xref: /freebsd/sys/powerpc/aim/aim_machdep.c (revision 2f513db72b034fd5ef7f080b11be5c711c15186a)
1 /*-
2  * Copyright (C) 1995, 1996 Wolfgang Solfrank.
3  * Copyright (C) 1995, 1996 TooLs GmbH.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *      This product includes software developed by TooLs GmbH.
17  * 4. The name of TooLs GmbH may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
29  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 /*-
32  * Copyright (C) 2001 Benno Rice
33  * All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  *
44  * THIS SOFTWARE IS PROVIDED BY Benno Rice ``AS IS'' AND ANY EXPRESS OR
45  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
46  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
47  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
48  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
49  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
50  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
51  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
52  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
53  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
54  *	$NetBSD: machdep.c,v 1.74.2.1 2000/11/01 16:13:48 tv Exp $
55  */
56 
57 #include <sys/cdefs.h>
58 __FBSDID("$FreeBSD$");
59 
60 #include "opt_ddb.h"
61 #include "opt_kstack_pages.h"
62 #include "opt_platform.h"
63 
64 #include <sys/param.h>
65 #include <sys/proc.h>
66 #include <sys/systm.h>
67 #include <sys/bio.h>
68 #include <sys/buf.h>
69 #include <sys/bus.h>
70 #include <sys/cons.h>
71 #include <sys/cpu.h>
72 #include <sys/eventhandler.h>
73 #include <sys/exec.h>
74 #include <sys/imgact.h>
75 #include <sys/kdb.h>
76 #include <sys/kernel.h>
77 #include <sys/ktr.h>
78 #include <sys/linker.h>
79 #include <sys/lock.h>
80 #include <sys/malloc.h>
81 #include <sys/mbuf.h>
82 #include <sys/msgbuf.h>
83 #include <sys/mutex.h>
84 #include <sys/ptrace.h>
85 #include <sys/reboot.h>
86 #include <sys/rwlock.h>
87 #include <sys/signalvar.h>
88 #include <sys/syscallsubr.h>
89 #include <sys/sysctl.h>
90 #include <sys/sysent.h>
91 #include <sys/sysproto.h>
92 #include <sys/ucontext.h>
93 #include <sys/uio.h>
94 #include <sys/vmmeter.h>
95 #include <sys/vnode.h>
96 
97 #include <net/netisr.h>
98 
99 #include <vm/vm.h>
100 #include <vm/vm_extern.h>
101 #include <vm/vm_kern.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_map.h>
104 #include <vm/vm_object.h>
105 #include <vm/vm_pager.h>
106 
107 #include <machine/altivec.h>
108 #ifndef __powerpc64__
109 #include <machine/bat.h>
110 #endif
111 #include <machine/cpu.h>
112 #include <machine/elf.h>
113 #include <machine/fpu.h>
114 #include <machine/hid.h>
115 #include <machine/kdb.h>
116 #include <machine/md_var.h>
117 #include <machine/metadata.h>
118 #include <machine/mmuvar.h>
119 #include <machine/pcb.h>
120 #include <machine/reg.h>
121 #include <machine/sigframe.h>
122 #include <machine/spr.h>
123 #include <machine/trap.h>
124 #include <machine/vmparam.h>
125 #include <machine/ofw_machdep.h>
126 
127 #include <ddb/ddb.h>
128 
129 #include <dev/ofw/openfirm.h>
130 
131 #ifdef __powerpc64__
132 #include "mmu_oea64.h"
133 #endif
134 
135 #ifndef __powerpc64__
136 struct bat	battable[16];
137 #endif
138 
139 #ifndef __powerpc64__
140 /* Bits for running on 64-bit systems in 32-bit mode. */
141 extern void	*testppc64, *testppc64size;
142 extern void	*restorebridge, *restorebridgesize;
