xref: /openbmc/linux/arch/powerpc/mm/nohash/tlb.c (revision 4bb1e4e7)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * This file contains the routines for TLB flushing.
4  * On machines where the MMU does not use a hash table to store virtual to
5  * physical translations (ie, SW loaded TLBs or Book3E compilant processors,
6  * this does -not- include 603 however which shares the implementation with
7  * hash based processors)
8  *
9  *  -- BenH
10  *
11  * Copyright 2008,2009 Ben Herrenschmidt <benh@kernel.crashing.org>
12  *                     IBM Corp.
13  *
14  *  Derived from arch/ppc/mm/init.c:
15  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
16  *
17  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
18  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
19  *    Copyright (C) 1996 Paul Mackerras
20  *
21  *  Derived from "arch/i386/mm/init.c"
22  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
23  */
24 
25 #include <linux/kernel.h>
26 #include <linux/export.h>
27 #include <linux/mm.h>
28 #include <linux/init.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/preempt.h>
32 #include <linux/spinlock.h>
33 #include <linux/memblock.h>
34 #include <linux/of_fdt.h>
35 #include <linux/hugetlb.h>
36 
37 #include <asm/tlbflush.h>
38 #include <asm/tlb.h>
39 #include <asm/code-patching.h>
40 #include <asm/cputhreads.h>
41 #include <asm/hugetlb.h>
42 #include <asm/paca.h>
43 
44 #include <mm/mmu_decl.h>
45 
46 /*
47  * This struct lists the sw-supported page sizes.  The hardawre MMU may support
48  * other sizes not listed here.   The .ind field is only used on MMUs that have
49  * indirect page table entries.
50  */
51 #if defined(CONFIG_PPC_BOOK3E_MMU) || defined(CONFIG_PPC_8xx)
52 #ifdef CONFIG_PPC_FSL_BOOK3E
53 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
54 	[MMU_PAGE_4K] = {
55 		.shift	= 12,
56 		.enc	= BOOK3E_PAGESZ_4K,
57 	},
58 	[MMU_PAGE_2M] = {
59 		.shift	= 21,
60 		.enc	= BOOK3E_PAGESZ_2M,
61 	},
62 	[MMU_PAGE_4M] = {
63 		.shift	= 22,
64 		.enc	= BOOK3E_PAGESZ_4M,
65 	},
66 	[MMU_PAGE_16M] = {
67 		.shift	= 24,
68 		.enc	= BOOK3E_PAGESZ_16M,
69 	},
70 	[MMU_PAGE_64M] = {
71 		.shift	= 26,
72 		.enc	= BOOK3E_PAGESZ_64M,
73 	},
74 	[MMU_PAGE_256M] = {
75 		.shift	= 28,
76 		.enc	= BOOK3E_PAGESZ_256M,
77 	},
78 	[MMU_PAGE_1G] = {
79 		.shift	= 30,
80 		.enc	= BOOK3E_PAGESZ_1GB,
81 	},
82 };
83 #elif defined(CONFIG_PPC_8xx)
84 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
85 	/* we only manage 4k and 16k pages as normal pages */
86 #ifdef CONFIG_PPC_4K_PAGES
87 	[MMU_PAGE_4K] = {
88 		.shift	= 12,
89 	},
90 #else
91 	[MMU_PAGE_16K] = {
92 		.shift	= 14,
93 	},
94 #endif
95 	[MMU_PAGE_512K] = {
96 		.shift	= 19,
97 	},
98 	[MMU_PAGE_8M] = {
99 		.shift	= 23,
100 	},
101 };
102 #else
103 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
104 	[MMU_PAGE_4K] = {
105 		.shift	= 12,
106 		.ind	= 20,
107 		.enc	= BOOK3E_PAGESZ_4K,
108 	},
109 	[MMU_PAGE_16K] = {
110 		.shift	= 14,
111 		.enc	= BOOK3E_PAGESZ_16K,
112 	},
113 	[MMU_PAGE_64K] = {
114 		.shift	= 16,
115 		.ind	= 28,
116 		.enc	= BOOK3E_PAGESZ_64K,
117 	},
118 	[MMU_PAGE_1M] = {
119 		.shift	= 20,
120 		.enc	= BOOK3E_PAGESZ_1M,
121 	},
122 	[MMU_PAGE_16M] = {
123 		.shift	= 24,
124 		.ind	= 36,
125 		.enc	= BOOK3E_PAGESZ_16M,
126 	},
127 	[MMU_PAGE_256M] = {
128 		.shift	= 28,
129 		.enc	= BOOK3E_PAGESZ_256M,
130 	},
131 	[MMU_PAGE_1G] = {
132 		.shift	= 30,
133 		.enc	= BOOK3E_PAGESZ_1GB,
134 	},
135 };
136 #endif /* CONFIG_FSL_BOOKE */
137 
138 static inline int mmu_get_tsize(int psize)
139 {
140 	return mmu_psize_defs[psize].enc;
