xref: /openbmc/u-boot/arch/arm/cpu/armv8/cache_v8.c (revision ad5b5801)
1 /*
2  * (C) Copyright 2013
3  * David Feng <fenghua@phytium.com.cn>
4  *
5  * (C) Copyright 2016
6  * Alexander Graf <agraf@suse.de>
7  *
8  * SPDX-License-Identifier:	GPL-2.0+
9  */
10 
11 #include <common.h>
12 #include <asm/system.h>
13 #include <asm/armv8/mmu.h>
14 
15 DECLARE_GLOBAL_DATA_PTR;
16 
17 #ifndef CONFIG_SYS_DCACHE_OFF
18 
19 /*
20  *  With 4k page granule, a virtual address is split into 4 lookup parts
21  *  spanning 9 bits each:
22  *
23  *    _______________________________________________
24  *   |       |       |       |       |       |       |
25  *   |   0   |  Lv0  |  Lv1  |  Lv2  |  Lv3  |  off  |
26  *   |_______|_______|_______|_______|_______|_______|
27  *     63-48   47-39   38-30   29-21   20-12   11-00
28  *
29  *             mask        page size
30  *
31  *    Lv0: FF8000000000       --
32  *    Lv1:   7FC0000000       1G
33  *    Lv2:     3FE00000       2M
34  *    Lv3:       1FF000       4K
35  *    off:          FFF
36  */
37 
38 static u64 get_tcr(int el, u64 *pips, u64 *pva_bits)
39 {
40 	u64 max_addr = 0;
41 	u64 ips, va_bits;
42 	u64 tcr;
43 	int i;
44 
45 	/* Find the largest address we need to support */
46 	for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
47 		max_addr = max(max_addr, mem_map[i].base + mem_map[i].size);
48 
49 	/* Calculate the maximum physical (and thus virtual) address */
50 	if (max_addr > (1ULL << 44)) {
51 		ips = 5;
52 		va_bits = 48;
53 	} else  if (max_addr > (1ULL << 42)) {
54 		ips = 4;
55 		va_bits = 44;
56 	} else  if (max_addr > (1ULL << 40)) {
57 		ips = 3;
58 		va_bits = 42;
59 	} else  if (max_addr > (1ULL << 36)) {
60 		ips = 2;
61 		va_bits = 40;
62 	} else  if (max_addr > (1ULL << 32)) {
63 		ips = 1;
64 		va_bits = 36;
65 	} else {
66 		ips = 0;
67 		va_bits = 32;
68 	}
69 
70 	if (el == 1) {
71 		tcr = TCR_EL1_RSVD | (ips << 32) | TCR_EPD1_DISABLE;
72 	} else if (el == 2) {
73 		tcr = TCR_EL2_RSVD | (ips << 16);
74 	} else {
75 		tcr = TCR_EL3_RSVD | (ips << 16);
76 	}
77 
78 	/* PTWs cacheable, inner/outer WBWA and inner shareable */
79 	tcr |= TCR_TG0_4K | TCR_SHARED_INNER | TCR_ORGN_WBWA | TCR_IRGN_WBWA;
80 	tcr |= TCR_T0SZ(va_bits);
81 
82 	if (pips)
83 		*pips = ips;
84 	if (pva_bits)
85 		*pva_bits = va_bits;
86 
87 	return tcr;
88 }
89 
90 #define MAX_PTE_ENTRIES 512
91 
92 static int pte_type(u64 *pte)
93 {
94 	return *pte & PTE_TYPE_MASK;
95 }
96 
97 /* Returns the LSB number for a PTE on level <level> */
98 static int level2shift(int level)
99 {
100 	/* Page is 12 bits wide, every level translates 9 bits */
101 	return (12 + 9 * (3 - level));
102 }
103 
104 static u64 *find_pte(u64 addr, int level)
105 {
106 	int start_level = 0;
107 	u64 *pte;
108 	u64 idx;
109 	u64 va_bits;
110 	int i;
111 
112 	debug("addr=%llx level=%d\n", addr, level);
113 
114 	get_tcr(0, NULL, &va_bits);
115 	if (va_bits < 39)
116 		start_level = 1;
117 
118 	if (level < start_level)
119 		return NULL;
120 
121 	/* Walk through all page table levels to find our PTE */
122 	pte = (u64*)gd->arch.tlb_addr;
123 	for (i = start_level; i < 4; i++) {
124 		idx = (addr >> level2shift(i)) & 0x1FF;
125 		pte += idx;
126 		debug("idx=%llx PTE %p at level %d: %llx\n", idx, pte, i, *pte);
127 
128 		/* Found it */
129 		if (i == level)
130 			return pte;
131 		/* PTE is no table (either invalid or block), can't traverse */
132 		if (pte_type(pte) != PTE_TYPE_TABLE)
133 			return NULL;
134 		/* Off to the next level */
135 		pte = (u64*)(*pte & 0x0000fffffffff000ULL);
136 	}
137 
138 	/* Should never reach here */
139 	return NULL;
