xref: /linux/arch/mips/mm/tlb-r4k.c (revision 127fa2ae9e2b1f9b9d876dfaa39fe3640cec5764)
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * Copyright (C) 1996 David S. Miller (davem@davemloft.net)
7  * Copyright (C) 1997, 1998, 1999, 2000 Ralf Baechle ralf@gnu.org
8  * Carsten Langgaard, carstenl@mips.com
9  * Copyright (C) 2002 MIPS Technologies, Inc.  All rights reserved.
10  */
11 #include <linux/cpu_pm.h>
12 #include <linux/init.h>
13 #include <linux/sched.h>
14 #include <linux/smp.h>
15 #include <linux/mm.h>
16 #include <linux/hugetlb.h>
17 #include <linux/export.h>
18 #include <linux/sort.h>
19 
20 #include <asm/cpu.h>
21 #include <asm/cpu-type.h>
22 #include <asm/bootinfo.h>
23 #include <asm/hazards.h>
24 #include <asm/mmu_context.h>
25 #include <asm/tlb.h>
26 #include <asm/tlbex.h>
27 #include <asm/tlbmisc.h>
28 #include <asm/setup.h>
29 
30 /*
31  * LOONGSON-2 has a 4 entry itlb which is a subset of jtlb, LOONGSON-3 has
32  * a 4 entry itlb and a 4 entry dtlb which are subsets of jtlb. Unfortunately,
33  * itlb/dtlb are not totally transparent to software.
34  */
35 static inline void flush_micro_tlb(void)
36 {
37 	switch (current_cpu_type()) {
38 	case CPU_LOONGSON2EF:
39 		write_c0_diag(LOONGSON_DIAG_ITLB);
40 		break;
41 	case CPU_LOONGSON64:
42 		write_c0_diag(LOONGSON_DIAG_ITLB | LOONGSON_DIAG_DTLB);
43 		break;
44 	default:
45 		break;
46 	}
47 }
48 
49 static inline void flush_micro_tlb_vm(struct vm_area_struct *vma)
50 {
51 	if (vma->vm_flags & VM_EXEC)
52 		flush_micro_tlb();
53 }
54 
55 void local_flush_tlb_all(void)
56 {
57 	unsigned long flags;
58 	unsigned long old_ctx;
59 	int entry, ftlbhighset;
60 
61 	local_irq_save(flags);
62 	/* Save old context and create impossible VPN2 value */
63 	old_ctx = read_c0_entryhi();
64 	htw_stop();
65 	write_c0_entrylo0(0);
66 	write_c0_entrylo1(0);
67 
68 	entry = num_wired_entries();
69 
70 	/*
71 	 * Blast 'em all away.
72 	 * If there are any wired entries, fall back to iterating
73 	 */
74 	if (cpu_has_tlbinv && !entry) {
75 		if (current_cpu_data.tlbsizevtlb) {
76 			write_c0_index(0);
77 			mtc0_tlbw_hazard();
78 			tlbinvf();  /* invalidate VTLB */
79 		}
80 		ftlbhighset = current_cpu_data.tlbsizevtlb +
81 			current_cpu_data.tlbsizeftlbsets;
82 		for (entry = current_cpu_data.tlbsizevtlb;
83 		     entry < ftlbhighset;
84 		     entry++) {
85 			write_c0_index(entry);
86 			mtc0_tlbw_hazard();
87 			tlbinvf();  /* invalidate one FTLB set */
88 		}
89 	} else {
90 		while (entry < current_cpu_data.tlbsize) {
91 			/* Make sure all entries differ. */
92 			write_c0_entryhi(UNIQUE_ENTRYHI(entry));
93 			write_c0_index(entry);
94 			mtc0_tlbw_hazard();
95 			tlb_write_indexed();
96 			entry++;
97 		}
98 	}
99 	tlbw_use_hazard();
100 	write_c0_entryhi(old_ctx);
101 	htw_start();
102 	flush_micro_tlb();
103 	local_irq_restore(flags);
104 }
105 EXPORT_SYMBOL(local_flush_tlb_all);
106 
107 void local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
108 	unsigned long end)
109 {
110 	struct mm_struct *mm = vma->vm_mm;
111 	int cpu = smp_processor_id();
112 
113 	if (cpu_context(cpu, mm) != 0) {
114 		unsigned long size, flags;
115 
116 		local_irq_save(flags);
117 		start = round_down(start, PAGE_SIZE << 1);
118 		end = round_up(end, PAGE_SIZE << 1);
119 		size = (end - start) >> (PAGE_SHIFT + 1);
120 		if (size <= (current_cpu_data.tlbsizeftlbsets ?
