blob: 647bf454a0fa5b4c8e6057fb0f03b9fab92d9e8c [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +00002/*
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 *
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +000011 * Copyright 2008,2009 Ben Herrenschmidt <benh@kernel.crashing.org>
12 * IBM Corp.
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +000013 *
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
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +000023 */
24
25#include <linux/kernel.h>
Paul Gortmaker93087942011-07-29 16:19:31 +100026#include <linux/export.h>
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +000027#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>
Yinghai Lu95f72d12010-07-12 14:36:09 +100033#include <linux/memblock.h>
Dave Kleikamp91b191c2011-07-04 18:38:03 +000034#include <linux/of_fdt.h>
Becky Bruce41151e72011-06-28 09:54:48 +000035#include <linux/hugetlb.h>
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +000036
Mike Rapoportca15ca42020-08-06 23:22:28 -070037#include <asm/pgalloc.h>
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +000038#include <asm/tlbflush.h>
39#include <asm/tlb.h>
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +000040#include <asm/code-patching.h>
Scott Woodd9e18312015-10-06 22:48:09 -050041#include <asm/cputhreads.h>
Becky Bruce41151e72011-06-28 09:54:48 +000042#include <asm/hugetlb.h>
Scott Wood28efc352013-10-11 19:22:38 -050043#include <asm/paca.h>
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +000044
Christophe Leroy9d9f2cc2019-03-29 09:59:59 +000045#include <mm/mmu_decl.h>
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +000046
Becky Bruce41151e72011-06-28 09:54:48 +000047/*
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 */
Christophe Leroy4b9142862016-12-07 08:47:28 +010052#if defined(CONFIG_PPC_BOOK3E_MMU) || defined(CONFIG_PPC_8xx)
Becky Bruce881fde12011-10-10 10:50:40 +000053#ifdef CONFIG_PPC_FSL_BOOK3E
Becky Bruce41151e72011-06-28 09:54:48 +000054struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
55 [MMU_PAGE_4K] = {
56 .shift = 12,
57 .enc = BOOK3E_PAGESZ_4K,
58 },
Scott Wood28efc352013-10-11 19:22:38 -050059 [MMU_PAGE_2M] = {
60 .shift = 21,
61 .enc = BOOK3E_PAGESZ_2M,
62 },
Becky Bruce41151e72011-06-28 09:54:48 +000063 [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};
Christophe Leroy4b9142862016-12-07 08:47:28 +010084#elif defined(CONFIG_PPC_8xx)
85struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
Christophe Leroy4b9142862016-12-07 08:47:28 +010086 [MMU_PAGE_4K] = {
87 .shift = 12,
88 },
Christophe Leroy4b9142862016-12-07 08:47:28 +010089 [MMU_PAGE_16K] = {
90 .shift = 14,
91 },
Christophe Leroy4b9142862016-12-07 08:47:28 +010092 [MMU_PAGE_512K] = {
93 .shift = 19,
94 },
95 [MMU_PAGE_8M] = {
96 .shift = 23,
97 },
98};
Becky Bruce41151e72011-06-28 09:54:48 +000099#else
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000100struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT] = {
101 [MMU_PAGE_4K] = {
102 .shift = 12,
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000103 .ind = 20,
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000104 .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,
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000112 .ind = 28,
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000113 .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,
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000121 .ind = 36,
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000122 .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};
Becky Bruce41151e72011-06-28 09:54:48 +0000133#endif /* CONFIG_FSL_BOOKE */
134
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000135static inline int mmu_get_tsize(int psize)
136{
137 return mmu_psize_defs[psize].enc;
138}
139#else
140static inline int mmu_get_tsize(int psize)
141{
142 /* This isn't used on !Book3E for now */
143 return 0;
144}
Becky Bruce41151e72011-06-28 09:54:48 +0000145#endif /* CONFIG_PPC_BOOK3E_MMU */
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000146
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
153int mmu_linear_psize; /* Page size used for the linear mapping */
154int mmu_pte_psize; /* Page size used for PTE pages */
Benjamin Herrenschmidt32a74942009-07-23 23:15:58 +0000155int mmu_vmemmap_psize; /* Page size used for the virtual mem map */
Scott Wood28efc352013-10-11 19:22:38 -0500156int book3e_htw_mode; /* HW tablewalk? Value is PPC_HTW_* */
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000157unsigned long linear_map_top; /* Top of linear mapping */
158
Scott Wood609af382014-03-10 17:29:38 -0500159
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 */
166int extlb_level_exc;
167
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000168#endif /* CONFIG_PPC64 */
169
