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/interface/memory.h>
  27 #include <xen/features.h>
  28
  29 #include "xen-ops.h"
  30 #include "vdso.h"
  31
  32 /* These are code, but not functions.  Defined in entry.S */
  33 extern const char xen_hypervisor_callback[];
  34 extern const char xen_failsafe_callback[];
  35 extern void xen_sysenter_target(void);
  36 extern void xen_syscall_target(void);
  37 extern void xen_syscall32_target(void);
  38
  39 /* Amount of extra memory space we add to the e820 ranges */
  40 phys_addr_t xen_extra_mem_start, xen_extra_mem_size;
  41
  42 /*
  43  * The maximum amount of extra memory compared to the base size.  The
  44  * main scaling factor is the size of struct page.  At extreme ratios
  45  * of base:extra, all the base memory can be filled with page
  46  * structures for the extra memory, leaving no space for anything
  47  * else.
  48  *
  49  * 10x seems like a reasonable balance between scaling flexibility and
  50  * leaving a practically usable system.
  51  */
  52 #define EXTRA_MEM_RATIO         (10)
  53
  54 static __init void xen_add_extra_mem(unsigned long pages)
  55 {
  56         u64 size = (u64)pages * PAGE_SIZE;
  57         u64 extra_start = xen_extra_mem_start + xen_extra_mem_size;
  58
  59         if (!pages)
  60                 return;
  61
  62         e820_add_region(extra_start, size, E820_RAM);
  63         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
  64
  65         memblock_x86_reserve_range(extra_start, extra_start + size, "XEN EXTRA");
  66
  67         xen_extra_mem_size += size;
  68
  69         xen_max_p2m_pfn = PFN_DOWN(extra_start + size);
  70 }
  71
  72 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
  73                                               phys_addr_t end_addr)
  74 {
  75         struct xen_memory_reservation reservation = {
  76                 .address_bits = 0,
  77                 .extent_order = 0,
  78                 .domid        = DOMID_SELF
  79         };
  80         unsigned long start, end;
  81         unsigned long len = 0;
  82         unsigned long pfn;
  83         int ret;
  84
  85         start = PFN_UP(start_addr);
  86         end = PFN_DOWN(end_addr);
  87
  88         if (end <= start)
  89                 return 0;
  90
  91         printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
  92                start, end);
  93         for(pfn = start; pfn < end; pfn++) {
  94                 unsigned long mfn = pfn_to_mfn(pfn);
  95
  96                 /* Make sure pfn exists to start with */
  97                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
  98                         continue;
  99
 100                 set_xen_guest_handle(reservation.extent_start, &mfn);
 101                 reservation.nr_extents = 1;
 102
 103                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
 104                                            &reservation);
 105                 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
 106                      start, end, ret);
 107                 if (ret == 1) {
 108                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 109                         len++;
 110                 }
 111         }
 112         printk(KERN_CONT "%ld pages freed\n", len);
 113
 114         return len;
 115 }
 116
 117 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
 118                                                      const struct e820map *e820)
 119 {
 120         phys_addr_t max_addr = PFN_PHYS(max_pfn);
 121         phys_addr_t last_end = ISA_END_ADDRESS;
 122         unsigned long released = 0;
 123         int i;
 124
 125         /* Free any unused memory above the low 1Mbyte. */
 126         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
 127                 phys_addr_t end = e820->map[i].addr;
 128                 end = min(max_addr, end);
 129
 130                 if (last_end < end)
 131                         released += xen_release_chunk(last_end, end);
 132                 last_end = max(last_end, e820->map[i].addr + e820->map[i].size);
 133         }
 134
 135         if (last_end < max_addr)
 136                 released += xen_release_chunk(last_end, max_addr);
 137
 138         printk(KERN_INFO "released %ld pages of unused memory\n", released);
 139         return released;
 140 }
 141
 142 /**
 143  * machine_specific_memory_setup - Hook for machine specific memory setup.
 144  **/
 145 char * __init xen_memory_setup(void)
 146 {
 147         static struct e820entry map[E820MAX] __initdata;
 148
 149         unsigned long max_pfn = xen_start_info->nr_pages;
 150         unsigned long long mem_end;
 151         int rc;
 152         struct xen_memory_map memmap;
 153         unsigned long extra_pages = 0;
 154         unsigned long extra_limit;
 155         int i;
 156         int op;
 157
 158         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 159         mem_end = PFN_PHYS(max_pfn);
 160
 161         memmap.nr_entries = E820MAX;
 162         set_xen_guest_handle(memmap.buffer, map);
 163
 164         op = xen_initial_domain() ?
 165                 XENMEM_machine_memory_map :
 166                 XENMEM_memory_map;
 167         rc = HYPERVISOR_memory_op(op, &memmap);
 168         if (rc == -ENOSYS) {
 169                 BUG_ON(xen_initial_domain());
 170                 memmap.nr_entries = 1;
 171                 map[0].addr = 0ULL;
 172                 map[0].size = mem_end;
 173                 /* 8MB slack (to balance backend allocations). */
 174                 map[0].size += 8ULL << 20;
 175                 map[0].type = E820_RAM;
 176                 rc = 0;
 177         }
 178         BUG_ON(rc);
 179
 180         e820.nr_map = 0;
 181         xen_extra_mem_start = mem_end;
 182         for (i = 0; i < memmap.nr_entries; i++) {
 183                 unsigned long long end = map[i].addr + map[i].size;
 184
 185                 if (map[i].type == E820_RAM) {
 186                         if (map[i].addr < mem_end && end > mem_end) {
 187                                 /* Truncate region to max_mem. */
 188                                 u64 delta = end - mem_end;
 189
 190                                 map[i].size -= delta;
 191                                 extra_pages += PFN_DOWN(delta);
 192
 193                                 end = mem_end;
 194                         }
 195                 }
 196
 197                 if (end > xen_extra_mem_start)
 198                         xen_extra_mem_start = end;
 199
 200                 /* If region is non-RAM or below mem_end, add what remains */
 201                 if ((map[i].type != E820_RAM || map[i].addr < mem_end) &&
 202                     map[i].size > 0)
 203                         e820_add_region(map[i].addr, map[i].size, map[i].type);
 204         }
 205
 206         /*
 207          * In domU, the ISA region is normal, usable memory, but we
 208          * reserve ISA memory anyway because too many things poke
 209          * about in there.
