]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
9476df7d DW |
2 | #ifndef _LINUX_MEMREMAP_H_ |
3 | #define _LINUX_MEMREMAP_H_ | |
5c2c2587 DW |
4 | #include <linux/ioport.h> |
5 | #include <linux/percpu-refcount.h> | |
9476df7d | 6 | |
5042db43 JG |
7 | #include <asm/pgtable.h> |
8 | ||
9476df7d DW |
9 | struct resource; |
10 | struct device; | |
4b94ffdc DW |
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 | ||
5042db43 JG |
28 | /* |
29 | * Specialize ZONE_DEVICE memory into multiple types each having differents | |
30 | * usage. | |
31 | * | |
5042db43 JG |
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 | |
ad56b738 | 40 | * include/linux/hmm.h and Documentation/vm/hmm.rst. |
df6ad698 JG |
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. | |
e7638488 DW |
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). | |
5042db43 JG |
56 | */ |
57 | enum memory_type { | |
e7638488 | 58 | MEMORY_DEVICE_PRIVATE = 1, |
df6ad698 | 59 | MEMORY_DEVICE_PUBLIC, |
e7638488 | 60 | MEMORY_DEVICE_FS_DAX, |
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 | |
ad56b738 | 70 | * include/linux/hmm.h and Documentation/vm/hmm.rst. There is also a brief |
5042db43 JG |
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 | |
e8d51348 | 126 | void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap); |
0822acb8 CH |
127 | struct dev_pagemap *get_dev_pagemap(unsigned long pfn, |
128 | struct dev_pagemap *pgmap); | |
7b2d55d2 | 129 | |
8e37d00a CH |
130 | unsigned long vmem_altmap_offset(struct vmem_altmap *altmap); |
131 | void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns); | |
9476df7d DW |
132 | #else |
133 | static inline void *devm_memremap_pages(struct device *dev, | |
e8d51348 | 134 | struct dev_pagemap *pgmap) |
9476df7d DW |
135 | { |
136 | /* | |
137 | * Fail attempts to call devm_memremap_pages() without | |
138 | * ZONE_DEVICE support enabled, this requires callers to fall | |
139 | * back to plain devm_memremap() based on config | |
140 | */ | |
141 | WARN_ON_ONCE(1); | |
142 | return ERR_PTR(-ENXIO); | |
143 | } | |
144 | ||
0822acb8 CH |
145 | static inline struct dev_pagemap *get_dev_pagemap(unsigned long pfn, |
146 | struct dev_pagemap *pgmap) | |
9476df7d DW |
147 | { |
148 | return NULL; | |
149 | } | |
8e37d00a CH |
150 | |
151 | static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap) | |
152 | { | |
153 | return 0; | |
154 | } | |
155 | ||
156 | static inline void vmem_altmap_free(struct vmem_altmap *altmap, | |
157 | unsigned long nr_pfns) | |
158 | { | |
159 | } | |
160 | #endif /* CONFIG_ZONE_DEVICE */ | |
7b2d55d2 | 161 | |
5c2c2587 DW |
162 | static inline void put_dev_pagemap(struct dev_pagemap *pgmap) |
163 | { | |
164 | if (pgmap) | |
165 | percpu_ref_put(pgmap->ref); | |
166 | } | |
9476df7d | 167 | #endif /* _LINUX_MEMREMAP_H_ */ |