1 /*
2  * This file contains the routines for TLB flushing.
3  * On machines where the MMU does not use a hash table to store virtual to
4  * physical translations (ie, SW loaded TLBs or Book3E compilant processors,
5  * this does -not- include 603 however which shares the implementation with
6  * hash based processors)
7  *
8  *  -- BenH
9  *
10  * Copyright 2008,2009 Ben Herrenschmidt <benh@kernel.crashing.org>
11  *                     IBM Corp.
12  *
13  *  Derived from arch/ppc/mm/init.c:
14  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
15  *
16  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
17  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
18  *    Copyright (C) 1996 Paul Mackerras
19  *
20  *  Derived from "arch/i386/mm/init.c"
21  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
22  *
23  *  This program is free software; you can redistribute it and/or
24  *  modify it under the terms of the GNU General Public License
25  *  as published by the Free Software Foundation; either version
26  *  2 of the License, or (at your option) any later version.
27  *
28  */
29 
30 #include <linux/kernel.h>
31 #include <linux/export.h>
32 #include <linux/mm.h>
33 #include <linux/init.h>
34 #include <linux/highmem.h>
35 #include <linux/pagemap.h>
36 #include <linux/preempt.h>
37 #include <linux/spinlock.h>
38 #include <linux/memblock.h>
39 #include <linux/of_fdt.h>
40 #include <linux/hugetlb.h>
41 
42 #include <asm/tlbflush.h>
43 #include <asm/tlb.h>
44 #include <asm/code-patching.h>
45 #include <asm/cputhreads.h>
46 #include <asm/hugetlb.h>
47 #include <asm/paca.h>
48 
49 #include "mmu_decl.h"
50 
51 /*
52  * This struct lists the sw-supported page sizes.  The hardawre MMU may support
53  * other sizes not listed here.   The .ind field is only used on MMUs that have
54  * indirect page table entries.
55  */
56 #if defined(CONFIG_PPC_BOOK3E_MMU) || defined(CONFIG_PPC_8xx)
57 #ifdef CONFIG_PPC_FSL_BOOK3E
58 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
59 	[MMU_PAGE_4K] = {
60 		.shift	= 12,
61 		.enc	= BOOK3E_PAGESZ_4K,
62 	},
63 	[MMU_PAGE_2M] = {
64 		.shift	= 21,
65 		.enc	= BOOK3E_PAGESZ_2M,
66 	},
67 	[MMU_PAGE_4M] = {
68 		.shift	= 22,
69 		.enc	= BOOK3E_PAGESZ_4M,
70 	},
71 	[MMU_PAGE_16M] = {
72 		.shift	= 24,
73 		.enc	= BOOK3E_PAGESZ_16M,
74 	},
75 	[MMU_PAGE_64M] = {
76 		.shift	= 26,
77 		.enc	= BOOK3E_PAGESZ_64M,
78 	},
79 	[MMU_PAGE_256M] = {
80 		.shift	= 28,
81 		.enc	= BOOK3E_PAGESZ_256M,
82 	},
83 	[MMU_PAGE_1G] = {
84 		.shift	= 30,
85 		.enc	= BOOK3E_PAGESZ_1GB,
86 	},
87 };
88 #elif defined(CONFIG_PPC_8xx)
89 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
90 	/* we only manage 4k and 16k pages as normal pages */
91 #ifdef CONFIG_PPC_4K_PAGES
92 	[MMU_PAGE_4K] = {
93 		.shift	= 12,
94 	},
95 #else
96 	[MMU_PAGE_16K] = {
97 		.shift	= 14,
98 	},
99 #endif
100 	[MMU_PAGE_512K] = {
101 		.shift	= 19,
102 	},
103 	[MMU_PAGE_8M] = {
104 		.shift	= 23,
105 	},
106 };