143 extern void	*rfid_patch, *rfi_patch1, *rfi_patch2;
144 extern void	*trapcode64;
145 
146 extern Elf_Addr	_GLOBAL_OFFSET_TABLE_[];
147 #endif
148 
149 extern void	*rstcode, *rstcodeend;
150 extern void	*trapcode, *trapcodeend;
151 extern void	*hypertrapcode, *hypertrapcodeend;
152 extern void	*generictrap, *generictrap64;
153 extern void	*alitrap, *aliend;
154 extern void	*dsitrap, *dsiend;
155 extern void	*decrint, *decrsize;
156 extern void     *extint, *extsize;
157 extern void	*dblow, *dbend;
158 extern void	*imisstrap, *imisssize;
159 extern void	*dlmisstrap, *dlmisssize;
160 extern void	*dsmisstrap, *dsmisssize;
161 
162 extern void *ap_pcpu;
163 extern void __restartkernel(vm_offset_t, vm_offset_t, vm_offset_t, void *, uint32_t, register_t offset, register_t msr);
164 
165 void aim_early_init(vm_offset_t fdt, vm_offset_t toc, vm_offset_t ofentry,
166     void *mdp, uint32_t mdp_cookie);
167 void aim_cpu_init(vm_offset_t toc);
168 
169 void
170 aim_early_init(vm_offset_t fdt, vm_offset_t toc, vm_offset_t ofentry, void *mdp,
171     uint32_t mdp_cookie)
172 {
173 	register_t	scratch;
174 
175 	/*
176 	 * If running from an FDT, make sure we are in real mode to avoid
177 	 * tromping on firmware page tables. Everything in the kernel assumes
178 	 * 1:1 mappings out of firmware, so this won't break anything not
179 	 * already broken. This doesn't work if there is live OF, since OF
180 	 * may internally use non-1:1 mappings.
181 	 */
182 	if (ofentry == 0)
183 		mtmsr(mfmsr() & ~(PSL_IR | PSL_DR));
184 
185 #ifdef __powerpc64__
186 	/*
187 	 * If in real mode, relocate to high memory so that the kernel
188 	 * can execute from the direct map.
189 	 */
190 	if (!(mfmsr() & PSL_DR) &&
191 	    (vm_offset_t)&aim_early_init < DMAP_BASE_ADDRESS)
192 		__restartkernel(fdt, 0, ofentry, mdp, mdp_cookie,
193 		    DMAP_BASE_ADDRESS, mfmsr());
194 #endif
195 
196 	/* Various very early CPU fix ups */
197 	switch (mfpvr() >> 16) {
198 		/*
199 		 * PowerPC 970 CPUs have a misfeature requested by Apple that
200 		 * makes them pretend they have a 32-byte cacheline. Turn this
201 		 * off before we measure the cacheline size.
202 		 */
203 		case IBM970:
204 		case IBM970FX:
205 		case IBM970MP:
206 		case IBM970GX:
207 			scratch = mfspr(SPR_HID5);
208 			scratch &= ~HID5_970_DCBZ_SIZE_HI;
209 			mtspr(SPR_HID5, scratch);
210 			break;
211 	#ifdef __powerpc64__
212 		case IBMPOWER7:
213 		case IBMPOWER7PLUS:
214 		case IBMPOWER8:
215 		case IBMPOWER8E:
216 		case IBMPOWER8NVL:
217 		case IBMPOWER9:
218 			/* XXX: get from ibm,slb-size in device tree */
219 			n_slbs = 32;
220 			break;
221 	#endif
222 	}
223 }
224 
225 void
226 aim_cpu_init(vm_offset_t toc)
227 {
228 	size_t		trap_offset, trapsize;
229 	vm_offset_t	trap;
230 	register_t	msr;
231 	uint8_t		*cache_check;
232 	int		cacheline_warn;
233 #ifndef __powerpc64__
234 	register_t	scratch;
235 	int		ppc64;
236 #endif
237 
238 	trap_offset = 0;
239 	cacheline_warn = 0;
240 
241 	/* General setup for AIM CPUs */
242 	psl_kernset = PSL_EE | PSL_ME | PSL_IR | PSL_DR | PSL_RI;
243 
244 #ifdef __powerpc64__
245 	psl_kernset |= PSL_SF;
246 	if (mfmsr() & PSL_HV)
247 		psl_kernset |= PSL_HV;
248 #endif
249 	psl_userset = psl_kernset | PSL_PR;
250 #ifdef __powerpc64__
251 	psl_userset32 = psl_userset & ~PSL_SF;
252 #endif
253 
254 	/*
255 	 * Zeroed bits in this variable signify that the value of the bit
256 	 * in its position is allowed to vary between userspace contexts.
