2 * Copyright (c) 2006-2008 Chelsio, Inc. All rights reserved.
4 * 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
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/list.h>
34 #include <linux/slab.h>
35 #include <net/neighbour.h>
36 #include <linux/notifier.h>
37 #include <linux/atomic.h>
38 #include <linux/proc_fs.h>
39 #include <linux/if_vlan.h>
40 #include <net/netevent.h>
41 #include <linux/highmem.h>
42 #include <linux/vmalloc.h>
43 #include <linux/export.h>
47 #include "cxgb3_ioctl.h"
48 #include "cxgb3_ctl_defs.h"
49 #include "cxgb3_defs.h"
51 #include "firmware_exports.h"
52 #include "cxgb3_offload.h"
54 static LIST_HEAD(client_list);
55 static LIST_HEAD(ofld_dev_list);
56 static DEFINE_MUTEX(cxgb3_db_lock);
58 static DEFINE_RWLOCK(adapter_list_lock);
59 static LIST_HEAD(adapter_list);
61 static const unsigned int MAX_ATIDS = 64 * 1024;
62 static const unsigned int ATID_BASE = 0x10000;
64 static void cxgb_neigh_update(struct neighbour *neigh);
65 static void cxgb_redirect(struct dst_entry *old, struct neighbour *old_neigh,
66 struct dst_entry *new, struct neighbour *new_neigh,
69 static inline int offload_activated(struct t3cdev *tdev)
71 const struct adapter *adapter = tdev2adap(tdev);
73 return test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
77 * cxgb3_register_client - register an offload client
80 * Add the client to the client list,
81 * and call backs the client for each activated offload device
83 void cxgb3_register_client(struct cxgb3_client *client)
87 mutex_lock(&cxgb3_db_lock);
88 list_add_tail(&client->client_list, &client_list);
91 list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
92 if (offload_activated(tdev))
96 mutex_unlock(&cxgb3_db_lock);
99 EXPORT_SYMBOL(cxgb3_register_client);
102 * cxgb3_unregister_client - unregister an offload client
103 * @client: the client
105 * Remove the client to the client list,
106 * and call backs the client for each activated offload device.
108 void cxgb3_unregister_client(struct cxgb3_client *client)
112 mutex_lock(&cxgb3_db_lock);
113 list_del(&client->client_list);
115 if (client->remove) {
116 list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
117 if (offload_activated(tdev))
118 client->remove(tdev);
121 mutex_unlock(&cxgb3_db_lock);
124 EXPORT_SYMBOL(cxgb3_unregister_client);
127 * cxgb3_add_clients - activate registered clients for an offload device
128 * @tdev: the offload device
130 * Call backs all registered clients once a offload device is activated
132 void cxgb3_add_clients(struct t3cdev *tdev)
134 struct cxgb3_client *client;
136 mutex_lock(&cxgb3_db_lock);
137 list_for_each_entry(client, &client_list, client_list) {
141 mutex_unlock(&cxgb3_db_lock);
145 * cxgb3_remove_clients - deactivates registered clients
146 * for an offload device
147 * @tdev: the offload device
149 * Call backs all registered clients once a offload device is deactivated
151 void cxgb3_remove_clients(struct t3cdev *tdev)
153 struct cxgb3_client *client;
155 mutex_lock(&cxgb3_db_lock);
156 list_for_each_entry(client, &client_list, client_list) {
158 client->remove(tdev);
160 mutex_unlock(&cxgb3_db_lock);
163 void cxgb3_event_notify(struct t3cdev *tdev, u32 event, u32 port)
165 struct cxgb3_client *client;
167 mutex_lock(&cxgb3_db_lock);
168 list_for_each_entry(client, &client_list, client_list) {
169 if (client->event_handler)
170 client->event_handler(tdev, event, port);
172 mutex_unlock(&cxgb3_db_lock);
175 static struct net_device *get_iff_from_mac(struct adapter *adapter,
176 const unsigned char *mac,
181 for_each_port(adapter, i) {
182 struct net_device *dev = adapter->port[i];
184 if (!memcmp(dev->dev_addr, mac, ETH_ALEN)) {
185 if (vlan && vlan != VLAN_VID_MASK) {
187 dev = __vlan_find_dev_deep(dev, vlan);
189 } else if (netif_is_bond_slave(dev)) {
199 static int cxgb_ulp_iscsi_ctl(struct adapter *adapter, unsigned int req,
204 unsigned int val = 0;
205 struct ulp_iscsi_info *uiip = data;
208 case ULP_ISCSI_GET_PARAMS:
209 uiip->pdev = adapter->pdev;
210 uiip->llimit = t3_read_reg(adapter, A_ULPRX_ISCSI_LLIMIT);
211 uiip->ulimit = t3_read_reg(adapter, A_ULPRX_ISCSI_ULIMIT);
212 uiip->tagmask = t3_read_reg(adapter, A_ULPRX_ISCSI_TAGMASK);
214 val = t3_read_reg(adapter, A_ULPRX_ISCSI_PSZ);
215 for (i = 0; i < 4; i++, val >>= 8)
216 uiip->pgsz_factor[i] = val & 0xFF;
218 val = t3_read_reg(adapter, A_TP_PARA_REG7);
220 uiip->max_rxsz = min((val >> S_PMMAXXFERLEN0)&M_PMMAXXFERLEN0,
221 (val >> S_PMMAXXFERLEN1)&M_PMMAXXFERLEN1);
223 * On tx, the iscsi pdu has to be <= tx page size and has to
224 * fit into the Tx PM FIFO.
