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 * MM_VM_SIZE(mm) gives the maximum address (plus 1) which may contain a mapping for
47 * address-space MM. Note that with 32-bit tasks, this is still DEFAULT_TASK_SIZE,
48 * because the kernel may have installed helper-mappings above TASK_SIZE. For example,
49 * for x86 emulation, the LDT and GDT are mapped above TASK_SIZE.
51 #define MM_VM_SIZE(mm) DEFAULT_TASK_SIZE
54 * This decides where the kernel will search for a free chunk of vm
55 * space during mmap's.
57 #define TASK_UNMAPPED_BASE (current->thread.map_base)
59 #define IA64_THREAD_FPH_VALID (__IA64_UL(1) << 0) /* floating-point high state valid? */
60 #define IA64_THREAD_DBG_VALID (__IA64_UL(1) << 1) /* debug registers valid? */
61 #define IA64_THREAD_PM_VALID (__IA64_UL(1) << 2) /* performance registers valid? */
62 #define IA64_THREAD_UAC_NOPRINT (__IA64_UL(1) << 3) /* don't log unaligned accesses */
63 #define IA64_THREAD_UAC_SIGBUS (__IA64_UL(1) << 4) /* generate SIGBUS on unaligned acc. */
64 /* bit 5 is currently unused */
65 #define IA64_THREAD_FPEMU_NOPRINT (__IA64_UL(1) << 6) /* don't log any fpswa faults */
66 #define IA64_THREAD_FPEMU_SIGFPE (__IA64_UL(1) << 7) /* send a SIGFPE for fpswa faults */
68 #define IA64_THREAD_UAC_SHIFT 3
69 #define IA64_THREAD_UAC_MASK (IA64_THREAD_UAC_NOPRINT | IA64_THREAD_UAC_SIGBUS)
70 #define IA64_THREAD_FPEMU_SHIFT 6
71 #define IA64_THREAD_FPEMU_MASK (IA64_THREAD_FPEMU_NOPRINT | IA64_THREAD_FPEMU_SIGFPE)
75 * This shift should be large enough to be able to represent 1000000000/itc_freq with good
76 * accuracy while being small enough to fit 10*1000000000<<IA64_NSEC_PER_CYC_SHIFT in 64 bits
77 * (this will give enough slack to represent 10 seconds worth of time as a scaled number).
79 #define IA64_NSEC_PER_CYC_SHIFT 30
83 #include <linux/cache.h>
84 #include <linux/compiler.h>
85 #include <linux/threads.h>
86 #include <linux/types.h>
90 #include <asm/percpu.h>
92 #include <asm/unwind.h>
93 #include <asm/atomic.h>
95 #include <asm/nodedata.h>
98 /* like above but expressed as bitfields for more efficient access: */
134 __u64 reserved4 : 19;
138 * CPU type, hardware bug flags, and per-CPU state. Frequently used
139 * state comes earlier:
141 struct cpuinfo_ia64 {
142 __u32 softirq_pending;
143 __u64 itm_delta; /* # of clock cycles between clock ticks */
144 __u64 itm_next; /* interval timer mask value to use for next clock tick */
145 __u64 nsec_per_cyc; /* (1000000000<<IA64_NSEC_PER_CYC_SHIFT)/itc_freq */
146 __u64 unimpl_va_mask; /* mask of unimplemented virtual address bits (from PAL) */
147 __u64 unimpl_pa_mask; /* mask of unimplemented physical address bits (from PAL) */
150 __u64 pgtable_cache_sz;
151 __u64 itc_freq; /* frequency of ITC counter */
152 __u64 proc_freq; /* frequency of processor */
153 __u64 cyc_per_usec; /* itc_freq/1000000 */
156 __u32 ptce_stride[2];
157 struct task_struct *ksoftirqd; /* kernel softirq daemon for this CPU */
160 __u64 loops_per_jiffy;
164 /* CPUID-derived information: */
175 struct ia64_node_data *node_data;
179 DECLARE_PER_CPU(struct cpuinfo_ia64, cpu_info);
182 * The "local" data variable. It refers to the per-CPU data of the currently executing
183 * CPU, much like "current" points to the per-task data of the currently executing task.
