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>
45 #include <linux/if_vlan.h>
46 #include <linux/init.h>
47 #include <linux/log2.h>
48 #include <linux/mdio.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/mutex.h>
52 #include <linux/netdevice.h>
53 #include <linux/pci.h>
54 #include <linux/aer.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/sched.h>
57 #include <linux/seq_file.h>
58 #include <linux/sockios.h>
59 #include <linux/vmalloc.h>
60 #include <linux/workqueue.h>
61 #include <net/neighbour.h>
62 #include <net/netevent.h>
63 #include <net/addrconf.h>
64 #include <asm/uaccess.h>
72 #include <../drivers/net/bonding/bonding.h>
77 #define DRV_VERSION "2.0.0-ko"
78 #define DRV_DESC "Chelsio T4/T5 Network Driver"
81 * Max interrupt hold-off timer value in us. Queues fall back to this value
82 * under extreme memory pressure so it's largish to give the system time to
85 #define MAX_SGE_TIMERVAL 200U
89 * Physical Function provisioning constants.
91 PFRES_NVI
= 4, /* # of Virtual Interfaces */
92 PFRES_NETHCTRL
= 128, /* # of EQs used for ETH or CTRL Qs */
93 PFRES_NIQFLINT
= 128, /* # of ingress Qs/w Free List(s)/intr
95 PFRES_NEQ
= 256, /* # of egress queues */
96 PFRES_NIQ
= 0, /* # of ingress queues */
97 PFRES_TC
= 0, /* PCI-E traffic class */
98 PFRES_NEXACTF
= 128, /* # of exact MPS filters */
100 PFRES_R_CAPS
= FW_CMD_CAP_PF
,
101 PFRES_WX_CAPS
= FW_CMD_CAP_PF
,
103 #ifdef CONFIG_PCI_IOV
105 * Virtual Function provisioning constants. We need two extra Ingress
106 * Queues with Interrupt capability to serve as the VF's Firmware
107 * Event Queue and Forwarded Interrupt Queue (when using MSI mode) --
108 * neither will have Free Lists associated with them). For each
109 * Ethernet/Control Egress Queue and for each Free List, we need an
112 VFRES_NPORTS
= 1, /* # of "ports" per VF */
113 VFRES_NQSETS
= 2, /* # of "Queue Sets" per VF */
115 VFRES_NVI
= VFRES_NPORTS
, /* # of Virtual Interfaces */
116 VFRES_NETHCTRL
= VFRES_NQSETS
, /* # of EQs used for ETH or CTRL Qs */
117 VFRES_NIQFLINT
= VFRES_NQSETS
+2,/* # of ingress Qs/w Free List(s)/intr */
118 VFRES_NEQ
= VFRES_NQSETS
*2, /* # of egress queues */
119 VFRES_NIQ
= 0, /* # of non-fl/int ingress queues */
120 VFRES_TC
= 0, /* PCI-E traffic class */
121 VFRES_NEXACTF
= 16, /* # of exact MPS filters */
123 VFRES_R_CAPS
= FW_CMD_CAP_DMAQ
|FW_CMD_CAP_VF
|FW_CMD_CAP_PORT
,
124 VFRES_WX_CAPS
= FW_CMD_CAP_DMAQ
|FW_CMD_CAP_VF
,
129 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
130 * static and likely not to be useful in the long run. We really need to
131 * implement some form of persistent configuration which the firmware
134 static unsigned int pfvfres_pmask(struct adapter
*adapter
,
135 unsigned int pf
, unsigned int vf
)
137 unsigned int portn
, portvec
;
140 * Give PF's access to all of the ports.
143 return FW_PFVF_CMD_PMASK_MASK
;
146 * For VFs, we'll assign them access to the ports based purely on the
147 * PF. We assign active ports in order, wrapping around if there are
148 * fewer active ports than PFs: e.g. active port[pf % nports].
149 * Unfortunately the adapter's port_info structs haven't been
150 * initialized yet so we have to compute this.
152 if (adapter
->params
.nports
== 0)
155 portn
= pf
% adapter
->params
.nports
;
156 portvec
= adapter
->params
.portvec
;
159 * Isolate the lowest set bit in the port vector. If we're at
160 * the port number that we want, return that as the pmask.
161 * otherwise mask that bit out of the port vector and
162 * decrement our port number ...
164 unsigned int pmask
= portvec
^ (portvec
& (portvec
-1));
174 MAX_TXQ_ENTRIES
= 16384,
175 MAX_CTRL_TXQ_ENTRIES
= 1024,
176 MAX_RSPQ_ENTRIES
= 16384,
177 MAX_RX_BUFFERS
= 16384,
178 MIN_TXQ_ENTRIES
= 32,
179 MIN_CTRL_TXQ_ENTRIES
= 32,
180 MIN_RSPQ_ENTRIES
= 128,
184 /* Host shadow copy of ingress filter entry. This is in host native format
185 * and doesn't match the ordering or bit order, etc. of the hardware of the
186 * firmware command. The use of bit-field structure elements is purely to
187 * remind ourselves of the field size limitations and save memory in the case
188 * where the filter table is large.
190 struct filter_entry
{
191 /* Administrative fields for filter.
193 u32 valid
:1; /* filter allocated and valid */
194 u32 locked
:1; /* filter is administratively locked */
196 u32 pending
:1; /* filter action is pending firmware reply */
197 u32 smtidx
:8; /* Source MAC Table index for smac */
198 struct l2t_entry
*l2t
; /* Layer Two Table entry for dmac */
200 /* The filter itself. Most of this is a straight copy of information
201 * provided by the extended ioctl(). Some fields are translated to
202 * internal forms -- for instance the Ingress Queue ID passed in from
203 * the ioctl() is translated into the Absolute Ingress Queue ID.
205 struct ch_filter_specification fs
;
208 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
209 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
210 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
212 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
214 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl
) = {
215 CH_DEVICE(0xa000, 0), /* PE10K */
216 CH_DEVICE(0x4001, -1),
217 CH_DEVICE(0x4002, -1),
218 CH_DEVICE(0x4003, -1),
219 CH_DEVICE(0x4004, -1),
220 CH_DEVICE(0x4005, -1),
221 CH_DEVICE(0x4006, -1),
222 CH_DEVICE(0x4007, -1),
223 CH_DEVICE(0x4008, -1),
224 CH_DEVICE(0x4009, -1),
225 CH_DEVICE(0x400a, -1),
226 CH_DEVICE(0x4401, 4),
227 CH_DEVICE(0x4402, 4),
228 CH_DEVICE(0x4403, 4),
229 CH_DEVICE(0x4404, 4),
230 CH_DEVICE(0x4405, 4),
231 CH_DEVICE(0x4406, 4),
232 CH_DEVICE(0x4407, 4),
233 CH_DEVICE(0x4408, 4),
234 CH_DEVICE(0x4409, 4),
235 CH_DEVICE(0x440a, 4),
236 CH_DEVICE(0x440d, 4),
237 CH_DEVICE(0x440e, 4),
238 CH_DEVICE(0x5001, 4),
239 CH_DEVICE(0x5002, 4),
240 CH_DEVICE(0x5003, 4),
241 CH_DEVICE(0x5004, 4),
242 CH_DEVICE(0x5005, 4),
243 CH_DEVICE(0x5006, 4),
244 CH_DEVICE(0x5007, 4),
245 CH_DEVICE(0x5008, 4),
246 CH_DEVICE(0x5009, 4),
247 CH_DEVICE(0x500A, 4),
248 CH_DEVICE(0x500B, 4),
249 CH_DEVICE(0x500C, 4),
250 CH_DEVICE(0x500D, 4),
251 CH_DEVICE(0x500E, 4),
252 CH_DEVICE(0x500F, 4),
253 CH_DEVICE(0x5010, 4),
254 CH_DEVICE(0x5011, 4),
255 CH_DEVICE(0x5012, 4),
256 CH_DEVICE(0x5013, 4),
257 CH_DEVICE(0x5401, 4),
258 CH_DEVICE(0x5402, 4),
259 CH_DEVICE(0x5403, 4),
260 CH_DEVICE(0x5404, 4),
261 CH_DEVICE(0x5405, 4),
262 CH_DEVICE(0x5406, 4),
263 CH_DEVICE(0x5407, 4),
264 CH_DEVICE(0x5408, 4),
265 CH_DEVICE(0x5409, 4),
266 CH_DEVICE(0x540A, 4),
267 CH_DEVICE(0x540B, 4),
268 CH_DEVICE(0x540C, 4),
269 CH_DEVICE(0x540D, 4),
270 CH_DEVICE(0x540E, 4),
271 CH_DEVICE(0x540F, 4),
272 CH_DEVICE(0x5410, 4),
273 CH_DEVICE(0x5411, 4),
274 CH_DEVICE(0x5412, 4),
275 CH_DEVICE(0x5413, 4),
279 #define FW4_FNAME "cxgb4/t4fw.bin"
280 #define FW5_FNAME "cxgb4/t5fw.bin"
281 #define FW4_CFNAME "cxgb4/t4-config.txt"
282 #define FW5_CFNAME "cxgb4/t5-config.txt"
284 MODULE_DESCRIPTION(DRV_DESC
);
285 MODULE_AUTHOR("Chelsio Communications");
286 MODULE_LICENSE("Dual BSD/GPL");
287 MODULE_VERSION(DRV_VERSION
);
288 MODULE_DEVICE_TABLE(pci
, cxgb4_pci_tbl
);
289 MODULE_FIRMWARE(FW4_FNAME
);
290 MODULE_FIRMWARE(FW5_FNAME
);
293 * Normally we're willing to become the firmware's Master PF but will be happy
294 * if another PF has already become the Master and initialized the adapter.
295 * Setting "force_init" will cause this driver to forcibly establish itself as
296 * the Master PF and initialize the adapter.
298 static uint force_init
;
300 module_param(force_init
, uint
, 0644);
301 MODULE_PARM_DESC(force_init
, "Forcibly become Master PF and initialize adapter");
304 * Normally if the firmware we connect to has Configuration File support, we
305 * use that and only fall back to the old Driver-based initialization if the
306 * Configuration File fails for some reason. If force_old_init is set, then
307 * we'll always use the old Driver-based initialization sequence.
