Loading...
1/*
2 * Machine specific setup for xen
3 *
4 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
5 */
6
7#include <linux/module.h>
8#include <linux/sched.h>
9#include <linux/mm.h>
10#include <linux/pm.h>
11#include <linux/memblock.h>
12#include <linux/cpuidle.h>
13#include <linux/cpufreq.h>
14
15#include <asm/elf.h>
16#include <asm/vdso.h>
17#include <asm/e820.h>
18#include <asm/setup.h>
19#include <asm/acpi.h>
20#include <asm/numa.h>
21#include <asm/xen/hypervisor.h>
22#include <asm/xen/hypercall.h>
23
24#include <xen/xen.h>
25#include <xen/page.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-ops.h"
31#include "vdso.h"
32
33/* These are code, but not functions. Defined in entry.S */
34extern const char xen_hypervisor_callback[];
35extern const char xen_failsafe_callback[];
36extern void xen_sysenter_target(void);
37extern void xen_syscall_target(void);
38extern void xen_syscall32_target(void);
39
40/* Amount of extra memory space we add to the e820 ranges */
41struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
42
43/* Number of pages released from the initial allocation. */
44unsigned long xen_released_pages;
45
46/*
47 * The maximum amount of extra memory compared to the base size. The
48 * main scaling factor is the size of struct page. At extreme ratios
49 * of base:extra, all the base memory can be filled with page
50 * structures for the extra memory, leaving no space for anything
51 * else.
52 *
53 * 10x seems like a reasonable balance between scaling flexibility and
54 * leaving a practically usable system.
55 */
56#define EXTRA_MEM_RATIO (10)
57
58static void __init xen_add_extra_mem(u64 start, u64 size)
59{
60 unsigned long pfn;
61 int i;
62
63 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
64 /* Add new region. */
65 if (xen_extra_mem[i].size == 0) {
66 xen_extra_mem[i].start = start;
67 xen_extra_mem[i].size = size;
68 break;
69 }
70 /* Append to existing region. */
71 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
72 xen_extra_mem[i].size += size;
73 break;
74 }
75 }
76 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
77 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
78
79 memblock_reserve(start, size);
80
81 xen_max_p2m_pfn = PFN_DOWN(start + size);
82 for (pfn = PFN_DOWN(start); pfn < xen_max_p2m_pfn; pfn++) {
83 unsigned long mfn = pfn_to_mfn(pfn);
84
85 if (WARN(mfn == pfn, "Trying to over-write 1-1 mapping (pfn: %lx)\n", pfn))
86 continue;
87 WARN(mfn != INVALID_P2M_ENTRY, "Trying to remove %lx which has %lx mfn!\n",
88 pfn, mfn);
89
90 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
91 }
92}
93
94static unsigned long __init xen_do_chunk(unsigned long start,
95 unsigned long end, bool release)
96{
97 struct xen_memory_reservation reservation = {
98 .address_bits = 0,
99 .extent_order = 0,
100 .domid = DOMID_SELF
101 };
102 unsigned long len = 0;
103 unsigned long pfn;
104 int ret;
105
106 for (pfn = start; pfn < end; pfn++) {
107 unsigned long frame;
108 unsigned long mfn = pfn_to_mfn(pfn);
109
110 if (release) {
111 /* Make sure pfn exists to start with */
112 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
113 continue;
114 frame = mfn;
115 } else {
116 if (mfn != INVALID_P2M_ENTRY)
117 continue;
118 frame = pfn;
119 }
120 set_xen_guest_handle(reservation.extent_start, &frame);
121 reservation.nr_extents = 1;
122
123 ret = HYPERVISOR_memory_op(release ? XENMEM_decrease_reservation : XENMEM_populate_physmap,
124 &reservation);
125 WARN(ret != 1, "Failed to %s pfn %lx err=%d\n",
126 release ? "release" : "populate", pfn, ret);
127
128 if (ret == 1) {
129 if (!early_set_phys_to_machine(pfn, release ? INVALID_P2M_ENTRY : frame)) {
130 if (release)
131 break;
132 set_xen_guest_handle(reservation.extent_start, &frame);
133 reservation.nr_extents = 1;
134 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
135 &reservation);
136 break;
137 }
138 len++;
139 } else
140 break;
141 }
142 if (len)
143 printk(KERN_INFO "%s %lx-%lx pfn range: %lu pages %s\n",
144 release ? "Freeing" : "Populating",
145 start, end, len,
146 release ? "freed" : "added");
147
148 return len;
149}
150
151static unsigned long __init xen_release_chunk(unsigned long start,
152 unsigned long end)
153{
154 return xen_do_chunk(start, end, true);
155}
156
157static unsigned long __init xen_populate_chunk(
158 const struct e820entry *list, size_t map_size,
159 unsigned long max_pfn, unsigned long *last_pfn,
160 unsigned long credits_left)
161{
162 const struct e820entry *entry;
163 unsigned int i;
164 unsigned long done = 0;
165 unsigned long dest_pfn;
166
167 for (i = 0, entry = list; i < map_size; i++, entry++) {
168 unsigned long credits = credits_left;
169 unsigned long s_pfn;
170 unsigned long e_pfn;
171 unsigned long pfns;
172 long capacity;
173
174 if (credits <= 0)
175 break;
176
177 if (entry->type != E820_RAM)
178 continue;
179
180 e_pfn = PFN_UP(entry->addr + entry->size);
181
182 /* We only care about E820 after the xen_start_info->nr_pages */
183 if (e_pfn <= max_pfn)
184 continue;
185
186 s_pfn = PFN_DOWN(entry->addr);
187 /* If the E820 falls within the nr_pages, we want to start
188 * at the nr_pages PFN.