140 }
141 
142 /* Returns and creates a new full table (512 entries) */
143 static u64 *create_table(void)
144 {
145 	u64 *new_table = (u64*)gd->arch.tlb_fillptr;
146 	u64 pt_len = MAX_PTE_ENTRIES * sizeof(u64);
147 
148 	/* Allocate MAX_PTE_ENTRIES pte entries */
149 	gd->arch.tlb_fillptr += pt_len;
150 
151 	if (gd->arch.tlb_fillptr - gd->arch.tlb_addr > gd->arch.tlb_size)
152 		panic("Insufficient RAM for page table: 0x%lx > 0x%lx. "
153 		      "Please increase the size in get_page_table_size()",
154 			gd->arch.tlb_fillptr - gd->arch.tlb_addr,
155 			gd->arch.tlb_size);
156 
157 	/* Mark all entries as invalid */
158 	memset(new_table, 0, pt_len);
159 
160 	return new_table;
161 }
162 
163 static void set_pte_table(u64 *pte, u64 *table)
164 {
165 	/* Point *pte to the new table */
166 	debug("Setting %p to addr=%p\n", pte, table);
167 	*pte = PTE_TYPE_TABLE | (ulong)table;
168 }
169 
170 /* Add one mm_region map entry to the page tables */
171 static void add_map(struct mm_region *map)
172 {
173 	u64 *pte;
174 	u64 addr = map->base;
175 	u64 size = map->size;
176 	u64 attrs = map->attrs | PTE_TYPE_BLOCK | PTE_BLOCK_AF;
177 	u64 blocksize;
178 	int level;
179 	u64 *new_table;
180 
181 	while (size) {
182 		pte = find_pte(addr, 0);
183 		if (pte && (pte_type(pte) == PTE_TYPE_FAULT)) {
184 			debug("Creating table for addr 0x%llx\n", addr);
185 			new_table = create_table();
186 			set_pte_table(pte, new_table);
187 		}
188 
189 		for (level = 1; level < 4; level++) {
190 			pte = find_pte(addr, level);
191 			blocksize = 1ULL << level2shift(level);
192 			debug("Checking if pte fits for addr=%llx size=%llx "
193 			      "blocksize=%llx\n", addr, size, blocksize);
194 			if (size >= blocksize && !(addr & (blocksize - 1))) {
195 				/* Page fits, create block PTE */
196 				debug("Setting PTE %p to block addr=%llx\n",
197 				      pte, addr);
198 				*pte = addr | attrs;
199 				addr += blocksize;
200 				size -= blocksize;
201 				break;
202 			} else if ((pte_type(pte) == PTE_TYPE_FAULT)) {
203 				/* Page doesn't fit, create subpages */
204 				debug("Creating subtable for addr 0x%llx "
205 				      "blksize=%llx\n", addr, blocksize);
206 				new_table = create_table();
207 				set_pte_table(pte, new_table);
208 			}
209 		}
210 	}
211 }
212 
213 /* Splits a block PTE into table with subpages spanning the old block */
214 static void split_block(u64 *pte, int level)
215 {
216 	u64 old_pte = *pte;
217 	u64 *new_table;
218 	u64 i = 0;
219 	/* level describes the parent level, we need the child ones */
220 	int levelshift = level2shift(level + 1);
221 
222 	if (pte_type(pte) != PTE_TYPE_BLOCK)
223 		panic("PTE %p (%llx) is not a block. Some driver code wants to "
224 		      "modify dcache settings for an range not covered in "
225 		      "mem_map.", pte, old_pte);
226 
227 	new_table = create_table();
228 	debug("Splitting pte %p (%llx) into %p\n", pte, old_pte, new_table);
229 
230 	for (i = 0; i < MAX_PTE_ENTRIES; i++) {
231 		new_table[i] = old_pte | (i << levelshift);
232 
233 		/* Level 3 block PTEs have the table type */
234 		if ((level + 1) == 3)
235 			new_table[i] |= PTE_TYPE_TABLE;
236 
237 		debug("Setting new_table[%lld] = %llx\n", i, new_table[i]);
238 	}
239 
240 	/* Set the new table into effect */
241 	set_pte_table(pte, new_table);
242 }
243 
244 enum pte_type {
245 	PTE_INVAL,
246 	PTE_BLOCK,
247 	PTE_LEVEL,
248 };
249 
250 /*
251  * This is a recursively called function to count the number of
252  * page tables we need to cover a particular PTE range. If you
253  * call this with level = -1 you basically get the full 48 bit
254  * coverage.
