?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
conn rcv_lock converted to spinlock, struct cor_sock created, kernel_packet skb_clone...
[cor_2_6_31.git] / arch / x86 / include / asm / paravirt.h
blob4fb37c8a0832da8cf51f7a2c771c54fa97af2708
1 #ifndef _ASM_X86_PARAVIRT_H
2 #define _ASM_X86_PARAVIRT_H
3 /* Various instructions on x86 need to be replaced for
4 * para-virtualization: those hooks are defined here. */
5
6 #ifdef CONFIG_PARAVIRT
7 #include <asm/pgtable_types.h>
8 #include <asm/asm.h>
9
10 /* Bitmask of what can be clobbered: usually at least eax. */
11 #define CLBR_NONE 0
12 #define CLBR_EAX (1 << 0)
13 #define CLBR_ECX (1 << 1)
14 #define CLBR_EDX (1 << 2)
15 #define CLBR_EDI (1 << 3)
16
17 #ifdef CONFIG_X86_32
18 /* CLBR_ANY should match all regs platform has. For i386, that's just it */
19 #define CLBR_ANY ((1 << 4) - 1)
20
21 #define CLBR_ARG_REGS (CLBR_EAX | CLBR_EDX | CLBR_ECX)
22 #define CLBR_RET_REG (CLBR_EAX | CLBR_EDX)
23 #define CLBR_SCRATCH (0)
24 #else
25 #define CLBR_RAX CLBR_EAX
26 #define CLBR_RCX CLBR_ECX
27 #define CLBR_RDX CLBR_EDX
28 #define CLBR_RDI CLBR_EDI
29 #define CLBR_RSI (1 << 4)
30 #define CLBR_R8 (1 << 5)
31 #define CLBR_R9 (1 << 6)
32 #define CLBR_R10 (1 << 7)
33 #define CLBR_R11 (1 << 8)
34
35 #define CLBR_ANY ((1 << 9) - 1)
36
37 #define CLBR_ARG_REGS (CLBR_RDI | CLBR_RSI | CLBR_RDX | \
38 CLBR_RCX | CLBR_R8 | CLBR_R9)
39 #define CLBR_RET_REG (CLBR_RAX)
40 #define CLBR_SCRATCH (CLBR_R10 | CLBR_R11)
41
42 #include <asm/desc_defs.h>
43 #endif /* X86_64 */
44
45 #define CLBR_CALLEE_SAVE ((CLBR_ARG_REGS | CLBR_SCRATCH) & ~CLBR_RET_REG)
46
47 #ifndef __ASSEMBLY__
48 #include <linux/types.h>
49 #include <linux/cpumask.h>
50 #include <asm/kmap_types.h>
51 #include <asm/desc_defs.h>
52
53 struct page;
54 struct thread_struct;
55 struct desc_ptr;
56 struct tss_struct;
57 struct mm_struct;
58 struct desc_struct;
59 struct task_struct;
60
61 /*
62 * Wrapper type for pointers to code which uses the non-standard
63 * calling convention. See PV_CALL_SAVE_REGS_THUNK below.
64 */
65 struct paravirt_callee_save {
66 void *func;
67 };
68
69 /* general info */
70 struct pv_info {
71 unsigned int kernel_rpl;
72 int shared_kernel_pmd;
73 int paravirt_enabled;
74 const char *name;
75 };
76
77 struct pv_init_ops {
78 /*
79 * Patch may replace one of the defined code sequences with
80 * arbitrary code, subject to the same register constraints.
81 * This generally means the code is not free to clobber any
82 * registers other than EAX. The patch function should return
83 * the number of bytes of code generated, as we nop pad the
84 * rest in generic code.
85 */
86 unsigned (*patch)(u8 type, u16 clobber, void *insnbuf,
87 unsigned long addr, unsigned len);
88
89 /* Basic arch-specific setup */
90 void (*arch_setup)(void);
91 char *(*memory_setup)(void);
92 void (*post_allocator_init)(void);
93
94 /* Print a banner to identify the environment */
95 void (*banner)(void);
96 };
97
98
99 struct pv_lazy_ops {
100 /* Set deferred update mode, used for batching operations. */
101 void (*enter)(void);
102 void (*leave)(void);
103 };
104
105 struct pv_time_ops {
106 void (*time_init)(void);
107
108 /* Set and set time of day */
109 unsigned long (*get_wallclock)(void);
110 int (*set_wallclock)(unsigned long);
111
112 unsigned long long (*sched_clock)(void);
113 unsigned long (*get_tsc_khz)(void);
114 };
115
116 struct pv_cpu_ops {
117 /* hooks for various privileged instructions */
118 unsigned long (*get_debugreg)(int regno);
119 void (*set_debugreg)(int regno, unsigned long value);
120
121 void (*clts)(void);
122
123 unsigned long (*read_cr0)(void);
124 void (*write_cr0)(unsigned long);
125
126 unsigned long (*read_cr4_safe)(void);
127 unsigned long (*read_cr4)(void);
128 void (*write_cr4)(unsigned long);
129
130 #ifdef CONFIG_X86_64
131 unsigned long (*read_cr8)(void);
132 void (*write_cr8)(unsigned long);
133 #endif
134
135 /* Segment descriptor handling */
136 void (*load_tr_desc)(void);
137 void (*load_gdt)(const struct desc_ptr *);
138 void (*load_idt)(const struct desc_ptr *);
139 void (*store_gdt)(struct desc_ptr *);
140 void (*store_idt)(struct desc_ptr *);
141 void (*set_ldt)(const void *desc, unsigned entries);
142 unsigned long (*store_tr)(void);
143 void (*load_tls)(struct thread_struct *t, unsigned int cpu);
144 #ifdef CONFIG_X86_64
145 void (*load_gs_index)(unsigned int idx);
146 #endif
147 void (*write_ldt_entry)(struct desc_struct *ldt, int entrynum,
148 const void *desc);
149 void (*write_gdt_entry)(struct desc_struct *,
150 int entrynum, const void *desc, int size);
151 void (*write_idt_entry)(gate_desc *,
152 int entrynum, const gate_desc *gate);
153 void (*alloc_ldt)(struct desc_struct *ldt, unsigned entries);
154 void (*free_ldt)(struct desc_struct *ldt, unsigned entries);
155
156 void (*load_sp0)(struct tss_struct *tss, struct thread_struct *t);
157
158 void (*set_iopl_mask)(unsigned mask);
159
160 void (*wbinvd)(void);
161 void (*io_delay)(void);
162
163 /* cpuid emulation, mostly so that caps bits can be disabled */
164 void (*cpuid)(unsigned int *eax, unsigned int *ebx,
165 unsigned int *ecx, unsigned int *edx);
166
167 /* MSR, PMC and TSR operations.
168 err = 0/-EFAULT. wrmsr returns 0/-EFAULT. */
169 u64 (*read_msr_amd)(unsigned int msr, int *err);
170 u64 (*read_msr)(unsigned int msr, int *err);
171 int (*write_msr)(unsigned int msr, unsigned low, unsigned high);
172
173 u64 (*read_tsc)(void);
174 u64 (*read_pmc)(int counter);
175 unsigned long long (*read_tscp)(unsigned int *aux);
176
177 /*
178 * Atomically enable interrupts and return to userspace. This
179 * is only ever used to return to 32-bit processes; in a
180 * 64-bit kernel, it's used for 32-on-64 compat processes, but
181 * never native 64-bit processes. (Jump, not call.)