309 static uint force_old_init
;
311 module_param(force_old_init
, uint
, 0644);
312 MODULE_PARM_DESC(force_old_init
, "Force old initialization sequence");
314 static int dflt_msg_enable
= DFLT_MSG_ENABLE
;
316 module_param(dflt_msg_enable
, int, 0644);
317 MODULE_PARM_DESC(dflt_msg_enable
, "Chelsio T4 default message enable bitmap");
320 * The driver uses the best interrupt scheme available on a platform in the
321 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
322 * of these schemes the driver may consider as follows:
324 * msi = 2: choose from among all three options
325 * msi = 1: only consider MSI and INTx interrupts
326 * msi = 0: force INTx interrupts
330 module_param(msi
, int, 0644);
331 MODULE_PARM_DESC(msi
, "whether to use INTx (0), MSI (1) or MSI-X (2)");
334 * Queue interrupt hold-off timer values. Queues default to the first of these
337 static unsigned int intr_holdoff
[SGE_NTIMERS
- 1] = { 5, 10, 20, 50, 100 };
339 module_param_array(intr_holdoff
, uint
, NULL
, 0644);
340 MODULE_PARM_DESC(intr_holdoff
, "values for queue interrupt hold-off timers "
341 "0..4 in microseconds");
343 static unsigned int intr_cnt
[SGE_NCOUNTERS
- 1] = { 4, 8, 16 };
345 module_param_array(intr_cnt
, uint
, NULL
, 0644);
346 MODULE_PARM_DESC(intr_cnt
,
347 "thresholds 1..3 for queue interrupt packet counters");
350 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
351 * offset by 2 bytes in order to have the IP headers line up on 4-byte
352 * boundaries. This is a requirement for many architectures which will throw
353 * a machine check fault if an attempt is made to access one of the 4-byte IP
354 * header fields on a non-4-byte boundary. And it's a major performance issue
355 * even on some architectures which allow it like some implementations of the
356 * x86 ISA. However, some architectures don't mind this and for some very
357 * edge-case performance sensitive applications (like forwarding large volumes
358 * of small packets), setting this DMA offset to 0 will decrease the number of
359 * PCI-E Bus transfers enough to measurably affect performance.
361 static int rx_dma_offset
= 2;
365 #ifdef CONFIG_PCI_IOV
366 module_param(vf_acls
, bool, 0644);
367 MODULE_PARM_DESC(vf_acls
, "if set enable virtualization L2 ACL enforcement");
369 /* Configure the number of PCI-E Virtual Function which are to be instantiated
370 * on SR-IOV Capable Physical Functions.
372 static unsigned int num_vf
[NUM_OF_PF_WITH_SRIOV
];
374 module_param_array(num_vf
, uint
, NULL
, 0644);
375 MODULE_PARM_DESC(num_vf
, "number of VFs for each of PFs 0-3");
379 * The filter TCAM has a fixed portion and a variable portion. The fixed
380 * portion can match on source/destination IP IPv4/IPv6 addresses and TCP/UDP
381 * ports. The variable portion is 36 bits which can include things like Exact
382 * Match MAC Index (9 bits), Ether Type (16 bits), IP Protocol (8 bits),
383 * [Inner] VLAN Tag (17 bits), etc. which, if all were somehow selected, would
384 * far exceed the 36-bit budget for this "compressed" header portion of the
385 * filter. Thus, we have a scarce resource which must be carefully managed.
387 * By default we set this up to mostly match the set of filter matching
388 * capabilities of T3 but with accommodations for some of T4's more
389 * interesting features:
391 * { IP Fragment (1), MPS Match Type (3), IP Protocol (8),
392 * [Inner] VLAN (17), Port (3), FCoE (1) }
395 TP_VLAN_PRI_MAP_DEFAULT
= HW_TPL_FR_MT_PR_IV_P_FC
,
396 TP_VLAN_PRI_MAP_FIRST
= FCOE_SHIFT
,
397 TP_VLAN_PRI_MAP_LAST
= FRAGMENTATION_SHIFT
,
400 static unsigned int tp_vlan_pri_map
= TP_VLAN_PRI_MAP_DEFAULT
;
402 module_param(tp_vlan_pri_map
, uint
, 0644);
403 MODULE_PARM_DESC(tp_vlan_pri_map
, "global compressed filter configuration");
405 static struct dentry
*cxgb4_debugfs_root
;
407 static LIST_HEAD(adapter_list
);
408 static DEFINE_MUTEX(uld_mutex
);
409 /* Adapter list to be accessed from atomic context */
410 static LIST_HEAD(adap_rcu_list
);
411 static DEFINE_SPINLOCK(adap_rcu_lock
);
412 static struct cxgb4_uld_info ulds
[CXGB4_ULD_MAX
];
413 static const char *uld_str
[] = { "RDMA", "iSCSI" };
415 static void link_report(struct net_device
*dev
)
417 if (!netif_carrier_ok(dev
))
418 netdev_info(dev
, "link down\n");
420 static const char *fc
[] = { "no", "Rx", "Tx", "Tx/Rx" };
422 const char *s
= "10Mbps";
423 const struct port_info
*p
= netdev_priv(dev
);
425 switch (p
->link_cfg
.speed
) {
435 case 40000: /* Need a SPEED_40000 in ethtool.h */
440 netdev_info(dev
, "link up, %s, full-duplex, %s PAUSE\n", s
,
445 void t4_os_link_changed(struct adapter
*adapter
, int port_id
, int link_stat
)
447 struct net_device
*dev
= adapter
->port
[port_id
];
449 /* Skip changes from disabled ports. */
450 if (netif_running(dev
) && link_stat
!= netif_carrier_ok(dev
)) {
452 netif_carrier_on(dev
);
454 netif_carrier_off(dev
);
460 void t4_os_portmod_changed(const struct adapter
*adap
, int port_id
)
462 static const char *mod_str
[] = {
463 NULL
, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
466 const struct net_device
*dev
= adap
->port
[port_id
];
467 const struct port_info
*pi
= netdev_priv(dev
);
469 if (pi
->mod_type
== FW_PORT_MOD_TYPE_NONE
)
470 netdev_info(dev
, "port module unplugged\n");
471 else if (pi
->mod_type
< ARRAY_SIZE(mod_str
))
472 netdev_info(dev
, "%s module inserted\n", mod_str
[pi
->mod_type
]);
476 * Configure the exact and hash address filters to handle a port's multicast
477 * and secondary unicast MAC addresses.
479 static int set_addr_filters(const struct net_device
*dev
, bool sleep
)
487 const struct netdev_hw_addr
*ha
;
488 int uc_cnt
= netdev_uc_count(dev
);
489 int mc_cnt
= netdev_mc_count(dev
);
490 const struct port_info
*pi
= netdev_priv(dev
);
491 unsigned int mb
= pi
->adapter
->fn
;
493 /* first do the secondary unicast addresses */
494 netdev_for_each_uc_addr(ha
, dev
) {
495 addr
[naddr
++] = ha
->addr
;
496 if (--uc_cnt
== 0 || naddr
>= ARRAY_SIZE(addr
)) {
497 ret
= t4_alloc_mac_filt(pi
->adapter
, mb
, pi
->viid
, free
,
498 naddr
, addr
, filt_idx
, &uhash
, sleep
);
507 /* next set up the multicast addresses */
508 netdev_for_each_mc_addr(ha
, dev
) {
509 addr
[naddr
++] = ha
->addr
;
510 if (--mc_cnt
== 0 || naddr
>= ARRAY_SIZE(addr
)) {
511 ret
= t4_alloc_mac_filt(pi
->adapter
, mb
, pi
->viid
, free
,
512 naddr
, addr
, filt_idx
, &mhash
, sleep
);
521 return t4_set_addr_hash(pi
->adapter
, mb
, pi
->viid
, uhash
!= 0,
522 uhash
| mhash
, sleep
);
525 int dbfifo_int_thresh
= 10; /* 10 == 640 entry threshold */
526 module_param(dbfifo_int_thresh
, int, 0644);
527 MODULE_PARM_DESC(dbfifo_int_thresh
, "doorbell fifo interrupt threshold");
530 * usecs to sleep while draining the dbfifo
532 static int dbfifo_drain_delay
= 1000;
533 module_param(dbfifo_drain_delay
, int, 0644);
534 MODULE_PARM_DESC(dbfifo_drain_delay
,
535 "usecs to sleep while draining the dbfifo");
538 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
539 * If @mtu is -1 it is left unchanged.
541 static int set_rxmode(struct net_device
*dev
, int mtu
, bool sleep_ok
)
544 struct port_info
*pi
= netdev_priv(dev
);
546 ret
= set_addr_filters(dev
, sleep_ok
);
548 ret
= t4_set_rxmode(pi
->adapter
, pi
->adapter
->fn
, pi
->viid
, mtu
,
549 (dev
->flags
& IFF_PROMISC
) ? 1 : 0,
550 (dev
->flags
& IFF_ALLMULTI
) ? 1 : 0, 1, -1,
555 static struct workqueue_struct
*workq
;
558 * link_start - enable a port
559 * @dev: the port to enable
561 * Performs the MAC and PHY actions needed to enable a port.
563 static int link_start(struct net_device
*dev
)
566 struct port_info
*pi
= netdev_priv(dev
);
567 unsigned int mb
= pi
->adapter
->fn
;
570 * We do not set address filters and promiscuity here, the stack does
571 * that step explicitly.
573 ret
= t4_set_rxmode(pi
->adapter
, mb
, pi
->viid
, dev
->mtu
, -1, -1, -1,
574 !!(dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
), true);
576 ret
= t4_change_mac(pi
->adapter
, mb
, pi
->viid
,
577 pi
->xact_addr_filt
, dev
->dev_addr
, true,
580 pi
->xact_addr_filt
= ret
;
585 ret
= t4_link_start(pi
->adapter
, mb
, pi
->tx_chan
,
588 ret
= t4_enable_vi(pi
->adapter
, mb
, pi
->viid
, true, true);
592 /* Clear a filter and release any of its resources that we own. This also
593 * clears the filter's "pending" status.
595 static void clear_filter(struct adapter
*adap
, struct filter_entry
*f
)
597 /* If the new or old filter have loopback rewriteing rules then we'll
598 * need to free any existing Layer Two Table (L2T) entries of the old
599 * filter rule. The firmware will handle freeing up any Source MAC
600 * Table (SMT) entries used for rewriting Source MAC Addresses in
604 cxgb4_l2t_release(f
->l2t
);
606 /* The zeroing of the filter rule below clears the filter valid,
607 * pending, locked flags, l2t pointer, etc. so it's all we need for
610 memset(f
, 0, sizeof(*f
));
613 /* Handle a filter write/deletion reply.
615 static void filter_rpl(struct adapter
*adap
, const struct cpl_set_tcb_rpl
*rpl
)
617 unsigned int idx
= GET_TID(rpl
);
618 unsigned int nidx
= idx
- adap
->tids
.ftid_base
;
620 struct filter_entry
*f
;
622 if (idx
>= adap
->tids
.ftid_base
&& nidx
<
623 (adap
->tids
.nftids
+ adap
->tids
.nsftids
)) {
625 ret
= GET_TCB_COOKIE(rpl
->cookie
);
626 f
= &adap
->tids
.ftid_tab
[idx
];
628 if (ret
== FW_FILTER_WR_FLT_DELETED
) {
629 /* Clear the filter when we get confirmation from the
630 * hardware that the filter has been deleted.
