]> git.ipfire.org Git - people/ms/u-boot.git/blob - arch/arm/cpu/armv8/cache_v8.c
arm64: dts: sun50i: h5: Order nodes in alphabetic for orangepi-prime
[people/ms/u-boot.git] / arch / arm / cpu / armv8 / cache_v8.c
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].virt + 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 virt = map->virt;
206 u64 phys = map->phys;
207 u64 size = map->size;
208 u64 attrs = map->attrs | PTE_TYPE_BLOCK | PTE_BLOCK_AF;
209 u64 blocksize;
210 int level;
211 u64 *new_table;
212
213 while (size) {
214 pte = find_pte(virt, 0);
215 if (pte && (pte_type(pte) == PTE_TYPE_FAULT)) {
216 debug("Creating table for virt 0x%llx\n", virt);
217 new_table = create_table();
218 set_pte_table(pte, new_table);
219 }
220
221 for (level = 1; level < 4; level++) {
222 pte = find_pte(virt, level);
223 if (!pte)
224 panic("pte not found\n");
225
226 blocksize = 1ULL << level2shift(level);
227 debug("Checking if pte fits for virt=%llx size=%llx blocksize=%llx\n",
228 virt, size, blocksize);
229 if (size >= blocksize && !(virt & (blocksize - 1))) {
230 /* Page fits, create block PTE */
231 debug("Setting PTE %p to block virt=%llx\n",
232 pte, virt);
233 *pte = phys | attrs;
234 virt += blocksize;
235 phys += blocksize;
236 size -= blocksize;
237 break;
238 } else if (pte_type(pte) == PTE_TYPE_FAULT) {
239 /* Page doesn't fit, create subpages */
240 debug("Creating subtable for virt 0x%llx blksize=%llx\n",
241 virt, blocksize);
242 new_table = create_table();
243 set_pte_table(pte, new_table);
244 } else if (pte_type(pte) == PTE_TYPE_BLOCK) {
245 debug("Split block into subtable for virt 0x%llx blksize=0x%llx\n",
246 virt, blocksize);
247 split_block(pte, level);
248 }
249 }
250 }
251 }
252
253 enum pte_type {
254 PTE_INVAL,
255 PTE_BLOCK,
256 PTE_LEVEL,
257 };
258
259 /*
260 * This is a recursively called function to count the number of
261 * page tables we need to cover a particular PTE range. If you
262 * call this with level = -1 you basically get the full 48 bit
263 * coverage.
264 */
265 static int count_required_pts(u64 addr, int level, u64 maxaddr)
266 {
267 int levelshift = level2shift(level);
268 u64 levelsize = 1ULL << levelshift;
269 u64 levelmask = levelsize - 1;
270 u64 levelend = addr + levelsize;
271 int r = 0;
272 int i;
273 enum pte_type pte_type = PTE_INVAL;
274
275 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++) {
276 struct mm_region *map = &mem_map[i];
277 u64 start = map->virt;
278 u64 end = start + map->size;
279
280 /* Check if the PTE would overlap with the map */
281 if (max(addr, start) <= min(levelend, end)) {
282 start = max(addr, start);
283 end = min(levelend, end);
284
285 /* We need a sub-pt for this level */
286 if ((start & levelmask) || (end & levelmask)) {
287 pte_type = PTE_LEVEL;
288 break;
289 }
290
291 /* Lv0 can not do block PTEs, so do levels here too */
292 if (level <= 0) {
293 pte_type = PTE_LEVEL;
294 break;
295 }
296
297 /* PTE is active, but fits into a block */
298 pte_type = PTE_BLOCK;
299 }
300 }
301
302 /*
303 * Block PTEs at this level are already covered by the parent page
304 * table, so we only need to count sub page tables.
305 */
306 if (pte_type == PTE_LEVEL) {
307 int sublevel = level + 1;
308 u64 sublevelsize = 1ULL << level2shift(sublevel);
309
310 /* Account for the new sub page table ... */
311 r = 1;
312
313 /* ... and for all child page tables that one might have */
314 for (i = 0; i < MAX_PTE_ENTRIES; i++) {
315 r += count_required_pts(addr, sublevel, maxaddr);
316 addr += sublevelsize;
317
318 if (addr >= maxaddr) {
319 /*
320 * We reached the end of address space, no need
321 * to look any further.
