linux/arch/x86/kernel/paravirt.c
<<
>>
Prefs
   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#include <linux/kprobes.h>
  27
  28#include <asm/bug.h>
  29#include <asm/paravirt.h>
  30#include <asm/debugreg.h>
  31#include <asm/desc.h>
  32#include <asm/setup.h>
  33#include <asm/pgtable.h>
  34#include <asm/time.h>
  35#include <asm/pgalloc.h>
  36#include <asm/irq.h>
  37#include <asm/delay.h>
  38#include <asm/fixmap.h>
  39#include <asm/apic.h>
  40#include <asm/tlbflush.h>
  41#include <asm/timer.h>
  42#include <asm/special_insns.h>
  43
  44/*
  45 * nop stub, which must not clobber anything *including the stack* to
  46 * avoid confusing the entry prologues.
  47 */
  48extern void _paravirt_nop(void);
  49asm (".pushsection .entry.text, \"ax\"\n"
  50     ".global _paravirt_nop\n"
  51     "_paravirt_nop:\n\t"
  52     "ret\n\t"
  53     ".size _paravirt_nop, . - _paravirt_nop\n\t"
  54     ".type _paravirt_nop, @function\n\t"
  55     ".popsection");
  56
  57/* identity function, which can be inlined */
  58u32 _paravirt_ident_32(u32 x)
  59{
  60        return x;
  61}
  62
  63u64 _paravirt_ident_64(u64 x)
  64{
  65        return x;
  66}
  67
  68void __init default_banner(void)
  69{
  70        printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  71               pv_info.name);
  72}
  73
  74/* Undefined instruction for dealing with missing ops pointers. */
  75static const unsigned char ud2a[] = { 0x0f, 0x0b };
  76
  77unsigned paravirt_patch_nop(void)
  78{
  79        return 0;
  80}
  81
  82unsigned paravirt_patch_ignore(unsigned len)
  83{
  84        return len;
  85}
  86
  87struct branch {
  88        unsigned char opcode;
  89        u32 delta;
  90} __attribute__((packed));
  91
  92unsigned paravirt_patch_call(void *insnbuf,
  93                             const void *target, u16 tgt_clobbers,
  94                             unsigned long addr, u16 site_clobbers,
  95                             unsigned len)
  96{
  97        struct branch *b = insnbuf;
  98        unsigned long delta = (unsigned long)target - (addr+5);
  99
 100        if (tgt_clobbers & ~site_clobbers)
 101                return len;     /* target would clobber too much for this site */
 102        if (len < 5)
 103                return len;     /* call too long for patch site */
 104
 105        b->opcode = 0xe8; /* call */
 106        b->delta = delta;
 107        BUILD_BUG_ON(sizeof(*b) != 5);
 108
 109        return 5;
 110}
 111
 112unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 113                            unsigned long addr, unsigned len)
 114{
 115        struct branch *b = insnbuf;
 116        unsigned long delta = (unsigned long)target - (addr+5);
 117
 118        if (len < 5)
 119                return len;     /* call too long for patch site */
 120
 121        b->opcode = 0xe9;       /* jmp */
 122        b->delta = delta;
 123
 124        return 5;
 125}
 126
 127/* Neat trick to map patch type back to the call within the
 128 * corresponding structure. */
 129static void *get_call_destination(u8 type)
 130{
 131        struct paravirt_patch_template tmpl = {
 132                .pv_init_ops = pv_init_ops,
 133                .pv_time_ops = pv_time_ops,
 134                .pv_cpu_ops = pv_cpu_ops,
 135                .pv_irq_ops = pv_irq_ops,
 136                .pv_apic_ops = pv_apic_ops,
 137                .pv_mmu_ops = pv_mmu_ops,
 138#ifdef CONFIG_PARAVIRT_SPINLOCKS
 139                .pv_lock_ops = pv_lock_ops,
