| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* Copyright (C) 2000 Ani Joshi <[email protected]> |
| 4 |
* Copyright (C) 2000, 2001, 06 Ralf Baechle <[email protected]> |
| 5 |
* swiped from i386, and cloned for MIPS by Geert, polished by Ralf. |
| 6 |
*/ |
| 7 |
#include <linux/dma-direct.h> |
| 8 |
#include <linux/dma-map-ops.h> |
| 9 |
#include <linux/highmem.h> |
| 10 |
|
| 11 |
#include <asm/cache.h> |
| 12 |
#include <asm/cpu-type.h> |
| 13 |
#include <asm/io.h> |
| 14 |
|
| 15 |
/* |
| 16 |
* The affected CPUs below in 'cpu_needs_post_dma_flush()' can speculatively |
| 17 |
* fill random cachelines with stale data at any time, requiring an extra |
| 18 |
* flush post-DMA. |
| 19 |
* |
| 20 |
* Warning on the terminology - Linux calls an uncached area coherent; MIPS |
| 21 |
* terminology calls memory areas with hardware maintained coherency coherent. |
| 22 |
* |
| 23 |
* Note that the R14000 and R16000 should also be checked for in this condition. |
| 24 |
* However this function is only called on non-I/O-coherent systems and only the |
| 25 |
* R10000 and R12000 are used in such systems, the SGI IP28 Indigo² rsp. |
| 26 |
* SGI IP32 aka O2. |
| 27 |
*/ |
| 28 |
static inline bool cpu_needs_post_dma_flush(void) |
| 29 |
{ |
| 30 |
switch (boot_cpu_type()) { |
| 31 |
case CPU_R10000: |
| 32 |
case CPU_R12000: |
| 33 |
case CPU_BMIPS5000: |
| 34 |
case CPU_LOONGSON2EF: |
| 35 |
case CPU_XBURST: |
| 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 *arch_dma_set_uncached(void *addr, size_t size) |
| 53 |
{ |
| 54 |
return (void *)(__pa(addr) + UNCAC_BASE); |
| 55 |
} |
| 56 |
|
| 57 |
static inline void dma_sync_virt_for_device(void *addr, size_t size, |
| 58 |
enum dma_data_direction dir) |
| 59 |
{ |
| 60 |
switch (dir) { |
| 61 |
case DMA_TO_DEVICE: |
| 62 |
dma_cache_wback((unsigned long)addr, size); |
| 63 |
break; |
| 64 |
case DMA_FROM_DEVICE: |
| 65 |
dma_cache_inv((unsigned long)addr, size); |
| 66 |
break; |
| 67 |
case DMA_BIDIRECTIONAL: |
| 68 |
dma_cache_wback_inv((unsigned long)addr, size); |
| 69 |
break; |
| 70 |
default: |
| 71 |
BUG(); |
| 72 |
} |
| 73 |
} |
| 74 |
|
| 75 |
static inline void dma_sync_virt_for_cpu(void *addr, size_t size, |
| 76 |
enum dma_data_direction dir) |
| 77 |
{ |
| 78 |
switch (dir) { |
| 79 |
case DMA_TO_DEVICE: |
| 80 |
break; |
| 81 |
case DMA_FROM_DEVICE: |
| 82 |
case DMA_BIDIRECTIONAL: |
| 83 |
dma_cache_inv((unsigned long)addr, size); |
| 84 |
break; |
| 85 |
default: |
| 86 |
BUG(); |
| 87 |
} |
| 88 |
} |
| 89 |
|
| 90 |
/* |
| 91 |
* A single sg entry may refer to multiple physically contiguous pages. But |
| 92 |
* we still need to process highmem pages individually. If highmem is not |
| 93 |
* configured then the bulk of this loop gets optimized out. |
| 94 |
*/ |
| 95 |
static inline void dma_sync_phys(phys_addr_t paddr, size_t size, |
| 96 |
enum dma_data_direction dir, bool for_device) |
| 97 |
{ |
| 98 |
struct page *page = pfn_to_page(paddr >> PAGE_SHIFT); |
| 99 |
unsigned long offset = paddr & ~PAGE_MASK; |
| 100 |
size_t left = size; |
| 101 |
|
| 102 |
do { |
| 103 |
size_t len = left; |
| 104 |
void *addr; |
| 105 |
|
| 106 |
if (PageHighMem(page)) { |
| 107 |
if (offset + len > PAGE_SIZE) |
| 108 |
len = PAGE_SIZE - offset; |
| 109 |
} |
| 110 |
|
| 111 |
addr = kmap_atomic(page); |
| 112 |
if (for_device) |
| 113 |
dma_sync_virt_for_device(addr + offset, len, dir); |
| 114 |
else |
| 115 |
dma_sync_virt_for_cpu(addr + offset, len, dir); |
| 116 |
kunmap_atomic(addr); |
| 117 |
|
| 118 |
offset = 0; |
| 119 |
page++; |
| 120 |
left -= len; |
| 121 |
} while (left); |
| 122 |
} |
| 123 |
|
| 124 |
void arch_sync_dma_for_device(phys_addr_t paddr, size_t size, |
| 125 |
enum dma_data_direction dir) |
| 126 |
{ |
| 127 |
dma_sync_phys(paddr, size, dir, true); |
| 128 |
} |
| 129 |
|
| 130 |
#ifdef CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU |
| 131 |
void arch_sync_dma_for_cpu(phys_addr_t paddr, size_t size, |
| 132 |
enum dma_data_direction dir) |
| 133 |
{ |
| 134 |
if (cpu_needs_post_dma_flush()) |
| 135 |
dma_sync_phys(paddr, size, dir, false); |
| 136 |
} |
| 137 |
#endif |
| 138 |
|
| 139 |
#ifdef CONFIG_ARCH_HAS_SETUP_DMA_OPS |
| 140 |
void arch_setup_dma_ops(struct device *dev, bool coherent) |
| 141 |
{ |
| 142 |
dev->dma_coherent = coherent; |
| 143 |
} |
| 144 |
#endif |