226 val = min(adapter->params.tp.tx_pg_size,
227 t3_read_reg(adapter, A_PM1_TX_CFG) >> 17);
228 uiip->max_txsz = min(val, uiip->max_txsz);
230 /* set MaxRxData to 16224 */
231 val = t3_read_reg(adapter, A_TP_PARA_REG2);
232 if ((val >> S_MAXRXDATA) != 0x3f60) {
233 val &= (M_RXCOALESCESIZE << S_RXCOALESCESIZE);
234 val |= V_MAXRXDATA(0x3f60);
236 "%s, iscsi set MaxRxData to 16224 (0x%x).\n",
238 t3_write_reg(adapter, A_TP_PARA_REG2, val);
242 * on rx, the iscsi pdu has to be < rx page size and the
243 * the max rx data length programmed in TP
245 val = min(adapter->params.tp.rx_pg_size,
246 ((t3_read_reg(adapter, A_TP_PARA_REG2)) >>
247 S_MAXRXDATA) & M_MAXRXDATA);
248 uiip->max_rxsz = min(val, uiip->max_rxsz);
250 case ULP_ISCSI_SET_PARAMS:
251 t3_write_reg(adapter, A_ULPRX_ISCSI_TAGMASK, uiip->tagmask);
252 /* program the ddp page sizes */
253 for (i = 0; i < 4; i++)
254 val |= (uiip->pgsz_factor[i] & 0xF) << (8 * i);
255 if (val && (val != t3_read_reg(adapter, A_ULPRX_ISCSI_PSZ))) {
257 "%s, setting iscsi pgsz 0x%x, %u,%u,%u,%u.\n",
258 adapter->name, val, uiip->pgsz_factor[0],
259 uiip->pgsz_factor[1], uiip->pgsz_factor[2],
260 uiip->pgsz_factor[3]);
261 t3_write_reg(adapter, A_ULPRX_ISCSI_PSZ, val);
270 /* Response queue used for RDMA events. */
271 #define ASYNC_NOTIF_RSPQ 0
273 static int cxgb_rdma_ctl(struct adapter *adapter, unsigned int req, void *data)
278 case RDMA_GET_PARAMS: {
279 struct rdma_info *rdma = data;
280 struct pci_dev *pdev = adapter->pdev;
282 rdma->udbell_physbase = pci_resource_start(pdev, 2);
283 rdma->udbell_len = pci_resource_len(pdev, 2);
285 t3_read_reg(adapter, A_ULPTX_TPT_LLIMIT);
286 rdma->tpt_top = t3_read_reg(adapter, A_ULPTX_TPT_ULIMIT);
288 t3_read_reg(adapter, A_ULPTX_PBL_LLIMIT);
289 rdma->pbl_top = t3_read_reg(adapter, A_ULPTX_PBL_ULIMIT);
290 rdma->rqt_base = t3_read_reg(adapter, A_ULPRX_RQ_LLIMIT);
291 rdma->rqt_top = t3_read_reg(adapter, A_ULPRX_RQ_ULIMIT);
292 rdma->kdb_addr = adapter->regs + A_SG_KDOORBELL;
298 struct rdma_cq_op *rdma = data;
300 /* may be called in any context */
301 spin_lock_irqsave(&adapter->sge.reg_lock, flags);
302 ret = t3_sge_cqcntxt_op(adapter, rdma->id, rdma->op,
304 spin_unlock_irqrestore(&adapter->sge.reg_lock, flags);
308 struct ch_mem_range *t = data;
311 if ((t->addr & 7) || (t->len & 7))
313 if (t->mem_id == MEM_CM)
315 else if (t->mem_id == MEM_PMRX)
316 mem = &adapter->pmrx;
317 else if (t->mem_id == MEM_PMTX)
318 mem = &adapter->pmtx;
323 t3_mc7_bd_read(mem, t->addr / 8, t->len / 8,
330 struct rdma_cq_setup *rdma = data;
332 spin_lock_irq(&adapter->sge.reg_lock);
334 t3_sge_init_cqcntxt(adapter, rdma->id,
335 rdma->base_addr, rdma->size,
337 rdma->ovfl_mode, rdma->credits,
339 spin_unlock_irq(&adapter->sge.reg_lock);
342 case RDMA_CQ_DISABLE:
343 spin_lock_irq(&adapter->sge.reg_lock);
344 ret = t3_sge_disable_cqcntxt(adapter, *(unsigned int *)data);
345 spin_unlock_irq(&adapter->sge.reg_lock);
347 case RDMA_CTRL_QP_SETUP:{
348 struct rdma_ctrlqp_setup *rdma = data;
350 spin_lock_irq(&adapter->sge.reg_lock);
351 ret = t3_sge_init_ecntxt(adapter, FW_RI_SGEEC_START, 0,
354 rdma->base_addr, rdma->size,
355 FW_RI_TID_START, 1, 0);
356 spin_unlock_irq(&adapter->sge.reg_lock);
360 spin_lock(&adapter->stats_lock);
361 t3_tp_get_mib_stats(adapter, (struct tp_mib_stats *)data);
362 spin_unlock(&adapter->stats_lock);
371 static int cxgb_offload_ctl(struct t3cdev *tdev, unsigned int req, void *data)
373 struct adapter *adapter = tdev2adap(tdev);
374 struct tid_range *tid;
376 struct iff_mac *iffmacp;
