1 #ifndef _ASM_IA64_PROCESSOR_H
2 #define _ASM_IA64_PROCESSOR_H
5 * Copyright (C) 1998-2004 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>
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
17 #include <asm/intrinsics.h>
18 #include <asm/kregs.h>
19 #include <asm/ptrace.h>
20 #include <asm/ustack.h>
22 #define __ARCH_WANT_UNLOCKED_CTXSW
23 #define ARCH_HAS_PREFETCH_SWITCH_STACK
25 #define IA64_NUM_PHYS_STACK_REG 96
26 #define IA64_NUM_DBG_REGS 8
28 #define DEFAULT_MAP_BASE __IA64_UL_CONST(0x2000000000000000)
29 #define DEFAULT_TASK_SIZE __IA64_UL_CONST(0xa000000000000000)
32 * TASK_SIZE really is a mis-named. It really is the maximum user
33 * space address (plus one). On IA-64, there are five regions of 2TB
34 * each (assuming 8KB page size), for a total of 8TB of user virtual
37 #define TASK_SIZE_OF(tsk) ((tsk)->thread.task_size)
38 #define TASK_SIZE TASK_SIZE_OF(current)
41 * This decides where the kernel will search for a free chunk of vm
42 * space during mmap's.
44 #define TASK_UNMAPPED_BASE (current->thread.map_base)
46 #define IA64_THREAD_FPH_VALID (__IA64_UL(1) << 0) /* floating-point high state valid? */
47 #define IA64_THREAD_DBG_VALID (__IA64_UL(1) << 1) /* debug registers valid? */
48 #define IA64_THREAD_PM_VALID (__IA64_UL(1) << 2) /* performance registers valid? */
49 #define IA64_THREAD_UAC_NOPRINT (__IA64_UL(1) << 3) /* don't log unaligned accesses */
50 #define IA64_THREAD_UAC_SIGBUS (__IA64_UL(1) << 4) /* generate SIGBUS on unaligned acc. */
51 #define IA64_THREAD_MIGRATION (__IA64_UL(1) << 5) /* require migration
53 #define IA64_THREAD_FPEMU_NOPRINT (__IA64_UL(1) << 6) /* don't log any fpswa faults */
54 #define IA64_THREAD_FPEMU_SIGFPE (__IA64_UL(1) << 7) /* send a SIGFPE for fpswa faults */
56 #define IA64_THREAD_UAC_SHIFT 3
57 #define IA64_THREAD_UAC_MASK (IA64_THREAD_UAC_NOPRINT | IA64_THREAD_UAC_SIGBUS)
58 #define IA64_THREAD_FPEMU_SHIFT 6
59 #define IA64_THREAD_FPEMU_MASK (IA64_THREAD_FPEMU_NOPRINT | IA64_THREAD_FPEMU_SIGFPE)
63 * This shift should be large enough to be able to represent 1000000000/itc_freq with good
64 * accuracy while being small enough to fit 10*1000000000<<IA64_NSEC_PER_CYC_SHIFT in 64 bits
65 * (this will give enough slack to represent 10 seconds worth of time as a scaled number).
