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

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

Version: ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ 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 
 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 
 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 
 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 
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 
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 
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 
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 
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 
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 void xen_ignore_unusable(struct e820entry *list, size_t map_size)
317 {
318         struct e820entry *entry;
319         unsigned int i;
320 
321         for (i = 0, entry = list; i < map_size; i++, entry++) {
322                 if (entry->type == E820_UNUSABLE)
323                         entry->type = E820_RAM;
324         }
325 }
326 
327 /**
328  * machine_specific_memory_setup - Hook for machine specific memory setup.
329  **/
330 char * __init xen_memory_setup(void)
331 {
332         static struct e820entry map[E820MAX] __initdata;
333 
334         unsigned long max_pfn = xen_start_info->nr_pages;
335         unsigned long long mem_end;
336         int rc;
337         struct xen_memory_map memmap;
338         unsigned long max_pages;
339         unsigned long last_pfn = 0;
340         unsigned long extra_pages = 0;
341         unsigned long populated;
342         int i;
343         int op;
344 
345         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
346         mem_end = PFN_PHYS(max_pfn);
347 
348         memmap.nr_entries = E820MAX;
349         set_xen_guest_handle(memmap.buffer, map);
350 
351         op = xen_initial_domain() ?
352                 XENMEM_machine_memory_map :
353                 XENMEM_memory_map;
354         rc = HYPERVISOR_memory_op(op, &memmap);
355         if (rc == -ENOSYS) {
356                 BUG_ON(xen_initial_domain());
357                 memmap.nr_entries = 1;
358                 map[0].addr = 0ULL;
359                 map[0].size = mem_end;
360                 /* 8MB slack (to balance backend allocations). */
361                 map[0].size += 8ULL << 20;
362                 map[0].type = E820_RAM;
363                 rc = 0;
364         }
365         BUG_ON(rc);
366 
367         /*
368          * Xen won't allow a 1:1 mapping to be created to UNUSABLE
369          * regions, so if we're using the machine memory map leave the
370          * region as RAM as it is in the pseudo-physical map.
371          *
372          * UNUSABLE regions in domUs are not handled and will need
373          * a patch in the future.
374          */
375         if (xen_initial_domain())
376                 xen_ignore_unusable(map, memmap.nr_entries);
377 
378         /* Make sure the Xen-supplied memory map is well-ordered. */
379         sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
380 
381         max_pages = xen_get_max_pages();
382         if (max_pages > max_pfn)
383                 extra_pages += max_pages - max_pfn;
384 
385         /*
386          * Set P2M for all non-RAM pages and E820 gaps to be identity
387          * type PFNs.  Any RAM pages that would be made inaccesible by
388          * this are first released.
389          */
390         xen_released_pages = xen_set_identity_and_release(
391                 map, memmap.nr_entries, max_pfn);
392 
393         /*
394          * Populate back the non-RAM pages and E820 gaps that had been
395          * released. */
396         populated = xen_populate_chunk(map, memmap.nr_entries,
397                         max_pfn, &last_pfn, xen_released_pages);
398 
399         xen_released_pages -= populated;
400         extra_pages += xen_released_pages;
401 
402         if (last_pfn > max_pfn) {
403                 max_pfn = min(MAX_DOMAIN_PAGES, last_pfn);
404                 mem_end = PFN_PHYS(max_pfn);
405         }
406         /*
407          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
408          * factor the base size.  On non-highmem systems, the base
409          * size is the full initial memory allocation; on highmem it
410          * is limited to the max size of lowmem, so that it doesn't
411          * get completely filled.
412          *
413          * In principle there could be a problem in lowmem systems if
414          * the initial memory is also very large with respect to
415          * lowmem, but we won't try to deal with that here.
416          */
417         extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
418                           extra_pages);
419         i = 0;
420         while (i < memmap.nr_entries) {
421                 u64 addr = map[i].addr;
422                 u64 size = map[i].size;
423                 u32 type = map[i].type;
424 
425                 if (type == E820_RAM) {
426                         if (addr < mem_end) {
427                                 size = min(size, mem_end - addr);
428                         } else if (extra_pages) {
429                                 size = min(size, (u64)extra_pages * PAGE_SIZE);
430                                 extra_pages -= size / PAGE_SIZE;
431                                 xen_add_extra_mem(addr, size);
432                         } else
433                                 type = E820_UNUSABLE;
434                 }
435 
436                 xen_align_and_add_e820_region(addr, size, type);
437 
438                 map[i].addr += size;
439                 map[i].size -= size;
440                 if (map[i].size == 0)
441                         i++;
442         }
443 
444         /*