189 * If that would mean going past the E820 entry, skip it
190 */
191 if (s_pfn <= max_pfn) {
192 capacity = e_pfn - max_pfn;
193 dest_pfn = max_pfn;
194 } else {
195 /* last_pfn MUST be within E820_RAM regions */
196 if (*last_pfn && e_pfn >= *last_pfn)
197 s_pfn = *last_pfn;
198 capacity = e_pfn - s_pfn;
199 dest_pfn = s_pfn;
200 }
201 /* If we had filled this E820_RAM entry, go to the next one. */
202 if (capacity <= 0)
203 continue;
204
205 if (credits > capacity)
206 credits = capacity;
207
208 pfns = xen_do_chunk(dest_pfn, dest_pfn + credits, false);
209 done += pfns;
210 credits_left -= pfns;
211 *last_pfn = (dest_pfn + pfns);
212 }
213 return done;
214}
215
216static void __init xen_set_identity_and_release_chunk(
217 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
218 unsigned long *released, unsigned long *identity)
219{
220 unsigned long pfn;
221
222 /*
223 * If the PFNs are currently mapped, the VA mapping also needs
224 * to be updated to be 1:1.
225 */
226 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
227 (void)HYPERVISOR_update_va_mapping(
228 (unsigned long)__va(pfn << PAGE_SHIFT),
229 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
230
231 if (start_pfn < nr_pages)
232 *released += xen_release_chunk(
233 start_pfn, min(end_pfn, nr_pages));
234
235 *identity += set_phys_range_identity(start_pfn, end_pfn);
236}
237
238static unsigned long __init xen_set_identity_and_release(
239 const struct e820entry *list, size_t map_size, unsigned long nr_pages)
240{
241 phys_addr_t start = 0;
242 unsigned long released = 0;
243 unsigned long identity = 0;
244 const struct e820entry *entry;
245 int i;
246
247 /*
248 * Combine non-RAM regions and gaps until a RAM region (or the
249 * end of the map) is reached, then set the 1:1 map and
250 * release the pages (if available) in those non-RAM regions.
251 *
252 * The combined non-RAM regions are rounded to a whole number
253 * of pages so any partial pages are accessible via the 1:1
254 * mapping. This is needed for some BIOSes that put (for
255 * example) the DMI tables in a reserved region that begins on
256 * a non-page boundary.
257 */
258 for (i = 0, entry = list; i < map_size; i++, entry++) {
259 phys_addr_t end = entry->addr + entry->size;
260 if (entry->type == E820_RAM || i == map_size - 1) {
261 unsigned long start_pfn = PFN_DOWN(start);
262 unsigned long end_pfn = PFN_UP(end);
263
264 if (entry->type == E820_RAM)
265 end_pfn = PFN_UP(entry->addr);
266
267 if (start_pfn < end_pfn)
268 xen_set_identity_and_release_chunk(
269 start_pfn, end_pfn, nr_pages,
270 &released, &identity);
271
272 start = end;
273 }
274 }
275
276 if (released)
277 printk(KERN_INFO "Released %lu pages of unused memory\n", released);
278 if (identity)
279 printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
280
281 return released;
282}
283
284static unsigned long __init xen_get_max_pages(void)
285{
286 unsigned long max_pages = MAX_DOMAIN_PAGES;
287 domid_t domid = DOMID_SELF;
288 int ret;
289
290 /*
291 * For the initial domain we use the maximum reservation as
292 * the maximum page.
293 *
294 * For guest domains the current maximum reservation reflects
295 * the current maximum rather than the static maximum. In this
296 * case the e820 map provided to us will cover the static
297 * maximum region.
298 */
299 if (xen_initial_domain()) {
300 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
301 if (ret > 0)
302 max_pages = ret;
303 }
304
305 return min(max_pages, MAX_DOMAIN_PAGES);
306}
307
308static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
309{
310 u64 end = start + size;
311
312 /* Align RAM regions to page boundaries. */
313 if (type == E820_RAM) {
314 start = PAGE_ALIGN(start);
315 end &= ~((u64)PAGE_SIZE - 1);
316 }
317
318 e820_add_region(start, end - start, type);
319}
320
321/**
322 * machine_specific_memory_setup - Hook for machine specific memory setup.
