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 */
34 extern const char xen_hypervisor_callback[];
35 extern const char xen_failsafe_callback[];
36 extern void xen_sysenter_target(void);
37 extern void xen_syscall_target(void);
38 extern void xen_syscall32_target(void);
39
40 /* Amount of extra memory space we add to the e820 ranges */
41 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
42
43 /* Number of pages released from the initial allocation. */
44 unsigned 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
xen_add_extra_mem(u64 start,u64 size)58 static 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
xen_do_chunk(unsigned long start,unsigned long end,bool release)94 static 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
xen_release_chunk(unsigned long start,unsigned long end)151 static unsigned long __init xen_release_chunk(unsigned long start,
152 unsigned long end)
153 {
154 return xen_do_chunk(start, end, true);
155 }
156
xen_populate_chunk(const struct e820entry * list,size_t map_size,unsigned long max_pfn,unsigned long * last_pfn,unsigned long credits_left)157 static 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 s_pfn;
169 unsigned long e_pfn;
170 unsigned long pfns;
171 long capacity;
172
173 if (credits_left <= 0)
174 break;
175
176 if (entry->type != E820_RAM)
177 continue;
178
179 e_pfn = PFN_DOWN(entry->addr + entry->size);
180
181 /* We only care about E820 after the xen_start_info->nr_pages */
182 if (e_pfn <= max_pfn)
183 continue;
184
185 s_pfn = PFN_UP(entry->addr);
186 /* If the E820 falls within the nr_pages, we want to start
187 * at the nr_pages PFN.
188 * If that would mean going past the E820 entry, skip it
189 */
190 if (s_pfn <= max_pfn) {
191 capacity = e_pfn - max_pfn;
192 dest_pfn = max_pfn;
193 } else {
194 capacity = e_pfn - s_pfn;
195 dest_pfn = s_pfn;
196 }
197
198 if (credits_left < capacity)
199 capacity = credits_left;
200
201 pfns = xen_do_chunk(dest_pfn, dest_pfn + capacity, false);
202 done += pfns;
203 *last_pfn = (dest_pfn + pfns);
204 if (pfns < capacity)
205 break;
206 credits_left -= pfns;
207 }
208 return done;
209 }
210
xen_set_identity_and_release_chunk(unsigned long start_pfn,unsigned long end_pfn,unsigned long nr_pages,unsigned long * released,unsigned long * identity)211 static void __init xen_set_identity_and_release_chunk(
212 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
213 unsigned long *released, unsigned long *identity)
214 {
215 unsigned long pfn;
216
217 /*
218 * If the PFNs are currently mapped, the VA mapping also needs
219 * to be updated to be 1:1.
220 */
221 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
222 (void)HYPERVISOR_update_va_mapping(
223 (unsigned long)__va(pfn << PAGE_SHIFT),
224 mfn_pte(pfn, PAGE_KERNEL_IO), 0);
225
226 if (start_pfn < nr_pages)
227 *released += xen_release_chunk(
228 start_pfn, min(end_pfn, nr_pages));
229
230 *identity += set_phys_range_identity(start_pfn, end_pfn);
231 }
232
xen_set_identity_and_release(const struct e820entry * list,size_t map_size,unsigned long nr_pages)233 static unsigned long __init xen_set_identity_and_release(
234 const struct e820entry *list, size_t map_size, unsigned long nr_pages)
235 {
236 phys_addr_t start = 0;
237 unsigned long released = 0;
238 unsigned long identity = 0;
239 const struct e820entry *entry;
240 int i;
241
242 /*
243 * Combine non-RAM regions and gaps until a RAM region (or the
244 * end of the map) is reached, then set the 1:1 map and
245 * release the pages (if available) in those non-RAM regions.
246 *
247 * The combined non-RAM regions are rounded to a whole number
248 * of pages so any partial pages are accessible via the 1:1
249 * mapping. This is needed for some BIOSes that put (for
250 * example) the DMI tables in a reserved region that begins on
251 * a non-page boundary.
252 */
253 for (i = 0, entry = list; i < map_size; i++, entry++) {
254 phys_addr_t end = entry->addr + entry->size;
255 if (entry->type == E820_RAM || i == map_size - 1) {
256 unsigned long start_pfn = PFN_DOWN(start);
257 unsigned long end_pfn = PFN_UP(end);
258
259 if (entry->type == E820_RAM)
260 end_pfn = PFN_UP(entry->addr);
261
262 if (start_pfn < end_pfn)
263 xen_set_identity_and_release_chunk(
264 start_pfn, end_pfn, nr_pages,
265 &released, &identity);
266
267 start = end;
268 }
269 }
270
271 if (released)
272 printk(KERN_INFO "Released %lu pages of unused memory\n", released);
273 if (identity)
274 printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
275
276 return released;
277 }
278
xen_get_max_pages(void)279 static unsigned long __init xen_get_max_pages(void)
280 {
281 unsigned long max_pages = MAX_DOMAIN_PAGES;
282 domid_t domid = DOMID_SELF;
283 int ret;
284
285 /*
286 * For the initial domain we use the maximum reservation as
287 * the maximum page.
