1 // SPDX-License-Identifier: GPL-2.0+
4 * David Feng <fenghua@phytium.com.cn>
7 * Alexander Graf <agraf@suse.de>
14 #include <asm/cache.h>
15 #include <asm/global_data.h>
16 #include <asm/system.h>
17 #include <asm/armv8/mmu.h>
19 DECLARE_GLOBAL_DATA_PTR;
21 #if !CONFIG_IS_ENABLED(SYS_DCACHE_OFF)
24 * With 4k page granule, a virtual address is split into 4 lookup parts
25 * spanning 9 bits each:
27 * _______________________________________________
29 * | 0 | Lv0 | Lv1 | Lv2 | Lv3 | off |
30 * |_______|_______|_______|_______|_______|_______|
31 * 63-48 47-39 38-30 29-21 20-12 11-00
35 * Lv0: FF8000000000 --
42 static int get_effective_el(void)
44 int el = current_el();
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.
53 __asm__ ("mrs %0, HCR_EL2\n" : "=r" (hcr_el2));
54 if (hcr_el2 & BIT(HCR_EL2_E2H_BIT))
61 u64 get_tcr(u64 *pips, u64 *pva_bits)
63 int el = get_effective_el();
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);
73 /* Calculate the maximum physical (and thus virtual) address */
74 if (max_addr > (1ULL << 44)) {
77 } else if (max_addr > (1ULL << 42)) {
80 } else if (max_addr > (1ULL << 40)) {
83 } else if (max_addr > (1ULL << 36)) {
86 } else if (max_addr > (1ULL << 32)) {
95 tcr = TCR_EL1_RSVD | (ips << 32) | TCR_EPD1_DISABLE;
96 if (gd->arch.has_hafdbs)
97 tcr |= TCR_EL1_HA | TCR_EL1_HD;
99 tcr = TCR_EL2_RSVD | (ips << 16);
100 if (gd->arch.has_hafdbs)
101 tcr |= TCR_EL2_HA | TCR_EL2_HD;
103 tcr = TCR_EL3_RSVD | (ips << 16);
104 if (gd->arch.has_hafdbs)
105 tcr |= TCR_EL3_HA | TCR_EL3_HD;
108 /* PTWs cacheable, inner/outer WBWA and inner shareable */
109 tcr |= TCR_TG0_4K | TCR_SHARED_INNER | TCR_ORGN_WBWA | TCR_IRGN_WBWA;
110 tcr |= TCR_T0SZ(va_bits);
120 #define MAX_PTE_ENTRIES 512
122 static int pte_type(u64 *pte)
124 return *pte & PTE_TYPE_MASK;
127 /* Returns the LSB number for a PTE on level <level> */
128 static int level2shift(int level)
130 /* Page is 12 bits wide, every level translates 9 bits */
131 return (12 + 9 * (3 - level));
134 static u64 *find_pte(u64 addr, int level)
142 debug("addr=%llx level=%d\n", addr, level);
144 get_tcr(NULL, &va_bits);
148 if (level < start_level)
151 /* Walk through all page table levels to find our PTE */
152 pte = (u64*)gd->arch.tlb_addr;
153 for (i = start_level; i < 4; i++) {
154 idx = (addr >> level2shift(i)) & 0x1FF;
156 debug("idx=%llx PTE %p at level %d: %llx\n", idx, pte, i, *pte);
161 /* PTE is no table (either invalid or block), can't traverse */
162 if (pte_type(pte) != PTE_TYPE_TABLE)
164 /* Off to the next level */
165 pte = (u64*)(*pte & 0x0000fffffffff000ULL);
168 /* Should never reach here */
172 #ifdef CONFIG_CMO_BY_VA_ONLY
173 static void __cmo_on_leaves(void (*cmo_fn)(unsigned long, unsigned long),
174 u64 pte, int level, u64 base)
179 ptep = (u64 *)(pte & GENMASK_ULL(47, PAGE_SHIFT));
180 for (i = 0; i < PAGE_SIZE / sizeof(u64); i++) {
181 u64 end, va = base + i * BIT(level2shift(level));
185 type = pte & PTE_TYPE_MASK;
186 attrs = pte & PMD_ATTRINDX_MASK;
187 debug("PTE %llx at level %d VA %llx\n", pte, level, va);
189 /* Not valid? next! */
190 if (!(type & PTE_TYPE_VALID))
193 /* Not a leaf? Recurse on the next level */
194 if (!(type == PTE_TYPE_BLOCK ||
195 (level == 3 && type == PTE_TYPE_PAGE))) {
196 __cmo_on_leaves(cmo_fn, pte, level + 1, va);
201 * From this point, this must be a leaf.