107 #else
108 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
109 	[MMU_PAGE_4K] = {
110 		.shift	= 12,
111 		.ind	= 20,
112 		.enc	= BOOK3E_PAGESZ_4K,
113 	},
114 	[MMU_PAGE_16K] = {
115 		.shift	= 14,
116 		.enc	= BOOK3E_PAGESZ_16K,
117 	},
118 	[MMU_PAGE_64K] = {
119 		.shift	= 16,
120 		.ind	= 28,
121 		.enc	= BOOK3E_PAGESZ_64K,
122 	},
123 	[MMU_PAGE_1M] = {
124 		.shift	= 20,
125 		.enc	= BOOK3E_PAGESZ_1M,
126 	},
127 	[MMU_PAGE_16M] = {
128 		.shift	= 24,
129 		.ind	= 36,
130 		.enc	= BOOK3E_PAGESZ_16M,
131 	},
132 	[MMU_PAGE_256M] = {
133 		.shift	= 28,
134 		.enc	= BOOK3E_PAGESZ_256M,
135 	},
136 	[MMU_PAGE_1G] = {
137 		.shift	= 30,
138 		.enc	= BOOK3E_PAGESZ_1GB,
139 	},
140 };
141 #endif /* CONFIG_FSL_BOOKE */
142 
mmu_get_tsize(int psize)143 static inline int mmu_get_tsize(int psize)
144 {
145 	return mmu_psize_defs[psize].enc;
146 }
147 #else
mmu_get_tsize(int psize)148 static inline int mmu_get_tsize(int psize)
149 {
150 	/* This isn't used on !Book3E for now */
151 	return 0;
152 }
153 #endif /* CONFIG_PPC_BOOK3E_MMU */
154 
155 /* The variables below are currently only used on 64-bit Book3E
156  * though this will probably be made common with other nohash
157  * implementations at some point
158  */
159 #ifdef CONFIG_PPC64
160 
161 int mmu_linear_psize;		/* Page size used for the linear mapping */
162 int mmu_pte_psize;		/* Page size used for PTE pages */
163 int mmu_vmemmap_psize;		/* Page size used for the virtual mem map */
164 int book3e_htw_mode;		/* HW tablewalk?  Value is PPC_HTW_* */
165 unsigned long linear_map_top;	/* Top of linear mapping */
166 
167 
168 /*
169  * Number of bytes to add to SPRN_SPRG_TLB_EXFRAME on crit/mcheck/debug
170  * exceptions.  This is used for bolted and e6500 TLB miss handlers which
171  * do not modify this SPRG in the TLB miss code; for other TLB miss handlers,
172  * this is set to zero.
173  */
174 int extlb_level_exc;
175 
176 #endif /* CONFIG_PPC64 */
177 
178 #ifdef CONFIG_PPC_FSL_BOOK3E
179 /* next_tlbcam_idx is used to round-robin tlbcam entry assignment */
180 DEFINE_PER_CPU(int, next_tlbcam_idx);
181 EXPORT_PER_CPU_SYMBOL(next_tlbcam_idx);
182 #endif
183 
184 /*
185  * Base TLB flushing operations:
186  *
187  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
188  *  - flush_tlb_page(vma, vmaddr) flushes one page
189  *  - flush_tlb_range(vma, start, end) flushes a range of pages
190  *  - flush_tlb_kernel_range(start, end) flushes kernel pages
191  *
192  *  - local_* variants of page and mm only apply to the current
193  *    processor
194  */
195 
196 /*
197  * These are the base non-SMP variants of page and mm flushing
198  */
local_flush_tlb_mm(struct mm_struct * mm)199 void local_flush_tlb_mm(struct mm_struct *mm)
200 {
201 	unsigned int pid;
202 
203 	preempt_disable();
204 	pid = mm->context.id;
205 	if (pid != MMU_NO_CONTEXT)
206 		_tlbil_pid(pid);
207 	preempt_enable();
208 }
209 EXPORT_SYMBOL(local_flush_tlb_mm);
210 
