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/start_kernel.h>
  23 #include <linux/highmem.h>
  24
  25 #include <asm/bug.h>
  26 #include <asm/paravirt.h>
  27 #include <asm/desc.h>
  28 #include <asm/setup.h>
  29 #include <asm/arch_hooks.h>
  30 #include <asm/time.h>
  31 #include <asm/irq.h>
  32 #include <asm/delay.h>
  33 #include <asm/fixmap.h>
  34 #include <asm/apic.h>
  35 #include <asm/tlbflush.h>
  36 #include <asm/timer.h>
  37
  38 /* nop stub */
  39 void _paravirt_nop(void)
  40 {
  41 }
  42
  43 static void __init default_banner(void)
  44 {
  45         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  46                paravirt_ops.name);
  47 }
  48
  49 char *memory_setup(void)
  50 {
  51         return paravirt_ops.memory_setup();
  52 }
  53
  54 /* Simple instruction patching code. */
  55 #define DEF_NATIVE(name, code)                                  \
  56         extern const char start_##name[], end_##name[];         \
  57         asm("start_" #name ": " code "; end_" #name ":")
  58
  59 DEF_NATIVE(irq_disable, "cli");
  60 DEF_NATIVE(irq_enable, "sti");
  61 DEF_NATIVE(restore_fl, "push %eax; popf");
  62 DEF_NATIVE(save_fl, "pushf; pop %eax");
  63 DEF_NATIVE(iret, "iret");
  64 DEF_NATIVE(irq_enable_sysexit, "sti; sysexit");
  65 DEF_NATIVE(read_cr2, "mov %cr2, %eax");
  66 DEF_NATIVE(write_cr3, "mov %eax, %cr3");
  67 DEF_NATIVE(read_cr3, "mov %cr3, %eax");
  68 DEF_NATIVE(clts, "clts");
  69 DEF_NATIVE(read_tsc, "rdtsc");
  70
  71 DEF_NATIVE(ud2a, "ud2a");
  72
  73 static unsigned native_patch(u8 type, u16 clobbers, void *insns, unsigned len)
  74 {
  75         const unsigned char *start, *end;
  76         unsigned ret;
  77
  78         switch(type) {
  79 #define SITE(x) case PARAVIRT_PATCH(x): start = start_##x; end = end_##x; goto patch_site
  80                 SITE(irq_disable);
  81                 SITE(irq_enable);
  82                 SITE(restore_fl);
  83                 SITE(save_fl);
  84                 SITE(iret);
  85                 SITE(irq_enable_sysexit);
  86                 SITE(read_cr2);
  87                 SITE(read_cr3);
  88                 SITE(write_cr3);
  89                 SITE(clts);
  90                 SITE(read_tsc);
  91 #undef SITE
  92
  93         patch_site:
  94                 ret = paravirt_patch_insns(insns, len, start, end);
  95                 break;
  96
  97         case PARAVIRT_PATCH(make_pgd):
  98         case PARAVIRT_PATCH(make_pte):
  99         case PARAVIRT_PATCH(pgd_val):
 100         case PARAVIRT_PATCH(pte_val):
 101 #ifdef CONFIG_X86_PAE
 102         case PARAVIRT_PATCH(make_pmd):
 103         case PARAVIRT_PATCH(pmd_val):
 104 #endif
 105                 /* These functions end up returning exactly what
 106                    they're passed, in the same registers. */
 107                 ret = paravirt_patch_nop();
 108                 break;
 109
 110         default:
 111                 ret = paravirt_patch_default(type, clobbers, insns, len);
 112                 break;
 113         }
 114
 115         return ret;
 116 }
 117
 118 unsigned paravirt_patch_nop(void)
 119 {
 120         return 0;
 121 }
 122
 123 unsigned paravirt_patch_ignore(unsigned len)
 124 {
 125         return len;
 126 }
 127
 128 unsigned paravirt_patch_call(void *target, u16 tgt_clobbers,
 129                              void *site, u16 site_clobbers,
 130                              unsigned len)
 131 {
 132         unsigned char *call = site;
 133         unsigned long delta = (unsigned long)target - (unsigned long)(call+5);
 134
 135         if (tgt_clobbers & ~site_clobbers)
 136                 return len;     /* target would clobber too much for this site */
 137         if (len < 5)
 138                 return len;     /* call too long for patch site */
 139
 140         *call++ = 0xe8;         /* call */
 141         *(unsigned long *)call = delta;
 142
 143         return 5;
 144 }
 145
 146 unsigned paravirt_patch_jmp(void *target, void *site, unsigned len)
 147 {
 148         unsigned char *jmp = site;
 149         unsigned long delta = (unsigned long)target - (unsigned long)(jmp+5);
 150
 151         if (len < 5)
 152                 return len;     /* call too long for patch site */
 153
 154         *jmp++ = 0xe9;          /* jmp */
 155         *(unsigned long *)jmp = delta;
 156
 157         return 5;
 158 }
 159
 160 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *site, unsigned len)
 161 {
 162         void *opfunc = *((void **)&paravirt_ops + type);
 163         unsigned ret;
 164
 165         if (opfunc == NULL)
 166                 /* If there's no function, patch it with a ud2a (BUG) */
 167                 ret = paravirt_patch_insns(site, len, start_ud2a, end_ud2a);
 168         else if (opfunc == paravirt_nop)
 169                 /* If the operation is a nop, then nop the callsite */
 170                 ret = paravirt_patch_nop();
 171         else if (type == PARAVIRT_PATCH(iret) ||
 172                  type == PARAVIRT_PATCH(irq_enable_sysexit))
 173                 /* If operation requires a jmp, then jmp */
 174                 ret = paravirt_patch_jmp(opfunc, site, len);
 175         else
 176                 /* Otherwise call the function; assume target could
 177                    clobber any caller-save reg */
 178                 ret = paravirt_patch_call(opfunc, CLBR_ANY,
 179                                           site, clobbers, len);
 180
 181         return ret;
 182 }
 183
 184 unsigned paravirt_patch_insns(void *site, unsigned len,
 185                               const char *start, const char *end)
 186 {
 187         unsigned insn_len = end - start;
 188
 189         if (insn_len > len || start == NULL)
 190                 insn_len = len;
 191         else
 192                 memcpy(site, start, insn_len);
 193
 194         return insn_len;
 195 }
 196
 197 void init_IRQ(void)
 198 {
 199         paravirt_ops.init_IRQ();
 200 }
 201
 202 static void native_flush_tlb(void)
 203 {
 204         __native_flush_tlb();
 205 }
 206
 207 /*
 208  * Global pages have to be flushed a bit differently. Not a real
 209  * performance problem because this does not happen often.
