niagara2.c revision 5615:94da8867d1d9
1214152Sed/*
2214152Sed * CDDL HEADER START
3214152Sed *
4214152Sed * The contents of this file are subject to the terms of the
5222656Sed * Common Development and Distribution License (the "License").
6222656Sed * You may not use this file except in compliance with the License.
7214152Sed *
8214152Sed * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9214152Sed * or http://www.opensolaris.org/os/licensing.
10214152Sed * See the License for the specific language governing permissions
11214152Sed * and limitations under the License.
12214152Sed *
13214152Sed * When distributing Covered Code, include this CDDL HEADER in each
14214152Sed * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15214152Sed * If applicable, add the following below this CDDL HEADER, with the
16214152Sed * fields enclosed by brackets "[]" replaced with your own identifying
17214152Sed * information: Portions Copyright [yyyy] [name of copyright owner]
18214152Sed *
19214152Sed * CDDL HEADER END
20214152Sed */
21214152Sed
22214152Sed/*
23214152Sed * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24214152Sed * Use is subject to license terms.
25214152Sed */
26214152Sed
27214152Sed#pragma ident	"%Z%%M%	%I%	%E% SMI"
28214152Sed
29214152Sed#include <sys/types.h>
30214152Sed#include <sys/systm.h>
31214152Sed#include <sys/archsystm.h>
32214152Sed#include <sys/machparam.h>
33214152Sed#include <sys/machsystm.h>
34214152Sed#include <sys/cpu.h>
35214152Sed#include <sys/elf_SPARC.h>
36214152Sed#include <vm/hat_sfmmu.h>
37214152Sed#include <vm/page.h>
38214152Sed#include <vm/vm_dep.h>
39214152Sed#include <sys/cpuvar.h>
40214152Sed#include <sys/async.h>
41214152Sed#include <sys/cmn_err.h>
42214152Sed#include <sys/debug.h>
43214152Sed#include <sys/dditypes.h>
44214152Sed#include <sys/sunddi.h>
45214152Sed#include <sys/cpu_module.h>
46214152Sed#include <sys/prom_debug.h>
47#include <sys/vmsystm.h>
48#include <sys/prom_plat.h>
49#include <sys/sysmacros.h>
50#include <sys/intreg.h>
51#include <sys/machtrap.h>
52#include <sys/ontrap.h>
53#include <sys/ivintr.h>
54#include <sys/atomic.h>
55#include <sys/panic.h>
56#include <sys/dtrace.h>
57#include <sys/simulate.h>
58#include <sys/fault.h>
59#include <sys/niagara2regs.h>
60#include <sys/hsvc.h>
61#include <sys/trapstat.h>
62
63uint_t root_phys_addr_lo_mask = 0xffffffffU;
64#if defined(NIAGARA2_IMPL)
65char cpu_module_name[] = "SUNW,UltraSPARC-T2";
66#elif defined(VFALLS_IMPL)
67char cpu_module_name[] = "SUNW,UltraSPARC-T2+";
68#endif
69
70/*
71 * Hypervisor services information for the NIAGARA2 and Victoria Falls
72 * CPU module
73 */
74static boolean_t cpu_hsvc_available = B_TRUE;
75static uint64_t cpu_sup_minor;		/* Supported minor number */
76#if defined(NIAGARA2_IMPL)
77static hsvc_info_t cpu_hsvc = {
78	HSVC_REV_1, NULL, HSVC_GROUP_NIAGARA2_CPU, NIAGARA2_HSVC_MAJOR,
79	NIAGARA2_HSVC_MINOR, cpu_module_name
80};
81#elif defined(VFALLS_IMPL)
82static hsvc_info_t cpu_hsvc = {
83	HSVC_REV_1, NULL, HSVC_GROUP_VFALLS_CPU, VFALLS_HSVC_MAJOR,
84	VFALLS_HSVC_MINOR, cpu_module_name
85};
86#endif
87
88void
89cpu_setup(void)
90{
91	extern int mmu_exported_pagesize_mask;
92	extern int cpc_has_overflow_intr;
93	extern size_t contig_mem_prealloc_base;
94	int status;
95
96	/*
97	 * Negotiate the API version for Niagara2 specific hypervisor
98	 * services.
