arch/i386/kernel/paravirt.c

   1 /*  Paravirtualization interfaces
   2     Copyright (C) 2006 Rusty Russell IBM Corporation
   3
   4     This program is free software; you can redistribute it and/or modify
   5     it under the terms of the GNU General Public License as published by
   6     the Free Software Foundation; either version 2 of the License, or
   7     (at your option) any later version.
   8
   9     This program is distributed in the hope that it will be useful,
  10     but WITHOUT ANY WARRANTY; without even the implied warranty of
  11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12     GNU General Public License for more details.
  13
  14     You should have received a copy of the GNU General Public License
  15     along with this program; if not, write to the Free Software
  16     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18 #include <linux/errno.h>
  19 #include <linux/module.h>
  20 #include <linux/efi.h>
  21 #include <linux/bcd.h>
  22 #include <linux/highmem.h>
  23
  24 #include <asm/bug.h>
  25 #include <asm/paravirt.h>
  26 #include <asm/desc.h>
  27 #include <asm/setup.h>
  28 #include <asm/arch_hooks.h>
  29 #include <asm/time.h>
  30 #include <asm/irq.h>
  31 #include <asm/delay.h>
  32 #include <asm/fixmap.h>
  33 #include <asm/apic.h>
  34 #include <asm/tlbflush.h>
  35 #include <asm/timer.h>
  36
  37 /* nop stub */
  38 void _paravirt_nop(void)
  39 {
  40 }
  41
  42 static void __init default_banner(void)
  43 {
  44         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  45                paravirt_ops.name);
  46 }
  47
  48 char *memory_setup(void)
  49 {
  50         return paravirt_ops.memory_setup();
  51 }
  52
  53 /* Simple instruction patching code. */
  54 #define DEF_NATIVE(name, code)                                  \
  55         extern const char start_##name[], end_##name[];         \
  56         asm("start_" #name ": " code "; end_" #name ":")
  57
  58 DEF_NATIVE(irq_disable, "cli");
  59 DEF_NATIVE(irq_enable, "sti");
  60 DEF_NATIVE(restore_fl, "push %eax; popf");
  61 DEF_NATIVE(save_fl, "pushf; pop %eax");
  62 DEF_NATIVE(iret, "iret");
  63 DEF_NATIVE(irq_enable_sysexit, "sti; sysexit");
  64 DEF_NATIVE(read_cr2, "mov %cr2, %eax");
  65 DEF_NATIVE(write_cr3, "mov %eax, %cr3");
  66 DEF_NATIVE(read_cr3, "mov %cr3, %eax");
  67 DEF_NATIVE(clts, "clts");
  68 DEF_NATIVE(read_tsc, "rdtsc");
  69
  70 DEF_NATIVE(ud2a, "ud2a");
  71
  72 static unsigned native_patch(u8 type, u16 clobbers, void *insns, unsigned len)
  73 {
  74         const unsigned char *start, *end;
  75         unsigned ret;
  76
  77         switch(type) {
  78 #define SITE(x) case PARAVIRT_PATCH(x): start = start_##x; end = end_##x; goto patch_site
  79                 SITE(irq_disable);
  80                 SITE(irq_enable);
  81                 SITE(restore_fl);
  82                 SITE(save_fl);
  83                 SITE(iret);
  84                 SITE(irq_enable_sysexit);
  85                 SITE(read_cr2);
  86                 SITE(read_cr3);
  87                 SITE(write_cr3);
  88                 SITE(clts);
  89                 SITE(read_tsc);
  90 #undef SITE
  91
  92         patch_site:
  93                 ret = paravirt_patch_insns(insns, len, start, end);
  94                 break;
  95
  96         case PARAVIRT_PATCH(make_pgd):
  97         case PARAVIRT_PATCH(make_pte):
  98         case PARAVIRT_PATCH(pgd_val):
  99         case PARAVIRT_PATCH(pte_val):
 100 #ifdef CONFIG_X86_PAE
 101         case PARAVIRT_PATCH(make_pmd):
