2 * Here is where the ball gets rolling as far as the kernel is concerned.
3 * When control is transferred to _start, the bootload has already
4 * loaded us to the correct address. All that's left to do here is
5 * to set up the kernel's global pointer and jump to the kernel
8 * Copyright (C) 1998-2001, 2003, 2005 Hewlett-Packard Co
9 * David Mosberger-Tang <davidm@hpl.hp.com>
10 * Stephane Eranian <eranian@hpl.hp.com>
11 * Copyright (C) 1999 VA Linux Systems
12 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
13 * Copyright (C) 1999 Intel Corp.
14 * Copyright (C) 1999 Asit Mallick <Asit.K.Mallick@intel.com>
15 * Copyright (C) 1999 Don Dugger <Don.Dugger@intel.com>
16 * Copyright (C) 2002 Fenghua Yu <fenghua.yu@intel.com>
17 * -Optimize __ia64_save_fpu() and __ia64_load_fpu() for Itanium 2.
18 * Copyright (C) 2004 Ashok Raj <ashok.raj@intel.com>
19 * Support for CPU Hotplug
23 #include <asm/asmmacro.h>
25 #include <asm/kregs.h>
26 #include <asm/mmu_context.h>
27 #include <asm/asm-offsets.h>
29 #include <asm/pgtable.h>
30 #include <asm/processor.h>
31 #include <asm/ptrace.h>
32 #include <asm/mca_asm.h>
33 #include <linux/init.h>
34 #include <linux/linkage.h>
36 #ifdef CONFIG_HOTPLUG_CPU
37 #define SAL_PSR_BITS_TO_SET \
38 (IA64_PSR_AC | IA64_PSR_BN | IA64_PSR_MFH | IA64_PSR_MFL)
40 #define SAVE_FROM_REG(src, ptr, dest) \
44 #define RESTORE_REG(reg, ptr, _tmp) \
45 ld8 _tmp=[ptr],0x08;; \
48 #define SAVE_BREAK_REGS(ptr, _idx, _breg, _dest)\
49 mov ar.lc=IA64_NUM_DBG_REGS-1;; \
52 SAVE_FROM_REG(_breg[_idx], ptr, _dest);; \
56 #define RESTORE_BREAK_REGS(ptr, _idx, _breg, _tmp, _lbl)\
57 mov ar.lc=IA64_NUM_DBG_REGS-1;; \
59 _lbl: RESTORE_REG(_breg[_idx], ptr, _tmp);; \
61 br.cloop.sptk.many _lbl
63 #define SAVE_ONE_RR(num, _reg, _tmp) \
64 movl _tmp=(num<<61);; \
67 #define SAVE_REGION_REGS(_tmp, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7) \
68 SAVE_ONE_RR(0,_r0, _tmp);; \
69 SAVE_ONE_RR(1,_r1, _tmp);; \
70 SAVE_ONE_RR(2,_r2, _tmp);; \
71 SAVE_ONE_RR(3,_r3, _tmp);; \
72 SAVE_ONE_RR(4,_r4, _tmp);; \
73 SAVE_ONE_RR(5,_r5, _tmp);; \
74 SAVE_ONE_RR(6,_r6, _tmp);; \
75 SAVE_ONE_RR(7,_r7, _tmp);;
77 #define STORE_REGION_REGS(ptr, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7) \
87 #define RESTORE_REGION_REGS(ptr, _idx1, _idx2, _tmp) \
91 dep.z _idx2=_idx1,61,3;; \
93 mov rr[_idx2]=_tmp;; \
96 br.cloop.sptk.few RestRR
98 #define SET_AREA_FOR_BOOTING_CPU(reg1, reg2) \
99 movl reg1=sal_state_for_booting_cpu;; \
103 * Adjust region registers saved before starting to save
104 * break regs and rest of the states that need to be preserved.
