xref: /rk3399_rockchip-uboot/arch/arm/cpu/armv8/cache_v8.c (revision f733d46620d0efb93091f147f81a4bf9588fad3f)
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 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 /* Splits a block PTE into table with subpages spanning the old block */
171 static void split_block(u64 *pte, int level)
172 {
173 	u64 old_pte = *pte;
174 	u64 *new_table;
175 	u64 i = 0;
176 	/* level describes the parent level, we need the child ones */
177 	int levelshift = level2shift(level + 1);
178 
179 	if (pte_type(pte) != PTE_TYPE_BLOCK)
180 		panic("PTE %p (%llx) is not a block. Some driver code wants to "
181 		      "modify dcache settings for an range not covered in "
182 		      "mem_map.", pte, old_pte);
183 
184 	new_table = create_table();
185 	debug("Splitting pte %p (%llx) into %p\n", pte, old_pte, new_table);
186 
187 	for (i = 0; i < MAX_PTE_ENTRIES; i++) {
188 		new_table[i] = old_pte | (i << levelshift);
189 
190 		/* Level 3 block PTEs have the table type */
191 		if ((level + 1) == 3)
192 			new_table[i] |= PTE_TYPE_TABLE;
193 
194 		debug("Setting new_table[%lld] = %llx\n", i, new_table[i]);
195 	}
196 
197 	/* Set the new table into effect */
198 	set_pte_table(pte, new_table);
199 }
200 
201 /* Add one mm_region map entry to the page tables */
202 static void add_map(struct mm_region *map)
203 {
204 	u64 *pte;
205 	u64 addr = map->base;
206 	u64 size = map->size;
207 	u64 attrs = map->attrs | PTE_TYPE_BLOCK | PTE_BLOCK_AF;
208 	u64 blocksize;
209 	int level;
210 	u64 *new_table;
211 
212 	while (size) {
213 		pte = find_pte(addr, 0);
214 		if (pte && (pte_type(pte) == PTE_TYPE_FAULT)) {
215 			debug("Creating table for addr 0x%llx\n", addr);
216 			new_table = create_table();
217 			set_pte_table(pte, new_table);
218 		}
219 
220 		for (level = 1; level < 4; level++) {
221 			pte = find_pte(addr, level);
222 			if (!pte)
223 				panic("pte not found\n");
224 			blocksize = 1ULL << level2shift(level);
225 			debug("Checking if pte fits for addr=%llx size=%llx "
226 			      "blocksize=%llx\n", addr, size, blocksize);
227 			if (size >= blocksize && !(addr & (blocksize - 1))) {
228 				/* Page fits, create block PTE */
229 				debug("Setting PTE %p to block addr=%llx\n",
230 				      pte, addr);
231 				*pte = addr | attrs;
232 				addr += blocksize;
233 				size -= blocksize;
234 				break;
235 			} else if (pte_type(pte) == PTE_TYPE_FAULT) {
236 				/* Page doesn't fit, create subpages */
237 				debug("Creating subtable for addr 0x%llx "
238 				      "blksize=%llx\n", addr, blocksize);
239 				new_table = create_table();
240 				set_pte_table(pte, new_table);
241 			} else if (pte_type(pte) == PTE_TYPE_BLOCK) {
242 				debug("Split block into subtable for addr 0x%llx blksize=0x%llx\n",
243 				      addr, blocksize);
244 				split_block(pte, level);
245 			}
246 		}
247 	}
248 }
249 
250 enum pte_type {
251 	PTE_INVAL,
252 	PTE_BLOCK,
253 	PTE_LEVEL,
254 };
255 
256 /*
257  * This is a recursively called function to count the number of
258  * page tables we need to cover a particular PTE range. If you
259  * call this with level = -1 you basically get the full 48 bit
260  * coverage.