445          * In domU, the ISA region is normal, usable memory, but we
446          * reserve ISA memory anyway because too many things poke
447          * about in there.
448          */
449         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
450                         E820_RESERVED);
451 
452         /*
453          * Reserve Xen bits:
454          *  - mfn_list
455          *  - xen_start_info
456          * See comment above "struct start_info" in <xen/interface/xen.h>
457          * We tried to make the the memblock_reserve more selective so
458          * that it would be clear what region is reserved. Sadly we ran
459          * in the problem wherein on a 64-bit hypervisor with a 32-bit
460          * initial domain, the pt_base has the cr3 value which is not
461          * neccessarily where the pagetable starts! As Jan put it: "
462          * Actually, the adjustment turns out to be correct: The page
463          * tables for a 32-on-64 dom0 get allocated in the order "first L1",
464          * "first L2", "first L3", so the offset to the page table base is
465          * indeed 2. When reading xen/include/public/xen.h's comment
466          * very strictly, this is not a violation (since there nothing is said
467          * that the first thing in the page table space is pointed to by
468          * pt_base; I admit that this seems to be implied though, namely
469          * do I think that it is implied that the page table space is the
470          * range [pt_base, pt_base + nt_pt_frames), whereas that
471          * range here indeed is [pt_base - 2, pt_base - 2 + nt_pt_frames),
472          * which - without a priori knowledge - the kernel would have
473          * difficulty to figure out)." - so lets just fall back to the
474          * easy way and reserve the whole region.
475          */
476         memblock_reserve(__pa(xen_start_info->mfn_list),
477                          xen_start_info->pt_base - xen_start_info->mfn_list);
478 
479         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
480 
481         return "Xen";
482 }
483 
484 /*
485  * Set the bit indicating "nosegneg" library variants should be used.
486  * We only need to bother in pure 32-bit mode; compat 32-bit processes
487  * can have un-truncated segments, so wrapping around is allowed.
488  */
489 static void __init fiddle_vdso(void)
490 {
491 #ifdef CONFIG_X86_32
492         u32 *mask;
493         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
494         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
495         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
496         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
497 #endif
498 }
499 
500 static int __cpuinit register_callback(unsigned type, const void *func)
501 {
502         struct callback_register callback = {
503                 .type = type,
504                 .address = XEN_CALLBACK(__KERNEL_CS, func),
505                 .flags = CALLBACKF_mask_events,
506         };
507 
508         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
509 }
510 
511 void __cpuinit xen_enable_sysenter(void)
512 {
513         int ret;
514         unsigned sysenter_feature;
515 
516 #ifdef CONFIG_X86_32
517         sysenter_feature = X86_FEATURE_SEP;
518 #else
519         sysenter_feature = X86_FEATURE_SYSENTER32;
520 #endif
521 
522         if (!boot_cpu_has(sysenter_feature))
523                 return;
524 
525         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
526         if(ret != 0)
527                 setup_clear_cpu_cap(sysenter_feature);
528 }
529 
530 void __cpuinit xen_enable_syscall(void)
531 {
532 #ifdef CONFIG_X86_64
533         int ret;
534 
535         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
536         if (ret != 0) {
537                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
538                 /* Pretty fatal; 64-bit userspace has no other
539                    mechanism for syscalls. */
540         }
541 
542         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
543                 ret = register_callback(CALLBACKTYPE_syscall32,
544                                         xen_syscall32_target);
545                 if (ret != 0)
546                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
547         }
548 #endif /* CONFIG_X86_64 */
549 }
550 
551 void __init xen_arch_setup(void)
552 {
553         xen_panic_handler_init();
554 
555         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
556         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
557 
558         if (!xen_feature(XENFEAT_auto_translated_physmap))
559                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
560                                      VMASST_TYPE_pae_extended_cr3);
561 
562         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
563             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
564                 BUG();
565 
566         xen_enable_sysenter();
567         xen_enable_syscall();
568 
569 #ifdef CONFIG_ACPI
570         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
571                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
572                 disable_acpi();
573         }
574 #endif
575 
576         memcpy(boot_command_line, xen_start_info->cmd_line,
577                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
578                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
579 
580         /* Set up idle, making sure it calls safe_halt() pvop */
581         disable_cpuidle();
582         disable_cpufreq();
583         WARN_ON(xen_set_default_idle());
584         fiddle_vdso();
585 #ifdef CONFIG_NUMA
586         numa_off = 1;
587 #endif
588 }
589 

~ [ 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