include/linux/memremap.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_MEMREMAP_H_
   3 #define _LINUX_MEMREMAP_H_
   4 #include <linux/mm.h>
   5 #include <linux/ioport.h>
   6 #include <linux/percpu-refcount.h>
   7
   8 #include <asm/pgtable.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         const unsigned long base_pfn;
  23         const unsigned long reserve;
  24         unsigned long free;
  25         unsigned long align;
  26         unsigned long alloc;
  27 };
  28
  29 /*
  30  * Specialize ZONE_DEVICE memory into multiple types each having differents
  31  * usage.
  32  *
  33  * MEMORY_DEVICE_HOST:
  34  * Persistent device memory (pmem): struct page might be allocated in different
  35  * memory and architecture might want to perform special actions. It is similar
  36  * to regular memory, in that the CPU can access it transparently. However,
  37  * it is likely to have different bandwidth and latency than regular memory.
  38  * See Documentation/nvdimm/nvdimm.txt for more information.
  39  *
  40  * MEMORY_DEVICE_PRIVATE:
  41  * Device memory that is not directly addressable by the CPU: CPU can neither
  42  * read nor write private memory. In this case, we do still have struct pages
  43  * backing the device memory. Doing so simplifies the implementation, but it is
  44  * important to remember that there are certain points at which the struct page
  45  * must be treated as an opaque object, rather than a "normal" struct page.
  46  *
  47  * A more complete discussion of unaddressable memory may be found in
  48  * include/linux/hmm.h and Documentation/vm/hmm.txt.
  49  *
  50  * MEMORY_DEVICE_PUBLIC:
  51  * Device memory that is cache coherent from device and CPU point of view. This
  52  * is use on platform that have an advance system bus (like CAPI or CCIX). A
  53  * driver can hotplug the device memory using ZONE_DEVICE and with that memory
  54  * type. Any page of a process can be migrated to such memory. However no one
  55  * should be allow to pin such memory so that it can always be evicted.
  56  */
  57 enum memory_type {
  58         MEMORY_DEVICE_HOST = 0,
  59         MEMORY_DEVICE_PRIVATE,
  60         MEMORY_DEVICE_PUBLIC,
  61 };
  62
  63 /*
  64  * For MEMORY_DEVICE_PRIVATE we use ZONE_DEVICE and extend it with two
  65  * callbacks:
  66  *   page_fault()
  67  *   page_free()
  68  *
  69  * Additional notes about MEMORY_DEVICE_PRIVATE may be found in
  70  * include/linux/hmm.h and Documentation/vm/hmm.txt. There is also a brief
  71  * explanation in include/linux/memory_hotplug.h.
  72  *
  73  * The page_fault() callback must migrate page back, from device memory to
  74  * system memory, so that the CPU can access it. This might fail for various
  75  * reasons (device issues,  device have been unplugged, ...). When such error
  76  * conditions happen, the page_fault() callback must return VM_FAULT_SIGBUS and
  77  * set the CPU page table entry to "poisoned".
  78  *
  79  * Note that because memory cgroup charges are transferred to the device memory,
  80  * this should never fail due to memory restrictions. However, allocation
  81  * of a regular system page might still fail because we are out of memory. If
  82  * that happens, the page_fault() callback must return VM_FAULT_OOM.
  83  *
  84  * The page_fault() callback can also try to migrate back multiple pages in one
  85  * chunk, as an optimization. It must, however, prioritize the faulting address
  86  * over all the others.
  87  *
  88  *
  89  * The page_free() callback is called once the page refcount reaches 1
  90  * (ZONE_DEVICE pages never reach 0 refcount unless there is a refcount bug.
  91  * This allows the device driver to implement its own memory management.)
  92  *
  93  * For MEMORY_DEVICE_PUBLIC only the page_free() callback matter.
  94  */
  95 typedef int (*dev_page_fault_t)(struct vm_area_struct *vma,
  96                                 unsigned long addr,
  97                                 const struct page *page,
  98                                 unsigned int flags,
  99                                 pmd_t *pmdp);
 100 typedef void (*dev_page_free_t)(struct page *page, void *data);
 101
 102 /**
 103  * struct dev_pagemap - metadata for ZONE_DEVICE mappings
 104  * @page_fault: callback when CPU fault on an unaddressable device page
 105  * @page_free: free page callback when page refcount reaches 1
 106  * @altmap: pre-allocated/reserved memory for vmemmap allocations
 107  * @res: physical address range covered by @ref
 108  * @ref: reference count that pins the devm_memremap_pages() mapping
 109  * @dev: host device of the mapping for debug
 110  * @data: private data pointer for page_free()
 111  * @type: memory type: see MEMORY_* in memory_hotplug.h
 112  */
 113 struct dev_pagemap {
 114         dev_page_fault_t page_fault;
 115         dev_page_free_t page_free;
 116         struct vmem_altmap altmap;
 117         bool altmap_valid;
 118         struct resource res;
 119         struct percpu_ref *ref;
 120         struct device *dev;
 121         void *data;
 122         enum memory_type type;
 123 };
 124
 125 #ifdef CONFIG_ZONE_DEVICE
 126 void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap);
 127 struct dev_pagemap *get_dev_pagemap(unsigned long pfn,
 128                 struct dev_pagemap *pgmap);
 129
 130 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap);
 131 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns);
 132
 133 static inline bool is_zone_device_page(const struct page *page);
 134 #else
 135 static inline void *devm_memremap_pages(struct device *dev,
 136                 struct dev_pagemap *pgmap)
 137 {
 138         /*
 139          * Fail attempts to call devm_memremap_pages() without
 140          * ZONE_DEVICE support enabled, this requires callers to fall
 141          * back to plain devm_memremap() based on config
 142          */
 143         WARN_ON_ONCE(1);
 144         return ERR_PTR(-ENXIO);
 145 }
 146
 147 static inline struct dev_pagemap *get_dev_pagemap(unsigned long pfn,
 148                 struct dev_pagemap *pgmap)
 149 {
 150         return NULL;
 151 }
 152
 153 static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
 154 {
 155         return 0;
 156 }
 157
 158 static inline void vmem_altmap_free(struct vmem_altmap *altmap,
 159                 unsigned long nr_pfns)
 160 {
 161 }
 162 #endif /* CONFIG_ZONE_DEVICE */
 163
 164 #if defined(CONFIG_DEVICE_PRIVATE) || defined(CONFIG_DEVICE_PUBLIC)
 165 static inline bool is_device_private_page(const struct page *page)
 166 {
 167         return is_zone_device_page(page) &&
 168                 page->pgmap->type == MEMORY_DEVICE_PRIVATE;
 169 }
 170
 171 static inline bool is_device_public_page(const struct page *page)
 172 {
 173         return is_zone_device_page(page) &&
 174                 page->pgmap->type == MEMORY_DEVICE_PUBLIC;
 175 }
 176 #endif /* CONFIG_DEVICE_PRIVATE || CONFIG_DEVICE_PUBLIC */
 177
 178 static inline void put_dev_pagemap(struct dev_pagemap *pgmap)
 179 {
 180         if (pgmap)
 181                 percpu_ref_put(pgmap->ref);
 182 }
 183 #endif /* _LINUX_MEMREMAP_H_ */