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
22 #include <linux/config.h>
24 #include <asm/asmmacro.h>
26 #include <asm/kregs.h>
27 #include <asm/mmu_context.h>
28 #include <asm/asm-offsets.h>
30 #include <asm/pgtable.h>
31 #include <asm/processor.h>
32 #include <asm/ptrace.h>
33 #include <asm/system.h>
34 #include <asm/mca_asm.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);; \
168 .section __special_page_section,"ax"
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 * Save the region registers, predicate before they get clobbered
204 SAVE_REGION_REGS(r2, r8,r9,r10,r11,r12,r13,r14,r15);
208 * Initialize kernel region registers:
209 * rr[0]: VHPT enabled, page size = PAGE_SHIFT
210 * rr[1]: VHPT enabled, page size = PAGE_SHIFT
211 * rr[2]: VHPT enabled, page size = PAGE_SHIFT
212 * rr[3]: VHPT enabled, page size = PAGE_SHIFT
213 * rr[4]: VHPT enabled, page size = PAGE_SHIFT
214 * rr[5]: VHPT enabled, page size = PAGE_SHIFT
215 * rr[6]: VHPT disabled, page size = IA64_GRANULE_SHIFT
216 * rr[7]: VHPT disabled, page size = IA64_GRANULE_SHIFT
217 * We initialize all of them to prevent inadvertently assuming
218 * something about the state of address translation early in boot.
220 SET_ONE_RR(0, PAGE_SHIFT, r2, r16, 1);;
221 SET_ONE_RR(1, PAGE_SHIFT, r2, r16, 1);;
222 SET_ONE_RR(2, PAGE_SHIFT, r2, r16, 1);;
223 SET_ONE_RR(3, PAGE_SHIFT, r2, r16, 1);;
224 SET_ONE_RR(4, PAGE_SHIFT, r2, r16, 1);;
225 SET_ONE_RR(5, PAGE_SHIFT, r2, r16, 1);;
226 SET_ONE_RR(6, IA64_GRANULE_SHIFT, r2, r16, 0);;
227 SET_ONE_RR(7, IA64_GRANULE_SHIFT, r2, r16, 0);;
229 * Now pin mappings into the TLB for kernel text and data
231 mov r18=KERNEL_TR_PAGE_SHIFT<<2
232 movl r17=KERNEL_START
236 mov r16=IA64_TR_KERNEL
240 dep r2=0,r3,0,KERNEL_TR_PAGE_SHIFT
253 * Switch into virtual mode:
255 movl r16=(IA64_PSR_IT|IA64_PSR_IC|IA64_PSR_DT|IA64_PSR_RT|IA64_PSR_DFH|IA64_PSR_BN \
266 1: // now we are in virtual mode
268 SET_AREA_FOR_BOOTING_CPU(r2, r16);
270 STORE_REGION_REGS(r16, r8,r9,r10,r11,r12,r13,r14,r15);
271 SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(r16,r17,r25)
274 // set IVT entry point---can't access I/O ports without it
286 #define isAP p2 // are we an Application Processor?
287 #define isBP p3 // are we the Bootstrap Processor?
291 * Find the init_task for the currently booting CPU. At poweron, and in
292 * UP mode, task_for_booting_cpu is NULL.
294 movl r3=task_for_booting_cpu
299 cmp.eq isBP,isAP=r3,r0
304 cmp.eq isBP,isAP=r0,r0
307 tpa r3=r2 // r3 == phys addr of task struct
309 (isBP) br.cond.dpnt .load_current // BP stack is on region 5 --- no need to map it
311 // load mapping for stack (virtaddr in r2, physaddr in r3)
319 dep r2=-1,r3,61,3 // IMVA of task
322 shr.u r16=r3,IA64_GRANULE_SHIFT
329 mov r19=IA64_TR_CURRENT_STACK
338 // load the "current" pointer (r13) and ar.k6 with the current task
339 mov IA64_KR(CURRENT)=r2 // virtual address
340 mov IA64_KR(CURRENT_STACK)=r16
343 * Reserve space at the top of the stack for "struct pt_regs". Kernel
344 * threads don't store interesting values in that structure, but the space
345 * still needs to be there because time-critical stuff such as the context
346 * switching can be implemented more efficiently (for example, __switch_to()
347 * always sets the psr.dfh bit of the task it is switching to).
