1 #ifndef __ASM_PARAVIRT_H
2 #define __ASM_PARAVIRT_H
3 /* Various instructions on x86 need to be replaced for
4 * para-virtualization: those hooks are defined here. */
10 /* Bitmask of what can be clobbered: usually at least eax. */
12 #define CLBR_EAX (1 << 0)
13 #define CLBR_ECX (1 << 1)
14 #define CLBR_EDX (1 << 2)
17 #define CLBR_RSI (1 << 3)
18 #define CLBR_RDI (1 << 4)
19 #define CLBR_R8 (1 << 5)
20 #define CLBR_R9 (1 << 6)
21 #define CLBR_R10 (1 << 7)
22 #define CLBR_R11 (1 << 8)
23 #define CLBR_ANY ((1 << 9) - 1)
24 #include <asm/desc_defs.h>
26 /* CLBR_ANY should match all regs platform has. For i386, that's just it */
27 #define CLBR_ANY ((1 << 3) - 1)
31 #include <linux/types.h>
32 #include <linux/cpumask.h>
33 #include <asm/kmap_types.h>
34 #include <asm/desc_defs.h>
45 unsigned int kernel_rpl
;
46 int shared_kernel_pmd
;
53 * Patch may replace one of the defined code sequences with
54 * arbitrary code, subject to the same register constraints.
55 * This generally means the code is not free to clobber any
56 * registers other than EAX. The patch function should return
57 * the number of bytes of code generated, as we nop pad the
58 * rest in generic code.
60 unsigned (*patch
)(u8 type
, u16 clobber
, void *insnbuf
,
61 unsigned long addr
, unsigned len
);
63 /* Basic arch-specific setup */
64 void (*arch_setup
)(void);
65 char *(*memory_setup
)(void);
66 void (*post_allocator_init
)(void);
68 /* Print a banner to identify the environment */
74 /* Set deferred update mode, used for batching operations. */
80 void (*time_init
)(void);
82 /* Set and set time of day */
83 unsigned long (*get_wallclock
)(void);
84 int (*set_wallclock
)(unsigned long);
86 unsigned long long (*sched_clock
)(void);
87 unsigned long (*get_tsc_khz
)(void);
91 /* hooks for various privileged instructions */
92 unsigned long (*get_debugreg
)(int regno
);
93 void (*set_debugreg
)(int regno
, unsigned long value
);
97 unsigned long (*read_cr0
)(void);
98 void (*write_cr0
)(unsigned long);
100 unsigned long (*read_cr4_safe
)(void);
101 unsigned long (*read_cr4
)(void);
102 void (*write_cr4
)(unsigned long);
105 unsigned long (*read_cr8
)(void);
106 void (*write_cr8
)(unsigned long);
109 /* Segment descriptor handling */
110 void (*load_tr_desc
)(void);
111 void (*load_gdt
)(const struct desc_ptr
*);
112 void (*load_idt
)(const struct desc_ptr
*);
113 void (*store_gdt
)(struct desc_ptr
*);
114 void (*store_idt
)(struct desc_ptr
*);
115 void (*set_ldt
)(const void *desc
, unsigned entries
);
116 unsigned long (*store_tr
)(void);
117 void (*load_tls
)(struct thread_struct
*t
, unsigned int cpu
);
119 void (*load_gs_index
)(unsigned int idx
);
121 void (*write_ldt_entry
)(struct desc_struct
*ldt
, int entrynum
,
123 void (*write_gdt_entry
)(struct desc_struct
*,
124 int entrynum
, const void *desc
, int size
);
125 void (*write_idt_entry
)(gate_desc
*,
126 int entrynum
, const gate_desc
*gate
);
127 void (*load_sp0
)(struct tss_struct
*tss
, struct thread_struct
*t
);
129 void (*set_iopl_mask
)(unsigned mask
);
131 void (*wbinvd
)(void);
132 void (*io_delay
)(void);
134 /* cpuid emulation, mostly so that caps bits can be disabled */
135 void (*cpuid
)(unsigned int *eax
, unsigned int *ebx
,
136 unsigned int *ecx
, unsigned int *edx
);
138 /* MSR, PMC and TSR operations.
139 err = 0/-EFAULT. wrmsr returns 0/-EFAULT. */
140 u64 (*read_msr
)(unsigned int msr
, int *err
);
141 int (*write_msr
)(unsigned int msr
, unsigned low
, unsigned high
);
143 u64 (*read_tsc
)(void);
144 u64 (*read_pmc
)(int counter
);
145 unsigned long long (*read_tscp
)(unsigned int *aux
);
148 * Atomically enable interrupts and return to userspace. This
149 * is only ever used to return to 32-bit processes; in a
150 * 64-bit kernel, it's used for 32-on-64 compat processes, but
151 * never native 64-bit processes. (Jump, not call.)
153 void (*irq_enable_sysexit
)(void);
156 * Switch to usermode gs and return to 64-bit usermode using
157 * sysret. Only used in 64-bit kernels to return to 64-bit
158 * processes. Usermode register state, including %rsp, must
159 * already be restored.
161 void (*usergs_sysret64
)(void);
164 * Switch to usermode gs and return to 32-bit usermode using
165 * sysret. Used to return to 32-on-64 compat processes.
166 * Other usermode register state, including %esp, must already
169 void (*usergs_sysret32
)(void);
171 /* Normal iret. Jump to this with the standard iret stack
175 void (*swapgs
)(void);
177 struct pv_lazy_ops lazy_mode
;
181 void (*init_IRQ
)(void);
184 * Get/set interrupt state. save_fl and restore_fl are only
185 * expected to use X86_EFLAGS_IF; all other bits
186 * returned from save_fl are undefined, and may be ignored by
189 unsigned long (*save_fl
)(void);
190 void (*restore_fl
)(unsigned long);
191 void (*irq_disable
)(void);
192 void (*irq_enable
)(void);
193 void (*safe_halt
)(void);
197 void (*adjust_exception_frame
)(void);
202 #ifdef CONFIG_X86_LOCAL_APIC
204 * Direct APIC operations, principally for VMI. Ideally
205 * these shouldn't be in this interface.