182 */
183 void (*irq_enable_sysexit)(void);
184
185 /*
186 * Switch to usermode gs and return to 64-bit usermode using
187 * sysret. Only used in 64-bit kernels to return to 64-bit
188 * processes. Usermode register state, including %rsp, must
189 * already be restored.
190 */
191 void (*usergs_sysret64)(void);
192
193 /*
194 * Switch to usermode gs and return to 32-bit usermode using
195 * sysret. Used to return to 32-on-64 compat processes.
196 * Other usermode register state, including %esp, must already
197 * be restored.
198 */
199 void (*usergs_sysret32)(void);
200
201 /* Normal iret. Jump to this with the standard iret stack
202 frame set up. */
203 void (*iret)(void);
204
205 void (*swapgs)(void);
206
207 void (*start_context_switch)(struct task_struct *prev);
208 void (*end_context_switch)(struct task_struct *next);
209 };
210
211 struct pv_irq_ops {
212 void (*init_IRQ)(void);
213
214 /*
215 * Get/set interrupt state. save_fl and restore_fl are only
216 * expected to use X86_EFLAGS_IF; all other bits
217 * returned from save_fl are undefined, and may be ignored by
218 * restore_fl.
219 *
220 * NOTE: These functions callers expect the callee to preserve
221 * more registers than the standard C calling convention.
222 */
223 struct paravirt_callee_save save_fl;
224 struct paravirt_callee_save restore_fl;
225 struct paravirt_callee_save irq_disable;
226 struct paravirt_callee_save irq_enable;
227
228 void (*safe_halt)(void);
229 void (*halt)(void);
230
231 #ifdef CONFIG_X86_64
232 void (*adjust_exception_frame)(void);
233 #endif
234 };
235
236 struct pv_apic_ops {
237 #ifdef CONFIG_X86_LOCAL_APIC
238 void (*setup_boot_clock)(void);
239 void (*setup_secondary_clock)(void);
240
241 void (*startup_ipi_hook)(int phys_apicid,
242 unsigned long start_eip,
243 unsigned long start_esp);
244 #endif
245 };
246
247 struct pv_mmu_ops {
248 /*
249 * Called before/after init_mm pagetable setup. setup_start
250 * may reset %cr3, and may pre-install parts of the pagetable;
251 * pagetable setup is expected to preserve any existing
252 * mapping.
253 */
254 void (*pagetable_setup_start)(pgd_t *pgd_base);
255 void (*pagetable_setup_done)(pgd_t *pgd_base);
256
257 unsigned long (*read_cr2)(void);
258 void (*write_cr2)(unsigned long);
259
260 unsigned long (*read_cr3)(void);
261 void (*write_cr3)(unsigned long);
262
263 /*
264 * Hooks for intercepting the creation/use/destruction of an
265 * mm_struct.
266 */
267 void (*activate_mm)(struct mm_struct *prev,
268 struct mm_struct *next);
269 void (*dup_mmap)(struct mm_struct *oldmm,
270 struct mm_struct *mm);
271 void (*exit_mmap)(struct mm_struct *mm);
272
273
274 /* TLB operations */
275 void (*flush_tlb_user)(void);
276 void (*flush_tlb_kernel)(void);
277 void (*flush_tlb_single)(unsigned long addr);
278 void (*flush_tlb_others)(const struct cpumask *cpus,
279 struct mm_struct *mm,
280 unsigned long va);
281
282 /* Hooks for allocating and freeing a pagetable top-level */
283 int (*pgd_alloc)(struct mm_struct *mm);
284 void (*pgd_free)(struct mm_struct *mm, pgd_t *pgd);
285
286 /*
287 * Hooks for allocating/releasing pagetable pages when they're
288 * attached to a pagetable
289 */
290 void (*alloc_pte)(struct mm_struct *mm, unsigned long pfn);
291 void (*alloc_pmd)(struct mm_struct *mm, unsigned long pfn);
292 void (*alloc_pmd_clone)(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count);
293 void (*alloc_pud)(struct mm_struct *mm, unsigned long pfn);
294 void (*release_pte)(unsigned long pfn);
295 void (*release_pmd)(unsigned long pfn);
296 void (*release_pud)(unsigned long pfn);
297
298 /* Pagetable manipulation functions */
299 void (*set_pte)(pte_t *ptep, pte_t pteval);
300 void (*set_pte_at)(struct mm_struct *mm, unsigned long addr,
301 pte_t *ptep, pte_t pteval);
302 void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval);
303 void (*pte_update)(struct mm_struct *mm, unsigned long addr,
304 pte_t *ptep);
305 void (*pte_update_defer)(struct mm_struct *mm,
306 unsigned long addr, pte_t *ptep);
307
308 pte_t (*ptep_modify_prot_start)(struct mm_struct *mm, unsigned long addr,
309 pte_t *ptep);
310 void (*ptep_modify_prot_commit)(struct mm_struct *mm, unsigned long addr,
311 pte_t *ptep, pte_t pte);
312
313 struct paravirt_callee_save pte_val;
314 struct paravirt_callee_save make_pte;
315
316 struct paravirt_callee_save pgd_val;
317 struct paravirt_callee_save make_pgd;
318
319 #if PAGETABLE_LEVELS >= 3
320 #ifdef CONFIG_X86_PAE
321 void (*set_pte_atomic)(pte_t *ptep, pte_t pteval);
322 void (*pte_clear)(struct mm_struct *mm, unsigned long addr,
323 pte_t *ptep);
324 void (*pmd_clear)(pmd_t *pmdp);
325
326 #endif /* CONFIG_X86_PAE */
327
328 void (*set_pud)(pud_t *pudp, pud_t pudval);
329
330 struct paravirt_callee_save pmd_val;
331 struct paravirt_callee_save make_pmd;
332
333 #if PAGETABLE_LEVELS == 4
334 struct paravirt_callee_save pud_val;
335 struct paravirt_callee_save make_pud;
336
337 void (*set_pgd)(pgd_t *pudp, pgd_t pgdval);
338 #endif /* PAGETABLE_LEVELS == 4 */
339 #endif /* PAGETABLE_LEVELS >= 3 */
340
341 #ifdef CONFIG_HIGHPTE
342 void *(*kmap_atomic_pte)(struct page *page, enum km_type type);
343 #endif
344
345 struct pv_lazy_ops lazy_mode;
346
347 /* dom0 ops */
348
349 /* Sometimes the physical address is a pfn, and sometimes its
350 an mfn. We can tell which is which from the index. */
351 void (*set_fixmap)(unsigned /* enum fixed_addresses */ idx,
352 phys_addr_t phys, pgprot_t flags);
353 };
354
355 struct raw_spinlock;
356 struct pv_lock_ops {
357 int (*spin_is_locked)(struct raw_spinlock *lock);
358 int (*spin_is_contended)(struct raw_spinlock *lock);
359 void (*spin_lock)(struct raw_spinlock *lock);
360 void (*spin_lock_flags)(struct raw_spinlock *lock, unsigned long flags);
361 int (*spin_trylock)(struct raw_spinlock *lock);
362 void (*spin_unlock)(struct raw_spinlock *lock);
363 };
364
365 /* This contains all the paravirt structures: we get a convenient
366 * number for each function using the offset which we use to indicate
367 * what to patch. */
368 struct paravirt_patch_template {
369 struct pv_init_ops pv_init_ops;
370 struct pv_time_ops pv_time_ops;
371 struct pv_cpu_ops pv_cpu_ops;
372 struct pv_irq_ops pv_irq_ops;
373 struct pv_apic_ops pv_apic_ops;
374 struct pv_mmu_ops pv_mmu_ops;
375 struct pv_lock_ops pv_lock_ops;
376 };
377
378 extern struct pv_info pv_info;
379 extern struct pv_init_ops pv_init_ops;
380 extern struct pv_time_ops pv_time_ops;
381 extern struct pv_cpu_ops pv_cpu_ops;
382 extern struct pv_irq_ops pv_irq_ops;
383 extern struct pv_apic_ops pv_apic_ops;
384 extern struct pv_mmu_ops pv_mmu_ops;
385 extern struct pv_lock_ops pv_lock_ops;
386
387 #define PARAVIRT_PATCH(x) \
388 (offsetof(struct paravirt_patch_template, x) / sizeof(void *))
389
390 #define paravirt_type(op) \
391 [paravirt_typenum] "i" (PARAVIRT_PATCH(op)), \
392 [paravirt_opptr] "i" (&(op))
393 #define paravirt_clobber(clobber) \
394 [paravirt_clobber] "i" (clobber)
395
396 /*
397 * Generate some code, and mark it as patchable by the
398 * apply_paravirt() alternate instruction patcher.
