arch/x86/xen/setup.c

   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
  13 #include <asm/elf.h>
  14 #include <asm/vdso.h>
  15 #include <asm/e820.h>
  16 #include <asm/setup.h>
  17 #include <asm/acpi.h>
  18 #include <asm/xen/hypervisor.h>
  19 #include <asm/xen/hypercall.h>
  20
  21 #include <xen/xen.h>
  22 #include <xen/page.h>
  23 #include <xen/interface/callback.h>
  24 #include <xen/interface/memory.h>
  25 #include <xen/interface/physdev.h>
  26 #include <xen/features.h>
  27
  28 #include "xen-ops.h"
  29 #include "vdso.h"
  30
  31 /* These are code, but not functions.  Defined in entry.S */
  32 extern const char xen_hypervisor_callback[];
  33 extern const char xen_failsafe_callback[];
  34 extern void xen_sysenter_target(void);
  35 extern void xen_syscall_target(void);
  36 extern void xen_syscall32_target(void);
  37
  38 /* Amount of extra memory space we add to the e820 ranges */
  39 phys_addr_t xen_extra_mem_start, xen_extra_mem_size;
  40
  41 /*
  42  * The maximum amount of extra memory compared to the base size.  The
  43  * main scaling factor is the size of struct page.  At extreme ratios
  44  * of base:extra, all the base memory can be filled with page
  45  * structures for the extra memory, leaving no space for anything
  46  * else.
  47  *
  48  * 10x seems like a reasonable balance between scaling flexibility and
  49  * leaving a practically usable system.
  50  */
  51 #define EXTRA_MEM_RATIO         (10)
  52
  53 static __init void xen_add_extra_mem(unsigned long pages)
  54 {
  55         u64 size = (u64)pages * PAGE_SIZE;
  56         u64 extra_start = xen_extra_mem_start + xen_extra_mem_size;
  57
  58         if (!pages)
  59                 return;
  60
  61         e820_add_region(extra_start, size, E820_RAM);
  62         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
  63
  64         memblock_x86_reserve_range(extra_start, extra_start + size, "XEN EXTRA");
  65
  66         xen_extra_mem_size += size;
  67
  68         xen_max_p2m_pfn = PFN_DOWN(extra_start + size);
  69 }
  70
  71 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
  72                                               phys_addr_t end_addr)
  73 {
  74         struct xen_memory_reservation reservation = {
  75                 .address_bits = 0,
  76                 .extent_order = 0,
  77                 .domid        = DOMID_SELF
  78         };
  79         unsigned long start, end;
  80         unsigned long len = 0;
  81         unsigned long pfn;
  82         int ret;
  83
  84         start = PFN_UP(start_addr);
  85         end = PFN_DOWN(end_addr);
  86
  87         if (end <= start)
  88                 return 0;
  89
  90         printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
  91                start, end);
  92         for(pfn = start; pfn < end; pfn++) {
  93                 unsigned long mfn = pfn_to_mfn(pfn);
  94
  95                 /* Make sure pfn exists to start with */
  96                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
  97                         continue;
  98
  99                 set_xen_guest_handle(reservation.extent_start, &mfn);
 100                 reservation.nr_extents = 1;
 101
 102                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
 103                                            &reservation);
 104                 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
 105                      start, end, ret);
 106                 if (ret == 1) {
 107                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 108                         len++;
 109                 }
 110         }
 111         printk(KERN_CONT "%ld pages freed\n", len);
 112
 113         return len;
 114 }
 115
 116 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
 117                                                      const struct e820map *e820)
 118 {
 119         phys_addr_t max_addr = PFN_PHYS(max_pfn);
 120         phys_addr_t last_end = ISA_END_ADDRESS;
 121         unsigned long released = 0;
 122         int i;
 123
 124         /* Free any unused memory above the low 1Mbyte. */
 125         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
 126                 phys_addr_t end = e820->map[i].addr;
 127                 end = min(max_addr, end);
 128
 129                 if (last_end < end)
 130                         released += xen_release_chunk(last_end, end);
 131                 last_end = max(last_end, e820->map[i].addr + e820->map[i].size);
 132         }
 133
 134         if (last_end < max_addr)
 135                 released += xen_release_chunk(last_end, max_addr);
 136
 137         printk(KERN_INFO "released %ld pages of unused memory\n", released);
 138         return released;
 139 }
 140
 141 /**
 142  * machine_specific_memory_setup - Hook for machine specific memory setup.
 143  **/
 144 char * __init xen_memory_setup(void)
 145 {
 146         static struct e820entry map[E820MAX] __initdata;
 147
 148         unsigned long max_pfn = xen_start_info->nr_pages;
 149         unsigned long long mem_end;
 150         int rc;
 151         struct xen_memory_map memmap;
 152         unsigned long extra_pages = 0;
 153         unsigned long extra_limit;
 154         int i;
 155         int op;
 156
 157         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 158         mem_end = PFN_PHYS(max_pfn);
 159
 160         memmap.nr_entries = E820MAX;
 161         set_xen_guest_handle(memmap.buffer, map);
 162
 163         op = xen_initial_domain() ?