Becky Bruce3160b092011-06-28 14:54:47 -0500170#ifdef CONFIG_PPC_FSL_BOOK3E
171/* next_tlbcam_idx is used to round-robin tlbcam entry assignment */
172DEFINE_PER_CPU(int, next_tlbcam_idx);
173EXPORT_PER_CPU_SYMBOL(next_tlbcam_idx);
174#endif
175
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000176/*
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
Christophe Leroy63f501e2021-09-17 15:57:12 +0200188#ifndef CONFIG_PPC_8xx
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000189/*
190 * These are the base non-SMP variants of page and mm flushing
191 */
192void local_flush_tlb_mm(struct mm_struct *mm)
193{
194 unsigned int pid;
195
196 preempt_disable();
197 pid = mm->context.id;
198 if (pid != MMU_NO_CONTEXT)
199 _tlbil_pid(pid);
200 preempt_enable();
201}
202EXPORT_SYMBOL(local_flush_tlb_mm);
203
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000204void __local_flush_tlb_page(struct mm_struct *mm, unsigned long vmaddr,
205 int tsize, int ind)
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000206{
207 unsigned int pid;
208
209 preempt_disable();
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000210 pid = mm ? mm->context.id : 0;
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000211 if (pid != MMU_NO_CONTEXT)
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000212 _tlbil_va(vmaddr, pid, tsize, ind);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000213 preempt_enable();
214}
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000215
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000216void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
217{
218 __local_flush_tlb_page(vma ? vma->vm_mm : NULL, vmaddr,
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000219 mmu_get_tsize(mmu_virtual_psize), 0);
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000220}
221EXPORT_SYMBOL(local_flush_tlb_page);
Christophe Leroy63f501e2021-09-17 15:57:12 +0200222#endif
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000223
224/*
225 * And here are the SMP non-local implementations
226 */
227#ifdef CONFIG_SMP
228
Thomas Gleixner3eb93c52010-02-18 02:22:44 +0000229static DEFINE_RAW_SPINLOCK(tlbivax_lock);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000230
231struct tlb_flush_param {
232 unsigned long addr;
233 unsigned int pid;
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000234 unsigned int tsize;
235 unsigned int ind;
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000236};
237
238static void do_flush_tlb_mm_ipi(void *param)
239{
240 struct tlb_flush_param *p = param;
241
242 _tlbil_pid(p ? p->pid : 0);
243}
244
245static void do_flush_tlb_page_ipi(void *param)
246{
247 struct tlb_flush_param *p = param;
248
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000249 _tlbil_va(p->addr, p->pid, p->tsize, p->ind);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000250}
251
252
253/* Note on invalidations and PID:
254 *
255 * We snapshot the PID with preempt disabled. At this point, it can still
256 * change either because:
257 * - our context is being stolen (PID -> NO_CONTEXT) on another CPU
258 * - we are invaliating some target that isn't currently running here
259 * and is concurrently acquiring a new PID on another CPU
260 * - some other CPU is re-acquiring a lost PID for this mm
261 * etc...
262 *
263 * However, this shouldn't be a problem as we only guarantee
264 * invalidation of TLB entries present prior to this call, so we
265 * don't care about the PID changing, and invalidating a stale PID
266 * is generally harmless.
267 */
268
269void flush_tlb_mm(struct mm_struct *mm)
270{
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000271 unsigned int pid;
272
273 preempt_disable();
274 pid = mm->context.id;
275 if (unlikely(pid == MMU_NO_CONTEXT))
276 goto no_context;
Benjamin Herrenschmidtfcce8102009-07-23 23:15:10 +0000277 if (!mm_is_core_local(mm)) {
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000278 struct tlb_flush_param p = { .pid = pid };
Rusty Russell56aa4122009-03-15 18:16:43 +0000279 /* Ignores smp_processor_id() even if set. */
280 smp_call_function_many(mm_cpumask(mm),
281 do_flush_tlb_mm_ipi, &p, 1);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000282 }
283 _tlbil_pid(pid);
284 no_context:
285 preempt_enable();
286}
287EXPORT_SYMBOL(flush_tlb_mm);
288
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000289void __flush_tlb_page(struct mm_struct *mm, unsigned long vmaddr,
290 int tsize, int ind)
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000291{
Rusty Russell56aa4122009-03-15 18:16:43 +0000292 struct cpumask *cpu_mask;
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000293 unsigned int pid;
294
Arseny Solokhac2c896b2015-02-04 13:18:02 +1100295 /*
296 * This function as well as __local_flush_tlb_page() must only be called
297 * for user contexts.
