1 /*
2 * Copyright (c) 2015 Oracle. All rights reserved.
3 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
4 */
5
6 /* Lightweight memory registration using Fast Memory Regions (FMR).
7 * Referred to sometimes as MTHCAFMR mode.
8 *
9 * FMR uses synchronous memory registration and deregistration.
10 * FMR registration is known to be fast, but FMR deregistration
11 * can take tens of usecs to complete.
12 */
13
14 #include "xprt_rdma.h"
15
16 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
17 # define RPCDBG_FACILITY RPCDBG_TRANS
18 #endif
19
20 /* Maximum scatter/gather per FMR */
21 #define RPCRDMA_MAX_FMR_SGES (64)
22
23 static int
fmr_op_open(struct rpcrdma_ia * ia,struct rpcrdma_ep * ep,struct rpcrdma_create_data_internal * cdata)24 fmr_op_open(struct rpcrdma_ia *ia, struct rpcrdma_ep *ep,
25 struct rpcrdma_create_data_internal *cdata)
26 {
27 return 0;
28 }
29
30 /* FMR mode conveys up to 64 pages of payload per chunk segment.
31 */
32 static size_t
fmr_op_maxpages(struct rpcrdma_xprt * r_xprt)33 fmr_op_maxpages(struct rpcrdma_xprt *r_xprt)
34 {
35 return min_t(unsigned int, RPCRDMA_MAX_DATA_SEGS,
36 rpcrdma_max_segments(r_xprt) * RPCRDMA_MAX_FMR_SGES);
37 }
38
39 static int
fmr_op_init(struct rpcrdma_xprt * r_xprt)40 fmr_op_init(struct rpcrdma_xprt *r_xprt)
41 {
42 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
43 int mr_access_flags = IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ;
44 struct ib_fmr_attr fmr_attr = {
45 .max_pages = RPCRDMA_MAX_FMR_SGES,
46 .max_maps = 1,
47 .page_shift = PAGE_SHIFT
48 };
49 struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
50 struct rpcrdma_mw *r;
51 int i, rc;
52
53 INIT_LIST_HEAD(&buf->rb_mws);
54 INIT_LIST_HEAD(&buf->rb_all);
55
56 i = (buf->rb_max_requests + 1) * RPCRDMA_MAX_SEGS;
57 dprintk("RPC: %s: initializing %d FMRs\n", __func__, i);
58
59 while (i--) {
60 r = kzalloc(sizeof(*r), GFP_KERNEL);
61 if (!r)
62 return -ENOMEM;
63
64 r->r.fmr = ib_alloc_fmr(pd, mr_access_flags, &fmr_attr);
65 if (IS_ERR(r->r.fmr))
66 goto out_fmr_err;
67
68 list_add(&r->mw_list, &buf->rb_mws);
69 list_add(&r->mw_all, &buf->rb_all);
70 }
71 return 0;
72
73 out_fmr_err:
74 rc = PTR_ERR(r->r.fmr);
75 dprintk("RPC: %s: ib_alloc_fmr status %i\n", __func__, rc);
76 kfree(r);
77 return rc;
78 }
79
80 /* Use the ib_map_phys_fmr() verb to register a memory region
81 * for remote access via RDMA READ or RDMA WRITE.