184 * Do not use the address of local_cpu_data, since it will be different from
185 * cpu_data(smp_processor_id())!
187 #define local_cpu_data (&__ia64_per_cpu_var(cpu_info))
188 #define cpu_data(cpu) (&per_cpu(cpu_info, cpu))
190 extern void identify_cpu (struct cpuinfo_ia64 *);
191 extern void print_cpu_info (struct cpuinfo_ia64 *);
197 #define SET_UNALIGN_CTL(task,value) \
199 (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_UAC_MASK) \
200 | (((value) << IA64_THREAD_UAC_SHIFT) & IA64_THREAD_UAC_MASK)); \
203 #define GET_UNALIGN_CTL(task,addr) \
205 put_user(((task)->thread.flags & IA64_THREAD_UAC_MASK) >> IA64_THREAD_UAC_SHIFT, \
206 (int __user *) (addr)); \
209 #define SET_FPEMU_CTL(task,value) \
211 (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_FPEMU_MASK) \
212 | (((value) << IA64_THREAD_FPEMU_SHIFT) & IA64_THREAD_FPEMU_MASK)); \
215 #define GET_FPEMU_CTL(task,addr) \
217 put_user(((task)->thread.flags & IA64_THREAD_FPEMU_MASK) >> IA64_THREAD_FPEMU_SHIFT, \
218 (int __user *) (addr)); \
221 #ifdef CONFIG_IA32_SUPPORT
226 #define desc_empty(desc) (!((desc)->a + (desc)->b))
227 #define desc_equal(desc1, desc2) (((desc1)->a == (desc2)->a) && ((desc1)->b == (desc2)->b))
229 #define GDT_ENTRY_TLS_ENTRIES 3
230 #define GDT_ENTRY_TLS_MIN 6
231 #define GDT_ENTRY_TLS_MAX (GDT_ENTRY_TLS_MIN + GDT_ENTRY_TLS_ENTRIES - 1)
233 #define TLS_SIZE (GDT_ENTRY_TLS_ENTRIES * 8)
235 struct partial_page_list;
238 struct thread_struct {
239 __u32 flags; /* various thread flags (see IA64_THREAD_*) */
240 /* writing on_ustack is performance-critical, so it's worth spending 8 bits on it... */
241 __u8 on_ustack; /* executing on user-stacks? */
243 __u64 ksp; /* kernel stack pointer */
244 __u64 map_base; /* base address for get_unmapped_area() */
245 __u64 task_size; /* limit for task size */
246 __u64 rbs_bot; /* the base address for the RBS */
247 int last_fph_cpu; /* CPU that may hold the contents of f32-f127 */
249 #ifdef CONFIG_IA32_SUPPORT
250 __u64 eflag; /* IA32 EFLAGS reg */
251 __u64 fsr; /* IA32 floating pt status reg */
252 __u64 fcr; /* IA32 floating pt control reg */
253 __u64 fir; /* IA32 fp except. instr. reg */
254 __u64 fdr; /* IA32 fp except. data reg */
255 __u64 old_k1; /* old value of ar.k1 */
256 __u64 old_iob; /* old IOBase value */
257 struct partial_page_list *ppl; /* partial page list for 4K page size issue */
258 /* cached TLS descriptors. */
259 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
261 # define INIT_THREAD_IA32 .eflag = 0, \
263 .fcr = 0x17800000037fULL, \
270 # define INIT_THREAD_IA32
271 #endif /* CONFIG_IA32_SUPPORT */
272 #ifdef CONFIG_PERFMON
273 __u64 pmcs[IA64_NUM_PMC_REGS];
274 __u64 pmds[IA64_NUM_PMD_REGS];
275 void *pfm_context; /* pointer to detailed PMU context */
276 unsigned long pfm_needs_checking; /* when >0, pending perfmon work on kernel exit */
277 # define INIT_THREAD_PM .pmcs = {0UL, }, \
279 .pfm_context = NULL, \
280 .pfm_needs_checking = 0UL,
282 # define INIT_THREAD_PM
284 __u64 dbr[IA64_NUM_DBG_REGS];
285 __u64 ibr[IA64_NUM_DBG_REGS];
286 struct ia64_fpreg fph[96]; /* saved/loaded on demand */
289 #define INIT_THREAD { \
293 .map_base = DEFAULT_MAP_BASE, \
294 .rbs_bot = STACK_TOP - DEFAULT_USER_STACK_SIZE, \
295 .task_size = DEFAULT_TASK_SIZE, \
296 .last_fph_cpu = -1, \
304 #define start_thread(regs,new_ip,new_sp) do { \
306 regs->cr_ipsr = ((regs->cr_ipsr | (IA64_PSR_BITS_TO_SET | IA64_PSR_CPL)) \