141 }
142 #else
143 static inline int mmu_get_tsize(int psize)
144 {
145 	/* This isn't used on !Book3E for now */
146 	return 0;
147 }
148 #endif /* CONFIG_PPC_BOOK3E_MMU */
149 
150 /* The variables below are currently only used on 64-bit Book3E
151  * though this will probably be made common with other nohash
152  * implementations at some point
153  */
154 #ifdef CONFIG_PPC64
155 
156 int mmu_linear_psize;		/* Page size used for the linear mapping */
157 int mmu_pte_psize;		/* Page size used for PTE pages */
158 int mmu_vmemmap_psize;		/* Page size used for the virtual mem map */
159 int book3e_htw_mode;		/* HW tablewalk?  Value is PPC_HTW_* */
160 unsigned long linear_map_top;	/* Top of linear mapping */
161 
162 
163 /*
164  * Number of bytes to add to SPRN_SPRG_TLB_EXFRAME on crit/mcheck/debug
165  * exceptions.  This is used for bolted and e6500 TLB miss handlers which
166  * do not modify this SPRG in the TLB miss code; for other TLB miss handlers,
167  * this is set to zero.
168  */
169 int extlb_level_exc;
170 
171 #endif /* CONFIG_PPC64 */
172 
173 #ifdef CONFIG_PPC_FSL_BOOK3E
174 /* next_tlbcam_idx is used to round-robin tlbcam entry assignment */
175 DEFINE_PER_CPU(int, next_tlbcam_idx);
176 EXPORT_PER_CPU_SYMBOL(next_tlbcam_idx);
177 #endif
178 
179 /*
180  * Base TLB flushing operations:
181  *
182  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
183  *  - flush_tlb_page(vma, vmaddr) flushes one page
184  *  - flush_tlb_range(vma, start, end) flushes a range of pages
185  *  - flush_tlb_kernel_range(start, end) flushes kernel pages
186  *
187  *  - local_* variants of page and mm only apply to the current
188  *    processor
189  */
190 
191 /*
192  * These are the base non-SMP variants of page and mm flushing
193  */
194 void local_flush_tlb_mm(struct mm_struct *mm)
195 {
196 	unsigned int pid;
197 
198 	preempt_disable();
199 	pid = mm->context.id;
200 	if (pid != MMU_NO_CONTEXT)
201 		_tlbil_pid(pid);
202 	preempt_enable();
203 }
204 EXPORT_SYMBOL(local_flush_tlb_mm);
205 
206 void __local_flush_tlb_page(struct mm_struct *mm, unsigned long vmaddr,
207 			    int tsize, int ind)
208 {
209 	unsigned int pid;
210 
211 	preempt_disable();
212 	pid = mm ? mm->context.id : 0;
213 	if (pid != MMU_NO_CONTEXT)
214 		_tlbil_va(vmaddr, pid, tsize, ind);
215 	preempt_enable();
216 }
217 
218 void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
219 {
220 	__local_flush_tlb_page(vma ? vma->vm_mm : NULL, vmaddr,
221 			       mmu_get_tsize(mmu_virtual_psize), 0);
222 }
223 EXPORT_SYMBOL(local_flush_tlb_page);
224 
225 /*
226  * And here are the SMP non-local implementations
227  */
228 #ifdef CONFIG_SMP
229 
230 static DEFINE_RAW_SPINLOCK(tlbivax_lock);
231 
232 struct tlb_flush_param {
233 	unsigned long addr;
234 	unsigned int pid;
235 	unsigned int tsize;
236 	unsigned int ind;
237 };
238 
239 static void do_flush_tlb_mm_ipi(void *param)
240 {
241 	struct tlb_flush_param *p = param;
242 
243 	_tlbil_pid(p ? p->pid : 0);
244 }
245 
246 static void do_flush_tlb_page_ipi(void *param)
247 {
248 	struct tlb_flush_param *p = param;
249 
250 	_tlbil_va(p->addr, p->pid, p->tsize, p->ind);
251 }
252 
253 
254 /* Note on invalidations and PID:
255  *
256  * We snapshot the PID with preempt disabled. At this point, it can still
257  * change either because:
258  * - our context is being stolen (PID -> NO_CONTEXT) on another CPU
259  * - we are invaliating some target that isn't currently running here
260  *   and is concurrently acquiring a new PID on another CPU
261  * - some other CPU is re-acquiring a lost PID for this mm
262  * etc...
