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 <net/neighbour.h>
35 #include <linux/notifier.h>
36 #include <asm/atomic.h>
37 #include <linux/proc_fs.h>
38 #include <linux/if_vlan.h>
39 #include <net/netevent.h>
40 #include <linux/highmem.h>
41 #include <linux/vmalloc.h>
45 #include "cxgb3_ioctl.h"
46 #include "cxgb3_ctl_defs.h"
47 #include "cxgb3_defs.h"
49 #include "firmware_exports.h"
50 #include "cxgb3_offload.h"
52 static LIST_HEAD(client_list);
53 static LIST_HEAD(ofld_dev_list);
54 static DEFINE_MUTEX(cxgb3_db_lock);
56 static DEFINE_RWLOCK(adapter_list_lock);
57 static LIST_HEAD(adapter_list);
59 static const unsigned int MAX_ATIDS = 64 * 1024;
60 static const unsigned int ATID_BASE = 0x10000;
62 static inline int offload_activated(struct t3cdev *tdev)
64 const struct adapter *adapter = tdev2adap(tdev);
66 return (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map));
70 * cxgb3_register_client - register an offload client
73 * Add the client to the client list,
74 * and call backs the client for each activated offload device
76 void cxgb3_register_client(struct cxgb3_client *client)
80 mutex_lock(&cxgb3_db_lock);
81 list_add_tail(&client->client_list, &client_list);
84 list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
85 if (offload_activated(tdev))
89 mutex_unlock(&cxgb3_db_lock);
92 EXPORT_SYMBOL(cxgb3_register_client);
95 * cxgb3_unregister_client - unregister an offload client
98 * Remove the client to the client list,
99 * and call backs the client for each activated offload device.
101 void cxgb3_unregister_client(struct cxgb3_client *client)
105 mutex_lock(&cxgb3_db_lock);
106 list_del(&client->client_list);
108 if (client->remove) {
109 list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
110 if (offload_activated(tdev))
111 client->remove(tdev);
114 mutex_unlock(&cxgb3_db_lock);
117 EXPORT_SYMBOL(cxgb3_unregister_client);
120 * cxgb3_add_clients - activate registered clients for an offload device
121 * @tdev: the offload device
123 * Call backs all registered clients once a offload device is activated
125 void cxgb3_add_clients(struct t3cdev *tdev)
127 struct cxgb3_client *client;
129 mutex_lock(&cxgb3_db_lock);
130 list_for_each_entry(client, &client_list, client_list) {
134 mutex_unlock(&cxgb3_db_lock);
138 * cxgb3_remove_clients - deactivates registered clients
139 * for an offload device
140 * @tdev: the offload device
142 * Call backs all registered clients once a offload device is deactivated
144 void cxgb3_remove_clients(struct t3cdev *tdev)
146 struct cxgb3_client *client;
148 mutex_lock(&cxgb3_db_lock);
149 list_for_each_entry(client, &client_list, client_list) {
151 client->remove(tdev);
153 mutex_unlock(&cxgb3_db_lock);
156 void cxgb3_err_notify(struct t3cdev *tdev, u32 status, u32 error)
158 struct cxgb3_client *client;
160 mutex_lock(&cxgb3_db_lock);
161 list_for_each_entry(client, &client_list, client_list) {
162 if (client->err_handler)
163 client->err_handler(tdev, status, error);
165 mutex_unlock(&cxgb3_db_lock);
168 static struct net_device *get_iff_from_mac(struct adapter *adapter,
169 const unsigned char *mac,
174 for_each_port(adapter, i) {
175 struct vlan_group *grp;
176 struct net_device *dev = adapter->port[i];
177 const struct port_info *p = netdev_priv(dev);
179 if (!memcmp(dev->dev_addr, mac, ETH_ALEN)) {
180 if (vlan && vlan != VLAN_VID_MASK) {
184 dev = vlan_group_get_device(grp, vlan);
194 static int cxgb_ulp_iscsi_ctl(struct adapter *adapter, unsigned int req,
199 unsigned int val = 0;
200 struct ulp_iscsi_info *uiip = data;
203 case ULP_ISCSI_GET_PARAMS:
204 uiip->pdev = adapter->pdev;
205 uiip->llimit = t3_read_reg(adapter, A_ULPRX_ISCSI_LLIMIT);
206 uiip->ulimit = t3_read_reg(adapter, A_ULPRX_ISCSI_ULIMIT);
207 uiip->tagmask = t3_read_reg(adapter, A_ULPRX_ISCSI_TAGMASK);
209 val = t3_read_reg(adapter, A_ULPRX_ISCSI_PSZ);
210 for (i = 0; i < 4; i++, val >>= 8)
211 uiip->pgsz_factor[i] = val & 0xFF;
213 val = t3_read_reg(adapter, A_TP_PARA_REG7);
215 uiip->max_rxsz = min((val >> S_PMMAXXFERLEN0)&M_PMMAXXFERLEN0,
216 (val >> S_PMMAXXFERLEN1)&M_PMMAXXFERLEN1);
218 * On tx, the iscsi pdu has to be <= tx page size and has to
219 * fit into the Tx PM FIFO.
