Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
9476df7d DW |
2 | #ifndef _LINUX_MEMREMAP_H_ |
3 | #define _LINUX_MEMREMAP_H_ | |
4 | #include <linux/mm.h> | |
5c2c2587 DW |
5 | #include <linux/ioport.h> |
6 | #include <linux/percpu-refcount.h> | |
9476df7d | 7 | |
5042db43 JG |
8 | #include <asm/pgtable.h> |
9 | ||
9476df7d DW |
10 | struct resource; |
11 | struct device; | |
4b94ffdc DW |
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 | ||
5042db43 JG |
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. | |
df6ad698 JG |
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. | |
5042db43 JG |
56 | */ |
57 | enum memory_type { | |
58 | MEMORY_DEVICE_HOST = 0, | |
59 | MEMORY_DEVICE_PRIVATE, | |
df6ad698 | 60 | MEMORY_DEVICE_PUBLIC, |
5042db43 JG |
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.) | |
df6ad698 JG |
92 | * |
93 | * For MEMORY_DEVICE_PUBLIC only the page_free() callback matter. | |
5042db43 JG |
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 | ||
9476df7d DW |
102 | /** |
103 | * struct dev_pagemap - metadata for ZONE_DEVICE mappings | |
5042db43 JG |
104 | * @page_fault: callback when CPU fault on an unaddressable device page |
105 | * @page_free: free page callback when page refcount reaches 1 | |
4b94ffdc | 106 | * @altmap: pre-allocated/reserved memory for vmemmap allocations |
5c2c2587 DW |
107 | * @res: physical address range covered by @ref |
108 | * @ref: reference count that pins the devm_memremap_pages() mapping | |
9476df7d | 109 | * @dev: host device of the mapping for debug |
5042db43 JG |
110 | * @data: private data pointer for page_free() |
111 | * @type: memory type: see MEMORY_* in memory_hotplug.h | |
9476df7d DW |
112 | */ |
113 | struct dev_pagemap { | |
5042db43 JG |
114 | dev_page_fault_t page_fault; |
115 | dev_page_free_t page_free; | |
e7744aa2 LG |
116 | struct vmem_altmap altmap; |
117 | bool altmap_valid; | |
118 | struct resource res; | |
5c2c2587 | 119 | struct percpu_ref *ref; |
9476df7d | 120 | struct device *dev; |
5042db43 JG |
121 | void *data; |
122 | enum memory_type type; | |
9476df7d DW |
123 | }; |
124 | ||
125 | #ifdef CONFIG_ZONE_DEVICE | |
4b94ffdc | 126 | void *devm_memremap_pages(struct device *dev, struct resource *res, |
5c2c2587 | 127 | struct percpu_ref *ref, struct vmem_altmap *altmap); |
0822acb8 CH |
128 | struct dev_pagemap *get_dev_pagemap(unsigned long pfn, |
129 | struct dev_pagemap *pgmap); | |
7b2d55d2 | 130 | |
8e37d00a CH |
131 | unsigned long vmem_altmap_offset(struct vmem_altmap *altmap); |
132 | void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns); | |
133 | ||
7b2d55d2 | 134 | static inline bool is_zone_device_page(const struct page *page); |
9476df7d DW |
135 | #else |
136 | static inline void *devm_memremap_pages(struct device *dev, | |
5c2c2587 DW |
137 | struct resource *res, struct percpu_ref *ref, |
138 | struct vmem_altmap *altmap) | |
9476df7d DW |
139 | { |
140 | /* | |
141 | * Fail attempts to call devm_memremap_pages() without | |
142 | * ZONE_DEVICE support enabled, this requires callers to fall | |
143 | * back to plain devm_memremap() based on config | |
144 | */ | |
145 | WARN_ON_ONCE(1); | |
146 | return ERR_PTR(-ENXIO); | |
147 | } | |
148 | ||
0822acb8 CH |
149 | static inline struct dev_pagemap *get_dev_pagemap(unsigned long pfn, |
150 | struct dev_pagemap *pgmap) | |
9476df7d DW |
151 | { |
152 | return NULL; | |
153 | } | |
8e37d00a CH |
154 | |
155 | static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap) | |
156 | { | |
157 | return 0; | |
158 | } | |
159 | ||
160 | static inline void vmem_altmap_free(struct vmem_altmap *altmap, | |
161 | unsigned long nr_pfns) | |
162 | { | |
163 | } | |
164 | #endif /* CONFIG_ZONE_DEVICE */ | |
7b2d55d2 | 165 | |
6b368cd4 | 166 | #if defined(CONFIG_DEVICE_PRIVATE) || defined(CONFIG_DEVICE_PUBLIC) |
7b2d55d2 JG |
167 | static inline bool is_device_private_page(const struct page *page) |
168 | { | |
6b368cd4 JG |
169 | return is_zone_device_page(page) && |
170 | page->pgmap->type == MEMORY_DEVICE_PRIVATE; | |
7b2d55d2 | 171 | } |
df6ad698 JG |
172 | |
173 | static inline bool is_device_public_page(const struct page *page) | |
174 | { | |
6b368cd4 JG |
175 | return is_zone_device_page(page) && |
176 | page->pgmap->type == MEMORY_DEVICE_PUBLIC; | |
df6ad698 | 177 | } |
6b368cd4 | 178 | #endif /* CONFIG_DEVICE_PRIVATE || CONFIG_DEVICE_PUBLIC */ |
5c2c2587 | 179 | |
5c2c2587 DW |
180 | static inline void put_dev_pagemap(struct dev_pagemap *pgmap) |
181 | { | |
182 | if (pgmap) | |
183 | percpu_ref_put(pgmap->ref); | |
184 | } | |
9476df7d | 185 | #endif /* _LINUX_MEMREMAP_H_ */ |