263  *
264  * However, this shouldn't be a problem as we only guarantee
265  * invalidation of TLB entries present prior to this call, so we
266  * don't care about the PID changing, and invalidating a stale PID
267  * is generally harmless.
268  */
269 
270 void flush_tlb_mm(struct mm_struct *mm)
271 {
272 	unsigned int pid;
273 
274 	preempt_disable();
275 	pid = mm->context.id;
276 	if (unlikely(pid == MMU_NO_CONTEXT))
277 		goto no_context;
278 	if (!mm_is_core_local(mm)) {
279 		struct tlb_flush_param p = { .pid = pid };
280 		/* Ignores smp_processor_id() even if set. */
281 		smp_call_function_many(mm_cpumask(mm),
282 				       do_flush_tlb_mm_ipi, &p, 1);
283 	}
284 	_tlbil_pid(pid);
285  no_context:
286 	preempt_enable();
287 }
288 EXPORT_SYMBOL(flush_tlb_mm);
289 
290 void __flush_tlb_page(struct mm_struct *mm, unsigned long vmaddr,
291 		      int tsize, int ind)
292 {
293 	struct cpumask *cpu_mask;
294 	unsigned int pid;
295 
296 	/*
297 	 * This function as well as __local_flush_tlb_page() must only be called
298 	 * for user contexts.
299 	 */
300 	if (WARN_ON(!mm))
301 		return;
302 
303 	preempt_disable();
304 	pid = mm->context.id;
305 	if (unlikely(pid == MMU_NO_CONTEXT))
306 		goto bail;
307 	cpu_mask = mm_cpumask(mm);
308 	if (!mm_is_core_local(mm)) {
309 		/* If broadcast tlbivax is supported, use it */
310 		if (mmu_has_feature(MMU_FTR_USE_TLBIVAX_BCAST)) {
311 			int lock = mmu_has_feature(MMU_FTR_LOCK_BCAST_INVAL);
312 			if (lock)
313 				raw_spin_lock(&tlbivax_lock);
314 			_tlbivax_bcast(vmaddr, pid, tsize, ind);
315 			if (lock)
316 				raw_spin_unlock(&tlbivax_lock);
317 			goto bail;
318 		} else {
319 			struct tlb_flush_param p = {
320 				.pid = pid,
321 				.addr = vmaddr,
322 				.tsize = tsize,
323 				.ind = ind,
324 			};
325 			/* Ignores smp_processor_id() even if set in cpu_mask */
326 			smp_call_function_many(cpu_mask,
327 					       do_flush_tlb_page_ipi, &p, 1);
328 		}
329 	}
330 	_tlbil_va(vmaddr, pid, tsize, ind);
331  bail:
332 	preempt_enable();
333 }
334 
335 void flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
336 {
337 #ifdef CONFIG_HUGETLB_PAGE
338 	if (vma && is_vm_hugetlb_page(vma))
339 		flush_hugetlb_page(vma, vmaddr);
340 #endif
341 
342 	__flush_tlb_page(vma ? vma->vm_mm : NULL, vmaddr,
343 			 mmu_get_tsize(mmu_virtual_psize), 0);
344 }
345 EXPORT_SYMBOL(flush_tlb_page);
346 
347 #endif /* CONFIG_SMP */
348 
349 #ifdef CONFIG_PPC_47x
350 void __init early_init_mmu_47x(void)
351 {
352 #ifdef CONFIG_SMP
353 	unsigned long root = of_get_flat_dt_root();
354 	if (of_get_flat_dt_prop(root, "cooperative-partition", NULL))
355 		mmu_clear_feature(MMU_FTR_USE_TLBIVAX_BCAST);
356 #endif /* CONFIG_SMP */
357 }
358 #endif /* CONFIG_PPC_47x */
359 
360 /*
361  * Flush kernel TLB entries in the given range
362  */
363 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
364 {
365 #ifdef CONFIG_SMP
366 	preempt_disable();
367 	smp_call_function(do_flush_tlb_mm_ipi, NULL, 1);
368 	_tlbil_pid(0);
369 	preempt_enable();
370 #else
371 	_tlbil_pid(0);
372 #endif
373 }
374 EXPORT_SYMBOL(flush_tlb_kernel_range);
375 
376 /*
377  * Currently, for range flushing, we just do a full mm flush. This should
378  * be optimized based on a threshold on the size of the range, since