350 addl r12=IA64_STK_OFFSET-IA64_PT_REGS_SIZE-16,r2
351 addl r2=IA64_RBS_OFFSET,r2 // initialize the RSE
352 mov ar.rsc=0 // place RSE in enforced lazy mode
354 loadrs // clear the dirty partition
356 mov ar.bspstore=r2 // establish the new RSE stack
358 mov ar.rsc=0x3 // place RSE in eager mode
360 (isBP) dep r28=-1,r28,61,3 // make address virtual
361 (isBP) movl r2=ia64_boot_param
363 (isBP) st8 [r2]=r28 // save the address of the boot param area passed by the bootloader
366 (isAP) br.call.sptk.many rp=start_secondary
368 (isAP) br.cond.sptk self
371 // This is executed by the bootstrap processor (bsp) only:
373 #ifdef CONFIG_IA64_FW_EMU
374 // initialize PAL & SAL emulator:
375 br.call.sptk.many rp=sys_fw_init
378 br.call.sptk.many rp=start_kernel
379 .ret2: addl r3=@ltoff(halt_msg),gp
381 alloc r2=ar.pfs,8,0,2,0
384 br.call.sptk.many b0=console_print
387 br.sptk.many self // endless loop
390 GLOBAL_ENTRY(ia64_save_debug_regs)
391 alloc r16=ar.pfs,1,0,0,0
392 mov r20=ar.lc // preserve ar.lc
393 mov ar.lc=IA64_NUM_DBG_REGS-1
395 add r19=IA64_NUM_DBG_REGS*8,in0
398 #ifdef CONFIG_ITANIUM
407 br.cloop.sptk.many 1b
409 mov ar.lc=r20 // restore ar.lc
411 END(ia64_save_debug_regs)
413 GLOBAL_ENTRY(ia64_load_debug_regs)
414 alloc r16=ar.pfs,1,0,0,0
416 mov r20=ar.lc // preserve ar.lc
417 add r19=IA64_NUM_DBG_REGS*8,in0
418 mov ar.lc=IA64_NUM_DBG_REGS-1
421 1: ld8.nta r16=[in0],8
426 #ifdef CONFIG_ITANIUM
428 srlz.d // Errata 132 (NoFix status)
431 br.cloop.sptk.many 1b
433 mov ar.lc=r20 // restore ar.lc
435 END(ia64_load_debug_regs)
437 GLOBAL_ENTRY(__ia64_save_fpu)
438 alloc r2=ar.pfs,1,4,0,0
439 adds loc0=96*16-16,in0
440 adds loc1=96*16-16-128,in0
442 stf.spill.nta [loc0]=f127,-256
443 stf.spill.nta [loc1]=f119,-256
445 stf.spill.nta [loc0]=f111,-256
446 stf.spill.nta [loc1]=f103,-256
448 stf.spill.nta [loc0]=f95,-256
449 stf.spill.nta [loc1]=f87,-256
451 stf.spill.nta [loc0]=f79,-256
452 stf.spill.nta [loc1]=f71,-256
454 stf.spill.nta [loc0]=f63,-256
455 stf.spill.nta [loc1]=f55,-256
456 adds loc2=96*16-32,in0
458 stf.spill.nta [loc0]=f47,-256
459 stf.spill.nta [loc1]=f39,-256
460 adds loc3=96*16-32-128,in0
462 stf.spill.nta [loc2]=f126,-256
463 stf.spill.nta [loc3]=f118,-256
465 stf.spill.nta [loc2]=f110,-256
466 stf.spill.nta [loc3]=f102,-256
468 stf.spill.nta [loc2]=f94,-256
