1 #ifndef _ASM_IA64_PROCESSOR_H
2 #define _ASM_IA64_PROCESSOR_H
3 
4 /*
5  * Copyright (C) 1998-2002 Hewlett-Packard Co
6  *	David Mosberger-Tang <davidm@hpl.hp.com>
7  *	Stephane Eranian <eranian@hpl.hp.com>
8  * Copyright (C) 1999 Asit Mallick <asit.k.mallick@intel.com>
9  * Copyright (C) 1999 Don Dugger <don.dugger@intel.com>
10  *
11  * 11/24/98	S.Eranian	added ia64_set_iva()
12  * 12/03/99	D. Mosberger	implement thread_saved_pc() via kernel unwind API
13  * 06/16/00	A. Mallick	added csd/ssd/tssd for ia32 support
14  */
15 
16 #include <linux/config.h>
17 #include <linux/cache.h>
18 
19 #include <asm/ptrace.h>
20 #include <asm/kregs.h>
21 #include <asm/types.h>
22 #include <asm/ustack.h>
23 
24 #define IA64_NUM_DBG_REGS	8
25 /*
26  * Limits for PMC and PMD are set to less than maximum architected values
27  * but should be sufficient for a while
28  */
29 #define IA64_NUM_PMC_REGS	32
30 #define IA64_NUM_PMD_REGS	32
31 
32 #define DEFAULT_MAP_BASE	0x2000000000000000
33 #define DEFAULT_TASK_SIZE	0xa000000000000000
34 
35 /*
36  * TASK_SIZE really is a mis-named.  It really is the maximum user
37  * space address (plus one).  On IA-64, there are five regions of 2TB
38  * each (assuming 8KB page size), for a total of 8TB of user virtual
39  * address space.
40  */
41 #define TASK_SIZE		(current->thread.task_size)
42 
43 /*
44  * This decides where the kernel will search for a free chunk of vm
45  * space during mmap's.
46  */
47 #define TASK_UNMAPPED_BASE	(current->thread.map_base)
48 
49 /*
50  * Bus types
51  */
52 #define EISA_bus 0
53 #define EISA_bus__is_a_macro /* for versions in ksyms.c */
54 #define MCA_bus 0
55 #define MCA_bus__is_a_macro /* for versions in ksyms.c */
56 
57 #define IA64_THREAD_FPH_VALID	(__IA64_UL(1) << 0)	/* floating-point high state valid? */
58 #define IA64_THREAD_DBG_VALID	(__IA64_UL(1) << 1)	/* debug registers valid? */
59 #define IA64_THREAD_PM_VALID	(__IA64_UL(1) << 2)	/* performance registers valid? */
60 #define IA64_THREAD_UAC_NOPRINT	(__IA64_UL(1) << 3)	/* don't log unaligned accesses */
61 #define IA64_THREAD_UAC_SIGBUS	(__IA64_UL(1) << 4)	/* generate SIGBUS on unaligned acc. */
62 #define IA64_THREAD_KRBS_SYNCED	(__IA64_UL(1) << 5)	/* krbs synced with process vm? */
63 #define IA64_THREAD_FPEMU_NOPRINT (__IA64_UL(1) << 6)	/* don't log any fpswa faults */
64 #define IA64_THREAD_FPEMU_SIGFPE  (__IA64_UL(1) << 7)	/* send a SIGFPE for fpswa faults */
65 #define IA64_THREAD_XSTACK	(__IA64_UL(1) << 8)	/* stack executable by default? */
66 
67 #define IA64_THREAD_UAC_SHIFT	3
68 #define IA64_THREAD_UAC_MASK	(IA64_THREAD_UAC_NOPRINT | IA64_THREAD_UAC_SIGBUS)
69 #define IA64_THREAD_FPEMU_SHIFT	6
70 #define IA64_THREAD_FPEMU_MASK	(IA64_THREAD_FPEMU_NOPRINT | IA64_THREAD_FPEMU_SIGFPE)
71 
72 
73 /*
74  * This shift should be large enough to be able to represent
75  * 1000000/itc_freq with good accuracy while being small enough to fit
76  * 1000000<<IA64_USEC_PER_CYC_SHIFT in 64 bits.
