~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/xen/setup.c

Version: ~ [ linux-5.13-rc1 ] ~ [ linux-5.12.2 ] ~ [ linux-5.11.19 ] ~ [ linux-5.10.35 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.117 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 #include "p2m.h"
 33 #include "mmu.h"
 34 
 35 /* Amount of extra memory space we add to the e820 ranges */
 36 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
 37 
 38 /* Number of pages released from the initial allocation. */
 39 unsigned long xen_released_pages;
 40 
 41 /*
 42  * Buffer used to remap identity mapped pages. We only need the virtual space.
 43  * The physical page behind this address is remapped as needed to different
 44  * buffer pages.
 45  */
 46 #define REMAP_SIZE      (P2M_PER_PAGE - 3)
 47 static struct {
 48         unsigned long   next_area_mfn;
 49         unsigned long   target_pfn;
 50         unsigned long   size;
 51         unsigned long   mfns[REMAP_SIZE];
 52 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
 53 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
 54 
 55 /* 
 56  * The maximum amount of extra memory compared to the base size.  The
 57  * main scaling factor is the size of struct page.  At extreme ratios
 58  * of base:extra, all the base memory can be filled with page
 59  * structures for the extra memory, leaving no space for anything
 60  * else.
 61  * 
 62  * 10x seems like a reasonable balance between scaling flexibility and
 63  * leaving a practically usable system.
 64  */
 65 #define EXTRA_MEM_RATIO         (10)
 66 
 67 static void __init xen_add_extra_mem(phys_addr_t start, phys_addr_t size)
 68 {
 69         int i;
 70 
 71         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 72                 /* Add new region. */
 73                 if (xen_extra_mem[i].size == 0) {
 74                         xen_extra_mem[i].start = start;
 75                         xen_extra_mem[i].size  = size;
 76                         break;
 77                 }
 78                 /* Append to existing region. */
 79                 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
 80                         xen_extra_mem[i].size += size;
 81                         break;
 82                 }
 83         }
 84         if (i == XEN_EXTRA_MEM_MAX_REGIONS)
 85                 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 86 
 87         memblock_reserve(start, size);
 88 }
 89 
 90 static void __init xen_del_extra_mem(phys_addr_t start, phys_addr_t size)
 91 {
 92         int i;
 93         phys_addr_t start_r, size_r;
 94 
 95         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 96                 start_r = xen_extra_mem[i].start;
 97                 size_r = xen_extra_mem[i].size;
 98 
 99                 /* Start of region. */
100                 if (start_r == start) {
101                         BUG_ON(size > size_r);
102                         xen_extra_mem[i].start += size;
103                         xen_extra_mem[i].size -= size;
104                         break;
105                 }
106                 /* End of region. */
107                 if (start_r + size_r == start + size) {
108                         BUG_ON(size > size_r);
109                         xen_extra_mem[i].size -= size;
110                         break;
111                 }
112                 /* Mid of region. */
113                 if (start > start_r && start < start_r + size_r) {
114                         BUG_ON(start + size > start_r + size_r);
115                         xen_extra_mem[i].size = start - start_r;
116                         /* Calling memblock_reserve() again is okay. */
117                         xen_add_extra_mem(start + size, start_r + size_r -
118                                           (start + size));
119                         break;
120                 }
121         }
122         memblock_free(start, size);
123 }
124 
125 /*
126  * Called during boot before the p2m list can take entries beyond the
127  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
128  * invalid.
129  */
130 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
131 {
132         int i;
133         phys_addr_t addr = PFN_PHYS(pfn);
134 
135         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
136                 if (addr >= xen_extra_mem[i].start &&
137                     addr < xen_extra_mem[i].start + xen_extra_mem[i].size)
138                         return INVALID_P2M_ENTRY;
139         }
140 
141         return IDENTITY_FRAME(pfn);
142 }
143 
144 /*
145  * Mark all pfns of extra mem as invalid in p2m list.