469 stf.spill.nta [loc3]=f86,-256
471 stf.spill.nta [loc2]=f78,-256
472 stf.spill.nta [loc3]=f70,-256
474 stf.spill.nta [loc2]=f62,-256
475 stf.spill.nta [loc3]=f54,-256
476 adds loc0=96*16-48,in0
478 stf.spill.nta [loc2]=f46,-256
479 stf.spill.nta [loc3]=f38,-256
480 adds loc1=96*16-48-128,in0
482 stf.spill.nta [loc0]=f125,-256
483 stf.spill.nta [loc1]=f117,-256
485 stf.spill.nta [loc0]=f109,-256
486 stf.spill.nta [loc1]=f101,-256
488 stf.spill.nta [loc0]=f93,-256
489 stf.spill.nta [loc1]=f85,-256
491 stf.spill.nta [loc0]=f77,-256
492 stf.spill.nta [loc1]=f69,-256
494 stf.spill.nta [loc0]=f61,-256
495 stf.spill.nta [loc1]=f53,-256
496 adds loc2=96*16-64,in0
498 stf.spill.nta [loc0]=f45,-256
499 stf.spill.nta [loc1]=f37,-256
500 adds loc3=96*16-64-128,in0
502 stf.spill.nta [loc2]=f124,-256
503 stf.spill.nta [loc3]=f116,-256
505 stf.spill.nta [loc2]=f108,-256
506 stf.spill.nta [loc3]=f100,-256
508 stf.spill.nta [loc2]=f92,-256
509 stf.spill.nta [loc3]=f84,-256
511 stf.spill.nta [loc2]=f76,-256
512 stf.spill.nta [loc3]=f68,-256
514 stf.spill.nta [loc2]=f60,-256
515 stf.spill.nta [loc3]=f52,-256
516 adds loc0=96*16-80,in0
518 stf.spill.nta [loc2]=f44,-256
519 stf.spill.nta [loc3]=f36,-256
520 adds loc1=96*16-80-128,in0
522 stf.spill.nta [loc0]=f123,-256
523 stf.spill.nta [loc1]=f115,-256
525 stf.spill.nta [loc0]=f107,-256
526 stf.spill.nta [loc1]=f99,-256
528 stf.spill.nta [loc0]=f91,-256
529 stf.spill.nta [loc1]=f83,-256
531 stf.spill.nta [loc0]=f75,-256
532 stf.spill.nta [loc1]=f67,-256
534 stf.spill.nta [loc0]=f59,-256
535 stf.spill.nta [loc1]=f51,-256
536 adds loc2=96*16-96,in0
538 stf.spill.nta [loc0]=f43,-256
539 stf.spill.nta [loc1]=f35,-256
540 adds loc3=96*16-96-128,in0
542 stf.spill.nta [loc2]=f122,-256
543 stf.spill.nta [loc3]=f114,-256
545 stf.spill.nta [loc2]=f106,-256
546 stf.spill.nta [loc3]=f98,-256
548 stf.spill.nta [loc2]=f90,-256
549 stf.spill.nta [loc3]=f82,-256
551 stf.spill.nta [loc2]=f74,-256
552 stf.spill.nta [loc3]=f66,-256
554 stf.spill.nta [loc2]=f58,-256
555 stf.spill.nta [loc3]=f50,-256
556 adds loc0=96*16-112,in0
558 stf.spill.nta [loc2]=f42,-256
559 stf.spill.nta [loc3]=f34,-256
560 adds loc1=96*16-112-128,in0
562 stf.spill.nta [loc0]=f121,-256
563 stf.spill.nta [loc1]=f113,-256
565 stf.spill.nta [loc0]=f105,-256
566 stf.spill.nta [loc1]=f97,-256
568 stf.spill.nta [loc0]=f89,-256