323 **/
324char * __init xen_memory_setup(void)
325{
326 static struct e820entry map[E820MAX] __initdata;
327
328 unsigned long max_pfn = xen_start_info->nr_pages;
329 unsigned long long mem_end;
330 int rc;
331 struct xen_memory_map memmap;
332 unsigned long max_pages;
333 unsigned long last_pfn = 0;
334 unsigned long extra_pages = 0;
335 unsigned long populated;
336 int i;
337 int op;
338
339 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
340 mem_end = PFN_PHYS(max_pfn);
341
342 memmap.nr_entries = E820MAX;
343 set_xen_guest_handle(memmap.buffer, map);
344
345 op = xen_initial_domain() ?
346 XENMEM_machine_memory_map :
347 XENMEM_memory_map;
348 rc = HYPERVISOR_memory_op(op, &memmap);
349 if (rc == -ENOSYS) {
350 BUG_ON(xen_initial_domain());
351 memmap.nr_entries = 1;
352 map[0].addr = 0ULL;
353 map[0].size = mem_end;
354 /* 8MB slack (to balance backend allocations). */
355 map[0].size += 8ULL << 20;
356 map[0].type = E820_RAM;
357 rc = 0;
358 }
359 BUG_ON(rc);
360
361 /* Make sure the Xen-supplied memory map is well-ordered. */
362 sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
363
364 max_pages = xen_get_max_pages();
365 if (max_pages > max_pfn)
366 extra_pages += max_pages - max_pfn;
367
368 /*
369 * Set P2M for all non-RAM pages and E820 gaps to be identity
370 * type PFNs. Any RAM pages that would be made inaccesible by
371 * this are first released.
372 */
373 xen_released_pages = xen_set_identity_and_release(
374 map, memmap.nr_entries, max_pfn);
375
376 /*
377 * Populate back the non-RAM pages and E820 gaps that had been
378 * released. */
379 populated = xen_populate_chunk(map, memmap.nr_entries,
380 max_pfn, &last_pfn, xen_released_pages);
381
382 xen_released_pages -= populated;
383 extra_pages += xen_released_pages;
384
385 if (last_pfn > max_pfn) {
386 max_pfn = min(MAX_DOMAIN_PAGES, last_pfn);
387 mem_end = PFN_PHYS(max_pfn);
388 }
389 /*
390 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
391 * factor the base size. On non-highmem systems, the base
392 * size is the full initial memory allocation; on highmem it
393 * is limited to the max size of lowmem, so that it doesn't
394 * get completely filled.
395 *
396 * In principle there could be a problem in lowmem systems if
397 * the initial memory is also very large with respect to
398 * lowmem, but we won't try to deal with that here.
399 */
400 extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
401 extra_pages);
402 i = 0;
403 while (i < memmap.nr_entries) {
404 u64 addr = map[i].addr;
405 u64 size = map[i].size;
406 u32 type = map[i].type;
407
408 if (type == E820_RAM) {
409 if (addr < mem_end) {
410 size = min(size, mem_end - addr);
411 } else if (extra_pages) {
412 size = min(size, (u64)extra_pages * PAGE_SIZE);
413 extra_pages -= size / PAGE_SIZE;
414 xen_add_extra_mem(addr, size);
415 } else
416 type = E820_UNUSABLE;
417 }
418
419 xen_align_and_add_e820_region(addr, size, type);
420
421 map[i].addr += size;
422 map[i].size -= size;
423 if (map[i].size == 0)
424 i++;
425 }
426
427 /*
428 * In domU, the ISA region is normal, usable memory, but we
429 * reserve ISA memory anyway because too many things poke
430 * about in there.
431 */
432 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
433 E820_RESERVED);
434
435 /*
436 * Reserve Xen bits:
437 * - mfn_list
438 * - xen_start_info
439 * See comment above "struct start_info" in <xen/interface/xen.h>
440 */
441 memblock_reserve(__pa(xen_start_info->mfn_list),
442 xen_start_info->pt_base - xen_start_info->mfn_list);
443
444 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
445
446 return "Xen";
447}
448
449/*
450 * Set the bit indicating "nosegneg" library variants should be used.
451 * We only need to bother in pure 32-bit mode; compat 32-bit processes
452 * can have un-truncated segments, so wrapping around is allowed.