322 */
323 break;
324 }
325 }
326 }
327
328 return r;
329 }
330
331 /* Returns the estimated required size of all page tables */
332 __weak u64 get_page_table_size(void)
333 {
334 u64 one_pt = MAX_PTE_ENTRIES * sizeof(u64);
335 u64 size = 0;
336 u64 va_bits;
337 int start_level = 0;
338
339 get_tcr(0, NULL, &va_bits);
340 if (va_bits < 39)
341 start_level = 1;
342
343 /* Account for all page tables we would need to cover our memory map */
344 size = one_pt * count_required_pts(0, start_level - 1, 1ULL << va_bits);
345
346 /*
347 * We need to duplicate our page table once to have an emergency pt to
348 * resort to when splitting page tables later on
349 */
350 size *= 2;
351
352 /*
353 * We may need to split page tables later on if dcache settings change,
354 * so reserve up to 4 (random pick) page tables for that.
355 */
356 size += one_pt * 4;
357
358 return size;
359 }
360
361 void setup_pgtables(void)
362 {
363 int i;
364
365 if (!gd->arch.tlb_fillptr || !gd->arch.tlb_addr)
366 panic("Page table pointer not setup.");
367
368 /*
369 * Allocate the first level we're on with invalidate entries.
370 * If the starting level is 0 (va_bits >= 39), then this is our
371 * Lv0 page table, otherwise it's the entry Lv1 page table.
372 */
373 create_table();
374
375 /* Now add all MMU table entries one after another to the table */
376 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
377 add_map(&mem_map[i]);
378 }
379
380 static void setup_all_pgtables(void)
381 {
382 u64 tlb_addr = gd->arch.tlb_addr;
383 u64 tlb_size = gd->arch.tlb_size;
384
385 /* Reset the fill ptr */
386 gd->arch.tlb_fillptr = tlb_addr;
387
388 /* Create normal system page tables */
389 setup_pgtables();
390
391 /* Create emergency page tables */
392 gd->arch.tlb_size -= (uintptr_t)gd->arch.tlb_fillptr -
393 (uintptr_t)gd->arch.tlb_addr;
394 gd->arch.tlb_addr = gd->arch.tlb_fillptr;
395 setup_pgtables();
396 gd->arch.tlb_emerg = gd->arch.tlb_addr;
397 gd->arch.tlb_addr = tlb_addr;
398 gd->arch.tlb_size = tlb_size;
399 }
400
401 /* to activate the MMU we need to set up virtual memory */
402 __weak void mmu_setup(void)
403 {
404 int el;
405
406 /* Set up page tables only once */
407 if (!gd->arch.tlb_fillptr)
408 setup_all_pgtables();
409
410 el = current_el();
411 set_ttbr_tcr_mair(el, gd->arch.tlb_addr, get_tcr(el, NULL, NULL),
412 MEMORY_ATTRIBUTES);
413
414 /* enable the mmu */
415 set_sctlr(get_sctlr() | CR_M);
416 }
417
418 /*
419 * Performs a invalidation of the entire data cache at all levels
420 */
421 void invalidate_dcache_all(void)
422 {
423 __asm_invalidate_dcache_all();
424 __asm_invalidate_l3_dcache();
425 }
426
427 /*
428 * Performs a clean & invalidation of the entire data cache at all levels.
429 * This function needs to be inline to avoid using stack.