__local_flush_tlb_page(struct mm_struct * mm,unsigned long vmaddr,int tsize,int ind)211 void __local_flush_tlb_page(struct mm_struct *mm, unsigned long vmaddr,
212 			    int tsize, int ind)
213 {
214 	unsigned int pid;
215 
216 	preempt_disable();
217 	pid = mm ? mm->context.id : 0;
218 	if (pid != MMU_NO_CONTEXT)
219 		_tlbil_va(vmaddr, pid, tsize, ind);
220 	preempt_enable();
221 }
222 
local_flush_tlb_page(struct vm_area_struct * vma,unsigned long vmaddr)223 void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
224 {
225 	__local_flush_tlb_page(vma ? vma->vm_mm : NULL, vmaddr,
226 			       mmu_get_tsize(mmu_virtual_psize), 0);
227 }
228 EXPORT_SYMBOL(local_flush_tlb_page);
229 
230 /*
231  * And here are the SMP non-local implementations
232  */
233 #ifdef CONFIG_SMP
234 
235 static DEFINE_RAW_SPINLOCK(tlbivax_lock);
236 
237 struct tlb_flush_param {
238 	unsigned long addr;
239 	unsigned int pid;
240 	unsigned int tsize;
241 	unsigned int ind;
242 };
243 
do_flush_tlb_mm_ipi(void * param)244 static void do_flush_tlb_mm_ipi(void *param)
245 {
246 	struct tlb_flush_param *p = param;
247 
248 	_tlbil_pid(p ? p->pid : 0);
249 }
250 
do_flush_tlb_page_ipi(void * param)251 static void do_flush_tlb_page_ipi(void *param)
252 {
253 	struct tlb_flush_param *p = param;
254 
255 	_tlbil_va(p->addr, p->pid, p->tsize, p->ind);
256 }
257 
258 
259 /* Note on invalidations and PID:
260  *
261  * We snapshot the PID with preempt disabled. At this point, it can still
262  * change either because:
263  * - our context is being stolen (PID -> NO_CONTEXT) on another CPU
264  * - we are invaliating some target that isn't currently running here
265  *   and is concurrently acquiring a new PID on another CPU
266  * - some other CPU is re-acquiring a lost PID for this mm
267  * etc...
268  *
269  * However, this shouldn't be a problem as we only guarantee
270  * invalidation of TLB entries present prior to this call, so we
271  * don't care about the PID changing, and invalidating a stale PID
272  * is generally harmless.
273  */
274 
flush_tlb_mm(struct mm_struct * mm)275 void flush_tlb_mm(struct mm_struct *mm)
276 {
277 	unsigned int pid;
278 
279 	preempt_disable();
280 	pid = mm->context.id;
281 	if (unlikely(pid == MMU_NO_CONTEXT))
282 		goto no_context;
283 	if (!mm_is_core_local(mm)) {
284 		struct tlb_flush_param p = { .pid = pid };
285 		/* Ignores smp_processor_id() even if set. */
286 		smp_call_function_many(mm_cpumask(mm),
287 				       do_flush_tlb_mm_ipi, &p, 1);
288 	}
289 	_tlbil_pid(pid);
290  no_context:
291 	preempt_enable();
292 }
293 EXPORT_SYMBOL(flush_tlb_mm);
294 
__flush_tlb_page(struct mm_struct * mm,unsigned long vmaddr,int tsize,int ind)295 void __flush_tlb_page(struct mm_struct *mm, unsigned long vmaddr,
296 		      int tsize, int ind)
297 {
298 	struct cpumask *cpu_mask;
299 	unsigned int pid;
300 
301 	/*
302 	 * This function as well as __local_flush_tlb_page() must only be called
303 	 * for user contexts.