67 #define IA64_NSEC_PER_CYC_SHIFT 30
71 #include <linux/cache.h>
72 #include <linux/compiler.h>
73 #include <linux/threads.h>
74 #include <linux/types.h>
78 #include <asm/percpu.h>
80 #include <asm/unwind.h>
81 #include <linux/atomic.h>
83 #include <asm/nodedata.h>
86 /* like above but expressed as bitfields for more efficient access: */
122 __u64 reserved4 : 19;
142 __u64 reserved2 : 20;
170 __u64 rv3 : 2; /* 0-1 */
171 __u64 ps : 6; /* 2-7 */
172 __u64 key : 24; /* 8-31 */
173 __u64 rv4 : 32; /* 32-63 */
180 __u64 ve : 1; /* enable hw walker */
181 __u64 reserved0: 1; /* reserved */
182 __u64 ps : 6; /* log page size */
183 __u64 rid : 24; /* region id */
184 __u64 reserved1: 32; /* reserved */
189 * CPU type, hardware bug flags, and per-CPU state. Frequently used
190 * state comes earlier:
192 struct cpuinfo_ia64 {
193 unsigned int softirq_pending;
194 unsigned long itm_delta; /* # of clock cycles between clock ticks */
195 unsigned long itm_next; /* interval timer mask value to use for next clock tick */
196 unsigned long nsec_per_cyc; /* (1000000000<<IA64_NSEC_PER_CYC_SHIFT)/itc_freq */
197 unsigned long unimpl_va_mask; /* mask of unimplemented virtual address bits (from PAL) */
198 unsigned long unimpl_pa_mask; /* mask of unimplemented physical address bits (from PAL) */
199 unsigned long itc_freq; /* frequency of ITC counter */
200 unsigned long proc_freq; /* frequency of processor */
201 unsigned long cyc_per_usec; /* itc_freq/1000000 */
202 unsigned long ptce_base;
203 unsigned int ptce_count[2];
204 unsigned int ptce_stride[2];
205 struct task_struct *ksoftirqd; /* kernel softirq daemon for this CPU */
208 unsigned long loops_per_jiffy;
210 unsigned int socket_id; /* physical processor socket id */
211 unsigned short core_id; /* core id */
212 unsigned short thread_id; /* thread id */
213 unsigned short num_log; /* Total number of logical processors on
214 * this socket that were successfully booted */
215 unsigned char cores_per_socket; /* Cores per processor socket */
216 unsigned char threads_per_core; /* Threads per core */
219 /* CPUID-derived information: */
221 unsigned long features;
222 unsigned char number;
223 unsigned char revision;
225 unsigned char family;
226 unsigned char archrev;
231 struct ia64_node_data *node_data;
235 DECLARE_PER_CPU(struct cpuinfo_ia64, ia64_cpu_info);
238 * The "local" data variable. It refers to the per-CPU data of the currently executing
239 * CPU, much like "current" points to the per-task data of the currently executing task.
240 * Do not use the address of local_cpu_data, since it will be different from
241 * cpu_data(smp_processor_id())!
243 #define local_cpu_data (&__ia64_per_cpu_var(ia64_cpu_info))
244 #define cpu_data(cpu) (&per_cpu(ia64_cpu_info, cpu))
246 extern void print_cpu_info (struct cpuinfo_ia64 *);
252 #define SET_UNALIGN_CTL(task,value) \
254 (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_UAC_MASK) \
255 | (((value) << IA64_THREAD_UAC_SHIFT) & IA64_THREAD_UAC_MASK)); \
258 #define GET_UNALIGN_CTL(task,addr) \
260 put_user(((task)->thread.flags & IA64_THREAD_UAC_MASK) >> IA64_THREAD_UAC_SHIFT, \
261 (int __user *) (addr)); \
264 #define SET_FPEMU_CTL(task,value) \
266 (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_FPEMU_MASK) \
267 | (((value) << IA64_THREAD_FPEMU_SHIFT) & IA64_THREAD_FPEMU_MASK)); \
270 #define GET_FPEMU_CTL(task,addr) \
272 put_user(((task)->thread.flags & IA64_THREAD_FPEMU_MASK) >> IA64_THREAD_FPEMU_SHIFT, \
273 (int __user *) (addr)); \
276 struct thread_struct {
277 __u32 flags; /* various thread flags (see IA64_THREAD_*) */
278 /* writing on_ustack is performance-critical, so it's worth spending 8 bits on it... */
279 __u8 on_ustack; /* executing on user-stacks? */
281 __u64 ksp; /* kernel stack pointer */
282 __u64 map_base; /* base address for get_unmapped_area() */
283 __u64 task_size; /* limit for task size */
284 __u64 rbs_bot; /* the base address for the RBS */