221 val = min(adapter->params.tp.tx_pg_size,
222 t3_read_reg(adapter, A_PM1_TX_CFG) >> 17);
223 uiip->max_txsz = min(val, uiip->max_txsz);
225 /* set MaxRxData to 16224 */
226 val = t3_read_reg(adapter, A_TP_PARA_REG2);
227 if ((val >> S_MAXRXDATA) != 0x3f60) {
228 val &= (M_RXCOALESCESIZE << S_RXCOALESCESIZE);
229 val |= V_MAXRXDATA(0x3f60);
231 "%s, iscsi set MaxRxData to 16224 (0x%x).\n",
233 t3_write_reg(adapter, A_TP_PARA_REG2, val);
237 * on rx, the iscsi pdu has to be < rx page size and the
238 * the max rx data length programmed in TP
240 val = min(adapter->params.tp.rx_pg_size,
241 ((t3_read_reg(adapter, A_TP_PARA_REG2)) >>
242 S_MAXRXDATA) & M_MAXRXDATA);
243 uiip->max_rxsz = min(val, uiip->max_rxsz);
245 case ULP_ISCSI_SET_PARAMS:
246 t3_write_reg(adapter, A_ULPRX_ISCSI_TAGMASK, uiip->tagmask);
247 /* program the ddp page sizes */
248 for (i = 0; i < 4; i++)
249 val |= (uiip->pgsz_factor[i] & 0xF) << (8 * i);
250 if (val && (val != t3_read_reg(adapter, A_ULPRX_ISCSI_PSZ))) {
252 "%s, setting iscsi pgsz 0x%x, %u,%u,%u,%u.\n",
253 adapter->name, val, uiip->pgsz_factor[0],
254 uiip->pgsz_factor[1], uiip->pgsz_factor[2],
255 uiip->pgsz_factor[3]);
256 t3_write_reg(adapter, A_ULPRX_ISCSI_PSZ, val);
265 /* Response queue used for RDMA events. */
266 #define ASYNC_NOTIF_RSPQ 0
268 static int cxgb_rdma_ctl(struct adapter *adapter, unsigned int req, void *data)
273 case RDMA_GET_PARAMS: {
274 struct rdma_info *rdma = data;
275 struct pci_dev *pdev = adapter->pdev;
277 rdma->udbell_physbase = pci_resource_start(pdev, 2);
278 rdma->udbell_len = pci_resource_len(pdev, 2);
280 t3_read_reg(adapter, A_ULPTX_TPT_LLIMIT);
281 rdma->tpt_top = t3_read_reg(adapter, A_ULPTX_TPT_ULIMIT);
283 t3_read_reg(adapter, A_ULPTX_PBL_LLIMIT);
284 rdma->pbl_top = t3_read_reg(adapter, A_ULPTX_PBL_ULIMIT);
285 rdma->rqt_base = t3_read_reg(adapter, A_ULPRX_RQ_LLIMIT);
286 rdma->rqt_top = t3_read_reg(adapter, A_ULPRX_RQ_ULIMIT);
287 rdma->kdb_addr = adapter->regs + A_SG_KDOORBELL;
293 struct rdma_cq_op *rdma = data;
295 /* may be called in any context */
296 spin_lock_irqsave(&adapter->sge.reg_lock, flags);
297 ret = t3_sge_cqcntxt_op(adapter, rdma->id, rdma->op,
299 spin_unlock_irqrestore(&adapter->sge.reg_lock, flags);
303 struct ch_mem_range *t = data;
306 if ((t->addr & 7) || (t->len & 7))
308 if (t->mem_id == MEM_CM)
310 else if (t->mem_id == MEM_PMRX)
311 mem = &adapter->pmrx;
312 else if (t->mem_id == MEM_PMTX)
313 mem = &adapter->pmtx;
318 t3_mc7_bd_read(mem, t->addr / 8, t->len / 8,
325 struct rdma_cq_setup *rdma = data;
327 spin_lock_irq(&adapter->sge.reg_lock);
329 t3_sge_init_cqcntxt(adapter, rdma->id,
330 rdma->base_addr, rdma->size,
332 rdma->ovfl_mode, rdma->credits,
334 spin_unlock_irq(&adapter->sge.reg_lock);
337 case RDMA_CQ_DISABLE:
338 spin_lock_irq(&adapter->sge.reg_lock);
339 ret = t3_sge_disable_cqcntxt(adapter, *(unsigned int *)data);
340 spin_unlock_irq(&adapter->sge.reg_lock);
342 case RDMA_CTRL_QP_SETUP:{
343 struct rdma_ctrlqp_setup *rdma = data;
345 spin_lock_irq(&adapter->sge.reg_lock);
346 ret = t3_sge_init_ecntxt(adapter, FW_RI_SGEEC_START, 0,
349 rdma->base_addr, rdma->size,
350 FW_RI_TID_START, 1, 0);
351 spin_unlock_irq(&adapter->sge.reg_lock);
355 spin_lock(&adapter->stats_lock);
356 t3_tp_get_mib_stats(adapter, (struct tp_mib_stats *)data);
357 spin_unlock(&adapter->stats_lock);
366 static int cxgb_offload_ctl(struct t3cdev *tdev, unsigned int req, void *data)
368 struct adapter *adapter = tdev2adap(tdev);
369 struct tid_range *tid;
371 struct iff_mac *iffmacp;