121 			     current_cpu_data.tlbsize / 8 :
122 			     current_cpu_data.tlbsize / 2)) {
123 			unsigned long old_entryhi, old_mmid;
124 			int newpid = cpu_asid(cpu, mm);
125 
126 			old_entryhi = read_c0_entryhi();
127 			if (cpu_has_mmid) {
128 				old_mmid = read_c0_memorymapid();
129 				write_c0_memorymapid(newpid);
130 			}
131 
132 			htw_stop();
133 			while (start < end) {
134 				int idx;
135 
136 				if (cpu_has_mmid)
137 					write_c0_entryhi(start);
138 				else
139 					write_c0_entryhi(start | newpid);
140 				start += (PAGE_SIZE << 1);
141 				mtc0_tlbw_hazard();
142 				tlb_probe();
143 				tlb_probe_hazard();
144 				idx = read_c0_index();
145 				write_c0_entrylo0(0);
146 				write_c0_entrylo1(0);
147 				if (idx < 0)
148 					continue;
149 				/* Make sure all entries differ. */
150 				write_c0_entryhi(UNIQUE_ENTRYHI(idx));
151 				mtc0_tlbw_hazard();
152 				tlb_write_indexed();
153 			}
154 			tlbw_use_hazard();
155 			write_c0_entryhi(old_entryhi);
156 			if (cpu_has_mmid)
157 				write_c0_memorymapid(old_mmid);
158 			htw_start();
159 		} else {
160 			drop_mmu_context(mm);
161 		}
162 		flush_micro_tlb();
163 		local_irq_restore(flags);
164 	}
165 }
166 
167 void local_flush_tlb_kernel_range(unsigned long start, unsigned long end)
168 {
169 	unsigned long size, flags;
170 
171 	local_irq_save(flags);
172 	size = (end - start + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
173 	size = (size + 1) >> 1;
174 	if (size <= (current_cpu_data.tlbsizeftlbsets ?
175 		     current_cpu_data.tlbsize / 8 :
176 		     current_cpu_data.tlbsize / 2)) {
177 		int pid = read_c0_entryhi();
178 
179 		start &= (PAGE_MASK << 1);
180 		end += ((PAGE_SIZE << 1) - 1);
181 		end &= (PAGE_MASK << 1);
182 		htw_stop();
183 
184 		while (start < end) {
185 			int idx;
186 
187 			write_c0_entryhi(start);
188 			start += (PAGE_SIZE << 1);
189 			mtc0_tlbw_hazard();
190 			tlb_probe();
191 			tlb_probe_hazard();
192 			idx = read_c0_index();
193 			write_c0_entrylo0(0);
194 			write_c0_entrylo1(0);
195 			if (idx < 0)
196 				continue;
197 			/* Make sure all entries differ. */
198 			write_c0_entryhi(UNIQUE_ENTRYHI(idx));
199 			mtc0_tlbw_hazard();
200 			tlb_write_indexed();
201 		}
202 		tlbw_use_hazard();
203 		write_c0_entryhi(pid);
204 		htw_start();
205 	} else {
206 		local_flush_tlb_all();
207 	}
208 	flush_micro_tlb();
209 	local_irq_restore(flags);
210 }
211 
212 void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
213 {
214 	int cpu = smp_processor_id();
215 
216 	if (cpu_context(cpu, vma->vm_mm) != 0) {
217 		unsigned long old_mmid;
218 		unsigned long flags, old_entryhi;
219 		int idx;
220 
221 		page &= (PAGE_MASK << 1);
222 		local_irq_save(flags);
223 		old_entryhi = read_c0_entryhi();
224 		htw_stop();
225 		if (cpu_has_mmid) {
226 			old_mmid = read_c0_memorymapid();
227 			write_c0_entryhi(page);
228 			write_c0_memorymapid(cpu_asid(cpu, vma->vm_mm));
229 		} else {
230 			write_c0_entryhi(page | cpu_asid(cpu, vma->vm_mm));
231 		}
232 		mtc0_tlbw_hazard();
233 		tlb_probe();
234 		tlb_probe_hazard();
235 		idx = read_c0_index();
236 		write_c0_entrylo0(0);
237 		write_c0_entrylo1(0);
238 		if (idx < 0)
239 			goto finish;
240 		/* Make sure all entries differ. */
241 		write_c0_entryhi(UNIQUE_ENTRYHI(idx));
242 		mtc0_tlbw_hazard();
243 		tlb_write_indexed();
244 		tlbw_use_hazard();
245 
246 	finish:
247 		write_c0_entryhi(old_entryhi);
248 		if (cpu_has_mmid)
249 			write_c0_memorymapid(old_mmid);
250 		htw_start();
251 		flush_micro_tlb_vm(vma);
252 		local_irq_restore(flags);
253 	}
254 }
255 
256 /*
257  * This one is only used for pages with the global bit set so we don't care
258  * much about the ASID.
