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