| .. | .. |
|---|
| 1 | +// SPDX-License-Identifier: GPL-2.0-only |
|---|
| 1 | 2 | /* |
|---|
| 2 | 3 | * Microblaze support for cache consistent memory. |
|---|
| 3 | 4 | * Copyright (C) 2010 Michal Simek <monstr@monstr.eu> |
|---|
| 4 | 5 | * Copyright (C) 2010 PetaLogix |
|---|
| 5 | 6 | * Copyright (C) 2005 John Williams <jwilliams@itee.uq.edu.au> |
|---|
| 6 | | - * |
|---|
| 7 | | - * Based on PowerPC version derived from arch/arm/mm/consistent.c |
|---|
| 8 | | - * Copyright (C) 2001 Dan Malek (dmalek@jlc.net) |
|---|
| 9 | | - * Copyright (C) 2000 Russell King |
|---|
| 10 | | - * |
|---|
| 11 | | - * This program is free software; you can redistribute it and/or modify |
|---|
| 12 | | - * it under the terms of the GNU General Public License version 2 as |
|---|
| 13 | | - * published by the Free Software Foundation. |
|---|
| 14 | 7 | */ |
|---|
| 15 | 8 | |
|---|
| 16 | | -#include <linux/export.h> |
|---|
| 17 | | -#include <linux/signal.h> |
|---|
| 18 | | -#include <linux/sched.h> |
|---|
| 19 | 9 | #include <linux/kernel.h> |
|---|
| 20 | | -#include <linux/errno.h> |
|---|
| 21 | 10 | #include <linux/string.h> |
|---|
| 22 | 11 | #include <linux/types.h> |
|---|
| 23 | | -#include <linux/ptrace.h> |
|---|
| 24 | | -#include <linux/mman.h> |
|---|
| 25 | 12 | #include <linux/mm.h> |
|---|
| 26 | | -#include <linux/swap.h> |
|---|
| 27 | | -#include <linux/stddef.h> |
|---|
| 28 | | -#include <linux/vmalloc.h> |
|---|
| 29 | 13 | #include <linux/init.h> |
|---|
| 30 | | -#include <linux/delay.h> |
|---|
| 31 | | -#include <linux/bootmem.h> |
|---|
| 32 | | -#include <linux/highmem.h> |
|---|
| 33 | | -#include <linux/pci.h> |
|---|
| 34 | | -#include <linux/interrupt.h> |
|---|
| 35 | | -#include <linux/gfp.h> |
|---|
| 36 | | -#include <linux/dma-noncoherent.h> |
|---|
| 37 | | - |
|---|
| 38 | | -#include <asm/pgalloc.h> |
|---|
| 39 | | -#include <linux/io.h> |
|---|
| 40 | | -#include <linux/hardirq.h> |
|---|
| 41 | | -#include <linux/mmu_context.h> |
|---|
| 42 | | -#include <asm/mmu.h> |
|---|
| 43 | | -#include <linux/uaccess.h> |
|---|
| 44 | | -#include <asm/pgtable.h> |
|---|
| 14 | +#include <linux/dma-map-ops.h> |
|---|
| 45 | 15 | #include <asm/cpuinfo.h> |
|---|
| 46 | | -#include <asm/tlbflush.h> |
|---|
| 16 | +#include <asm/cacheflush.h> |
|---|
| 47 | 17 | |
|---|
| 48 | | -#ifndef CONFIG_MMU |
|---|
| 49 | | -/* I have to use dcache values because I can't relate on ram size */ |
|---|
| 50 | | -# define UNCACHED_SHADOW_MASK (cpuinfo.dcache_high - cpuinfo.dcache_base + 1) |
|---|
| 51 | | -#endif |
|---|
| 52 | | - |
|---|
| 53 | | -/* |
|---|
| 54 | | - * Consistent memory allocators. Used for DMA devices that want to |
|---|
| 55 | | - * share uncached memory with the processor core. |
|---|
| 56 | | - * My crufty no-MMU approach is simple. In the HW platform we can optionally |
|---|