207 void (*apic_write
)(unsigned long reg
, u32 v
);
208 void (*apic_write_atomic
)(unsigned long reg
, u32 v
);
209 u32 (*apic_read
)(unsigned long reg
);
210 void (*setup_boot_clock
)(void);
211 void (*setup_secondary_clock
)(void);
213 void (*startup_ipi_hook
)(int phys_apicid
,
214 unsigned long start_eip
,
215 unsigned long start_esp
);
221 * Called before/after init_mm pagetable setup. setup_start
222 * may reset %cr3, and may pre-install parts of the pagetable;
223 * pagetable setup is expected to preserve any existing
226 void (*pagetable_setup_start
)(pgd_t
*pgd_base
);
227 void (*pagetable_setup_done
)(pgd_t
*pgd_base
);
229 unsigned long (*read_cr2
)(void);
230 void (*write_cr2
)(unsigned long);
232 unsigned long (*read_cr3
)(void);
233 void (*write_cr3
)(unsigned long);
236 * Hooks for intercepting the creation/use/destruction of an
239 void (*activate_mm
)(struct mm_struct
*prev
,
240 struct mm_struct
*next
);
241 void (*dup_mmap
)(struct mm_struct
*oldmm
,
242 struct mm_struct
*mm
);
243 void (*exit_mmap
)(struct mm_struct
*mm
);
247 void (*flush_tlb_user
)(void);
248 void (*flush_tlb_kernel
)(void);
249 void (*flush_tlb_single
)(unsigned long addr
);
250 void (*flush_tlb_others
)(const cpumask_t
*cpus
, struct mm_struct
*mm
,
253 /* Hooks for allocating and freeing a pagetable top-level */
254 int (*pgd_alloc
)(struct mm_struct
*mm
);
255 void (*pgd_free
)(struct mm_struct
*mm
, pgd_t
*pgd
);
258 * Hooks for allocating/releasing pagetable pages when they're
259 * attached to a pagetable
261 void (*alloc_pte
)(struct mm_struct
*mm
, u32 pfn
);
262 void (*alloc_pmd
)(struct mm_struct
*mm
, u32 pfn
);
263 void (*alloc_pmd_clone
)(u32 pfn
, u32 clonepfn
, u32 start
, u32 count
);
264 void (*alloc_pud
)(struct mm_struct
*mm
, u32 pfn
);
265 void (*release_pte
)(u32 pfn
);
266 void (*release_pmd
)(u32 pfn
);
267 void (*release_pud
)(u32 pfn
);
269 /* Pagetable manipulation functions */
270 void (*set_pte
)(pte_t
*ptep
, pte_t pteval
);
271 void (*set_pte_at
)(struct mm_struct
*mm
, unsigned long addr
,
272 pte_t
*ptep
, pte_t pteval
);
273 void (*set_pmd
)(pmd_t
*pmdp
, pmd_t pmdval
);
274 void (*pte_update
)(struct mm_struct
*mm
, unsigned long addr
,
276 void (*pte_update_defer
)(struct mm_struct
*mm
,
277 unsigned long addr
, pte_t
*ptep
);
279 pte_t (*ptep_modify_prot_start
)(struct mm_struct
*mm
, unsigned long addr
,
281 void (*ptep_modify_prot_commit
)(struct mm_struct
*mm
, unsigned long addr
,
282 pte_t
*ptep
, pte_t pte
);
284 pteval_t (*pte_val
)(pte_t
);
285 pteval_t (*pte_flags
)(pte_t
);
286 pte_t (*make_pte
)(pteval_t pte
);
288 pgdval_t (*pgd_val
)(pgd_t
);
289 pgd_t (*make_pgd
)(pgdval_t pgd
);
291 #if PAGETABLE_LEVELS >= 3
292 #ifdef CONFIG_X86_PAE
293 void (*set_pte_atomic
)(pte_t
*ptep
, pte_t pteval
);
294 void (*set_pte_present
)(struct mm_struct
*mm
, unsigned long addr
,
295 pte_t
*ptep
, pte_t pte
);
296 void (*pte_clear
)(struct mm_struct
*mm
, unsigned long addr
,
298 void (*pmd_clear
)(pmd_t
*pmdp
);
300 #endif /* CONFIG_X86_PAE */
302 void (*set_pud
)(pud_t
*pudp
, pud_t pudval
);
304 pmdval_t (*pmd_val
)(pmd_t
);
305 pmd_t (*make_pmd
)(pmdval_t pmd
);
307 #if PAGETABLE_LEVELS == 4
308 pudval_t (*pud_val
)(pud_t
);
309 pud_t (*make_pud
)(pudval_t pud
);
311 void (*set_pgd
)(pgd_t
*pudp
, pgd_t pgdval
);
312 #endif /* PAGETABLE_LEVELS == 4 */
313 #endif /* PAGETABLE_LEVELS >= 3 */
315 #ifdef CONFIG_HIGHPTE
316 void *(*kmap_atomic_pte
)(struct page
*page
, enum km_type type
);
319 struct pv_lazy_ops lazy_mode
;
323 /* Sometimes the physical address is a pfn, and sometimes its
324 an mfn. We can tell which is which from the index. */
325 void (*set_fixmap
)(unsigned /* enum fixed_addresses */ idx
,
326 unsigned long phys
, pgprot_t flags
);
329 /* This contains all the paravirt structures: we get a convenient
330 * number for each function using the offset which we use to indicate
332 struct paravirt_patch_template
{
333 struct pv_init_ops pv_init_ops
;
334 struct pv_time_ops pv_time_ops
;
335 struct pv_cpu_ops pv_cpu_ops
;
336 struct pv_irq_ops pv_irq_ops
;
337 struct pv_apic_ops pv_apic_ops
;
338 struct pv_mmu_ops pv_mmu_ops
;
341 extern struct pv_info pv_info
;
342 extern struct pv_init_ops pv_init_ops
;
343 extern struct pv_time_ops pv_time_ops
;
344 extern struct pv_cpu_ops pv_cpu_ops
;
345 extern struct pv_irq_ops pv_irq_ops
;
346 extern struct pv_apic_ops pv_apic_ops
;
347 extern struct pv_mmu_ops pv_mmu_ops
;
349 #define PARAVIRT_PATCH(x) \
350 (offsetof(struct paravirt_patch_template, x) / sizeof(void *))
352 #define paravirt_type(op) \
353 [paravirt_typenum] "i" (PARAVIRT_PATCH(op)), \
354 [paravirt_opptr] "m" (op)
355 #define paravirt_clobber(clobber) \
356 [paravirt_clobber] "i" (clobber)
359 * Generate some code, and mark it as patchable by the
360 * apply_paravirt() alternate instruction patcher.