 210          *
 211          * In Dom0, the host E820 information can leave gaps in the
 212          * ISA range, which would cause us to release those pages.  To
 213          * avoid this, we unconditionally reserve them here.
 214          */
 215         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 216                         E820_RESERVED);
 217
 218         /*
 219          * Reserve Xen bits:
 220          *  - mfn_list
 221          *  - xen_start_info
 222          * See comment above "struct start_info" in <xen/interface/xen.h>
 223          */
 224         memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
 225                       __pa(xen_start_info->pt_base),
 226                         "XEN START INFO");
 227
 228         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 229
 230         extra_pages += xen_return_unused_memory(xen_start_info->nr_pages, &e820);
 231
 232         /*
 233          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
 234          * factor the base size.  On non-highmem systems, the base
 235          * size is the full initial memory allocation; on highmem it
 236          * is limited to the max size of lowmem, so that it doesn't
 237          * get completely filled.
 238          *
 239          * In principle there could be a problem in lowmem systems if
 240          * the initial memory is also very large with respect to
 241          * lowmem, but we won't try to deal with that here.
 242          */
 243         extra_limit = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
 244                           max_pfn + extra_pages);
 245
 246         if (extra_limit >= max_pfn)
 247                 extra_pages = extra_limit - max_pfn;
 248         else
 249                 extra_pages = 0;
 250
 251         if (!xen_initial_domain())
 252                 xen_add_extra_mem(extra_pages);
 253
 254         return "Xen";
 255 }
 256
 257 static void xen_idle(void)
 258 {
 259         local_irq_disable();
 260
 261         if (need_resched())
 262                 local_irq_enable();
 263         else {
 264                 current_thread_info()->status &= ~TS_POLLING;
 265                 smp_mb__after_clear_bit();
 266                 safe_halt();
 267                 current_thread_info()->status |= TS_POLLING;
 268         }
 269 }
 270
 271 /*
 272  * Set the bit indicating "nosegneg" library variants should be used.
 273  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 274  * can have un-truncated segments, so wrapping around is allowed.
 275  */
 276 static void __init fiddle_vdso(void)
 277 {
 278 #ifdef CONFIG_X86_32
 279         u32 *mask;
 280         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 281         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 282         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 283         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 284 #endif
 285 }
 286
 287 static __cpuinit int register_callback(unsigned type, const void *func)
 288 {
 289         struct callback_register callback = {
 290                 .type = type,
 291                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 292                 .flags = CALLBACKF_mask_events,
 293         };
 294
 295         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 296 }
 297
 298 void __cpuinit xen_enable_sysenter(void)
 299 {
 300         int ret;
 301         unsigned sysenter_feature;
 302
 303 #ifdef CONFIG_X86_32
 304         sysenter_feature = X86_FEATURE_SEP;
 305 #else
 306         sysenter_feature = X86_FEATURE_SYSENTER32;
 307 #endif
 308
 309         if (!boot_cpu_has(sysenter_feature))
 310                 return;
 311
 312         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 313         if(ret != 0)
 314                 setup_clear_cpu_cap(sysenter_feature);
 315 }
 316
 317 void __cpuinit xen_enable_syscall(void)
 318 {
 319 #ifdef CONFIG_X86_64
 320         int ret;
 321
 322         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 323         if (ret != 0) {
 324                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 325                 /* Pretty fatal; 64-bit userspace has no other
 326                    mechanism for syscalls. */
 327         }
 328
 329         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 330                 ret = register_callback(CALLBACKTYPE_syscall32,
 331                                         xen_syscall32_target);
 332                 if (ret != 0)
 333                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 334         }
 335 #endif /* CONFIG_X86_64 */
 336 }
 337
 338 void __init xen_arch_setup(void)
 339 {
 340         struct physdev_set_iopl set_iopl;
 341         int rc;
 342
 343         xen_panic_handler_init();
 344
 345         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 346         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 347
 348         if (!xen_feature(XENFEAT_auto_translated_physmap))
 349                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 350                                      VMASST_TYPE_pae_extended_cr3);
 351
 352         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 353             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 354                 BUG();
 355
 356         xen_enable_sysenter();
 357         xen_enable_syscall();
 358
 359         set_iopl.iopl = 1;
 360         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 361         if (rc != 0)
 362                 printk(KERN_INFO "physdev_op failed %d\n", rc);
 363
 364 #ifdef CONFIG_ACPI
 365         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 366                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 367                 disable_acpi();
 368         }
 369 #endif
 370
 371         memcpy(boot_command_line, xen_start_info->cmd_line,
 372                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 373                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 374
 375         pm_idle = xen_idle;
 376
 377         fiddle_vdso();
 378 }