298 */
Igor Stoppa63da6ca2018-09-07 18:35:26 +0300299 if (WARN_ON(!mm))
Arseny Solokha0dc294f2015-01-30 19:08:27 +0700300 return;
301
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000302 preempt_disable();
Arseny Solokha0dc294f2015-01-30 19:08:27 +0700303 pid = mm->context.id;
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000304 if (unlikely(pid == MMU_NO_CONTEXT))
305 goto bail;
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000306 cpu_mask = mm_cpumask(mm);
Benjamin Herrenschmidtfcce8102009-07-23 23:15:10 +0000307 if (!mm_is_core_local(mm)) {
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000308 /* If broadcast tlbivax is supported, use it */
309 if (mmu_has_feature(MMU_FTR_USE_TLBIVAX_BCAST)) {
310 int lock = mmu_has_feature(MMU_FTR_LOCK_BCAST_INVAL);
311 if (lock)
Thomas Gleixner3eb93c52010-02-18 02:22:44 +0000312 raw_spin_lock(&tlbivax_lock);
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000313 _tlbivax_bcast(vmaddr, pid, tsize, ind);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000314 if (lock)
Thomas Gleixner3eb93c52010-02-18 02:22:44 +0000315 raw_spin_unlock(&tlbivax_lock);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000316 goto bail;
317 } else {
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000318 struct tlb_flush_param p = {
319 .pid = pid,
320 .addr = vmaddr,
321 .tsize = tsize,
322 .ind = ind,
323 };
Rusty Russell56aa4122009-03-15 18:16:43 +0000324 /* Ignores smp_processor_id() even if set in cpu_mask */
325 smp_call_function_many(cpu_mask,
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000326 do_flush_tlb_page_ipi, &p, 1);
327 }
328 }
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000329 _tlbil_va(vmaddr, pid, tsize, ind);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000330 bail:
331 preempt_enable();
332}
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000333
334void flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
335{
Becky Bruce41151e72011-06-28 09:54:48 +0000336#ifdef CONFIG_HUGETLB_PAGE
Scott Woodd742aa12013-11-21 18:26:42 -0600337 if (vma && is_vm_hugetlb_page(vma))
Becky Bruce41151e72011-06-28 09:54:48 +0000338 flush_hugetlb_page(vma, vmaddr);
339#endif
340
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000341 __flush_tlb_page(vma ? vma->vm_mm : NULL, vmaddr,
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000342 mmu_get_tsize(mmu_virtual_psize), 0);
Benjamin Herrenschmidtd4e167d2009-07-23 23:15:24 +0000343}
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000344EXPORT_SYMBOL(flush_tlb_page);
345
346#endif /* CONFIG_SMP */
347
Dave Kleikamp91b191c2011-07-04 18:38:03 +0000348#ifdef CONFIG_PPC_47x
349void __init early_init_mmu_47x(void)
350{
351#ifdef CONFIG_SMP
352 unsigned long root = of_get_flat_dt_root();
353 if (of_get_flat_dt_prop(root, "cooperative-partition", NULL))
354 mmu_clear_feature(MMU_FTR_USE_TLBIVAX_BCAST);
355#endif /* CONFIG_SMP */
356}
357#endif /* CONFIG_PPC_47x */
358
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000359/*
360 * Flush kernel TLB entries in the given range
361 */
362void flush_tlb_kernel_range(unsigned long start, unsigned long end)
363{
364#ifdef CONFIG_SMP
365 preempt_disable();