259  */
260 void local_flush_tlb_one(unsigned long page)
261 {
262 	unsigned long flags;
263 	int oldpid, idx;
264 
265 	local_irq_save(flags);
266 	oldpid = read_c0_entryhi();
267 	htw_stop();
268 	page &= (PAGE_MASK << 1);
269 	write_c0_entryhi(page);
270 	mtc0_tlbw_hazard();
271 	tlb_probe();
272 	tlb_probe_hazard();
273 	idx = read_c0_index();
274 	write_c0_entrylo0(0);
275 	write_c0_entrylo1(0);
276 	if (idx >= 0) {
277 		/* Make sure all entries differ. */
278 		write_c0_entryhi(UNIQUE_ENTRYHI(idx));
279 		mtc0_tlbw_hazard();
280 		tlb_write_indexed();
281 		tlbw_use_hazard();
282 	}
283 	write_c0_entryhi(oldpid);
284 	htw_start();
285 	flush_micro_tlb();
286 	local_irq_restore(flags);
287 }
288 
289 /*
290  * We will need multiple versions of update_mmu_cache(), one that just
291  * updates the TLB with the new pte(s), and another which also checks
292  * for the R4k "end of page" hardware bug and does the needy.
293  */
294 void __update_tlb(struct vm_area_struct * vma, unsigned long address, pte_t pte)
295 {
296 	unsigned long flags;
297 	pgd_t *pgdp;
298 	p4d_t *p4dp;
299 	pud_t *pudp;
300 	pmd_t *pmdp;
301 	pte_t *ptep, *ptemap = NULL;
302 	int idx, pid;
303 
304 	/*
305 	 * Handle debugger faulting in for debuggee.
306 	 */
307 	if (current->active_mm != vma->vm_mm)
308 		return;
309 
310 	local_irq_save(flags);
311 
312 	htw_stop();
313 	address &= (PAGE_MASK << 1);
314 	if (cpu_has_mmid) {
315 		write_c0_entryhi(address);
316 	} else {
317 		pid = read_c0_entryhi() & cpu_asid_mask(&current_cpu_data);
318 		write_c0_entryhi(address | pid);
319 	}
320 	pgdp = pgd_offset(vma->vm_mm, address);
321 	mtc0_tlbw_hazard();
322 	tlb_probe();
323 	tlb_probe_hazard();
324 	p4dp = p4d_offset(pgdp, address);
325 	pudp = pud_offset(p4dp, address);
326 	pmdp = pmd_offset(pudp, address);
327 	idx = read_c0_index();
328 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
329 	/* this could be a huge page  */
330 	if (pmd_leaf(*pmdp)) {
331 		unsigned long lo;
332 		write_c0_pagemask(PM_HUGE_MASK);
333 		ptep = (pte_t *)pmdp;
334 		lo = pte_to_entrylo(pte_val(*ptep));
335 		write_c0_entrylo0(lo);
336 		write_c0_entrylo1(lo + (HPAGE_SIZE >> 7));
337 
338 		mtc0_tlbw_hazard();
339 		if (idx < 0)
340 			tlb_write_random();
341 		else
342 			tlb_write_indexed();
343 		tlbw_use_hazard();
344 		write_c0_pagemask(PM_DEFAULT_MASK);
345 	} else
346 #endif
347 	{
348 		ptemap = ptep = pte_offset_map(pmdp, address);
349 		/*
350 		 * update_mmu_cache() is called between pte_offset_map_lock()
351 		 * and pte_unmap_unlock(), so we can assume that ptep is not
352 		 * NULL here: and what should be done below if it were NULL?