 164                 XENMEM_machine_memory_map :
 165                 XENMEM_memory_map;
 166         rc = HYPERVISOR_memory_op(op, &memmap);
 167         if (rc == -ENOSYS) {
 168                 BUG_ON(xen_initial_domain());
 169                 memmap.nr_entries = 1;
 170                 map[0].addr = 0ULL;
 171                 map[0].size = mem_end;
 172                 /* 8MB slack (to balance backend allocations). */
 173                 map[0].size += 8ULL << 20;
 174                 map[0].type = E820_RAM;
 175                 rc = 0;
 176         }
 177         BUG_ON(rc);
 178
 179         e820.nr_map = 0;
 180         xen_extra_mem_start = mem_end;
 181         for (i = 0; i < memmap.nr_entries; i++) {
 182                 unsigned long long end = map[i].addr + map[i].size;
 183
 184                 if (map[i].type == E820_RAM) {
 185                         if (map[i].addr < mem_end && end > mem_end) {
 186                                 /* Truncate region to max_mem. */
 187                                 u64 delta = end - mem_end;
 188
 189                                 map[i].size -= delta;
 190                                 extra_pages += PFN_DOWN(delta);
 191
 192                                 end = mem_end;
 193                         }
 194                 }
 195
 196                 if (end > xen_extra_mem_start)
 197                         xen_extra_mem_start = end;
 198
 199                 /* If region is non-RAM or below mem_end, add what remains */
 200                 if ((map[i].type != E820_RAM || map[i].addr < mem_end) &&
 201                     map[i].size > 0)
 202                         e820_add_region(map[i].addr, map[i].size, map[i].type);
 203         }
 204
 205         /*
 206          * In domU, the ISA region is normal, usable memory, but we
 207          * reserve ISA memory anyway because too many things poke
 208          * about in there.
 209          *
 210          * In Dom0, the host E820 information can leave gaps in the
 211          * ISA range, which would cause us to release those pages.  To
 212          * avoid this, we unconditionally reserve them here.
 213          */
 214         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 215                         E820_RESERVED);
 216
 217         /*
 218          * Reserve Xen bits:
 219          *  - mfn_list
 220          *  - xen_start_info
 221          * See comment above "struct start_info" in <xen/interface/xen.h>
 222          */
 223         memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
 224                       __pa(xen_start_info->pt_base),
 225                         "XEN START INFO");
 226
 227         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 228
 229         extra_pages += xen_return_unused_memory(xen_start_info->nr_pages, &e820);
 230
 231         /*
 232          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
 233          * factor the base size.  On non-highmem systems, the base
 234          * size is the full initial memory allocation; on highmem it
 235          * is limited to the max size of lowmem, so that it doesn't
 236          * get completely filled.
 237          *
 238          * In principle there could be a problem in lowmem systems if
 239          * the initial memory is also very large with respect to
 240          * lowmem, but we won't try to deal with that here.
 241          */
 242         extra_limit = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
 243                           max_pfn + extra_pages);
 244
 245         if (extra_limit >= max_pfn)
 246                 extra_pages = extra_limit - max_pfn;
 247         else
 248                 extra_pages = 0;
 249
 250         xen_add_extra_mem(extra_pages);
 251
 252         return "Xen";
 253 }
 254
 255 static void xen_idle(void)
 256 {
 257         local_irq_disable();
 258
 259         if (need_resched())
 260                 local_irq_enable();
 261         else {
 262                 current_thread_info()->status &= ~TS_POLLING;
 263                 smp_mb__after_clear_bit();
 264                 safe_halt();
 265                 current_thread_info()->status |= TS_POLLING;
 266         }
 267 }
 268
 269 /*
 270  * Set the bit indicating "nosegneg" library variants should be used.
 271  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 272  * can have un-truncated segments, so wrapping around is allowed.
 273  */
 274 static void __init fiddle_vdso(void)
 275 {
 276 #ifdef CONFIG_X86_32
 277         u32 *mask;
 278         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 279         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 280         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 281         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 282 #endif
 283 }
 284
 285 static __cpuinit int register_callback(unsigned type, const void *func)
 286 {
 287         struct callback_register callback = {
 288                 .type = type,
 289                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 290                 .flags = CALLBACKF_mask_events,
 291         };
 292
 293         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 294 }
 295
 296 void __cpuinit xen_enable_sysenter(void)
 297 {
 298         int ret;
 299         unsigned sysenter_feature;
 300
 301 #ifdef CONFIG_X86_32
 302         sysenter_feature = X86_FEATURE_SEP;
 303 #else
 304         sysenter_feature = X86_FEATURE_SYSENTER32;
 305 #endif
 306
 307         if (!boot_cpu_has(sysenter_feature))
 308                 return;
 309
 310         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 311         if(ret != 0)
 312                 setup_clear_cpu_cap(sysenter_feature);
 313 }
 314
 315 void __cpuinit xen_enable_syscall(void)
 316 {
 317 #ifdef CONFIG_X86_64
 318         int ret;
 319
 320         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 321         if (ret != 0) {
 322                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 323                 /* Pretty fatal; 64-bit userspace has no other
 324                    mechanism for syscalls. */
 325         }
 326
 327         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 328                 ret = register_callback(CALLBACKTYPE_syscall32,
 329                                         xen_syscall32_target);
 330                 if (ret != 0)
 331                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 332         }
 333 #endif /* CONFIG_X86_64 */
 334 }
 335
 336 void __init xen_arch_setup(void)
 337 {
 338         xen_panic_handler_init();
 339
 340         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 341         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 342
 343         if (!xen_feature(XENFEAT_auto_translated_physmap))
 344                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 345                                      VMASST_TYPE_pae_extended_cr3);
 346
 347         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 348             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 349                 BUG();
 350
 351         xen_enable_sysenter();
 352         xen_enable_syscall();
 353
 354 #ifdef CONFIG_ACPI
 355         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 356                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 357                 disable_acpi();
 358         }
 359 #endif
 360
 361         memcpy(boot_command_line, xen_start_info->cmd_line,
 362                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 363                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 364
 365         pm_idle = xen_idle;
 366
 367         fiddle_vdso();
 368 }