257 	 *
258 	 * All other bits are required to be identical for every userspace
259 	 * context. The actual *value* of the bit is determined by
260 	 * psl_userset and/or psl_userset32, and is not allowed to change.
261 	 *
262 	 * Remember to update this set when implementing support for
263 	 * *conditionally* enabling a processor facility. Failing to do
264 	 * this will cause swapcontext() in userspace to break when a
265 	 * process uses a conditionally-enabled facility.
266 	 *
267 	 * When *unconditionally* implementing support for a processor
268 	 * facility, update psl_userset / psl_userset32 instead.
269 	 *
270 	 * See the access control check in set_mcontext().
271 	 */
272 	psl_userstatic = ~(PSL_VSX | PSL_VEC | PSL_FP | PSL_FE0 | PSL_FE1);
273 	/*
274 	 * Mask bits from the SRR1 that aren't really the MSR:
275 	 * Bits 1-4, 10-15 (ppc32), 33-36, 42-47 (ppc64)
276 	 */
277 	psl_userstatic &= ~0x783f0000UL;
278 
279 	/*
280 	 * Initialize the interrupt tables and figure out our cache line
281 	 * size and whether or not we need the 64-bit bridge code.
282 	 */
283 
284 	/*
285 	 * Disable translation in case the vector area hasn't been
286 	 * mapped (G5). Note that no OFW calls can be made until
287 	 * translation is re-enabled.
288 	 */
289 
290 	msr = mfmsr();
291 	mtmsr((msr & ~(PSL_IR | PSL_DR)) | PSL_RI);
292 
293 	/*
294 	 * Measure the cacheline size using dcbz
295 	 *
296 	 * Use EXC_PGM as a playground. We are about to overwrite it
297 	 * anyway, we know it exists, and we know it is cache-aligned.
298 	 */
299 
300 	cache_check = (void *)EXC_PGM;
301 
302 	for (cacheline_size = 0; cacheline_size < 0x100; cacheline_size++)
303 		cache_check[cacheline_size] = 0xff;
304 
305 	__asm __volatile("dcbz 0,%0":: "r" (cache_check) : "memory");
306 
307 	/* Find the first byte dcbz did not zero to get the cache line size */
308 	for (cacheline_size = 0; cacheline_size < 0x100 &&
309 	    cache_check[cacheline_size] == 0; cacheline_size++);
310 
311 	/* Work around psim bug */
312 	if (cacheline_size == 0) {
313 		cacheline_warn = 1;
314 		cacheline_size = 32;
315 	}
316 
317 	#ifndef __powerpc64__
318 	/*
319 	 * Figure out whether we need to use the 64 bit PMAP. This works by
320 	 * executing an instruction that is only legal on 64-bit PPC (mtmsrd),
321 	 * and setting ppc64 = 0 if that causes a trap.
322 	 */
323 
324 	ppc64 = 1;
325 
326 	bcopy(&testppc64, (void *)EXC_PGM,  (size_t)&testppc64size);
327 	__syncicache((void *)EXC_PGM, (size_t)&testppc64size);
328 
329 	__asm __volatile("\
330 		mfmsr %0;	\
331 		mtsprg2 %1;	\
332 				\
333 		mtmsrd %0;	\
334 		mfsprg2 %1;"
335 	    : "=r"(scratch), "=r"(ppc64));
336 
337 	if (ppc64)
338 		cpu_features |= PPC_FEATURE_64;
339 
340 	/*
341 	 * Now copy restorebridge into all the handlers, if necessary,
342 	 * and set up the trap tables.