285 int last_fph_cpu; /* CPU that may hold the contents of f32-f127 */
287 #ifdef CONFIG_PERFMON
288 void *pfm_context; /* pointer to detailed PMU context */
289 unsigned long pfm_needs_checking; /* when >0, pending perfmon work on kernel exit */
290 # define INIT_THREAD_PM .pfm_context = NULL, \
291 .pfm_needs_checking = 0UL,
293 # define INIT_THREAD_PM
295 unsigned long dbr[IA64_NUM_DBG_REGS];
296 unsigned long ibr[IA64_NUM_DBG_REGS];
297 struct ia64_fpreg fph[96]; /* saved/loaded on demand */
300 #define INIT_THREAD { \
304 .map_base = DEFAULT_MAP_BASE, \
305 .rbs_bot = STACK_TOP - DEFAULT_USER_STACK_SIZE, \
306 .task_size = DEFAULT_TASK_SIZE, \
307 .last_fph_cpu = -1, \
314 #define start_thread(regs,new_ip,new_sp) do { \
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 */ \
320 regs->ar_bspstore = current->thread.rbs_bot; \
321 regs->ar_fpsr = FPSR_DEFAULT; \
323 regs->r8 = get_dumpable(current->mm); /* set "don't zap registers" flag */ \
324 regs->r12 = new_sp - 16; /* allocate 16 byte scratch area */ \
325 if (unlikely(!get_dumpable(current->mm))) { \
327 * Zap scratch regs to avoid leaking bits between processes with different \
330 regs->ar_pfs = 0; regs->b0 = 0; regs->pr = 0; \
331 regs->r1 = 0; regs->r9 = 0; regs->r11 = 0; regs->r13 = 0; regs->r15 = 0; \
335 /* Forward declarations, a strange C thing... */
340 * Free all resources held by a thread. This is called after the
341 * parent of DEAD_TASK has collected the exit status of the task via
344 #define release_thread(dead_task)
346 /* Prepare to copy thread state - unlazy all lazy status */
347 #define prepare_to_copy(tsk) do { } while (0)
350 * This is the mechanism for creating a new kernel thread.
352 * NOTE 1: Only a kernel-only process (ie the swapper or direct
353 * descendants who haven't done an "execve()") should use this: it
354 * will work within a system call from a "real" process, but the
355 * process memory space will not be free'd until both the parent and
356 * the child have exited.
358 * NOTE 2: This MUST NOT be an inlined function. Otherwise, we get
359 * into trouble in init/main.c when the child thread returns to
360 * do_basic_setup() and the timing is such that free_initmem() has
361 * been called already.
363 extern pid_t kernel_thread (int (*fn)(void *), void *arg, unsigned long flags);
365 /* Get wait channel for task P. */
366 extern unsigned long get_wchan (struct task_struct *p);
368 /* Return instruction pointer of blocked task TSK. */
369 #define KSTK_EIP(tsk) \
371 struct pt_regs *_regs = task_pt_regs(tsk); \
372 _regs->cr_iip + ia64_psr(_regs)->ri; \
375 /* Return stack pointer of blocked task TSK. */
376 #define KSTK_ESP(tsk) ((tsk)->thread.ksp)
378 extern void ia64_getreg_unknown_kr (void);
379 extern void ia64_setreg_unknown_kr (void);
381 #define ia64_get_kr(regnum) \
383 unsigned long r = 0; \
386 case 0: r = ia64_getreg(_IA64_REG_AR_KR0); break; \
387 case 1: r = ia64_getreg(_IA64_REG_AR_KR1); break; \
388 case 2: r = ia64_getreg(_IA64_REG_AR_KR2); break; \
389 case 3: r = ia64_getreg(_IA64_REG_AR_KR3); break; \
390 case 4: r = ia64_getreg(_IA64_REG_AR_KR4); break; \
391 case 5: r = ia64_getreg(_IA64_REG_AR_KR5); break; \
392 case 6: r = ia64_getreg(_IA64_REG_AR_KR6); break; \
393 case 7: r = ia64_getreg(_IA64_REG_AR_KR7); break; \
394 default: ia64_getreg_unknown_kr(); break; \
399 #define ia64_set_kr(regnum, r) \
402 case 0: ia64_setreg(_IA64_REG_AR_KR0, r); break; \
403 case 1: ia64_setreg(_IA64_REG_AR_KR1, r); break; \
404 case 2: ia64_setreg(_IA64_REG_AR_KR2, r); break; \
405 case 3: ia64_setreg(_IA64_REG_AR_KR3, r); break; \
406 case 4: ia64_setreg(_IA64_REG_AR_KR4, r); break; \
407 case 5: ia64_setreg(_IA64_REG_AR_KR5, r); break; \
408 case 6: ia64_setreg(_IA64_REG_AR_KR6, r); break; \
409 case 7: ia64_setreg(_IA64_REG_AR_KR7, r); break; \
410 default: ia64_setreg_unknown_kr(); break; \
415 * The following three macros can't be inline functions because we don't have struct
416 * task_struct at this point.