399 */
400 #define _paravirt_alt(insn_string, type, clobber) \
401 "771:\n\t" insn_string "\n" "772:\n" \
402 ".pushsection .parainstructions,\"a\"\n" \
403 _ASM_ALIGN "\n" \
404 _ASM_PTR " 771b\n" \
405 " .byte " type "\n" \
406 " .byte 772b-771b\n" \
407 " .short " clobber "\n" \
408 ".popsection\n"
409
410 /* Generate patchable code, with the default asm parameters. */
411 #define paravirt_alt(insn_string) \
412 _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
413
414 /* Simple instruction patching code. */
415 #define DEF_NATIVE(ops, name, code) \
416 extern const char start_##ops##_##name[], end_##ops##_##name[]; \
417 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
418
419 unsigned paravirt_patch_nop(void);
420 unsigned paravirt_patch_ident_32(void *insnbuf, unsigned len);
421 unsigned paravirt_patch_ident_64(void *insnbuf, unsigned len);
422 unsigned paravirt_patch_ignore(unsigned len);
423 unsigned paravirt_patch_call(void *insnbuf,
424 const void *target, u16 tgt_clobbers,
425 unsigned long addr, u16 site_clobbers,
426 unsigned len);
427 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
428 unsigned long addr, unsigned len);
429 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
430 unsigned long addr, unsigned len);
431
432 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
433 const char *start, const char *end);
434
435 unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
436 unsigned long addr, unsigned len);
437
438 int paravirt_disable_iospace(void);
439
440 /*
441 * This generates an indirect call based on the operation type number.
442 * The type number, computed in PARAVIRT_PATCH, is derived from the
443 * offset into the paravirt_patch_template structure, and can therefore be
444 * freely converted back into a structure offset.
445 */
446 #define PARAVIRT_CALL "call *%c[paravirt_opptr];"
447
448 /*
449 * These macros are intended to wrap calls through one of the paravirt
450 * ops structs, so that they can be later identified and patched at
451 * runtime.
452 *
453 * Normally, a call to a pv_op function is a simple indirect call:
454 * (pv_op_struct.operations)(args...).
455 *
456 * Unfortunately, this is a relatively slow operation for modern CPUs,
457 * because it cannot necessarily determine what the destination
458 * address is. In this case, the address is a runtime constant, so at
459 * the very least we can patch the call to e a simple direct call, or
460 * ideally, patch an inline implementation into the callsite. (Direct
461 * calls are essentially free, because the call and return addresses
462 * are completely predictable.)
463 *
464 * For i386, these macros rely on the standard gcc "regparm(3)" calling
465 * convention, in which the first three arguments are placed in %eax,
466 * %edx, %ecx (in that order), and the remaining arguments are placed
467 * on the stack. All caller-save registers (eax,edx,ecx) are expected
468 * to be modified (either clobbered or used for return values).
469 * X86_64, on the other hand, already specifies a register-based calling
470 * conventions, returning at %rax, with parameteres going on %rdi, %rsi,
471 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
472 * special handling for dealing with 4 arguments, unlike i386.
473 * However, x86_64 also have to clobber all caller saved registers, which
474 * unfortunately, are quite a bit (r8 - r11)
475 *
476 * The call instruction itself is marked by placing its start address
477 * and size into the .parainstructions section, so that
478 * apply_paravirt() in arch/i386/kernel/alternative.c can do the
479 * appropriate patching under the control of the backend pv_init_ops
480 * implementation.
481 *
482 * Unfortunately there's no way to get gcc to generate the args setup
483 * for the call, and then allow the call itself to be generated by an
484 * inline asm. Because of this, we must do the complete arg setup and
485 * return value handling from within these macros. This is fairly
486 * cumbersome.
487 *
488 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
489 * It could be extended to more arguments, but there would be little
490 * to be gained from that. For each number of arguments, there are
491 * the two VCALL and CALL variants for void and non-void functions.
492 *
493 * When there is a return value, the invoker of the macro must specify
494 * the return type. The macro then uses sizeof() on that type to
495 * determine whether its a 32 or 64 bit value, and places the return
496 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
497 * 64-bit). For x86_64 machines, it just returns at %rax regardless of
498 * the return value size.
499 *
500 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
501 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
502 * in low,high order
503 *
504 * Small structures are passed and returned in registers. The macro
505 * calling convention can't directly deal with this, so the wrapper
506 * functions must do this.
507 *
508 * These PVOP_* macros are only defined within this header. This
509 * means that all uses must be wrapped in inline functions. This also
510 * makes sure the incoming and outgoing types are always correct.