343 	 */
344 
345 	if (cpu_features & PPC_FEATURE_64) {
346 		/* Patch the two instances of rfi -> rfid */
347 		bcopy(&rfid_patch,&rfi_patch1,4);
348 	#ifdef KDB
349 		/* rfi_patch2 is at the end of dbleave */
350 		bcopy(&rfid_patch,&rfi_patch2,4);
351 	#endif
352 	}
353 	#else /* powerpc64 */
354 	cpu_features |= PPC_FEATURE_64;
355 	#endif
356 
357 	trapsize = (size_t)&trapcodeend - (size_t)&trapcode;
358 
359 	/*
360 	 * Copy generic handler into every possible trap. Special cases will get
361 	 * different ones in a minute.
362 	 */
363 	for (trap = EXC_RST; trap < EXC_LAST; trap += 0x20)
364 		bcopy(&trapcode, (void *)trap, trapsize);
365 
366 	#ifndef __powerpc64__
367 	if (cpu_features & PPC_FEATURE_64) {
368 		/*
369 		 * Copy a code snippet to restore 32-bit bridge mode
370 		 * to the top of every non-generic trap handler
371 		 */
372 
373 		trap_offset += (size_t)&restorebridgesize;
374 		bcopy(&restorebridge, (void *)EXC_RST, trap_offset);
375 		bcopy(&restorebridge, (void *)EXC_DSI, trap_offset);
376 		bcopy(&restorebridge, (void *)EXC_ALI, trap_offset);
377 		bcopy(&restorebridge, (void *)EXC_PGM, trap_offset);
378 		bcopy(&restorebridge, (void *)EXC_MCHK, trap_offset);
379 		bcopy(&restorebridge, (void *)EXC_TRC, trap_offset);
380 		bcopy(&restorebridge, (void *)EXC_BPT, trap_offset);
381 	}
382 	#else
383 	trapsize = (size_t)&hypertrapcodeend - (size_t)&hypertrapcode;
384 	bcopy(&hypertrapcode, (void *)(EXC_HEA + trap_offset), trapsize);
385 	bcopy(&hypertrapcode, (void *)(EXC_HMI + trap_offset), trapsize);
386 	bcopy(&hypertrapcode, (void *)(EXC_HVI + trap_offset), trapsize);
387 	bcopy(&hypertrapcode, (void *)(EXC_SOFT_PATCH + trap_offset), trapsize);
388 	#endif
389 
390 	bcopy(&rstcode, (void *)(EXC_RST + trap_offset), (size_t)&rstcodeend -
391 	    (size_t)&rstcode);
392 
393 #ifdef KDB
394 	bcopy(&dblow, (void *)(EXC_MCHK + trap_offset), (size_t)&dbend -
395 	    (size_t)&dblow);
396 	bcopy(&dblow, (void *)(EXC_PGM + trap_offset), (size_t)&dbend -
397 	    (size_t)&dblow);
398 	bcopy(&dblow, (void *)(EXC_TRC + trap_offset), (size_t)&dbend -
399 	    (size_t)&dblow);
400 	bcopy(&dblow, (void *)(EXC_BPT + trap_offset), (size_t)&dbend -
401 	    (size_t)&dblow);
402 #endif
403 	bcopy(&alitrap,  (void *)(EXC_ALI + trap_offset),  (size_t)&aliend -
404 	    (size_t)&alitrap);
405 	bcopy(&dsitrap,  (void *)(EXC_DSI + trap_offset),  (size_t)&dsiend -
406 	    (size_t)&dsitrap);
407 
408 	/* Set address of generictrap for self-reloc calculations */
409 	*((void **)TRAP_GENTRAP) = &generictrap;
410 	#ifdef __powerpc64__
411 	/* Set TOC base so that the interrupt code can get at it */
412 	*((void **)TRAP_ENTRY) = &generictrap;