304 	 */
305 	if (unlikely(WARN_ON(!mm)))
306 		return;
307 
308 	preempt_disable();
309 	pid = mm->context.id;
310 	if (unlikely(pid == MMU_NO_CONTEXT))
311 		goto bail;
312 	cpu_mask = mm_cpumask(mm);
313 	if (!mm_is_core_local(mm)) {
314 		/* If broadcast tlbivax is supported, use it */
315 		if (mmu_has_feature(MMU_FTR_USE_TLBIVAX_BCAST)) {
316 			int lock = mmu_has_feature(MMU_FTR_LOCK_BCAST_INVAL);
317 			if (lock)
318 				raw_spin_lock(&tlbivax_lock);
319 			_tlbivax_bcast(vmaddr, pid, tsize, ind);
320 			if (lock)
321 				raw_spin_unlock(&tlbivax_lock);
322 			goto bail;
323 		} else {
324 			struct tlb_flush_param p = {
325 				.pid = pid,
326 				.addr = vmaddr,
327 				.tsize = tsize,
328 				.ind = ind,
329 			};
330 			/* Ignores smp_processor_id() even if set in cpu_mask */
331 			smp_call_function_many(cpu_mask,
332 					       do_flush_tlb_page_ipi, &p, 1);
333 		}
334 	}
335 	_tlbil_va(vmaddr, pid, tsize, ind);
336  bail:
337 	preempt_enable();
338 }
339 
flush_tlb_page(struct vm_area_struct * vma,unsigned long vmaddr)340 void flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
341 {
342 #ifdef CONFIG_HUGETLB_PAGE
343 	if (vma && is_vm_hugetlb_page(vma))
344 		flush_hugetlb_page(vma, vmaddr);
345 #endif
346 
347 	__flush_tlb_page(vma ? vma->vm_mm : NULL, vmaddr,
348 			 mmu_get_tsize(mmu_virtual_psize), 0);
349 }
350 EXPORT_SYMBOL(flush_tlb_page);
351 
352 #endif /* CONFIG_SMP */
353 
354 #ifdef CONFIG_PPC_47x
early_init_mmu_47x(void)355 void __init early_init_mmu_47x(void)
356 {
357 #ifdef CONFIG_SMP
358 	unsigned long root = of_get_flat_dt_root();
359 	if (of_get_flat_dt_prop(root, "cooperative-partition", NULL))
360 		mmu_clear_feature(MMU_FTR_USE_TLBIVAX_BCAST);
361 #endif /* CONFIG_SMP */
362 }
363 #endif /* CONFIG_PPC_47x */
364 
365 /*
366  * Flush kernel TLB entries in the given range
367  */
flush_tlb_kernel_range(unsigned long start,unsigned long end)368 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
369 {
370 #ifdef CONFIG_SMP
371 	preempt_disable();
372 	smp_call_function(do_flush_tlb_mm_ipi, NULL, 1);
373 	_tlbil_pid(0);
374 	preempt_enable();
375 #else
376 	_tlbil_pid(0);
377 #endif
378 }
379 EXPORT_SYMBOL(flush_tlb_kernel_range);
380 
381 /*
382  * Currently, for range flushing, we just do a full mm flush. This should
383  * be optimized based on a threshold on the size of the range, since
384  * some implementation can stack multiple tlbivax before a tlbsync but
385  * for now, we keep it that way
386  */
flush_tlb_range(struct vm_area_struct * vma,unsigned long start,unsigned long end)387 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
388 		     unsigned long end)
389 
390 {
391 	if (end - start == PAGE_SIZE && !(start & ~PAGE_MASK))
392 		flush_tlb_page(vma, start);
393 	else
394 		flush_tlb_mm(vma->vm_mm);
395 }
396 EXPORT_SYMBOL(flush_tlb_range);
397 
tlb_flush(struct mmu_gather * tlb)398 void tlb_flush(struct mmu_gather *tlb)
399 {
400 	flush_tlb_mm(tlb->mm);
401 }
402 
403 /*