372 struct ddp_params *ddpp;
373 struct adap_ports *ports;
374 struct ofld_page_info *rx_page_info;
375 struct tp_params *tp = &adapter->params.tp;
379 case GET_MAX_OUTSTANDING_WR:
380 *(unsigned int *)data = FW_WR_NUM;
383 *(unsigned int *)data = WR_FLITS;
385 case GET_TX_MAX_CHUNK:
386 *(unsigned int *)data = 1 << 20; /* 1MB */
390 tid->num = t3_mc5_size(&adapter->mc5) -
391 adapter->params.mc5.nroutes -
392 adapter->params.mc5.nfilters - adapter->params.mc5.nservers;
397 tid->num = adapter->params.mc5.nservers;
398 tid->base = t3_mc5_size(&adapter->mc5) - tid->num -
399 adapter->params.mc5.nfilters - adapter->params.mc5.nroutes;
401 case GET_L2T_CAPACITY:
402 *(unsigned int *)data = 2048;
407 mtup->mtus = adapter->params.mtus;
409 case GET_IFF_FROM_MAC:
411 iffmacp->dev = get_iff_from_mac(adapter, iffmacp->mac_addr,
417 ddpp->llimit = t3_read_reg(adapter, A_ULPRX_TDDP_LLIMIT);
418 ddpp->ulimit = t3_read_reg(adapter, A_ULPRX_TDDP_ULIMIT);
419 ddpp->tag_mask = t3_read_reg(adapter, A_ULPRX_TDDP_TAGMASK);
423 ports->nports = adapter->params.nports;
424 for_each_port(adapter, i)
425 ports->lldevs[i] = adapter->port[i];
427 case ULP_ISCSI_GET_PARAMS:
428 case ULP_ISCSI_SET_PARAMS:
429 if (!offload_running(adapter))
431 return cxgb_ulp_iscsi_ctl(adapter, req, data);
432 case RDMA_GET_PARAMS:
435 case RDMA_CQ_DISABLE:
436 case RDMA_CTRL_QP_SETUP:
439 if (!offload_running(adapter))
441 return cxgb_rdma_ctl(adapter, req, data);
442 case GET_RX_PAGE_INFO:
444 rx_page_info->page_size = tp->rx_pg_size;
445 rx_page_info->num = tp->rx_num_pgs;
447 case GET_ISCSI_IPV4ADDR: {
448 struct iscsi_ipv4addr *p = data;
449 struct port_info *pi = netdev_priv(p->dev);
450 p->ipv4addr = pi->iscsi_ipv4addr;
453 case GET_EMBEDDED_INFO: {
454 struct ch_embedded_info *e = data;
456 spin_lock(&adapter->stats_lock);
457 t3_get_fw_version(adapter, &e->fw_vers);
458 t3_get_tp_version(adapter, &e->tp_vers);
459 spin_unlock(&adapter->stats_lock);
469 * Dummy handler for Rx offload packets in case we get an offload packet before
470 * proper processing is setup. This complains and drops the packet as it isn't
471 * normal to get offload packets at this stage.
473 static int rx_offload_blackhole(struct t3cdev *dev, struct sk_buff **skbs,
477 dev_kfree_skb_any(skbs[n]);
481 static void dummy_neigh_update(struct t3cdev *dev, struct neighbour *neigh)
485 void cxgb3_set_dummy_ops(struct t3cdev *dev)
487 dev->recv = rx_offload_blackhole;
488 dev->neigh_update = dummy_neigh_update;
492 * Free an active-open TID.
494 void *cxgb3_free_atid(struct t3cdev *tdev, int atid)
496 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
497 union active_open_entry *p = atid2entry(t, atid);
498 void *ctx = p->t3c_tid.ctx;
500 spin_lock_bh(&t->atid_lock);
504 spin_unlock_bh(&t->atid_lock);
509 EXPORT_SYMBOL(cxgb3_free_atid);
512 * Free a server TID and return it to the free pool.
514 void cxgb3_free_stid(struct t3cdev *tdev, int stid)
516 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
517 union listen_entry *p = stid2entry(t, stid);
519 spin_lock_bh(&t->stid_lock);
523 spin_unlock_bh(&t->stid_lock);
526 EXPORT_SYMBOL(cxgb3_free_stid);
528 void cxgb3_insert_tid(struct t3cdev *tdev, struct cxgb3_client *client,
529 void *ctx, unsigned int tid)
531 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
533 t->tid_tab[tid].client = client;
534 t->tid_tab[tid].ctx = ctx;
535 atomic_inc(&t->tids_in_use);
538 EXPORT_SYMBOL(cxgb3_insert_tid);
541 * Populate a TID_RELEASE WR. The skb must be already propely sized.