420 * Return TRUE if task T owns the fph partition of the CPU we're running on.
421 * Must be called from code that has preemption disabled.
423 #define ia64_is_local_fpu_owner(t) \
425 struct task_struct *__ia64_islfo_task = (t); \
426 (__ia64_islfo_task->thread.last_fph_cpu == smp_processor_id() \
427 && __ia64_islfo_task == (struct task_struct *) ia64_get_kr(IA64_KR_FPU_OWNER)); \
431 * Mark task T as owning the fph partition of the CPU we're running on.
432 * Must be called from code that has preemption disabled.
434 #define ia64_set_local_fpu_owner(t) do { \
435 struct task_struct *__ia64_slfo_task = (t); \
436 __ia64_slfo_task->thread.last_fph_cpu = smp_processor_id(); \
437 ia64_set_kr(IA64_KR_FPU_OWNER, (unsigned long) __ia64_slfo_task); \
440 /* Mark the fph partition of task T as being invalid on all CPUs. */
441 #define ia64_drop_fpu(t) ((t)->thread.last_fph_cpu = -1)
443 extern void __ia64_init_fpu (void);
444 extern void __ia64_save_fpu (struct ia64_fpreg *fph);
445 extern void __ia64_load_fpu (struct ia64_fpreg *fph);
446 extern void ia64_save_debug_regs (unsigned long *save_area);
447 extern void ia64_load_debug_regs (unsigned long *save_area);
449 #define ia64_fph_enable() do { ia64_rsm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
450 #define ia64_fph_disable() do { ia64_ssm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
452 /* load fp 0.0 into fph */
454 ia64_init_fpu (void) {
460 /* save f32-f127 at FPH */
462 ia64_save_fpu (struct ia64_fpreg *fph) {
464 __ia64_save_fpu(fph);
468 /* load f32-f127 from FPH */
470 ia64_load_fpu (struct ia64_fpreg *fph) {
472 __ia64_load_fpu(fph);
480 psr = ia64_getreg(_IA64_REG_PSR);
482 ia64_rsm(IA64_PSR_I | IA64_PSR_IC);
491 ia64_set_psr (__u64 psr)
494 ia64_setreg(_IA64_REG_PSR_L, psr);
499 * Insert a translation into an instruction and/or data translation
503 ia64_itr (__u64 target_mask, __u64 tr_num,
504 __u64 vmaddr, __u64 pte,
507 ia64_setreg(_IA64_REG_CR_ITIR, (log_page_size << 2));
508 ia64_setreg(_IA64_REG_CR_IFA, vmaddr);
510 if (target_mask & 0x1)
511 ia64_itri(tr_num, pte);
512 if (target_mask & 0x2)
513 ia64_itrd(tr_num, pte);
517 * Insert a translation into the instruction and/or data translation
521 ia64_itc (__u64 target_mask, __u64 vmaddr, __u64 pte,
524 ia64_setreg(_IA64_REG_CR_ITIR, (log_page_size << 2));
525 ia64_setreg(_IA64_REG_CR_IFA, vmaddr);
527 /* as per EAS2.6, itc must be the last instruction in an instruction group */
528 if (target_mask & 0x1)
530 if (target_mask & 0x2)
535 * Purge a range of addresses from instruction and/or data translation
539 ia64_ptr (__u64 target_mask, __u64 vmaddr, __u64 log_size)
541 if (target_mask & 0x1)
542 ia64_ptri(vmaddr, (log_size << 2));
543 if (target_mask & 0x2)
544 ia64_ptrd(vmaddr, (log_size << 2));
547 /* Set the interrupt vector address. The address must be suitably aligned (32KB). */
549 ia64_set_iva (void *ivt_addr)
551 ia64_setreg(_IA64_REG_CR_IVA, (__u64) ivt_addr);
555 /* Set the page table address and control bits. */