99	 */
100	status = hsvc_register(&cpu_hsvc, &cpu_sup_minor);
101	if (status != 0) {
102		cmn_err(CE_WARN, "%s: cannot negotiate hypervisor services "
103		    "group: 0x%lx major: 0x%lx minor: 0x%lx errno: %d",
104		    cpu_hsvc.hsvc_modname, cpu_hsvc.hsvc_group,
105		    cpu_hsvc.hsvc_major, cpu_hsvc.hsvc_minor, status);
106		cpu_hsvc_available = B_FALSE;
107	}
108
109	/*
110	 * The setup common to all CPU modules is done in cpu_setup_common
111	 * routine.
112	 */
113	cpu_setup_common(NULL);
114
115	cache |= (CACHE_PTAG | CACHE_IOCOHERENT);
116
117	if ((mmu_exported_pagesize_mask &
118	    DEFAULT_SUN4V_MMU_PAGESIZE_MASK) !=
119	    DEFAULT_SUN4V_MMU_PAGESIZE_MASK)
120		cmn_err(CE_PANIC, "machine description"
121		    " does not have required sun4v page sizes"
122		    " 8K, 64K and 4M: MD mask is 0x%x",
123		    mmu_exported_pagesize_mask);
124
125	cpu_hwcap_flags = AV_SPARC_VIS | AV_SPARC_VIS2 | AV_SPARC_ASI_BLK_INIT;
126
127	/*
128	 * Niagara2 supports a 48-bit subset of the full 64-bit virtual
129	 * address space. Virtual addresses between 0x0000800000000000
130	 * and 0xffff.7fff.ffff.ffff inclusive lie within a "VA Hole"
131	 * and must never be mapped. In addition, software must not use
132	 * pages within 4GB of the VA hole as instruction pages to
133	 * avoid problems with prefetching into the VA hole.
134	 */
135	hole_start = (caddr_t)((1ull << (va_bits - 1)) - (1ull << 32));
136	hole_end = (caddr_t)((0ull - (1ull << (va_bits - 1))) + (1ull << 32));
137
138	/*
139	 * Niagara2 has a performance counter overflow interrupt
140	 */
141	cpc_has_overflow_intr = 1;
142
143	/*
144	 * Enable 4M pages for OOB.
145	 */
146	max_uheap_lpsize = MMU_PAGESIZE4M;
147	max_ustack_lpsize = MMU_PAGESIZE4M;
148	max_privmap_lpsize = MMU_PAGESIZE4M;
149
150	contig_mem_prealloc_base = NIAGARA2_PREALLOC_BASE;
151}
152
153/*
154 * Set the magic constants of the implementation.
155 */
156void
157cpu_fiximp(struct cpu_node *cpunode)
158{
159	/*
160	 * The Cache node is optional in MD. Therefore in case "Cache"
161	 * node does not exists in MD, set the default L2 cache associativity,
162	 * size, linesize.
163	 */
164	if (cpunode->ecache_size == 0)
165		cpunode->ecache_size = L2CACHE_SIZE;
166	if (cpunode->ecache_linesize == 0)
167		cpunode->ecache_linesize = L2CACHE_LINESIZE;
168	if (cpunode->ecache_associativity == 0)
169		cpunode->ecache_associativity = L2CACHE_ASSOCIATIVITY;
170}
171
172void
173cpu_map_exec_units(struct cpu *cp)
174{
175	ASSERT(MUTEX_HELD(&cpu_lock));
176
177	/*
178	 * The cpu_ipipe and cpu_fpu fields are initialized based on
179	 * the execution unit sharing information from the MD. They
180	 * default to the CPU id in the absence of such information.
181	 */
182	cp->cpu_m.cpu_ipipe = cpunodes[cp->cpu_id].exec_unit_mapping;
183	if (cp->cpu_m.cpu_ipipe == NO_EU_MAPPING_FOUND)
184		cp->cpu_m.cpu_ipipe = (id_t)(cp->cpu_id);
185
186	cp->cpu_m.cpu_fpu = cpunodes[cp->cpu_id].fpu_mapping;
187	if (cp->cpu_m.cpu_fpu == NO_EU_MAPPING_FOUND)
188		cp->cpu_m.cpu_fpu = (id_t)(cp->cpu_id);
189
190	/*
191	 * Niagara 2 defines the core to be at the FPU level
192	 */
193	cp->cpu_m.cpu_core = cp->cpu_m.cpu_fpu;
194
195	/*
196	 * The cpu_chip field is initialized based on the information
197	 * in the MD and assume that all cpus within a chip
198	 * share the same L2 cache. If no such info is available, we
199	 * set the cpu to belong to the defacto chip 0.