413 	*((register_t *)TRAP_TOCBASE) = toc;
414 	#else
415 	/* Set branch address for trap code */
416 	if (cpu_features & PPC_FEATURE_64)
417 		*((void **)TRAP_ENTRY) = &generictrap64;
418 	else
419 		*((void **)TRAP_ENTRY) = &generictrap;
420 	*((void **)TRAP_TOCBASE) = _GLOBAL_OFFSET_TABLE_;
421 
422 	/* G2-specific TLB miss helper handlers */
423 	bcopy(&imisstrap, (void *)EXC_IMISS,  (size_t)&imisssize);
424 	bcopy(&dlmisstrap, (void *)EXC_DLMISS,  (size_t)&dlmisssize);
425 	bcopy(&dsmisstrap, (void *)EXC_DSMISS,  (size_t)&dsmisssize);
426 	#endif
427 	__syncicache(EXC_RSVD, EXC_LAST - EXC_RSVD);
428 
429 	/*
430 	 * Restore MSR
431 	 */
432 	mtmsr(msr);
433 
434 	/* Warn if cachline size was not determined */
435 	if (cacheline_warn == 1) {
436 		printf("WARNING: cacheline size undetermined, setting to 32\n");
437 	}
438 
439 	/*
440 	 * Initialise virtual memory. Use BUS_PROBE_GENERIC priority
441 	 * in case the platform module had a better idea of what we
442 	 * should do.
443 	 */
444 	if (cpu_features & PPC_FEATURE_64)
445 		pmap_mmu_install(MMU_TYPE_G5, BUS_PROBE_GENERIC);
446 	else
447 		pmap_mmu_install(MMU_TYPE_OEA, BUS_PROBE_GENERIC);
448 }
449 
450 /*
451  * Shutdown the CPU as much as possible.
452  */
453 void
454 cpu_halt(void)
455 {
456 
457 	OF_exit();
458 }
459 
460 int
461 ptrace_single_step(struct thread *td)
462 {
463 	struct trapframe *tf;
464 
465 	tf = td->td_frame;
466 	tf->srr1 |= PSL_SE;
467 
468 	return (0);
469 }
470 
471 int
472 ptrace_clear_single_step(struct thread *td)
473 {
474 	struct trapframe *tf;
475 
476 	tf = td->td_frame;
477 	tf->srr1 &= ~PSL_SE;
478 
479 	return (0);
480 }
481 
482 void
483 kdb_cpu_clear_singlestep(void)
484 {
485 
486 	kdb_frame->srr1 &= ~PSL_SE;
487 }
488 
489 void
490 kdb_cpu_set_singlestep(void)
491 {
492 
493 	kdb_frame->srr1 |= PSL_SE;
494 }
495 
496 /*
497  * Initialise a struct pcpu.
498  */
499 void
500 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t sz)
501 {
502 #ifdef __powerpc64__
503 /* Copy the SLB contents from the current CPU */
504 memcpy(pcpu->pc_aim.slb, PCPU_GET(aim.slb), sizeof(pcpu->pc_aim.slb));
505 #endif
506 }
507 
508 #ifndef __powerpc64__
509 uint64_t
510 va_to_vsid(pmap_t pm, vm_offset_t va)
511 {
512 	return ((pm->pm_sr[(uintptr_t)va >> ADDR_SR_SHFT]) & SR_VSID_MASK);
513 }
514 
515 #endif
516 
517 /*
518  * These functions need to provide addresses that both (a) work in real mode
519  * (or whatever mode/circumstances the kernel is in in early boot (now)) and
520  * (b) can still, in principle, work once the kernel is going. Because these
521  * rely on existing mappings/real mode, unmap is a no-op.