 102         case PARAVIRT_PATCH(pmd_val):
 103 #endif
 104                 /* These functions end up returning exactly what
 105                    they're passed, in the same registers. */
 106                 ret = paravirt_patch_nop();
 107                 break;
 108
 109         default:
 110                 ret = paravirt_patch_default(type, clobbers, insns, len);
 111                 break;
 112         }
 113
 114         return ret;
 115 }
 116
 117 unsigned paravirt_patch_nop(void)
 118 {
 119         return 0;
 120 }
 121
 122 unsigned paravirt_patch_ignore(unsigned len)
 123 {
 124         return len;
 125 }
 126
 127 unsigned paravirt_patch_call(void *target, u16 tgt_clobbers,
 128                              void *site, u16 site_clobbers,
 129                              unsigned len)
 130 {
 131         unsigned char *call = site;
 132         unsigned long delta = (unsigned long)target - (unsigned long)(call+5);
 133
 134         if (tgt_clobbers & ~site_clobbers)
 135                 return len;     /* target would clobber too much for this site */
 136         if (len < 5)
 137                 return len;     /* call too long for patch site */
 138
 139         *call++ = 0xe8;         /* call */
 140         *(unsigned long *)call = delta;
 141
 142         return 5;
 143 }
 144
 145 unsigned paravirt_patch_jmp(void *target, void *site, unsigned len)
 146 {
 147         unsigned char *jmp = site;
 148         unsigned long delta = (unsigned long)target - (unsigned long)(jmp+5);
 149
 150         if (len < 5)
 151                 return len;     /* call too long for patch site */
 152
 153         *jmp++ = 0xe9;          /* jmp */
 154         *(unsigned long *)jmp = delta;
 155
 156         return 5;
 157 }
 158
 159 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *site, unsigned len)
 160 {
 161         void *opfunc = *((void **)&paravirt_ops + type);
 162         unsigned ret;
 163
 164         if (opfunc == NULL)
 165                 /* If there's no function, patch it with a ud2a (BUG) */
 166                 ret = paravirt_patch_insns(site, len, start_ud2a, end_ud2a);
 167         else if (opfunc == paravirt_nop)
 168                 /* If the operation is a nop, then nop the callsite */
 169                 ret = paravirt_patch_nop();
 170         else if (type == PARAVIRT_PATCH(iret) ||
 171                  type == PARAVIRT_PATCH(irq_enable_sysexit))
 172                 /* If operation requires a jmp, then jmp */
 173                 ret = paravirt_patch_jmp(opfunc, site, len);
 174         else
 175                 /* Otherwise call the function; assume target could
 176                    clobber any caller-save reg */
 177                 ret = paravirt_patch_call(opfunc, CLBR_ANY,
 178                                           site, clobbers, len);
 179
 180         return ret;
 181 }
 182
 183 unsigned paravirt_patch_insns(void *site, unsigned len,
 184                               const char *start, const char *end)
 185 {
 186         unsigned insn_len = end - start;
 187
 188         if (insn_len > len || start == NULL)
 189                 insn_len = len;
 190         else
 191                 memcpy(site, start, insn_len);
 192
 193         return insn_len;
 194 }
 195
 196 void init_IRQ(void)
 197 {
 198         paravirt_ops.init_IRQ();
 199 }
 200
 201 static void native_flush_tlb(void)
 202 {
 203         __native_flush_tlb();
 204 }
 205
 206 /*
 207  * Global pages have to be flushed a bit differently. Not a real
 208  * performance problem because this does not happen often.