543 static inline void mk_tid_release(struct sk_buff *skb, unsigned int tid)
545 struct cpl_tid_release *req;
547 skb->priority = CPL_PRIORITY_SETUP;
548 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
549 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
550 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
553 static void t3_process_tid_release_list(struct work_struct *work)
555 struct t3c_data *td = container_of(work, struct t3c_data,
558 struct t3cdev *tdev = td->dev;
561 spin_lock_bh(&td->tid_release_lock);
562 while (td->tid_release_list) {
563 struct t3c_tid_entry *p = td->tid_release_list;
565 td->tid_release_list = (struct t3c_tid_entry *)p->ctx;
566 spin_unlock_bh(&td->tid_release_lock);
568 skb = alloc_skb(sizeof(struct cpl_tid_release),
571 skb = td->nofail_skb;
573 spin_lock_bh(&td->tid_release_lock);
574 p->ctx = (void *)td->tid_release_list;
575 td->tid_release_list = (struct t3c_tid_entry *)p;
578 mk_tid_release(skb, p - td->tid_maps.tid_tab);
579 cxgb3_ofld_send(tdev, skb);
581 if (skb == td->nofail_skb)
583 alloc_skb(sizeof(struct cpl_tid_release),
585 spin_lock_bh(&td->tid_release_lock);
587 td->release_list_incomplete = (td->tid_release_list == NULL) ? 0 : 1;
588 spin_unlock_bh(&td->tid_release_lock);
592 alloc_skb(sizeof(struct cpl_tid_release),
596 /* use ctx as a next pointer in the tid release list */
597 void cxgb3_queue_tid_release(struct t3cdev *tdev, unsigned int tid)
599 struct t3c_data *td = T3C_DATA(tdev);
600 struct t3c_tid_entry *p = &td->tid_maps.tid_tab[tid];
602 spin_lock_bh(&td->tid_release_lock);
603 p->ctx = (void *)td->tid_release_list;
605 td->tid_release_list = p;
606 if (!p->ctx || td->release_list_incomplete)
607 schedule_work(&td->tid_release_task);
608 spin_unlock_bh(&td->tid_release_lock);
611 EXPORT_SYMBOL(cxgb3_queue_tid_release);
614 * Remove a tid from the TID table. A client may defer processing its last
615 * CPL message if it is locked at the time it arrives, and while the message
616 * sits in the client's backlog the TID may be reused for another connection.
617 * To handle this we atomically switch the TID association if it still points
618 * to the original client context.
620 void cxgb3_remove_tid(struct t3cdev *tdev, void *ctx, unsigned int tid)
622 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
624 BUG_ON(tid >= t->ntids);
625 if (tdev->type == T3A)
626 (void)cmpxchg(&t->tid_tab[tid].ctx, ctx, NULL);
630 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
632 mk_tid_release(skb, tid);
633 cxgb3_ofld_send(tdev, skb);
634 t->tid_tab[tid].ctx = NULL;
636 cxgb3_queue_tid_release(tdev, tid);
638 atomic_dec(&t->tids_in_use);
641 EXPORT_SYMBOL(cxgb3_remove_tid);
643 int cxgb3_alloc_atid(struct t3cdev *tdev, struct cxgb3_client *client,
647 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
649 spin_lock_bh(&t->atid_lock);
651 t->atids_in_use + atomic_read(&t->tids_in_use) + MC5_MIN_TIDS <=
653 union active_open_entry *p = t->afree;
655 atid = (p - t->atid_tab) + t->atid_base;
657 p->t3c_tid.ctx = ctx;
658 p->t3c_tid.client = client;
661 spin_unlock_bh(&t->atid_lock);
665 EXPORT_SYMBOL(cxgb3_alloc_atid);
667 int cxgb3_alloc_stid(struct t3cdev *tdev, struct cxgb3_client *client,
671 struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
673 spin_lock_bh(&t->stid_lock);
675 union listen_entry *p = t->sfree;
677 stid = (p - t->stid_tab) + t->stid_base;
679 p->t3c_tid.ctx = ctx;
680 p->t3c_tid.client = client;
683 spin_unlock_bh(&t->stid_lock);
687 EXPORT_SYMBOL(cxgb3_alloc_stid);
689 /* Get the t3cdev associated with a net_device */
690 struct t3cdev *dev2t3cdev(struct net_device *dev)
692 const struct port_info *pi = netdev_priv(dev);
694 return (struct t3cdev *)pi->adapter;
697 EXPORT_SYMBOL(dev2t3cdev);
699 static int do_smt_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
701 struct cpl_smt_write_rpl *rpl = cplhdr(skb);
703 if (rpl->status != CPL_ERR_NONE)
705 "Unexpected SMT_WRITE_RPL status %u for entry %u\n",
706 rpl->status, GET_TID(rpl));
708 return CPL_RET_BUF_DONE;
711 static int do_l2t_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
713 struct cpl_l2t_write_rpl *rpl = cplhdr(skb);
715 if (rpl->status != CPL_ERR_NONE)
717 "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
718 rpl->status, GET_TID(rpl));
720 return CPL_RET_BUF_DONE;
723 static int do_rte_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
725 struct cpl_rte_write_rpl *rpl = cplhdr(skb);
727 if (rpl->status != CPL_ERR_NONE)
729 "Unexpected RTE_WRITE_RPL status %u for entry %u\n",
730 rpl->status, GET_TID(rpl));
732 return CPL_RET_BUF_DONE;
735 static int do_act_open_rpl(struct t3cdev *dev, struct sk_buff *skb)
737 struct cpl_act_open_rpl *rpl = cplhdr(skb);
738 unsigned int atid = G_TID(ntohl(rpl->atid));
739 struct t3c_tid_entry *t3c_tid;
741 t3c_tid = lookup_atid(&(T3C_DATA(dev))->tid_maps, atid);
742 if (t3c_tid && t3c_tid->ctx && t3c_tid->client &&
743 t3c_tid->client->handlers &&
744 t3c_tid->client->handlers[CPL_ACT_OPEN_RPL]) {
745 return t3c_tid->client->handlers[CPL_ACT_OPEN_RPL] (dev, skb,
749 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