511 */
512 #ifdef CONFIG_X86_32
513 #define PVOP_VCALL_ARGS \
514 unsigned long __eax = __eax, __edx = __edx, __ecx = __ecx
515 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS
516
517 #define PVOP_CALL_ARG1(x) "a" ((unsigned long)(x))
518 #define PVOP_CALL_ARG2(x) "d" ((unsigned long)(x))
519 #define PVOP_CALL_ARG3(x) "c" ((unsigned long)(x))
520
521 #define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \
522 "=c" (__ecx)
523 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS
524
525 #define PVOP_VCALLEE_CLOBBERS "=a" (__eax), "=d" (__edx)
526 #define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS
527
528 #define EXTRA_CLOBBERS
529 #define VEXTRA_CLOBBERS
530 #else /* CONFIG_X86_64 */
531 #define PVOP_VCALL_ARGS \
532 unsigned long __edi = __edi, __esi = __esi, \
533 __edx = __edx, __ecx = __ecx
534 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS, __eax
535
536 #define PVOP_CALL_ARG1(x) "D" ((unsigned long)(x))
537 #define PVOP_CALL_ARG2(x) "S" ((unsigned long)(x))
538 #define PVOP_CALL_ARG3(x) "d" ((unsigned long)(x))
539 #define PVOP_CALL_ARG4(x) "c" ((unsigned long)(x))
540
541 #define PVOP_VCALL_CLOBBERS "=D" (__edi), \
542 "=S" (__esi), "=d" (__edx), \
543 "=c" (__ecx)
544 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax)
545
546 #define PVOP_VCALLEE_CLOBBERS "=a" (__eax)
547 #define PVOP_CALLEE_CLOBBERS PVOP_VCALLEE_CLOBBERS
548
549 #define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11"
550 #define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11"
551 #endif /* CONFIG_X86_32 */
552
553 #ifdef CONFIG_PARAVIRT_DEBUG
554 #define PVOP_TEST_NULL(op) BUG_ON(op == NULL)
555 #else
556 #define PVOP_TEST_NULL(op) ((void)op)
557 #endif
558
559 #define ____PVOP_CALL(rettype, op, clbr, call_clbr, extra_clbr, \
560 pre, post, ...) \
561 ({ \
562 rettype __ret; \
563 PVOP_CALL_ARGS; \
564 PVOP_TEST_NULL(op); \
565 /* This is 32-bit specific, but is okay in 64-bit */ \
566 /* since this condition will never hold */ \
567 if (sizeof(rettype) > sizeof(unsigned long)) { \
568 asm volatile(pre \
569 paravirt_alt(PARAVIRT_CALL) \
570 post \
571 : call_clbr \
572 : paravirt_type(op), \
573 paravirt_clobber(clbr), \
574 ##__VA_ARGS__ \
575 : "memory", "cc" extra_clbr); \
576 __ret = (rettype)((((u64)__edx) << 32) | __eax); \
577 } else { \
578 asm volatile(pre \
579 paravirt_alt(PARAVIRT_CALL) \
580 post \
581 : call_clbr \
582 : paravirt_type(op), \
583 paravirt_clobber(clbr), \
584 ##__VA_ARGS__ \
585 : "memory", "cc" extra_clbr); \
586 __ret = (rettype)__eax; \
587 } \
588 __ret; \
589 })
590
591 #define __PVOP_CALL(rettype, op, pre, post, ...) \
592 ____PVOP_CALL(rettype, op, CLBR_ANY, PVOP_CALL_CLOBBERS, \
593 EXTRA_CLOBBERS, pre, post, ##__VA_ARGS__)
594
595 #define __PVOP_CALLEESAVE(rettype, op, pre, post, ...) \
596 ____PVOP_CALL(rettype, op.func, CLBR_RET_REG, \
597 PVOP_CALLEE_CLOBBERS, , \
598 pre, post, ##__VA_ARGS__)
599
600
601 #define ____PVOP_VCALL(op, clbr, call_clbr, extra_clbr, pre, post, ...) \
602 ({ \
603 PVOP_VCALL_ARGS; \
604 PVOP_TEST_NULL(op); \
605 asm volatile(pre \
606 paravirt_alt(PARAVIRT_CALL) \
607 post \
608 : call_clbr \
609 : paravirt_type(op), \
610 paravirt_clobber(clbr), \
611 ##__VA_ARGS__ \
612 : "memory", "cc" extra_clbr); \
613 })
614
615 #define __PVOP_VCALL(op, pre, post, ...) \
616 ____PVOP_VCALL(op, CLBR_ANY, PVOP_VCALL_CLOBBERS, \
617 VEXTRA_CLOBBERS, \
618 pre, post, ##__VA_ARGS__)
619
620 #define __PVOP_VCALLEESAVE(rettype, op, pre, post, ...) \
621 ____PVOP_CALL(rettype, op.func, CLBR_RET_REG, \
622 PVOP_VCALLEE_CLOBBERS, , \
623 pre, post, ##__VA_ARGS__)
624
625
626
627 #define PVOP_CALL0(rettype, op) \
628 __PVOP_CALL(rettype, op, "", "")
629 #define PVOP_VCALL0(op) \
630 __PVOP_VCALL(op, "", "")
631
632 #define PVOP_CALLEE0(rettype, op) \
633 __PVOP_CALLEESAVE(rettype, op, "", "")
634 #define PVOP_VCALLEE0(op) \
635 __PVOP_VCALLEESAVE(op, "", "")
636
637
638 #define PVOP_CALL1(rettype, op, arg1) \
639 __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1))
640 #define PVOP_VCALL1(op, arg1) \
641 __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1))
642
643 #define PVOP_CALLEE1(rettype, op, arg1) \
644 __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1))
645 #define PVOP_VCALLEE1(op, arg1) \
646 __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1))
647
648
649 #define PVOP_CALL2(rettype, op, arg1, arg2) \
650 __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1), \
651 PVOP_CALL_ARG2(arg2))
652 #define PVOP_VCALL2(op, arg1, arg2) \
653 __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1), \
654 PVOP_CALL_ARG2(arg2))
655
656 #define PVOP_CALLEE2(rettype, op, arg1, arg2) \
657 __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1), \
658 PVOP_CALL_ARG2(arg2))
659 #define PVOP_VCALLEE2(op, arg1, arg2) \
660 __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1), \
661 PVOP_CALL_ARG2(arg2))
662
663
664 #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \
665 __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1), \
666 PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
667 #define PVOP_VCALL3(op, arg1, arg2, arg3) \
668 __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1), \
669 PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
670
671 /* This is the only difference in x86_64. We can make it much simpler */
672 #ifdef CONFIG_X86_32
673 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
674 __PVOP_CALL(rettype, op, \
675 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
676 PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \
677 PVOP_CALL_ARG3(arg3), [_arg4] "mr" ((u32)(arg4)))
678 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
679 __PVOP_VCALL(op, \
680 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
681 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
682 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
683 #else
684 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
685 __PVOP_CALL(rettype, op, "", "", \
686 PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \
687 PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
688 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
689 __PVOP_VCALL(op, "", "", \
690 PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2), \
691 PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
692 #endif
693
694 static inline int paravirt_enabled(void)
695 {
696 return pv_info.paravirt_enabled;
697 }
698
699 static inline void load_sp0(struct tss_struct *tss,
700 struct thread_struct *thread)
701 {
702 PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
703 }
704
705 #define ARCH_SETUP pv_init_ops.arch_setup();
706 static inline unsigned long get_wallclock(void)
707 {
708 return PVOP_CALL0(unsigned long, pv_time_ops.get_wallclock);
709 }
710
711 static inline int set_wallclock(unsigned long nowtime)
712 {
713 return PVOP_CALL1(int, pv_time_ops.set_wallclock, nowtime);
714 }
715
716 static inline void (*choose_time_init(void))(void)
717 {
718 return pv_time_ops.time_init;
719 }
720
721 /* The paravirtualized CPUID instruction. */
722 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
723 unsigned int *ecx, unsigned int *edx)
724 {
725 PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
726 }
727
728 /*
729 * These special macros can be used to get or set a debugging register
730 */
731 static inline unsigned long paravirt_get_debugreg(int reg)
732 {