569 stf.spill.nta [loc1]=f81,-256
571 stf.spill.nta [loc0]=f73,-256
572 stf.spill.nta [loc1]=f65,-256
574 stf.spill.nta [loc0]=f57,-256
575 stf.spill.nta [loc1]=f49,-256
576 adds loc2=96*16-128,in0
578 stf.spill.nta [loc0]=f41,-256
579 stf.spill.nta [loc1]=f33,-256
580 adds loc3=96*16-128-128,in0
582 stf.spill.nta [loc2]=f120,-256
583 stf.spill.nta [loc3]=f112,-256
585 stf.spill.nta [loc2]=f104,-256
586 stf.spill.nta [loc3]=f96,-256
588 stf.spill.nta [loc2]=f88,-256
589 stf.spill.nta [loc3]=f80,-256
591 stf.spill.nta [loc2]=f72,-256
592 stf.spill.nta [loc3]=f64,-256
594 stf.spill.nta [loc2]=f56,-256
595 stf.spill.nta [loc3]=f48,-256
597 stf.spill.nta [loc2]=f40
598 stf.spill.nta [loc3]=f32
602 GLOBAL_ENTRY(__ia64_load_fpu)
603 alloc r2=ar.pfs,1,2,0,0
610 ldf.fill.nta f32=[in0],loc0
611 ldf.fill.nta f40=[ r3],loc0
612 ldf.fill.nta f48=[r14],loc0
613 ldf.fill.nta f56=[r15],loc0
615 ldf.fill.nta f64=[in0],loc0
616 ldf.fill.nta f72=[ r3],loc0
617 ldf.fill.nta f80=[r14],loc0
618 ldf.fill.nta f88=[r15],loc0
620 ldf.fill.nta f96=[in0],loc1
621 ldf.fill.nta f104=[ r3],loc1
622 ldf.fill.nta f112=[r14],loc1
623 ldf.fill.nta f120=[r15],loc1
625 ldf.fill.nta f33=[in0],loc0
626 ldf.fill.nta f41=[ r3],loc0
627 ldf.fill.nta f49=[r14],loc0
628 ldf.fill.nta f57=[r15],loc0
630 ldf.fill.nta f65=[in0],loc0
631 ldf.fill.nta f73=[ r3],loc0
632 ldf.fill.nta f81=[r14],loc0
633 ldf.fill.nta f89=[r15],loc0
635 ldf.fill.nta f97=[in0],loc1
636 ldf.fill.nta f105=[ r3],loc1
637 ldf.fill.nta f113=[r14],loc1
638 ldf.fill.nta f121=[r15],loc1
640 ldf.fill.nta f34=[in0],loc0
641 ldf.fill.nta f42=[ r3],loc0
642 ldf.fill.nta f50=[r14],loc0
643 ldf.fill.nta f58=[r15],loc0
645 ldf.fill.nta f66=[in0],loc0
646 ldf.fill.nta f74=[ r3],loc0
647 ldf.fill.nta f82=[r14],loc0
648 ldf.fill.nta f90=[r15],loc0
650 ldf.fill.nta f98=[in0],loc1
651 ldf.fill.nta f106=[ r3],loc1
652 ldf.fill.nta f114=[r14],loc1
653 ldf.fill.nta f122=[r15],loc1
655 ldf.fill.nta f35=[in0],loc0
656 ldf.fill.nta f43=[ r3],loc0
657 ldf.fill.nta f51=[r14],loc0
658 ldf.fill.nta f59=[r15],loc0
660 ldf.fill.nta f67=[in0],loc0
661 ldf.fill.nta f75=[ r3],loc0
662 ldf.fill.nta f83=[r14],loc0
663 ldf.fill.nta f91=[r15],loc0
665 ldf.fill.nta f99=[in0],loc1
666 ldf.fill.nta f107=[ r3],loc1
667 ldf.fill.nta f115=[r14],loc1
668 ldf.fill.nta f123=[r15],loc1
