2 * Copyright (c) 2015 Oracle. All rights reserved.
3 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
6 /* Lightweight memory registration using Fast Registration Work
7 * Requests (FRWR). Also referred to sometimes as FRMR mode.
9 * FRWR features ordered asynchronous registration and deregistration
10 * of arbitrarily sized memory regions. This is the fastest and safest
11 * but most complex memory registration mode.
14 #include "xprt_rdma.h"
16 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
17 # define RPCDBG_FACILITY RPCDBG_TRANS
21 __frwr_init(struct rpcrdma_mw *r, struct ib_pd *pd, struct ib_device *device,
24 struct rpcrdma_frmr *f = &r->r.frmr;
27 f->fr_mr = ib_alloc_fast_reg_mr(pd, depth);
30 f->fr_pgl = ib_alloc_fast_reg_page_list(device, depth);
31 if (IS_ERR(f->fr_pgl))
36 rc = PTR_ERR(f->fr_mr);
37 dprintk("RPC: %s: ib_alloc_fast_reg_mr status %i\n",
42 rc = PTR_ERR(f->fr_pgl);
43 dprintk("RPC: %s: ib_alloc_fast_reg_page_list status %i\n",
45 ib_dereg_mr(f->fr_mr);
50 __frwr_release(struct rpcrdma_mw *r)
54 rc = ib_dereg_mr(r->r.frmr.fr_mr);
56 dprintk("RPC: %s: ib_dereg_mr status %i\n",
58 ib_free_fast_reg_page_list(r->r.frmr.fr_pgl);
61 /* FRWR mode conveys a list of pages per chunk segment. The
62 * maximum length of that list is the FRWR page list depth.
65 frwr_op_maxpages(struct rpcrdma_xprt *r_xprt)
67 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
69 return min_t(unsigned int, RPCRDMA_MAX_DATA_SEGS,
70 rpcrdma_max_segments(r_xprt) * ia->ri_max_frmr_depth);
74 frwr_op_init(struct rpcrdma_xprt *r_xprt)
76 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
77 struct ib_device *device = r_xprt->rx_ia.ri_id->device;
78 unsigned int depth = r_xprt->rx_ia.ri_max_frmr_depth;
79 struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
82 INIT_LIST_HEAD(&buf->rb_mws);
83 INIT_LIST_HEAD(&buf->rb_all);
85 i = (buf->rb_max_requests + 1) * RPCRDMA_MAX_SEGS;
86 dprintk("RPC: %s: initalizing %d FRMRs\n", __func__, i);
92 r = kzalloc(sizeof(*r), GFP_KERNEL);
96 rc = __frwr_init(r, pd, device, depth);
102 list_add(&r->mw_list, &buf->rb_mws);
103 list_add(&r->mw_all, &buf->rb_all);
109 /* Post a FAST_REG Work Request to register a memory region
110 * for remote access via RDMA READ or RDMA WRITE.
113 frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
114 int nsegs, bool writing)
116 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
117 struct rpcrdma_mr_seg *seg1 = seg;
118 struct rpcrdma_mw *mw = seg1->rl_mw;
119 struct rpcrdma_frmr *frmr = &mw->r.frmr;
120 struct ib_mr *mr = frmr->fr_mr;
121 struct ib_send_wr fastreg_wr, *bad_wr;
129 pageoff = offset_in_page(seg1->mr_offset);
130 seg1->mr_offset -= pageoff; /* start of page */
131 seg1->mr_len += pageoff;
133 if (nsegs > ia->ri_max_frmr_depth)
134 nsegs = ia->ri_max_frmr_depth;
135 for (page_no = i = 0; i < nsegs;) {
136 rpcrdma_map_one(ia, seg, writing);
138 for (seg_len = seg->mr_len; seg_len > 0; seg_len -= PAGE_SIZE) {
139 frmr->fr_pgl->page_list[page_no++] = pa;
145 /* Check for holes */
146 if ((i < nsegs && offset_in_page(seg->mr_offset)) ||
147 offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
150 dprintk("RPC: %s: Using frmr %p to map %d segments (%d bytes)\n",
151 __func__, mw, i, len);
153 frmr->fr_state = FRMR_IS_VALID;
155 memset(&fastreg_wr, 0, sizeof(fastreg_wr));
156 fastreg_wr.wr_id = (unsigned long)(void *)mw;
157 fastreg_wr.opcode = IB_WR_FAST_REG_MR;
158 fastreg_wr.wr.fast_reg.iova_start = seg1->mr_dma + pageoff;
159 fastreg_wr.wr.fast_reg.page_list = frmr->fr_pgl;
160 fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
161 fastreg_wr.wr.fast_reg.page_list_len = page_no;
162 fastreg_wr.wr.fast_reg.length = len;
163 fastreg_wr.wr.fast_reg.access_flags = writing ?