632 clear_filter(adap
, f
);
633 } else if (ret
== FW_FILTER_WR_SMT_TBL_FULL
) {
634 dev_err(adap
->pdev_dev
, "filter %u setup failed due to full SMT\n",
636 clear_filter(adap
, f
);
637 } else if (ret
== FW_FILTER_WR_FLT_ADDED
) {
638 f
->smtidx
= (be64_to_cpu(rpl
->oldval
) >> 24) & 0xff;
639 f
->pending
= 0; /* asynchronous setup completed */
642 /* Something went wrong. Issue a warning about the
643 * problem and clear everything out.
645 dev_err(adap
->pdev_dev
, "filter %u setup failed with error %u\n",
647 clear_filter(adap
, f
);
652 /* Response queue handler for the FW event queue.
654 static int fwevtq_handler(struct sge_rspq
*q
, const __be64
*rsp
,
655 const struct pkt_gl
*gl
)
657 u8 opcode
= ((const struct rss_header
*)rsp
)->opcode
;
659 rsp
++; /* skip RSS header */
661 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
663 if (unlikely(opcode
== CPL_FW4_MSG
&&
664 ((const struct cpl_fw4_msg
*)rsp
)->type
== FW_TYPE_RSSCPL
)) {
666 opcode
= ((const struct rss_header
*)rsp
)->opcode
;
668 if (opcode
!= CPL_SGE_EGR_UPDATE
) {
669 dev_err(q
->adap
->pdev_dev
, "unexpected FW4/CPL %#x on FW event queue\n"
675 if (likely(opcode
== CPL_SGE_EGR_UPDATE
)) {
676 const struct cpl_sge_egr_update
*p
= (void *)rsp
;
677 unsigned int qid
= EGR_QID(ntohl(p
->opcode_qid
));
680 txq
= q
->adap
->sge
.egr_map
[qid
- q
->adap
->sge
.egr_start
];
682 if ((u8
*)txq
< (u8
*)q
->adap
->sge
.ofldtxq
) {
683 struct sge_eth_txq
*eq
;
685 eq
= container_of(txq
, struct sge_eth_txq
, q
);
686 netif_tx_wake_queue(eq
->txq
);
688 struct sge_ofld_txq
*oq
;
690 oq
= container_of(txq
, struct sge_ofld_txq
, q
);
691 tasklet_schedule(&oq
->qresume_tsk
);
693 } else if (opcode
== CPL_FW6_MSG
|| opcode
== CPL_FW4_MSG
) {
694 const struct cpl_fw6_msg
*p
= (void *)rsp
;
697 t4_handle_fw_rpl(q
->adap
, p
->data
);
698 } else if (opcode
== CPL_L2T_WRITE_RPL
) {
699 const struct cpl_l2t_write_rpl
*p
= (void *)rsp
;
701 do_l2t_write_rpl(q
->adap
, p
);
702 } else if (opcode
== CPL_SET_TCB_RPL
) {
703 const struct cpl_set_tcb_rpl
*p
= (void *)rsp
;
705 filter_rpl(q
->adap
, p
);
707 dev_err(q
->adap
->pdev_dev
,
708 "unexpected CPL %#x on FW event queue\n", opcode
);
714 * uldrx_handler - response queue handler for ULD queues
715 * @q: the response queue that received the packet
716 * @rsp: the response queue descriptor holding the offload message
717 * @gl: the gather list of packet fragments
719 * Deliver an ingress offload packet to a ULD. All processing is done by
720 * the ULD, we just maintain statistics.
722 static int uldrx_handler(struct sge_rspq
*q
, const __be64
*rsp
,
723 const struct pkt_gl
*gl
)
725 struct sge_ofld_rxq
*rxq
= container_of(q
, struct sge_ofld_rxq
, rspq
);
727 /* FW can send CPLs encapsulated in a CPL_FW4_MSG.
729 if (((const struct rss_header
*)rsp
)->opcode
== CPL_FW4_MSG
&&
730 ((const struct cpl_fw4_msg
*)(rsp
+ 1))->type
== FW_TYPE_RSSCPL
)
733 if (ulds
[q
->uld
].rx_handler(q
->adap
->uld_handle
[q
->uld
], rsp
, gl
)) {
739 else if (gl
== CXGB4_MSG_AN
)
746 static void disable_msi(struct adapter
*adapter
)
748 if (adapter
->flags
& USING_MSIX
) {
749 pci_disable_msix(adapter
->pdev
);
750 adapter
->flags
&= ~USING_MSIX
;
751 } else if (adapter
->flags
& USING_MSI
) {
752 pci_disable_msi(adapter
->pdev
);
753 adapter
->flags
&= ~USING_MSI
;
758 * Interrupt handler for non-data events used with MSI-X.
760 static irqreturn_t
t4_nondata_intr(int irq
, void *cookie
)
762 struct adapter
*adap
= cookie
;
764 u32 v
= t4_read_reg(adap
, MYPF_REG(PL_PF_INT_CAUSE
));
767 t4_write_reg(adap
, MYPF_REG(PL_PF_INT_CAUSE
), v
);
769 t4_slow_intr_handler(adap
);
774 * Name the MSI-X interrupts.
776 static void name_msix_vecs(struct adapter
*adap
)
778 int i
, j
, msi_idx
= 2, n
= sizeof(adap
->msix_info
[0].desc
);
780 /* non-data interrupts */
781 snprintf(adap
->msix_info
[0].desc
, n
, "%s", adap
->port
[0]->name
);
784 snprintf(adap
->msix_info
[1].desc
, n
, "%s-FWeventq",
785 adap
->port
[0]->name
);
787 /* Ethernet queues */
788 for_each_port(adap
, j
) {
789 struct net_device
*d
= adap
->port
[j
];
790 const struct port_info
*pi
= netdev_priv(d
);
792 for (i
= 0; i
< pi
->nqsets
; i
++, msi_idx
++)
793 snprintf(adap
->msix_info
[msi_idx
].desc
, n
, "%s-Rx%d",
798 for_each_ofldrxq(&adap
->sge
, i
)
799 snprintf(adap
->msix_info
[msi_idx
++].desc
, n
, "%s-ofld%d",
800 adap
->port
[0]->name
, i
);
802 for_each_rdmarxq(&adap
->sge
, i
)
803 snprintf(adap
->msix_info
[msi_idx
++].desc
, n
, "%s-rdma%d",
804 adap
->port
[0]->name
, i
);
807 static int request_msix_queue_irqs(struct adapter
*adap
)
809 struct sge
*s
= &adap
->sge
;
810 int err
, ethqidx
, ofldqidx
= 0, rdmaqidx
= 0, msi_index
= 2;
812 err
= request_irq(adap
->msix_info
[1].vec
, t4_sge_intr_msix
, 0,
813 adap
->msix_info
[1].desc
, &s
->fw_evtq
);
817 for_each_ethrxq(s
, ethqidx
) {
818 err
= request_irq(adap
->msix_info
[msi_index
].vec
,
820 adap
->msix_info
[msi_index
].desc
,
821 &s
->ethrxq
[ethqidx
].rspq
);
826 for_each_ofldrxq(s
, ofldqidx
) {
827 err
= request_irq(adap
->msix_info
[msi_index
].vec
,
829 adap
->msix_info
[msi_index
].desc
,
830 &s
->ofldrxq
[ofldqidx
].rspq
);
835 for_each_rdmarxq(s
, rdmaqidx
) {
836 err
= request_irq(adap
->msix_info
[msi_index
].vec
,
838 adap
->msix_info
[msi_index
].desc
,
839 &s
->rdmarxq
[rdmaqidx
].rspq
);
847 while (--rdmaqidx
>= 0)
848 free_irq(adap
->msix_info
[--msi_index
].vec
,
849 &s
->rdmarxq
[rdmaqidx
].rspq
);
850 while (--ofldqidx
>= 0)
851 free_irq(adap
->msix_info
[--msi_index
].vec
,
852 &s
->ofldrxq
[ofldqidx
].rspq
);
853 while (--ethqidx
>= 0)
854 free_irq(adap
->msix_info
[--msi_index
].vec
,
855 &s
->ethrxq
[ethqidx
].rspq
);
856 free_irq(adap
->msix_info
[1].vec
, &s
->fw_evtq
);
860 static void free_msix_queue_irqs(struct adapter
*adap
)
862 int i
, msi_index
= 2;
863 struct sge
*s
= &adap
->sge
;
865 free_irq(adap
->msix_info
[1].vec
, &s
->fw_evtq
);
866 for_each_ethrxq(s
, i
)
867 free_irq(adap
->msix_info
[msi_index
++].vec
, &s
->ethrxq
[i
].rspq
);
868 for_each_ofldrxq(s
, i
)
869 free_irq(adap
->msix_info
[msi_index
++].vec
, &s
->ofldrxq
[i
].rspq
);
870 for_each_rdmarxq(s
, i
)
871 free_irq(adap
->msix_info
[msi_index
++].vec
, &s
->rdmarxq
[i
].rspq
);
875 * write_rss - write the RSS table for a given port
877 * @queues: array of queue indices for RSS
879 * Sets up the portion of the HW RSS table for the port's VI to distribute
880 * packets to the Rx queues in @queues.
882 static int write_rss(const struct port_info
*pi
, const u16
*queues
)
886 const struct sge_eth_rxq
*q
= &pi
->adapter
->sge
.ethrxq
[pi
->first_qset
];
888 rss
= kmalloc(pi
->rss_size
* sizeof(u16
), GFP_KERNEL
);
892 /* map the queue indices to queue ids */
893 for (i
= 0; i
< pi
->rss_size
; i
++, queues
++)
894 rss
[i
] = q
[*queues
].rspq
.abs_id
;
896 err
= t4_config_rss_range(pi
->adapter
, pi
->adapter
->fn
, pi
->viid
, 0,
897 pi
->rss_size
, rss
, pi
->rss_size
);
903 * setup_rss - configure RSS
906 * Sets up RSS for each port.
908 static int setup_rss(struct adapter
*adap
)
912 for_each_port(adap
, i
) {
913 const struct port_info
*pi
= adap2pinfo(adap
, i
);
915 err
= write_rss(pi
, pi
->rss
);
923 * Return the channel of the ingress queue with the given qid.
925 static unsigned int rxq_to_chan(const struct sge
*p
, unsigned int qid
)
927 qid
-= p
->ingr_start
;
928 return netdev2pinfo(p
->ingr_map
[qid
]->netdev
)->tx_chan
;
932 * Wait until all NAPI handlers are descheduled.
934 static void quiesce_rx(struct adapter
*adap
)
938 for (i
= 0; i
< ARRAY_SIZE(adap
->sge
.ingr_map
); i
++) {
939 struct sge_rspq
*q
= adap
->sge
.ingr_map
[i
];
942 napi_disable(&q
->napi
);
947 * Enable NAPI scheduling and interrupt generation for all Rx queues.
