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
  12 #include <asm/elf.h>
  13 #include <asm/vdso.h>
  14 #include <asm/e820.h>
  15 #include <asm/setup.h>
  16 #include <asm/acpi.h>
  17 #include <asm/xen/hypervisor.h>
  18 #include <asm/xen/hypercall.h>
  19
  20 #include <xen/page.h>
  21 #include <xen/interface/callback.h>
  22 #include <xen/interface/physdev.h>
  23 #include <xen/features.h>
  24
  25 #include "xen-ops.h"
  26 #include "vdso.h"
  27
  28 /* These are code, but not functions.  Defined in entry.S */
  29 extern const char xen_hypervisor_callback[];
  30 extern const char xen_failsafe_callback[];
  31
  32
  33 /**
  34  * machine_specific_memory_setup - Hook for machine specific memory setup.
  35  **/
  36
  37 char * __init xen_memory_setup(void)
  38 {
  39         unsigned long max_pfn = xen_start_info->nr_pages;
  40
  41         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
  42
  43         e820.nr_map = 0;
  44
  45         e820_add_region(0, PFN_PHYS((u64)max_pfn), E820_RAM);
  46
  47         /*
  48          * Even though this is normal, usable memory under Xen, reserve
  49          * ISA memory anyway because too many things think they can poke
  50          * about in there.
  51          */
  52         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
  53                         E820_RESERVED);
  54
  55         /*
  56          * Reserve Xen bits:
  57          *  - mfn_list
  58          *  - xen_start_info
  59          * See comment above "struct start_info" in <xen/interface/xen.h>
  60          */
  61         e820_add_region(__pa(xen_start_info->mfn_list),
  62                         xen_start_info->pt_base - xen_start_info->mfn_list,
  63                         E820_RESERVED);
  64
  65         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
  66
  67         return "Xen";
  68 }
  69
  70 static void xen_idle(void)
  71 {
  72         local_irq_disable();
  73
  74         if (need_resched())
  75                 local_irq_enable();
  76         else {
  77                 current_thread_info()->status &= ~TS_POLLING;
  78                 smp_mb__after_clear_bit();
  79                 safe_halt();
  80                 current_thread_info()->status |= TS_POLLING;
  81         }
  82 }
  83
  84 /*
  85  * Set the bit indicating "nosegneg" library variants should be used.
  86  * We only need to bother in pure 32-bit mode; compat 32-bit processes
  87  * can have un-truncated segments, so wrapping around is allowed.
  88  */
  89 static void __init fiddle_vdso(void)
  90 {
  91 #ifdef CONFIG_X86_32
  92         u32 *mask;
  93         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
  94         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
  95         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
  96         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
  97 #endif
  98 }
  99
 100 static __cpuinit int register_callback(unsigned type, const void *func)
 101 {
 102         struct callback_register callback = {
 103                 .type = type,
 104                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 105                 .flags = CALLBACKF_mask_events,
 106         };
 107
 108         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 109 }
 110
 111 void __cpuinit xen_enable_sysenter(void)
 112 {
 113         extern void xen_sysenter_target(void);
 114         int ret;
 115         unsigned sysenter_feature;
 116
 117 #ifdef CONFIG_X86_32
 118         sysenter_feature = X86_FEATURE_SEP;
 119 #else
 120         sysenter_feature = X86_FEATURE_SYSENTER32;
 121 #endif
 122
 123         if (!boot_cpu_has(sysenter_feature))
 124                 return;
 125
 126         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 127         if(ret != 0)
 128                 setup_clear_cpu_cap(sysenter_feature);
 129 }
 130
 131 void __cpuinit xen_enable_syscall(void)
 132 {
 133 #ifdef CONFIG_X86_64
 134         int ret;
 135         extern void xen_syscall_target(void);
 136         extern void xen_syscall32_target(void);
 137
 138         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 139         if (ret != 0) {
 140                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 141                 /* Pretty fatal; 64-bit userspace has no other
 142                    mechanism for syscalls. */
 143         }
 144
 145         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 146                 ret = register_callback(CALLBACKTYPE_syscall32,
 147                                         xen_syscall32_target);
 148                 if (ret != 0)
 149                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 150         }
 151 #endif /* CONFIG_X86_64 */
 152 }
 153
 154 void __init xen_arch_setup(void)
 155 {
 156         struct physdev_set_iopl set_iopl;
 157         int rc;
 158
 159         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 160         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 161
 162         if (!xen_feature(XENFEAT_auto_translated_physmap))
 163                 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_pae_extended_cr3);
 164
 165         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 166             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 167                 BUG();
 168
 169         xen_enable_sysenter();
 170         xen_enable_syscall();
 171
 172         set_iopl.iopl = 1;
 173         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 174         if (rc != 0)
 175                 printk(KERN_INFO "physdev_op failed %d\n", rc);
 176
 177 #ifdef CONFIG_ACPI
 178         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 179                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 180                 disable_acpi();
 181         }
 182 #endif
 183
 184         memcpy(boot_command_line, xen_start_info->cmd_line,
 185                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 186                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 187
 188         pm_idle = xen_idle;
 189
 190         paravirt_disable_iospace();
 191
 192         fiddle_vdso();
 193 }