200	 */
201	cp->cpu_m.cpu_mpipe = cpunodes[cp->cpu_id].l2_cache_mapping;
202	if (cp->cpu_m.cpu_mpipe == NO_L2_CACHE_MAPPING_FOUND)
203		cp->cpu_m.cpu_mpipe = CPU_L2_CACHEID_INVALID;
204
205	cp->cpu_m.cpu_chip = cpunodes[cp->cpu_id].l2_cache_mapping;
206	if (cp->cpu_m.cpu_chip == NO_L2_CACHE_MAPPING_FOUND)
207		cp->cpu_m.cpu_chip = CPU_CHIPID_INVALID;
208}
209
210static int cpucnt;
211
212void
213cpu_init_private(struct cpu *cp)
214{
215	extern void niagara_kstat_init(void);
216
217	ASSERT(MUTEX_HELD(&cpu_lock));
218
219	cpu_map_exec_units(cp);
220
221	if ((cpucnt++ == 0) && (cpu_hsvc_available == B_TRUE))
222		(void) niagara_kstat_init();
223}
224
225/*ARGSUSED*/
226void
227cpu_uninit_private(struct cpu *cp)
228{
229	extern void niagara_kstat_fini(void);
230
231	ASSERT(MUTEX_HELD(&cpu_lock));
232	if ((--cpucnt == 0) && (cpu_hsvc_available == B_TRUE))
233		(void) niagara_kstat_fini();
234}
235
236/*
237 * On Niagara2, any flush will cause all preceding stores to be
238 * synchronized wrt the i$, regardless of address or ASI.  In fact,
239 * the address is ignored, so we always flush address 0.
240 */
241/*ARGSUSED*/
242void
243dtrace_flush_sec(uintptr_t addr)
244{
245	doflush(0);
246}
247
248/*
249 * Trapstat support for Niagara2 processor
250 * The Niagara2 provides HWTW support for TSB lookup and with HWTW
251 * enabled no TSB hit information will be available. Therefore setting
252 * the time spent in TLB miss handler for TSB hits to 0.
253 */
254int
255cpu_trapstat_conf(int cmd)
256{
257	int status = 0;
258
259	switch (cmd) {
260	case CPU_TSTATCONF_INIT:
261	case CPU_TSTATCONF_FINI:
262	case CPU_TSTATCONF_ENABLE:
263	case CPU_TSTATCONF_DISABLE:
264		break;
265	default:
266		status = EINVAL;
267		break;
268	}
269	return (status);
270}
271
272void
273cpu_trapstat_data(void *buf, uint_t tstat_pgszs)
274{
275	tstat_pgszdata_t	*tstatp = (tstat_pgszdata_t *)buf;
276	int	i;
277
278	for (i = 0; i < tstat_pgszs; i++, tstatp++) {
279		tstatp->tpgsz_kernel.tmode_itlb.ttlb_tlb.tmiss_count = 0;
280		tstatp->tpgsz_kernel.tmode_itlb.ttlb_tlb.tmiss_time = 0;
281		tstatp->tpgsz_user.tmode_itlb.ttlb_tlb.tmiss_count = 0;
282		tstatp->tpgsz_user.tmode_itlb.ttlb_tlb.tmiss_time = 0;
283		tstatp->tpgsz_kernel.tmode_dtlb.ttlb_tlb.tmiss_count = 0;
284		tstatp->tpgsz_kernel.tmode_dtlb.ttlb_tlb.tmiss_time = 0;
285		tstatp->tpgsz_user.tmode_dtlb.ttlb_tlb.tmiss_count = 0;
286		tstatp->tpgsz_user.tmode_dtlb.ttlb_tlb.tmiss_time = 0;
287	}
288}
289
290/*
291 * Page coloring support for hashed cache index mode
292 */
293
294/*
295 * Node id bits from machine description (MD).  Node id distinguishes
296 * local versus remote memory. Because of MPO, page allocation does
297 * not cross node boundaries. Therefore, remove the node id bits from
298 * the color, since they are fixed. Either bit 30, or 31:30 in
299 * Victoria Falls processors.
