]> git.ipfire.org Git - thirdparty/u-boot.git/blob - arch/arm/cpu/armv8/cache_v8.c
Prepare v2023.04
[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 <log.h>
14 #include <asm/cache.h>
15 #include <asm/global_data.h>
16 #include <asm/system.h>
17 #include <asm/armv8/mmu.h>
18
19 DECLARE_GLOBAL_DATA_PTR;
20
21 #if !CONFIG_IS_ENABLED(SYS_DCACHE_OFF)
22
23 /*
24 * With 4k page granule, a virtual address is split into 4 lookup parts
25 * spanning 9 bits each:
26 *
27 * _______________________________________________
28 * | | | | | | |
29 * | 0 | Lv0 | Lv1 | Lv2 | Lv3 | off |
30 * |_______|_______|_______|_______|_______|_______|
31 * 63-48 47-39 38-30 29-21 20-12 11-00
32 *
33 * mask page size
34 *
35 * Lv0: FF8000000000 --
36 * Lv1: 7FC0000000 1G
37 * Lv2: 3FE00000 2M
38 * Lv3: 1FF000 4K
39 * off: FFF
40 */
41
42 static int get_effective_el(void)
43 {
44 int el = current_el();
45
46 if (el == 2) {
47 u64 hcr_el2;
48
49 /*
50 * If we are using the EL2&0 translation regime, the TCR_EL2
51 * looks like the EL1 version, even though we are in EL2.
52 */
53 __asm__ ("mrs %0, HCR_EL2\n" : "=r" (hcr_el2));
54 if (hcr_el2 & BIT(HCR_EL2_E2H_BIT))
55 return 1;
56 }
57
58 return el;
59 }
60
61 u64 get_tcr(u64 *pips, u64 *pva_bits)
62 {
63 int el = get_effective_el();
64 u64 max_addr = 0;
65 u64 ips, va_bits;
66 u64 tcr;
67 int i;
68
69 /* Find the largest address we need to support */
70 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
71 max_addr = max(max_addr, mem_map[i].virt + mem_map[i].size);
72
73 /* Calculate the maximum physical (and thus virtual) address */
74 if (max_addr > (1ULL << 44)) {
75 ips = 5;
76 va_bits = 48;
77 } else if (max_addr > (1ULL << 42)) {
78 ips = 4;
79 va_bits = 44;
80 } else if (max_addr > (1ULL << 40)) {
81 ips = 3;
82 va_bits = 42;
83 } else if (max_addr > (1ULL << 36)) {
84 ips = 2;
85 va_bits = 40;
86 } else if (max_addr > (1ULL << 32)) {
87 ips = 1;
88 va_bits = 36;
89 } else {
90 ips = 0;
91 va_bits = 32;
92 }
93
94 if (el == 1) {
95 tcr = TCR_EL1_RSVD | (ips << 32) | TCR_EPD1_DISABLE;
96 } else if (el == 2) {
97 tcr = TCR_EL2_RSVD | (ips << 16);
98 } else {
99 tcr = TCR_EL3_RSVD | (ips << 16);
100 }
101
102 /* PTWs cacheable, inner/outer WBWA and inner shareable */
103 tcr |= TCR_TG0_4K | TCR_SHARED_INNER | TCR_ORGN_WBWA | TCR_IRGN_WBWA;
104 tcr |= TCR_T0SZ(va_bits);
105
106 if (pips)
107 *pips = ips;
108 if (pva_bits)
109 *pva_bits = va_bits;
110
111 return tcr;
112 }
113
114 #define MAX_PTE_ENTRIES 512
115
116 static int pte_type(u64 *pte)
117 {
118 return *pte & PTE_TYPE_MASK;
119 }
120
121 /* Returns the LSB number for a PTE on level <level> */
122 static int level2shift(int level)
123 {
124 /* Page is 12 bits wide, every level translates 9 bits */
125 return (12 + 9 * (3 - level));
126 }
127
128 static u64 *find_pte(u64 addr, int level)
129 {
130 int start_level = 0;
131 u64 *pte;
132 u64 idx;
133 u64 va_bits;
134 int i;
135
136 debug("addr=%llx level=%d\n", addr, level);
137
138 get_tcr(NULL, &va_bits);
139 if (va_bits < 39)
140 start_level = 1;
141
142 if (level < start_level)
143 return NULL;
144
145 /* Walk through all page table levels to find our PTE */
146 pte = (u64*)gd->arch.tlb_addr;
147 for (i = start_level; i < 4; i++) {
148 idx = (addr >> level2shift(i)) & 0x1FF;
149 pte += idx;