404  * Below are functions specific to the 64-bit variant of Book3E though that
405  * may change in the future
406  */
407 
408 #ifdef CONFIG_PPC64
409 
410 /*
411  * Handling of virtual linear page tables or indirect TLB entries
412  * flushing when PTE pages are freed
413  */
tlb_flush_pgtable(struct mmu_gather * tlb,unsigned long address)414 void tlb_flush_pgtable(struct mmu_gather *tlb, unsigned long address)
415 {
416 	int tsize = mmu_psize_defs[mmu_pte_psize].enc;
417 
418 	if (book3e_htw_mode != PPC_HTW_NONE) {
419 		unsigned long start = address & PMD_MASK;
420 		unsigned long end = address + PMD_SIZE;
421 		unsigned long size = 1UL << mmu_psize_defs[mmu_pte_psize].shift;
422 
423 		/* This isn't the most optimal, ideally we would factor out the
424 		 * while preempt & CPU mask mucking around, or even the IPI but
425 		 * it will do for now
426 		 */
427 		while (start < end) {
428 			__flush_tlb_page(tlb->mm, start, tsize, 1);
429 			start += size;
430 		}
431 	} else {
432 		unsigned long rmask = 0xf000000000000000ul;
433 		unsigned long rid = (address & rmask) | 0x1000000000000000ul;
434 		unsigned long vpte = address & ~rmask;
435 
436 #ifdef CONFIG_PPC_64K_PAGES
437 		vpte = (vpte >> (PAGE_SHIFT - 4)) & ~0xfffful;
438 #else
439 		vpte = (vpte >> (PAGE_SHIFT - 3)) & ~0xffful;
440 #endif
441 		vpte |= rid;
442 		__flush_tlb_page(tlb->mm, vpte, tsize, 0);
443 	}
444 }
445 
setup_page_sizes(void)446 static void setup_page_sizes(void)
447 {
448 	unsigned int tlb0cfg;
449 	unsigned int tlb0ps;
450 	unsigned int eptcfg;
451 	int i, psize;
452 
453 #ifdef CONFIG_PPC_FSL_BOOK3E
454 	unsigned int mmucfg = mfspr(SPRN_MMUCFG);
455 	int fsl_mmu = mmu_has_feature(MMU_FTR_TYPE_FSL_E);
456 
457 	if (fsl_mmu && (mmucfg & MMUCFG_MAVN) == MMUCFG_MAVN_V1) {
458 		unsigned int tlb1cfg = mfspr(SPRN_TLB1CFG);
459 		unsigned int min_pg, max_pg;
460 
461 		min_pg = (tlb1cfg & TLBnCFG_MINSIZE) >> TLBnCFG_MINSIZE_SHIFT;
462 		max_pg = (tlb1cfg & TLBnCFG_MAXSIZE) >> TLBnCFG_MAXSIZE_SHIFT;
463 
464 		for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
465 			struct mmu_psize_def *def;
466 			unsigned int shift;
467 
468 			def = &mmu_psize_defs[psize];
469 			shift = def->shift;
470 
471 			if (shift == 0 || shift & 1)
472 				continue;
473 
474 			/* adjust to be in terms of 4^shift Kb */
475 			shift = (shift - 10) >> 1;
476 
477 			if ((shift >= min_pg) && (shift <= max_pg))
478 				def->flags |= MMU_PAGE_SIZE_DIRECT;
479 		}
480 
481 		goto out;
482 	}
483 
484 	if (fsl_mmu && (mmucfg & MMUCFG_MAVN) == MMUCFG_MAVN_V2) {
485 		u32 tlb1cfg, tlb1ps;
486 
487 		tlb0cfg = mfspr(SPRN_TLB0CFG);
488 		tlb1cfg = mfspr(SPRN_TLB1CFG);
489 		tlb1ps = mfspr(SPRN_TLB1PS);
490 		eptcfg = mfspr(SPRN_EPTCFG);
491 
492 		if ((tlb1cfg & TLBnCFG_IND) && (tlb0cfg & TLBnCFG_PT))
493 			book3e_htw_mode = PPC_HTW_E6500;
494 
495 		/*
496 		 * We expect 4K subpage size and unrestricted indirect size.
497 		 * The lack of a restriction on indirect size is a Freescale
498 		 * extension, indicated by PSn = 0 but SPSn != 0.
