]> git.ipfire.org Git - people/ms/u-boot.git/blob - arch/arm/cpu/armv8/cache_v8.c
Merge git://git.denx.de/u-boot-spi
[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 if (level == 3)
234 *pte = phys | attrs | PTE_TYPE_PAGE;
235 else
236 *pte = phys | attrs;
237 virt += blocksize;
238 phys += blocksize;
239 size -= blocksize;
240 break;
241 } else if (pte_type(pte) == PTE_TYPE_FAULT) {
242 /* Page doesn't fit, create subpages */
243 debug("Creating subtable for virt 0x%llx blksize=%llx\n",
244 virt, blocksize);
245 new_table = create_table();
246 set_pte_table(pte, new_table);
247 } else if (pte_type(pte) == PTE_TYPE_BLOCK) {
248 debug("Split block into subtable for virt 0x%llx blksize=0x%llx\n",
249 virt, blocksize);
250 split_block(pte, level);
251 }
252 }
253 }
254 }
255
256 enum pte_type {
257 PTE_INVAL,
258 PTE_BLOCK,
259 PTE_LEVEL,
260 };
261
262 /*
263 * This is a recursively called function to count the number of
264 * page tables we need to cover a particular PTE range. If you
265 * call this with level = -1 you basically get the full 48 bit
266 * coverage.
267 */
268 static int count_required_pts(u64 addr, int level, u64 maxaddr)
269 {
270 int levelshift = level2shift(level);
271 u64 levelsize = 1ULL << levelshift;
272 u64 levelmask = levelsize - 1;
273 u64 levelend = addr + levelsize;
274 int r = 0;
275 int i;
276 enum pte_type pte_type = PTE_INVAL;
277
278 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++) {
279 struct mm_region *map = &mem_map[i];
280 u64 start = map->virt;
281 u64 end = start + map->size;
282
283 /* Check if the PTE would overlap with the map */
284 if (max(addr, start) <= min(levelend, end)) {
285 start = max(addr, start);
286 end = min(levelend, end);
287
288 /* We need a sub-pt for this level */
289 if ((start & levelmask) || (end & levelmask)) {
290 pte_type = PTE_LEVEL;
291 break;
292 }
293
294 /* Lv0 can not do block PTEs, so do levels here too */
295 if (level <= 0) {
296 pte_type = PTE_LEVEL;
297 break;
298 }
299
300 /* PTE is active, but fits into a block */
301 pte_type = PTE_BLOCK;
302 }
303 }
304
305 /*
306 * Block PTEs at this level are already covered by the parent page
307 * table, so we only need to count sub page tables.
308 */
309 if (pte_type == PTE_LEVEL) {
310 int sublevel = level + 1;
311 u64 sublevelsize = 1ULL << level2shift(sublevel);
312
313 /* Account for the new sub page table ... */
314 r = 1;
315
316 /* ... and for all child page tables that one might have */
317 for (i = 0; i < MAX_PTE_ENTRIES; i++) {
318 r += count_required_pts(addr, sublevel, maxaddr);
319 addr += sublevelsize;
320
321 if (addr >= maxaddr) {
322 /*
323 * We reached the end of address space, no need
324 * to look any further.
325 */
326 break;
327 }
328 }
329 }
330
331 return r;
332 }
333
334 /* Returns the estimated required size of all page tables */
335 __weak u64 get_page_table_size(void)
336 {
337 u64 one_pt = MAX_PTE_ENTRIES * sizeof(u64);
338 u64 size = 0;
339 u64 va_bits;
340 int start_level = 0;
341
342 get_tcr(0, NULL, &va_bits);
343 if (va_bits < 39)
344 start_level = 1;
345
346 /* Account for all page tables we would need to cover our memory map */
347 size = one_pt * count_required_pts(0, start_level - 1, 1ULL << va_bits);
348
349 /*
350 * We need to duplicate our page table once to have an emergency pt to
351 * resort to when splitting page tables later on
352 */
353 size *= 2;
354
355 /*
356 * We may need to split page tables later on if dcache settings change,
357 * so reserve up to 4 (random pick) page tables for that.
358 */
359 size += one_pt * 4;
360
361 return size;
362 }
363
364 void setup_pgtables(void)
365 {
366 int i;
367
368 if (!gd->arch.tlb_fillptr || !gd->arch.tlb_addr)
369 panic("Page table pointer not setup.");
370
371 /*
372 * Allocate the first level we're on with invalidate entries.
373 * If the starting level is 0 (va_bits >= 39), then this is our
374 * Lv0 page table, otherwise it's the entry Lv1 page table.