203 * Start excluding non memory mappings
205 if (attrs != PTE_BLOCK_MEMTYPE(MT_NORMAL) &&
206 attrs != PTE_BLOCK_MEMTYPE(MT_NORMAL_NC))
209 if (gd->arch.has_hafdbs && (pte & (PTE_RDONLY | PTE_DBM)) != PTE_DBM)
212 end = va + BIT(level2shift(level)) - 1;
214 /* No intersection with RAM? */
215 if (end < gd->ram_base ||
216 va >= (gd->ram_base + gd->ram_size))
220 * OK, we have a partial RAM mapping. However, this
221 * can cover *more* than the RAM. Yes, u-boot is
222 * *that* braindead. Compute the intersection we care
223 * about, and not a byte more.
225 va = max(va, (u64)gd->ram_base);
226 end = min(end, gd->ram_base + gd->ram_size);
228 debug("Flush PTE %llx at level %d: %llx-%llx\n",
229 pte, level, va, end);
234 static void apply_cmo_to_mappings(void (*cmo_fn)(unsigned long, unsigned long))
239 if (!gd->arch.tlb_addr)
242 get_tcr(NULL, &va_bits);
246 __cmo_on_leaves(cmo_fn, gd->arch.tlb_addr, sl, 0);
249 static inline void apply_cmo_to_mappings(void *dummy) {}
252 /* Returns and creates a new full table (512 entries) */
253 static u64 *create_table(void)
255 u64 *new_table = (u64*)gd->arch.tlb_fillptr;
256 u64 pt_len = MAX_PTE_ENTRIES * sizeof(u64);
258 /* Allocate MAX_PTE_ENTRIES pte entries */
259 gd->arch.tlb_fillptr += pt_len;
261 if (gd->arch.tlb_fillptr - gd->arch.tlb_addr > gd->arch.tlb_size)
262 panic("Insufficient RAM for page table: 0x%lx > 0x%lx. "
263 "Please increase the size in get_page_table_size()",
264 gd->arch.tlb_fillptr - gd->arch.tlb_addr,
267 /* Mark all entries as invalid */
268 memset(new_table, 0, pt_len);
273 static void set_pte_table(u64 *pte, u64 *table)
275 /* Point *pte to the new table */
276 debug("Setting %p to addr=%p\n", pte, table);
277 *pte = PTE_TYPE_TABLE | (ulong)table;
280 /* Splits a block PTE into table with subpages spanning the old block */
281 static void split_block(u64 *pte, int level)
286 /* level describes the parent level, we need the child ones */
287 int levelshift = level2shift(level + 1);
289 if (pte_type(pte) != PTE_TYPE_BLOCK)
290 panic("PTE %p (%llx) is not a block. Some driver code wants to "
291 "modify dcache settings for an range not covered in "
292 "mem_map.", pte, old_pte);
294 new_table = create_table();
295 debug("Splitting pte %p (%llx) into %p\n", pte, old_pte, new_table);
297 for (i = 0; i < MAX_PTE_ENTRIES; i++) {
298 new_table[i] = old_pte | (i << levelshift);
300 /* Level 3 block PTEs have the table type */
301 if ((level + 1) == 3)
302 new_table[i] |= PTE_TYPE_TABLE;
304 debug("Setting new_table[%lld] = %llx\n", i, new_table[i]);
307 /* Set the new table into effect */
308 set_pte_table(pte, new_table);
311 static void map_range(u64 virt, u64 phys, u64 size, int level,
312 u64 *table, u64 attrs)
314 u64 map_size = BIT_ULL(level2shift(level));
317 idx = (virt >> level2shift(level)) & (MAX_PTE_ENTRIES - 1);
318 for (i = idx; size; i++) {
319 u64 next_size, *next_table;
321 if (level >= gd->arch.first_block_level &&
322 size >= map_size && !(virt & (map_size - 1))) {
324 table[i] = phys | attrs | PTE_TYPE_PAGE;
326 table[i] = phys | attrs;
335 /* Going one level down */
336 if (pte_type(&table[i]) == PTE_TYPE_FAULT)
337 set_pte_table(&table[i], create_table());
339 next_table = (u64 *)(table[i] & GENMASK_ULL(47, PAGE_SHIFT));
340 next_size = min(map_size - (virt & (map_size - 1)), size);
342 map_range(virt, phys, next_size, level + 1, next_table, attrs);