499 		 */
500 		if (eptcfg != 2)
501 			book3e_htw_mode = PPC_HTW_NONE;
502 
503 		for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
504 			struct mmu_psize_def *def = &mmu_psize_defs[psize];
505 
506 			if (tlb1ps & (1U << (def->shift - 10))) {
507 				def->flags |= MMU_PAGE_SIZE_DIRECT;
508 
509 				if (book3e_htw_mode && psize == MMU_PAGE_2M)
510 					def->flags |= MMU_PAGE_SIZE_INDIRECT;
511 			}
512 		}
513 
514 		goto out;
515 	}
516 #endif
517 
518 	tlb0cfg = mfspr(SPRN_TLB0CFG);
519 	tlb0ps = mfspr(SPRN_TLB0PS);
520 	eptcfg = mfspr(SPRN_EPTCFG);
521 
522 	/* Look for supported direct sizes */
523 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
524 		struct mmu_psize_def *def = &mmu_psize_defs[psize];
525 
526 		if (tlb0ps & (1U << (def->shift - 10)))
527 			def->flags |= MMU_PAGE_SIZE_DIRECT;
528 	}
529 
530 	/* Indirect page sizes supported ? */
531 	if ((tlb0cfg & TLBnCFG_IND) == 0 ||
532 	    (tlb0cfg & TLBnCFG_PT) == 0)
533 		goto out;
534 
535 	book3e_htw_mode = PPC_HTW_IBM;
536 
537 	/* Now, we only deal with one IND page size for each
538 	 * direct size. Hopefully all implementations today are
539 	 * unambiguous, but we might want to be careful in the
540 	 * future.
541 	 */
542 	for (i = 0; i < 3; i++) {
543 		unsigned int ps, sps;
544 
545 		sps = eptcfg & 0x1f;
546 		eptcfg >>= 5;
547 		ps = eptcfg & 0x1f;
548 		eptcfg >>= 5;
549 		if (!ps || !sps)
550 			continue;
551 		for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
552 			struct mmu_psize_def *def = &mmu_psize_defs[psize];
553 
554 			if (ps == (def->shift - 10))
555 				def->flags |= MMU_PAGE_SIZE_INDIRECT;
556 			if (sps == (def->shift - 10))
557 				def->ind = ps + 10;
558 		}
559 	}
560 
561 out:
562 	/* Cleanup array and print summary */
563 	pr_info("MMU: Supported page sizes\n");
564 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
565 		struct mmu_psize_def *def = &mmu_psize_defs[psize];
566 		const char *__page_type_names[] = {
567 			"unsupported",
568 			"direct",
569 			"indirect",
570 			"direct & indirect"
571 		};
572 		if (def->flags == 0) {
573 			def->shift = 0;
574 			continue;
575 		}
576 		pr_info("  %8ld KB as %s\n", 1ul << (def->shift - 10),
577 			__page_type_names[def->flags & 0x3]);
578 	}
579 }
580 
setup_mmu_htw(void)581 static void setup_mmu_htw(void)
582 {
583 	/*
584 	 * If we want to use HW tablewalk, enable it by patching the TLB miss
585 	 * handlers to branch to the one dedicated to it.