77  */
78 #define IA64_USEC_PER_CYC_SHIFT	41
79 
80 #ifndef __ASSEMBLY__
81 
82 #include <linux/threads.h>
83 #include <linux/cache.h>
84 
85 #include <asm/fpu.h>
86 #include <asm/offsets.h>
87 #include <asm/page.h>
88 #include <asm/rse.h>
89 #include <asm/unwind.h>
90 #include <asm/atomic.h>
91 #ifdef CONFIG_NUMA
92 #include <asm/nodedata.h>
93 #endif
94 
95 /* like above but expressed as bitfields for more efficient access: */
96 struct ia64_psr {
97 	__u64 reserved0 : 1;
98 	__u64 be : 1;
99 	__u64 up : 1;
100 	__u64 ac : 1;
101 	__u64 mfl : 1;
102 	__u64 mfh : 1;
103 	__u64 reserved1 : 7;
104 	__u64 ic : 1;
105 	__u64 i : 1;
106 	__u64 pk : 1;
107 	__u64 reserved2 : 1;
108 	__u64 dt : 1;
109 	__u64 dfl : 1;
110 	__u64 dfh : 1;
111 	__u64 sp : 1;
112 	__u64 pp : 1;
113 	__u64 di : 1;
114 	__u64 si : 1;
115 	__u64 db : 1;
116 	__u64 lp : 1;
117 	__u64 tb : 1;
118 	__u64 rt : 1;
119 	__u64 reserved3 : 4;
120 	__u64 cpl : 2;
121 	__u64 is : 1;
122 	__u64 mc : 1;
123 	__u64 it : 1;
124 	__u64 id : 1;
125 	__u64 da : 1;
126 	__u64 dd : 1;
127 	__u64 ss : 1;
128 	__u64 ri : 2;
129 	__u64 ed : 1;
130 	__u64 bn : 1;
131 	__u64 reserved4 : 19;
132 };
133 
134 /*
135  * CPU type, hardware bug flags, and per-CPU state.  Frequently used
136  * state comes earlier:
137  */
138 struct cpuinfo_ia64 {
139 	/* irq_stat must be 64-bit aligned */
140 	union {
141 		struct {
142 			__u32 irq_count;
143 			__u32 bh_count;
144 		} f;
145 		__u64 irq_and_bh_counts;
146 	} irq_stat;
147 	__u32 softirq_pending;
148 	__u32 phys_stacked_size_p8;	/* size of physical stacked registers + 8 */
149 	__u64 itm_delta;	/* # of clock cycles between clock ticks */
150 	__u64 itm_next;		/* interval timer mask value to use for next clock tick */
151 	__u64 *pgd_quick;
152 	__u64 *pmd_quick;
153 	__u64 *pte_quick;
154 	__u64 pgtable_cache_sz;
155 	/* CPUID-derived information: */
156 	__u64 ppn;
157 	__u64 features;
158 	__u8 number;
159 	__u8 revision;
160 	__u8 model;
161 	__u8 family;
162 	__u8 archrev;
163 	char vendor[16];
164 	__u8 need_tlb_flush;
165 	__u64 itc_freq;		/* frequency of ITC counter */
166 	__u64 proc_freq;	/* frequency of processor */
167 	__u64 cyc_per_usec;	/* itc_freq/1000000 */
168 	__u64 usec_per_cyc;	/* 2^IA64_USEC_PER_CYC_SHIFT*1000000/itc_freq */
169 	__u64 unimpl_va_mask;	/* mask of unimplemented virtual address bits (from PAL) */
170 	__u64 unimpl_pa_mask;	/* mask of unimplemented physical address bits (from PAL) */
171 	__u64 ptce_base;
172 	__u32 ptce_count[2];
173 	__u32 ptce_stride[2];
174 	struct task_struct *ksoftirqd;	/* kernel softirq daemon for this CPU */
175 	void *mmu_gathers;
176 # ifdef CONFIG_PERFMON
177 	unsigned long pfm_syst_info;
178 # endif
179 #ifdef CONFIG_SMP
180 	int processor;
181 	__u64 loops_per_jiffy;
182 	__u64 ipi_count;
183 	__u64 prof_counter;
184 	__u64 prof_multiplier;
185 	union {
186 		/*
187 		 *  This is written to by *other* CPUs,
188 		 *  so isolate it in its own cacheline.
189 		 */
190 		__u64 operation;
191 		char pad[SMP_CACHE_BYTES] ____cacheline_aligned;
192 	} ipi;
193 #endif
194 #ifdef CONFIG_NUMA
195 	struct ia64_node_data *node_data;
196 	int nodeid;
197 	struct cpuinfo_ia64 *cpu_data[NR_CPUS];
198 #endif
199 	/* Platform specific word.  MUST BE LAST IN STRUCT */
200 	__u64 platform_specific;
201 } __attribute__ ((aligned (PAGE_SIZE)));
202 
203 /*
204  * The "local" data pointer.  It points to the per-CPU data of the currently executing
205  * CPU, much like "current" points to the per-task data of the currently executing task.
206  */
207 #define local_cpu_data		((struct cpuinfo_ia64 *) PERCPU_ADDR)
208 
209 /*
210  * On NUMA systems, cpu_data for each cpu is allocated during cpu_init() & is allocated on
211  * the node that contains the cpu. This minimizes off-node memory references.  cpu_data
212  * for each cpu contains an array of pointers to the cpu_data structures of each of the
213  * other cpus.
214  *
215  * On non-NUMA systems, cpu_data is a static array allocated at compile time.  References
216  * to the cpu_data of another cpu is done by direct references to the appropriate entry of
217  * the array.