379  * some implementation can stack multiple tlbivax before a tlbsync but
380  * for now, we keep it that way
381  */
382 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
383 		     unsigned long end)
384 
385 {
386 	if (end - start == PAGE_SIZE && !(start & ~PAGE_MASK))
387 		flush_tlb_page(vma, start);
388 	else
389 		flush_tlb_mm(vma->vm_mm);
390 }
391 EXPORT_SYMBOL(flush_tlb_range);
392 
393 void tlb_flush(struct mmu_gather *tlb)
394 {
395 	flush_tlb_mm(tlb->mm);
396 }
397 
398 /*
399  * Below are functions specific to the 64-bit variant of Book3E though that
400  * may change in the future
401  */
402 
403 #ifdef CONFIG_PPC64
404 
405 /*
406  * Handling of virtual linear page tables or indirect TLB entries
407  * flushing when PTE pages are freed
408  */
409 void tlb_flush_pgtable(struct mmu_gather *tlb, unsigned long address)
410 {
411 	int tsize = mmu_psize_defs[mmu_pte_psize].enc;
412 
413 	if (book3e_htw_mode != PPC_HTW_NONE) {
414 		unsigned long start = address & PMD_MASK;
415 		unsigned long end = address + PMD_SIZE;
416 		unsigned long size = 1UL << mmu_psize_defs[mmu_pte_psize].shift;
417 
418 		/* This isn't the most optimal, ideally we would factor out the
419 		 * while preempt & CPU mask mucking around, or even the IPI but
420 		 * it will do for now
421 		 */
422 		while (start < end) {
423 			__flush_tlb_page(tlb->mm, start, tsize, 1);
424 			start += size;
425 		}
426 	} else {
427 		unsigned long rmask = 0xf000000000000000ul;
428 		unsigned long rid = (address & rmask) | 0x1000000000000000ul;
429 		unsigned long vpte = address & ~rmask;
430 
431 		vpte = (vpte >> (PAGE_SHIFT - 3)) & ~0xffful;
432 		vpte |= rid;
433 		__flush_tlb_page(tlb->mm, vpte, tsize, 0);
434 	}
435 }
436 
437 static void setup_page_sizes(void)
438 {
439 	unsigned int tlb0cfg;
440 	unsigned int tlb0ps;
441 	unsigned int eptcfg;
442 	int i, psize;
443 
444 #ifdef CONFIG_PPC_FSL_BOOK3E
445 	unsigned int mmucfg = mfspr(SPRN_MMUCFG);
446 	int fsl_mmu = mmu_has_feature(MMU_FTR_TYPE_FSL_E);
447 
448 	if (fsl_mmu && (mmucfg & MMUCFG_MAVN) == MMUCFG_MAVN_V1) {
449 		unsigned int tlb1cfg = mfspr(SPRN_TLB1CFG);
450 		unsigned int min_pg, max_pg;
451 
452 		min_pg = (tlb1cfg & TLBnCFG_MINSIZE) >> TLBnCFG_MINSIZE_SHIFT;
453 		max_pg = (tlb1cfg & TLBnCFG_MAXSIZE) >> TLBnCFG_MAXSIZE_SHIFT;
454 
455 		for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
456 			struct mmu_psize_def *def;
457 			unsigned int shift;
458 
459 			def = &mmu_psize_defs[psize];
460 			shift = def->shift;
461 
462 			if (shift == 0 || shift & 1)
463 				continue;
464 
465 			/* adjust to be in terms of 4^shift Kb */
466 			shift = (shift - 10) >> 1;
467 
468 			if ((shift >= min_pg) && (shift <= max_pg))
469 				def->flags |= MMU_PAGE_SIZE_DIRECT;
470 		}
471 
472 		goto out;
473 	}
474 
475 	if (fsl_mmu && (mmucfg & MMUCFG_MAVN) == MMUCFG_MAVN_V2) {
476 		u32 tlb1cfg, tlb1ps;
477 
478 		tlb0cfg = mfspr(SPRN_TLB0CFG);
479 		tlb1cfg = mfspr(SPRN_TLB1CFG);
480 		tlb1ps = mfspr(SPRN_TLB1PS);
481 		eptcfg = mfspr(SPRN_EPTCFG);
482 
483 		if ((tlb1cfg & TLBnCFG_IND) && (tlb0cfg & TLBnCFG_PT))
484 			book3e_htw_mode = PPC_HTW_E6500;
485 
486 		/*
487 		 * We expect 4K subpage size and unrestricted indirect size.