362 #define _paravirt_alt(insn_string, type, clobber) \
363 "771:\n\t" insn_string "\n" "772:\n" \
364 ".pushsection .parainstructions,\"a\"\n" \
367 " .byte " type "\n" \
368 " .byte 772b-771b\n" \
369 " .short " clobber "\n" \
372 /* Generate patchable code, with the default asm parameters. */
373 #define paravirt_alt(insn_string) \
374 _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
376 /* Simple instruction patching code. */
377 #define DEF_NATIVE(ops, name, code) \
378 extern const char start_##ops##_##name[], end_##ops##_##name[]; \
379 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
381 unsigned paravirt_patch_nop(void);
382 unsigned paravirt_patch_ignore(unsigned len
);
383 unsigned paravirt_patch_call(void *insnbuf
,
384 const void *target
, u16 tgt_clobbers
,
385 unsigned long addr
, u16 site_clobbers
,
387 unsigned paravirt_patch_jmp(void *insnbuf
, const void *target
,
388 unsigned long addr
, unsigned len
);
389 unsigned paravirt_patch_default(u8 type
, u16 clobbers
, void *insnbuf
,
390 unsigned long addr
, unsigned len
);
392 unsigned paravirt_patch_insns(void *insnbuf
, unsigned len
,
393 const char *start
, const char *end
);
395 unsigned native_patch(u8 type
, u16 clobbers
, void *ibuf
,
396 unsigned long addr
, unsigned len
);
398 int paravirt_disable_iospace(void);
401 * This generates an indirect call based on the operation type number.
402 * The type number, computed in PARAVIRT_PATCH, is derived from the
403 * offset into the paravirt_patch_template structure, and can therefore be
404 * freely converted back into a structure offset.
406 #define PARAVIRT_CALL "call *%[paravirt_opptr];"
409 * These macros are intended to wrap calls through one of the paravirt
410 * ops structs, so that they can be later identified and patched at
413 * Normally, a call to a pv_op function is a simple indirect call:
414 * (pv_op_struct.operations)(args...).
416 * Unfortunately, this is a relatively slow operation for modern CPUs,
417 * because it cannot necessarily determine what the destination
418 * address is. In this case, the address is a runtime constant, so at
419 * the very least we can patch the call to e a simple direct call, or
420 * ideally, patch an inline implementation into the callsite. (Direct
421 * calls are essentially free, because the call and return addresses
422 * are completely predictable.)
424 * For i386, these macros rely on the standard gcc "regparm(3)" calling
425 * convention, in which the first three arguments are placed in %eax,
426 * %edx, %ecx (in that order), and the remaining arguments are placed
427 * on the stack. All caller-save registers (eax,edx,ecx) are expected
428 * to be modified (either clobbered or used for return values).
429 * X86_64, on the other hand, already specifies a register-based calling
430 * conventions, returning at %rax, with parameteres going on %rdi, %rsi,
431 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
432 * special handling for dealing with 4 arguments, unlike i386.
433 * However, x86_64 also have to clobber all caller saved registers, which
434 * unfortunately, are quite a bit (r8 - r11)
436 * The call instruction itself is marked by placing its start address
437 * and size into the .parainstructions section, so that
438 * apply_paravirt() in arch/i386/kernel/alternative.c can do the
439 * appropriate patching under the control of the backend pv_init_ops
442 * Unfortunately there's no way to get gcc to generate the args setup
443 * for the call, and then allow the call itself to be generated by an
444 * inline asm. Because of this, we must do the complete arg setup and
445 * return value handling from within these macros. This is fairly
448 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
449 * It could be extended to more arguments, but there would be little
450 * to be gained from that. For each number of arguments, there are
451 * the two VCALL and CALL variants for void and non-void functions.
453 * When there is a return value, the invoker of the macro must specify
454 * the return type. The macro then uses sizeof() on that type to
455 * determine whether its a 32 or 64 bit value, and places the return
456 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
457 * 64-bit). For x86_64 machines, it just returns at %rax regardless of
458 * the return value size.
460 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
461 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
464 * Small structures are passed and returned in registers. The macro
465 * calling convention can't directly deal with this, so the wrapper
466 * functions must do this.
468 * These PVOP_* macros are only defined within this header. This
469 * means that all uses must be wrapped in inline functions. This also
470 * makes sure the incoming and outgoing types are always correct.