522  */
523 vm_offset_t
524 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
525 {
526 	KASSERT(!pmap_bootstrapped, ("Not available after PMAP started!"));
527 
528 	/*
529 	 * If we have the MMU up in early boot, assume it is 1:1. Otherwise,
530 	 * try to get the address in a memory region compatible with the
531 	 * direct map for efficiency later.
532 	 */
533 	if (mfmsr() & PSL_DR)
534 		return (pa);
535 	else
536 		return (DMAP_BASE_ADDRESS + pa);
537 }
538 
539 void
540 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
541 {
542 
543 	KASSERT(!pmap_bootstrapped, ("Not available after PMAP started!"));
544 }
545 
546 /* From p3-53 of the MPC7450 RISC Microprocessor Family Reference Manual */
547 void
548 flush_disable_caches(void)
549 {
550 	register_t msr;
551 	register_t msscr0;
552 	register_t cache_reg;
553 	volatile uint32_t *memp;
554 	uint32_t temp;
555 	int i;
556 	int x;
557 
558 	msr = mfmsr();
559 	powerpc_sync();
560 	mtmsr(msr & ~(PSL_EE | PSL_DR));
561 	msscr0 = mfspr(SPR_MSSCR0);
562 	msscr0 &= ~MSSCR0_L2PFE;
563 	mtspr(SPR_MSSCR0, msscr0);
564 	powerpc_sync();
565 	isync();
566 	__asm__ __volatile__("dssall; sync");
567 	powerpc_sync();
568 	isync();
569 	__asm__ __volatile__("dcbf 0,%0" :: "r"(0));
570 	__asm__ __volatile__("dcbf 0,%0" :: "r"(0));
571 	__asm__ __volatile__("dcbf 0,%0" :: "r"(0));
572 
573 	/* Lock the L1 Data cache. */
574 	mtspr(SPR_LDSTCR, mfspr(SPR_LDSTCR) | 0xFF);
575 	powerpc_sync();
576 	isync();
577 
578 	mtspr(SPR_LDSTCR, 0);
579 
580 	/*
581 	 * Perform this in two stages: Flush the cache starting in RAM, then do it
582 	 * from ROM.
583 	 */
584 	memp = (volatile uint32_t *)0x00000000;
585 	for (i = 0; i < 128 * 1024; i++) {
586 		temp = *memp;
587 		__asm__ __volatile__("dcbf 0,%0" :: "r"(memp));
588 		memp += 32/sizeof(*memp);
589 	}
590 
591 	memp = (volatile uint32_t *)0xfff00000;
592 	x = 0xfe;
593 
594 	for (; x != 0xff;) {
595 		mtspr(SPR_LDSTCR, x);
596 		for (i = 0; i < 128; i++) {
597 			temp = *memp;
598 			__asm__ __volatile__("dcbf 0,%0" :: "r"(memp));
599 			memp += 32/sizeof(*memp);
600 		}
601 		x = ((x << 1) | 1) & 0xff;
602 	}
603 	mtspr(SPR_LDSTCR, 0);
604 
605 	cache_reg = mfspr(SPR_L2CR);
606 	if (cache_reg & L2CR_L2E) {
607 		cache_reg &= ~(L2CR_L2IO_7450 | L2CR_L2DO_7450);
608 		mtspr(SPR_L2CR, cache_reg);
609 		powerpc_sync();
610 		mtspr(SPR_L2CR, cache_reg | L2CR_L2HWF);
611 		while (mfspr(SPR_L2CR) & L2CR_L2HWF)
612 			; /* Busy wait for cache to flush */
613 		powerpc_sync();
614 		cache_reg &= ~L2CR_L2E;
615 		mtspr(SPR_L2CR, cache_reg);