255  */
256 static int count_required_pts(u64 addr, int level, u64 maxaddr)
257 {
258 	int levelshift = level2shift(level);
259 	u64 levelsize = 1ULL << levelshift;
260 	u64 levelmask = levelsize - 1;
261 	u64 levelend = addr + levelsize;
262 	int r = 0;
263 	int i;
264 	enum pte_type pte_type = PTE_INVAL;
265 
266 	for (i = 0; mem_map[i].size || mem_map[i].attrs; i++) {
267 		struct mm_region *map = &mem_map[i];
268 		u64 start = map->base;
269 		u64 end = start + map->size;
270 
271 		/* Check if the PTE would overlap with the map */
272 		if (max(addr, start) <= min(levelend, end)) {
273 			start = max(addr, start);
274 			end = min(levelend, end);
275 
276 			/* We need a sub-pt for this level */
277 			if ((start & levelmask) || (end & levelmask)) {
278 				pte_type = PTE_LEVEL;
279 				break;
280 			}
281 
282 			/* Lv0 can not do block PTEs, so do levels here too */
283 			if (level <= 0) {
284 				pte_type = PTE_LEVEL;
285 				break;
286 			}
287 
288 			/* PTE is active, but fits into a block */
289 			pte_type = PTE_BLOCK;
290 		}
291 	}
292 
293 	/*
294 	 * Block PTEs at this level are already covered by the parent page
295 	 * table, so we only need to count sub page tables.
296 	 */
297 	if (pte_type == PTE_LEVEL) {
298 		int sublevel = level + 1;
299 		u64 sublevelsize = 1ULL << level2shift(sublevel);
300 
301 		/* Account for the new sub page table ... */
302 		r = 1;
303 
304 		/* ... and for all child page tables that one might have */
305 		for (i = 0; i < MAX_PTE_ENTRIES; i++) {
306 			r += count_required_pts(addr, sublevel, maxaddr);
307 			addr += sublevelsize;
308 
309 			if (addr >= maxaddr) {
310 				/*
311 				 * We reached the end of address space, no need
312 				 * to look any further.
313 				 */
314 				break;
315 			}
316 		}
317 	}
318 
319 	return r;
320 }
321 
322 /* Returns the estimated required size of all page tables */
323 __weak u64 get_page_table_size(void)
324 {
325 	u64 one_pt = MAX_PTE_ENTRIES * sizeof(u64);
326 	u64 size = 0;
327 	u64 va_bits;
328 	int start_level = 0;
329 
330 	get_tcr(0, NULL, &va_bits);
331 	if (va_bits < 39)
332 		start_level = 1;
333 
334 	/* Account for all page tables we would need to cover our memory map */
335 	size = one_pt * count_required_pts(0, start_level - 1, 1ULL << va_bits);
336 
337 	/*
338 	 * We need to duplicate our page table once to have an emergency pt to
339 	 * resort to when splitting page tables later on
340 	 */
341 	size *= 2;
342 
343 	/*
344 	 * We may need to split page tables later on if dcache settings change,
345 	 * so reserve up to 4 (random pick) page tables for that.
346 	 */
347 	size += one_pt * 4;
348 
349 	return size;
350 }
351 
352 static void setup_pgtables(void)
353 {
354 	int i;
355 
356 	/*
357 	 * Allocate the first level we're on with invalidate entries.
358 	 * If the starting level is 0 (va_bits >= 39), then this is our
359 	 * Lv0 page table, otherwise it's the entry Lv1 page table.