949 static void enable_rx(struct adapter
*adap
)
953 for (i
= 0; i
< ARRAY_SIZE(adap
->sge
.ingr_map
); i
++) {
954 struct sge_rspq
*q
= adap
->sge
.ingr_map
[i
];
959 napi_enable(&q
->napi
);
960 /* 0-increment GTS to start the timer and enable interrupts */
961 t4_write_reg(adap
, MYPF_REG(SGE_PF_GTS
),
962 SEINTARM(q
->intr_params
) |
963 INGRESSQID(q
->cntxt_id
));
968 * setup_sge_queues - configure SGE Tx/Rx/response queues
971 * Determines how many sets of SGE queues to use and initializes them.
972 * We support multiple queue sets per port if we have MSI-X, otherwise
973 * just one queue set per port.
975 static int setup_sge_queues(struct adapter
*adap
)
977 int err
, msi_idx
, i
, j
;
978 struct sge
*s
= &adap
->sge
;
980 bitmap_zero(s
->starving_fl
, MAX_EGRQ
);
981 bitmap_zero(s
->txq_maperr
, MAX_EGRQ
);
983 if (adap
->flags
& USING_MSIX
)
984 msi_idx
= 1; /* vector 0 is for non-queue interrupts */
986 err
= t4_sge_alloc_rxq(adap
, &s
->intrq
, false, adap
->port
[0], 0,
990 msi_idx
= -((int)s
->intrq
.abs_id
+ 1);
993 err
= t4_sge_alloc_rxq(adap
, &s
->fw_evtq
, true, adap
->port
[0],
994 msi_idx
, NULL
, fwevtq_handler
);
996 freeout
: t4_free_sge_resources(adap
);
1000 for_each_port(adap
, i
) {
1001 struct net_device
*dev
= adap
->port
[i
];
1002 struct port_info
*pi
= netdev_priv(dev
);
1003 struct sge_eth_rxq
*q
= &s
->ethrxq
[pi
->first_qset
];
1004 struct sge_eth_txq
*t
= &s
->ethtxq
[pi
->first_qset
];
1006 for (j
= 0; j
< pi
->nqsets
; j
++, q
++) {
1009 err
= t4_sge_alloc_rxq(adap
, &q
->rspq
, false, dev
,
1015 memset(&q
->stats
, 0, sizeof(q
->stats
));
1017 for (j
= 0; j
< pi
->nqsets
; j
++, t
++) {
1018 err
= t4_sge_alloc_eth_txq(adap
, t
, dev
,
1019 netdev_get_tx_queue(dev
, j
),
1020 s
->fw_evtq
.cntxt_id
);
1026 j
= s
->ofldqsets
/ adap
->params
.nports
; /* ofld queues per channel */
1027 for_each_ofldrxq(s
, i
) {
1028 struct sge_ofld_rxq
*q
= &s
->ofldrxq
[i
];
1029 struct net_device
*dev
= adap
->port
[i
/ j
];
1033 err
= t4_sge_alloc_rxq(adap
, &q
->rspq
, false, dev
, msi_idx
,
1034 &q
->fl
, uldrx_handler
);
1037 memset(&q
->stats
, 0, sizeof(q
->stats
));
1038 s
->ofld_rxq
[i
] = q
->rspq
.abs_id
;
1039 err
= t4_sge_alloc_ofld_txq(adap
, &s
->ofldtxq
[i
], dev
,
1040 s
->fw_evtq
.cntxt_id
);
1045 for_each_rdmarxq(s
, i
) {
1046 struct sge_ofld_rxq
*q
= &s
->rdmarxq
[i
];
1050 err
= t4_sge_alloc_rxq(adap
, &q
->rspq
, false, adap
->port
[i
],
1051 msi_idx
, &q
->fl
, uldrx_handler
);
1054 memset(&q
->stats
, 0, sizeof(q
->stats
));
1055 s
->rdma_rxq
[i
] = q
->rspq
.abs_id
;
1058 for_each_port(adap
, i
) {
1060 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
1061 * have RDMA queues, and that's the right value.
1063 err
= t4_sge_alloc_ctrl_txq(adap
, &s
->ctrlq
[i
], adap
->port
[i
],
1064 s
->fw_evtq
.cntxt_id
,
1065 s
->rdmarxq
[i
].rspq
.cntxt_id
);
1070 t4_write_reg(adap
, MPS_TRC_RSS_CONTROL
,
1071 RSSCONTROL(netdev2pinfo(adap
->port
[0])->tx_chan
) |
1072 QUEUENUMBER(s
->ethrxq
[0].rspq
.abs_id
));
1077 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1078 * The allocated memory is cleared.
1080 void *t4_alloc_mem(size_t size
)
1082 void *p
= kzalloc(size
, GFP_KERNEL
| __GFP_NOWARN
);
1090 * Free memory allocated through alloc_mem().
1092 static void t4_free_mem(void *addr
)
1094 if (is_vmalloc_addr(addr
))
1100 /* Send a Work Request to write the filter at a specified index. We construct
1101 * a Firmware Filter Work Request to have the work done and put the indicated
1102 * filter into "pending" mode which will prevent any further actions against
1103 * it till we get a reply from the firmware on the completion status of the
1106 static int set_filter_wr(struct adapter
*adapter
, int fidx
)
1108 struct filter_entry
*f
= &adapter
->tids
.ftid_tab
[fidx
];
1109 struct sk_buff
*skb
;
1110 struct fw_filter_wr
*fwr
;
1113 /* If the new filter requires loopback Destination MAC and/or VLAN
1114 * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
1117 if (f
->fs
.newdmac
|| f
->fs
.newvlan
) {
1118 /* allocate L2T entry for new filter */
1119 f
->l2t
= t4_l2t_alloc_switching(adapter
->l2t
);
1122 if (t4_l2t_set_switching(adapter
, f
->l2t
, f
->fs
.vlan
,
1123 f
->fs
.eport
, f
->fs
.dmac
)) {
1124 cxgb4_l2t_release(f
->l2t
);
1130 ftid
= adapter
->tids
.ftid_base
+ fidx
;
1132 skb
= alloc_skb(sizeof(*fwr
), GFP_KERNEL
| __GFP_NOFAIL
);
1133 fwr
= (struct fw_filter_wr
*)__skb_put(skb
, sizeof(*fwr
));
1134 memset(fwr
, 0, sizeof(*fwr
));
1136 /* It would be nice to put most of the following in t4_hw.c but most
1137 * of the work is translating the cxgbtool ch_filter_specification
1138 * into the Work Request and the definition of that structure is
1139 * currently in cxgbtool.h which isn't appropriate to pull into the
1140 * common code. We may eventually try to come up with a more neutral
1141 * filter specification structure but for now it's easiest to simply
1142 * put this fairly direct code in line ...
1144 fwr
->op_pkd
= htonl(FW_WR_OP(FW_FILTER_WR
));
1145 fwr
->len16_pkd
= htonl(FW_WR_LEN16(sizeof(*fwr
)/16));
1147 htonl(V_FW_FILTER_WR_TID(ftid
) |
1148 V_FW_FILTER_WR_RQTYPE(f
->fs
.type
) |
1149 V_FW_FILTER_WR_NOREPLY(0) |
1150 V_FW_FILTER_WR_IQ(f
->fs
.iq
));
1151 fwr
->del_filter_to_l2tix
=
1152 htonl(V_FW_FILTER_WR_RPTTID(f
->fs
.rpttid
) |
1153 V_FW_FILTER_WR_DROP(f
->fs
.action
== FILTER_DROP
) |
1154 V_FW_FILTER_WR_DIRSTEER(f
->fs
.dirsteer
) |
1155 V_FW_FILTER_WR_MASKHASH(f
->fs
.maskhash
) |
1156 V_FW_FILTER_WR_DIRSTEERHASH(f
->fs
.dirsteerhash
) |
1157 V_FW_FILTER_WR_LPBK(f
->fs
.action
== FILTER_SWITCH
) |
1158 V_FW_FILTER_WR_DMAC(f
->fs
.newdmac
) |
1159 V_FW_FILTER_WR_SMAC(f
->fs
.newsmac
) |
1160 V_FW_FILTER_WR_INSVLAN(f
->fs
.newvlan
== VLAN_INSERT
||
1161 f
->fs
.newvlan
== VLAN_REWRITE
) |
1162 V_FW_FILTER_WR_RMVLAN(f
->fs
.newvlan
== VLAN_REMOVE
||
1163 f
->fs
.newvlan
== VLAN_REWRITE
) |
1164 V_FW_FILTER_WR_HITCNTS(f
->fs
.hitcnts
) |
1165 V_FW_FILTER_WR_TXCHAN(f
->fs
.eport
) |
1166 V_FW_FILTER_WR_PRIO(f
->fs
.prio
) |
1167 V_FW_FILTER_WR_L2TIX(f
->l2t
? f
->l2t
->idx
: 0));
1168 fwr
->ethtype
= htons(f
->fs
.val
.ethtype
);
1169 fwr
->ethtypem
= htons(f
->fs
.mask
.ethtype
);
1170 fwr
->frag_to_ovlan_vldm
=
1171 (V_FW_FILTER_WR_FRAG(f
->fs
.val
.frag
) |
1172 V_FW_FILTER_WR_FRAGM(f
->fs
.mask
.frag
) |
1173 V_FW_FILTER_WR_IVLAN_VLD(f
->fs
.val
.ivlan_vld
) |
1174 V_FW_FILTER_WR_OVLAN_VLD(f
->fs
.val
.ovlan_vld
) |
1175 V_FW_FILTER_WR_IVLAN_VLDM(f
->fs
.mask
.ivlan_vld
) |
1176 V_FW_FILTER_WR_OVLAN_VLDM(f
->fs
.mask
.ovlan_vld
));
1178 fwr
->rx_chan_rx_rpl_iq
=
1179 htons(V_FW_FILTER_WR_RX_CHAN(0) |
1180 V_FW_FILTER_WR_RX_RPL_IQ(adapter
->sge
.fw_evtq
.abs_id
));
1181 fwr
->maci_to_matchtypem
=
1182 htonl(V_FW_FILTER_WR_MACI(f
->fs
.val
.macidx
) |
1183 V_FW_FILTER_WR_MACIM(f
->fs
.mask
.macidx
) |
1184 V_FW_FILTER_WR_FCOE(f
->fs
.val
.fcoe
) |
1185 V_FW_FILTER_WR_FCOEM(f
->fs
.mask
.fcoe
) |
1186 V_FW_FILTER_WR_PORT(f
->fs
.val
.iport
) |
1187 V_FW_FILTER_WR_PORTM(f
->fs
.mask
.iport
) |
1188 V_FW_FILTER_WR_MATCHTYPE(f
->fs
.val
.matchtype
) |
1189 V_FW_FILTER_WR_MATCHTYPEM(f
->fs
.mask
.matchtype
));
1190 fwr
->ptcl
= f
->fs
.val
.proto
;
1191 fwr
->ptclm
= f
->fs
.mask
.proto
;
1192 fwr
->ttyp
= f
->fs
.val
.tos
;
1193 fwr
->ttypm
= f
->fs
.mask
.tos
;
1194 fwr
->ivlan
= htons(f
->fs
.val
.ivlan
);
1195 fwr
->ivlanm
= htons(f
->fs
.mask
.ivlan
);
1196 fwr
->ovlan
= htons(f
->fs
.val
.ovlan
);
1197 fwr
->ovlanm
= htons(f
->fs
.mask
.ovlan
);
1198 memcpy(fwr
->lip
, f
->fs
.val
.lip
, sizeof(fwr
->lip
));
1199 memcpy(fwr
->lipm
, f
->fs
.mask
.lip
, sizeof(fwr
->lipm
));
1200 memcpy(fwr
->fip
, f
->fs
.val
.fip
, sizeof(fwr
->fip
));
1201 memcpy(fwr
->fipm
, f
->fs
.mask
.fip
, sizeof(fwr
->fipm
));
1202 fwr
->lp
= htons(f
->fs
.val
.lport
);
1203 fwr
->lpm
= htons(f
->fs
.mask
.lport
);
1204 fwr
->fp
= htons(f
->fs
.val
.fport
);
1205 fwr
->fpm
= htons(f
->fs
.mask
.fport
);
1207 memcpy(fwr
->sma
, f
->fs
.smac
, sizeof(fwr
->sma
));
1209 /* Mark the filter as "pending" and ship off the Filter Work Request.
