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.usersp_sysret))
 145                 /* If operation requires a jmp, then jmp */
 146                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
 147         else
 148                 /* Otherwise call the function; assume target could
 149                    clobber any caller-save reg */
 150                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
 151                                           addr, clobbers, len);
 152
 153         return ret;
 154 }
 155
 156 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 157                               const char *start, const char *end)
 158 {
 159         unsigned insn_len = end - start;
 160
 161         if (insn_len > len || start == NULL)
 162                 insn_len = len;
 163         else
 164                 memcpy(insnbuf, start, insn_len);
 165
 166         return insn_len;
 167 }
 168
 169 void init_IRQ(void)
 170 {
 171         pv_irq_ops.init_IRQ();
 172 }
 173
 174 static void native_flush_tlb(void)
 175 {
 176         __native_flush_tlb();
 177 }
 178
 179 /*
 180  * Global pages have to be flushed a bit differently. Not a real
 181  * performance problem because this does not happen often.
 182  */
 183 static void native_flush_tlb_global(void)
 184 {
 185         __native_flush_tlb_global();
 186 }
 187
 188 static void native_flush_tlb_single(unsigned long addr)
 189 {
 190         __native_flush_tlb_single(addr);
 191 }
 192
 193 /* These are in entry.S */
 194 extern void native_iret(void);
 195 extern void native_irq_enable_sysexit(void);
 196 extern void native_usersp_sysret(void);
 197
 198 static int __init print_banner(void)
 199 {
 200         pv_init_ops.banner();
 201         return 0;
 202 }
 203 core_initcall(print_banner);
 204
 205 static struct resource reserve_ioports = {
 206         .start = 0,
 207         .end = IO_SPACE_LIMIT,
 208         .name = "paravirt-ioport",
 209         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
 210 };
 211
 212 /*
 213  * Reserve the whole legacy IO space to prevent any legacy drivers
 214  * from wasting time probing for their hardware.  This is a fairly
 215  * brute-force approach to disabling all non-virtual drivers.
 216  *
 217  * Note that this must be called very early to have any effect.
 218  */
 219 int paravirt_disable_iospace(void)
 220 {
 221         return request_resource(&ioport_resource, &reserve_ioports);
 222 }
 223
 224 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 225
 226 static inline void enter_lazy(enum paravirt_lazy_mode mode)
 227 {
 228         BUG_ON(__get_cpu_var(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
 229         BUG_ON(preemptible());
 230
 231         __get_cpu_var(paravirt_lazy_mode) = mode;
 232 }
 233
 234 void paravirt_leave_lazy(enum paravirt_lazy_mode mode)
 235 {
 236         BUG_ON(__get_cpu_var(paravirt_lazy_mode) != mode);
 237         BUG_ON(preemptible());
 238
 239         __get_cpu_var(paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
 240 }
 241
 242 void paravirt_enter_lazy_mmu(void)
 243 {
 244         enter_lazy(PARAVIRT_LAZY_MMU);
 245 }
 246
 247 void paravirt_leave_lazy_mmu(void)
 248 {
 249         paravirt_leave_lazy(PARAVIRT_LAZY_MMU);
 250 }
 251
 252 void paravirt_enter_lazy_cpu(void)
 253 {
 254         enter_lazy(PARAVIRT_LAZY_CPU);
 255 }
 256
 257 void paravirt_leave_lazy_cpu(void)
 258 {
 259         paravirt_leave_lazy(PARAVIRT_LAZY_CPU);
 260 }
 261
 262 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
 263 {
 264         return __get_cpu_var(paravirt_lazy_mode);
 265 }
 266
 267 struct pv_info pv_info = {
 268         .name = "bare hardware",
 269         .paravirt_enabled = 0,
 270         .kernel_rpl = 0,
 271         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
 272 };
 273
 274 struct pv_init_ops pv_init_ops = {
 275         .patch = native_patch,
 276         .banner = default_banner,
 277         .arch_setup = paravirt_nop,
 278         .memory_setup = machine_specific_memory_setup,
 279 };
 280
 281 struct pv_time_ops pv_time_ops = {
 282         .time_init = hpet_time_init,
 283         .get_wallclock = native_get_wallclock,
 284         .set_wallclock = native_set_wallclock,
 285         .sched_clock = native_sched_clock,
 286         .get_cpu_khz = native_calculate_cpu_khz,
 287 };
 288
 289 struct pv_irq_ops pv_irq_ops = {
 290         .init_IRQ = native_init_IRQ,
 291         .save_fl = native_save_fl,
 292         .restore_fl = native_restore_fl,
 293         .irq_disable = native_irq_disable,
 294         .irq_enable = native_irq_enable,
 295         .safe_halt = native_safe_halt,
 296         .halt = native_halt,
 297 };
 298
 299 struct pv_cpu_ops pv_cpu_ops = {
 300         .cpuid = native_cpuid,
 301         .get_debugreg = native_get_debugreg,
 302         .set_debugreg = native_set_debugreg,