261  */
262 static int count_required_pts(u64 addr, int level, u64 maxaddr)
263 {
264 	int levelshift = level2shift(level);
265 	u64 levelsize = 1ULL << levelshift;
266 	u64 levelmask = levelsize - 1;
267 	u64 levelend = addr + levelsize;
268 	int r = 0;
269 	int i;
270 	enum pte_type pte_type = PTE_INVAL;
271 
272 	for (i = 0; mem_map[i].size || mem_map[i].attrs; i++) {
273 		struct mm_region *map = &mem_map[i];
274 		u64 start = map->base;
275 		u64 end = start + map->size;
276 
277 		/* Check if the PTE would overlap with the map */
278 		if (max(addr, start) <= min(levelend, end)) {
279 			start = max(addr, start);
280 			end = min(levelend, end);
281 
282 			/* We need a sub-pt for this level */
283 			if ((start & levelmask) || (end & levelmask)) {
284 				pte_type = PTE_LEVEL;
285 				break;
286 			}
287 
288 			/* Lv0 can not do block PTEs, so do levels here too */
289 			if (level <= 0) {
290 				pte_type = PTE_LEVEL;
291 				break;
292 			}
293 
294 			/* PTE is active, but fits into a block */
295 			pte_type = PTE_BLOCK;
296 		}
297 	}
298 
299 	/*
300 	 * Block PTEs at this level are already covered by the parent page
301 	 * table, so we only need to count sub page tables.
302 	 */
303 	if (pte_type == PTE_LEVEL) {
304 		int sublevel = level + 1;
305 		u64 sublevelsize = 1ULL << level2shift(sublevel);
306 
307 		/* Account for the new sub page table ... */
308 		r = 1;
309 
310 		/* ... and for all child page tables that one might have */
311 		for (i = 0; i < MAX_PTE_ENTRIES; i++) {
312 			r += count_required_pts(addr, sublevel, maxaddr);
313 			addr += sublevelsize;
314 
315 			if (addr >= maxaddr) {
316 				/*
317 				 * We reached the end of address space, no need
318 				 * to look any further.
319 				 */
320 				break;
321 			}
322 		}
323 	}
324 
325 	return r;
326 }
327 
328 /* Returns the estimated required size of all page tables */
329 __weak u64 get_page_table_size(void)
330 {
331 	u64 one_pt = MAX_PTE_ENTRIES * sizeof(u64);
332 	u64 size = 0;
333 	u64 va_bits;
334 	int start_level = 0;
335 
336 	get_tcr(0, NULL, &va_bits);
337 	if (va_bits < 39)
338 		start_level = 1;
339 
340 	/* Account for all page tables we would need to cover our memory map */
341 	size = one_pt * count_required_pts(0, start_level - 1, 1ULL << va_bits);
342 
343 	/*
344 	 * We need to duplicate our page table once to have an emergency pt to
345 	 * resort to when splitting page tables later on
346 	 */
347 	size *= 2;
348 
349 	/*
350 	 * We may need to split page tables later on if dcache settings change,
351 	 * so reserve up to 4 (random pick) page tables for that.
352 	 */
353 	size += one_pt * 4;
354 
355 	return size;
356 }
357 
358 void setup_pgtables(void)
359 {
360 	int i;
361 
362 	if (!gd->arch.tlb_fillptr || !gd->arch.tlb_addr)
363 		panic("Page table pointer not setup.");
364 
365 	/*
366 	 * Allocate the first level we're on with invalidate entries.
367 	 * If the starting level is 0 (va_bits >= 39), then this is our
368 	 * Lv0 page table, otherwise it's the entry Lv1 page table.