488 		 * The lack of a restriction on indirect size is a Freescale
489 		 * extension, indicated by PSn = 0 but SPSn != 0.
490 		 */
491 		if (eptcfg != 2)
492 			book3e_htw_mode = PPC_HTW_NONE;
493 
494 		for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
495 			struct mmu_psize_def *def = &mmu_psize_defs[psize];
496 
497 			if (!def->shift)
498 				continue;
499 
500 			if (tlb1ps & (1U << (def->shift - 10))) {
501 				def->flags |= MMU_PAGE_SIZE_DIRECT;
502 
503 				if (book3e_htw_mode && psize == MMU_PAGE_2M)
504 					def->flags |= MMU_PAGE_SIZE_INDIRECT;
505 			}
506 		}
507 
508 		goto out;
509 	}
510 #endif
511 
512 	tlb0cfg = mfspr(SPRN_TLB0CFG);
513 	tlb0ps = mfspr(SPRN_TLB0PS);
514 	eptcfg = mfspr(SPRN_EPTCFG);
515 
516 	/* Look for supported direct sizes */
517 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
518 		struct mmu_psize_def *def = &mmu_psize_defs[psize];
519 
520 		if (tlb0ps & (1U << (def->shift - 10)))
521 			def->flags |= MMU_PAGE_SIZE_DIRECT;
522 	}
523 
524 	/* Indirect page sizes supported ? */
525 	if ((tlb0cfg & TLBnCFG_IND) == 0 ||
526 	    (tlb0cfg & TLBnCFG_PT) == 0)
527 		goto out;
528 
529 	book3e_htw_mode = PPC_HTW_IBM;
530 
531 	/* Now, we only deal with one IND page size for each
532 	 * direct size. Hopefully all implementations today are
533 	 * unambiguous, but we might want to be careful in the
534 	 * future.
535 	 */
536 	for (i = 0; i < 3; i++) {
537 		unsigned int ps, sps;
538 
539 		sps = eptcfg & 0x1f;
540 		eptcfg >>= 5;
541 		ps = eptcfg & 0x1f;
542 		eptcfg >>= 5;
543 		if (!ps || !sps)
544 			continue;
545 		for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
546 			struct mmu_psize_def *def = &mmu_psize_defs[psize];
547 
548 			if (ps == (def->shift - 10))
549 				def->flags |= MMU_PAGE_SIZE_INDIRECT;
550 			if (sps == (def->shift - 10))
551 				def->ind = ps + 10;
552 		}
553 	}
554 
555 out:
556 	/* Cleanup array and print summary */
557 	pr_info("MMU: Supported page sizes\n");
558 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
559 		struct mmu_psize_def *def = &mmu_psize_defs[psize];
560 		const char *__page_type_names[] = {
561 			"unsupported",
562 			"direct",
563 			"indirect",
564 			"direct & indirect"
565 		};
566 		if (def->flags == 0) {
567 			def->shift = 0;
568 			continue;
569 		}
570 		pr_info("  %8ld KB as %s\n", 1ul << (def->shift - 10),
571 			__page_type_names[def->flags & 0x3]);
572 	}
573 }
574 
575 static void setup_mmu_htw(void)
576 {
577 	/*
578 	 * If we want to use HW tablewalk, enable it by patching the TLB miss
579 	 * handlers to branch to the one dedicated to it.
