2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/bitmap.h>
38 #include <linux/crc32.h>
39 #include <linux/ctype.h>
40 #include <linux/debugfs.h>
41 #include <linux/err.h>
42 #include <linux/etherdevice.h>
43 #include <linux/firmware.h>
44 #include <linux/if_vlan.h>
45 #include <linux/init.h>
46 #include <linux/log2.h>
47 #include <linux/mdio.h>
48 #include <linux/module.h>
49 #include <linux/moduleparam.h>
50 #include <linux/mutex.h>
51 #include <linux/netdevice.h>
52 #include <linux/pci.h>
53 #include <linux/aer.h>
54 #include <linux/rtnetlink.h>
55 #include <linux/sched.h>
56 #include <linux/seq_file.h>
57 #include <linux/sockios.h>
58 #include <linux/vmalloc.h>
59 #include <linux/workqueue.h>
60 #include <net/neighbour.h>
61 #include <net/netevent.h>
62 #include <asm/uaccess.h>
70 #define DRV_VERSION "1.0.0-ko"
71 #define DRV_DESC "Chelsio T4 Network Driver"
74 * Max interrupt hold-off timer value in us. Queues fall back to this value
75 * under extreme memory pressure so it's largish to give the system time to
78 #define MAX_SGE_TIMERVAL 200U
82 * Virtual Function provisioning constants. We need two extra Ingress Queues
83 * with Interrupt capability to serve as the VF's Firmware Event Queue and
84 * Forwarded Interrupt Queue (when using MSI mode) -- neither will have Free
85 * Lists associated with them). For each Ethernet/Control Egress Queue and
86 * for each Free List, we need an Egress Context.
89 VFRES_NPORTS
= 1, /* # of "ports" per VF */
90 VFRES_NQSETS
= 2, /* # of "Queue Sets" per VF */
92 VFRES_NVI
= VFRES_NPORTS
, /* # of Virtual Interfaces */
93 VFRES_NETHCTRL
= VFRES_NQSETS
, /* # of EQs used for ETH or CTRL Qs */
94 VFRES_NIQFLINT
= VFRES_NQSETS
+2,/* # of ingress Qs/w Free List(s)/intr */
95 VFRES_NIQ
= 0, /* # of non-fl/int ingress queues */
96 VFRES_NEQ
= VFRES_NQSETS
*2, /* # of egress queues */
97 VFRES_TC
= 0, /* PCI-E traffic class */
98 VFRES_NEXACTF
= 16, /* # of exact MPS filters */
100 VFRES_R_CAPS
= FW_CMD_CAP_DMAQ
|FW_CMD_CAP_VF
|FW_CMD_CAP_PORT
,
101 VFRES_WX_CAPS
= FW_CMD_CAP_DMAQ
|FW_CMD_CAP_VF
,
105 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
106 * static and likely not to be useful in the long run. We really need to
107 * implement some form of persistent configuration which the firmware
110 static unsigned int pfvfres_pmask(struct adapter
*adapter
,
111 unsigned int pf
, unsigned int vf
)
113 unsigned int portn
, portvec
;
116 * Give PF's access to all of the ports.
119 return FW_PFVF_CMD_PMASK_MASK
;
122 * For VFs, we'll assign them access to the ports based purely on the
123 * PF. We assign active ports in order, wrapping around if there are
124 * fewer active ports than PFs: e.g. active port[pf % nports].
125 * Unfortunately the adapter's port_info structs haven't been
126 * initialized yet so we have to compute this.
128 if (adapter
->params
.nports
== 0)
131 portn
= pf
% adapter
->params
.nports
;
132 portvec
= adapter
->params
.portvec
;
135 * Isolate the lowest set bit in the port vector. If we're at
136 * the port number that we want, return that as the pmask.
137 * otherwise mask that bit out of the port vector and
138 * decrement our port number ...
140 unsigned int pmask
= portvec
^ (portvec
& (portvec
-1));
151 MEMWIN0_APERTURE
= 65536,
152 MEMWIN0_BASE
= 0x30000,
153 MEMWIN1_APERTURE
= 32768,
154 MEMWIN1_BASE
= 0x28000,
155 MEMWIN2_APERTURE
= 2048,
156 MEMWIN2_BASE
= 0x1b800,
160 MAX_TXQ_ENTRIES
= 16384,
161 MAX_CTRL_TXQ_ENTRIES
= 1024,
162 MAX_RSPQ_ENTRIES
= 16384,
163 MAX_RX_BUFFERS
= 16384,
164 MIN_TXQ_ENTRIES
= 32,
165 MIN_CTRL_TXQ_ENTRIES
= 32,
166 MIN_RSPQ_ENTRIES
= 128,
170 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
171 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
172 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
174 #define CH_DEVICE(devid) { PCI_VDEVICE(CHELSIO, devid), 0 }
176 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl
) = {
177 CH_DEVICE(0xa000), /* PE10K */
181 #define FW_FNAME "cxgb4/t4fw.bin"
183 MODULE_DESCRIPTION(DRV_DESC
);
184 MODULE_AUTHOR("Chelsio Communications");
185 MODULE_LICENSE("Dual BSD/GPL");
186 MODULE_VERSION(DRV_VERSION
);
187 MODULE_DEVICE_TABLE(pci
, cxgb4_pci_tbl
);
188 MODULE_FIRMWARE(FW_FNAME
);
190 static int dflt_msg_enable
= DFLT_MSG_ENABLE
;
192 module_param(dflt_msg_enable
, int, 0644);
193 MODULE_PARM_DESC(dflt_msg_enable
, "Chelsio T4 default message enable bitmap");
196 * The driver uses the best interrupt scheme available on a platform in the
197 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
198 * of these schemes the driver may consider as follows:
200 * msi = 2: choose from among all three options
201 * msi = 1: only consider MSI and INTx interrupts
202 * msi = 0: force INTx interrupts
206 module_param(msi
, int, 0644);
207 MODULE_PARM_DESC(msi
, "whether to use INTx (0), MSI (1) or MSI-X (2)");
210 * Queue interrupt hold-off timer values. Queues default to the first of these
213 static unsigned int intr_holdoff
[SGE_NTIMERS
- 1] = { 5, 10, 20, 50, 100 };
215 module_param_array(intr_holdoff
, uint
, NULL
, 0644);
216 MODULE_PARM_DESC(intr_holdoff
, "values for queue interrupt hold-off timers "
217 "0..4 in microseconds");
219 static unsigned int intr_cnt
[SGE_NCOUNTERS
- 1] = { 4, 8, 16 };
221 module_param_array(intr_cnt
, uint
, NULL
, 0644);
222 MODULE_PARM_DESC(intr_cnt
,
223 "thresholds 1..3 for queue interrupt packet counters");
227 #ifdef CONFIG_PCI_IOV
228 module_param(vf_acls
, bool, 0644);
229 MODULE_PARM_DESC(vf_acls
, "if set enable virtualization L2 ACL enforcement");
231 static unsigned int num_vf
[4];
233 module_param_array(num_vf
, uint
, NULL
, 0644);
234 MODULE_PARM_DESC(num_vf
, "number of VFs for each of PFs 0-3");
237 static struct dentry
*cxgb4_debugfs_root
;
239 static LIST_HEAD(adapter_list
);
240 static DEFINE_MUTEX(uld_mutex
);
241 static struct cxgb4_uld_info ulds
[CXGB4_ULD_MAX
];
242 static const char *uld_str
[] = { "RDMA", "iSCSI" };
244 static void link_report(struct net_device
*dev
)
246 if (!netif_carrier_ok(dev
))
247 netdev_info(dev
, "link down\n");
249 static const char *fc
[] = { "no", "Rx", "Tx", "Tx/Rx" };
251 const char *s
= "10Mbps";
252 const struct port_info
*p
= netdev_priv(dev
);
254 switch (p
->link_cfg
.speed
) {
266 netdev_info(dev
, "link up, %s, full-duplex, %s PAUSE\n", s
,
271 void t4_os_link_changed(struct adapter
*adapter
, int port_id
, int link_stat
)
273 struct net_device
*dev
= adapter
->port
[port_id
];
275 /* Skip changes from disabled ports. */
276 if (netif_running(dev
) && link_stat
!= netif_carrier_ok(dev
)) {
278 netif_carrier_on(dev
);
280 netif_carrier_off(dev
);
286 void t4_os_portmod_changed(const struct adapter
*adap
, int port_id
)
288 static const char *mod_str
[] = {
289 NULL
, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
292 const struct net_device
*dev
= adap
->port
[port_id
];
293 const struct port_info
*pi
= netdev_priv(dev
);
295 if (pi
->mod_type
== FW_PORT_MOD_TYPE_NONE
)
296 netdev_info(dev
, "port module unplugged\n");
297 else if (pi
->mod_type
< ARRAY_SIZE(mod_str
))
298 netdev_info(dev
, "%s module inserted\n", mod_str
[pi
->mod_type
]);
302 * Configure the exact and hash address filters to handle a port's multicast
303 * and secondary unicast MAC addresses.
305 static int set_addr_filters(const struct net_device
*dev
, bool sleep
)
313 const struct netdev_hw_addr
*ha
;
314 int uc_cnt
= netdev_uc_count(dev
);
315 int mc_cnt
= netdev_mc_count(dev
);
316 const struct port_info
*pi
= netdev_priv(dev
);
318 /* first do the secondary unicast addresses */
319 netdev_for_each_uc_addr(ha
, dev
) {
320 addr
[naddr
++] = ha
->addr
;
321 if (--uc_cnt
== 0 || naddr
>= ARRAY_SIZE(addr
)) {
322 ret
= t4_alloc_mac_filt(pi
->adapter
, 0, pi
->viid
, free
,
323 naddr
, addr
, filt_idx
, &uhash
, sleep
);
332 /* next set up the multicast addresses */
333 netdev_for_each_mc_addr(ha
, dev
) {
334 addr
[naddr
++] = ha
->addr
;
335 if (--mc_cnt
== 0 || naddr
>= ARRAY_SIZE(addr
)) {
336 ret
= t4_alloc_mac_filt(pi
->adapter
, 0, pi
->viid
, free
,
337 naddr
, addr
, filt_idx
, &mhash
, sleep
);
346 return t4_set_addr_hash(pi
->adapter
, 0, pi
->viid
, uhash
!= 0,
347 uhash
| mhash
, sleep
);
351 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
352 * If @mtu is -1 it is left unchanged.
354 static int set_rxmode(struct net_device
*dev
, int mtu
, bool sleep_ok
)
357 struct port_info
*pi
= netdev_priv(dev
);
359 ret
= set_addr_filters(dev
, sleep_ok
);
361 ret
= t4_set_rxmode(pi
->adapter
, 0, pi
->viid
, mtu
,
362 (dev
->flags
& IFF_PROMISC
) ? 1 : 0,
363 (dev
->flags
& IFF_ALLMULTI
) ? 1 : 0, 1, -1,
369 * link_start - enable a port
370 * @dev: the port to enable
372 * Performs the MAC and PHY actions needed to enable a port.
374 static int link_start(struct net_device
*dev
)
377 struct port_info
*pi
= netdev_priv(dev
);
380 * We do not set address filters and promiscuity here, the stack does
381 * that step explicitly.
383 ret
= t4_set_rxmode(pi
->adapter
, 0, pi
->viid
, dev
->mtu
, -1, -1, -1,
384 pi
->vlan_grp
!= NULL
, true);
386 ret
= t4_change_mac(pi
->adapter
, 0, pi
->viid
,
387 pi
->xact_addr_filt
, dev
->dev_addr
, true,
390 pi
->xact_addr_filt
= ret
;
395 ret
= t4_link_start(pi
->adapter
, 0, pi
->tx_chan
, &pi
->link_cfg
);
397 ret
= t4_enable_vi(pi
->adapter
, 0, pi
->viid
, true, true);
402 * Response queue handler for the FW event queue.
404 static int fwevtq_handler(struct sge_rspq
*q
, const __be64
*rsp
,
405 const struct pkt_gl
*gl
)
407 u8 opcode
= ((const struct rss_header
*)rsp
)->opcode
;
409 rsp
++; /* skip RSS header */
410 if (likely(opcode
== CPL_SGE_EGR_UPDATE
)) {
411 const struct cpl_sge_egr_update
*p
= (void *)rsp
;
412 unsigned int qid
= EGR_QID(ntohl(p
->opcode_qid
));
413 struct sge_txq
*txq
= q
->adap
->sge
.egr_map
[qid
];
416 if ((u8
*)txq
< (u8
*)q
->adap
->sge
.ethrxq
) {
417 struct sge_eth_txq
*eq
;
419 eq
= container_of(txq
, struct sge_eth_txq
, q
);
420 netif_tx_wake_queue(eq
->txq
);
422 struct sge_ofld_txq
*oq
;
424 oq
= container_of(txq
, struct sge_ofld_txq
, q
);
425 tasklet_schedule(&oq
->qresume_tsk
);
427 } else if (opcode
== CPL_FW6_MSG
|| opcode
== CPL_FW4_MSG
) {
428 const struct cpl_fw6_msg
*p
= (void *)rsp
;
431 t4_handle_fw_rpl(q
->adap
, p
->data
);
432 } else if (opcode
== CPL_L2T_WRITE_RPL
) {
433 const struct cpl_l2t_write_rpl
*p
= (void *)rsp
;
435 do_l2t_write_rpl(q
->adap
, p
);
437 dev_err(q
->adap
->pdev_dev
,
438 "unexpected CPL %#x on FW event queue\n", opcode
);
443 * uldrx_handler - response queue handler for ULD queues
444 * @q: the response queue that received the packet
445 * @rsp: the response queue descriptor holding the offload message
446 * @gl: the gather list of packet fragments
448 * Deliver an ingress offload packet to a ULD. All processing is done by
449 * the ULD, we just maintain statistics.
451 static int uldrx_handler(struct sge_rspq
*q
, const __be64
*rsp
,
452 const struct pkt_gl
*gl
)
454 struct sge_ofld_rxq
*rxq
= container_of(q
, struct sge_ofld_rxq
, rspq
);
456 if (ulds
[q
->uld
].rx_handler(q
->adap
->uld_handle
[q
->uld
], rsp
, gl
)) {
462 else if (gl
== CXGB4_MSG_AN
)
469 static void disable_msi(struct adapter
*adapter
)
471 if (adapter
->flags
& USING_MSIX
) {
472 pci_disable_msix(adapter
->pdev
);
473 adapter
->flags
&= ~USING_MSIX
;
474 } else if (adapter
->flags
& USING_MSI
) {
475 pci_disable_msi(adapter
->pdev
);
476 adapter
->flags
&= ~USING_MSI
;
481 * Interrupt handler for non-data events used with MSI-X.