106 #define SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(_reg1,_reg2,_pred) \
107 SAVE_FROM_REG(b0,_reg1,_reg2);; \
108 SAVE_FROM_REG(b1,_reg1,_reg2);; \
109 SAVE_FROM_REG(b2,_reg1,_reg2);; \
110 SAVE_FROM_REG(b3,_reg1,_reg2);; \
111 SAVE_FROM_REG(b4,_reg1,_reg2);; \
112 SAVE_FROM_REG(b5,_reg1,_reg2);; \
113 st8 [_reg1]=r1,0x08;; \
114 st8 [_reg1]=r12,0x08;; \
115 st8 [_reg1]=r13,0x08;; \
116 SAVE_FROM_REG(ar.fpsr,_reg1,_reg2);; \
117 SAVE_FROM_REG(ar.pfs,_reg1,_reg2);; \
118 SAVE_FROM_REG(ar.rnat,_reg1,_reg2);; \
119 SAVE_FROM_REG(ar.unat,_reg1,_reg2);; \
120 SAVE_FROM_REG(ar.bspstore,_reg1,_reg2);; \
121 SAVE_FROM_REG(cr.dcr,_reg1,_reg2);; \
122 SAVE_FROM_REG(cr.iva,_reg1,_reg2);; \
123 SAVE_FROM_REG(cr.pta,_reg1,_reg2);; \
124 SAVE_FROM_REG(cr.itv,_reg1,_reg2);; \
125 SAVE_FROM_REG(cr.pmv,_reg1,_reg2);; \
126 SAVE_FROM_REG(cr.cmcv,_reg1,_reg2);; \
127 SAVE_FROM_REG(cr.lrr0,_reg1,_reg2);; \
128 SAVE_FROM_REG(cr.lrr1,_reg1,_reg2);; \
129 st8 [_reg1]=r4,0x08;; \
130 st8 [_reg1]=r5,0x08;; \
131 st8 [_reg1]=r6,0x08;; \
132 st8 [_reg1]=r7,0x08;; \
133 st8 [_reg1]=_pred,0x08;; \
134 SAVE_FROM_REG(ar.lc, _reg1, _reg2);; \
135 stf.spill.nta [_reg1]=f2,16;; \
136 stf.spill.nta [_reg1]=f3,16;; \
137 stf.spill.nta [_reg1]=f4,16;; \
138 stf.spill.nta [_reg1]=f5,16;; \
139 stf.spill.nta [_reg1]=f16,16;; \
140 stf.spill.nta [_reg1]=f17,16;; \
141 stf.spill.nta [_reg1]=f18,16;; \
142 stf.spill.nta [_reg1]=f19,16;; \
143 stf.spill.nta [_reg1]=f20,16;; \
144 stf.spill.nta [_reg1]=f21,16;; \
145 stf.spill.nta [_reg1]=f22,16;; \
146 stf.spill.nta [_reg1]=f23,16;; \
147 stf.spill.nta [_reg1]=f24,16;; \
148 stf.spill.nta [_reg1]=f25,16;; \
149 stf.spill.nta [_reg1]=f26,16;; \
150 stf.spill.nta [_reg1]=f27,16;; \
151 stf.spill.nta [_reg1]=f28,16;; \
152 stf.spill.nta [_reg1]=f29,16;; \
153 stf.spill.nta [_reg1]=f30,16;; \
154 stf.spill.nta [_reg1]=f31,16;;
157 #define SET_AREA_FOR_BOOTING_CPU(a1, a2)
158 #define SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(a1,a2, a3)
159 #define SAVE_REGION_REGS(_tmp, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7)
160 #define STORE_REGION_REGS(ptr, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7)
163 #define SET_ONE_RR(num, pgsize, _tmp1, _tmp2, vhpt) \
164 movl _tmp1=(num << 61);; \
165 mov _tmp2=((ia64_rid(IA64_REGION_ID_KERNEL, (num<<61)) << 8) | (pgsize << 2) | vhpt);; \
170 .global empty_zero_page
174 .global swapper_pg_dir
180 stringz "Halting kernel\n"
187 * Start the kernel. When the bootloader passes control to _start(), r28
188 * points to the address of the boot parameter area. Execution reaches
189 * here in physical mode.
194 .save rp, r0 // terminate unwind chain with a NULL rp
202 flushrs // must be first insn in group
207 * Save the region registers, predicate before they get clobbered
209 SAVE_REGION_REGS(r2, r8,r9,r10,r11,r12,r13,r14,r15);
213 * Initialize kernel region registers:
214 * rr[0]: VHPT enabled, page size = PAGE_SHIFT
215 * rr[1]: VHPT enabled, page size = PAGE_SHIFT
216 * rr[2]: VHPT enabled, page size = PAGE_SHIFT
217 * rr[3]: VHPT enabled, page size = PAGE_SHIFT
218 * rr[4]: VHPT enabled, page size = PAGE_SHIFT
219 * rr[5]: VHPT enabled, page size = PAGE_SHIFT
220 * rr[6]: VHPT disabled, page size = IA64_GRANULE_SHIFT
221 * rr[7]: VHPT disabled, page size = IA64_GRANULE_SHIFT
222 * We initialize all of them to prevent inadvertently assuming
223 * something about the state of address translation early in boot.
