Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 1 | #ifndef _ASM_X86_PARAVIRT_TYPES_H |
| 2 | #define _ASM_X86_PARAVIRT_TYPES_H |
| 3 | |
| 4 | /* Bitmask of what can be clobbered: usually at least eax. */ |
| 5 | #define CLBR_NONE 0 |
| 6 | #define CLBR_EAX (1 << 0) |
| 7 | #define CLBR_ECX (1 << 1) |
| 8 | #define CLBR_EDX (1 << 2) |
| 9 | #define CLBR_EDI (1 << 3) |
| 10 | |
| 11 | #ifdef CONFIG_X86_32 |
| 12 | /* CLBR_ANY should match all regs platform has. For i386, that's just it */ |
| 13 | #define CLBR_ANY ((1 << 4) - 1) |
| 14 | |
| 15 | #define CLBR_ARG_REGS (CLBR_EAX | CLBR_EDX | CLBR_ECX) |
| 16 | #define CLBR_RET_REG (CLBR_EAX | CLBR_EDX) |
| 17 | #define CLBR_SCRATCH (0) |
| 18 | #else |
| 19 | #define CLBR_RAX CLBR_EAX |
| 20 | #define CLBR_RCX CLBR_ECX |
| 21 | #define CLBR_RDX CLBR_EDX |
| 22 | #define CLBR_RDI CLBR_EDI |
| 23 | #define CLBR_RSI (1 << 4) |
| 24 | #define CLBR_R8 (1 << 5) |
| 25 | #define CLBR_R9 (1 << 6) |
| 26 | #define CLBR_R10 (1 << 7) |
| 27 | #define CLBR_R11 (1 << 8) |
| 28 | |
| 29 | #define CLBR_ANY ((1 << 9) - 1) |
| 30 | |
| 31 | #define CLBR_ARG_REGS (CLBR_RDI | CLBR_RSI | CLBR_RDX | \ |
| 32 | CLBR_RCX | CLBR_R8 | CLBR_R9) |
| 33 | #define CLBR_RET_REG (CLBR_RAX) |
| 34 | #define CLBR_SCRATCH (CLBR_R10 | CLBR_R11) |
| 35 | |
| 36 | #endif /* X86_64 */ |
| 37 | |
| 38 | #define CLBR_CALLEE_SAVE ((CLBR_ARG_REGS | CLBR_SCRATCH) & ~CLBR_RET_REG) |
| 39 | |
| 40 | #ifndef __ASSEMBLY__ |
| 41 | |
| 42 | #include <asm/desc_defs.h> |
| 43 | #include <asm/kmap_types.h> |
| 44 | |
| 45 | struct page; |
| 46 | struct thread_struct; |
| 47 | struct desc_ptr; |
| 48 | struct tss_struct; |
| 49 | struct mm_struct; |
| 50 | struct desc_struct; |
| 51 | struct task_struct; |
| 52 | struct cpumask; |
| 53 | |
| 54 | /* |
| 55 | * Wrapper type for pointers to code which uses the non-standard |
| 56 | * calling convention. See PV_CALL_SAVE_REGS_THUNK below. |
| 57 | */ |
| 58 | struct paravirt_callee_save { |
| 59 | void *func; |
| 60 | }; |
| 61 | |
| 62 | /* general info */ |
| 63 | struct pv_info { |
| 64 | unsigned int kernel_rpl; |
| 65 | int shared_kernel_pmd; |
| 66 | int paravirt_enabled; |
| 67 | const char *name; |
| 68 | }; |
| 69 | |
| 70 | struct pv_init_ops { |
| 71 | /* |
| 72 | * Patch may replace one of the defined code sequences with |
| 73 | * arbitrary code, subject to the same register constraints. |
| 74 | * This generally means the code is not free to clobber any |
| 75 | * registers other than EAX. The patch function should return |
| 76 | * the number of bytes of code generated, as we nop pad the |
| 77 | * rest in generic code. |
| 78 | */ |
| 79 | unsigned (*patch)(u8 type, u16 clobber, void *insnbuf, |
| 80 | unsigned long addr, unsigned len); |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 81 | }; |
| 82 | |
| 83 | |
| 84 | struct pv_lazy_ops { |
| 85 | /* Set deferred update mode, used for batching operations. */ |
| 86 | void (*enter)(void); |
| 87 | void (*leave)(void); |
| 88 | }; |
| 89 | |
| 90 | struct pv_time_ops { |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 91 | unsigned long long (*sched_clock)(void); |
| 92 | unsigned long (*get_tsc_khz)(void); |
| 93 | }; |
| 94 | |
| 95 | struct pv_cpu_ops { |
| 96 | /* hooks for various privileged instructions */ |
| 97 | unsigned long (*get_debugreg)(int regno); |
| 98 | void (*set_debugreg)(int regno, unsigned long value); |
| 99 | |
| 100 | void (*clts)(void); |
| 101 | |
| 102 | unsigned long (*read_cr0)(void); |
| 103 | void (*write_cr0)(unsigned long); |
| 104 | |
| 105 | unsigned long (*read_cr4_safe)(void); |
| 106 | unsigned long (*read_cr4)(void); |
| 107 | void (*write_cr4)(unsigned long); |
| 108 | |
| 109 | #ifdef CONFIG_X86_64 |
| 110 | unsigned long (*read_cr8)(void); |