453 */
454static void __init fiddle_vdso(void)
455{
456#ifdef CONFIG_X86_32
457 u32 *mask;
458 mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
459 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
460 mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
461 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
462#endif
463}
464
465static int __cpuinit register_callback(unsigned type, const void *func)
466{
467 struct callback_register callback = {
468 .type = type,
469 .address = XEN_CALLBACK(__KERNEL_CS, func),
470 .flags = CALLBACKF_mask_events,
471 };
472
473 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
474}
475
476void __cpuinit xen_enable_sysenter(void)
477{
478 int ret;
479 unsigned sysenter_feature;
480
481#ifdef CONFIG_X86_32
482 sysenter_feature = X86_FEATURE_SEP;
483#else
484 sysenter_feature = X86_FEATURE_SYSENTER32;
485#endif
486
487 if (!boot_cpu_has(sysenter_feature))
488 return;
489
490 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
491 if(ret != 0)
492 setup_clear_cpu_cap(sysenter_feature);
493}
494
495void __cpuinit xen_enable_syscall(void)
496{
497#ifdef CONFIG_X86_64
498 int ret;
499
500 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
501 if (ret != 0) {
502 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
503 /* Pretty fatal; 64-bit userspace has no other
504 mechanism for syscalls. */
505 }
506
507 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
508 ret = register_callback(CALLBACKTYPE_syscall32,
509 xen_syscall32_target);
510 if (ret != 0)
511 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
512 }
513#endif /* CONFIG_X86_64 */
514}
515
516void __init xen_arch_setup(void)
517{
518 xen_panic_handler_init();
519
520 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
521 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
522
523 if (!xen_feature(XENFEAT_auto_translated_physmap))
524 HYPERVISOR_vm_assist(VMASST_CMD_enable,
525 VMASST_TYPE_pae_extended_cr3);
526
527 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
528 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
529 BUG();
530
531 xen_enable_sysenter();
532 xen_enable_syscall();
533
534#ifdef CONFIG_ACPI
535 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
536 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
537 disable_acpi();
538 }
539#endif
540
541 memcpy(boot_command_line, xen_start_info->cmd_line,
542 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
543 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
544
545 /* Set up idle, making sure it calls safe_halt() pvop */
546#ifdef CONFIG_X86_32
547 boot_cpu_data.hlt_works_ok = 1;
548#endif
549 disable_cpuidle();
550 disable_cpufreq();
551 WARN_ON(set_pm_idle_to_default());
552 fiddle_vdso();
553#ifdef CONFIG_NUMA
554 numa_off = 1;
555#endif
556}
1/*
2 * Machine specific setup for xen
3 *
4 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
5 */
6
7#include <linux/module.h>
8#include <linux/sched.h>
9#include <linux/mm.h>
10#include <linux/pm.h>
11#include <linux/memblock.h>
12#include <linux/cpuidle.h>
13#include <linux/cpufreq.h>
14
15#include <asm/elf.h>
16#include <asm/vdso.h>
17#include <asm/e820.h>
18#include <asm/setup.h>
19#include <asm/acpi.h>
20#include <asm/numa.h>
21#include <asm/xen/hypervisor.h>
22#include <asm/xen/hypercall.h>
23
24#include <xen/xen.h>
25#include <xen/page.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>
31#include "xen-ops.h"
32#include "vdso.h"
33#include "mmu.h"
34
35#define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
36
37/* Amount of extra memory space we add to the e820 ranges */
38struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
39
40/* Number of pages released from the initial allocation. */
41unsigned long xen_released_pages;
42
43/* E820 map used during setting up memory. */
44static struct e820entry xen_e820_map[E820MAX] __initdata;
45static u32 xen_e820_map_entries __initdata;
46
47/*
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
50 * buffer pages.
51 */
52#define REMAP_SIZE (P2M_PER_PAGE - 3)
53static struct {
54 unsigned long next_area_mfn;
55 unsigned long target_pfn;
56 unsigned long size;
57 unsigned long mfns[REMAP_SIZE];
58} xen_remap_buf __initdata __aligned(PAGE_SIZE);
59static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
60
61/*
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
66 * else.
67 *
68 * 10x seems like a reasonable balance between scaling flexibility and
69 * leaving a practically usable system.
70 */
71#define EXTRA_MEM_RATIO (10)
72
73static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
74
75static void __init xen_parse_512gb(void)
76{
77 bool val = false;
78 char *arg;
79
80 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
81 if (!arg)
82 return;
83
84 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
85 if (!arg)
86 val = true;
87 else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
88 return;
89
90 xen_512gb_limit = val;
91}
92
93static void __init xen_add_extra_mem(unsigned long start_pfn,
94 unsigned long n_pfns)
95{
96 int i;
97
98 /*
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.
101 */
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;
107 break;
108 }
109 /* Append to existing region. */
110 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
111 start_pfn) {
112 xen_extra_mem[i].n_pfns += n_pfns;
113 break;
114 }
115 }
116 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
117 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
118
119 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
120}
121
122static void __init xen_del_extra_mem(unsigned long start_pfn,
123 unsigned long n_pfns)
124{
125 int i;
126 unsigned long start_r, size_r;
127
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;
131
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;
137 break;
138 }
139 /* End of region. */
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;
143 break;
144 }
145 /* Mid of region. */
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));
152 break;
153 }
154 }
155 memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
156}
157
158/*
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
161 * invalid.
162 */
163unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
164{
165 int i;
166
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;
171 }
172
173 return IDENTITY_FRAME(pfn);
174}
175
176/*
177 * Mark all pfns of extra mem as invalid in p2m list.