353 		 */
354 
355 #if defined(CONFIG_PHYS_ADDR_T_64BIT) && defined(CONFIG_CPU_MIPS32)
356 #ifdef CONFIG_XPA
357 		write_c0_entrylo0(pte_to_entrylo(ptep->pte_high));
358 		if (cpu_has_xpa)
359 			writex_c0_entrylo0(ptep->pte_low & _PFNX_MASK);
360 		ptep++;
361 		write_c0_entrylo1(pte_to_entrylo(ptep->pte_high));
362 		if (cpu_has_xpa)
363 			writex_c0_entrylo1(ptep->pte_low & _PFNX_MASK);
364 #else
365 		write_c0_entrylo0(ptep->pte_high);
366 		ptep++;
367 		write_c0_entrylo1(ptep->pte_high);
368 #endif
369 #else
370 		write_c0_entrylo0(pte_to_entrylo(pte_val(*ptep++)));
371 		write_c0_entrylo1(pte_to_entrylo(pte_val(*ptep)));
372 #endif
373 		mtc0_tlbw_hazard();
374 		if (idx < 0)
375 			tlb_write_random();
376 		else
377 			tlb_write_indexed();
378 	}
379 	tlbw_use_hazard();
380 	htw_start();
381 	flush_micro_tlb_vm(vma);
382 
383 	if (ptemap)
384 		pte_unmap(ptemap);
385 	local_irq_restore(flags);
386 }
387 
388 void add_wired_entry(unsigned long entrylo0, unsigned long entrylo1,
389 		     unsigned long entryhi, unsigned long pagemask)
390 {
391 #ifdef CONFIG_XPA
392 	panic("Broken for XPA kernels");
393 #else
394 	unsigned int old_mmid;
395 	unsigned long flags;
396 	unsigned long wired;
397 	unsigned long old_pagemask;
398 	unsigned long old_ctx;
399 
400 	local_irq_save(flags);
401 	if (cpu_has_mmid) {
402 		old_mmid = read_c0_memorymapid();
403 		write_c0_memorymapid(MMID_KERNEL_WIRED);
404 	}
405 	/* Save old context and create impossible VPN2 value */
406 	old_ctx = read_c0_entryhi();
407 	htw_stop();
408 	old_pagemask = read_c0_pagemask();
409 	wired = num_wired_entries();
410 	write_c0_wired(wired + 1);
411 	write_c0_index(wired);
412 	tlbw_use_hazard();	/* What is the hazard here? */
413 	write_c0_pagemask(pagemask);
414 	write_c0_entryhi(entryhi);
415 	write_c0_entrylo0(entrylo0);
416 	write_c0_entrylo1(entrylo1);
417 	mtc0_tlbw_hazard();
418 	tlb_write_indexed();
419 	tlbw_use_hazard();
420 
421 	write_c0_entryhi(old_ctx);
422 	if (cpu_has_mmid)
423 		write_c0_memorymapid(old_mmid);
424 	tlbw_use_hazard();	/* What is the hazard here? */
425 	htw_start();
426 	write_c0_pagemask(old_pagemask);
427 	local_flush_tlb_all();
428 	local_irq_restore(flags);
429 #endif
430 }
431 
432 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
433 
434 int has_transparent_hugepage(void)
435 {
436 	static unsigned int mask = -1;
437 
438 	if (mask == -1) {	/* first call comes during __init */
439 		unsigned long flags;
440 
441 		local_irq_save(flags);
442 		write_c0_pagemask(PM_HUGE_MASK);
443 		back_to_back_c0_hazard();
444 		mask = read_c0_pagemask();
445 		write_c0_pagemask(PM_DEFAULT_MASK);
446 		local_irq_restore(flags);
447 	}
448 	return mask == PM_HUGE_MASK;
449 }
450 EXPORT_SYMBOL(has_transparent_hugepage);
451 
452 #endif /* CONFIG_TRANSPARENT_HUGEPAGE  */
453 
454 /*
455  * Used for loading TLB entries before trap_init() has started, when we
456  * don't actually want to add a wired entry which remains throughout the
457  * lifetime of the system
458  */
459 
460 int temp_tlb_entry;
461 
462 #ifndef CONFIG_64BIT
463 __init int add_temporary_entry(unsigned long entrylo0, unsigned long entrylo1,
464 			       unsigned long entryhi, unsigned long pagemask)
465 {
466 	int ret = 0;
467 	unsigned long flags;