150 debug("idx=%llx PTE %p at level %d: %llx\n", idx, pte, i, *pte);
151
152 /* Found it */
153 if (i == level)
154 return pte;
155 /* PTE is no table (either invalid or block), can't traverse */
156 if (pte_type(pte) != PTE_TYPE_TABLE)
157 return NULL;
158 /* Off to the next level */
159 pte = (u64*)(*pte & 0x0000fffffffff000ULL);
160 }
161
162 /* Should never reach here */
163 return NULL;
164 }
165
166 /* Returns and creates a new full table (512 entries) */
167 static u64 *create_table(void)
168 {
169 u64 *new_table = (u64*)gd->arch.tlb_fillptr;
170 u64 pt_len = MAX_PTE_ENTRIES * sizeof(u64);
171
172 /* Allocate MAX_PTE_ENTRIES pte entries */
173 gd->arch.tlb_fillptr += pt_len;
174
175 if (gd->arch.tlb_fillptr - gd->arch.tlb_addr > gd->arch.tlb_size)
176 panic("Insufficient RAM for page table: 0x%lx > 0x%lx. "
177 "Please increase the size in get_page_table_size()",
178 gd->arch.tlb_fillptr - gd->arch.tlb_addr,
179 gd->arch.tlb_size);
180
181 /* Mark all entries as invalid */
182 memset(new_table, 0, pt_len);
183
184 return new_table;
185 }
186
187 static void set_pte_table(u64 *pte, u64 *table)
188 {
189 /* Point *pte to the new table */
190 debug("Setting %p to addr=%p\n", pte, table);
191 *pte = PTE_TYPE_TABLE | (ulong)table;
192 }
193
194 /* Splits a block PTE into table with subpages spanning the old block */
195 static void split_block(u64 *pte, int level)
196 {
197 u64 old_pte = *pte;
198 u64 *new_table;
199 u64 i = 0;
200 /* level describes the parent level, we need the child ones */
201 int levelshift = level2shift(level + 1);
202
203 if (pte_type(pte) != PTE_TYPE_BLOCK)
204 panic("PTE %p (%llx) is not a block. Some driver code wants to "
205 "modify dcache settings for an range not covered in "
206 "mem_map.", pte, old_pte);
207
208 new_table = create_table();
209 debug("Splitting pte %p (%llx) into %p\n", pte, old_pte, new_table);
210
211 for (i = 0; i < MAX_PTE_ENTRIES; i++) {
212 new_table[i] = old_pte | (i << levelshift);
213
214 /* Level 3 block PTEs have the table type */
215 if ((level + 1) == 3)
216 new_table[i] |= PTE_TYPE_TABLE;
217
218 debug("Setting new_table[%lld] = %llx\n", i, new_table[i]);
219 }
220
221 /* Set the new table into effect */
222 set_pte_table(pte, new_table);
223 }
224
225 /* Add one mm_region map entry to the page tables */
226 static void add_map(struct mm_region *map)
227 {
228 u64 *pte;
229 u64 virt = map->virt;
230 u64 phys = map->phys;
231 u64 size = map->size;
232 u64 attrs = map->attrs | PTE_TYPE_BLOCK | PTE_BLOCK_AF;
233 u64 blocksize;
234 int level;
235 u64 *new_table;
236
237 while (size) {
238 pte = find_pte(virt, 0);
239 if (pte && (pte_type(pte) == PTE_TYPE_FAULT)) {
240 debug("Creating table for virt 0x%llx\n", virt);
241 new_table = create_table();
242 set_pte_table(pte, new_table);
243 }
244
245 for (level = 1; level < 4; level++) {
246 pte = find_pte(virt, level);
247 if (!pte)
248 panic("pte not found\n");
249
250 blocksize = 1ULL << level2shift(level);
251 debug("Checking if pte fits for virt=%llx size=%llx blocksize=%llx\n",
252 virt, size, blocksize);
253 if (size >= blocksize && !(virt & (blocksize - 1))) {
254 /* Page fits, create block PTE */
255 debug("Setting PTE %p to block virt=%llx\n",
256 pte, virt);
257 if (level == 3)
258 *pte = phys | attrs | PTE_TYPE_PAGE;
259 else