225 SET_ONE_RR(0, PAGE_SHIFT, r2, r16, 1);;
226 SET_ONE_RR(1, PAGE_SHIFT, r2, r16, 1);;
227 SET_ONE_RR(2, PAGE_SHIFT, r2, r16, 1);;
228 SET_ONE_RR(3, PAGE_SHIFT, r2, r16, 1);;
229 SET_ONE_RR(4, PAGE_SHIFT, r2, r16, 1);;
230 SET_ONE_RR(5, PAGE_SHIFT, r2, r16, 1);;
231 SET_ONE_RR(6, IA64_GRANULE_SHIFT, r2, r16, 0);;
232 SET_ONE_RR(7, IA64_GRANULE_SHIFT, r2, r16, 0);;
234 * Now pin mappings into the TLB for kernel text and data
236 mov r18=KERNEL_TR_PAGE_SHIFT<<2
237 movl r17=KERNEL_START
241 mov r16=IA64_TR_KERNEL
245 dep r2=0,r3,0,KERNEL_TR_PAGE_SHIFT
258 * Switch into virtual mode:
260 movl r16=(IA64_PSR_IT|IA64_PSR_IC|IA64_PSR_DT|IA64_PSR_RT|IA64_PSR_DFH|IA64_PSR_BN \
271 1: // now we are in virtual mode
273 SET_AREA_FOR_BOOTING_CPU(r2, r16);
275 STORE_REGION_REGS(r16, r8,r9,r10,r11,r12,r13,r14,r15);
276 SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(r16,r17,r25)
279 // set IVT entry point---can't access I/O ports without it
291 #define isAP p2 // are we an Application Processor?
292 #define isBP p3 // are we the Bootstrap Processor?
296 * Find the init_task for the currently booting CPU. At poweron, and in
297 * UP mode, task_for_booting_cpu is NULL.
299 movl r3=task_for_booting_cpu
304 cmp.eq isBP,isAP=r3,r0
309 cmp.eq isBP,isAP=r0,r0
312 tpa r3=r2 // r3 == phys addr of task struct
314 (isBP) br.cond.dpnt .load_current // BP stack is on region 5 --- no need to map it
316 // load mapping for stack (virtaddr in r2, physaddr in r3)
324 dep r2=-1,r3,61,3 // IMVA of task
327 shr.u r16=r3,IA64_GRANULE_SHIFT
334 mov r19=IA64_TR_CURRENT_STACK
343 // load the "current" pointer (r13) and ar.k6 with the current task
344 mov IA64_KR(CURRENT)=r2 // virtual address
345 mov IA64_KR(CURRENT_STACK)=r16
348 * Reserve space at the top of the stack for "struct pt_regs". Kernel
349 * threads don't store interesting values in that structure, but the space
350 * still needs to be there because time-critical stuff such as the context
351 * switching can be implemented more efficiently (for example, __switch_to()
352 * always sets the psr.dfh bit of the task it is switching to).
355 addl r12=IA64_STK_OFFSET-IA64_PT_REGS_SIZE-16,r2
356 addl r2=IA64_RBS_OFFSET,r2 // initialize the RSE
357 mov ar.rsc=0 // place RSE in enforced lazy mode
359 loadrs // clear the dirty partition
360 movl r19=__phys_per_cpu_start
361 mov r18=PERCPU_PAGE_SIZE
368 movl r20=__cpu0_per_cpu
376 (p7) br.cond.dptk.few 1b
383 .pred.rel.mutex isBP,isAP
384 (isBP) mov IA64_KR(PER_CPU_DATA)=r19 // per-CPU base for cpu0
385 (isAP) mov IA64_KR(PER_CPU_DATA)=r0 // clear physical per-CPU base
387 mov ar.bspstore=r2 // establish the new RSE stack
389 mov ar.rsc=0x3 // place RSE in eager mode
391 (isBP) dep r28=-1,r28,61,3 // make address virtual
392 (isBP) movl r2=ia64_boot_param
394 (isBP) st8 [r2]=r28 // save the address of the boot param area passed by the bootloader
397 (isAP) br.call.sptk.many rp=start_secondary
