aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/infiniband
diff options
context:
space:
mode:
authorHal Rosenstock <halr@voltaire.com>2005-07-27 14:45:42 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2005-07-27 19:26:14 -0400
commitcb183a06b381652b7637fedfa7ef85ec0baf2a1f (patch)
tree8b559d3fd574af62dfbe39751980d5cb71e32598 /drivers/infiniband
parent3f75daddb4fc6b695faa4e12e76894389e913dcb (diff)
[PATCH] IB: Implementation for RMPP support in user MAD
Implementation for RMPP support in user MAD Signed-off-by: Hal Rosenstock <halr@voltaire.com> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'drivers/infiniband')
-rw-r--r--drivers/infiniband/core/user_mad.c300
1 files changed, 194 insertions, 106 deletions
diff --git a/drivers/infiniband/core/user_mad.c b/drivers/infiniband/core/user_mad.c
index 088bb1f0f514..2e38792df533 100644
--- a/drivers/infiniband/core/user_mad.c
+++ b/drivers/infiniband/core/user_mad.c
@@ -1,5 +1,7 @@
1/* 1/*
2 * Copyright (c) 2004 Topspin Communications. All rights reserved. 2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
4 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
3 * 5 *
4 * This software is available to you under a choice of one of two 6 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU 7 * licenses. You may choose to be licensed under the terms of the GNU
@@ -29,7 +31,7 @@
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE. 32 * SOFTWARE.
31 * 33 *
32 * $Id: user_mad.c 1389 2004-12-27 22:56:47Z roland $ 34 * $Id: user_mad.c 2814 2005-07-06 19:14:09Z halr $
33 */ 35 */
34 36
35#include <linux/module.h> 37#include <linux/module.h>
@@ -94,10 +96,12 @@ struct ib_umad_file {
94}; 96};
95 97
96struct ib_umad_packet { 98struct ib_umad_packet {
97 struct ib_user_mad mad;
98 struct ib_ah *ah; 99 struct ib_ah *ah;
100 struct ib_mad_send_buf *msg;
99 struct list_head list; 101 struct list_head list;
102 int length;
100 DECLARE_PCI_UNMAP_ADDR(mapping) 103 DECLARE_PCI_UNMAP_ADDR(mapping)
104 struct ib_user_mad mad;
101}; 105};
102 106
103static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE); 107static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
@@ -114,10 +118,10 @@ static int queue_packet(struct ib_umad_file *file,
114 int ret = 1; 118 int ret = 1;
115 119
116 down_read(&file->agent_mutex); 120 down_read(&file->agent_mutex);
117 for (packet->mad.id = 0; 121 for (packet->mad.hdr.id = 0;
118 packet->mad.id < IB_UMAD_MAX_AGENTS; 122 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
119 packet->mad.id++) 123 packet->mad.hdr.id++)
120 if (agent == file->agent[packet->mad.id]) { 124 if (agent == file->agent[packet->mad.hdr.id]) {
121 spin_lock_irq(&file->recv_lock); 125 spin_lock_irq(&file->recv_lock);
122 list_add_tail(&packet->list, &file->recv_list); 126 list_add_tail(&packet->list, &file->recv_list);
123 spin_unlock_irq(&file->recv_lock); 127 spin_unlock_irq(&file->recv_lock);
@@ -135,22 +139,30 @@ static void send_handler(struct ib_mad_agent *agent,
135 struct ib_mad_send_wc *send_wc) 139 struct ib_mad_send_wc *send_wc)
136{ 140{
137 struct ib_umad_file *file = agent->context; 141 struct ib_umad_file *file = agent->context;
138 struct ib_umad_packet *packet = 142 struct ib_umad_packet *timeout, *packet =
139 (void *) (unsigned long) send_wc->wr_id; 143 (void *) (unsigned long) send_wc->wr_id;
140 144
141 dma_unmap_single(agent->device->dma_device, 145 ib_destroy_ah(packet->msg->send_wr.wr.ud.ah);