288 *
289 * For guest domains the current maximum reservation reflects
290 * the current maximum rather than the static maximum. In this
291 * case the e820 map provided to us will cover the static
292 * maximum region.
293 */
294 if (xen_initial_domain()) {
295 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
296 if (ret > 0)
297 max_pages = ret;
298 }
299
300 return min(max_pages, MAX_DOMAIN_PAGES);
301 }
302
xen_align_and_add_e820_region(u64 start,u64 size,int type)303 static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
304 {
305 u64 end = start + size;
306
307 /* Align RAM regions to page boundaries. */
308 if (type == E820_RAM) {
309 start = PAGE_ALIGN(start);
310 end &= ~((u64)PAGE_SIZE - 1);
311 }
312
313 e820_add_region(start, end - start, type);
314 }
315
316 /**
317 * machine_specific_memory_setup - Hook for machine specific memory setup.
318 **/
xen_memory_setup(void)319 char * __init xen_memory_setup(void)
320 {
321 static struct e820entry map[E820MAX] __initdata;
322
323 unsigned long max_pfn = xen_start_info->nr_pages;
324 unsigned long long mem_end;
325 int rc;
326 struct xen_memory_map memmap;
327 unsigned long max_pages;
328 unsigned long last_pfn = 0;
329 unsigned long extra_pages = 0;
330 unsigned long populated;
331 int i;
332 int op;
333
334 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
335 mem_end = PFN_PHYS(max_pfn);
336
337 memmap.nr_entries = E820MAX;
338 set_xen_guest_handle(memmap.buffer, map);
339
340 op = xen_initial_domain() ?
341 XENMEM_machine_memory_map :
342 XENMEM_memory_map;
343 rc = HYPERVISOR_memory_op(op, &memmap);
344 if (rc == -ENOSYS) {
345 BUG_ON(xen_initial_domain());
346 memmap.nr_entries = 1;
347 map[0].addr = 0ULL;
348 map[0].size = mem_end;
349 /* 8MB slack (to balance backend allocations). */
350 map[0].size += 8ULL << 20;
351 map[0].type = E820_RAM;
352 rc = 0;
353 }
354 BUG_ON(rc);
355
356 /* Make sure the Xen-supplied memory map is well-ordered. */
357 sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
358
359 max_pages = xen_get_max_pages();
360 if (max_pages > max_pfn)
361 extra_pages += max_pages - max_pfn;
362
363 /*
364 * Set P2M for all non-RAM pages and E820 gaps to be identity
365 * type PFNs. Any RAM pages that would be made inaccesible by
366 * this are first released.
367 */
368 xen_released_pages = xen_set_identity_and_release(
369 map, memmap.nr_entries, max_pfn);
370
371 /*
372 * Populate back the non-RAM pages and E820 gaps that had been
373 * released. */
374 populated = xen_populate_chunk(map, memmap.nr_entries,
375 max_pfn, &last_pfn, xen_released_pages);
376
377 xen_released_pages -= populated;
378 extra_pages += xen_released_pages;
379
380 if (last_pfn > max_pfn) {
381 max_pfn = min(MAX_DOMAIN_PAGES, last_pfn);
382 mem_end = PFN_PHYS(max_pfn);
383 }
384 /*
385 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
386 * factor the base size. On non-highmem systems, the base
387 * size is the full initial memory allocation; on highmem it
388 * is limited to the max size of lowmem, so that it doesn't
389 * get completely filled.
390 *
391 * In principle there could be a problem in lowmem systems if
392 * the initial memory is also very large with respect to
393 * lowmem, but we won't try to deal with that here.
394 */
395 extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
396 extra_pages);
397 i = 0;
398 while (i < memmap.nr_entries) {
399 u64 addr = map[i].addr;
400 u64 size = map[i].size;
401 u32 type = map[i].type;
402
403 if (type == E820_RAM) {
404 if (addr < mem_end) {
405 size = min(size, mem_end - addr);
406 } else if (extra_pages) {
407 size = min(size, (u64)extra_pages * PAGE_SIZE);
408 extra_pages -= size / PAGE_SIZE;
409 xen_add_extra_mem(addr, size);
410 } else
411 type = E820_UNUSABLE;
412 }
413
414 xen_align_and_add_e820_region(addr, size, type);
415
416 map[i].addr += size;
417 map[i].size -= size;
418 if (map[i].size == 0)
419 i++;
420 }
421
422 /*
423 * In domU, the ISA region is normal, usable memory, but we
424 * reserve ISA memory anyway because too many things poke
425 * about in there.