146  */
147 void __init xen_inv_extra_mem(void)
148 {
149         unsigned long pfn, pfn_s, pfn_e;
150         int i;
151 
152         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
153                 if (!xen_extra_mem[i].size)
154                         continue;
155                 pfn_s = PFN_DOWN(xen_extra_mem[i].start);
156                 pfn_e = PFN_UP(xen_extra_mem[i].start + xen_extra_mem[i].size);
157                 for (pfn = pfn_s; pfn < pfn_e; pfn++)
158                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
159         }
160 }
161 
162 /*
163  * Finds the next RAM pfn available in the E820 map after min_pfn.
164  * This function updates min_pfn with the pfn found and returns
165  * the size of that range or zero if not found.
166  */
167 static unsigned long __init xen_find_pfn_range(
168         const struct e820entry *list, size_t map_size,
169         unsigned long *min_pfn)
170 {
171         const struct e820entry *entry;
172         unsigned int i;
173         unsigned long done = 0;
174 
175         for (i = 0, entry = list; i < map_size; i++, entry++) {
176                 unsigned long s_pfn;
177                 unsigned long e_pfn;
178 
179                 if (entry->type != E820_RAM)
180                         continue;
181 
182                 e_pfn = PFN_DOWN(entry->addr + entry->size);
183 
184                 /* We only care about E820 after this */
185                 if (e_pfn < *min_pfn)
186                         continue;
187 
188                 s_pfn = PFN_UP(entry->addr);
189 
190                 /* If min_pfn falls within the E820 entry, we want to start
191                  * at the min_pfn PFN.
192                  */
193                 if (s_pfn <= *min_pfn) {
194                         done = e_pfn - *min_pfn;
195                 } else {
196                         done = e_pfn - s_pfn;
197                         *min_pfn = s_pfn;
198                 }
199                 break;
200         }
201 
202         return done;
203 }
204 
205 static int __init xen_free_mfn(unsigned long mfn)
206 {
207         struct xen_memory_reservation reservation = {
208                 .address_bits = 0,
209                 .extent_order = 0,
210                 .domid        = DOMID_SELF
211         };
212 
213         set_xen_guest_handle(reservation.extent_start, &mfn);
214         reservation.nr_extents = 1;
215 
216         return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
217 }
218 
219 /*
220  * This releases a chunk of memory and then does the identity map. It's used
221  * as a fallback if the remapping fails.
222  */
223 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
224         unsigned long end_pfn, unsigned long nr_pages, unsigned long *released)
225 {
226         unsigned long pfn, end;
227         int ret;
228 
229         WARN_ON(start_pfn > end_pfn);
230 
231         /* Release pages first. */
232         end = min(end_pfn, nr_pages);
233         for (pfn = start_pfn; pfn < end; pfn++) {
234                 unsigned long mfn = pfn_to_mfn(pfn);
235 
236                 /* Make sure pfn exists to start with */
237                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
238                         continue;
239 
240                 ret = xen_free_mfn(mfn);
241                 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
242 
243                 if (ret == 1) {
244                         (*released)++;
245                         if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
246                                 break;
247                 } else
248                         break;
249         }
250 
251         set_phys_range_identity(start_pfn, end_pfn);
252 }
253 
254 /*
255  * Helper function to update the p2m and m2p tables and kernel mapping.
256  */
257 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
258 {
259         struct mmu_update update = {
260                 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
261                 .val = pfn
262         };
263 
264         /* Update p2m */
265         if (!set_phys_to_machine(pfn, mfn)) {
266                 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
267                      pfn, mfn);
268                 BUG();
269         }
270 
271         /* Update m2p */
272         if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
273                 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
274                      mfn, pfn);
275                 BUG();
276         }
277 
278         /* Update kernel mapping, but not for highmem. */
279         if (pfn >= PFN_UP(__pa(high_memory - 1)))
280                 return;
281 
282         if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
283                                          mfn_pte(mfn, PAGE_KERNEL), 0)) {
284                 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
285                       mfn, pfn);
286                 BUG();
287         }
288 }
289 
290 /*
291  * This function updates the p2m and m2p tables with an identity map from
292  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
293  * original allocation at remap_pfn. The information needed for remapping is
294  * saved in the memory itself to avoid the need for allocating buffers. The
295  * complete remap information is contained in a list of MFNs each containing
296  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
297  * This enables us to preserve the original mfn sequence while doing the
298  * remapping at a time when the memory management is capable of allocating
299  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
300  * its callers.