366 smp_call_function(do_flush_tlb_mm_ipi, NULL, 1);
367 _tlbil_pid(0);
368 preempt_enable();
Dave Liud6a09e02008-12-30 23:42:55 +0000369#else
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000370 _tlbil_pid(0);
Dave Liud6a09e02008-12-30 23:42:55 +0000371#endif
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000372}
373EXPORT_SYMBOL(flush_tlb_kernel_range);
374
375/*
376 * Currently, for range flushing, we just do a full mm flush. This should
377 * be optimized based on a threshold on the size of the range, since
378 * some implementation can stack multiple tlbivax before a tlbsync but
379 * for now, we keep it that way
380 */
381void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
382 unsigned long end)
383
384{
Christophe Leroy5c8136f2018-01-23 14:22:50 +0100385 if (end - start == PAGE_SIZE && !(start & ~PAGE_MASK))
386 flush_tlb_page(vma, start);
387 else
388 flush_tlb_mm(vma->vm_mm);
Benjamin Herrenschmidtf048aac2008-12-18 19:13:38 +0000389}
390EXPORT_SYMBOL(flush_tlb_range);
Benjamin Herrenschmidtc7cc58a12009-07-23 23:15:28 +0000391
392void tlb_flush(struct mmu_gather *tlb)
393{
394 flush_tlb_mm(tlb->mm);
Benjamin Herrenschmidtc7cc58a12009-07-23 23:15:28 +0000395}
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000396
397/*
398 * Below are functions specific to the 64-bit variant of Book3E though that
399 * may change in the future
400 */
401
402#ifdef CONFIG_PPC64
403
404/*
405 * Handling of virtual linear page tables or indirect TLB entries
406 * flushing when PTE pages are freed
407 */
408void tlb_flush_pgtable(struct mmu_gather *tlb, unsigned long address)
409{
410 int tsize = mmu_psize_defs[mmu_pte_psize].enc;
411
Scott Wood28efc352013-10-11 19:22:38 -0500412 if (book3e_htw_mode != PPC_HTW_NONE) {
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000413 unsigned long start = address & PMD_MASK;
414 unsigned long end = address + PMD_SIZE;
415 unsigned long size = 1UL << mmu_psize_defs[mmu_pte_psize].shift;
416
417 /* This isn't the most optimal, ideally we would factor out the
418 * while preempt & CPU mask mucking around, or even the IPI but
419 * it will do for now
420 */
421 while (start < end) {
422 __flush_tlb_page(tlb->mm, start, tsize, 1);
423 start += size;
424 }
425 } else {
426 unsigned long rmask = 0xf000000000000000ul;
427 unsigned long rid = (address & rmask) | 0x1000000000000000ul;
428 unsigned long vpte = address & ~rmask;
429
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000430 vpte = (vpte >> (PAGE_SHIFT - 3)) & ~0xffful;
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000431 vpte |= rid;
432 __flush_tlb_page(tlb->mm, vpte, tsize, 0);
433 }
434}
435
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000436static void setup_page_sizes(void)
437{
Kumar Gala988cf862010-10-08 02:13:25 -0500438 unsigned int tlb0cfg;
439 unsigned int tlb0ps;
440 unsigned int eptcfg;
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000441 int i, psize;
442
Kumar Gala988cf862010-10-08 02:13:25 -0500443#ifdef CONFIG_PPC_FSL_BOOK3E
444 unsigned int mmucfg = mfspr(SPRN_MMUCFG);
Kumar Gala1b291872013-03-05 12:08:32 -0600445 int fsl_mmu = mmu_has_feature(MMU_FTR_TYPE_FSL_E);
Kumar Gala988cf862010-10-08 02:13:25 -0500446