260 *pte = phys | attrs;
261 virt += blocksize;
262 phys += blocksize;
263 size -= blocksize;
264 break;
265 } else if (pte_type(pte) == PTE_TYPE_FAULT) {
266 /* Page doesn't fit, create subpages */
267 debug("Creating subtable for virt 0x%llx blksize=%llx\n",
268 virt, blocksize);
269 new_table = create_table();
270 set_pte_table(pte, new_table);
271 } else if (pte_type(pte) == PTE_TYPE_BLOCK) {
272 debug("Split block into subtable for virt 0x%llx blksize=0x%llx\n",
273 virt, blocksize);
274 split_block(pte, level);
275 }
276 }
277 }
278 }
279
280 enum pte_type {
281 PTE_INVAL,
282 PTE_BLOCK,
283 PTE_LEVEL,
284 };
285
286 /*
287 * This is a recursively called function to count the number of
288 * page tables we need to cover a particular PTE range. If you
289 * call this with level = -1 you basically get the full 48 bit
290 * coverage.
291 */
292 static int count_required_pts(u64 addr, int level, u64 maxaddr)
293 {
294 int levelshift = level2shift(level);
295 u64 levelsize = 1ULL << levelshift;
296 u64 levelmask = levelsize - 1;
297 u64 levelend = addr + levelsize;
298 int r = 0;
299 int i;
300 enum pte_type pte_type = PTE_INVAL;
301
302 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++) {
303 struct mm_region *map = &mem_map[i];
304 u64 start = map->virt;
305 u64 end = start + map->size;
306
307 /* Check if the PTE would overlap with the map */
308 if (max(addr, start) <= min(levelend, end)) {
309 start = max(addr, start);
310 end = min(levelend, end);
311
312 /* We need a sub-pt for this level */
313 if ((start & levelmask) || (end & levelmask)) {
314 pte_type = PTE_LEVEL;
315 break;
316 }
317
318 /* Lv0 can not do block PTEs, so do levels here too */
319 if (level <= 0) {
320 pte_type = PTE_LEVEL;
321 break;
322 }
323
324 /* PTE is active, but fits into a block */
325 pte_type = PTE_BLOCK;
326 }
327 }
328
329 /*
330 * Block PTEs at this level are already covered by the parent page
331 * table, so we only need to count sub page tables.
332 */
333 if (pte_type == PTE_LEVEL) {
334 int sublevel = level + 1;
335 u64 sublevelsize = 1ULL << level2shift(sublevel);
336
337 /* Account for the new sub page table ... */
338 r = 1;
339
340 /* ... and for all child page tables that one might have */
341 for (i = 0; i < MAX_PTE_ENTRIES; i++) {
342 r += count_required_pts(addr, sublevel, maxaddr);
343 addr += sublevelsize;
344
345 if (addr >= maxaddr) {
346 /*
347 * We reached the end of address space, no need
348 * to look any further.
349 */
350 break;
351 }
352 }
353 }
354
355 return r;
356 }
357
358 /* Returns the estimated required size of all page tables */
359 __weak u64 get_page_table_size(void)
360 {
361 u64 one_pt = MAX_PTE_ENTRIES * sizeof(u64);
362 u64 size = 0;
363 u64 va_bits;
364 int start_level = 0;
365
366 get_tcr(NULL, &va_bits);
367 if (va_bits < 39)
368 start_level = 1;
369
370 /* Account for all page tables we would need to cover our memory map */
371 size = one_pt * count_required_pts(0, start_level - 1, 1ULL << va_bits);
372
373 /*
374 * We need to duplicate our page table once to have an emergency pt to
375 * resort to when splitting page tables later on
376 */
377 size *= 2;
378
379 /*
380 * We may need to split page tables later on if dcache settings change,
381 * so reserve up to 4 (random pick) page tables for that.