369 	 */
370 	create_table();
371 
372 	/* Now add all MMU table entries one after another to the table */
373 	for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
374 		add_map(&mem_map[i]);
375 }
376 
377 static void setup_all_pgtables(void)
378 {
379 	u64 tlb_addr = gd->arch.tlb_addr;
380 
381 	/* Reset the fill ptr */
382 	gd->arch.tlb_fillptr = tlb_addr;
383 
384 	/* Create normal system page tables */
385 	setup_pgtables();
386 
387 	/* Create emergency page tables */
388 	gd->arch.tlb_addr = gd->arch.tlb_fillptr;
389 	setup_pgtables();
390 	gd->arch.tlb_emerg = gd->arch.tlb_addr;
391 	gd->arch.tlb_addr = tlb_addr;
392 }
393 
394 /* to activate the MMU we need to set up virtual memory */
395 __weak void mmu_setup(void)
396 {
397 	int el;
398 
399 	/* Set up page tables only once */
400 	if (!gd->arch.tlb_fillptr)
401 		setup_all_pgtables();
402 
403 	el = current_el();
404 	set_ttbr_tcr_mair(el, gd->arch.tlb_addr, get_tcr(el, NULL, NULL),
405 			  MEMORY_ATTRIBUTES);
406 
407 	/* enable the mmu */
408 	set_sctlr(get_sctlr() | CR_M);
409 }
410 
411 /*
412  * Performs a invalidation of the entire data cache at all levels
413  */
414 void invalidate_dcache_all(void)
415 {
416 	__asm_invalidate_dcache_all();
417 }
418 
419 /*
420  * Performs a clean & invalidation of the entire data cache at all levels.
421  * This function needs to be inline to avoid using stack.
422  * __asm_flush_l3_cache return status of timeout
423  */
424 inline void flush_dcache_all(void)
425 {
426 	int ret;
427 
428 	__asm_flush_dcache_all();
429 	ret = __asm_flush_l3_cache();
430 	if (ret)
431 		debug("flushing dcache returns 0x%x\n", ret);
432 	else
433 		debug("flushing dcache successfully.\n");
434 }
435 
436 /*
437  * Invalidates range in all levels of D-cache/unified cache
438  */
439 void invalidate_dcache_range(unsigned long start, unsigned long stop)
440 {
441 	__asm_flush_dcache_range(start, stop);
442 }
443 
444 /*
445  * Flush range(clean & invalidate) from all levels of D-cache/unified cache
446  */
447 void flush_dcache_range(unsigned long start, unsigned long stop)
448 {
449 	__asm_flush_dcache_range(start, stop);
450 }
451 
452 void dcache_enable(void)
453 {
454 	/* The data cache is not active unless the mmu is enabled */
455 	if (!(get_sctlr() & CR_M)) {
456 		invalidate_dcache_all();
457 		__asm_invalidate_tlb_all();
458 		mmu_setup();
459 	}
460 
461 	set_sctlr(get_sctlr() | CR_C);
462 }
463 
464 void dcache_disable(void)
465 {
466 	uint32_t sctlr;
467 
468 	sctlr = get_sctlr();
469 
470 	/* if cache isn't enabled no need to disable */
471 	if (!(sctlr & CR_C))
472 		return;
473 
474 	set_sctlr(sctlr & ~(CR_C|CR_M));
475 
476 	flush_dcache_all();
477 	__asm_invalidate_tlb_all();
478 }
479 
480 int dcache_status(void)
481 {
482 	return (get_sctlr() & CR_C) != 0;
483 }
484 
485 u64 *__weak arch_get_page_table(void) {
486 	puts("No page table offset defined\n");
487 
488 	return NULL;
489 }
490 
491 static bool is_aligned(u64 addr, u64 size, u64 align)
492 {
493 	return !(addr & (align - 1)) && !(size & (align - 1));
494 }
495 
496 static u64 set_one_region(u64 start, u64 size, u64 attrs, int level)
497 {
498 	int levelshift = level2shift(level);
499 	u64 levelsize = 1ULL << levelshift;
500 	u64 *pte = find_pte(start, level);
501 
502 	/* Can we can just modify the current level block PTE? */
503 	if (is_aligned(start, size, levelsize)) {
504 		*pte &= ~PMD_ATTRINDX_MASK;
505 		*pte |= attrs;
506 		debug("Set attrs=%llx pte=%p level=%d\n", attrs, pte, level);
507 
508 		return levelsize;
509 	}
510 
511 	/* Unaligned or doesn't fit, maybe split block into table */
