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#include <linux/kprobes.h> 27 28#include <asm/bug.h> 29#include <asm/paravirt.h> 30#include <asm/debugreg.h> 31#include <asm/desc.h> 32#include <asm/setup.h> 33#include <asm/pgtable.h> 34#include <asm/time.h> 35#include <asm/pgalloc.h> 36#include <asm/irq.h> 37#include <asm/delay.h> 38#include <asm/fixmap.h> 39#include <asm/apic.h> 40#include <asm/tlbflush.h> 41#include <asm/timer.h> 42#include <asm/special_insns.h> 43 44/* nop stub */ 45void _paravirt_nop(void) 46{ 47} 48 49/* identity function, which can be inlined */ 50u32 _paravirt_ident_32(u32 x) 51{ 52 return x; 53} 54 55u64 _paravirt_ident_64(u64 x) 56{ 57 return x; 58} 59 60void __init default_banner(void) 61{ 62 printk(KERN_INFO "Booting paravirtualized kernel on %s\n", 63 pv_info.name); 64} 65 66/* Undefined instruction for dealing with missing ops pointers. */ 67static const unsigned char ud2a[] = { 0x0f, 0x0b }; 68 69unsigned paravirt_patch_nop(void) 70{ 71 return 0; 72} 73 74unsigned paravirt_patch_ignore(unsigned len) 75{ 76 return len; 77} 78 79struct branch { 80 unsigned char opcode; 81 u32 delta; 82} __attribute__((packed)); 83 84unsigned paravirt_patch_call(void *insnbuf, 85 const void *target, u16 tgt_clobbers, 86 unsigned long addr, u16 site_clobbers, 87 unsigned len) 88{ 89 struct branch *b = insnbuf; 90 unsigned long delta = (unsigned long)target - (addr+5); 91 92 if (tgt_clobbers & ~site_clobbers) 93 return len; /* target would clobber too much for this site */ 94 if (len < 5) 95 return len; /* call too long for patch site */ 96 97 b->opcode = 0xe8; /* call */ 98 b->delta = delta; 99 BUILD_BUG_ON(sizeof(*b) != 5); 100 101 return 5; 102} 103 104unsigned paravirt_patch_jmp(void *insnbuf, const void *target, 105 unsigned long addr, unsigned len) 106{ 107 struct branch *b = insnbuf; 108 unsigned long delta = (unsigned long)target - (addr+5); 109 110 if (len < 5) 111 return len; /* call too long for patch site */ 112 113 b->opcode = 0xe9; /* jmp */ 114 b->delta = delta; 115 116 return 5; 117} 118 119/* Neat trick to map patch type back to the call within the 120 * corresponding structure. */ 121static void *get_call_destination(u8 type) 122{ 123 struct paravirt_patch_template tmpl = { 124 .pv_init_ops = pv_init_ops, 125 .pv_time_ops = pv_time_ops, 126 .pv_cpu_ops = pv_cpu_ops, 127 .pv_irq_ops = pv_irq_ops, 128 .pv_apic_ops = pv_apic_ops, 129 .pv_mmu_ops = pv_mmu_ops, 130#ifdef CONFIG_PARAVIRT_SPINLOCKS 131 .pv_lock_ops = pv_lock_ops, 132#endif 133 }; 134 return *((void **)&tmpl + type); 135} 136 137unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf, 138 unsigned long addr, unsigned len) 139{ 140 void *opfunc = get_call_destination(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 /* If the operation is a nop, then nop the callsite */ 148 ret = paravirt_patch_nop(); 149 150 /* identity functions just return their single argument */ 151 else if (opfunc == _paravirt_ident_32) 152 ret = paravirt_patch_ident_32(insnbuf, len); 153 else if (opfunc == _paravirt_ident_64) 154 ret = paravirt_patch_ident_64(insnbuf, len); 155 156 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) || 157 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) || 158 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) || 159 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64)) 160 /* If operation requires a jmp, then jmp */ 161 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len); 162 else 163 /* Otherwise call the function; assume target could 164 clobber any caller-save reg */ 165 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY, 166 addr, clobbers, len); 167 168 return ret; 169} 170 171unsigned paravirt_patch_insns(void *insnbuf, unsigned len, 172 const char *start, const char *end) 173{ 174 unsigned insn_len = end - start; 175 176 if (insn_len > len || start == NULL) 177 insn_len = len; 178 else 179 memcpy(insnbuf, start, insn_len); 180 181 return insn_len; 182} 183 184static void native_flush_tlb(void) 185{ 186 __native_flush_tlb(); 187} 188 189/* 190 * Global pages have to be flushed a bit differently. Not a real 191 * performance problem because this does not happen often. 