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: stable/12/sys/powerpc/aim/aim_machdep.c 372438 2022-08-21 11:31:16Z dim $");
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
aim_early_init(vm_offset_t fdt,vm_offset_t toc,vm_offset_t ofentry,void * mdp,uint32_t mdp_cookie)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
aim_cpu_init(vm_offset_t toc)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 	/* Bits that users aren't allowed to change */
255 	psl_userstatic = ~(PSL_VEC | PSL_FP | PSL_FE0 | PSL_FE1);
256 	/*
257 	 * Mask bits from the SRR1 that aren't really the MSR:
258 	 * Bits 1-4, 10-15 (ppc32), 33-36, 42-47 (ppc64)
259 	 */
260 	psl_userstatic &= ~0x783f0000UL;
261 
262 	/*
263 	 * Initialize the interrupt tables and figure out our cache line
264 	 * size and whether or not we need the 64-bit bridge code.
265 	 */
266 
267 	/*
268 	 * Disable translation in case the vector area hasn't been
269 	 * mapped (G5). Note that no OFW calls can be made until
270 	 * translation is re-enabled.
271 	 */
272 
273 	msr = mfmsr();
274 	mtmsr((msr & ~(PSL_IR | PSL_DR)) | PSL_RI);
275 
276 	/*
277 	 * Measure the cacheline size using dcbz
278 	 *
279 	 * Use EXC_PGM as a playground. We are about to overwrite it
280 	 * anyway, we know it exists, and we know it is cache-aligned.
281 	 */
282 
283 	cache_check = (void *)EXC_PGM;
284 
285 	for (cacheline_size = 0; cacheline_size < 0x100; cacheline_size++)
286 		cache_check[cacheline_size] = 0xff;
287 
288 	__asm __volatile("dcbz 0,%0":: "r" (cache_check) : "memory");
289 
290 	/* Find the first byte dcbz did not zero to get the cache line size */
291 	for (cacheline_size = 0; cacheline_size < 0x100 &&
292 	    cache_check[cacheline_size] == 0; cacheline_size++);
293 
294 	/* Work around psim bug */
295 	if (cacheline_size == 0) {
296 		cacheline_warn = 1;
297 		cacheline_size = 32;
298 	}
299 
300 	#ifndef __powerpc64__
301 	/*
302 	 * Figure out whether we need to use the 64 bit PMAP. This works by
303 	 * executing an instruction that is only legal on 64-bit PPC (mtmsrd),
304 	 * and setting ppc64 = 0 if that causes a trap.
305 	 */
306 
307 	ppc64 = 1;
308 
309 	bcopy(&testppc64, (void *)EXC_PGM,  (size_t)&testppc64size);
310 	__syncicache((void *)EXC_PGM, (size_t)&testppc64size);
311 
312 	__asm __volatile("\
313 		mfmsr %0;	\
314 		mtsprg2 %1;	\
315 				\
316 		mtmsrd %0;	\
317 		mfsprg2 %1;"
318 	    : "=r"(scratch), "=r"(ppc64));
319 
320 	if (ppc64)
321 		cpu_features |= PPC_FEATURE_64;
322 
323 	/*
324 	 * Now copy restorebridge into all the handlers, if necessary,
325 	 * and set up the trap tables.
326 	 */
327 
328 	if (cpu_features & PPC_FEATURE_64) {
329 		/* Patch the two instances of rfi -> rfid */
330 		bcopy(&rfid_patch,&rfi_patch1,4);
331 	#ifdef KDB
332 		/* rfi_patch2 is at the end of dbleave */
333 		bcopy(&rfid_patch,&rfi_patch2,4);
334 	#endif
335 	}
336 	#else /* powerpc64 */
337 	cpu_features |= PPC_FEATURE_64;
338 	#endif
339 
340 	trapsize = (size_t)&trapcodeend - (size_t)&trapcode;
341 
342 	/*
343 	 * Copy generic handler into every possible trap. Special cases will get
344 	 * different ones in a minute.
