2 * This file is part of the Chelsio T4 PCI-E SR-IOV Virtual Function Ethernet
5 * Copyright (c) 2009-2010 Chelsio Communications, Inc. All rights reserved.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #include <linux/version.h>
37 #include <linux/module.h>
38 #include <linux/moduleparam.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/netdevice.h>
43 #include <linux/etherdevice.h>
44 #include <linux/debugfs.h>
45 #include <linux/ethtool.h>
47 #include "t4vf_common.h"
48 #include "t4vf_defs.h"
50 #include "../cxgb4/t4_regs.h"
51 #include "../cxgb4/t4_msg.h"
54 * Generic information about the driver.
56 #define DRV_VERSION "1.0.0"
57 #define DRV_DESC "Chelsio T4 Virtual Function (VF) Network Driver"
65 * Default ethtool "message level" for adapters.
67 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
68 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
69 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
71 static int dflt_msg_enable
= DFLT_MSG_ENABLE
;
73 module_param(dflt_msg_enable
, int, 0644);
74 MODULE_PARM_DESC(dflt_msg_enable
,
75 "default adapter ethtool message level bitmap");
78 * The driver uses the best interrupt scheme available on a platform in the
79 * order MSI-X then MSI. This parameter determines which of these schemes the
80 * driver may consider as follows:
82 * msi = 2: choose from among MSI-X and MSI
83 * msi = 1: only consider MSI interrupts
85 * Note that unlike the Physical Function driver, this Virtual Function driver
86 * does _not_ support legacy INTx interrupts (this limitation is mandated by
87 * the PCI-E SR-IOV standard).
91 #define MSI_DEFAULT MSI_MSIX
93 static int msi
= MSI_DEFAULT
;
95 module_param(msi
, int, 0644);
96 MODULE_PARM_DESC(msi
, "whether to use MSI-X or MSI");
99 * Fundamental constants.
100 * ======================
104 MAX_TXQ_ENTRIES
= 16384,
105 MAX_RSPQ_ENTRIES
= 16384,
106 MAX_RX_BUFFERS
= 16384,
108 MIN_TXQ_ENTRIES
= 32,
109 MIN_RSPQ_ENTRIES
= 128,
113 * For purposes of manipulating the Free List size we need to
114 * recognize that Free Lists are actually Egress Queues (the host
115 * produces free buffers which the hardware consumes), Egress Queues
116 * indices are all in units of Egress Context Units bytes, and free
117 * list entries are 64-bit PCI DMA addresses. And since the state of
118 * the Producer Index == the Consumer Index implies an EMPTY list, we
119 * always have at least one Egress Unit's worth of Free List entries
120 * unused. See sge.c for more details ...
122 EQ_UNIT
= SGE_EQ_IDXSIZE
,
123 FL_PER_EQ_UNIT
= EQ_UNIT
/ sizeof(__be64
),
124 MIN_FL_RESID
= FL_PER_EQ_UNIT
,
128 * Global driver state.
129 * ====================
132 static struct dentry
*cxgb4vf_debugfs_root
;
135 * OS "Callback" functions.
136 * ========================
140 * The link status has changed on the indicated "port" (Virtual Interface).
142 void t4vf_os_link_changed(struct adapter
*adapter
, int pidx
, int link_ok
)
144 struct net_device
*dev
= adapter
->port
[pidx
];
147 * If the port is disabled or the current recorded "link up"
148 * status matches the new status, just return.
150 if (!netif_running(dev
) || link_ok
== netif_carrier_ok(dev
))
154 * Tell the OS that the link status has changed and print a short
155 * informative message on the console about the event.
160 const struct port_info
*pi
= netdev_priv(dev
);
162 netif_carrier_on(dev
);
164 switch (pi
->link_cfg
.speed
) {
182 switch (pi
->link_cfg
.fc
) {
191 case PAUSE_RX
|PAUSE_TX
:
200 printk(KERN_INFO
"%s: link up, %s, full-duplex, %s PAUSE\n",
203 netif_carrier_off(dev
);
204 printk(KERN_INFO
"%s: link down\n", dev
->name
);
209 * Net device operations.
210 * ======================
214 * Record our new VLAN Group and enable/disable hardware VLAN Tag extraction
215 * based on whether the specified VLAN Group pointer is NULL or not.
217 static void cxgb4vf_vlan_rx_register(struct net_device
*dev
,
218 struct vlan_group
*grp
)
220 struct port_info
*pi
= netdev_priv(dev
);
223 t4vf_set_rxmode(pi
->adapter
, pi
->viid
, -1, -1, -1, -1, grp
!= NULL
, 0);
227 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
230 static int link_start(struct net_device
*dev
)
233 struct port_info
*pi
= netdev_priv(dev
);
236 * We do not set address filters and promiscuity here, the stack does
237 * that step explicitly.
239 ret
= t4vf_set_rxmode(pi
->adapter
, pi
->viid
, dev
->mtu
, -1, -1, -1, -1,
242 ret
= t4vf_change_mac(pi
->adapter
, pi
->viid
,
243 pi
->xact_addr_filt
, dev
->dev_addr
, true);
245 pi
->xact_addr_filt
= ret
;
251 * We don't need to actually "start the link" itself since the
252 * firmware will do that for us when the first Virtual Interface
253 * is enabled on a port.
256 ret
= t4vf_enable_vi(pi
->adapter
, pi
->viid
, true, true);
261 * Name the MSI-X interrupts.
263 static void name_msix_vecs(struct adapter
*adapter
)
265 int namelen
= sizeof(adapter
->msix_info
[0].desc
) - 1;
271 snprintf(adapter
->msix_info
[MSIX_FW
].desc
, namelen
,
272 "%s-FWeventq", adapter
->name
);
273 adapter
->msix_info
[MSIX_FW
].desc
[namelen
] = 0;
278 for_each_port(adapter
, pidx
) {
279 struct net_device
*dev
= adapter
->port
[pidx
];
280 const struct port_info
*pi
= netdev_priv(dev
);
283 for (qs
= 0, msi
= MSIX_NIQFLINT
;
286 snprintf(adapter
->msix_info
[msi
].desc
, namelen
,
287 "%s-%d", dev
->name
, qs
);
288 adapter
->msix_info
[msi
].desc
[namelen
] = 0;
294 * Request all of our MSI-X resources.
296 static int request_msix_queue_irqs(struct adapter
*adapter
)
298 struct sge
*s
= &adapter
->sge
;
304 err
= request_irq(adapter
->msix_info
[MSIX_FW
].vec
, t4vf_sge_intr_msix
,
305 0, adapter
->msix_info
[MSIX_FW
].desc
, &s
->fw_evtq
);
313 for_each_ethrxq(s
, rxq
) {
314 err
= request_irq(adapter
->msix_info
[msi
].vec
,
315 t4vf_sge_intr_msix
, 0,
316 adapter
->msix_info
[msi
].desc
,
317 &s
->ethrxq
[rxq
].rspq
);
326 free_irq(adapter
->msix_info
[--msi
].vec
, &s
->ethrxq
[rxq
].rspq
);
327 free_irq(adapter
->msix_info
[MSIX_FW
].vec
, &s
->fw_evtq
);
332 * Free our MSI-X resources.
334 static void free_msix_queue_irqs(struct adapter
*adapter
)
336 struct sge
*s
= &adapter
->sge
;
339 free_irq(adapter
->msix_info
[MSIX_FW
].vec
, &s
->fw_evtq
);
341 for_each_ethrxq(s
, rxq
)
342 free_irq(adapter
->msix_info
[msi
++].vec
,
343 &s
->ethrxq
[rxq
].rspq
);
347 * Turn on NAPI and start up interrupts on a response queue.
349 static void qenable(struct sge_rspq
*rspq
)
351 napi_enable(&rspq
->napi
);
354 * 0-increment the Going To Sleep register to start the timer and
357 t4_write_reg(rspq
->adapter
, T4VF_SGE_BASE_ADDR
+ SGE_VF_GTS
,
359 SEINTARM(rspq
->intr_params
) |
360 INGRESSQID(rspq
->cntxt_id
));
364 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
366 static void enable_rx(struct adapter
*adapter
)
369 struct sge
*s
= &adapter
->sge
;
371 for_each_ethrxq(s
, rxq
)
372 qenable(&s
->ethrxq
[rxq
].rspq
);
373 qenable(&s
->fw_evtq
);
376 * The interrupt queue doesn't use NAPI so we do the 0-increment of
377 * its Going To Sleep register here to get it started.
379 if (adapter
->flags
& USING_MSI
)
380 t4_write_reg(adapter
, T4VF_SGE_BASE_ADDR
+ SGE_VF_GTS
,
382 SEINTARM(s
->intrq
.intr_params
) |
383 INGRESSQID(s
->intrq
.cntxt_id
));
388 * Wait until all NAPI handlers are descheduled.
390 static void quiesce_rx(struct adapter
*adapter
)
392 struct sge
*s
= &adapter
->sge
;
395 for_each_ethrxq(s
, rxq
)
396 napi_disable(&s
->ethrxq
[rxq
].rspq
.napi
);
397 napi_disable(&s
->fw_evtq
.napi
);
401 * Response queue handler for the firmware event queue.
403 static int fwevtq_handler(struct sge_rspq
*rspq
, const __be64
*rsp
,
404 const struct pkt_gl
*gl
)
407 * Extract response opcode and get pointer to CPL message body.
409 struct adapter
*adapter
= rspq
->adapter
;
410 u8 opcode
= ((const struct rss_header
*)rsp
)->opcode
;
411 void *cpl
= (void *)(rsp
+ 1);
416 * We've received an asynchronous message from the firmware.
418 const struct cpl_fw6_msg
*fw_msg
= cpl
;
419 if (fw_msg
->type
== FW6_TYPE_CMD_RPL
)
420 t4vf_handle_fw_rpl(adapter
, fw_msg
->data
);
424 case CPL_SGE_EGR_UPDATE
: {
426 * We've received an Egress Queue Status Update message. We
427 * get these, if the SGE is configured to send these when the
428 * firmware passes certain points in processing our TX
429 * Ethernet Queue or if we make an explicit request for one.
430 * We use these updates to determine when we may need to
431 * restart a TX Ethernet Queue which was stopped for lack of
432 * free TX Queue Descriptors ...
434 const struct cpl_sge_egr_update
*p
= (void *)cpl
;
435 unsigned int qid
= EGR_QID(be32_to_cpu(p
->opcode_qid
));
436 struct sge
*s
= &adapter
->sge
;
438 struct sge_eth_txq
*txq
;
442 * Perform sanity checking on the Queue ID to make sure it
443 * really refers to one of our TX Ethernet Egress Queues which
444 * is active and matches the queue's ID. None of these error
445 * conditions should ever happen so we may want to either make
446 * them fatal and/or conditionalized under DEBUG.
448 eq_idx
= EQ_IDX(s
, qid
);
449 if (unlikely(eq_idx
>= MAX_EGRQ
)) {
450 dev_err(adapter
->pdev_dev
,
451 "Egress Update QID %d out of range\n", qid
);
454 tq
= s
->egr_map
[eq_idx
];
455 if (unlikely(tq
== NULL
)) {
456 dev_err(adapter
->pdev_dev
,
457 "Egress Update QID %d TXQ=NULL\n", qid
);
460 txq
= container_of(tq
, struct sge_eth_txq
, q
);
461 if (unlikely(tq
->abs_id
!= qid
)) {
462 dev_err(adapter
->pdev_dev
,
463 "Egress Update QID %d refers to TXQ %d\n",
469 * Restart a stopped TX Queue which has less than half of its
473 netif_tx_wake_queue(txq
->txq
);
478 dev_err(adapter
->pdev_dev
,
479 "unexpected CPL %#x on FW event queue\n", opcode
);
486 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
487 * to use and initializes them. We support multiple "Queue Sets" per port if
488 * we have MSI-X, otherwise just one queue set per port.