557 ia64_set_pta (__u64 pta)
559 /* Note: srlz.i implies srlz.d */
560 ia64_setreg(_IA64_REG_CR_PTA, pta);
567 ia64_setreg(_IA64_REG_CR_EOI, 0);
571 #define cpu_relax() ia64_hint(ia64_hint_pause)
574 ia64_get_irr(unsigned int vector)
576 unsigned int reg = vector / 64;
577 unsigned int bit = vector % 64;
581 case 0: irr = ia64_getreg(_IA64_REG_CR_IRR0); break;
582 case 1: irr = ia64_getreg(_IA64_REG_CR_IRR1); break;
583 case 2: irr = ia64_getreg(_IA64_REG_CR_IRR2); break;
584 case 3: irr = ia64_getreg(_IA64_REG_CR_IRR3); break;
587 return test_bit(bit, &irr);
591 ia64_set_lrr0 (unsigned long val)
593 ia64_setreg(_IA64_REG_CR_LRR0, val);
598 ia64_set_lrr1 (unsigned long val)
600 ia64_setreg(_IA64_REG_CR_LRR1, val);
606 * Given the address to which a spill occurred, return the unat bit
607 * number that corresponds to this address.
610 ia64_unat_pos (void *spill_addr)
612 return ((__u64) spill_addr >> 3) & 0x3f;
616 * Set the NaT bit of an integer register which was spilled at address
617 * SPILL_ADDR. UNAT is the mask to be updated.
620 ia64_set_unat (__u64 *unat, void *spill_addr, unsigned long nat)
622 __u64 bit = ia64_unat_pos(spill_addr);
623 __u64 mask = 1UL << bit;
625 *unat = (*unat & ~mask) | (nat << bit);
629 * Return saved PC of a blocked thread.
630 * Note that the only way T can block is through a call to schedule() -> switch_to().
632 static inline unsigned long
633 thread_saved_pc (struct task_struct *t)
635 struct unw_frame_info info;
638 unw_init_from_blocked_task(&info, t);
639 if (unw_unwind(&info) < 0)
641 unw_get_ip(&info, &ip);
646 * Get the current instruction/program counter value.
648 #define current_text_addr() \
649 ({ void *_pc; _pc = (void *)ia64_getreg(_IA64_REG_IP); _pc; })
656 r = ia64_getreg(_IA64_REG_CR_IVR);
662 ia64_set_dbr (__u64 regnum, __u64 value)
664 __ia64_set_dbr(regnum, value);
665 #ifdef CONFIG_ITANIUM
671 ia64_get_dbr (__u64 regnum)
675 retval = __ia64_get_dbr(regnum);
676 #ifdef CONFIG_ITANIUM
683 ia64_rotr (__u64 w, __u64 n)
685 return (w >> n) | (w << (64 - n));
688 #define ia64_rotl(w,n) ia64_rotr((w), (64) - (n))
691 * Take a mapped kernel address and return the equivalent address
692 * in the region 7 identity mapped virtual area.
695 ia64_imva (void *addr)
698 result = (void *) ia64_tpa(addr);
702 #define ARCH_HAS_PREFETCH
703 #define ARCH_HAS_PREFETCHW
704 #define ARCH_HAS_SPINLOCK_PREFETCH
705 #define PREFETCH_STRIDE L1_CACHE_BYTES
708 prefetch (const void *x)
710 ia64_lfetch(ia64_lfhint_none, x);
714 prefetchw (const void *x)
716 ia64_lfetch_excl(ia64_lfhint_none, x);
719 #define spin_lock_prefetch(x) prefetchw(x)
721 extern unsigned long boot_option_idle_override;
723 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_FORCE_MWAIT,
724 IDLE_NOMWAIT, IDLE_POLL};
726 void cpu_idle_wait(void);
727 void default_idle(void);
729 #define ia64_platform_is(x) (strcmp(x, platform_name) == 0)
731 #endif /* !__ASSEMBLY__ */
733 #endif /* _ASM_IA64_PROCESSOR_H */