Kumar Gala1b291872013-03-05 12:08:32 -0600447 if (fsl_mmu && (mmucfg & MMUCFG_MAVN) == MMUCFG_MAVN_V1) {
Kumar Gala988cf862010-10-08 02:13:25 -0500448 unsigned int tlb1cfg = mfspr(SPRN_TLB1CFG);
449 unsigned int min_pg, max_pg;
450
451 min_pg = (tlb1cfg & TLBnCFG_MINSIZE) >> TLBnCFG_MINSIZE_SHIFT;
452 max_pg = (tlb1cfg & TLBnCFG_MAXSIZE) >> TLBnCFG_MAXSIZE_SHIFT;
453
454 for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
455 struct mmu_psize_def *def;
456 unsigned int shift;
457
458 def = &mmu_psize_defs[psize];
459 shift = def->shift;
460
Scott Wood28efc352013-10-11 19:22:38 -0500461 if (shift == 0 || shift & 1)
Kumar Gala988cf862010-10-08 02:13:25 -0500462 continue;
463
464 /* adjust to be in terms of 4^shift Kb */
465 shift = (shift - 10) >> 1;
466
467 if ((shift >= min_pg) && (shift <= max_pg))
468 def->flags |= MMU_PAGE_SIZE_DIRECT;
469 }
470
Scott Wood28efc352013-10-11 19:22:38 -0500471 goto out;
Kumar Gala988cf862010-10-08 02:13:25 -0500472 }
Kumar Gala1b291872013-03-05 12:08:32 -0600473
474 if (fsl_mmu && (mmucfg & MMUCFG_MAVN) == MMUCFG_MAVN_V2) {
Scott Wood28efc352013-10-11 19:22:38 -0500475 u32 tlb1cfg, tlb1ps;
476
477 tlb0cfg = mfspr(SPRN_TLB0CFG);
478 tlb1cfg = mfspr(SPRN_TLB1CFG);
479 tlb1ps = mfspr(SPRN_TLB1PS);
480 eptcfg = mfspr(SPRN_EPTCFG);
481
482 if ((tlb1cfg & TLBnCFG_IND) && (tlb0cfg & TLBnCFG_PT))
483 book3e_htw_mode = PPC_HTW_E6500;
484
485 /*
486 * We expect 4K subpage size and unrestricted indirect size.
487 * The lack of a restriction on indirect size is a Freescale
488 * extension, indicated by PSn = 0 but SPSn != 0.
489 */
490 if (eptcfg != 2)
491 book3e_htw_mode = PPC_HTW_NONE;
Kumar Gala1b291872013-03-05 12:08:32 -0600492
493 for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
494 struct mmu_psize_def *def = &mmu_psize_defs[psize];
495
Daniel Axtensf5e28482018-10-01 16:21:51 +1000496 if (!def->shift)
497 continue;
498
Kumar Gala1b291872013-03-05 12:08:32 -0600499 if (tlb1ps & (1U << (def->shift - 10))) {
500 def->flags |= MMU_PAGE_SIZE_DIRECT;
Scott Wood28efc352013-10-11 19:22:38 -0500501
502 if (book3e_htw_mode && psize == MMU_PAGE_2M)
503 def->flags |= MMU_PAGE_SIZE_INDIRECT;
Kumar Gala1b291872013-03-05 12:08:32 -0600504 }
505 }
506
Scott Wood28efc352013-10-11 19:22:38 -0500507 goto out;
Kumar Gala1b291872013-03-05 12:08:32 -0600508 }
Kumar Gala988cf862010-10-08 02:13:25 -0500509#endif
510
511 tlb0cfg = mfspr(SPRN_TLB0CFG);
512 tlb0ps = mfspr(SPRN_TLB0PS);
513 eptcfg = mfspr(SPRN_EPTCFG);
514
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000515 /* Look for supported direct sizes */
516 for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
517 struct mmu_psize_def *def = &mmu_psize_defs[psize];
518
519 if (tlb0ps & (1U << (def->shift - 10)))
520 def->flags |= MMU_PAGE_SIZE_DIRECT;
521 }
522
523 /* Indirect page sizes supported ? */
Scott Wood28efc352013-10-11 19:22:38 -0500524 if ((tlb0cfg & TLBnCFG_IND) == 0 ||
525 (tlb0cfg & TLBnCFG_PT) == 0)
526 goto out;
527
528 book3e_htw_mode = PPC_HTW_IBM;
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000529
530 /* Now, we only deal with one IND page size for each
531 * direct size. Hopefully all implementations today are
532 * unambiguous, but we might want to be careful in the
533 * future.