218  */
219 #ifdef CONFIG_NUMA
220 # define cpu_data(cpu)		local_cpu_data->cpu_data[cpu]
221 # define numa_node_id()		(local_cpu_data->nodeid)
222 #else
223   extern struct cpuinfo_ia64	_cpu_data[NR_CPUS];
224 # define cpu_data(cpu)		(&_cpu_data[cpu])
225 #endif
226 
227 extern void identify_cpu (struct cpuinfo_ia64 *);
228 extern void print_cpu_info (struct cpuinfo_ia64 *);
229 
230 typedef struct {
231 	unsigned long seg;
232 } mm_segment_t;
233 
234 #define SET_UNALIGN_CTL(task,value)								\
235 ({												\
236 	(task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_UAC_MASK)			\
237 				| (((value) << IA64_THREAD_UAC_SHIFT) & IA64_THREAD_UAC_MASK));	\
238 	0;											\
239 })
240 #define GET_UNALIGN_CTL(task,addr)								\
241 ({												\
242 	put_user(((task)->thread.flags & IA64_THREAD_UAC_MASK) >> IA64_THREAD_UAC_SHIFT,	\
243 		 (int *) (addr));								\
244 })
245 
246 #define SET_FPEMU_CTL(task,value)								\
247 ({												\
248 	(task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_FPEMU_MASK)		\
249 			  | (((value) << IA64_THREAD_FPEMU_SHIFT) & IA64_THREAD_FPEMU_MASK));	\
250 	0;											\
251 })
252 #define GET_FPEMU_CTL(task,addr)								\
253 ({												\
254 	put_user(((task)->thread.flags & IA64_THREAD_FPEMU_MASK) >> IA64_THREAD_FPEMU_SHIFT,	\
255 		 (int *) (addr));								\
256 })
257 
258 struct siginfo;
259 
260 struct thread_struct {
261 	__u64 ksp;			/* kernel stack pointer */
262 	unsigned long flags;		/* various flags */
263 	__u64 map_base;			/* base address for get_unmapped_area() */
264 	__u64 task_size;		/* limit for task size */
265 	__u64 rbs_bot;			/* the base address for the RBS */
266 	struct siginfo *siginfo;	/* current siginfo struct for ptrace() */
267 
268 #ifdef CONFIG_IA32_SUPPORT
269 	__u64 eflag;			/* IA32 EFLAGS reg */
270 	__u64 fsr;			/* IA32 floating pt status reg */
271 	__u64 fcr;			/* IA32 floating pt control reg */
272 	__u64 fir;			/* IA32 fp except. instr. reg */
273 	__u64 fdr;			/* IA32 fp except. data reg */
274 	__u64 old_k1;			/* old value of ar.k1 */
275 	__u64 old_iob;			/* old IOBase value */
276 # define INIT_THREAD_IA32	0, 0, 0x17800000037fULL, 0, 0, 0, 0,
277 #else
278 # define INIT_THREAD_IA32
279 #endif /* CONFIG_IA32_SUPPORT */
280 #ifdef CONFIG_PERFMON
281 	__u64 pmc[IA64_NUM_PMC_REGS];
282 	__u64 pmd[IA64_NUM_PMD_REGS];
283 	unsigned long pfm_ovfl_block_reset;/* non-zero if we need to block or reset regs on ovfl */
284 	void *pfm_context;		/* pointer to detailed PMU context */
285 	atomic_t pfm_notifiers_check;	/* when >0, will cleanup ctx_notify_task in tasklist */
286 	atomic_t pfm_owners_check;	/* when >0, will cleanup ctx_owner in tasklist */
287 	void *pfm_smpl_buf_list;	/* list of sampling buffers to vfree */
288 # define INIT_THREAD_PM		{0, }, {0, }, 0, NULL, {0}, {0}, NULL,
289 #else
290 # define INIT_THREAD_PM
291 #endif
292 	__u64 dbr[IA64_NUM_DBG_REGS];
293 	__u64 ibr[IA64_NUM_DBG_REGS];
294 	struct ia64_fpreg fph[96];	/* saved/loaded on demand */
295 	int last_fph_cpu;
296 };
297 
298 #define INIT_THREAD {					\
299 	0,				/* ksp */	\
300 	0,				/* flags */	\
301 	DEFAULT_MAP_BASE,		/* map_base */	\
302 	DEFAULT_TASK_SIZE,		/* task_size */	\
303 	DEFAULT_USER_STACK_SIZE,	/* rbs_bot */	\
304 	0,				/* siginfo */	\
305 	INIT_THREAD_IA32				\
306 	INIT_THREAD_PM					\
307 	{0, },				/* dbr */	\
308 	{0, },				/* ibr */	\
309 	{{{{0}}}, },			/* fph */	\
310 	-1				/* last_fph_cpu*/	\
311 }
312 
313 #define start_thread(regs,new_ip,new_sp) do {							\
314 	set_fs(USER_DS);									\
315 	regs->cr_ipsr = ((regs->cr_ipsr | (IA64_PSR_BITS_TO_SET | IA64_PSR_CPL))		\
316 			 & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_RI | IA64_PSR_IS));		\
317 	regs->cr_iip = new_ip;									\
318 	regs->ar_rsc = 0xf;		/* eager mode, privilege level 3 */			\
319 	regs->ar_rnat = 0;									\
320 	regs->ar_bspstore = current->thread.rbs_bot;						\
321 	regs->ar_fpsr = FPSR_DEFAULT;								\
322 	regs->loadrs = 0;									\
323 	regs->r8 = current->mm->dumpable;	/* set "don't zap registers" flag */		\
324 	regs->r12 = new_sp - 16;	/* allocate 16 byte scratch area */			\
325 	if (!__builtin_expect (current->mm->dumpable, 1)) {					\
326 		/*										\
327 		 * Zap scratch regs to avoid leaking bits between processes with different	\
328 		 * uid/privileges.								\