300 * The number of node id bits is always 0 in Niagara2.
301 */
302typedef struct n2color {
303	uchar_t nnbits;	/* number of node id bits */
304	uchar_t nnmask; /* mask for node id bits */
305	uchar_t	lomask;	/* mask for bits below node id */
306	uchar_t lobits;	/* number of bits below node id */
307} n2color_t;
308
309n2color_t n2color[MMU_PAGE_SIZES];
310static uchar_t nhbits[] = {7, 7, 6, 5, 5, 5};
311
312/*
313 * Remove node id bits from color bits 32:28.
314 * This will reduce the number of colors.
315 * No change if number of node bits is zero.
316 */
317static inline uint_t
318n2_hash2color(uint_t color, uchar_t szc)
319{
320	n2color_t m = n2color[szc];
321
322	if (m.nnbits > 0) {
323		color = ((color >> m.nnbits) & ~m.lomask) | (color & m.lomask);
324		ASSERT((color & ~(hw_page_array[szc].hp_colors - 1)) == 0);
325	}
326
327	return (color);
328}
329
330/*
331 * Restore node id bits into page color.
332 * This will increase the number of colors to match N2.
333 * No change if number of node bits is zero.
334 */
335static inline uint_t
336n2_color2hash(uint_t color, uchar_t szc, uint_t node)
337{
338	n2color_t m = n2color[szc];
339
340	if (m.nnbits > 0) {
341		color = ((color & ~m.lomask) << m.nnbits) | (color & m.lomask);
342		color |= (node & m.nnmask) << m.lobits;
343	}
344
345	return (color);
346}
347
348/* NI2 L2$ index is pa[32:28]^pa[17:13].pa[19:18]^pa[12:11].pa[10:6] */
349
350/*
351 * iterator NULL means pfn is VA, do not adjust ra_to_pa
352 * iterator (-1) means pfn is RA, need to convert to PA
353 * iterator non-null means pfn is RA, use ra_to_pa
354 */
355uint_t
356page_pfn_2_color_cpu(pfn_t pfn, uchar_t szc, void *cookie)
357{
358	mem_node_iterator_t *it = cookie;
359	uint_t color;
360
361	ASSERT(szc <= TTE256M);
362
363	if (it == ((mem_node_iterator_t *)(-1))) {
364		pfn = plat_rapfn_to_papfn(pfn);
365	} else if (it != NULL) {
366		ASSERT(pfn >= it->mi_mblock_base && pfn <= it->mi_mblock_end);
367		pfn = pfn + it->mi_ra_to_pa;
368	}
369	pfn = PFN_BASE(pfn, szc);
370	color = ((pfn >> 15) ^ pfn) & 0x1f;
371	if (szc < TTE4M) {
372		/* 19:18 */
373		color = (color << 2) | ((pfn >> 5) & 0x3);
374		if (szc > TTE64K)
375			color >>= 1;    /* 19 */
376	}
377	return (n2_hash2color(color, szc));
378}
379
380static uint_t
381page_papfn_2_color_cpu(pfn_t papfn, uchar_t szc)
382{
383	uint_t color;
384
385	ASSERT(szc <= TTE256M);
386
387	papfn = PFN_BASE(papfn, szc);
388	color = ((papfn >> 15) ^ papfn) & 0x1f;
389	if (szc < TTE4M) {
390		/* 19:18 */
391		color = (color << 2) | ((papfn >> 5) & 0x3);
392		if (szc > TTE64K)
393			color >>= 1;    /* 19 */
394	}
395	return (color);
396}
397
398#if TTE256M != 5
399#error TTE256M is not 5
400#endif
401
402uint_t
403page_get_nsz_color_mask_cpu(uchar_t szc, uint_t mask)
404{
405	static uint_t ni2_color_masks[5] = {0x63, 0x1e, 0x3e, 0x1f, 0x1f};
406	ASSERT(szc < TTE256M);
407	mask = n2_color2hash(mask, szc, 0);