| 111 | void (*write_cr8)(unsigned long); |
| 112 | #endif |
| 113 | |
| 114 | /* Segment descriptor handling */ |
| 115 | void (*load_tr_desc)(void); |
| 116 | void (*load_gdt)(const struct desc_ptr *); |
| 117 | void (*load_idt)(const struct desc_ptr *); |
| 118 | void (*store_gdt)(struct desc_ptr *); |
| 119 | void (*store_idt)(struct desc_ptr *); |
| 120 | void (*set_ldt)(const void *desc, unsigned entries); |
| 121 | unsigned long (*store_tr)(void); |
| 122 | void (*load_tls)(struct thread_struct *t, unsigned int cpu); |
| 123 | #ifdef CONFIG_X86_64 |
| 124 | void (*load_gs_index)(unsigned int idx); |
| 125 | #endif |
| 126 | void (*write_ldt_entry)(struct desc_struct *ldt, int entrynum, |
| 127 | const void *desc); |
| 128 | void (*write_gdt_entry)(struct desc_struct *, |
| 129 | int entrynum, const void *desc, int size); |
| 130 | void (*write_idt_entry)(gate_desc *, |
| 131 | int entrynum, const gate_desc *gate); |
| 132 | void (*alloc_ldt)(struct desc_struct *ldt, unsigned entries); |
| 133 | void (*free_ldt)(struct desc_struct *ldt, unsigned entries); |
| 134 | |
| 135 | void (*load_sp0)(struct tss_struct *tss, struct thread_struct *t); |
| 136 | |
| 137 | void (*set_iopl_mask)(unsigned mask); |
| 138 | |
| 139 | void (*wbinvd)(void); |
| 140 | void (*io_delay)(void); |
| 141 | |
| 142 | /* cpuid emulation, mostly so that caps bits can be disabled */ |
| 143 | void (*cpuid)(unsigned int *eax, unsigned int *ebx, |
| 144 | unsigned int *ecx, unsigned int *edx); |
| 145 | |
| 146 | /* MSR, PMC and TSR operations. |
| 147 | err = 0/-EFAULT. wrmsr returns 0/-EFAULT. */ |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 148 | u64 (*read_msr)(unsigned int msr, int *err); |
Ingo Molnar | c931aaf | 2009-09-01 12:13:30 +0200 | [diff] [blame] | 149 | int (*rdmsr_regs)(u32 *regs); |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 150 | int (*write_msr)(unsigned int msr, unsigned low, unsigned high); |
Ingo Molnar | c931aaf | 2009-09-01 12:13:30 +0200 | [diff] [blame] | 151 | int (*wrmsr_regs)(u32 *regs); |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 152 | |
| 153 | u64 (*read_tsc)(void); |
| 154 | u64 (*read_pmc)(int counter); |
| 155 | unsigned long long (*read_tscp)(unsigned int *aux); |
| 156 | |
| 157 | /* |
| 158 | * Atomically enable interrupts and return to userspace. This |
| 159 | * is only ever used to return to 32-bit processes; in a |
| 160 | * 64-bit kernel, it's used for 32-on-64 compat processes, but |
| 161 | * never native 64-bit processes. (Jump, not call.) |
| 162 | */ |
| 163 | void (*irq_enable_sysexit)(void); |
| 164 | |
| 165 | /* |
| 166 | * Switch to usermode gs and return to 64-bit usermode using |
| 167 | * sysret. Only used in 64-bit kernels to return to 64-bit |
| 168 | * processes. Usermode register state, including %rsp, must |
| 169 | * already be restored. |
| 170 | */ |
| 171 | void (*usergs_sysret64)(void); |
| 172 | |
| 173 | /* |
| 174 | * Switch to usermode gs and return to 32-bit usermode using |
| 175 | * sysret. Used to return to 32-on-64 compat processes. |
| 176 | * Other usermode register state, including %esp, must already |
| 177 | * be restored. |
| 178 | */ |
| 179 | void (*usergs_sysret32)(void); |
| 180 | |
| 181 | /* Normal iret. Jump to this with the standard iret stack |
| 182 | frame set up. */ |
| 183 | void (*iret)(void); |
| 184 | |
| 185 | void (*swapgs)(void); |
| 186 | |
| 187 | void (*start_context_switch)(struct task_struct *prev); |
| 188 | void (*end_context_switch)(struct task_struct *next); |
| 189 | }; |
| 190 | |
| 191 | struct pv_irq_ops { |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 192 | /* |
| 193 | * Get/set interrupt state. save_fl and restore_fl are only |
| 194 | * expected to use X86_EFLAGS_IF; all other bits |
