arch/x86/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     2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
  19 */
  20
  21 #include <linux/errno.h>
  22 #include <linux/module.h>
  23 #include <linux/efi.h>
  24 #include <linux/bcd.h>
  25 #include <linux/highmem.h>
  26
  27 #include <asm/bug.h>
  28 #include <asm/paravirt.h>
  29 #include <asm/desc.h>
  30 #include <asm/setup.h>
  31 #include <asm/arch_hooks.h>
  32 #include <asm/time.h>
  33 #include <asm/pgalloc.h>
  34 #include <asm/irq.h>
  35 #include <asm/delay.h>
  36 #include <asm/fixmap.h>
  37 #include <asm/apic.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/timer.h>
  40
  41 /* nop stub */
  42 void _paravirt_nop(void)
  43 {
  44 }
  45
  46 static void __init default_banner(void)
  47 {
  48         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  49                pv_info.name);
  50 }
  51
  52 char *memory_setup(void)
  53 {
  54         return pv_init_ops.memory_setup();
  55 }
  56
  57 /* Simple instruction patching code. */
  58 #define DEF_NATIVE(ops, name, code)                                     \
  59         extern const char start_##ops##_##name[], end_##ops##_##name[]; \
  60         asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
  61
  62 /* Undefined instruction for dealing with missing ops pointers. */
  63 static const unsigned char ud2a[] = { 0x0f, 0x0b };
  64
  65 unsigned paravirt_patch_nop(void)
  66 {
  67         return 0;
  68 }
  69
  70 unsigned paravirt_patch_ignore(unsigned len)
  71 {
  72         return len;
  73 }
  74
  75 struct branch {
  76         unsigned char opcode;
  77         u32 delta;
  78 } __attribute__((packed));
  79
  80 unsigned paravirt_patch_call(void *insnbuf,
  81                              const void *target, u16 tgt_clobbers,
  82                              unsigned long addr, u16 site_clobbers,
  83                              unsigned len)
  84 {
  85         struct branch *b = insnbuf;
  86         unsigned long delta = (unsigned long)target - (addr+5);
  87
  88         if (tgt_clobbers & ~site_clobbers)
  89                 return len;     /* target would clobber too much for this site */
  90         if (len < 5)
  91                 return len;     /* call too long for patch site */
  92
  93         b->opcode = 0xe8; /* call */
  94         b->delta = delta;
  95         BUILD_BUG_ON(sizeof(*b) != 5);
  96
  97         return 5;
  98 }
  99
 100 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 101                             unsigned long addr, unsigned len)
 102 {
 103         struct branch *b = insnbuf;
 104         unsigned long delta = (unsigned long)target - (addr+5);
 105
 106         if (len < 5)
 107                 return len;     /* call too long for patch site */
 108
 109         b->opcode = 0xe9;       /* jmp */
 110         b->delta = delta;
 111
 112         return 5;
 113 }
 114
 115 /* Neat trick to map patch type back to the call within the
 116  * corresponding structure. */
 117 static void *get_call_destination(u8 type)
 118 {
 119         struct paravirt_patch_template tmpl = {
 120                 .pv_init_ops = pv_init_ops,
 121                 .pv_time_ops = pv_time_ops,
 122                 .pv_cpu_ops = pv_cpu_ops,
 123                 .pv_irq_ops = pv_irq_ops,
 124                 .pv_apic_ops = pv_apic_ops,
 125                 .pv_mmu_ops = pv_mmu_ops,
 126         };
 127         return *((void **)&tmpl + type);
 128 }
 129
 130 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
 131                                 unsigned long addr, unsigned len)
 132 {
 133         void *opfunc = get_call_destination(type);
 134         unsigned ret;
 135
 136         if (opfunc == NULL)
 137                 /* If there's no function, patch it with a ud2a (BUG) */
 138                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
 139         else if (opfunc == paravirt_nop)
 140                 /* If the operation is a nop, then nop the callsite */
 141                 ret = paravirt_patch_nop();
 142         else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
 143                  type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
 144                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
 145                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
 146                 /* If operation requires a jmp, then jmp */
 147                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 148         else
 149                 /* Otherwise call the function; assume target could
 150                    clobber any caller-save reg */
 151                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
 152                                           addr, clobbers, len);
 153
 154         return ret;
 155 }
 156
 157 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 158                               const char *start, const char *end)
 159 {
 160         unsigned insn_len = end - start;
 161
 162         if (insn_len > len || start == NULL)
 163                 insn_len = len;
 164         else
 165                 memcpy(insnbuf, start, insn_len);
 166
 167         return insn_len;
 168 }
 169
 170 void init_IRQ(void)
 171 {
 172         pv_irq_ops.init_IRQ();
 173 }
 174
 175 static void native_flush_tlb(void)
 176 {
 177         __native_flush_tlb();
 178 }
 179
 180 /*
 181  * Global pages have to be flushed a bit differently. Not a real
 182  * performance problem because this does not happen often.