468 	unsigned long wired;
469 	unsigned long old_pagemask;
470 	unsigned long old_ctx;
471 
472 	local_irq_save(flags);
473 	/* Save old context and create impossible VPN2 value */
474 	htw_stop();
475 	old_ctx = read_c0_entryhi();
476 	old_pagemask = read_c0_pagemask();
477 	wired = num_wired_entries();
478 	if (--temp_tlb_entry < wired) {
479 		printk(KERN_WARNING
480 		       "No TLB space left for add_temporary_entry\n");
481 		ret = -ENOSPC;
482 		goto out;
483 	}
484 
485 	write_c0_index(temp_tlb_entry);
486 	write_c0_pagemask(pagemask);
487 	write_c0_entryhi(entryhi);
488 	write_c0_entrylo0(entrylo0);
489 	write_c0_entrylo1(entrylo1);
490 	mtc0_tlbw_hazard();
491 	tlb_write_indexed();
492 	tlbw_use_hazard();
493 
494 	write_c0_entryhi(old_ctx);
495 	write_c0_pagemask(old_pagemask);
496 	htw_start();
497 out:
498 	local_irq_restore(flags);
499 	return ret;
500 }
501 #endif
502 
503 static int ntlb;
504 static int __init set_ntlb(char *str)
505 {
506 	get_option(&str, &ntlb);
507 	return 1;
508 }
509 
510 __setup("ntlb=", set_ntlb);
511 
512 
513 /* Comparison function for EntryHi VPN fields.  */
514 static int r4k_vpn_cmp(const void *a, const void *b)
515 {
516 	long v = *(unsigned long *)a - *(unsigned long *)b;
517 	int s = sizeof(long) > sizeof(int) ? sizeof(long) * 8 - 1: 0;
518 	return s ? (v != 0) | v >> s : v;
519 }
520 
521 /*
522  * Initialise all TLB entries with unique values that do not clash with
523  * what we have been handed over and what we'll be using ourselves.
524  */
525 static void r4k_tlb_uniquify(void)
526 {
527 	unsigned long tlb_vpns[1 << MIPS_CONF1_TLBS_SIZE];
528 	int tlbsize = current_cpu_data.tlbsize;
529 	int start = num_wired_entries();
530 	unsigned long vpn_mask;
531 	int cnt, ent, idx, i;
532 
533 	vpn_mask = GENMASK(cpu_vmbits - 1, 13);
534 	vpn_mask |= IS_ENABLED(CONFIG_64BIT) ? 3ULL << 62 : 1 << 31;
535 
536 	htw_stop();
537 
538 	for (i = start, cnt = 0; i < tlbsize; i++, cnt++) {
539 		unsigned long vpn;
540 
541 		write_c0_index(i);
542 		mtc0_tlbr_hazard();
543 		tlb_read();
544 		tlb_read_hazard();
545 		vpn = read_c0_entryhi();
546 		vpn &= vpn_mask & PAGE_MASK;
547 		tlb_vpns[cnt] = vpn;
548 
549 		/* Prevent any large pages from overlapping regular ones.  */
550 		write_c0_pagemask(read_c0_pagemask() & PM_DEFAULT_MASK);
551 		mtc0_tlbw_hazard();
552 		tlb_write_indexed();
553 		tlbw_use_hazard();
554 	}
555 
556 	sort(tlb_vpns, cnt, sizeof(tlb_vpns[0]), r4k_vpn_cmp, NULL);
557 
558 	write_c0_pagemask(PM_DEFAULT_MASK);
559 	write_c0_entrylo0(0);
560 	write_c0_entrylo1(0);
561 
562 	idx = 0;
563 	ent = tlbsize;
564 	for (i = start; i < tlbsize; i++)
565 		while (1) {
566 			unsigned long entryhi, vpn;
567 
568 			entryhi = UNIQUE_ENTRYHI(ent);
569 			vpn = entryhi & vpn_mask & PAGE_MASK;
570 
571 			if (idx >= cnt || vpn < tlb_vpns[idx]) {
572 				write_c0_entryhi(entryhi);
573 				write_c0_index(i);
574 				mtc0_tlbw_hazard();
575 				tlb_write_indexed();
576 				ent++;
577 				break;
578 			} else if (vpn == tlb_vpns[idx]) {
579 				ent++;
580 			} else {
581 				idx++;
582 			}
583 		}
584 
585 	tlbw_use_hazard();
586 	htw_start();
587 	flush_micro_tlb();
588 }
589 
590 /*
591  * Configure TLB (for init or after a CPU has been powered off).