 140#endif
 141        };
 142        return *((void **)&tmpl + type);
 143}
 144
 145unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
 146                                unsigned long addr, unsigned len)
 147{
 148        void *opfunc = get_call_destination(type);
 149        unsigned ret;
 150
 151        if (opfunc == NULL)
 152                /* If there's no function, patch it with a ud2a (BUG) */
 153                ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
 154        else if (opfunc == _paravirt_nop)
 155                /* If the operation is a nop, then nop the callsite */
 156                ret = paravirt_patch_nop();
 157
 158        /* identity functions just return their single argument */
 159        else if (opfunc == _paravirt_ident_32)
 160                ret = paravirt_patch_ident_32(insnbuf, len);
 161        else if (opfunc == _paravirt_ident_64)
 162                ret = paravirt_patch_ident_64(insnbuf, len);
 163
 164        else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
 165#ifdef CONFIG_X86_32
 166                 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
 167#endif
 168                 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
 169                 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
 170                /* If operation requires a jmp, then jmp */
 171                ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 172        else
 173                /* Otherwise call the function; assume target could
 174                   clobber any caller-save reg */
 175                ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
 176                                          addr, clobbers, len);
 177
 178        return ret;
 179}
 180
 181unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 182                              const char *start, const char *end)
 183{
 184        unsigned insn_len = end - start;
 185
 186        if (insn_len > len || start == NULL)
 187                insn_len = len;
 188        else
 189                memcpy(insnbuf, start, insn_len);
 190
 191        return insn_len;
 192}
 193
 194static void native_flush_tlb(void)
 195{
 196        __native_flush_tlb();
 197}
 198
 199/*
 200 * Global pages have to be flushed a bit differently. Not a real
 201 * performance problem because this does not happen often.
 202 */
 203static void native_flush_tlb_global(void)
 204{
 205        __native_flush_tlb_global();
 206}
 207
 208static void native_flush_tlb_single(unsigned long addr)
 209{
 210        __native_flush_tlb_single(addr);
 211}
 212
 213struct static_key paravirt_steal_enabled;
 214struct static_key paravirt_steal_rq_enabled;
 215
 216static u64 native_steal_clock(int cpu)
 217{
 218        return 0;
 219}
 220
 221/* These are in entry.S */
 222extern void native_iret(void);
 223extern void native_irq_enable_sysexit(void);
 224extern void native_usergs_sysret32(void);
 225extern void native_usergs_sysret64(void);
 226
 227static struct resource reserve_ioports = {
 228        .start = 0,
 229        .end = IO_SPACE_LIMIT,
 230        .name = "paravirt-ioport",
 231        .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 232};
 233
 234/*
 235 * Reserve the whole legacy IO space to prevent any legacy drivers
 236 * from wasting time probing for their hardware.  This is a fairly
 237 * brute-force approach to disabling all non-virtual drivers.
 238 *
 239 * Note that this must be called very early to have any effect.