 210  */
 211 static void native_flush_tlb_global(void)
 212 {
 213         __native_flush_tlb_global();
 214 }
 215
 216 static void native_flush_tlb_single(unsigned long addr)
 217 {
 218         __native_flush_tlb_single(addr);
 219 }
 220
 221 /* These are in entry.S */
 222 extern void native_iret(void);
 223 extern void native_irq_enable_sysexit(void);
 224
 225 static int __init print_banner(void)
 226 {
 227         paravirt_ops.banner();
 228         return 0;
 229 }
 230 core_initcall(print_banner);
 231
 232 struct paravirt_ops paravirt_ops = {
 233         .name = "bare hardware",
 234         .paravirt_enabled = 0,
 235         .kernel_rpl = 0,
 236         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 237
 238         .patch = native_patch,
 239         .banner = default_banner,
 240         .arch_setup = paravirt_nop,
 241         .memory_setup = machine_specific_memory_setup,
 242         .get_wallclock = native_get_wallclock,
 243         .set_wallclock = native_set_wallclock,
 244         .time_init = hpet_time_init,
 245         .init_IRQ = native_init_IRQ,
 246
 247         .cpuid = native_cpuid,
 248         .get_debugreg = native_get_debugreg,
 249         .set_debugreg = native_set_debugreg,
 250         .clts = native_clts,
 251         .read_cr0 = native_read_cr0,
 252         .write_cr0 = native_write_cr0,
 253         .read_cr2 = native_read_cr2,
 254         .write_cr2 = native_write_cr2,
 255         .read_cr3 = native_read_cr3,
 256         .write_cr3 = native_write_cr3,
 257         .read_cr4 = native_read_cr4,
 258         .read_cr4_safe = native_read_cr4_safe,
 259         .write_cr4 = native_write_cr4,
 260         .save_fl = native_save_fl,
 261         .restore_fl = native_restore_fl,
 262         .irq_disable = native_irq_disable,
 263         .irq_enable = native_irq_enable,
 264         .safe_halt = native_safe_halt,
 265         .halt = native_halt,
 266         .wbinvd = native_wbinvd,
 267         .read_msr = native_read_msr_safe,
 268         .write_msr = native_write_msr_safe,
 269         .read_tsc = native_read_tsc,
 270         .read_pmc = native_read_pmc,
 271         .get_scheduled_cycles = native_read_tsc,
 272         .get_cpu_khz = native_calculate_cpu_khz,
 273         .load_tr_desc = native_load_tr_desc,
 274         .set_ldt = native_set_ldt,
 275         .load_gdt = native_load_gdt,
 276         .load_idt = native_load_idt,
 277         .store_gdt = native_store_gdt,
 278         .store_idt = native_store_idt,
 279         .store_tr = native_store_tr,
 280         .load_tls = native_load_tls,
 281         .write_ldt_entry = write_dt_entry,
 282         .write_gdt_entry = write_dt_entry,
 283         .write_idt_entry = write_dt_entry,
 284         .load_esp0 = native_load_esp0,
 285
 286         .set_iopl_mask = native_set_iopl_mask,
 287         .io_delay = native_io_delay,
 288
 289 #ifdef CONFIG_X86_LOCAL_APIC
 290         .apic_write = native_apic_write,
 291         .apic_write_atomic = native_apic_write_atomic,
 292         .apic_read = native_apic_read,
 293         .setup_boot_clock = setup_boot_APIC_clock,
 294         .setup_secondary_clock = setup_secondary_APIC_clock,
 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         .startup_ipi_hook = paravirt_nop,
 347 };
 348
 349 /*
 350  * NOTE: CONFIG_PARAVIRT is experimental and the paravirt_ops
 351  * semantics are subject to change. Hence we only do this
 352  * internal-only export of this, until it gets sorted out and
 353  * all lowlevel CPU ops used by modules are separately exported.
 354  */
 355 EXPORT_SYMBOL_GPL(paravirt_ops);