 183  */
 184 static void native_flush_tlb_global(void)
 185 {
 186         __native_flush_tlb_global();
 187 }
 188
 189 static void native_flush_tlb_single(unsigned long addr)
 190 {
 191         __native_flush_tlb_single(addr);
 192 }
 193
 194 /* These are in entry.S */
 195 extern void native_iret(void);
 196 extern void native_irq_enable_sysexit(void);
 197 extern void native_usergs_sysret32(void);
 198 extern void native_usergs_sysret64(void);
 199
 200 static int __init print_banner(void)
 201 {
 202         pv_init_ops.banner();
 203         return 0;
 204 }
 205 core_initcall(print_banner);
 206
 207 static struct resource reserve_ioports = {
 208         .start = 0,
 209         .end = IO_SPACE_LIMIT,
 210         .name = "paravirt-ioport",
 211         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 212 };
 213
 214 /*
 215  * Reserve the whole legacy IO space to prevent any legacy drivers
 216  * from wasting time probing for their hardware.  This is a fairly
 217  * brute-force approach to disabling all non-virtual drivers.
 218  *
 219  * Note that this must be called very early to have any effect.
 220  */
 221 int paravirt_disable_iospace(void)
 222 {
 223         return request_resource(&ioport_resource, &reserve_ioports);
 224 }
 225
 226 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 227
 228 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 229 {
 230         BUG_ON(__get_cpu_var(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 231         BUG_ON(preemptible());
 232
 233         __get_cpu_var(paravirt_lazy_mode) = mode;
 234 }
 235
 236 void paravirt_leave_lazy(enum paravirt_lazy_mode mode)
 237 {
 238         BUG_ON(__get_cpu_var(paravirt_lazy_mode) != mode);
 239         BUG_ON(preemptible());
 240
 241         __get_cpu_var(paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 242 }
 243
 244 void paravirt_enter_lazy_mmu(void)
 245 {
 246         enter_lazy(PARAVIRT_LAZY_MMU);
 247 }
 248
 249 void paravirt_leave_lazy_mmu(void)
 250 {
 251         paravirt_leave_lazy(PARAVIRT_LAZY_MMU);
 252 }
 253
 254 void paravirt_enter_lazy_cpu(void)
 255 {
 256         enter_lazy(PARAVIRT_LAZY_CPU);
 257 }
 258
 259 void paravirt_leave_lazy_cpu(void)
 260 {
 261         paravirt_leave_lazy(PARAVIRT_LAZY_CPU);
 262 }
 263
 264 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 265 {
 266         return __get_cpu_var(paravirt_lazy_mode);
 267 }
 268
 269 struct pv_info pv_info = {
 270         .name = "bare hardware",
 271         .paravirt_enabled = 0,
 272         .kernel_rpl = 0,
 273         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 274 };
 275
 276 struct pv_init_ops pv_init_ops = {
 277         .patch = native_patch,
 278         .banner = default_banner,
 279         .arch_setup = paravirt_nop,
 280         .memory_setup = machine_specific_memory_setup,
 281 };
 282
 283 struct pv_time_ops pv_time_ops = {
 284         .time_init = hpet_time_init,
 285         .get_wallclock = native_get_wallclock,
 286         .set_wallclock = native_set_wallclock,
 287         .sched_clock = native_sched_clock,
 288         .get_tsc_khz = native_calibrate_tsc,
 289 };
 290
 291 struct pv_irq_ops pv_irq_ops = {
 292         .init_IRQ = native_init_IRQ,
 293         .save_fl = native_save_fl,
 294         .restore_fl = native_restore_fl,
 295         .irq_disable = native_irq_disable,
 296         .irq_enable = native_irq_enable,
 297         .safe_halt = native_safe_halt,
 298         .halt = native_halt,
 299 #ifdef CONFIG_X86_64
 300         .adjust_exception_frame = paravirt_nop,
 301 #endif
 302 };
 303
 304 struct pv_cpu_ops pv_cpu_ops = {
 305         .cpuid = native_cpuid,
 306         .get_debugreg = native_get_debugreg,
 307         .set_debugreg = native_set_debugreg,
 308         .clts = native_clts,
 309         .read_cr0 = native_read_cr0,
 310         .write_cr0 = native_write_cr0,
 311         .read_cr4 = native_read_cr4,
 312         .read_cr4_safe = native_read_cr4_safe,
 313         .write_cr4 = native_write_cr4,
 314 #ifdef CONFIG_X86_64
 315         .read_cr8 = native_read_cr8,
 316         .write_cr8 = native_write_cr8,
 317 #endif
 318         .wbinvd = native_wbinvd,
 319         .read_msr = native_read_msr_safe,
 320         .write_msr = native_write_msr_safe,
 321         .read_tsc = native_read_tsc,
 322         .read_pmc = native_read_pmc,
 323         .read_tscp = native_read_tscp,
 324         .load_tr_desc = native_load_tr_desc,