670 ldf.fill.nta f36=[in0],loc0
671 ldf.fill.nta f44=[ r3],loc0
672 ldf.fill.nta f52=[r14],loc0
673 ldf.fill.nta f60=[r15],loc0
675 ldf.fill.nta f68=[in0],loc0
676 ldf.fill.nta f76=[ r3],loc0
677 ldf.fill.nta f84=[r14],loc0
678 ldf.fill.nta f92=[r15],loc0
680 ldf.fill.nta f100=[in0],loc1
681 ldf.fill.nta f108=[ r3],loc1
682 ldf.fill.nta f116=[r14],loc1
683 ldf.fill.nta f124=[r15],loc1
685 ldf.fill.nta f37=[in0],loc0
686 ldf.fill.nta f45=[ r3],loc0
687 ldf.fill.nta f53=[r14],loc0
688 ldf.fill.nta f61=[r15],loc0
690 ldf.fill.nta f69=[in0],loc0
691 ldf.fill.nta f77=[ r3],loc0
692 ldf.fill.nta f85=[r14],loc0
693 ldf.fill.nta f93=[r15],loc0
695 ldf.fill.nta f101=[in0],loc1
696 ldf.fill.nta f109=[ r3],loc1
697 ldf.fill.nta f117=[r14],loc1
698 ldf.fill.nta f125=[r15],loc1
700 ldf.fill.nta f38 =[in0],loc0
701 ldf.fill.nta f46 =[ r3],loc0
702 ldf.fill.nta f54 =[r14],loc0
703 ldf.fill.nta f62 =[r15],loc0
705 ldf.fill.nta f70 =[in0],loc0
706 ldf.fill.nta f78 =[ r3],loc0
707 ldf.fill.nta f86 =[r14],loc0
708 ldf.fill.nta f94 =[r15],loc0
710 ldf.fill.nta f102=[in0],loc1
711 ldf.fill.nta f110=[ r3],loc1
712 ldf.fill.nta f118=[r14],loc1
713 ldf.fill.nta f126=[r15],loc1
715 ldf.fill.nta f39 =[in0],loc0
716 ldf.fill.nta f47 =[ r3],loc0
717 ldf.fill.nta f55 =[r14],loc0
718 ldf.fill.nta f63 =[r15],loc0
720 ldf.fill.nta f71 =[in0],loc0
721 ldf.fill.nta f79 =[ r3],loc0
722 ldf.fill.nta f87 =[r14],loc0
723 ldf.fill.nta f95 =[r15],loc0
725 ldf.fill.nta f103=[in0]
726 ldf.fill.nta f111=[ r3]
727 ldf.fill.nta f119=[r14]
728 ldf.fill.nta f127=[r15]
732 GLOBAL_ENTRY(__ia64_init_fpu)
733 stf.spill [sp]=f0 // M3
737 ldfps f33,f34=[sp] // M0
738 ldfps f35,f36=[sp] // M1
746 ldfps f41,f42=[sp] // M0
747 ldfps f43,f44=[sp] // M1
754 ldfps f49,f50=[sp] // M0
755 ldfps f51,f52=[sp] // M1
762 ldfps f57,f58=[sp] // M0
763 ldfps f59,f60=[sp] // M1
770 ldfps f65,f66=[sp] // M0
771 ldfps f67,f68=[sp] // M1
778 ldfps f73,f74=[sp] // M0
779 ldfps f75,f76=[sp] // M1
786 ldfps f81,f82=[sp] // M0
787 ldfps f83,f84=[sp] // M1
795 * When the instructions are cached, it would be faster to initialize
796 * the remaining registers with simply mov instructions (F-unit).
797 * This gets the time down to ~29 cycles. However, this would use up
798 * 33 bundles, whereas continuing with the above pattern yields
799 * 10 bundles and ~30 cycles.
