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