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. */
18 #include <linux/types.h>
19 #include <linux/cpumask.h>
20 #include <asm/kmap_types.h>
21 #include <asm/desc_defs.h>
32 unsigned int kernel_rpl
;
33 int shared_kernel_pmd
;
40 * Patch may replace one of the defined code sequences with
41 * arbitrary code, subject to the same register constraints.
42 * This generally means the code is not free to clobber any
43 * registers other than EAX. The patch function should return
44 * the number of bytes of code generated, as we nop pad the
45 * rest in generic code.
47 unsigned (*patch
)(u8 type
, u16 clobber
, void *insnbuf
,
48 unsigned long addr
, unsigned len
);
50 /* Basic arch-specific setup */
51 void (*arch_setup
)(void);
52 char *(*memory_setup
)(void);
53 void (*post_allocator_init
)(void);
55 /* Print a banner to identify the environment */
61 /* Set deferred update mode, used for batching operations. */
67 void (*time_init
)(void);
69 /* Set and set time of day */
70 unsigned long (*get_wallclock
)(void);
71 int (*set_wallclock
)(unsigned long);
73 unsigned long long (*sched_clock
)(void);
74 unsigned long (*get_cpu_khz
)(void);
78 /* hooks for various privileged instructions */
79 unsigned long (*get_debugreg
)(int regno
);
80 void (*set_debugreg
)(int regno
, unsigned long value
);
84 unsigned long (*read_cr0
)(void);
85 void (*write_cr0
)(unsigned long);
87 unsigned long (*read_cr4_safe
)(void);
88 unsigned long (*read_cr4
)(void);
89 void (*write_cr4
)(unsigned long);
91 /* Segment descriptor handling */
92 void (*load_tr_desc
)(void);
93 void (*load_gdt
)(const struct desc_ptr
*);
94 void (*load_idt
)(const struct desc_ptr
*);
95 void (*store_gdt
)(struct desc_ptr
*);
96 void (*store_idt
)(struct desc_ptr
*);
97 void (*set_ldt
)(const void *desc
, unsigned entries
);
98 unsigned long (*store_tr
)(void);
99 void (*load_tls
)(struct thread_struct
*t
, unsigned int cpu
);
100 void (*write_ldt_entry
)(struct desc_struct
*ldt
, int entrynum
,
102 void (*write_gdt_entry
)(struct desc_struct
*,
103 int entrynum
, const void *desc
, int size
);
104 void (*write_idt_entry
)(gate_desc
*,
105 int entrynum
, const gate_desc
*gate
);
106 void (*load_sp0
)(struct tss_struct
*tss
, struct thread_struct
*t
);
108 void (*set_iopl_mask
)(unsigned mask
);
110 void (*wbinvd
)(void);
111 void (*io_delay
)(void);
113 /* cpuid emulation, mostly so that caps bits can be disabled */
114 void (*cpuid
)(unsigned int *eax
, unsigned int *ebx
,
115 unsigned int *ecx
, unsigned int *edx
);
117 /* MSR, PMC and TSR operations.
118 err = 0/-EFAULT. wrmsr returns 0/-EFAULT. */
119 u64 (*read_msr
)(unsigned int msr
, int *err
);
120 int (*write_msr
)(unsigned int msr
, unsigned low
, unsigned high
);
122 u64 (*read_tsc
)(void);
123 u64 (*read_pmc
)(int counter
);
124 unsigned long long (*read_tscp
)(unsigned int *aux
);
126 /* These two are jmp to, not actually called. */
127 void (*irq_enable_syscall_ret
)(void);
130 struct pv_lazy_ops lazy_mode
;
134 void (*init_IRQ
)(void);
137 * Get/set interrupt state. save_fl and restore_fl are only
138 * expected to use X86_EFLAGS_IF; all other bits
139 * returned from save_fl are undefined, and may be ignored by
142 unsigned long (*save_fl
)(void);
143 void (*restore_fl
)(unsigned long);
144 void (*irq_disable
)(void);
145 void (*irq_enable
)(void);
146 void (*safe_halt
)(void);
151 #ifdef CONFIG_X86_LOCAL_APIC
153 * Direct APIC operations, principally for VMI. Ideally
154 * these shouldn't be in this interface.