178 */
179void __init xen_inv_extra_mem(void)
180{
181 unsigned long pfn, pfn_s, pfn_e;
182 int i;
183
184 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
185 if (!xen_extra_mem[i].n_pfns)
186 continue;
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);
191 }
192}
193
194/*
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.
198 */
199static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
200{
201 const struct e820entry *entry = xen_e820_map;
202 unsigned int i;
203 unsigned long done = 0;
204
205 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
206 unsigned long s_pfn;
207 unsigned long e_pfn;
208
209 if (entry->type != E820_RAM)
210 continue;
211
212 e_pfn = PFN_DOWN(entry->addr + entry->size);
213
214 /* We only care about E820 after this */
215 if (e_pfn <= *min_pfn)
216 continue;
217
218 s_pfn = PFN_UP(entry->addr);
219
220 /* If min_pfn falls within the E820 entry, we want to start
221 * at the min_pfn PFN.
222 */
223 if (s_pfn <= *min_pfn) {
224 done = e_pfn - *min_pfn;
225 } else {
226 done = e_pfn - s_pfn;
227 *min_pfn = s_pfn;
228 }
229 break;
230 }
231
232 return done;
233}
234
235static int __init xen_free_mfn(unsigned long mfn)
236{
237 struct xen_memory_reservation reservation = {
238 .address_bits = 0,
239 .extent_order = 0,
240 .domid = DOMID_SELF
241 };
242
243 set_xen_guest_handle(reservation.extent_start, &mfn);
244 reservation.nr_extents = 1;
245
246 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
247}
248
249/*
250 * This releases a chunk of memory and then does the identity map. It's used
251 * as a fallback if the remapping fails.
252 */
253static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
254 unsigned long end_pfn, unsigned long nr_pages)
255{
256 unsigned long pfn, end;
257 int ret;
258
259 WARN_ON(start_pfn > end_pfn);
260
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);
265
266 /* Make sure pfn exists to start with */
267 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
268 continue;
269
270 ret = xen_free_mfn(mfn);
271 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
272
273 if (ret == 1) {
274 xen_released_pages++;
275 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
276 break;
277 } else
278 break;
279 }
280
281 set_phys_range_identity(start_pfn, end_pfn);
282}
283
284/*
285 * Helper function to update the p2m and m2p tables and kernel mapping.
286 */
287static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
288{
289 struct mmu_update update = {
290 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
291 .val = pfn
292 };
293
294 /* Update p2m */
295 if (!set_phys_to_machine(pfn, mfn)) {
296 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
297 pfn, mfn);
298 BUG();
299 }
300
301 /* Update m2p */
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",
304 mfn, pfn);
305 BUG();
306 }
307
308 /* Update kernel mapping, but not for highmem. */
309 if (pfn >= PFN_UP(__pa(high_memory - 1)))
310 return;
311
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",
315 mfn, pfn);
316 BUG();
317 }
318}
319
320/*
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
330 * its callers.
331 */
332static void __init xen_do_set_identity_and_remap_chunk(
333 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
334{
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;
341
342 WARN_ON(size == 0);
343
344 BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
345
346 mfn_save = virt_to_mfn(buf);
347
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;
352
353 /* Map first pfn to xen_remap_buf */
354 mfn = pfn_to_mfn(ident_pfn_iter);
355 set_pte_mfn(buf, mfn, PAGE_KERNEL);
356
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);
363
364 /* Put remap buf into list. */
365 xen_remap_mfn = mfn;
366
367 /* Set identity map */
368 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
369
370 left -= chunk;
371 }
372
373 /* Restore old xen_remap_buf mapping */
374 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
375}
376
377/*
378 * This function takes a contiguous pfn range that needs to be identity mapped
379 * and:
380 *
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.
383 *
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.
387 */
388static 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)
391{
392 unsigned long pfn;
393 unsigned long i = 0;
394 unsigned long n = end_pfn - start_pfn;
395
396 while (i < n) {
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;
401
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);
406 break;
407 }
408 if (cur_pfn + size > nr_pages)
409 size = nr_pages - cur_pfn;
410
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);
416 break;
417 }
418 /* Adjust size to fit in current e820 RAM region */
419 if (size > remap_range_size)
420 size = remap_range_size;
421
422 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
423
424 /* Update variables to reflect new mappings. */
425 i += size;
426 remap_pfn += size;
427 }
428
429 /*
430 * If the PFNs are currently mapped, the VA mapping also needs
431 * to be updated to be 1:1.
432 */
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);
437
438 return remap_pfn;
439}
440
441static void __init xen_set_identity_and_remap(unsigned long nr_pages)
442{
443 phys_addr_t start = 0;
444 unsigned long last_pfn = nr_pages;
445 const struct e820entry *entry = xen_e820_map;
446 int i;
447
448 /*
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.
452 *
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.