 303         .clts = native_clts,
 304         .read_cr0 = native_read_cr0,
 305         .write_cr0 = native_write_cr0,
 306         .read_cr4 = native_read_cr4,
 307         .read_cr4_safe = native_read_cr4_safe,
 308         .write_cr4 = native_write_cr4,
 309 #ifdef CONFIG_X86_64
 310         .read_cr8 = native_read_cr8,
 311         .write_cr8 = native_write_cr8,
 312 #endif
 313         .wbinvd = native_wbinvd,
 314         .read_msr = native_read_msr_safe,
 315         .write_msr = native_write_msr_safe,
 316         .read_tsc = native_read_tsc,
 317         .read_pmc = native_read_pmc,
 318         .read_tscp = native_read_tscp,
 319         .load_tr_desc = native_load_tr_desc,
 320         .set_ldt = native_set_ldt,
 321         .load_gdt = native_load_gdt,
 322         .load_idt = native_load_idt,
 323         .store_gdt = native_store_gdt,
 324         .store_idt = native_store_idt,
 325         .store_tr = native_store_tr,
 326         .load_tls = native_load_tls,
 327         .write_ldt_entry = native_write_ldt_entry,
 328         .write_gdt_entry = native_write_gdt_entry,
 329         .write_idt_entry = native_write_idt_entry,
 330         .load_sp0 = native_load_sp0,
 331
 332 #ifdef CONFIG_X86_32
 333         .irq_enable_sysexit = native_irq_enable_sysexit,
 334 #else
 335         .usersp_sysret = native_usersp_sysret,
 336 #endif
 337         .iret = native_iret,
 338         .swapgs = native_swapgs,
 339
 340         .set_iopl_mask = native_set_iopl_mask,
 341         .io_delay = native_io_delay,
 342
 343         .lazy_mode = {
 344                 .enter = paravirt_nop,
 345                 .leave = paravirt_nop,
 346         },
 347 };
 348
 349 struct pv_apic_ops pv_apic_ops = {
 350 #ifdef CONFIG_X86_LOCAL_APIC
 351         .apic_write = native_apic_write,
 352         .apic_write_atomic = native_apic_write_atomic,
 353         .apic_read = native_apic_read,
 354         .setup_boot_clock = setup_boot_APIC_clock,
 355         .setup_secondary_clock = setup_secondary_APIC_clock,
 356         .startup_ipi_hook = paravirt_nop,
 357 #endif
 358 };
 359
 360 struct pv_mmu_ops pv_mmu_ops = {
 361 #ifndef CONFIG_X86_64
 362         .pagetable_setup_start = native_pagetable_setup_start,
 363         .pagetable_setup_done = native_pagetable_setup_done,
 364 #endif
 365
 366         .read_cr2 = native_read_cr2,
 367         .write_cr2 = native_write_cr2,
 368         .read_cr3 = native_read_cr3,
 369         .write_cr3 = native_write_cr3,
 370
 371         .flush_tlb_user = native_flush_tlb,
 372         .flush_tlb_kernel = native_flush_tlb_global,
 373         .flush_tlb_single = native_flush_tlb_single,
 374         .flush_tlb_others = native_flush_tlb_others,
 375
 376         .pgd_alloc = __paravirt_pgd_alloc,
 377         .pgd_free = paravirt_nop,
 378
 379         .alloc_pte = paravirt_nop,
 380         .alloc_pmd = paravirt_nop,
 381         .alloc_pmd_clone = paravirt_nop,
 382         .alloc_pud = paravirt_nop,
 383         .release_pte = paravirt_nop,
 384         .release_pmd = paravirt_nop,
 385         .release_pud = paravirt_nop,
 386
 387         .set_pte = native_set_pte,
 388         .set_pte_at = native_set_pte_at,
 389         .set_pmd = native_set_pmd,
 390         .pte_update = paravirt_nop,
 391         .pte_update_defer = paravirt_nop,
 392
 393         .ptep_modify_prot_start = __ptep_modify_prot_start,
 394         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
 395
 396 #ifdef CONFIG_HIGHPTE
 397         .kmap_atomic_pte = kmap_atomic,
 398 #endif
 399
 400 #if PAGETABLE_LEVELS >= 3
 401 #ifdef CONFIG_X86_PAE
 402         .set_pte_atomic = native_set_pte_atomic,
 403         .set_pte_present = native_set_pte_present,
 404         .pte_clear = native_pte_clear,
 405         .pmd_clear = native_pmd_clear,
 406 #endif
 407         .set_pud = native_set_pud,
 408         .pmd_val = native_pmd_val,
 409         .make_pmd = native_make_pmd,
 410
 411 #if PAGETABLE_LEVELS == 4
 412         .pud_val = native_pud_val,
 413         .make_pud = native_make_pud,
 414         .set_pgd = native_set_pgd,
 415 #endif
 416 #endif /* PAGETABLE_LEVELS >= 3 */
 417
 418         .pte_val = native_pte_val,
 419         .pte_flags = native_pte_val,
 420         .pgd_val = native_pgd_val,
 421
 422         .make_pte = native_make_pte,
 423         .make_pgd = native_make_pgd,
 424
 425         .dup_mmap = paravirt_nop,
 426         .exit_mmap = paravirt_nop,
 427         .activate_mm = paravirt_nop,
 428
 429         .lazy_mode = {
 430                 .enter = paravirt_nop,
 431                 .leave = paravirt_nop,
 432         },
 433
 434         .set_fixmap = native_set_fixmap,
 435 };
 436
 437 EXPORT_SYMBOL_GPL(pv_time_ops);
 438 EXPORT_SYMBOL    (pv_cpu_ops);
 439 EXPORT_SYMBOL    (pv_mmu_ops);
 440 EXPORT_SYMBOL_GPL(pv_apic_ops);
 441 EXPORT_SYMBOL_GPL(pv_info);
 442 EXPORT_SYMBOL    (pv_irq_ops);