1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_MEMREMAP_H_ 3 #define _LINUX_MEMREMAP_H_ 4 #include <linux/ioport.h> 5 #include <linux/percpu-refcount.h> 6 7 #include <asm/pgtable.h> 8 9 struct resource; 10 struct device; 11 12 /** 13 * struct vmem_altmap - pre-allocated storage for vmemmap_populate 14 * @base_pfn: base of the entire dev_pagemap mapping 15 * @reserve: pages mapped, but reserved for driver use (relative to @base) 16 * @free: free pages set aside in the mapping for memmap storage 17 * @align: pages reserved to meet allocation alignments 18 * @alloc: track pages consumed, private to vmemmap_populate() 19 */ 20 struct vmem_altmap { 21 const unsigned long base_pfn; 22 const unsigned long reserve; 23 unsigned long free; 24 unsigned long align; 25 unsigned long alloc; 26 }; 27 28 /* 29 * Specialize ZONE_DEVICE memory into multiple types each having differents 30 * usage. 31 * 32 * MEMORY_DEVICE_PRIVATE: 33 * Device memory that is not directly addressable by the CPU: CPU can neither 34 * read nor write private memory. In this case, we do still have struct pages 35 * backing the device memory. Doing so simplifies the implementation, but it is 36 * important to remember that there are certain points at which the struct page 37 * must be treated as an opaque object, rather than a "normal" struct page. 38 * 39 * A more complete discussion of unaddressable memory may be found in 40 * include/linux/hmm.h and Documentation/vm/hmm.rst. 41 * 42 * MEMORY_DEVICE_PUBLIC: 43 * Device memory that is cache coherent from device and CPU point of view. This 44 * is use on platform that have an advance system bus (like CAPI or CCIX). A 45 * driver can hotplug the device memory using ZONE_DEVICE and with that memory 46 * type. Any page of a process can be migrated to such memory. However no one 47 * should be allow to pin such memory so that it can always be evicted. 48 * 49 * MEMORY_DEVICE_FS_DAX: 50 * Host memory that has similar access semantics as System RAM i.e. DMA 51 * coherent and supports page pinning. In support of coordinating page 52 * pinning vs other operations MEMORY_DEVICE_FS_DAX arranges for a 53 * wakeup event whenever a page is unpinned and becomes idle. This 54 * wakeup is used to coordinate physical address space management (ex: 55 * fs truncate/hole punch) vs pinned pages (ex: device dma). 56 * 57 * MEMORY_DEVICE_PCI_P2PDMA: 58 * Device memory residing in a PCI BAR intended for use with Peer-to-Peer 59 * transactions. 60 */ 61 enum memory_type { 62 MEMORY_DEVICE_PRIVATE = 1, 63 MEMORY_DEVICE_PUBLIC, 64 MEMORY_DEVICE_FS_DAX, 65 MEMORY_DEVICE_PCI_P2PDMA, 66 }; 67 68 /* 69 * For MEMORY_DEVICE_PRIVATE we use ZONE_DEVICE and extend it with two 70 * callbacks: 71 * page_fault() 72 * page_free() 73 * 74 * Additional notes about MEMORY_DEVICE_PRIVATE may be found in 75 * include/linux/hmm.h and Documentation/vm/hmm.rst. There is also a brief 76 * explanation in include/linux/memory_hotplug.h. 77 * 78 * The page_fault() callback must migrate page back, from device memory to 79 * system memory, so that the CPU can access it. This might fail for various 80 * reasons (device issues, device have been unplugged, ...). When such error 81 * conditions happen, the page_fault() callback must return VM_FAULT_SIGBUS and 82 * set the CPU page table entry to "poisoned". 83 * 84 * Note that because memory cgroup charges are transferred to the device memory, 85 * this should never fail due to memory restrictions. However, allocation 86 * of a regular system page might still fail because we are out of memory. If 87 * that happens, the page_fault() callback must return VM_FAULT_OOM. 88 * 89 * The page_fault() callback can also try to migrate back multiple pages in one 90 * chunk, as an optimization. It must, however, prioritize the faulting address 91 * over all the others. 92 * 93 * 94 * The page_free() callback is called once the page refcount reaches 1 95 * (ZONE_DEVICE pages never reach 0 refcount unless there is a refcount bug. 96 * This allows the device driver to implement its own memory management.) 97 * 98 * For MEMORY_DEVICE_PUBLIC only the page_free() callback matter. 99 */ 100 typedef int (*dev_page_fault_t)(struct vm_area_struct *vma, 101 unsigned long addr, 102 const struct page *page, 103 unsigned int flags, 104 pmd_t *pmdp); 105 typedef void (*dev_page_free_t)(struct page *page, void *data); 106 107 /** 108 * struct dev_pagemap - metadata for ZONE_DEVICE mappings 109 * @page_fault: callback when CPU fault on an unaddressable device page 110 * @page_free: free page callback when page refcount reaches 1 111 * @altmap: pre-allocated/reserved memory for vmemmap allocations 112 * @res: physical address range covered by @ref 113 * @ref: reference count that pins the devm_memremap_pages() mapping 114 * @dev: host device of the mapping for debug 115 * @data: private data pointer for page_free() 116 * @type: memory type: see MEMORY_* in memory_hotplug.h 117 */ 118 struct dev_pagemap { 119 dev_page_fault_t page_fault; 120 dev_page_free_t page_free; 121 struct vmem_altmap altmap; 122 bool altmap_valid; 123 struct resource res; 124 struct percpu_ref *ref; 125 struct device *dev; 126 void *data; 127 enum memory_type type; 128 u64 pci_p2pdma_bus_offset; 129 }; 130 131 #ifdef CONFIG_ZONE_DEVICE 132 void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap); 133 struct dev_pagemap *get_dev_pagemap(unsigned long pfn, 134 struct dev_pagemap *pgmap); 135 136 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap); 137 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns); 138 #else 139 static inline void *devm_memremap_pages(struct device *dev, 140 struct dev_pagemap *pgmap) 141 { 142 /* 143 * Fail attempts to call devm_memremap_pages() without 144 * ZONE_DEVICE support enabled, this requires callers to fall 145 * back to plain devm_memremap() based on config 146 */ 147 WARN_ON_ONCE(1); 148 return ERR_PTR(-ENXIO); 149 } 150 151 static inline struct dev_pagemap *get_dev_pagemap(unsigned long pfn, 152 struct dev_pagemap *pgmap) 153 { 154 return NULL; 155 } 156 157 static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap) 158 { 159 return 0; 160 } 161 162 static inline void vmem_altmap_free(struct vmem_altmap *altmap, 163 unsigned long nr_pfns) 164 { 165 } 166 #endif /* CONFIG_ZONE_DEVICE */ 167 168 static inline void put_dev_pagemap(struct dev_pagemap *pgmap) 169 { 170 if (pgmap) 171 percpu_ref_put(pgmap->ref); 172 } 173 #endif /* _LINUX_MEMREMAP_H_ */ 174