82 */
83 static int
fmr_op_map(struct rpcrdma_xprt * r_xprt,struct rpcrdma_mr_seg * seg,int nsegs,bool writing)84 fmr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
85 int nsegs, bool writing)
86 {
87 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
88 struct ib_device *device = ia->ri_id->device;
89 enum dma_data_direction direction = rpcrdma_data_dir(writing);
90 struct rpcrdma_mr_seg *seg1 = seg;
91 struct rpcrdma_mw *mw = seg1->rl_mw;
92 u64 physaddrs[RPCRDMA_MAX_DATA_SEGS];
93 int len, pageoff, i, rc;
94
95 pageoff = offset_in_page(seg1->mr_offset);
96 seg1->mr_offset -= pageoff; /* start of page */
97 seg1->mr_len += pageoff;
98 len = -pageoff;
99 if (nsegs > RPCRDMA_MAX_FMR_SGES)
100 nsegs = RPCRDMA_MAX_FMR_SGES;
101 for (i = 0; i < nsegs;) {
102 rpcrdma_map_one(device, seg, direction);
103 physaddrs[i] = seg->mr_dma;
104 len += seg->mr_len;
105 ++seg;
106 ++i;
107 /* Check for holes */
108 if ((i < nsegs && offset_in_page(seg->mr_offset)) ||
109 offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
110 break;
111 }
112
113 rc = ib_map_phys_fmr(mw->r.fmr, physaddrs, i, seg1->mr_dma);
114 if (rc)
115 goto out_maperr;
116
117 seg1->mr_rkey = mw->r.fmr->rkey;
118 seg1->mr_base = seg1->mr_dma + pageoff;
119 seg1->mr_nsegs = i;
120 seg1->mr_len = len;
121 return i;
122
123 out_maperr:
124 dprintk("RPC: %s: ib_map_phys_fmr %u@0x%llx+%i (%d) status %i\n",
125 __func__, len, (unsigned long long)seg1->mr_dma,
126 pageoff, i, rc);
127 while (i--)
128 rpcrdma_unmap_one(device, --seg);
129 return rc;
130 }
131
132 /* Use the ib_unmap_fmr() verb to prevent further remote
133 * access via RDMA READ or RDMA WRITE.
134 */
135 static int
fmr_op_unmap(struct rpcrdma_xprt * r_xprt,struct rpcrdma_mr_seg * seg)136 fmr_op_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg)
137 {
138 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
139 struct rpcrdma_mr_seg *seg1 = seg;
140 struct ib_device *device;
141 int rc, nsegs = seg->mr_nsegs;
142 LIST_HEAD(l);
143
144 list_add(&seg1->rl_mw->r.fmr->list, &l);
145 rc = ib_unmap_fmr(&l);
146 read_lock(&ia->ri_qplock);
147 device = ia->ri_id->device;
148 while (seg1->mr_nsegs--)
149 rpcrdma_unmap_one(device, seg++);
150 read_unlock(&ia->ri_qplock);
151 if (rc)
152 goto out_err;
153 return nsegs;
154
155 out_err:
156 dprintk("RPC: %s: ib_unmap_fmr status %i\n", __func__, rc);
157 return nsegs;
158 }
159
160 /* After a disconnect, unmap all FMRs.
161 *
162 * This is invoked only in the transport connect worker in order
163 * to serialize with rpcrdma_register_fmr_external().
164 */
165 static void
fmr_op_reset(struct rpcrdma_xprt * r_xprt)166 fmr_op_reset(struct rpcrdma_xprt *r_xprt)
167 {
168 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
169 struct rpcrdma_mw *r;
170 LIST_HEAD(list);
171 int rc;
172
173 list_for_each_entry(r, &buf->rb_all, mw_all)
174 list_add(&r->r.fmr->list, &list);
175
176 rc = ib_unmap_fmr(&list);
177 if (rc)
178 dprintk("RPC: %s: ib_unmap_fmr failed %i\n",
179 __func__, rc);
180 }
181
182 static void
fmr_op_destroy(struct rpcrdma_buffer * buf)183 fmr_op_destroy(struct rpcrdma_buffer *buf)
184 {
185 struct rpcrdma_mw *r;
186 int rc;
187
188 while (!list_empty(&buf->rb_all)) {
189 r = list_entry(buf->rb_all.next, struct rpcrdma_mw, mw_all);
190 list_del(&r->mw_all);
191 rc = ib_dealloc_fmr(r->r.fmr);
192 if (rc)
193 dprintk("RPC: %s: ib_dealloc_fmr failed %i\n",
194 __func__, rc);
195 kfree(r);
196 }
197 }
198
199 const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops = {
200 .ro_map = fmr_op_map,
201 .ro_unmap = fmr_op_unmap,
202 .ro_open = fmr_op_open,
203 .ro_maxpages = fmr_op_maxpages,
204 .ro_init = fmr_op_init,
205 .ro_reset = fmr_op_reset,
206 .ro_destroy = fmr_op_destroy,
207 .ro_displayname = "fmr",
208 };
209