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