802 ldfps f89,f90=[sp] // M0
803 ldfps f91,f92=[sp] // M1
810 ldfps f97,f98=[sp] // M0
811 ldfps f99,f100=[sp] // M1
818 ldfps f105,f106=[sp] // M0
819 ldfps f107,f108=[sp] // M1
826 ldfps f113,f114=[sp] // M0
827 ldfps f115,f116=[sp] // M1
834 ldfps f121,f122=[sp] // M0
835 ldfps f123,f124=[sp] // M1
840 br.ret.sptk.many rp // F
844 * Switch execution mode from virtual to physical
847 * r16 = new psr to establish
849 * r19 = old virtual address of ar.bsp
850 * r20 = old virtual address of sp
852 * Note: RSE must already be in enforced lazy mode
854 GLOBAL_ENTRY(ia64_switch_mode_phys)
856 alloc r2=ar.pfs,0,0,0,0
857 rsm psr.i | psr.ic // disable interrupts and interrupt collection
862 flushrs // must be first insn in group
866 mov cr.ipsr=r16 // set new PSR
867 add r3=1f-ia64_switch_mode_phys,r15
871 mov r14=rp // get return address into a general register
874 // going to physical mode, use tpa to translate virt->phys
881 mov r18=ar.rnat // save ar.rnat
882 mov ar.bspstore=r17 // this steps on ar.rnat
886 mov ar.rnat=r18 // restore ar.rnat
887 rfi // must be last insn in group
891 END(ia64_switch_mode_phys)
894 * Switch execution mode from physical to virtual
897 * r16 = new psr to establish
898 * r19 = new bspstore to establish
899 * r20 = new sp to establish
901 * Note: RSE must already be in enforced lazy mode
903 GLOBAL_ENTRY(ia64_switch_mode_virt)
905 alloc r2=ar.pfs,0,0,0,0
906 rsm psr.i | psr.ic // disable interrupts and interrupt collection
911 flushrs // must be first insn in group
915 mov cr.ipsr=r16 // set new PSR
916 add r3=1f-ia64_switch_mode_virt,r15
918 mov r14=rp // get return address into a general register
922 // - for code addresses, set upper bits of addr to KERNEL_START
923 // - for stack addresses, copy from input argument
924 movl r18=KERNEL_START
925 dep r3=0,r3,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
926 dep r14=0,r14,KERNEL_TR_PAGE_SHIFT,64-KERNEL_TR_PAGE_SHIFT
933 mov r18=ar.rnat // save ar.rnat
934 mov ar.bspstore=r19 // this steps on ar.rnat
938 mov ar.rnat=r18 // restore ar.rnat
939 rfi // must be last insn in group
943 END(ia64_switch_mode_virt)
945 GLOBAL_ENTRY(ia64_delay_loop)
947 { nop 0 // work around GAS unwind info generation bug...
955 // force loop to be 32-byte aligned (GAS bug means we cannot use .align
956 // inside function body without corrupting unwind info).
958 1: br.cloop.sptk.few 1b
965 * Return a CPU-local timestamp in nano-seconds. This timestamp is
966 * NOT synchronized across CPUs its return value must never be
967 * compared against the values returned on another CPU. The usage in
968 * kernel/sched.c ensures that.
970 * The return-value of sched_clock() is NOT supposed to wrap-around.
971 * If it did, it would cause some scheduling hiccups (at the worst).
972 * Fortunately, with a 64-bit cycle-counter ticking at 100GHz, even
973 * that would happen only once every 5+ years.
975 * The code below basically calculates:
977 * (ia64_get_itc() * local_cpu_data->nsec_per_cyc) >> IA64_NSEC_PER_CYC_SHIFT
979 * except that the multiplication and the shift are done with 128-bit
980 * intermediate precision so that we can produce a full 64-bit result.
982 GLOBAL_ENTRY(sched_clock)
983 addl r8=THIS_CPU(cpu_info) + IA64_CPUINFO_NSEC_PER_CYC_OFFSET,r0
984 mov.m r9=ar.itc // fetch cycle-counter (35 cyc)
988 setf.sig f9=r9 // certain to stall, so issue it _after_ ldf8...