329 		 */										\
330 		regs->ar_pfs = 0;								\
331 		regs->pr = 0;									\
332 		/*										\
333 		 * XXX fix me: everything below can go away once we stop preserving scratch	\
334 		 * regs on a system call.							\
335 		 */										\
336 		regs->b6 = 0;									\
337 		regs->r1 = 0; regs->r2 = 0; regs->r3 = 0;					\
338 		regs->r13 = 0; regs->r14 = 0; regs->r15 = 0;					\
339 		regs->r9  = 0; regs->r11 = 0;							\
340 		regs->r16 = 0; regs->r17 = 0; regs->r18 = 0; regs->r19 = 0;			\
341 		regs->r20 = 0; regs->r21 = 0; regs->r22 = 0; regs->r23 = 0;			\
342 		regs->r24 = 0; regs->r25 = 0; regs->r26 = 0; regs->r27 = 0;			\
343 		regs->r28 = 0; regs->r29 = 0; regs->r30 = 0; regs->r31 = 0;			\
344 		regs->ar_ccv = 0;								\
345 		regs->ar_csd = 0;                                                               \
346 		regs->ar_ssd = 0;                                                               \
347 		regs->b0 = 0; regs->b7 = 0;							\
348 		regs->f6.u.bits[0] = 0; regs->f6.u.bits[1] = 0;					\
349 		regs->f7.u.bits[0] = 0; regs->f7.u.bits[1] = 0;					\
350 		regs->f8.u.bits[0] = 0; regs->f8.u.bits[1] = 0;					\
351 		regs->f9.u.bits[0] = 0; regs->f9.u.bits[1] = 0;					\
352 		regs->f10.u.bits[0] = 0; regs->f10.u.bits[1] = 0;				\
353 		regs->f11.u.bits[0] = 0; regs->f11.u.bits[1] = 0;				\
354 	}											\
355 } while (0)
356 
357 /* Forward declarations, a strange C thing... */
358 struct mm_struct;
359 struct task_struct;
360 
361 /*
362  * Free all resources held by a thread. This is called after the
363  * parent of DEAD_TASK has collected the exist status of the task via
364  * wait().
365  */
366 #ifdef CONFIG_PERFMON
367   extern void release_thread (struct task_struct *task);
368 #else
369 # define release_thread(dead_task)
370 #endif
371 
372 /*
373  * This is the mechanism for creating a new kernel thread.
374  *
375  * NOTE 1: Only a kernel-only process (ie the swapper or direct
376  * descendants who haven't done an "execve()") should use this: it
377  * will work within a system call from a "real" process, but the
378  * process memory space will not be free'd until both the parent and
379  * the child have exited.
380  *
381  * NOTE 2: This MUST NOT be an inlined function.  Otherwise, we get
382  * into trouble in init/main.c when the child thread returns to
383  * do_basic_setup() and the timing is such that free_initmem() has
384  * been called already.
385  */
386 extern int arch_kernel_thread (int (*fn)(void *), void *arg, unsigned long flags);
387 
388 /* Copy and release all segment info associated with a VM */
389 #define copy_segments(tsk, mm)			do { } while (0)
390 #define release_segments(mm)			do { } while (0)
391 
392 /* Get wait channel for task P.  */
393 extern unsigned long get_wchan (struct task_struct *p);
394 
395 /* Return instruction pointer of blocked task TSK.  */
396 #define KSTK_EIP(tsk)					\
397   ({							\
398 	struct pt_regs *_regs = ia64_task_regs(tsk);	\
399 	_regs->cr_iip + ia64_psr(_regs)->ri;		\
400   })
401 
402 /* Return stack pointer of blocked task TSK.  */
403 #define KSTK_ESP(tsk)  ((tsk)->thread.ksp)
404 
405 static inline unsigned long
ia64_get_kr(unsigned long regnum)406 ia64_get_kr (unsigned long regnum)
407 {
408 	unsigned long r = 0;
409 
410 	switch (regnum) {
411 	      case 0: asm volatile ("mov %0=ar.k0" : "=r"(r)); break;
412 	      case 1: asm volatile ("mov %0=ar.k1" : "=r"(r)); break;
413 	      case 2: asm volatile ("mov %0=ar.k2" : "=r"(r)); break;
414 	      case 3: asm volatile ("mov %0=ar.k3" : "=r"(r)); break;
415 	      case 4: asm volatile ("mov %0=ar.k4" : "=r"(r)); break;
416 	      case 5: asm volatile ("mov %0=ar.k5" : "=r"(r)); break;
417 	      case 6: asm volatile ("mov %0=ar.k6" : "=r"(r)); break;
418 	      case 7: asm volatile ("mov %0=ar.k7" : "=r"(r)); break;
419 	}
420 	return r;
421 }
422 
423 static inline void
ia64_set_kr(unsigned long regnum,unsigned long r)424 ia64_set_kr (unsigned long regnum, unsigned long r)
425 {
426 	switch (regnum) {
427 	      case 0: asm volatile ("mov ar.k0=%0" :: "r"(r)); break;
428 	      case 1: asm volatile ("mov ar.k1=%0" :: "r"(r)); break;
429 	      case 2: asm volatile ("mov ar.k2=%0" :: "r"(r)); break;
430 	      case 3: asm volatile ("mov ar.k3=%0" :: "r"(r)); break;
431 	      case 4: asm volatile ("mov ar.k4=%0" :: "r"(r)); break;