360 	 */
361 	create_table();
362 
363 	/* Now add all MMU table entries one after another to the table */
364 	for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
365 		add_map(&mem_map[i]);
366 
367 	/* Create the same thing once more for our emergency page table */
368 	create_table();
369 }
370 
371 static void setup_all_pgtables(void)
372 {
373 	u64 tlb_addr = gd->arch.tlb_addr;
374 
375 	/* Reset the fill ptr */
376 	gd->arch.tlb_fillptr = tlb_addr;
377 
378 	/* Create normal system page tables */
379 	setup_pgtables();
380 
381 	/* Create emergency page tables */
382 	gd->arch.tlb_addr = gd->arch.tlb_fillptr;
383 	setup_pgtables();
384 	gd->arch.tlb_emerg = gd->arch.tlb_addr;
385 	gd->arch.tlb_addr = tlb_addr;
386 }
387 
388 /* to activate the MMU we need to set up virtual memory */
389 __weak void mmu_setup(void)
390 {
391 	int el;
392 
393 	/* Set up page tables only once */
394 	if (!gd->arch.tlb_fillptr)
395 		setup_all_pgtables();
396 
397 	el = current_el();
398 	set_ttbr_tcr_mair(el, gd->arch.tlb_addr, get_tcr(el, NULL, NULL),
399 			  MEMORY_ATTRIBUTES);
400 
401 	/* enable the mmu */
402 	set_sctlr(get_sctlr() | CR_M);
403 }
404 
405 /*
406  * Performs a invalidation of the entire data cache at all levels
407  */
408 void invalidate_dcache_all(void)
409 {
410 	__asm_invalidate_dcache_all();
411 }
412 
413 /*
414  * Performs a clean & invalidation of the entire data cache at all levels.
415  * This function needs to be inline to avoid using stack.
416  * __asm_flush_l3_cache return status of timeout
417  */
418 inline void flush_dcache_all(void)
419 {
420 	int ret;
421 
422 	__asm_flush_dcache_all();
423 	ret = __asm_flush_l3_cache();
424 	if (ret)
425 		debug("flushing dcache returns 0x%x\n", ret);
426 	else
427 		debug("flushing dcache successfully.\n");
428 }
429 
430 /*
431  * Invalidates range in all levels of D-cache/unified cache
432  */
433 void invalidate_dcache_range(unsigned long start, unsigned long stop)
434 {
435 	__asm_flush_dcache_range(start, stop);
436 }
437 
438 /*
439  * Flush range(clean & invalidate) from all levels of D-cache/unified cache
440  */
441 void flush_dcache_range(unsigned long start, unsigned long stop)
442 {
443 	__asm_flush_dcache_range(start, stop);
444 }
445 
446 void dcache_enable(void)
447 {
448 	/* The data cache is not active unless the mmu is enabled */
449 	if (!(get_sctlr() & CR_M)) {
450 		invalidate_dcache_all();
451 		__asm_invalidate_tlb_all();
452 		mmu_setup();
453 	}
454 
455 	set_sctlr(get_sctlr() | CR_C);
456 }
457 
458 void dcache_disable(void)
459 {
460 	uint32_t sctlr;
461 
462 	sctlr = get_sctlr();
463 
464 	/* if cache isn't enabled no need to disable */
465 	if (!(sctlr & CR_C))
466 		return;
467 
468 	set_sctlr(sctlr & ~(CR_C|CR_M));
469 
470 	flush_dcache_all();
471 	__asm_invalidate_tlb_all();
472 }
473 
474 int dcache_status(void)
475 {
476 	return (get_sctlr() & CR_C) != 0;
477 }
478 
479 u64 *__weak arch_get_page_table(void) {
480 	puts("No page table offset defined\n");
481 
482 	return NULL;
483 }
484 
485 static bool is_aligned(u64 addr, u64 size, u64 align)
486 {
487 	return !(addr & (align - 1)) && !(size & (align - 1));
488 }
489 
490 static u64 set_one_region(u64 start, u64 size, u64 attrs, int level)
491 {
492 	int levelshift = level2shift(level);
493 	u64 levelsize = 1ULL << levelshift;
494 	u64 *pte = find_pte(start, level);
495 
496 	/* Can we can just modify the current level block PTE? */
497 	if (is_aligned(start, size, levelsize)) {
498 		*pte &= ~PMD_ATTRINDX_MASK;
499 		*pte |= attrs;
500 		debug("Set attrs=%llx pte=%p level=%d\n", attrs, pte, level);
501 
502 		return levelsize;
503 	}
504 