142 pci_unmap_addr(packet, mapping), 146 ib_free_send_mad(packet->msg);
143 sizeof packet->mad.data,
144 DMA_TO_DEVICE);
145 ib_destroy_ah(packet->ah);
146 147
147 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) { 148 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
148 packet->mad.status = ETIMEDOUT; 149 timeout = kmalloc(sizeof *timeout + sizeof (struct ib_mad_hdr),
150 GFP_KERNEL);
151 if (!timeout)
152 goto out;
149 153
150 if (!queue_packet(file, agent, packet)) 154 memset(timeout, 0, sizeof *timeout + sizeof (struct ib_mad_hdr));
151 return;
152 }
153 155
156 timeout->length = sizeof (struct ib_mad_hdr);
157 timeout->mad.hdr.id = packet->mad.hdr.id;
158 timeout->mad.hdr.status = ETIMEDOUT;
159 memcpy(timeout->mad.data, packet->mad.data,
160 sizeof (struct ib_mad_hdr));
161
162 if (!queue_packet(file, agent, timeout))
163 return;
164 }
165out:
154 kfree(packet); 166 kfree(packet);
155} 167}
156 168
@@ -159,30 +171,35 @@ static void recv_handler(struct ib_mad_agent *agent,
159{ 171{
160 struct ib_umad_file *file = agent->context; 172 struct ib_umad_file *file = agent->context;
161 struct ib_umad_packet *packet; 173 struct ib_umad_packet *packet;
174 int length;
162 175
163 if (mad_recv_wc->wc->status != IB_WC_SUCCESS) 176 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
164 goto out; 177 goto out;
165 178
166 packet = kmalloc(sizeof *packet, GFP_KERNEL); 179 length = mad_recv_wc->mad_len;
180 packet = kmalloc(sizeof *packet + length, GFP_KERNEL);
167 if (!packet) 181 if (!packet)
168 goto out; 182 goto out;
169 183
170 memset(packet, 0, sizeof *packet); 184 memset(packet, 0, sizeof *packet + length);
185 packet->length = length;
186
187 ib_coalesce_recv_mad(mad_recv_wc, packet->mad.data);
171 188
172 memcpy(packet->mad.data, mad_recv_wc->recv_buf.mad, sizeof packet->mad.data); 189 packet->mad.hdr.status = 0;
173 packet->mad.status = 0; 190 packet->mad.hdr.length = length + sizeof (struct ib_user_mad);
174 packet->mad.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp); 191 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
175 packet->mad.lid = cpu_to_be16(mad_recv_wc->wc->slid); 192 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
176 packet->mad.sl = mad_recv_wc->wc->sl; 193 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
177 packet->mad.path_bits = mad_recv_wc->wc->dlid_path_bits; 194 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
178 packet->mad.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH); 195 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
179 if (packet->mad.grh_present) { 196 if (packet->mad.hdr.grh_present) {
180 /* XXX parse GRH */ 197 /* XXX parse GRH */
181 packet->mad.gid_index = 0; 198 packet->mad.hdr.gid_index = 0;
182 packet->mad.hop_limit = 0; 199 packet->mad.hdr.hop_limit = 0;
183 packet->mad.traffic_class = 0; 200 packet->mad.hdr.traffic_class = 0;
184 memset(packet->mad.gid, 0, 16); 201 memset(packet->mad.hdr.gid, 0, 16);
185 packet->mad.flow_label = 0; 202 packet->mad.hdr.flow_label = 0;
186 } 203 }
187 204
188 if (queue_packet(file, agent, packet)) 205 if (queue_packet(file, agent, packet))
@@ -199,7 +216,7 @@ static ssize_t ib_umad_read(struct file *filp, char __user *buf,
199 struct ib_umad_packet *packet; 216 struct ib_umad_packet *packet;
200 ssize_t ret; 217 ssize_t ret;
201 218