164 IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
165 IB_ACCESS_REMOTE_READ;
166 key = (u8)(mr->rkey & 0x000000FF);
167 ib_update_fast_reg_key(mr, ++key);
168 fastreg_wr.wr.fast_reg.rkey = mr->rkey;
170 DECR_CQCOUNT(&r_xprt->rx_ep);
171 rc = ib_post_send(ia->ri_id->qp, &fastreg_wr, &bad_wr);
175 seg1->mr_rkey = mr->rkey;
176 seg1->mr_base = seg1->mr_dma + pageoff;
182 dprintk("RPC: %s: ib_post_send status %i\n", __func__, rc);
183 ib_update_fast_reg_key(mr, --key);
184 frmr->fr_state = FRMR_IS_INVALID;
186 rpcrdma_unmap_one(ia, --seg);
190 /* Post a LOCAL_INV Work Request to prevent further remote access
191 * via RDMA READ or RDMA WRITE.
194 frwr_op_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg)
196 struct rpcrdma_mr_seg *seg1 = seg;
197 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
198 struct ib_send_wr invalidate_wr, *bad_wr;
199 int rc, nsegs = seg->mr_nsegs;
201 seg1->rl_mw->r.frmr.fr_state = FRMR_IS_INVALID;
203 memset(&invalidate_wr, 0, sizeof(invalidate_wr));
204 invalidate_wr.wr_id = (unsigned long)(void *)seg1->rl_mw;
205 invalidate_wr.opcode = IB_WR_LOCAL_INV;
206 invalidate_wr.ex.invalidate_rkey = seg1->rl_mw->r.frmr.fr_mr->rkey;
207 DECR_CQCOUNT(&r_xprt->rx_ep);
209 read_lock(&ia->ri_qplock);
210 while (seg1->mr_nsegs--)
211 rpcrdma_unmap_one(ia, seg++);
212 rc = ib_post_send(ia->ri_id->qp, &invalidate_wr, &bad_wr);
213 read_unlock(&ia->ri_qplock);
219 /* Force rpcrdma_buffer_get() to retry */
220 seg1->rl_mw->r.frmr.fr_state = FRMR_IS_STALE;
221 dprintk("RPC: %s: ib_post_send status %i\n", __func__, rc);
225 /* After a disconnect, a flushed FAST_REG_MR can leave an FRMR in
226 * an unusable state. Find FRMRs in this state and dereg / reg
227 * each. FRMRs that are VALID and attached to an rpcrdma_req are
230 * This gives all in-use FRMRs a fresh rkey and leaves them INVALID.
232 * This is invoked only in the transport connect worker in order
233 * to serialize with rpcrdma_register_frmr_external().
236 frwr_op_reset(struct rpcrdma_xprt *r_xprt)
238 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
239 struct ib_device *device = r_xprt->rx_ia.ri_id->device;
240 unsigned int depth = r_xprt->rx_ia.ri_max_frmr_depth;
241 struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
242 struct rpcrdma_mw *r;
245 list_for_each_entry(r, &buf->rb_all, mw_all) {
246 if (r->r.frmr.fr_state == FRMR_IS_INVALID)
250 rc = __frwr_init(r, pd, device, depth);
252 dprintk("RPC: %s: mw %p left %s\n",
254 (r->r.frmr.fr_state == FRMR_IS_STALE ?
259 r->r.frmr.fr_state = FRMR_IS_INVALID;
263 const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops = {
264 .ro_map = frwr_op_map,
265 .ro_unmap = frwr_op_unmap,
266 .ro_maxpages = frwr_op_maxpages,
267 .ro_init = frwr_op_init,
268 .ro_reset = frwr_op_reset,
269 .ro_displayname = "frwr",