473 #define PVOP_VCALL_ARGS unsigned long __eax, __edx, __ecx
474 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS
475 #define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \
477 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS
478 #define EXTRA_CLOBBERS
479 #define VEXTRA_CLOBBERS
481 #define PVOP_VCALL_ARGS unsigned long __edi, __esi, __edx, __ecx
482 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS, __eax
483 #define PVOP_VCALL_CLOBBERS "=D" (__edi), \
484 "=S" (__esi), "=d" (__edx), \
487 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax)
489 #define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11"
490 #define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11"
493 #ifdef CONFIG_PARAVIRT_DEBUG
494 #define PVOP_TEST_NULL(op) BUG_ON(op == NULL)
496 #define PVOP_TEST_NULL(op) ((void)op)
499 #define __PVOP_CALL(rettype, op, pre, post, ...) \
503 PVOP_TEST_NULL(op); \
504 /* This is 32-bit specific, but is okay in 64-bit */ \
505 /* since this condition will never hold */ \
506 if (sizeof(rettype) > sizeof(unsigned long)) { \
508 paravirt_alt(PARAVIRT_CALL) \
510 : PVOP_CALL_CLOBBERS \
511 : paravirt_type(op), \
512 paravirt_clobber(CLBR_ANY), \
514 : "memory", "cc" EXTRA_CLOBBERS); \
515 __ret = (rettype)((((u64)__edx) << 32) | __eax); \
518 paravirt_alt(PARAVIRT_CALL) \
520 : PVOP_CALL_CLOBBERS \
521 : paravirt_type(op), \
522 paravirt_clobber(CLBR_ANY), \
524 : "memory", "cc" EXTRA_CLOBBERS); \
525 __ret = (rettype)__eax; \
529 #define __PVOP_VCALL(op, pre, post, ...) \
532 PVOP_TEST_NULL(op); \
534 paravirt_alt(PARAVIRT_CALL) \
536 : PVOP_VCALL_CLOBBERS \
537 : paravirt_type(op), \
538 paravirt_clobber(CLBR_ANY), \
540 : "memory", "cc" VEXTRA_CLOBBERS); \
543 #define PVOP_CALL0(rettype, op) \
544 __PVOP_CALL(rettype, op, "", "")
545 #define PVOP_VCALL0(op) \
546 __PVOP_VCALL(op, "", "")
548 #define PVOP_CALL1(rettype, op, arg1) \
549 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)))
550 #define PVOP_VCALL1(op, arg1) \
551 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)))
553 #define PVOP_CALL2(rettype, op, arg1, arg2) \
554 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
555 "1" ((unsigned long)(arg2)))
556 #define PVOP_VCALL2(op, arg1, arg2) \
557 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
558 "1" ((unsigned long)(arg2)))
560 #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \
561 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
562 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
563 #define PVOP_VCALL3(op, arg1, arg2, arg3) \
564 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
565 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
567 /* This is the only difference in x86_64. We can make it much simpler */
569 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
570 __PVOP_CALL(rettype, op, \
571 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
572 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
573 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
574 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
576 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
577 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
578 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
580 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
581 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
582 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
583 "3"((unsigned long)(arg4)))
584 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
585 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
586 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
587 "3"((unsigned long)(arg4)))
590 static inline int paravirt_enabled(void)
592 return pv_info
.paravirt_enabled
;
595 static inline void load_sp0(struct tss_struct
*tss
,
596 struct thread_struct
*thread
)
598 PVOP_VCALL2(pv_cpu_ops
.load_sp0
, tss
, thread
);
601 #define ARCH_SETUP pv_init_ops.arch_setup();
602 static inline unsigned long get_wallclock(void)
604 return PVOP_CALL0(unsigned long, pv_time_ops
.get_wallclock
);
607 static inline int set_wallclock(unsigned long nowtime
)
609 return PVOP_CALL1(int, pv_time_ops
.set_wallclock
, nowtime
);
612 static inline void (*choose_time_init(void))(void)
614 return pv_time_ops
.time_init
;
617 /* The paravirtualized CPUID instruction. */
618 static inline void __cpuid(unsigned int *eax
, unsigned int *ebx
,
619 unsigned int *ecx
, unsigned int *edx
)
621 PVOP_VCALL4(pv_cpu_ops
.cpuid
, eax
, ebx
, ecx
, edx
);
625 * These special macros can be used to get or set a debugging register
627 static inline unsigned long paravirt_get_debugreg(int reg
)
629 return PVOP_CALL1(unsigned long, pv_cpu_ops
.get_debugreg
, reg
);
631 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
632 static inline void set_debugreg(unsigned long val
, int reg
)
634 PVOP_VCALL2(pv_cpu_ops
.set_debugreg
, reg
, val
);
637 static inline void clts(void)
639 PVOP_VCALL0(pv_cpu_ops
.clts
);
642 static inline unsigned long read_cr0(void)
644 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr0
);
647 static inline void write_cr0(unsigned long x
)
649 PVOP_VCALL1(pv_cpu_ops
.write_cr0
, x
);
652 static inline unsigned long read_cr2(void)
654 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr2
