]> git.ipfire.org Git - thirdparty/u-boot.git/blob - arch/arm/cpu/armv8/cache_v8.c
common: Drop net.h from common header
[thirdparty/u-boot.git] / arch / arm / cpu / armv8 / cache_v8.c
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * (C) Copyright 2013
4 * David Feng <fenghua@phytium.com.cn>
5 *
6 * (C) Copyright 2016
7 * Alexander Graf <agraf@suse.de>
8 */
9
10 #include <common.h>
11 #include <cpu_func.h>
12 #include <hang.h>
13 #include <asm/cache.h>
14 #include <asm/system.h>
15 #include <asm/armv8/mmu.h>
16
17 DECLARE_GLOBAL_DATA_PTR;
18
19 #if !CONFIG_IS_ENABLED(SYS_DCACHE_OFF)
20
21 /*
22 * With 4k page granule, a virtual address is split into 4 lookup parts
23 * spanning 9 bits each:
24 *
25 * _______________________________________________
26 * | | | | | | |
27 * | 0 | Lv0 | Lv1 | Lv2 | Lv3 | off |
28 * |_______|_______|_______|_______|_______|_______|
29 * 63-48 47-39 38-30 29-21 20-12 11-00
30 *
31 * mask page size
32 *
33 * Lv0: FF8000000000 --
34 * Lv1: 7FC0000000 1G
35 * Lv2: 3FE00000 2M
36 * Lv3: 1FF000 4K
37 * off: FFF
38 */
39
40 u64 get_tcr(int el, u64 *pips, u64 *pva_bits)
41 {
42 u64 max_addr = 0;
43 u64 ips, va_bits;
44 u64 tcr;
45 int i;
46
47 /* Find the largest address we need to support */
48 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
49 max_addr = max(max_addr, mem_map[i].virt + mem_map[i].size);
50
51 /* Calculate the maximum physical (and thus virtual) address */
52 if (max_addr > (1ULL << 44)) {
53 ips = 5;
54 va_bits = 48;
55 } else if (max_addr > (1ULL << 42)) {
56 ips = 4;
57 va_bits = 44;
58 } else if (max_addr > (1ULL << 40)) {
59 ips = 3;
60 va_bits = 42;
61 } else if (max_addr > (1ULL << 36)) {
62 ips = 2;
63 va_bits = 40;
64 } else if (max_addr > (1ULL << 32)) {
65 ips = 1;
66 va_bits = 36;
67 } else {
68 ips = 0;
69 va_bits = 32;
70 }
71
72 if (el == 1) {
73 tcr = TCR_EL1_RSVD | (ips << 32) | TCR_EPD1_DISABLE;
74 } else if (el == 2) {
75 tcr = TCR_EL2_RSVD | (ips << 16);
76 } else {
77 tcr = TCR_EL3_RSVD | (ips << 16);
78 }
79
80 /* PTWs cacheable, inner/outer WBWA and inner shareable */
81 tcr |= TCR_TG0_4K | TCR_SHARED_INNER | TCR_ORGN_WBWA | TCR_IRGN_WBWA;
82 tcr |= TCR_T0SZ(va_bits);
83
84 if (pips)
85 *pips = ips;
86 if (pva_bits)
87 *pva_bits = va_bits;
88
89 return tcr;
90 }
91
92 #define MAX_PTE_ENTRIES 512
93
94 static int pte_type(u64 *pte)
95 {
96 return *pte & PTE_TYPE_MASK;
97 }
98
99 /* Returns the LSB number for a PTE on level <level> */
100 static int level2shift(int level)
101 {
102 /* Page is 12 bits wide, every level translates 9 bits */
103 return (12 + 9 * (3 - level));
104 }
105
106 static u64 *find_pte(u64 addr, int level)
107 {
108 int start_level = 0;
109 u64 *pte;
110 u64 idx;
111 u64 va_bits;
112 int i;
113
114 debug("addr=%llx level=%d\n", addr, level);
115
116 get_tcr(0, NULL, &va_bits);
117 if (va_bits < 39)
118 start_level = 1;
119
120 if (level < start_level)
121 return NULL;
122
123 /* Walk through all page table levels to find our PTE */
124 pte = (u64*)gd->arch.tlb_addr;
125 for (i = start_level; i < 4; i++) {
126 idx = (addr >> level2shift(i)) & 0x1FF;
127 pte += idx;
128 debug("idx=%llx PTE %p at level %d: %llx\n", idx, pte, i, *pte);
129
130 /* Found it */
131 if (i == level)
132 return pte;