733 return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
734 }
735 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
736 static inline void set_debugreg(unsigned long val, int reg)
737 {
738 PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
739 }
740
741 static inline void clts(void)
742 {
743 PVOP_VCALL0(pv_cpu_ops.clts);
744 }
745
746 static inline unsigned long read_cr0(void)
747 {
748 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
749 }
750
751 static inline void write_cr0(unsigned long x)
752 {
753 PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
754 }
755
756 static inline unsigned long read_cr2(void)
757 {
758 return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
759 }
760
761 static inline void write_cr2(unsigned long x)
762 {
763 PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
764 }
765
766 static inline unsigned long read_cr3(void)
767 {
768 return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
769 }
770
771 static inline void write_cr3(unsigned long x)
772 {
773 PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
774 }
775
776 static inline unsigned long read_cr4(void)
777 {
778 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
779 }
780 static inline unsigned long read_cr4_safe(void)
781 {
782 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
783 }
784
785 static inline void write_cr4(unsigned long x)
786 {
787 PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
788 }
789
790 #ifdef CONFIG_X86_64
791 static inline unsigned long read_cr8(void)
792 {
793 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
794 }
795
796 static inline void write_cr8(unsigned long x)
797 {
798 PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
799 }
800 #endif
801
802 static inline void raw_safe_halt(void)
803 {
804 PVOP_VCALL0(pv_irq_ops.safe_halt);
805 }
806
807 static inline void halt(void)
808 {
809 PVOP_VCALL0(pv_irq_ops.safe_halt);
810 }
811
812 static inline void wbinvd(void)
813 {
814 PVOP_VCALL0(pv_cpu_ops.wbinvd);
815 }
816
817 #define get_kernel_rpl() (pv_info.kernel_rpl)
818
819 static inline u64 paravirt_read_msr(unsigned msr, int *err)
820 {
821 return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
822 }
823 static inline u64 paravirt_read_msr_amd(unsigned msr, int *err)
824 {
825 return PVOP_CALL2(u64, pv_cpu_ops.read_msr_amd, msr, err);
826 }
827 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
828 {
829 return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
830 }
831
832 /* These should all do BUG_ON(_err), but our headers are too tangled. */
833 #define rdmsr(msr, val1, val2) \
834 do { \
835 int _err; \
836 u64 _l = paravirt_read_msr(msr, &_err); \
837 val1 = (u32)_l; \
838 val2 = _l >> 32; \
839 } while (0)
840
841 #define wrmsr(msr, val1, val2) \
842 do { \
843 paravirt_write_msr(msr, val1, val2); \
844 } while (0)
845
846 #define rdmsrl(msr, val) \
847 do { \
848 int _err; \
849 val = paravirt_read_msr(msr, &_err); \
850 } while (0)
851
852 #define wrmsrl(msr, val) wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
853 #define wrmsr_safe(msr, a, b) paravirt_write_msr(msr, a, b)
854
855 /* rdmsr with exception handling */
856 #define rdmsr_safe(msr, a, b) \
857 ({ \
858 int _err; \
859 u64 _l = paravirt_read_msr(msr, &_err); \
860 (*a) = (u32)_l; \
861 (*b) = _l >> 32; \
862 _err; \
863 })
864
865 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
866 {
867 int err;
868
869 *p = paravirt_read_msr(msr, &err);
870 return err;
871 }
872 static inline int rdmsrl_amd_safe(unsigned msr, unsigned long long *p)
873 {
874 int err;
875
876 *p = paravirt_read_msr_amd(msr, &err);
877 return err;
878 }
879
880 static inline u64 paravirt_read_tsc(void)
881 {
882 return PVOP_CALL0(u64, pv_cpu_ops.read_tsc);
883 }
884
885 #define rdtscl(low) \
886 do { \
887 u64 _l = paravirt_read_tsc(); \
888 low = (int)_l; \
889 } while (0)
890
891 #define rdtscll(val) (val = paravirt_read_tsc())
892
893 static inline unsigned long long paravirt_sched_clock(void)
894 {
895 return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
896 }
897 #define calibrate_tsc() (pv_time_ops.get_tsc_khz())
898
899 static inline unsigned long long paravirt_read_pmc(int counter)
900 {
901 return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
902 }
903
904 #define rdpmc(counter, low, high) \
905 do { \
906 u64 _l = paravirt_read_pmc(counter); \
907 low = (u32)_l; \
908 high = _l >> 32; \
909 } while (0)
910
911 static inline unsigned long long paravirt_rdtscp(unsigned int *aux)
912 {
913 return PVOP_CALL1(u64, pv_cpu_ops.read_tscp, aux);
914 }
915
916 #define rdtscp(low, high, aux) \
917 do { \
918 int __aux; \
919 unsigned long __val = paravirt_rdtscp(&__aux); \
920 (low) = (u32)__val; \
921 (high) = (u32)(__val >> 32); \
922 (aux) = __aux; \
923 } while (0)
924
925 #define rdtscpll(val, aux) \
926 do { \
927 unsigned long __aux; \
928 val = paravirt_rdtscp(&__aux); \
929 (aux) = __aux; \
930 } while (0)
931
932 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
933 {
934 PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
935 }
936
937 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
938 {
939 PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
940 }
941
942 static inline void load_TR_desc(void)
943 {
944 PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
945 }
946 static inline void load_gdt(const struct desc_ptr *dtr)
947 {
948 PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
949 }
950 static inline void load_idt(const struct desc_ptr *dtr)
951 {
952 PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
953 }
954 static inline void set_ldt(const void *addr, unsigned entries)
955 {
956 PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
957 }
958 static inline void store_gdt(struct desc_ptr *dtr)
959 {
960 PVOP_VCALL1(pv_cpu_ops.store_gdt, dtr);
961 }
962 static inline void store_idt(struct desc_ptr *dtr)
963 {
964 PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
965 }
966 static inline unsigned long paravirt_store_tr(void)
967 {
968 return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
969 }
970 #define store_tr(tr) ((tr) = paravirt_store_tr())
971 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
972 {
973 PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
974 }
975
976 #ifdef CONFIG_X86_64
977 static inline void load_gs_index(unsigned int gs)
978 {
979 PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
980 }
981 #endif
982
983 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
984 const void *desc)
985 {
986 PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
987 }
988
989 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
990 void *desc, int type)
991 {
992 PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
993 }
994
995 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
996 {
997 PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
998 }
999 static inline void set_iopl_mask(unsigned mask)
1000 {
1001 PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
1002 }
1003
1004 /* The paravirtualized I/O functions */
1005 static inline void slow_down_io(void)
1006 {
1007 pv_cpu_ops.io_delay();
1008 #ifdef REALLY_SLOW_IO
1009 pv_cpu_ops.io_delay();
1010 pv_cpu_ops.io_delay();
1011 pv_cpu_ops.io_delay();
1012 #endif
1013 }
1014
1015 #ifdef CONFIG_X86_LOCAL_APIC
1016 static inline void setup_boot_clock(void)
1017 {
1018 PVOP_VCALL0(pv_apic_ops.setup_boot_clock);
1019 }
1020