616 		powerpc_sync();
617 		mtspr(SPR_L2CR, cache_reg | L2CR_L2I);
618 		powerpc_sync();
619 		while (mfspr(SPR_L2CR) & L2CR_L2I)
620 			; /* Busy wait for L2 cache invalidate */
621 		powerpc_sync();
622 	}
623 
624 	cache_reg = mfspr(SPR_L3CR);
625 	if (cache_reg & L3CR_L3E) {
626 		cache_reg &= ~(L3CR_L3IO | L3CR_L3DO);
627 		mtspr(SPR_L3CR, cache_reg);
628 		powerpc_sync();
629 		mtspr(SPR_L3CR, cache_reg | L3CR_L3HWF);
630 		while (mfspr(SPR_L3CR) & L3CR_L3HWF)
631 			; /* Busy wait for cache to flush */
632 		powerpc_sync();
633 		cache_reg &= ~L3CR_L3E;
634 		mtspr(SPR_L3CR, cache_reg);
635 		powerpc_sync();
636 		mtspr(SPR_L3CR, cache_reg | L3CR_L3I);
637 		powerpc_sync();
638 		while (mfspr(SPR_L3CR) & L3CR_L3I)
639 			; /* Busy wait for L3 cache invalidate */
640 		powerpc_sync();
641 	}
642 
643 	mtspr(SPR_HID0, mfspr(SPR_HID0) & ~HID0_DCE);
644 	powerpc_sync();
645 	isync();
646 
647 	mtmsr(msr);
648 }
649 
650 void
651 cpu_sleep()
652 {
653 	static u_quad_t timebase = 0;
654 	static register_t sprgs[4];
655 	static register_t srrs[2];
656 
657 	jmp_buf resetjb;
658 	struct thread *fputd;
659 	struct thread *vectd;
660 	register_t hid0;
661 	register_t msr;
662 	register_t saved_msr;
663 
664 	ap_pcpu = pcpup;
665 
666 	PCPU_SET(restore, &resetjb);
667 
668 	saved_msr = mfmsr();
669 	fputd = PCPU_GET(fputhread);
670 	vectd = PCPU_GET(vecthread);
671 	if (fputd != NULL)
672 		save_fpu(fputd);
673 	if (vectd != NULL)
674 		save_vec(vectd);
675 	if (setjmp(resetjb) == 0) {
676 		sprgs[0] = mfspr(SPR_SPRG0);
677 		sprgs[1] = mfspr(SPR_SPRG1);
678 		sprgs[2] = mfspr(SPR_SPRG2);
679 		sprgs[3] = mfspr(SPR_SPRG3);
680 		srrs[0] = mfspr(SPR_SRR0);
681 		srrs[1] = mfspr(SPR_SRR1);
682 		timebase = mftb();
683 		powerpc_sync();
684 		flush_disable_caches();
685 		hid0 = mfspr(SPR_HID0);
686 		hid0 = (hid0 & ~(HID0_DOZE | HID0_NAP)) | HID0_SLEEP;
687 		powerpc_sync();
688 		isync();
689 		msr = mfmsr() | PSL_POW;
690 		mtspr(SPR_HID0, hid0);
691 		powerpc_sync();
692 
693 		while (1)
694 			mtmsr(msr);
695 	}
696 	platform_smp_timebase_sync(timebase, 0);
697 	PCPU_SET(curthread, curthread);
698 	PCPU_SET(curpcb, curthread->td_pcb);
699 	pmap_activate(curthread);
700 	powerpc_sync();
701 	mtspr(SPR_SPRG0, sprgs[0]);
702 	mtspr(SPR_SPRG1, sprgs[1]);
703 	mtspr(SPR_SPRG2, sprgs[2]);
704 	mtspr(SPR_SPRG3, sprgs[3]);
705 	mtspr(SPR_SRR0, srrs[0]);
706 	mtspr(SPR_SRR1, srrs[1]);
707 	mtmsr(saved_msr);
708 	if (fputd == curthread)
709 		enable_fpu(curthread);
710 	if (vectd == curthread)
711 		enable_vec(curthread);
712 	powerpc_sync();
713 }
714 
715