133 /* PTE is no table (either invalid or block), can't traverse */
134 if (pte_type(pte) != PTE_TYPE_TABLE)
135 return NULL;
136 /* Off to the next level */
137 pte = (u64*)(*pte & 0x0000fffffffff000ULL);
138 }
139
140 /* Should never reach here */
141 return NULL;
142 }
143
144 /* Returns and creates a new full table (512 entries) */
145 static u64 *create_table(void)
146 {
147 u64 *new_table = (u64*)gd->arch.tlb_fillptr;
148 u64 pt_len = MAX_PTE_ENTRIES * sizeof(u64);
149
150 /* Allocate MAX_PTE_ENTRIES pte entries */
151 gd->arch.tlb_fillptr += pt_len;
152
153 if (gd->arch.tlb_fillptr - gd->arch.tlb_addr > gd->arch.tlb_size)
154 panic("Insufficient RAM for page table: 0x%lx > 0x%lx. "
155 "Please increase the size in get_page_table_size()",
156 gd->arch.tlb_fillptr - gd->arch.tlb_addr,
157 gd->arch.tlb_size);
158
159 /* Mark all entries as invalid */
160 memset(new_table, 0, pt_len);
161
162 return new_table;
163 }
164
165 static void set_pte_table(u64 *pte, u64 *table)
166 {
167 /* Point *pte to the new table */
168 debug("Setting %p to addr=%p\n", pte, table);
169 *pte = PTE_TYPE_TABLE | (ulong)table;
170 }
171
172 /* Splits a block PTE into table with subpages spanning the old block */
173 static void split_block(u64 *pte, int level)
174 {
175 u64 old_pte = *pte;
176 u64 *new_table;
177 u64 i = 0;
178 /* level describes the parent level, we need the child ones */
179 int levelshift = level2shift(level + 1);
180
181 if (pte_type(pte) != PTE_TYPE_BLOCK)
182 panic("PTE %p (%llx) is not a block. Some driver code wants to "
183 "modify dcache settings for an range not covered in "
184 "mem_map.", pte, old_pte);
185
186 new_table = create_table();
187 debug("Splitting pte %p (%llx) into %p\n", pte, old_pte, new_table);
188
189 for (i = 0; i < MAX_PTE_ENTRIES; i++) {
190 new_table[i] = old_pte | (i << levelshift);
191
192 /* Level 3 block PTEs have the table type */
193 if ((level + 1) == 3)
194 new_table[i] |= PTE_TYPE_TABLE;
195
196 debug("Setting new_table[%lld] = %llx\n", i, new_table[i]);
197 }
198
199 /* Set the new table into effect */
200 set_pte_table(pte, new_table);
201 }
202
203 /* Add one mm_region map entry to the page tables */
204 static void add_map(struct mm_region *map)
205 {
206 u64 *pte;
207 u64 virt = map->virt;
208 u64 phys = map->phys;
209 u64 size = map->size;
210 u64 attrs = map->attrs | PTE_TYPE_BLOCK | PTE_BLOCK_AF;
211 u64 blocksize;
212 int level;
213 u64 *new_table;
214
215 while (size) {
216 pte = find_pte(virt, 0);
217 if (pte && (pte_type(pte) == PTE_TYPE_FAULT)) {
218 debug("Creating table for virt 0x%llx\n", virt);
219 new_table = create_table();
220 set_pte_table(pte, new_table);
221 }
222
223 for (level = 1; level < 4; level++) {
224 pte = find_pte(virt, level);
225 if (!pte)
226 panic("pte not found\n");
227
228 blocksize = 1ULL << level2shift(level);
229 debug("Checking if pte fits for virt=%llx size=%llx blocksize=%llx\n",
230 virt, size, blocksize);
231 if (size >= blocksize && !(virt & (blocksize - 1))) {
232 /* Page fits, create block PTE */
233 debug("Setting PTE %p to block virt=%llx\n",
234 pte, virt);
235 if (level == 3)
236 *pte = phys | attrs | PTE_TYPE_PAGE;
237 else
238 *pte = phys | attrs;
239 virt += blocksize;
240 phys += blocksize;
