1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3 *
4 * Copyright (c) 2014 Mellanox Technologies. All rights reserved.
5 *
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
33 */
34
35 #ifndef IB_UMEM_ODP_H
36 #define IB_UMEM_ODP_H
37
38 #include <linux/rbtree.h>
39
40 #include <rdma/ib_umem.h>
41 #include <rdma/ib_verbs.h>
42
43 struct umem_odp_node {
44 u64 __subtree_last;
45 struct rb_node rb;
46 };
47
48 struct ib_umem_odp {
49 /*
50 * An array of the pages included in the on-demand paging umem.
51 * Indices of pages that are currently not mapped into the device will
52 * contain NULL.
53 */
54 struct page **page_list;
55 /*
56 * An array of the same size as page_list, with DMA addresses mapped
57 * for pages the pages in page_list. The lower two bits designate
58 * access permissions. See ODP_READ_ALLOWED_BIT and
59 * ODP_WRITE_ALLOWED_BIT.
60 */
61 dma_addr_t *dma_list;
62 /*
63 * The umem_mutex protects the page_list and dma_list fields of an ODP
64 * umem, allowing only a single thread to map/unmap pages. The mutex
65 * also protects access to the mmu notifier counters.
66 */
67 struct mutex umem_mutex;
68 void *private; /* for the HW driver to use. */
69
70 /* When false, use the notifier counter in the ucontext struct. */
71 bool mn_counters_active;
72 int notifiers_seq;
73 int notifiers_count;
74
75 /* A linked list of umems that don't have private mmu notifier
76 * counters yet. */
77 struct list_head no_private_counters;
78 struct ib_umem *umem;
79
80 /* Tree tracking */
81 struct umem_odp_node interval_tree;
82
83 struct completion notifier_completion;
84 int dying;
85 };
86
87 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
88
89 int ib_umem_odp_get(struct ib_ucontext *context, struct ib_umem *umem);
90
91 void ib_umem_odp_release(struct ib_umem *umem);
92
93 /*
94 * The lower 2 bits of the DMA address signal the R/W permissions for
95 * the entry. To upgrade the permissions, provide the appropriate
96 * bitmask to the map_dma_pages function.
97 *
98 * Be aware that upgrading a mapped address might result in change of
99 * the DMA address for the page.
100 */
101 #define ODP_READ_ALLOWED_BIT (1<<0ULL)
102 #define ODP_WRITE_ALLOWED_BIT (1<<1ULL)
103
104 #define ODP_DMA_ADDR_MASK (~(ODP_READ_ALLOWED_BIT | ODP_WRITE_ALLOWED_BIT))
105
106 int ib_umem_odp_map_dma_pages(struct ib_umem *umem, u64 start_offset, u64 bcnt,
107 u64 access_mask, unsigned long current_seq);
108
109 void ib_umem_odp_unmap_dma_pages(struct ib_umem *umem, u64 start_offset,
110 u64 bound);
111
112 void rbt_ib_umem_insert(struct umem_odp_node *node, struct rb_root *root);
113 void rbt_ib_umem_remove(struct umem_odp_node *node, struct rb_root *root);
114 typedef int (*umem_call_back)(struct ib_umem *item, u64 start, u64 end,
115 void *cookie);
116 /*
117 * Call the callback on each ib_umem in the range. Returns the logical or of
118 * the return values of the functions called.
119 */
120 int rbt_ib_umem_for_each_in_range(struct rb_root *root, u64 start, u64 end,
121 umem_call_back cb, void *cookie);
122
123 struct umem_odp_node *rbt_ib_umem_iter_first(struct rb_root *root,
124 u64 start, u64 last);
125 struct umem_odp_node *rbt_ib_umem_iter_next(struct umem_odp_node *node,
126 u64 start, u64 last);
127
ib_umem_mmu_notifier_retry(struct ib_umem * item,unsigned long mmu_seq)128 static inline int ib_umem_mmu_notifier_retry(struct ib_umem *item,
129 unsigned long mmu_seq)
130 {
131 /*
132 * This code is strongly based on the KVM code from
133 * mmu_notifier_retry. Should be called with
134 * the relevant locks taken (item->odp_data->umem_mutex
135 * and the ucontext umem_mutex semaphore locked for read).
136 */
137
138 /* Do not allow page faults while the new ib_umem hasn't seen a state
139 * with zero notifiers yet, and doesn't have its own valid set of
140 * private counters. */
141 if (!item->odp_data->mn_counters_active)
142 return 1;
143
144 if (unlikely(item->odp_data->notifiers_count))
145 return 1;
146 if (item->odp_data->notifiers_seq != mmu_seq)
147 return 1;
148 return 0;
149 }
150
151 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
152
ib_umem_odp_get(struct ib_ucontext * context,struct ib_umem * umem)153 static inline int ib_umem_odp_get(struct ib_ucontext *context,
154 struct ib_umem *umem)
155 {
156 return -EINVAL;
157 }
158
ib_umem_odp_release(struct ib_umem * umem)159 static inline void ib_umem_odp_release(struct ib_umem *umem) {}
160
161 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
162
163 #endif /* IB_UMEM_ODP_H */
164