1210 * When we get the Work Request Reply we'll clear the pending status.
1213 set_wr_txq(skb
, CPL_PRIORITY_CONTROL
, f
->fs
.val
.iport
& 0x3);
1214 t4_ofld_send(adapter
, skb
);
1218 /* Delete the filter at a specified index.
1220 static int del_filter_wr(struct adapter
*adapter
, int fidx
)
1222 struct filter_entry
*f
= &adapter
->tids
.ftid_tab
[fidx
];
1223 struct sk_buff
*skb
;
1224 struct fw_filter_wr
*fwr
;
1225 unsigned int len
, ftid
;
1228 ftid
= adapter
->tids
.ftid_base
+ fidx
;
1230 skb
= alloc_skb(len
, GFP_KERNEL
| __GFP_NOFAIL
);
1231 fwr
= (struct fw_filter_wr
*)__skb_put(skb
, len
);
1232 t4_mk_filtdelwr(ftid
, fwr
, adapter
->sge
.fw_evtq
.abs_id
);
1234 /* Mark the filter as "pending" and ship off the Filter Work Request.
1235 * When we get the Work Request Reply we'll clear the pending status.
1238 t4_mgmt_tx(adapter
, skb
);
1242 static inline int is_offload(const struct adapter
*adap
)
1244 return adap
->params
.offload
;
1248 * Implementation of ethtool operations.
1251 static u32
get_msglevel(struct net_device
*dev
)
1253 return netdev2adap(dev
)->msg_enable
;
1256 static void set_msglevel(struct net_device
*dev
, u32 val
)
1258 netdev2adap(dev
)->msg_enable
= val
;
1261 static char stats_strings
[][ETH_GSTRING_LEN
] = {
1264 "TxBroadcastFrames ",
1265 "TxMulticastFrames ",
1271 "TxFrames128To255 ",
1272 "TxFrames256To511 ",
1273 "TxFrames512To1023 ",
1274 "TxFrames1024To1518 ",
1275 "TxFrames1519ToMax ",
1290 "RxBroadcastFrames ",
1291 "RxMulticastFrames ",
1303 "RxFrames128To255 ",
1304 "RxFrames256To511 ",
1305 "RxFrames512To1023 ",
1306 "RxFrames1024To1518 ",
1307 "RxFrames1519ToMax ",
1319 "RxBG0FramesDropped ",
1320 "RxBG1FramesDropped ",
1321 "RxBG2FramesDropped ",
1322 "RxBG3FramesDropped ",
1323 "RxBG0FramesTrunc ",
1324 "RxBG1FramesTrunc ",
1325 "RxBG2FramesTrunc ",
1326 "RxBG3FramesTrunc ",
1335 "WriteCoalSuccess ",
1339 static int get_sset_count(struct net_device
*dev
, int sset
)
1343 return ARRAY_SIZE(stats_strings
);
1349 #define T4_REGMAP_SIZE (160 * 1024)
1350 #define T5_REGMAP_SIZE (332 * 1024)
1352 static int get_regs_len(struct net_device
*dev
)
1354 struct adapter
*adap
= netdev2adap(dev
);
1355 if (is_t4(adap
->params
.chip
))
1356 return T4_REGMAP_SIZE
;
1358 return T5_REGMAP_SIZE
;
1361 static int get_eeprom_len(struct net_device
*dev
)
1366 static void get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
1368 struct adapter
*adapter
= netdev2adap(dev
);
1370 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1371 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
1372 strlcpy(info
->bus_info
, pci_name(adapter
->pdev
),
1373 sizeof(info
->bus_info
));
1375 if (adapter
->params
.fw_vers
)
1376 snprintf(info
->fw_version
, sizeof(info
->fw_version
),
1377 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1378 FW_HDR_FW_VER_MAJOR_GET(adapter
->params
.fw_vers
),
1379 FW_HDR_FW_VER_MINOR_GET(adapter
->params
.fw_vers
),
1380 FW_HDR_FW_VER_MICRO_GET(adapter
->params
.fw_vers
),
1381 FW_HDR_FW_VER_BUILD_GET(adapter
->params
.fw_vers
),
1382 FW_HDR_FW_VER_MAJOR_GET(adapter
->params
.tp_vers
),
1383 FW_HDR_FW_VER_MINOR_GET(adapter
->params
.tp_vers
),
1384 FW_HDR_FW_VER_MICRO_GET(adapter
->params
.tp_vers
),
1385 FW_HDR_FW_VER_BUILD_GET(adapter
->params
.tp_vers
));
1388 static void get_strings(struct net_device
*dev
, u32 stringset
, u8
*data
)
1390 if (stringset
== ETH_SS_STATS
)
1391 memcpy(data
, stats_strings
, sizeof(stats_strings
));
1395 * port stats maintained per queue of the port. They should be in the same
1396 * order as in stats_strings above.
1398 struct queue_port_stats
{
1408 static void collect_sge_port_stats(const struct adapter
*adap
,
1409 const struct port_info
*p
, struct queue_port_stats
*s
)
1412 const struct sge_eth_txq
*tx
= &adap
->sge
.ethtxq
[p
->first_qset
];
1413 const struct sge_eth_rxq
*rx
= &adap
->sge
.ethrxq
[p
->first_qset
];
1415 memset(s
, 0, sizeof(*s
));
1416 for (i
= 0; i
< p
->nqsets
; i
++, rx
++, tx
++) {
1418 s
->tx_csum
+= tx
->tx_cso
;
1419 s
->rx_csum
+= rx
->stats
.rx_cso
;
1420 s
->vlan_ex
+= rx
->stats
.vlan_ex
;
1421 s
->vlan_ins
+= tx
->vlan_ins
;
1422 s
->gro_pkts
+= rx
->stats
.lro_pkts
;
1423 s
->gro_merged
+= rx
->stats
.lro_merged
;
1427 static void get_stats(struct net_device
*dev
, struct ethtool_stats
*stats
,
1430 struct port_info
*pi
= netdev_priv(dev
);
1431 struct adapter
*adapter
= pi
->adapter
;
1434 t4_get_port_stats(adapter
, pi
->tx_chan
, (struct port_stats
*)data
);
1436 data
+= sizeof(struct port_stats
) / sizeof(u64
);
1437 collect_sge_port_stats(adapter
, pi
, (struct queue_port_stats
*)data
);
1438 data
+= sizeof(struct queue_port_stats
) / sizeof(u64
);
1439 if (!is_t4(adapter
->params
.chip
)) {
1440 t4_write_reg(adapter
, SGE_STAT_CFG
, STATSOURCE_T5(7));
1441 val1
= t4_read_reg(adapter
, SGE_STAT_TOTAL
);
1442 val2
= t4_read_reg(adapter
, SGE_STAT_MATCH
);
1443 *data
= val1
- val2
;
1448 memset(data
, 0, 2 * sizeof(u64
));
1454 * Return a version number to identify the type of adapter. The scheme is:
1455 * - bits 0..9: chip version
1456 * - bits 10..15: chip revision
1457 * - bits 16..23: register dump version
1459 static inline unsigned int mk_adap_vers(const struct adapter
*ap
)
1461 return CHELSIO_CHIP_VERSION(ap
->params
.chip
) |
1462 (CHELSIO_CHIP_RELEASE(ap
->params
.chip
) << 10) | (1 << 16);
1465 static void reg_block_dump(struct adapter
*ap
, void *buf
, unsigned int start
,
1468 u32
*p
= buf
+ start
;
1470 for ( ; start
<= end
; start
+= sizeof(u32
))
1471 *p
++ = t4_read_reg(ap
, start
);
1474 static void get_regs(struct net_device
*dev
, struct ethtool_regs
*regs
,
1477 static const unsigned int t4_reg_ranges
[] = {
1697 static const unsigned int t5_reg_ranges
[] = {
2125 struct adapter
*ap
= netdev2adap(dev
);
2126 static const unsigned int *reg_ranges
;
2127 int arr_size
= 0, buf_size
= 0;
2129 if (is_t4(ap
->params
.chip
)) {
2130 reg_ranges
= &t4_reg_ranges
[0];
2131 arr_size
= ARRAY_SIZE(t4_reg_ranges
);
2132 buf_size
= T4_REGMAP_SIZE
;
2134 reg_ranges
= &t5_reg_ranges
[0];
2135 arr_size
= ARRAY_SIZE(t5_reg_ranges
);
2136 buf_size
= T5_REGMAP_SIZE
;
2139 regs
->version
= mk_adap_vers(ap
);
2141 memset(buf
, 0, buf_size
);
2142 for (i
= 0; i
< arr_size
; i
+= 2)
2143 reg_block_dump(ap
, buf
, reg_ranges
[i
], reg_ranges
[i
+ 1]);
2146 static int restart_autoneg(struct net_device
*dev
)
2148 struct port_info
*p
= netdev_priv(dev
);
2150 if (!netif_running(dev
))
2152 if (p
->link_cfg
.autoneg
!= AUTONEG_ENABLE
)
2154 t4_restart_aneg(p
->adapter
, p
->adapter
->fn
, p
->tx_chan
);
2158 static int identify_port(struct net_device
*dev
,
2159 enum ethtool_phys_id_state state
)
2162 struct adapter
*adap
= netdev2adap(dev
);
2164 if (state
== ETHTOOL_ID_ACTIVE
)
2166 else if (state
== ETHTOOL_ID_INACTIVE
)
2171 return t4_identify_port(adap
, adap
->fn
, netdev2pinfo(dev
)->viid
, val
);
2174 static unsigned int from_fw_linkcaps(unsigned int type
, unsigned int caps
)
2178 if (type
== FW_PORT_TYPE_BT_SGMII
|| type
== FW_PORT_TYPE_BT_XFI
||
2179 type
== FW_PORT_TYPE_BT_XAUI
) {
2181 if (caps
& FW_PORT_CAP_SPEED_100M
)
2182 v
|= SUPPORTED_100baseT_Full
;
2183 if (caps
& FW_PORT_CAP_SPEED_1G
