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
16 #include <linux/config.h>
18 #include <asm/intrinsics.h>
19 #include <asm/kregs.h>
20 #include <asm/ptrace.h>
21 #include <asm/ustack.h>
23 /* Our arch specific arch_init_sched_domain is in arch/ia64/kernel/domain.c */
24 #define ARCH_HAS_SCHED_DOMAIN
26 #define IA64_NUM_DBG_REGS 8
28 * Limits for PMC and PMD are set to less than maximum architected values
29 * but should be sufficient for a while
31 #define IA64_NUM_PMC_REGS 32
32 #define IA64_NUM_PMD_REGS 32
34 #define DEFAULT_MAP_BASE __IA64_UL_CONST(0x2000000000000000)
35 #define DEFAULT_TASK_SIZE __IA64_UL_CONST(0xa000000000000000)
38 * TASK_SIZE really is a mis-named. It really is the maximum user
39 * space address (plus one). On IA-64, there are five regions of 2TB
40 * each (assuming 8KB page size), for a total of 8TB of user virtual
43 #define TASK_SIZE (current->thread.task_size)
46 * This decides where the kernel will search for a free chunk of vm
47 * space during mmap's.
49 #define TASK_UNMAPPED_BASE (current->thread.map_base)
51 #define IA64_THREAD_FPH_VALID (__IA64_UL(1) << 0) /* floating-point high state valid? */
52 #define IA64_THREAD_DBG_VALID (__IA64_UL(1) << 1) /* debug registers valid? */
53 #define IA64_THREAD_PM_VALID (__IA64_UL(1) << 2) /* performance registers valid? */
54 #define IA64_THREAD_UAC_NOPRINT (__IA64_UL(1) << 3) /* don't log unaligned accesses */
55 #define IA64_THREAD_UAC_SIGBUS (__IA64_UL(1) << 4) /* generate SIGBUS on unaligned acc. */
56 /* bit 5 is currently unused */
57 #define IA64_THREAD_FPEMU_NOPRINT (__IA64_UL(1) << 6) /* don't log any fpswa faults */
58 #define IA64_THREAD_FPEMU_SIGFPE (__IA64_UL(1) << 7) /* send a SIGFPE for fpswa faults */
60 #define IA64_THREAD_UAC_SHIFT 3
61 #define IA64_THREAD_UAC_MASK (IA64_THREAD_UAC_NOPRINT | IA64_THREAD_UAC_SIGBUS)
62 #define IA64_THREAD_FPEMU_SHIFT 6
63 #define IA64_THREAD_FPEMU_MASK (IA64_THREAD_FPEMU_NOPRINT | IA64_THREAD_FPEMU_SIGFPE)
67 * This shift should be large enough to be able to represent 1000000000/itc_freq with good
68 * accuracy while being small enough to fit 10*1000000000<<IA64_NSEC_PER_CYC_SHIFT in 64 bits
69 * (this will give enough slack to represent 10 seconds worth of time as a scaled number).
71 #define IA64_NSEC_PER_CYC_SHIFT 30
75 #include <linux/cache.h>
76 #include <linux/compiler.h>
77 #include <linux/threads.h>
78 #include <linux/types.h>
82 #include <asm/percpu.h>
84 #include <asm/unwind.h>
85 #include <asm/atomic.h>
87 #include <asm/nodedata.h>
90 /* like above but expressed as bitfields for more efficient access: */
126 __u64 reserved4
: 19;
130 * CPU type, hardware bug flags, and per-CPU state. Frequently used
131 * state comes earlier:
133 struct cpuinfo_ia64
{
134 __u32 softirq_pending
;
135 __u64 itm_delta
; /* # of clock cycles between clock ticks */
136 __u64 itm_next
; /* interval timer mask value to use for next clock tick */
137 __u64 nsec_per_cyc
; /* (1000000000<<IA64_NSEC_PER_CYC_SHIFT)/itc_freq */
138 __u64 unimpl_va_mask
; /* mask of unimplemented virtual address bits (from PAL) */
139 __u64 unimpl_pa_mask
; /* mask of unimplemented physical address bits (from PAL) */
140 __u64 itc_freq
; /* frequency of ITC counter */
141 __u64 proc_freq
; /* frequency of processor */
142 __u64 cyc_per_usec
; /* itc_freq/1000000 */
145 __u32 ptce_stride
[2];
146 struct task_struct
*ksoftirqd
; /* kernel softirq daemon for this CPU */
149 __u64 loops_per_jiffy
;
151 __u32 socket_id
; /* physical processor socket id */
152 __u16 core_id
; /* core id */
153 __u16 thread_id
; /* thread id */
154 __u16 num_log
; /* Total number of logical processors on
155 * this socket that were successfully booted */
156 __u8 cores_per_socket
; /* Cores per processor socket */
157 __u8 threads_per_core
; /* Threads per core */
160 /* CPUID-derived information: */
171 struct ia64_node_data
*node_data
;
175 DECLARE_PER_CPU(struct cpuinfo_ia64
, cpu_info
);
178 * The "local" data variable. It refers to the per-CPU data of the currently executing
179 * CPU, much like "current" points to the per-task data of the currently executing task.