382 */
383 size += one_pt * 4;
384
385 return size;
386 }
387
388 void setup_pgtables(void)
389 {
390 int i;
391
392 if (!gd->arch.tlb_fillptr || !gd->arch.tlb_addr)
393 panic("Page table pointer not setup.");
394
395 /*
396 * Allocate the first level we're on with invalidate entries.
397 * If the starting level is 0 (va_bits >= 39), then this is our
398 * Lv0 page table, otherwise it's the entry Lv1 page table.
399 */
400 create_table();
401
402 /* Now add all MMU table entries one after another to the table */
403 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
404 add_map(&mem_map[i]);
405 }
406
407 static void setup_all_pgtables(void)
408 {
409 u64 tlb_addr = gd->arch.tlb_addr;
410 u64 tlb_size = gd->arch.tlb_size;
411
412 /* Reset the fill ptr */
413 gd->arch.tlb_fillptr = tlb_addr;
414
415 /* Create normal system page tables */
416 setup_pgtables();
417
418 /* Create emergency page tables */
419 gd->arch.tlb_size -= (uintptr_t)gd->arch.tlb_fillptr -
420 (uintptr_t)gd->arch.tlb_addr;
421 gd->arch.tlb_addr = gd->arch.tlb_fillptr;
422 setup_pgtables();
423 gd->arch.tlb_emerg = gd->arch.tlb_addr;
424 gd->arch.tlb_addr = tlb_addr;
425 gd->arch.tlb_size = tlb_size;
426 }
427
428 /* to activate the MMU we need to set up virtual memory */
429 __weak void mmu_setup(void)
430 {
431 int el;
432
433 /* Set up page tables only once */
434 if (!gd->arch.tlb_fillptr)
435 setup_all_pgtables();
436
437 el = current_el();
438 set_ttbr_tcr_mair(el, gd->arch.tlb_addr, get_tcr(NULL, NULL),
439 MEMORY_ATTRIBUTES);
440
441 /* enable the mmu */
442 set_sctlr(get_sctlr() | CR_M);
443 }
444
445 /*
446 * Performs a invalidation of the entire data cache at all levels
447 */
448 void invalidate_dcache_all(void)
449 {
450 __asm_invalidate_dcache_all();
451 __asm_invalidate_l3_dcache();
452 }
453
454 /*
455 * Performs a clean & invalidation of the entire data cache at all levels.
456 * This function needs to be inline to avoid using stack.
457 * __asm_flush_l3_dcache return status of timeout
458 */
459 inline void flush_dcache_all(void)
460 {
461 int ret;
462
463 __asm_flush_dcache_all();
464 ret = __asm_flush_l3_dcache();
465 if (ret)
466 debug("flushing dcache returns 0x%x\n", ret);
467 else
468 debug("flushing dcache successfully.\n");
469 }
470
471 #ifndef CONFIG_SYS_DISABLE_DCACHE_OPS
472 /*
473 * Invalidates range in all levels of D-cache/unified cache
474 */
475 void invalidate_dcache_range(unsigned long start, unsigned long stop)
476 {
477 __asm_invalidate_dcache_range(start, stop);
478 }
479
480 /*
481 * Flush range(clean & invalidate) from all levels of D-cache/unified cache
482 */
483 void flush_dcache_range(unsigned long start, unsigned long stop)
484 {
485 __asm_flush_dcache_range(start, stop);
486 }
487 #else
488 void invalidate_dcache_range(unsigned long start, unsigned long stop)
489 {
490 }
491
492 void flush_dcache_range(unsigned long start, unsigned long stop)
493 {
494 }
495 #endif /* CONFIG_SYS_DISABLE_DCACHE_OPS */
496
497 void dcache_enable(void)
498 {
499 /* The data cache is not active unless the mmu is enabled */
500 if (!(get_sctlr() & CR_M)) {
501 invalidate_dcache_all();
502 __asm_invalidate_tlb_all();
503 mmu_setup();
504 }
505
506 /* Set up page tables only once (it is done also by mmu_setup()) */
507 if (!gd->arch.tlb_fillptr)
508 setup_all_pgtables();
509