| 195 | * returned from save_fl are undefined, and may be ignored by |
| 196 | * restore_fl. |
| 197 | * |
| 198 | * NOTE: These functions callers expect the callee to preserve |
| 199 | * more registers than the standard C calling convention. |
| 200 | */ |
| 201 | struct paravirt_callee_save save_fl; |
| 202 | struct paravirt_callee_save restore_fl; |
| 203 | struct paravirt_callee_save irq_disable; |
| 204 | struct paravirt_callee_save irq_enable; |
| 205 | |
| 206 | void (*safe_halt)(void); |
| 207 | void (*halt)(void); |
| 208 | |
| 209 | #ifdef CONFIG_X86_64 |
| 210 | void (*adjust_exception_frame)(void); |
| 211 | #endif |
| 212 | }; |
| 213 | |
| 214 | struct pv_apic_ops { |
| 215 | #ifdef CONFIG_X86_LOCAL_APIC |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 216 | void (*startup_ipi_hook)(int phys_apicid, |
| 217 | unsigned long start_eip, |
| 218 | unsigned long start_esp); |
| 219 | #endif |
| 220 | }; |
| 221 | |
| 222 | struct pv_mmu_ops { |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 223 | unsigned long (*read_cr2)(void); |
| 224 | void (*write_cr2)(unsigned long); |
| 225 | |
| 226 | unsigned long (*read_cr3)(void); |
| 227 | void (*write_cr3)(unsigned long); |
| 228 | |
| 229 | /* |
| 230 | * Hooks for intercepting the creation/use/destruction of an |
| 231 | * mm_struct. |
| 232 | */ |
| 233 | void (*activate_mm)(struct mm_struct *prev, |
| 234 | struct mm_struct *next); |
| 235 | void (*dup_mmap)(struct mm_struct *oldmm, |
| 236 | struct mm_struct *mm); |
| 237 | void (*exit_mmap)(struct mm_struct *mm); |
| 238 | |
| 239 | |
| 240 | /* TLB operations */ |
| 241 | void (*flush_tlb_user)(void); |
| 242 | void (*flush_tlb_kernel)(void); |
| 243 | void (*flush_tlb_single)(unsigned long addr); |
| 244 | void (*flush_tlb_others)(const struct cpumask *cpus, |
| 245 | struct mm_struct *mm, |
| 246 | unsigned long va); |
| 247 | |
| 248 | /* Hooks for allocating and freeing a pagetable top-level */ |
| 249 | int (*pgd_alloc)(struct mm_struct *mm); |
| 250 | void (*pgd_free)(struct mm_struct *mm, pgd_t *pgd); |
| 251 | |
| 252 | /* |
| 253 | * Hooks for allocating/releasing pagetable pages when they're |
| 254 | * attached to a pagetable |
| 255 | */ |
| 256 | void (*alloc_pte)(struct mm_struct *mm, unsigned long pfn); |
| 257 | void (*alloc_pmd)(struct mm_struct *mm, unsigned long pfn); |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 258 | void (*alloc_pud)(struct mm_struct *mm, unsigned long pfn); |
| 259 | void (*release_pte)(unsigned long pfn); |
| 260 | void (*release_pmd)(unsigned long pfn); |
| 261 | void (*release_pud)(unsigned long pfn); |
| 262 | |
| 263 | /* Pagetable manipulation functions */ |
| 264 | void (*set_pte)(pte_t *ptep, pte_t pteval); |
| 265 | void (*set_pte_at)(struct mm_struct *mm, unsigned long addr, |
| 266 | pte_t *ptep, pte_t pteval); |
| 267 | void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval); |
Andrea Arcangeli | 331127f | 2011-01-13 15:46:36 -0800 | [diff] [blame] | 268 | void (*set_pmd_at)(struct mm_struct *mm, unsigned long addr, |
| 269 | pmd_t *pmdp, pmd_t pmdval); |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 270 | void (*pte_update)(struct mm_struct *mm, unsigned long addr, |
| 271 | pte_t *ptep); |
| 272 | void (*pte_update_defer)(struct mm_struct *mm, |
| 273 | unsigned long addr, pte_t *ptep); |
Andrea Arcangeli | 331127f | 2011-01-13 15:46:36 -0800 | [diff] [blame] | 274 | void (*pmd_update)(struct mm_struct *mm, unsigned long addr, |
| 275 | pmd_t *pmdp); |
| 276 | void (*pmd_update_defer)(struct mm_struct *mm, |
| 277 | unsigned long addr, pmd_t *pmdp); |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 278 | |
| 279 | pte_t (*ptep_modify_prot_start)(struct mm_struct *mm, unsigned long addr, |
| 280 | pte_t *ptep); |