505 	/* Unaligned or doesn't fit, maybe split block into table */
506 	debug("addr=%llx level=%d pte=%p (%llx)\n", start, level, pte, *pte);
507 
508 	/* Maybe we need to split the block into a table */
509 	if (pte_type(pte) == PTE_TYPE_BLOCK)
510 		split_block(pte, level);
511 
512 	/* And then double-check it became a table or already is one */
513 	if (pte_type(pte) != PTE_TYPE_TABLE)
514 		panic("PTE %p (%llx) for addr=%llx should be a table",
515 		      pte, *pte, start);
516 
517 	/* Roll on to the next page table level */
518 	return 0;
519 }
520 
521 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
522 				     enum dcache_option option)
523 {
524 	u64 attrs = PMD_ATTRINDX(option);
525 	u64 real_start = start;
526 	u64 real_size = size;
527 
528 	debug("start=%lx size=%lx\n", (ulong)start, (ulong)size);
529 
530 	/*
531 	 * We can not modify page tables that we're currently running on,
532 	 * so we first need to switch to the "emergency" page tables where
533 	 * we can safely modify our primary page tables and then switch back
534 	 */
535 	__asm_switch_ttbr(gd->arch.tlb_emerg);
536 
537 	/*
538 	 * Loop through the address range until we find a page granule that fits
539 	 * our alignment constraints, then set it to the new cache attributes
540 	 */
541 	while (size > 0) {
542 		int level;
543 		u64 r;
544 
545 		for (level = 1; level < 4; level++) {
546 			r = set_one_region(start, size, attrs, level);
547 			if (r) {
548 				/* PTE successfully replaced */
549 				size -= r;
550 				start += r;
551 				break;
552 			}
553 		}
554 
555 	}
556 
557 	/* We're done modifying page tables, switch back to our primary ones */
558 	__asm_switch_ttbr(gd->arch.tlb_addr);
559 
560 	/*
561 	 * Make sure there's nothing stale in dcache for a region that might
562 	 * have caches off now
563 	 */
564 	flush_dcache_range(real_start, real_start + real_size);
565 }
566 
567 #else	/* CONFIG_SYS_DCACHE_OFF */
568 
569 /*
570  * For SPL builds, we may want to not have dcache enabled. Any real U-Boot
571  * running however really wants to have dcache and the MMU active. Check that
572  * everything is sane and give the developer a hint if it isn't.
573  */
574 #ifndef CONFIG_SPL_BUILD
575 #error Please describe your MMU layout in CONFIG_SYS_MEM_MAP and enable dcache.
576 #endif
577 
578 void invalidate_dcache_all(void)
579 {
580 }
581 
582 void flush_dcache_all(void)
583 {
584 }
585 
586 void dcache_enable(void)
587 {
588 }
589 
590 void dcache_disable(void)
591 {
592 }
593 
594 int dcache_status(void)
595 {
596 	return 0;
597 }
598 
599 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
600 				     enum dcache_option option)
601 {
602 }
603 
604 #endif	/* CONFIG_SYS_DCACHE_OFF */
605 
606 #ifndef CONFIG_SYS_ICACHE_OFF
607 
608 void icache_enable(void)
609 {
610 	__asm_invalidate_icache_all();
611 	set_sctlr(get_sctlr() | CR_I);
612 }
613 
614 void icache_disable(void)
615 {
616 	set_sctlr(get_sctlr() & ~CR_I);
617 }
618 
619 int icache_status(void)
620 {
621 	return (get_sctlr() & CR_I) != 0;
622 }
623 
624 void invalidate_icache_all(void)
625 {
626 	__asm_invalidate_icache_all();
627 }
628 
629 #else	/* CONFIG_SYS_ICACHE_OFF */
630 
631 void icache_enable(void)
632 {
633 }
634 
635 void icache_disable(void)
636 {
637 }
638 
639 int icache_status(void)
640 {
641 	return 0;
642 }
643 
644 void invalidate_icache_all(void)
645 {
646 }
647 
648 #endif	/* CONFIG_SYS_ICACHE_OFF */
649 
650 /*
651  * Enable dCache & iCache, whether cache is actually enabled
652  * depend on CONFIG_SYS_DCACHE_OFF and CONFIG_SYS_ICACHE_OFF
653  */
654 void __weak enable_caches(void)
655 {
656 	icache_enable();
657 	dcache_enable();
658 }
659