534 */
535 for (i = 0; i < 3; i++) {
536 unsigned int ps, sps;
537
538 sps = eptcfg & 0x1f;
539 eptcfg >>= 5;
540 ps = eptcfg & 0x1f;
541 eptcfg >>= 5;
542 if (!ps || !sps)
543 continue;
544 for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
545 struct mmu_psize_def *def = &mmu_psize_defs[psize];
546
547 if (ps == (def->shift - 10))
548 def->flags |= MMU_PAGE_SIZE_INDIRECT;
549 if (sps == (def->shift - 10))
550 def->ind = ps + 10;
551 }
552 }
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000553
Scott Wood28efc352013-10-11 19:22:38 -0500554out:
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000555 /* Cleanup array and print summary */
556 pr_info("MMU: Supported page sizes\n");
557 for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
558 struct mmu_psize_def *def = &mmu_psize_defs[psize];
559 const char *__page_type_names[] = {
560 "unsupported",
561 "direct",
562 "indirect",
563 "direct & indirect"
564 };
565 if (def->flags == 0) {
566 def->shift = 0;
567 continue;
568 }
569 pr_info(" %8ld KB as %s\n", 1ul << (def->shift - 10),
570 __page_type_names[def->flags & 0x3]);
571 }
572}
573
Scott Woodf67f4ef2011-06-22 11:25:42 +0000574static void setup_mmu_htw(void)
575{
Scott Wood28efc352013-10-11 19:22:38 -0500576 /*
577 * If we want to use HW tablewalk, enable it by patching the TLB miss
578 * handlers to branch to the one dedicated to it.
579 */
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000580
Scott Wood28efc352013-10-11 19:22:38 -0500581 switch (book3e_htw_mode) {
582 case PPC_HTW_IBM:
Scott Woodf67f4ef2011-06-22 11:25:42 +0000583 patch_exception(0x1c0, exc_data_tlb_miss_htw_book3e);
584 patch_exception(0x1e0, exc_instruction_tlb_miss_htw_book3e);
Scott Wood28efc352013-10-11 19:22:38 -0500585 break;
Scott Wood9841c792014-01-17 18:36:38 -0600586#ifdef CONFIG_PPC_FSL_BOOK3E
Scott Wood28efc352013-10-11 19:22:38 -0500587 case PPC_HTW_E6500:
Scott Wood609af382014-03-10 17:29:38 -0500588 extlb_level_exc = EX_TLB_SIZE;
Scott Wood28efc352013-10-11 19:22:38 -0500589 patch_exception(0x1c0, exc_data_tlb_miss_e6500_book3e);
590 patch_exception(0x1e0, exc_instruction_tlb_miss_e6500_book3e);
591 break;
Scott Wood9841c792014-01-17 18:36:38 -0600592#endif
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000593 }
Kumar Gala32d206e2011-05-19 20:09:28 +0000594 pr_info("MMU: Book3E HW tablewalk %s\n",
Scott Wood28efc352013-10-11 19:22:38 -0500595 book3e_htw_mode != PPC_HTW_NONE ? "enabled" : "not supported");
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000596}
597
598/*
599 * Early initialization of the MMU TLB code
600 */
Scott Wood5d61a212014-08-08 18:44:01 -0500601static void early_init_this_mmu(void)
Benjamin Herrenschmidtf2b26c92010-07-09 14:57:43 +1000602{
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000603 unsigned int mas4;
604
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000605 /* Set MAS4 based on page table setting */
606
607 mas4 = 0x4 << MAS4_WIMGED_SHIFT;
Scott Wood28efc352013-10-11 19:22:38 -0500608 switch (book3e_htw_mode) {
609 case PPC_HTW_E6500:
610 mas4 |= MAS4_INDD;
611 mas4 |= BOOK3E_PAGESZ_2M << MAS4_TSIZED_SHIFT;
612 mas4 |= MAS4_TLBSELD(1);
613 mmu_pte_psize = MMU_PAGE_2M;
614 break;
615
616 case PPC_HTW_IBM:
617 mas4 |= MAS4_INDD;
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000618 mas4 |= BOOK3E_PAGESZ_1M << MAS4_TSIZED_SHIFT;
619 mmu_pte_psize = MMU_PAGE_1M;
Scott Wood28efc352013-10-11 19:22:38 -0500620 break;
621
622 case PPC_HTW_NONE:
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000623 mas4 |= BOOK3E_PAGESZ_4K << MAS4_TSIZED_SHIFT;
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000624 mmu_pte_psize = mmu_virtual_psize;
Scott Wood28efc352013-10-11 19:22:38 -0500625 break;
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000626 }
627 mtspr(SPRN_MAS4, mas4);
628
Kumar Gala55fd7662009-10-16 18:48:40 -0500629#ifdef CONFIG_PPC_FSL_BOOK3E
630 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
631 unsigned int num_cams;
Scott Woodd9e18312015-10-06 22:48:09 -0500632 bool map = true;
Kumar Gala55fd7662009-10-16 18:48:40 -0500633
634 /* use a quarter of the TLBCAM for bolted linear map */
635 num_cams = (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) / 4;
Scott Woodd9e18312015-10-06 22:48:09 -0500636
637 /*
638 * Only do the mapping once per core, or else the
639 * transient mapping would cause problems.