375 */
376 create_table();
377
378 /* Now add all MMU table entries one after another to the table */
379 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
380 add_map(&mem_map[i]);
381 }
382
383 static void setup_all_pgtables(void)
384 {
385 u64 tlb_addr = gd->arch.tlb_addr;
386 u64 tlb_size = gd->arch.tlb_size;
387
388 /* Reset the fill ptr */
389 gd->arch.tlb_fillptr = tlb_addr;
390
391 /* Create normal system page tables */
392 setup_pgtables();
393
394 /* Create emergency page tables */
395 gd->arch.tlb_size -= (uintptr_t)gd->arch.tlb_fillptr -
396 (uintptr_t)gd->arch.tlb_addr;
397 gd->arch.tlb_addr = gd->arch.tlb_fillptr;
398 setup_pgtables();
399 gd->arch.tlb_emerg = gd->arch.tlb_addr;
400 gd->arch.tlb_addr = tlb_addr;
401 gd->arch.tlb_size = tlb_size;
402 }
403
404 /* to activate the MMU we need to set up virtual memory */
405 __weak void mmu_setup(void)
406 {
407 int el;
408
409 /* Set up page tables only once */
410 if (!gd->arch.tlb_fillptr)
411 setup_all_pgtables();
412
413 el = current_el();
414 set_ttbr_tcr_mair(el, gd->arch.tlb_addr, get_tcr(el, NULL, NULL),
415 MEMORY_ATTRIBUTES);
416
417 /* enable the mmu */
418 set_sctlr(get_sctlr() | CR_M);
419 }
420
421 /*
422 * Performs a invalidation of the entire data cache at all levels
423 */
424 void invalidate_dcache_all(void)
425 {
426 __asm_invalidate_dcache_all();
427 __asm_invalidate_l3_dcache();
428 }
429
430 /*
431 * Performs a clean & invalidation of the entire data cache at all levels.
432 * This function needs to be inline to avoid using stack.
433 * __asm_flush_l3_dcache return status of timeout
434 */
435 inline void flush_dcache_all(void)
436 {
437 int ret;
438
439 __asm_flush_dcache_all();
440 ret = __asm_flush_l3_dcache();
441 if (ret)
442 debug("flushing dcache returns 0x%x\n", ret);
443 else
444 debug("flushing dcache successfully.\n");
445 }
446
447 /*
448 * Invalidates range in all levels of D-cache/unified cache
449 */
450 void invalidate_dcache_range(unsigned long start, unsigned long stop)
451 {
452 __asm_invalidate_dcache_range(start, stop);
453 }
454
455 /*
456 * Flush range(clean & invalidate) from all levels of D-cache/unified cache
457 */
458 void flush_dcache_range(unsigned long start, unsigned long stop)
459 {
460 __asm_flush_dcache_range(start, stop);
461 }
462
463 void dcache_enable(void)
464 {
465 /* The data cache is not active unless the mmu is enabled */
466 if (!(get_sctlr() & CR_M)) {
467 invalidate_dcache_all();
468 __asm_invalidate_tlb_all();
469 mmu_setup();
470 }
471
472 set_sctlr(get_sctlr() | CR_C);
473 }
474
475 void dcache_disable(void)
476 {
477 uint32_t sctlr;
478
479 sctlr = get_sctlr();
480
481 /* if cache isn't enabled no need to disable */
482 if (!(sctlr & CR_C))
483 return;
484
485 set_sctlr(sctlr & ~(CR_C|CR_M));
486
487 flush_dcache_all();
488 __asm_invalidate_tlb_all();
489 }
490
491 int dcache_status(void)
492 {
493 return (get_sctlr() & CR_C) != 0;
494 }
495
496 u64 *__weak arch_get_page_table(void) {
497 puts("No page table offset defined\n");
498
499 return NULL;
500 }
501
502 static bool is_aligned(u64 addr, u64 size, u64 align)
503 {
504 return !(addr & (align - 1)) && !(size & (align - 1));
505 }
506
507 /* Use flag to indicate if attrs has more than d-cache attributes */
508 static u64 set_one_region(u64 start, u64 size, u64 attrs, bool flag, int level)
509 {
510 int levelshift = level2shift(level);
511 u64 levelsize = 1ULL << levelshift;
512 u64 *pte = find_pte(start, level);
513
514 /* Can we can just modify the current level block PTE? */
515 if (is_aligned(start, size, levelsize)) {
516 if (flag) {
517 *pte &= ~PMD_ATTRMASK;
518 *pte |= attrs & PMD_ATTRMASK;
519 } else {
520 *pte &= ~PMD_ATTRINDX_MASK;
521 *pte |= attrs & PMD_ATTRINDX_MASK;
522 }
523 debug("Set attrs=%llx pte=%p level=%d\n", attrs, pte, level);
524
525 return levelsize;
526 }
527
528 /* Unaligned or doesn't fit, maybe split block into table */
529 debug("addr=%llx level=%d pte=%p (%llx)\n", start, level, pte, *pte);
530
531 /* Maybe we need to split the block into a table */
532 if (pte_type(pte) == PTE_TYPE_BLOCK)
533 split_block(pte, level);
534
535 /* And then double-check it became a table or already is one */
536 if (pte_type(pte) != PTE_TYPE_TABLE)
537 panic("PTE %p (%llx) for addr=%llx should be a table",
538 pte, *pte, start);
539
540 /* Roll on to the next page table level */
541 return 0;
542 }
543
544 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
545 enum dcache_option option)
546 {
547 u64 attrs = PMD_ATTRINDX(option);
548 u64 real_start = start;
549 u64 real_size = size;
550
551 debug("start=%lx size=%lx\n", (ulong)start, (ulong)size);
552
553 if (!gd->arch.tlb_emerg)
554 panic("Emergency page table not setup.");
555
556 /*
557 * We can not modify page tables that we're currently running on,
558 * so we first need to switch to the "emergency" page tables where
559 * we can safely modify our primary page tables and then switch back
560 */
561 __asm_switch_ttbr(gd->arch.tlb_emerg);
562
563 /*
564 * Loop through the address range until we find a page granule that fits
565 * our alignment constraints, then set it to the new cache attributes
566 */
567 while (size > 0) {
568 int level;
569 u64 r;
570
571 for (level = 1; level < 4; level++) {
572 /* Set d-cache attributes only */
573 r = set_one_region(start, size, attrs, false, level);
574 if (r) {
575 /* PTE successfully replaced */
576 size -= r;
577 start += r;
578 break;
579 }
580 }
581
582 }
583
584 /* We're done modifying page tables, switch back to our primary ones */
585 __asm_switch_ttbr(gd->arch.tlb_addr);
586
587 /*
588 * Make sure there's nothing stale in dcache for a region that might
589 * have caches off now
590 */
591 flush_dcache_range(real_start, real_start + real_size);
592 }
593
594 /*
595 * Modify MMU table for a region with updated PXN/UXN/Memory type/valid bits.