307 & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_RI | IA64_PSR_IS)); \
308 regs->cr_iip = new_ip; \
309 regs->ar_rsc = 0xf; /* eager mode, privilege level 3 */ \
311 regs->ar_bspstore = current->thread.rbs_bot; \
312 regs->ar_fpsr = FPSR_DEFAULT; \
314 regs->r8 = current->mm->dumpable; /* set "don't zap registers" flag */ \
315 regs->r12 = new_sp - 16; /* allocate 16 byte scratch area */ \
316 if (unlikely(!current->mm->dumpable)) { \
318 * Zap scratch regs to avoid leaking bits between processes with different \
321 regs->ar_pfs = 0; regs->b0 = 0; regs->pr = 0; \
322 regs->r1 = 0; regs->r9 = 0; regs->r11 = 0; regs->r13 = 0; regs->r15 = 0; \
326 /* Forward declarations, a strange C thing... */
331 * Free all resources held by a thread. This is called after the
332 * parent of DEAD_TASK has collected the exit status of the task via
335 #define release_thread(dead_task)
337 /* Prepare to copy thread state - unlazy all lazy status */
338 #define prepare_to_copy(tsk) do { } while (0)
341 * This is the mechanism for creating a new kernel thread.
343 * NOTE 1: Only a kernel-only process (ie the swapper or direct
344 * descendants who haven't done an "execve()") should use this: it
345 * will work within a system call from a "real" process, but the
346 * process memory space will not be free'd until both the parent and
347 * the child have exited.
349 * NOTE 2: This MUST NOT be an inlined function. Otherwise, we get
350 * into trouble in init/main.c when the child thread returns to
351 * do_basic_setup() and the timing is such that free_initmem() has
352 * been called already.
354 extern pid_t kernel_thread (int (*fn)(void *), void *arg, unsigned long flags);
356 /* Get wait channel for task P. */
357 extern unsigned long get_wchan (struct task_struct *p);
359 /* Return instruction pointer of blocked task TSK. */
360 #define KSTK_EIP(tsk) \
362 struct pt_regs *_regs = ia64_task_regs(tsk); \
363 _regs->cr_iip + ia64_psr(_regs)->ri; \
366 /* Return stack pointer of blocked task TSK. */
367 #define KSTK_ESP(tsk) ((tsk)->thread.ksp)
369 extern void ia64_getreg_unknown_kr (void);
370 extern void ia64_setreg_unknown_kr (void);
372 #define ia64_get_kr(regnum) \
374 unsigned long r = 0; \
377 case 0: r = ia64_getreg(_IA64_REG_AR_KR0); break; \
378 case 1: r = ia64_getreg(_IA64_REG_AR_KR1); break; \
379 case 2: r = ia64_getreg(_IA64_REG_AR_KR2); break; \
380 case 3: r = ia64_getreg(_IA64_REG_AR_KR3); break; \
381 case 4: r = ia64_getreg(_IA64_REG_AR_KR4); break; \
382 case 5: r = ia64_getreg(_IA64_REG_AR_KR5); break; \
383 case 6: r = ia64_getreg(_IA64_REG_AR_KR6); break; \
384 case 7: r = ia64_getreg(_IA64_REG_AR_KR7); break; \
385 default: ia64_getreg_unknown_kr(); break; \
390 #define ia64_set_kr(regnum, r) \
393 case 0: ia64_setreg(_IA64_REG_AR_KR0, r); break; \
394 case 1: ia64_setreg(_IA64_REG_AR_KR1, r); break; \
395 case 2: ia64_setreg(_IA64_REG_AR_KR2, r); break; \
396 case 3: ia64_setreg(_IA64_REG_AR_KR3, r); break; \
397 case 4: ia64_setreg(_IA64_REG_AR_KR4, r); break; \
398 case 5: ia64_setreg(_IA64_REG_AR_KR5, r); break; \
399 case 6: ia64_setreg(_IA64_REG_AR_KR6, r); break; \
400 case 7: ia64_setreg(_IA64_REG_AR_KR7, r); break; \
401 default: ia64_setreg_unknown_kr(); break; \
406 * The following three macros can't be inline functions because we don't have struct
407 * task_struct at this point.