483 static irqreturn_t
t4_nondata_intr(int irq
, void *cookie
)
485 struct adapter
*adap
= cookie
;
487 u32 v
= t4_read_reg(adap
, MYPF_REG(PL_PF_INT_CAUSE
));
490 t4_write_reg(adap
, MYPF_REG(PL_PF_INT_CAUSE
), v
);
492 t4_slow_intr_handler(adap
);
497 * Name the MSI-X interrupts.
499 static void name_msix_vecs(struct adapter
*adap
)
501 int i
, j
, msi_idx
= 2, n
= sizeof(adap
->msix_info
[0].desc
) - 1;
503 /* non-data interrupts */
504 snprintf(adap
->msix_info
[0].desc
, n
, "%s", adap
->name
);
505 adap
->msix_info
[0].desc
[n
] = 0;
508 snprintf(adap
->msix_info
[1].desc
, n
, "%s-FWeventq", adap
->name
);
509 adap
->msix_info
[1].desc
[n
] = 0;
511 /* Ethernet queues */
512 for_each_port(adap
, j
) {
513 struct net_device
*d
= adap
->port
[j
];
514 const struct port_info
*pi
= netdev_priv(d
);
516 for (i
= 0; i
< pi
->nqsets
; i
++, msi_idx
++) {
517 snprintf(adap
->msix_info
[msi_idx
].desc
, n
, "%s-Rx%d",
519 adap
->msix_info
[msi_idx
].desc
[n
] = 0;
524 for_each_ofldrxq(&adap
->sge
, i
) {
525 snprintf(adap
->msix_info
[msi_idx
].desc
, n
, "%s-ofld%d",
527 adap
->msix_info
[msi_idx
++].desc
[n
] = 0;
529 for_each_rdmarxq(&adap
->sge
, i
) {
530 snprintf(adap
->msix_info
[msi_idx
].desc
, n
, "%s-rdma%d",
532 adap
->msix_info
[msi_idx
++].desc
[n
] = 0;
536 static int request_msix_queue_irqs(struct adapter
*adap
)
538 struct sge
*s
= &adap
->sge
;
539 int err
, ethqidx
, ofldqidx
= 0, rdmaqidx
= 0, msi
= 2;
541 err
= request_irq(adap
->msix_info
[1].vec
, t4_sge_intr_msix
, 0,
542 adap
->msix_info
[1].desc
, &s
->fw_evtq
);
546 for_each_ethrxq(s
, ethqidx
) {
547 err
= request_irq(adap
->msix_info
[msi
].vec
, t4_sge_intr_msix
, 0,
548 adap
->msix_info
[msi
].desc
,
549 &s
->ethrxq
[ethqidx
].rspq
);
554 for_each_ofldrxq(s
, ofldqidx
) {
555 err
= request_irq(adap
->msix_info
[msi
].vec
, t4_sge_intr_msix
, 0,
556 adap
->msix_info
[msi
].desc
,
557 &s
->ofldrxq
[ofldqidx
].rspq
);
562 for_each_rdmarxq(s
, rdmaqidx
) {
563 err
= request_irq(adap
->msix_info
[msi
].vec
, t4_sge_intr_msix
, 0,
564 adap
->msix_info
[msi
].desc
,
565 &s
->rdmarxq
[rdmaqidx
].rspq
);
573 while (--rdmaqidx
>= 0)
574 free_irq(adap
->msix_info
[--msi
].vec
,
575 &s
->rdmarxq
[rdmaqidx
].rspq
);
576 while (--ofldqidx
>= 0)
577 free_irq(adap
->msix_info
[--msi
].vec
,
578 &s
->ofldrxq
[ofldqidx
].rspq
);
579 while (--ethqidx
>= 0)
580 free_irq(adap
->msix_info
[--msi
].vec
, &s
->ethrxq
[ethqidx
].rspq
);
581 free_irq(adap
->msix_info
[1].vec
, &s
->fw_evtq
);
585 static void free_msix_queue_irqs(struct adapter
*adap
)
588 struct sge
*s
= &adap
->sge
;
590 free_irq(adap
->msix_info
[1].vec
, &s
->fw_evtq
);
591 for_each_ethrxq(s
, i
)
592 free_irq(adap
->msix_info
[msi
++].vec
, &s
->ethrxq
[i
].rspq
);
593 for_each_ofldrxq(s
, i
)
594 free_irq(adap
->msix_info
[msi
++].vec
, &s
->ofldrxq
[i
].rspq
);
595 for_each_rdmarxq(s
, i
)
596 free_irq(adap
->msix_info
[msi
++].vec
, &s
->rdmarxq
[i
].rspq
);
600 * setup_rss - configure RSS
603 * Sets up RSS to distribute packets to multiple receive queues. We
604 * configure the RSS CPU lookup table to distribute to the number of HW
605 * receive queues, and the response queue lookup table to narrow that
606 * down to the response queues actually configured for each port.
607 * We always configure the RSS mapping for all ports since the mapping
608 * table has plenty of entries.
610 static int setup_rss(struct adapter
*adap
)
613 u16 rss
[MAX_ETH_QSETS
];
615 for_each_port(adap
, i
) {
616 const struct port_info
*pi
= adap2pinfo(adap
, i
);
617 const struct sge_eth_rxq
*q
= &adap
->sge
.ethrxq
[pi
->first_qset
];
619 for (j
= 0; j
< pi
->nqsets
; j
++)
620 rss
[j
] = q
[j
].rspq
.abs_id
;
622 err
= t4_config_rss_range(adap
, 0, pi
->viid
, 0, pi
->rss_size
,
631 * Wait until all NAPI handlers are descheduled.
633 static void quiesce_rx(struct adapter
*adap
)
637 for (i
= 0; i
< ARRAY_SIZE(adap
->sge
.ingr_map
); i
++) {
638 struct sge_rspq
*q
= adap
->sge
.ingr_map
[i
];
641 napi_disable(&q
->napi
);
646 * Enable NAPI scheduling and interrupt generation for all Rx queues.
648 static void enable_rx(struct adapter
*adap
)
652 for (i
= 0; i
< ARRAY_SIZE(adap
->sge
.ingr_map
); i
++) {
653 struct sge_rspq
*q
= adap
->sge
.ingr_map
[i
];
658 napi_enable(&q
->napi
);
659 /* 0-increment GTS to start the timer and enable interrupts */
660 t4_write_reg(adap
, MYPF_REG(SGE_PF_GTS
),
661 SEINTARM(q
->intr_params
) |
662 INGRESSQID(q
->cntxt_id
));
667 * setup_sge_queues - configure SGE Tx/Rx/response queues
670 * Determines how many sets of SGE queues to use and initializes them.
671 * We support multiple queue sets per port if we have MSI-X, otherwise
672 * just one queue set per port.
674 static int setup_sge_queues(struct adapter
*adap
)
676 int err
, msi_idx
, i
, j
;
677 struct sge
*s
= &adap
->sge
;
679 bitmap_zero(s
->starving_fl
, MAX_EGRQ
);
680 bitmap_zero(s
->txq_maperr
, MAX_EGRQ
);
682 if (adap
->flags
& USING_MSIX
)
683 msi_idx
= 1; /* vector 0 is for non-queue interrupts */
685 err
= t4_sge_alloc_rxq(adap
, &s
->intrq
, false, adap
->port
[0], 0,
689 msi_idx
= -((int)s
->intrq
.abs_id
+ 1);
692 err
= t4_sge_alloc_rxq(adap
, &s
->fw_evtq
, true, adap
->port
[0],
693 msi_idx
, NULL
, fwevtq_handler
);
695 freeout
: t4_free_sge_resources(adap
);
699 for_each_port(adap
, i
) {
700 struct net_device
*dev
= adap
->port
[i
];
701 struct port_info
*pi
= netdev_priv(dev
);
702 struct sge_eth_rxq
*q
= &s
->ethrxq
[pi
->first_qset
];
703 struct sge_eth_txq
*t
= &s
->ethtxq
[pi
->first_qset
];
705 for (j
= 0; j
< pi
->nqsets
; j
++, q
++) {
708 err
= t4_sge_alloc_rxq(adap
, &q
->rspq
, false, dev
,
714 memset(&q
->stats
, 0, sizeof(q
->stats
));
716 for (j
= 0; j
< pi
->nqsets
; j
++, t
++) {
717 err
= t4_sge_alloc_eth_txq(adap
, t
, dev
,
718 netdev_get_tx_queue(dev
, j
),
719 s
->fw_evtq
.cntxt_id
);
725 j
= s
->ofldqsets
/ adap
->params
.nports
; /* ofld queues per channel */
726 for_each_ofldrxq(s
, i
) {
727 struct sge_ofld_rxq
*q
= &s
->ofldrxq
[i
];
728 struct net_device
*dev
= adap
->port
[i
/ j
];
732 err
= t4_sge_alloc_rxq(adap
, &q
->rspq
, false, dev
, msi_idx
,
733 &q
->fl
, uldrx_handler
);
736 memset(&q
->stats
, 0, sizeof(q
->stats
));
737 s
->ofld_rxq
[i
] = q
->rspq
.abs_id
;
738 err
= t4_sge_alloc_ofld_txq(adap
, &s
->ofldtxq
[i
], dev
,
739 s
->fw_evtq
.cntxt_id
);
744 for_each_rdmarxq(s
, i
) {
745 struct sge_ofld_rxq
*q
= &s
->rdmarxq
[i
];
749 err
= t4_sge_alloc_rxq(adap
, &q
->rspq
, false, adap
->port
[i
],
750 msi_idx
, &q
->fl
, uldrx_handler
);
753 memset(&q
->stats
, 0, sizeof(q
->stats
));
754 s
->rdma_rxq
[i
] = q
->rspq
.abs_id
;
757 for_each_port(adap
, i
) {
759 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
760 * have RDMA queues, and that's the right value.
762 err
= t4_sge_alloc_ctrl_txq(adap
, &s
->ctrlq
[i
], adap
->port
[i
],
764 s
->rdmarxq
[i
].rspq
.cntxt_id
);
769 t4_write_reg(adap
, MPS_TRC_RSS_CONTROL
,
770 RSSCONTROL(netdev2pinfo(adap
->port
[0])->tx_chan
) |
771 QUEUENUMBER(s
->ethrxq
[0].rspq
.abs_id
));
776 * Returns 0 if new FW was successfully loaded, a positive errno if a load was
777 * started but failed, and a negative errno if flash load couldn't start.
779 static int upgrade_fw(struct adapter
*adap
)
783 const struct fw_hdr
*hdr
;
784 const struct firmware
*fw
;
785 struct device
*dev
= adap
->pdev_dev
;
787 ret
= request_firmware(&fw
, FW_FNAME
, dev
);
789 dev_err(dev
, "unable to load firmware image " FW_FNAME
790 ", error %d\n", ret
);
794 hdr
= (const struct fw_hdr
*)fw
->data
;
795 vers
= ntohl(hdr
->fw_ver
);
796 if (FW_HDR_FW_VER_MAJOR_GET(vers
) != FW_VERSION_MAJOR
) {
797 ret
= -EINVAL
; /* wrong major version, won't do */
802 * If the flash FW is unusable or we found something newer, load it.
804 if (FW_HDR_FW_VER_MAJOR_GET(adap
->params
.fw_vers
) != FW_VERSION_MAJOR
||
805 vers
> adap
->params
.fw_vers
) {
806 ret
= -t4_load_fw(adap
, fw
->data
, fw
->size
);
808 dev_info(dev
, "firmware upgraded to version %pI4 from "
809 FW_FNAME
"\n", &hdr
->fw_ver
);
811 out
: release_firmware(fw
);
816 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
817 * The allocated memory is cleared.
819 void *t4_alloc_mem(size_t size
)
821 void *p
= kmalloc(size
, GFP_KERNEL
);
831 * Free memory allocated through alloc_mem().
833 void t4_free_mem(void *addr
)
835 if (is_vmalloc_addr(addr
))
841 static inline int is_offload(const struct adapter
*adap
)
843 return adap
->params
.offload
;
847 * Implementation of ethtool operations.
850 static u32
get_msglevel(struct net_device
*dev
)
852 return netdev2adap(dev
)->msg_enable
;
855 static void set_msglevel(struct net_device
*dev
, u32 val
)
857 netdev2adap(dev
)->msg_enable
= val
;
860 static char stats_strings
[][ETH_GSTRING_LEN
] = {
863 "TxBroadcastFrames ",
864 "TxMulticastFrames ",
872 "TxFrames512To1023 ",
873 "TxFrames1024To1518 ",
874 "TxFrames1519ToMax ",
889 "RxBroadcastFrames ",
890 "RxMulticastFrames ",
904 "RxFrames512To1023 ",
905 "RxFrames1024To1518 ",
906 "RxFrames1519ToMax ",
918 "RxBG0FramesDropped ",
919 "RxBG1FramesDropped ",
920 "RxBG2FramesDropped ",
921 "RxBG3FramesDropped ",
936 static int get_sset_count(struct net_device
*dev
, int sset
)
940 return ARRAY_SIZE(stats_strings
);
946 #define T4_REGMAP_SIZE (160 * 1024)
948 static int get_regs_len(struct net_device
*dev
)
950 return T4_REGMAP_SIZE
;
953 static int get_eeprom_len(struct net_device
*dev
)
958 static void get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
960 struct adapter
*adapter
= netdev2adap(dev
);
962 strcpy(info
->driver
, KBUILD_MODNAME
);
963 strcpy(info
->version
, DRV_VERSION
);
964 strcpy(info
->bus_info
, pci_name(adapter
->pdev
));
966 if (!adapter
->params
.fw_vers
)
967 strcpy(info
->fw_version
, "N/A");
969 snprintf(info
->fw_version
, sizeof(info
->fw_version
),
970 "%u.%u.%u.%u, TP %u.%u.%u.%u",
971 FW_HDR_FW_VER_MAJOR_GET(adapter
->params
.fw_vers
),
972 FW_HDR_FW_VER_MINOR_GET(adapter
->params
.fw_vers
),
973 FW_HDR_FW_VER_MICRO_GET(adapter
->params
.fw_vers
),
974 FW_HDR_FW_VER_BUILD_GET(adapter
->params
.fw_vers
),
975 FW_HDR_FW_VER_MAJOR_GET(adapter
->params
.tp_vers
),
976 FW_HDR_FW_VER_MINOR_GET(adapter
->params
.tp_vers
),
977 FW_HDR_FW_VER_MICRO_GET(adapter
->params
.tp_vers
),
978 FW_HDR_FW_VER_BUILD_GET(adapter
->params
.tp_vers
));
981 static void get_strings(struct net_device
*dev
, u32 stringset
, u8
*data
)
983 if (stringset
== ETH_SS_STATS
)
984 memcpy(data
, stats_strings
, sizeof(stats_strings
));
988 * port stats maintained per queue of the port. They should be in the same
989 * order as in stats_strings above.