 325         .set_ldt = native_set_ldt,
 326         .load_gdt = native_load_gdt,
 327         .load_idt = native_load_idt,
 328         .store_gdt = native_store_gdt,
 329         .store_idt = native_store_idt,
 330         .store_tr = native_store_tr,
 331         .load_tls = native_load_tls,
 332 #ifdef CONFIG_X86_64
 333         .load_gs_index = native_load_gs_index,
 334 #endif
 335         .write_ldt_entry = native_write_ldt_entry,
 336         .write_gdt_entry = native_write_gdt_entry,
 337         .write_idt_entry = native_write_idt_entry,
 338         .load_sp0 = native_load_sp0,
 339
 340 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 341         .irq_enable_sysexit = native_irq_enable_sysexit,
 342 #endif
 343 #ifdef CONFIG_X86_64
 344 #ifdef CONFIG_IA32_EMULATION
 345         .usergs_sysret32 = native_usergs_sysret32,
 346 #endif
 347         .usergs_sysret64 = native_usergs_sysret64,
 348 #endif
 349         .iret = native_iret,
 350         .swapgs = native_swapgs,
 351
 352         .set_iopl_mask = native_set_iopl_mask,
 353         .io_delay = native_io_delay,
 354
 355         .lazy_mode = {
 356                 .enter = paravirt_nop,
 357                 .leave = paravirt_nop,
 358         },
 359 };
 360
 361 struct pv_apic_ops pv_apic_ops = {
 362 #ifdef CONFIG_X86_LOCAL_APIC
 363         .apic_write = native_apic_write,
 364         .apic_write_atomic = native_apic_write_atomic,
 365         .apic_read = native_apic_read,
 366         .setup_boot_clock = setup_boot_APIC_clock,
 367         .setup_secondary_clock = setup_secondary_APIC_clock,
 368         .startup_ipi_hook = paravirt_nop,
 369 #endif
 370 };
 371
 372 struct pv_mmu_ops pv_mmu_ops = {
 373 #ifndef CONFIG_X86_64
 374         .pagetable_setup_start = native_pagetable_setup_start,
 375         .pagetable_setup_done = native_pagetable_setup_done,
 376 #endif
 377
 378         .read_cr2 = native_read_cr2,
 379         .write_cr2 = native_write_cr2,
 380         .read_cr3 = native_read_cr3,
 381         .write_cr3 = native_write_cr3,
 382
 383         .flush_tlb_user = native_flush_tlb,
 384         .flush_tlb_kernel = native_flush_tlb_global,
 385         .flush_tlb_single = native_flush_tlb_single,
 386         .flush_tlb_others = native_flush_tlb_others,
 387
 388         .pgd_alloc = __paravirt_pgd_alloc,
 389         .pgd_free = paravirt_nop,
 390
 391         .alloc_pte = paravirt_nop,
 392         .alloc_pmd = paravirt_nop,
 393         .alloc_pmd_clone = paravirt_nop,
 394         .alloc_pud = paravirt_nop,
 395         .release_pte = paravirt_nop,
 396         .release_pmd = paravirt_nop,
 397         .release_pud = paravirt_nop,
 398
 399         .set_pte = native_set_pte,
 400         .set_pte_at = native_set_pte_at,
 401         .set_pmd = native_set_pmd,
 402         .pte_update = paravirt_nop,
 403         .pte_update_defer = paravirt_nop,
 404
 405         .ptep_modify_prot_start = __ptep_modify_prot_start,
 406         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 407
 408 #ifdef CONFIG_HIGHPTE
 409         .kmap_atomic_pte = kmap_atomic,
 410 #endif
 411
 412 #if PAGETABLE_LEVELS >= 3
 413 #ifdef CONFIG_X86_PAE
 414         .set_pte_atomic = native_set_pte_atomic,
 415         .set_pte_present = native_set_pte_present,
 416         .pte_clear = native_pte_clear,
 417         .pmd_clear = native_pmd_clear,
 418 #endif
 419         .set_pud = native_set_pud,
 420         .pmd_val = native_pmd_val,
 421         .make_pmd = native_make_pmd,
 422
 423 #if PAGETABLE_LEVELS == 4
 424         .pud_val = native_pud_val,
 425         .make_pud = native_make_pud,
 426         .set_pgd = native_set_pgd,
 427 #endif
 428 #endif /* PAGETABLE_LEVELS >= 3 */
 429
 430         .pte_val = native_pte_val,
 431         .pte_flags = native_pte_val,
 432         .pgd_val = native_pgd_val,
 433
 434         .make_pte = native_make_pte,
 435         .make_pgd = native_make_pgd,
 436
 437         .dup_mmap = paravirt_nop,
 438         .exit_mmap = paravirt_nop,
 439         .activate_mm = paravirt_nop,
 440
 441         .lazy_mode = {
 442                 .enter = paravirt_nop,
 443                 .leave = paravirt_nop,
 444         },
 445
 446         .set_fixmap = native_set_fixmap,
 447 };
 448
 449 EXPORT_SYMBOL_GPL(pv_time_ops);
 450 EXPORT_SYMBOL    (pv_cpu_ops);
 451 EXPORT_SYMBOL    (pv_mmu_ops);
 452 EXPORT_SYMBOL_GPL(pv_apic_ops);
 453 EXPORT_SYMBOL_GPL(pv_info);
 454 EXPORT_SYMBOL    (pv_irq_ops);