| 57 | | - * mirror the DDR up above the processor cacheable region. So, memory accessed |
|---|
| 58 | | - * in this mirror region will not be cached. It's alloced from the same |
|---|
| 59 | | - * pool as normal memory, but the handle we return is shifted up into the |
|---|
| 60 | | - * uncached region. This will no doubt cause big problems if memory allocated |
|---|
| 61 | | - * here is not also freed properly. -- JW |
|---|
| 62 | | - */ |
|---|
| 63 | | -void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle, |
|---|
| 64 | | - gfp_t gfp, unsigned long attrs) |
|---|
| 18 | +void arch_dma_prep_coherent(struct page *page, size_t size) |
|---|
| 65 | 19 | { |
|---|
| 66 | | - unsigned long order, vaddr; |
|---|
| 67 | | - void *ret; |
|---|
| 68 | | - unsigned int i, err = 0; |
|---|
| 69 | | - struct page *page, *end; |
|---|
| 20 | + phys_addr_t paddr = page_to_phys(page); |
|---|
| 70 | 21 | |
|---|
| 71 | | -#ifdef CONFIG_MMU |
|---|
| 72 | | - phys_addr_t pa; |
|---|
| 73 | | - struct vm_struct *area; |
|---|
| 74 | | - unsigned long va; |
|---|
| 75 | | -#endif |
|---|
| 76 | | - |
|---|
| 77 | | - if (in_interrupt()) |
|---|
| 78 | | - BUG(); |
|---|
| 79 | | - |
|---|
| 80 | | - /* Only allocate page size areas. */ |
|---|
| 81 | | - size = PAGE_ALIGN(size); |
|---|
| 82 | | - order = get_order(size); |
|---|
| 83 | | - |
|---|
| 84 | | - vaddr = __get_free_pages(gfp, order); |
|---|
| 85 | | - if (!vaddr) |
|---|
| 86 | | - return NULL; |
|---|
| 87 | | - |
|---|
| 88 | | - /* |
|---|
| 89 | | - * we need to ensure that there are no cachelines in use, |
|---|
| 90 | | - * or worse dirty in this area. |
|---|
| 91 | | - */ |
|---|
| 92 | | - flush_dcache_range(virt_to_phys((void *)vaddr), |
|---|
| 93 | | - virt_to_phys((void *)vaddr) + size); |
|---|
| 22 | + flush_dcache_range(paddr, paddr + size); |
|---|
| 23 | +} |
|---|
| 94 | 24 | |
|---|
| 95 | 25 | #ifndef CONFIG_MMU |
|---|
| 96 | | - ret = (void *)vaddr; |
|---|
| 97 | | - /* |
|---|
| 98 | | - * Here's the magic! Note if the uncached shadow is not implemented, |
|---|
| 99 | | - * it's up to the calling code to also test that condition and make |
|---|
| 100 | | - * other arranegments, such as manually flushing the cache and so on. |
|---|
| 101 | | - */ |
|---|
| 102 | | -# ifdef CONFIG_XILINX_UNCACHED_SHADOW |
|---|
| 103 | | - ret = (void *)((unsigned) ret | UNCACHED_SHADOW_MASK); |
|---|
| 104 | | -# endif |
|---|
| 105 | | - if ((unsigned int)ret > cpuinfo.dcache_base && |
|---|
| 106 | | - (unsigned int)ret < cpuinfo.dcache_high) |
|---|
| 26 | +/* |
|---|
| 27 | + * Consistent memory allocators. Used for DMA devices that want to share |
|---|
| 28 | + * uncached memory with the processor core. My crufty no-MMU approach is |
|---|
| 29 | + * simple. In the HW platform we can optionally mirror the DDR up above the |
|---|
| 30 | + * processor cacheable region. So, memory accessed in this mirror region will |