458 */
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);
464
465 if (entry->type == E820_RAM)
466 end_pfn = PFN_UP(entry->addr);
467
468 if (start_pfn < end_pfn)
469 last_pfn = xen_set_identity_and_remap_chunk(
470 start_pfn, end_pfn, nr_pages,
471 last_pfn);
472 start = end;
473 }
474 }
475
476 pr_info("Released %ld page(s)\n", xen_released_pages);
477}
478
479/*
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.
485 */
486void __init xen_remap_memory(void)
487{
488 unsigned long buf = (unsigned long)&xen_remap_buf;
489 unsigned long mfn_save, mfn, pfn;
490 unsigned long remapped = 0;
491 unsigned int i;
492 unsigned long pfn_s = ~0UL;
493 unsigned long len = 0;
494
495 mfn_save = virt_to_mfn(buf);
496
497 while (xen_remap_mfn != INVALID_P2M_ENTRY) {
498 /* Map the remap information */
499 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
500
501 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
502
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);
507 remapped++;
508 pfn++;
509 }
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;
515 } else {
516 xen_del_extra_mem(pfn_s, len);
517 pfn_s = xen_remap_buf.target_pfn;
518 len = xen_remap_buf.size;
519 }
520
521 mfn = xen_remap_mfn;
522 xen_remap_mfn = xen_remap_buf.next_area_mfn;
523 }
524
525 if (pfn_s != ~0UL && len)
526 xen_del_extra_mem(pfn_s, len);
527
528 set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
529
530 pr_info("Remapped %ld page(s)\n", remapped);
531}
532
533static unsigned long __init xen_get_pages_limit(void)
534{
535 unsigned long limit;
536
537#ifdef CONFIG_X86_32
538 limit = GB(64) / PAGE_SIZE;
539#else
540 limit = MAXMEM / PAGE_SIZE;
541 if (!xen_initial_domain() && xen_512gb_limit)
542 limit = GB(512) / PAGE_SIZE;
543#endif
544 return limit;
545}
546
547static unsigned long __init xen_get_max_pages(void)
548{
549 unsigned long max_pages, limit;
550 domid_t domid = DOMID_SELF;
551 long ret;
552
553 limit = xen_get_pages_limit();
554 max_pages = limit;
555
556 /*
557 * For the initial domain we use the maximum reservation as
558 * the maximum page.
559 *
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
563 * maximum region.
564 */
565 if (xen_initial_domain()) {
566 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
567 if (ret > 0)
568 max_pages = ret;
569 }
570
571 return min(max_pages, limit);
572}
573
574static void __init xen_align_and_add_e820_region(phys_addr_t start,
575 phys_addr_t size, int type)
576{
577 phys_addr_t end = start + size;
578
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);
583 }
584
585 e820_add_region(start, end - start, type);
586}
587
588static void __init xen_ignore_unusable(void)
589{
590 struct e820entry *entry = xen_e820_map;
591 unsigned int i;
592
593 for (i = 0; i < xen_e820_map_entries; i++, entry++) {
594 if (entry->type == E820_UNUSABLE)
595 entry->type = E820_RAM;
596 }
597}
598
599static unsigned long __init xen_count_remap_pages(unsigned long max_pfn)
600{
601 unsigned long extra = 0;
602 unsigned long start_pfn, end_pfn;
603 const struct e820entry *entry = xen_e820_map;
604 int i;
605
606 end_pfn = 0;
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);
611
612 if (start_pfn >= max_pfn)
613 return extra + max_pfn - end_pfn;
614
615 /* Add any holes in map to result. */
616 extra += start_pfn - end_pfn;
617
618 end_pfn = PFN_UP(entry->addr + entry->size);
619 end_pfn = min(end_pfn, max_pfn);
620
621 if (entry->type != E820_RAM)
622 extra += end_pfn - start_pfn;
623 }
624
625 return extra;
626}
627
628bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
629{
630 struct e820entry *entry;
631 unsigned mapcnt;
632 phys_addr_t end;
633
634 if (!size)
635 return false;
636
637 end = start + size;
638 entry = xen_e820_map;
639
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)
643 return false;
644
645 entry++;
646 }
647
648 return true;
649}
650
651/*
652 * Find a free area in physical memory not yet reserved and compliant with
653 * E820 map.
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.
658 */
659phys_addr_t __init xen_find_free_area(phys_addr_t size)
660{
661 unsigned mapcnt;
662 phys_addr_t addr, start;
663 struct e820entry *entry = xen_e820_map;
664
665 for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++, entry++) {
666 if (entry->type != E820_RAM || entry->size < size)
667 continue;
668 start = entry->addr;
669 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
670 if (!memblock_is_reserved(addr))
671 continue;
672 start = addr + PAGE_SIZE;
673 if (start + size > entry->addr + entry->size)
674 break;
675 }
676 if (addr >= start + size) {
677 memblock_reserve(start, size);
678 return start;
679 }
680 }
681
682 return 0;
683}
684
685/*
686 * Like memcpy, but with physical addresses for dest and src.