991 struct queue_port_stats
{
1001 static void collect_sge_port_stats(const struct adapter
*adap
,
1002 const struct port_info
*p
, struct queue_port_stats
*s
)
1005 const struct sge_eth_txq
*tx
= &adap
->sge
.ethtxq
[p
->first_qset
];
1006 const struct sge_eth_rxq
*rx
= &adap
->sge
.ethrxq
[p
->first_qset
];
1008 memset(s
, 0, sizeof(*s
));
1009 for (i
= 0; i
< p
->nqsets
; i
++, rx
++, tx
++) {
1011 s
->tx_csum
+= tx
->tx_cso
;
1012 s
->rx_csum
+= rx
->stats
.rx_cso
;
1013 s
->vlan_ex
+= rx
->stats
.vlan_ex
;
1014 s
->vlan_ins
+= tx
->vlan_ins
;
1015 s
->gro_pkts
+= rx
->stats
.lro_pkts
;
1016 s
->gro_merged
+= rx
->stats
.lro_merged
;
1020 static void get_stats(struct net_device
*dev
, struct ethtool_stats
*stats
,
1023 struct port_info
*pi
= netdev_priv(dev
);
1024 struct adapter
*adapter
= pi
->adapter
;
1026 t4_get_port_stats(adapter
, pi
->tx_chan
, (struct port_stats
*)data
);
1028 data
+= sizeof(struct port_stats
) / sizeof(u64
);
1029 collect_sge_port_stats(adapter
, pi
, (struct queue_port_stats
*)data
);
1033 * Return a version number to identify the type of adapter. The scheme is:
1034 * - bits 0..9: chip version
1035 * - bits 10..15: chip revision
1037 static inline unsigned int mk_adap_vers(const struct adapter
*ap
)
1039 return 4 | (ap
->params
.rev
<< 10);
1042 static void reg_block_dump(struct adapter
*ap
, void *buf
, unsigned int start
,
1045 u32
*p
= buf
+ start
;
1047 for ( ; start
<= end
; start
+= sizeof(u32
))
1048 *p
++ = t4_read_reg(ap
, start
);
1051 static void get_regs(struct net_device
*dev
, struct ethtool_regs
*regs
,
1054 static const unsigned int reg_ranges
[] = {
1273 struct adapter
*ap
= netdev2adap(dev
);
1275 regs
->version
= mk_adap_vers(ap
);
1277 memset(buf
, 0, T4_REGMAP_SIZE
);
1278 for (i
= 0; i
< ARRAY_SIZE(reg_ranges
); i
+= 2)
1279 reg_block_dump(ap
, buf
, reg_ranges
[i
], reg_ranges
[i
+ 1]);
1282 static int restart_autoneg(struct net_device
*dev
)
1284 struct port_info
*p
= netdev_priv(dev
);
1286 if (!netif_running(dev
))
1288 if (p
->link_cfg
.autoneg
!= AUTONEG_ENABLE
)
1290 t4_restart_aneg(p
->adapter
, 0, p
->tx_chan
);
1294 static int identify_port(struct net_device
*dev
, u32 data
)
1297 data
= 2; /* default to 2 seconds */
1299 return t4_identify_port(netdev2adap(dev
), 0, netdev2pinfo(dev
)->viid
,
1303 static unsigned int from_fw_linkcaps(unsigned int type
, unsigned int caps
)
1307 if (type
== FW_PORT_TYPE_BT_SGMII
|| type
== FW_PORT_TYPE_BT_XFI
||
1308 type
== FW_PORT_TYPE_BT_XAUI
) {
1310 if (caps
& FW_PORT_CAP_SPEED_100M
)
1311 v
|= SUPPORTED_100baseT_Full
;
1312 if (caps
& FW_PORT_CAP_SPEED_1G
)
1313 v
|= SUPPORTED_1000baseT_Full
;
1314 if (caps
& FW_PORT_CAP_SPEED_10G
)
1315 v
|= SUPPORTED_10000baseT_Full
;
1316 } else if (type
== FW_PORT_TYPE_KX4
|| type
== FW_PORT_TYPE_KX
) {
1317 v
|= SUPPORTED_Backplane
;
1318 if (caps
& FW_PORT_CAP_SPEED_1G
)
1319 v
|= SUPPORTED_1000baseKX_Full
;
1320 if (caps
& FW_PORT_CAP_SPEED_10G
)
1321 v
|= SUPPORTED_10000baseKX4_Full
;
1322 } else if (type
== FW_PORT_TYPE_KR
)
1323 v
|= SUPPORTED_Backplane
| SUPPORTED_10000baseKR_Full
;
1324 else if (type
== FW_PORT_TYPE_BP_AP
)
1325 v
|= SUPPORTED_Backplane
| SUPPORTED_10000baseR_FEC
;
1326 else if (type
== FW_PORT_TYPE_FIBER_XFI
||
1327 type
== FW_PORT_TYPE_FIBER_XAUI
|| type
== FW_PORT_TYPE_SFP
)
1328 v
|= SUPPORTED_FIBRE
;
1330 if (caps
& FW_PORT_CAP_ANEG
)
1331 v
|= SUPPORTED_Autoneg
;
1335 static unsigned int to_fw_linkcaps(unsigned int caps
)
1339 if (caps
& ADVERTISED_100baseT_Full
)
1340 v
|= FW_PORT_CAP_SPEED_100M
;
1341 if (caps
& ADVERTISED_1000baseT_Full
)
1342 v
|= FW_PORT_CAP_SPEED_1G
;
1343 if (caps
& ADVERTISED_10000baseT_Full
)
1344 v
|= FW_PORT_CAP_SPEED_10G
;
1348 static int get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1350 const struct port_info
*p
= netdev_priv(dev
);
1352 if (p
->port_type
== FW_PORT_TYPE_BT_SGMII
||
1353 p
->port_type
== FW_PORT_TYPE_BT_XFI
||
1354 p
->port_type
== FW_PORT_TYPE_BT_XAUI
)
1355 cmd
->port
= PORT_TP
;
1356 else if (p
->port_type
== FW_PORT_TYPE_FIBER_XFI
||
1357 p
->port_type
== FW_PORT_TYPE_FIBER_XAUI
)
1358 cmd
->port
= PORT_FIBRE
;
1359 else if (p
->port_type
== FW_PORT_TYPE_SFP
) {
1360 if (p
->mod_type
== FW_PORT_MOD_TYPE_TWINAX_PASSIVE
||
1361 p
->mod_type
== FW_PORT_MOD_TYPE_TWINAX_ACTIVE
)
1362 cmd
->port
= PORT_DA
;
1364 cmd
->port
= PORT_FIBRE
;
1366 cmd
->port
= PORT_OTHER
;
1368 if (p
->mdio_addr
>= 0) {
1369 cmd
->phy_address
= p
->mdio_addr
;
1370 cmd
->transceiver
= XCVR_EXTERNAL
;
1371 cmd
->mdio_support
= p
->port_type
== FW_PORT_TYPE_BT_SGMII
?
1372 MDIO_SUPPORTS_C22
: MDIO_SUPPORTS_C45
;
1374 cmd
->phy_address
= 0; /* not really, but no better option */
1375 cmd
->transceiver
= XCVR_INTERNAL
;
1376 cmd
->mdio_support
= 0;
1379 cmd
->supported
= from_fw_linkcaps(p
->port_type
, p
->link_cfg
.supported
);
1380 cmd
->advertising
= from_fw_linkcaps(p
->port_type
,
1381 p
->link_cfg
.advertising
);
1382 cmd
->speed
= netif_carrier_ok(dev
) ? p
->link_cfg
.speed
: 0;
1383 cmd
->duplex
= DUPLEX_FULL
;
1384 cmd
->autoneg
= p
->link_cfg
.autoneg
;
1390 static unsigned int speed_to_caps(int speed
)
1392 if (speed
== SPEED_100
)
1393 return FW_PORT_CAP_SPEED_100M
;
1394 if (speed
== SPEED_1000
)
1395 return FW_PORT_CAP_SPEED_1G
;
1396 if (speed
== SPEED_10000
)
1397 return FW_PORT_CAP_SPEED_10G
;
1401 static int set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1404 struct port_info
*p
= netdev_priv(dev
);
1405 struct link_config
*lc
= &p
->link_cfg
;
1407 if (cmd
->duplex
!= DUPLEX_FULL
) /* only full-duplex supported */
1410 if (!(lc
->supported
& FW_PORT_CAP_ANEG
)) {
1412 * PHY offers a single speed. See if that's what's
1415 if (cmd
->autoneg
== AUTONEG_DISABLE
&&
1416 (lc
->supported
& speed_to_caps(cmd
->speed
)))
1421 if (cmd
->autoneg
== AUTONEG_DISABLE
) {
1422 cap
= speed_to_caps(cmd
->speed
);
1424 if (!(lc
->supported
& cap
) || cmd
->speed
== SPEED_1000
||
1425 cmd
->speed
== SPEED_10000
)
1427 lc
->requested_speed
= cap
;
1428 lc
->advertising
= 0;
1430 cap
= to_fw_linkcaps(cmd
->advertising
);
1431 if (!(lc
->supported
& cap
))
1433 lc
->requested_speed
= 0;
1434 lc
->advertising
= cap
| FW_PORT_CAP_ANEG
;
1436 lc
->autoneg
= cmd
->autoneg
;
1438 if (netif_running(dev
))
1439 return t4_link_start(p
->adapter
, 0, p
->tx_chan
, lc
);
1443 static void get_pauseparam(struct net_device
*dev
,
1444 struct ethtool_pauseparam
*epause
)
1446 struct port_info
*p
= netdev_priv(dev
);
1448 epause
->autoneg
= (p
->link_cfg
.requested_fc
& PAUSE_AUTONEG
) != 0;
1449 epause
->rx_pause
= (p
->link_cfg
.fc
& PAUSE_RX
) != 0;
1450 epause
->tx_pause
= (p
->link_cfg
.fc
& PAUSE_TX
) != 0;
1453 static int set_pauseparam(struct net_device
*dev
,
1454 struct ethtool_pauseparam
*epause
)
1456 struct port_info
*p
= netdev_priv(dev
);
1457 struct link_config
*lc
= &p
->link_cfg
;
1459 if (epause
->autoneg
== AUTONEG_DISABLE
)
1460 lc
->requested_fc
= 0;
1461 else if (lc
->supported
& FW_PORT_CAP_ANEG
)
1462 lc
->requested_fc
= PAUSE_AUTONEG
;
1466 if (epause
->rx_pause
)
1467 lc
->requested_fc
|= PAUSE_RX
;
1468 if (epause
->tx_pause
)
1469 lc
->requested_fc
|= PAUSE_TX
;
1470 if (netif_running(dev
))
1471 return t4_link_start(p
->adapter
, 0, p
->tx_chan
, lc
);
1475 static u32
get_rx_csum(struct net_device
*dev
)
1477 struct port_info
*p
= netdev_priv(dev
);
1479 return p
->rx_offload
& RX_CSO
;
1482 static int set_rx_csum(struct net_device
*dev
, u32 data
)
1484 struct port_info
*p
= netdev_priv(dev
);
1487 p
->rx_offload
|= RX_CSO
;
1489 p
->rx_offload
&= ~RX_CSO
;
1493 static void get_sge_param(struct net_device
*dev
, struct ethtool_ringparam
*e
)
1495 const struct port_info
*pi
= netdev_priv(dev
);
1496 const struct sge
*s
= &pi
->adapter
->sge
;
1498 e
->rx_max_pending
= MAX_RX_BUFFERS
;
1499 e
->rx_mini_max_pending
= MAX_RSPQ_ENTRIES
;
1500 e
->rx_jumbo_max_pending
= 0;
1501 e
->tx_max_pending
= MAX_TXQ_ENTRIES
;
1503 e
->rx_pending
= s
->ethrxq
[pi
->first_qset
].fl
.size
- 8;
1504 e
->rx_mini_pending
= s
->ethrxq
[pi
->first_qset
].rspq
.size
;
1505 e
->rx_jumbo_pending
= 0;
1506 e
->tx_pending
= s
->ethtxq
[pi
->first_qset
].q
.size
;
1509 static int set_sge_param(struct net_device
*dev
, struct ethtool_ringparam
*e
)
1512 const struct port_info
*pi
= netdev_priv(dev
);
1513 struct adapter
*adapter
= pi
->adapter
;
1514 struct sge
*s
= &adapter
->sge
;
1516 if (e
->rx_pending
> MAX_RX_BUFFERS
|| e
->rx_jumbo_pending
||
1517 e
->tx_pending
> MAX_TXQ_ENTRIES
||
1518 e
->rx_mini_pending
> MAX_RSPQ_ENTRIES
||
1519 e
->rx_mini_pending
< MIN_RSPQ_ENTRIES
||
1520 e
->rx_pending
< MIN_FL_ENTRIES
|| e
->tx_pending
< MIN_TXQ_ENTRIES
)
1523 if (adapter
->flags
& FULL_INIT_DONE
)
1526 for (i
= 0; i
< pi
->nqsets
; ++i
) {
1527 s
->ethtxq
[pi
->first_qset
+ i
].q
.size
= e
->tx_pending
;
1528 s
->ethrxq
[pi
->first_qset
+ i
].fl
.size
= e
->rx_pending
+ 8;
1529 s
->ethrxq
[pi
->first_qset
+ i
].rspq
.size
= e
->rx_mini_pending
;
1534 static int closest_timer(const struct sge
*s
, int time
)
1536 int i
, delta
, match
= 0, min_delta
= INT_MAX
;
1538 for (i
= 0; i
< ARRAY_SIZE(s
->timer_val
); i
++) {
1539 delta
= time
- s
->timer_val
[i
];
1542 if (delta
< min_delta
) {
1550 static int closest_thres(const struct sge
*s
, int thres
)
1552 int i
, delta
, match
= 0, min_delta
= INT_MAX
;
1554 for (i
= 0; i
< ARRAY_SIZE(s
->counter_val
); i
++) {
1555 delta
= thres
- s
->counter_val
[i
];
1558 if (delta
< min_delta
) {
1567 * Return a queue's interrupt hold-off time in us. 0 means no timer.
1569 static unsigned int qtimer_val(const struct adapter
*adap
,
1570 const struct sge_rspq
*q
)
1572 unsigned int idx
= q
->intr_params
>> 1;
1574 return idx
< SGE_NTIMERS
? adap
->sge
.timer_val
[idx
] : 0;
1578 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1579 * @adap: the adapter
1581 * @us: the hold-off time in us, or 0 to disable timer
1582 * @cnt: the hold-off packet count, or 0 to disable counter
1584 * Sets an Rx queue's interrupt hold-off time and packet count. At least
1585 * one of the two needs to be enabled for the queue to generate interrupts.