990 xmpy.lu f10=f9,f8 // calculate low 64 bits of 128-bit product (4 cyc)
991 xmpy.hu f11=f9,f8 // calculate high 64 bits of 128-bit product
993 getf.sig r8=f10 // (5 cyc)
996 shrp r8=r9,r8,IA64_NSEC_PER_CYC_SHIFT
1000 GLOBAL_ENTRY(start_kernel_thread)
1002 .save rp, r0 // this is the end of the call-chain
1004 alloc r2 = ar.pfs, 0, 0, 2, 0
1007 br.call.sptk.many rp = kernel_thread_helper;;
1009 br.call.sptk.many rp = sys_exit;;
1010 1: br.sptk.few 1b // not reached
1011 END(start_kernel_thread)
1013 #ifdef CONFIG_IA64_BRL_EMU
1016 * Assembly routines used by brl_emu.c to set preserved register state.
1019 #define SET_REG(reg) \
1020 GLOBAL_ENTRY(ia64_set_##reg); \
1021 alloc r16=ar.pfs,1,0,0,0; \
1024 br.ret.sptk.many rp; \
1033 #endif /* CONFIG_IA64_BRL_EMU */
1037 * This routine handles spinlock contention. It uses a non-standard calling
1038 * convention to avoid converting leaf routines into interior routines. Because
1039 * of this special convention, there are several restrictions:
1041 * - do not use gp relative variables, this code is called from the kernel
1042 * and from modules, r1 is undefined.
1043 * - do not use stacked registers, the caller owns them.
1044 * - do not use the scratch stack space, the caller owns it.
1045 * - do not use any registers other than the ones listed below
1048 * ar.pfs - saved CFM of caller
1049 * ar.ccv - 0 (and available for use)
1050 * r27 - flags from spin_lock_irqsave or 0. Must be preserved.
1051 * r28 - available for use.
1052 * r29 - available for use.
1053 * r30 - available for use.
1054 * r31 - address of lock, available for use.
1055 * b6 - return address
1056 * p14 - available for use.
1057 * p15 - used to track flag status.
1059 * If you patch this code to use more registers, do not forget to update
1060 * the clobber lists for spin_lock() in include/asm-ia64/spinlock.h.
1063 #if __GNUC__ < 3 || (__GNUC__ == 3 && __GNUC_MINOR__ < 3)
1065 GLOBAL_ENTRY(ia64_spinlock_contention_pre3_4)
1067 .save ar.pfs, r0 // this code effectively has a zero frame size
1071 tbit.nz p15,p0=r27,IA64_PSR_I_BIT
1072 .restore sp // pop existing prologue after next insn
1079 (p15) ssm psr.i // reenable interrupts if they were on
1080 // DavidM says that srlz.d is slow and is not required in this case
1082 // exponential backoff, kdb, lockmeter etc. go in here
1084 ld4 r30=[r31] // don't use ld4.bias; if it's contended, we won't write the word
1087 cmp4.ne p14,p0=r30,r0
1088 (p14) br.cond.sptk.few .wait
1089 (p15) rsm psr.i // disable interrupts if we reenabled them
1090 br.cond.sptk.few b6 // lock is now free, try to acquire
1091 .global ia64_spinlock_contention_pre3_4_end // for kernprof
1092 ia64_spinlock_contention_pre3_4_end:
1093 END(ia64_spinlock_contention_pre3_4)
1097 GLOBAL_ENTRY(ia64_spinlock_contention)
1101 tbit.nz p15,p0=r27,IA64_PSR_I_BIT
