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 #include <asm/tlb.h> 45 46 /* 47 * nop stub, which must not clobber anything *including the stack* to 48 * avoid confusing the entry prologues. 49 */ 50 extern void _paravirt_nop(void); 51 asm (".pushsection .entry.text, \"ax\"\n" 52 ".global _paravirt_nop\n" 53 "_paravirt_nop:\n\t" 54 "ret\n\t" 55 ".size _paravirt_nop, . - _paravirt_nop\n\t" 56 ".type _paravirt_nop, @function\n\t" 57 ".popsection"); 58 59 /* identity function, which can be inlined */ 60 u32 notrace _paravirt_ident_32(u32 x) 61 { 62 return x; 63 } 64 65 u64 notrace _paravirt_ident_64(u64 x) 66 { 67 return x; 68 } 69 70 void __init default_banner(void) 71 { 72 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 73 pv_info.name); 74 } 75 76 /* Undefined instruction for dealing with missing ops pointers. */ 77 static const unsigned char ud2a[] = { 0x0f, 0x0b }; 78 79 struct branch { 80 unsigned char opcode; 81 u32 delta; 82 } __attribute__((packed)); 83 84 static unsigned paravirt_patch_call(void *insnbuf, const void *target, 85 unsigned long addr, unsigned len) 86 { 87 struct branch *b = insnbuf; 88 unsigned long delta = (unsigned long)target - (addr+5); 89 90 if (len < 5) { 91 #ifdef CONFIG_RETPOLINE 92 WARN_ONCE(1, "Failing to patch indirect CALL in %ps\n", (void *)addr); 93 #endif 94 return len; /* call too long for patch site */ 95 } 96 97 b->opcode = 0xe8; /* call */ 98 b->delta = delta; 99 BUILD_BUG_ON(sizeof(*b) != 5); 100 101 return 5; 102 } 103 104 #ifdef CONFIG_PARAVIRT_XXL 105 static unsigned paravirt_patch_jmp(void *insnbuf, const void *target, 106 unsigned long addr, unsigned len) 107 { 108 struct branch *b = insnbuf; 109 unsigned long delta = (unsigned long)target - (addr+5); 110 111 if (len < 5) { 112 #ifdef CONFIG_RETPOLINE 113 WARN_ONCE(1, "Failing to patch indirect JMP in %ps\n", (void *)addr); 114 #endif 115 return len; /* call too long for patch site */ 116 } 117 118 b->opcode = 0xe9; /* jmp */ 119 b->delta = delta; 120 121 return 5; 122 } 123 #endif 124 125 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key); 126 127 void __init native_pv_lock_init(void) 128 { 129 if (!static_cpu_has(X86_FEATURE_HYPERVISOR)) 130 static_branch_disable(&virt_spin_lock_key); 131 } 132 133 unsigned paravirt_patch_default(u8 type, void *insnbuf, 134 unsigned long addr, unsigned len) 135 { 136 /* 137 * Neat trick to map patch type back to the call within the 138 * corresponding structure. 139 */ 140 void *opfunc = *((void **)&pv_ops + type); 141 unsigned ret; 142 143 if (opfunc == NULL) 144 /* If there's no function, patch it with a ud2a (BUG) */ 145 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a)); 146 else if (opfunc == _paravirt_nop) 147 ret = 0; 148 149 /* identity functions just return their single argument */ 150 else if (opfunc == _paravirt_ident_32) 151 ret = paravirt_patch_ident_32(insnbuf, len); 152 else if (opfunc == _paravirt_ident_64) 153 ret = paravirt_patch_ident_64(insnbuf, len); 154 155 #ifdef CONFIG_PARAVIRT_XXL 156 else if (type == PARAVIRT_PATCH(cpu.iret) || 157 type == PARAVIRT_PATCH(cpu.usergs_sysret64)) 158 /* If operation requires a jmp, then jmp */ 159 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len); 160 #endif 161 else 162 /* Otherwise call the function. */ 163 ret = paravirt_patch_call(insnbuf, opfunc, addr, len); 164 165 return ret; 166 } 167 168 unsigned paravirt_patch_insns(void *insnbuf, unsigned len, 169 const char *start, const char *end) 170 { 171 unsigned insn_len = end - start; 172 173 if (insn_len > len || start == NULL) 174 insn_len = len; 175 else 176 memcpy(insnbuf, start, insn_len); 177 178 return insn_len; 179 } 180 181 static void native_flush_tlb(void) 182 { 183 __native_flush_tlb(); 184 } 185 186 /* 187 * Global pages have to be flushed a bit differently. Not a real 188 * performance problem because this does not happen often. 189 */ 190 static void native_flush_tlb_global(void) 191 { 192 __native_flush_tlb_global(); 193 } 194 195 static void native_flush_tlb_one_user(unsigned long addr) 196 { 197 __native_flush_tlb_one_user(addr); 198 } 199 200 struct static_key paravirt_steal_enabled; 201 struct static_key paravirt_steal_rq_enabled; 202 203 static u64 native_steal_clock(int cpu) 204 { 205 return 0; 206 } 207 208 /* These are in entry.S */ 209 extern void native_iret(void); 210 extern void native_usergs_sysret64(void); 211 212 static struct resource reserve_ioports = { 213 .start = 0, 214 .end = IO_SPACE_LIMIT, 215 .name = "paravirt-ioport", 216 .flags = IORESOURCE_IO | IORESOURCE_BUSY, 217 }; 218 219 /* 220 * Reserve the whole legacy IO space to prevent any legacy drivers 221 * from wasting time probing for their hardware. This is a fairly 222 * brute-force approach to disabling all non-virtual drivers. 223 * 224 * Note that this must be called very early to have any effect. 225 */ 226 int paravirt_disable_iospace(void) 227 { 228 return request_resource(&ioport_resource, &reserve_ioports); 229 } 230 231 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE; 232 233 static inline void enter_lazy(enum paravirt_lazy_mode mode) 234 { 235 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE); 236 237 this_cpu_write(paravirt_lazy_mode, mode); 238 } 239 240 static void leave_lazy(enum paravirt_lazy_mode mode) 241 { 242 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode); 243 244 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE); 245 } 246 247 void paravirt_enter_lazy_mmu(void) 248 { 249 enter_lazy(PARAVIRT_LAZY_MMU); 250 } 251 252 void paravirt_leave_lazy_mmu(void) 253 { 254 leave_lazy(PARAVIRT_LAZY_MMU); 255 } 256 257 void paravirt_flush_lazy_mmu(void) 258 { 259 preempt_disable(); 260 261 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { 262 arch_leave_lazy_mmu_mode(); 263 arch_enter_lazy_mmu_mode(); 264 } 265 266 preempt_enable(); 267 } 268 269 #ifdef CONFIG_PARAVIRT_XXL 270 void paravirt_start_context_switch(struct task_struct *prev) 271 { 272 BUG_ON(preemptible()); 273 274 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) { 275 arch_leave_lazy_mmu_mode(); 276 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES); 277 } 278 enter_lazy(PARAVIRT_LAZY_CPU); 279 } 280 281 void paravirt_end_context_switch(struct task_struct *next) 282 { 283 BUG_ON(preemptible()); 284 285 leave_lazy(PARAVIRT_LAZY_CPU); 286 287 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES)) 288 arch_enter_lazy_mmu_mode(); 289 } 290 #endif 291 292 enum paravirt_lazy_mode paravirt_get_lazy_mode(void) 293 { 294 if (in_interrupt()) 295 return PARAVIRT_LAZY_NONE; 296 297 return this_cpu_read(paravirt_lazy_mode); 298 } 299 300 struct pv_info pv_info = { 301 .name = "bare hardware", 302 #ifdef CONFIG_PARAVIRT_XXL 303 .kernel_rpl = 0, 304 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */ 305 306 #ifdef CONFIG_X86_64 307 .extra_user_64bit_cs = __USER_CS, 308 #endif 309 #endif 310 }; 311 312 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE) 313 /* 32-bit pagetable entries */ 314 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32) 315 #else 316 /* 64-bit pagetable entries */ 317 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64) 318 #endif 319 320 struct paravirt_patch_template pv_ops = { 321 /* Init ops. */ 322 .init.patch = native_patch, 323 324 /* Time ops. */ 325 .time.sched_clock = native_sched_clock, 326 .time.steal_clock = native_steal_clock, 327 328 /* Cpu ops. */ 329 .cpu.io_delay = native_io_delay, 330 331 #ifdef CONFIG_PARAVIRT_XXL 332 .cpu.cpuid = native_cpuid, 333 .cpu.get_debugreg = native_get_debugreg, 334 .cpu.set_debugreg = native_set_debugreg, 335 .cpu.read_cr0 = native_read_cr0, 336 .cpu.write_cr0 = native_write_cr0, 337 .cpu.write_cr4 = native_write_cr4, 338 #ifdef CONFIG_X86_64 339 .cpu.read_cr8 = native_read_cr8, 340 .cpu.write_cr8 = native_write_cr8, 341 #endif 342 .cpu.wbinvd = native_wbinvd, 343 .cpu.read_msr = native_read_msr, 344 .cpu.write_msr = native_write_msr, 345 .cpu.read_msr_safe = native_read_msr_safe, 346 .cpu.write_msr_safe = native_write_msr_safe, 347 .cpu.read_pmc = native_read_pmc, 348 .cpu.load_tr_desc = native_load_tr_desc, 349 .cpu.set_ldt = native_set_ldt, 350 .cpu.load_gdt = native_load_gdt, 351 .cpu.load_idt = native_load_idt, 352 .cpu.store_tr = native_store_tr, 353 .cpu.load_tls = native_load_tls, 354 #ifdef CONFIG_X86_64 355 .cpu.load_gs_index = native_load_gs_index, 356 #endif 357 .cpu.write_ldt_entry = native_write_ldt_entry, 358 .cpu.write_gdt_entry = native_write_gdt_entry, 359 .cpu.write_idt_entry = native_write_idt_entry, 360 361 .cpu.alloc_ldt = paravirt_nop, 362 .cpu.free_ldt = paravirt_nop, 363 364 .cpu.load_sp0 = native_load_sp0, 365 366 #ifdef CONFIG_X86_64 367 .cpu.usergs_sysret64 = native_usergs_sysret64, 368 #endif 369 .cpu.iret = native_iret, 370 .cpu.swapgs = native_swapgs, 371 372 .cpu.set_iopl_mask = native_set_iopl_mask, 373 374 .cpu.start_context_switch = paravirt_nop, 375 .cpu.end_context_switch = paravirt_nop, 376 377 /* Irq ops. */ 378 .irq.save_fl = __PV_IS_CALLEE_SAVE(native_save_fl), 379 .irq.restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl), 380 .irq.irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable), 381 .irq.irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable), 382 .irq.safe_halt = native_safe_halt, 383 .irq.halt = native_halt, 384 #endif /* CONFIG_PARAVIRT_XXL */ 385 386 /* Mmu ops. */ 387 .mmu.flush_tlb_user = native_flush_tlb, 388 .mmu.flush_tlb_kernel = native_flush_tlb_global, 389 .mmu.flush_tlb_one_user = native_flush_tlb_one_user, 390 .mmu.flush_tlb_others = native_flush_tlb_others, 391 .mmu.tlb_remove_table = 392 (void (*)(struct mmu_gather *, void *))tlb_remove_page, 393 394 .mmu.exit_mmap = paravirt_nop, 395 396 #ifdef CONFIG_PARAVIRT_XXL 397 .mmu.read_cr2 = native_read_cr2, 398 .mmu.write_cr2 = native_write_cr2, 399 .mmu.read_cr3 = __native_read_cr3, 400 .mmu.write_cr3 = native_write_cr3, 401 402 .mmu.pgd_alloc = __paravirt_pgd_alloc, 403 .mmu.pgd_free = paravirt_nop, 404 405 .mmu.alloc_pte = paravirt_nop, 406 .mmu.alloc_pmd = paravirt_nop, 407 .mmu.alloc_pud = paravirt_nop, 408 .mmu.alloc_p4d = paravirt_nop, 409 .mmu.release_pte = paravirt_nop, 410 .mmu.release_pmd = paravirt_nop, 411 .mmu.release_pud = paravirt_nop, 412 .mmu.release_p4d = paravirt_nop, 413 414 .mmu.set_pte = native_set_pte, 415 .mmu.set_pte_at = native_set_pte_at, 416 .mmu.set_pmd = native_set_pmd, 417 418 .mmu.ptep_modify_prot_start = __ptep_modify_prot_start, 419 .mmu.ptep_modify_prot_commit = __ptep_modify_prot_commit, 420 421 #if CONFIG_PGTABLE_LEVELS >= 3 422 #ifdef CONFIG_X86_PAE 423 .mmu.set_pte_atomic = native_set_pte_atomic, 424 .mmu.pte_clear = native_pte_clear, 425 .mmu.pmd_clear = native_pmd_clear, 426 #endif 427 .mmu.set_pud = native_set_pud, 428 429 .mmu.pmd_val = PTE_IDENT, 430 .mmu.make_pmd = PTE_IDENT, 431 432 #if CONFIG_PGTABLE_LEVELS >= 4 433 .mmu.pud_val = PTE_IDENT, 434 .mmu.make_pud = PTE_IDENT, 435 436 .mmu.set_p4d = native_set_p4d, 437 438 #if CONFIG_PGTABLE_LEVELS >= 5 439 .mmu.p4d_val = PTE_IDENT, 440 .mmu.make_p4d = PTE_IDENT, 441 442 .mmu.set_pgd = native_set_pgd, 443 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */ 444 #endif /* CONFIG_PGTABLE_LEVELS >= 4 */ 445 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */ 446 447 .mmu.pte_val = PTE_IDENT, 448 .mmu.pgd_val = PTE_IDENT, 449 450 .mmu.make_pte = PTE_IDENT, 451 .mmu.make_pgd = PTE_IDENT, 452 453 .mmu.dup_mmap = paravirt_nop, 454 .mmu.activate_mm = paravirt_nop, 455 456 .mmu.lazy_mode = { 457 .enter = paravirt_nop, 458 .leave = paravirt_nop, 459 .flush = paravirt_nop, 460 }, 461 462 .mmu.set_fixmap = native_set_fixmap, 463 #endif /* CONFIG_PARAVIRT_XXL */ 464 465 #if defined(CONFIG_PARAVIRT_SPINLOCKS) 466 /* Lock ops. */ 467 #ifdef CONFIG_SMP 468 .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath, 469 .lock.queued_spin_unlock = 470 PV_CALLEE_SAVE(__native_queued_spin_unlock), 471 .lock.wait = paravirt_nop, 472 .lock.kick = paravirt_nop, 473 .lock.vcpu_is_preempted = 474 PV_CALLEE_SAVE(__native_vcpu_is_preempted), 475 #endif /* SMP */ 476 #endif 477 }; 478 479 #ifdef CONFIG_PARAVIRT_XXL 480 /* At this point, native_get/set_debugreg has real function entries */ 481 NOKPROBE_SYMBOL(native_get_debugreg); 482 NOKPROBE_SYMBOL(native_set_debugreg); 483 NOKPROBE_SYMBOL(native_load_idt); 484 #endif 485 486 EXPORT_SYMBOL_GPL(pv_ops); 487 EXPORT_SYMBOL_GPL(pv_info); 488