399 (isAP) br.cond.sptk self
402 // This is executed by the bootstrap processor (bsp) only:
404 #ifdef CONFIG_IA64_FW_EMU
405 // initialize PAL & SAL emulator:
406 br.call.sptk.many rp=sys_fw_init
409 br.call.sptk.many rp=start_kernel
410 .ret2: addl r3=@ltoff(halt_msg),gp
412 alloc r2=ar.pfs,8,0,2,0
415 br.call.sptk.many b0=console_print
418 br.sptk.many self // endless loop
423 GLOBAL_ENTRY(ia64_save_debug_regs)
424 alloc r16=ar.pfs,1,0,0,0
425 mov r20=ar.lc // preserve ar.lc
426 mov ar.lc=IA64_NUM_DBG_REGS-1
428 add r19=IA64_NUM_DBG_REGS*8,in0
431 #ifdef CONFIG_ITANIUM
440 br.cloop.sptk.many 1b
442 mov ar.lc=r20 // restore ar.lc
444 END(ia64_save_debug_regs)
446 GLOBAL_ENTRY(ia64_load_debug_regs)
447 alloc r16=ar.pfs,1,0,0,0
449 mov r20=ar.lc // preserve ar.lc
450 add r19=IA64_NUM_DBG_REGS*8,in0
451 mov ar.lc=IA64_NUM_DBG_REGS-1
454 1: ld8.nta r16=[in0],8
459 #ifdef CONFIG_ITANIUM
461 srlz.d // Errata 132 (NoFix status)
464 br.cloop.sptk.many 1b
466 mov ar.lc=r20 // restore ar.lc
468 END(ia64_load_debug_regs)
470 GLOBAL_ENTRY(__ia64_save_fpu)
471 alloc r2=ar.pfs,1,4,0,0
472 adds loc0=96*16-16,in0
473 adds loc1=96*16-16-128,in0
475 stf.spill.nta [loc0]=f127,-256
476 stf.spill.nta [loc1]=f119,-256
478 stf.spill.nta [loc0]=f111,-256
479 stf.spill.nta [loc1]=f103,-256
481 stf.spill.nta [loc0]=f95,-256
482 stf.spill.nta [loc1]=f87,-256
484 stf.spill.nta [loc0]=f79,-256
485 stf.spill.nta [loc1]=f71,-256
487 stf.spill.nta [loc0]=f63,-256
488 stf.spill.nta [loc1]=f55,-256
489 adds loc2=96*16-32,in0
491 stf.spill.nta [loc0]=f47,-256
492 stf.spill.nta [loc1]=f39,-256
493 adds loc3=96*16-32-128,in0
495 stf.spill.nta [loc2]=f126,-256
496 stf.spill.nta [loc3]=f118,-256
498 stf.spill.nta [loc2]=f110,-256
499 stf.spill.nta [loc3]=f102,-256
501 stf.spill.nta [loc2]=f94,-256
502 stf.spill.nta [loc3]=f86,-256
504 stf.spill.nta [loc2]=f78,-256
505 stf.spill.nta [loc3]=f70,-256
507 stf.spill.nta [loc2]=f62,-256
508 stf.spill.nta [loc3]=f54,-256
509 adds loc0=96*16-48,in0
511 stf.spill.nta [loc2]=f46,-256
512 stf.spill.nta [loc3]=f38,-256
513 adds loc1=96*16-48-128,in0
515 stf.spill.nta [loc0]=f125,-256
516 stf.spill.nta [loc1]=f117,-256
518 stf.spill.nta [loc0]=f109,-256
519 stf.spill.nta [loc1]=f101,-256
521 stf.spill.nta [loc0]=f93,-256
522 stf.spill.nta [loc1]=f85,-256
524 stf.spill.nta [loc0]=f77,-256
525 stf.spill.nta [loc1]=f69,-256
527 stf.spill.nta [loc0]=f61,-256
528 stf.spill.nta [loc1]=f53,-256
529 adds loc2=96*16-64,in0
531 stf.spill.nta [loc0]=f45,-256
532 stf.spill.nta [loc1]=f37,-256
533 adds loc3=96*16-64-128,in0
535 stf.spill.nta [loc2]=f124,-256
536 stf.spill.nta [loc3]=f116,-256
538 stf.spill.nta [loc2]=f108,-256
539 stf.spill.nta [loc3]=f100,-256
541 stf.spill.nta [loc2]=f92,-256
542 stf.spill.nta [loc3]=f84,-256
544 stf.spill.nta [loc2]=f76,-256
545 stf.spill.nta [loc3]=f68,-256
547 stf.spill.nta [loc2]=f60,-256
548 stf.spill.nta [loc3]=f52,-256
549 adds loc0=96*16-80,in0