510 set_sctlr(get_sctlr() | CR_C);
511 }
512
513 void dcache_disable(void)
514 {
515 uint32_t sctlr;
516
517 sctlr = get_sctlr();
518
519 /* if cache isn't enabled no need to disable */
520 if (!(sctlr & CR_C))
521 return;
522
523 set_sctlr(sctlr & ~(CR_C|CR_M));
524
525 flush_dcache_all();
526 __asm_invalidate_tlb_all();
527 }
528
529 int dcache_status(void)
530 {
531 return (get_sctlr() & CR_C) != 0;
532 }
533
534 u64 *__weak arch_get_page_table(void) {
535 puts("No page table offset defined\n");
536
537 return NULL;
538 }
539
540 static bool is_aligned(u64 addr, u64 size, u64 align)
541 {
542 return !(addr & (align - 1)) && !(size & (align - 1));
543 }
544
545 /* Use flag to indicate if attrs has more than d-cache attributes */
546 static u64 set_one_region(u64 start, u64 size, u64 attrs, bool flag, int level)
547 {
548 int levelshift = level2shift(level);
549 u64 levelsize = 1ULL << levelshift;
550 u64 *pte = find_pte(start, level);
551
552 /* Can we can just modify the current level block PTE? */
553 if (is_aligned(start, size, levelsize)) {
554 if (flag) {
555 *pte &= ~PMD_ATTRMASK;
556 *pte |= attrs & PMD_ATTRMASK;
557 } else {
558 *pte &= ~PMD_ATTRINDX_MASK;
559 *pte |= attrs & PMD_ATTRINDX_MASK;
560 }
561 debug("Set attrs=%llx pte=%p level=%d\n", attrs, pte, level);
562
563 return levelsize;
564 }
565
566 /* Unaligned or doesn't fit, maybe split block into table */
567 debug("addr=%llx level=%d pte=%p (%llx)\n", start, level, pte, *pte);
568
569 /* Maybe we need to split the block into a table */
570 if (pte_type(pte) == PTE_TYPE_BLOCK)
571 split_block(pte, level);
572
573 /* And then double-check it became a table or already is one */
574 if (pte_type(pte) != PTE_TYPE_TABLE)
575 panic("PTE %p (%llx) for addr=%llx should be a table",
576 pte, *pte, start);
577
578 /* Roll on to the next page table level */
579 return 0;
580 }
581
582 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
583 enum dcache_option option)
584 {
585 u64 attrs = PMD_ATTRINDX(option >> 2);
586 u64 real_start = start;
587 u64 real_size = size;
588
589 debug("start=%lx size=%lx\n", (ulong)start, (ulong)size);
590
591 if (!gd->arch.tlb_emerg)
592 panic("Emergency page table not setup.");
593
594 /*
595 * We can not modify page tables that we're currently running on,
596 * so we first need to switch to the "emergency" page tables where
597 * we can safely modify our primary page tables and then switch back
598 */
599 __asm_switch_ttbr(gd->arch.tlb_emerg);
600
601 /*
602 * Loop through the address range until we find a page granule that fits
603 * our alignment constraints, then set it to the new cache attributes
604 */
605 while (size > 0) {
606 int level;
607 u64 r;
608
609 for (level = 1; level < 4; level++) {
610 /* Set d-cache attributes only */
611 r = set_one_region(start, size, attrs, false, level);
612 if (r) {
613 /* PTE successfully replaced */
614 size -= r;
615 start += r;
616 break;
617 }
618 }
619
620 }
621
622 /* We're done modifying page tables, switch back to our primary ones */
623 __asm_switch_ttbr(gd->arch.tlb_addr);
624
625 /*
626 * Make sure there's nothing stale in dcache for a region that might
627 * have caches off now
628 */
629 flush_dcache_range(real_start, real_start + real_size);
630 }
631
632 /*
633 * Modify MMU table for a region with updated PXN/UXN/Memory type/valid bits.