202 if (count < sizeof (struct ib_user_mad)) 219 if (count < sizeof (struct ib_user_mad) + sizeof (struct ib_mad))
203 return -EINVAL; 220 return -EINVAL;
204 221
205 spin_lock_irq(&file->recv_lock); 222 spin_lock_irq(&file->recv_lock);
@@ -222,12 +239,25 @@ static ssize_t ib_umad_read(struct file *filp, char __user *buf,
222 239
223 spin_unlock_irq(&file->recv_lock); 240 spin_unlock_irq(&file->recv_lock);
224 241
225 if (copy_to_user(buf, &packet->mad, sizeof packet->mad)) 242 if (count < packet->length + sizeof (struct ib_user_mad)) {
243 /* Return length needed (and first RMPP segment) if too small */
244 if (copy_to_user(buf, &packet->mad,
245 sizeof (struct ib_user_mad) + sizeof (struct ib_mad)))
246 ret = -EFAULT;
247 else
248 ret = -ENOSPC;
249 } else if (copy_to_user(buf, &packet->mad,
250 packet->length + sizeof (struct ib_user_mad)))
226 ret = -EFAULT; 251 ret = -EFAULT;
227 else 252 else
228 ret = sizeof packet->mad; 253 ret = packet->length + sizeof (struct ib_user_mad);
229 254 if (ret < 0) {
230 kfree(packet); 255 /* Requeue packet */
256 spin_lock_irq(&file->recv_lock);
257 list_add(&packet->list, &file->recv_list);
258 spin_unlock_irq(&file->recv_lock);
259 } else
260 kfree(packet);
231 return ret; 261 return ret;
232} 262}
233 263
@@ -238,69 +268,57 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
238 struct ib_umad_packet *packet; 268 struct ib_umad_packet *packet;
239 struct ib_mad_agent *agent; 269 struct ib_mad_agent *agent;
240 struct ib_ah_attr ah_attr; 270 struct ib_ah_attr ah_attr;
241 struct ib_sge gather_list; 271 struct ib_send_wr *bad_wr;
242 struct ib_send_wr *bad_wr, wr = { 272 struct ib_rmpp_mad *rmpp_mad;
243 .opcode = IB_WR_SEND,
244 .sg_list = &gather_list,
245 .num_sge = 1,
246 .send_flags = IB_SEND_SIGNALED,
247 };
248 u8 method; 273 u8 method;
249 u64 *tid; 274 u64 *tid;
250 int ret; 275 int ret, length, hdr_len, data_len, rmpp_hdr_size;
276 int rmpp_active = 0;
251 277
252 if (count < sizeof (struct ib_user_mad)) 278 if (count < sizeof (struct ib_user_mad))
253 return -EINVAL; 279 return -EINVAL;
254 280
255 packet = kmalloc(sizeof *packet, GFP_KERNEL); 281 length = count - sizeof (struct ib_user_mad);
282 packet = kmalloc(sizeof *packet + sizeof(struct ib_mad_hdr) +
283 sizeof(struct ib_rmpp_hdr), GFP_KERNEL);
256 if (!packet) 284 if (!packet)
257 return -ENOMEM; 285 return -ENOMEM;
258 286
259 if (copy_from_user(&packet->mad, buf, sizeof packet->mad)) { 287 if (copy_from_user(&packet->mad, buf,
260 kfree(packet); 288 sizeof (struct ib_user_mad) +
261 return -EFAULT; 289 sizeof(struct ib_mad_hdr) +
290 sizeof(struct ib_rmpp_hdr))) {
291 ret = -EFAULT;
292 goto err;
262 } 293 }
263 294
264 if (packet->mad.id < 0 || packet->mad.id >= IB_UMAD_MAX_AGENTS) { 295 if (packet->mad.hdr.id < 0 ||
296 packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
265 ret = -EINVAL; 297 ret = -EINVAL;
266 goto err; 298 goto err;
267 } 299 }
268 300
301 packet->length = length;
302
269 down_read(&file->agent_mutex); 303 down_read(&file->agent_mutex);
270 304
271 agent = file->agent[packet->mad.id]; 305 agent = file->agent[packet->mad.hdr.id];
272 if (!agent) { 306 if (!agent) {
273 ret = -EINVAL; 307 ret = -EINVAL;
274 goto err_up; 308 goto err_up;
275 } 309 }
276 310
277 /*
278 * If userspace is generating a request that will generate a
279 * response, we need to make sure the high-order part of the
280 * transaction ID matches the agent being used to send the
281 * MAD.