750 dev->name, CPL_ACT_OPEN_RPL);
751 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
755 static int do_stid_rpl(struct t3cdev *dev, struct sk_buff *skb)
757 union opcode_tid *p = cplhdr(skb);
758 unsigned int stid = G_TID(ntohl(p->opcode_tid));
759 struct t3c_tid_entry *t3c_tid;
761 t3c_tid = lookup_stid(&(T3C_DATA(dev))->tid_maps, stid);
762 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
763 t3c_tid->client->handlers[p->opcode]) {
764 return t3c_tid->client->handlers[p->opcode] (dev, skb,
767 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
768 dev->name, p->opcode);
769 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
773 static int do_hwtid_rpl(struct t3cdev *dev, struct sk_buff *skb)
775 union opcode_tid *p = cplhdr(skb);
776 unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
777 struct t3c_tid_entry *t3c_tid;
779 t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
780 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
781 t3c_tid->client->handlers[p->opcode]) {
782 return t3c_tid->client->handlers[p->opcode]
783 (dev, skb, t3c_tid->ctx);
785 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
786 dev->name, p->opcode);
787 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
791 static int do_cr(struct t3cdev *dev, struct sk_buff *skb)
793 struct cpl_pass_accept_req *req = cplhdr(skb);
794 unsigned int stid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
795 struct tid_info *t = &(T3C_DATA(dev))->tid_maps;
796 struct t3c_tid_entry *t3c_tid;
797 unsigned int tid = GET_TID(req);
799 if (unlikely(tid >= t->ntids)) {
800 printk("%s: passive open TID %u too large\n",
802 t3_fatal_err(tdev2adap(dev));
803 return CPL_RET_BUF_DONE;
806 t3c_tid = lookup_stid(t, stid);
807 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
808 t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]) {
809 return t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]
810 (dev, skb, t3c_tid->ctx);
812 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
813 dev->name, CPL_PASS_ACCEPT_REQ);
814 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
819 * Returns an sk_buff for a reply CPL message of size len. If the input
820 * sk_buff has no other users it is trimmed and reused, otherwise a new buffer
821 * is allocated. The input skb must be of size at least len. Note that this
822 * operation does not destroy the original skb data even if it decides to reuse
825 static struct sk_buff *cxgb3_get_cpl_reply_skb(struct sk_buff *skb, size_t len,
828 if (likely(!skb_cloned(skb))) {
829 BUG_ON(skb->len < len);
830 __skb_trim(skb, len);
833 skb = alloc_skb(len, gfp);
840 static int do_abort_req_rss(struct t3cdev *dev, struct sk_buff *skb)
842 union opcode_tid *p = cplhdr(skb);
843 unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
844 struct t3c_tid_entry *t3c_tid;
846 t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
847 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
848 t3c_tid->client->handlers[p->opcode]) {
849 return t3c_tid->client->handlers[p->opcode]
850 (dev, skb, t3c_tid->ctx);
852 struct cpl_abort_req_rss *req = cplhdr(skb);
853 struct cpl_abort_rpl *rpl;
854 struct sk_buff *reply_skb;
855 unsigned int tid = GET_TID(req);
856 u8 cmd = req->status;
858 if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
859 req->status == CPL_ERR_PERSIST_NEG_ADVICE)
862 reply_skb = cxgb3_get_cpl_reply_skb(skb,
868 printk("do_abort_req_rss: couldn't get skb!\n");
871 reply_skb->priority = CPL_PRIORITY_DATA;
872 __skb_put(reply_skb, sizeof(struct cpl_abort_rpl));
873 rpl = cplhdr(reply_skb);
875 htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
876 rpl->wr.wr_lo = htonl(V_WR_TID(tid));
877 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, tid));
879 cxgb3_ofld_send(dev, reply_skb);
881 return CPL_RET_BUF_DONE;
885 static int do_act_establish(struct t3cdev *dev, struct sk_buff *skb)
887 struct cpl_act_establish *req = cplhdr(skb);
888 unsigned int atid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
889 struct tid_info *t = &(T3C_DATA(dev))->tid_maps;
890 struct t3c_tid_entry *t3c_tid;
891 unsigned int tid = GET_TID(req);
893 if (unlikely(tid >= t->ntids)) {
894 printk("%s: active establish TID %u too large\n",
896 t3_fatal_err(tdev2adap(dev));
897 return CPL_RET_BUF_DONE;
900 t3c_tid = lookup_atid(t, atid);
901 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
902 t3c_tid->client->handlers[CPL_ACT_ESTABLISH]) {
903 return t3c_tid->client->handlers[CPL_ACT_ESTABLISH]
904 (dev, skb, t3c_tid->ctx);
906 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
907 dev->name, CPL_ACT_ESTABLISH);
908 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
912 static int do_trace(struct t3cdev *dev, struct sk_buff *skb)
914 struct cpl_trace_pkt *p = cplhdr(skb);
916 skb->protocol = htons(0xffff);
917 skb->dev = dev->lldev;
918 skb_pull(skb, sizeof(*p));
919 skb_reset_mac_header(skb);
920 netif_receive_skb(skb);
925 * That skb would better have come from process_responses() where we abuse
926 * ->priority and ->csum to carry our data. NB: if we get to per-arch
927 * ->csum, the things might get really interesting here.