592  */
593 static void r4k_tlb_configure(void)
594 {
595 	/*
596 	 * You should never change this register:
597 	 *   - On R4600 1.7 the tlbp never hits for pages smaller than
598 	 *     the value in the c0_pagemask register.
599 	 *   - The entire mm handling assumes the c0_pagemask register to
600 	 *     be set to fixed-size pages.
601 	 */
602 	write_c0_pagemask(PM_DEFAULT_MASK);
603 	back_to_back_c0_hazard();
604 	if (read_c0_pagemask() != PM_DEFAULT_MASK)
605 		panic("MMU doesn't support PAGE_SIZE=0x%lx", PAGE_SIZE);
606 
607 	write_c0_wired(0);
608 	if (current_cpu_type() == CPU_R10000 ||
609 	    current_cpu_type() == CPU_R12000 ||
610 	    current_cpu_type() == CPU_R14000 ||
611 	    current_cpu_type() == CPU_R16000)
612 		write_c0_framemask(0);
613 
614 	if (cpu_has_rixi) {
615 		/*
616 		 * Enable the no read, no exec bits, and enable large physical
617 		 * address.
618 		 */
619 #ifdef CONFIG_64BIT
620 		set_c0_pagegrain(PG_RIE | PG_XIE | PG_ELPA);
621 #else
622 		set_c0_pagegrain(PG_RIE | PG_XIE);
623 #endif
624 	}
625 
626 	temp_tlb_entry = current_cpu_data.tlbsize - 1;
627 
628 	/* From this point on the ARC firmware is dead.	 */
629 	r4k_tlb_uniquify();
630 	local_flush_tlb_all();
631 
632 	/* Did I tell you that ARC SUCKS?  */
633 }
634 
635 void tlb_init(void)
636 {
637 	r4k_tlb_configure();
638 
639 	if (ntlb) {
640 		if (ntlb > 1 && ntlb <= current_cpu_data.tlbsize) {
641 			int wired = current_cpu_data.tlbsize - ntlb;
642 			write_c0_wired(wired);
643 			write_c0_index(wired-1);
644 			printk("Restricting TLB to %d entries\n", ntlb);
645 		} else
646 			printk("Ignoring invalid argument ntlb=%d\n", ntlb);
647 	}
648 
649 	build_tlb_refill_handler();
650 }
651 
652 static int r4k_tlb_pm_notifier(struct notifier_block *self, unsigned long cmd,
653 			       void *v)
654 {
655 	switch (cmd) {
656 	case CPU_PM_ENTER_FAILED:
657 	case CPU_PM_EXIT:
658 		r4k_tlb_configure();
659 		break;
660 	}
661 
662 	return NOTIFY_OK;
663 }
664 
665 static struct notifier_block r4k_tlb_pm_notifier_block = {
666 	.notifier_call = r4k_tlb_pm_notifier,
667 };
668 
669 static int __init r4k_tlb_init_pm(void)
670 {
671 	return cpu_pm_register_notifier(&r4k_tlb_pm_notifier_block);
672 }
673 arch_initcall(r4k_tlb_init_pm);
674