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/page.h>
  22 #include <xen/interface/callback.h>
  23 #include <xen/interface/physdev.h>
  24 #include <xen/interface/memory.h>
  25 #include <xen/features.h>
  26
  27 #include "xen-ops.h"
  28 #include "vdso.h"
  29
  30 /* These are code, but not functions.  Defined in entry.S */
  31 extern const char xen_hypervisor_callback[];
  32 extern const char xen_failsafe_callback[];
  33 extern void xen_sysenter_target(void);
  34 extern void xen_syscall_target(void);
  35 extern void xen_syscall32_target(void);
  36
  37 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
  38                                               phys_addr_t end_addr)
  39 {
  40         struct xen_memory_reservation reservation = {
  41                 .address_bits = 0,
  42                 .extent_order = 0,
  43                 .domid        = DOMID_SELF
  44         };
  45         unsigned long start, end;
  46         unsigned long len = 0;
  47         unsigned long pfn;
  48         int ret;
  49
  50         start = PFN_UP(start_addr);
  51         end = PFN_DOWN(end_addr);
  52
  53         if (end <= start)
  54                 return 0;
  55
  56         printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
  57                start, end);
  58         for(pfn = start; pfn < end; pfn++) {
  59                 unsigned long mfn = pfn_to_mfn(pfn);
  60
  61                 /* Make sure pfn exists to start with */
  62                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
  63                         continue;
  64
  65                 set_xen_guest_handle(reservation.extent_start, &mfn);
  66                 reservation.nr_extents = 1;
  67
  68                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
  69                                            &reservation);
  70                 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
  71                      start, end, ret);
  72                 if (ret == 1) {
  73                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
  74                         len++;
  75                 }
  76         }
  77         printk(KERN_CONT "%ld pages freed\n", len);
  78
  79         return len;
  80 }
  81
  82 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
  83                                                      const struct e820map *e820)
  84 {
  85         phys_addr_t max_addr = PFN_PHYS(max_pfn);
  86         phys_addr_t last_end = 0;
  87         unsigned long released = 0;
  88         int i;
  89
  90         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
  91                 phys_addr_t end = e820->map[i].addr;
  92                 end = min(max_addr, end);
  93
  94                 released += xen_release_chunk(last_end, end);
  95                 last_end = e820->map[i].addr + e820->map[i].size;
  96         }
  97
  98         if (last_end < max_addr)
  99                 released += xen_release_chunk(last_end, max_addr);
 100
 101         printk(KERN_INFO "released %ld pages of unused memory\n", released);
 102         return released;
 103 }
 104
 105 /**
 106  * machine_specific_memory_setup - Hook for machine specific memory setup.
 107  **/
 108
 109 char * __init xen_memory_setup(void)
 110 {
 111         unsigned long max_pfn = xen_start_info->nr_pages;
 112
 113         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 114
 115         e820.nr_map = 0;
 116
 117         e820_add_region(0, PFN_PHYS((u64)max_pfn), E820_RAM);
 118
 119         /*
 120          * Even though this is normal, usable memory under Xen, reserve
 121          * ISA memory anyway because too many things think they can poke
 122          * about in there.
 123          */
 124         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 125                         E820_RESERVED);
 126
 127         /*
 128          * Reserve Xen bits:
 129          *  - mfn_list
 130          *  - xen_start_info
 131          * See comment above "struct start_info" in <xen/interface/xen.h>
 132          */
 133         memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
 134                       __pa(xen_start_info->pt_base),
 135                         "XEN START INFO");
 136
 137         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 138
 139         xen_return_unused_memory(xen_start_info->nr_pages, &e820);
 140
 141         return "Xen";
 142 }
 143
 144 static void xen_idle(void)
 145 {
 146         local_irq_disable();
 147
 148         if (need_resched())
 149                 local_irq_enable();
 150         else {
 151                 current_thread_info()->status &= ~TS_POLLING;
 152                 smp_mb__after_clear_bit();
 153                 safe_halt();
 154                 current_thread_info()->status |= TS_POLLING;
 155         }
 156 }
 157
 158 /*
 159  * Set the bit indicating "nosegneg" library variants should be used.
 160  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 161  * can have un-truncated segments, so wrapping around is allowed.
 162  */
 163 static void __init fiddle_vdso(void)
 164 {
 165 #ifdef CONFIG_X86_32
 166         u32 *mask;
 167         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 168         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 169         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 170         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 171 #endif
 172 }
 173
 174 static __cpuinit int register_callback(unsigned type, const void *func)
 175 {
 176         struct callback_register callback = {
 177                 .type = type,
 178                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 179                 .flags = CALLBACKF_mask_events,
 180         };
 181
 182         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 183 }
 184
 185 void __cpuinit xen_enable_sysenter(void)
 186 {
 187         int ret;
 188         unsigned sysenter_feature;
 189
 190 #ifdef CONFIG_X86_32
 191         sysenter_feature = X86_FEATURE_SEP;
 192 #else
 193         sysenter_feature = X86_FEATURE_SYSENTER32;
 194 #endif
 195
 196         if (!boot_cpu_has(sysenter_feature))
 197                 return;
 198
 199         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 200         if(ret != 0)
 201                 setup_clear_cpu_cap(sysenter_feature);
 202 }
 203
 204 void __cpuinit xen_enable_syscall(void)
 205 {
 206 #ifdef CONFIG_X86_64
 207         int ret;
 208
 209         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 210         if (ret != 0) {
 211                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 212                 /* Pretty fatal; 64-bit userspace has no other
 213                    mechanism for syscalls. */
 214         }
 215
 216         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 217                 ret = register_callback(CALLBACKTYPE_syscall32,
 218                                         xen_syscall32_target);
 219                 if (ret != 0)
 220                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 221         }
 222 #endif /* CONFIG_X86_64 */
 223 }
 224
 225 void __init xen_arch_setup(void)
 226 {
 227         struct physdev_set_iopl set_iopl;
 228         int rc;
 229
 230         xen_panic_handler_init();
 231
 232         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 233         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 234
 235         if (!xen_feature(XENFEAT_auto_translated_physmap))
 236                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 237                                      VMASST_TYPE_pae_extended_cr3);
 238
 239         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 240             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 241                 BUG();
 242
 243         xen_enable_sysenter();
 244         xen_enable_syscall();
 245
 246         set_iopl.iopl = 1;
 247         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 248         if (rc != 0)
 249                 printk(KERN_INFO "physdev_op failed %d\n", rc);
 250
 251 #ifdef CONFIG_ACPI
 252         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 253                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 254                 disable_acpi();
 255         }
 256 #endif
 257
 258         memcpy(boot_command_line, xen_start_info->cmd_line,
 259                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 260                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 261
 262         pm_idle = xen_idle;
 263
 264         paravirt_disable_iospace();
 265
 266         fiddle_vdso();
 267 }