551 stf.spill.nta [loc2]=f44,-256
552 stf.spill.nta [loc3]=f36,-256
553 adds loc1=96*16-80-128,in0
555 stf.spill.nta [loc0]=f123,-256
556 stf.spill.nta [loc1]=f115,-256
558 stf.spill.nta [loc0]=f107,-256
559 stf.spill.nta [loc1]=f99,-256
561 stf.spill.nta [loc0]=f91,-256
562 stf.spill.nta [loc1]=f83,-256
564 stf.spill.nta [loc0]=f75,-256
565 stf.spill.nta [loc1]=f67,-256
567 stf.spill.nta [loc0]=f59,-256
568 stf.spill.nta [loc1]=f51,-256
569 adds loc2=96*16-96,in0
571 stf.spill.nta [loc0]=f43,-256
572 stf.spill.nta [loc1]=f35,-256
573 adds loc3=96*16-96-128,in0
575 stf.spill.nta [loc2]=f122,-256
576 stf.spill.nta [loc3]=f114,-256
578 stf.spill.nta [loc2]=f106,-256
579 stf.spill.nta [loc3]=f98,-256
581 stf.spill.nta [loc2]=f90,-256
582 stf.spill.nta [loc3]=f82,-256
584 stf.spill.nta [loc2]=f74,-256
585 stf.spill.nta [loc3]=f66,-256
587 stf.spill.nta [loc2]=f58,-256
588 stf.spill.nta [loc3]=f50,-256
589 adds loc0=96*16-112,in0
591 stf.spill.nta [loc2]=f42,-256
592 stf.spill.nta [loc3]=f34,-256
593 adds loc1=96*16-112-128,in0
595 stf.spill.nta [loc0]=f121,-256
596 stf.spill.nta [loc1]=f113,-256
598 stf.spill.nta [loc0]=f105,-256
599 stf.spill.nta [loc1]=f97,-256
601 stf.spill.nta [loc0]=f89,-256
602 stf.spill.nta [loc1]=f81,-256
604 stf.spill.nta [loc0]=f73,-256
605 stf.spill.nta [loc1]=f65,-256
607 stf.spill.nta [loc0]=f57,-256
608 stf.spill.nta [loc1]=f49,-256
609 adds loc2=96*16-128,in0
611 stf.spill.nta [loc0]=f41,-256
612 stf.spill.nta [loc1]=f33,-256
613 adds loc3=96*16-128-128,in0
615 stf.spill.nta [loc2]=f120,-256
616 stf.spill.nta [loc3]=f112,-256
618 stf.spill.nta [loc2]=f104,-256
619 stf.spill.nta [loc3]=f96,-256
621 stf.spill.nta [loc2]=f88,-256
622 stf.spill.nta [loc3]=f80,-256
624 stf.spill.nta [loc2]=f72,-256
625 stf.spill.nta [loc3]=f64,-256
627 stf.spill.nta [loc2]=f56,-256
628 stf.spill.nta [loc3]=f48,-256
630 stf.spill.nta [loc2]=f40
631 stf.spill.nta [loc3]=f32
635 GLOBAL_ENTRY(__ia64_load_fpu)
636 alloc r2=ar.pfs,1,2,0,0
643 ldf.fill.nta f32=[in0],loc0
644 ldf.fill.nta f40=[ r3],loc0
645 ldf.fill.nta f48=[r14],loc0
646 ldf.fill.nta f56=[r15],loc0
648 ldf.fill.nta f64=[in0],loc0
649 ldf.fill.nta f72=[ r3],loc0
650 ldf.fill.nta f80=[r14],loc0
651 ldf.fill.nta f88=[r15],loc0
653 ldf.fill.nta f96=[in0],loc1
654 ldf.fill.nta f104=[ r3],loc1
655 ldf.fill.nta f112=[r14],loc1
656 ldf.fill.nta f120=[r15],loc1
658 ldf.fill.nta f33=[in0],loc0
659 ldf.fill.nta f41=[ r3],loc0
660 ldf.fill.nta f49=[r14],loc0
661 ldf.fill.nta f57=[r15],loc0
663 ldf.fill.nta f65=[in0],loc0
664 ldf.fill.nta f73=[ r3],loc0
665 ldf.fill.nta f81=[r14],loc0
666 ldf.fill.nta f89=[r15],loc0
668 ldf.fill.nta f97=[in0],loc1
669 ldf.fill.nta f105=[ r3],loc1
670 ldf.fill.nta f113=[r14],loc1
671 ldf.fill.nta f121=[r15],loc1
673 ldf.fill.nta f34=[in0],loc0
674 ldf.fill.nta f42=[ r3],loc0
675 ldf.fill.nta f50=[r14],loc0
676 ldf.fill.nta f58=[r15],loc0
678 ldf.fill.nta f66=[in0],loc0
679 ldf.fill.nta f74=[ r3],loc0