|---|
| 31 | + * not be cached. It's alloced from the same pool as normal memory, but the |
|---|
| 32 | + * handle we return is shifted up into the uncached region. This will no doubt |
|---|
| 33 | + * cause big problems if memory allocated here is not also freed properly. -- JW |
|---|
| 34 | + * |
|---|
| 35 | + * I have to use dcache values because I can't relate on ram size: |
|---|
| 36 | + */ |
|---|
| 37 | +#ifdef CONFIG_XILINX_UNCACHED_SHADOW |
|---|
| 38 | +#define UNCACHED_SHADOW_MASK (cpuinfo.dcache_high - cpuinfo.dcache_base + 1) |
|---|
| 39 | +#else |
|---|
| 40 | +#define UNCACHED_SHADOW_MASK 0 |
|---|
| 41 | +#endif /* CONFIG_XILINX_UNCACHED_SHADOW */ |
|---|
| 42 | + |
|---|
| 43 | +void *arch_dma_set_uncached(void *ptr, size_t size) |
|---|
| 44 | +{ |
|---|
| 45 | + unsigned long addr = (unsigned long)ptr; |
|---|
| 46 | + |
|---|
| 47 | + addr |= UNCACHED_SHADOW_MASK; |
|---|
| 48 | + if (addr > cpuinfo.dcache_base && addr < cpuinfo.dcache_high) |
|---|
| 107 | 49 | pr_warn("ERROR: Your cache coherent area is CACHED!!!\n"); |
|---|
| 108 | | - |
|---|
| 109 | | - /* dma_handle is same as physical (shadowed) address */ |
|---|
| 110 | | - *dma_handle = (dma_addr_t)ret; |
|---|
| 111 | | -#else |
|---|
| 112 | | - /* Allocate some common virtual space to map the new pages. */ |
|---|
| 113 | | - area = get_vm_area(size, VM_ALLOC); |
|---|
| 114 | | - if (!area) { |
|---|
| 115 | | - free_pages(vaddr, order); |
|---|
| 116 | | - return NULL; |
|---|
| 117 | | - } |
|---|
| 118 | | - va = (unsigned long) area->addr; |
|---|
| 119 | | - ret = (void *)va; |
|---|
| 120 | | - |
|---|
| 121 | | - /* This gives us the real physical address of the first page. */ |
|---|
| 122 | | - *dma_handle = pa = __virt_to_phys(vaddr); |
|---|
| 123 | | -#endif |
|---|
| 124 | | - |
|---|
| 125 | | - /* |
|---|
| 126 | | - * free wasted pages. We skip the first page since we know |
|---|
| 127 | | - * that it will have count = 1 and won't require freeing. |
|---|
| 128 | | - * We also mark the pages in use as reserved so that |
|---|
| 129 | | - * remap_page_range works. |
|---|
| 130 | | - */ |
|---|
| 131 | | - page = virt_to_page(vaddr); |
|---|
| 132 | | - end = page + (1 << order); |
|---|
| 133 | | - |
|---|
| 134 | | - split_page(page, order); |
|---|
| 135 | | - |
|---|
| 136 | | - for (i = 0; i < size && err == 0; i += PAGE_SIZE) { |
|---|
| 137 | | -#ifdef CONFIG_MMU |
|---|
| 138 | | - /* MS: This is the whole magic - use cache inhibit pages */ |
|---|
| 139 | | - err = map_page(va + i, pa + i, _PAGE_KERNEL | _PAGE_NO_CACHE); |
|---|
| 140 | | -#endif |
|---|
| 141 | | - |
|---|
| 142 | | - SetPageReserved(page); |
|---|
| 143 | | - page++; |
|---|
| 144 | | - } |
|---|
| 145 | | - |
|---|
| 146 | | - /* Free the otherwise unused pages. */ |
|---|
| 147 | | - while (page < end) { |
|---|