);
657 static inline void write_cr2(unsigned long x
)
659 PVOP_VCALL1(pv_mmu_ops
.write_cr2
, x
);
662 static inline unsigned long read_cr3(void)
664 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr3
);
667 static inline void write_cr3(unsigned long x
)
669 PVOP_VCALL1(pv_mmu_ops
.write_cr3
, x
);
672 static inline unsigned long read_cr4(void)
674 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4
);
676 static inline unsigned long read_cr4_safe(void)
678 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4_safe
);
681 static inline void write_cr4(unsigned long x
)
683 PVOP_VCALL1(pv_cpu_ops
.write_cr4
, x
);
687 static inline unsigned long read_cr8(void)
689 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr8
);
692 static inline void write_cr8(unsigned long x
)
694 PVOP_VCALL1(pv_cpu_ops
.write_cr8
, x
);
698 static inline void raw_safe_halt(void)
700 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
703 static inline void halt(void)
705 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
708 static inline void wbinvd(void)
710 PVOP_VCALL0(pv_cpu_ops
.wbinvd
);
713 #define get_kernel_rpl() (pv_info.kernel_rpl)
715 static inline u64
paravirt_read_msr(unsigned msr
, int *err
)
717 return PVOP_CALL2(u64
, pv_cpu_ops
.read_msr
, msr
, err
);
719 static inline int paravirt_write_msr(unsigned msr
, unsigned low
, unsigned high
)
721 return PVOP_CALL3(int, pv_cpu_ops
.write_msr
, msr
, low
, high
);
724 /* These should all do BUG_ON(_err), but our headers are too tangled. */
725 #define rdmsr(msr, val1, val2) \
728 u64 _l = paravirt_read_msr(msr, &_err); \
733 #define wrmsr(msr, val1, val2) \
735 paravirt_write_msr(msr, val1, val2); \
738 #define rdmsrl(msr, val) \
741 val = paravirt_read_msr(msr, &_err); \
744 #define wrmsrl(msr, val) wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
745 #define wrmsr_safe(msr, a, b) paravirt_write_msr(msr, a, b)
747 /* rdmsr with exception handling */
748 #define rdmsr_safe(msr, a, b) \
751 u64 _l = paravirt_read_msr(msr, &_err); \
757 static inline int rdmsrl_safe(unsigned msr
, unsigned long long *p
)
761 *p
= paravirt_read_msr(msr
, &err
);
765 static inline u64
paravirt_read_tsc(void)
767 return PVOP_CALL0(u64
, pv_cpu_ops
.read_tsc
);
770 #define rdtscl(low) \
772 u64 _l = paravirt_read_tsc(); \
776 #define rdtscll(val) (val = paravirt_read_tsc())
778 static inline unsigned long long paravirt_sched_clock(void)
780 return PVOP_CALL0(unsigned long long, pv_time_ops
.sched_clock
);
782 #define calibrate_tsc() (pv_time_ops.get_tsc_khz())
784 static inline unsigned long long paravirt_read_pmc(int counter
)
786 return PVOP_CALL1(u64
, pv_cpu_ops
.read_pmc
, counter
);
789 #define rdpmc(counter, low, high) \
791 u64 _l = paravirt_read_pmc(counter); \
796 static inline unsigned long long paravirt_rdtscp(unsigned int *aux
)
798 return PVOP_CALL1(u64
, pv_cpu_ops
.read_tscp
, aux
);
801 #define rdtscp(low, high, aux) \
804 unsigned long __val = paravirt_rdtscp(&__aux); \
805 (low) = (u32)__val; \
806 (high) = (u32)(__val >> 32); \
810 #define rdtscpll(val, aux) \
812 unsigned long __aux; \
813 val = paravirt_rdtscp(&__aux); \
817 static inline void load_TR_desc(void)
819 PVOP_VCALL0(pv_cpu_ops
.load_tr_desc
);
821 static inline void load_gdt(const struct desc_ptr
*dtr
)
823 PVOP_VCALL1(pv_cpu_ops
.load_gdt
, dtr
);
825 static inline void load_idt(const struct desc_ptr
*dtr
)
827 PVOP_VCALL1(pv_cpu_ops
.load_idt
, dtr
);
829 static inline void set_ldt(const void *addr
, unsigned entries
)
831 PVOP_VCALL2(pv_cpu_ops
.set_ldt
, addr
, entries
);
833 static inline void store_gdt(struct desc_ptr
*dtr
)
835 PVOP_VCALL1(pv_cpu_ops
.store_gdt
, dtr
);
837 static inline void store_idt(struct desc_ptr
*dtr
)
839 PVOP_VCALL1(pv_cpu_ops
.store_idt
, dtr
);
841 static inline unsigned long paravirt_store_tr(void)
843 return PVOP_CALL0(unsigned long, pv_cpu_ops
.store_tr
);
845 #define store_tr(tr) ((tr) = paravirt_store_tr())
846 static inline void load_TLS(struct thread_struct
*t
, unsigned cpu
)
848 PVOP_VCALL2(pv_cpu_ops
.load_tls
, t
, cpu
);
852 static inline void load_gs_index(unsigned int gs
)
854 PVOP_VCALL1(pv_cpu_ops
.load_gs_index
, gs
);
858 static inline void write_ldt_entry(struct desc_struct
*dt
, int entry
,
861 PVOP_VCALL3(pv_cpu_ops
.write_ldt_entry
, dt
, entry
, desc
);
864 static inline void write_gdt_entry(struct desc_struct
*dt
, int entry
,
865 void *desc
, int type
)
867 PVOP_VCALL4(pv_cpu_ops
.write_gdt_entry
, dt
, entry
, desc
, type
);
870 static inline void write_idt_entry(gate_desc
*dt
, int entry
, const gate_desc
*g
)
872 PVOP_VCALL3(pv_cpu_ops
.write_idt_entry
, dt
, entry
, g
);
874 static inline void set_iopl_mask(unsigned mask
)
876 PVOP_VCALL1(pv_cpu_ops
.set_iopl_mask
, mask
);
879 /* The paravirtualized I/O functions */
880 static inline void slow_down_io(void)
882 pv_cpu_ops
.io_delay();
883 #ifdef REALLY_SLOW_IO
884 pv_cpu_ops
.io_delay();
885 pv_cpu_ops
.io_delay();
886 pv_cpu_ops
.io_delay();
890 #ifdef CONFIG_X86_LOCAL_APIC
892 * Basic functions accessing APICs.
894 static inline void apic_write(unsigned long reg
, u32 v
)
896 PVOP_VCALL2(pv_apic_ops
.apic_write
, reg
, v
);
899 static inline void apic_write_atomic(unsigned long reg
, u32 v