)
2184 v
|= SUPPORTED_1000baseT_Full
;
2185 if (caps
& FW_PORT_CAP_SPEED_10G
)
2186 v
|= SUPPORTED_10000baseT_Full
;
2187 } else if (type
== FW_PORT_TYPE_KX4
|| type
== FW_PORT_TYPE_KX
) {
2188 v
|= SUPPORTED_Backplane
;
2189 if (caps
& FW_PORT_CAP_SPEED_1G
)
2190 v
|= SUPPORTED_1000baseKX_Full
;
2191 if (caps
& FW_PORT_CAP_SPEED_10G
)
2192 v
|= SUPPORTED_10000baseKX4_Full
;
2193 } else if (type
== FW_PORT_TYPE_KR
)
2194 v
|= SUPPORTED_Backplane
| SUPPORTED_10000baseKR_Full
;
2195 else if (type
== FW_PORT_TYPE_BP_AP
)
2196 v
|= SUPPORTED_Backplane
| SUPPORTED_10000baseR_FEC
|
2197 SUPPORTED_10000baseKR_Full
| SUPPORTED_1000baseKX_Full
;
2198 else if (type
== FW_PORT_TYPE_BP4_AP
)
2199 v
|= SUPPORTED_Backplane
| SUPPORTED_10000baseR_FEC
|
2200 SUPPORTED_10000baseKR_Full
| SUPPORTED_1000baseKX_Full
|
2201 SUPPORTED_10000baseKX4_Full
;
2202 else if (type
== FW_PORT_TYPE_FIBER_XFI
||
2203 type
== FW_PORT_TYPE_FIBER_XAUI
|| type
== FW_PORT_TYPE_SFP
)
2204 v
|= SUPPORTED_FIBRE
;
2205 else if (type
== FW_PORT_TYPE_BP40_BA
)
2206 v
|= SUPPORTED_40000baseSR4_Full
;
2208 if (caps
& FW_PORT_CAP_ANEG
)
2209 v
|= SUPPORTED_Autoneg
;
2213 static unsigned int to_fw_linkcaps(unsigned int caps
)
2217 if (caps
& ADVERTISED_100baseT_Full
)
2218 v
|= FW_PORT_CAP_SPEED_100M
;
2219 if (caps
& ADVERTISED_1000baseT_Full
)
2220 v
|= FW_PORT_CAP_SPEED_1G
;
2221 if (caps
& ADVERTISED_10000baseT_Full
)
2222 v
|= FW_PORT_CAP_SPEED_10G
;
2223 if (caps
& ADVERTISED_40000baseSR4_Full
)
2224 v
|= FW_PORT_CAP_SPEED_40G
;
2228 static int get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2230 const struct port_info
*p
= netdev_priv(dev
);
2232 if (p
->port_type
== FW_PORT_TYPE_BT_SGMII
||
2233 p
->port_type
== FW_PORT_TYPE_BT_XFI
||
2234 p
->port_type
== FW_PORT_TYPE_BT_XAUI
)
2235 cmd
->port
= PORT_TP
;
2236 else if (p
->port_type
== FW_PORT_TYPE_FIBER_XFI
||
2237 p
->port_type
== FW_PORT_TYPE_FIBER_XAUI
)
2238 cmd
->port
= PORT_FIBRE
;
2239 else if (p
->port_type
== FW_PORT_TYPE_SFP
) {
2240 if (p
->mod_type
== FW_PORT_MOD_TYPE_TWINAX_PASSIVE
||
2241 p
->mod_type
== FW_PORT_MOD_TYPE_TWINAX_ACTIVE
)
2242 cmd
->port
= PORT_DA
;
2244 cmd
->port
= PORT_FIBRE
;
2246 cmd
->port
= PORT_OTHER
;
2248 if (p
->mdio_addr
>= 0) {
2249 cmd
->phy_address
= p
->mdio_addr
;
2250 cmd
->transceiver
= XCVR_EXTERNAL
;
2251 cmd
->mdio_support
= p
->port_type
== FW_PORT_TYPE_BT_SGMII
?
2252 MDIO_SUPPORTS_C22
: MDIO_SUPPORTS_C45
;
2254 cmd
->phy_address
= 0; /* not really, but no better option */
2255 cmd
->transceiver
= XCVR_INTERNAL
;
2256 cmd
->mdio_support
= 0;
2259 cmd
->supported
= from_fw_linkcaps(p
->port_type
, p
->link_cfg
.supported
);
2260 cmd
->advertising
= from_fw_linkcaps(p
->port_type
,
2261 p
->link_cfg
.advertising
);
2262 ethtool_cmd_speed_set(cmd
,
2263 netif_carrier_ok(dev
) ? p
->link_cfg
.speed
: 0);
2264 cmd
->duplex
= DUPLEX_FULL
;
2265 cmd
->autoneg
= p
->link_cfg
.autoneg
;
2271 static unsigned int speed_to_caps(int speed
)
2273 if (speed
== SPEED_100
)
2274 return FW_PORT_CAP_SPEED_100M
;
2275 if (speed
== SPEED_1000
)
2276 return FW_PORT_CAP_SPEED_1G
;
2277 if (speed
== SPEED_10000
)
2278 return FW_PORT_CAP_SPEED_10G
;
2279 if (speed
== 40000) /* Need SPEED_40000 in ethtool.h */
2280 return FW_PORT_CAP_SPEED_40G
;
2284 static int set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2287 struct port_info
*p
= netdev_priv(dev
);
2288 struct link_config
*lc
= &p
->link_cfg
;
2289 u32 speed
= ethtool_cmd_speed(cmd
);
2291 if (cmd
->duplex
!= DUPLEX_FULL
) /* only full-duplex supported */
2294 if (!(lc
->supported
& FW_PORT_CAP_ANEG
)) {
2296 * PHY offers a single speed. See if that's what's
2299 if (cmd
->autoneg
== AUTONEG_DISABLE
&&
2300 (lc
->supported
& speed_to_caps(speed
)))
2305 if (cmd
->autoneg
== AUTONEG_DISABLE
) {
2306 cap
= speed_to_caps(speed
);
2308 if (!(lc
->supported
& cap
) ||
2309 (speed
== SPEED_1000
) ||
2310 (speed
== SPEED_10000
) ||
2313 lc
->requested_speed
= cap
;
2314 lc
->advertising
= 0;
2316 cap
= to_fw_linkcaps(cmd
->advertising
);
2317 if (!(lc
->supported
& cap
))
2319 lc
->requested_speed
= 0;
2320 lc
->advertising
= cap
| FW_PORT_CAP_ANEG
;
2322 lc
->autoneg
= cmd
->autoneg
;
2324 if (netif_running(dev
))
2325 return t4_link_start(p
->adapter
, p
->adapter
->fn
, p
->tx_chan
,
2330 static void get_pauseparam(struct net_device
*dev
,
2331 struct ethtool_pauseparam
*epause
)
2333 struct port_info
*p
= netdev_priv(dev
);
2335 epause
->autoneg
= (p
->link_cfg
.requested_fc
& PAUSE_AUTONEG
) != 0;
2336 epause
->rx_pause
= (p
->link_cfg
.fc
& PAUSE_RX
) != 0;
2337 epause
->tx_pause
= (p
->link_cfg
.fc
& PAUSE_TX
) != 0;
2340 static int set_pauseparam(struct net_device
*dev
,
2341 struct ethtool_pauseparam
*epause
)
2343 struct port_info
*p
= netdev_priv(dev
);
2344 struct link_config
*lc
= &p
->link_cfg
;
2346 if (epause
->autoneg
== AUTONEG_DISABLE
)
2347 lc
->requested_fc
= 0;
2348 else if (lc
->supported
& FW_PORT_CAP_ANEG
)
2349 lc
->requested_fc
= PAUSE_AUTONEG
;
2353 if (epause
->rx_pause
)
2354 lc
->requested_fc
|= PAUSE_RX
;
2355 if (epause
->tx_pause
)
2356 lc
->requested_fc
|= PAUSE_TX
;
2357 if (netif_running(dev
))
2358 return t4_link_start(p
->adapter
, p
->adapter
->fn
, p
->tx_chan
,
2363 static void get_sge_param(struct net_device
*dev
, struct ethtool_ringparam
*e
)
2365 const struct port_info
*pi
= netdev_priv(dev
);
2366 const struct sge
*s
= &pi
->adapter
->sge
;
2368 e
->rx_max_pending
= MAX_RX_BUFFERS
;
2369 e
->rx_mini_max_pending
= MAX_RSPQ_ENTRIES
;
2370 e
->rx_jumbo_max_pending
= 0;
2371 e
->tx_max_pending
= MAX_TXQ_ENTRIES
;
2373 e
->rx_pending
= s
->ethrxq
[pi
->first_qset
].fl
.size
- 8;
2374 e
->rx_mini_pending
= s
->ethrxq
[pi
->first_qset
].rspq
.size
;
2375 e
->rx_jumbo_pending
= 0;
2376 e
->tx_pending
= s
->ethtxq
[pi
->first_qset
].q
.size
;
2379 static int set_sge_param(struct net_device
*dev
, struct ethtool_ringparam
*e
)
2382 const struct port_info
*pi
= netdev_priv(dev
);
2383 struct adapter
*adapter
= pi
->adapter
;
2384 struct sge
*s
= &adapter
->sge
;
2386 if (e
->rx_pending
> MAX_RX_BUFFERS
|| e
->rx_jumbo_pending
||
2387 e
->tx_pending
> MAX_TXQ_ENTRIES
||
2388 e
->rx_mini_pending
> MAX_RSPQ_ENTRIES
||
2389 e
->rx_mini_pending
< MIN_RSPQ_ENTRIES
||
2390 e
->rx_pending
< MIN_FL_ENTRIES
|| e
->tx_pending
< MIN_TXQ_ENTRIES
)
2393 if (adapter
->flags
& FULL_INIT_DONE
)
2396 for (i
= 0; i
< pi
->nqsets
; ++i
) {
2397 s
->ethtxq
[pi
->first_qset
+ i
].q
.size
= e
->tx_pending
;
2398 s
->ethrxq
[pi
->first_qset
+ i
].fl
.size
= e
->rx_pending
+ 8;
2399 s
->ethrxq
[pi
->first_qset
+ i
].rspq
.size
= e
->rx_mini_pending
;
2404 static int closest_timer(const struct sge
*s
, int time
)
2406 int i
, delta
, match
= 0, min_delta
= INT_MAX
;
2408 for (i
= 0; i
< ARRAY_SIZE(s
->timer_val
); i
++) {
2409 delta
= time
- s
->timer_val
[i
];
2412 if (delta
< min_delta
) {
2420 static int closest_thres(const struct sge
*s
, int thres
)
2422 int i
, delta
, match
= 0, min_delta
= INT_MAX
;
2424 for (i
= 0; i
< ARRAY_SIZE(s
->counter_val
); i
++) {
2425 delta
= thres
- s
->counter_val
[i
];
2428 if (delta
< min_delta
) {
2437 * Return a queue's interrupt hold-off time in us. 0 means no timer.