408	mask &= ni2_color_masks[szc];
409	if (szc == TTE64K || szc == TTE512K)
410		mask >>= 1;
411	return (n2_hash2color(mask, szc + 1));
412}
413
414uint_t
415page_get_nsz_color_cpu(uchar_t szc, uint_t color)
416{
417	ASSERT(szc < TTE256M);
418	color = n2_color2hash(color, szc, 0);
419	if (szc == TTE64K || szc == TTE512K)
420		color >>= 1;
421	return (n2_hash2color(color, szc + 1));
422}
423
424uint_t
425page_get_color_shift_cpu(uchar_t szc, uchar_t nszc)
426{
427	uint_t s;
428	ASSERT(nszc >= szc);
429	ASSERT(nszc <= TTE256M);
430
431	s = nhbits[szc] - n2color[szc].nnbits;
432	s -= nhbits[nszc] - n2color[nszc].nnbits;
433
434	return (s);
435}
436
437uint_t
438page_convert_color_cpu(uint_t ncolor, uchar_t szc, uchar_t nszc)
439{
440	uint_t color;
441
442	ASSERT(nszc > szc);
443	ASSERT(nszc <= TTE256M);
444	ncolor = n2_color2hash(ncolor, nszc, 0);
445	color = ncolor << (nhbits[szc] - nhbits[nszc]);
446	color = n2_hash2color(color, szc);
447	return (color);
448}
449
450#define	PAPFN_2_MNODE(pfn) \
451	(((pfn) & it->mi_mnode_pfn_mask) >> it->mi_mnode_pfn_shift)
452
453/*ARGSUSED*/
454pfn_t
455page_next_pfn_for_color_cpu(pfn_t pfn, uchar_t szc, uint_t color,
456    uint_t ceq_mask, uint_t color_mask, void *cookie)
457{
458	mem_node_iterator_t *it = cookie;
459	pfn_t pstep = PNUM_SIZE(szc);
460	pfn_t npfn, pfn_ceq_mask, pfn_color;
461	pfn_t tmpmask, mask = (pfn_t)-1;
462	uint_t pfnmn;
463
464	ASSERT((color & ~ceq_mask) == 0);
465	ASSERT(pfn >= it->mi_mblock_base && pfn <= it->mi_mblock_end);
466
467	/* convert RA to PA for accurate color calculation */
468	if (it->mi_init) {
469		/* first call after it, so cache these values */
470		it->mi_hash_ceq_mask =
471		    n2_color2hash(ceq_mask, szc, it->mi_mnode_mask);
472		it->mi_hash_color =
473		    n2_color2hash(color, szc, it->mi_mnode);
474		it->mi_init = 0;
475	} else {
476		ASSERT(it->mi_hash_ceq_mask ==
477		    n2_color2hash(ceq_mask, szc, it->mi_mnode_mask));
478		ASSERT(it->mi_hash_color ==
479		    n2_color2hash(color, szc, it->mi_mnode));
480	}
481	ceq_mask = it->mi_hash_ceq_mask;
482	color = it->mi_hash_color;
483	pfn += it->mi_ra_to_pa;
484
485	/* restart here when we switch memblocks */
486next_mem_block:
487	if (szc <= TTE64K) {
488		pfnmn = PAPFN_2_MNODE(pfn);
489	}
490	if (((page_papfn_2_color_cpu(pfn, szc) ^ color) & ceq_mask) == 0 &&
491	    (szc > TTE64K || pfnmn == it->mi_mnode)) {
492
493		/* we start from the page with correct color */
494		if (szc >= TTE512K) {
495			if (szc >= TTE4M) {
496				/* page color is PA[32:28] */
497				pfn_ceq_mask = ceq_mask << 15;
498			} else {
499				/* page color is PA[32:28].PA[19:19] */
500				pfn_ceq_mask = ((ceq_mask & 1) << 6) |
501				    ((ceq_mask >> 1) << 15);
502			}
503			npfn = ADD_MASKED(pfn, pstep, pfn_ceq_mask, mask);
504			goto done;
505		} else {
506			/*
507			 * We deal 64K or 8K page. Check if we could the