377 struct ddp_params *ddpp;
378 struct adap_ports *ports;
379 struct ofld_page_info *rx_page_info;
380 struct tp_params *tp = &adapter->params.tp;
384 case GET_MAX_OUTSTANDING_WR:
385 *(unsigned int *)data = FW_WR_NUM;
388 *(unsigned int *)data = WR_FLITS;
390 case GET_TX_MAX_CHUNK:
391 *(unsigned int *)data = 1 << 20; /* 1MB */
395 tid->num = t3_mc5_size(&adapter->mc5) -
396 adapter->params.mc5.nroutes -
397 adapter->params.mc5.nfilters - adapter->params.mc5.nservers;
402 tid->num = adapter->params.mc5.nservers;
403 tid->base = t3_mc5_size(&adapter->mc5) - tid->num -
404 adapter->params.mc5.nfilters - adapter->params.mc5.nroutes;
406 case GET_L2T_CAPACITY:
407 *(unsigned int *)data = 2048;
412 mtup->mtus = adapter->params.mtus;
414 case GET_IFF_FROM_MAC:
416 iffmacp->dev = get_iff_from_mac(adapter, iffmacp->mac_addr,
422 ddpp->llimit = t3_read_reg(adapter, A_ULPRX_TDDP_LLIMIT);
423 ddpp->ulimit = t3_read_reg(adapter, A_ULPRX_TDDP_ULIMIT);
424 ddpp->tag_mask = t3_read_reg(adapter, A_ULPRX_TDDP_TAGMASK);
428 ports->nports = adapter->params.nports;
429 for_each_port(adapter, i)
430 ports->lldevs[i] = adapter->port[i];
432 case ULP_ISCSI_GET_PARAMS:
433 case ULP_ISCSI_SET_PARAMS:
434 if (!offload_running(adapter))
436 return cxgb_ulp_iscsi_ctl(adapter, req, data);
437 case RDMA_GET_PARAMS:
440 case RDMA_CQ_DISABLE:
441 case RDMA_CTRL_QP_SETUP:
444 if (!offload_running(adapter))
446 return cxgb_rdma_ctl(adapter, req, data);
447 case GET_RX_PAGE_INFO:
449 rx_page_info->page_size = tp->rx_pg_size;
450 rx_page_info->num = tp->rx_num_pgs;
452 case GET_ISCSI_IPV4ADDR: {
453 struct iscsi_ipv4addr *p = data;
454 struct port_info *pi = netdev_priv(p->dev);
455 p->ipv4addr = pi->iscsi_ipv4addr;
458 case GET_EMBEDDED_INFO: {
459 struct ch_embedded_info *e = data;
461 spin_lock(&adapter->stats_lock);
462 t3_get_fw_version(adapter, &e->fw_vers);
463 t3_get_tp_version(adapter, &e->tp_vers);
464 spin_unlock(&adapter->stats_lock);
474 * Dummy handler for Rx offload packets in case we get an offload packet before
475 * proper processing is setup. This complains and drops the packet as it isn't
476 * normal to get offload packets at this stage.
478 static int rx_offload_blackhole(struct t3cdev *dev, struct sk_buff **skbs,
482 dev_kfree_skb_any(skbs[n]);
486 static void dummy_neigh_update(struct t3cdev *dev, struct neighbour *neigh)
490 void cxgb3_set_dummy_ops(struct t3cdev *dev)
492 dev->recv = rx_offload_blackhole;
493 dev->neigh_update = dummy_neigh_update;
497 * Free an active-open TID.
499 void *cxgb3_free_atid(struct t3cdev *tdev, int atid)
501 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
502 union active_open_entry *p = atid2entry(t, atid);
503 void *ctx = p->t3c_tid.ctx;
505 spin_lock_bh(&t->atid_lock);
509 spin_unlock_bh(&t->atid_lock);
514 EXPORT_SYMBOL(cxgb3_free_atid);
517 * Free a server TID and return it to the free pool.
519 void cxgb3_free_stid(struct t3cdev *tdev, int stid)
521 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
522 union listen_entry *p = stid2entry(t, stid);
524 spin_lock_bh(&t->stid_lock);
528 spin_unlock_bh(&t->stid_lock);
531 EXPORT_SYMBOL(cxgb3_free_stid);
533 void cxgb3_insert_tid(struct t3cdev *tdev, struct cxgb3_client *client,
534 void *ctx, unsigned int tid)
536 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
538 t->tid_tab[tid].client = client;
539 t->tid_tab[tid].ctx = ctx;
540 atomic_inc(&t->tids_in_use);
543 EXPORT_SYMBOL(cxgb3_insert_tid);
546 * Populate a TID_RELEASE WR. The skb must be already propely sized.