2439 static unsigned int qtimer_val(const struct adapter
*adap
,
2440 const struct sge_rspq
*q
)
2442 unsigned int idx
= q
->intr_params
>> 1;
2444 return idx
< SGE_NTIMERS
? adap
->sge
.timer_val
[idx
] : 0;
2448 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
2449 * @adap: the adapter
2451 * @us: the hold-off time in us, or 0 to disable timer
2452 * @cnt: the hold-off packet count, or 0 to disable counter
2454 * Sets an Rx queue's interrupt hold-off time and packet count. At least
2455 * one of the two needs to be enabled for the queue to generate interrupts.
2457 static int set_rxq_intr_params(struct adapter
*adap
, struct sge_rspq
*q
,
2458 unsigned int us
, unsigned int cnt
)
2460 if ((us
| cnt
) == 0)
2467 new_idx
= closest_thres(&adap
->sge
, cnt
);
2468 if (q
->desc
&& q
->pktcnt_idx
!= new_idx
) {
2469 /* the queue has already been created, update it */
2470 v
= FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ
) |
2471 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH
) |
2472 FW_PARAMS_PARAM_YZ(q
->cntxt_id
);
2473 err
= t4_set_params(adap
, adap
->fn
, adap
->fn
, 0, 1, &v
,
2478 q
->pktcnt_idx
= new_idx
;
2481 us
= us
== 0 ? 6 : closest_timer(&adap
->sge
, us
);
2482 q
->intr_params
= QINTR_TIMER_IDX(us
) | (cnt
> 0 ? QINTR_CNT_EN
: 0);
2486 static int set_coalesce(struct net_device
*dev
, struct ethtool_coalesce
*c
)
2488 const struct port_info
*pi
= netdev_priv(dev
);
2489 struct adapter
*adap
= pi
->adapter
;
2494 for (i
= pi
->first_qset
; i
< pi
->first_qset
+ pi
->nqsets
; i
++) {
2495 q
= &adap
->sge
.ethrxq
[i
].rspq
;
2496 r
= set_rxq_intr_params(adap
, q
, c
->rx_coalesce_usecs
,
2497 c
->rx_max_coalesced_frames
);
2499 dev_err(&dev
->dev
, "failed to set coalesce %d\n", r
);
2506 static int get_coalesce(struct net_device
*dev
, struct ethtool_coalesce
*c
)
2508 const struct port_info
*pi
= netdev_priv(dev
);
2509 const struct adapter
*adap
= pi
->adapter
;
2510 const struct sge_rspq
*rq
= &adap
->sge
.ethrxq
[pi
->first_qset
].rspq
;
2512 c
->rx_coalesce_usecs
= qtimer_val(adap
, rq
);
2513 c
->rx_max_coalesced_frames
= (rq
->intr_params
& QINTR_CNT_EN
) ?
2514 adap
->sge
.counter_val
[rq
->pktcnt_idx
] : 0;
2519 * eeprom_ptov - translate a physical EEPROM address to virtual
2520 * @phys_addr: the physical EEPROM address
2521 * @fn: the PCI function number
2522 * @sz: size of function-specific area
2524 * Translate a physical EEPROM address to virtual. The first 1K is
2525 * accessed through virtual addresses starting at 31K, the rest is
2526 * accessed through virtual addresses starting at 0.
2528 * The mapping is as follows:
2529 * [0..1K) -> [31K..32K)
2530 * [1K..1K+A) -> [31K-A..31K)
2531 * [1K+A..ES) -> [0..ES-A-1K)
2533 * where A = @fn * @sz, and ES = EEPROM size.
2535 static int eeprom_ptov(unsigned int phys_addr
, unsigned int fn
, unsigned int sz
)
2538 if (phys_addr
< 1024)
2539 return phys_addr
+ (31 << 10);
2540 if (phys_addr
< 1024 + fn
)
2541 return 31744 - fn
+ phys_addr
- 1024;
2542 if (phys_addr
< EEPROMSIZE
)
2543 return phys_addr
- 1024 - fn
;
2548 * The next two routines implement eeprom read/write from physical addresses.
2550 static int eeprom_rd_phys(struct adapter
*adap
, unsigned int phys_addr
, u32
*v
)
2552 int vaddr
= eeprom_ptov(phys_addr
, adap
->fn
, EEPROMPFSIZE
);
2555 vaddr
= pci_read_vpd(adap
->pdev
, vaddr
, sizeof(u32
), v
);
2556 return vaddr
< 0 ? vaddr
: 0;
2559 static int eeprom_wr_phys(struct adapter
*adap
, unsigned int phys_addr
, u32 v
)
2561 int vaddr
= eeprom_ptov(phys_addr
, adap
->fn
, EEPROMPFSIZE
);
2564 vaddr
= pci_write_vpd(adap
->pdev
, vaddr
, sizeof(u32
), &v
);
2565 return vaddr
< 0 ? vaddr
: 0;
2568 #define EEPROM_MAGIC 0x38E2F10C
2570 static int get_eeprom(struct net_device
*dev
, struct ethtool_eeprom
*e
,
2574 struct adapter
*adapter
= netdev2adap(dev
);
2576 u8
*buf
= kmalloc(EEPROMSIZE
, GFP_KERNEL
);
2580 e
->magic
= EEPROM_MAGIC
;
2581 for (i
= e
->offset
& ~3; !err
&& i
< e
->offset
+ e
->len
; i
+= 4)
2582 err
= eeprom_rd_phys(adapter
, i
, (u32
*)&buf
[i
]);
2585 memcpy(data
, buf
+ e
->offset
, e
->len
);
2590 static int set_eeprom(struct net_device
*dev
, struct ethtool_eeprom
*eeprom
,
2595 u32 aligned_offset
, aligned_len
, *p
;
2596 struct adapter
*adapter
= netdev2adap(dev
);
2598 if (eeprom
->magic
!= EEPROM_MAGIC
)
2601 aligned_offset
= eeprom
->offset
& ~3;
2602 aligned_len
= (eeprom
->len
+ (eeprom
->offset
& 3) + 3) & ~3;
2604 if (adapter
->fn
> 0) {
2605 u32 start
= 1024 + adapter
->fn
* EEPROMPFSIZE
;
2607 if (aligned_offset
< start
||
2608 aligned_offset
+ aligned_len
> start
+ EEPROMPFSIZE
)
2612 if (aligned_offset
!= eeprom
->offset
|| aligned_len
!= eeprom
->len
) {
2614 * RMW possibly needed for first or last words.
2616 buf
= kmalloc(aligned_len
, GFP_KERNEL
);
2619 err
= eeprom_rd_phys(adapter
, aligned_offset
, (u32
*)buf
);
2620 if (!err
&& aligned_len
> 4)
2621 err
= eeprom_rd_phys(adapter
,
2622 aligned_offset
+ aligned_len
- 4,
2623 (u32
*)&buf
[aligned_len
- 4]);
2626 memcpy(buf
+ (eeprom
->offset
& 3), data
, eeprom
->len
);
2630 err
= t4_seeprom_wp(adapter
, false);
2634 for (p
= (u32
*)buf
; !err
&& aligned_len
; aligned_len
-= 4, p
++) {
2635 err
= eeprom_wr_phys(adapter
, aligned_offset
, *p
);
2636 aligned_offset
+= 4;
2640 err
= t4_seeprom_wp(adapter
, true);
2647 static int set_flash(struct net_device
*netdev
, struct ethtool_flash
*ef
)
2650 const struct firmware
*fw
;
2651 struct adapter
*adap
= netdev2adap(netdev
);
2653 ef
->data
[sizeof(ef
->data
) - 1] = '\0';
2654 ret
= request_firmware(&fw
, ef
->data
, adap
->pdev_dev
);
2658 ret
= t4_load_fw(adap
, fw
->data
, fw
->size
);
2659 release_firmware(fw
);
2661 dev_info(adap
->pdev_dev
, "loaded firmware %s\n", ef
->data
);
2665 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2666 #define BCAST_CRC 0xa0ccc1a6
2668 static void get_wol(struct net_device
*dev
, struct ethtool_wolinfo
*wol
)
2670 wol
->supported
= WAKE_BCAST
| WAKE_MAGIC
;
2671 wol
->wolopts
= netdev2adap(dev
)->wol
;
2672 memset(&wol
->sopass
, 0, sizeof(wol
->sopass
));
2675 static int set_wol(struct net_device
*dev
, struct ethtool_wolinfo
*wol
)
2678 struct port_info
*pi
= netdev_priv(dev
);
2680 if (wol
->wolopts
& ~WOL_SUPPORTED
)
2682 t4_wol_magic_enable(pi
->adapter
, pi
->tx_chan
,
2683 (wol
->wolopts
& WAKE_MAGIC
) ? dev
->dev_addr
: NULL
);
2684 if (wol
->wolopts
& WAKE_BCAST
) {
2685 err
= t4_wol_pat_enable(pi
->adapter
, pi
->tx_chan
, 0xfe, ~0ULL,
2688 err
= t4_wol_pat_enable(pi
->adapter
, pi
->tx_chan
, 1,
2689 ~6ULL, ~0ULL, BCAST_CRC
, true);
2691 t4_wol_pat_enable(pi
->adapter
, pi
->tx_chan
, 0, 0, 0, 0, false);
2695 static int cxgb_set_features(struct net_device
*dev
, netdev_features_t features
)
2697 const struct port_info
*pi
= netdev_priv(dev
);
2698 netdev_features_t changed
= dev
->features
^ features
;
2701 if (!(changed
& NETIF_F_HW_VLAN_CTAG_RX
))
2704 err
= t4_set_rxmode(pi
->adapter
, pi
->adapter
->fn
, pi
->viid
, -1,
2706 !!(features
& NETIF_F_HW_VLAN_CTAG_RX
), true);
2708 dev
->features
= features
^ NETIF_F_HW_VLAN_CTAG_RX
;
2712 static u32
get_rss_table_size(struct net_device
*dev
)
2714 const struct port_info
*pi
= netdev_priv(dev
);
2716 return pi
->rss_size
;
2719 static int get_rss_table(struct net_device
*dev
, u32
*p
)
2721 const struct port_info
*pi
= netdev_priv(dev
);
2722 unsigned int n
= pi
->rss_size
;
2729 static int set_rss_table(struct net_device
*dev
, const u32
*p
)
2732 struct port_info
*pi
= netdev_priv(dev
);
2734 for (i
= 0; i
< pi
->rss_size
; i
++)
2736 if (pi
->adapter
->flags
& FULL_INIT_DONE
)
2737 return write_rss(pi
, pi
->rss
);