301  */
302 static void __init xen_do_set_identity_and_remap_chunk(
303         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
304 {
305         unsigned long buf = (unsigned long)&xen_remap_buf;
306         unsigned long mfn_save, mfn;
307         unsigned long ident_pfn_iter, remap_pfn_iter;
308         unsigned long ident_end_pfn = start_pfn + size;
309         unsigned long left = size;
310         unsigned int i, chunk;
311 
312         WARN_ON(size == 0);
313 
314         BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
315 
316         mfn_save = virt_to_mfn(buf);
317 
318         for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
319              ident_pfn_iter < ident_end_pfn;
320              ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
321                 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
322 
323                 /* Map first pfn to xen_remap_buf */
324                 mfn = pfn_to_mfn(ident_pfn_iter);
325                 set_pte_mfn(buf, mfn, PAGE_KERNEL);
326 
327                 /* Save mapping information in page */
328                 xen_remap_buf.next_area_mfn = xen_remap_mfn;
329                 xen_remap_buf.target_pfn = remap_pfn_iter;
330                 xen_remap_buf.size = chunk;
331                 for (i = 0; i < chunk; i++)
332                         xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
333 
334                 /* Put remap buf into list. */
335                 xen_remap_mfn = mfn;
336 
337                 /* Set identity map */
338                 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
339 
340                 left -= chunk;
341         }
342 
343         /* Restore old xen_remap_buf mapping */
344         set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
345 }
346 
347 /*
348  * This function takes a contiguous pfn range that needs to be identity mapped
349  * and:
350  *
351  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
352  *  2) Calls the do_ function to actually do the mapping/remapping work.
353  *
354  * The goal is to not allocate additional memory but to remap the existing
355  * pages. In the case of an error the underlying memory is simply released back
356  * to Xen and not remapped.
357  */
358 static unsigned long __init xen_set_identity_and_remap_chunk(
359         const struct e820entry *list, size_t map_size, unsigned long start_pfn,
360         unsigned long end_pfn, unsigned long nr_pages, unsigned long remap_pfn,
361         unsigned long *released, unsigned long *remapped)
362 {
363         unsigned long pfn;
364         unsigned long i = 0;
365         unsigned long n = end_pfn - start_pfn;
366 
367         while (i < n) {
368                 unsigned long cur_pfn = start_pfn + i;
369                 unsigned long left = n - i;
370                 unsigned long size = left;
371                 unsigned long remap_range_size;
372 
373                 /* Do not remap pages beyond the current allocation */
374                 if (cur_pfn >= nr_pages) {
375                         /* Identity map remaining pages */
376                         set_phys_range_identity(cur_pfn, cur_pfn + size);
377                         break;
378                 }
379                 if (cur_pfn + size > nr_pages)
380                         size = nr_pages - cur_pfn;
381 
382                 remap_range_size = xen_find_pfn_range(list, map_size,
383                                                       &remap_pfn);
384                 if (!remap_range_size) {
385                         pr_warning("Unable to find available pfn range, not remapping identity pages\n");
386                         xen_set_identity_and_release_chunk(cur_pfn,
387                                 cur_pfn + left, nr_pages, released);
388                         break;
389                 }
390                 /* Adjust size to fit in current e820 RAM region */
391                 if (size > remap_range_size)
392                         size = remap_range_size;
393 
394                 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
395 
396                 /* Update variables to reflect new mappings. */
397                 i += size;
398                 remap_pfn += size;
399                 *remapped += size;
400         }
401 
402         /*
403          * If the PFNs are currently mapped, the VA mapping also needs
404          * to be updated to be 1:1.
