1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_MEMREMAP_H_ 3 #define _LINUX_MEMREMAP_H_ 4 5 #include <linux/mmzone.h> 6 #include <linux/range.h> 7 #include <linux/ioport.h> 8 #include <linux/percpu-refcount.h> 9 10 struct resource; 11 struct device; 12 13 /** 14 * struct vmem_altmap - pre-allocated storage for vmemmap_populate 15 * @base_pfn: base of the entire dev_pagemap mapping 16 * @reserve: pages mapped, but reserved for driver use (relative to @base) 17 * @free: free pages set aside in the mapping for memmap storage 18 * @align: pages reserved to meet allocation alignments 19 * @alloc: track pages consumed, private to vmemmap_populate() 20 */ 21 struct vmem_altmap { 22 unsigned long base_pfn; 23 const unsigned long end_pfn; 24 const unsigned long reserve; 25 unsigned long free; 26 unsigned long align; 27 unsigned long alloc; 28 }; 29 30 /* 31 * Specialize ZONE_DEVICE memory into multiple types each has a different 32 * usage. 33 * 34 * MEMORY_DEVICE_PRIVATE: 35 * Device memory that is not directly addressable by the CPU: CPU can neither 36 * read nor write private memory. In this case, we do still have struct pages 37 * backing the device memory. Doing so simplifies the implementation, but it is 38 * important to remember that there are certain points at which the struct page 39 * must be treated as an opaque object, rather than a "normal" struct page. 40 * 41 * A more complete discussion of unaddressable memory may be found in 42 * include/linux/hmm.h and Documentation/mm/hmm.rst. 43 * 44 * MEMORY_DEVICE_COHERENT: 45 * Device memory that is cache coherent from device and CPU point of view. This 46 * is used on platforms that have an advanced system bus (like CAPI or CXL). A 47 * driver can hotplug the device memory using ZONE_DEVICE and with that memory 48 * type. Any page of a process can be migrated to such memory. However no one 49 * should be allowed to pin such memory so that it can always be evicted. 50 * 51 * MEMORY_DEVICE_FS_DAX: 52 * Host memory that has similar access semantics as System RAM i.e. DMA 53 * coherent and supports page pinning. In support of coordinating page 54 * pinning vs other operations MEMORY_DEVICE_FS_DAX arranges for a 55 * wakeup event whenever a page is unpinned and becomes idle. This 56 * wakeup is used to coordinate physical address space management (ex: 57 * fs truncate/hole punch) vs pinned pages (ex: device dma). 58 * 59 * MEMORY_DEVICE_GENERIC: 60 * Host memory that has similar access semantics as System RAM i.e. DMA 61 * coherent and supports page pinning. This is for example used by DAX devices 62 * that expose memory using a character device. 63 * 64 * MEMORY_DEVICE_PCI_P2PDMA: 65 * Device memory residing in a PCI BAR intended for use with Peer-to-Peer 66 * transactions. 67 */ 68 enum memory_type { 69 /* 0 is reserved to catch uninitialized type fields */ 70 MEMORY_DEVICE_PRIVATE = 1, 71 MEMORY_DEVICE_COHERENT, 72 MEMORY_DEVICE_FS_DAX, 73 MEMORY_DEVICE_GENERIC, 74 MEMORY_DEVICE_PCI_P2PDMA, 75 }; 76 77 struct dev_pagemap_ops { 78 /* 79 * Called once the page refcount reaches 0. The reference count will be 80 * reset to one by the core code after the method is called to prepare 81 * for handing out the page again. 82 */ 83 void (*page_free)(struct page *page); 84 85 /* 86 * Used for private (un-addressable) device memory only. Must migrate 87 * the page back to a CPU accessible page. 88 */ 89 vm_fault_t (*migrate_to_ram)(struct vm_fault *vmf); 90 }; 91 92 #define PGMAP_ALTMAP_VALID (1 << 0) 93 94 /** 95 * struct dev_pagemap - metadata for ZONE_DEVICE mappings 96 * @altmap: pre-allocated/reserved memory for vmemmap allocations 97 * @ref: reference count that pins the devm_memremap_pages() mapping 98 * @done: completion for @ref 99 * @type: memory type: see MEMORY_* in memory_hotplug.h 100 * @flags: PGMAP_* flags to specify defailed behavior 101 * @vmemmap_shift: structural definition of how the vmemmap page metadata 102 * is populated, specifically the metadata page order. 