282 */
283 method = ((struct ib_mad_hdr *) packet->mad.data)->method;
284
285 if (!(method & IB_MGMT_METHOD_RESP) &&
286 method != IB_MGMT_METHOD_TRAP_REPRESS &&
287 method != IB_MGMT_METHOD_SEND) {
288 tid = &((struct ib_mad_hdr *) packet->mad.data)->tid;
289 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
290 (be64_to_cpup(tid) & 0xffffffff));
291 }
292
293 memset(&ah_attr, 0, sizeof ah_attr); 311 memset(&ah_attr, 0, sizeof ah_attr);
294 ah_attr.dlid = be16_to_cpu(packet->mad.lid); 312 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
295 ah_attr.sl = packet->mad.sl; 313 ah_attr.sl = packet->mad.hdr.sl;
296 ah_attr.src_path_bits = packet->mad.path_bits; 314 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
297 ah_attr.port_num = file->port->port_num; 315 ah_attr.port_num = file->port->port_num;
298 if (packet->mad.grh_present) { 316 if (packet->mad.hdr.grh_present) {
299 ah_attr.ah_flags = IB_AH_GRH; 317 ah_attr.ah_flags = IB_AH_GRH;
300 memcpy(ah_attr.grh.dgid.raw, packet->mad.gid, 16); 318 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
301 ah_attr.grh.flow_label = packet->mad.flow_label; 319 ah_attr.grh.flow_label = packet->mad.hdr.flow_label;
302 ah_attr.grh.hop_limit = packet->mad.hop_limit; 320 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
303 ah_attr.grh.traffic_class = packet->mad.traffic_class; 321 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
304 } 322 }
305 323
306 packet->ah = ib_create_ah(agent->qp->pd, &ah_attr); 324 packet->ah = ib_create_ah(agent->qp->pd, &ah_attr);
@@ -309,35 +327,104 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
309 goto err_up; 327 goto err_up;
310 } 328 }
311 329
312 gather_list.addr = dma_map_single(agent->device->dma_device, 330 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
313 packet->mad.data, 331 if (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE) {
314 sizeof packet->mad.data, 332 /* RMPP active */
315 DMA_TO_DEVICE); 333 if (!agent->rmpp_version) {
316 gather_list.length = sizeof packet->mad.data; 334 ret = -EINVAL;
317 gather_list.lkey = file->mr[packet->mad.id]->lkey; 335 goto err_ah;
318 pci_unmap_addr_set(packet, mapping, gather_list.addr); 336 }
337 /* Validate that management class can support RMPP */
338 if (rmpp_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_ADM) {
339 hdr_len = offsetof(struct ib_sa_mad, data);
340 data_len = length;
341 } else if ((rmpp_mad->mad_hdr.mgmt_class >= IB_MGMT_CLASS_VENDOR_RANGE2_START) &&
342 (rmpp_mad->mad_hdr.mgmt_class <= IB_MGMT_CLASS_VENDOR_RANGE2_END)) {
343 hdr_len = offsetof(struct ib_vendor_mad, data);
344 data_len = length - hdr_len;
345 } else {
346 ret = -EINVAL;
347 goto err_ah;
348 }
349 rmpp_active = 1;
350 } else {
351 if (length > sizeof(struct ib_mad)) {
352 ret = -EINVAL;
353 goto err_ah;
354 }
355 hdr_len = offsetof(struct ib_mad, data);
356 data_len = length - hdr_len;
357 }
358
359 packet->msg = ib_create_send_mad(agent,
360 be32_to_cpu(packet->mad.hdr.qpn),
361 0, packet->ah, rmpp_active,
362 hdr_len, data_len,
363 GFP_KERNEL);
364 if (IS_ERR(packet->msg)) {
365 ret = PTR_ERR(packet->msg);
366 goto err_ah;
367 }
319 368
320 wr.wr.ud.mad_hdr = (struct ib_mad_hdr *) packet->mad.data; 369 packet->msg->send_wr.wr.ud.timeout_ms = packet->mad.hdr.timeout_ms;
321 wr.wr.ud.ah = packet->ah; 370 packet->msg->send_wr.wr.ud.retries = packet->mad.hdr.retries;
322 wr.wr.ud.remote_qpn = be32_to_cpu(packet->mad.qpn);
323 wr.wr.ud.remote_qkey = be32_to_cpu(packet->mad.qkey);
324 wr.wr.ud.timeout_ms = packet->mad.timeout_ms;
325 wr.wr.ud.retries = 0;
326 371
327 wr.wr_id = (unsigned long) packet; 372 /* Override send WR WRID initialized in ib_create_send_mad */
373 packet->msg->send_wr.wr_id = (unsigned long) packet;
328 374
329 ret = ib_post_send_mad(agent, &wr, &bad_wr); 375 if (!rmpp_active) {
330 if (ret) { 376 /* Copy message from user into send buffer */
331 dma_unmap_single(agent->device->dma_device, 377 if (copy_from_user(packet->msg->mad,
332 pci_unmap_addr(packet, mapping), 378 buf + sizeof(struct ib_user_mad), length)) {
333 sizeof packet->mad.data, 379 ret = -EFAULT;
334 DMA_TO_DEVICE); 380 goto err_msg;
335 goto err_up; 381 }
382 } else {
383 rmpp_hdr_size = sizeof(struct ib_mad_hdr) +
384 sizeof(struct ib_rmpp_hdr);
385
386 /* Only copy MAD headers (RMPP header in place) */
387 memcpy(packet->msg->mad, packet->mad.data,
388 sizeof(struct ib_mad_hdr));
389
390 /* Now, copy rest of message from user into send buffer */
391 if (copy_from_user(((struct ib_rmpp_mad *) packet->msg->mad)->data,
392 buf + sizeof (struct ib_user_mad) + rmpp_hdr_size,
393 length - rmpp_hdr_size)) {
394 ret = -EFAULT;
395 goto err_msg;
396 }
397 }
398
399 /*
400 * If userspace is generating a request that will generate a
401 * response, we need to make sure the high-order part of the
402 * transaction ID matches the agent being used to send the
403 * MAD.