345 	 */
346 	for (trap = EXC_RST; trap < EXC_LAST; trap += 0x20)
347 		bcopy(&trapcode, (void *)trap, trapsize);
348 
349 	#ifndef __powerpc64__
350 	if (cpu_features & PPC_FEATURE_64) {
351 		/*
352 		 * Copy a code snippet to restore 32-bit bridge mode
353 		 * to the top of every non-generic trap handler
354 		 */
355 
356 		trap_offset += (size_t)&restorebridgesize;
357 		bcopy(&restorebridge, (void *)EXC_RST, trap_offset);
358 		bcopy(&restorebridge, (void *)EXC_DSI, trap_offset);
359 		bcopy(&restorebridge, (void *)EXC_ALI, trap_offset);
360 		bcopy(&restorebridge, (void *)EXC_PGM, trap_offset);
361 		bcopy(&restorebridge, (void *)EXC_MCHK, trap_offset);
362 		bcopy(&restorebridge, (void *)EXC_TRC, trap_offset);
363 		bcopy(&restorebridge, (void *)EXC_BPT, trap_offset);
364 	} else {
365 
366 		/*
367 		 * Use an IBAT and a DBAT to map the bottom 256M segment.
368 		 *
369 		 * It is very important to do it *now* to avoid taking a
370 		 * fault in .text / .data before the MMU is bootstrapped,
371 		 * because until then, the translation data has not been
372 		 * copied over from OpenFirmware, so our DSI/ISI will fail
373 		 * to find a match.
374 		 */
375 
376 		battable[0x0].batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
377 		battable[0x0].batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
378 
379 		__asm (".balign 32; \n"
380 		    "mtibatu 0,%0; mtibatl 0,%1; isync; \n"
381 		    "mtdbatu 0,%0; mtdbatl 0,%1; isync"
382 		    :: "r"(battable[0].batu), "r"(battable[0].batl));
383 	}
384 	#else
385 	trapsize = (size_t)&hypertrapcodeend - (size_t)&hypertrapcode;
386 	bcopy(&hypertrapcode, (void *)(EXC_HEA + trap_offset), trapsize);
387 	bcopy(&hypertrapcode, (void *)(EXC_HMI + trap_offset), trapsize);
388 	bcopy(&hypertrapcode, (void *)(EXC_HVI + trap_offset), trapsize);
389 	#endif
390 
391 	bcopy(&rstcode, (void *)(EXC_RST + trap_offset), (size_t)&rstcodeend -
392 	    (size_t)&rstcode);
393 
394 #ifdef KDB
395 	bcopy(&dblow, (void *)(EXC_MCHK + trap_offset), (size_t)&dbend -
396 	    (size_t)&dblow);
397 	bcopy(&dblow, (void *)(EXC_PGM + trap_offset), (size_t)&dbend -
398 	    (size_t)&dblow);
399 	bcopy(&dblow, (void *)(EXC_TRC + trap_offset), (size_t)&dbend -
400 	    (size_t)&dblow);
401 	bcopy(&dblow, (void *)(EXC_BPT + trap_offset), (size_t)&dbend -
402 	    (size_t)&dblow);
403 #endif
404 	bcopy(&alitrap,  (void *)(EXC_ALI + trap_offset),  (size_t)&aliend -
405 	    (size_t)&alitrap);
406 	bcopy(&dsitrap,  (void *)(EXC_DSI + trap_offset),  (size_t)&dsiend -
407 	    (size_t)&dsitrap);
408 
409 	#ifdef __powerpc64__
410 	/* Set TOC base so that the interrupt code can get at it */
411 	*((void **)TRAP_GENTRAP) = &generictrap;
412 	*((register_t *)TRAP_TOCBASE) = toc;
413 	#else
414 	/* Set branch address for trap code */
415 	if (cpu_features & PPC_FEATURE_64)
416 		*((void **)TRAP_GENTRAP) = &generictrap64;
417 	else
418 		*((void **)TRAP_GENTRAP) = &generictrap;
419 	*((void **)TRAP_TOCBASE) = _GLOBAL_OFFSET_TABLE_;
420 
421 	/* G2-specific TLB miss helper handlers */
422 	bcopy(&imisstrap, (void *)EXC_IMISS,  (size_t)&imisssize);
423 	bcopy(&dlmisstrap, (void *)EXC_DLMISS,  (size_t)&dlmisssize);
424 	bcopy(&dsmisstrap, (void *)EXC_DSMISS,  (size_t)&dsmisssize);
425 	#endif
426 	__syncicache(EXC_RSVD, EXC_LAST - EXC_RSVD);
427 
428 	/*
429 	 * Restore MSR
430 	 */
431 	mtmsr(msr);
432 
433 	/* Warn if cachline size was not determined */
434 	if (cacheline_warn == 1) {
435 		printf("WARNING: cacheline size undetermined, setting to 32\n");
436 	}
437 
438 	/*
439 	 * Initialise virtual memory. Use BUS_PROBE_GENERIC priority
440 	 * in case the platform module had a better idea of what we
441 	 * should do.