586 	 */
587 
588 	switch (book3e_htw_mode) {
589 	case PPC_HTW_IBM:
590 		patch_exception(0x1c0, exc_data_tlb_miss_htw_book3e);
591 		patch_exception(0x1e0, exc_instruction_tlb_miss_htw_book3e);
592 		break;
593 #ifdef CONFIG_PPC_FSL_BOOK3E
594 	case PPC_HTW_E6500:
595 		extlb_level_exc = EX_TLB_SIZE;
596 		patch_exception(0x1c0, exc_data_tlb_miss_e6500_book3e);
597 		patch_exception(0x1e0, exc_instruction_tlb_miss_e6500_book3e);
598 		break;
599 #endif
600 	}
601 	pr_info("MMU: Book3E HW tablewalk %s\n",
602 		book3e_htw_mode != PPC_HTW_NONE ? "enabled" : "not supported");
603 }
604 
605 /*
606  * Early initialization of the MMU TLB code
607  */
early_init_this_mmu(void)608 static void early_init_this_mmu(void)
609 {
610 	unsigned int mas4;
611 
612 	/* Set MAS4 based on page table setting */
613 
614 	mas4 = 0x4 << MAS4_WIMGED_SHIFT;
615 	switch (book3e_htw_mode) {
616 	case PPC_HTW_E6500:
617 		mas4 |= MAS4_INDD;
618 		mas4 |= BOOK3E_PAGESZ_2M << MAS4_TSIZED_SHIFT;
619 		mas4 |= MAS4_TLBSELD(1);
620 		mmu_pte_psize = MMU_PAGE_2M;
621 		break;
622 
623 	case PPC_HTW_IBM:
624 		mas4 |= MAS4_INDD;
625 #ifdef CONFIG_PPC_64K_PAGES
626 		mas4 |=	BOOK3E_PAGESZ_256M << MAS4_TSIZED_SHIFT;
627 		mmu_pte_psize = MMU_PAGE_256M;
628 #else
629 		mas4 |=	BOOK3E_PAGESZ_1M << MAS4_TSIZED_SHIFT;
630 		mmu_pte_psize = MMU_PAGE_1M;
631 #endif
632 		break;
633 
634 	case PPC_HTW_NONE:
635 #ifdef CONFIG_PPC_64K_PAGES
636 		mas4 |=	BOOK3E_PAGESZ_64K << MAS4_TSIZED_SHIFT;
637 #else
638 		mas4 |=	BOOK3E_PAGESZ_4K << MAS4_TSIZED_SHIFT;
639 #endif
640 		mmu_pte_psize = mmu_virtual_psize;
641 		break;
642 	}
643 	mtspr(SPRN_MAS4, mas4);
644 
645 #ifdef CONFIG_PPC_FSL_BOOK3E
646 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
647 		unsigned int num_cams;
648 		int __maybe_unused cpu = smp_processor_id();
649 		bool map = true;
650 
651 		/* use a quarter of the TLBCAM for bolted linear map */
652 		num_cams = (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) / 4;
653 
654 		/*
655 		 * Only do the mapping once per core, or else the
656 		 * transient mapping would cause problems.
657 		 */
658 #ifdef CONFIG_SMP
659 		if (hweight32(get_tensr()) > 1)
660 			map = false;
661 #endif
662 
663 		if (map)
664 			linear_map_top = map_mem_in_cams(linear_map_top,
665 							 num_cams, false);
666 	}
667 #endif
668 
669 	/* A sync won't hurt us after mucking around with
670 	 * the MMU configuration
671 	 */
672 	mb();
673 }
674 
early_init_mmu_global(void)675 static void __init early_init_mmu_global(void)
676 {
677 	/* XXX This will have to be decided at runtime, but right
678 	 * now our boot and TLB miss code hard wires it. Ideally
679 	 * we should find out a suitable page size and patch the
680 	 * TLB miss code (either that or use the PACA to store
681 	 * the value we want)
682 	 */
683 	mmu_linear_psize = MMU_PAGE_1G;
684 
685 	/* XXX This should be decided at runtime based on supported
686 	 * page sizes in the TLB, but for now let's assume 16M is
687 	 * always there and a good fit (which it probably is)
688 	 *
689 	 * Freescale booke only supports 4K pages in TLB0, so use that.
690 	 */
691 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
692 		mmu_vmemmap_psize = MMU_PAGE_4K;
693 	else
694 		mmu_vmemmap_psize = MMU_PAGE_16M;
695 
696 	/* XXX This code only checks for TLB 0 capabilities and doesn't
697 	 *     check what page size combos are supported by the HW. It
698 	 *     also doesn't handle the case where a separate array holds
699 	 *     the IND entries from the array loaded by the PT.