548 static inline void mk_tid_release(struct sk_buff *skb, unsigned int tid)
550 struct cpl_tid_release *req;
552 skb->priority = CPL_PRIORITY_SETUP;
553 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
554 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
555 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
558 static void t3_process_tid_release_list(struct work_struct *work)
560 struct t3c_data *td = container_of(work, struct t3c_data,
563 struct t3cdev *tdev = td->dev;
566 spin_lock_bh(&td->tid_release_lock);
567 while (td->tid_release_list) {
568 struct t3c_tid_entry *p = td->tid_release_list;
570 td->tid_release_list = p->ctx;
571 spin_unlock_bh(&td->tid_release_lock);
573 skb = alloc_skb(sizeof(struct cpl_tid_release),
576 skb = td->nofail_skb;
578 spin_lock_bh(&td->tid_release_lock);
579 p->ctx = (void *)td->tid_release_list;
580 td->tid_release_list = p;
583 mk_tid_release(skb, p - td->tid_maps.tid_tab);
584 cxgb3_ofld_send(tdev, skb);
586 if (skb == td->nofail_skb)
588 alloc_skb(sizeof(struct cpl_tid_release),
590 spin_lock_bh(&td->tid_release_lock);
592 td->release_list_incomplete = (td->tid_release_list == NULL) ? 0 : 1;
593 spin_unlock_bh(&td->tid_release_lock);
597 alloc_skb(sizeof(struct cpl_tid_release),
601 /* use ctx as a next pointer in the tid release list */
602 void cxgb3_queue_tid_release(struct t3cdev *tdev, unsigned int tid)
604 struct t3c_data *td = T3C_DATA(tdev);
605 struct t3c_tid_entry *p = &td->tid_maps.tid_tab[tid];
607 spin_lock_bh(&td->tid_release_lock);
608 p->ctx = (void *)td->tid_release_list;
610 td->tid_release_list = p;
611 if (!p->ctx || td->release_list_incomplete)
612 schedule_work(&td->tid_release_task);
613 spin_unlock_bh(&td->tid_release_lock);
616 EXPORT_SYMBOL(cxgb3_queue_tid_release);
619 * Remove a tid from the TID table. A client may defer processing its last
620 * CPL message if it is locked at the time it arrives, and while the message
621 * sits in the client's backlog the TID may be reused for another connection.
622 * To handle this we atomically switch the TID association if it still points
623 * to the original client context.
625 void cxgb3_remove_tid(struct t3cdev *tdev, void *ctx, unsigned int tid)
627 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
629 BUG_ON(tid >= t->ntids);
630 if (tdev->type == T3A)
631 (void)cmpxchg(&t->tid_tab[tid].ctx, ctx, NULL);
635 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
637 mk_tid_release(skb, tid);
638 cxgb3_ofld_send(tdev, skb);
639 t->tid_tab[tid].ctx = NULL;
641 cxgb3_queue_tid_release(tdev, tid);
643 atomic_dec(&t->tids_in_use);
646 EXPORT_SYMBOL(cxgb3_remove_tid);
648 int cxgb3_alloc_atid(struct t3cdev *tdev, struct cxgb3_client *client,
652 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
654 spin_lock_bh(&t->atid_lock);
656 t->atids_in_use + atomic_read(&t->tids_in_use) + MC5_MIN_TIDS <=
658 union active_open_entry *p = t->afree;
660 atid = (p - t->atid_tab) + t->atid_base;
662 p->t3c_tid.ctx = ctx;
663 p->t3c_tid.client = client;
666 spin_unlock_bh(&t->atid_lock);
670 EXPORT_SYMBOL(cxgb3_alloc_atid);
672 int cxgb3_alloc_stid(struct t3cdev *tdev, struct cxgb3_client *client,
676 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
678 spin_lock_bh(&t->stid_lock);
680 union listen_entry *p = t->sfree;
682 stid = (p - t->stid_tab) + t->stid_base;
684 p->t3c_tid.ctx = ctx;
685 p->t3c_tid.client = client;
688 spin_unlock_bh(&t->stid_lock);
692 EXPORT_SYMBOL(cxgb3_alloc_stid);
694 /* Get the t3cdev associated with a net_device */
695 struct t3cdev *dev2t3cdev(struct net_device *dev)
697 const struct port_info *pi = netdev_priv(dev);
699 return (struct t3cdev *)pi->adapter;
702 EXPORT_SYMBOL(dev2t3cdev);
704 static int do_smt_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
706 struct cpl_smt_write_rpl *rpl = cplhdr(skb);
708 if (rpl->status != CPL_ERR_NONE)
710 "Unexpected SMT_WRITE_RPL status %u for entry %u\n",
711 rpl->status, GET_TID(rpl));
713 return CPL_RET_BUF_DONE;
716 static int do_l2t_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
718 struct cpl_l2t_write_rpl *rpl = cplhdr(skb);
720 if (rpl->status != CPL_ERR_NONE)
722 "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
723 rpl->status, GET_TID(rpl));
725 return CPL_RET_BUF_DONE;
728 static int do_rte_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
730 struct cpl_rte_write_rpl *rpl = cplhdr(skb);
732 if (rpl->status != CPL_ERR_NONE)
734 "Unexpected RTE_WRITE_RPL status %u for entry %u\n",
735 rpl->status, GET_TID(rpl));
737 return CPL_RET_BUF_DONE;
740 static int do_act_open_rpl(struct t3cdev *dev, struct sk_buff *skb)
742 struct cpl_act_open_rpl *rpl = cplhdr(skb);
743 unsigned int atid = G_TID(ntohl(rpl->atid));
744 struct t3c_tid_entry *t3c_tid;
746 t3c_tid = lookup_atid(&(T3C_DATA(dev))->tid_maps, atid);
747 if (t3c_tid && t3c_tid->ctx && t3c_tid->client &&
748 t3c_tid->client->handlers &&
749 t3c_tid->client->handlers[CPL_ACT_OPEN_RPL]) {
750 return t3c_tid->client->handlers[CPL_ACT_OPEN_RPL] (dev, skb,
754 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
755 dev->name, CPL_ACT_OPEN_RPL);
