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