| 281 | void (*ptep_modify_prot_commit)(struct mm_struct *mm, unsigned long addr, |
| 282 | pte_t *ptep, pte_t pte); |
| 283 | |
| 284 | struct paravirt_callee_save pte_val; |
| 285 | struct paravirt_callee_save make_pte; |
| 286 | |
| 287 | struct paravirt_callee_save pgd_val; |
| 288 | struct paravirt_callee_save make_pgd; |
| 289 | |
| 290 | #if PAGETABLE_LEVELS >= 3 |
| 291 | #ifdef CONFIG_X86_PAE |
| 292 | void (*set_pte_atomic)(pte_t *ptep, pte_t pteval); |
| 293 | void (*pte_clear)(struct mm_struct *mm, unsigned long addr, |
| 294 | pte_t *ptep); |
| 295 | void (*pmd_clear)(pmd_t *pmdp); |
| 296 | |
| 297 | #endif /* CONFIG_X86_PAE */ |
| 298 | |
| 299 | void (*set_pud)(pud_t *pudp, pud_t pudval); |
| 300 | |
| 301 | struct paravirt_callee_save pmd_val; |
| 302 | struct paravirt_callee_save make_pmd; |
| 303 | |
| 304 | #if PAGETABLE_LEVELS == 4 |
| 305 | struct paravirt_callee_save pud_val; |
| 306 | struct paravirt_callee_save make_pud; |
| 307 | |
| 308 | void (*set_pgd)(pgd_t *pudp, pgd_t pgdval); |
| 309 | #endif /* PAGETABLE_LEVELS == 4 */ |
| 310 | #endif /* PAGETABLE_LEVELS >= 3 */ |
| 311 | |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 312 | struct pv_lazy_ops lazy_mode; |
| 313 | |
| 314 | /* dom0 ops */ |
| 315 | |
| 316 | /* Sometimes the physical address is a pfn, and sometimes its |
| 317 | an mfn. We can tell which is which from the index. */ |
| 318 | void (*set_fixmap)(unsigned /* enum fixed_addresses */ idx, |
| 319 | phys_addr_t phys, pgprot_t flags); |
| 320 | }; |
| 321 | |
Thomas Gleixner | 445c895 | 2009-12-02 19:49:50 +0100 | [diff] [blame] | 322 | struct arch_spinlock; |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 323 | struct pv_lock_ops { |
Thomas Gleixner | 445c895 | 2009-12-02 19:49:50 +0100 | [diff] [blame] | 324 | int (*spin_is_locked)(struct arch_spinlock *lock); |
| 325 | int (*spin_is_contended)(struct arch_spinlock *lock); |
| 326 | void (*spin_lock)(struct arch_spinlock *lock); |
| 327 | void (*spin_lock_flags)(struct arch_spinlock *lock, unsigned long flags); |
| 328 | int (*spin_trylock)(struct arch_spinlock *lock); |
| 329 | void (*spin_unlock)(struct arch_spinlock *lock); |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 330 | }; |
| 331 | |
| 332 | /* This contains all the paravirt structures: we get a convenient |
| 333 | * number for each function using the offset which we use to indicate |
| 334 | * what to patch. */ |
| 335 | struct paravirt_patch_template { |
| 336 | struct pv_init_ops pv_init_ops; |
| 337 | struct pv_time_ops pv_time_ops; |
| 338 | struct pv_cpu_ops pv_cpu_ops; |
| 339 | struct pv_irq_ops pv_irq_ops; |
| 340 | struct pv_apic_ops pv_apic_ops; |
| 341 | struct pv_mmu_ops pv_mmu_ops; |
| 342 | struct pv_lock_ops pv_lock_ops; |
| 343 | }; |
| 344 | |
| 345 | extern struct pv_info pv_info; |
| 346 | extern struct pv_init_ops pv_init_ops; |
| 347 | extern struct pv_time_ops pv_time_ops; |
| 348 | extern struct pv_cpu_ops pv_cpu_ops; |
| 349 | extern struct pv_irq_ops pv_irq_ops; |
| 350 | extern struct pv_apic_ops pv_apic_ops; |
| 351 | extern struct pv_mmu_ops pv_mmu_ops; |
| 352 | extern struct pv_lock_ops pv_lock_ops; |
| 353 | |
| 354 | #define PARAVIRT_PATCH(x) \ |
| 355 | (offsetof(struct paravirt_patch_template, x) / sizeof(void *)) |
| 356 | |
| 357 | #define paravirt_type(op) \ |
| 358 | [paravirt_typenum] "i" (PARAVIRT_PATCH(op)), \ |
| 359 | [paravirt_opptr] "i" (&(op)) |
| 360 | #define paravirt_clobber(clobber) \ |
| 361 | [paravirt_clobber] "i" (clobber) |
| 362 | |
| 363 | /* |
| 364 | * Generate some code, and mark it as patchable by the |
| 365 | * apply_paravirt() alternate instruction patcher. |
| 366 | */ |
| 367 | #define _paravirt_alt(insn_string, type, clobber) \ |
| 368 | "771:\n\t" insn_string "\n" "772:\n" \ |
