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/pgtable.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 /* identity function, which can be inlined */ 47 u32 _paravirt_ident_32(u32 x) 48 { 49 return x; 50 } 51 52 u64 _paravirt_ident_64(u64 x) 53 { 54 return x; 55 } 56 57 static void __init default_banner(void) 58 { 59 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 60 pv_info.name); 61 } 62 63 char *memory_setup(void) 64 { 65 return pv_init_ops.memory_setup(); 66 } 67 68 /* Simple instruction patching code. */ 69 #define DEF_NATIVE(ops, name, code) \ 70 extern const char start_##ops##_##name[], end_##ops##_##name[]; \ 71 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":") 72 73 /* Undefined instruction for dealing with missing ops pointers. */ 74 static const unsigned char ud2a[] = { 0x0f, 0x0b }; 75 76 unsigned paravirt_patch_nop(void) 77 { 78 return 0; 79 } 80 81 unsigned paravirt_patch_ignore(unsigned len) 82 { 83 return len; 84 } 85 86 struct branch { 87 unsigned char opcode; 88 u32 delta; 89 } __attribute__((packed)); 90 91 unsigned paravirt_patch_call(void *insnbuf, 92 const void *target, u16 tgt_clobbers, 93 unsigned long addr, u16 site_clobbers, 94 unsigned len) 95 { 96 struct branch *b = insnbuf; 97 unsigned long delta = (unsigned long)target - (addr+5); 98 99 if (tgt_clobbers & ~site_clobbers) 100 return len; /* target would clobber too much for this site */ 101 if (len < 5) 102 return len; /* call too long for patch site */ 103 104 b->opcode = 0xe8; /* call */ 105 b->delta = delta; 106 BUILD_BUG_ON(sizeof(*b) != 5); 107 108 return 5; 109 } 110 111 unsigned paravirt_patch_jmp(void *insnbuf, const void *target, 112 unsigned long addr, unsigned len) 113 { 114 struct branch *b = insnbuf; 115 unsigned long delta = (unsigned long)target - (addr+5); 116 117 if (len < 5) 118 return len; /* call too long for patch site */ 119 120 b->opcode = 0xe9; /* jmp */ 121 b->delta = delta; 122 123 return 5; 124 } 125 126 /* Neat trick to map patch type back to the call within the 127 * corresponding structure. */ 128 static void *get_call_destination(u8 type) 129 { 130 struct paravirt_patch_template tmpl = { 131 .pv_init_ops = pv_init_ops, 132 .pv_time_ops = pv_time_ops, 133 .pv_cpu_ops = pv_cpu_ops, 134 .pv_irq_ops = pv_irq_ops, 135 .pv_apic_ops = pv_apic_ops, 136 .pv_mmu_ops = pv_mmu_ops, 137 .pv_lock_ops = pv_lock_ops, 138 }; 139 return *((void **)&tmpl + type); 140 } 141 142 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf, 143 unsigned long addr, unsigned len) 144 { 145 void *opfunc = get_call_destination(type); 146 unsigned ret; 147 148 if (opfunc == NULL) 149 /* If there's no function, patch it with a ud2a (BUG) */ 150 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a)); 151 else if (opfunc == _paravirt_nop) 152 /* If the operation is a nop, then nop the callsite */ 153 ret = paravirt_patch_nop(); 154 155 /* identity functions just return their single argument */ 156 else if (opfunc == _paravirt_ident_32) 157 ret = paravirt_patch_ident_32(insnbuf, len); 158 else if (opfunc == _paravirt_ident_64) 159 ret = paravirt_patch_ident_64(insnbuf, len); 160 161 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) || 162 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) || 163 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) || 164 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64)) 165 /* If operation requires a jmp, then jmp */ 166 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len); 167 else 168 /* Otherwise call the function; assume target could 169 clobber any caller-save reg */ 170 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY, 171 addr, clobbers, len); 172 173 return ret; 174 } 175 176 unsigned paravirt_patch_insns(void *insnbuf, unsigned len, 177 const char *start, const char *end) 178 { 179 unsigned insn_len = end - start; 180 181 if (insn_len > len || start == NULL) 182 insn_len = len; 183 else 184 memcpy(insnbuf, start, insn_len); 185 186 return insn_len; 187 } 188 189 void init_IRQ(void) 190 { 191 pv_irq_ops.init_IRQ(); 192 } 193 194 static 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 */ 203 static void native_flush_tlb_global(void) 204 { 205 __native_flush_tlb_global(); 206 } 207 208 static void native_flush_tlb_single(unsigned long addr) 209 { 210 __native_flush_tlb_single(addr); 211 } 212 213 /* These are in entry.S */ 214 extern void native_iret(void); 215 extern void native_irq_enable_sysexit(void); 216 extern void native_usergs_sysret32(void); 217 extern void native_usergs_sysret64(void); 218 219 static int __init print_banner(void) 220 { 221 pv_init_ops.banner(); 222 return 0; 223 } 224 core_initcall(print_banner); 225 226 static struct resource reserve_ioports = { 227 .start = 0, 228 .end = IO_SPACE_LIMIT, 229 .name = "paravirt-ioport", 230 .flags = IORESOURCE_IO | IORESOURCE_BUSY, 231 }; 232 233 /* 234 * Reserve the whole legacy IO space to prevent any legacy drivers 235 * from wasting time probing for their hardware. This is a fairly 236 * brute-force approach to disabling all non-virtual drivers. 237 * 238 * Note that this must be called very early to have any effect. 239 */ 240 int paravirt_disable_iospace(void) 241 { 242 return request_resource(&ioport_resource, &reserve_ioports); 243 } 244 245 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE; 246 247 static inline void enter_lazy(enum paravirt_lazy_mode mode) 248 { 249 BUG_ON(__get_cpu_var(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE); 250 BUG_ON(preemptible()); 251 252 __get_cpu_var(paravirt_lazy_mode) = mode; 253 } 254 255 void paravirt_leave_lazy(enum paravirt_lazy_mode mode) 256 { 257 BUG_ON(__get_cpu_var(paravirt_lazy_mode) != mode); 258 BUG_ON(preemptible()); 259 260 __get_cpu_var(paravirt_lazy_mode) = PARAVIRT_LAZY_NONE; 261 } 262 263 void paravirt_enter_lazy_mmu(void) 264 { 265 enter_lazy(PARAVIRT_LAZY_MMU); 266 } 267 268 void paravirt_leave_lazy_mmu(void) 269 { 270 paravirt_leave_lazy(PARAVIRT_LAZY_MMU); 271 } 272 273 void paravirt_enter_lazy_cpu(void) 274 { 275 enter_lazy(PARAVIRT_LAZY_CPU); 276 } 277 278 void paravirt_leave_lazy_cpu(void) 279 { 280 paravirt_leave_lazy(PARAVIRT_LAZY_CPU); 281 } 282 283 enum paravirt_lazy_mode paravirt_get_lazy_mode(void) 284 { 285 return __get_cpu_var(paravirt_lazy_mode); 286 } 287 288 void arch_flush_lazy_mmu_mode(void) 289 { 290 preempt_disable(); 291 292 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { 293 WARN_ON(preempt_count() == 1); 294 arch_leave_lazy_mmu_mode(); 295 arch_enter_lazy_mmu_mode(); 296 } 297 298 preempt_enable(); 299 } 300 301 void arch_flush_lazy_cpu_mode(void) 302 { 303 preempt_disable(); 304 305 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { 306 WARN_ON(preempt_count() == 1); 307 arch_leave_lazy_cpu_mode(); 308 arch_enter_lazy_cpu_mode(); 309 } 310 311 preempt_enable(); 312 } 313 314 struct pv_info pv_info = { 315 .name = "bare hardware", 316 .paravirt_enabled = 0, 317 .kernel_rpl = 0, 318 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */ 319 }; 320 321 struct pv_init_ops pv_init_ops = { 322 .patch = native_patch, 323 .banner = default_banner, 324 .arch_setup = paravirt_nop, 325 .memory_setup = machine_specific_memory_setup, 326 }; 327 328 struct pv_time_ops pv_time_ops = { 329 .time_init = hpet_time_init, 330 .get_wallclock = native_get_wallclock, 331 .set_wallclock = native_set_wallclock, 332 .sched_clock = native_sched_clock, 333 .get_tsc_khz = native_calibrate_tsc, 334 }; 335 336 struct pv_irq_ops pv_irq_ops = { 337 .init_IRQ = native_init_IRQ, 338 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl), 339 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl), 340 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable), 341 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable), 342 .safe_halt = native_safe_halt, 343 .halt = native_halt, 344 #ifdef CONFIG_X86_64 345 .adjust_exception_frame = paravirt_nop, 346 #endif 347 }; 348 349 struct pv_cpu_ops pv_cpu_ops = { 350 .cpuid = native_cpuid, 351 .get_debugreg = native_get_debugreg, 352 .set_debugreg = native_set_debugreg, 353 .clts = native_clts, 354 .read_cr0 = native_read_cr0, 355 .write_cr0 = native_write_cr0, 356 .read_cr4 = native_read_cr4, 357 .read_cr4_safe = native_read_cr4_safe, 358 .write_cr4 = native_write_cr4, 359 #ifdef CONFIG_X86_64 360 .read_cr8 = native_read_cr8, 361 .write_cr8 = native_write_cr8, 