 209  */
 210 static void native_flush_tlb_global(void)
 211 {
 212         __native_flush_tlb_global();
 213 }
 214
 215 static void native_flush_tlb_single(unsigned long addr)
 216 {
 217         __native_flush_tlb_single(addr);
 218 }
 219
 220 /* These are in entry.S */
 221 extern void native_iret(void);
 222 extern void native_irq_enable_sysexit(void);
 223
 224 static int __init print_banner(void)
 225 {
 226         paravirt_ops.banner();
 227         return 0;
 228 }
 229 core_initcall(print_banner);
 230
 231 struct paravirt_ops paravirt_ops = {
 232         .name = "bare hardware",
 233         .paravirt_enabled = 0,
 234         .kernel_rpl = 0,
 235         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 236
 237         .patch = native_patch,
 238         .banner = default_banner,
 239         .arch_setup = paravirt_nop,
 240         .memory_setup = machine_specific_memory_setup,
 241         .get_wallclock = native_get_wallclock,
 242         .set_wallclock = native_set_wallclock,
 243         .time_init = hpet_time_init,
 244         .init_IRQ = native_init_IRQ,
 245
 246         .cpuid = native_cpuid,
 247         .get_debugreg = native_get_debugreg,
 248         .set_debugreg = native_set_debugreg,
 249         .clts = native_clts,
 250         .read_cr0 = native_read_cr0,
 251         .write_cr0 = native_write_cr0,
 252         .read_cr2 = native_read_cr2,
 253         .write_cr2 = native_write_cr2,
 254         .read_cr3 = native_read_cr3,
 255         .write_cr3 = native_write_cr3,
 256         .read_cr4 = native_read_cr4,
 257         .read_cr4_safe = native_read_cr4_safe,
 258         .write_cr4 = native_write_cr4,
 259         .save_fl = native_save_fl,
 260         .restore_fl = native_restore_fl,
 261         .irq_disable = native_irq_disable,
 262         .irq_enable = native_irq_enable,
 263         .safe_halt = native_safe_halt,
 264         .halt = native_halt,
 265         .wbinvd = native_wbinvd,
 266         .read_msr = native_read_msr_safe,
 267         .write_msr = native_write_msr_safe,
 268         .read_tsc = native_read_tsc,
 269         .read_pmc = native_read_pmc,
 270         .get_scheduled_cycles = native_read_tsc,
 271         .get_cpu_khz = native_calculate_cpu_khz,
 272         .load_tr_desc = native_load_tr_desc,
 273         .set_ldt = native_set_ldt,
 274         .load_gdt = native_load_gdt,
 275         .load_idt = native_load_idt,
 276         .store_gdt = native_store_gdt,
 277         .store_idt = native_store_idt,
 278         .store_tr = native_store_tr,
 279         .load_tls = native_load_tls,
 280         .write_ldt_entry = write_dt_entry,
 281         .write_gdt_entry = write_dt_entry,
 282         .write_idt_entry = write_dt_entry,
 283         .load_esp0 = native_load_esp0,
 284
 285         .set_iopl_mask = native_set_iopl_mask,
 286         .io_delay = native_io_delay,
 287
 288 #ifdef CONFIG_X86_LOCAL_APIC
 289         .apic_write = native_apic_write,
 290         .apic_write_atomic = native_apic_write_atomic,
 291         .apic_read = native_apic_read,
 292         .setup_boot_clock = setup_boot_APIC_clock,
 293         .setup_secondary_clock = setup_secondary_APIC_clock,
 294         .startup_ipi_hook = paravirt_nop,
 295 #endif
 296         .set_lazy_mode = paravirt_nop,
 297
 298         .pagetable_setup_start = native_pagetable_setup_start,
 299         .pagetable_setup_done = native_pagetable_setup_done,
 300
 301         .flush_tlb_user = native_flush_tlb,
 302         .flush_tlb_kernel = native_flush_tlb_global,
 303         .flush_tlb_single = native_flush_tlb_single,
 304         .flush_tlb_others = native_flush_tlb_others,
 305
 306         .alloc_pt = paravirt_nop,
 307         .alloc_pd = paravirt_nop,
 308         .alloc_pd_clone = paravirt_nop,
 309         .release_pt = paravirt_nop,
 310         .release_pd = paravirt_nop,
 311
 312         .set_pte = native_set_pte,
 313         .set_pte_at = native_set_pte_at,
 314         .set_pmd = native_set_pmd,
 315         .pte_update = paravirt_nop,
 316         .pte_update_defer = paravirt_nop,
 317
 318 #ifdef CONFIG_HIGHPTE
 319         .kmap_atomic_pte = kmap_atomic,
 320 #endif
 321
 322 #ifdef CONFIG_X86_PAE
 323         .set_pte_atomic = native_set_pte_atomic,
 324         .set_pte_present = native_set_pte_present,
 325         .set_pud = native_set_pud,
 326         .pte_clear = native_pte_clear,
 327         .pmd_clear = native_pmd_clear,
 328
 329         .pmd_val = native_pmd_val,
 330         .make_pmd = native_make_pmd,
 331 #endif
 332
 333         .pte_val = native_pte_val,
 334         .pgd_val = native_pgd_val,
 335
 336         .make_pte = native_make_pte,
 337         .make_pgd = native_make_pgd,
 338
 339         .irq_enable_sysexit = native_irq_enable_sysexit,
 340         .iret = native_iret,
 341
 342         .dup_mmap = paravirt_nop,
 343         .exit_mmap = paravirt_nop,
 344         .activate_mm = paravirt_nop,
 345 };
 346
 347 EXPORT_SYMBOL(paravirt_ops);