180 * Do not use the address of local_cpu_data, since it will be different from
181 * cpu_data(smp_processor_id())!
183 #define local_cpu_data (&__ia64_per_cpu_var(cpu_info))
184 #define cpu_data(cpu) (&per_cpu(cpu_info, cpu))
186 extern void identify_cpu (struct cpuinfo_ia64
*);
187 extern void print_cpu_info (struct cpuinfo_ia64
*);
193 #define SET_UNALIGN_CTL(task,value) \
195 (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_UAC_MASK) \
196 | (((value) << IA64_THREAD_UAC_SHIFT) & IA64_THREAD_UAC_MASK)); \
199 #define GET_UNALIGN_CTL(task,addr) \
201 put_user(((task)->thread.flags & IA64_THREAD_UAC_MASK) >> IA64_THREAD_UAC_SHIFT, \
202 (int __user *) (addr)); \
205 #define SET_FPEMU_CTL(task,value) \
207 (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_FPEMU_MASK) \
208 | (((value) << IA64_THREAD_FPEMU_SHIFT) & IA64_THREAD_FPEMU_MASK)); \
211 #define GET_FPEMU_CTL(task,addr) \
213 put_user(((task)->thread.flags & IA64_THREAD_FPEMU_MASK) >> IA64_THREAD_FPEMU_SHIFT, \
214 (int __user *) (addr)); \
217 #ifdef CONFIG_IA32_SUPPORT
222 #define desc_empty(desc) (!((desc)->a + (desc)->b))
223 #define desc_equal(desc1, desc2) (((desc1)->a == (desc2)->a) && ((desc1)->b == (desc2)->b))
225 #define GDT_ENTRY_TLS_ENTRIES 3
226 #define GDT_ENTRY_TLS_MIN 6
227 #define GDT_ENTRY_TLS_MAX (GDT_ENTRY_TLS_MIN + GDT_ENTRY_TLS_ENTRIES - 1)
229 #define TLS_SIZE (GDT_ENTRY_TLS_ENTRIES * 8)
231 struct partial_page_list
;
234 struct thread_struct
{
235 __u32 flags
; /* various thread flags (see IA64_THREAD_*) */
236 /* writing on_ustack is performance-critical, so it's worth spending 8 bits on it... */
237 __u8 on_ustack
; /* executing on user-stacks? */
239 __u64 ksp
; /* kernel stack pointer */
240 __u64 map_base
; /* base address for get_unmapped_area() */
241 __u64 task_size
; /* limit for task size */
242 __u64 rbs_bot
; /* the base address for the RBS */
243 int last_fph_cpu
; /* CPU that may hold the contents of f32-f127 */
245 #ifdef CONFIG_IA32_SUPPORT
246 __u64 eflag
; /* IA32 EFLAGS reg */
247 __u64 fsr
; /* IA32 floating pt status reg */
248 __u64 fcr
; /* IA32 floating pt control reg */
249 __u64 fir
; /* IA32 fp except. instr. reg */
250 __u64 fdr
; /* IA32 fp except. data reg */
251 __u64 old_k1
; /* old value of ar.k1 */
252 __u64 old_iob
; /* old IOBase value */
253 struct partial_page_list
*ppl
; /* partial page list for 4K page size issue */
254 /* cached TLS descriptors. */
255 struct desc_struct tls_array
[GDT_ENTRY_TLS_ENTRIES
];
257 # define INIT_THREAD_IA32 .eflag = 0, \