350 static void add_map(struct mm_region *map)
352 u64 attrs = map->attrs | PTE_TYPE_BLOCK | PTE_BLOCK_AF;
356 get_tcr(NULL, &va_bits);
360 if (!gd->arch.first_block_level)
361 gd->arch.first_block_level = 1;
363 if (gd->arch.has_hafdbs)
364 attrs |= PTE_DBM | PTE_RDONLY;
366 map_range(map->virt, map->phys, map->size, level,
367 (u64 *)gd->arch.tlb_addr, attrs);
370 static void count_range(u64 virt, u64 size, int level, int *cntp)
372 u64 map_size = BIT_ULL(level2shift(level));
375 idx = (virt >> level2shift(level)) & (MAX_PTE_ENTRIES - 1);
376 for (i = idx; size; i++) {
379 if (level >= gd->arch.first_block_level &&
380 size >= map_size && !(virt & (map_size - 1))) {
387 /* Going one level down */
389 next_size = min(map_size - (virt & (map_size - 1)), size);
391 count_range(virt, next_size, level + 1, cntp);
398 static int count_ranges(void)
400 int i, count = 0, level = 0;
403 get_tcr(NULL, &va_bits);
407 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
408 count_range(mem_map[i].virt, mem_map[i].size, level, &count);
413 /* Returns the estimated required size of all page tables */
414 __weak u64 get_page_table_size(void)
416 u64 one_pt = MAX_PTE_ENTRIES * sizeof(u64);
419 asm volatile("mrs %0, id_aa64mmfr1_el1" : "=r" (mmfr1));
420 if ((mmfr1 & 0xf) == 2) {
421 gd->arch.has_hafdbs = true;
422 gd->arch.first_block_level = 2;
424 gd->arch.has_hafdbs = false;
425 gd->arch.first_block_level = 1;
428 /* Account for all page tables we would need to cover our memory map */
429 size = one_pt * count_ranges();
432 * We need to duplicate our page table once to have an emergency pt to
433 * resort to when splitting page tables later on
438 * We may need to split page tables later on if dcache settings change,
439 * so reserve up to 4 (random pick) page tables for that.
446 void setup_pgtables(void)
450 if (!gd->arch.tlb_fillptr || !gd->arch.tlb_addr)
451 panic("Page table pointer not setup.");
454 * Allocate the first level we're on with invalidate entries.
455 * If the starting level is 0 (va_bits >= 39), then this is our
456 * Lv0 page table, otherwise it's the entry Lv1 page table.
460 /* Now add all MMU table entries one after another to the table */
461 for (i = 0; mem_map[i].size || mem_map[i].attrs; i++)
462 add_map(&mem_map[i]);
465 static void setup_all_pgtables(void)
467 u64 tlb_addr = gd->arch.tlb_addr;
468 u64 tlb_size = gd->arch.tlb_size;
470 /* Reset the fill ptr */
471 gd->arch.tlb_fillptr = tlb_addr;
473 /* Create normal system page tables */
476 /* Create emergency page tables */
477 gd->arch.tlb_size -= (uintptr_t)gd->arch.tlb_fillptr -
478 (uintptr_t)gd->arch.tlb_addr;
479 gd->arch.tlb_addr = gd->arch.tlb_fillptr;
481 gd->arch.tlb_emerg = gd->arch.tlb_addr;
482 gd->arch.tlb_addr = tlb_addr;
483 gd->arch.tlb_size = tlb_size;
486 /* to activate the MMU we need to set up virtual memory */
487 __weak void mmu_setup(void)
491 /* Set up page tables only once */
492 if (!gd->arch.tlb_fillptr)
493 setup_all_pgtables();
496 set_ttbr_tcr_mair(el, gd->arch.tlb_addr, get_tcr(NULL, NULL),
500 set_sctlr(get_sctlr() | CR_M);
504 * Performs a invalidation of the entire data cache at all levels
506 void invalidate_dcache_all(void)
508 #ifndef CONFIG_CMO_BY_VA_ONLY
509 __asm_invalidate_dcache_all();
510 __asm_invalidate_l3_dcache();
512 apply_cmo_to_mappings(invalidate_dcache_range);
517 * Performs a clean & invalidation of the entire data cache at all levels.