432 	      case 5: asm volatile ("mov ar.k5=%0" :: "r"(r)); break;
433 	      case 6: asm volatile ("mov ar.k6=%0" :: "r"(r)); break;
434 	      case 7: asm volatile ("mov ar.k7=%0" :: "r"(r)); break;
435 	}
436 }
437 /* Return TRUE if task T owns the fph partition of the CPU we're running on. */
438 #define ia64_is_local_fpu_owner(t)								\
439 ({												\
440 	struct task_struct *__ia64_islfo_task = (t);						\
441 	(__ia64_islfo_task->thread.last_fph_cpu == smp_processor_id()				\
442 	 && __ia64_islfo_task == (struct task_struct *) ia64_get_kr(IA64_KR_FPU_OWNER));	\
443 })
444 
445 /* Mark task T as owning the fph partition of the CPU we're running on. */
446 #define ia64_set_local_fpu_owner(t) do {						\
447 	struct task_struct *__ia64_slfo_task = (t);					\
448 	__ia64_slfo_task->thread.last_fph_cpu = smp_processor_id();			\
449 	ia64_set_kr(IA64_KR_FPU_OWNER, (unsigned long) __ia64_slfo_task);		\
450 } while (0)
451 
452 /* Mark the fph partition of task T as being invalid on all CPUs.  */
453 #define ia64_drop_fpu(t)	((t)->thread.last_fph_cpu = -1)
454 
455 extern void __ia64_init_fpu (void);
456 extern void __ia64_save_fpu (struct ia64_fpreg *fph);
457 extern void __ia64_load_fpu (struct ia64_fpreg *fph);
458 extern void ia64_save_debug_regs (unsigned long *save_area);
459 extern void ia64_load_debug_regs (unsigned long *save_area);
460 
461 #ifdef CONFIG_IA32_SUPPORT
462 extern void ia32_save_state (struct task_struct *task);
463 extern void ia32_load_state (struct task_struct *task);
464 #endif
465 
466 #define ia64_fph_enable()	asm volatile (";; rsm psr.dfh;; srlz.d;;" ::: "memory");
467 #define ia64_fph_disable()	asm volatile (";; ssm psr.dfh;; srlz.d;;" ::: "memory");
468 
469 /* load fp 0.0 into fph */
470 static inline void
ia64_init_fpu(void)471 ia64_init_fpu (void) {
472 	ia64_fph_enable();
473 	__ia64_init_fpu();
474 	ia64_fph_disable();
475 }
476 
477 /* save f32-f127 at FPH */
478 static inline void
ia64_save_fpu(struct ia64_fpreg * fph)479 ia64_save_fpu (struct ia64_fpreg *fph) {
480 	ia64_fph_enable();
481 	__ia64_save_fpu(fph);
482 	ia64_fph_disable();
483 }
484 
485 /* load f32-f127 from FPH */
486 static inline void
ia64_load_fpu(struct ia64_fpreg * fph)487 ia64_load_fpu (struct ia64_fpreg *fph) {
488 	ia64_fph_enable();
489 	__ia64_load_fpu(fph);
490 	ia64_fph_disable();
491 }
492 
493 static inline void
ia64_fc(void * addr)494 ia64_fc (void *addr)
495 {
496 	asm volatile ("fc %0" :: "r"(addr) : "memory");
497 }
498 
499 static inline void
ia64_sync_i(void)500 ia64_sync_i (void)
501 {
502 	asm volatile (";; sync.i" ::: "memory");
503 }
504 
505 static inline void
ia64_srlz_i(void)506 ia64_srlz_i (void)
507 {
508 	asm volatile (";; srlz.i ;;" ::: "memory");
509 }
510 
511 static inline void
ia64_srlz_d(void)512 ia64_srlz_d (void)
513 {
514 	asm volatile (";; srlz.d" ::: "memory");
515 }
516 
517 static inline __u64
ia64_get_rr(__u64 reg_bits)518 ia64_get_rr (__u64 reg_bits)
519 {
520 	__u64 r;
521 	asm volatile ("mov %0=rr[%1]" : "=r"(r) : "r"(reg_bits) : "memory");
522 	return r;
523 }
524 
525 static inline void
ia64_set_rr(__u64 reg_bits,__u64 rr_val)526 ia64_set_rr (__u64 reg_bits, __u64 rr_val)
527 {
528 	asm volatile ("mov rr[%0]=%1" :: "r"(reg_bits), "r"(rr_val) : "memory");
529 }
530 
531 static inline __u64
ia64_get_dcr(void)532 ia64_get_dcr (void)
533 {
534 	__u64 r;
535 	asm volatile ("mov %0=cr.dcr" : "=r"(r));
536 	return r;
537 }
538 
539 static inline void
ia64_set_dcr(__u64 val)540 ia64_set_dcr (__u64 val)
541 {
542 	asm volatile ("mov cr.dcr=%0;;" :: "r"(val) : "memory");
543 	ia64_srlz_d();
544 }
545 
546 static inline __u64
ia64_get_lid(void)547 ia64_get_lid (void)
548 {
549 	__u64 r;
550 	asm volatile ("mov %0=cr.lid" : "=r"(r));
551 	return r;
552 }
553 
554 static inline void
ia64_invala(void)555 ia64_invala (void)
556 {
557 	asm volatile ("invala" ::: "memory");
558 }
559 
560 /*
561  * Save the processor status flags in FLAGS and then clear the interrupt collection and
562  * interrupt enable bits.  Don't trigger any mandatory RSE references while this bit is
563  * off!
564  */
565 static inline __u64
ia64_clear_ic(void)566 ia64_clear_ic (void)
567 {
568 	__u64 psr;
569 	asm volatile ("mov %0=psr;; rsm psr.i | psr.ic;; srlz.i;;" : "=r"(psr) :: "memory");
570 	return psr;
571 }
572 
573 /*
574  * Restore the psr.