430 * __asm_flush_l3_dcache return status of timeout
431 */
432 inline void flush_dcache_all(void)
433 {
434 int ret;
435
436 __asm_flush_dcache_all();
437 ret = __asm_flush_l3_dcache();
438 if (ret)
439 debug("flushing dcache returns 0x%x\n", ret);
440 else
441 debug("flushing dcache successfully.\n");
442 }
443
444 /*
445 * Invalidates range in all levels of D-cache/unified cache
446 */
447 void invalidate_dcache_range(unsigned long start, unsigned long stop)
448 {
449 __asm_invalidate_dcache_range(start, stop);
450 }
451
452 /*
453 * Flush range(clean & invalidate) from all levels of D-cache/unified cache
454 */
455 void flush_dcache_range(unsigned long start, unsigned long stop)
456 {
457 __asm_flush_dcache_range(start, stop);
458 }
459
460 void dcache_enable(void)
461 {
462 /* The data cache is not active unless the mmu is enabled */
463 if (!(get_sctlr() & CR_M)) {
464 invalidate_dcache_all();
465 __asm_invalidate_tlb_all();
466 mmu_setup();
467 }
468
469 set_sctlr(get_sctlr() | CR_C);
470 }
471
472 void dcache_disable(void)
473 {
474 uint32_t sctlr;
475
476 sctlr = get_sctlr();
477
478 /* if cache isn't enabled no need to disable */
479 if (!(sctlr & CR_C))
480 return;
481
482 set_sctlr(sctlr & ~(CR_C|CR_M));
483
484 flush_dcache_all();
485 __asm_invalidate_tlb_all();
486 }
487
488 int dcache_status(void)
489 {
490 return (get_sctlr() & CR_C) != 0;
491 }
492
493 u64 *__weak arch_get_page_table(void) {
494 puts("No page table offset defined\n");
495
496 return NULL;
497 }
498
499 static bool is_aligned(u64 addr, u64 size, u64 align)
500 {
501 return !(addr & (align - 1)) && !(size & (align - 1));
502 }
503
504 /* Use flag to indicate if attrs has more than d-cache attributes */
505 static u64 set_one_region(u64 start, u64 size, u64 attrs, bool flag, int level)
506 {
507 int levelshift = level2shift(level);
508 u64 levelsize = 1ULL << levelshift;
509 u64 *pte = find_pte(start, level);
510
511 /* Can we can just modify the current level block PTE? */
512 if (is_aligned(start, size, levelsize)) {
513 if (flag) {
514 *pte &= ~PMD_ATTRMASK;
515 *pte |= attrs & PMD_ATTRMASK;
516 } else {
517 *pte &= ~PMD_ATTRINDX_MASK;
518 *pte |= attrs & PMD_ATTRINDX_MASK;
519 }
520 debug("Set attrs=%llx pte=%p level=%d\n", attrs, pte, level);
521
522 return levelsize;
523 }
524
525 /* Unaligned or doesn't fit, maybe split block into table */
526 debug("addr=%llx level=%d pte=%p (%llx)\n", start, level, pte, *pte);
527
528 /* Maybe we need to split the block into a table */
529 if (pte_type(pte) == PTE_TYPE_BLOCK)
530 split_block(pte, level);
531
532 /* And then double-check it became a table or already is one */
533 if (pte_type(pte) != PTE_TYPE_TABLE)
534 panic("PTE %p (%llx) for addr=%llx should be a table",
535 pte, *pte, start);
536
537 /* Roll on to the next page table level */
538 return 0;
539 }
540
541 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
542 enum dcache_option option)
543 {
544 u64 attrs = PMD_ATTRINDX(option);
545 u64 real_start = start;
546 u64 real_size = size;
547
548 debug("start=%lx size=%lx\n", (ulong)start, (ulong)size);
549
550 if (!gd->arch.tlb_emerg)
551 panic("Emergency page table not setup.");
552
553 /*
554 * We can not modify page tables that we're currently running on,
555 * so we first need to switch to the "emergency" page tables where
556 * we can safely modify our primary page tables and then switch back
557 */
558 __asm_switch_ttbr(gd->arch.tlb_emerg);
559
560 /*
561 * Loop through the address range until we find a page granule that fits
562 * our alignment constraints, then set it to the new cache attributes
563 */
564 while (size > 0) {
565 int level;
566 u64 r;
567
568 for (level = 1; level < 4; level++) {
569 /* Set d-cache attributes only */
570 r = set_one_region(start, size, attrs, false, level);
571 if (r) {
572 /* PTE successfully replaced */
573 size -= r;
574 start += r;
575 break;
576 }
577 }
578
579 }
580
581 /* We're done modifying page tables, switch back to our primary ones */
582 __asm_switch_ttbr(gd->arch.tlb_addr);
583
584 /*
585 * Make sure there's nothing stale in dcache for a region that might
586 * have caches off now
587 */
588 flush_dcache_range(real_start, real_start + real_size);
589 }
590
591 /*
592 * Modify MMU table for a region with updated PXN/UXN/Memory type/valid bits.