1021 static inline void setup_secondary_clock(void)
1022 {
1023 PVOP_VCALL0(pv_apic_ops.setup_secondary_clock);
1024 }
1025 #endif
1026
1027 static inline void paravirt_post_allocator_init(void)
1028 {
1029 if (pv_init_ops.post_allocator_init)
1030 (*pv_init_ops.post_allocator_init)();
1031 }
1032
1033 static inline void paravirt_pagetable_setup_start(pgd_t *base)
1034 {
1035 (*pv_mmu_ops.pagetable_setup_start)(base);
1036 }
1037
1038 static inline void paravirt_pagetable_setup_done(pgd_t *base)
1039 {
1040 (*pv_mmu_ops.pagetable_setup_done)(base);
1041 }
1042
1043 #ifdef CONFIG_SMP
1044 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
1045 unsigned long start_esp)
1046 {
1047 PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
1048 phys_apicid, start_eip, start_esp);
1049 }
1050 #endif
1051
1052 static inline void paravirt_activate_mm(struct mm_struct *prev,
1053 struct mm_struct *next)
1054 {
1055 PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
1056 }
1057
1058 static inline void arch_dup_mmap(struct mm_struct *oldmm,
1059 struct mm_struct *mm)
1060 {
1061 PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
1062 }
1063
1064 static inline void arch_exit_mmap(struct mm_struct *mm)
1065 {
1066 PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
1067 }
1068
1069 static inline void __flush_tlb(void)
1070 {
1071 PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
1072 }
1073 static inline void __flush_tlb_global(void)
1074 {
1075 PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
1076 }
1077 static inline void __flush_tlb_single(unsigned long addr)
1078 {
1079 PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
1080 }
1081
1082 static inline void flush_tlb_others(const struct cpumask *cpumask,
1083 struct mm_struct *mm,
1084 unsigned long va)
1085 {
1086 PVOP_VCALL3(pv_mmu_ops.flush_tlb_others, cpumask, mm, va);
1087 }
1088
1089 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
1090 {
1091 return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
1092 }
1093
1094 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1095 {
1096 PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
1097 }
1098
1099 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1100 {
1101 PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
1102 }
1103 static inline void paravirt_release_pte(unsigned long pfn)
1104 {
1105 PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
1106 }
1107
1108 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1109 {
1110 PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
1111 }
1112
1113 static inline void paravirt_alloc_pmd_clone(unsigned long pfn, unsigned long clonepfn,
1114 unsigned long start, unsigned long count)
1115 {
1116 PVOP_VCALL4(pv_mmu_ops.alloc_pmd_clone, pfn, clonepfn, start, count);
1117 }
1118 static inline void paravirt_release_pmd(unsigned long pfn)
1119 {
1120 PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
1121 }
1122
1123 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1124 {
1125 PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
1126 }
1127 static inline void paravirt_release_pud(unsigned long pfn)
1128 {
1129 PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
1130 }
1131
1132 #ifdef CONFIG_HIGHPTE
1133 static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
1134 {
1135 unsigned long ret;
1136 ret = PVOP_CALL2(unsigned long, pv_mmu_ops.kmap_atomic_pte, page, type);
1137 return (void *)ret;
1138 }
1139 #endif
1140
1141 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
1142 pte_t *ptep)
1143 {
1144 PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
1145 }
1146
1147 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
1148 pte_t *ptep)
1149 {
1150 PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
1151 }
1152
1153 static inline pte_t __pte(pteval_t val)
1154 {
1155 pteval_t ret;
1156
1157 if (sizeof(pteval_t) > sizeof(long))
1158 ret = PVOP_CALLEE2(pteval_t,
1159 pv_mmu_ops.make_pte,
1160 val, (u64)val >> 32);
1161 else
1162 ret = PVOP_CALLEE1(pteval_t,
1163 pv_mmu_ops.make_pte,
1164 val);
1165
1166 return (pte_t) { .pte = ret };
1167 }
1168
1169 static inline pteval_t pte_val(pte_t pte)
1170 {
1171 pteval_t ret;
1172
1173 if (sizeof(pteval_t) > sizeof(long))
1174 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
1175 pte.pte, (u64)pte.pte >> 32);
1176 else
1177 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
1178 pte.pte);
1179
1180 return ret;
1181 }
1182
1183 static inline pgd_t __pgd(pgdval_t val)
1184 {
1185 pgdval_t ret;
1186
1187 if (sizeof(pgdval_t) > sizeof(long))
1188 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
1189 val, (u64)val >> 32);
1190 else
1191 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
1192 val);
1193
1194 return (pgd_t) { ret };
1195 }
1196
1197 static inline pgdval_t pgd_val(pgd_t pgd)
1198 {
1199 pgdval_t ret;
1200
1201 if (sizeof(pgdval_t) > sizeof(long))
1202 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
1203 pgd.pgd, (u64)pgd.pgd >> 32);
1204 else
1205 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
1206 pgd.pgd);
1207
1208 return ret;
1209 }
1210
1211 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1212 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
1213 pte_t *ptep)
1214 {
1215 pteval_t ret;
1216
1217 ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
1218 mm, addr, ptep);
1219
1220 return (pte_t) { .pte = ret };
1221 }
1222
1223 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
1224 pte_t *ptep, pte_t pte)
1225 {
1226 if (sizeof(pteval_t) > sizeof(long))
1227 /* 5 arg words */
1228 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
1229 else
1230 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
1231 mm, addr, ptep, pte.pte);
1232 }
1233
1234 static inline void set_pte(pte_t *ptep, pte_t pte)
1235 {
1236 if (sizeof(pteval_t) > sizeof(long))
1237 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
1238 pte.pte, (u64)pte.pte >> 32);
1239 else
1240 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
1241 pte.pte);
1242 }
1243
1244 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
1245 pte_t *ptep, pte_t pte)
1246 {
1247 if (sizeof(pteval_t) > sizeof(long))
1248 /* 5 arg words */
1249 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
1250 else
1251 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
1252 }
1253
1254 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
1255 {
1256 pmdval_t val = native_pmd_val(pmd);
1257
1258 if (sizeof(pmdval_t) > sizeof(long))
1259 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
1260 else
1261 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
1262 }
1263
1264 #if PAGETABLE_LEVELS >= 3
1265 static inline pmd_t __pmd(pmdval_t val)
1266 {
1267 pmdval_t ret;
1268
1269 if (sizeof(pmdval_t) > sizeof(long))
1270 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
1271 val, (u64)val >> 32);
1272 else
1273 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
1274 val);
1275
1276 return (pmd_t) { ret };
1277 }
1278
1279 static inline pmdval_t pmd_val(pmd_t pmd)
1280 {
1281 pmdval_t ret;
1282
1283 if (sizeof(pmdval_t) > sizeof(long))
1284 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
1285 pmd.pmd, (u64)pmd.pmd >> 32);
1286 else
1287 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
1288 pmd.pmd);
1289
1290 return ret;
1291 }
1292
1293 static inline void set_pud(pud_t *pudp, pud_t pud)
1294 {
1295 pudval_t val = native_pud_val(pud);
1296