103 * A zero value (default) uses base pages as the vmemmap metadata 104 * representation. A bigger value will set up compound struct pages 105 * of the requested order value. 106 * @ops: method table 107 * @owner: an opaque pointer identifying the entity that manages this 108 * instance. Used by various helpers to make sure that no 109 * foreign ZONE_DEVICE memory is accessed. 110 * @nr_range: number of ranges to be mapped 111 * @range: range to be mapped when nr_range == 1 112 * @ranges: array of ranges to be mapped when nr_range > 1 113 */ 114 struct dev_pagemap { 115 struct vmem_altmap altmap; 116 struct percpu_ref ref; 117 struct completion done; 118 enum memory_type type; 119 unsigned int flags; 120 unsigned long vmemmap_shift; 121 const struct dev_pagemap_ops *ops; 122 void *owner; 123 int nr_range; 124 union { 125 struct range range; 126 struct range ranges[0]; 127 }; 128 }; 129 130 static inline struct vmem_altmap *pgmap_altmap(struct dev_pagemap *pgmap) 131 { 132 if (pgmap->flags & PGMAP_ALTMAP_VALID) 133 return &pgmap->altmap; 134 return NULL; 135 } 136 137 static inline unsigned long pgmap_vmemmap_nr(struct dev_pagemap *pgmap) 138 { 139 return 1 << pgmap->vmemmap_shift; 140 } 141 142 static inline bool is_device_private_page(const struct page *page) 143 { 144 return IS_ENABLED(CONFIG_DEVICE_PRIVATE) && 145 is_zone_device_page(page) && 146 page->pgmap->type == MEMORY_DEVICE_PRIVATE; 147 } 148 149 static inline bool folio_is_device_private(const struct folio *folio) 150 { 151 return is_device_private_page(&folio->page); 152 } 153 154 static inline bool is_pci_p2pdma_page(const struct page *page) 155 { 156 return IS_ENABLED(CONFIG_PCI_P2PDMA) && 157 is_zone_device_page(page) && 158 page->pgmap->type == MEMORY_DEVICE_PCI_P2PDMA; 159 } 160 161 static inline bool is_device_coherent_page(const struct page *page) 162 { 163 return is_zone_device_page(page) && 164 page->pgmap->type == MEMORY_DEVICE_COHERENT; 165 } 166 167 static inline bool folio_is_device_coherent(const struct folio *folio) 168 { 169 return is_device_coherent_page(&folio->page); 170 } 171 172 #ifdef CONFIG_ZONE_DEVICE 173 void *memremap_pages(struct dev_pagemap *pgmap, int nid); 174 void memunmap_pages(struct dev_pagemap *pgmap); 175 void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap); 176 void devm_memunmap_pages(struct device *dev, struct dev_pagemap *pgmap); 177 struct dev_pagemap *get_dev_pagemap(unsigned long pfn, 178 struct dev_pagemap *pgmap); 179 bool pgmap_pfn_valid(struct dev_pagemap *pgmap, unsigned long pfn); 180 181 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap); 182 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns); 183 unsigned long memremap_compat_align(void); 184 #else 185 static inline void *devm_memremap_pages(struct device *dev, 186 struct dev_pagemap *pgmap) 187 { 188 /* 189 * Fail attempts to call devm_memremap_pages() without 190 * ZONE_DEVICE support enabled, this requires callers to fall 191 * back to plain devm_memremap() based on config 192 */ 193 WARN_ON_ONCE(1); 194 return ERR_PTR(-ENXIO); 195 } 196 197 static inline void devm_memunmap_pages(struct device *dev, 198 struct dev_pagemap *pgmap) 199 { 200 } 201 202 static inline struct dev_pagemap *get_dev_pagemap(unsigned long pfn, 203 struct dev_pagemap *pgmap) 204 { 205 return NULL; 206 } 207 208 static inline bool pgmap_pfn_valid(struct dev_pagemap *pgmap, unsigned long pfn) 209 { 210 return false; 211 } 212 213 static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap) 214 { 215 return 0; 216 } 217 218 static inline void vmem_altmap_free(struct vmem_altmap *altmap, 219 unsigned long nr_pfns) 220 { 221 } 222 223 /* when memremap_pages() is disabled all archs can remap a single page */ 224 static inline unsigned long memremap_compat_align(void) 225 { 226 return PAGE_SIZE; 227 } 228 #endif /* CONFIG_ZONE_DEVICE */ 229 230 static inline void put_dev_pagemap(struct dev_pagemap *pgmap) 231 { 232 if (pgmap) 233 percpu_ref_put(&pgmap->ref); 234 } 235 236 #endif /* _LINUX_MEMREMAP_H_ */ 237