410 /* Return TRUE if task T owns the fph partition of the CPU we're running on. */
411 #define ia64_is_local_fpu_owner(t) \
413 struct task_struct *__ia64_islfo_task = (t); \
414 (__ia64_islfo_task->thread.last_fph_cpu == smp_processor_id() \
415 && __ia64_islfo_task == (struct task_struct *) ia64_get_kr(IA64_KR_FPU_OWNER)); \
418 /* Mark task T as owning the fph partition of the CPU we're running on. */
419 #define ia64_set_local_fpu_owner(t) do { \
420 struct task_struct *__ia64_slfo_task = (t); \
421 __ia64_slfo_task->thread.last_fph_cpu = smp_processor_id(); \
422 ia64_set_kr(IA64_KR_FPU_OWNER, (unsigned long) __ia64_slfo_task); \
425 /* Mark the fph partition of task T as being invalid on all CPUs. */
426 #define ia64_drop_fpu(t) ((t)->thread.last_fph_cpu = -1)
428 extern void __ia64_init_fpu (void);
429 extern void __ia64_save_fpu (struct ia64_fpreg *fph);
430 extern void __ia64_load_fpu (struct ia64_fpreg *fph);
431 extern void ia64_save_debug_regs (unsigned long *save_area);
432 extern void ia64_load_debug_regs (unsigned long *save_area);
434 #ifdef CONFIG_IA32_SUPPORT
435 extern void ia32_save_state (struct task_struct *task);
436 extern void ia32_load_state (struct task_struct *task);
439 #define ia64_fph_enable() do { ia64_rsm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
440 #define ia64_fph_disable() do { ia64_ssm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
442 /* load fp 0.0 into fph */
444 ia64_init_fpu (void) {
450 /* save f32-f127 at FPH */
452 ia64_save_fpu (struct ia64_fpreg *fph) {
454 __ia64_save_fpu(fph);
458 /* load f32-f127 from FPH */
460 ia64_load_fpu (struct ia64_fpreg *fph) {
462 __ia64_load_fpu(fph);
470 psr = ia64_getreg(_IA64_REG_PSR);
472 ia64_rsm(IA64_PSR_I | IA64_PSR_IC);
481 ia64_set_psr (__u64 psr)
484 ia64_setreg(_IA64_REG_PSR_L, psr);
489 * Insert a translation into an instruction and/or data translation
493 ia64_itr (__u64 target_mask, __u64 tr_num,
494 __u64 vmaddr, __u64 pte,
497 ia64_setreg(_IA64_REG_CR_ITIR, (log_page_size << 2));
498 ia64_setreg(_IA64_REG_CR_IFA, vmaddr);
500 if (target_mask & 0x1)
501 ia64_itri(tr_num, pte);
502 if (target_mask & 0x2)
503 ia64_itrd(tr_num, pte);
507 * Insert a translation into the instruction and/or data translation
511 ia64_itc (__u64 target_mask, __u64 vmaddr, __u64 pte,
514 ia64_setreg(_IA64_REG_CR_ITIR, (log_page_size << 2));
515 ia64_setreg(_IA64_REG_CR_IFA, vmaddr);
517 /* as per EAS2.6, itc must be the last instruction in an instruction group */
518 if (target_mask & 0x1)
520 if (target_mask & 0x2)
525 * Purge a range of addresses from instruction and/or data translation
529 ia64_ptr (__u64 target_mask, __u64 vmaddr, __u64 log_size)
531 if (target_mask & 0x1)