508			 * satisfy the request without changing PA[32:28]
509			 */
510			pfn_ceq_mask = ((ceq_mask & 3) << 5) | (ceq_mask >> 2);
511			pfn_ceq_mask |= it->mi_mnode_pfn_mask;
512			npfn = ADD_MASKED(pfn, pstep, pfn_ceq_mask, mask);
513
514			if ((((npfn ^ pfn) >> 15) & 0x1f) == 0)
515				goto done;
516
517			/*
518			 * for next pfn we have to change bits PA[32:28]
519			 * set PA[63:28] and PA[19:18] of the next pfn
520			 */
521			npfn = (pfn >> 15) << 15;
522			npfn |= (ceq_mask & color & 3) << 5;
523			pfn_ceq_mask = (szc == TTE8K) ? 0 :
524			    (ceq_mask & 0x1c) << 13;
525			pfn_ceq_mask |= it->mi_mnode_pfn_mask;
526			npfn = ADD_MASKED(npfn, (1 << 15), pfn_ceq_mask, mask);
527
528			/*
529			 * set bits PA[17:13] to match the color
530			 */
531			npfn |= ((npfn >> 15) ^ (color >> 2)) & (ceq_mask >> 2);
532			goto done;
533		}
534	}
535
536	/*
537	 * we start from the page with incorrect color - rare case
538	 */
539	if (szc >= TTE512K) {
540		if (szc >= TTE4M) {
541			/* page color is in bits PA[32:28] */
542			npfn = ((pfn >> 20) << 20) | (color << 15);
543			pfn_ceq_mask = (ceq_mask << 15) | 0x7fff;
544		} else {
545			/* try get the right color by changing bit PA[19:19] */
546			npfn = pfn + pstep;
547			if (((page_papfn_2_color_cpu(npfn, szc) ^ color) &
548			    ceq_mask) == 0)
549				goto done;
550
551			/* page color is PA[32:28].PA[19:19] */
552			pfn_ceq_mask = ((ceq_mask & 1) << 6) |
553			    ((ceq_mask >> 1) << 15) | (0xff << 7);
554			pfn_color = ((color & 1) << 6) | ((color >> 1) << 15);
555			npfn = ((pfn >> 20) << 20) | pfn_color;
556		}
557
558		while (npfn <= pfn) {
559			npfn = ADD_MASKED(npfn, pstep, pfn_ceq_mask, mask);
560		}
561		goto done;
562	}
563
564	/*
565	 *  We deal 64K or 8K page of incorrect color.
566	 * Try correcting color without changing PA[32:28]
567	 */
568	pfn_ceq_mask = ((ceq_mask & 3) << 5) | (ceq_mask >> 2);
569	pfn_color = ((color & 3) << 5) | (color >> 2);
570	if (pfnmn == it->mi_mnode) {
571		npfn = (pfn & ~(pfn_t)0x7f);
572		npfn |= (((pfn >> 15) & 0x1f) ^ pfn_color) & pfn_ceq_mask;
573		npfn = (szc == TTE64K) ? (npfn & ~(pfn_t)0x7) : npfn;
574
575		if (((page_papfn_2_color_cpu(npfn, szc) ^ color) &
576		    ceq_mask) == 0) {
577			/* the color is fixed - find the next page */
578			pfn_ceq_mask |= it->mi_mnode_pfn_mask;
579			while (npfn <= pfn) {
580				npfn = ADD_MASKED(npfn, pstep, pfn_ceq_mask,
581				    mask);
582			}
583			if ((((npfn ^ pfn) >> 15) & 0x1f) == 0)
584				goto done;
585		}
586	}
587
588	/* to fix the color need to touch PA[32:28] */
589	npfn = (szc == TTE8K) ? ((pfn >> 15) << 15) :
590	    (((pfn >> 18) << 18) | ((color & 0x1c) << 13));
591
592	/* fix mnode if input pfn is in the wrong mnode. */
593	if ((pfnmn = PAPFN_2_MNODE(npfn)) != it->mi_mnode) {
594		npfn += ((it->mi_mnode - pfnmn) & it->mi_mnode_mask) <<
595		    it->mi_mnode_pfn_shift;
596	}
597