 240 */
 241int paravirt_disable_iospace(void)
 242{
 243        return request_resource(&ioport_resource, &reserve_ioports);
 244}
 245
 246static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 247
 248static inline void enter_lazy(enum paravirt_lazy_mode mode)
 249{
 250        BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 251
 252        this_cpu_write(paravirt_lazy_mode, mode);
 253}
 254
 255static void leave_lazy(enum paravirt_lazy_mode mode)
 256{
 257        BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
 258
 259        this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
 260}
 261
 262void paravirt_enter_lazy_mmu(void)
 263{
 264        enter_lazy(PARAVIRT_LAZY_MMU);
 265}
 266
 267void paravirt_leave_lazy_mmu(void)
 268{
 269        leave_lazy(PARAVIRT_LAZY_MMU);
 270}
 271
 272void paravirt_flush_lazy_mmu(void)
 273{
 274        preempt_disable();
 275
 276        if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
 277                arch_leave_lazy_mmu_mode();
 278                arch_enter_lazy_mmu_mode();
 279        }
 280
 281        preempt_enable();
 282}
 283
 284void paravirt_start_context_switch(struct task_struct *prev)
 285{
 286        BUG_ON(preemptible());
 287
 288        if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
 289                arch_leave_lazy_mmu_mode();
 290                set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
 291        }
 292        enter_lazy(PARAVIRT_LAZY_CPU);
 293}
 294
 295void paravirt_end_context_switch(struct task_struct *next)
 296{
 297        BUG_ON(preemptible());
 298
 299        leave_lazy(PARAVIRT_LAZY_CPU);
 300
 301        if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
 302                arch_enter_lazy_mmu_mode();
 303}
 304
 305enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 306{
 307        if (in_interrupt())
 308                return PARAVIRT_LAZY_NONE;
 309
 310        return this_cpu_read(paravirt_lazy_mode);
 311}
 312
 313struct pv_info pv_info = {
 314        .name = "bare hardware",
 315        .paravirt_enabled = 0,
 316        .kernel_rpl = 0,
 317        .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 318
 319#ifdef CONFIG_X86_64
 320        .extra_user_64bit_cs = __USER_CS,
 321#endif
 322};
 323
 324struct pv_init_ops pv_init_ops = {
 325        .patch = native_patch,
 326};
 327
 328struct pv_time_ops pv_time_ops = {
 329        .sched_clock = native_sched_clock,
 330        .steal_clock = native_steal_clock,
 331};
 332
 333__visible struct pv_irq_ops pv_irq_ops = {
 334        .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
 335        .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
 336        .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
 337        .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
 338        .safe_halt = native_safe_halt,
 339        .halt = native_halt,
 340#ifdef CONFIG_X86_64
 341        .adjust_exception_frame = paravirt_nop,
 342#endif
 343};
 344
 345__visible struct pv_cpu_ops pv_cpu_ops = {
 346        .cpuid = native_cpuid,
 347        .get_debugreg = native_get_debugreg,
 348        .set_debugreg = native_set_debugreg,
 349        .clts = native_clts,
 350        .read_cr0 = native_read_cr0,
 351        .write_cr0 = native_write_cr0,
 352        .read_cr4 = native_read_cr4,
 353        .read_cr4_safe = native_read_cr4_safe,
 354        .write_cr4 = native_write_cr4,
 355#ifdef CONFIG_X86_64
 356        .read_cr8 = native_read_cr8,
 357        .write_cr8 = native_write_cr8,
 358#endif
 359        .wbinvd = native_wbinvd,
 360        .read_msr = native_read_msr_safe,
 361        .write_msr = native_write_msr_safe,
 362        .read_pmc = native_read_pmc,
 363        .load_tr_desc = native_load_tr_desc,
 364        .set_ldt = native_set_ldt,
 365        .load_gdt = native_load_gdt,
 366        .load_idt = native_load_idt,
 367        .store_idt = native_store_idt,
 368        .store_tr = native_store_tr,
 369        .load_tls = native_load_tls,
 370#ifdef CONFIG_X86_64