930 static inline u32 get_hwtid(struct sk_buff *skb)
932 return ntohl((__force __be32)skb->priority) >> 8 & 0xfffff;
935 static inline u32 get_opcode(struct sk_buff *skb)
937 return G_OPCODE(ntohl((__force __be32)skb->csum));
940 static int do_term(struct t3cdev *dev, struct sk_buff *skb)
942 unsigned int hwtid = get_hwtid(skb);
943 unsigned int opcode = get_opcode(skb);
944 struct t3c_tid_entry *t3c_tid;
946 t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
947 if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
948 t3c_tid->client->handlers[opcode]) {
949 return t3c_tid->client->handlers[opcode] (dev, skb,
952 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
954 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
958 static int nb_callback(struct notifier_block *self, unsigned long event,
962 case (NETEVENT_NEIGH_UPDATE):{
963 cxgb_neigh_update((struct neighbour *)ctx);
966 case (NETEVENT_PMTU_UPDATE):
968 case (NETEVENT_REDIRECT):{
969 struct netevent_redirect *nr = ctx;
970 cxgb_redirect(nr->old, nr->new);
971 cxgb_neigh_update(nr->new->neighbour);
980 static struct notifier_block nb = {
981 .notifier_call = nb_callback
985 * Process a received packet with an unknown/unexpected CPL opcode.
987 static int do_bad_cpl(struct t3cdev *dev, struct sk_buff *skb)
989 printk(KERN_ERR "%s: received bad CPL command 0x%x\n", dev->name,
991 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
995 * Handlers for each CPL opcode
997 static cpl_handler_func cpl_handlers[NUM_CPL_CMDS];
1000 * Add a new handler to the CPL dispatch table. A NULL handler may be supplied
1001 * to unregister an existing handler.
1003 void t3_register_cpl_handler(unsigned int opcode, cpl_handler_func h)
1005 if (opcode < NUM_CPL_CMDS)
1006 cpl_handlers[opcode] = h ? h : do_bad_cpl;
1008 printk(KERN_ERR "T3C: handler registration for "
1009 "opcode %x failed\n", opcode);
1012 EXPORT_SYMBOL(t3_register_cpl_handler);
1015 * T3CDEV's receive method.
1017 int process_rx(struct t3cdev *dev, struct sk_buff **skbs, int n)
1020 struct sk_buff *skb = *skbs++;
1021 unsigned int opcode = get_opcode(skb);
1022 int ret = cpl_handlers[opcode] (dev, skb);
1025 if (ret & CPL_RET_UNKNOWN_TID) {
1026 union opcode_tid *p = cplhdr(skb);
1028 printk(KERN_ERR "%s: CPL message (opcode %u) had "
1029 "unknown TID %u\n", dev->name, opcode,
1030 G_TID(ntohl(p->opcode_tid)));
1033 if (ret & CPL_RET_BUF_DONE)
1040 * Sends an sk_buff to a T3C driver after dealing with any active network taps.
1042 int cxgb3_ofld_send(struct t3cdev *dev, struct sk_buff *skb)
1047 r = dev->send(dev, skb);
1052 EXPORT_SYMBOL(cxgb3_ofld_send);
1054 static int is_offloading(struct net_device *dev)
1056 struct adapter *adapter;
1059 read_lock_bh(&adapter_list_lock);
1060 list_for_each_entry(adapter, &adapter_list, adapter_list) {
1061 for_each_port(adapter, i) {
1062 if (dev == adapter->port[i]) {
1063 read_unlock_bh(&adapter_list_lock);
1068 read_unlock_bh(&adapter_list_lock);
1072 void cxgb_neigh_update(struct neighbour *neigh)
1074 struct net_device *dev = neigh->dev;
1076 if (dev && (is_offloading(dev))) {
1077 struct t3cdev *tdev = dev2t3cdev(dev);
1080 t3_l2t_update(tdev, neigh);
1084 static void set_l2t_ix(struct t3cdev *tdev, u32 tid, struct l2t_entry *e)
1086 struct sk_buff *skb;
1087 struct cpl_set_tcb_field *req;
1089 skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
1091 printk(KERN_ERR "%s: cannot allocate skb!\n", __func__);
1094 skb->priority = CPL_PRIORITY_CONTROL;
1095 req = (struct cpl_set_tcb_field *)skb_put(skb, sizeof(*req));
1096 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1097 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1100 req->word = htons(W_TCB_L2T_IX);
1101 req->mask = cpu_to_be64(V_TCB_L2T_IX(M_TCB_L2T_IX));
