This source file includes following definitions.
- mca_btl_ugni_smsg_mbox_construct
- mca_btl_ugni_smsg_init
- mca_btl_ugni_smsg_process
- mca_btl_ugni_handle_remote_smsg_overrun
- mca_btl_ugni_progress_remote_smsg
1
2
3
4
5
6
7
8
9
10
11
12
13 #include "btl_ugni_smsg.h"
14 #include "btl_ugni_rdma.h"
15
16 static void mca_btl_ugni_smsg_mbox_construct (mca_btl_ugni_smsg_mbox_t *mbox) {
17 struct mca_btl_ugni_reg_t *ugni_reg =
18 (struct mca_btl_ugni_reg_t *) mbox->super.registration;
19 mca_rcache_base_registration_t *base_reg =
20 (mca_rcache_base_registration_t *) ugni_reg;
21
22
23 mbox->attr.smsg_attr.msg_type = GNI_SMSG_TYPE_MBOX_AUTO_RETRANSMIT;
24 mbox->attr.smsg_attr.msg_maxsize = mca_btl_ugni_component.ugni_smsg_limit;
25 mbox->attr.smsg_attr.mbox_maxcredit = mca_btl_ugni_component.smsg_max_credits;
26 mbox->attr.smsg_attr.mbox_offset = (uintptr_t) mbox->super.ptr - (uintptr_t) base_reg->base;
27 mbox->attr.smsg_attr.msg_buffer = base_reg->base;
28 mbox->attr.smsg_attr.buff_size = mca_btl_ugni_component.smsg_mbox_size;
29 mbox->attr.smsg_attr.mem_hndl = ugni_reg->handle.gni_handle;
30 mbox->attr.proc_name = OPAL_PROC_MY_NAME;
31 mbox->attr.rmt_irq_mem_hndl = mca_btl_ugni_component.modules[0].devices[0].smsg_irq_mhndl;
32 }
33
34 OBJ_CLASS_INSTANCE(mca_btl_ugni_smsg_mbox_t, opal_free_list_item_t,
35 mca_btl_ugni_smsg_mbox_construct, NULL);
36
37
38 int mca_btl_ugni_smsg_init (mca_btl_ugni_module_t *ugni_module)
39 {
40 gni_return_t rc;
41
42 for (int i = 0 ; i < mca_btl_ugni_component.virtual_device_count ; ++i) {
43 rc = GNI_SmsgSetMaxRetrans (ugni_module->devices[i].dev_handle,
44 mca_btl_ugni_component.smsg_max_retries);
45 if (GNI_RC_SUCCESS != rc) {
46 BTL_ERROR(("error setting maximum SMSG retries %s",gni_err_str[rc]));
47 return mca_btl_rc_ugni_to_opal (rc);
48 }
49 }
50
51 return OPAL_SUCCESS;
52 }
53
54
55 int mca_btl_ugni_smsg_process (mca_btl_base_endpoint_t *ep)
56 {
57 mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
58 mca_btl_active_message_callback_t *reg;
59 mca_btl_ugni_base_frag_t frag;
60 mca_btl_base_segment_t seg;
61 bool disconnect = false;
62 int32_t _tmp_value = 0;
63 uintptr_t data_ptr;
64 gni_return_t rc;
65 uint32_t len;
66 int count = 0;
67
68 if (!opal_atomic_compare_exchange_strong_32 (&ep->smsg_progressing, &_tmp_value, 1)) {
69
70 return 0;
71 }
72
73
74 do {
75 uint8_t tag = GNI_SMSG_ANY_TAG;
76
77 rc = mca_btl_ugni_smsg_get_next_wtag (&ep->smsg_ep_handle, &data_ptr, &tag);
78 if (GNI_RC_SUCCESS != rc) {
79 if (OPAL_LIKELY(GNI_RC_NOT_DONE == rc)) {
80 BTL_VERBOSE(("no smsg message waiting. rc = %s", gni_err_str[rc]));
81
82 ep->smsg_progressing = 0;
83 return count;
84 }
85
86 BTL_ERROR(("unhandled GNI_SmsgGetNextWTag error"));
87 return OPAL_ERROR;
88 }
89
90 assert (0 != data_ptr);
91
92 count++;
93
94 BTL_VERBOSE(("got smsg fragment. tag = %d\n", tag));
95
96 switch (tag) {
97 case MCA_BTL_UGNI_TAG_SEND:
98 frag.hdr.send = ((mca_btl_ugni_send_frag_hdr_t *) data_ptr)[0];
99
100 tag = frag.hdr.send.lag >> 24;
101 len = frag.hdr.send.lag & 0x00ffffff;
102
103 BTL_VERBOSE(("received smsg fragment. hdr = {len = %u, tag = %d}", len, tag));
104
105 reg = mca_btl_base_active_message_trigger + tag;
106 frag.base.des_segments = &seg;
107 frag.base.des_segment_count = 1;
108
109 seg.seg_addr.pval = (void *)((uintptr_t)data_ptr + sizeof (mca_btl_ugni_send_frag_hdr_t));
110 seg.seg_len = len;
111
112 assert (NULL != reg->cbfunc);
113
114 reg->cbfunc(&ugni_module->super, tag, &(frag.base), reg->cbdata);