192 */ 193static void native_flush_tlb_global(void) 194{ 195 __native_flush_tlb_global(); 196} 197 198static void native_flush_tlb_single(unsigned long addr) 199{ 200 __native_flush_tlb_single(addr); 201} 202 203struct static_key paravirt_steal_enabled; 204struct static_key paravirt_steal_rq_enabled; 205 206static u64 native_steal_clock(int cpu) 207{ 208 return 0; 209} 210 211/* These are in entry.S */ 212extern void native_iret(void); 213extern void native_irq_enable_sysexit(void); 214extern void native_usergs_sysret32(void); 215extern void native_usergs_sysret64(void); 216 217static struct resource reserve_ioports = { 218 .start = 0, 219 .end = IO_SPACE_LIMIT, 220 .name = "paravirt-ioport", 221 .flags = IORESOURCE_IO | IORESOURCE_BUSY, 222}; 223 224/* 225 * Reserve the whole legacy IO space to prevent any legacy drivers 226 * from wasting time probing for their hardware. This is a fairly 227 * brute-force approach to disabling all non-virtual drivers. 228 * 229 * Note that this must be called very early to have any effect. 230 */ 231int paravirt_disable_iospace(void) 232{ 233 return request_resource(&ioport_resource, &reserve_ioports); 234} 235 236static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE; 237 238static inline void enter_lazy(enum paravirt_lazy_mode mode) 239{ 240 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE); 241 242 this_cpu_write(paravirt_lazy_mode, mode); 243} 244 245static void leave_lazy(enum paravirt_lazy_mode mode) 246{ 247 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode); 248 249 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE); 250} 251 252void paravirt_enter_lazy_mmu(void) 253{ 254 enter_lazy(PARAVIRT_LAZY_MMU); 255} 256 257void paravirt_leave_lazy_mmu(void) 258{ 259 leave_lazy(PARAVIRT_LAZY_MMU); 260} 261 262void paravirt_flush_lazy_mmu(void) 263{ 264 preempt_disable(); 265 266 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { 267 arch_leave_lazy_mmu_mode(); 268 arch_enter_lazy_mmu_mode(); 269 } 270 271 preempt_enable(); 272} 273 274void paravirt_start_context_switch(struct task_struct *prev) 275{ 276 BUG_ON(preemptible()); 277 278 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) { 279 arch_leave_lazy_mmu_mode(); 280 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES); 281 } 282 enter_lazy(PARAVIRT_LAZY_CPU); 283} 284 285void paravirt_end_context_switch(struct task_struct *next) 286{ 287 BUG_ON(preemptible()); 288 289 leave_lazy(PARAVIRT_LAZY_CPU); 290 291 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES)) 292 arch_enter_lazy_mmu_mode(); 293} 294 295enum paravirt_lazy_mode paravirt_get_lazy_mode(void) 296{ 297 if (in_interrupt()) 298 return PARAVIRT_LAZY_NONE; 299 300 return this_cpu_read(paravirt_lazy_mode); 301} 302 303struct pv_info pv_info = { 304 .name = "bare hardware", 305 .paravirt_enabled = 0, 306 .kernel_rpl = 0, 307 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */ 308 309#ifdef CONFIG_X86_64 310 .extra_user_64bit_cs = __USER_CS, 311#endif 312}; 313 314struct pv_init_ops pv_init_ops = { 315 .patch = native_patch, 316}; 317 318struct pv_time_ops pv_time_ops = { 319 .sched_clock = native_sched_clock, 320 .steal_clock = native_steal_clock, 321}; 322 323__visible struct pv_irq_ops pv_irq_ops = { 324 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl), 325 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl), 326 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable), 327 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable), 328 .safe_halt = native_safe_halt, 329 .halt = native_halt, 330#ifdef CONFIG_X86_64 331 .adjust_exception_frame = paravirt_nop, 332#endif 333}; 334 335__visible struct pv_cpu_ops pv_cpu_ops = { 336 .cpuid = native_cpuid, 337 .get_debugreg = native_get_debugreg, 338 .set_debugreg = native_set_debugreg, 339 .clts = native_clts, 340 .read_cr0 = native_read_cr0, 341 .write_cr0 = native_write_cr0, 342 .read_cr4 = native_read_cr4, 343 .read_cr4_safe = native_read_cr4_safe, 344 .write_cr4 = native_write_cr4, 345#ifdef CONFIG_X86_64 346 .read_cr8 = native_read_cr8, 347 .write_cr8 = native_write_cr8, 348#endif 349 .wbinvd = native_wbinvd, 350 .read_msr = native_read_msr_safe, 351 .write_msr = native_write_msr_safe, 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_idt = native_store_idt, 360 .store_tr = native_store_tr, 361 .load_tls = native_load_tls, 362#ifdef CONFIG_X86_64 363 .load_gs_index = native_load_gs_index, 364#endif 365 .write_ldt_entry = native_write_ldt_entry, 366 .write_gdt_entry = native_write_gdt_entry, 367 .write_idt_entry = native_write_idt_entry, 368 369 .alloc_ldt = paravirt_nop, 370 .free_ldt = paravirt_nop, 371 372 .load_sp0 = native_load_sp0, 373 374#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 375 .irq_enable_sysexit = native_irq_enable_sysexit, 376#endif 377#ifdef CONFIG_X86_64 378#ifdef CONFIG_IA32_EMULATION 379 .usergs_sysret32 = native_usergs_sysret32, 380#endif 381 .usergs_sysret64 = native_usergs_sysret64, 382#endif 383 .iret = native_iret, 384 .swapgs = native_swapgs, 385 386 .set_iopl_mask = native_set_iopl_mask, 387 .io_delay = native_io_delay, 388 389 .start_context_switch = paravirt_nop, 390 .end_context_switch = paravirt_nop, 391}; 392 393/* At this point, native_get/set_debugreg has real function entries */ 394NOKPROBE_SYMBOL(native_get_debugreg); 395NOKPROBE_SYMBOL(native_set_debugreg); 396NOKPROBE_SYMBOL(native_load_idt); 397 398struct pv_apic_ops pv_apic_ops = { 399#ifdef CONFIG_X86_LOCAL_APIC 400 .startup_ipi_hook = paravirt_nop, 401#endif 402}; 403 404#if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE) 405/* 32-bit pagetable entries */ 406#define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32) 407#else 408/* 64-bit pagetable entries */ 409#define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64) 410#endif 411 412struct pv_mmu_ops pv_mmu_ops = { 413 414 .read_cr2 = native_read_cr2, 415 .write_cr2 = native_write_cr2, 416 .read_cr3 = native_read_cr3, 417 .write_cr3 = native_write_cr3, 418 419 .flush_tlb_user = native_flush_tlb, 420 .flush_tlb_kernel = native_flush_tlb_global, 421 .flush_tlb_single = native_flush_tlb_single, 422 .flush_tlb_others = native_flush_tlb_others, 423 424 .pgd_alloc = __paravirt_pgd_alloc, 425 .pgd_free = paravirt_nop, 426 427 .alloc_pte = paravirt_nop, 428 .alloc_pmd = paravirt_nop, 429 .alloc_pud = paravirt_nop, 430 .release_pte = paravirt_nop, 431 .release_pmd = paravirt_nop, 432 .release_pud = paravirt_nop, 433 434 .set_pte = native_set_pte, 435 .set_pte_at = native_set_pte_at, 436 .set_pmd = native_set_pmd, 437 .set_pmd_at = native_set_pmd_at, 438 .pte_update = paravirt_nop, 439 .pte_update_defer = paravirt_nop, 440 .pmd_update = paravirt_nop, 441 .pmd_update_defer = paravirt_nop, 442 443 .ptep_modify_prot_start = __ptep_modify_prot_start, 444 .ptep_modify_prot_commit = __ptep_modify_prot_commit, 445 446#if PAGETABLE_LEVELS >= 3 447#ifdef CONFIG_X86_PAE 448 .set_pte_atomic = native_set_pte_atomic, 449 .pte_clear = native_pte_clear, 450 .pmd_clear = native_pmd_clear, 451#endif 452 .set_pud = native_set_pud, 453 454 .pmd_val = PTE_IDENT, 455 .make_pmd = PTE_IDENT, 456 457#if PAGETABLE_LEVELS == 4 458 .pud_val = PTE_IDENT, 459 .make_pud = PTE_IDENT, 460 461 .set_pgd = native_set_pgd, 462#endif 463#endif /* PAGETABLE_LEVELS >= 3 */ 464 465 .pte_val = PTE_IDENT, 466 .pgd_val = PTE_IDENT, 467 468 .make_pte = PTE_IDENT, 469 .make_pgd = PTE_IDENT, 470 471 .dup_mmap = paravirt_nop, 472 .exit_mmap = paravirt_nop, 473 .activate_mm = paravirt_nop, 474 475 .lazy_mode = { 476 .enter = paravirt_nop, 477 .leave = paravirt_nop, 478 .flush = paravirt_nop, 479 }, 480 481 .set_fixmap = native_set_fixmap, 482}; 483 484EXPORT_SYMBOL_GPL(pv_time_ops); 485EXPORT_SYMBOL (pv_cpu_ops); 486EXPORT_SYMBOL (pv_mmu_ops); 487EXPORT_SYMBOL_GPL(pv_apic_ops); 488EXPORT_SYMBOL_GPL(pv_info); 489EXPORT_SYMBOL (pv_irq_ops); 490