700 	 */
701 	/* Look for supported page sizes */
702 	setup_page_sizes();
703 
704 	/* Look for HW tablewalk support */
705 	setup_mmu_htw();
706 
707 #ifdef CONFIG_PPC_FSL_BOOK3E
708 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
709 		if (book3e_htw_mode == PPC_HTW_NONE) {
710 			extlb_level_exc = EX_TLB_SIZE;
711 			patch_exception(0x1c0, exc_data_tlb_miss_bolted_book3e);
712 			patch_exception(0x1e0,
713 				exc_instruction_tlb_miss_bolted_book3e);
714 		}
715 	}
716 #endif
717 
718 	/* Set the global containing the top of the linear mapping
719 	 * for use by the TLB miss code
720 	 */
721 	linear_map_top = memblock_end_of_DRAM();
722 }
723 
early_mmu_set_memory_limit(void)724 static void __init early_mmu_set_memory_limit(void)
725 {
726 #ifdef CONFIG_PPC_FSL_BOOK3E
727 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
728 		/*
729 		 * Limit memory so we dont have linear faults.
730 		 * Unlike memblock_set_current_limit, which limits
731 		 * memory available during early boot, this permanently
732 		 * reduces the memory available to Linux.  We need to
733 		 * do this because highmem is not supported on 64-bit.
734 		 */
735 		memblock_enforce_memory_limit(linear_map_top);
736 	}
737 #endif
738 
739 	memblock_set_current_limit(linear_map_top);
740 }
741 
742 /* boot cpu only */
early_init_mmu(void)743 void __init early_init_mmu(void)
744 {
745 	early_init_mmu_global();
746 	early_init_this_mmu();
747 	early_mmu_set_memory_limit();
748 }
749 
early_init_mmu_secondary(void)750 void early_init_mmu_secondary(void)
751 {
752 	early_init_this_mmu();
753 }
754 
setup_initial_memory_limit(phys_addr_t first_memblock_base,phys_addr_t first_memblock_size)755 void setup_initial_memory_limit(phys_addr_t first_memblock_base,
756 				phys_addr_t first_memblock_size)
757 {
758 	/* On non-FSL Embedded 64-bit, we adjust the RMA size to match
759 	 * the bolted TLB entry. We know for now that only 1G
760 	 * entries are supported though that may eventually
761 	 * change.
762 	 *
763 	 * on FSL Embedded 64-bit, usually all RAM is bolted, but with
764 	 * unusual memory sizes it's possible for some RAM to not be mapped
765 	 * (such RAM is not used at all by Linux, since we don't support
766 	 * highmem on 64-bit).  We limit ppc64_rma_size to what would be
767 	 * mappable if this memblock is the only one.  Additional memblocks
768 	 * can only increase, not decrease, the amount that ends up getting
769 	 * mapped.  We still limit max to 1G even if we'll eventually map
770 	 * more.  This is due to what the early init code is set up to do.
771 	 *
772 	 * We crop it to the size of the first MEMBLOCK to
773 	 * avoid going over total available memory just in case...
774 	 */
775 #ifdef CONFIG_PPC_FSL_BOOK3E
776 	if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
777 		unsigned long linear_sz;
778 		unsigned int num_cams;
779 
780 		/* use a quarter of the TLBCAM for bolted linear map */
781 		num_cams = (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) / 4;
782 
783 		linear_sz = map_mem_in_cams(first_memblock_size, num_cams,
784 					    true);
785 
786 		ppc64_rma_size = min_t(u64, linear_sz, 0x40000000);
787 	} else
788 #endif
789 		ppc64_rma_size = min_t(u64, first_memblock_size, 0x40000000);
790 
791 	/* Finally limit subsequent allocations */
792 	memblock_set_current_limit(first_memblock_base + ppc64_rma_size);
793 }
794 #else /* ! CONFIG_PPC64 */
early_init_mmu(void)795 void __init early_init_mmu(void)
796 {
797 #ifdef CONFIG_PPC_47x
798 	early_init_mmu_47x();
799 #endif
800 }
801 #endif /* CONFIG_PPC64 */
802