575  */
576 static inline void
ia64_set_psr(__u64 psr)577 ia64_set_psr (__u64 psr)
578 {
579 	asm volatile (";; mov psr.l=%0;; srlz.d" :: "r" (psr) : "memory");
580 }
581 
582 /*
583  * Insert a translation into an instruction and/or data translation
584  * register.
585  */
586 static inline void
ia64_itr(__u64 target_mask,__u64 tr_num,__u64 vmaddr,__u64 pte,__u64 log_page_size)587 ia64_itr (__u64 target_mask, __u64 tr_num,
588 	  __u64 vmaddr, __u64 pte,
589 	  __u64 log_page_size)
590 {
591 	asm volatile ("mov cr.itir=%0" :: "r"(log_page_size << 2) : "memory");
592 	asm volatile ("mov cr.ifa=%0;;" :: "r"(vmaddr) : "memory");
593 	if (target_mask & 0x1)
594 		asm volatile ("itr.i itr[%0]=%1"
595 				      :: "r"(tr_num), "r"(pte) : "memory");
596 	if (target_mask & 0x2)
597 		asm volatile (";;itr.d dtr[%0]=%1"
598 				      :: "r"(tr_num), "r"(pte) : "memory");
599 }
600 
601 /*
602  * Insert a translation into the instruction and/or data translation
603  * cache.
604  */
605 static inline void
ia64_itc(__u64 target_mask,__u64 vmaddr,__u64 pte,__u64 log_page_size)606 ia64_itc (__u64 target_mask, __u64 vmaddr, __u64 pte,
607 	  __u64 log_page_size)
608 {
609 	asm volatile ("mov cr.itir=%0" :: "r"(log_page_size << 2) : "memory");
610 	asm volatile ("mov cr.ifa=%0;;" :: "r"(vmaddr) : "memory");
611 	/* as per EAS2.6, itc must be the last instruction in an instruction group */
612 	if (target_mask & 0x1)
613 		asm volatile ("itc.i %0;;" :: "r"(pte) : "memory");
614 	if (target_mask & 0x2)
615 		asm volatile (";;itc.d %0;;" :: "r"(pte) : "memory");
616 }
617 
618 /*
619  * Purge a range of addresses from instruction and/or data translation
620  * register(s).
621  */
622 static inline void
ia64_ptr(__u64 target_mask,__u64 vmaddr,__u64 log_size)623 ia64_ptr (__u64 target_mask, __u64 vmaddr, __u64 log_size)
624 {
625 	if (target_mask & 0x1)
626 		asm volatile ("ptr.i %0,%1" :: "r"(vmaddr), "r"(log_size << 2));
627 	if (target_mask & 0x2)
628 		asm volatile ("ptr.d %0,%1" :: "r"(vmaddr), "r"(log_size << 2));
629 }
630 
631 /* Set the interrupt vector address.  The address must be suitably aligned (32KB).  */
632 static inline void
ia64_set_iva(void * ivt_addr)633 ia64_set_iva (void *ivt_addr)
634 {
635 	asm volatile ("mov cr.iva=%0;; srlz.i;;" :: "r"(ivt_addr) : "memory");
636 }
637 
638 /* Set the page table address and control bits.  */
639 static inline void
ia64_set_pta(__u64 pta)640 ia64_set_pta (__u64 pta)
641 {
642 	/* Note: srlz.i implies srlz.d */
643 	asm volatile ("mov cr.pta=%0;; srlz.i;;" :: "r"(pta) : "memory");
644 }
645 
646 static inline __u64
ia64_get_cpuid(__u64 regnum)647 ia64_get_cpuid (__u64 regnum)
648 {
649 	__u64 r;
650 
651 	asm ("mov %0=cpuid[%r1]" : "=r"(r) : "rO"(regnum));
652 	return r;
653 }
654 
655 static inline void
ia64_eoi(void)656 ia64_eoi (void)
657 {
658 	asm ("mov cr.eoi=r0;; srlz.d;;" ::: "memory");
659 }
660 
661 static inline void
ia64_set_lrr0(unsigned long val)662 ia64_set_lrr0 (unsigned long val)
663 {
664 	asm volatile ("mov cr.lrr0=%0;; srlz.d" :: "r"(val) : "memory");
665 }
666 
667 #ifdef GAS_HAS_HINT_INSN
668 static inline void
ia64_hint_pause(void)669 ia64_hint_pause (void)
670 {
671 	asm volatile ("hint @pause" ::: "memory");
672 }
673 
674 #define cpu_relax()	ia64_hint_pause()
675 #else
676 #define cpu_relax()	barrier()
677 #endif
678 
679 static inline void
ia64_set_lrr1(unsigned long val)680 ia64_set_lrr1 (unsigned long val)
681 {
682 	asm volatile ("mov cr.lrr1=%0;; srlz.d" :: "r"(val) : "memory");
683 }
684 
685 static inline void
ia64_set_pmv(__u64 val)686 ia64_set_pmv (__u64 val)
687 {
688 	asm volatile ("mov cr.pmv=%0" :: "r"(val) : "memory");
689 }
690 
691 static inline __u64
ia64_get_pmc(__u64 regnum)692 ia64_get_pmc (__u64 regnum)
693 {
694 	__u64 retval;
695 
696 	asm volatile ("mov %0=pmc[%1]" : "=r"(retval) : "r"(regnum));
697 	return retval;
698 }
699 
700 static inline void
ia64_set_pmc(__u64 regnum,__u64 value)701 ia64_set_pmc (__u64 regnum, __u64 value)
702 {
703 	asm volatile ("mov pmc[%0]=%1" :: "r"(regnum), "r"(value));
704 }
705 
706 static inline __u64
ia64_get_pmd(__u64 regnum)707 ia64_get_pmd (__u64 regnum)
708 {
709 	__u64 retval;
710 
711 	asm volatile ("mov %0=pmd[%1]" : "=r"(retval) : "r"(regnum));
712 	return retval;
713 }
714 
715 static inline void
ia64_set_pmd(__u64 regnum,__u64 value)716 ia64_set_pmd (__u64 regnum, __u64 value)
717 {
718 	asm volatile ("mov pmd[%0]=%1" :: "r"(regnum), "r"(value));
719 }
720 
721 /*
722  * Given the address to which a spill occurred, return the unat bit
723  * number that corresponds to this address.