598	tmpmask = (szc == TTE8K) ? 0 : (ceq_mask & 0x1c) << 13;
599	tmpmask |= it->mi_mnode_pfn_mask;
600
601	while (npfn <= pfn) {
602		npfn = ADD_MASKED(npfn, (1 << 15), tmpmask, mask);
603	}
604
605	/* set bits PA[19:13] to match the color */
606	npfn |= (((npfn >> 15) & 0x1f) ^ pfn_color) & pfn_ceq_mask;
607	npfn = (szc == TTE64K) ? (npfn & ~(pfn_t)0x7) : npfn;
608
609done:
610	ASSERT(((page_papfn_2_color_cpu(npfn, szc) ^ color) & ceq_mask) == 0);
611	ASSERT(PAPFN_2_MNODE(npfn) == it->mi_mnode);
612
613	/* PA to RA */
614	npfn -= it->mi_ra_to_pa;
615
616	/* check for possible memblock switch */
617	if (npfn > it->mi_mblock_end) {
618		pfn = plat_mem_node_iterator_init(npfn, it->mi_mnode, it, 0);
619		if (pfn == (pfn_t)-1)
620			return (pfn);
621		ASSERT(pfn >= it->mi_mblock_base && pfn <= it->mi_mblock_end);
622		pfn += it->mi_ra_to_pa;
623		goto next_mem_block;
624	}
625
626	return (npfn);
627}
628
629/*
630 * init page coloring
631 * VF encodes node_id for an L-group in either bit 30 or 31:30,
632 * which effectively reduces the number of colors available per mnode.
633 */
634void
635page_coloring_init_cpu()
636{
637	int i;
638	uchar_t id;
639	uchar_t lo;
640	uchar_t hi;
641	n2color_t m;
642	mem_node_iterator_t it;
643	static uchar_t idmask[] = {0, 0x7, 0x1f, 0x1f, 0x1f, 0x1f};
644
645	for (i = 0; i < max_mem_nodes; i++) {
646		memset(&it, 0, sizeof (it));
647		if (plat_mem_node_iterator_init(0, i, &it, 1) != (pfn_t)-1)
648			break;
649	}
650	ASSERT(i < max_mem_nodes);
651	for (i = 0; i < mmu_page_sizes; i++) {
652		(void) memset(&m, 0, sizeof (m));
653		id = it.mi_mnode_pfn_mask >> 15;	/* node id mask */
654		id &= idmask[i];
655		lo = lowbit(id);
656		if (lo > 0) {
657			hi = highbit(id);
658			m.nnbits = hi - lo + 1;
659			m.nnmask = (1 << m.nnbits) - 1;
660			lo += nhbits[i] - 5;
661			m.lomask = (1 << (lo - 1)) - 1;
662			m.lobits = lo - 1;
663		}
664		hw_page_array[i].hp_colors = 1 << (nhbits[i] - m.nnbits);
665		n2color[i] = m;
666	}
667}
668
669/*
670 * group colorequiv colors on N2 by low order bits of the color first
671 */
672void
673page_set_colorequiv_arr_cpu(void)
674{
675	static uint_t nequiv_shades_log2[MMU_PAGE_SIZES] = {2, 5, 0, 0, 0, 0};
676
677	nequiv_shades_log2[1] -= n2color[1].nnbits;
678	if (colorequiv > 1) {
679		int i;
680		uint_t sv_a = lowbit(colorequiv) - 1;
681
682		if (sv_a > 15)
683			sv_a = 15;
684
685		for (i = 0; i < MMU_PAGE_SIZES; i++) {
686			uint_t colors;
687			uint_t a = sv_a;
688
689			if ((colors = hw_page_array[i].hp_colors) <= 1)
690				continue;
691			while ((colors >> a) == 0)
692				a--;
693			if (a > (colorequivszc[i] & 0xf) +
694			    (colorequivszc[i] >> 4)) {
695				if (a <= nequiv_shades_log2[i]) {
696					colorequivszc[i] = (uchar_t)a;
697				} else {
698					colorequivszc[i] =
699					    ((a - nequiv_shades_log2[i]) << 4) |
700					    nequiv_shades_log2[i];
701				}
702			}
703		}
704	}
705}
706