442 	 */
443 	if (cpu_features2 & PPC_FEATURE2_ARCH_3_00)
444 		pmap_mmu_install(MMU_TYPE_P9H, BUS_PROBE_GENERIC);
445 	else if (cpu_features & PPC_FEATURE_64)
446 		pmap_mmu_install(MMU_TYPE_G5, BUS_PROBE_GENERIC);
447 	else
448 		pmap_mmu_install(MMU_TYPE_OEA, BUS_PROBE_GENERIC);
449 }
450 
451 /*
452  * Shutdown the CPU as much as possible.
453  */
454 void
cpu_halt(void)455 cpu_halt(void)
456 {
457 
458 	OF_exit();
459 }
460 
461 int
ptrace_single_step(struct thread * td)462 ptrace_single_step(struct thread *td)
463 {
464 	struct trapframe *tf;
465 
466 	tf = td->td_frame;
467 	tf->srr1 |= PSL_SE;
468 
469 	return (0);
470 }
471 
472 int
ptrace_clear_single_step(struct thread * td)473 ptrace_clear_single_step(struct thread *td)
474 {
475 	struct trapframe *tf;
476 
477 	tf = td->td_frame;
478 	tf->srr1 &= ~PSL_SE;
479 
480 	return (0);
481 }
482 
483 void
kdb_cpu_clear_singlestep(void)484 kdb_cpu_clear_singlestep(void)
485 {
486 
487 	kdb_frame->srr1 &= ~PSL_SE;
488 }
489 
490 void
kdb_cpu_set_singlestep(void)491 kdb_cpu_set_singlestep(void)
492 {
493 
494 	kdb_frame->srr1 |= PSL_SE;
495 }
496 
497 /*
498  * Initialise a struct pcpu.
499  */
500 void
cpu_pcpu_init(struct pcpu * pcpu,int cpuid,size_t sz)501 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t sz)
502 {
503 #ifdef __powerpc64__
504 /* Copy the SLB contents from the current CPU */
505 memcpy(pcpu->pc_aim.slb, PCPU_GET(aim.slb), sizeof(pcpu->pc_aim.slb));
506 #endif
507 }
508 
509 #ifndef __powerpc64__
510 uint64_t
va_to_vsid(pmap_t pm,vm_offset_t va)511 va_to_vsid(pmap_t pm, vm_offset_t va)
512 {
513 	return ((pm->pm_sr[(uintptr_t)va >> ADDR_SR_SHFT]) & SR_VSID_MASK);
514 }
515 
516 #endif
517 
518 /*
519  * These functions need to provide addresses that both (a) work in real mode
520  * (or whatever mode/circumstances the kernel is in in early boot (now)) and
521  * (b) can still, in principle, work once the kernel is going. Because these
522  * rely on existing mappings/real mode, unmap is a no-op.