405          */
406         for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
407                 (void)HYPERVISOR_update_va_mapping(
408                         (unsigned long)__va(pfn << PAGE_SHIFT),
409                         mfn_pte(pfn, PAGE_KERNEL_IO), 0);
410 
411         return remap_pfn;
412 }
413 
414 static void __init xen_set_identity_and_remap(
415         const struct e820entry *list, size_t map_size, unsigned long nr_pages,
416         unsigned long *released, unsigned long *remapped)
417 {
418         phys_addr_t start = 0;
419         unsigned long last_pfn = nr_pages;
420         const struct e820entry *entry;
421         unsigned long num_released = 0;
422         unsigned long num_remapped = 0;
423         int i;
424 
425         /*
426          * Combine non-RAM regions and gaps until a RAM region (or the
427          * end of the map) is reached, then set the 1:1 map and
428          * remap the memory in those non-RAM regions.
429          *
430          * The combined non-RAM regions are rounded to a whole number
431          * of pages so any partial pages are accessible via the 1:1
432          * mapping.  This is needed for some BIOSes that put (for
433          * example) the DMI tables in a reserved region that begins on
434          * a non-page boundary.
435          */
436         for (i = 0, entry = list; i < map_size; i++, entry++) {
437                 phys_addr_t end = entry->addr + entry->size;
438                 if (entry->type == E820_RAM || i == map_size - 1) {
439                         unsigned long start_pfn = PFN_DOWN(start);
440                         unsigned long end_pfn = PFN_UP(end);
441 
442                         if (entry->type == E820_RAM)
443                                 end_pfn = PFN_UP(entry->addr);
444 
445                         if (start_pfn < end_pfn)
446                                 last_pfn = xen_set_identity_and_remap_chunk(
447                                                 list, map_size, start_pfn,
448                                                 end_pfn, nr_pages, last_pfn,
449                                                 &num_released, &num_remapped);
450                         start = end;
451                 }
452         }
453 
454         *released = num_released;
455         *remapped = num_remapped;
456 
457         pr_info("Released %ld page(s)\n", num_released);
458 }
459 
460 /*
461  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
462  * The remap information (which mfn remap to which pfn) is contained in the
463  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
464  * This scheme allows to remap the different chunks in arbitrary order while
465  * the resulting mapping will be independant from the order.
466  */
467 void __init xen_remap_memory(void)
468 {
469         unsigned long buf = (unsigned long)&xen_remap_buf;
470         unsigned long mfn_save, mfn, pfn;
471         unsigned long remapped = 0;
472         unsigned int i;
473         unsigned long pfn_s = ~0UL;
474         unsigned long len = 0;
475 
476         mfn_save = virt_to_mfn(buf);
477 
478         while (xen_remap_mfn != INVALID_P2M_ENTRY) {
479                 /* Map the remap information */
480                 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
481 
482                 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
483 
484                 pfn = xen_remap_buf.target_pfn;
485                 for (i = 0; i < xen_remap_buf.size; i++) {
486                         mfn = xen_remap_buf.mfns[i];
487                         xen_update_mem_tables(pfn, mfn);
488                         remapped++;
489                         pfn++;
490                 }
491                 if (pfn_s == ~0UL || pfn == pfn_s) {
492                         pfn_s = xen_remap_buf.target_pfn;
493                         len += xen_remap_buf.size;
494                 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
495                         len += xen_remap_buf.size;
496                 } else {
497                         xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
498                         pfn_s = xen_remap_buf.target_pfn;
499                         len = xen_remap_buf.size;
500                 }
501 
502                 mfn = xen_remap_mfn;
503                 xen_remap_mfn = xen_remap_buf.next_area_mfn;
504         }
505 
506         if (pfn_s != ~0UL && len)
507                 xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len));
508 
509         set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
510 
511         pr_info("Remapped %ld page(s)\n", remapped);
512 }
513 
514 static unsigned long __init xen_get_max_pages(void)
515 {
516         unsigned long max_pages = MAX_DOMAIN_PAGES;
517         domid_t domid = DOMID_SELF;
518         int ret;
519 
520         /*
521          * For the initial domain we use the maximum reservation as
522          * the maximum page.