680 ldf.fill.nta f82=[r14],loc0
681 ldf.fill.nta f90=[r15],loc0
683 ldf.fill.nta f98=[in0],loc1
684 ldf.fill.nta f106=[ r3],loc1
685 ldf.fill.nta f114=[r14],loc1
686 ldf.fill.nta f122=[r15],loc1
688 ldf.fill.nta f35=[in0],loc0
689 ldf.fill.nta f43=[ r3],loc0
690 ldf.fill.nta f51=[r14],loc0
691 ldf.fill.nta f59=[r15],loc0
693 ldf.fill.nta f67=[in0],loc0
694 ldf.fill.nta f75=[ r3],loc0
695 ldf.fill.nta f83=[r14],loc0
696 ldf.fill.nta f91=[r15],loc0
698 ldf.fill.nta f99=[in0],loc1
699 ldf.fill.nta f107=[ r3],loc1
700 ldf.fill.nta f115=[r14],loc1
701 ldf.fill.nta f123=[r15],loc1
703 ldf.fill.nta f36=[in0],loc0
704 ldf.fill.nta f44=[ r3],loc0
705 ldf.fill.nta f52=[r14],loc0
706 ldf.fill.nta f60=[r15],loc0
708 ldf.fill.nta f68=[in0],loc0
709 ldf.fill.nta f76=[ r3],loc0
710 ldf.fill.nta f84=[r14],loc0
711 ldf.fill.nta f92=[r15],loc0
713 ldf.fill.nta f100=[in0],loc1
714 ldf.fill.nta f108=[ r3],loc1
715 ldf.fill.nta f116=[r14],loc1
716 ldf.fill.nta f124=[r15],loc1
718 ldf.fill.nta f37=[in0],loc0
719 ldf.fill.nta f45=[ r3],loc0
720 ldf.fill.nta f53=[r14],loc0
721 ldf.fill.nta f61=[r15],loc0
723 ldf.fill.nta f69=[in0],loc0
724 ldf.fill.nta f77=[ r3],loc0
725 ldf.fill.nta f85=[r14],loc0
726 ldf.fill.nta f93=[r15],loc0
728 ldf.fill.nta f101=[in0],loc1
729 ldf.fill.nta f109=[ r3],loc1
730 ldf.fill.nta f117=[r14],loc1
731 ldf.fill.nta f125=[r15],loc1
733 ldf.fill.nta f38 =[in0],loc0
734 ldf.fill.nta f46 =[ r3],loc0
735 ldf.fill.nta f54 =[r14],loc0
736 ldf.fill.nta f62 =[r15],loc0
738 ldf.fill.nta f70 =[in0],loc0
739 ldf.fill.nta f78 =[ r3],loc0
740 ldf.fill.nta f86 =[r14],loc0
741 ldf.fill.nta f94 =[r15],loc0
743 ldf.fill.nta f102=[in0],loc1
744 ldf.fill.nta f110=[ r3],loc1
745 ldf.fill.nta f118=[r14],loc1
746 ldf.fill.nta f126=[r15],loc1
748 ldf.fill.nta f39 =[in0],loc0
749 ldf.fill.nta f47 =[ r3],loc0
750 ldf.fill.nta f55 =[r14],loc0
751 ldf.fill.nta f63 =[r15],loc0
753 ldf.fill.nta f71 =[in0],loc0
754 ldf.fill.nta f79 =[ r3],loc0
755 ldf.fill.nta f87 =[r14],loc0
756 ldf.fill.nta f95 =[r15],loc0
758 ldf.fill.nta f103=[in0]
759 ldf.fill.nta f111=[ r3]
760 ldf.fill.nta f119=[r14]
761 ldf.fill.nta f127=[r15]
765 GLOBAL_ENTRY(__ia64_init_fpu)
766 stf.spill [sp]=f0 // M3
770 ldfps f33,f34=[sp] // M0
771 ldfps f35,f36=[sp] // M1
779 ldfps f41,f42=[sp] // M0
780 ldfps f43,f44=[sp] // M1
787 ldfps f49,f50=[sp] // M0
788 ldfps f51,f52=[sp] // M1
795 ldfps f57,f58=[sp] // M0
796 ldfps f59,f60=[sp] // M1
803 ldfps f65,f66=[sp] // M0
804 ldfps f67,f68=[sp] // M1
811 ldfps f73,f74=[sp] // M0
812 ldfps f75,f76=[sp] // M1
819 ldfps f81,f82=[sp] // M0
820 ldfps f83,f84=[sp] // M1
828 * When the instructions are cached, it would be faster to initialize
829 * the remaining registers with simply mov instructions (F-unit).
830 * This gets the time down to ~29 cycles. However, this would use up
831 * 33 bundles, whereas continuing with the above pattern yields
832 * 10 bundles and ~30 cycles.