523  */
524 vm_offset_t
pmap_early_io_map(vm_paddr_t pa,vm_size_t size)525 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
526 {
527 	KASSERT(!pmap_bootstrapped, ("Not available after PMAP started!"));
528 
529 	/*
530 	 * If we have the MMU up in early boot, assume it is 1:1. Otherwise,
531 	 * try to get the address in a memory region compatible with the
532 	 * direct map for efficiency later.
533 	 */
534 	if (mfmsr() & PSL_DR)
535 		return (pa);
536 	else
537 		return (DMAP_BASE_ADDRESS + pa);
538 }
539 
540 void
pmap_early_io_unmap(vm_offset_t va,vm_size_t size)541 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
542 {
543 
544 	KASSERT(!pmap_bootstrapped, ("Not available after PMAP started!"));
545 }
546 
547 /* From p3-53 of the MPC7450 RISC Microprocessor Family Reference Manual */
548 void
flush_disable_caches(void)549 flush_disable_caches(void)
550 {
551 	register_t msr;
552 	register_t msscr0;
553 	register_t cache_reg;
554 	volatile uint32_t *memp;
555 	uint32_t temp;
556 	int i;
557 	int x;
558 
559 	msr = mfmsr();
560 	powerpc_sync();
561 	mtmsr(msr & ~(PSL_EE | PSL_DR));
562 	msscr0 = mfspr(SPR_MSSCR0);
563 	msscr0 &= ~MSSCR0_L2PFE;
564 	mtspr(SPR_MSSCR0, msscr0);
565 	powerpc_sync();
566 	isync();
567 	__asm__ __volatile__("dssall; sync");
568 	powerpc_sync();
569 	isync();
570 	__asm__ __volatile__("dcbf 0,%0" :: "r"(0));
571 	__asm__ __volatile__("dcbf 0,%0" :: "r"(0));
572 	__asm__ __volatile__("dcbf 0,%0" :: "r"(0));
573 
574 	/* Lock the L1 Data cache. */
575 	mtspr(SPR_LDSTCR, mfspr(SPR_LDSTCR) | 0xFF);
576 	powerpc_sync();
577 	isync();
578 
579 	mtspr(SPR_LDSTCR, 0);
580 
581 	/*
582 	 * Perform this in two stages: Flush the cache starting in RAM, then do it
583 	 * from ROM.
584 	 */
585 	memp = (volatile uint32_t *)0x00000000;
586 	for (i = 0; i < 128 * 1024; i++) {
587 		temp = *memp;
588 		__asm__ __volatile__("dcbf 0,%0" :: "r"(memp));
589 		memp += 32/sizeof(*memp);
590 	}
591 
592 	memp = (volatile uint32_t *)0xfff00000;
593 	x = 0xfe;
594 
595 	for (; x != 0xff;) {
596 		mtspr(SPR_LDSTCR, x);
597 		for (i = 0; i < 128; i++) {
598 			temp = *memp;
599 			__asm__ __volatile__("dcbf 0,%0" :: "r"(memp));
600 			memp += 32/sizeof(*memp);
601 		}
602 		x = ((x << 1) | 1) & 0xff;
603 	}
604 	mtspr(SPR_LDSTCR, 0);
605 
606 	cache_reg = mfspr(SPR_L2CR);
607 	if (cache_reg & L2CR_L2E) {
608 		cache_reg &= ~(L2CR_L2IO_7450 | L2CR_L2DO_7450);
609 		mtspr(SPR_L2CR, cache_reg);
610 		powerpc_sync();
611 		mtspr(SPR_L2CR, cache_reg | L2CR_L2HWF);
612 		while (mfspr(SPR_L2CR) & L2CR_L2HWF)
613 			; /* Busy wait for cache to flush */
614 		powerpc_sync();
615 		cache_reg &= ~L2CR_L2E;