1587 static int set_rxq_intr_params(struct adapter
*adap
, struct sge_rspq
*q
,
1588 unsigned int us
, unsigned int cnt
)
1590 if ((us
| cnt
) == 0)
1597 new_idx
= closest_thres(&adap
->sge
, cnt
);
1598 if (q
->desc
&& q
->pktcnt_idx
!= new_idx
) {
1599 /* the queue has already been created, update it */
1600 v
= FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ
) |
1601 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH
) |
1602 FW_PARAMS_PARAM_YZ(q
->cntxt_id
);
1603 err
= t4_set_params(adap
, 0, 0, 0, 1, &v
, &new_idx
);
1607 q
->pktcnt_idx
= new_idx
;
1610 us
= us
== 0 ? 6 : closest_timer(&adap
->sge
, us
);
1611 q
->intr_params
= QINTR_TIMER_IDX(us
) | (cnt
> 0 ? QINTR_CNT_EN
: 0);
1615 static int set_coalesce(struct net_device
*dev
, struct ethtool_coalesce
*c
)
1617 const struct port_info
*pi
= netdev_priv(dev
);
1618 struct adapter
*adap
= pi
->adapter
;
1620 return set_rxq_intr_params(adap
, &adap
->sge
.ethrxq
[pi
->first_qset
].rspq
,
1621 c
->rx_coalesce_usecs
, c
->rx_max_coalesced_frames
);
1624 static int get_coalesce(struct net_device
*dev
, struct ethtool_coalesce
*c
)
1626 const struct port_info
*pi
= netdev_priv(dev
);
1627 const struct adapter
*adap
= pi
->adapter
;
1628 const struct sge_rspq
*rq
= &adap
->sge
.ethrxq
[pi
->first_qset
].rspq
;
1630 c
->rx_coalesce_usecs
= qtimer_val(adap
, rq
);
1631 c
->rx_max_coalesced_frames
= (rq
->intr_params
& QINTR_CNT_EN
) ?
1632 adap
->sge
.counter_val
[rq
->pktcnt_idx
] : 0;
1637 * Translate a physical EEPROM address to virtual. The first 1K is accessed
1638 * through virtual addresses starting at 31K, the rest is accessed through
1639 * virtual addresses starting at 0. This mapping is correct only for PF0.
1641 static int eeprom_ptov(unsigned int phys_addr
)
1643 if (phys_addr
< 1024)
1644 return phys_addr
+ (31 << 10);
1645 if (phys_addr
< EEPROMSIZE
)
1646 return phys_addr
- 1024;
1651 * The next two routines implement eeprom read/write from physical addresses.
1652 * The physical->virtual translation is correct only for PF0.
1654 static int eeprom_rd_phys(struct adapter
*adap
, unsigned int phys_addr
, u32
*v
)
1656 int vaddr
= eeprom_ptov(phys_addr
);
1659 vaddr
= pci_read_vpd(adap
->pdev
, vaddr
, sizeof(u32
), v
);
1660 return vaddr
< 0 ? vaddr
: 0;
1663 static int eeprom_wr_phys(struct adapter
*adap
, unsigned int phys_addr
, u32 v
)
1665 int vaddr
= eeprom_ptov(phys_addr
);
1668 vaddr
= pci_write_vpd(adap
->pdev
, vaddr
, sizeof(u32
), &v
);
1669 return vaddr
< 0 ? vaddr
: 0;
1672 #define EEPROM_MAGIC 0x38E2F10C
1674 static int get_eeprom(struct net_device
*dev
, struct ethtool_eeprom
*e
,
1678 struct adapter
*adapter
= netdev2adap(dev
);
1680 u8
*buf
= kmalloc(EEPROMSIZE
, GFP_KERNEL
);
1684 e
->magic
= EEPROM_MAGIC
;
1685 for (i
= e
->offset
& ~3; !err
&& i
< e
->offset
+ e
->len
; i
+= 4)
1686 err
= eeprom_rd_phys(adapter
, i
, (u32
*)&buf
[i
]);
1689 memcpy(data
, buf
+ e
->offset
, e
->len
);
1694 static int set_eeprom(struct net_device
*dev
, struct ethtool_eeprom
*eeprom
,
1699 u32 aligned_offset
, aligned_len
, *p
;
1700 struct adapter
*adapter
= netdev2adap(dev
);
1702 if (eeprom
->magic
!= EEPROM_MAGIC
)
1705 aligned_offset
= eeprom
->offset
& ~3;
1706 aligned_len
= (eeprom
->len
+ (eeprom
->offset
& 3) + 3) & ~3;
1708 if (aligned_offset
!= eeprom
->offset
|| aligned_len
!= eeprom
->len
) {
1710 * RMW possibly needed for first or last words.
1712 buf
= kmalloc(aligned_len
, GFP_KERNEL
);
1715 err
= eeprom_rd_phys(adapter
, aligned_offset
, (u32
*)buf
);
1716 if (!err
&& aligned_len
> 4)
1717 err
= eeprom_rd_phys(adapter
,
1718 aligned_offset
+ aligned_len
- 4,
1719 (u32
*)&buf
[aligned_len
- 4]);
1722 memcpy(buf
+ (eeprom
->offset
& 3), data
, eeprom
->len
);
1726 err
= t4_seeprom_wp(adapter
, false);
1730 for (p
= (u32
*)buf
; !err
&& aligned_len
; aligned_len
-= 4, p
++) {
1731 err
= eeprom_wr_phys(adapter
, aligned_offset
, *p
);
1732 aligned_offset
+= 4;
1736 err
= t4_seeprom_wp(adapter
, true);
1743 static int set_flash(struct net_device
*netdev
, struct ethtool_flash
*ef
)
1746 const struct firmware
*fw
;
1747 struct adapter
*adap
= netdev2adap(netdev
);
1749 ef
->data
[sizeof(ef
->data
) - 1] = '\0';
1750 ret
= request_firmware(&fw
, ef
->data
, adap
->pdev_dev
);
1754 ret
= t4_load_fw(adap
, fw
->data
, fw
->size
);
1755 release_firmware(fw
);
1757 dev_info(adap
->pdev_dev
, "loaded firmware %s\n", ef
->data
);
1761 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
1762 #define BCAST_CRC 0xa0ccc1a6
1764 static void get_wol(struct net_device
*dev
, struct ethtool_wolinfo
*wol
)
1766 wol
->supported
= WAKE_BCAST
| WAKE_MAGIC
;
1767 wol
->wolopts
= netdev2adap(dev
)->wol
;
1768 memset(&wol
->sopass
, 0, sizeof(wol
->sopass
));
1771 static int set_wol(struct net_device
*dev
, struct ethtool_wolinfo
*wol
)
1774 struct port_info
*pi
= netdev_priv(dev
);
1776 if (wol
->wolopts
& ~WOL_SUPPORTED
)
1778 t4_wol_magic_enable(pi
->adapter
, pi
->tx_chan
,
1779 (wol
->wolopts
& WAKE_MAGIC
) ? dev
->dev_addr
: NULL
);
1780 if (wol
->wolopts
& WAKE_BCAST
) {
1781 err
= t4_wol_pat_enable(pi
->adapter
, pi
->tx_chan
, 0xfe, ~0ULL,
1784 err
= t4_wol_pat_enable(pi
->adapter
, pi
->tx_chan
, 1,
1785 ~6ULL, ~0ULL, BCAST_CRC
, true);
1787 t4_wol_pat_enable(pi
->adapter
, pi
->tx_chan
, 0, 0, 0, 0, false);
1791 static int set_tso(struct net_device
*dev
, u32 value
)
1794 dev
->features
|= NETIF_F_TSO
| NETIF_F_TSO6
;
1796 dev
->features
&= ~(NETIF_F_TSO
| NETIF_F_TSO6
);
1800 static int set_flags(struct net_device
*dev
, u32 flags
)
1802 return ethtool_op_set_flags(dev
, flags
, ETH_FLAG_RXHASH
);
1805 static struct ethtool_ops cxgb_ethtool_ops
= {
1806 .get_settings
= get_settings
,
1807 .set_settings
= set_settings
,
1808 .get_drvinfo
= get_drvinfo
,
1809 .get_msglevel
= get_msglevel
,
1810 .set_msglevel
= set_msglevel
,
1811 .get_ringparam
= get_sge_param
,
1812 .set_ringparam
= set_sge_param
,
1813 .get_coalesce
= get_coalesce
,
1814 .set_coalesce
= set_coalesce
,
1815 .get_eeprom_len
= get_eeprom_len
,
1816 .get_eeprom
= get_eeprom
,
1817 .set_eeprom
= set_eeprom
,
1818 .get_pauseparam
= get_pauseparam
,
1819 .set_pauseparam
= set_pauseparam
,
1820 .get_rx_csum
= get_rx_csum
,
1821 .set_rx_csum
= set_rx_csum
,
1822 .set_tx_csum
= ethtool_op_set_tx_ipv6_csum
,
1823 .set_sg
= ethtool_op_set_sg
,
1824 .get_link
= ethtool_op_get_link
,
1825 .get_strings
= get_strings
,
1826 .phys_id
= identify_port
,
1827 .nway_reset
= restart_autoneg
,
1828 .get_sset_count
= get_sset_count
,
1829 .get_ethtool_stats
= get_stats
,
1830 .get_regs_len
= get_regs_len
,
1831 .get_regs
= get_regs
,
1835 .set_flags
= set_flags
,
1836 .flash_device
= set_flash
,
1843 static int mem_open(struct inode
*inode
, struct file
*file
)
1845 file
->private_data
= inode
->i_private
;
1849 static ssize_t
mem_read(struct file
*file
, char __user
*buf
, size_t count
,
1853 loff_t avail
= file
->f_path
.dentry
->d_inode
->i_size
;
1854 unsigned int mem
= (uintptr_t)file
->private_data
& 3;
1855 struct adapter
*adap
= file
->private_data
- mem
;
1861 if (count
> avail
- pos
)
1862 count
= avail
- pos
;
1870 ret
= t4_mc_read(adap
, pos
, data
, NULL
);
1872 ret
= t4_edc_read(adap
, mem
, pos
, data
, NULL
);
1876 ofst
= pos
% sizeof(data
);
1877 len
= min(count
, sizeof(data
) - ofst
);
1878 if (copy_to_user(buf
, (u8
*)data
+ ofst
, len
))
1885 count
= pos
- *ppos
;
1890 static const struct file_operations mem_debugfs_fops
= {
1891 .owner
= THIS_MODULE
,
1896 static void __devinit
add_debugfs_mem(struct adapter
*adap
, const char *name
,
1897 unsigned int idx
, unsigned int size_mb
)
1901 de
= debugfs_create_file(name
, S_IRUSR
, adap
->debugfs_root
,
1902 (void *)adap
+ idx
, &mem_debugfs_fops
);
1903 if (de
&& de
->d_inode
)
1904 de
->d_inode
->i_size
= size_mb
<< 20;
1907 static int __devinit
setup_debugfs(struct adapter
*adap
)
1911 if (IS_ERR_OR_NULL(adap
->debugfs_root
))
1914 i
= t4_read_reg(adap
, MA_TARGET_MEM_ENABLE
);
1915 if (i
& EDRAM0_ENABLE
)
1916 add_debugfs_mem(adap
, "edc0", MEM_EDC0
, 5);
1917 if (i
& EDRAM1_ENABLE
)
1918 add_debugfs_mem(adap
, "edc1", MEM_EDC1
, 5);
1919 if (i
& EXT_MEM_ENABLE
)
1920 add_debugfs_mem(adap
, "mc", MEM_MC
,
1921 EXT_MEM_SIZE_GET(t4_read_reg(adap
, MA_EXT_MEMORY_BAR
)));
1923 debugfs_create_file("l2t", S_IRUSR
, adap
->debugfs_root
, adap
,
1929 * upper-layer driver support
1933 * Allocate an active-open TID and set it to the supplied value.
1935 int cxgb4_alloc_atid(struct tid_info
*t
, void *data
)
1939 spin_lock_bh(&t
->atid_lock
);
1941 union aopen_entry
*p
= t
->afree
;
1943 atid
= p
- t
->atid_tab
;
1948 spin_unlock_bh(&t
->atid_lock
);
1951 EXPORT_SYMBOL(cxgb4_alloc_atid
);
1954 * Release an active-open TID.
1956 void cxgb4_free_atid(struct tid_info
*t
, unsigned int atid
)
1958 union aopen_entry
*p
= &t
->atid_tab
[atid
];
1960 spin_lock_bh(&t
->atid_lock
);
1964 spin_unlock_bh(&t
->atid_lock
);
1966 EXPORT_SYMBOL(cxgb4_free_atid
);
1969 * Allocate a server TID and set it to the supplied value.
1971 int cxgb4_alloc_stid(struct tid_info
*t
, int family
, void *data
)
1975 spin_lock_bh(&t
->stid_lock
);
1976 if (family
== PF_INET
) {
1977 stid
= find_first_zero_bit(t
->stid_bmap
, t
->nstids
);
1978 if (stid
< t
->nstids
)
1979 __set_bit(stid
, t
->stid_bmap
);
1983 stid
= bitmap_find_free_region(t
->stid_bmap
, t
->nstids
, 2);
1988 t
->stid_tab
[stid
].data
= data
;
1989 stid
+= t
->stid_base
;
1992 spin_unlock_bh(&t
->stid_lock
);
1995 EXPORT_SYMBOL(cxgb4_alloc_stid
);
1998 * Release a server TID.
2000 void cxgb4_free_stid(struct tid_info
*t
, unsigned int stid
, int family
)
2002 stid
-= t
->stid_base
;
2003 spin_lock_bh(&t
->stid_lock
);
2004 if (family
== PF_INET
)
2005 __clear_bit(stid
, t
->stid_bmap
);
2007 bitmap_release_region(t
->stid_bmap
, stid
, 2);
2008 t
->stid_tab
[stid
].data
= NULL
;
2010 spin_unlock_bh(&t
->stid_lock
);
2012 EXPORT_SYMBOL(cxgb4_free_stid
);
2015 * Populate a TID_RELEASE WR. Caller must properly size the skb.
2017 static void mk_tid_release(struct sk_buff
*skb
, unsigned int chan
,
2020 struct cpl_tid_release
*req
;
2022 set_wr_txq(skb
, CPL_PRIORITY_SETUP
, chan
);
2023 req
= (struct cpl_tid_release
*)__skb_put(skb
, sizeof(*req
));
2024 INIT_TP_WR(req
, tid
);
2025 OPCODE_TID(req
) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE
, tid
));
2029 * Queue a TID release request and if necessary schedule a work queue to
2032 void cxgb4_queue_tid_release(struct tid_info
*t
, unsigned int chan
,
2035 void **p
= &t
->tid_tab
[tid
];
2036 struct adapter
*adap
= container_of(t
, struct adapter
, tids
);
2038 spin_lock_bh(&adap
->tid_release_lock
);
2039 *p
= adap
->tid_release_head
;
2040 /* Low 2 bits encode the Tx channel number */
2041 adap
->tid_release_head
= (void **)((uintptr_t)p
| chan
);
2042 if (!adap
->tid_release_task_busy
) {
2043 adap
->tid_release_task_busy
= true;
2044 schedule_work(&adap
->tid_release_task
);
2046 spin_unlock_bh(&adap
->tid_release_lock
);
2048 EXPORT_SYMBOL(cxgb4_queue_tid_release
);
2051 * Process the list of pending TID release requests.
2053 static void process_tid_release_list(struct work_struct
*work
)
2055 struct sk_buff
*skb
;
2056 struct adapter
*adap
;
2058 adap
= container_of(work
, struct adapter
, tid_release_task
);
2060 spin_lock_bh(&adap
->tid_release_lock
);
2061 while (adap
->tid_release_head
) {
2062 void **p
= adap
->tid_release_head
;
2063 unsigned int chan
= (uintptr_t)p
& 3;
2064 p
= (void *)p
- chan
;
2066 adap
->tid_release_head
= *p
;
2068 spin_unlock_bh(&adap
->tid_release_lock
);
2070 while (!(skb
= alloc_skb(sizeof(struct cpl_tid_release
),
2072 schedule_timeout_uninterruptible(1);
2074 mk_tid_release(skb
, chan
, p
- adap
->tids
.tid_tab
);
2075 t4_ofld_send(adap
, skb
);
2076 spin_lock_bh(&adap
->tid_release_lock
);
2078 adap
->tid_release_task_busy
= false;
2079 spin_unlock_bh(&adap
->tid_release_lock
);
2083 * Release a TID and inform HW. If we are unable to allocate the release
2084 * message we defer to a work queue.
2086 void cxgb4_remove_tid(struct tid_info
*t
, unsigned int chan
, unsigned int tid
)
2089 struct sk_buff
*skb
;
2090 struct adapter
*adap
= container_of(t
, struct adapter
, tids
);
2092 old
= t
->tid_tab
[tid
];
2093 skb
= alloc_skb(sizeof(struct cpl_tid_release
), GFP_ATOMIC
);
2095 t
->tid_tab
[tid
] = NULL
;
2096 mk_tid_release(skb
, chan
, tid
);
2097 t4_ofld_send(adap
, skb
);
2099 cxgb4_queue_tid_release(t
, chan
, tid
);
2101 atomic_dec(&t
->tids_in_use
);
2103 EXPORT_SYMBOL(cxgb4_remove_tid
);
2106 * Allocate and initialize the TID tables. Returns 0 on success.
2108 static int tid_init(struct tid_info
*t
)
2111 unsigned int natids
= t
->natids
;
2113 size
= t
->ntids
* sizeof(*t
->tid_tab
) + natids
* sizeof(*t
->atid_tab
) +
2114 t
->nstids
* sizeof(*t
->stid_tab
) +
2115 BITS_TO_LONGS(t
->nstids
) * sizeof(long);
2116 t
->tid_tab
= t4_alloc_mem(size
);
2120 t
->atid_tab
= (union aopen_entry
*)&t
->tid_tab
[t
->ntids
];
2121 t
->stid_tab
= (struct serv_entry
*)&t
->atid_tab
[natids
];
2122 t
->stid_bmap
= (unsigned long *)&t
->stid_tab
[t
->nstids
];
2123 spin_lock_init(&t
->stid_lock
);
2124 spin_lock_init(&t
->atid_lock
);
2126 t
->stids_in_use
= 0;
2128 t
->atids_in_use
= 0;
2129 atomic_set(&t
->tids_in_use
, 0);
2131 /* Setup the free list for atid_tab and clear the stid bitmap. */
2134 t
->atid_tab
[natids
- 1].next
= &t
->atid_tab
[natids
];
2135 t
->afree
= t
->atid_tab
;
2137 bitmap_zero(t
->stid_bmap
, t
->nstids
);
2142 * cxgb4_create_server - create an IP server
2144 * @stid: the server TID
2145 * @sip: local IP address to bind server to
2146 * @sport: the server's TCP port
2147 * @queue: queue to direct messages from this server to
2149 * Create an IP server for the given port and address.
2150 * Returns <0 on error and one of the %NET_XMIT_* values on success.
2152 int cxgb4_create_server(const struct net_device
*dev
, unsigned int stid
,
2153 __be32 sip
, __be16 sport
, unsigned int queue
)
2156 struct sk_buff
*skb
;
2157 struct adapter
*adap
;
2158 struct cpl_pass_open_req
*req
;
2160 skb
= alloc_skb(sizeof(*req
), GFP_KERNEL
);
2164 adap
= netdev2adap(dev
);
2165 req
= (struct cpl_pass_open_req
*)__skb_put(skb
, sizeof(*req
));
2167 OPCODE_TID(req
) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ
, stid
));
2168 req
->local_port
= sport
;
2169 req
->peer_port
= htons(0);
2170 req
->local_ip
= sip
;
2171 req
->peer_ip
= htonl(0);
2172 chan
= netdev2pinfo(adap
->sge
.ingr_map
[queue
]->netdev
)->tx_chan
;
2173 req
->opt0
= cpu_to_be64(TX_CHAN(chan
));
2174 req
->opt1
= cpu_to_be64(CONN_POLICY_ASK
|
2175 SYN_RSS_ENABLE
| SYN_RSS_QUEUE(queue
));
2176 return t4_mgmt_tx(adap
, skb
);
2178 EXPORT_SYMBOL(cxgb4_create_server
);
2181 * cxgb4_create_server6 - create an IPv6 server
2183 * @stid: the server TID
2184 * @sip: local IPv6 address to bind server to
2185 * @sport: the server's TCP port
2186 * @queue: queue to direct messages from this server to
2188 * Create an IPv6 server for the given port and address.
2189 * Returns <0 on error and one of the %NET_XMIT_* values on success.
2191 int cxgb4_create_server6(const struct net_device
*dev
, unsigned int stid
,
2192 const struct in6_addr
*sip
, __be16 sport
,
2196 struct sk_buff
*skb
;
2197 struct adapter
*adap
;
2198 struct cpl_pass_open_req6
*req
;
2200 skb
= alloc_skb(sizeof(*req
), GFP_KERNEL
);
2204 adap
= netdev2adap(dev
);
2205 req
= (struct cpl_pass_open_req6
*)__skb_put(skb
, sizeof(*req
));
2207 OPCODE_TID(req
) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6
, stid
));
2208 req
->local_port
= sport
;
2209 req
->peer_port
= htons(0);
2210 req
->local_ip_hi
= *(__be64
*)(sip
->s6_addr
);
2211 req
->local_ip_lo
= *(__be64
*)(sip
->s6_addr
+ 8);
2212 req
->peer_ip_hi
= cpu_to_be64(0);
2213 req
->peer_ip_lo
= cpu_to_be64(0);
2214 chan
= netdev2pinfo(adap
->sge
.ingr_map
[queue
]->netdev
)->tx_chan
;
2215 req
->opt0
= cpu_to_be64(TX_CHAN(chan
));
2216 req
->opt1
= cpu_to_be64(CONN_POLICY_ASK
|
2217 SYN_RSS_ENABLE
| SYN_RSS_QUEUE(queue
));
2218 return t4_mgmt_tx(adap
, skb
);
2220 EXPORT_SYMBOL(cxgb4_create_server6
);
2223 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
2224 * @mtus: the HW MTU table
2225 * @mtu: the target MTU
2226 * @idx: index of selected entry in the MTU table
2228 * Returns the index and the value in the HW MTU table that is closest to
2229 * but does not exceed @mtu, unless @mtu is smaller than any value in the
2230 * table, in which case that smallest available value is selected.
2232 unsigned int cxgb4_best_mtu(const unsigned short *mtus
, unsigned short mtu
,
2237 while (i
< NMTUS
- 1 && mtus
[i
+ 1] <= mtu
)
2243 EXPORT_SYMBOL(cxgb4_best_mtu
);
2246 * cxgb4_port_chan - get the HW channel of a port
2247 * @dev: the net device for the port
2249 * Return the HW Tx channel of the given port.
2251 unsigned int cxgb4_port_chan(const struct net_device
*dev
)
2253 return netdev2pinfo(dev
)->tx_chan
;
2255 EXPORT_SYMBOL(cxgb4_port_chan
);
2258 * cxgb4_port_viid - get the VI id of a port
2259 * @dev: the net device for the port
2261 * Return the VI id of the given port.
2263 unsigned int cxgb4_port_viid(const struct net_device
*dev
)
2265 return netdev2pinfo(dev
)->viid
;
2267 EXPORT_SYMBOL(cxgb4_port_viid
);
2270 * cxgb4_port_idx - get the index of a port
2271 * @dev: the net device for the port
2273 * Return the index of the given port.
2275 unsigned int cxgb4_port_idx(const struct net_device
*dev
)
2277 return netdev2pinfo(dev
)->port_id
;
2279 EXPORT_SYMBOL(cxgb4_port_idx
);
2282 * cxgb4_netdev_by_hwid - return the net device of a HW port
2283 * @pdev: identifies the adapter
2284 * @id: the HW port id
2286 * Return the net device associated with the interface with the given HW
2289 struct net_device
*cxgb4_netdev_by_hwid(struct pci_dev
*pdev
, unsigned int id
)
2291 const struct adapter
*adap
= pci_get_drvdata(pdev
);
2293 if (!adap
|| id
>= NCHAN
)
2295 id
= adap
->chan_map
[id
];
2296 return id
< MAX_NPORTS
? adap
->port
[id
] : NULL
;
2298 EXPORT_SYMBOL(cxgb4_netdev_by_hwid
);
2300 void cxgb4_get_tcp_stats(struct pci_dev
*pdev
, struct tp_tcp_stats
*v4
,
2301 struct tp_tcp_stats
*v6
)
2303 struct adapter
*adap
= pci_get_drvdata(pdev
);
2305 spin_lock(&adap
->stats_lock
);
2306 t4_tp_get_tcp_stats(adap
, v4
, v6
);
2307 spin_unlock(&adap
->stats_lock
);
2309 EXPORT_SYMBOL(cxgb4_get_tcp_stats
);
2311 void cxgb4_iscsi_init(struct net_device
*dev
, unsigned int tag_mask
,
2312 const unsigned int *pgsz_order
)
2314 struct adapter
*adap
= netdev2adap(dev
);
2316 t4_write_reg(adap
, ULP_RX_ISCSI_TAGMASK
, tag_mask
);
2317 t4_write_reg(adap
, ULP_RX_ISCSI_PSZ
, HPZ0(pgsz_order
[0]) |
2318 HPZ1(pgsz_order
[1]) | HPZ2(pgsz_order
[2]) |
2319 HPZ3(pgsz_order
[3]));
2321 EXPORT_SYMBOL(cxgb4_iscsi_init
);
2323 static struct pci_driver cxgb4_driver
;
2325 static void check_neigh_update(struct neighbour
*neigh
)
2327 const struct device
*parent
;
2328 const struct net_device
*netdev
= neigh
->dev
;
2330 if (netdev
->priv_flags
& IFF_802_1Q_VLAN
)
2331 netdev
= vlan_dev_real_dev(netdev
);
2332 parent
= netdev
->dev
.parent
;
2333 if (parent
&& parent
->driver
== &cxgb4_driver
.driver
)
2334 t4_l2t_update(dev_get_drvdata(parent
), neigh
);
2337 static int netevent_cb(struct notifier_block
*nb
, unsigned long event
,
2341 case NETEVENT_NEIGH_UPDATE
:
2342 check_neigh_update(data
);
2344 case NETEVENT_PMTU_UPDATE
:
2345 case NETEVENT_REDIRECT
:
2352 static bool netevent_registered
;
2353 static struct notifier_block cxgb4_netevent_nb
= {
2354 .notifier_call
= netevent_cb
2357 static void uld_attach(struct adapter
*adap
, unsigned int uld
)
2360 struct cxgb4_lld_info lli
;
2362 lli
.pdev
= adap
->pdev
;
2363 lli
.l2t
= adap
->l2t
;
2364 lli
.tids
= &adap
->tids
;
2365 lli
.ports
= adap
->port
;
2366 lli
.vr
= &adap
->vres
;
2367 lli
.mtus
= adap
->params
.mtus
;
2368 if (uld
== CXGB4_ULD_RDMA
) {
2369 lli
.rxq_ids
= adap
->sge
.rdma_rxq
;
2370 lli
.nrxq
= adap
->sge
.rdmaqs
;
2371 } else if (uld
== CXGB4_ULD_ISCSI
) {
2372 lli
.rxq_ids
= adap
->sge
.ofld_rxq
;
2373 lli
.nrxq
= adap
->sge
.ofldqsets
;
2375 lli
.ntxq
= adap
->sge
.ofldqsets
;
2376 lli
.nchan
= adap
->params
.nports
;
2377 lli
.nports
= adap
->params
.nports
;
2378 lli
.wr_cred
= adap
->params
.ofldq_wr_cred
;
2379 lli
.adapter_type
= adap
->params
.rev
;
2380 lli
.iscsi_iolen
= MAXRXDATA_GET(t4_read_reg(adap
, TP_PARA_REG2
));
2381 lli
.udb_density
= 1 << QUEUESPERPAGEPF0_GET(
2382 t4_read_reg(adap
, SGE_EGRESS_QUEUES_PER_PAGE_PF
));
2383 lli
.ucq_density
= 1 << QUEUESPERPAGEPF0_GET(
2384 t4_read_reg(adap
, SGE_INGRESS_QUEUES_PER_PAGE_PF
));
2385 lli
.gts_reg
= adap
->regs
+ MYPF_REG(SGE_PF_GTS
);
2386 lli
.db_reg
= adap
->regs
+ MYPF_REG(SGE_PF_KDOORBELL
);
2387 lli
.fw_vers
= adap
->params
.fw_vers
;
2389 handle
= ulds
[uld
].add(&lli
);
2390 if (IS_ERR(handle
)) {
2391 dev_warn(adap
->pdev_dev
,
2392 "could not attach to the %s driver, error %ld\n",
2393 uld_str
[uld
], PTR_ERR(handle
));
2397 adap
->uld_handle
[uld
] = handle
;
2399 if (!netevent_registered
) {
2400 register_netevent_notifier(&cxgb4_netevent_nb
);
2401 netevent_registered
= true;
2404 if (adap
->flags
& FULL_INIT_DONE
)
2405 ulds
[uld
].state_change(handle
, CXGB4_STATE_UP
);
2408 static void attach_ulds(struct adapter
*adap
)
2412 mutex_lock(&uld_mutex
);
2413 list_add_tail(&adap
->list_node
, &adapter_list
);
2414 for (i
= 0; i
< CXGB4_ULD_MAX
; i
++)
2416 uld_attach(adap
, i
);
2417 mutex_unlock(&uld_mutex
);
2420 static void detach_ulds(struct adapter
*adap
)
2424 mutex_lock(&uld_mutex
);
2425 list_del(&adap
->list_node
);
2426 for (i
= 0; i
< CXGB4_ULD_MAX
; i
++)
2427 if (adap
->uld_handle
[i
]) {
2428 ulds
[i
].state_change(adap
->uld_handle
[i
],
2429 CXGB4_STATE_DETACH
);
2430 adap
->uld_handle
[i
] = NULL
;
2432 if (netevent_registered
&& list_empty(&adapter_list
)) {
2433 unregister_netevent_notifier(&cxgb4_netevent_nb
);
2434 netevent_registered
= false;
2436 mutex_unlock(&uld_mutex
);
2439 static void notify_ulds(struct adapter
*adap
, enum cxgb4_state new_state
)
2443 mutex_lock(&uld_mutex
);
2444 for (i
= 0; i
< CXGB4_ULD_MAX
; i
++)
2445 if (adap
->uld_handle
[i
])
2446 ulds
[i
].state_change(adap
->uld_handle
[i
], new_state
);
2447 mutex_unlock(&uld_mutex
);
2451 * cxgb4_register_uld - register an upper-layer driver
2452 * @type: the ULD type
2453 * @p: the ULD methods
2455 * Registers an upper-layer driver with this driver and notifies the ULD
2456 * about any presently available devices that support its type. Returns
2457 * %-EBUSY if a ULD of the same type is already registered.
2459 int cxgb4_register_uld(enum cxgb4_uld type
, const struct cxgb4_uld_info
*p
)
2462 struct adapter
*adap
;
2464 if (type
>= CXGB4_ULD_MAX
)
2466 mutex_lock(&uld_mutex
);
2467 if (ulds
[type
].add
) {
2472 list_for_each_entry(adap
, &adapter_list
, list_node
)
2473 uld_attach(adap
, type
);
2474 out
: mutex_unlock(&uld_mutex
);
2477 EXPORT_SYMBOL(cxgb4_register_uld
);
2480 * cxgb4_unregister_uld - unregister an upper-layer driver
2481 * @type: the ULD type
2483 * Unregisters an existing upper-layer driver.
2485 int cxgb4_unregister_uld(enum cxgb4_uld type
)
2487 struct adapter
*adap
;
2489 if (type
>= CXGB4_ULD_MAX
)
2491 mutex_lock(&uld_mutex
);
2492 list_for_each_entry(adap
, &adapter_list
, list_node
)
2493 adap
->uld_handle
[type
] = NULL
;
2494 ulds
[type
].add
= NULL
;
2495 mutex_unlock(&uld_mutex
);
2498 EXPORT_SYMBOL(cxgb4_unregister_uld
);
2501 * cxgb_up - enable the adapter
2502 * @adap: adapter being enabled
2504 * Called when the first port is enabled, this function performs the
2505 * actions necessary to make an adapter operational, such as completing
2506 * the initialization of HW modules, and enabling interrupts.
2508 * Must be called with the rtnl lock held.
2510 static int cxgb_up(struct adapter
*adap
)
2514 err
= setup_sge_queues(adap
);
2517 err
= setup_rss(adap
);
2521 if (adap
->flags
& USING_MSIX
) {
2522 name_msix_vecs(adap
);
2523 err
= request_irq(adap
->msix_info
[0].vec
, t4_nondata_intr
, 0,
2524 adap
->msix_info
[0].desc
, adap
);
2528 err
= request_msix_queue_irqs(adap
);
2530 free_irq(adap
->msix_info
[0].vec
, adap
);
2534 err
= request_irq(adap
->pdev
->irq
, t4_intr_handler(adap
),
2535 (adap
->flags
& USING_MSI
) ? 0 : IRQF_SHARED
,
2542 t4_intr_enable(adap
);
2543 adap
->flags
|= FULL_INIT_DONE
;
2544 notify_ulds(adap
, CXGB4_STATE_UP
);
2548 dev_err(adap
->pdev_dev
, "request_irq failed, err %d\n", err
);
2550 t4_free_sge_resources(adap
);
2554 static void cxgb_down(struct adapter
*adapter
)
2556 t4_intr_disable(adapter
);
2557 cancel_work_sync(&adapter
->tid_release_task
);
2558 adapter
->tid_release_task_busy
= false;
2559 adapter
->tid_release_head
= NULL
;
2561 if (adapter
->flags
& USING_MSIX
) {
2562 free_msix_queue_irqs(adapter
);
2563 free_irq(adapter
->msix_info
[0].vec
, adapter
);
2565 free_irq(adapter
->pdev
->irq
, adapter
);
2566 quiesce_rx(adapter
);
2567 t4_sge_stop(adapter
);
2568 t4_free_sge_resources(adapter
);
2569 adapter
->flags
&= ~FULL_INIT_DONE
;
2573 * net_device operations
2575 static int cxgb_open(struct net_device
*dev
)
2578 struct port_info
*pi
= netdev_priv(dev
);
2579 struct adapter
*adapter
= pi
->adapter
;
2581 if (!(adapter
->flags
& FULL_INIT_DONE
)) {
2582 err
= cxgb_up(adapter
);
2587 dev
->real_num_tx_queues
= pi
->nqsets
;
2588 err
= link_start(dev
);
2590 netif_tx_start_all_queues(dev
);
2594 static int cxgb_close(struct net_device
*dev
)
2596 struct port_info
*pi
= netdev_priv(dev
);
2597 struct adapter
*adapter
= pi
->adapter
;
2599 netif_tx_stop_all_queues(dev
);
2600 netif_carrier_off(dev
);
2601 return t4_enable_vi(adapter
, 0, pi
->viid
, false, false);
2604 static struct rtnl_link_stats64
*cxgb_get_stats(struct net_device
*dev
)
2606 struct port_stats stats
;
2607 struct port_info
*p
= netdev_priv(dev
);
2608 struct adapter
*adapter
= p
->adapter
;
2609 struct rtnl_link_stats64
*ns
= &dev
->stats64
;
2611 spin_lock(&adapter
->stats_lock
);
2612 t4_get_port_stats(adapter
, p
->tx_chan
, &stats
);
2613 spin_unlock(&adapter
->stats_lock
);
2615 ns
->tx_bytes
= stats
.tx_octets
;
2616 ns
->tx_packets
= stats
.tx_frames
;
2617 ns
->rx_bytes
= stats
.rx_octets
;
2618 ns
->rx_packets
= stats
.rx_frames
;
2619 ns
->multicast
= stats
.rx_mcast_frames
;
2621 /* detailed rx_errors */
2622 ns
->rx_length_errors
= stats
.rx_jabber
+ stats
.rx_too_long
+
2624 ns
->rx_over_errors
= 0;
2625 ns
->rx_crc_errors
= stats
.rx_fcs_err
;
2626 ns
->rx_frame_errors
= stats
.rx_symbol_err
;
2627 ns
->rx_fifo_errors
= stats
.rx_ovflow0
+ stats
.rx_ovflow1
+
2628 stats
.rx_ovflow2
+ stats
.rx_ovflow3
+
2629 stats
.rx_trunc0
+ stats
.rx_trunc1
+
2630 stats
.rx_trunc2
+ stats
.rx_trunc3
;
2631 ns
->rx_missed_errors
= 0;
2633 /* detailed tx_errors */
2634 ns
->tx_aborted_errors
= 0;
2635 ns
->tx_carrier_errors
= 0;
2636 ns
->tx_fifo_errors
= 0;
2637 ns
->tx_heartbeat_errors
= 0;
2638 ns
->tx_window_errors
= 0;
2640 ns
->tx_errors
= stats
.tx_error_frames
;
2641 ns
->rx_errors
= stats
.rx_symbol_err
+ stats
.rx_fcs_err
+
2642 ns
->rx_length_errors
+ stats
.rx_len_err
+ ns
->rx_fifo_errors
;
2646 static int cxgb_ioctl(struct net_device
*dev
, struct ifreq
*req
, int cmd
)
2648 int ret
= 0, prtad
, devad
;
2649 struct port_info
*pi
= netdev_priv(dev
);
2650 struct mii_ioctl_data
*data
= (struct mii_ioctl_data
*)&req
->ifr_data
;
2654 if (pi
->mdio_addr
< 0)
2656 data
->phy_id
= pi
->mdio_addr
;
2660 if (mdio_phy_id_is_c45(data
->phy_id
)) {
2661 prtad
= mdio_phy_id_prtad(data
->phy_id
);
2662 devad
= mdio_phy_id_devad(data
->phy_id
);
2663 } else if (data
->phy_id
< 32) {
2664 prtad
= data
->phy_id
;
2666 data
->reg_num
&= 0x1f;
2670 if (cmd
== SIOCGMIIREG
)
2671 ret
= t4_mdio_rd(pi
->adapter
, 0, prtad
, devad
,
2672 data
->reg_num
, &data
->val_out
);
2674 ret
= t4_mdio_wr(pi
->adapter
, 0, prtad
, devad
,
2675 data
->reg_num
, data
->val_in
);
2683 static void cxgb_set_rxmode(struct net_device
*dev
)
2685 /* unfortunately we can't return errors to the stack */
2686 set_rxmode(dev
, -1, false);
2689 static int cxgb_change_mtu(struct net_device
*dev
, int new_mtu
)
2692 struct port_info
*pi
= netdev_priv(dev
);
2694 if (new_mtu
< 81 || new_mtu
> MAX_MTU
) /* accommodate SACK */
2696 ret
= t4_set_rxmode(pi
->adapter
, 0, pi
->viid
, new_mtu
, -1, -1, -1, -1,
2703 static int cxgb_set_mac_addr(struct net_device
*dev
, void *p
)
2706 struct sockaddr
*addr
= p
;
2707 struct port_info
*pi
= netdev_priv(dev
);
2709 if (!is_valid_ether_addr(addr
->sa_data
))
2712 ret
= t4_change_mac(pi
->adapter
, 0, pi
->viid
, pi
->xact_addr_filt
,
2713 addr
->sa_data
, true, true);
2717 memcpy(dev
->dev_addr
, addr
->sa_data
, dev
->addr_len
);
2718 pi
->xact_addr_filt
= ret
;
2722 static void vlan_rx_register(struct net_device
*dev
, struct vlan_group
*grp
)
2724 struct port_info
*pi
= netdev_priv(dev
);
2727 t4_set_rxmode(pi
->adapter
, 0, pi
->viid
, -1, -1, -1, -1, grp
!= NULL
,
2731 #ifdef CONFIG_NET_POLL_CONTROLLER
2732 static void cxgb_netpoll(struct net_device
*dev
)
2734 struct port_info
*pi
= netdev_priv(dev
);
2735 struct adapter
*adap
= pi
->adapter
;
2737 if (adap
->flags
& USING_MSIX
) {
2739 struct sge_eth_rxq
*rx
= &adap
->sge
.ethrxq
[pi
->first_qset
];
2741 for (i
= pi
->nqsets
; i
; i
--, rx
++)
2742 t4_sge_intr_msix(0, &rx
->rspq
);
2744 t4_intr_handler(adap
)(0, adap
);
2748 static const struct net_device_ops cxgb4_netdev_ops
= {
2749 .ndo_open
= cxgb_open
,
2750 .ndo_stop
= cxgb_close
,
2751 .ndo_start_xmit
= t4_eth_xmit
,
2752 .ndo_get_stats64
= cxgb_get_stats
,
2753 .ndo_set_rx_mode
= cxgb_set_rxmode
,
2754 .ndo_set_mac_address
= cxgb_set_mac_addr
,
2755 .ndo_validate_addr
= eth_validate_addr
,
2756 .ndo_do_ioctl
= cxgb_ioctl
,
2757 .ndo_change_mtu
= cxgb_change_mtu
,
2758 .ndo_vlan_rx_register
= vlan_rx_register
,
2759 #ifdef CONFIG_NET_POLL_CONTROLLER
2760 .ndo_poll_controller
= cxgb_netpoll
,
2764 void t4_fatal_err(struct adapter
*adap
)
2766 t4_set_reg_field(adap
, SGE_CONTROL
, GLOBALENABLE
, 0);
2767 t4_intr_disable(adap
);
2768 dev_alert(adap
->pdev_dev
, "encountered fatal error, adapter stopped\n");
2771 static void setup_memwin(struct adapter
*adap
)
2775 bar0
= pci_resource_start(adap
->pdev
, 0); /* truncation intentional */
2776 t4_write_reg(adap
, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN
, 0),
2777 (bar0
+ MEMWIN0_BASE
) | BIR(0) |
2778 WINDOW(ilog2(MEMWIN0_APERTURE
) - 10));
2779 t4_write_reg(adap
, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN
, 1),
2780 (bar0
+ MEMWIN1_BASE
) | BIR(0) |
2781 WINDOW(ilog2(MEMWIN1_APERTURE
) - 10));
2782 t4_write_reg(adap
, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN
, 2),
2783 (bar0
+ MEMWIN2_BASE
) | BIR(0) |
2784 WINDOW(ilog2(MEMWIN2_APERTURE
) - 10));
2787 static int adap_init1(struct adapter
*adap
, struct fw_caps_config_cmd
*c
)
2792 /* get device capabilities */
2793 memset(c
, 0, sizeof(*c
));
2794 c
->op_to_write
= htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
2795 FW_CMD_REQUEST
| FW_CMD_READ
);
2796 c
->retval_len16
= htonl(FW_LEN16(*c
));
2797 ret
= t4_wr_mbox(adap
, 0, c
, sizeof(*c
), c
);
2801 /* select capabilities we'll be using */
2802 if (c
->niccaps
& htons(FW_CAPS_CONFIG_NIC_VM
)) {
2804 c
->niccaps
^= htons(FW_CAPS_CONFIG_NIC_VM
);
2806 c
->niccaps
= htons(FW_CAPS_CONFIG_NIC_VM
);
2807 } else if (vf_acls
) {
2808 dev_err(adap
->pdev_dev
, "virtualization ACLs not supported");
2811 c
->op_to_write
= htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
2812 FW_CMD_REQUEST
| FW_CMD_WRITE
);
2813 ret
= t4_wr_mbox(adap
, 0, c
, sizeof(*c
), NULL
);
2817 ret
= t4_config_glbl_rss(adap
, 0,
2818 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL
,
2819 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN
|
2820 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP
);
2824 ret
= t4_cfg_pfvf(adap
, 0, 0, 0, MAX_EGRQ
, 64, MAX_INGQ
, 0, 0, 4,
2825 0xf, 0xf, 16, FW_CMD_CAP_PF
, FW_CMD_CAP_PF
);
2831 /* get basic stuff going */
2832 ret
= t4_early_init(adap
, 0);
2836 /* tweak some settings */
2837 t4_write_reg(adap
, TP_SHIFT_CNT
, 0x64f8849);
2838 t4_write_reg(adap
, ULP_RX_TDDP_PSZ
, HPZ0(PAGE_SHIFT
- 12));
2839 t4_write_reg(adap
, TP_PIO_ADDR
, TP_INGRESS_CONFIG
);
2840 v
= t4_read_reg(adap
, TP_PIO_DATA
);
2841 t4_write_reg(adap
, TP_PIO_DATA
, v
& ~CSUM_HAS_PSEUDO_HDR
);
2847 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
2849 #define MAX_ATIDS 8192U
2852 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
2854 static int adap_init0(struct adapter
*adap
)
2858 enum dev_state state
;
2859 u32 params
[7], val
[7];
2860 struct fw_caps_config_cmd c
;
2862 ret
= t4_check_fw_version(adap
);
2863 if (ret
== -EINVAL
|| ret
> 0) {
2864 if (upgrade_fw(adap
) >= 0) /* recache FW version */
2865 ret
= t4_check_fw_version(adap
);
2870 /* contact FW, request master */
2871 ret
= t4_fw_hello(adap
, 0, 0, MASTER_MUST
, &state
);
2873 dev_err(adap
->pdev_dev
, "could not connect to FW, error %d\n",
2879 ret
= t4_fw_reset(adap
, 0, PIORSTMODE
| PIORST
);
2883 for (v
= 0; v
< SGE_NTIMERS
- 1; v
++)
2884 adap
->sge
.timer_val
[v
] = min(intr_holdoff
[v
], MAX_SGE_TIMERVAL
);
2885 adap
->sge
.timer_val
[SGE_NTIMERS
- 1] = MAX_SGE_TIMERVAL
;
2886 adap
->sge
.counter_val
[0] = 1;
2887 for (v
= 1; v
< SGE_NCOUNTERS
; v
++)
2888 adap
->sge
.counter_val
[v
] = min(intr_cnt
[v
- 1],
2890 #define FW_PARAM_DEV(param) \
2891 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2892 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
2894 params
[0] = FW_PARAM_DEV(CCLK
);
2895 ret
= t4_query_params(adap
, 0, 0, 0, 1, params
, val
);
2898 adap
->params
.vpd
.cclk
= val
[0];
2900 ret
= adap_init1(adap
, &c
);
2904 #define FW_PARAM_PFVF(param) \
2905 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
2906 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
2908 params
[0] = FW_PARAM_DEV(PORTVEC
);
2909 params
[1] = FW_PARAM_PFVF(L2T_START
);
2910 params
[2] = FW_PARAM_PFVF(L2T_END
);
2911 params
[3] = FW_PARAM_PFVF(FILTER_START
);
2912 params
[4] = FW_PARAM_PFVF(FILTER_END
);
2913 ret
= t4_query_params(adap
, 0, 0, 0, 5, params
, val
);
2917 adap
->tids
.ftid_base
= val
[3];
2918 adap
->tids
.nftids
= val
[4] - val
[3] + 1;
2921 /* query offload-related parameters */
2922 params
[0] = FW_PARAM_DEV(NTID
);
2923 params
[1] = FW_PARAM_PFVF(SERVER_START
);
2924 params
[2] = FW_PARAM_PFVF(SERVER_END
);
2925 params
[3] = FW_PARAM_PFVF(TDDP_START
);
2926 params
[4] = FW_PARAM_PFVF(TDDP_END
);
2927 params
[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ
);
2928 ret
= t4_query_params(adap
, 0, 0, 0, 6, params
, val
);
2931 adap
->tids
.ntids
= val
[0];
2932 adap
->tids
.natids
= min(adap
->tids
.ntids
/ 2, MAX_ATIDS
);
2933 adap
->tids
.stid_base
= val
[1];
2934 adap
->tids
.nstids
= val
[2] - val
[1] + 1;
2935 adap
->vres
.ddp
.start
= val
[3];
2936 adap
->vres
.ddp
.size
= val
[4] - val
[3] + 1;
2937 adap
->params
.ofldq_wr_cred
= val
[5];
2938 adap
->params
.offload
= 1;
2941 params
[0] = FW_PARAM_PFVF(STAG_START
);
2942 params
[1] = FW_PARAM_PFVF(STAG_END
);
2943 params
[2] = FW_PARAM_PFVF(RQ_START
);
2944 params
[3] = FW_PARAM_PFVF(RQ_END
);
2945 params
[4] = FW_PARAM_PFVF(PBL_START
);
2946 params
[5] = FW_PARAM_PFVF(PBL_END
);
2947 ret
= t4_query_params(adap
, 0, 0, 0, 6, params
, val
);
2950 adap
->vres
.stag
.start
= val
[0];
2951 adap
->vres
.stag
.size
= val
[1] - val
[0] + 1;
2952 adap
->vres
.rq
.start
= val
[2];
2953 adap
->vres
.rq
.size
= val
[3] - val
[2] + 1;
2954 adap
->vres
.pbl
.start
= val
[4];
2955 adap
->vres
.pbl
.size
= val
[5] - val
[4] + 1;
2957 params
[0] = FW_PARAM_PFVF(SQRQ_START
);
2958 params
[1] = FW_PARAM_PFVF(SQRQ_END
);
2959 params
[2] = FW_PARAM_PFVF(CQ_START
);
2960 params
[3] = FW_PARAM_PFVF(CQ_END
);
2961 ret
= t4_query_params(adap
, 0, 0, 0, 4, params
, val
);
2964 adap
->vres
.qp
.start
= val
[0];
2965 adap
->vres
.qp
.size
= val
[1] - val
[0] + 1;
2966 adap
->vres
.cq
.start
= val
[2];
2967 adap
->vres
.cq
.size
= val
[3] - val
[2] + 1;
2970 params
[0] = FW_PARAM_PFVF(ISCSI_START
);
2971 params
[1] = FW_PARAM_PFVF(ISCSI_END
);
2972 ret
= t4_query_params(adap
, 0, 0, 0, 2, params
, val
);
2975 adap
->vres
.iscsi
.start
= val
[0];
2976 adap
->vres
.iscsi
.size
= val
[1] - val
[0] + 1;
2978 #undef FW_PARAM_PFVF
2981 adap
->params
.nports
= hweight32(port_vec
);
2982 adap
->params
.portvec
= port_vec
;
2983 adap
->flags
|= FW_OK
;
2985 /* These are finalized by FW initialization, load their values now */
2986 v
= t4_read_reg(adap
, TP_TIMER_RESOLUTION
);
2987 adap
->params
.tp
.tre
= TIMERRESOLUTION_GET(v
);
2988 t4_read_mtu_tbl(adap
, adap
->params
.mtus
, NULL
);
2989 t4_load_mtus(adap
, adap
->params
.mtus
, adap
->params
.a_wnd
,
2990 adap
->params
.b_wnd
);
2992 #ifdef CONFIG_PCI_IOV
2994 * Provision resource limits for Virtual Functions. We currently
2995 * grant them all the same static resource limits except for the Port
2996 * Access Rights Mask which we're assigning based on the PF. All of
2997 * the static provisioning stuff for both the PF and VF really needs
2998 * to be managed in a persistent manner for each device which the
2999 * firmware controls.
3004 for (pf
= 0; pf
< ARRAY_SIZE(num_vf
); pf
++) {
3005 if (num_vf
[pf
] <= 0)
3008 /* VF numbering starts at 1! */
3009 for (vf
= 1; vf
<= num_vf
[pf
]; vf
++) {
3010 ret
= t4_cfg_pfvf(adap
, 0, pf
, vf
,
3011 VFRES_NEQ
, VFRES_NETHCTRL
,
3012 VFRES_NIQFLINT
, VFRES_NIQ
,
3013 VFRES_TC
, VFRES_NVI
,
3014 FW_PFVF_CMD_CMASK_MASK
,
3015 pfvfres_pmask(adap
, pf
, vf
),
3017 VFRES_R_CAPS
, VFRES_WX_CAPS
);
3019 dev_warn(adap
->pdev_dev
, "failed to "
3020 "provision pf/vf=%d/%d; "
3021 "err=%d\n", pf
, vf
, ret
);
3030 * If a command timed out or failed with EIO FW does not operate within
3031 * its spec or something catastrophic happened to HW/FW, stop issuing
3034 bye
: if (ret
!= -ETIMEDOUT
&& ret
!= -EIO
)
3041 static pci_ers_result_t
eeh_err_detected(struct pci_dev
*pdev
,
3042 pci_channel_state_t state
)
3045 struct adapter
*adap
= pci_get_drvdata(pdev
);
3051 adap
->flags
&= ~FW_OK
;
3052 notify_ulds(adap
, CXGB4_STATE_START_RECOVERY
);
3053 for_each_port(adap
, i
) {
3054 struct net_device
*dev
= adap
->port
[i
];
3056 netif_device_detach(dev
);
3057 netif_carrier_off(dev
);
3059 if (adap
->flags
& FULL_INIT_DONE
)
3062 pci_disable_device(pdev
);
3063 out
: return state
== pci_channel_io_perm_failure
?
3064 PCI_ERS_RESULT_DISCONNECT
: PCI_ERS_RESULT_NEED_RESET
;
3067 static pci_ers_result_t
eeh_slot_reset(struct pci_dev
*pdev
)
3070 struct fw_caps_config_cmd c
;
3071 struct adapter
*adap
= pci_get_drvdata(pdev
);
3074 pci_restore_state(pdev
);
3075 pci_save_state(pdev
);
3076 return PCI_ERS_RESULT_RECOVERED
;
3079 if (pci_enable_device(pdev
)) {
3080 dev_err(&pdev
->dev
, "cannot reenable PCI device after reset\n");
3081 return PCI_ERS_RESULT_DISCONNECT
;
3084 pci_set_master(pdev
);
3085 pci_restore_state(pdev
);
3086 pci_save_state(pdev
);
3087 pci_cleanup_aer_uncorrect_error_status(pdev
);
3089 if (t4_wait_dev_ready(adap
) < 0)
3090 return PCI_ERS_RESULT_DISCONNECT
;
3091 if (t4_fw_hello(adap
, 0, 0, MASTER_MUST
, NULL
))
3092 return PCI_ERS_RESULT_DISCONNECT
;
3093 adap
->flags
|= FW_OK
;
3094 if (adap_init1(adap
, &c
))
3095 return PCI_ERS_RESULT_DISCONNECT
;
3097 for_each_port(adap
, i
) {
3098 struct port_info
*p
= adap2pinfo(adap
, i
);
3100 ret
= t4_alloc_vi(adap
, 0, p
->tx_chan
, 0, 0, 1, NULL
, NULL
);
3102 return PCI_ERS_RESULT_DISCONNECT
;
3104 p
->xact_addr_filt
= -1;
3107 t4_load_mtus(adap
, adap
->params
.mtus
, adap
->params
.a_wnd
,
3108 adap
->params
.b_wnd
);
3110 return PCI_ERS_RESULT_DISCONNECT
;
3111 return PCI_ERS_RESULT_RECOVERED
;
3114 static void eeh_resume(struct pci_dev
*pdev
)
3117 struct adapter
*adap
= pci_get_drvdata(pdev
);
3123 for_each_port(adap
, i
) {
3124 struct net_device
*dev
= adap
->port
[i
];
3126 if (netif_running(dev
)) {
3128 cxgb_set_rxmode(dev
);
3130 netif_device_attach(dev
);
3135 static struct pci_error_handlers cxgb4_eeh
= {
3136 .error_detected
= eeh_err_detected
,
3137 .slot_reset
= eeh_slot_reset
,
3138 .resume
= eeh_resume
,
3141 static inline bool is_10g_port(const struct link_config
*lc
)
3143 return (lc
->supported
& FW_PORT_CAP_SPEED_10G
) != 0;
3146 static inline void init_rspq(struct sge_rspq
*q
, u8 timer_idx
, u8 pkt_cnt_idx
,
3147 unsigned int size
, unsigned int iqe_size
)
3149 q
->intr_params
= QINTR_TIMER_IDX(timer_idx
) |
3150 (pkt_cnt_idx
< SGE_NCOUNTERS
? QINTR_CNT_EN
: 0);
3151 q
->pktcnt_idx
= pkt_cnt_idx
< SGE_NCOUNTERS
? pkt_cnt_idx
: 0;
3152 q
->iqe_len
= iqe_size
;
3157 * Perform default configuration of DMA queues depending on the number and type
3158 * of ports we found and the number of available CPUs. Most settings can be
3159 * modified by the admin prior to actual use.
3161 static void __devinit
cfg_queues(struct adapter
*adap
)
3163 struct sge
*s
= &adap
->sge
;
3164 int i
, q10g
= 0, n10g
= 0, qidx
= 0;
3166 for_each_port(adap
, i
)
3167 n10g
+= is_10g_port(&adap2pinfo(adap
, i
)->link_cfg
);
3170 * We default to 1 queue per non-10G port and up to # of cores queues
3174 q10g
= (MAX_ETH_QSETS
- (adap
->params
.nports
- n10g
)) / n10g
;
3175 if (q10g
> num_online_cpus())
3176 q10g
= num_online_cpus();
3178 for_each_port(adap
, i
) {
3179 struct port_info
*pi
= adap2pinfo(adap
, i
);
3181 pi
->first_qset
= qidx
;
3182 pi
->nqsets
= is_10g_port(&pi
->link_cfg
) ? q10g
: 1;
3187 s
->max_ethqsets
= qidx
; /* MSI-X may lower it later */
3189 if (is_offload(adap
)) {
3191 * For offload we use 1 queue/channel if all ports are up to 1G,
3192 * otherwise we divide all available queues amongst the channels
3193 * capped by the number of available cores.
3196 i
= min_t(int, ARRAY_SIZE(s
->ofldrxq
),
3198 s
->ofldqsets
= roundup(i
, adap
->params
.nports
);
3200 s
->ofldqsets
= adap
->params
.nports
;
3201 /* For RDMA one Rx queue per channel suffices */
3202 s
->rdmaqs
= adap
->params
.nports
;
3205 for (i
= 0; i
< ARRAY_SIZE(s
->ethrxq
); i
++) {
3206 struct sge_eth_rxq
*r
= &s
->ethrxq
[i
];
3208 init_rspq(&r
->rspq
, 0, 0, 1024, 64);
3212 for (i
= 0; i
< ARRAY_SIZE(s
->ethtxq
); i
++)
3213 s
->ethtxq
[i
].q
.size
= 1024;
3215 for (i
= 0; i
< ARRAY_SIZE(s
->ctrlq
); i
++)
3216 s
->ctrlq
[i
].q
.size
= 512;
3218 for (i
= 0; i
< ARRAY_SIZE(s
->ofldtxq
); i
++)
3219 s
->ofldtxq
[i
].q
.size
= 1024;
3221 for (i
= 0; i
< ARRAY_SIZE(s
->ofldrxq
); i
++) {
3222 struct sge_ofld_rxq
*r
= &s
->ofldrxq
[i
];
3224 init_rspq(&r
->rspq
, 0, 0, 1024, 64);
3225 r
->rspq
.uld
= CXGB4_ULD_ISCSI
;
3229 for (i
= 0; i
< ARRAY_SIZE(s
->rdmarxq
); i
++) {
3230 struct sge_ofld_rxq
*r
= &s
->rdmarxq
[i
];
3232 init_rspq(&r
->rspq
, 0, 0, 511, 64);
3233 r
->rspq
.uld
= CXGB4_ULD_RDMA
;
3237 init_rspq(&s
->fw_evtq
, 6, 0, 512, 64);
3238 init_rspq(&s
->intrq
, 6, 0, 2 * MAX_INGQ
, 64);
3242 * Reduce the number of Ethernet queues across all ports to at most n.
3243 * n provides at least one queue per port.
3245 static void __devinit
reduce_ethqs(struct adapter
*adap
, int n
)
3248 struct port_info
*pi
;
3250 while (n
< adap
->sge
.ethqsets
)
3251 for_each_port(adap
, i
) {
3252 pi
= adap2pinfo(adap
, i
);
3253 if (pi
->nqsets
> 1) {
3255 adap
->sge
.ethqsets
--;
3256 if (adap
->sge
.ethqsets
<= n
)
3262 for_each_port(adap
, i
) {
3263 pi
= adap2pinfo(adap
, i
);
3269 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
3270 #define EXTRA_VECS 2
3272 static int __devinit
enable_msix(struct adapter
*adap
)
3275 int i
, err
, want
, need
;
3276 struct sge
*s
= &adap
->sge
;
3277 unsigned int nchan
= adap
->params
.nports
;
3278 struct msix_entry entries
[MAX_INGQ
+ 1];
3280 for (i
= 0; i
< ARRAY_SIZE(entries
); ++i
)
3281 entries
[i
].entry
= i
;
3283 want
= s
->max_ethqsets
+ EXTRA_VECS
;
3284 if (is_offload(adap
)) {
3285 want
+= s
->rdmaqs
+ s
->ofldqsets
;
3286 /* need nchan for each possible ULD */
3287 ofld_need
= 2 * nchan
;
3289 need
= adap
->params
.nports
+ EXTRA_VECS
+ ofld_need
;
3291 while ((err
= pci_enable_msix(adap
->pdev
, entries
, want
)) >= need
)
3296 * Distribute available vectors to the various queue groups.
3297 * Every group gets its minimum requirement and NIC gets top
3298 * priority for leftovers.
3300 i
= want
- EXTRA_VECS
- ofld_need
;
3301 if (i
< s
->max_ethqsets
) {
3302 s
->max_ethqsets
= i
;
3303 if (i
< s
->ethqsets
)
3304 reduce_ethqs(adap
, i
);
3306 if (is_offload(adap
)) {
3307 i
= want
- EXTRA_VECS
- s
->max_ethqsets
;
3308 i
-= ofld_need
- nchan
;
3309 s
->ofldqsets
= (i
/ nchan
) * nchan
; /* round down */
3311 for (i
= 0; i
< want
; ++i
)
3312 adap
->msix_info
[i
].vec
= entries
[i
].vector
;
3314 dev_info(adap
->pdev_dev
,
3315 "only %d MSI-X vectors left, not using MSI-X\n", err
);
3321 static void __devinit
print_port_info(struct adapter
*adap
)
3323 static const char *base
[] = {
3324 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
3325 "KX", "KR", "KR SFP+", "KR FEC"
3330 const char *spd
= "";
3332 if (adap
->params
.pci
.speed
== PCI_EXP_LNKSTA_CLS_2_5GB
)
3334 else if (adap
->params
.pci
.speed
== PCI_EXP_LNKSTA_CLS_5_0GB
)
3337 for_each_port(adap
, i
) {
3338 struct net_device
*dev
= adap
->port
[i
];
3339 const struct port_info
*pi
= netdev_priv(dev
);
3342 if (!test_bit(i
, &adap
->registered_device_map
))
3345 if (pi
->link_cfg
.supported
& FW_PORT_CAP_SPEED_100M
)
3346 bufp
+= sprintf(bufp
, "100/");
3347 if (pi
->link_cfg
.supported
& FW_PORT_CAP_SPEED_1G
)
3348 bufp
+= sprintf(bufp
, "1000/");
3349 if (pi
->link_cfg
.supported
& FW_PORT_CAP_SPEED_10G
)
3350 bufp
+= sprintf(bufp
, "10G/");
3353 sprintf(bufp
, "BASE-%s", base
[pi
->port_type
]);
3355 netdev_info(dev
, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
3356 adap
->params
.vpd
.id
, adap
->params
.rev
,
3357 buf
, is_offload(adap
) ? "R" : "",
3358 adap
->params
.pci
.width
, spd
,
3359 (adap
->flags
& USING_MSIX
) ? " MSI-X" :
3360 (adap
->flags
& USING_MSI
) ? " MSI" : "");
3361 if (adap
->name
== dev
->name
)
3362 netdev_info(dev
, "S/N: %s, E/C: %s\n",
3363 adap
->params
.vpd
.sn
, adap
->params
.vpd
.ec
);
3367 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO | NETIF_F_TSO6 |\
3368 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
3370 static int __devinit
init_one(struct pci_dev
*pdev
,
3371 const struct pci_device_id
*ent
)
3374 struct port_info
*pi
;
3375 unsigned int highdma
= 0;
3376 struct adapter
*adapter
= NULL
;
3378 printk_once(KERN_INFO
"%s - version %s\n", DRV_DESC
, DRV_VERSION
);
3380 err
= pci_request_regions(pdev
, KBUILD_MODNAME
);
3382 /* Just info, some other driver may have claimed the device. */
3383 dev_info(&pdev
->dev
, "cannot obtain PCI resources\n");
3387 /* We control everything through PF 0 */
3388 func
= PCI_FUNC(pdev
->devfn
);
3390 pci_save_state(pdev
); /* to restore SR-IOV later */
3394 err
= pci_enable_device(pdev
);
3396 dev_err(&pdev
->dev
, "cannot enable PCI device\n");
3397 goto out_release_regions
;
3400 if (!pci_set_dma_mask(pdev
, DMA_BIT_MASK(64))) {
3401 highdma
= NETIF_F_HIGHDMA
;
3402 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
3404 dev_err(&pdev
->dev
, "unable to obtain 64-bit DMA for "
3405 "coherent allocations\n");
3406 goto out_disable_device
;
3409 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
3411 dev_err(&pdev
->dev
, "no usable DMA configuration\n");
3412 goto out_disable_device
;
3416 pci_enable_pcie_error_reporting(pdev
);
3417 pci_set_master(pdev
);
3418 pci_save_state(pdev
);
3420 adapter
= kzalloc(sizeof(*adapter
), GFP_KERNEL
);
3423 goto out_disable_device
;
3426 adapter
->regs
= pci_ioremap_bar(pdev
, 0);
3427 if (!adapter
->regs
) {
3428 dev_err(&pdev
->dev
, "cannot map device registers\n");
3430 goto out_free_adapter
;
3433 adapter
->pdev
= pdev
;
3434 adapter
->pdev_dev
= &pdev
->dev
;
3435 adapter
->name
= pci_name(pdev
);
3436 adapter
->msg_enable
= dflt_msg_enable
;
3437 memset(adapter
->chan_map
, 0xff, sizeof(adapter
->chan_map
));
3439 spin_lock_init(&adapter
->stats_lock
);
3440 spin_lock_init(&adapter
->tid_release_lock
);
3442 INIT_WORK(&adapter
->tid_release_task
, process_tid_release_list
);
3444 err
= t4_prep_adapter(adapter
);
3447 err
= adap_init0(adapter
);
3451 for_each_port(adapter
, i
) {
3452 struct net_device
*netdev
;
3454 netdev
= alloc_etherdev_mq(sizeof(struct port_info
),
3461 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
3463 adapter
->port
[i
] = netdev
;
3464 pi
= netdev_priv(netdev
);
3465 pi
->adapter
= adapter
;
3466 pi
->xact_addr_filt
= -1;
3467 pi
->rx_offload
= RX_CSO
;
3469 netif_carrier_off(netdev
);
3470 netif_tx_stop_all_queues(netdev
);
3471 netdev
->irq
= pdev
->irq
;
3473 netdev
->features
|= NETIF_F_SG
| NETIF_F_TSO
| NETIF_F_TSO6
;
3474 netdev
->features
|= NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
;
3475 netdev
->features
|= NETIF_F_GRO
| NETIF_F_RXHASH
| highdma
;
3476 netdev
->features
|= NETIF_F_HW_VLAN_TX
| NETIF_F_HW_VLAN_RX
;
3477 netdev
->vlan_features
= netdev
->features
& VLAN_FEAT
;
3479 netdev
->netdev_ops
= &cxgb4_netdev_ops
;
3480 SET_ETHTOOL_OPS(netdev
, &cxgb_ethtool_ops
);
3483 pci_set_drvdata(pdev
, adapter
);
3485 if (adapter
->flags
& FW_OK
) {
3486 err
= t4_port_init(adapter
, 0, 0, 0);
3492 * Configure queues and allocate tables now, they can be needed as
3493 * soon as the first register_netdev completes.
3495 cfg_queues(adapter
);
3497 adapter
->l2t
= t4_init_l2t();
3498 if (!adapter
->l2t
) {
3499 /* We tolerate a lack of L2T, giving up some functionality */
3500 dev_warn(&pdev
->dev
, "could not allocate L2T, continuing\n");
3501 adapter
->params
.offload
= 0;
3504 if (is_offload(adapter
) && tid_init(&adapter
->tids
) < 0) {
3505 dev_warn(&pdev
->dev
, "could not allocate TID table, "
3507 adapter
->params
.offload
= 0;
3511 * The card is now ready to go. If any errors occur during device
3512 * registration we do not fail the whole card but rather proceed only
3513 * with the ports we manage to register successfully. However we must
3514 * register at least one net device.
3516 for_each_port(adapter
, i
) {
3517 err
= register_netdev(adapter
->port
[i
]);
3519 dev_warn(&pdev
->dev
,
3520 "cannot register net device %s, skipping\n",
3521 adapter
->port
[i
]->name
);
3524 * Change the name we use for messages to the name of
3525 * the first successfully registered interface.
3527 if (!adapter
->registered_device_map
)
3528 adapter
->name
= adapter
->port
[i
]->name
;
3530 __set_bit(i
, &adapter
->registered_device_map
);
3531 adapter
->chan_map
[adap2pinfo(adapter
, i
)->tx_chan
] = i
;
3534 if (!adapter
->registered_device_map
) {
3535 dev_err(&pdev
->dev
, "could not register any net devices\n");
3539 if (cxgb4_debugfs_root
) {
3540 adapter
->debugfs_root
= debugfs_create_dir(pci_name(pdev
),
3541 cxgb4_debugfs_root
);
3542 setup_debugfs(adapter
);
3545 /* See what interrupts we'll be using */
3546 if (msi
> 1 && enable_msix(adapter
) == 0)
3547 adapter
->flags
|= USING_MSIX
;
3548 else if (msi
> 0 && pci_enable_msi(pdev
) == 0)
3549 adapter
->flags
|= USING_MSI
;
3551 if (is_offload(adapter
))
3552 attach_ulds(adapter
);
3554 print_port_info(adapter
);
3557 #ifdef CONFIG_PCI_IOV
3558 if (func
< ARRAY_SIZE(num_vf
) && num_vf
[func
] > 0)
3559 if (pci_enable_sriov(pdev
, num_vf
[func
]) == 0)
3560 dev_info(&pdev
->dev
,
3561 "instantiated %u virtual functions\n",
3567 t4_free_mem(adapter
->tids
.tid_tab
);
3568 t4_free_mem(adapter
->l2t
);
3569 for_each_port(adapter
, i
)
3570 if (adapter
->port
[i
])
3571 free_netdev(adapter
->port
[i
]);
3572 if (adapter
->flags
& FW_OK
)
3573 t4_fw_bye(adapter
, 0);
3575 iounmap(adapter
->regs
);
3579 pci_disable_pcie_error_reporting(pdev
);
3580 pci_disable_device(pdev
);
3581 out_release_regions
:
3582 pci_release_regions(pdev
);
3583 pci_set_drvdata(pdev
, NULL
);
3587 static void __devexit
remove_one(struct pci_dev
*pdev
)
3589 struct adapter
*adapter
= pci_get_drvdata(pdev
);
3591 pci_disable_sriov(pdev
);
3596 if (is_offload(adapter
))
3597 detach_ulds(adapter
);
3599 for_each_port(adapter
, i
)
3600 if (test_bit(i
, &adapter
->registered_device_map
))
3601 unregister_netdev(adapter
->port
[i
]);
3603 if (adapter
->debugfs_root
)
3604 debugfs_remove_recursive(adapter
->debugfs_root
);
3606 if (adapter
->flags
& FULL_INIT_DONE
)
3608 t4_free_mem(adapter
->l2t
);
3609 t4_free_mem(adapter
->tids
.tid_tab
);
3610 disable_msi(adapter
);
3612 for_each_port(adapter
, i
)
3613 if (adapter
->port
[i
])
3614 free_netdev(adapter
->port
[i
]);
3616 if (adapter
->flags
& FW_OK
)
3617 t4_fw_bye(adapter
, 0);
3618 iounmap(adapter
->regs
);
3620 pci_disable_pcie_error_reporting(pdev
);
3621 pci_disable_device(pdev
);
3622 pci_release_regions(pdev
);
3623 pci_set_drvdata(pdev
, NULL
);
3624 } else if (PCI_FUNC(pdev
->devfn
) > 0)
3625 pci_release_regions(pdev
);
3628 static struct pci_driver cxgb4_driver
= {
3629 .name
= KBUILD_MODNAME
,
3630 .id_table
= cxgb4_pci_tbl
,
3632 .remove
= __devexit_p(remove_one
),
3633 .err_handler
= &cxgb4_eeh
,
3636 static int __init
cxgb4_init_module(void)
3640 /* Debugfs support is optional, just warn if this fails */
3641 cxgb4_debugfs_root
= debugfs_create_dir(KBUILD_MODNAME
, NULL
);
3642 if (!cxgb4_debugfs_root
)
3643 pr_warning("could not create debugfs entry, continuing\n");
3645 ret
= pci_register_driver(&cxgb4_driver
);
3647 debugfs_remove(cxgb4_debugfs_root
);
3651 static void __exit
cxgb4_cleanup_module(void)
3653 pci_unregister_driver(&cxgb4_driver
);
3654 debugfs_remove(cxgb4_debugfs_root
); /* NULL ok */
3657 module_init(cxgb4_init_module
);
3658 module_exit(cxgb4_cleanup_module
);