835 ldfps f89,f90=[sp] // M0
836 ldfps f91,f92=[sp] // M1
843 ldfps f97,f98=[sp] // M0
844 ldfps f99,f100=[sp] // M1
851 ldfps f105,f106=[sp] // M0
852 ldfps f107,f108=[sp] // M1
859 ldfps f113,f114=[sp] // M0
860 ldfps f115,f116=[sp] // M1
867 ldfps f121,f122=[sp] // M0
868 ldfps f123,f124=[sp] // M1
873 br.ret.sptk.many rp // F
877 * Switch execution mode from virtual to physical
880 * r16 = new psr to establish
882 * r19 = old virtual address of ar.bsp
883 * r20 = old virtual address of sp
885 * Note: RSE must already be in enforced lazy mode
887 GLOBAL_ENTRY(ia64_switch_mode_phys)
889 rsm psr.i | psr.ic // disable interrupts and interrupt collection
894 flushrs // must be first insn in group
898 mov cr.ipsr=r16 // set new PSR
899 add r3=1f-ia64_switch_mode_phys,r15
903 mov r14=rp // get return address into a general register
906 // going to physical mode, use tpa to translate virt->phys
913 mov r18=ar.rnat // save ar.rnat
914 mov ar.bspstore=r17 // this steps on ar.rnat
918 mov ar.rnat=r18 // restore ar.rnat
919 rfi // must be last insn in group
923 END(ia64_switch_mode_phys)
926 * Switch execution mode from physical to virtual
929 * r16 = new psr to establish
930 * r19 = new bspstore to establish
931 * r20 = new sp to establish
933 * Note: RSE must already be in enforced lazy mode
935 GLOBAL_ENTRY(ia64_switch_mode_virt)
937 rsm psr.i | psr.ic // disable interrupts and interrupt collection
942 flushrs // must be first insn in group
946 mov cr.ipsr=r16 // set new PSR
947 add r3=1f-ia64_switch_mode_virt,r15
949 mov r14=rp // get return address into a general register
953 // - for code addresses, set upper bits of addr to KERNEL_START
954 // - for stack addresses, copy from input argument
955 movl r18=KERNEL_START
956 dep r3=0,r3,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
957 dep r14=0,r14,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
964 mov r18=ar.rnat // save ar.rnat
965 mov ar.bspstore=r19 // this steps on ar.rnat
969 mov ar.rnat=r18 // restore ar.rnat
970 rfi // must be last insn in group
974 END(ia64_switch_mode_virt)
976 GLOBAL_ENTRY(ia64_delay_loop)
978 { nop 0 // work around GAS unwind info generation bug...
986 // force loop to be 32-byte aligned (GAS bug means we cannot use .align
987 // inside function body without corrupting unwind info).
989 1: br.cloop.sptk.few 1b
996 * Return a CPU-local timestamp in nano-seconds. This timestamp is
997 * NOT synchronized across CPUs its return value must never be
998 * compared against the values returned on another CPU. The usage in
999 * kernel/sched/core.c ensures that.
1001 * The return-value of sched_clock() is NOT supposed to wrap-around.
1002 * If it did, it would cause some scheduling hiccups (at the worst).
1003 * Fortunately, with a 64-bit cycle-counter ticking at 100GHz, even
1004 * that would happen only once every 5+ years.
1006 * The code below basically calculates:
1008 * (ia64_get_itc() * local_cpu_data->nsec_per_cyc) >> IA64_NSEC_PER_CYC_SHIFT
1010 * except that the multiplication and the shift are done with 128-bit
1011 * intermediate precision so that we can produce a full 64-bit result.
1013 GLOBAL_ENTRY(ia64_native_sched_clock)
1014 addl r8=THIS_CPU(ia64_cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
1015 mov.m r9=ar.itc // fetch cycle-counter (35 cyc)
1019 setf.sig f9=r9 // certain to stall, so issue it _after_ ldf8...
1021 xmpy.lu f10=f9,f8 // calculate low 64 bits of 128-bit product (4 cyc)
1022 xmpy.hu f11=f9,f8 // calculate high 64 bits of 128-bit product
1024 getf.sig r8=f10 // (5 cyc)
1027 shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
1029 END(ia64_native_sched_clock)
1031 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1032 GLOBAL_ENTRY(cycle_to_cputime)
1033 alloc r16=ar.pfs,1,0,0,0
1034 addl r8=THIS_CPU(ia64_cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
1040 xmpy.lu f10=f9,f8 // calculate low 64 bits of 128-bit product (4 cyc)
1041 xmpy.hu f11=f9,f8 // calculate high 64 bits of 128-bit product
1043 getf.sig r8=f10 // (5 cyc)
1046 shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
1048 END(cycle_to_cputime)
1049 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
1051 #ifdef CONFIG_IA64_BRL_EMU
1054 * Assembly routines used by brl_emu.c to set preserved register state.