115
116 break;
117 case MCA_BTL_UGNI_TAG_GET_INIT:
118 frag.hdr.eager_ex = ((mca_btl_ugni_eager_ex_frag_hdr_t *) data_ptr)[0];
119
120 mca_btl_ugni_start_eager_get (ep, frag.hdr.eager_ex, NULL);
121 break;
122 case MCA_BTL_UGNI_TAG_RDMA_COMPLETE:
123 frag.hdr.rdma = ((mca_btl_ugni_rdma_frag_hdr_t *) data_ptr)[0];
124
125 if (((mca_btl_ugni_base_frag_t *)frag.hdr.rdma.ctx)->flags & MCA_BTL_UGNI_FRAG_SMSG_COMPLETE) {
126 mca_btl_ugni_frag_complete (frag.hdr.rdma.ctx, OPAL_SUCCESS);
127 } else {
128
129 ((mca_btl_ugni_base_frag_t *)frag.hdr.rdma.ctx)->flags &= ~MCA_BTL_UGNI_FRAG_IGNORE;
130 }
131 break;
132 case MCA_BTL_UGNI_TAG_DISCONNECT:
133
134 disconnect = true;
135 break;
136 default:
137 BTL_ERROR(("unknown tag %d\n", tag));
138 break;
139 }
140
141 rc = mca_btl_ugni_smsg_release (&ep->smsg_ep_handle);
142 if (OPAL_UNLIKELY(GNI_RC_SUCCESS != rc)) {
143 BTL_ERROR(("Smsg release failed! rc = %d", rc));
144 return OPAL_ERROR;
145 }
146 } while (!disconnect);
147
148 ep->smsg_progressing = 0;
149
150
151 opal_mutex_lock (&ep->lock);
152
153 mca_btl_ugni_ep_disconnect (ep, false);
154
155 opal_mutex_unlock (&ep->lock);
156
157 return count;
158 }
159
160 static inline int
161 mca_btl_ugni_handle_remote_smsg_overrun (mca_btl_ugni_module_t *btl)
162 {
163 size_t endpoint_count;
164 unsigned int ep_index;
165 int count, rc;
166
167 BTL_VERBOSE(("btl/ugni_component detected SMSG CQ overrun. "
168 "processing message backlog..."));
169
170
171
172
173
174 mca_btl_ugni_cq_clear (btl->devices, btl->smsg_remote_cq);
175
176 endpoint_count = opal_pointer_array_get_size (&btl->endpoints);
177
178 for (ep_index = 0, count = 0 ; ep_index < endpoint_count ; ++ep_index) {
179 mca_btl_base_endpoint_t *ep;
180
181 ep = (mca_btl_base_endpoint_t *) opal_pointer_array_get_item (&btl->endpoints,
182 ep_index);
183
184 if (NULL == ep || MCA_BTL_UGNI_EP_STATE_CONNECTED != ep->state) {
185 continue;
186 }
187
188
189 rc = mca_btl_ugni_smsg_process (ep);
190 if (OPAL_LIKELY(rc >= 0)) {
191 count += rc;
192 }
193 }
194
195 return count;
196 }
197
198 int mca_btl_ugni_progress_remote_smsg (mca_btl_ugni_module_t *btl)
199 {
200 mca_btl_base_endpoint_t *ep;
201 gni_cq_entry_t event_data;
202 gni_return_t grc;
203 uint64_t inst_id;
204
205 grc = mca_btl_ugni_gni_cq_get_event (btl->devices, btl->smsg_remote_cq, &event_data);
206 if (GNI_RC_NOT_DONE == grc) {
207 return 0;
208 }
209
210 if (OPAL_UNLIKELY(GNI_RC_SUCCESS != grc || !GNI_CQ_STATUS_OK(event_data) ||
211 GNI_CQ_OVERRUN(event_data))) {
212 if (GNI_RC_ERROR_RESOURCE == grc ||
213 (GNI_RC_SUCCESS == grc && GNI_CQ_OVERRUN(event_data))) {
214
215 return mca_btl_ugni_handle_remote_smsg_overrun (btl);
216 }
217
218 BTL_ERROR(("unhandled error in GNI_CqGetEvent"));
219
220
221 assert (0);
222 return mca_btl_rc_ugni_to_opal (grc);
223 }
224
225 BTL_VERBOSE(("REMOTE CQ: Got event 0x%" PRIx64 ". msg id = %" PRIu64
226 ". ok = %d, type = %" PRIu64, (uint64_t) event_data,
227 GNI_CQ_GET_INST_ID(event_data), GNI_CQ_STATUS_OK(event_data),
228 GNI_CQ_GET_TYPE(event_data)));
229
230 inst_id = GNI_CQ_GET_INST_ID(event_data);
231
232 ep = (mca_btl_base_endpoint_t *) opal_pointer_array_get_item (&btl->endpoints, inst_id);
233
234 if (OPAL_UNLIKELY(MCA_BTL_UGNI_EP_STATE_CONNECTED != ep->state)) {
235
236
237 BTL_VERBOSE(("event occurred on an unconnected endpoint! ep state = %d", ep->state));
238 return 0;
239 }
240
241 return mca_btl_ugni_smsg_process (ep);
242 }
243