1102 req->val = cpu_to_be64(V_TCB_L2T_IX(e->idx));
1103 tdev->send(tdev, skb);
1106 void cxgb_redirect(struct dst_entry *old, struct dst_entry *new)
1108 struct net_device *olddev, *newdev;
1109 struct tid_info *ti;
1110 struct t3cdev *tdev;
1113 struct l2t_entry *e;
1114 struct t3c_tid_entry *te;
1116 olddev = old->neighbour->dev;
1117 newdev = new->neighbour->dev;
1118 if (!is_offloading(olddev))
1120 if (!is_offloading(newdev)) {
1121 printk(KERN_WARNING "%s: Redirect to non-offload "
1122 "device ignored.\n", __func__);
1125 tdev = dev2t3cdev(olddev);
1127 if (tdev != dev2t3cdev(newdev)) {
1128 printk(KERN_WARNING "%s: Redirect to different "
1129 "offload device ignored.\n", __func__);
1133 /* Add new L2T entry */
1134 e = t3_l2t_get(tdev, new->neighbour, newdev);
1136 printk(KERN_ERR "%s: couldn't allocate new l2t entry!\n",
1141 /* Walk tid table and notify clients of dst change. */
1142 ti = &(T3C_DATA(tdev))->tid_maps;
1143 for (tid = 0; tid < ti->ntids; tid++) {
1144 te = lookup_tid(ti, tid);
1146 if (te && te->ctx && te->client && te->client->redirect) {
1147 update_tcb = te->client->redirect(te->ctx, old, new, e);
1149 l2t_hold(L2DATA(tdev), e);
1150 set_l2t_ix(tdev, tid, e);
1154 l2t_release(L2DATA(tdev), e);
1158 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1159 * The allocated memory is cleared.
1161 void *cxgb_alloc_mem(unsigned long size)
1163 void *p = kmalloc(size, GFP_KERNEL);
1173 * Free memory allocated through t3_alloc_mem().
1175 void cxgb_free_mem(void *addr)
1177 if (is_vmalloc_addr(addr))
1184 * Allocate and initialize the TID tables. Returns 0 on success.
1186 static int init_tid_tabs(struct tid_info *t, unsigned int ntids,
1187 unsigned int natids, unsigned int nstids,
1188 unsigned int atid_base, unsigned int stid_base)
1190 unsigned long size = ntids * sizeof(*t->tid_tab) +
1191 natids * sizeof(*t->atid_tab) + nstids * sizeof(*t->stid_tab);
1193 t->tid_tab = cxgb_alloc_mem(size);
1197 t->stid_tab = (union listen_entry *)&t->tid_tab[ntids];
1198 t->atid_tab = (union active_open_entry *)&t->stid_tab[nstids];
1201 t->stid_base = stid_base;
1204 t->atid_base = atid_base;
1206 t->stids_in_use = t->atids_in_use = 0;
1207 atomic_set(&t->tids_in_use, 0);
1208 spin_lock_init(&t->stid_lock);
1209 spin_lock_init(&t->atid_lock);
1212 * Setup the free lists for stid_tab and atid_tab.
1216 t->stid_tab[nstids - 1].next = &t->stid_tab[nstids];
1217 t->sfree = t->stid_tab;
1221 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1222 t->afree = t->atid_tab;
1227 static void free_tid_maps(struct tid_info *t)
1229 cxgb_free_mem(t->tid_tab);
1232 static inline void add_adapter(struct adapter *adap)
1234 write_lock_bh(&adapter_list_lock);
1235 list_add_tail(&adap->adapter_list, &adapter_list);
1236 write_unlock_bh(&adapter_list_lock);
1239 static inline void remove_adapter(struct adapter *adap)
1241 write_lock_bh(&adapter_list_lock);
1242 list_del(&adap->adapter_list);
1243 write_unlock_bh(&adapter_list_lock);
1246 int cxgb3_offload_activate(struct adapter *adapter)
1248 struct t3cdev *dev = &adapter->tdev;
1251 struct tid_range stid_range, tid_range;
1252 struct mtutab mtutab;
1253 unsigned int l2t_capacity;
1255 t = kcalloc(1, sizeof(*t), GFP_KERNEL);
1260 if (dev->ctl(dev, GET_TX_MAX_CHUNK, &t->tx_max_chunk) < 0 ||
1261 dev->ctl(dev, GET_MAX_OUTSTANDING_WR, &t->max_wrs) < 0 ||
1262 dev->ctl(dev, GET_L2T_CAPACITY, &l2t_capacity) < 0 ||
1263 dev->ctl(dev, GET_MTUS, &mtutab) < 0 ||
1264 dev->ctl(dev, GET_TID_RANGE, &tid_range) < 0 ||
1265 dev->ctl(dev, GET_STID_RANGE, &stid_range) < 0)
1269 L2DATA(dev) = t3_init_l2t(l2t_capacity);
1273 natids = min(tid_range.num / 2, MAX_ATIDS);
1274 err = init_tid_tabs(&t->tid_maps, tid_range.num, natids,