)
901 PVOP_VCALL2(pv_apic_ops
.apic_write_atomic
, reg
, v
);
904 static inline u32
apic_read(unsigned long reg
)
906 return PVOP_CALL1(unsigned long, pv_apic_ops
.apic_read
, reg
);
909 static inline void setup_boot_clock(void)
911 PVOP_VCALL0(pv_apic_ops
.setup_boot_clock
);
914 static inline void setup_secondary_clock(void)
916 PVOP_VCALL0(pv_apic_ops
.setup_secondary_clock
);
920 static inline void paravirt_post_allocator_init(void)
922 if (pv_init_ops
.post_allocator_init
)
923 (*pv_init_ops
.post_allocator_init
)();
926 static inline void paravirt_pagetable_setup_start(pgd_t
*base
)
928 (*pv_mmu_ops
.pagetable_setup_start
)(base
);
931 static inline void paravirt_pagetable_setup_done(pgd_t
*base
)
933 (*pv_mmu_ops
.pagetable_setup_done
)(base
);
937 static inline void startup_ipi_hook(int phys_apicid
, unsigned long start_eip
,
938 unsigned long start_esp
)
940 PVOP_VCALL3(pv_apic_ops
.startup_ipi_hook
,
941 phys_apicid
, start_eip
, start_esp
);
945 static inline void paravirt_activate_mm(struct mm_struct
*prev
,
946 struct mm_struct
*next
)
948 PVOP_VCALL2(pv_mmu_ops
.activate_mm
, prev
, next
);
951 static inline void arch_dup_mmap(struct mm_struct
*oldmm
,
952 struct mm_struct
*mm
)
954 PVOP_VCALL2(pv_mmu_ops
.dup_mmap
, oldmm
, mm
);
957 static inline void arch_exit_mmap(struct mm_struct
*mm
)
959 PVOP_VCALL1(pv_mmu_ops
.exit_mmap
, mm
);
962 static inline void __flush_tlb(void)
964 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_user
);
966 static inline void __flush_tlb_global(void)
968 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_kernel
);
970 static inline void __flush_tlb_single(unsigned long addr
)
972 PVOP_VCALL1(pv_mmu_ops
.flush_tlb_single
, addr
);
975 static inline void flush_tlb_others(cpumask_t cpumask
, struct mm_struct
*mm
,
978 PVOP_VCALL3(pv_mmu_ops
.flush_tlb_others
, &cpumask
, mm
, va
);
981 static inline int paravirt_pgd_alloc(struct mm_struct
*mm
)
983 return PVOP_CALL1(int, pv_mmu_ops
.pgd_alloc
, mm
);
986 static inline void paravirt_pgd_free(struct mm_struct
*mm
, pgd_t
*pgd
)
988 PVOP_VCALL2(pv_mmu_ops
.pgd_free
, mm
, pgd
);
991 static inline void paravirt_alloc_pte(struct mm_struct
*mm
, unsigned pfn
)
993 PVOP_VCALL2(pv_mmu_ops
.alloc_pte
, mm
, pfn
);
995 static inline void paravirt_release_pte(unsigned pfn
)
997 PVOP_VCALL1(pv_mmu_ops
.release_pte
, pfn
);
1000 static inline void paravirt_alloc_pmd(struct mm_struct
*mm
, unsigned pfn
)
1002 PVOP_VCALL2(pv_mmu_ops
.alloc_pmd
, mm
, pfn
);
1005 static inline void paravirt_alloc_pmd_clone(unsigned pfn
, unsigned clonepfn
,
1006 unsigned start
, unsigned count
)
1008 PVOP_VCALL4(pv_mmu_ops
.alloc_pmd_clone
, pfn
, clonepfn
, start
, count
);
1010 static inline void paravirt_release_pmd(unsigned pfn
)
1012 PVOP_VCALL1(pv_mmu_ops
.release_pmd
, pfn
);
1015 static inline void paravirt_alloc_pud(struct mm_struct
*mm
, unsigned pfn
)
1017 PVOP_VCALL2(pv_mmu_ops
.alloc_pud
, mm
, pfn
);
1019 static inline void paravirt_release_pud(unsigned pfn
)
1021 PVOP_VCALL1(pv_mmu_ops
.release_pud
, pfn
);
1024 #ifdef CONFIG_HIGHPTE
1025 static inline void *kmap_atomic_pte(struct page
*page
, enum km_type type
)
1028 ret
= PVOP_CALL2(unsigned long, pv_mmu_ops
.kmap_atomic_pte
, page
, type
);
1033 static inline void pte_update(struct mm_struct
*mm
, unsigned long addr
,
1036 PVOP_VCALL3(pv_mmu_ops
.pte_update
, mm
, addr
, ptep
);
1039 static inline void pte_update_defer(struct mm_struct
*mm
, unsigned long addr
,
1042 PVOP_VCALL3(pv_mmu_ops
.pte_update_defer
, mm
, addr
, ptep
);
1045 static inline pte_t
__pte(pteval_t val
)
1049 if (sizeof(pteval_t
) > sizeof(long))
1050 ret
= PVOP_CALL2(pteval_t
,
1051 pv_mmu_ops
.make_pte
,
1052 val
, (u64
)val
>> 32);
1054 ret
= PVOP_CALL1(pteval_t
,
1055 pv_mmu_ops
.make_pte
,
1058 return (pte_t
) { .pte
= ret
};
1061 static inline pteval_t
pte_val(pte_t pte
)
1065 if (sizeof(pteval_t
) > sizeof(long))
1066 ret
= PVOP_CALL2(pteval_t
, pv_mmu_ops
.pte_val
,
1067 pte
.pte
, (u64
)pte
.pte
>> 32);
1069 ret
= PVOP_CALL1(pteval_t
, pv_mmu_ops
.pte_val
,
1075 static inline pteval_t
pte_flags(pte_t pte
)
1079 if (sizeof(pteval_t
) > sizeof(long))
1080 ret
= PVOP_CALL2(pteval_t
, pv_mmu_ops
.pte_flags
,
1081 pte
.pte
, (u64
)pte
.pte
>> 32);
1083 ret
= PVOP_CALL1(pteval_t
, pv_mmu_ops
.pte_flags
,
1089 static inline pgd_t
__pgd(pgdval_t val
)
1093 if (sizeof(pgdval_t
) > sizeof(long))
1094 ret
= PVOP_CALL2(pgdval_t
, pv_mmu_ops
.make_pgd
,
1095 val
, (u64
)val
>> 32);
1097 ret
= PVOP_CALL1(pgdval_t
, pv_mmu_ops
.make_pgd
,
1100 return (pgd_t
) { ret
};
1103 static inline pgdval_t
pgd_val(pgd_t pgd
)
1107 if (sizeof(pgdval_t
) > sizeof(long))
1108 ret
= PVOP_CALL2(pgdval_t
, pv_mmu_ops
.pgd_val
,
1109 pgd
.pgd
, (u64
)pgd
.pgd
>> 32);
1111 ret
= PVOP_CALL1(pgdval_t
, pv_mmu_ops
.pgd_val
,
1117 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1118 static inline pte_t
ptep_modify_prot_start(struct mm_struct
*mm
, unsigned long addr
,
1123 ret
= PVOP_CALL3(pteval_t
, pv_mmu_ops
.ptep_modify_prot_start
,
1126 return (pte_t
) { .pte
= ret
};
1129 static inline void ptep_modify_prot_commit(struct mm_struct
*mm
, unsigned long addr
,
1130 pte_t
*ptep
, pte_t pte
)
1132 if (sizeof(pteval_t
) > sizeof(long))
1134 pv_mmu_ops
.ptep_modify_prot_commit(mm