156 void (*apic_write
)(unsigned long reg
, u32 v
);
157 void (*apic_write_atomic
)(unsigned long reg
, u32 v
);
158 u32 (*apic_read
)(unsigned long reg
);
159 void (*setup_boot_clock
)(void);
160 void (*setup_secondary_clock
)(void);
162 void (*startup_ipi_hook
)(int phys_apicid
,
163 unsigned long start_eip
,
164 unsigned long start_esp
);
170 * Called before/after init_mm pagetable setup. setup_start
171 * may reset %cr3, and may pre-install parts of the pagetable;
172 * pagetable setup is expected to preserve any existing
175 void (*pagetable_setup_start
)(pgd_t
*pgd_base
);
176 void (*pagetable_setup_done
)(pgd_t
*pgd_base
);
178 unsigned long (*read_cr2
)(void);
179 void (*write_cr2
)(unsigned long);
181 unsigned long (*read_cr3
)(void);
182 void (*write_cr3
)(unsigned long);
185 * Hooks for intercepting the creation/use/destruction of an
188 void (*activate_mm
)(struct mm_struct
*prev
,
189 struct mm_struct
*next
);
190 void (*dup_mmap
)(struct mm_struct
*oldmm
,
191 struct mm_struct
*mm
);
192 void (*exit_mmap
)(struct mm_struct
*mm
);
196 void (*flush_tlb_user
)(void);
197 void (*flush_tlb_kernel
)(void);
198 void (*flush_tlb_single
)(unsigned long addr
);
199 void (*flush_tlb_others
)(const cpumask_t
*cpus
, struct mm_struct
*mm
,
202 /* Hooks for allocating/releasing pagetable pages */
203 void (*alloc_pt
)(struct mm_struct
*mm
, u32 pfn
);
204 void (*alloc_pd
)(u32 pfn
);
205 void (*alloc_pd_clone
)(u32 pfn
, u32 clonepfn
, u32 start
, u32 count
);
206 void (*release_pt
)(u32 pfn
);
207 void (*release_pd
)(u32 pfn
);
209 /* Pagetable manipulation functions */
210 void (*set_pte
)(pte_t
*ptep
, pte_t pteval
);
211 void (*set_pte_at
)(struct mm_struct
*mm
, unsigned long addr
,
212 pte_t
*ptep
, pte_t pteval
);
213 void (*set_pmd
)(pmd_t
*pmdp
, pmd_t pmdval
);
214 void (*pte_update
)(struct mm_struct
*mm
, unsigned long addr
, pte_t
*ptep
);
215 void (*pte_update_defer
)(struct mm_struct
*mm
,
216 unsigned long addr
, pte_t
*ptep
);
218 #ifdef CONFIG_X86_PAE
219 void (*set_pte_atomic
)(pte_t
*ptep
, pte_t pteval
);
220 void (*set_pte_present
)(struct mm_struct
*mm
, unsigned long addr
,
221 pte_t
*ptep
, pte_t pte
);
222 void (*set_pud
)(pud_t
*pudp
, pud_t pudval
);
223 void (*pte_clear
)(struct mm_struct
*mm
, unsigned long addr
, pte_t
*ptep
);
224 void (*pmd_clear
)(pmd_t
*pmdp
);
226 unsigned long long (*pte_val
)(pte_t
);
227 unsigned long long (*pmd_val
)(pmd_t
);
228 unsigned long long (*pgd_val
)(pgd_t
);
230 pte_t (*make_pte
)(unsigned long long pte
);
231 pmd_t (*make_pmd
)(unsigned long long pmd
);
232 pgd_t (*make_pgd
)(unsigned long long pgd
);
234 unsigned long (*pte_val
)(pte_t
);
235 unsigned long (*pgd_val
)(pgd_t
);
237 pte_t (*make_pte
)(unsigned long pte
);
238 pgd_t (*make_pgd
)(unsigned long pgd
);
241 #ifdef CONFIG_HIGHPTE
242 void *(*kmap_atomic_pte
)(struct page
*page
, enum km_type type
);
245 struct pv_lazy_ops lazy_mode
;
248 /* This contains all the paravirt structures: we get a convenient
249 * number for each function using the offset which we use to indicate
251 struct paravirt_patch_template
253 struct pv_init_ops pv_init_ops
;
254 struct pv_time_ops pv_time_ops
;
255 struct pv_cpu_ops pv_cpu_ops
;
256 struct pv_irq_ops pv_irq_ops
;
257 struct pv_apic_ops pv_apic_ops
;
258 struct pv_mmu_ops pv_mmu_ops
;
261 extern struct pv_info pv_info
;
262 extern struct pv_init_ops pv_init_ops
;
263 extern struct pv_time_ops pv_time_ops
;
264 extern struct pv_cpu_ops pv_cpu_ops
;
265 extern struct pv_irq_ops pv_irq_ops
;
266 extern struct pv_apic_ops pv_apic_ops
;
267 extern struct pv_mmu_ops pv_mmu_ops
;
269 #define PARAVIRT_PATCH(x) \
270 (offsetof(struct paravirt_patch_template, x) / sizeof(void *))
272 #define paravirt_type(op) \
273 [paravirt_typenum] "i" (PARAVIRT_PATCH(op)), \
274 [paravirt_opptr] "m" (op)
275 #define paravirt_clobber(clobber) \
276 [paravirt_clobber] "i" (clobber)
279 * Generate some code, and mark it as patchable by the
280 * apply_paravirt() alternate instruction patcher.
282 #define _paravirt_alt(insn_string, type, clobber) \
283 "771:\n\t" insn_string "\n" "772:\n" \
284 ".pushsection .parainstructions,\"a\"\n" \
287 " .byte " type "\n" \
288 " .byte 772b-771b\n" \
289 " .short " clobber "\n" \
292 /* Generate patchable code, with the default asm parameters. */
293 #define paravirt_alt(insn_string) \
294 _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
296 unsigned paravirt_patch_nop(void);
297 unsigned paravirt_patch_ignore(unsigned len
);
298 unsigned paravirt_patch_call(void *insnbuf
,
299 const void *target
, u16 tgt_clobbers
,
300 unsigned long addr
, u16 site_clobbers
,
302 unsigned paravirt_patch_jmp(void *insnbuf
, const void *target
,
303 unsigned long addr
, unsigned len
);
304 unsigned paravirt_patch_default(u8 type
, u16 clobbers
, void *insnbuf
,
305 unsigned long addr
, unsigned len
);
307 unsigned paravirt_patch_insns(void *insnbuf
, unsigned len
,
308 const char *start
, const char *end
);
310 int paravirt_disable_iospace(void);
313 * This generates an indirect call based on the operation type number.
314 * The type number, computed in PARAVIRT_PATCH, is derived from the
315 * offset into the paravirt_patch_template structure, and can therefore be
316 * freely converted back into a structure offset.
318 #define PARAVIRT_CALL "call *%[paravirt_opptr];"
321 * These macros are intended to wrap calls through one of the paravirt
322 * ops structs, so that they can be later identified and patched at
325 * Normally, a call to a pv_op function is a simple indirect call:
326 * (pv_op_struct.operations)(args...).
328 * Unfortunately, this is a relatively slow operation for modern CPUs,
329 * because it cannot necessarily determine what the destination
330 * address is. In this case, the address is a runtime constant, so at
331 * the very least we can patch the call to e a simple direct call, or
332 * ideally, patch an inline implementation into the callsite. (Direct
333 * calls are essentially free, because the call and return addresses
334 * are completely predictable.)
336 * For i386, these macros rely on the standard gcc "regparm(3)" calling
337 * convention, in which the first three arguments are placed in %eax,
338 * %edx, %ecx (in that order), and the remaining arguments are placed
339 * on the stack. All caller-save registers (eax,edx,ecx) are expected
340 * to be modified (either clobbered or used for return values).
341 * X86_64, on the other hand, already specifies a register-based calling
342 * conventions, returning at %rax, with parameteres going on %rdi, %rsi,
343 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
344 * special handling for dealing with 4 arguments, unlike i386.
345 * However, x86_64 also have to clobber all caller saved registers, which
346 * unfortunately, are quite a bit (r8 - r11)
348 * The call instruction itself is marked by placing its start address
349 * and size into the .parainstructions section, so that
350 * apply_paravirt() in arch/i386/kernel/alternative.c can do the
351 * appropriate patching under the control of the backend pv_init_ops
354 * Unfortunately there's no way to get gcc to generate the args setup
355 * for the call, and then allow the call itself to be generated by an
356 * inline asm. Because of this, we must do the complete arg setup and
357 * return value handling from within these macros. This is fairly
360 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
361 * It could be extended to more arguments, but there would be little
362 * to be gained from that. For each number of arguments, there are
363 * the two VCALL and CALL variants for void and non-void functions.
365 * When there is a return value, the invoker of the macro must specify
366 * the return type. The macro then uses sizeof() on that type to
367 * determine whether its a 32 or 64 bit value, and places the return
368 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
369 * 64-bit). For x86_64 machines, it just returns at %rax regardless of
370 * the return value size.
372 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
373 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
376 * Small structures are passed and returned in registers. The macro
377 * calling convention can't directly deal with this, so the wrapper
378 * functions must do this.
380 * These PVOP_* macros are only defined within this header. This
381 * means that all uses must be wrapped in inline functions. This also
382 * makes sure the incoming and outgoing types are always correct.
385 #define PVOP_VCALL_ARGS unsigned long __eax, __edx, __ecx
386 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS
387 #define PVOP_VCALL_CLOBBERS "=a" (__eax), "=d" (__edx), \
389 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS
390 #define EXTRA_CLOBBERS
391 #define VEXTRA_CLOBBERS
393 #define PVOP_VCALL_ARGS unsigned long __edi, __esi, __edx, __ecx
394 #define PVOP_CALL_ARGS PVOP_VCALL_ARGS, __eax
395 #define PVOP_VCALL_CLOBBERS "=D" (__edi), \
396 "=S" (__esi), "=d" (__edx), \
399 #define PVOP_CALL_CLOBBERS PVOP_VCALL_CLOBBERS, "=a" (__eax)
401 #define EXTRA_CLOBBERS , "r8", "r9", "r10", "r11"
402 #define VEXTRA_CLOBBERS , "rax", "r8", "r9", "r10", "r11"
405 #define __PVOP_CALL(rettype, op, pre, post, ...) \
409 /* This is 32-bit specific, but is okay in 64-bit */ \
410 /* since this condition will never hold */ \
411 if (sizeof(rettype) > sizeof(unsigned long)) { \
413 paravirt_alt(PARAVIRT_CALL) \
415 : PVOP_CALL_CLOBBERS \
416 : paravirt_type(op), \
417 paravirt_clobber(CLBR_ANY), \
419 : "memory", "cc" EXTRA_CLOBBERS); \
420 __ret = (rettype)((((u64)__edx) << 32) | __eax); \
423 paravirt_alt(PARAVIRT_CALL) \
425 : PVOP_CALL_CLOBBERS \
426 : paravirt_type(op), \
427 paravirt_clobber(CLBR_ANY), \
429 : "memory", "cc" EXTRA_CLOBBERS); \
430 __ret = (rettype)__eax; \
434 #define __PVOP_VCALL(op, pre, post, ...) \
438 paravirt_alt(PARAVIRT_CALL) \
440 : PVOP_VCALL_CLOBBERS \
441 : paravirt_type(op), \
442 paravirt_clobber(CLBR_ANY), \
444 : "memory", "cc" VEXTRA_CLOBBERS); \
447 #define PVOP_CALL0(rettype, op) \
448 __PVOP_CALL(rettype, op, "", "")
449 #define PVOP_VCALL0(op) \
450 __PVOP_VCALL(op, "", "")
452 #define PVOP_CALL1(rettype, op, arg1) \
453 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)))
454 #define PVOP_VCALL1(op, arg1) \
455 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)))
457 #define PVOP_CALL2(rettype, op, arg1, arg2) \
458 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
459 "1" ((unsigned long)(arg2)))
460 #define PVOP_VCALL2(op, arg1, arg2) \
461 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
462 "1" ((unsigned long)(arg2)))
464 #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \
465 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
466 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
467 #define PVOP_VCALL3(op, arg1, arg2, arg3) \
468 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
469 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)))
471 /* This is the only difference in x86_64. We can make it much simpler */
473 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
474 __PVOP_CALL(rettype, op, \
475 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
476 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
477 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
478 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
480 "push %[_arg4];", "lea 4(%%esp),%%esp;", \
481 "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
482 "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
484 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
485 __PVOP_CALL(rettype, op, "", "", "0" ((unsigned long)(arg1)), \
486 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
487 "3"((unsigned long)(arg4)))
488 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
489 __PVOP_VCALL(op, "", "", "0" ((unsigned long)(arg1)), \
490 "1"((unsigned long)(arg2)), "2"((unsigned long)(arg3)), \
491 "3"((unsigned long)(arg4)))
494 static inline int paravirt_enabled(void)
496 return pv_info
.paravirt_enabled
;
499 static inline void load_sp0(struct tss_struct
*tss
,
500 struct thread_struct
*thread
)
502 PVOP_VCALL2(pv_cpu_ops
.load_sp0
, tss
, thread
);
505 #define ARCH_SETUP pv_init_ops.arch_setup();
506 static inline unsigned long get_wallclock(void)
508 return PVOP_CALL0(unsigned long, pv_time_ops
.get_wallclock
);
511 static inline int set_wallclock(unsigned long nowtime
)
513 return PVOP_CALL1(int, pv_time_ops
.set_wallclock
, nowtime
);
516 static inline void (*choose_time_init(void))(void)
518 return pv_time_ops
.time_init
;
521 /* The paravirtualized CPUID instruction. */
522 static inline void __cpuid(unsigned int *eax
, unsigned int *ebx
,
523 unsigned int *ecx
, unsigned int *edx
)
525 PVOP_VCALL4(pv_cpu_ops
.cpuid
, eax
, ebx
, ecx
, edx
);
529 * These special macros can be used to get or set a debugging register
531 static inline unsigned long paravirt_get_debugreg(int reg
)
533 return PVOP_CALL1(unsigned long, pv_cpu_ops
.get_debugreg
, reg
);
535 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
536 static inline void set_debugreg(unsigned long val
, int reg
)
538 PVOP_VCALL2(pv_cpu_ops
.set_debugreg
, reg
, val
);
541 static inline void clts(void)
543 PVOP_VCALL0(pv_cpu_ops
.clts
);
546 static inline unsigned long read_cr0(void)
548 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr0
);
551 static inline void write_cr0(unsigned long x
)
553 PVOP_VCALL1(pv_cpu_ops
.write_cr0
, x
);
556 static inline unsigned long read_cr2(void)
558 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr2
);
561 static inline void write_cr2(unsigned long x
)
563 PVOP_VCALL1(pv_mmu_ops
.write_cr2
, x
);
566 static inline unsigned long read_cr3(void)
568 return PVOP_CALL0(unsigned long, pv_mmu_ops
.read_cr3
);
571 static inline void write_cr3(unsigned long x
)
573 PVOP_VCALL1(pv_mmu_ops
.write_cr3
, x
);
576 static inline unsigned long read_cr4(void)
578 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4
);
580 static inline unsigned long read_cr4_safe(void)
582 return PVOP_CALL0(unsigned long, pv_cpu_ops
.read_cr4_safe
);
585 static inline void write_cr4(unsigned long x
)
587 PVOP_VCALL1(pv_cpu_ops
.write_cr4
, x
);
590 static inline void raw_safe_halt(void)
592 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
595 static inline void halt(void)
597 PVOP_VCALL0(pv_irq_ops
.safe_halt
);
600 static inline void wbinvd(void)
602 PVOP_VCALL0(pv_cpu_ops
.wbinvd
);
605 #define get_kernel_rpl() (pv_info.kernel_rpl)
607 static inline u64
paravirt_read_msr(unsigned msr
, int *err
)
609 return PVOP_CALL2(u64
, pv_cpu_ops
.read_msr
, msr
, err
);
611 static inline int paravirt_write_msr(unsigned msr
, unsigned low
, unsigned high
)
613 return PVOP_CALL3(int, pv_cpu_ops
.write_msr
, msr
, low
, high
);
616 /* These should all do BUG_ON(_err), but our headers are too tangled. */
617 #define rdmsr(msr,val1,val2) do { \
619 u64 _l = paravirt_read_msr(msr, &_err); \
624 #define wrmsr(msr,val1,val2) do { \
625 paravirt_write_msr(msr, val1, val2); \
628 #define rdmsrl(msr,val) do { \
630 val = paravirt_read_msr(msr, &_err); \
633 #define wrmsrl(msr,val) wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
634 #define wrmsr_safe(msr,a,b) paravirt_write_msr(msr, a, b)
636 /* rdmsr with exception handling */
637 #define rdmsr_safe(msr,a,b) ({ \
639 u64 _l = paravirt_read_msr(msr, &_err); \
645 static inline u64
paravirt_read_tsc(void)
647 return PVOP_CALL0(u64
, pv_cpu_ops
.read_tsc
);
650 #define rdtscl(low) do { \
651 u64 _l = paravirt_read_tsc(); \
655 #define rdtscll(val) (val = paravirt_read_tsc())
657 static inline unsigned long long paravirt_sched_clock(void)
659 return PVOP_CALL0(unsigned long long, pv_time_ops
.sched_clock
);
661 #define calculate_cpu_khz() (pv_time_ops.get_cpu_khz())
663 static inline unsigned long long paravirt_read_pmc(int counter
)
665 return PVOP_CALL1(u64
, pv_cpu_ops
.read_pmc
, counter
);
668 #define rdpmc(counter,low,high) do { \
669 u64 _l = paravirt_read_pmc(counter); \
674 static inline unsigned long long paravirt_rdtscp(unsigned int *aux
)
676 return PVOP_CALL1(u64
, pv_cpu_ops
.read_tscp
, aux
);
679 #define rdtscp(low, high, aux) \
682 unsigned long __val = paravirt_rdtscp(&__aux); \
683 (low) = (u32)__val; \
684 (high) = (u32)(__val >> 32); \
688 #define rdtscpll(val, aux) \
690 unsigned long __aux; \
691 val = paravirt_rdtscp(&__aux); \
695 static inline void load_TR_desc(void)
697 PVOP_VCALL0(pv_cpu_ops
.load_tr_desc
);
699 static inline void load_gdt(const struct desc_ptr
*dtr
)
701 PVOP_VCALL1(pv_cpu_ops
.load_gdt
, dtr
);
703 static inline void load_idt(const struct desc_ptr
*dtr
)
705 PVOP_VCALL1(pv_cpu_ops
.load_idt
, dtr
);
707 static inline void set_ldt(const void *addr
, unsigned entries
)
709 PVOP_VCALL2(pv_cpu_ops
.set_ldt
, addr
, entries
);
711 static inline void store_gdt(struct desc_ptr
*dtr
)
713 PVOP_VCALL1(pv_cpu_ops
.store_gdt
, dtr
);
715 static inline void store_idt(struct desc_ptr
*dtr
)
717 PVOP_VCALL1(pv_cpu_ops
.store_idt
, dtr
);
719 static inline unsigned long paravirt_store_tr(void)
721 return PVOP_CALL0(unsigned long, pv_cpu_ops
.store_tr
);
723 #define store_tr(tr) ((tr) = paravirt_store_tr())
724 static inline void load_TLS(struct thread_struct
*t
, unsigned cpu
)
726 PVOP_VCALL2(pv_cpu_ops
.load_tls
, t
, cpu
);
729 static inline void write_ldt_entry(struct desc_struct
*dt
, int entry
,
732 PVOP_VCALL3(pv_cpu_ops
.write_ldt_entry
, dt
, entry
, desc
);
735 static inline void write_gdt_entry(struct desc_struct
*dt
, int entry
,
736 void *desc
, int type
)
738 PVOP_VCALL4(pv_cpu_ops
.write_gdt_entry
, dt
, entry
, desc
, type
);
741 static inline void write_idt_entry(gate_desc
*dt
, int entry
, const gate_desc
*g
)
743 PVOP_VCALL3(pv_cpu_ops
.write_idt_entry
, dt
, entry
, g
);
745 static inline void set_iopl_mask(unsigned mask
)
747 PVOP_VCALL1(pv_cpu_ops
.set_iopl_mask
, mask
);
750 /* The paravirtualized I/O functions */
751 static inline void slow_down_io(void) {
752 pv_cpu_ops
.io_delay();
753 #ifdef REALLY_SLOW_IO
754 pv_cpu_ops
.io_delay();
755 pv_cpu_ops
.io_delay();
756 pv_cpu_ops
.io_delay();
760 #ifdef CONFIG_X86_LOCAL_APIC
762 * Basic functions accessing APICs.
764 static inline void apic_write(unsigned long reg
, u32 v
)
766 PVOP_VCALL2(pv_apic_ops
.apic_write
, reg
, v
);
769 static inline void apic_write_atomic(unsigned long reg
, u32 v
)
771 PVOP_VCALL2(pv_apic_ops
.apic_write_atomic
, reg
, v
);
774 static inline u32
apic_read(unsigned long reg
)
776 return PVOP_CALL1(unsigned long, pv_apic_ops
.apic_read
, reg
);
779 static inline void setup_boot_clock(void)
781 PVOP_VCALL0(pv_apic_ops
.setup_boot_clock
);
784 static inline void setup_secondary_clock(void)
786 PVOP_VCALL0(pv_apic_ops
.setup_secondary_clock
);
790 static inline void paravirt_post_allocator_init(void)
792 if (pv_init_ops
.post_allocator_init
)
793 (*pv_init_ops
.post_allocator_init
)();
796 static inline void paravirt_pagetable_setup_start(pgd_t
*base
)
798 (*pv_mmu_ops
.pagetable_setup_start
)(base
);
801 static inline void paravirt_pagetable_setup_done(pgd_t
*base
)
803 (*pv_mmu_ops
.pagetable_setup_done
)(base
);
807 static inline void startup_ipi_hook(int phys_apicid
, unsigned long start_eip
,
808 unsigned long start_esp
)
810 PVOP_VCALL3(pv_apic_ops
.startup_ipi_hook
,
811 phys_apicid
, start_eip
, start_esp
);
815 static inline void paravirt_activate_mm(struct mm_struct
*prev
,
816 struct mm_struct
*next
)
818 PVOP_VCALL2(pv_mmu_ops
.activate_mm
, prev
, next
);
821 static inline void arch_dup_mmap(struct mm_struct
*oldmm
,
822 struct mm_struct
*mm
)
824 PVOP_VCALL2(pv_mmu_ops
.dup_mmap
, oldmm
, mm
);
827 static inline void arch_exit_mmap(struct mm_struct
*mm
)
829 PVOP_VCALL1(pv_mmu_ops
.exit_mmap
, mm
);
832 static inline void __flush_tlb(void)
834 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_user
);
836 static inline void __flush_tlb_global(void)
838 PVOP_VCALL0(pv_mmu_ops
.flush_tlb_kernel
);
840 static inline void __flush_tlb_single(unsigned long addr
)
842 PVOP_VCALL1(pv_mmu_ops
.flush_tlb_single
, addr
);
845 static inline void flush_tlb_others(cpumask_t cpumask
, struct mm_struct
*mm
,
848 PVOP_VCALL3(pv_mmu_ops
.flush_tlb_others
, &cpumask
, mm
, va
);
851 static inline void paravirt_alloc_pt(struct mm_struct
*mm
, unsigned pfn
)
853 PVOP_VCALL2(pv_mmu_ops
.alloc_pt
, mm
, pfn
);
855 static inline void paravirt_release_pt(unsigned pfn
)
857 PVOP_VCALL1(pv_mmu_ops
.release_pt
, pfn
);
860 static inline void paravirt_alloc_pd(unsigned pfn
)
862 PVOP_VCALL1(pv_mmu_ops
.alloc_pd
, pfn
);
865 static inline void paravirt_alloc_pd_clone(unsigned pfn
, unsigned clonepfn
,
866 unsigned start
, unsigned count
)
868 PVOP_VCALL4(pv_mmu_ops
.alloc_pd_clone
, pfn
, clonepfn
, start
, count
);
870 static inline void paravirt_release_pd(unsigned pfn
)
872 PVOP_VCALL1(pv_mmu_ops
.release_pd
, pfn
);
875 #ifdef CONFIG_HIGHPTE
876 static inline void *kmap_atomic_pte(struct page
*page
, enum km_type type
)
879 ret
= PVOP_CALL2(unsigned long, pv_mmu_ops
.kmap_atomic_pte
, page
, type
);
884 static inline void pte_update(struct mm_struct
*mm
, unsigned long addr
,
887 PVOP_VCALL3(pv_mmu_ops
.pte_update
, mm
, addr
, ptep
);
890 static inline void pte_update_defer(struct mm_struct
*mm
, unsigned long addr
,
893 PVOP_VCALL3(pv_mmu_ops
.pte_update_defer
, mm
, addr
, ptep
);
896 #ifdef CONFIG_X86_PAE
897 static inline pte_t
__pte(unsigned long long val
)
899 unsigned long long ret
= PVOP_CALL2(unsigned long long,
902 return (pte_t
) { ret
, ret
>> 32 };
905 static inline pmd_t
__pmd(unsigned long long val
)
907 return (pmd_t
) { PVOP_CALL2(unsigned long long, pv_mmu_ops
.make_pmd
,
911 static inline pgd_t
__pgd(unsigned long long val
)
913 return (pgd_t
) { PVOP_CALL2(unsigned long long, pv_mmu_ops
.make_pgd
,
917 static inline unsigned long long pte_val(pte_t x
)
919 return PVOP_CALL2(unsigned long long, pv_mmu_ops
.pte_val
,
920 x
.pte_low
, x
.pte_high
);
923 static inline unsigned long long pmd_val(pmd_t x
)
925 return PVOP_CALL2(unsigned long long, pv_mmu_ops
.pmd_val
,
929 static inline unsigned long long pgd_val(pgd_t x
)
931 return PVOP_CALL2(unsigned long long, pv_mmu_ops
.pgd_val
,
935 static inline void set_pte(pte_t
*ptep
, pte_t pteval
)
937 PVOP_VCALL3(pv_mmu_ops
.set_pte
, ptep
, pteval
.pte_low
, pteval
.pte_high
);
940 static inline void set_pte_at(struct mm_struct
*mm
, unsigned long addr
,
941 pte_t
*ptep
, pte_t pteval
)
944 pv_mmu_ops
.set_pte_at(mm
, addr
, ptep
, pteval
);
947 static inline void set_pte_atomic(pte_t
*ptep
, pte_t pteval
)
949 PVOP_VCALL3(pv_mmu_ops
.set_pte_atomic
, ptep
,
950 pteval
.pte_low
, pteval
.pte_high
);
953 static inline void set_pte_present(struct mm_struct
*mm
, unsigned long addr
,
954 pte_t
*ptep
, pte_t pte
)
957 pv_mmu_ops
.set_pte_present(mm
, addr
, ptep
, pte
);
960 static inline void set_pmd(pmd_t
*pmdp
, pmd_t pmdval
)
962 PVOP_VCALL3(pv_mmu_ops
.set_pmd
, pmdp
,
963 pmdval
.pmd
, pmdval
.pmd
>> 32);
966 static inline void set_pud(pud_t
*pudp
, pud_t pudval
)
968 PVOP_VCALL3(pv_mmu_ops
.set_pud
, pudp
,
969 pudval
.pgd
.pgd
, pudval
.pgd
.pgd
>> 32);
972 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
, pte_t
*ptep
)
974 PVOP_VCALL3(pv_mmu_ops
.pte_clear
, mm
, addr
, ptep
);
977 static inline void pmd_clear(pmd_t
*pmdp
)
979 PVOP_VCALL1(pv_mmu_ops
.pmd_clear
, pmdp
);
982 #else /* !CONFIG_X86_PAE */
984 static inline pte_t
__pte(unsigned long val
)
986 return (pte_t
) { PVOP_CALL1(unsigned long, pv_mmu_ops
.make_pte
, val
) };
989 static inline pgd_t
__pgd(unsigned long val
)
991 return (pgd_t
) { PVOP_CALL1(unsigned long, pv_mmu_ops
.make_pgd
, val
) };
994 static inline unsigned long pte_val(pte_t x
)
996 return PVOP_CALL1(unsigned long, pv_mmu_ops
.pte_val
, x
.pte_low
);
999 static inline unsigned long pgd_val(pgd_t x
)
1001 return PVOP_CALL1(unsigned long, pv_mmu_ops
.pgd_val
, x
.pgd
);
1004 static inline void set_pte(pte_t
*ptep
, pte_t pteval
)
1006 PVOP_VCALL2(pv_mmu_ops
.set_pte
, ptep
, pteval
.pte_low
);
1009 static inline void set_pte_at(struct mm_struct
*mm
, unsigned long addr
,
1010 pte_t
*ptep
, pte_t pteval
)
1012 PVOP_VCALL4(pv_mmu_ops
.set_pte_at
, mm
, addr
, ptep
, pteval
.pte_low
);
1015 static inline void set_pmd(pmd_t
*pmdp
, pmd_t pmdval
)
1017 PVOP_VCALL2(pv_mmu_ops
.set_pmd
, pmdp
, pmdval
.pud
.pgd
.pgd
);
1019 #endif /* CONFIG_X86_PAE */
1021 /* Lazy mode for batching updates / context switch */
1022 enum paravirt_lazy_mode
{
1028 enum paravirt_lazy_mode
paravirt_get_lazy_mode(void);
1029 void paravirt_enter_lazy_cpu(void);
1030 void paravirt_leave_lazy_cpu(void);
1031 void paravirt_enter_lazy_mmu(void);
1032 void paravirt_leave_lazy_mmu(void);
1033 void paravirt_leave_lazy(enum paravirt_lazy_mode mode
);
1035 #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
1036 static inline void arch_enter_lazy_cpu_mode(void)
1038 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.enter
);
1041 static inline void arch_leave_lazy_cpu_mode(void)
1043 PVOP_VCALL0(pv_cpu_ops
.lazy_mode
.leave
);
1046 static inline void arch_flush_lazy_cpu_mode(void)
1048 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
)) {
1049 arch_leave_lazy_cpu_mode();
1050 arch_enter_lazy_cpu_mode();
1055 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1056 static inline void arch_enter_lazy_mmu_mode(void)
1058 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.enter
);
1061 static inline void arch_leave_lazy_mmu_mode(void)
1063 PVOP_VCALL0(pv_mmu_ops
.lazy_mode
.leave
);
1066 static inline void arch_flush_lazy_mmu_mode(void)
1068 if (unlikely(paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
)) {
1069 arch_leave_lazy_mmu_mode();
1070 arch_enter_lazy_mmu_mode();
1074 void _paravirt_nop(void);
1075 #define paravirt_nop ((void *)_paravirt_nop)
1077 /* These all sit in the .parainstructions section to tell us what to patch. */
1078 struct paravirt_patch_site
{
1079 u8
*instr
; /* original instructions */
1080 u8 instrtype
; /* type of this instruction */
1081 u8 len
; /* length of original instruction */
1082 u16 clobbers
; /* what registers you may clobber */
1085 extern struct paravirt_patch_site __parainstructions
[],
1086 __parainstructions_end
[];
1088 #ifdef CONFIG_X86_32
1089 #define PV_SAVE_REGS "pushl %%ecx; pushl %%edx;"
1090 #define PV_RESTORE_REGS "popl %%edx; popl %%ecx"
1091 #define PV_FLAGS_ARG "0"
1092 #define PV_EXTRA_CLOBBERS
1093 #define PV_VEXTRA_CLOBBERS
1095 /* We save some registers, but all of them, that's too much. We clobber all
1096 * caller saved registers but the argument parameter */
1097 #define PV_SAVE_REGS "pushq %%rdi;"
1098 #define PV_RESTORE_REGS "popq %%rdi;"
1099 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx"
1100 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx"
1101 #define PV_FLAGS_ARG "D"
1104 static inline unsigned long __raw_local_save_flags(void)
1108 asm volatile(paravirt_alt(PV_SAVE_REGS
1112 : paravirt_type(pv_irq_ops
.save_fl
),
1113 paravirt_clobber(CLBR_EAX
)
1114 : "memory", "cc" PV_VEXTRA_CLOBBERS
);
1118 static inline void raw_local_irq_restore(unsigned long f
)
1120 asm volatile(paravirt_alt(PV_SAVE_REGS
1125 paravirt_type(pv_irq_ops
.restore_fl
),
1126 paravirt_clobber(CLBR_EAX
)
1127 : "memory", "cc" PV_EXTRA_CLOBBERS
);
1130 static inline void raw_local_irq_disable(void)
1132 asm volatile(paravirt_alt(PV_SAVE_REGS
1136 : paravirt_type(pv_irq_ops
.irq_disable
),
1137 paravirt_clobber(CLBR_EAX
)
1138 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1141 static inline void raw_local_irq_enable(void)
1143 asm volatile(paravirt_alt(PV_SAVE_REGS
1147 : paravirt_type(pv_irq_ops
.irq_enable
),
1148 paravirt_clobber(CLBR_EAX
)
1149 : "memory", "eax", "cc" PV_EXTRA_CLOBBERS
);
1152 static inline unsigned long __raw_local_irq_save(void)
1156 f
= __raw_local_save_flags();
1157 raw_local_irq_disable();
1161 /* Make sure as little as possible of this mess escapes. */
1162 #undef PARAVIRT_CALL
1176 #else /* __ASSEMBLY__ */
1178 #define _PVSITE(ptype, clobbers, ops, word, algn) \
1182 .pushsection .parainstructions,"a"; \
1191 #ifdef CONFIG_X86_64
1192 #define PV_SAVE_REGS pushq %rax; pushq %rdi; pushq %rcx; pushq %rdx
1193 #define PV_RESTORE_REGS popq %rdx; popq %rcx; popq %rdi; popq %rax
1194 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 8)
1195 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
1197 #define PV_SAVE_REGS pushl %eax; pushl %edi; pushl %ecx; pushl %edx
1198 #define PV_RESTORE_REGS popl %edx; popl %ecx; popl %edi; popl %eax
1199 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 4)
1200 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
1203 #define INTERRUPT_RETURN \
1204 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE, \
1205 jmp *%cs:pv_cpu_ops+PV_CPU_iret)
1207 #define DISABLE_INTERRUPTS(clobbers) \
1208 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
1210 call *%cs:pv_irq_ops+PV_IRQ_irq_disable; \
1213 #define ENABLE_INTERRUPTS(clobbers) \
1214 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers, \
1216 call *%cs:pv_irq_ops+PV_IRQ_irq_enable; \
1219 #define ENABLE_INTERRUPTS_SYSCALL_RET \
1220 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_syscall_ret),\
1222 jmp *%cs:pv_cpu_ops+PV_CPU_irq_enable_syscall_ret)
1225 #ifdef CONFIG_X86_32
1226 #define GET_CR0_INTO_EAX \
1227 push %ecx; push %edx; \
1228 call *pv_cpu_ops+PV_CPU_read_cr0; \
1231 #define GET_CR2_INTO_RCX \
1232 call *pv_mmu_ops+PV_MMU_read_cr2; \
1238 #endif /* __ASSEMBLY__ */
1239 #endif /* CONFIG_PARAVIRT */
1240 #endif /* __ASM_PARAVIRT_H */