580 	 */
581 
582 	switch (book3e_htw_mode) {
583 	case PPC_HTW_IBM:
584 		patch_exception(0x1c0, exc_data_tlb_miss_htw_book3e);
585 		patch_exception(0x1e0, exc_instruction_tlb_miss_htw_book3e);
586 		break;
587 #ifdef CONFIG_PPC_FSL_BOOK3E
588 	case PPC_HTW_E6500:
589 		extlb_level_exc = EX_TLB_SIZE;
590 		patch_exception(0x1c0, exc_data_tlb_miss_e6500_book3e);
591 		patch_exception(0x1e0, exc_instruction_tlb_miss_e6500_book3e);
592 		break;
593 #endif
594 	}
595 	pr_info("MMU: Book3E HW tablewalk %s\n",
596 		book3e_htw_mode != PPC_HTW_NONE ? "enabled" : "not supported");
597 }
598 
599 /*
600  * Early initialization of the MMU TLB code
601  */
602 static void early_init_this_mmu(void)
603 {
604 	unsigned int mas4;
605 
606 	/* Set MAS4 based on page table setting */
607 
608 	mas4 = 0x4 << MAS4_WIMGED_SHIFT;
609 	switch (book3e_htw_mode) {
610 	case PPC_HTW_E6500:
611 		mas4 |= MAS4_INDD;
612 		mas4 |= BOOK3E_PAGESZ_2M << MAS4_TSIZED_SHIFT;
613 		mas4 |= MAS4_TLBSELD(1);
614 		mmu_pte_psize = MMU_PAGE_2M;
615 		break;
616 
617 	case PPC_HTW_IBM:
618 		mas4 |= MAS4_INDD;
619 		mas4 |=	BOOK3E_PAGESZ_1M << MAS4_TSIZED_SHIFT;
620 		mmu_pte_psize = MMU_PAGE_1M;
621 		break;
622 
623 	case PPC_HTW_NONE:
624 		mas4 |=	BOOK3E_PAGESZ_4K << MAS4_TSIZED_SHIFT;
625 		mmu_pte_psize = mmu_virtual_psize;
626 		break;
627 	}
628 	mtspr(SPRN_MAS4, mas4);
629 
630 #ifdef CONFIG_PPC_FSL_BOOK3E
631 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
632 		unsigned int num_cams;
633 		bool map = true;
634 
635 		/* use a quarter of the TLBCAM for bolted linear map */
636 		num_cams = (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) / 4;
637 
638 		/*
639 		 * Only do the mapping once per core, or else the
640 		 * transient mapping would cause problems.
641 		 */
642 #ifdef CONFIG_SMP
643 		if (hweight32(get_tensr()) > 1)
644 			map = false;
645 #endif
646 
647 		if (map)
648 			linear_map_top = map_mem_in_cams(linear_map_top,
649 							 num_cams, false);
650 	}
651 #endif
652 
653 	/* A sync won't hurt us after mucking around with
654 	 * the MMU configuration
655 	 */
656 	mb();
657 }
658 
659 static void __init early_init_mmu_global(void)
660 {
661 	/* XXX This will have to be decided at runtime, but right
662 	 * now our boot and TLB miss code hard wires it. Ideally
663 	 * we should find out a suitable page size and patch the
664 	 * TLB miss code (either that or use the PACA to store
665 	 * the value we want)
666 	 */
667 	mmu_linear_psize = MMU_PAGE_1G;
668 
669 	/* XXX This should be decided at runtime based on supported
670 	 * page sizes in the TLB, but for now let's assume 16M is
671 	 * always there and a good fit (which it probably is)
672 	 *
673 	 * Freescale booke only supports 4K pages in TLB0, so use that.
674 	 */
675 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
676 		mmu_vmemmap_psize = MMU_PAGE_4K;
677 	else
678 		mmu_vmemmap_psize = MMU_PAGE_16M;
679 
680 	/* XXX This code only checks for TLB 0 capabilities and doesn't
681 	 *     check what page size combos are supported by the HW. It
682 	 *     also doesn't handle the case where a separate array holds
683 	 *     the IND entries from the array loaded by the PT.