241 size -= blocksize;
242 break;
243 } else if (pte_type(pte) == PTE_TYPE_FAULT) {
244 /* Page doesn't fit, create subpages */
245 debug("Creating subtable for virt 0x%llx blksize=%llx\n",
246 virt, blocksize);
247 new_table = create_table();
248 set_pte_table(pte, new_table);
249 } else if (pte_type(pte) == PTE_TYPE_BLOCK) {
250 debug("Split block into subtable for virt 0x%llx blksize=0x%llx\n",
251 virt, blocksize);
252 split_block(pte, level);
253 }
254 }
255 }
256 }
257
258 enum pte_type {
259 PTE_INVAL,
260 PTE_BLOCK,
261 PTE_LEVEL,
262 };
263
264 /*
265 * This is a recursively called function to count the number of
266 * page tables we need to cover a particular PTE range. If you
267 * call this with level = -1 you basically get the full 48 bit
268 * coverage.
269 */
270 static int count_required_pts(u64 addr, int level, u64 maxaddr)
271 {
272 int levelshift = level2shift(level);
273 u64 levelsize = 1ULL << levelshift;
274 u64 levelmask = levelsize - 1;
275 u64 levelend = addr + levelsize;
276 int r = 0;
277 int i;
278 enum pte_type pte_type = PTE_INVAL;
279
280 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++) {
281 struct mm_region *map = &mem_map[i];
282 u64 start = map->virt;
283 u64 end = start + map->size;
284
285 /* Check if the PTE would overlap with the map */
286 if (max(addr, start) <= min(levelend, end)) {
287 start = max(addr, start);
288 end = min(levelend, end);
289
290 /* We need a sub-pt for this level */
291 if ((start & levelmask) || (end & levelmask)) {
292 pte_type = PTE_LEVEL;
293 break;
294 }
295
296 /* Lv0 can not do block PTEs, so do levels here too */
297 if (level <= 0) {
298 pte_type = PTE_LEVEL;
299 break;
300 }
301
302 /* PTE is active, but fits into a block */
303 pte_type = PTE_BLOCK;
304 }
305 }
306
307 /*
308 * Block PTEs at this level are already covered by the parent page
309 * table, so we only need to count sub page tables.
310 */
311 if (pte_type == PTE_LEVEL) {
312 int sublevel = level + 1;
313 u64 sublevelsize = 1ULL << level2shift(sublevel);
314
315 /* Account for the new sub page table ... */
316 r = 1;
317
318 /* ... and for all child page tables that one might have */
319 for (i = 0; i < MAX_PTE_ENTRIES; i++) {
320 r += count_required_pts(addr, sublevel, maxaddr);
321 addr += sublevelsize;
322
323 if (addr >= maxaddr) {
324 /*
325 * We reached the end of address space, no need
326 * to look any further.
327 */
328 break;
329 }
330 }
331 }
332
333 return r;
334 }
335
336 /* Returns the estimated required size of all page tables */
337 __weak u64 get_page_table_size(void)
338 {
339 u64 one_pt = MAX_PTE_ENTRIES * sizeof(u64);
340 u64 size = 0;
341 u64 va_bits;
342 int start_level = 0;
343
344 get_tcr(0, NULL, &va_bits);
345 if (va_bits < 39)
346 start_level = 1;
347
348 /* Account for all page tables we would need to cover our memory map */
349 size = one_pt * count_required_pts(0, start_level - 1, 1ULL << va_bits);
350
351 /*
352 * We need to duplicate our page table once to have an emergency pt to
353 * resort to when splitting page tables later on
354 */
355 size *= 2;
356
357 /*
358 * We may need to split page tables later on if dcache settings change,
359 * so reserve up to 4 (random pick) page tables for that.