259 .fcr = 0x17800000037fULL, \
266 # define INIT_THREAD_IA32
267 #endif /* CONFIG_IA32_SUPPORT */
268 #ifdef CONFIG_PERFMON
269 __u64 pmcs
[IA64_NUM_PMC_REGS
];
270 __u64 pmds
[IA64_NUM_PMD_REGS
];
271 void *pfm_context
; /* pointer to detailed PMU context */
272 unsigned long pfm_needs_checking
; /* when >0, pending perfmon work on kernel exit */
273 # define INIT_THREAD_PM .pmcs = {0UL, }, \
275 .pfm_context = NULL, \
276 .pfm_needs_checking = 0UL,
278 # define INIT_THREAD_PM
280 __u64 dbr
[IA64_NUM_DBG_REGS
];
281 __u64 ibr
[IA64_NUM_DBG_REGS
];
282 struct ia64_fpreg fph
[96]; /* saved/loaded on demand */
285 #define INIT_THREAD { \
289 .map_base = DEFAULT_MAP_BASE, \
290 .rbs_bot = STACK_TOP - DEFAULT_USER_STACK_SIZE, \
291 .task_size = DEFAULT_TASK_SIZE, \
292 .last_fph_cpu = -1, \
300 #define start_thread(regs,new_ip,new_sp) do { \
302 regs->cr_ipsr = ((regs->cr_ipsr | (IA64_PSR_BITS_TO_SET | IA64_PSR_CPL)) \
303 & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_RI | IA64_PSR_IS)); \
304 regs->cr_iip = new_ip; \
305 regs->ar_rsc = 0xf; /* eager mode, privilege level 3 */ \
307 regs->ar_bspstore = current->thread.rbs_bot; \
308 regs->ar_fpsr = FPSR_DEFAULT; \
310 regs->r8 = current->mm->dumpable; /* set "don't zap registers" flag */ \
311 regs->r12 = new_sp - 16; /* allocate 16 byte scratch area */ \
312 if (unlikely(!current->mm->dumpable)) { \
314 * Zap scratch regs to avoid leaking bits between processes with different \
317 regs->ar_pfs = 0; regs->b0 = 0; regs->pr = 0; \
318 regs->r1 = 0; regs->r9 = 0; regs->r11 = 0; regs->r13 = 0; regs->r15 = 0; \
322 /* Forward declarations, a strange C thing... */
327 * Free all resources held by a thread. This is called after the
328 * parent of DEAD_TASK has collected the exit status of the task via
331 #define release_thread(dead_task)
333 /* Prepare to copy thread state - unlazy all lazy status */
334 #define prepare_to_copy(tsk) do { } while (0)
337 * This is the mechanism for creating a new kernel thread.
339 * NOTE 1: Only a kernel-only process (ie the swapper or direct
340 * descendants who haven't done an "execve()") should use this: it
341 * will work within a system call from a "real" process, but the
342 * process memory space will not be free'd until both the parent and
343 * the child have exited.
345 * NOTE 2: This MUST NOT be an inlined function. Otherwise, we get
346 * into trouble in init/main.c when the child thread returns to
347 * do_basic_setup() and the timing is such that free_initmem() has
348 * been called already.