640 */
641#ifdef CONFIG_SMP
chenhui zhaoebb9d302015-12-24 08:39:57 +0800642 if (hweight32(get_tensr()) > 1)
Scott Woodd9e18312015-10-06 22:48:09 -0500643 map = false;
644#endif
645
646 if (map)
647 linear_map_top = map_mem_in_cams(linear_map_top,
Christophe Leroy5b548602021-11-15 10:05:32 +0100648 num_cams, false, true);
Scott Wood5d61a212014-08-08 18:44:01 -0500649 }
650#endif
Kumar Gala55fd7662009-10-16 18:48:40 -0500651
Scott Wood5d61a212014-08-08 18:44:01 -0500652 /* A sync won't hurt us after mucking around with
653 * the MMU configuration
654 */
655 mb();
656}
Scott Woodf67f4ef2011-06-22 11:25:42 +0000657
Scott Wood5d61a212014-08-08 18:44:01 -0500658static void __init early_init_mmu_global(void)
659{
660 /* XXX This will have to be decided at runtime, but right
661 * now our boot and TLB miss code hard wires it. Ideally
662 * we should find out a suitable page size and patch the
663 * TLB miss code (either that or use the PACA to store
664 * the value we want)
665 */
666 mmu_linear_psize = MMU_PAGE_1G;
667
668 /* XXX This should be decided at runtime based on supported
669 * page sizes in the TLB, but for now let's assume 16M is
670 * always there and a good fit (which it probably is)
671 *
672 * Freescale booke only supports 4K pages in TLB0, so use that.
673 */
674 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
675 mmu_vmemmap_psize = MMU_PAGE_4K;
676 else
677 mmu_vmemmap_psize = MMU_PAGE_16M;
678
679 /* XXX This code only checks for TLB 0 capabilities and doesn't
680 * check what page size combos are supported by the HW. It
681 * also doesn't handle the case where a separate array holds
682 * the IND entries from the array loaded by the PT.
683 */
684 /* Look for supported page sizes */
685 setup_page_sizes();
686
687 /* Look for HW tablewalk support */
688 setup_mmu_htw();
689
690#ifdef CONFIG_PPC_FSL_BOOK3E
691 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
Scott Wood28efc352013-10-11 19:22:38 -0500692 if (book3e_htw_mode == PPC_HTW_NONE) {
Scott Wood609af382014-03-10 17:29:38 -0500693 extlb_level_exc = EX_TLB_SIZE;
Scott Wood28efc352013-10-11 19:22:38 -0500694 patch_exception(0x1c0, exc_data_tlb_miss_bolted_book3e);
695 patch_exception(0x1e0,
696 exc_instruction_tlb_miss_bolted_book3e);
697 }
Kumar Gala55fd7662009-10-16 18:48:40 -0500698 }
699#endif
700
Scott Wood5d61a212014-08-08 18:44:01 -0500701 /* Set the global containing the top of the linear mapping
702 * for use by the TLB miss code
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000703 */
Scott Wood5d61a212014-08-08 18:44:01 -0500704 linear_map_top = memblock_end_of_DRAM();
Christophe Leroy7cd9b3172019-08-20 14:07:16 +0000705
706 ioremap_bot = IOREMAP_BASE;
Scott Wood5d61a212014-08-08 18:44:01 -0500707}
708
709static void __init early_mmu_set_memory_limit(void)
710{
711#ifdef CONFIG_PPC_FSL_BOOK3E
712 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
713 /*
714 * Limit memory so we dont have linear faults.
715 * Unlike memblock_set_current_limit, which limits
716 * memory available during early boot, this permanently
717 * reduces the memory available to Linux. We need to
718 * do this because highmem is not supported on 64-bit.