634 * The procecess is break-before-make. The target region will be marked as
635 * invalid during the process of changing.
636 */
637 void mmu_change_region_attr(phys_addr_t addr, size_t siz, u64 attrs)
638 {
639 int level;
640 u64 r, size, start;
641
642 start = addr;
643 size = siz;
644 /*
645 * Loop through the address range until we find a page granule that fits
646 * our alignment constraints, then set it to "invalid".
647 */
648 while (size > 0) {
649 for (level = 1; level < 4; level++) {
650 /* Set PTE to fault */
651 r = set_one_region(start, size, PTE_TYPE_FAULT, true,
652 level);
653 if (r) {
654 /* PTE successfully invalidated */
655 size -= r;
656 start += r;
657 break;
658 }
659 }
660 }
661
662 flush_dcache_range(gd->arch.tlb_addr,
663 gd->arch.tlb_addr + gd->arch.tlb_size);
664 __asm_invalidate_tlb_all();
665
666 /*
667 * Loop through the address range until we find a page granule that fits
668 * our alignment constraints, then set it to the new cache attributes
669 */
670 start = addr;
671 size = siz;
672 while (size > 0) {
673 for (level = 1; level < 4; level++) {
674 /* Set PTE to new attributes */
675 r = set_one_region(start, size, attrs, true, level);
676 if (r) {
677 /* PTE successfully updated */
678 size -= r;
679 start += r;
680 break;
681 }
682 }
683 }
684 flush_dcache_range(gd->arch.tlb_addr,
685 gd->arch.tlb_addr + gd->arch.tlb_size);
686 __asm_invalidate_tlb_all();
687 }
688
689 #else /* !CONFIG_IS_ENABLED(SYS_DCACHE_OFF) */
690
691 /*
692 * For SPL builds, we may want to not have dcache enabled. Any real U-Boot
693 * running however really wants to have dcache and the MMU active. Check that
694 * everything is sane and give the developer a hint if it isn't.
695 */
696 #ifndef CONFIG_SPL_BUILD
697 #error Please describe your MMU layout in CONFIG_SYS_MEM_MAP and enable dcache.
698 #endif
699
700 void invalidate_dcache_all(void)
701 {
702 }
703
704 void flush_dcache_all(void)
705 {
706 }
707
708 void dcache_enable(void)
709 {
710 }
711
712 void dcache_disable(void)
713 {
714 }
715
716 int dcache_status(void)
717 {
718 return 0;
719 }
720
721 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
722 enum dcache_option option)
723 {
724 }
725
726 #endif /* !CONFIG_IS_ENABLED(SYS_DCACHE_OFF) */
727
728 #if !CONFIG_IS_ENABLED(SYS_ICACHE_OFF)
729
730 void icache_enable(void)
731 {
732 invalidate_icache_all();
733 set_sctlr(get_sctlr() | CR_I);
734 }
735
736 void icache_disable(void)
737 {
738 set_sctlr(get_sctlr() & ~CR_I);
739 }
740
741 int icache_status(void)
742 {
743 return (get_sctlr() & CR_I) != 0;
744 }
745
746 int mmu_status(void)
747 {
748 return (get_sctlr() & CR_M) != 0;
749 }
750
751 void invalidate_icache_all(void)
752 {
753 __asm_invalidate_icache_all();
754 __asm_invalidate_l3_icache();
755 }
756
757 #else /* !CONFIG_IS_ENABLED(SYS_ICACHE_OFF) */
758
759 void icache_enable(void)
760 {
761 }
762
763 void icache_disable(void)
764 {
765 }
766
767 int icache_status(void)
768 {
769 return 0;
770 }
771
772 int mmu_status(void)
773 {
774 return 0;
775 }
776
777 void invalidate_icache_all(void)
778 {
779 }
780
781 #endif /* !CONFIG_IS_ENABLED(SYS_ICACHE_OFF) */
782
783 /*
784 * Enable dCache & iCache, whether cache is actually enabled
785 * depend on CONFIG_SYS_DCACHE_OFF and CONFIG_SYS_ICACHE_OFF
786 */
787 void __weak enable_caches(void)
788 {
789 icache_enable();
790 dcache_enable();
791 }