756 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
760 static int do_stid_rpl(struct t3cdev *dev, struct sk_buff *skb)
762 union opcode_tid *p = cplhdr(skb);
763 unsigned int stid = G_TID(ntohl(p->opcode_tid));
764 struct t3c_tid_entry *t3c_tid;
766 t3c_tid = lookup_stid(&(T3C_DATA(dev))->tid_maps, stid);
767 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
768 t3c_tid->client->handlers[p->opcode]) {
769 return t3c_tid->client->handlers[p->opcode] (dev, skb,
772 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
773 dev->name, p->opcode);
774 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
778 static int do_hwtid_rpl(struct t3cdev *dev, struct sk_buff *skb)
780 union opcode_tid *p = cplhdr(skb);
781 unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
782 struct t3c_tid_entry *t3c_tid;
784 t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
785 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
786 t3c_tid->client->handlers[p->opcode]) {
787 return t3c_tid->client->handlers[p->opcode]
788 (dev, skb, t3c_tid->ctx);
790 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
791 dev->name, p->opcode);
792 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
796 static int do_cr(struct t3cdev *dev, struct sk_buff *skb)
798 struct cpl_pass_accept_req *req = cplhdr(skb);
799 unsigned int stid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
800 struct tid_info *t = &(T3C_DATA(dev))->tid_maps;
801 struct t3c_tid_entry *t3c_tid;
802 unsigned int tid = GET_TID(req);
804 if (unlikely(tid >= t->ntids)) {
805 printk("%s: passive open TID %u too large\n",
807 t3_fatal_err(tdev2adap(dev));
808 return CPL_RET_BUF_DONE;
811 t3c_tid = lookup_stid(t, stid);
812 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
813 t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]) {
814 return t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]
815 (dev, skb, t3c_tid->ctx);
817 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
818 dev->name, CPL_PASS_ACCEPT_REQ);
819 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
824 * Returns an sk_buff for a reply CPL message of size len. If the input
825 * sk_buff has no other users it is trimmed and reused, otherwise a new buffer
826 * is allocated. The input skb must be of size at least len. Note that this
827 * operation does not destroy the original skb data even if it decides to reuse
830 static struct sk_buff *cxgb3_get_cpl_reply_skb(struct sk_buff *skb, size_t len,
833 if (likely(!skb_cloned(skb))) {
834 BUG_ON(skb->len < len);
835 __skb_trim(skb, len);
838 skb = alloc_skb(len, gfp);
845 static int do_abort_req_rss(struct t3cdev *dev, struct sk_buff *skb)
847 union opcode_tid *p = cplhdr(skb);
848 unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
849 struct t3c_tid_entry *t3c_tid;
851 t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
852 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
853 t3c_tid->client->handlers[p->opcode]) {
854 return t3c_tid->client->handlers[p->opcode]
855 (dev, skb, t3c_tid->ctx);
857 struct cpl_abort_req_rss *req = cplhdr(skb);
858 struct cpl_abort_rpl *rpl;
859 struct sk_buff *reply_skb;
860 unsigned int tid = GET_TID(req);
861 u8 cmd = req->status;
863 if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
864 req->status == CPL_ERR_PERSIST_NEG_ADVICE)
867 reply_skb = cxgb3_get_cpl_reply_skb(skb,
873 printk("do_abort_req_rss: couldn't get skb!\n");
876 reply_skb->priority = CPL_PRIORITY_DATA;
877 __skb_put(reply_skb, sizeof(struct cpl_abort_rpl));
878 rpl = cplhdr(reply_skb);
880 htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
881 rpl->wr.wr_lo = htonl(V_WR_TID(tid));
882 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, tid));
884 cxgb3_ofld_send(dev, reply_skb);
886 return CPL_RET_BUF_DONE;
890 static int do_act_establish(struct t3cdev *dev, struct sk_buff *skb)
892 struct cpl_act_establish *req = cplhdr(skb);
893 unsigned int atid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
894 struct tid_info *t = &(T3C_DATA(dev))->tid_maps;
895 struct t3c_tid_entry *t3c_tid;
896 unsigned int tid = GET_TID(req);
898 if (unlikely(tid >= t->ntids)) {
899 printk("%s: active establish TID %u too large\n",
901 t3_fatal_err(tdev2adap(dev));
902 return CPL_RET_BUF_DONE;
905 t3c_tid = lookup_atid(t, atid);
906 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
907 t3c_tid->client->handlers[CPL_ACT_ESTABLISH]) {
908 return t3c_tid->client->handlers[CPL_ACT_ESTABLISH]
909 (dev, skb, t3c_tid->ctx);
911 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
912 dev->name, CPL_ACT_ESTABLISH);
913 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
917 static int do_trace(struct t3cdev *dev, struct sk_buff *skb)
919 struct cpl_trace_pkt *p = cplhdr(skb);
921 skb->protocol = htons(0xffff);
922 skb->dev = dev->lldev;
923 skb_pull(skb, sizeof(*p));
924 skb_reset_mac_header(skb);
925 netif_receive_skb(skb);
930 * That skb would better have come from process_responses() where we abuse
931 * ->priority and ->csum to carry our data. NB: if we get to per-arch
932 * ->csum, the things might get really interesting here.