518 * This function needs to be inline to avoid using stack.
519 * __asm_flush_l3_dcache return status of timeout
521 inline void flush_dcache_all(void)
523 #ifndef CONFIG_CMO_BY_VA_ONLY
526 __asm_flush_dcache_all();
527 ret = __asm_flush_l3_dcache();
529 debug("flushing dcache returns 0x%x\n", ret);
531 debug("flushing dcache successfully.\n");
533 apply_cmo_to_mappings(flush_dcache_range);
537 #ifndef CONFIG_SYS_DISABLE_DCACHE_OPS
539 * Invalidates range in all levels of D-cache/unified cache
541 void invalidate_dcache_range(unsigned long start, unsigned long stop)
543 __asm_invalidate_dcache_range(start, stop);
547 * Flush range(clean & invalidate) from all levels of D-cache/unified cache
549 void flush_dcache_range(unsigned long start, unsigned long stop)
551 __asm_flush_dcache_range(start, stop);
554 void invalidate_dcache_range(unsigned long start, unsigned long stop)
558 void flush_dcache_range(unsigned long start, unsigned long stop)
561 #endif /* CONFIG_SYS_DISABLE_DCACHE_OPS */
563 void dcache_enable(void)
565 /* The data cache is not active unless the mmu is enabled */
566 if (!(get_sctlr() & CR_M)) {
567 invalidate_dcache_all();
568 __asm_invalidate_tlb_all();
572 /* Set up page tables only once (it is done also by mmu_setup()) */
573 if (!gd->arch.tlb_fillptr)
574 setup_all_pgtables();
576 set_sctlr(get_sctlr() | CR_C);
579 void dcache_disable(void)
585 /* if cache isn't enabled no need to disable */
589 if (IS_ENABLED(CONFIG_CMO_BY_VA_ONLY)) {
591 * When invalidating by VA, do it *before* turning the MMU
592 * off, so that at least our stack is coherent.
597 set_sctlr(sctlr & ~(CR_C|CR_M));
599 if (!IS_ENABLED(CONFIG_CMO_BY_VA_ONLY))
602 __asm_invalidate_tlb_all();
605 int dcache_status(void)
607 return (get_sctlr() & CR_C) != 0;
610 u64 *__weak arch_get_page_table(void) {
611 puts("No page table offset defined\n");
616 static bool is_aligned(u64 addr, u64 size, u64 align)
618 return !(addr & (align - 1)) && !(size & (align - 1));
621 /* Use flag to indicate if attrs has more than d-cache attributes */
622 static u64 set_one_region(u64 start, u64 size, u64 attrs, bool flag, int level)
624 int levelshift = level2shift(level);
625 u64 levelsize = 1ULL << levelshift;
626 u64 *pte = find_pte(start, level);
628 /* Can we can just modify the current level block PTE? */
629 if (is_aligned(start, size, levelsize)) {
631 *pte &= ~PMD_ATTRMASK;
632 *pte |= attrs & PMD_ATTRMASK;
634 *pte &= ~PMD_ATTRINDX_MASK;
635 *pte |= attrs & PMD_ATTRINDX_MASK;
637 debug("Set attrs=%llx pte=%p level=%d\n", attrs, pte, level);
642 /* Unaligned or doesn't fit, maybe split block into table */
643 debug("addr=%llx level=%d pte=%p (%llx)\n", start, level, pte, *pte);
645 /* Maybe we need to split the block into a table */
646 if (pte_type(pte) == PTE_TYPE_BLOCK)
647 split_block(pte, level);
649 /* And then double-check it became a table or already is one */
650 if (pte_type(pte) != PTE_TYPE_TABLE)
651 panic("PTE %p (%llx) for addr=%llx should be a table",
654 /* Roll on to the next page table level */
658 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
659 enum dcache_option option)
661 u64 attrs = PMD_ATTRINDX(option >> 2);
662 u64 real_start = start;
663 u64 real_size = size;
665 debug("start=%lx size=%lx\n", (ulong)start, (ulong)size);
667 if (!gd->arch.tlb_emerg)
668 panic("Emergency page table not setup.");
671 * We can not modify page tables that we're currently running on,
672 * so we first need to switch to the "emergency" page tables where
673 * we can safely modify our primary page tables and then switch back
675 __asm_switch_ttbr(gd->arch.tlb_emerg);
678 * Loop through the address range until we find a page granule that fits
679 * our alignment constraints, then set it to the new cache attributes
685 for (level = 1; level < 4; level++) {
686 /* Set d-cache attributes only */
687 r = set_one_region(start, size, attrs, false, level);
689 /* PTE successfully replaced */
698 /* We're done modifying page tables, switch back to our primary ones */
699 __asm_switch_ttbr(gd->arch.tlb_addr);
702 * Make sure there's nothing stale in dcache for a region that might
703 * have caches off now
705 flush_dcache_range(real_start, real_start + real_size);
709 * Modify MMU table for a region with updated PXN/UXN/Memory type/valid bits.