350 extern pid_t
kernel_thread (int (*fn
)(void *), void *arg
, unsigned long flags
);
352 /* Get wait channel for task P. */
353 extern unsigned long get_wchan (struct task_struct
*p
);
355 /* Return instruction pointer of blocked task TSK. */
356 #define KSTK_EIP(tsk) \
358 struct pt_regs *_regs = ia64_task_regs(tsk); \
359 _regs->cr_iip + ia64_psr(_regs)->ri; \
362 /* Return stack pointer of blocked task TSK. */
363 #define KSTK_ESP(tsk) ((tsk)->thread.ksp)
365 extern void ia64_getreg_unknown_kr (void);
366 extern void ia64_setreg_unknown_kr (void);
368 #define ia64_get_kr(regnum) \
370 unsigned long r = 0; \
373 case 0: r = ia64_getreg(_IA64_REG_AR_KR0); break; \
374 case 1: r = ia64_getreg(_IA64_REG_AR_KR1); break; \
375 case 2: r = ia64_getreg(_IA64_REG_AR_KR2); break; \
376 case 3: r = ia64_getreg(_IA64_REG_AR_KR3); break; \
377 case 4: r = ia64_getreg(_IA64_REG_AR_KR4); break; \
378 case 5: r = ia64_getreg(_IA64_REG_AR_KR5); break; \
379 case 6: r = ia64_getreg(_IA64_REG_AR_KR6); break; \
380 case 7: r = ia64_getreg(_IA64_REG_AR_KR7); break; \
381 default: ia64_getreg_unknown_kr(); break; \
386 #define ia64_set_kr(regnum, r) \
389 case 0: ia64_setreg(_IA64_REG_AR_KR0, r); break; \
390 case 1: ia64_setreg(_IA64_REG_AR_KR1, r); break; \
391 case 2: ia64_setreg(_IA64_REG_AR_KR2, r); break; \
392 case 3: ia64_setreg(_IA64_REG_AR_KR3, r); break; \
393 case 4: ia64_setreg(_IA64_REG_AR_KR4, r); break; \
394 case 5: ia64_setreg(_IA64_REG_AR_KR5, r); break; \
395 case 6: ia64_setreg(_IA64_REG_AR_KR6, r); break; \
396 case 7: ia64_setreg(_IA64_REG_AR_KR7, r); break; \
397 default: ia64_setreg_unknown_kr(); break; \
402 * The following three macros can't be inline functions because we don't have struct
403 * task_struct at this point.
406 /* Return TRUE if task T owns the fph partition of the CPU we're running on. */
407 #define ia64_is_local_fpu_owner(t) \
409 struct task_struct *__ia64_islfo_task = (t); \
410 (__ia64_islfo_task->thread.last_fph_cpu == smp_processor_id() \
411 && __ia64_islfo_task == (struct task_struct *) ia64_get_kr(IA64_KR_FPU_OWNER)); \
414 /* Mark task T as owning the fph partition of the CPU we're running on. */
415 #define ia64_set_local_fpu_owner(t) do { \
416 struct task_struct *__ia64_slfo_task = (t); \
417 __ia64_slfo_task->thread.last_fph_cpu = smp_processor_id(); \
418 ia64_set_kr(IA64_KR_FPU_OWNER, (unsigned long) __ia64_slfo_task); \
421 /* Mark the fph partition of task T as being invalid on all CPUs. */
422 #define ia64_drop_fpu(t) ((t)->thread.last_fph_cpu = -1)
424 extern void __ia64_init_fpu (void);
425 extern void __ia64_save_fpu (struct ia64_fpreg
*fph
);
426 extern void __ia64_load_fpu (struct ia64_fpreg
*fph
);
427 extern void ia64_save_debug_regs (unsigned long *save_area
);
428 extern void ia64_load_debug_regs (unsigned long *save_area
);
430 #ifdef CONFIG_IA32_SUPPORT
431 extern void ia32_save_state (struct task_struct
*task
);
432 extern void ia32_load_state (struct task_struct
*task
);
435 #define ia64_fph_enable() do { ia64_rsm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
436 #define ia64_fph_disable() do { ia64_ssm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
438 /* load fp 0.0 into fph */
440 ia64_init_fpu (void) {
446 /* save f32-f127 at FPH */
448 ia64_save_fpu (struct ia64_fpreg
*fph
) {
450 __ia64_save_fpu(fph
);
454 /* load f32-f127 from FPH */
456 ia64_load_fpu (struct ia64_fpreg
*fph
) {
458 __ia64_load_fpu(fph
);
466 psr
= ia64_getreg(_IA64_REG_PSR
);
468 ia64_rsm(IA64_PSR_I
| IA64_PSR_IC
);
477 ia64_set_psr (__u64 psr
)
480 ia64_setreg(_IA64_REG_PSR_L
, psr
);