512 	debug("addr=%llx level=%d pte=%p (%llx)\n", start, level, pte, *pte);
513 
514 	/* Maybe we need to split the block into a table */
515 	if (pte_type(pte) == PTE_TYPE_BLOCK)
516 		split_block(pte, level);
517 
518 	/* And then double-check it became a table or already is one */
519 	if (pte_type(pte) != PTE_TYPE_TABLE)
520 		panic("PTE %p (%llx) for addr=%llx should be a table",
521 		      pte, *pte, start);
522 
523 	/* Roll on to the next page table level */
524 	return 0;
525 }
526 
527 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
528 				     enum dcache_option option)
529 {
530 	u64 attrs = PMD_ATTRINDX(option);
531 	u64 real_start = start;
532 	u64 real_size = size;
533 
534 	debug("start=%lx size=%lx\n", (ulong)start, (ulong)size);
535 
536 	if (!gd->arch.tlb_emerg)
537 		panic("Emergency page table not setup.");
538 
539 	/*
540 	 * We can not modify page tables that we're currently running on,
541 	 * so we first need to switch to the "emergency" page tables where
542 	 * we can safely modify our primary page tables and then switch back
543 	 */
544 	__asm_switch_ttbr(gd->arch.tlb_emerg);
545 
546 	/*
547 	 * Loop through the address range until we find a page granule that fits
548 	 * our alignment constraints, then set it to the new cache attributes
549 	 */
550 	while (size > 0) {
551 		int level;
552 		u64 r;
553 
554 		for (level = 1; level < 4; level++) {
555 			r = set_one_region(start, size, attrs, level);
556 			if (r) {
557 				/* PTE successfully replaced */
558 				size -= r;
559 				start += r;
560 				break;
561 			}
562 		}
563 
564 	}
565 
566 	/* We're done modifying page tables, switch back to our primary ones */
567 	__asm_switch_ttbr(gd->arch.tlb_addr);
568 
569 	/*
570 	 * Make sure there's nothing stale in dcache for a region that might
571 	 * have caches off now
572 	 */
573 	flush_dcache_range(real_start, real_start + real_size);
574 }
575 
576 #else	/* CONFIG_SYS_DCACHE_OFF */
577 
578 /*
579  * For SPL builds, we may want to not have dcache enabled. Any real U-Boot
580  * running however really wants to have dcache and the MMU active. Check that
581  * everything is sane and give the developer a hint if it isn't.
582  */
583 #ifndef CONFIG_SPL_BUILD
584 #error Please describe your MMU layout in CONFIG_SYS_MEM_MAP and enable dcache.
585 #endif
586 
587 void invalidate_dcache_all(void)
588 {
589 }
590 
591 void flush_dcache_all(void)
592 {
593 }
594 
595 void dcache_enable(void)
596 {
597 }
598 
599 void dcache_disable(void)
600 {
601 }
602 
603 int dcache_status(void)
604 {
605 	return 0;
606 }
607 
608 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
609 				     enum dcache_option option)
610 {
611 }
612 
613 #endif	/* CONFIG_SYS_DCACHE_OFF */
614 
615 #ifndef CONFIG_SYS_ICACHE_OFF
616 
617 void icache_enable(void)
618 {
619 	__asm_invalidate_icache_all();
620 	set_sctlr(get_sctlr() | CR_I);
621 }
622 
623 void icache_disable(void)
624 {
625 	set_sctlr(get_sctlr() & ~CR_I);
626 }
627 
628 int icache_status(void)
629 {
630 	return (get_sctlr() & CR_I) != 0;
631 }
632 
633 void invalidate_icache_all(void)
634 {
635 	__asm_invalidate_icache_all();
636 }
637 
638 #else	/* CONFIG_SYS_ICACHE_OFF */
639 
640 void icache_enable(void)
641 {
642 }
643 
644 void icache_disable(void)
645 {
646 }
647 
648 int icache_status(void)
649 {
650 	return 0;
651 }
652 
653 void invalidate_icache_all(void)
654 {
655 }
656 
657 #endif	/* CONFIG_SYS_ICACHE_OFF */
658 
659 /*
660  * Enable dCache & iCache, whether cache is actually enabled
661  * depend on CONFIG_SYS_DCACHE_OFF and CONFIG_SYS_ICACHE_OFF
662  */
663 void __weak enable_caches(void)
664 {
665 	icache_enable();
666 	dcache_enable();
667 }
668