935 static inline u32 get_hwtid(struct sk_buff *skb)
937 return ntohl((__force __be32)skb->priority) >> 8 & 0xfffff;
940 static inline u32 get_opcode(struct sk_buff *skb)
942 return G_OPCODE(ntohl((__force __be32)skb->csum));
945 static int do_term(struct t3cdev *dev, struct sk_buff *skb)
947 unsigned int hwtid = get_hwtid(skb);
948 unsigned int opcode = get_opcode(skb);
949 struct t3c_tid_entry *t3c_tid;
951 t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
952 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
953 t3c_tid->client->handlers[opcode]) {
954 return t3c_tid->client->handlers[opcode] (dev, skb,
957 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
959 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
963 static int nb_callback(struct notifier_block *self, unsigned long event,
967 case (NETEVENT_NEIGH_UPDATE):{
968 cxgb_neigh_update((struct neighbour *)ctx);
971 case (NETEVENT_REDIRECT):{
972 struct netevent_redirect *nr = ctx;
973 cxgb_redirect(nr->old, nr->old_neigh,
974 nr->new, nr->new_neigh,
976 cxgb_neigh_update(nr->new_neigh);
985 static struct notifier_block nb = {
986 .notifier_call = nb_callback
990 * Process a received packet with an unknown/unexpected CPL opcode.
992 static int do_bad_cpl(struct t3cdev *dev, struct sk_buff *skb)
994 printk(KERN_ERR "%s: received bad CPL command 0x%x\n", dev->name,
996 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
1000 * Handlers for each CPL opcode
1002 static cpl_handler_func cpl_handlers[NUM_CPL_CMDS];
1005 * Add a new handler to the CPL dispatch table. A NULL handler may be supplied
1006 * to unregister an existing handler.
1008 void t3_register_cpl_handler(unsigned int opcode, cpl_handler_func h)
1010 if (opcode < NUM_CPL_CMDS)
1011 cpl_handlers[opcode] = h ? h : do_bad_cpl;
1013 printk(KERN_ERR "T3C: handler registration for "
1014 "opcode %x failed\n", opcode);
1017 EXPORT_SYMBOL(t3_register_cpl_handler);
1020 * T3CDEV's receive method.
1022 static int process_rx(struct t3cdev *dev, struct sk_buff **skbs, int n)
1025 struct sk_buff *skb = *skbs++;
1026 unsigned int opcode = get_opcode(skb);
1027 int ret = cpl_handlers[opcode] (dev, skb);
1030 if (ret & CPL_RET_UNKNOWN_TID) {
1031 union opcode_tid *p = cplhdr(skb);
1033 printk(KERN_ERR "%s: CPL message (opcode %u) had "
1034 "unknown TID %u\n", dev->name, opcode,
1035 G_TID(ntohl(p->opcode_tid)));
1038 if (ret & CPL_RET_BUF_DONE)
1045 * Sends an sk_buff to a T3C driver after dealing with any active network taps.
1047 int cxgb3_ofld_send(struct t3cdev *dev, struct sk_buff *skb)
1052 r = dev->send(dev, skb);
1057 EXPORT_SYMBOL(cxgb3_ofld_send);
1059 static int is_offloading(struct net_device *dev)
1061 struct adapter *adapter;
1064 read_lock_bh(&adapter_list_lock);
1065 list_for_each_entry(adapter, &adapter_list, adapter_list) {
1066 for_each_port(adapter, i) {
1067 if (dev == adapter->port[i]) {
1068 read_unlock_bh(&adapter_list_lock);
1073 read_unlock_bh(&adapter_list_lock);
1077 static void cxgb_neigh_update(struct neighbour *neigh)
1079 struct net_device *dev;
1084 if (dev && (is_offloading(dev))) {
1085 struct t3cdev *tdev = dev2t3cdev(dev);
1088 t3_l2t_update(tdev, neigh);
1092 static void set_l2t_ix(struct t3cdev *tdev, u32 tid, struct l2t_entry *e)
1094 struct sk_buff *skb;
1095 struct cpl_set_tcb_field *req;
1097 skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
1099 printk(KERN_ERR "%s: cannot allocate skb!\n", __func__);
1102 skb->priority = CPL_PRIORITY_CONTROL;
1103 req = (struct cpl_set_tcb_field *)skb_put(skb, sizeof(*req));
1104 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1105 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1108 req->word = htons(W_TCB_L2T_IX);
1109 req->mask = cpu_to_be64(V_TCB_L2T_IX(M_TCB_L2T_IX));
1110 req->val = cpu_to_be64(V_TCB_L2T_IX(e->idx));
1111 tdev->send(tdev, skb);
1114 static void cxgb_redirect(struct dst_entry *old, struct neighbour *old_neigh,
1115 struct dst_entry *new, struct neighbour *new_neigh,
1118 struct net_device *olddev, *newdev;
1119 struct tid_info *ti;
1120 struct t3cdev *tdev;
1123 struct l2t_entry *e;
1124 struct t3c_tid_entry *te;
1126 olddev = old_neigh->dev;
1127 newdev = new_neigh->dev;
1129 if (!is_offloading(olddev))
1131 if (!is_offloading(newdev)) {
1132 printk(KERN_WARNING "%s: Redirect to non-offload "
1133 "device ignored.\n", __func__);
1136 tdev = dev2t3cdev(olddev);
1138 if (tdev != dev2t3cdev(newdev)) {
1139 printk(KERN_WARNING "%s: Redirect to different "
1140 "offload device ignored.\n", __func__);
1144 /* Add new L2T entry */
1145 e = t3_l2t_get(tdev, new, newdev, daddr);
1147 printk(KERN_ERR "%s: couldn't allocate new l2t entry!\n",
1152 /* Walk tid table and notify clients of dst change. */
1153 ti = &(T3C_DATA(tdev))->tid_maps;
1154 for (tid = 0; tid < ti->ntids; tid++) {
1155 te = lookup_tid(ti, tid);
1157 if (te && te->ctx && te->client && te->client->redirect) {
1158 update_tcb = te->client->redirect(te->ctx, old, new, e);
1161 l2t_hold(L2DATA(tdev), e);
1163 set_l2t_ix(tdev, tid, e);
1167 l2t_release(tdev, e);
1171 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1172 * The allocated memory is cleared.