523          *
524          * For guest domains the current maximum reservation reflects
525          * the current maximum rather than the static maximum. In this
526          * case the e820 map provided to us will cover the static
527          * maximum region.
528          */
529         if (xen_initial_domain()) {
530                 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
531                 if (ret > 0)
532                         max_pages = ret;
533         }
534 
535         return min(max_pages, MAX_DOMAIN_PAGES);
536 }
537 
538 static void __init xen_align_and_add_e820_region(phys_addr_t start,
539                                                  phys_addr_t size, int type)
540 {
541         phys_addr_t end = start + size;
542 
543         /* Align RAM regions to page boundaries. */
544         if (type == E820_RAM) {
545                 start = PAGE_ALIGN(start);
546                 end &= ~((phys_addr_t)PAGE_SIZE - 1);
547         }
548 
549         e820_add_region(start, end - start, type);
550 }
551 
552 static void __init xen_ignore_unusable(struct e820entry *list, size_t map_size)
553 {
554         struct e820entry *entry;
555         unsigned int i;
556 
557         for (i = 0, entry = list; i < map_size; i++, entry++) {
558                 if (entry->type == E820_UNUSABLE)
559                         entry->type = E820_RAM;
560         }
561 }
562 
563 /**
564  * machine_specific_memory_setup - Hook for machine specific memory setup.
565  **/
566 char * __init xen_memory_setup(void)
567 {
568         static struct e820entry map[E820MAX] __initdata;
569 
570         unsigned long max_pfn = xen_start_info->nr_pages;
571         phys_addr_t mem_end;
572         int rc;
573         struct xen_memory_map memmap;
574         unsigned long max_pages;
575         unsigned long extra_pages = 0;
576         unsigned long remapped_pages;
577         int i;
578         int op;
579 
580         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
581         mem_end = PFN_PHYS(max_pfn);
582 
583         memmap.nr_entries = E820MAX;
584         set_xen_guest_handle(memmap.buffer, map);
585 
586         op = xen_initial_domain() ?
587                 XENMEM_machine_memory_map :
588                 XENMEM_memory_map;
589         rc = HYPERVISOR_memory_op(op, &memmap);
590         if (rc == -ENOSYS) {
591                 BUG_ON(xen_initial_domain());
592                 memmap.nr_entries = 1;
593                 map[0].addr = 0ULL;
594                 map[0].size = mem_end;
595                 /* 8MB slack (to balance backend allocations). */
596                 map[0].size += 8ULL << 20;
597                 map[0].type = E820_RAM;
598                 rc = 0;
599         }
600         BUG_ON(rc);
601         BUG_ON(memmap.nr_entries == 0);
602 
603         /*
604          * Xen won't allow a 1:1 mapping to be created to UNUSABLE
605          * regions, so if we're using the machine memory map leave the
606          * region as RAM as it is in the pseudo-physical map.
607          *
608          * UNUSABLE regions in domUs are not handled and will need
609          * a patch in the future.
610          */
611         if (xen_initial_domain())
612                 xen_ignore_unusable(map, memmap.nr_entries);
613 
614         /* Make sure the Xen-supplied memory map is well-ordered. */
615         sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
616 
617         max_pages = xen_get_max_pages();
618         if (max_pages > max_pfn)
619                 extra_pages += max_pages - max_pfn;
620 
621         /*
622          * Set identity map on non-RAM pages and prepare remapping the
623          * underlying RAM.