687 */
688static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
689 phys_addr_t n)
690{
691 phys_addr_t dest_off, src_off, dest_len, src_len, len;
692 void *from, *to;
693
694 while (n) {
695 dest_off = dest & ~PAGE_MASK;
696 src_off = src & ~PAGE_MASK;
697 dest_len = n;
698 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
699 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
700 src_len = n;
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);
709 n -= len;
710 dest += len;
711 src += len;
712 }
713}
714
715/*
716 * Reserve Xen mfn_list.
717 */
718static void __init xen_reserve_xen_mfnlist(void)
719{
720 phys_addr_t start, size;
721
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));
726 } else {
727 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
728 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
729 }
730
731 if (!xen_is_e820_reserved(start, size)) {
732 memblock_reserve(start, size);
733 return;
734 }
735
736#ifdef CONFIG_X86_32
737 /*
738 * Relocating the p2m on 32 bit system to an arbitrary virtual address
739 * is not supported, so just give up.
740 */
741 xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n");
742 BUG();
743#else
744 xen_relocate_p2m();
745#endif
746}
747
748/**
749 * machine_specific_memory_setup - Hook for machine specific memory setup.
750 **/
751char * __init xen_memory_setup(void)
752{
753 unsigned long max_pfn, pfn_s, n_pfns;
754 phys_addr_t mem_end, addr, size, chunk_size;
755 u32 type;
756 int rc;
757 struct xen_memory_map memmap;
758 unsigned long max_pages;
759 unsigned long extra_pages = 0;
760 int i;
761 int op;
762
763 xen_parse_512gb();
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);
767
768 memmap.nr_entries = E820MAX;
769 set_xen_guest_handle(memmap.buffer, xen_e820_map);
770
771 op = xen_initial_domain() ?
772 XENMEM_machine_memory_map :
773 XENMEM_memory_map;
774 rc = HYPERVISOR_memory_op(op, &memmap);
775 if (rc == -ENOSYS) {
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;
783 rc = 0;
784 }
785 BUG_ON(rc);
786 BUG_ON(memmap.nr_entries == 0);
787 xen_e820_map_entries = memmap.nr_entries;
788
789 /*
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.
793 *
794 * UNUSABLE regions in domUs are not handled and will need
795 * a patch in the future.
796 */
797 if (xen_initial_domain())
798 xen_ignore_unusable();
799
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);
803
804 max_pages = xen_get_max_pages();
805
806 /* How many extra pages do we need due to remapping? */
807 max_pages += xen_count_remap_pages(max_pfn);
808
809 if (max_pages > max_pfn)
810 extra_pages += max_pages - max_pfn;
811
812 /*
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.
818 *
819 * Make sure we have no memory above max_pages, as this area
820 * isn't handled by the p2m management.
821 *
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.
825 */
826 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
827 extra_pages, max_pages - max_pfn);
828 i = 0;
829 addr = xen_e820_map[0].addr;
830 size = xen_e820_map[0].size;
831 while (i < xen_e820_map_entries) {
832 bool discard = false;
833
834 chunk_size = size;
835 type = xen_e820_map[i].type;
836
837 if (type == E820_RAM) {
838 if (addr < mem_end) {
839 chunk_size = min(size, mem_end - addr);
840 } else if (extra_pages) {
841 chunk_size = min(size, PFN_PHYS(extra_pages));
842 pfn_s = PFN_UP(addr);
843 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
844 extra_pages -= n_pfns;
845 xen_add_extra_mem(pfn_s, n_pfns);
846 xen_max_p2m_pfn = pfn_s + n_pfns;
847 } else
848 discard = true;
849 }
850
851 if (!discard)
852 xen_align_and_add_e820_region(addr, chunk_size, type);
853
854 addr += chunk_size;
855 size -= chunk_size;
856 if (size == 0) {
857 i++;
858 if (i < xen_e820_map_entries) {
859 addr = xen_e820_map[i].addr;
860 size = xen_e820_map[i].size;
861 }
862 }
863 }
864
865 /*
866 * Set the rest as identity mapped, in case PCI BARs are
867 * located here.
868 */
869 set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
870
871 /*
872 * In domU, the ISA region is normal, usable memory, but we
873 * reserve ISA memory anyway because too many things poke
874 * about in there.
875 */
876 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
877 E820_RESERVED);
878
879 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
880
881 /*
882 * Check whether the kernel itself conflicts with the target E820 map.
883 * Failing now is better than running into weird problems later due
884 * to relocating (and even reusing) pages with kernel text or data.
885 */
886 if (xen_is_e820_reserved(__pa_symbol(_text),
887 __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
888 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
889 BUG();
890 }
891
892 /*
893 * Check for a conflict of the hypervisor supplied page tables with
894 * the target E820 map.
