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