| 369 | ".pushsection .parainstructions,\"a\"\n" \ |
| 370 | _ASM_ALIGN "\n" \ |
| 371 | _ASM_PTR " 771b\n" \ |
| 372 | " .byte " type "\n" \ |
| 373 | " .byte 772b-771b\n" \ |
| 374 | " .short " clobber "\n" \ |
| 375 | ".popsection\n" |
| 376 | |
| 377 | /* Generate patchable code, with the default asm parameters. */ |
| 378 | #define paravirt_alt(insn_string) \ |
| 379 | _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]") |
| 380 | |
| 381 | /* Simple instruction patching code. */ |
| 382 | #define DEF_NATIVE(ops, name, code) \ |
| 383 | extern const char start_##ops##_##name[], end_##ops##_##name[]; \ |
| 384 | asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":") |
| 385 | |
| 386 | unsigned paravirt_patch_nop(void); |
| 387 | unsigned paravirt_patch_ident_32(void *insnbuf, unsigned len); |
| 388 | unsigned paravirt_patch_ident_64(void *insnbuf, unsigned len); |
| 389 | unsigned paravirt_patch_ignore(unsigned len); |
| 390 | unsigned paravirt_patch_call(void *insnbuf, |
| 391 | const void *target, u16 tgt_clobbers, |
| 392 | unsigned long addr, u16 site_clobbers, |
| 393 | unsigned len); |
| 394 | unsigned paravirt_patch_jmp(void *insnbuf, const void *target, |
| 395 | unsigned long addr, unsigned len); |
| 396 | unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf, |
| 397 | unsigned long addr, unsigned len); |
| 398 | |
| 399 | unsigned paravirt_patch_insns(void *insnbuf, unsigned len, |
| 400 | const char *start, const char *end); |
| 401 | |
| 402 | unsigned native_patch(u8 type, u16 clobbers, void *ibuf, |
| 403 | unsigned long addr, unsigned len); |
| 404 | |
| 405 | int paravirt_disable_iospace(void); |
| 406 | |
| 407 | /* |
| 408 | * This generates an indirect call based on the operation type number. |
| 409 | * The type number, computed in PARAVIRT_PATCH, is derived from the |
| 410 | * offset into the paravirt_patch_template structure, and can therefore be |
| 411 | * freely converted back into a structure offset. |
| 412 | */ |
| 413 | #define PARAVIRT_CALL "call *%c[paravirt_opptr];" |
| 414 | |
| 415 | /* |
| 416 | * These macros are intended to wrap calls through one of the paravirt |
| 417 | * ops structs, so that they can be later identified and patched at |
| 418 | * runtime. |
| 419 | * |
| 420 | * Normally, a call to a pv_op function is a simple indirect call: |
| 421 | * (pv_op_struct.operations)(args...). |
| 422 | * |
| 423 | * Unfortunately, this is a relatively slow operation for modern CPUs, |
| 424 | * because it cannot necessarily determine what the destination |
| 425 | * address is. In this case, the address is a runtime constant, so at |
| 426 | * the very least we can patch the call to e a simple direct call, or |
| 427 | * ideally, patch an inline implementation into the callsite. (Direct |
| 428 | * calls are essentially free, because the call and return addresses |
| 429 | * are completely predictable.) |
| 430 | * |
| 431 | * For i386, these macros rely on the standard gcc "regparm(3)" calling |
| 432 | * convention, in which the first three arguments are placed in %eax, |
| 433 | * %edx, %ecx (in that order), and the remaining arguments are placed |
| 434 | * on the stack. All caller-save registers (eax,edx,ecx) are expected |
| 435 | * to be modified (either clobbered or used for return values). |
| 436 | * X86_64, on the other hand, already specifies a register-based calling |
| 437 | * conventions, returning at %rax, with parameteres going on %rdi, %rsi, |
| 438 | * %rdx, and %rcx. Note that for this reason, x86_64 does not need any |
| 439 | * special handling for dealing with 4 arguments, unlike i386. |
| 440 | * However, x86_64 also have to clobber all caller saved registers, which |
| 441 | * unfortunately, are quite a bit (r8 - r11) |
| 442 | * |
