.nr_entries, max_pfn);
403 * Populate back the non-RAM pages and E820 gaps that had been
405 populated = xen_populate_chunk(map, memmap.nr_entries,
406 max_pfn, &last_pfn, xen_released_pages);
408 xen_released_pages -= populated;
409 extra_pages += xen_released_pages;
411 if (last_pfn > max_pfn) {
412 max_pfn = min(MAX_DOMAIN_PAGES, last_pfn);
413 mem_end = PFN_PHYS(max_pfn);
416 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
417 * factor the base size. On non-highmem systems, the base
418 * size is the full initial memory allocation; on highmem it
419 * is limited to the max size of lowmem, so that it doesn't
420 * get completely filled.
422 * In principle there could be a problem in lowmem systems if
423 * the initial memory is also very large with respect to
424 * lowmem, but we won't try to deal with that here.
426 extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
429 while (i < memmap.nr_entries) {
430 u64 addr = map[i].addr;
431 u64 size = map[i].size;
432 u32 type = map[i].type;
434 if (type == E820_RAM) {
435 if (addr < mem_end) {
436 size = min(size, mem_end - addr);
437 } else if (extra_pages) {
438 size = min(size, (u64)extra_pages * PAGE_SIZE);
439 extra_pages -= size / PAGE_SIZE;
440 xen_add_extra_mem(addr, size);
442 type = E820_UNUSABLE;
445 xen_align_and_add_e820_region(addr, size, type);
449 if (map[i].size == 0)
454 * Set the rest as identity mapped, in case PCI BARs are
457 * PFNs above MAX_P2M_PFN are considered identity mapped as
460 set_phys_range_identity(map[i-1].addr / PAGE_SIZE, ~0ul);
463 * In domU, the ISA region is normal, usable memory, but we
464 * reserve ISA memory anyway because too many things poke
467 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
474 * See comment above "struct start_info" in <xen/interface/xen.h>
475 * We tried to make the the memblock_reserve more selective so
476 * that it would be clear what region is reserved. Sadly we ran
477 * in the problem wherein on a 64-bit hypervisor with a 32-bit
478 * initial domain, the pt_base has the cr3 value which is not
479 * neccessarily where the pagetable starts! As Jan put it: "
480 * Actually, the adjustment turns out to be correct: The page
481 * tables for a 32-on-64 dom0 get allocated in the order "first L1",
482 * "first L2", "first L3", so the offset to the page table base is
483 * indeed 2. When reading xen/include/public/xen.h's comment
484 * very strictly, this is not a violation (since there nothing is said
485 * that the first thing in the page table space is pointed to by
486 * pt_base; I admit that this seems to be implied though, namely
487 * do I think that it is implied that the page table space is the
488 * range [pt_base, pt_base + nt_pt_frames), whereas that
489 * range here indeed is [pt_base - 2, pt_base - 2 + nt_pt_frames),
490 * which - without a priori knowledge - the kernel would have
491 * difficulty to figure out)." - so lets just fall back to the
492 * easy way and reserve the whole region.
494 memblock_reserve(__pa(xen_start_info->mfn_list),
495 xen_start_info->pt_base - xen_start_info->mfn_list);
497 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
503 * Machine specific memory setup for auto-translated guests.
505 char * __init xen_auto_xlated_memory_setup(void)
507 static struct e820entry map[E820MAX] __initdata;
509 struct xen_memory_map memmap;
513 memmap.nr_entries = E820MAX;
514 set_xen_guest_handle(memmap.buffer, map);
516 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
518 panic("No memory map (%d)\n", rc);
520 sanitize_e820_map(map, ARRAY_SIZE(map), &memmap.nr_entries);
522 for (i = 0; i < memmap.nr_entries; i++)
523 e820_add_region(map[i].addr, map[i].size, map[i].type);
525 memblock_reserve(__pa(xen_start_info->mfn_list),
526 xen_start_info->pt_base - xen_start_info->mfn_list);
532 * Set the bit indicating "nosegneg" library variants should be used.
533 * We only need to bother in pure 32-bit mode; compat 32-bit processes
534 * can have un-truncated segments, so wrapping around is allowed.
536 static void __init fiddle_vdso(void)
540 * This could be called before selected_vdso32 is initialized, so
541 * just fiddle with both possible images. vdso_image_32_syscall
542 * can't be selected, since it only exists on 64-bit systems.
545 mask = vdso_image_32_int80.data +
546 vdso_image_32_int80.sym_VDSO32_NOTE_MASK;
547 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
548 mask = vdso_image_32_sysenter.data +
549 vdso_image_32_sysenter.sym_VDSO32_NOTE_MASK;
550 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
554 static int register_callback(unsigned type, const void *func)
556 struct callback_register callback = {
558 .address = XEN_CALLBACK(__KERNEL_CS, func),
559 .flags = CALLBACKF_mask_events,
562 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
565 void xen_enable_sysenter(void)
568 unsigned sysenter_feature;
571 sysenter_feature = X86_FEATURE_SEP;
573 sysenter_feature = X86_FEATURE_SYSENTER32;
576 if (!boot_cpu_has(sysenter_feature))
579 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
581 setup_clear_cpu_cap(sysenter_feature);
584 void xen_enable_syscall(void)
589 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
591 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
592 /* Pretty fatal; 64-bit userspace has no other
593 mechanism for syscalls. */
596 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
597 ret = register_callback(CALLBACKTYPE_syscall32,
598 xen_syscall32_target);
600 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
602 #endif /* CONFIG_X86_64 */
605 void __init xen_pvmmu_arch_setup(void)
607 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
608 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
610 HYPERVISOR_vm_assist(VMASST_CMD_enable,
611 VMASST_TYPE_pae_extended_cr3);
613 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
614 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
617 xen_enable_sysenter();
618 xen_enable_syscall();
621 /* This function is not called for HVM domains */
622 void __init xen_arch_setup(void)
624 xen_panic_handler_init();
625 if (!xen_feature(XENFEAT_auto_translated_physmap))
626 xen_pvmmu_arch_setup();
629 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
630 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
635 memcpy(boot_command_line, xen_start_info->cmd_line,
636 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
637 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
639 /* Set up idle, making sure it calls safe_halt() pvop */
642 WARN_ON(xen_set_default_idle());