1104 (p15) ssm psr.i // reenable interrupts if they were on
1105 // DavidM says that srlz.d is slow and is not required in this case
1107 // exponential backoff, kdb, lockmeter etc. go in here
1109 ld4 r30=[r31] // don't use ld4.bias; if it's contended, we won't write the word
1111 cmp4.ne p14,p0=r30,r0
1113 (p14) br.cond.sptk.few .wait2
1114 (p15) rsm psr.i // disable interrupts if we reenabled them
1116 cmpxchg4.acq r30=[r31], r30, ar.ccv
1118 cmp4.ne p14,p0=r0,r30
1119 (p14) br.cond.sptk.few .wait
1121 br.ret.sptk.many b6 // lock is now taken
1122 END(ia64_spinlock_contention)
1126 #ifdef CONFIG_HOTPLUG_CPU
1127 GLOBAL_ENTRY(ia64_jump_to_sal)
1128 alloc r16=ar.pfs,1,0,0,0;;
1135 movl r18=tlb_purge_done;;
1136 DATA_VA_TO_PA(r18);;
1137 mov b1=r18 // Return location
1138 movl r18=ia64_do_tlb_purge;;
1139 DATA_VA_TO_PA(r18);;
1140 mov b2=r18 // doing tlb_flush work
1141 mov ar.rsc=0 // Put RSE in enforced lazy, LE mode
1143 DATA_VA_TO_PA(r17);;
1145 movl r16=SAL_PSR_BITS_TO_SET;;
1151 * Invalidate all TLB data/inst
1153 br.sptk.many b2;; // jump to tlb purge code
1156 RESTORE_REGION_REGS(r25, r17,r18,r19);;
1157 RESTORE_REG(b0, r25, r17);;
1158 RESTORE_REG(b1, r25, r17);;
1159 RESTORE_REG(b2, r25, r17);;
1160 RESTORE_REG(b3, r25, r17);;
1161 RESTORE_REG(b4, r25, r17);;
1162 RESTORE_REG(b5, r25, r17);;
1164 ld8 r12=[r25],0x08;;
1165 ld8 r13=[r25],0x08;;
1166 RESTORE_REG(ar.fpsr, r25, r17);;
1167 RESTORE_REG(ar.pfs, r25, r17);;
1168 RESTORE_REG(ar.rnat, r25, r17);;
1169 RESTORE_REG(ar.unat, r25, r17);;
1170 RESTORE_REG(ar.bspstore, r25, r17);;
1171 RESTORE_REG(cr.dcr, r25, r17);;
1172 RESTORE_REG(cr.iva, r25, r17);;
1173 RESTORE_REG(cr.pta, r25, r17);;
1174 RESTORE_REG(cr.itv, r25, r17);;
1175 RESTORE_REG(cr.pmv, r25, r17);;
1176 RESTORE_REG(cr.cmcv, r25, r17);;
1177 RESTORE_REG(cr.lrr0, r25, r17);;
1178 RESTORE_REG(cr.lrr1, r25, r17);;
1183 ld8 r17=[r25],0x08;;
1185 RESTORE_REG(ar.lc, r25, r17);;
1187 * Now Restore floating point regs
1189 ldf.fill.nta f2=[r25],16;;
1190 ldf.fill.nta f3=[r25],16;;
1191 ldf.fill.nta f4=[r25],16;;
1192 ldf.fill.nta f5=[r25],16;;
1193 ldf.fill.nta f16=[r25],16;;
1194 ldf.fill.nta f17=[r25],16;;
1195 ldf.fill.nta f18=[r25],16;;
1196 ldf.fill.nta f19=[r25],16;;
1197 ldf.fill.nta f20=[r25],16;;
1198 ldf.fill.nta f21=[r25],16;;
1199 ldf.fill.nta f22=[r25],16;;
1200 ldf.fill.nta f23=[r25],16;;
1201 ldf.fill.nta f24=[r25],16;;
1202 ldf.fill.nta f25=[r25],16;;
1203 ldf.fill.nta f26=[r25],16;;
1204 ldf.fill.nta f27=[r25],16;;
1205 ldf.fill.nta f28=[r25],16;;
1206 ldf.fill.nta f29=[r25],16;;
1207 ldf.fill.nta f30=[r25],16;;
1208 ldf.fill.nta f31=[r25],16;;
1211 * Now that we have done all the register restores
1212 * we are now ready for the big DIVE to SAL Land
1216 br.ret.sptk.many b0;;
1217 END(ia64_jump_to_sal)
1218 #endif /* CONFIG_HOTPLUG_CPU */
1220 #endif /* CONFIG_SMP */