| 443 | * The call instruction itself is marked by placing its start address |
| 444 | * and size into the .parainstructions section, so that |
| 445 | * apply_paravirt() in arch/i386/kernel/alternative.c can do the |
| 446 | * appropriate patching under the control of the backend pv_init_ops |
| 447 | * implementation. |
| 448 | * |
| 449 | * Unfortunately there's no way to get gcc to generate the args setup |
| 450 | * for the call, and then allow the call itself to be generated by an |
| 451 | * inline asm. Because of this, we must do the complete arg setup and |
| 452 | * return value handling from within these macros. This is fairly |
| 453 | * cumbersome. |
| 454 | * |
| 455 | * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments. |
| 456 | * It could be extended to more arguments, but there would be little |
| 457 | * to be gained from that. For each number of arguments, there are |
| 458 | * the two VCALL and CALL variants for void and non-void functions. |
| 459 | * |
| 460 | * When there is a return value, the invoker of the macro must specify |
| 461 | * the return type. The macro then uses sizeof() on that type to |
| 462 | * determine whether its a 32 or 64 bit value, and places the return |
| 463 | * in the right register(s) (just %eax for 32-bit, and %edx:%eax for |
| 464 | * 64-bit). For x86_64 machines, it just returns at %rax regardless of |
| 465 | * the return value size. |
| 466 | * |
| 467 | * 64-bit arguments are passed as a pair of adjacent 32-bit arguments |
| 468 | * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments |
| 469 | * in low,high order |
| 470 | * |
| 471 | * Small structures are passed and returned in registers. The macro |
| 472 | * calling convention can't directly deal with this, so the wrapper |
| 473 | * functions must do this. |
| 474 | * |
| 475 | * These PVOP_* macros are only defined within this header. This |
| 476 | * means that all uses must be wrapped in inline functions. This also |
| 477 | * makes sure the incoming and outgoing types are always correct. |
| 478 | */ |
| 479 | #ifdef CONFIG_X86_32 |
| 480 | #define PVOP_VCALL_ARGS \ |
| 481 | unsigned long __eax = __eax, __edx = __edx, __ecx = __ecx |
| 482 | #define PVOP_CALL_ARGS PVOP_VCALL_ARGS |
| 483 | |
| 484 | #define PVOP_CALL_ARG1(x) "a" ((unsigned long)(x)) |
| 485 | #define PVOP_CALL_ARG2(x) "d" ((unsigned long)(x)) |
| 486 | #define PVOP_CALL_ARG3(x) "c" ((unsigned long)(x)) |
| 487 | |
| 488 | #define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \ |
| 489 | "=c" (__ecx) |
| 490 | #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS |
| 491 | |
| 492 | #define PVOP_VCALLEE_CLOBBERS "=a" (__eax), "=d" (__edx) |
| 493 | #define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS |
| 494 | |
| 495 | #define EXTRA_CLOBBERS |
| 496 | #define VEXTRA_CLOBBERS |
| 497 | #else /* CONFIG_X86_64 */ |
Jeremy Fitzhardinge | 71999d9 | 2009-10-12 16:32:43 -0700 | [diff] [blame] | 498 | /* [re]ax isn't an arg, but the return val */ |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 499 | #define PVOP_VCALL_ARGS \ |
| 500 | unsigned long __edi = __edi, __esi = __esi, \ |
Jeremy Fitzhardinge | 71999d9 | 2009-10-12 16:32:43 -0700 | [diff] [blame] | 501 | __edx = __edx, __ecx = __ecx, __eax = __eax |
| 502 | #define PVOP_CALL_ARGS PVOP_VCALL_ARGS |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 503 | |
| 504 | #define PVOP_CALL_ARG1(x) "D" ((unsigned long)(x)) |
| 505 | #define PVOP_CALL_ARG2(x) "S" ((unsigned long)(x)) |
| 506 | #define PVOP_CALL_ARG3(x) "d" ((unsigned long)(x)) |
| 507 | #define PVOP_CALL_ARG4(x) "c" ((unsigned long)(x)) |
| 508 | |
| 509 | #define PVOP_VCALL_CLOBBERS "=D" (__edi), \ |
| 510 | "=S" (__esi), "=d" (__edx), \ |
| 511 | "=c" (__ecx) |
| 512 | #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax) |