360 */
361 size += one_pt * 4;
362
363 return size;
364 }
365
366 void setup_pgtables(void)
367 {
368 int i;
369
370 if (!gd->arch.tlb_fillptr || !gd->arch.tlb_addr)
371 panic("Page table pointer not setup.");
372
373 /*
374 * Allocate the first level we're on with invalidate entries.
375 * If the starting level is 0 (va_bits >= 39), then this is our
376 * Lv0 page table, otherwise it's the entry Lv1 page table.
377 */
378 create_table();
379
380 /* Now add all MMU table entries one after another to the table */
381 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
382 add_map(&mem_map[i]);
383 }
384
385 static void setup_all_pgtables(void)
386 {
387 u64 tlb_addr = gd->arch.tlb_addr;
388 u64 tlb_size = gd->arch.tlb_size;
389
390 /* Reset the fill ptr */
391 gd->arch.tlb_fillptr = tlb_addr;
392
393 /* Create normal system page tables */
394 setup_pgtables();
395
396 /* Create emergency page tables */
397 gd->arch.tlb_size -= (uintptr_t)gd->arch.tlb_fillptr -
398 (uintptr_t)gd->arch.tlb_addr;
399 gd->arch.tlb_addr = gd->arch.tlb_fillptr;
400 setup_pgtables();
401 gd->arch.tlb_emerg = gd->arch.tlb_addr;
402 gd->arch.tlb_addr = tlb_addr;
403 gd->arch.tlb_size = tlb_size;
404 }
405
406 /* to activate the MMU we need to set up virtual memory */
407 __weak void mmu_setup(void)
408 {
409 int el;
410
411 /* Set up page tables only once */
412 if (!gd->arch.tlb_fillptr)
413 setup_all_pgtables();
414
415 el = current_el();
416 set_ttbr_tcr_mair(el, gd->arch.tlb_addr, get_tcr(el, NULL, NULL),
417 MEMORY_ATTRIBUTES);
418
419 /* enable the mmu */
420 set_sctlr(get_sctlr() | CR_M);
421 }
422
423 /*
424 * Performs a invalidation of the entire data cache at all levels
425 */
426 void invalidate_dcache_all(void)
427 {
428 __asm_invalidate_dcache_all();
429 __asm_invalidate_l3_dcache();
430 }
431
432 /*
433 * Performs a clean & invalidation of the entire data cache at all levels.
434 * This function needs to be inline to avoid using stack.
435 * __asm_flush_l3_dcache return status of timeout
436 */
437 inline void flush_dcache_all(void)
438 {
439 int ret;
440
441 __asm_flush_dcache_all();
442 ret = __asm_flush_l3_dcache();
443 if (ret)
444 debug("flushing dcache returns 0x%x\n", ret);
445 else
446 debug("flushing dcache successfully.\n");
447 }
448
449 #ifndef CONFIG_SYS_DISABLE_DCACHE_OPS
450 /*
451 * Invalidates range in all levels of D-cache/unified cache
452 */
453 void invalidate_dcache_range(unsigned long start, unsigned long stop)
454 {
455 __asm_invalidate_dcache_range(start, stop);
456 }
457
458 /*
459 * Flush range(clean & invalidate) from all levels of D-cache/unified cache
460 */
461 void flush_dcache_range(unsigned long start, unsigned long stop)
462 {
463 __asm_flush_dcache_range(start, stop);
464 }
465 #else
466 void invalidate_dcache_range(unsigned long start, unsigned long stop)
467 {
468 }
469
470 void flush_dcache_range(unsigned long start, unsigned long stop)
471 {
472 }
473 #endif /* CONFIG_SYS_DISABLE_DCACHE_OPS */
474
475 void dcache_enable(void)
476 {
477 /* The data cache is not active unless the mmu is enabled */
478 if (!(get_sctlr() & CR_M)) {
479 invalidate_dcache_all();
480 __asm_invalidate_tlb_all();
481 mmu_setup();
482 }
483
484 set_sctlr(get_sctlr() | CR_C);
485 }
486
487 void dcache_disable(void)