624          */
625         xen_set_identity_and_remap(map, memmap.nr_entries, max_pfn,
626                                    &xen_released_pages, &remapped_pages);
627 
628         extra_pages += xen_released_pages;
629         extra_pages += remapped_pages;
630 
631         /*
632          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
633          * factor the base size.  On non-highmem systems, the base
634          * size is the full initial memory allocation; on highmem it
635          * is limited to the max size of lowmem, so that it doesn't
636          * get completely filled.
637          *
638          * In principle there could be a problem in lowmem systems if
639          * the initial memory is also very large with respect to
640          * lowmem, but we won't try to deal with that here.
641          */
642         extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
643                           extra_pages);
644         i = 0;
645         while (i < memmap.nr_entries) {
646                 phys_addr_t addr = map[i].addr;
647                 phys_addr_t size = map[i].size;
648                 u32 type = map[i].type;
649 
650                 if (type == E820_RAM) {
651                         if (addr < mem_end) {
652                                 size = min(size, mem_end - addr);
653                         } else if (extra_pages) {
654                                 size = min(size, PFN_PHYS(extra_pages));
655                                 extra_pages -= PFN_DOWN(size);
656                                 xen_add_extra_mem(addr, size);
657                                 xen_max_p2m_pfn = PFN_DOWN(addr + size);
658                         } else
659                                 type = E820_UNUSABLE;
660                 }
661 
662                 xen_align_and_add_e820_region(addr, size, type);
663 
664                 map[i].addr += size;
665                 map[i].size -= size;
666                 if (map[i].size == 0)
667                         i++;
668         }
669 
670         /*
671          * Set the rest as identity mapped, in case PCI BARs are
672          * located here.
673          *
674          * PFNs above MAX_P2M_PFN are considered identity mapped as
675          * well.
676          */
677         set_phys_range_identity(map[i-1].addr / PAGE_SIZE, ~0ul);
678 
679         /*
680          * In domU, the ISA region is normal, usable memory, but we
681          * reserve ISA memory anyway because too many things poke
682          * about in there.
683          */
684         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
685                         E820_RESERVED);
686 
687         /*
688          * Reserve Xen bits:
689          *  - mfn_list
690          *  - xen_start_info
691          * See comment above "struct start_info" in <xen/interface/xen.h>
692          * We tried to make the the memblock_reserve more selective so
693          * that it would be clear what region is reserved. Sadly we ran
694          * in the problem wherein on a 64-bit hypervisor with a 32-bit
695          * initial domain, the pt_base has the cr3 value which is not
696          * neccessarily where the pagetable starts! As Jan put it: "
697          * Actually, the adjustment turns out to be correct: The page
698          * tables for a 32-on-64 dom0 get allocated in the order "first L1",
699          * "first L2", "first L3", so the offset to the page table base is
700          * indeed 2. When reading xen/include/public/xen.h's comment
701          * very strictly, this is not a violation (since there nothing is said
702          * that the first thing in the page table space is pointed to by
703          * pt_base; I admit that this seems to be implied though, namely
704          * do I think that it is implied that the page table space is the
705          * range [pt_base, pt_base + nt_pt_frames), whereas that
706          * range here indeed is [pt_base - 2, pt_base - 2 + nt_pt_frames),
707          * which - without a priori knowledge - the kernel would have
708          * difficulty to figure out)." - so lets just fall back to the
709          * easy way and reserve the whole region.
710          */
711         memblock_reserve(__pa(xen_start_info->mfn_list),
712                          xen_start_info->pt_base - xen_start_info->mfn_list);
713 
714         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
715 
716         return "Xen";
717 }
718 
719 /*
720  * Machine specific memory setup for auto-translated guests.