1275 stid_range.num, ATID_BASE, stid_range.base);
1279 t->mtus = mtutab.mtus;
1280 t->nmtus = mtutab.size;
1282 INIT_WORK(&t->tid_release_task, t3_process_tid_release_list);
1283 spin_lock_init(&t->tid_release_lock);
1284 INIT_LIST_HEAD(&t->list_node);
1288 dev->recv = process_rx;
1289 dev->neigh_update = t3_l2t_update;
1291 /* Register netevent handler once */
1292 if (list_empty(&adapter_list))
1293 register_netevent_notifier(&nb);
1295 t->nofail_skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_KERNEL);
1296 t->release_list_incomplete = 0;
1298 add_adapter(adapter);
1302 t3_free_l2t(L2DATA(dev));
1309 void cxgb3_offload_deactivate(struct adapter *adapter)
1311 struct t3cdev *tdev = &adapter->tdev;
1312 struct t3c_data *t = T3C_DATA(tdev);
1314 remove_adapter(adapter);
1315 if (list_empty(&adapter_list))
1316 unregister_netevent_notifier(&nb);
1318 free_tid_maps(&t->tid_maps);
1319 T3C_DATA(tdev) = NULL;
1320 t3_free_l2t(L2DATA(tdev));
1321 L2DATA(tdev) = NULL;
1323 kfree_skb(t->nofail_skb);
1327 static inline void register_tdev(struct t3cdev *tdev)
1331 mutex_lock(&cxgb3_db_lock);
1332 snprintf(tdev->name, sizeof(tdev->name), "ofld_dev%d", unit++);
1333 list_add_tail(&tdev->ofld_dev_list, &ofld_dev_list);
1334 mutex_unlock(&cxgb3_db_lock);
1337 static inline void unregister_tdev(struct t3cdev *tdev)
1339 mutex_lock(&cxgb3_db_lock);
1340 list_del(&tdev->ofld_dev_list);
1341 mutex_unlock(&cxgb3_db_lock);
1344 static inline int adap2type(struct adapter *adapter)
1348 switch (adapter->params.rev) {
1363 void __devinit cxgb3_adapter_ofld(struct adapter *adapter)
1365 struct t3cdev *tdev = &adapter->tdev;
1367 INIT_LIST_HEAD(&tdev->ofld_dev_list);
1369 cxgb3_set_dummy_ops(tdev);
1370 tdev->send = t3_offload_tx;
1371 tdev->ctl = cxgb_offload_ctl;
1372 tdev->type = adap2type(adapter);
1374 register_tdev(tdev);
1377 void __devexit cxgb3_adapter_unofld(struct adapter *adapter)
1379 struct t3cdev *tdev = &adapter->tdev;
1382 tdev->neigh_update = NULL;
1384 unregister_tdev(tdev);
1387 void __init cxgb3_offload_init(void)
1391 for (i = 0; i < NUM_CPL_CMDS; ++i)
1392 cpl_handlers[i] = do_bad_cpl;
1394 t3_register_cpl_handler(CPL_SMT_WRITE_RPL, do_smt_write_rpl);
1395 t3_register_cpl_handler(CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
1396 t3_register_cpl_handler(CPL_RTE_WRITE_RPL, do_rte_write_rpl);
1397 t3_register_cpl_handler(CPL_PASS_OPEN_RPL, do_stid_rpl);
1398 t3_register_cpl_handler(CPL_CLOSE_LISTSRV_RPL, do_stid_rpl);
1399 t3_register_cpl_handler(CPL_PASS_ACCEPT_REQ, do_cr);
1400 t3_register_cpl_handler(CPL_PASS_ESTABLISH, do_hwtid_rpl);
1401 t3_register_cpl_handler(CPL_ABORT_RPL_RSS, do_hwtid_rpl);
1402 t3_register_cpl_handler(CPL_ABORT_RPL, do_hwtid_rpl);
1403 t3_register_cpl_handler(CPL_RX_URG_NOTIFY, do_hwtid_rpl);
1404 t3_register_cpl_handler(CPL_RX_DATA, do_hwtid_rpl);
1405 t3_register_cpl_handler(CPL_TX_DATA_ACK, do_hwtid_rpl);
1406 t3_register_cpl_handler(CPL_TX_DMA_ACK, do_hwtid_rpl);
1407 t3_register_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl);
1408 t3_register_cpl_handler(CPL_PEER_CLOSE, do_hwtid_rpl);
1409 t3_register_cpl_handler(CPL_CLOSE_CON_RPL, do_hwtid_rpl);
1410 t3_register_cpl_handler(CPL_ABORT_REQ_RSS, do_abort_req_rss);
1411 t3_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish);
1412 t3_register_cpl_handler(CPL_SET_TCB_RPL, do_hwtid_rpl);
1413 t3_register_cpl_handler(CPL_GET_TCB_RPL, do_hwtid_rpl);
1414 t3_register_cpl_handler(CPL_RDMA_TERMINATE, do_term);
1415 t3_register_cpl_handler(CPL_RDMA_EC_STATUS, do_hwtid_rpl);
1416 t3_register_cpl_handler(CPL_TRACE_PKT, do_trace);
1417 t3_register_cpl_handler(CPL_RX_DATA_DDP, do_hwtid_rpl);
1418 t3_register_cpl_handler(CPL_RX_DDP_COMPLETE, do_hwtid_rpl);
1419 t3_register_cpl_handler(CPL_ISCSI_HDR, do_hwtid_rpl);