488 {
489 uint32_t sctlr;
490
491 sctlr = get_sctlr();
492
493 /* if cache isn't enabled no need to disable */
494 if (!(sctlr & CR_C))
495 return;
496
497 set_sctlr(sctlr & ~(CR_C|CR_M));
498
499 flush_dcache_all();
500 __asm_invalidate_tlb_all();
501 }
502
503 int dcache_status(void)
504 {
505 return (get_sctlr() & CR_C) != 0;
506 }
507
508 u64 *__weak arch_get_page_table(void) {
509 puts("No page table offset defined\n");
510
511 return NULL;
512 }
513
514 static bool is_aligned(u64 addr, u64 size, u64 align)
515 {
516 return !(addr & (align - 1)) && !(size & (align - 1));
517 }
518
519 /* Use flag to indicate if attrs has more than d-cache attributes */
520 static u64 set_one_region(u64 start, u64 size, u64 attrs, bool flag, int level)
521 {
522 int levelshift = level2shift(level);
523 u64 levelsize = 1ULL << levelshift;
524 u64 *pte = find_pte(start, level);
525
526 /* Can we can just modify the current level block PTE? */
527 if (is_aligned(start, size, levelsize)) {
528 if (flag) {
529 *pte &= ~PMD_ATTRMASK;
530 *pte |= attrs & PMD_ATTRMASK;
531 } else {
532 *pte &= ~PMD_ATTRINDX_MASK;
533 *pte |= attrs & PMD_ATTRINDX_MASK;
534 }
535 debug("Set attrs=%llx pte=%p level=%d\n", attrs, pte, level);
536
537 return levelsize;
538 }
539
540 /* Unaligned or doesn't fit, maybe split block into table */
541 debug("addr=%llx level=%d pte=%p (%llx)\n", start, level, pte, *pte);
542
543 /* Maybe we need to split the block into a table */
544 if (pte_type(pte) == PTE_TYPE_BLOCK)
545 split_block(pte, level);
546
547 /* And then double-check it became a table or already is one */
548 if (pte_type(pte) != PTE_TYPE_TABLE)
549 panic("PTE %p (%llx) for addr=%llx should be a table",
550 pte, *pte, start);
551
552 /* Roll on to the next page table level */
553 return 0;
554 }
555
556 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
557 enum dcache_option option)
558 {
559 u64 attrs = PMD_ATTRINDX(option);
560 u64 real_start = start;
561 u64 real_size = size;
562
563 debug("start=%lx size=%lx\n", (ulong)start, (ulong)size);
564
565 if (!gd->arch.tlb_emerg)
566 panic("Emergency page table not setup.");
567
568 /*
569 * We can not modify page tables that we're currently running on,
570 * so we first need to switch to the "emergency" page tables where
571 * we can safely modify our primary page tables and then switch back
572 */
573 __asm_switch_ttbr(gd->arch.tlb_emerg);
574
575 /*
576 * Loop through the address range until we find a page granule that fits
577 * our alignment constraints, then set it to the new cache attributes
578 */
579 while (size > 0) {
580 int level;
581 u64 r;
582
583 for (level = 1; level < 4; level++) {
584 /* Set d-cache attributes only */
585 r = set_one_region(start, size, attrs, false, level);
586 if (r) {
587 /* PTE successfully replaced */
588 size -= r;
589 start += r;
590 break;
591 }
592 }
593
594 }
595
596 /* We're done modifying page tables, switch back to our primary ones */
597 __asm_switch_ttbr(gd->arch.tlb_addr);
598
599 /*
600 * Make sure there's nothing stale in dcache for a region that might
601 * have caches off now
602 */
603 flush_dcache_range(real_start, real_start + real_size);
604 }
605
606 /*
607 * Modify MMU table for a region with updated PXN/UXN/Memory type/valid bits.