, addr
, ptep
, pte
);
1136 PVOP_VCALL4(pv_mmu_ops
.ptep_modify_prot_commit
,
1137 mm
, addr
, ptep
, pte
.pte
);
1140 static inline void set_pte(pte_t
*ptep
, pte_t pte
)
1142 if (sizeof(pteval_t
) > sizeof(long))
1143 PVOP_VCALL3(pv_mmu_ops
.set_pte
, ptep
,
1144 pte
.pte
, (u64
)pte
.pte
>> 32);
1146 PVOP_VCALL2(pv_mmu_ops
.set_pte
, ptep
,
1150 static inline void set_pte_at(struct mm_struct
*mm
, unsigned long addr
,
1151 pte_t
*ptep
, pte_t pte
)
1153 if (sizeof(pteval_t
) > sizeof(long))
1155 pv_mmu_ops
.set_pte_at(mm
, addr
, ptep
, pte
);
1157 PVOP_VCALL4(pv_mmu_ops
.set_pte_at
, mm
, addr
, ptep
, pte
.pte
);
1160 static inline void set_pmd(pmd_t
*pmdp
, pmd_t pmd
)
1162 pmdval_t val
= native_pmd_val(pmd
);
1164 if (sizeof(pmdval_t
) > sizeof(long))
1165 PVOP_VCALL3(pv_mmu_ops
.set_pmd
, pmdp
, val
, (u64
)val
>> 32);
1167 PVOP_VCALL2(pv_mmu_ops
.set_pmd
, pmdp
, val
);
1170 #if PAGETABLE_LEVELS >= 3
1171 static inline pmd_t
__pmd(pmdval_t val
)
1175 if (sizeof(pmdval_t
) > sizeof(long))
1176 ret
= PVOP_CALL2(pmdval_t
, pv_mmu_ops
.make_pmd
,
1177 val
, (u64
)val
>> 32);
1179 ret
= PVOP_CALL1(pmdval_t
, pv_mmu_ops
.make_pmd
,
1182 return (pmd_t
) { ret
};
1185 static inline pmdval_t
pmd_val(pmd_t pmd
)
1189 if (sizeof(pmdval_t
) > sizeof(long))
1190 ret
= PVOP_CALL2(pmdval_t
, pv_mmu_ops
.pmd_val
,
1191 pmd
.pmd
, (u64
)pmd
.pmd
>> 32);
1193 ret
= PVOP_CALL1(pmdval_t
, pv_mmu_ops
.pmd_val
,
1199 static inline void set_pud(pud_t
*pudp
, pud_t pud
)
1201 pudval_t val
= native_pud_val(pud
);
1203 if (sizeof(pudval_t
) > sizeof(long))
1204 PVOP_VCALL3(pv_mmu_ops
.set_pud
, pudp
,
1205 val
, (u64
)val
>> 32);
1207 PVOP_VCALL2(pv_mmu_ops
.set_pud
, pudp
,
1210 #if PAGETABLE_LEVELS == 4
1211 static inline pud_t
__pud(pudval_t val
)
1215 if (sizeof(pudval_t
) > sizeof(long))
1216 ret
= PVOP_CALL2(pudval_t
, pv_mmu_ops
.make_pud
,
1217 val
, (u64
)val
>> 32);
1219 ret
= PVOP_CALL1(pudval_t
, pv_mmu_ops
.make_pud
,
1222 return (pud_t
) { ret
};
1225 static inline pudval_t
pud_val(pud_t pud
)
1229 if (sizeof(pudval_t
) > sizeof(long))
1230 ret
= PVOP_CALL2(pudval_t
, pv_mmu_ops
.pud_val
,
1231 pud
.pud
, (u64
)pud
.pud
>> 32);
1233 ret
= PVOP_CALL1(pudval_t
, pv_mmu_ops
.pud_val
,
1239 static inline void set_pgd(pgd_t
*pgdp
, pgd_t pgd
)
1241 pgdval_t val
= native_pgd_val(pgd
);
1243 if (sizeof(pgdval_t
) > sizeof(long))
1244 PVOP_VCALL3(pv_mmu_ops
.set_pgd
, pgdp
,
1245 val
, (u64
)val
>> 32);
1247 PVOP_VCALL2(pv_mmu_ops
.set_pgd
, pgdp
,
1251 static inline void pgd_clear(pgd_t
*pgdp
)
1253 set_pgd(pgdp
, __pgd(0));
1256 static inline void pud_clear(pud_t
*pudp
)
1258 set_pud(pudp
, __pud(0));
1261 #endif /* PAGETABLE_LEVELS == 4 */
1263 #endif /* PAGETABLE_LEVELS >= 3 */
1265 #ifdef CONFIG_X86_PAE
1266 /* Special-case pte-setting operations for PAE, which can't update a
1267 64-bit pte atomically */
1268 static inline void set_pte_atomic(pte_t
*ptep
, pte_t pte
)
1270 PVOP_VCALL3(pv_mmu_ops
.set_pte_atomic
, ptep
,
1271 pte
.pte
, pte
.pte
>> 32);
1274 static inline void set_pte_present(struct mm_struct
*mm
, unsigned long addr
,
1275 pte_t
*ptep
, pte_t pte
)
1278 pv_mmu_ops
.set_pte_present(mm
, addr
, ptep
, pte
);
1281 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
,
1284 PVOP_VCALL3(pv_mmu_ops
.pte_clear
, mm
, addr
, ptep
);
1287 static inline void pmd_clear(pmd_t
*pmdp
)
1289 PVOP_VCALL1(pv_mmu_ops
.pmd_clear
, pmdp
);
1291 #else /* !CONFIG_X86_PAE */
1292 static inline void set_pte_atomic(pte_t
*ptep
, pte_t pte
)
1297 static inline void set_pte_present(struct mm_struct
*mm
, unsigned long addr
,
1298 pte_t
*ptep
, pte_t pte
)
1303 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
,
1306 set_pte_at(mm
, addr
, ptep
, __pte(0));
1309 static inline void pmd_clear(pmd_t
*pmdp
)
1311 set_pmd(pmdp
, __pmd(0));
1313 #endif /* CONFIG_X86_PAE */
1315 /* Lazy mode for batching updates / context switch */
1316 enum paravirt_lazy_mode
{
1322 enum paravirt_lazy_mode
paravirt_get_lazy_mode(void);
1323 void paravirt_enter_lazy_cpu(void);
1324 void paravirt_leave_lazy_cpu(void);
1325 void paravirt_enter_lazy_mmu(void);
1326 void paravirt_leave_lazy_mmu(void);
1327 void paravirt_leave_lazy(enum paravirt_lazy_mode mode
);
1329 #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
1330 static inline void arch_enter_lazy_cpu_mode(void)
1332 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.enter
);
1335 static inline void arch_leave_lazy_cpu_mode(void)
1337 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.leave
);
1340 static inline void arch_flush_lazy_cpu_mode(void)
1342 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
)) {
1343 arch_leave_lazy_cpu_mode();
1344 arch_enter_lazy_cpu_mode();
1349 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1350 static inline void arch_enter_lazy_mmu_mode(void)
1352 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.enter
);
1355 static inline void arch_leave_lazy_mmu_mode(void)
1357 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.leave
);
1360 static inline void arch_flush_lazy_mmu_mode(void)
1362 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
)) {
1363 arch_leave_lazy_mmu_mode();
1364 arch_enter_lazy_mmu_mode();
1368 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx
,
1369 unsigned long phys
, pgprot_t flags
)
1371 pv_mmu_ops
.set_fixmap(idx
, phys
, flags
);
1374 void _paravirt_nop(void);
1375 #define paravirt_nop ((void *)_paravirt_nop)
1377 /* These all sit in the .parainstructions section to tell us what to patch. */
1378 struct paravirt_patch_site
{
1379 u8
*instr
; /* original instructions */
1380 u8 instrtype
; /* type of this instruction */
1381 u8 len
; /* length of original instruction */
1382 u16 clobbers
; /* what registers you may clobber */
1385 extern struct paravirt_patch_site __parainstructions
[],
1386 __parainstructions_end
[];
1388 #ifdef CONFIG_X86_32
1389 #define PV_SAVE_REGS "pushl %%ecx; pushl %%edx;"
1390 #define PV_RESTORE_REGS "popl %%edx; popl %%ecx"
1391 #define PV_FLAGS_ARG "0"
1392 #define PV_EXTRA_CLOBBERS
1393 #define PV_VEXTRA_CLOBBERS
1395 /* We save some registers, but all of them, that's too much. We clobber all
1396 * caller saved registers but the argument parameter */
1397 #define PV_SAVE_REGS "pushq %%rdi;"
1398 #define PV_RESTORE_REGS "popq %%rdi;"
1399 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx"
1400 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx"
1401 #define PV_FLAGS_ARG "D"
1404 static inline unsigned long __raw_local_save_flags(void)
1408 asm volatile(paravirt_alt(PV_SAVE_REGS
1412 : paravirt_type(pv_irq_ops
.save_fl
),
1413 paravirt_clobber(CLBR_EAX
)
1414 : "memory", "cc" PV_VEXTRA_CLOBBERS
);
1418 static inline void raw_local_irq_restore(unsigned long f
)
1420 asm volatile(paravirt_alt(PV_SAVE_REGS
1425 paravirt_type(pv_irq_ops
.restore_fl
),
1426 paravirt_clobber(CLBR_EAX
)
1427 : "memory", "cc" PV_EXTRA_CLOBBERS
);
1430 static inline void raw_local_irq_disable(void)
1432 asm volatile(paravirt_alt(PV_SAVE_REGS
1436 : paravirt_type(pv_irq_ops
.irq_disable
),
1437 paravirt_clobber(CLBR_EAX
)
1438 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1441 static inline void raw_local_irq_enable(void)
1443 asm volatile(paravirt_alt(PV_SAVE_REGS
1447 : paravirt_type(pv_irq_ops
.irq_enable
),
1448 paravirt_clobber(CLBR_EAX
)
1449 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1452 static inline unsigned long __raw_local_irq_save(void)
1456 f
= __raw_local_save_flags();
1457 raw_local_irq_disable();
1461 /* Make sure as little as possible of this mess escapes. */
1462 #undef PARAVIRT_CALL
1476 #else /* __ASSEMBLY__ */
1478 #define _PVSITE(ptype, clobbers, ops, word, algn) \
1482 .pushsection .parainstructions,"a"; \
1491 #ifdef CONFIG_X86_64
1492 #define PV_SAVE_REGS pushq %rax; pushq %rdi; pushq %rcx; pushq %rdx
1493 #define PV_RESTORE_REGS popq %rdx; popq %rcx; popq %rdi; popq %rax
1494 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 8)
1495 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
1496 #define PARA_INDIRECT(addr) *addr(%rip)
1498 #define PV_SAVE_REGS pushl %eax; pushl %edi; pushl %ecx; pushl %edx
1499 #define PV_RESTORE_REGS popl %edx; popl %ecx; popl %edi; popl %eax
1500 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 4)
1501 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
1502 #define PARA_INDIRECT(addr) *%cs:addr
1505 #define INTERRUPT_RETURN \
1506 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE, \
1507 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
1509 #define DISABLE_INTERRUPTS(clobbers) \
1510 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
1512 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable); \
1515 #define ENABLE_INTERRUPTS(clobbers) \
1516 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers, \
1518 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable); \
1521 #define USERGS_SYSRET32 \
1522 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32), \
1524 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
1526 #ifdef CONFIG_X86_32
1527 #define GET_CR0_INTO_EAX \
1528 push %ecx; push %edx; \
1529 call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
1532 #define ENABLE_INTERRUPTS_SYSEXIT \
1533 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
1535 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1538 #else /* !CONFIG_X86_32 */
1541 * If swapgs is used while the userspace stack is still current,
1542 * there's no way to call a pvop. The PV replacement *must* be
1543 * inlined, or the swapgs instruction must be trapped and emulated.
1545 #define SWAPGS_UNSAFE_STACK \
1546 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1550 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1552 call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs); \
1556 #define GET_CR2_INTO_RCX \
1557 call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2); \
1561 #define PARAVIRT_ADJUST_EXCEPTION_FRAME \
1562 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
1564 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
1566 #define USERGS_SYSRET64 \
1567 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64), \
1569 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1571 #define ENABLE_INTERRUPTS_SYSEXIT32 \
1572 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
1574 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1575 #endif /* CONFIG_X86_32 */
1577 #endif /* __ASSEMBLY__ */
1578 #endif /* CONFIG_PARAVIRT */
1579 #endif /* __ASM_PARAVIRT_H */