724  */
725 static inline __u64
ia64_unat_pos(void * spill_addr)726 ia64_unat_pos (void *spill_addr)
727 {
728 	return ((__u64) spill_addr >> 3) & 0x3f;
729 }
730 
731 /*
732  * Set the NaT bit of an integer register which was spilled at address
733  * SPILL_ADDR.  UNAT is the mask to be updated.
734  */
735 static inline void
ia64_set_unat(__u64 * unat,void * spill_addr,unsigned long nat)736 ia64_set_unat (__u64 *unat, void *spill_addr, unsigned long nat)
737 {
738 	__u64 bit = ia64_unat_pos(spill_addr);
739 	__u64 mask = 1UL << bit;
740 
741 	*unat = (*unat & ~mask) | (nat << bit);
742 }
743 
744 /*
745  * Return saved PC of a blocked thread.
746  * Note that the only way T can block is through a call to schedule() -> switch_to().
747  */
748 static inline unsigned long
thread_saved_pc(struct thread_struct * t)749 thread_saved_pc (struct thread_struct *t)
750 {
751 	struct unw_frame_info info;
752 	unsigned long ip;
753 
754 	/* XXX ouch: Linus, please pass the task pointer to thread_saved_pc() instead! */
755 	struct task_struct *p = (void *) ((unsigned long) t - IA64_TASK_THREAD_OFFSET);
756 
757 	unw_init_from_blocked_task(&info, p);
758 	if (unw_unwind(&info) < 0)
759 		return 0;
760 	unw_get_ip(&info, &ip);
761 	return ip;
762 }
763 
764 /*
765  * Get the current instruction/program counter value.
766  */
767 #define current_text_addr() \
768 	({ void *_pc; asm volatile ("mov %0=ip" : "=r" (_pc)); _pc; })
769 
770 #define THREAD_SIZE	IA64_STK_OFFSET
771 /* NOTE: The task struct and the stacks are allocated together.  */
772 #define alloc_task_struct() \
773         ((struct task_struct *) __get_free_pages(GFP_KERNEL, IA64_TASK_STRUCT_LOG_NUM_PAGES))
774 #define free_task_struct(p)	free_pages((unsigned long)(p), IA64_TASK_STRUCT_LOG_NUM_PAGES)
775 #define get_task_struct(tsk)	atomic_inc(&virt_to_page(tsk)->count)
776 
777 #define init_task	(init_task_union.task)
778 #define init_stack	(init_task_union.stack)
779 
780 static inline void
ia64_set_cmcv(__u64 val)781 ia64_set_cmcv (__u64 val)
782 {
783 	asm volatile ("mov cr.cmcv=%0" :: "r"(val) : "memory");
784 }
785 
786 static inline __u64
ia64_get_cmcv(void)787 ia64_get_cmcv (void)
788 {
789 	__u64 val;
790 
791 	asm volatile ("mov %0=cr.cmcv" : "=r"(val) :: "memory");
792 	return val;
793 }
794 
795 static inline __u64
ia64_get_ivr(void)796 ia64_get_ivr (void)
797 {
798 	__u64 r;
799 	asm volatile ("srlz.d;; mov %0=cr.ivr;; srlz.d;;" : "=r"(r));
800 	return r;
801 }
802 
803 static inline void
ia64_set_tpr(__u64 val)804 ia64_set_tpr (__u64 val)
805 {
806 	asm volatile ("mov cr.tpr=%0" :: "r"(val));
807 }
808 
809 static inline __u64
ia64_get_tpr(void)810 ia64_get_tpr (void)
811 {
812 	__u64 r;
813 	asm volatile ("mov %0=cr.tpr" : "=r"(r));
814 	return r;
815 }
816 
817 static inline void
ia64_set_irr0(__u64 val)818 ia64_set_irr0 (__u64 val)
819 {
820 	asm volatile("mov cr.irr0=%0;;" :: "r"(val) : "memory");
821 	ia64_srlz_d();
822 }
823 
824 static inline __u64
ia64_get_irr0(void)825 ia64_get_irr0 (void)
826 {
827 	__u64 val;
828 
829 	/* this is volatile because irr may change unbeknownst to gcc... */
830 	asm volatile("mov %0=cr.irr0" : "=r"(val));
831 	return val;
832 }
833 
834 static inline void
ia64_set_irr1(__u64 val)835 ia64_set_irr1 (__u64 val)
836 {
837 	asm volatile("mov cr.irr1=%0;;" :: "r"(val) : "memory");
838 	ia64_srlz_d();
839 }
840 
841 static inline __u64
ia64_get_irr1(void)842 ia64_get_irr1 (void)
843 {
844 	__u64 val;
845 
846 	/* this is volatile because irr may change unbeknownst to gcc... */