1174 void *cxgb_alloc_mem(unsigned long size)
1176 void *p = kzalloc(size, GFP_KERNEL);
1184 * Free memory allocated through t3_alloc_mem().
1186 void cxgb_free_mem(void *addr)
1188 if (is_vmalloc_addr(addr))
1195 * Allocate and initialize the TID tables. Returns 0 on success.
1197 static int init_tid_tabs(struct tid_info *t, unsigned int ntids,
1198 unsigned int natids, unsigned int nstids,
1199 unsigned int atid_base, unsigned int stid_base)
1201 unsigned long size = ntids * sizeof(*t->tid_tab) +
1202 natids * sizeof(*t->atid_tab) + nstids * sizeof(*t->stid_tab);
1204 t->tid_tab = cxgb_alloc_mem(size);
1208 t->stid_tab = (union listen_entry *)&t->tid_tab[ntids];
1209 t->atid_tab = (union active_open_entry *)&t->stid_tab[nstids];
1212 t->stid_base = stid_base;
1215 t->atid_base = atid_base;
1217 t->stids_in_use = t->atids_in_use = 0;
1218 atomic_set(&t->tids_in_use, 0);
1219 spin_lock_init(&t->stid_lock);
1220 spin_lock_init(&t->atid_lock);
1223 * Setup the free lists for stid_tab and atid_tab.
1227 t->stid_tab[nstids - 1].next = &t->stid_tab[nstids];
1228 t->sfree = t->stid_tab;
1232 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1233 t->afree = t->atid_tab;
1238 static void free_tid_maps(struct tid_info *t)
1240 cxgb_free_mem(t->tid_tab);
1243 static inline void add_adapter(struct adapter *adap)
1245 write_lock_bh(&adapter_list_lock);
1246 list_add_tail(&adap->adapter_list, &adapter_list);
1247 write_unlock_bh(&adapter_list_lock);
1250 static inline void remove_adapter(struct adapter *adap)
1252 write_lock_bh(&adapter_list_lock);
1253 list_del(&adap->adapter_list);
1254 write_unlock_bh(&adapter_list_lock);
1257 int cxgb3_offload_activate(struct adapter *adapter)
1259 struct t3cdev *dev = &adapter->tdev;
1262 struct tid_range stid_range, tid_range;
1263 struct mtutab mtutab;
1264 unsigned int l2t_capacity;
1266 t = kzalloc(sizeof(*t), GFP_KERNEL);
1271 if (dev->ctl(dev, GET_TX_MAX_CHUNK, &t->tx_max_chunk) < 0 ||
1272 dev->ctl(dev, GET_MAX_OUTSTANDING_WR, &t->max_wrs) < 0 ||
1273 dev->ctl(dev, GET_L2T_CAPACITY, &l2t_capacity) < 0 ||
1274 dev->ctl(dev, GET_MTUS, &mtutab) < 0 ||
1275 dev->ctl(dev, GET_TID_RANGE, &tid_range) < 0 ||
1276 dev->ctl(dev, GET_STID_RANGE, &stid_range) < 0)
1280 RCU_INIT_POINTER(dev->l2opt, t3_init_l2t(l2t_capacity));
1284 natids = min(tid_range.num / 2, MAX_ATIDS);
1285 err = init_tid_tabs(&t->tid_maps, tid_range.num, natids,
1286 stid_range.num, ATID_BASE, stid_range.base);
1290 t->mtus = mtutab.mtus;
1291 t->nmtus = mtutab.size;
1293 INIT_WORK(&t->tid_release_task, t3_process_tid_release_list);
1294 spin_lock_init(&t->tid_release_lock);
1295 INIT_LIST_HEAD(&t->list_node);
1299 dev->recv = process_rx;
1300 dev->neigh_update = t3_l2t_update;
1302 /* Register netevent handler once */
1303 if (list_empty(&adapter_list))
1304 register_netevent_notifier(&nb);
1306 t->nofail_skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_KERNEL);
1307 t->release_list_incomplete = 0;
1309 add_adapter(adapter);
1313 t3_free_l2t(L2DATA(dev));
1314 RCU_INIT_POINTER(dev->l2opt, NULL);
1320 static void clean_l2_data(struct rcu_head *head)
1322 struct l2t_data *d = container_of(head, struct l2t_data, rcu_head);
