b.liu | e958203 | 2025-04-17 19:18:16 +0800 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 2000 Ani Joshi <ajoshi@unixbox.com> |
| 4 | * Copyright (C) 2000, 2001, 06 Ralf Baechle <ralf@linux-mips.org> |
| 5 | * swiped from i386, and cloned for MIPS by Geert, polished by Ralf. |
| 6 | */ |
| 7 | #include <linux/dma-direct.h> |
| 8 | #include <linux/dma-noncoherent.h> |
| 9 | #include <linux/dma-contiguous.h> |
| 10 | #include <linux/highmem.h> |
| 11 | |
| 12 | #include <asm/cache.h> |
| 13 | #include <asm/cpu-type.h> |
| 14 | #include <asm/dma-coherence.h> |
| 15 | #include <asm/io.h> |
| 16 | |
| 17 | /* |
| 18 | * The affected CPUs below in 'cpu_needs_post_dma_flush()' can speculatively |
| 19 | * fill random cachelines with stale data at any time, requiring an extra |
| 20 | * flush post-DMA. |
| 21 | * |
| 22 | * Warning on the terminology - Linux calls an uncached area coherent; MIPS |
| 23 | * terminology calls memory areas with hardware maintained coherency coherent. |
| 24 | * |
| 25 | * Note that the R14000 and R16000 should also be checked for in this condition. |
| 26 | * However this function is only called on non-I/O-coherent systems and only the |
| 27 | * R10000 and R12000 are used in such systems, the SGI IP28 Indigo² rsp. |
| 28 | * SGI IP32 aka O2. |
| 29 | */ |
| 30 | static inline bool cpu_needs_post_dma_flush(void) |
| 31 | { |
| 32 | switch (boot_cpu_type()) { |
| 33 | case CPU_R10000: |
| 34 | case CPU_R12000: |
| 35 | case CPU_BMIPS5000: |
| 36 | return true; |
| 37 | default: |
| 38 | /* |
| 39 | * Presence of MAARs suggests that the CPU supports |
| 40 | * speculatively prefetching data, and therefore requires |
| 41 | * the post-DMA flush/invalidate. |
| 42 | */ |
| 43 | return cpu_has_maar; |
| 44 | } |
| 45 | } |
| 46 | |
| 47 | void arch_dma_prep_coherent(struct page *page, size_t size) |
| 48 | { |
| 49 | dma_cache_wback_inv((unsigned long)page_address(page), size); |
| 50 | } |
| 51 | |
| 52 | void *uncached_kernel_address(void *addr) |
| 53 | { |
| 54 | return (void *)(__pa(addr) + UNCAC_BASE); |
| 55 | } |
| 56 | |
| 57 | void *cached_kernel_address(void *addr) |
| 58 | { |
| 59 | return __va(addr) - UNCAC_BASE; |
| 60 | } |
| 61 | |
| 62 | long arch_dma_coherent_to_pfn(struct device *dev, void *cpu_addr, |
| 63 | dma_addr_t dma_addr) |
| 64 | { |
| 65 | return page_to_pfn(virt_to_page(cached_kernel_address(cpu_addr))); |
| 66 | } |
| 67 | |
| 68 | static inline void dma_sync_virt(void *addr, size_t size, |
| 69 | enum dma_data_direction dir) |
| 70 | { |
| 71 | switch (dir) { |
| 72 | case DMA_TO_DEVICE: |
| 73 | dma_cache_wback((unsigned long)addr, size); |
| 74 | break; |
| 75 | |
| 76 | case DMA_FROM_DEVICE: |
| 77 | dma_cache_inv((unsigned long)addr, size); |
| 78 | break; |
| 79 | |
| 80 | case DMA_BIDIRECTIONAL: |
| 81 | dma_cache_wback_inv((unsigned long)addr, size); |
| 82 | break; |
| 83 | |
| 84 | default: |
| 85 | BUG(); |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | /* |
| 90 | * A single sg entry may refer to multiple physically contiguous pages. But |
| 91 | * we still need to process highmem pages individually. If highmem is not |
| 92 | * configured then the bulk of this loop gets optimized out. |
| 93 | */ |
| 94 | static inline void dma_sync_phys(phys_addr_t paddr, size_t size, |
| 95 | enum dma_data_direction dir) |
| 96 | { |
| 97 | struct page *page = pfn_to_page(paddr >> PAGE_SHIFT); |
| 98 | unsigned long offset = paddr & ~PAGE_MASK; |
| 99 | size_t left = size; |
| 100 | |
| 101 | do { |
| 102 | size_t len = left; |
| 103 | |
| 104 | if (PageHighMem(page)) { |
| 105 | void *addr; |
| 106 | |
| 107 | if (offset + len > PAGE_SIZE) |
| 108 | len = PAGE_SIZE - offset; |
| 109 | |
| 110 | addr = kmap_atomic(page); |
| 111 | dma_sync_virt(addr + offset, len, dir); |
| 112 | kunmap_atomic(addr); |
| 113 | } else |
| 114 | dma_sync_virt(page_address(page) + offset, size, dir); |
| 115 | offset = 0; |
| 116 | page++; |
| 117 | left -= len; |
| 118 | } while (left); |
| 119 | } |
| 120 | |
| 121 | void arch_sync_dma_for_device(phys_addr_t paddr, size_t size, |
| 122 | enum dma_data_direction dir) |
| 123 | { |
| 124 | dma_sync_phys(paddr, size, dir); |
| 125 | } |
| 126 | |
| 127 | #ifdef CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU |
| 128 | void arch_sync_dma_for_cpu(phys_addr_t paddr, size_t size, |
| 129 | enum dma_data_direction dir) |
| 130 | { |
| 131 | if (cpu_needs_post_dma_flush()) |
| 132 | dma_sync_phys(paddr, size, dir); |
| 133 | } |
| 134 | #endif |
| 135 | |
| 136 | void arch_dma_cache_sync(struct device *dev, void *vaddr, size_t size, |
| 137 | enum dma_data_direction direction) |
| 138 | { |
| 139 | BUG_ON(direction == DMA_NONE); |
| 140 | |
| 141 | dma_sync_virt(vaddr, size, direction); |
| 142 | } |
| 143 | |
| 144 | #ifdef CONFIG_DMA_PERDEV_COHERENT |
| 145 | void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size, |
| 146 | const struct iommu_ops *iommu, bool coherent) |
| 147 | { |
| 148 | dev->dma_coherent = coherent; |
| 149 | } |
| 150 | EXPORT_SYMBOL_GPL(arch_setup_dma_ops); |
| 151 | #endif |