593 * The procecess is break-before-make. The target region will be marked as
594 * invalid during the process of changing.
595 */
596 void mmu_change_region_attr(phys_addr_t addr, size_t siz, u64 attrs)
597 {
598 int level;
599 u64 r, size, start;
600
601 start = addr;
602 size = siz;
603 /*
604 * Loop through the address range until we find a page granule that fits
605 * our alignment constraints, then set it to "invalid".
606 */
607 while (size > 0) {
608 for (level = 1; level < 4; level++) {
609 /* Set PTE to fault */
610 r = set_one_region(start, size, PTE_TYPE_FAULT, true,
611 level);
612 if (r) {
613 /* PTE successfully invalidated */
614 size -= r;
615 start += r;
616 break;
617 }
618 }
619 }
620
621 flush_dcache_range(gd->arch.tlb_addr,
622 gd->arch.tlb_addr + gd->arch.tlb_size);
623 __asm_invalidate_tlb_all();
624
625 /*
626 * Loop through the address range until we find a page granule that fits
627 * our alignment constraints, then set it to the new cache attributes
628 */
629 start = addr;
630 size = siz;
631 while (size > 0) {
632 for (level = 1; level < 4; level++) {
633 /* Set PTE to new attributes */
634 r = set_one_region(start, size, attrs, true, level);
635 if (r) {
636 /* PTE successfully updated */
637 size -= r;
638 start += r;
639 break;
640 }
641 }
642 }
643 flush_dcache_range(gd->arch.tlb_addr,
644 gd->arch.tlb_addr + gd->arch.tlb_size);
645 __asm_invalidate_tlb_all();
646 }
647
648 #else /* CONFIG_SYS_DCACHE_OFF */
649
650 /*
651 * For SPL builds, we may want to not have dcache enabled. Any real U-Boot
652 * running however really wants to have dcache and the MMU active. Check that
653 * everything is sane and give the developer a hint if it isn't.
654 */
655 #ifndef CONFIG_SPL_BUILD
656 #error Please describe your MMU layout in CONFIG_SYS_MEM_MAP and enable dcache.
657 #endif
658
659 void invalidate_dcache_all(void)
660 {
661 }
662
663 void flush_dcache_all(void)
664 {
665 }
666
667 void dcache_enable(void)
668 {
669 }
670
671 void dcache_disable(void)
672 {
673 }
674
675 int dcache_status(void)
676 {
677 return 0;
678 }
679
680 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
681 enum dcache_option option)
682 {
683 }
684
685 #endif /* CONFIG_SYS_DCACHE_OFF */
686
687 #ifndef CONFIG_SYS_ICACHE_OFF
688
689 void icache_enable(void)
690 {
691 invalidate_icache_all();
692 set_sctlr(get_sctlr() | CR_I);
693 }
694
695 void icache_disable(void)
696 {
697 set_sctlr(get_sctlr() & ~CR_I);
698 }
699
700 int icache_status(void)
701 {
702 return (get_sctlr() & CR_I) != 0;
703 }
704
705 void invalidate_icache_all(void)
706 {
707 __asm_invalidate_icache_all();
708 __asm_invalidate_l3_icache();
709 }
710
711 #else /* CONFIG_SYS_ICACHE_OFF */
712
713 void icache_enable(void)
714 {
715 }
716
717 void icache_disable(void)
718 {
719 }
720
721 int icache_status(void)
722 {
723 return 0;
724 }
725
726 void invalidate_icache_all(void)
727 {
728 }
729
730 #endif /* CONFIG_SYS_ICACHE_OFF */
731
732 /*
733 * Enable dCache & iCache, whether cache is actually enabled
734 * depend on CONFIG_SYS_DCACHE_OFF and CONFIG_SYS_ICACHE_OFF
735 */
736 void __weak enable_caches(void)
737 {
738 icache_enable();
739 dcache_enable();
740 }