596 * The procecess is break-before-make. The target region will be marked as
597 * invalid during the process of changing.
598 */
599 void mmu_change_region_attr(phys_addr_t addr, size_t siz, u64 attrs)
600 {
601 int level;
602 u64 r, size, start;
603
604 start = addr;
605 size = siz;
606 /*
607 * Loop through the address range until we find a page granule that fits
608 * our alignment constraints, then set it to "invalid".
609 */
610 while (size > 0) {
611 for (level = 1; level < 4; level++) {
612 /* Set PTE to fault */
613 r = set_one_region(start, size, PTE_TYPE_FAULT, true,
614 level);
615 if (r) {
616 /* PTE successfully invalidated */
617 size -= r;
618 start += r;
619 break;
620 }
621 }
622 }
623
624 flush_dcache_range(gd->arch.tlb_addr,
625 gd->arch.tlb_addr + gd->arch.tlb_size);
626 __asm_invalidate_tlb_all();
627
628 /*
629 * Loop through the address range until we find a page granule that fits
630 * our alignment constraints, then set it to the new cache attributes
631 */
632 start = addr;
633 size = siz;
634 while (size > 0) {
635 for (level = 1; level < 4; level++) {
636 /* Set PTE to new attributes */
637 r = set_one_region(start, size, attrs, true, level);
638 if (r) {
639 /* PTE successfully updated */
640 size -= r;
641 start += r;
642 break;
643 }
644 }
645 }
646 flush_dcache_range(gd->arch.tlb_addr,
647 gd->arch.tlb_addr + gd->arch.tlb_size);
648 __asm_invalidate_tlb_all();
649 }
650
651 #else /* CONFIG_SYS_DCACHE_OFF */
652
653 /*
654 * For SPL builds, we may want to not have dcache enabled. Any real U-Boot
655 * running however really wants to have dcache and the MMU active. Check that
656 * everything is sane and give the developer a hint if it isn't.
657 */
658 #ifndef CONFIG_SPL_BUILD
659 #error Please describe your MMU layout in CONFIG_SYS_MEM_MAP and enable dcache.
660 #endif
661
662 void invalidate_dcache_all(void)
663 {
664 }
665
666 void flush_dcache_all(void)
667 {
668 }
669
670 void dcache_enable(void)
671 {
672 }
673
674 void dcache_disable(void)
675 {
676 }
677
678 int dcache_status(void)
679 {
680 return 0;
681 }
682
683 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
684 enum dcache_option option)
685 {
686 }
687
688 #endif /* CONFIG_SYS_DCACHE_OFF */
689
690 #ifndef CONFIG_SYS_ICACHE_OFF
691
692 void icache_enable(void)
693 {
694 invalidate_icache_all();
695 set_sctlr(get_sctlr() | CR_I);
696 }
697
698 void icache_disable(void)
699 {
700 set_sctlr(get_sctlr() & ~CR_I);
701 }
702
703 int icache_status(void)
704 {
705 return (get_sctlr() & CR_I) != 0;
706 }
707
708 void invalidate_icache_all(void)
709 {
710 __asm_invalidate_icache_all();
711 __asm_invalidate_l3_icache();
712 }
713
714 #else /* CONFIG_SYS_ICACHE_OFF */
715
716 void icache_enable(void)
717 {
718 }
719
720 void icache_disable(void)
721 {
722 }
723
724 int icache_status(void)
725 {
726 return 0;
727 }
728
729 void invalidate_icache_all(void)
730 {
731 }
732
733 #endif /* CONFIG_SYS_ICACHE_OFF */
734
735 /*
736 * Enable dCache & iCache, whether cache is actually enabled
737 * depend on CONFIG_SYS_DCACHE_OFF and CONFIG_SYS_ICACHE_OFF
738 */
739 void __weak enable_caches(void)
740 {
741 icache_enable();
742 dcache_enable();
743 }