1327 void cxgb3_offload_deactivate(struct adapter *adapter)
1329 struct t3cdev *tdev = &adapter->tdev;
1330 struct t3c_data *t = T3C_DATA(tdev);
1333 remove_adapter(adapter);
1334 if (list_empty(&adapter_list))
1335 unregister_netevent_notifier(&nb);
1337 free_tid_maps(&t->tid_maps);
1338 T3C_DATA(tdev) = NULL;
1342 RCU_INIT_POINTER(tdev->l2opt, NULL);
1343 call_rcu(&d->rcu_head, clean_l2_data);
1345 kfree_skb(t->nofail_skb);
1349 static inline void register_tdev(struct t3cdev *tdev)
1353 mutex_lock(&cxgb3_db_lock);
1354 snprintf(tdev->name, sizeof(tdev->name), "ofld_dev%d", unit++);
1355 list_add_tail(&tdev->ofld_dev_list, &ofld_dev_list);
1356 mutex_unlock(&cxgb3_db_lock);
1359 static inline void unregister_tdev(struct t3cdev *tdev)
1361 mutex_lock(&cxgb3_db_lock);
1362 list_del(&tdev->ofld_dev_list);
1363 mutex_unlock(&cxgb3_db_lock);
1366 static inline int adap2type(struct adapter *adapter)
1370 switch (adapter->params.rev) {
1385 void cxgb3_adapter_ofld(struct adapter *adapter)
1387 struct t3cdev *tdev = &adapter->tdev;
1389 INIT_LIST_HEAD(&tdev->ofld_dev_list);
1391 cxgb3_set_dummy_ops(tdev);
1392 tdev->send = t3_offload_tx;
1393 tdev->ctl = cxgb_offload_ctl;
1394 tdev->type = adap2type(adapter);
1396 register_tdev(tdev);
1399 void cxgb3_adapter_unofld(struct adapter *adapter)
1401 struct t3cdev *tdev = &adapter->tdev;
1404 tdev->neigh_update = NULL;
1406 unregister_tdev(tdev);
1409 void __init cxgb3_offload_init(void)
1413 for (i = 0; i < NUM_CPL_CMDS; ++i)
1414 cpl_handlers[i] = do_bad_cpl;
1416 t3_register_cpl_handler(CPL_SMT_WRITE_RPL, do_smt_write_rpl);
1417 t3_register_cpl_handler(CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
1418 t3_register_cpl_handler(CPL_RTE_WRITE_RPL, do_rte_write_rpl);
1419 t3_register_cpl_handler(CPL_PASS_OPEN_RPL, do_stid_rpl);
1420 t3_register_cpl_handler(CPL_CLOSE_LISTSRV_RPL, do_stid_rpl);
1421 t3_register_cpl_handler(CPL_PASS_ACCEPT_REQ, do_cr);
1422 t3_register_cpl_handler(CPL_PASS_ESTABLISH, do_hwtid_rpl);
1423 t3_register_cpl_handler(CPL_ABORT_RPL_RSS, do_hwtid_rpl);
1424 t3_register_cpl_handler(CPL_ABORT_RPL, do_hwtid_rpl);
1425 t3_register_cpl_handler(CPL_RX_URG_NOTIFY, do_hwtid_rpl);
1426 t3_register_cpl_handler(CPL_RX_DATA, do_hwtid_rpl);
1427 t3_register_cpl_handler(CPL_TX_DATA_ACK, do_hwtid_rpl);
1428 t3_register_cpl_handler(CPL_TX_DMA_ACK, do_hwtid_rpl);
1429 t3_register_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl);
1430 t3_register_cpl_handler(CPL_PEER_CLOSE, do_hwtid_rpl);
1431 t3_register_cpl_handler(CPL_CLOSE_CON_RPL, do_hwtid_rpl);
1432 t3_register_cpl_handler(CPL_ABORT_REQ_RSS, do_abort_req_rss);
1433 t3_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish);
1434 t3_register_cpl_handler(CPL_SET_TCB_RPL, do_hwtid_rpl);
1435 t3_register_cpl_handler(CPL_GET_TCB_RPL, do_hwtid_rpl);
1436 t3_register_cpl_handler(CPL_RDMA_TERMINATE, do_term);
1437 t3_register_cpl_handler(CPL_RDMA_EC_STATUS, do_hwtid_rpl);
1438 t3_register_cpl_handler(CPL_TRACE_PKT, do_trace);
1439 t3_register_cpl_handler(CPL_RX_DATA_DDP, do_hwtid_rpl);
1440 t3_register_cpl_handler(CPL_RX_DDP_COMPLETE, do_hwtid_rpl);
1441 t3_register_cpl_handler(CPL_ISCSI_HDR, do_hwtid_rpl);