532 ia64_ptri(vmaddr, (log_size << 2));
533 if (target_mask & 0x2)
534 ia64_ptrd(vmaddr, (log_size << 2));
537 /* Set the interrupt vector address. The address must be suitably aligned (32KB). */
539 ia64_set_iva (void *ivt_addr)
541 ia64_setreg(_IA64_REG_CR_IVA, (__u64) ivt_addr);
545 /* Set the page table address and control bits. */
547 ia64_set_pta (__u64 pta)
549 /* Note: srlz.i implies srlz.d */
550 ia64_setreg(_IA64_REG_CR_PTA, pta);
557 ia64_setreg(_IA64_REG_CR_EOI, 0);
561 #define cpu_relax() ia64_hint(ia64_hint_pause)
564 ia64_set_lrr0 (unsigned long val)
566 ia64_setreg(_IA64_REG_CR_LRR0, val);
571 ia64_set_lrr1 (unsigned long val)
573 ia64_setreg(_IA64_REG_CR_LRR1, val);
579 * Given the address to which a spill occurred, return the unat bit
580 * number that corresponds to this address.
583 ia64_unat_pos (void *spill_addr)
585 return ((__u64) spill_addr >> 3) & 0x3f;
589 * Set the NaT bit of an integer register which was spilled at address
590 * SPILL_ADDR. UNAT is the mask to be updated.
593 ia64_set_unat (__u64 *unat, void *spill_addr, unsigned long nat)
595 __u64 bit = ia64_unat_pos(spill_addr);
596 __u64 mask = 1UL << bit;
598 *unat = (*unat & ~mask) | (nat << bit);
602 * Return saved PC of a blocked thread.
603 * Note that the only way T can block is through a call to schedule() -> switch_to().
605 static inline unsigned long
606 thread_saved_pc (struct task_struct *t)
608 struct unw_frame_info info;
611 unw_init_from_blocked_task(&info, t);
612 if (unw_unwind(&info) < 0)
614 unw_get_ip(&info, &ip);
619 * Get the current instruction/program counter value.
621 #define current_text_addr() \
622 ({ void *_pc; _pc = (void *)ia64_getreg(_IA64_REG_IP); _pc; })
629 r = ia64_getreg(_IA64_REG_CR_IVR);
635 ia64_set_dbr (__u64 regnum, __u64 value)
637 __ia64_set_dbr(regnum, value);
638 #ifdef CONFIG_ITANIUM
644 ia64_get_dbr (__u64 regnum)
648 retval = __ia64_get_dbr(regnum);
649 #ifdef CONFIG_ITANIUM
656 ia64_rotr (__u64 w, __u64 n)
658 return (w >> n) | (w << (64 - n));
661 #define ia64_rotl(w,n) ia64_rotr((w), (64) - (n))
664 * Take a mapped kernel address and return the equivalent address
665 * in the region 7 identity mapped virtual area.
668 ia64_imva (void *addr)
671 result = (void *) ia64_tpa(addr);
675 #define ARCH_HAS_PREFETCH
676 #define ARCH_HAS_PREFETCHW
677 #define ARCH_HAS_SPINLOCK_PREFETCH
678 #define PREFETCH_STRIDE L1_CACHE_BYTES
681 prefetch (const void *x)
683 ia64_lfetch(ia64_lfhint_none, x);
687 prefetchw (const void *x)
689 ia64_lfetch_excl(ia64_lfhint_none, x);
692 #define spin_lock_prefetch(x) prefetchw(x)
694 extern unsigned long boot_option_idle_override;
696 #endif /* !__ASSEMBLY__ */
698 #endif /* _ASM_IA64_PROCESSOR_H */