1057 #define SET_REG(reg) \
1058 GLOBAL_ENTRY(ia64_set_##reg); \
1059 alloc r16=ar.pfs,1,0,0,0; \
1062 br.ret.sptk.many rp; \
1071 #endif /* CONFIG_IA64_BRL_EMU */
1075 #ifdef CONFIG_HOTPLUG_CPU
1076 GLOBAL_ENTRY(ia64_jump_to_sal)
1077 alloc r16=ar.pfs,1,0,0,0;;
1084 movl r18=tlb_purge_done;;
1085 DATA_VA_TO_PA(r18);;
1086 mov b1=r18 // Return location
1087 movl r18=ia64_do_tlb_purge;;
1088 DATA_VA_TO_PA(r18);;
1089 mov b2=r18 // doing tlb_flush work
1090 mov ar.rsc=0 // Put RSE in enforced lazy, LE mode
1092 DATA_VA_TO_PA(r17);;
1094 movl r16=SAL_PSR_BITS_TO_SET;;
1097 rfi;; // note: this unmask MCA/INIT (psr.mc)
1100 * Invalidate all TLB data/inst
1102 br.sptk.many b2;; // jump to tlb purge code
1105 RESTORE_REGION_REGS(r25, r17,r18,r19);;
1106 RESTORE_REG(b0, r25, r17);;
1107 RESTORE_REG(b1, r25, r17);;
1108 RESTORE_REG(b2, r25, r17);;
1109 RESTORE_REG(b3, r25, r17);;
1110 RESTORE_REG(b4, r25, r17);;
1111 RESTORE_REG(b5, r25, r17);;
1113 ld8 r12=[r25],0x08;;
1114 ld8 r13=[r25],0x08;;
1115 RESTORE_REG(ar.fpsr, r25, r17);;
1116 RESTORE_REG(ar.pfs, r25, r17);;
1117 RESTORE_REG(ar.rnat, r25, r17);;
1118 RESTORE_REG(ar.unat, r25, r17);;
1119 RESTORE_REG(ar.bspstore, r25, r17);;
1120 RESTORE_REG(cr.dcr, r25, r17);;
1121 RESTORE_REG(cr.iva, r25, r17);;
1122 RESTORE_REG(cr.pta, r25, r17);;
1123 srlz.d;; // required not to violate RAW dependency
1124 RESTORE_REG(cr.itv, r25, r17);;
1125 RESTORE_REG(cr.pmv, r25, r17);;
1126 RESTORE_REG(cr.cmcv, r25, r17);;
1127 RESTORE_REG(cr.lrr0, r25, r17);;
1128 RESTORE_REG(cr.lrr1, r25, r17);;
1133 ld8 r17=[r25],0x08;;
1135 RESTORE_REG(ar.lc, r25, r17);;
1137 * Now Restore floating point regs
1139 ldf.fill.nta f2=[r25],16;;
1140 ldf.fill.nta f3=[r25],16;;
1141 ldf.fill.nta f4=[r25],16;;
1142 ldf.fill.nta f5=[r25],16;;
1143 ldf.fill.nta f16=[r25],16;;
1144 ldf.fill.nta f17=[r25],16;;
1145 ldf.fill.nta f18=[r25],16;;
1146 ldf.fill.nta f19=[r25],16;;
1147 ldf.fill.nta f20=[r25],16;;
1148 ldf.fill.nta f21=[r25],16;;
1149 ldf.fill.nta f22=[r25],16;;
1150 ldf.fill.nta f23=[r25],16;;
1151 ldf.fill.nta f24=[r25],16;;
1152 ldf.fill.nta f25=[r25],16;;
1153 ldf.fill.nta f26=[r25],16;;
1154 ldf.fill.nta f27=[r25],16;;
1155 ldf.fill.nta f28=[r25],16;;
1156 ldf.fill.nta f29=[r25],16;;
1157 ldf.fill.nta f30=[r25],16;;
1158 ldf.fill.nta f31=[r25],16;;
1161 * Now that we have done all the register restores
1162 * we are now ready for the big DIVE to SAL Land
1166 br.ret.sptk.many b0;;
1167 END(ia64_jump_to_sal)
1168 #endif /* CONFIG_HOTPLUG_CPU */
1170 #endif /* CONFIG_SMP */