490 static int setup_sge_queues(struct adapter
*adapter
)
492 struct sge
*s
= &adapter
->sge
;
496 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
499 bitmap_zero(s
->starving_fl
, MAX_EGRQ
);
502 * If we're using MSI interrupt mode we need to set up a "forwarded
503 * interrupt" queue which we'll set up with our MSI vector. The rest
504 * of the ingress queues will be set up to forward their interrupts to
505 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
506 * the intrq's queue ID as the interrupt forwarding queue for the
507 * subsequent calls ...
509 if (adapter
->flags
& USING_MSI
) {
510 err
= t4vf_sge_alloc_rxq(adapter
, &s
->intrq
, false,
511 adapter
->port
[0], 0, NULL
, NULL
);
513 goto err_free_queues
;
517 * Allocate our ingress queue for asynchronous firmware messages.
519 err
= t4vf_sge_alloc_rxq(adapter
, &s
->fw_evtq
, true, adapter
->port
[0],
520 MSIX_FW
, NULL
, fwevtq_handler
);
522 goto err_free_queues
;
525 * Allocate each "port"'s initial Queue Sets. These can be changed
526 * later on ... up to the point where any interface on the adapter is
527 * brought up at which point lots of things get nailed down
530 msix
= MSIX_NIQFLINT
;
531 for_each_port(adapter
, pidx
) {
532 struct net_device
*dev
= adapter
->port
[pidx
];
533 struct port_info
*pi
= netdev_priv(dev
);
534 struct sge_eth_rxq
*rxq
= &s
->ethrxq
[pi
->first_qset
];
535 struct sge_eth_txq
*txq
= &s
->ethtxq
[pi
->first_qset
];
538 for (qs
= 0; qs
< pi
->nqsets
; qs
++, rxq
++, txq
++) {
539 err
= t4vf_sge_alloc_rxq(adapter
, &rxq
->rspq
, false,
541 &rxq
->fl
, t4vf_ethrx_handler
);
543 goto err_free_queues
;
545 err
= t4vf_sge_alloc_eth_txq(adapter
, txq
, dev
,
546 netdev_get_tx_queue(dev
, qs
),
547 s
->fw_evtq
.cntxt_id
);
549 goto err_free_queues
;
552 memset(&rxq
->stats
, 0, sizeof(rxq
->stats
));
557 * Create the reverse mappings for the queues.
559 s
->egr_base
= s
->ethtxq
[0].q
.abs_id
- s
->ethtxq
[0].q
.cntxt_id
;
560 s
->ingr_base
= s
->ethrxq
[0].rspq
.abs_id
- s
->ethrxq
[0].rspq
.cntxt_id
;
561 IQ_MAP(s
, s
->fw_evtq
.abs_id
) = &s
->fw_evtq
;
562 for_each_port(adapter
, pidx
) {
563 struct net_device
*dev
= adapter
->port
[pidx
];
564 struct port_info
*pi
= netdev_priv(dev
);
565 struct sge_eth_rxq
*rxq
= &s
->ethrxq
[pi
->first_qset
];
566 struct sge_eth_txq
*txq
= &s
->ethtxq
[pi
->first_qset
];
569 for (qs
= 0; qs
< pi
->nqsets
; qs
++, rxq
++, txq
++) {
570 IQ_MAP(s
, rxq
->rspq
.abs_id
) = &rxq
->rspq
;
571 EQ_MAP(s
, txq
->q
.abs_id
) = &txq
->q
;
574 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
575 * for Free Lists but since all of the Egress Queues
576 * (including Free Lists) have Relative Queue IDs
577 * which are computed as Absolute - Base Queue ID, we
578 * can synthesize the Absolute Queue IDs for the Free
579 * Lists. This is useful for debugging purposes when
580 * we want to dump Queue Contexts via the PF Driver.
582 rxq
->fl
.abs_id
= rxq
->fl
.cntxt_id
+ s
->egr_base
;
583 EQ_MAP(s
, rxq
->fl
.abs_id
) = &rxq
->fl
;
589 t4vf_free_sge_resources(adapter
);
594 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
595 * queues. We configure the RSS CPU lookup table to distribute to the number
596 * of HW receive queues, and the response queue lookup table to narrow that
597 * down to the response queues actually configured for each "port" (Virtual
598 * Interface). We always configure the RSS mapping for all ports since the
599 * mapping table has plenty of entries.
601 static int setup_rss(struct adapter
*adapter
)
605 for_each_port(adapter
, pidx
) {
606 struct port_info
*pi
= adap2pinfo(adapter
, pidx
);
607 struct sge_eth_rxq
*rxq
= &adapter
->sge
.ethrxq
[pi
->first_qset
];
608 u16 rss
[MAX_PORT_QSETS
];
611 for (qs
= 0; qs
< pi
->nqsets
; qs
++)
612 rss
[qs
] = rxq
[qs
].rspq
.abs_id
;
614 err
= t4vf_config_rss_range(adapter
, pi
->viid
,
615 0, pi
->rss_size
, rss
, pi
->nqsets
);
620 * Perform Global RSS Mode-specific initialization.
622 switch (adapter
->params
.rss
.mode
) {
623 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL
:
625 * If Tunnel All Lookup isn't specified in the global
626 * RSS Configuration, then we need to specify a
627 * default Ingress Queue for any ingress packets which
628 * aren't hashed. We'll use our first ingress queue
631 if (!adapter
->params
.rss
.u
.basicvirtual
.tnlalllookup
) {
632 union rss_vi_config config
;
633 err
= t4vf_read_rss_vi_config(adapter
,
638 config
.basicvirtual
.defaultq
=
640 err
= t4vf_write_rss_vi_config(adapter
,
654 * Bring the adapter up. Called whenever we go from no "ports" open to having
655 * one open. This function performs the actions necessary to make an adapter
656 * operational, such as completing the initialization of HW modules, and
657 * enabling interrupts. Must be called with the rtnl lock held. (Note that
658 * this is called "cxgb_up" in the PF Driver.)
660 static int adapter_up(struct adapter
*adapter
)
665 * If this is the first time we've been called, perform basic
666 * adapter setup. Once we've done this, many of our adapter
667 * parameters can no longer be changed ...
669 if ((adapter
->flags
& FULL_INIT_DONE
) == 0) {
670 err
= setup_sge_queues(adapter
);
673 err
= setup_rss(adapter
);
675 t4vf_free_sge_resources(adapter
);
679 if (adapter
->flags
& USING_MSIX
)
680 name_msix_vecs(adapter
);
681 adapter
->flags
|= FULL_INIT_DONE
;
685 * Acquire our interrupt resources. We only support MSI-X and MSI.
687 BUG_ON((adapter
->flags
& (USING_MSIX
|USING_MSI
)) == 0);
688 if (adapter
->flags
& USING_MSIX
)
689 err
= request_msix_queue_irqs(adapter
);
691 err
= request_irq(adapter
->pdev
->irq
,
692 t4vf_intr_handler(adapter
), 0,
693 adapter
->name
, adapter
);
695 dev_err(adapter
->pdev_dev
, "request_irq failed, err %d\n",
701 * Enable NAPI ingress processing and return success.
704 t4vf_sge_start(adapter
);
709 * Bring the adapter down. Called whenever the last "port" (Virtual
710 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
713 static void adapter_down(struct adapter
*adapter
)
716 * Free interrupt resources.
718 if (adapter
->flags
& USING_MSIX
)
719 free_msix_queue_irqs(adapter
);
721 free_irq(adapter
->pdev
->irq
, adapter
);
724 * Wait for NAPI handlers to finish.
730 * Start up a net device.
732 static int cxgb4vf_open(struct net_device
*dev
)
735 struct port_info
*pi
= netdev_priv(dev
);
736 struct adapter
*adapter
= pi
->adapter
;
739 * If this is the first interface that we're opening on the "adapter",
740 * bring the "adapter" up now.
742 if (adapter
->open_device_map
== 0) {
743 err
= adapter_up(adapter
);
749 * Note that this interface is up and start everything up ...
751 netif_set_real_num_tx_queues(dev
, pi
->nqsets
);
752 err
= netif_set_real_num_rx_queues(dev
, pi
->nqsets
);
755 set_bit(pi
->port_id
, &adapter
->open_device_map
);
757 netif_tx_start_all_queues(dev
);
762 * Shut down a net device. This routine is called "cxgb_close" in the PF
765 static int cxgb4vf_stop(struct net_device
*dev
)
768 struct port_info
*pi
= netdev_priv(dev
);
769 struct adapter
*adapter
= pi
->adapter
;
771 netif_tx_stop_all_queues(dev
);
772 netif_carrier_off(dev
);
773 ret
= t4vf_enable_vi(adapter
, pi
->viid
, false, false);
774 pi
->link_cfg
.link_ok
= 0;
776 clear_bit(pi
->port_id
, &adapter
->open_device_map
);
777 if (adapter
->open_device_map
== 0)
778 adapter_down(adapter
);
783 * Translate our basic statistics into the standard "ifconfig" statistics.
785 static struct net_device_stats
*cxgb4vf_get_stats(struct net_device
*dev
)
787 struct t4vf_port_stats stats
;
788 struct port_info
*pi
= netdev2pinfo(dev
);
789 struct adapter
*adapter
= pi
->adapter
;
790 struct net_device_stats
*ns
= &dev
->stats
;
793 spin_lock(&adapter
->stats_lock
);
794 err
= t4vf_get_port_stats(adapter
, pi
->pidx
, &stats
);
795 spin_unlock(&adapter
->stats_lock
);
797 memset(ns
, 0, sizeof(*ns
));
801 ns
->tx_bytes
= (stats
.tx_bcast_bytes
+ stats
.tx_mcast_bytes
+
802 stats
.tx_ucast_bytes
+ stats
.tx_offload_bytes
);
803 ns
->tx_packets
= (stats
.tx_bcast_frames
+ stats
.tx_mcast_frames
+
804 stats
.tx_ucast_frames
+ stats
.tx_offload_frames
);
805 ns
->rx_bytes
= (stats
.rx_bcast_bytes
+ stats
.rx_mcast_bytes
+
806 stats
.rx_ucast_bytes
);
807 ns
->rx_packets
= (stats
.rx_bcast_frames
+ stats
.rx_mcast_frames
+
808 stats
.rx_ucast_frames
);
809 ns
->multicast
= stats
.rx_mcast_frames
;
810 ns
->tx_errors
= stats
.tx_drop_frames
;
811 ns
->rx_errors
= stats
.rx_err_frames
;
817 * Collect up to maxaddrs worth of a netdevice's unicast addresses into an
818 * array of addrss pointers and return the number collected.
820 static inline int collect_netdev_uc_list_addrs(const struct net_device
*dev
,
822 unsigned int maxaddrs
)
824 unsigned int naddr
= 0;
825 const struct netdev_hw_addr
*ha
;
827 for_each_dev_addr(dev
, ha
) {
828 addr
[naddr
++] = ha
->addr
;
829 if (naddr
>= maxaddrs
)
836 * Collect up to maxaddrs worth of a netdevice's multicast addresses into an
837 * array of addrss pointers and return the number collected.
839 static inline int collect_netdev_mc_list_addrs(const struct net_device
*dev
,
841 unsigned int maxaddrs
)
843 unsigned int naddr
= 0;
844 const struct netdev_hw_addr
*ha
;
846 netdev_for_each_mc_addr(ha
, dev
) {
847 addr
[naddr
++] = ha
->addr
;
848 if (naddr
>= maxaddrs
)
855 * Configure the exact and hash address filters to handle a port's multicast
856 * and secondary unicast MAC addresses.
858 static int set_addr_filters(const struct net_device
*dev
, bool sleep
)
866 const struct port_info
*pi
= netdev_priv(dev
);
868 /* first do the secondary unicast addresses */
869 naddr
= collect_netdev_uc_list_addrs(dev
, addr
, ARRAY_SIZE(addr
));
871 ret
= t4vf_alloc_mac_filt(pi
->adapter
, pi
->viid
, free
,
872 naddr
, addr
, filt_idx
, &uhash
, sleep
);
879 /* next set up the multicast addresses */
880 naddr
= collect_netdev_mc_list_addrs(dev
, addr
, ARRAY_SIZE(addr
));
882 ret
= t4vf_alloc_mac_filt(pi
->adapter
, pi
->viid
, free
,
883 naddr
, addr
, filt_idx
, &mhash
, sleep
);
888 return t4vf_set_addr_hash(pi
->adapter
, pi
->viid
, uhash
!= 0,
889 uhash
| mhash
, sleep
);
893 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
894 * If @mtu is -1 it is left unchanged.
896 static int set_rxmode(struct net_device
*dev
, int mtu
, bool sleep_ok
)
899 struct port_info
*pi
= netdev_priv(dev
);
901 ret
= set_addr_filters(dev
, sleep_ok
);
903 ret
= t4vf_set_rxmode(pi
->adapter
, pi
->viid
, -1,
904 (dev
->flags
& IFF_PROMISC
) != 0,
905 (dev
->flags
& IFF_ALLMULTI
) != 0,
911 * Set the current receive modes on the device.
913 static void cxgb4vf_set_rxmode(struct net_device
*dev
)
915 /* unfortunately we can't return errors to the stack */
916 set_rxmode(dev
, -1, false);
920 * Find the entry in the interrupt holdoff timer value array which comes
921 * closest to the specified interrupt holdoff value.
923 static int closest_timer(const struct sge
*s
, int us
)
925 int i
, timer_idx
= 0, min_delta
= INT_MAX
;
927 for (i
= 0; i
< ARRAY_SIZE(s
->timer_val
); i
++) {
928 int delta
= us
- s
->timer_val
[i
];
931 if (delta
< min_delta
) {
939 static int closest_thres(const struct sge
*s
, int thres
)
941 int i
, delta
, pktcnt_idx
= 0, min_delta
= INT_MAX
;
943 for (i
= 0; i
< ARRAY_SIZE(s
->counter_val
); i
++) {
944 delta
= thres
- s
->counter_val
[i
];
947 if (delta
< min_delta
) {
956 * Return a queue's interrupt hold-off time in us. 0 means no timer.
958 static unsigned int qtimer_val(const struct adapter
*adapter
,
959 const struct sge_rspq
*rspq
)
961 unsigned int timer_idx
= QINTR_TIMER_IDX_GET(rspq
->intr_params
);
963 return timer_idx
< SGE_NTIMERS
964 ? adapter
->sge
.timer_val
[timer_idx
]
969 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
970 * @adapter: the adapter
971 * @rspq: the RX response queue
972 * @us: the hold-off time in us, or 0 to disable timer
973 * @cnt: the hold-off packet count, or 0 to disable counter
975 * Sets an RX response queue's interrupt hold-off time and packet count.
976 * At least one of the two needs to be enabled for the queue to generate
979 static int set_rxq_intr_params(struct adapter
*adapter
, struct sge_rspq
*rspq
,
980 unsigned int us
, unsigned int cnt
)
982 unsigned int timer_idx
;
985 * If both the interrupt holdoff timer and count are specified as
986 * zero, default to a holdoff count of 1 ...
992 * If an interrupt holdoff count has been specified, then find the
993 * closest configured holdoff count and use that. If the response
994 * queue has already been created, then update its queue context
1001 pktcnt_idx
= closest_thres(&adapter
->sge
, cnt
);
1002 if (rspq
->desc
&& rspq
->pktcnt_idx
!= pktcnt_idx
) {
1003 v
= FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ
) |
1005 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH
) |
1006 FW_PARAMS_PARAM_YZ(rspq
->cntxt_id
);
1007 err
= t4vf_set_params(adapter
, 1, &v
, &pktcnt_idx
);
1011 rspq
->pktcnt_idx
= pktcnt_idx
;
1015 * Compute the closest holdoff timer index from the supplied holdoff
1018 timer_idx
= (us
== 0
1019 ? SGE_TIMER_RSTRT_CNTR
1020 : closest_timer(&adapter
->sge
, us
));
1023 * Update the response queue's interrupt coalescing parameters and
1026 rspq
->intr_params
= (QINTR_TIMER_IDX(timer_idx
) |
1027 (cnt
> 0 ? QINTR_CNT_EN
: 0));
1032 * Return a version number to identify the type of adapter. The scheme is:
1033 * - bits 0..9: chip version
1034 * - bits 10..15: chip revision
1036 static inline unsigned int mk_adap_vers(const struct adapter
*adapter
)
1039 * Chip version 4, revision 0x3f (cxgb4vf).
1041 return 4 | (0x3f << 10);
1045 * Execute the specified ioctl command.
1047 static int cxgb4vf_do_ioctl(struct net_device
*dev
, struct ifreq
*ifr
, int cmd
)
1053 * The VF Driver doesn't have access to any of the other
1054 * common Ethernet device ioctl()'s (like reading/writing
1055 * PHY registers, etc.
1066 * Change the device's MTU.
1068 static int cxgb4vf_change_mtu(struct net_device
*dev
, int new_mtu
)
1071 struct port_info
*pi
= netdev_priv(dev
);
1073 /* accommodate SACK */
1077 ret
= t4vf_set_rxmode(pi
->adapter
, pi
->viid
, new_mtu
,
1078 -1, -1, -1, -1, true);
1085 * Change the devices MAC address.
1087 static int cxgb4vf_set_mac_addr(struct net_device
*dev
, void *_addr
)
1090 struct sockaddr
*addr
= _addr
;
1091 struct port_info
*pi
= netdev_priv(dev
);
1093 if (!is_valid_ether_addr(addr
->sa_data
))
1096 ret
= t4vf_change_mac(pi
->adapter
, pi
->viid
, pi
->xact_addr_filt
,
1097 addr
->sa_data
, true);
1101 memcpy(dev
->dev_addr
, addr
->sa_data
, dev
->addr_len
);
1102 pi
->xact_addr_filt
= ret
;
1107 * Return a TX Queue on which to send the specified skb.
1109 static u16
cxgb4vf_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
1112 * XXX For now just use the default hash but we probably want to
1113 * XXX look at other possibilities ...
1115 return skb_tx_hash(dev
, skb
);
1118 #ifdef CONFIG_NET_POLL_CONTROLLER
1120 * Poll all of our receive queues. This is called outside of normal interrupt
1123 static void cxgb4vf_poll_controller(struct net_device
*dev
)
1125 struct port_info
*pi
= netdev_priv(dev
);
1126 struct adapter
*adapter
= pi
->adapter
;
1128 if (adapter
->flags
& USING_MSIX
) {
1129 struct sge_eth_rxq
*rxq
;
1132 rxq
= &adapter
->sge
.ethrxq
[pi
->first_qset
];
1133 for (nqsets
= pi
->nqsets
; nqsets
; nqsets
--) {
1134 t4vf_sge_intr_msix(0, &rxq
->rspq
);
1138 t4vf_intr_handler(adapter
)(0, adapter
);
1143 * Ethtool operations.
1144 * ===================
1146 * Note that we don't support any ethtool operations which change the physical
1147 * state of the port to which we're linked.
1151 * Return current port link settings.
1153 static int cxgb4vf_get_settings(struct net_device
*dev
,
1154 struct ethtool_cmd
*cmd
)
1156 const struct port_info
*pi
= netdev_priv(dev
);
1158 cmd
->supported
= pi
->link_cfg
.supported
;
1159 cmd
->advertising
= pi
->link_cfg
.advertising
;
1160 cmd
->speed
= netif_carrier_ok(dev
) ? pi
->link_cfg
.speed
: -1;
1161 cmd
->duplex
= DUPLEX_FULL
;
1163 cmd
->port
= (cmd
->supported
& SUPPORTED_TP
) ? PORT_TP
: PORT_FIBRE
;
1164 cmd
->phy_address
= pi
->port_id
;
1165 cmd
->transceiver
= XCVR_EXTERNAL
;
1166 cmd
->autoneg
= pi
->link_cfg
.autoneg
;
1173 * Return our driver information.
1175 static void cxgb4vf_get_drvinfo(struct net_device
*dev
,
1176 struct ethtool_drvinfo
*drvinfo
)
1178 struct adapter
*adapter
= netdev2adap(dev
);
1180 strcpy(drvinfo
->driver
, KBUILD_MODNAME
);
1181 strcpy(drvinfo
->version
, DRV_VERSION
);
1182 strcpy(drvinfo
->bus_info
, pci_name(to_pci_dev(dev
->dev
.parent
)));
1183 snprintf(drvinfo
->fw_version
, sizeof(drvinfo
->fw_version
),
1184 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1185 FW_HDR_FW_VER_MAJOR_GET(adapter
->params
.dev
.fwrev
),
1186 FW_HDR_FW_VER_MINOR_GET(adapter
->params
.dev
.fwrev
),
1187 FW_HDR_FW_VER_MICRO_GET(adapter
->params
.dev
.fwrev
),
1188 FW_HDR_FW_VER_BUILD_GET(adapter
->params
.dev
.fwrev
),
1189 FW_HDR_FW_VER_MAJOR_GET(adapter
->params
.dev
.tprev
),
1190 FW_HDR_FW_VER_MINOR_GET(adapter
->params
.dev
.tprev
),
1191 FW_HDR_FW_VER_MICRO_GET(adapter
->params
.dev
.tprev
),
1192 FW_HDR_FW_VER_BUILD_GET(adapter
->params
.dev
.tprev
));
1196 * Return current adapter message level.
1198 static u32
cxgb4vf_get_msglevel(struct net_device
*dev
)
1200 return netdev2adap(dev
)->msg_enable
;
1204 * Set current adapter message level.
1206 static void cxgb4vf_set_msglevel(struct net_device
*dev
, u32 msglevel
)
1208 netdev2adap(dev
)->msg_enable
= msglevel
;
1212 * Return the device's current Queue Set ring size parameters along with the
1213 * allowed maximum values. Since ethtool doesn't understand the concept of
1214 * multi-queue devices, we just return the current values associated with the
1217 static void cxgb4vf_get_ringparam(struct net_device
*dev
,
1218 struct ethtool_ringparam
*rp
)
1220 const struct port_info
*pi
= netdev_priv(dev
);
1221 const struct sge
*s
= &pi
->adapter
->sge
;
1223 rp
->rx_max_pending
= MAX_RX_BUFFERS
;
1224 rp
->rx_mini_max_pending
= MAX_RSPQ_ENTRIES
;
1225 rp
->rx_jumbo_max_pending
= 0;
1226 rp
->tx_max_pending
= MAX_TXQ_ENTRIES
;
1228 rp
->rx_pending
= s
->ethrxq
[pi
->first_qset
].fl
.size
- MIN_FL_RESID
;
1229 rp
->rx_mini_pending
= s
->ethrxq
[pi
->first_qset
].rspq
.size
;
1230 rp
->rx_jumbo_pending
= 0;
1231 rp
->tx_pending
= s
->ethtxq
[pi
->first_qset
].q
.size
;
1235 * Set the Queue Set ring size parameters for the device. Again, since
1236 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1237 * apply these new values across all of the Queue Sets associated with the
1238 * device -- after vetting them of course!
1240 static int cxgb4vf_set_ringparam(struct net_device
*dev
,
1241 struct ethtool_ringparam
*rp
)
1243 const struct port_info
*pi
= netdev_priv(dev
);
1244 struct adapter
*adapter
= pi
->adapter
;
1245 struct sge
*s
= &adapter
->sge
;
1248 if (rp
->rx_pending
> MAX_RX_BUFFERS
||
1249 rp
->rx_jumbo_pending
||
1250 rp
->tx_pending
> MAX_TXQ_ENTRIES
||
1251 rp
->rx_mini_pending
> MAX_RSPQ_ENTRIES
||
1252 rp
->rx_mini_pending
< MIN_RSPQ_ENTRIES
||
1253 rp
->rx_pending
< MIN_FL_ENTRIES
||
1254 rp
->tx_pending
< MIN_TXQ_ENTRIES
)
1257 if (adapter
->flags
& FULL_INIT_DONE
)
1260 for (qs
= pi
->first_qset
; qs
< pi
->first_qset
+ pi
->nqsets
; qs
++) {
1261 s
->ethrxq
[qs
].fl
.size
= rp
->rx_pending
+ MIN_FL_RESID
;
1262 s
->ethrxq
[qs
].rspq
.size
= rp
->rx_mini_pending
;
1263 s
->ethtxq
[qs
].q
.size
= rp
->tx_pending
;
1269 * Return the interrupt holdoff timer and count for the first Queue Set on the
1270 * device. Our extension ioctl() (the cxgbtool interface) allows the
1271 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1273 static int cxgb4vf_get_coalesce(struct net_device
*dev
,
1274 struct ethtool_coalesce
*coalesce
)
1276 const struct port_info
*pi
= netdev_priv(dev
);
1277 const struct adapter
*adapter
= pi
->adapter
;
1278 const struct sge_rspq
*rspq
= &adapter
->sge
.ethrxq
[pi
->first_qset
].rspq
;
1280 coalesce
->rx_coalesce_usecs
= qtimer_val(adapter
, rspq
);
1281 coalesce
->rx_max_coalesced_frames
=
1282 ((rspq
->intr_params
& QINTR_CNT_EN
)
1283 ? adapter
->sge
.counter_val
[rspq
->pktcnt_idx
]
1289 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1290 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1291 * the interrupt holdoff timer on any of the device's Queue Sets.
1293 static int cxgb4vf_set_coalesce(struct net_device
*dev
,
1294 struct ethtool_coalesce
*coalesce
)
1296 const struct port_info
*pi
= netdev_priv(dev
);
1297 struct adapter
*adapter
= pi
->adapter
;
1299 return set_rxq_intr_params(adapter
,
1300 &adapter
->sge
.ethrxq
[pi
->first_qset
].rspq
,
1301 coalesce
->rx_coalesce_usecs
,
1302 coalesce
->rx_max_coalesced_frames
);
1306 * Report current port link pause parameter settings.
1308 static void cxgb4vf_get_pauseparam(struct net_device
*dev
,
1309 struct ethtool_pauseparam
*pauseparam
)
1311 struct port_info
*pi
= netdev_priv(dev
);
1313 pauseparam
->autoneg
= (pi
->link_cfg
.requested_fc
& PAUSE_AUTONEG
) != 0;
1314 pauseparam
->rx_pause
= (pi
->link_cfg
.fc
& PAUSE_RX
) != 0;
1315 pauseparam
->tx_pause
= (pi
->link_cfg
.fc
& PAUSE_TX
) != 0;
1319 * Return whether RX Checksum Offloading is currently enabled for the device.
1321 static u32
cxgb4vf_get_rx_csum(struct net_device
*dev
)
1323 struct port_info
*pi
= netdev_priv(dev
);
1325 return (pi
->rx_offload
& RX_CSO
) != 0;
1329 * Turn RX Checksum Offloading on or off for the device.
1331 static int cxgb4vf_set_rx_csum(struct net_device
*dev
, u32 csum
)
1333 struct port_info
*pi
= netdev_priv(dev
);
1336 pi
->rx_offload
|= RX_CSO
;
1338 pi
->rx_offload
&= ~RX_CSO
;
1343 * Identify the port by blinking the port's LED.
1345 static int cxgb4vf_phys_id(struct net_device
*dev
, u32 id
)
1347 struct port_info
*pi
= netdev_priv(dev
);
1349 return t4vf_identify_port(pi
->adapter
, pi
->viid
, 5);
1353 * Port stats maintained per queue of the port.
1355 struct queue_port_stats
{
1364 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1365 * these need to match the order of statistics returned by
1366 * t4vf_get_port_stats().
1368 static const char stats_strings
[][ETH_GSTRING_LEN
] = {
1370 * These must match the layout of the t4vf_port_stats structure.
1372 "TxBroadcastBytes ",
1373 "TxBroadcastFrames ",
1374 "TxMulticastBytes ",
1375 "TxMulticastFrames ",
1381 "RxBroadcastBytes ",
1382 "RxBroadcastFrames ",
1383 "RxMulticastBytes ",
1384 "RxMulticastFrames ",
1390 * These are accumulated per-queue statistics and must match the
1391 * order of the fields in the queue_port_stats structure.
1401 * Return the number of statistics in the specified statistics set.
1403 static int cxgb4vf_get_sset_count(struct net_device
*dev
, int sset
)
1407 return ARRAY_SIZE(stats_strings
);
1415 * Return the strings for the specified statistics set.
1417 static void cxgb4vf_get_strings(struct net_device
*dev
,
1423 memcpy(data
, stats_strings
, sizeof(stats_strings
));
1429 * Small utility routine to accumulate queue statistics across the queues of
1432 static void collect_sge_port_stats(const struct adapter
*adapter
,
1433 const struct port_info
*pi
,
1434 struct queue_port_stats
*stats
)
1436 const struct sge_eth_txq
*txq
= &adapter
->sge
.ethtxq
[pi
->first_qset
];
1437 const struct sge_eth_rxq
*rxq
= &adapter
->sge
.ethrxq
[pi
->first_qset
];
1440 memset(stats
, 0, sizeof(*stats
));
1441 for (qs
= 0; qs
< pi
->nqsets
; qs
++, rxq
++, txq
++) {
1442 stats
->tso
+= txq
->tso
;
1443 stats
->tx_csum
+= txq
->tx_cso
;
1444 stats
->rx_csum
+= rxq
->stats
.rx_cso
;
1445 stats
->vlan_ex
+= rxq
->stats
.vlan_ex
;
1446 stats
->vlan_ins
+= txq
->vlan_ins
;
1451 * Return the ETH_SS_STATS statistics set.
1453 static void cxgb4vf_get_ethtool_stats(struct net_device
*dev
,
1454 struct ethtool_stats
*stats
,
1457 struct port_info
*pi
= netdev2pinfo(dev
);
1458 struct adapter
*adapter
= pi
->adapter
;
1459 int err
= t4vf_get_port_stats(adapter
, pi
->pidx
,
1460 (struct t4vf_port_stats
*)data
);
1462 memset(data
, 0, sizeof(struct t4vf_port_stats
));
1464 data
+= sizeof(struct t4vf_port_stats
) / sizeof(u64
);
1465 collect_sge_port_stats(adapter
, pi
, (struct queue_port_stats
*)data
);
1469 * Return the size of our register map.
1471 static int cxgb4vf_get_regs_len(struct net_device
*dev
)
1473 return T4VF_REGMAP_SIZE
;
1477 * Dump a block of registers, start to end inclusive, into a buffer.
1479 static void reg_block_dump(struct adapter
*adapter
, void *regbuf
,
1480 unsigned int start
, unsigned int end
)
1482 u32
*bp
= regbuf
+ start
- T4VF_REGMAP_START
;
1484 for ( ; start
<= end
; start
+= sizeof(u32
)) {
1486 * Avoid reading the Mailbox Control register since that
1487 * can trigger a Mailbox Ownership Arbitration cycle and
1488 * interfere with communication with the firmware.
1490 if (start
== T4VF_CIM_BASE_ADDR
+ CIM_VF_EXT_MAILBOX_CTRL
)
1493 *bp
++ = t4_read_reg(adapter
, start
);
1498 * Copy our entire register map into the provided buffer.
1500 static void cxgb4vf_get_regs(struct net_device
*dev
,
1501 struct ethtool_regs
*regs
,
1504 struct adapter
*adapter
= netdev2adap(dev
);
1506 regs
->version
= mk_adap_vers(adapter
);
1509 * Fill in register buffer with our register map.
1511 memset(regbuf
, 0, T4VF_REGMAP_SIZE
);
1513 reg_block_dump(adapter
, regbuf
,
1514 T4VF_SGE_BASE_ADDR
+ T4VF_MOD_MAP_SGE_FIRST
,
1515 T4VF_SGE_BASE_ADDR
+ T4VF_MOD_MAP_SGE_LAST
);
1516 reg_block_dump(adapter
, regbuf
,
1517 T4VF_MPS_BASE_ADDR
+ T4VF_MOD_MAP_MPS_FIRST
,
1518 T4VF_MPS_BASE_ADDR
+ T4VF_MOD_MAP_MPS_LAST
);
1519 reg_block_dump(adapter
, regbuf
,
1520 T4VF_PL_BASE_ADDR
+ T4VF_MOD_MAP_PL_FIRST
,
1521 T4VF_PL_BASE_ADDR
+ T4VF_MOD_MAP_PL_LAST
);
1522 reg_block_dump(adapter
, regbuf
,
1523 T4VF_CIM_BASE_ADDR
+ T4VF_MOD_MAP_CIM_FIRST
,
1524 T4VF_CIM_BASE_ADDR
+ T4VF_MOD_MAP_CIM_LAST
);
1526 reg_block_dump(adapter
, regbuf
,
1527 T4VF_MBDATA_BASE_ADDR
+ T4VF_MBDATA_FIRST
,
1528 T4VF_MBDATA_BASE_ADDR
+ T4VF_MBDATA_LAST
);
1532 * Report current Wake On LAN settings.
1534 static void cxgb4vf_get_wol(struct net_device
*dev
,
1535 struct ethtool_wolinfo
*wol
)
1539 memset(&wol
->sopass
, 0, sizeof(wol
->sopass
));
1543 * Set TCP Segmentation Offloading feature capabilities.
1545 static int cxgb4vf_set_tso(struct net_device
*dev
, u32 tso
)
1548 dev
->features
|= NETIF_F_TSO
| NETIF_F_TSO6
;
1550 dev
->features
&= ~(NETIF_F_TSO
| NETIF_F_TSO6
);
1554 static struct ethtool_ops cxgb4vf_ethtool_ops
= {
1555 .get_settings
= cxgb4vf_get_settings
,
1556 .get_drvinfo
= cxgb4vf_get_drvinfo
,
1557 .get_msglevel
= cxgb4vf_get_msglevel
,
1558 .set_msglevel
= cxgb4vf_set_msglevel
,
1559 .get_ringparam
= cxgb4vf_get_ringparam
,
1560 .set_ringparam
= cxgb4vf_set_ringparam
,
1561 .get_coalesce
= cxgb4vf_get_coalesce
,
1562 .set_coalesce
= cxgb4vf_set_coalesce
,
1563 .get_pauseparam
= cxgb4vf_get_pauseparam
,
1564 .get_rx_csum
= cxgb4vf_get_rx_csum
,
1565 .set_rx_csum
= cxgb4vf_set_rx_csum
,
1566 .set_tx_csum
= ethtool_op_set_tx_ipv6_csum
,
1567 .set_sg
= ethtool_op_set_sg
,
1568 .get_link
= ethtool_op_get_link
,
1569 .get_strings
= cxgb4vf_get_strings
,
1570 .phys_id
= cxgb4vf_phys_id
,
1571 .get_sset_count
= cxgb4vf_get_sset_count
,
1572 .get_ethtool_stats
= cxgb4vf_get_ethtool_stats
,
1573 .get_regs_len
= cxgb4vf_get_regs_len
,
1574 .get_regs
= cxgb4vf_get_regs
,
1575 .get_wol
= cxgb4vf_get_wol
,
1576 .set_tso
= cxgb4vf_set_tso
,
1580 * /sys/kernel/debug/cxgb4vf support code and data.
1581 * ================================================
1585 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1589 static int sge_qinfo_show(struct seq_file
*seq
, void *v
)
1591 struct adapter
*adapter
= seq
->private;
1592 int eth_entries
= DIV_ROUND_UP(adapter
->sge
.ethqsets
, QPL
);
1593 int qs
, r
= (uintptr_t)v
- 1;
1596 seq_putc(seq
, '\n');
1598 #define S3(fmt_spec, s, v) \
1600 seq_printf(seq, "%-12s", s); \
1601 for (qs = 0; qs < n; ++qs) \
1602 seq_printf(seq, " %16" fmt_spec, v); \
1603 seq_putc(seq, '\n'); \
1605 #define S(s, v) S3("s", s, v)
1606 #define T(s, v) S3("u", s, txq[qs].v)
1607 #define R(s, v) S3("u", s, rxq[qs].v)
1609 if (r
< eth_entries
) {
1610 const struct sge_eth_rxq
*rxq
= &adapter
->sge
.ethrxq
[r
* QPL
];
1611 const struct sge_eth_txq
*txq
= &adapter
->sge
.ethtxq
[r
* QPL
];
1612 int n
= min(QPL
, adapter
->sge
.ethqsets
- QPL
* r
);
1614 S("QType:", "Ethernet");
1616 (rxq
[qs
].rspq
.netdev
1617 ? rxq
[qs
].rspq
.netdev
->name
1620 (rxq
[qs
].rspq
.netdev
1621 ? ((struct port_info
*)
1622 netdev_priv(rxq
[qs
].rspq
.netdev
))->port_id
1624 T("TxQ ID:", q
.abs_id
);
1625 T("TxQ size:", q
.size
);
1626 T("TxQ inuse:", q
.in_use
);
1627 T("TxQ PIdx:", q
.pidx
);
1628 T("TxQ CIdx:", q
.cidx
);
1629 R("RspQ ID:", rspq
.abs_id
);
1630 R("RspQ size:", rspq
.size
);
1631 R("RspQE size:", rspq
.iqe_len
);
1632 S3("u", "Intr delay:", qtimer_val(adapter
, &rxq
[qs
].rspq
));
1633 S3("u", "Intr pktcnt:",
1634 adapter
->sge
.counter_val
[rxq
[qs
].rspq
.pktcnt_idx
]);
1635 R("RspQ CIdx:", rspq
.cidx
);
1636 R("RspQ Gen:", rspq
.gen
);
1637 R("FL ID:", fl
.abs_id
);
1638 R("FL size:", fl
.size
- MIN_FL_RESID
);
1639 R("FL avail:", fl
.avail
);
1640 R("FL PIdx:", fl
.pidx
);
1641 R("FL CIdx:", fl
.cidx
);
1647 const struct sge_rspq
*evtq
= &adapter
->sge
.fw_evtq
;
1649 seq_printf(seq
, "%-12s %16s\n", "QType:", "FW event queue");
1650 seq_printf(seq
, "%-12s %16u\n", "RspQ ID:", evtq
->abs_id
);
1651 seq_printf(seq
, "%-12s %16u\n", "Intr delay:",
1652 qtimer_val(adapter
, evtq
));
1653 seq_printf(seq
, "%-12s %16u\n", "Intr pktcnt:",
1654 adapter
->sge
.counter_val
[evtq
->pktcnt_idx
]);
1655 seq_printf(seq
, "%-12s %16u\n", "RspQ Cidx:", evtq
->cidx
);
1656 seq_printf(seq
, "%-12s %16u\n", "RspQ Gen:", evtq
->gen
);
1657 } else if (r
== 1) {
1658 const struct sge_rspq
*intrq
= &adapter
->sge
.intrq
;
1660 seq_printf(seq
, "%-12s %16s\n", "QType:", "Interrupt Queue");
1661 seq_printf(seq
, "%-12s %16u\n", "RspQ ID:", intrq
->abs_id
);
1662 seq_printf(seq
, "%-12s %16u\n", "Intr delay:",
1663 qtimer_val(adapter
, intrq
));
1664 seq_printf(seq
, "%-12s %16u\n", "Intr pktcnt:",
1665 adapter
->sge
.counter_val
[intrq
->pktcnt_idx
]);
1666 seq_printf(seq
, "%-12s %16u\n", "RspQ Cidx:", intrq
->cidx
);
1667 seq_printf(seq
, "%-12s %16u\n", "RspQ Gen:", intrq
->gen
);
1679 * Return the number of "entries" in our "file". We group the multi-Queue
1680 * sections with QPL Queue Sets per "entry". The sections of the output are:
1682 * Ethernet RX/TX Queue Sets
1683 * Firmware Event Queue
1684 * Forwarded Interrupt Queue (if in MSI mode)
1686 static int sge_queue_entries(const struct adapter
*adapter
)
1688 return DIV_ROUND_UP(adapter
->sge
.ethqsets
, QPL
) + 1 +
1689 ((adapter
->flags
& USING_MSI
) != 0);
1692 static void *sge_queue_start(struct seq_file
*seq
, loff_t
*pos
)
1694 int entries
= sge_queue_entries(seq
->private);
1696 return *pos
< entries
? (void *)((uintptr_t)*pos
+ 1) : NULL
;
1699 static void sge_queue_stop(struct seq_file
*seq
, void *v
)
1703 static void *sge_queue_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1705 int entries
= sge_queue_entries(seq
->private);
1708 return *pos
< entries
? (void *)((uintptr_t)*pos
+ 1) : NULL
;
1711 static const struct seq_operations sge_qinfo_seq_ops
= {
1712 .start
= sge_queue_start
,
1713 .next
= sge_queue_next
,
1714 .stop
= sge_queue_stop
,
1715 .show
= sge_qinfo_show
1718 static int sge_qinfo_open(struct inode
*inode
, struct file
*file
)
1720 int res
= seq_open(file
, &sge_qinfo_seq_ops
);
1723 struct seq_file
*seq
= file
->private_data
;
1724 seq
->private = inode
->i_private
;
1729 static const struct file_operations sge_qinfo_debugfs_fops
= {
1730 .owner
= THIS_MODULE
,
1731 .open
= sge_qinfo_open
,
1733 .llseek
= seq_lseek
,
1734 .release
= seq_release
,
1738 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1742 static int sge_qstats_show(struct seq_file
*seq
, void *v
)
1744 struct adapter
*adapter
= seq
->private;
1745 int eth_entries
= DIV_ROUND_UP(adapter
->sge
.ethqsets
, QPL
);
1746 int qs
, r
= (uintptr_t)v
- 1;
1749 seq_putc(seq
, '\n');
1751 #define S3(fmt, s, v) \
1753 seq_printf(seq, "%-16s", s); \
1754 for (qs = 0; qs < n; ++qs) \
1755 seq_printf(seq, " %8" fmt, v); \
1756 seq_putc(seq, '\n'); \
1758 #define S(s, v) S3("s", s, v)
1760 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1761 #define T(s, v) T3("lu", s, v)
1763 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1764 #define R(s, v) R3("lu", s, v)
1766 if (r
< eth_entries
) {
1767 const struct sge_eth_rxq
*rxq
= &adapter
->sge
.ethrxq
[r
* QPL
];
1768 const struct sge_eth_txq
*txq
= &adapter
->sge
.ethtxq
[r
* QPL
];
1769 int n
= min(QPL
, adapter
->sge
.ethqsets
- QPL
* r
);
1771 S("QType:", "Ethernet");
1773 (rxq
[qs
].rspq
.netdev
1774 ? rxq
[qs
].rspq
.netdev
->name
1776 R3("u", "RspQNullInts:", rspq
.unhandled_irqs
);
1777 R("RxPackets:", stats
.pkts
);
1778 R("RxCSO:", stats
.rx_cso
);
1779 R("VLANxtract:", stats
.vlan_ex
);
1780 R("LROmerged:", stats
.lro_merged
);
1781 R("LROpackets:", stats
.lro_pkts
);
1782 R("RxDrops:", stats
.rx_drops
);
1784 T("TxCSO:", tx_cso
);
1785 T("VLANins:", vlan_ins
);
1786 T("TxQFull:", q
.stops
);
1787 T("TxQRestarts:", q
.restarts
);
1788 T("TxMapErr:", mapping_err
);
1789 R("FLAllocErr:", fl
.alloc_failed
);
1790 R("FLLrgAlcErr:", fl
.large_alloc_failed
);
1791 R("FLStarving:", fl
.starving
);
1797 const struct sge_rspq
*evtq
= &adapter
->sge
.fw_evtq
;
1799 seq_printf(seq
, "%-8s %16s\n", "QType:", "FW event queue");
1800 seq_printf(seq
, "%-16s %8u\n", "RspQNullInts:",
1801 evtq
->unhandled_irqs
);
1802 seq_printf(seq
, "%-16s %8u\n", "RspQ CIdx:", evtq
->cidx
);
1803 seq_printf(seq
, "%-16s %8u\n", "RspQ Gen:", evtq
->gen
);
1804 } else if (r
== 1) {
1805 const struct sge_rspq
*intrq
= &adapter
->sge
.intrq
;
1807 seq_printf(seq
, "%-8s %16s\n", "QType:", "Interrupt Queue");
1808 seq_printf(seq
, "%-16s %8u\n", "RspQNullInts:",
1809 intrq
->unhandled_irqs
);
1810 seq_printf(seq
, "%-16s %8u\n", "RspQ CIdx:", intrq
->cidx
);
1811 seq_printf(seq
, "%-16s %8u\n", "RspQ Gen:", intrq
->gen
);
1825 * Return the number of "entries" in our "file". We group the multi-Queue
1826 * sections with QPL Queue Sets per "entry". The sections of the output are:
1828 * Ethernet RX/TX Queue Sets
1829 * Firmware Event Queue
1830 * Forwarded Interrupt Queue (if in MSI mode)
1832 static int sge_qstats_entries(const struct adapter
*adapter
)
1834 return DIV_ROUND_UP(adapter
->sge
.ethqsets
, QPL
) + 1 +
1835 ((adapter
->flags
& USING_MSI
) != 0);
1838 static void *sge_qstats_start(struct seq_file
*seq
, loff_t
*pos
)
1840 int entries
= sge_qstats_entries(seq
->private);
1842 return *pos
< entries
? (void *)((uintptr_t)*pos
+ 1) : NULL
;
1845 static void sge_qstats_stop(struct seq_file
*seq
, void *v
)
1849 static void *sge_qstats_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1851 int entries
= sge_qstats_entries(seq
->private);
1854 return *pos
< entries
? (void *)((uintptr_t)*pos
+ 1) : NULL
;
1857 static const struct seq_operations sge_qstats_seq_ops
= {
1858 .start
= sge_qstats_start
,
1859 .next
= sge_qstats_next
,
1860 .stop
= sge_qstats_stop
,
1861 .show
= sge_qstats_show
1864 static int sge_qstats_open(struct inode
*inode
, struct file
*file
)
1866 int res
= seq_open(file
, &sge_qstats_seq_ops
);
1869 struct seq_file
*seq
= file
->private_data
;
1870 seq
->private = inode
->i_private
;
1875 static const struct file_operations sge_qstats_proc_fops
= {
1876 .owner
= THIS_MODULE
,
1877 .open
= sge_qstats_open
,
1879 .llseek
= seq_lseek
,
1880 .release
= seq_release
,
1884 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1886 static int resources_show(struct seq_file
*seq
, void *v
)
1888 struct adapter
*adapter
= seq
->private;
1889 struct vf_resources
*vfres
= &adapter
->params
.vfres
;
1891 #define S(desc, fmt, var) \
1892 seq_printf(seq, "%-60s " fmt "\n", \
1893 desc " (" #var "):", vfres->var)
1895 S("Virtual Interfaces", "%d", nvi
);
1896 S("Egress Queues", "%d", neq
);
1897 S("Ethernet Control", "%d", nethctrl
);
1898 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint
);
1899 S("Ingress Queues", "%d", niq
);
1900 S("Traffic Class", "%d", tc
);
1901 S("Port Access Rights Mask", "%#x", pmask
);
1902 S("MAC Address Filters", "%d", nexactf
);
1903 S("Firmware Command Read Capabilities", "%#x", r_caps
);
1904 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps
);
1911 static int resources_open(struct inode
*inode
, struct file
*file
)
1913 return single_open(file
, resources_show
, inode
->i_private
);
1916 static const struct file_operations resources_proc_fops
= {
1917 .owner
= THIS_MODULE
,
1918 .open
= resources_open
,
1920 .llseek
= seq_lseek
,
1921 .release
= single_release
,
1925 * Show Virtual Interfaces.
1927 static int interfaces_show(struct seq_file
*seq
, void *v
)
1929 if (v
== SEQ_START_TOKEN
) {
1930 seq_puts(seq
, "Interface Port VIID\n");
1932 struct adapter
*adapter
= seq
->private;
1933 int pidx
= (uintptr_t)v
- 2;
1934 struct net_device
*dev
= adapter
->port
[pidx
];
1935 struct port_info
*pi
= netdev_priv(dev
);
1937 seq_printf(seq
, "%9s %4d %#5x\n",
1938 dev
->name
, pi
->port_id
, pi
->viid
);
1943 static inline void *interfaces_get_idx(struct adapter
*adapter
, loff_t pos
)
1945 return pos
<= adapter
->params
.nports
1946 ? (void *)(uintptr_t)(pos
+ 1)
1950 static void *interfaces_start(struct seq_file
*seq
, loff_t
*pos
)
1953 ? interfaces_get_idx(seq
->private, *pos
)
1957 static void *interfaces_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1960 return interfaces_get_idx(seq
->private, *pos
);
1963 static void interfaces_stop(struct seq_file
*seq
, void *v
)
1967 static const struct seq_operations interfaces_seq_ops
= {
1968 .start
= interfaces_start
,
1969 .next
= interfaces_next
,
1970 .stop
= interfaces_stop
,
1971 .show
= interfaces_show
1974 static int interfaces_open(struct inode
*inode
, struct file
*file
)
1976 int res
= seq_open(file
, &interfaces_seq_ops
);
1979 struct seq_file
*seq
= file
->private_data
;
1980 seq
->private = inode
->i_private
;
1985 static const struct file_operations interfaces_proc_fops
= {
1986 .owner
= THIS_MODULE
,
1987 .open
= interfaces_open
,
1989 .llseek
= seq_lseek
,
1990 .release
= seq_release
,
1994 * /sys/kernel/debugfs/cxgb4vf/ files list.
1996 struct cxgb4vf_debugfs_entry
{
1997 const char *name
; /* name of debugfs node */
1998 mode_t mode
; /* file system mode */
1999 const struct file_operations
*fops
;
2002 static struct cxgb4vf_debugfs_entry debugfs_files
[] = {
2003 { "sge_qinfo", S_IRUGO
, &sge_qinfo_debugfs_fops
},
2004 { "sge_qstats", S_IRUGO
, &sge_qstats_proc_fops
},
2005 { "resources", S_IRUGO
, &resources_proc_fops
},
2006 { "interfaces", S_IRUGO
, &interfaces_proc_fops
},
2010 * Module and device initialization and cleanup code.
2011 * ==================================================
2015 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2016 * directory (debugfs_root) has already been set up.
2018 static int __devinit
setup_debugfs(struct adapter
*adapter
)
2022 BUG_ON(adapter
->debugfs_root
== NULL
);
2025 * Debugfs support is best effort.
2027 for (i
= 0; i
< ARRAY_SIZE(debugfs_files
); i
++)
2028 (void)debugfs_create_file(debugfs_files
[i
].name
,
2029 debugfs_files
[i
].mode
,
2030 adapter
->debugfs_root
,
2032 debugfs_files
[i
].fops
);
2038 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2039 * it to our caller to tear down the directory (debugfs_root).
2041 static void __devexit
cleanup_debugfs(struct adapter
*adapter
)
2043 BUG_ON(adapter
->debugfs_root
== NULL
);
2046 * Unlike our sister routine cleanup_proc(), we don't need to remove
2047 * individual entries because a call will be made to
2048 * debugfs_remove_recursive(). We just need to clean up any ancillary
2055 * Perform early "adapter" initialization. This is where we discover what
2056 * adapter parameters we're going to be using and initialize basic adapter
2059 static int adap_init0(struct adapter
*adapter
)
2061 struct vf_resources
*vfres
= &adapter
->params
.vfres
;
2062 struct sge_params
*sge_params
= &adapter
->params
.sge
;
2063 struct sge
*s
= &adapter
->sge
;
2064 unsigned int ethqsets
;
2068 * Wait for the device to become ready before proceeding ...
2070 err
= t4vf_wait_dev_ready(adapter
);
2072 dev_err(adapter
->pdev_dev
, "device didn't become ready:"
2078 * Grab basic operational parameters. These will predominantly have
2079 * been set up by the Physical Function Driver or will be hard coded
2080 * into the adapter. We just have to live with them ... Note that
2081 * we _must_ get our VPD parameters before our SGE parameters because
2082 * we need to know the adapter's core clock from the VPD in order to
2083 * properly decode the SGE Timer Values.
2085 err
= t4vf_get_dev_params(adapter
);
2087 dev_err(adapter
->pdev_dev
, "unable to retrieve adapter"
2088 " device parameters: err=%d\n", err
);
2091 err
= t4vf_get_vpd_params(adapter
);
2093 dev_err(adapter
->pdev_dev
, "unable to retrieve adapter"
2094 " VPD parameters: err=%d\n", err
);
2097 err
= t4vf_get_sge_params(adapter
);
2099 dev_err(adapter
->pdev_dev
, "unable to retrieve adapter"
2100 " SGE parameters: err=%d\n", err
);
2103 err
= t4vf_get_rss_glb_config(adapter
);
2105 dev_err(adapter
->pdev_dev
, "unable to retrieve adapter"
2106 " RSS parameters: err=%d\n", err
);
2109 if (adapter
->params
.rss
.mode
!=
2110 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL
) {
2111 dev_err(adapter
->pdev_dev
, "unable to operate with global RSS"
2112 " mode %d\n", adapter
->params
.rss
.mode
);
2115 err
= t4vf_sge_init(adapter
);
2117 dev_err(adapter
->pdev_dev
, "unable to use adapter parameters:"
2123 * Retrieve our RX interrupt holdoff timer values and counter
2124 * threshold values from the SGE parameters.
2126 s
->timer_val
[0] = core_ticks_to_us(adapter
,
2127 TIMERVALUE0_GET(sge_params
->sge_timer_value_0_and_1
));
2128 s
->timer_val
[1] = core_ticks_to_us(adapter
,
2129 TIMERVALUE1_GET(sge_params
->sge_timer_value_0_and_1
));
2130 s
->timer_val
[2] = core_ticks_to_us(adapter
,
2131 TIMERVALUE0_GET(sge_params
->sge_timer_value_2_and_3
));
2132 s
->timer_val
[3] = core_ticks_to_us(adapter
,
2133 TIMERVALUE1_GET(sge_params
->sge_timer_value_2_and_3
));
2134 s
->timer_val
[4] = core_ticks_to_us(adapter
,
2135 TIMERVALUE0_GET(sge_params
->sge_timer_value_4_and_5
));
2136 s
->timer_val
[5] = core_ticks_to_us(adapter
,
2137 TIMERVALUE1_GET(sge_params
->sge_timer_value_4_and_5
));
2140 THRESHOLD_0_GET(sge_params
->sge_ingress_rx_threshold
);
2142 THRESHOLD_1_GET(sge_params
->sge_ingress_rx_threshold
);
2144 THRESHOLD_2_GET(sge_params
->sge_ingress_rx_threshold
);
2146 THRESHOLD_3_GET(sge_params
->sge_ingress_rx_threshold
);
2149 * Grab our Virtual Interface resource allocation, extract the
2150 * features that we're interested in and do a bit of sanity testing on
2153 err
= t4vf_get_vfres(adapter
);
2155 dev_err(adapter
->pdev_dev
, "unable to get virtual interface"
2156 " resources: err=%d\n", err
);
2161 * The number of "ports" which we support is equal to the number of
2162 * Virtual Interfaces with which we've been provisioned.
2164 adapter
->params
.nports
= vfres
->nvi
;
2165 if (adapter
->params
.nports
> MAX_NPORTS
) {
2166 dev_warn(adapter
->pdev_dev
, "only using %d of %d allowed"
2167 " virtual interfaces\n", MAX_NPORTS
,
2168 adapter
->params
.nports
);
2169 adapter
->params
.nports
= MAX_NPORTS
;
2173 * We need to reserve a number of the ingress queues with Free List
2174 * and Interrupt capabilities for special interrupt purposes (like
2175 * asynchronous firmware messages, or forwarded interrupts if we're
2176 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2177 * matched up one-for-one with Ethernet/Control egress queues in order
2178 * to form "Queue Sets" which will be aportioned between the "ports".
2179 * For each Queue Set, we'll need the ability to allocate two Egress
2180 * Contexts -- one for the Ingress Queue Free List and one for the TX
2183 ethqsets
= vfres
->niqflint
- INGQ_EXTRAS
;
2184 if (vfres
->nethctrl
!= ethqsets
) {
2185 dev_warn(adapter
->pdev_dev
, "unequal number of [available]"
2186 " ingress/egress queues (%d/%d); using minimum for"
2187 " number of Queue Sets\n", ethqsets
, vfres
->nethctrl
);
2188 ethqsets
= min(vfres
->nethctrl
, ethqsets
);
2190 if (vfres
->neq
< ethqsets
*2) {
2191 dev_warn(adapter
->pdev_dev
, "Not enough Egress Contexts (%d)"
2192 " to support Queue Sets (%d); reducing allowed Queue"
2193 " Sets\n", vfres
->neq
, ethqsets
);
2194 ethqsets
= vfres
->neq
/2;
2196 if (ethqsets
> MAX_ETH_QSETS
) {
2197 dev_warn(adapter
->pdev_dev
, "only using %d of %d allowed Queue"
2198 " Sets\n", MAX_ETH_QSETS
, adapter
->sge
.max_ethqsets
);
2199 ethqsets
= MAX_ETH_QSETS
;
2201 if (vfres
->niq
!= 0 || vfres
->neq
> ethqsets
*2) {
2202 dev_warn(adapter
->pdev_dev
, "unused resources niq/neq (%d/%d)"
2203 " ignored\n", vfres
->niq
, vfres
->neq
- ethqsets
*2);
2205 adapter
->sge
.max_ethqsets
= ethqsets
;
2208 * Check for various parameter sanity issues. Most checks simply
2209 * result in us using fewer resources than our provissioning but we
2210 * do need at least one "port" with which to work ...
2212 if (adapter
->sge
.max_ethqsets
< adapter
->params
.nports
) {
2213 dev_warn(adapter
->pdev_dev
, "only using %d of %d available"
2214 " virtual interfaces (too few Queue Sets)\n",
2215 adapter
->sge
.max_ethqsets
, adapter
->params
.nports
);
2216 adapter
->params
.nports
= adapter
->sge
.max_ethqsets
;
2218 if (adapter
->params
.nports
== 0) {
2219 dev_err(adapter
->pdev_dev
, "no virtual interfaces configured/"
2226 static inline void init_rspq(struct sge_rspq
*rspq
, u8 timer_idx
,
2227 u8 pkt_cnt_idx
, unsigned int size
,
2228 unsigned int iqe_size
)
2230 rspq
->intr_params
= (QINTR_TIMER_IDX(timer_idx
) |
2231 (pkt_cnt_idx
< SGE_NCOUNTERS
? QINTR_CNT_EN
: 0));
2232 rspq
->pktcnt_idx
= (pkt_cnt_idx
< SGE_NCOUNTERS
2235 rspq
->iqe_len
= iqe_size
;
2240 * Perform default configuration of DMA queues depending on the number and
2241 * type of ports we found and the number of available CPUs. Most settings can
2242 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2243 * being brought up for the first time.
2245 static void __devinit
cfg_queues(struct adapter
*adapter
)
2247 struct sge
*s
= &adapter
->sge
;
2248 int q10g
, n10g
, qidx
, pidx
, qs
;
2251 * We should not be called till we know how many Queue Sets we can
2252 * support. In particular, this means that we need to know what kind
2253 * of interrupts we'll be using ...
2255 BUG_ON((adapter
->flags
& (USING_MSIX
|USING_MSI
)) == 0);
2258 * Count the number of 10GbE Virtual Interfaces that we have.
2261 for_each_port(adapter
, pidx
)
2262 n10g
+= is_10g_port(&adap2pinfo(adapter
, pidx
)->link_cfg
);
2265 * We default to 1 queue per non-10G port and up to # of cores queues
2271 int n1g
= (adapter
->params
.nports
- n10g
);
2272 q10g
= (adapter
->sge
.max_ethqsets
- n1g
) / n10g
;
2273 if (q10g
> num_online_cpus())
2274 q10g
= num_online_cpus();
2278 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2279 * The layout will be established in setup_sge_queues() when the
2280 * adapter is brough up for the first time.
2283 for_each_port(adapter
, pidx
) {
2284 struct port_info
*pi
= adap2pinfo(adapter
, pidx
);
2286 pi
->first_qset
= qidx
;
2287 pi
->nqsets
= is_10g_port(&pi
->link_cfg
) ? q10g
: 1;
2293 * Set up default Queue Set parameters ... Start off with the
2294 * shortest interrupt holdoff timer.
2296 for (qs
= 0; qs
< s
->max_ethqsets
; qs
++) {
2297 struct sge_eth_rxq
*rxq
= &s
->ethrxq
[qs
];
2298 struct sge_eth_txq
*txq
= &s
->ethtxq
[qs
];
2300 init_rspq(&rxq
->rspq
, 0, 0, 1024, L1_CACHE_BYTES
);
2306 * The firmware event queue is used for link state changes and
2307 * notifications of TX DMA completions.
2309 init_rspq(&s
->fw_evtq
, SGE_TIMER_RSTRT_CNTR
, 0, 512,
2313 * The forwarded interrupt queue is used when we're in MSI interrupt
2314 * mode. In this mode all interrupts associated with RX queues will
2315 * be forwarded to a single queue which we'll associate with our MSI
2316 * interrupt vector. The messages dropped in the forwarded interrupt
2317 * queue will indicate which ingress queue needs servicing ... This
2318 * queue needs to be large enough to accommodate all of the ingress
2319 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2320 * from equalling the CIDX if every ingress queue has an outstanding
2321 * interrupt). The queue doesn't need to be any larger because no
2322 * ingress queue will ever have more than one outstanding interrupt at
2325 init_rspq(&s
->intrq
, SGE_TIMER_RSTRT_CNTR
, 0, MSIX_ENTRIES
+ 1,
2330 * Reduce the number of Ethernet queues across all ports to at most n.
2331 * n provides at least one queue per port.
2333 static void __devinit
reduce_ethqs(struct adapter
*adapter
, int n
)
2336 struct port_info
*pi
;
2339 * While we have too many active Ether Queue Sets, interate across the
2340 * "ports" and reduce their individual Queue Set allocations.
2342 BUG_ON(n
< adapter
->params
.nports
);
2343 while (n
< adapter
->sge
.ethqsets
)
2344 for_each_port(adapter
, i
) {
2345 pi
= adap2pinfo(adapter
, i
);
2346 if (pi
->nqsets
> 1) {
2348 adapter
->sge
.ethqsets
--;
2349 if (adapter
->sge
.ethqsets
<= n
)
2355 * Reassign the starting Queue Sets for each of the "ports" ...
2358 for_each_port(adapter
, i
) {
2359 pi
= adap2pinfo(adapter
, i
);
2366 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2367 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2368 * need. Minimally we need one for every Virtual Interface plus those needed
2369 * for our "extras". Note that this process may lower the maximum number of
2370 * allowed Queue Sets ...
2372 static int __devinit
enable_msix(struct adapter
*adapter
)
2374 int i
, err
, want
, need
;
2375 struct msix_entry entries
[MSIX_ENTRIES
];
2376 struct sge
*s
= &adapter
->sge
;
2378 for (i
= 0; i
< MSIX_ENTRIES
; ++i
)
2379 entries
[i
].entry
= i
;
2382 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2383 * plus those needed for our "extras" (for example, the firmware
2384 * message queue). We _need_ at least one "Queue Set" per Virtual
2385 * Interface plus those needed for our "extras". So now we get to see
2386 * if the song is right ...
2388 want
= s
->max_ethqsets
+ MSIX_EXTRAS
;
2389 need
= adapter
->params
.nports
+ MSIX_EXTRAS
;
2390 while ((err
= pci_enable_msix(adapter
->pdev
, entries
, want
)) >= need
)
2394 int nqsets
= want
- MSIX_EXTRAS
;
2395 if (nqsets
< s
->max_ethqsets
) {
2396 dev_warn(adapter
->pdev_dev
, "only enough MSI-X vectors"
2397 " for %d Queue Sets\n", nqsets
);
2398 s
->max_ethqsets
= nqsets
;
2399 if (nqsets
< s
->ethqsets
)
2400 reduce_ethqs(adapter
, nqsets
);
2402 for (i
= 0; i
< want
; ++i
)
2403 adapter
->msix_info
[i
].vec
= entries
[i
].vector
;
2404 } else if (err
> 0) {
2405 pci_disable_msix(adapter
->pdev
);
2406 dev_info(adapter
->pdev_dev
, "only %d MSI-X vectors left,"
2407 " not using MSI-X\n", err
);
2412 #ifdef HAVE_NET_DEVICE_OPS
2413 static const struct net_device_ops cxgb4vf_netdev_ops
= {
2414 .ndo_open
= cxgb4vf_open
,
2415 .ndo_stop
= cxgb4vf_stop
,
2416 .ndo_start_xmit
= t4vf_eth_xmit
,
2417 .ndo_get_stats
= cxgb4vf_get_stats
,
2418 .ndo_set_rx_mode
= cxgb4vf_set_rxmode
,
2419 .ndo_set_mac_address
= cxgb4vf_set_mac_addr
,
2420 .ndo_select_queue
= cxgb4vf_select_queue
,
2421 .ndo_validate_addr
= eth_validate_addr
,
2422 .ndo_do_ioctl
= cxgb4vf_do_ioctl
,
2423 .ndo_change_mtu
= cxgb4vf_change_mtu
,
2424 .ndo_vlan_rx_register
= cxgb4vf_vlan_rx_register
,
2425 #ifdef CONFIG_NET_POLL_CONTROLLER
2426 .ndo_poll_controller
= cxgb4vf_poll_controller
,
2432 * "Probe" a device: initialize a device and construct all kernel and driver
2433 * state needed to manage the device. This routine is called "init_one" in
2436 static int __devinit
cxgb4vf_pci_probe(struct pci_dev
*pdev
,
2437 const struct pci_device_id
*ent
)
2439 static int version_printed
;
2444 struct adapter
*adapter
;
2445 struct port_info
*pi
;
2446 struct net_device
*netdev
;
2449 * Vet our module parameters.
2451 if (msi
!= MSI_MSIX
&& msi
!= MSI_MSI
) {
2452 dev_err(&pdev
->dev
, "bad module parameter msi=%d; must be %d"
2453 " (MSI-X or MSI) or %d (MSI)\n", msi
, MSI_MSIX
,
2460 * Print our driver banner the first time we're called to initialize a
2463 if (version_printed
== 0) {
2464 printk(KERN_INFO
"%s - version %s\n", DRV_DESC
, DRV_VERSION
);
2465 version_printed
= 1;
2470 * Initialize generic PCI device state.
2472 err
= pci_enable_device(pdev
);
2474 dev_err(&pdev
->dev
, "cannot enable PCI device\n");
2479 * Reserve PCI resources for the device. If we can't get them some
2480 * other driver may have already claimed the device ...
2482 err
= pci_request_regions(pdev
, KBUILD_MODNAME
);
2484 dev_err(&pdev
->dev
, "cannot obtain PCI resources\n");
2485 goto err_disable_device
;
2489 * Set up our DMA mask: try for 64-bit address masking first and
2490 * fall back to 32-bit if we can't get 64 bits ...
2492 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
2494 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
2496 dev_err(&pdev
->dev
, "unable to obtain 64-bit DMA for"
2497 " coherent allocations\n");
2498 goto err_release_regions
;
2502 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
2504 dev_err(&pdev
->dev
, "no usable DMA configuration\n");
2505 goto err_release_regions
;
2511 * Enable bus mastering for the device ...
2513 pci_set_master(pdev
);
2516 * Allocate our adapter data structure and attach it to the device.
2518 adapter
= kzalloc(sizeof(*adapter
), GFP_KERNEL
);
2521 goto err_release_regions
;
2523 pci_set_drvdata(pdev
, adapter
);
2524 adapter
->pdev
= pdev
;
2525 adapter
->pdev_dev
= &pdev
->dev
;
2528 * Initialize SMP data synchronization resources.
2530 spin_lock_init(&adapter
->stats_lock
);
2533 * Map our I/O registers in BAR0.
2535 adapter
->regs
= pci_ioremap_bar(pdev
, 0);
2536 if (!adapter
->regs
) {
2537 dev_err(&pdev
->dev
, "cannot map device registers\n");
2539 goto err_free_adapter
;
2543 * Initialize adapter level features.
2545 adapter
->name
= pci_name(pdev
);
2546 adapter
->msg_enable
= dflt_msg_enable
;
2547 err
= adap_init0(adapter
);
2552 * Allocate our "adapter ports" and stitch everything together.
2554 pmask
= adapter
->params
.vfres
.pmask
;
2555 for_each_port(adapter
, pidx
) {
2559 * We simplistically allocate our virtual interfaces
2560 * sequentially across the port numbers to which we have
2561 * access rights. This should be configurable in some manner
2566 port_id
= ffs(pmask
) - 1;
2567 pmask
&= ~(1 << port_id
);
2568 viid
= t4vf_alloc_vi(adapter
, port_id
);
2570 dev_err(&pdev
->dev
, "cannot allocate VI for port %d:"
2571 " err=%d\n", port_id
, viid
);
2577 * Allocate our network device and stitch things together.
2579 netdev
= alloc_etherdev_mq(sizeof(struct port_info
),
2581 if (netdev
== NULL
) {
2582 dev_err(&pdev
->dev
, "cannot allocate netdev for"
2583 " port %d\n", port_id
);
2584 t4vf_free_vi(adapter
, viid
);
2588 adapter
->port
[pidx
] = netdev
;
2589 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2590 pi
= netdev_priv(netdev
);
2591 pi
->adapter
= adapter
;
2593 pi
->port_id
= port_id
;
2597 * Initialize the starting state of our "port" and register
2600 pi
->xact_addr_filt
= -1;
2601 pi
->rx_offload
= RX_CSO
;
2602 netif_carrier_off(netdev
);
2603 netif_tx_stop_all_queues(netdev
);
2604 netdev
->irq
= pdev
->irq
;
2606 netdev
->features
= (NETIF_F_SG
| NETIF_F_TSO
| NETIF_F_TSO6
|
2607 NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
|
2608 NETIF_F_HW_VLAN_TX
| NETIF_F_HW_VLAN_RX
|
2611 netdev
->features
|= NETIF_F_HIGHDMA
;
2612 netdev
->vlan_features
=
2614 ~(NETIF_F_HW_VLAN_TX
| NETIF_F_HW_VLAN_RX
));
2616 #ifdef HAVE_NET_DEVICE_OPS
2617 netdev
->netdev_ops
= &cxgb4vf_netdev_ops
;
2619 netdev
->vlan_rx_register
= cxgb4vf_vlan_rx_register
;
2620 netdev
->open
= cxgb4vf_open
;
2621 netdev
->stop
= cxgb4vf_stop
;
2622 netdev
->hard_start_xmit
= t4vf_eth_xmit
;
2623 netdev
->get_stats
= cxgb4vf_get_stats
;
2624 netdev
->set_rx_mode
= cxgb4vf_set_rxmode
;
2625 netdev
->do_ioctl
= cxgb4vf_do_ioctl
;
2626 netdev
->change_mtu
= cxgb4vf_change_mtu
;
2627 netdev
->set_mac_address
= cxgb4vf_set_mac_addr
;
2628 netdev
->select_queue
= cxgb4vf_select_queue
;
2629 #ifdef CONFIG_NET_POLL_CONTROLLER
2630 netdev
->poll_controller
= cxgb4vf_poll_controller
;
2633 SET_ETHTOOL_OPS(netdev
, &cxgb4vf_ethtool_ops
);
2636 * Initialize the hardware/software state for the port.
2638 err
= t4vf_port_init(adapter
, pidx
);
2640 dev_err(&pdev
->dev
, "cannot initialize port %d\n",
2647 * The "card" is now ready to go. If any errors occur during device
2648 * registration we do not fail the whole "card" but rather proceed
2649 * only with the ports we manage to register successfully. However we
2650 * must register at least one net device.
2652 for_each_port(adapter
, pidx
) {
2653 netdev
= adapter
->port
[pidx
];
2657 err
= register_netdev(netdev
);
2659 dev_warn(&pdev
->dev
, "cannot register net device %s,"
2660 " skipping\n", netdev
->name
);
2664 set_bit(pidx
, &adapter
->registered_device_map
);
2666 if (adapter
->registered_device_map
== 0) {
2667 dev_err(&pdev
->dev
, "could not register any net devices\n");
2672 * Set up our debugfs entries.
2674 if (cxgb4vf_debugfs_root
) {
2675 adapter
->debugfs_root
=
2676 debugfs_create_dir(pci_name(pdev
),
2677 cxgb4vf_debugfs_root
);
2678 if (adapter
->debugfs_root
== NULL
)
2679 dev_warn(&pdev
->dev
, "could not create debugfs"
2682 setup_debugfs(adapter
);
2686 * See what interrupts we'll be using. If we've been configured to
2687 * use MSI-X interrupts, try to enable them but fall back to using
2688 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2689 * get MSI interrupts we bail with the error.
2691 if (msi
== MSI_MSIX
&& enable_msix(adapter
) == 0)
2692 adapter
->flags
|= USING_MSIX
;
2694 err
= pci_enable_msi(pdev
);
2696 dev_err(&pdev
->dev
, "Unable to allocate %s interrupts;"
2698 msi
== MSI_MSIX
? "MSI-X or MSI" : "MSI", err
);
2699 goto err_free_debugfs
;
2701 adapter
->flags
|= USING_MSI
;
2705 * Now that we know how many "ports" we have and what their types are,
2706 * and how many Queue Sets we can support, we can configure our queue
2709 cfg_queues(adapter
);
2712 * Print a short notice on the existance and configuration of the new
2713 * VF network device ...
2715 for_each_port(adapter
, pidx
) {
2716 dev_info(adapter
->pdev_dev
, "%s: Chelsio VF NIC PCIe %s\n",
2717 adapter
->port
[pidx
]->name
,
2718 (adapter
->flags
& USING_MSIX
) ? "MSI-X" :
2719 (adapter
->flags
& USING_MSI
) ? "MSI" : "");
2728 * Error recovery and exit code. Unwind state that's been created
2729 * so far and return the error.
2733 if (adapter
->debugfs_root
) {
2734 cleanup_debugfs(adapter
);
2735 debugfs_remove_recursive(adapter
->debugfs_root
);
2739 for_each_port(adapter
, pidx
) {
2740 netdev
= adapter
->port
[pidx
];
2743 pi
= netdev_priv(netdev
);
2744 t4vf_free_vi(adapter
, pi
->viid
);
2745 if (test_bit(pidx
, &adapter
->registered_device_map
))
2746 unregister_netdev(netdev
);
2747 free_netdev(netdev
);
2751 iounmap(adapter
->regs
);
2755 pci_set_drvdata(pdev
, NULL
);
2757 err_release_regions
:
2758 pci_release_regions(pdev
);
2759 pci_set_drvdata(pdev
, NULL
);
2760 pci_clear_master(pdev
);
2763 pci_disable_device(pdev
);
2770 * "Remove" a device: tear down all kernel and driver state created in the
2771 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2772 * that this is called "remove_one" in the PF Driver.)
2774 static void __devexit
cxgb4vf_pci_remove(struct pci_dev
*pdev
)
2776 struct adapter
*adapter
= pci_get_drvdata(pdev
);
2779 * Tear down driver state associated with device.
2785 * Stop all of our activity. Unregister network port,
2786 * disable interrupts, etc.
2788 for_each_port(adapter
, pidx
)
2789 if (test_bit(pidx
, &adapter
->registered_device_map
))
2790 unregister_netdev(adapter
->port
[pidx
]);
2791 t4vf_sge_stop(adapter
);
2792 if (adapter
->flags
& USING_MSIX
) {
2793 pci_disable_msix(adapter
->pdev
);
2794 adapter
->flags
&= ~USING_MSIX
;
2795 } else if (adapter
->flags
& USING_MSI
) {
2796 pci_disable_msi(adapter
->pdev
);
2797 adapter
->flags
&= ~USING_MSI
;
2801 * Tear down our debugfs entries.
2803 if (adapter
->debugfs_root
) {
2804 cleanup_debugfs(adapter
);
2805 debugfs_remove_recursive(adapter
->debugfs_root
);
2809 * Free all of the various resources which we've acquired ...
2811 t4vf_free_sge_resources(adapter
);
2812 for_each_port(adapter
, pidx
) {
2813 struct net_device
*netdev
= adapter
->port
[pidx
];
2814 struct port_info
*pi
;
2819 pi
= netdev_priv(netdev
);
2820 t4vf_free_vi(adapter
, pi
->viid
);
2821 free_netdev(netdev
);
2823 iounmap(adapter
->regs
);
2825 pci_set_drvdata(pdev
, NULL
);
2829 * Disable the device and release its PCI resources.
2831 pci_disable_device(pdev
);
2832 pci_clear_master(pdev
);
2833 pci_release_regions(pdev
);
2837 * PCI Device registration data structures.
2839 #define CH_DEVICE(devid, idx) \
2840 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
2842 static struct pci_device_id cxgb4vf_pci_tbl
[] = {
2843 CH_DEVICE(0xb000, 0), /* PE10K FPGA */
2844 CH_DEVICE(0x4800, 0), /* T440-dbg */
2845 CH_DEVICE(0x4801, 0), /* T420-cr */
2846 CH_DEVICE(0x4802, 0), /* T422-cr */
2850 MODULE_DESCRIPTION(DRV_DESC
);
2851 MODULE_AUTHOR("Chelsio Communications");
2852 MODULE_LICENSE("Dual BSD/GPL");
2853 MODULE_VERSION(DRV_VERSION
);
2854 MODULE_DEVICE_TABLE(pci
, cxgb4vf_pci_tbl
);
2856 static struct pci_driver cxgb4vf_driver
= {
2857 .name
= KBUILD_MODNAME
,
2858 .id_table
= cxgb4vf_pci_tbl
,
2859 .probe
= cxgb4vf_pci_probe
,
2860 .remove
= __devexit_p(cxgb4vf_pci_remove
),
2864 * Initialize global driver state.
2866 static int __init
cxgb4vf_module_init(void)
2870 /* Debugfs support is optional, just warn if this fails */
2871 cxgb4vf_debugfs_root
= debugfs_create_dir(KBUILD_MODNAME
, NULL
);
2872 if (!cxgb4vf_debugfs_root
)
2873 printk(KERN_WARNING KBUILD_MODNAME
": could not create"
2874 " debugfs entry, continuing\n");
2876 ret
= pci_register_driver(&cxgb4vf_driver
);
2878 debugfs_remove(cxgb4vf_debugfs_root
);
2883 * Tear down global driver state.
2885 static void __exit
cxgb4vf_module_exit(void)
2887 pci_unregister_driver(&cxgb4vf_driver
);
2888 debugfs_remove(cxgb4vf_debugfs_root
);
2891 module_init(cxgb4vf_module_init
);
2892 module_exit(cxgb4vf_module_exit
);