684 	 */
685 	/* Look for supported page sizes */
686 	setup_page_sizes();
687 
688 	/* Look for HW tablewalk support */
689 	setup_mmu_htw();
690 
691 #ifdef CONFIG_PPC_FSL_BOOK3E
692 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
693 		if (book3e_htw_mode == PPC_HTW_NONE) {
694 			extlb_level_exc = EX_TLB_SIZE;
695 			patch_exception(0x1c0, exc_data_tlb_miss_bolted_book3e);
696 			patch_exception(0x1e0,
697 				exc_instruction_tlb_miss_bolted_book3e);
698 		}
699 	}
700 #endif
701 
702 	/* Set the global containing the top of the linear mapping
703 	 * for use by the TLB miss code
704 	 */
705 	linear_map_top = memblock_end_of_DRAM();
706 
707 	ioremap_bot = IOREMAP_BASE;
708 }
709 
710 static void __init early_mmu_set_memory_limit(void)
711 {
712 #ifdef CONFIG_PPC_FSL_BOOK3E
713 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
714 		/*
715 		 * Limit memory so we dont have linear faults.
716 		 * Unlike memblock_set_current_limit, which limits
717 		 * memory available during early boot, this permanently
718 		 * reduces the memory available to Linux.  We need to
719 		 * do this because highmem is not supported on 64-bit.
720 		 */
721 		memblock_enforce_memory_limit(linear_map_top);
722 	}
723 #endif
724 
725 	memblock_set_current_limit(linear_map_top);
726 }
727 
728 /* boot cpu only */
729 void __init early_init_mmu(void)
730 {
731 	early_init_mmu_global();
732 	early_init_this_mmu();
733 	early_mmu_set_memory_limit();
734 }
735 
736 void early_init_mmu_secondary(void)
737 {
738 	early_init_this_mmu();
739 }
740 
741 void setup_initial_memory_limit(phys_addr_t first_memblock_base,
742 				phys_addr_t first_memblock_size)
743 {
744 	/* On non-FSL Embedded 64-bit, we adjust the RMA size to match
745 	 * the bolted TLB entry. We know for now that only 1G
746 	 * entries are supported though that may eventually
747 	 * change.
748 	 *
749 	 * on FSL Embedded 64-bit, usually all RAM is bolted, but with
750 	 * unusual memory sizes it's possible for some RAM to not be mapped
751 	 * (such RAM is not used at all by Linux, since we don't support
752 	 * highmem on 64-bit).  We limit ppc64_rma_size to what would be
753 	 * mappable if this memblock is the only one.  Additional memblocks
754 	 * can only increase, not decrease, the amount that ends up getting
755 	 * mapped.  We still limit max to 1G even if we'll eventually map
756 	 * more.  This is due to what the early init code is set up to do.
757 	 *
758 	 * We crop it to the size of the first MEMBLOCK to
759 	 * avoid going over total available memory just in case...
760 	 */
761 #ifdef CONFIG_PPC_FSL_BOOK3E
762 	if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
763 		unsigned long linear_sz;
764 		unsigned int num_cams;
765 
766 		/* use a quarter of the TLBCAM for bolted linear map */
767 		num_cams = (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) / 4;
768 
769 		linear_sz = map_mem_in_cams(first_memblock_size, num_cams,
770 					    true);
771 
772 		ppc64_rma_size = min_t(u64, linear_sz, 0x40000000);
773 	} else
774 #endif
775 		ppc64_rma_size = min_t(u64, first_memblock_size, 0x40000000);
776 
777 	/* Finally limit subsequent allocations */
778 	memblock_set_current_limit(first_memblock_base + ppc64_rma_size);
779 }
780 #else /* ! CONFIG_PPC64 */
781 void __init early_init_mmu(void)
782 {
783 #ifdef CONFIG_PPC_47x
784 	early_init_mmu_47x();
785 #endif
786 
787 #ifdef CONFIG_PPC_MM_SLICES
788 	mm_ctx_set_slb_addr_limit(&init_mm.context, SLB_ADDR_LIMIT_DEFAULT);
789 #endif
790 }
791 #endif /* CONFIG_PPC64 */
792