485 * Insert a translation into an instruction and/or data translation
489 ia64_itr (__u64 target_mask
, __u64 tr_num
,
490 __u64 vmaddr
, __u64 pte
,
493 ia64_setreg(_IA64_REG_CR_ITIR
, (log_page_size
<< 2));
494 ia64_setreg(_IA64_REG_CR_IFA
, vmaddr
);
496 if (target_mask
& 0x1)
497 ia64_itri(tr_num
, pte
);
498 if (target_mask
& 0x2)
499 ia64_itrd(tr_num
, pte
);
503 * Insert a translation into the instruction and/or data translation
507 ia64_itc (__u64 target_mask
, __u64 vmaddr
, __u64 pte
,
510 ia64_setreg(_IA64_REG_CR_ITIR
, (log_page_size
<< 2));
511 ia64_setreg(_IA64_REG_CR_IFA
, vmaddr
);
513 /* as per EAS2.6, itc must be the last instruction in an instruction group */
514 if (target_mask
& 0x1)
516 if (target_mask
& 0x2)
521 * Purge a range of addresses from instruction and/or data translation
525 ia64_ptr (__u64 target_mask
, __u64 vmaddr
, __u64 log_size
)
527 if (target_mask
& 0x1)
528 ia64_ptri(vmaddr
, (log_size
<< 2));
529 if (target_mask
& 0x2)
530 ia64_ptrd(vmaddr
, (log_size
<< 2));
533 /* Set the interrupt vector address. The address must be suitably aligned (32KB). */
535 ia64_set_iva (void *ivt_addr
)
537 ia64_setreg(_IA64_REG_CR_IVA
, (__u64
) ivt_addr
);
541 /* Set the page table address and control bits. */
543 ia64_set_pta (__u64 pta
)
545 /* Note: srlz.i implies srlz.d */
546 ia64_setreg(_IA64_REG_CR_PTA
, pta
);
553 ia64_setreg(_IA64_REG_CR_EOI
, 0);
557 #define cpu_relax() ia64_hint(ia64_hint_pause)
560 ia64_set_lrr0 (unsigned long val
)
562 ia64_setreg(_IA64_REG_CR_LRR0
, val
);
567 ia64_set_lrr1 (unsigned long val
)
569 ia64_setreg(_IA64_REG_CR_LRR1
, val
);
575 * Given the address to which a spill occurred, return the unat bit
576 * number that corresponds to this address.
579 ia64_unat_pos (void *spill_addr
)
581 return ((__u64
) spill_addr
>> 3) & 0x3f;
585 * Set the NaT bit of an integer register which was spilled at address
586 * SPILL_ADDR. UNAT is the mask to be updated.
589 ia64_set_unat (__u64
*unat
, void *spill_addr
, unsigned long nat
)
591 __u64 bit
= ia64_unat_pos(spill_addr
);
592 __u64 mask
= 1UL << bit
;
594 *unat
= (*unat
& ~mask
) | (nat
<< bit
);
598 * Return saved PC of a blocked thread.
599 * Note that the only way T can block is through a call to schedule() -> switch_to().
601 static inline unsigned long
602 thread_saved_pc (struct task_struct
*t
)
604 struct unw_frame_info info
;
607 unw_init_from_blocked_task(&info
, t
);
608 if (unw_unwind(&info
) < 0)
610 unw_get_ip(&info
, &ip
);
615 * Get the current instruction/program counter value.
617 #define current_text_addr() \
618 ({ void *_pc; _pc = (void *)ia64_getreg(_IA64_REG_IP); _pc; })
625 r
= ia64_getreg(_IA64_REG_CR_IVR
);
631 ia64_set_dbr (__u64 regnum
, __u64 value
)
633 __ia64_set_dbr(regnum
, value
);
634 #ifdef CONFIG_ITANIUM
640 ia64_get_dbr (__u64 regnum
)
644 retval
= __ia64_get_dbr(regnum
);
645 #ifdef CONFIG_ITANIUM
652 ia64_rotr (__u64 w
, __u64 n
)
654 return (w
>> n
) | (w
<< (64 - n
));
657 #define ia64_rotl(w,n) ia64_rotr((w), (64) - (n))
660 * Take a mapped kernel address and return the equivalent address
661 * in the region 7 identity mapped virtual area.
664 ia64_imva (void *addr
)
667 result
= (void *) ia64_tpa(addr
);
671 #define ARCH_HAS_PREFETCH
672 #define ARCH_HAS_PREFETCHW
673 #define ARCH_HAS_SPINLOCK_PREFETCH
674 #define PREFETCH_STRIDE L1_CACHE_BYTES
677 prefetch (const void *x
)
679 ia64_lfetch(ia64_lfhint_none
, x
);
683 prefetchw (const void *x
)
685 ia64_lfetch_excl(ia64_lfhint_none
, x
);
688 #define spin_lock_prefetch(x) prefetchw(x)
690 extern unsigned long boot_option_idle_override
;
692 #endif /* !__ASSEMBLY__ */
694 #endif /* _ASM_IA64_PROCESSOR_H */