1297 if (sizeof(pudval_t) > sizeof(long))
1298 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
1299 val, (u64)val >> 32);
1300 else
1301 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
1302 val);
1303 }
1304 #if PAGETABLE_LEVELS == 4
1305 static inline pud_t __pud(pudval_t val)
1306 {
1307 pudval_t ret;
1308
1309 if (sizeof(pudval_t) > sizeof(long))
1310 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
1311 val, (u64)val >> 32);
1312 else
1313 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
1314 val);
1315
1316 return (pud_t) { ret };
1317 }
1318
1319 static inline pudval_t pud_val(pud_t pud)
1320 {
1321 pudval_t ret;
1322
1323 if (sizeof(pudval_t) > sizeof(long))
1324 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
1325 pud.pud, (u64)pud.pud >> 32);
1326 else
1327 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
1328 pud.pud);
1329
1330 return ret;
1331 }
1332
1333 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
1334 {
1335 pgdval_t val = native_pgd_val(pgd);
1336
1337 if (sizeof(pgdval_t) > sizeof(long))
1338 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
1339 val, (u64)val >> 32);
1340 else
1341 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
1342 val);
1343 }
1344
1345 static inline void pgd_clear(pgd_t *pgdp)
1346 {
1347 set_pgd(pgdp, __pgd(0));
1348 }
1349
1350 static inline void pud_clear(pud_t *pudp)
1351 {
1352 set_pud(pudp, __pud(0));
1353 }
1354
1355 #endif /* PAGETABLE_LEVELS == 4 */
1356
1357 #endif /* PAGETABLE_LEVELS >= 3 */
1358
1359 #ifdef CONFIG_X86_PAE
1360 /* Special-case pte-setting operations for PAE, which can't update a
1361 64-bit pte atomically */
1362 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
1363 {
1364 PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
1365 pte.pte, pte.pte >> 32);
1366 }
1367
1368 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
1369 pte_t *ptep)
1370 {
1371 PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
1372 }
1373
1374 static inline void pmd_clear(pmd_t *pmdp)
1375 {
1376 PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
1377 }
1378 #else /* !CONFIG_X86_PAE */
1379 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
1380 {
1381 set_pte(ptep, pte);
1382 }
1383
1384 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
1385 pte_t *ptep)
1386 {
1387 set_pte_at(mm, addr, ptep, __pte(0));
1388 }
1389
1390 static inline void pmd_clear(pmd_t *pmdp)
1391 {
1392 set_pmd(pmdp, __pmd(0));
1393 }
1394 #endif /* CONFIG_X86_PAE */
1395
1396 /* Lazy mode for batching updates / context switch */
1397 enum paravirt_lazy_mode {
1398 PARAVIRT_LAZY_NONE,
1399 PARAVIRT_LAZY_MMU,
1400 PARAVIRT_LAZY_CPU,
1401 };
1402
1403 enum paravirt_lazy_mode paravirt_get_lazy_mode(void);
1404 void paravirt_start_context_switch(struct task_struct *prev);
1405 void paravirt_end_context_switch(struct task_struct *next);
1406
1407 void paravirt_enter_lazy_mmu(void);
1408 void paravirt_leave_lazy_mmu(void);
1409
1410 #define __HAVE_ARCH_START_CONTEXT_SWITCH
1411 static inline void arch_start_context_switch(struct task_struct *prev)
1412 {
1413 PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
1414 }
1415
1416 static inline void arch_end_context_switch(struct task_struct *next)
1417 {
1418 PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
1419 }
1420
1421 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1422 static inline void arch_enter_lazy_mmu_mode(void)
1423 {
1424 PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
1425 }
1426
1427 static inline void arch_leave_lazy_mmu_mode(void)
1428 {
1429 PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
1430 }
1431
1432 void arch_flush_lazy_mmu_mode(void);
1433
1434 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
1435 phys_addr_t phys, pgprot_t flags)
1436 {
1437 pv_mmu_ops.set_fixmap(idx, phys, flags);
1438 }
1439
1440 void _paravirt_nop(void);
1441 u32 _paravirt_ident_32(u32);
1442 u64 _paravirt_ident_64(u64);
1443
1444 #define paravirt_nop ((void *)_paravirt_nop)
1445
1446 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
1447
1448 static inline int __raw_spin_is_locked(struct raw_spinlock *lock)
1449 {
1450 return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
1451 }
1452
1453 static inline int __raw_spin_is_contended(struct raw_spinlock *lock)
1454 {
1455 return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
1456 }
1457 #define __raw_spin_is_contended __raw_spin_is_contended
1458
1459 static __always_inline void __raw_spin_lock(struct raw_spinlock *lock)
1460 {
1461 PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
1462 }
1463
1464 static __always_inline void __raw_spin_lock_flags(struct raw_spinlock *lock,
1465 unsigned long flags)
1466 {
1467 PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
1468 }
1469
1470 static __always_inline int __raw_spin_trylock(struct raw_spinlock *lock)
1471 {
1472 return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
1473 }
1474
1475 static __always_inline void __raw_spin_unlock(struct raw_spinlock *lock)
1476 {
1477 PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
1478 }
1479
1480 #endif
1481
1482 /* These all sit in the .parainstructions section to tell us what to patch. */
1483 struct paravirt_patch_site {
1484 u8 *instr; /* original instructions */
1485 u8 instrtype; /* type of this instruction */
1486 u8 len; /* length of original instruction */
1487 u16 clobbers; /* what registers you may clobber */
1488 };
1489
1490 extern struct paravirt_patch_site __parainstructions[],
1491 __parainstructions_end[];
1492
1493 #ifdef CONFIG_X86_32
1494 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
1495 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
1496
1497 /* save and restore all caller-save registers, except return value */
1498 #define PV_SAVE_ALL_CALLER_REGS "pushl %ecx;"
1499 #define PV_RESTORE_ALL_CALLER_REGS "popl %ecx;"
1500
1501 #define PV_FLAGS_ARG "0"
1502 #define PV_EXTRA_CLOBBERS
1503 #define PV_VEXTRA_CLOBBERS
1504 #else
1505 /* save and restore all caller-save registers, except return value */
1506 #define PV_SAVE_ALL_CALLER_REGS \
1507 "push %rcx;" \
1508 "push %rdx;" \
1509 "push %rsi;" \
1510 "push %rdi;" \
1511 "push %r8;" \
1512 "push %r9;" \
1513 "push %r10;" \
1514 "push %r11;"
1515 #define PV_RESTORE_ALL_CALLER_REGS \
1516 "pop %r11;" \
1517 "pop %r10;" \
1518 "pop %r9;" \
1519 "pop %r8;" \
1520 "pop %rdi;" \
1521 "pop %rsi;" \
1522 "pop %rdx;" \
1523 "pop %rcx;"
1524
1525 /* We save some registers, but all of them, that's too much. We clobber all
1526 * caller saved registers but the argument parameter */
1527 #define PV_SAVE_REGS "pushq %%rdi;"
1528 #define PV_RESTORE_REGS "popq %%rdi;"
1529 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
1530 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
1531 #define PV_FLAGS_ARG "D"
1532 #endif
1533
1534 /*
1535 * Generate a thunk around a function which saves all caller-save
1536 * registers except for the return value. This allows C functions to
1537 * be called from assembler code where fewer than normal registers are
1538 * available. It may also help code generation around calls from C
1539 * code if the common case doesn't use many registers.
1540 *
1541 * When a callee is wrapped in a thunk, the caller can assume that all
1542 * arg regs and all scratch registers are preserved across the
1543 * call. The return value in rax/eax will not be saved, even for void
1544 * functions.