| 513 | |
Jeremy Fitzhardinge | 71999d9 | 2009-10-12 16:32:43 -0700 | [diff] [blame] | 514 | /* void functions are still allowed [re]ax for scratch */ |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 515 | #define PVOP_VCALLEE_CLOBBERS "=a" (__eax) |
| 516 | #define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS |
| 517 | |
| 518 | #define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11" |
| 519 | #define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11" |
| 520 | #endif /* CONFIG_X86_32 */ |
| 521 | |
| 522 | #ifdef CONFIG_PARAVIRT_DEBUG |
| 523 | #define PVOP_TEST_NULL(op) BUG_ON(op == NULL) |
| 524 | #else |
| 525 | #define PVOP_TEST_NULL(op) ((void)op) |
| 526 | #endif |
| 527 | |
| 528 | #define ____PVOP_CALL(rettype, op, clbr, call_clbr, extra_clbr, \ |
| 529 | pre, post, ...) \ |
| 530 | ({ \ |
| 531 | rettype __ret; \ |
| 532 | PVOP_CALL_ARGS; \ |
| 533 | PVOP_TEST_NULL(op); \ |
| 534 | /* This is 32-bit specific, but is okay in 64-bit */ \ |
| 535 | /* since this condition will never hold */ \ |
| 536 | if (sizeof(rettype) > sizeof(unsigned long)) { \ |
| 537 | asm volatile(pre \ |
| 538 | paravirt_alt(PARAVIRT_CALL) \ |
| 539 | post \ |
| 540 | : call_clbr \ |
| 541 | : paravirt_type(op), \ |
| 542 | paravirt_clobber(clbr), \ |
| 543 | ##__VA_ARGS__ \ |
| 544 | : "memory", "cc" extra_clbr); \ |
| 545 | __ret = (rettype)((((u64)__edx) << 32) | __eax); \ |
| 546 | } else { \ |
| 547 | asm volatile(pre \ |
| 548 | paravirt_alt(PARAVIRT_CALL) \ |
| 549 | post \ |
| 550 | : call_clbr \ |
| 551 | : paravirt_type(op), \ |
| 552 | paravirt_clobber(clbr), \ |
| 553 | ##__VA_ARGS__ \ |
| 554 | : "memory", "cc" extra_clbr); \ |
| 555 | __ret = (rettype)__eax; \ |
| 556 | } \ |
| 557 | __ret; \ |
| 558 | }) |
| 559 | |
| 560 | #define __PVOP_CALL(rettype, op, pre, post, ...) \ |
| 561 | ____PVOP_CALL(rettype, op, CLBR_ANY, PVOP_CALL_CLOBBERS, \ |
| 562 | EXTRA_CLOBBERS, pre, post, ##__VA_ARGS__) |
| 563 | |
| 564 | #define __PVOP_CALLEESAVE(rettype, op, pre, post, ...) \ |
| 565 | ____PVOP_CALL(rettype, op.func, CLBR_RET_REG, \ |
| 566 | PVOP_CALLEE_CLOBBERS, , \ |
| 567 | pre, post, ##__VA_ARGS__) |
| 568 | |
| 569 | |
| 570 | #define ____PVOP_VCALL(op, clbr, call_clbr, extra_clbr, pre, post, ...) \ |
| 571 | ({ \ |
| 572 | PVOP_VCALL_ARGS; \ |
| 573 | PVOP_TEST_NULL(op); \ |
| 574 | asm volatile(pre \ |
| 575 | paravirt_alt(PARAVIRT_CALL) \ |
| 576 | post \ |
| 577 | : call_clbr \ |
| 578 | : paravirt_type(op), \ |
| 579 | paravirt_clobber(clbr), \ |
| 580 | ##__VA_ARGS__ \ |
| 581 | : "memory", "cc" extra_clbr); \ |
| 582 | }) |
| 583 | |
| 584 | #define __PVOP_VCALL(op, pre, post, ...) \ |
| 585 | ____PVOP_VCALL(op, CLBR_ANY, PVOP_VCALL_CLOBBERS, \ |
| 586 | VEXTRA_CLOBBERS, \ |
| 587 | pre, post, ##__VA_ARGS__) |
| 588 | |
Jeremy Fitzhardinge | 71999d9 | 2009-10-12 16:32:43 -0700 | [diff] [blame] | 589 | #define __PVOP_VCALLEESAVE(op, pre, post, ...) \ |
| 590 | ____PVOP_VCALL(op.func, CLBR_RET_REG, \ |
Jeremy Fitzhardinge | ac5672f | 2009-04-14 14:29:44 -0700 | [diff] [blame] | 591 | PVOP_VCALLEE_CLOBBERS, , \ |
| 592 | pre, post, ##__VA_ARGS__) |
| 593 | |
| 594 | |
| 595 | |
| 596 | #define PVOP_CALL0(rettype, op) \ |
| 597 | __PVOP_CALL(rettype, op, "", "") |
| 598 | #define PVOP_VCALL0(op) \ |
| 599 | __PVOP_VCALL(op, "", "") |
| 600 | |
| 601 | #define PVOP_CALLEE0(rettype, op) \ |
| 602 | __PVOP_CALLEESAVE(rettype, op, "", "") |
| 603 | #define PVOP_VCALLEE0(op) \ |
| 604 | __PVOP_VCALLEESAVE(op, "", "") |
| 605 | |
| 606 | |
| 607 | #define PVOP_CALL1(rettype, op, arg1) \ |
| 608 | __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1)) |
| 609 | #define PVOP_VCALL1(op, arg1) \ |
| 610 | __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1)) |
| 611 | |