719 */
720 memblock_enforce_memory_limit(linear_map_top);
721 }
722#endif
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700723
724 memblock_set_current_limit(linear_map_top);
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000725}
726
Scott Wood5d61a212014-08-08 18:44:01 -0500727/* boot cpu only */
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000728void __init early_init_mmu(void)
729{
Scott Wood5d61a212014-08-08 18:44:01 -0500730 early_init_mmu_global();
731 early_init_this_mmu();
732 early_mmu_set_memory_limit();
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000733}
734
Paul Gortmaker061d19f2013-06-24 15:30:09 -0400735void early_init_mmu_secondary(void)
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000736{
Scott Wood5d61a212014-08-08 18:44:01 -0500737 early_init_this_mmu();
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000738}
739
Benjamin Herrenschmidtcd3db0c2010-07-06 15:39:02 -0700740void setup_initial_memory_limit(phys_addr_t first_memblock_base,
741 phys_addr_t first_memblock_size)
742{
Kumar Gala1dc91c32011-09-16 10:39:59 -0500743 /* On non-FSL Embedded 64-bit, we adjust the RMA size to match
Benjamin Herrenschmidtcd3db0c2010-07-06 15:39:02 -0700744 * the bolted TLB entry. We know for now that only 1G
745 * entries are supported though that may eventually
Kumar Gala1dc91c32011-09-16 10:39:59 -0500746 * change.
747 *
Scott Woodeba5de82015-10-06 22:48:10 -0500748 * on FSL Embedded 64-bit, usually all RAM is bolted, but with
749 * unusual memory sizes it's possible for some RAM to not be mapped
750 * (such RAM is not used at all by Linux, since we don't support
751 * highmem on 64-bit). We limit ppc64_rma_size to what would be
752 * mappable if this memblock is the only one. Additional memblocks
753 * can only increase, not decrease, the amount that ends up getting
754 * mapped. We still limit max to 1G even if we'll eventually map
755 * more. This is due to what the early init code is set up to do.
Kumar Gala1dc91c32011-09-16 10:39:59 -0500756 *
757 * We crop it to the size of the first MEMBLOCK to
Benjamin Herrenschmidtcd3db0c2010-07-06 15:39:02 -0700758 * avoid going over total available memory just in case...
759 */
Kumar Gala1dc91c32011-09-16 10:39:59 -0500760#ifdef CONFIG_PPC_FSL_BOOK3E
Michael Ellerman4868e352017-04-03 12:05:55 +1000761 if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E)) {
Kumar Gala1dc91c32011-09-16 10:39:59 -0500762 unsigned long linear_sz;
Scott Woodeba5de82015-10-06 22:48:10 -0500763 unsigned int num_cams;
764
765 /* use a quarter of the TLBCAM for bolted linear map */
766 num_cams = (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) / 4;
767
768 linear_sz = map_mem_in_cams(first_memblock_size, num_cams,
Christophe Leroy5b548602021-11-15 10:05:32 +0100769 true, true);
Scott Woodeba5de82015-10-06 22:48:10 -0500770
Kumar Gala1dc91c32011-09-16 10:39:59 -0500771 ppc64_rma_size = min_t(u64, linear_sz, 0x40000000);
772 } else
773#endif
774 ppc64_rma_size = min_t(u64, first_memblock_size, 0x40000000);
Benjamin Herrenschmidtcd3db0c2010-07-06 15:39:02 -0700775
776 /* Finally limit subsequent allocations */
Kumar Gala4a892612010-11-10 12:29:49 +0000777 memblock_set_current_limit(first_memblock_base + ppc64_rma_size);
Benjamin Herrenschmidtcd3db0c2010-07-06 15:39:02 -0700778}
Dave Kleikamp91b191c2011-07-04 18:38:03 +0000779#else /* ! CONFIG_PPC64 */
780void __init early_init_mmu(void)
781{
782#ifdef CONFIG_PPC_47x
783 early_init_mmu_47x();
784#endif
Aneesh Kumar K.V67fda38f2019-04-17 18:33:49 +0530785
786#ifdef CONFIG_PPC_MM_SLICES
Christophe Leroy5953fb42019-04-25 14:29:36 +0000787 mm_ctx_set_slb_addr_limit(&init_mm.context, SLB_ADDR_LIMIT_DEFAULT);
Aneesh Kumar K.V67fda38f2019-04-17 18:33:49 +0530788#endif
Dave Kleikamp91b191c2011-07-04 18:38:03 +0000789}
Benjamin Herrenschmidt25d21ad2009-07-23 23:15:47 +0000790#endif /* CONFIG_PPC64 */