608 * The procecess is break-before-make. The target region will be marked as
609 * invalid during the process of changing.
610 */
611 void mmu_change_region_attr(phys_addr_t addr, size_t siz, u64 attrs)
612 {
613 int level;
614 u64 r, size, start;
615
616 start = addr;
617 size = siz;
618 /*
619 * Loop through the address range until we find a page granule that fits
620 * our alignment constraints, then set it to "invalid".
621 */
622 while (size > 0) {
623 for (level = 1; level < 4; level++) {
624 /* Set PTE to fault */
625 r = set_one_region(start, size, PTE_TYPE_FAULT, true,
626 level);
627 if (r) {
628 /* PTE successfully invalidated */
629 size -= r;
630 start += r;
631 break;
632 }
633 }
634 }
635
636 flush_dcache_range(gd->arch.tlb_addr,
637 gd->arch.tlb_addr + gd->arch.tlb_size);
638 __asm_invalidate_tlb_all();
639
640 /*
641 * Loop through the address range until we find a page granule that fits
642 * our alignment constraints, then set it to the new cache attributes
643 */
644 start = addr;
645 size = siz;
646 while (size > 0) {
647 for (level = 1; level < 4; level++) {
648 /* Set PTE to new attributes */
649 r = set_one_region(start, size, attrs, true, level);
650 if (r) {
651 /* PTE successfully updated */
652 size -= r;
653 start += r;
654 break;
655 }
656 }
657 }
658 flush_dcache_range(gd->arch.tlb_addr,
659 gd->arch.tlb_addr + gd->arch.tlb_size);
660 __asm_invalidate_tlb_all();
661 }
662
663 #else /* !CONFIG_IS_ENABLED(SYS_DCACHE_OFF) */
664
665 /*
666 * For SPL builds, we may want to not have dcache enabled. Any real U-Boot
667 * running however really wants to have dcache and the MMU active. Check that
668 * everything is sane and give the developer a hint if it isn't.
669 */
670 #ifndef CONFIG_SPL_BUILD
671 #error Please describe your MMU layout in CONFIG_SYS_MEM_MAP and enable dcache.
672 #endif
673
674 void invalidate_dcache_all(void)
675 {
676 }
677
678 void flush_dcache_all(void)
679 {
680 }
681
682 void dcache_enable(void)
683 {
684 }
685
686 void dcache_disable(void)
687 {
688 }
689
690 int dcache_status(void)
691 {
692 return 0;
693 }
694
695 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
696 enum dcache_option option)
697 {
698 }
699
700 #endif /* !CONFIG_IS_ENABLED(SYS_DCACHE_OFF) */
701
702 #if !CONFIG_IS_ENABLED(SYS_ICACHE_OFF)
703
704 void icache_enable(void)
705 {
706 invalidate_icache_all();
707 set_sctlr(get_sctlr() | CR_I);
708 }
709
710 void icache_disable(void)
711 {
712 set_sctlr(get_sctlr() & ~CR_I);
713 }
714
715 int icache_status(void)
716 {
717 return (get_sctlr() & CR_I) != 0;
718 }
719
720 void invalidate_icache_all(void)
721 {
722 __asm_invalidate_icache_all();
723 __asm_invalidate_l3_icache();
724 }
725
726 #else /* !CONFIG_IS_ENABLED(SYS_ICACHE_OFF) */
727
728 void icache_enable(void)
729 {
730 }
731
732 void icache_disable(void)
733 {
734 }
735
736 int icache_status(void)
737 {
738 return 0;
739 }
740
741 void invalidate_icache_all(void)
742 {
743 }
744
745 #endif /* !CONFIG_IS_ENABLED(SYS_ICACHE_OFF) */
746
747 /*
748 * Enable dCache & iCache, whether cache is actually enabled
749 * depend on CONFIG_SYS_DCACHE_OFF and CONFIG_SYS_ICACHE_OFF
750 */
751 void __weak enable_caches(void)
752 {
753 icache_enable();
754 dcache_enable();
755 }