426 */
427 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
428 E820_RESERVED);
429
430 /*
431 * Reserve Xen bits:
432 * - mfn_list
433 * - xen_start_info
434 * See comment above "struct start_info" in <xen/interface/xen.h>
435 * We tried to make the the memblock_reserve more selective so
436 * that it would be clear what region is reserved. Sadly we ran
437 * in the problem wherein on a 64-bit hypervisor with a 32-bit
438 * initial domain, the pt_base has the cr3 value which is not
439 * neccessarily where the pagetable starts! As Jan put it: "
440 * Actually, the adjustment turns out to be correct: The page
441 * tables for a 32-on-64 dom0 get allocated in the order "first L1",
442 * "first L2", "first L3", so the offset to the page table base is
443 * indeed 2. When reading xen/include/public/xen.h's comment
444 * very strictly, this is not a violation (since there nothing is said
445 * that the first thing in the page table space is pointed to by
446 * pt_base; I admit that this seems to be implied though, namely
447 * do I think that it is implied that the page table space is the
448 * range [pt_base, pt_base + nt_pt_frames), whereas that
449 * range here indeed is [pt_base - 2, pt_base - 2 + nt_pt_frames),
450 * which - without a priori knowledge - the kernel would have
451 * difficulty to figure out)." - so lets just fall back to the
452 * easy way and reserve the whole region.
453 */
454 memblock_reserve(__pa(xen_start_info->mfn_list),
455 xen_start_info->pt_base - xen_start_info->mfn_list);
456
457 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
458
459 return "Xen";
460 }
461
462 /*
463 * Set the bit indicating "nosegneg" library variants should be used.
464 * We only need to bother in pure 32-bit mode; compat 32-bit processes
465 * can have un-truncated segments, so wrapping around is allowed.
466 */
fiddle_vdso(void)467 static void __init fiddle_vdso(void)
468 {
469 #ifdef CONFIG_X86_32
470 u32 *mask;
471 mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
472 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
473 mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
474 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
475 #endif
476 }
477
register_callback(unsigned type,const void * func)478 static int __cpuinit register_callback(unsigned type, const void *func)
479 {
480 struct callback_register callback = {
481 .type = type,
482 .address = XEN_CALLBACK(__KERNEL_CS, func),
483 .flags = CALLBACKF_mask_events,
484 };
485
486 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
487 }
488
xen_enable_sysenter(void)489 void __cpuinit xen_enable_sysenter(void)
490 {
491 int ret;
492 unsigned sysenter_feature;
493
494 #ifdef CONFIG_X86_32
495 sysenter_feature = X86_FEATURE_SEP;
496 #else
497 sysenter_feature = X86_FEATURE_SYSENTER32;
498 #endif
499
500 if (!boot_cpu_has(sysenter_feature))
501 return;
502
503 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
504 if(ret != 0)
505 setup_clear_cpu_cap(sysenter_feature);
506 }
507
xen_enable_syscall(void)508 void __cpuinit xen_enable_syscall(void)
509 {
510 #ifdef CONFIG_X86_64
511 int ret;
512
513 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
514 if (ret != 0) {
515 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
516 /* Pretty fatal; 64-bit userspace has no other
517 mechanism for syscalls. */
518 }
519
520 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
521 ret = register_callback(CALLBACKTYPE_syscall32,
522 xen_syscall32_target);
523 if (ret != 0)
524 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
525 }
526 #endif /* CONFIG_X86_64 */
527 }
528
xen_arch_setup(void)529 void __init xen_arch_setup(void)
530 {
531 xen_panic_handler_init();
532
533 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
534 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
535
536 if (!xen_feature(XENFEAT_auto_translated_physmap))
537 HYPERVISOR_vm_assist(VMASST_CMD_enable,
538 VMASST_TYPE_pae_extended_cr3);
539
540 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
541 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
542 BUG();
543
544 xen_enable_sysenter();
545 xen_enable_syscall();
546
547 #ifdef CONFIG_ACPI
548 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
549 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
550 disable_acpi();
551 }
552 #endif
553
554 memcpy(boot_command_line, xen_start_info->cmd_line,
555 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
556 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
557
558 /* Set up idle, making sure it calls safe_halt() pvop */
559 disable_cpuidle();
560 disable_cpufreq();
561 WARN_ON(xen_set_default_idle());
562 fiddle_vdso();
563 #ifdef CONFIG_NUMA
564 numa_off = 1;
565 #endif
566 }
567