721  */
722 char * __init xen_auto_xlated_memory_setup(void)
723 {
724         static struct e820entry map[E820MAX] __initdata;
725 
726         struct xen_memory_map memmap;
727         int i;
728         int rc;
729 
730         memmap.nr_entries = E820MAX;
731         set_xen_guest_handle(memmap.buffer, map);
732 
733         rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
734         if (rc < 0)
735                 panic("No memory map (%d)\n", rc);
736 
737         sanitize_e820_map(map, ARRAY_SIZE(map), &memmap.nr_entries);
738 
739         for (i = 0; i < memmap.nr_entries; i++)
740                 e820_add_region(map[i].addr, map[i].size, map[i].type);
741 
742         memblock_reserve(__pa(xen_start_info->mfn_list),
743                          xen_start_info->pt_base - xen_start_info->mfn_list);
744 
745         return "Xen";
746 }
747 
748 /*
749  * Set the bit indicating "nosegneg" library variants should be used.
750  * We only need to bother in pure 32-bit mode; compat 32-bit processes
751  * can have un-truncated segments, so wrapping around is allowed.
752  */
753 static void __init fiddle_vdso(void)
754 {
755 #ifdef CONFIG_X86_32
756         /*
757          * This could be called before selected_vdso32 is initialized, so
758          * just fiddle with both possible images.  vdso_image_32_syscall
759          * can't be selected, since it only exists on 64-bit systems.
760          */
761         u32 *mask;
762         mask = vdso_image_32_int80.data +
763                 vdso_image_32_int80.sym_VDSO32_NOTE_MASK;
764         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
765         mask = vdso_image_32_sysenter.data +
766                 vdso_image_32_sysenter.sym_VDSO32_NOTE_MASK;
767         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
768 #endif
769 }
770 
771 static int register_callback(unsigned type, const void *func)
772 {
773         struct callback_register callback = {
774                 .type = type,
775                 .address = XEN_CALLBACK(__KERNEL_CS, func),
776                 .flags = CALLBACKF_mask_events,
777         };
778 
779         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
780 }
781 
782 void xen_enable_sysenter(void)
783 {
784         int ret;
785         unsigned sysenter_feature;
786 
787 #ifdef CONFIG_X86_32
788         sysenter_feature = X86_FEATURE_SEP;
789 #else
790         sysenter_feature = X86_FEATURE_SYSENTER32;
791 #endif
792 
793         if (!boot_cpu_has(sysenter_feature))
794                 return;
795 
796         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
797         if(ret != 0)
798                 setup_clear_cpu_cap(sysenter_feature);
799 }
800 
801 void xen_enable_syscall(void)
802 {
803 #ifdef CONFIG_X86_64
804         int ret;
805 
806         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
807         if (ret != 0) {
808                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
809                 /* Pretty fatal; 64-bit userspace has no other
810                    mechanism for syscalls. */
811         }
812 
813         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
814                 ret = register_callback(CALLBACKTYPE_syscall32,
815                                         xen_syscall32_target);
816                 if (ret != 0)
817                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
818         }
819 #endif /* CONFIG_X86_64 */
820 }
821 
822 void __init xen_pvmmu_arch_setup(void)
823 {
824         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
825         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
826 
827         HYPERVISOR_vm_assist(VMASST_CMD_enable,
828                              VMASST_TYPE_pae_extended_cr3);
829 
830         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
831             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
832                 BUG();
833 
834         xen_enable_sysenter();
835         xen_enable_syscall();
836 }
837 
838 /* This function is not called for HVM domains */
839 void __init xen_arch_setup(void)
840 {
841         xen_panic_handler_init();
842         if (!xen_feature(XENFEAT_auto_translated_physmap))
843                 xen_pvmmu_arch_setup();
844 
845 #ifdef CONFIG_ACPI
846         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
847                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
848                 disable_acpi();
849         }
850 #endif
851 
852         memcpy(boot_command_line, xen_start_info->cmd_line,
853                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
854                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
855 
856         /* Set up idle, making sure it calls safe_halt() pvop */
857         disable_cpuidle();
858         disable_cpufreq();
859         WARN_ON(xen_set_default_idle());
860         fiddle_vdso();
861 #ifdef CONFIG_NUMA
862         numa_off = 1;
863 #endif
864 }
865 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp