4 #include <linux/kernel.h>
5 #include <asm/segment.h>
6 #include <asm/alternative.h>
11 #define STR(x) __STR(x)
13 #define __SAVE(reg,offset) "movq %%" #reg ",(14-" #offset ")*8(%%rsp)\n\t"
14 #define __RESTORE(reg,offset) "movq (14-" #offset ")*8(%%rsp),%%" #reg "\n\t"
16 /* frame pointer must be last for get_wchan */
17 #define SAVE_CONTEXT "pushf ; pushq %%rbp ; movq %%rsi,%%rbp\n\t"
18 #define RESTORE_CONTEXT "movq %%rbp,%%rsi ; popq %%rbp ; popf\t"
20 #define __EXTRA_CLOBBER \
21 ,"rcx","rbx","rdx","r8","r9","r10","r11","r12","r13","r14","r15"
23 /* Save restore flags to clear handle leaking NT */
24 #define switch_to(prev,next,last) \
25 asm volatile(SAVE_CONTEXT \
26 "movq %%rsp,%P[threadrsp](%[prev])\n\t" /* save RSP */ \
27 "movq %P[threadrsp](%[next]),%%rsp\n\t" /* restore RSP */ \
28 "call __switch_to\n\t" \
29 ".globl thread_return\n" \
30 "thread_return:\n\t" \
31 "movq %%gs:%P[pda_pcurrent],%%rsi\n\t" \
32 "movq %P[thread_info](%%rsi),%%r8\n\t" \
33 LOCK_PREFIX "btr %[tif_fork],%P[ti_flags](%%r8)\n\t" \
34 "movq %%rax,%%rdi\n\t" \
35 "jc ret_from_fork\n\t" \
38 : [next] "S" (next), [prev] "D" (prev), \
39 [threadrsp] "i" (offsetof(struct task_struct, thread.rsp)), \
40 [ti_flags] "i" (offsetof(struct thread_info, flags)),\
41 [tif_fork] "i" (TIF_FORK), \
42 [thread_info] "i" (offsetof(struct task_struct, thread_info)), \
43 [pda_pcurrent] "i" (offsetof(struct x8664_pda, pcurrent)) \
44 : "memory", "cc" __EXTRA_CLOBBER)
46 extern void load_gs_index(unsigned);
49 * Load a segment. Fall back on loading the zero
50 * segment if something goes wrong..
52 #define loadsegment(seg,value) \
55 "movl %k0,%%" #seg "\n" \
57 ".section .fixup,\"ax\"\n" \
59 "movl %1,%%" #seg "\n\t" \
62 ".section __ex_table,\"a\"\n\t" \
66 : :"r" (value), "r" (0))
69 * Clear and set 'TS' bit respectively
71 #define clts() __asm__ __volatile__ ("clts")
73 static inline unsigned long read_cr0(void)
76 asm volatile("movq %%cr0,%0" : "=r" (cr0));
80 static inline void write_cr0(unsigned long val)
82 asm volatile("movq %0,%%cr0" :: "r" (val));
85 static inline unsigned long read_cr3(void)
88 asm("movq %%cr3,%0" : "=r" (cr3));
92 static inline unsigned long read_cr4(void)
95 asm("movq %%cr4,%0" : "=r" (cr4));
99 static inline void write_cr4(unsigned long val)
101 asm volatile("movq %0,%%cr4" :: "r" (val));
104 #define stts() write_cr0(8 | read_cr0())
107 __asm__ __volatile__ ("wbinvd": : :"memory");
110 * On SMP systems, when the scheduler does migration-cost autodetection,
111 * it needs a way to flush as much of the CPU's caches as possible.
113 static inline void sched_cacheflush(void)
118 #endif /* __KERNEL__ */
120 #define nop() __asm__ __volatile__ ("nop")
122 #define xchg(ptr,v) ((__typeof__(*(ptr)))__xchg((unsigned long)(v),(ptr),sizeof(*(ptr))))
124 #define tas(ptr) (xchg((ptr),1))
126 #define __xg(x) ((volatile long *)(x))
128 static inline void set_64bit(volatile unsigned long *ptr, unsigned long val)
133 #define _set_64bit set_64bit
136 * Note: no "lock" prefix even on SMP: xchg always implies lock anyway
137 * Note 2: xchg has side effect, so that attribute volatile is necessary,
138 * but generally the primitive is invalid, *ptr is output argument. --ANK
140 static inline unsigned long __xchg(unsigned long x, volatile void * ptr, int size)
144 __asm__ __volatile__("xchgb %b0,%1"
146 :"m" (*__xg(ptr)), "0" (x)
150 __asm__ __volatile__("xchgw %w0,%1"
152 :"m" (*__xg(ptr)), "0" (x)
156 __asm__ __volatile__("xchgl %k0,%1"
158 :"m" (*__xg(ptr)), "0" (x)
162 __asm__ __volatile__("xchgq %0,%1"
164 :"m" (*__xg(ptr)), "0" (x)
172 * Atomic compare and exchange. Compare OLD with MEM, if identical,
173 * store NEW in MEM. Return the initial value in MEM. Success is
174 * indicated by comparing RETURN with OLD.
177 #define __HAVE_ARCH_CMPXCHG 1
179 static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old,
180 unsigned long new, int size)
185 __asm__ __volatile__(LOCK_PREFIX "cmpxchgb %b1,%2"
187 : "q"(new), "m"(*__xg(ptr)), "0"(old)
191 __asm__ __volatile__(LOCK_PREFIX "cmpxchgw %w1,%2"
193 : "r"(new), "m"(*__xg(ptr)), "0"(old)
197 __asm__ __volatile__(LOCK_PREFIX "cmpxchgl %k1,%2"
199 : "r"(new), "m"(*__xg(ptr)), "0"(old)
203 __asm__ __volatile__(LOCK_PREFIX "cmpxchgq %1,%2"
205 : "r"(new), "m"(*__xg(ptr)), "0"(old)
212 #define cmpxchg(ptr,o,n)\
213 ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(o),\
214 (unsigned long)(n),sizeof(*(ptr))))
217 #define smp_mb() mb()
218 #define smp_rmb() rmb()
219 #define smp_wmb() wmb()
220 #define smp_read_barrier_depends() do {} while(0)
222 #define smp_mb() barrier()
223 #define smp_rmb() barrier()
224 #define smp_wmb() barrier()
225 #define smp_read_barrier_depends() do {} while(0)
230 * Force strict CPU ordering.
231 * And yes, this is required on UP too when we're talking
234 #define mb() asm volatile("mfence":::"memory")
235 #define rmb() asm volatile("lfence":::"memory")
237 #ifdef CONFIG_UNORDERED_IO
238 #define wmb() asm volatile("sfence" ::: "memory")
240 #define wmb() asm volatile("" ::: "memory")
242 #define read_barrier_depends() do {} while(0)
243 #define set_mb(var, value) do { (void) xchg(&var, value); } while (0)
245 #define warn_if_not_ulong(x) do { unsigned long foo; (void) (&(x) == &foo); } while (0)
247 #include <linux/irqflags.h>
249 void cpu_idle_wait(void);
251 extern unsigned long arch_align_stack(unsigned long sp);
252 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);