895 */
896 xen_pt_check_e820();
897
898 xen_reserve_xen_mfnlist();
899
900 /* Check for a conflict of the initrd with the target E820 map. */
901 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
902 boot_params.hdr.ramdisk_size)) {
903 phys_addr_t new_area, start, size;
904
905 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
906 if (!new_area) {
907 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
908 BUG();
909 }
910
911 start = boot_params.hdr.ramdisk_image;
912 size = boot_params.hdr.ramdisk_size;
913 xen_phys_memcpy(new_area, start, size);
914 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
915 start, start + size, new_area, new_area + size);
916 memblock_free(start, size);
917 boot_params.hdr.ramdisk_image = new_area;
918 boot_params.ext_ramdisk_image = new_area >> 32;
919 }
920
921 /*
922 * Set identity map on non-RAM pages and prepare remapping the
923 * underlying RAM.
924 */
925 xen_set_identity_and_remap(max_pfn);
926
927 return "Xen";
928}
929
930/*
931 * Machine specific memory setup for auto-translated guests.
932 */
933char * __init xen_auto_xlated_memory_setup(void)
934{
935 struct xen_memory_map memmap;
936 int i;
937 int rc;
938
939 memmap.nr_entries = E820MAX;
940 set_xen_guest_handle(memmap.buffer, xen_e820_map);
941
942 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
943 if (rc < 0)
944 panic("No memory map (%d)\n", rc);
945
946 xen_e820_map_entries = memmap.nr_entries;
947
948 sanitize_e820_map(xen_e820_map, ARRAY_SIZE(xen_e820_map),
949 &xen_e820_map_entries);
950
951 for (i = 0; i < xen_e820_map_entries; i++)
952 e820_add_region(xen_e820_map[i].addr, xen_e820_map[i].size,
953 xen_e820_map[i].type);
954
955 /* Remove p2m info, it is not needed. */
956 xen_start_info->mfn_list = 0;
957 xen_start_info->first_p2m_pfn = 0;
958 xen_start_info->nr_p2m_frames = 0;
959
960 return "Xen";
961}
962
963/*
964 * Set the bit indicating "nosegneg" library variants should be used.
965 * We only need to bother in pure 32-bit mode; compat 32-bit processes
966 * can have un-truncated segments, so wrapping around is allowed.
967 */
968static void __init fiddle_vdso(void)
969{
970#ifdef CONFIG_X86_32
971 u32 *mask = vdso_image_32.data +
972 vdso_image_32.sym_VDSO32_NOTE_MASK;
973 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
974#endif
975}
976
977static int register_callback(unsigned type, const void *func)
978{
979 struct callback_register callback = {
980 .type = type,
981 .address = XEN_CALLBACK(__KERNEL_CS, func),
982 .flags = CALLBACKF_mask_events,
983 };
984
985 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
986}
987
988void xen_enable_sysenter(void)
989{
990 int ret;
991 unsigned sysenter_feature;
992
993#ifdef CONFIG_X86_32
994 sysenter_feature = X86_FEATURE_SEP;
995#else
996 sysenter_feature = X86_FEATURE_SYSENTER32;
997#endif
998
999 if (!boot_cpu_has(sysenter_feature))
1000 return;
1001
1002 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
1003 if(ret != 0)
1004 setup_clear_cpu_cap(sysenter_feature);
1005}
1006
1007void xen_enable_syscall(void)
1008{
1009#ifdef CONFIG_X86_64
1010 int ret;
1011
1012 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
1013 if (ret != 0) {
1014 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
1015 /* Pretty fatal; 64-bit userspace has no other
1016 mechanism for syscalls. */
1017 }
1018
1019 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
1020 ret = register_callback(CALLBACKTYPE_syscall32,
1021 xen_syscall32_target);
1022 if (ret != 0)
1023 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
1024 }
1025#endif /* CONFIG_X86_64 */
1026}
1027
1028void __init xen_pvmmu_arch_setup(void)
1029{
1030 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
1031 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
1032
1033 HYPERVISOR_vm_assist(VMASST_CMD_enable,
1034 VMASST_TYPE_pae_extended_cr3);
1035
1036 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
1037 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
1038 BUG();
1039
1040 xen_enable_sysenter();
1041 xen_enable_syscall();
1042}
1043
1044/* This function is not called for HVM domains */
1045void __init xen_arch_setup(void)
1046{
1047 xen_panic_handler_init();
1048 if (!xen_feature(XENFEAT_auto_translated_physmap))
1049 xen_pvmmu_arch_setup();
1050
1051#ifdef CONFIG_ACPI
1052 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1053 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1054 disable_acpi();
1055 }
1056#endif
1057
1058 memcpy(boot_command_line, xen_start_info->cmd_line,
1059 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1060 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1061
1062 /* Set up idle, making sure it calls safe_halt() pvop */
1063 disable_cpuidle();
1064 disable_cpufreq();
1065 WARN_ON(xen_set_default_idle());
1066 fiddle_vdso();
1067#ifdef CONFIG_NUMA
1068 numa_off = 1;
1069#endif
1070}