404 */
405 method = packet->msg->mad->mad_hdr.method;
406
407 if (!(method & IB_MGMT_METHOD_RESP) &&
408 method != IB_MGMT_METHOD_TRAP_REPRESS &&
409 method != IB_MGMT_METHOD_SEND) {
410 tid = &packet->msg->mad->mad_hdr.tid;
411 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
412 (be64_to_cpup(tid) & 0xffffffff));
336 } 413 }
337 414
415 ret = ib_post_send_mad(agent, &packet->msg->send_wr, &bad_wr);
416 if (ret)
417 goto err_msg;
418
338 up_read(&file->agent_mutex); 419 up_read(&file->agent_mutex);
339 420
340 return sizeof packet->mad; 421 return sizeof (struct ib_user_mad_hdr) + packet->length;
422
423err_msg:
424 ib_free_send_mad(packet->msg);
425
426err_ah:
427 ib_destroy_ah(packet->ah);
341 428
342err_up: 429err_up:
343 up_read(&file->agent_mutex); 430 up_read(&file->agent_mutex);
@@ -400,7 +487,8 @@ found:
400 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num, 487 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
401 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI, 488 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
402 ureq.mgmt_class ? &req : NULL, 489 ureq.mgmt_class ? &req : NULL,
403 0, send_handler, recv_handler, file); 490 ureq.rmpp_version,
491 send_handler, recv_handler, file);
404 if (IS_ERR(agent)) { 492 if (IS_ERR(agent)) {
405 ret = PTR_ERR(agent); 493 ret = PTR_ERR(agent);
406 goto out; 494 goto out;
@@ -461,8 +549,8 @@ out:
461 return ret; 549 return ret;
462} 550}
463 551
464static long ib_umad_ioctl(struct file *filp, 552static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
465 unsigned int cmd, unsigned long arg) 553 unsigned long arg)
466{ 554{
467 switch (cmd) { 555 switch (cmd) {
468 case IB_USER_MAD_REGISTER_AGENT: 556 case IB_USER_MAD_REGISTER_AGENT:
@@ -518,14 +606,14 @@ static int ib_umad_close(struct inode *inode, struct file *filp)
518} 606}
519 607
520static struct file_operations umad_fops = { 608static struct file_operations umad_fops = {
521 .owner = THIS_MODULE, 609 .owner = THIS_MODULE,
522 .read = ib_umad_read, 610 .read = ib_umad_read,
523 .write = ib_umad_write, 611 .write = ib_umad_write,
524 .poll = ib_umad_poll, 612 .poll = ib_umad_poll,
525 .unlocked_ioctl = ib_umad_ioctl, 613 .unlocked_ioctl = ib_umad_ioctl,
526 .compat_ioctl = ib_umad_ioctl, 614 .compat_ioctl = ib_umad_ioctl,
527 .open = ib_umad_open, 615 .open = ib_umad_open,
528 .release = ib_umad_close 616 .release = ib_umad_close
529}; 617};
530 618
531static int ib_umad_sm_open(struct inode *inode, struct file *filp) 619static int ib_umad_sm_open(struct inode *inode, struct file *filp)