 371        .load_gs_index = native_load_gs_index,
 372#endif
 373        .write_ldt_entry = native_write_ldt_entry,
 374        .write_gdt_entry = native_write_gdt_entry,
 375        .write_idt_entry = native_write_idt_entry,
 376
 377        .alloc_ldt = paravirt_nop,
 378        .free_ldt = paravirt_nop,
 379
 380        .load_sp0 = native_load_sp0,
 381
 382#if defined(CONFIG_X86_32)
 383        .irq_enable_sysexit = native_irq_enable_sysexit,
 384#endif
 385#ifdef CONFIG_X86_64
 386#ifdef CONFIG_IA32_EMULATION
 387        .usergs_sysret32 = native_usergs_sysret32,
 388#endif
 389        .usergs_sysret64 = native_usergs_sysret64,
 390#endif
 391        .iret = native_iret,
 392        .swapgs = native_swapgs,
 393
 394        .set_iopl_mask = native_set_iopl_mask,
 395        .io_delay = native_io_delay,
 396
 397        .start_context_switch = paravirt_nop,
 398        .end_context_switch = paravirt_nop,
 399};
 400
 401/* At this point, native_get/set_debugreg has real function entries */
 402NOKPROBE_SYMBOL(native_get_debugreg);
 403NOKPROBE_SYMBOL(native_set_debugreg);
 404NOKPROBE_SYMBOL(native_load_idt);
 405
 406struct pv_apic_ops pv_apic_ops = {
 407#ifdef CONFIG_X86_LOCAL_APIC
 408        .startup_ipi_hook = paravirt_nop,
 409#endif
 410};
 411
 412#if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
 413/* 32-bit pagetable entries */
 414#define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
 415#else
 416/* 64-bit pagetable entries */
 417#define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
 418#endif
 419
 420struct pv_mmu_ops pv_mmu_ops = {
 421
 422        .read_cr2 = native_read_cr2,
 423        .write_cr2 = native_write_cr2,
 424        .read_cr3 = native_read_cr3,
 425        .write_cr3 = native_write_cr3,
 426
 427        .flush_tlb_user = native_flush_tlb,
 428        .flush_tlb_kernel = native_flush_tlb_global,
 429        .flush_tlb_single = native_flush_tlb_single,
 430        .flush_tlb_others = native_flush_tlb_others,
 431
 432        .pgd_alloc = __paravirt_pgd_alloc,
 433        .pgd_free = paravirt_nop,
 434
 435        .alloc_pte = paravirt_nop,
 436        .alloc_pmd = paravirt_nop,
 437        .alloc_pud = paravirt_nop,
 438        .release_pte = paravirt_nop,
 439        .release_pmd = paravirt_nop,
 440        .release_pud = paravirt_nop,
 441
 442        .set_pte = native_set_pte,
 443        .set_pte_at = native_set_pte_at,
 444        .set_pmd = native_set_pmd,
 445        .set_pmd_at = native_set_pmd_at,
 446        .pte_update = paravirt_nop,
 447        .pte_update_defer = paravirt_nop,
 448        .pmd_update = paravirt_nop,
 449        .pmd_update_defer = paravirt_nop,
 450
 451        .ptep_modify_prot_start = __ptep_modify_prot_start,
 452        .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 453
 454#if CONFIG_PGTABLE_LEVELS >= 3
 455#ifdef CONFIG_X86_PAE
 456        .set_pte_atomic = native_set_pte_atomic,
 457        .pte_clear = native_pte_clear,
 458        .pmd_clear = native_pmd_clear,
 459#endif
 460        .set_pud = native_set_pud,
 461
 462        .pmd_val = PTE_IDENT,
 463        .make_pmd = PTE_IDENT,
 464
 465#if CONFIG_PGTABLE_LEVELS == 4
 466        .pud_val = PTE_IDENT,
 467        .make_pud = PTE_IDENT,
 468
 469        .set_pgd = native_set_pgd,
 470#endif
 471#endif /* CONFIG_PGTABLE_LEVELS >= 3 */
 472
 473        .pte_val = PTE_IDENT,
 474        .pgd_val = PTE_IDENT,
 475
 476        .make_pte = PTE_IDENT,
 477        .make_pgd = PTE_IDENT,
 478
 479        .dup_mmap = paravirt_nop,
 480        .exit_mmap = paravirt_nop,
 481        .activate_mm = paravirt_nop,
 482
 483        .lazy_mode = {
 484                .enter = paravirt_nop,
 485                .leave = paravirt_nop,
 486                .flush = paravirt_nop,
 487        },
 488
 489        .set_fixmap = native_set_fixmap,
 490};
 491
 492EXPORT_SYMBOL_GPL(pv_time_ops);
 493EXPORT_SYMBOL    (pv_cpu_ops);
 494EXPORT_SYMBOL    (pv_mmu_ops);
 495EXPORT_SYMBOL_GPL(pv_apic_ops);
 496EXPORT_SYMBOL_GPL(pv_info);
 497EXPORT_SYMBOL    (pv_irq_ops);
 498