| 612 | #define PVOP_CALLEE1(rettype, op, arg1) \ |
| 613 | __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1)) |
| 614 | #define PVOP_VCALLEE1(op, arg1) \ |
| 615 | __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1)) |
| 616 | |
| 617 | |
| 618 | #define PVOP_CALL2(rettype, op, arg1, arg2) \ |
| 619 | __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1), \ |
| 620 | PVOP_CALL_ARG2(arg2)) |
| 621 | #define PVOP_VCALL2(op, arg1, arg2) \ |
| 622 | __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1), \ |
| 623 | PVOP_CALL_ARG2(arg2)) |
| 624 | |
| 625 | #define PVOP_CALLEE2(rettype, op, arg1, arg2) \ |
| 626 | __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1), \ |
| 627 | PVOP_CALL_ARG2(arg2)) |
| 628 | #define PVOP_VCALLEE2(op, arg1, arg2) \ |
| 629 | __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1), \ |
| 630 | PVOP_CALL_ARG2(arg2)) |
| 631 | |
| 632 | |
| 633 | #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \ |
| 634 | __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1), \ |
| 635 | PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3)) |
| 636 | #define PVOP_VCALL3(op, arg1, arg2, arg3) \ |
| 637 | __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1), \ |
| 638 | PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3)) |
| 639 | |
| 640 | /* This is the only difference in x86_64. We can make it much simpler */ |
| 641 | #ifdef CONFIG_X86_32 |
| 642 | #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \ |
| 643 | __PVOP_CALL(rettype, op, \ |
| 644 | "push %[_arg4];", "lea 4(%%esp),%%esp;", \ |
| 645 | PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \ |
| 646 | PVOP_CALL_ARG3(arg3), [_arg4] "mr" ((u32)(arg4))) |
| 647 | #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \ |
| 648 | __PVOP_VCALL(op, \ |
| 649 | "push %[_arg4];", "lea 4(%%esp),%%esp;", \ |
| 650 | "0" ((u32)(arg1)), "1" ((u32)(arg2)), \ |
| 651 | "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4))) |
| 652 | #else |
| 653 | #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \ |
| 654 | __PVOP_CALL(rettype, op, "", "", \ |
| 655 | PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \ |
| 656 | PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4)) |
| 657 | #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \ |
| 658 | __PVOP_VCALL(op, "", "", \ |
| 659 | PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \ |
| 660 | PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4)) |
| 661 | #endif |
| 662 | |
| 663 | /* Lazy mode for batching updates / context switch */ |
| 664 | enum paravirt_lazy_mode { |
| 665 | PARAVIRT_LAZY_NONE, |
| 666 | PARAVIRT_LAZY_MMU, |
| 667 | PARAVIRT_LAZY_CPU, |
| 668 | }; |
| 669 | |
| 670 | enum paravirt_lazy_mode paravirt_get_lazy_mode(void); |
| 671 | void paravirt_start_context_switch(struct task_struct *prev); |
| 672 | void paravirt_end_context_switch(struct task_struct *next); |
| 673 | |
| 674 | void paravirt_enter_lazy_mmu(void); |
| 675 | void paravirt_leave_lazy_mmu(void); |
| 676 | |
| 677 | void _paravirt_nop(void); |
| 678 | u32 _paravirt_ident_32(u32); |
| 679 | u64 _paravirt_ident_64(u64); |
| 680 | |
| 681 | #define paravirt_nop ((void *)_paravirt_nop) |
| 682 | |
| 683 | /* These all sit in the .parainstructions section to tell us what to patch. */ |
| 684 | struct paravirt_patch_site { |
| 685 | u8 *instr; /* original instructions */ |
| 686 | u8 instrtype; /* type of this instruction */ |
| 687 | u8 len; /* length of original instruction */ |
| 688 | u16 clobbers; /* what registers you may clobber */ |
| 689 | }; |
| 690 | |
| 691 | extern struct paravirt_patch_site __parainstructions[], |
| 692 | __parainstructions_end[]; |
| 693 | |
| 694 | #endif /* __ASSEMBLY__ */ |
| 695 | |
| 696 | #endif /* _ASM_X86_PARAVIRT_TYPES_H */ |