| 148 | | - __free_page(page); |
|---|
| 149 | | - page++; |
|---|
| 150 | | - } |
|---|
| 151 | | - |
|---|
| 152 | | - if (err) { |
|---|
| 153 | | - free_pages(vaddr, order); |
|---|
| 154 | | - return NULL; |
|---|
| 155 | | - } |
|---|
| 156 | | - |
|---|
| 157 | | - return ret; |
|---|
| 50 | + return (void *)addr; |
|---|
| 158 | 51 | } |
|---|
| 159 | | - |
|---|
| 160 | | -#ifdef CONFIG_MMU |
|---|
| 161 | | -static pte_t *consistent_virt_to_pte(void *vaddr) |
|---|
| 162 | | -{ |
|---|
| 163 | | - unsigned long addr = (unsigned long)vaddr; |
|---|
| 164 | | - |
|---|
| 165 | | - return pte_offset_kernel(pmd_offset(pgd_offset_k(addr), addr), addr); |
|---|
| 166 | | -} |
|---|
| 167 | | - |
|---|
| 168 | | -unsigned long consistent_virt_to_pfn(void *vaddr) |
|---|
| 169 | | -{ |
|---|
| 170 | | - pte_t *ptep = consistent_virt_to_pte(vaddr); |
|---|
| 171 | | - |
|---|
| 172 | | - if (pte_none(*ptep) || !pte_present(*ptep)) |
|---|
| 173 | | - return 0; |
|---|
| 174 | | - |
|---|
| 175 | | - return pte_pfn(*ptep); |
|---|
| 176 | | -} |
|---|
| 177 | | -#endif |
|---|
| 178 | | - |
|---|
| 179 | | -/* |
|---|
| 180 | | - * free page(s) as defined by the above mapping. |
|---|
| 181 | | - */ |
|---|
| 182 | | -void arch_dma_free(struct device *dev, size_t size, void *vaddr, |
|---|
| 183 | | - dma_addr_t dma_addr, unsigned long attrs) |
|---|
| 184 | | -{ |
|---|
| 185 | | - struct page *page; |
|---|
| 186 | | - |
|---|
| 187 | | - if (in_interrupt()) |
|---|
| 188 | | - BUG(); |
|---|
| 189 | | - |
|---|
| 190 | | - size = PAGE_ALIGN(size); |
|---|
| 191 | | - |
|---|
| 192 | | -#ifndef CONFIG_MMU |
|---|
| 193 | | - /* Clear SHADOW_MASK bit in address, and free as per usual */ |
|---|
| 194 | | -# ifdef CONFIG_XILINX_UNCACHED_SHADOW |
|---|
| 195 | | - vaddr = (void *)((unsigned)vaddr & ~UNCACHED_SHADOW_MASK); |
|---|
| 196 | | -# endif |
|---|
| 197 | | - page = virt_to_page(vaddr); |
|---|
| 198 | | - |
|---|
| 199 | | - do { |
|---|
| 200 | | - __free_reserved_page(page); |
|---|
| 201 | | - page++; |
|---|
| 202 | | - } while (size -= PAGE_SIZE); |
|---|
| 203 | | -#else |
|---|
| 204 | | - do { |
|---|
| 205 | | - pte_t *ptep = consistent_virt_to_pte(vaddr); |
|---|
| 206 | | - unsigned long pfn; |
|---|
| 207 | | - |
|---|
| 208 | | - if (!pte_none(*ptep) && pte_present(*ptep)) { |
|---|
| 209 | | - pfn = pte_pfn(*ptep); |
|---|
| 210 | | - pte_clear(&init_mm, (unsigned int)vaddr, ptep); |
|---|
| 211 | | - if (pfn_valid(pfn)) { |
|---|
| 212 | | - page = pfn_to_page(pfn); |
|---|
| 213 | | - __free_reserved_page(page); |
|---|
| 214 | | - } |
|---|
| 215 | | - } |
|---|
| 216 | | - vaddr += PAGE_SIZE; |
|---|
| 217 | | - } while (size -= PAGE_SIZE); |
|---|
| 218 | | - |
|---|
| 219 | | - /* flush tlb */ |
|---|
| 220 | | - flush_tlb_all(); |
|---|
| 221 | | -#endif |
|---|
| 222 | | -} |
|---|
| 52 | +#endif /* CONFIG_MMU */ |
|---|