847 	asm volatile("mov %0=cr.irr1" : "=r"(val));
848 	return val;
849 }
850 
851 static inline void
ia64_set_irr2(__u64 val)852 ia64_set_irr2 (__u64 val)
853 {
854 	asm volatile("mov cr.irr2=%0;;" :: "r"(val) : "memory");
855 	ia64_srlz_d();
856 }
857 
858 static inline __u64
ia64_get_irr2(void)859 ia64_get_irr2 (void)
860 {
861 	__u64 val;
862 
863 	/* this is volatile because irr may change unbeknownst to gcc... */
864 	asm volatile("mov %0=cr.irr2" : "=r"(val));
865 	return val;
866 }
867 
868 static inline void
ia64_set_irr3(__u64 val)869 ia64_set_irr3 (__u64 val)
870 {
871 	asm volatile("mov cr.irr3=%0;;" :: "r"(val) : "memory");
872 	ia64_srlz_d();
873 }
874 
875 static inline __u64
ia64_get_irr3(void)876 ia64_get_irr3 (void)
877 {
878 	__u64 val;
879 
880 	/* this is volatile because irr may change unbeknownst to gcc... */
881 	asm volatile ("mov %0=cr.irr3" : "=r"(val));
882 	return val;
883 }
884 
885 static inline __u64
ia64_get_gp(void)886 ia64_get_gp(void)
887 {
888 	__u64 val;
889 
890 	asm ("mov %0=gp" : "=r"(val));
891 	return val;
892 }
893 
894 static inline void
ia64_set_ibr(__u64 regnum,__u64 value)895 ia64_set_ibr (__u64 regnum, __u64 value)
896 {
897 	asm volatile ("mov ibr[%0]=%1" :: "r"(regnum), "r"(value));
898 }
899 
900 static inline void
ia64_set_dbr(__u64 regnum,__u64 value)901 ia64_set_dbr (__u64 regnum, __u64 value)
902 {
903 	asm volatile ("mov dbr[%0]=%1" :: "r"(regnum), "r"(value));
904 #ifdef CONFIG_ITANIUM
905 	asm volatile (";; srlz.d");
906 #endif
907 }
908 
909 static inline __u64
ia64_get_ibr(__u64 regnum)910 ia64_get_ibr (__u64 regnum)
911 {
912 	__u64 retval;
913 
914 	asm volatile ("mov %0=ibr[%1]" : "=r"(retval) : "r"(regnum));
915 	return retval;
916 }
917 
918 static inline __u64
ia64_get_dbr(__u64 regnum)919 ia64_get_dbr (__u64 regnum)
920 {
921 	__u64 retval;
922 
923 	asm volatile ("mov %0=dbr[%1]" : "=r"(retval) : "r"(regnum));
924 #ifdef CONFIG_ITANIUM
925 	asm volatile (";; srlz.d");
926 #endif
927 	return retval;
928 }
929 
930 static inline __u64
ia64_rotr(__u64 w,__u64 n)931 ia64_rotr (__u64 w, __u64 n)
932 {
933 	return (w >> n) | (w << (64 - n));
934 }
935 
936 #define ia64_rotl(w,n)	ia64_rotr((w), (64) - (n))
937 
938 static inline __u64
ia64_thash(__u64 addr)939 ia64_thash (__u64 addr)
940 {
941 	__u64 result;
942 	asm ("thash %0=%1" : "=r"(result) : "r" (addr));
943 	return result;
944 }
945 
946 static inline __u64
ia64_tpa(__u64 addr)947 ia64_tpa (__u64 addr)
948 {
949 	__u64 result;
950 	asm ("tpa %0=%1" : "=r"(result) : "r"(addr));
951 	return result;
952 }
953 
954 /*
955  * Take a mapped kernel address and return the equivalent address
956  * in the region 7 identity mapped virtual area.
957  */
958 static inline void *
ia64_imva(void * addr)959 ia64_imva (void *addr)
960 {
961 	void *result;
962 	asm ("tpa %0=%1" : "=r"(result) : "r"(addr));
963 	return __va(result);
964 }
965 
966 #define ARCH_HAS_PREFETCH
967 #define ARCH_HAS_PREFETCHW
968 #define ARCH_HAS_SPINLOCK_PREFETCH
969 #define PREFETCH_STRIDE 256
970 
971 extern inline void
prefetch(const void * x)972 prefetch (const void *x)
973 {
974          __asm__ __volatile__ ("lfetch [%0]" : : "r"(x));
975 }
976 
977 extern inline void
prefetchw(const void * x)978 prefetchw (const void *x)
979 {
980 	__asm__ __volatile__ ("lfetch.excl [%0]" : : "r"(x));
981 }
982 
983 #define spin_lock_prefetch(x)	prefetchw(x)
984 
985 #endif /* !__ASSEMBLY__ */
986 
987 #endif /* _ASM_IA64_PROCESSOR_H */
988