616 		mtspr(SPR_L2CR, cache_reg);
617 		powerpc_sync();
618 		mtspr(SPR_L2CR, cache_reg | L2CR_L2I);
619 		powerpc_sync();
620 		while (mfspr(SPR_L2CR) & L2CR_L2I)
621 			; /* Busy wait for L2 cache invalidate */
622 		powerpc_sync();
623 	}
624 
625 	cache_reg = mfspr(SPR_L3CR);
626 	if (cache_reg & L3CR_L3E) {
627 		cache_reg &= ~(L3CR_L3IO | L3CR_L3DO);
628 		mtspr(SPR_L3CR, cache_reg);
629 		powerpc_sync();
630 		mtspr(SPR_L3CR, cache_reg | L3CR_L3HWF);
631 		while (mfspr(SPR_L3CR) & L3CR_L3HWF)
632 			; /* Busy wait for cache to flush */
633 		powerpc_sync();
634 		cache_reg &= ~L3CR_L3E;
635 		mtspr(SPR_L3CR, cache_reg);
636 		powerpc_sync();
637 		mtspr(SPR_L3CR, cache_reg | L3CR_L3I);
638 		powerpc_sync();
639 		while (mfspr(SPR_L3CR) & L3CR_L3I)
640 			; /* Busy wait for L3 cache invalidate */
641 		powerpc_sync();
642 	}
643 
644 	mtspr(SPR_HID0, mfspr(SPR_HID0) & ~HID0_DCE);
645 	powerpc_sync();
646 	isync();
647 
648 	mtmsr(msr);
649 }
650 
651 void
cpu_sleep(void)652 cpu_sleep(void)
653 {
654 	static u_quad_t timebase = 0;
655 	static register_t sprgs[4];
656 	static register_t srrs[2];
657 
658 	jmp_buf resetjb;
659 	struct thread *fputd;
660 	struct thread *vectd;
661 	register_t hid0;
662 	register_t msr;
663 	register_t saved_msr;
664 
665 	ap_pcpu = pcpup;
666 
667 	PCPU_SET(restore, &resetjb);
668 
669 	saved_msr = mfmsr();
670 	fputd = PCPU_GET(fputhread);
671 	vectd = PCPU_GET(vecthread);
672 	if (fputd != NULL)
673 		save_fpu(fputd);
674 	if (vectd != NULL)
675 		save_vec(vectd);
676 	if (setjmp(resetjb) == 0) {
677 		sprgs[0] = mfspr(SPR_SPRG0);
678 		sprgs[1] = mfspr(SPR_SPRG1);
679 		sprgs[2] = mfspr(SPR_SPRG2);
680 		sprgs[3] = mfspr(SPR_SPRG3);
681 		srrs[0] = mfspr(SPR_SRR0);
682 		srrs[1] = mfspr(SPR_SRR1);
683 		timebase = mftb();
684 		powerpc_sync();
685 		flush_disable_caches();
686 		hid0 = mfspr(SPR_HID0);
687 		hid0 = (hid0 & ~(HID0_DOZE | HID0_NAP)) | HID0_SLEEP;
688 		powerpc_sync();
689 		isync();
690 		msr = mfmsr() | PSL_POW;
691 		mtspr(SPR_HID0, hid0);
692 		powerpc_sync();
693 
694 		while (1)
695 			mtmsr(msr);
696 	}
697 	platform_smp_timebase_sync(timebase, 0);
698 	PCPU_SET(curthread, curthread);
699 	PCPU_SET(curpcb, curthread->td_pcb);
700 	pmap_activate(curthread);
701 	powerpc_sync();
702 	mtspr(SPR_SPRG0, sprgs[0]);
703 	mtspr(SPR_SPRG1, sprgs[1]);
704 	mtspr(SPR_SPRG2, sprgs[2]);
705 	mtspr(SPR_SPRG3, sprgs[3]);
706 	mtspr(SPR_SRR0, srrs[0]);
707 	mtspr(SPR_SRR1, srrs[1]);
708 	mtmsr(saved_msr);
709 	if (fputd == curthread)
710 		enable_fpu(curthread);
711 	if (vectd == curthread)
712 		enable_vec(curthread);
713 	powerpc_sync();
714 }
715 
716