362 #endif 363 .wbinvd = native_wbinvd, 364 .read_msr = native_read_msr_safe, 365 .read_msr_amd = native_read_msr_amd_safe, 366 .write_msr = native_write_msr_safe, 367 .read_tsc = native_read_tsc, 368 .read_pmc = native_read_pmc, 369 .read_tscp = native_read_tscp, 370 .load_tr_desc = native_load_tr_desc, 371 .set_ldt = native_set_ldt, 372 .load_gdt = native_load_gdt, 373 .load_idt = native_load_idt, 374 .store_gdt = native_store_gdt, 375 .store_idt = native_store_idt, 376 .store_tr = native_store_tr, 377 .load_tls = native_load_tls, 378 #ifdef CONFIG_X86_64 379 .load_gs_index = native_load_gs_index, 380 #endif 381 .write_ldt_entry = native_write_ldt_entry, 382 .write_gdt_entry = native_write_gdt_entry, 383 .write_idt_entry = native_write_idt_entry, 384 385 .alloc_ldt = paravirt_nop, 386 .free_ldt = paravirt_nop, 387 388 .load_sp0 = native_load_sp0, 389 390 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 391 .irq_enable_sysexit = native_irq_enable_sysexit, 392 #endif 393 #ifdef CONFIG_X86_64 394 #ifdef CONFIG_IA32_EMULATION 395 .usergs_sysret32 = native_usergs_sysret32, 396 #endif 397 .usergs_sysret64 = native_usergs_sysret64, 398 #endif 399 .iret = native_iret, 400 .swapgs = native_swapgs, 401 402 .set_iopl_mask = native_set_iopl_mask, 403 .io_delay = native_io_delay, 404 405 .lazy_mode = { 406 .enter = paravirt_nop, 407 .leave = paravirt_nop, 408 }, 409 }; 410 411 struct pv_apic_ops pv_apic_ops = { 412 #ifdef CONFIG_X86_LOCAL_APIC 413 .setup_boot_clock = setup_boot_APIC_clock, 414 .setup_secondary_clock = setup_secondary_APIC_clock, 415 .startup_ipi_hook = paravirt_nop, 416 #endif 417 }; 418 419 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE) 420 /* 32-bit pagetable entries */ 421 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32) 422 #else 423 /* 64-bit pagetable entries */ 424 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64) 425 #endif 426 427 struct pv_mmu_ops pv_mmu_ops = { 428 #ifndef CONFIG_X86_64 429 .pagetable_setup_start = native_pagetable_setup_start, 430 .pagetable_setup_done = native_pagetable_setup_done, 431 #else 432 .pagetable_setup_start = paravirt_nop, 433 .pagetable_setup_done = paravirt_nop, 434 #endif 435 436 .read_cr2 = native_read_cr2, 437 .write_cr2 = native_write_cr2, 438 .read_cr3 = native_read_cr3, 439 .write_cr3 = native_write_cr3, 440 441 .flush_tlb_user = native_flush_tlb, 442 .flush_tlb_kernel = native_flush_tlb_global, 443 .flush_tlb_single = native_flush_tlb_single, 444 .flush_tlb_others = native_flush_tlb_others, 445 446 .pgd_alloc = __paravirt_pgd_alloc, 447 .pgd_free = paravirt_nop, 448 449 .alloc_pte = paravirt_nop, 450 .alloc_pmd = paravirt_nop, 451 .alloc_pmd_clone = paravirt_nop, 452 .alloc_pud = paravirt_nop, 453 .release_pte = paravirt_nop, 454 .release_pmd = paravirt_nop, 455 .release_pud = paravirt_nop, 456 457 .set_pte = native_set_pte, 458 .set_pte_at = native_set_pte_at, 459 .set_pmd = native_set_pmd, 460 .pte_update = paravirt_nop, 461 .pte_update_defer = paravirt_nop, 462 463 .ptep_modify_prot_start = __ptep_modify_prot_start, 464 .ptep_modify_prot_commit = __ptep_modify_prot_commit, 465 466 #ifdef CONFIG_HIGHPTE 467 .kmap_atomic_pte = kmap_atomic, 468 #endif 469 470 #if PAGETABLE_LEVELS >= 3 471 #ifdef CONFIG_X86_PAE 472 .set_pte_atomic = native_set_pte_atomic, 473 .pte_clear = native_pte_clear, 474 .pmd_clear = native_pmd_clear, 475 #endif 476 .set_pud = native_set_pud, 477 478 .pmd_val = PTE_IDENT, 479 .make_pmd = PTE_IDENT, 480 481 #if PAGETABLE_LEVELS == 4 482 .pud_val = PTE_IDENT, 483 .make_pud = PTE_IDENT, 484 485 .set_pgd = native_set_pgd, 486 #endif 487 #endif /* PAGETABLE_LEVELS >= 3 */ 488 489 .pte_val = PTE_IDENT, 490 .pgd_val = PTE_IDENT, 491 492 .make_pte = PTE_IDENT, 493 .make_pgd = PTE_IDENT, 494 495 .dup_mmap = paravirt_nop, 496 .exit_mmap = paravirt_nop, 497 .activate_mm = paravirt_nop, 498 499 .lazy_mode = { 500 .enter = paravirt_nop, 501 .leave = paravirt_nop, 502 }, 503 504 .set_fixmap = native_set_fixmap, 505 }; 506 507 EXPORT_SYMBOL_GPL(pv_time_ops); 508 EXPORT_SYMBOL (pv_cpu_ops); 509 EXPORT_SYMBOL (pv_mmu_ops); 510 EXPORT_SYMBOL_GPL(pv_apic_ops); 511 EXPORT_SYMBOL_GPL(pv_info); 512 EXPORT_SYMBOL (pv_irq_ops); 513