710 * The procecess is break-before-make. The target region will be marked as
711 * invalid during the process of changing.
713 void mmu_change_region_attr(phys_addr_t addr, size_t siz, u64 attrs)
721 * Loop through the address range until we find a page granule that fits
722 * our alignment constraints, then set it to "invalid".
725 for (level = 1; level < 4; level++) {
726 /* Set PTE to fault */
727 r = set_one_region(start, size, PTE_TYPE_FAULT, true,
730 /* PTE successfully invalidated */
738 flush_dcache_range(gd->arch.tlb_addr,
739 gd->arch.tlb_addr + gd->arch.tlb_size);
740 __asm_invalidate_tlb_all();
743 * Loop through the address range until we find a page granule that fits
744 * our alignment constraints, then set it to the new cache attributes
749 for (level = 1; level < 4; level++) {
750 /* Set PTE to new attributes */
751 r = set_one_region(start, size, attrs, true, level);
753 /* PTE successfully updated */
760 flush_dcache_range(gd->arch.tlb_addr,
761 gd->arch.tlb_addr + gd->arch.tlb_size);
762 __asm_invalidate_tlb_all();
765 #else /* !CONFIG_IS_ENABLED(SYS_DCACHE_OFF) */
768 * For SPL builds, we may want to not have dcache enabled. Any real U-Boot
769 * running however really wants to have dcache and the MMU active. Check that
770 * everything is sane and give the developer a hint if it isn't.
772 #ifndef CONFIG_SPL_BUILD
773 #error Please describe your MMU layout in CONFIG_SYS_MEM_MAP and enable dcache.
776 void invalidate_dcache_all(void)
780 void flush_dcache_all(void)
784 void dcache_enable(void)
788 void dcache_disable(void)
792 int dcache_status(void)
797 void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
798 enum dcache_option option)
802 #endif /* !CONFIG_IS_ENABLED(SYS_DCACHE_OFF) */
804 #if !CONFIG_IS_ENABLED(SYS_ICACHE_OFF)
806 void icache_enable(void)
808 invalidate_icache_all();
809 set_sctlr(get_sctlr() | CR_I);
812 void icache_disable(void)
814 set_sctlr(get_sctlr() & ~CR_I);
817 int icache_status(void)
819 return (get_sctlr() & CR_I) != 0;
824 return (get_sctlr() & CR_M) != 0;
827 void invalidate_icache_all(void)
829 __asm_invalidate_icache_all();
830 __asm_invalidate_l3_icache();
833 #else /* !CONFIG_IS_ENABLED(SYS_ICACHE_OFF) */
835 void icache_enable(void)
839 void icache_disable(void)
843 int icache_status(void)
853 void invalidate_icache_all(void)
857 #endif /* !CONFIG_IS_ENABLED(SYS_ICACHE_OFF) */
860 * Enable dCache & iCache, whether cache is actually enabled
861 * depend on CONFIG_SYS_DCACHE_OFF and CONFIG_SYS_ICACHE_OFF
863 void __weak enable_caches(void)