1545 */
1546 #define PV_CALLEE_SAVE_REGS_THUNK(func) \
1547 extern typeof(func) __raw_callee_save_##func; \
1548 static void *__##func##__ __used = func; \
1549 \
1550 asm(".pushsection .text;" \
1551 "__raw_callee_save_" #func ": " \
1552 PV_SAVE_ALL_CALLER_REGS \
1553 "call " #func ";" \
1554 PV_RESTORE_ALL_CALLER_REGS \
1555 "ret;" \
1556 ".popsection")
1557
1558 /* Get a reference to a callee-save function */
1559 #define PV_CALLEE_SAVE(func) \
1560 ((struct paravirt_callee_save) { __raw_callee_save_##func })
1561
1562 /* Promise that "func" already uses the right calling convention */
1563 #define __PV_IS_CALLEE_SAVE(func) \
1564 ((struct paravirt_callee_save) { func })
1565
1566 static inline unsigned long __raw_local_save_flags(void)
1567 {
1568 unsigned long f;
1569
1570 asm volatile(paravirt_alt(PARAVIRT_CALL)
1571 : "=a"(f)
1572 : paravirt_type(pv_irq_ops.save_fl),
1573 paravirt_clobber(CLBR_EAX)
1574 : "memory", "cc");
1575 return f;
1576 }
1577
1578 static inline void raw_local_irq_restore(unsigned long f)
1579 {
1580 asm volatile(paravirt_alt(PARAVIRT_CALL)
1581 : "=a"(f)
1582 : PV_FLAGS_ARG(f),
1583 paravirt_type(pv_irq_ops.restore_fl),
1584 paravirt_clobber(CLBR_EAX)
1585 : "memory", "cc");
1586 }
1587
1588 static inline void raw_local_irq_disable(void)
1589 {
1590 asm volatile(paravirt_alt(PARAVIRT_CALL)
1591 :
1592 : paravirt_type(pv_irq_ops.irq_disable),
1593 paravirt_clobber(CLBR_EAX)
1594 : "memory", "eax", "cc");
1595 }
1596
1597 static inline void raw_local_irq_enable(void)
1598 {
1599 asm volatile(paravirt_alt(PARAVIRT_CALL)
1600 :
1601 : paravirt_type(pv_irq_ops.irq_enable),
1602 paravirt_clobber(CLBR_EAX)
1603 : "memory", "eax", "cc");
1604 }
1605
1606 static inline unsigned long __raw_local_irq_save(void)
1607 {
1608 unsigned long f;
1609
1610 f = __raw_local_save_flags();
1611 raw_local_irq_disable();
1612 return f;
1613 }
1614
1615
1616 /* Make sure as little as possible of this mess escapes. */
1617 #undef PARAVIRT_CALL
1618 #undef __PVOP_CALL
1619 #undef __PVOP_VCALL
1620 #undef PVOP_VCALL0
1621 #undef PVOP_CALL0
1622 #undef PVOP_VCALL1
1623 #undef PVOP_CALL1
1624 #undef PVOP_VCALL2
1625 #undef PVOP_CALL2
1626 #undef PVOP_VCALL3
1627 #undef PVOP_CALL3
1628 #undef PVOP_VCALL4
1629 #undef PVOP_CALL4
1630
1631 #else /* __ASSEMBLY__ */
1632
1633 #define _PVSITE(ptype, clobbers, ops, word, algn) \
1634 771:; \
1635 ops; \
1636 772:; \
1637 .pushsection .parainstructions,"a"; \
1638 .align algn; \
1639 word 771b; \
1640 .byte ptype; \
1641 .byte 772b-771b; \
1642 .short clobbers; \
1643 .popsection
1644
1645
1646 #define COND_PUSH(set, mask, reg) \
1647 .if ((~(set)) & mask); push %reg; .endif
1648 #define COND_POP(set, mask, reg) \
1649 .if ((~(set)) & mask); pop %reg; .endif
1650
1651 #ifdef CONFIG_X86_64
1652
1653 #define PV_SAVE_REGS(set) \
1654 COND_PUSH(set, CLBR_RAX, rax); \
1655 COND_PUSH(set, CLBR_RCX, rcx); \
1656 COND_PUSH(set, CLBR_RDX, rdx); \
1657 COND_PUSH(set, CLBR_RSI, rsi); \
1658 COND_PUSH(set, CLBR_RDI, rdi); \
1659 COND_PUSH(set, CLBR_R8, r8); \
1660 COND_PUSH(set, CLBR_R9, r9); \
1661 COND_PUSH(set, CLBR_R10, r10); \
1662 COND_PUSH(set, CLBR_R11, r11)
1663 #define PV_RESTORE_REGS(set) \
1664 COND_POP(set, CLBR_R11, r11); \
1665 COND_POP(set, CLBR_R10, r10); \
1666 COND_POP(set, CLBR_R9, r9); \
1667 COND_POP(set, CLBR_R8, r8); \
1668 COND_POP(set, CLBR_RDI, rdi); \
1669 COND_POP(set, CLBR_RSI, rsi); \
1670 COND_POP(set, CLBR_RDX, rdx); \
1671 COND_POP(set, CLBR_RCX, rcx); \
1672 COND_POP(set, CLBR_RAX, rax)
1673
1674 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 8)
1675 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
1676 #define PARA_INDIRECT(addr) *addr(%rip)
1677 #else
1678 #define PV_SAVE_REGS(set) \
1679 COND_PUSH(set, CLBR_EAX, eax); \
1680 COND_PUSH(set, CLBR_EDI, edi); \
1681 COND_PUSH(set, CLBR_ECX, ecx); \
1682 COND_PUSH(set, CLBR_EDX, edx)
1683 #define PV_RESTORE_REGS(set) \
1684 COND_POP(set, CLBR_EDX, edx); \
1685 COND_POP(set, CLBR_ECX, ecx); \
1686 COND_POP(set, CLBR_EDI, edi); \
1687 COND_POP(set, CLBR_EAX, eax)
1688
1689 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 4)
1690 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
1691 #define PARA_INDIRECT(addr) *%cs:addr
1692 #endif
1693
1694 #define INTERRUPT_RETURN \
1695 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE, \
1696 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
1697
1698 #define DISABLE_INTERRUPTS(clobbers) \
1699 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
1700 PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \
1701 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable); \
1702 PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
1703
1704 #define ENABLE_INTERRUPTS(clobbers) \
1705 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers, \
1706 PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \
1707 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable); \
1708 PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
1709
1710 #define USERGS_SYSRET32 \
1711 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32), \
1712 CLBR_NONE, \
1713 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
1714
1715 #ifdef CONFIG_X86_32
1716 #define GET_CR0_INTO_EAX \
1717 push %ecx; push %edx; \
1718 call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
1719 pop %edx; pop %ecx
1720
1721 #define ENABLE_INTERRUPTS_SYSEXIT \
1722 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
1723 CLBR_NONE, \
1724 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1725
1726
1727 #else /* !CONFIG_X86_32 */
1728
1729 /*
1730 * If swapgs is used while the userspace stack is still current,
1731 * there's no way to call a pvop. The PV replacement *must* be
1732 * inlined, or the swapgs instruction must be trapped and emulated.
1733 */
1734 #define SWAPGS_UNSAFE_STACK \
1735 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1736 swapgs)
1737
1738 /*
1739 * Note: swapgs is very special, and in practise is either going to be
1740 * implemented with a single "swapgs" instruction or something very
1741 * special. Either way, we don't need to save any registers for
1742 * it.
1743 */
1744 #define SWAPGS \
1745 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1746 call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs) \
1747 )
1748
1749 #define GET_CR2_INTO_RCX \
1750 call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2); \
1751 movq %rax, %rcx; \
1752 xorq %rax, %rax;
1753
1754 #define PARAVIRT_ADJUST_EXCEPTION_FRAME \
1755 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
1756 CLBR_NONE, \
1757 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
1758
1759 #define USERGS_SYSRET64 \
1760 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64), \
1761 CLBR_NONE, \
1762 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1763
1764 #define ENABLE_INTERRUPTS_SYSEXIT32 \
1765 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
1766 CLBR_NONE, \
1767 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1768 #endif /* CONFIG_X86_32 */
1769
1770 #endif /* __ASSEMBLY__ */
1771 #endif /* CONFIG_PARAVIRT */
1772 #endif /* _ASM_X86_PARAVIRT_H */
connection oriented routing - ARCHIVE NO UPDATES HERE