2741 static int get_rxnfc(struct net_device
*dev
, struct ethtool_rxnfc
*info
,
2744 const struct port_info
*pi
= netdev_priv(dev
);
2746 switch (info
->cmd
) {
2747 case ETHTOOL_GRXFH
: {
2748 unsigned int v
= pi
->rss_mode
;
2751 switch (info
->flow_type
) {
2753 if (v
& FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN
)
2754 info
->data
= RXH_IP_SRC
| RXH_IP_DST
|
2755 RXH_L4_B_0_1
| RXH_L4_B_2_3
;
2756 else if (v
& FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN
)
2757 info
->data
= RXH_IP_SRC
| RXH_IP_DST
;
2760 if ((v
& FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN
) &&
2761 (v
& FW_RSS_VI_CONFIG_CMD_UDPEN
))
2762 info
->data
= RXH_IP_SRC
| RXH_IP_DST
|
2763 RXH_L4_B_0_1
| RXH_L4_B_2_3
;
2764 else if (v
& FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN
)
2765 info
->data
= RXH_IP_SRC
| RXH_IP_DST
;
2768 case AH_ESP_V4_FLOW
:
2770 if (v
& FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN
)
2771 info
->data
= RXH_IP_SRC
| RXH_IP_DST
;
2774 if (v
& FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN
)
2775 info
->data
= RXH_IP_SRC
| RXH_IP_DST
|
2776 RXH_L4_B_0_1
| RXH_L4_B_2_3
;
2777 else if (v
& FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN
)
2778 info
->data
= RXH_IP_SRC
| RXH_IP_DST
;
2781 if ((v
& FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN
) &&
2782 (v
& FW_RSS_VI_CONFIG_CMD_UDPEN
))
2783 info
->data
= RXH_IP_SRC
| RXH_IP_DST
|
2784 RXH_L4_B_0_1
| RXH_L4_B_2_3
;
2785 else if (v
& FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN
)
2786 info
->data
= RXH_IP_SRC
| RXH_IP_DST
;
2789 case AH_ESP_V6_FLOW
:
2791 if (v
& FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN
)
2792 info
->data
= RXH_IP_SRC
| RXH_IP_DST
;
2797 case ETHTOOL_GRXRINGS
:
2798 info
->data
= pi
->nqsets
;
2804 static const struct ethtool_ops cxgb_ethtool_ops
= {
2805 .get_settings
= get_settings
,
2806 .set_settings
= set_settings
,
2807 .get_drvinfo
= get_drvinfo
,
2808 .get_msglevel
= get_msglevel
,
2809 .set_msglevel
= set_msglevel
,
2810 .get_ringparam
= get_sge_param
,
2811 .set_ringparam
= set_sge_param
,
2812 .get_coalesce
= get_coalesce
,
2813 .set_coalesce
= set_coalesce
,
2814 .get_eeprom_len
= get_eeprom_len
,
2815 .get_eeprom
= get_eeprom
,
2816 .set_eeprom
= set_eeprom
,
2817 .get_pauseparam
= get_pauseparam
,
2818 .set_pauseparam
= set_pauseparam
,
2819 .get_link
= ethtool_op_get_link
,
2820 .get_strings
= get_strings
,
2821 .set_phys_id
= identify_port
,
2822 .nway_reset
= restart_autoneg
,
2823 .get_sset_count
= get_sset_count
,
2824 .get_ethtool_stats
= get_stats
,
2825 .get_regs_len
= get_regs_len
,
2826 .get_regs
= get_regs
,
2829 .get_rxnfc
= get_rxnfc
,
2830 .get_rxfh_indir_size
= get_rss_table_size
,
2831 .get_rxfh_indir
= get_rss_table
,
2832 .set_rxfh_indir
= set_rss_table
,
2833 .flash_device
= set_flash
,
2839 static ssize_t
mem_read(struct file
*file
, char __user
*buf
, size_t count
,
2843 loff_t avail
= file_inode(file
)->i_size
;
2844 unsigned int mem
= (uintptr_t)file
->private_data
& 3;
2845 struct adapter
*adap
= file
->private_data
- mem
;
2851 if (count
> avail
- pos
)
2852 count
= avail
- pos
;
2859 if ((mem
== MEM_MC
) || (mem
== MEM_MC1
))
2860 ret
= t4_mc_read(adap
, mem
% MEM_MC
, pos
, data
, NULL
);
2862 ret
= t4_edc_read(adap
, mem
, pos
, data
, NULL
);
2866 ofst
= pos
% sizeof(data
);
2867 len
= min(count
, sizeof(data
) - ofst
);
2868 if (copy_to_user(buf
, (u8
*)data
+ ofst
, len
))
2875 count
= pos
- *ppos
;
2880 static const struct file_operations mem_debugfs_fops
= {
2881 .owner
= THIS_MODULE
,
2882 .open
= simple_open
,
2884 .llseek
= default_llseek
,
2887 static void add_debugfs_mem(struct adapter
*adap
, const char *name
,
2888 unsigned int idx
, unsigned int size_mb
)
2892 de
= debugfs_create_file(name
, S_IRUSR
, adap
->debugfs_root
,
2893 (void *)adap
+ idx
, &mem_debugfs_fops
);
2894 if (de
&& de
->d_inode
)
2895 de
->d_inode
->i_size
= size_mb
<< 20;
2898 static int setup_debugfs(struct adapter
*adap
)
2903 if (IS_ERR_OR_NULL(adap
->debugfs_root
))
2906 i
= t4_read_reg(adap
, MA_TARGET_MEM_ENABLE
);
2907 if (i
& EDRAM0_ENABLE
) {
2908 size
= t4_read_reg(adap
, MA_EDRAM0_BAR
);
2909 add_debugfs_mem(adap
, "edc0", MEM_EDC0
, EDRAM_SIZE_GET(size
));
2911 if (i
& EDRAM1_ENABLE
) {
2912 size
= t4_read_reg(adap
, MA_EDRAM1_BAR
);
2913 add_debugfs_mem(adap
, "edc1", MEM_EDC1
, EDRAM_SIZE_GET(size
));
2915 if (is_t4(adap
->params
.chip
)) {
2916 size
= t4_read_reg(adap
, MA_EXT_MEMORY_BAR
);
2917 if (i
& EXT_MEM_ENABLE
)
2918 add_debugfs_mem(adap
, "mc", MEM_MC
,
2919 EXT_MEM_SIZE_GET(size
));
2921 if (i
& EXT_MEM_ENABLE
) {
2922 size
= t4_read_reg(adap
, MA_EXT_MEMORY_BAR
);
2923 add_debugfs_mem(adap
, "mc0", MEM_MC0
,
2924 EXT_MEM_SIZE_GET(size
));
2926 if (i
& EXT_MEM1_ENABLE
) {
2927 size
= t4_read_reg(adap
, MA_EXT_MEMORY1_BAR
);
2928 add_debugfs_mem(adap
, "mc1", MEM_MC1
,
2929 EXT_MEM_SIZE_GET(size
));
2933 debugfs_create_file("l2t", S_IRUSR
, adap
->debugfs_root
, adap
,
2939 * upper-layer driver support
2943 * Allocate an active-open TID and set it to the supplied value.
2945 int cxgb4_alloc_atid(struct tid_info
*t
, void *data
)
2949 spin_lock_bh(&t
->atid_lock
);
2951 union aopen_entry
*p
= t
->afree
;
2953 atid
= (p
- t
->atid_tab
) + t
->atid_base
;
2958 spin_unlock_bh(&t
->atid_lock
);
2961 EXPORT_SYMBOL(cxgb4_alloc_atid
);
2964 * Release an active-open TID.
2966 void cxgb4_free_atid(struct tid_info
*t
, unsigned int atid
)
2968 union aopen_entry
*p
= &t
->atid_tab
[atid
- t
->atid_base
];
2970 spin_lock_bh(&t
->atid_lock
);
2974 spin_unlock_bh(&t
->atid_lock
);
2976 EXPORT_SYMBOL(cxgb4_free_atid
);
2979 * Allocate a server TID and set it to the supplied value.
2981 int cxgb4_alloc_stid(struct tid_info
*t
, int family
, void *data
)
2985 spin_lock_bh(&t
->stid_lock
);
2986 if (family
== PF_INET
) {
2987 stid
= find_first_zero_bit(t
->stid_bmap
, t
->nstids
);
2988 if (stid
< t
->nstids
)
2989 __set_bit(stid
, t
->stid_bmap
);
2993 stid
= bitmap_find_free_region(t
->stid_bmap
, t
->nstids
, 2);
2998 t
->stid_tab
[stid
].data
= data
;
2999 stid
+= t
->stid_base
;
3000 /* IPv6 requires max of 520 bits or 16 cells in TCAM
3001 * This is equivalent to 4 TIDs. With CLIP enabled it
3004 if (family
== PF_INET
)
3007 t
->stids_in_use
+= 4;
3009 spin_unlock_bh(&t
->stid_lock
);
3012 EXPORT_SYMBOL(cxgb4_alloc_stid
);
3014 /* Allocate a server filter TID and set it to the supplied value.
3016 int cxgb4_alloc_sftid(struct tid_info
*t
, int family
, void *data
)
3020 spin_lock_bh(&t
->stid_lock
);
3021 if (family
== PF_INET
) {
3022 stid
= find_next_zero_bit(t
->stid_bmap
,
3023 t
->nstids
+ t
->nsftids
, t
->nstids
);
3024 if (stid
< (t
->nstids
+ t
->nsftids
))
3025 __set_bit(stid
, t
->stid_bmap
);
3032 t
->stid_tab
[stid
].data
= data
;
3034 stid
+= t
->sftid_base
;
3037 spin_unlock_bh(&t
->stid_lock
);
3040 EXPORT_SYMBOL(cxgb4_alloc_sftid
);
3042 /* Release a server TID.
3044 void cxgb4_free_stid(struct tid_info
*t
, unsigned int stid
, int family
)
3046 /* Is it a server filter TID? */
3047 if (t
->nsftids
&& (stid
>= t
->sftid_base
)) {
3048 stid
-= t
->sftid_base
;
3051 stid
-= t
->stid_base
;
3054 spin_lock_bh(&t
->stid_lock
);
3055 if (family
== PF_INET
)
3056 __clear_bit(stid
, t
->stid_bmap
);
3058 bitmap_release_region(t
->stid_bmap
, stid
, 2);
3059 t
->stid_tab
[stid
].data
= NULL
;
3060 if (family
== PF_INET
)
3063 t
->stids_in_use
-= 4;
3064 spin_unlock_bh(&t
->stid_lock
);
3066 EXPORT_SYMBOL(cxgb4_free_stid
);
3069 * Populate a TID_RELEASE WR. Caller must properly size the skb.
3071 static void mk_tid_release(struct sk_buff
*skb
, unsigned int chan
,
3074 struct cpl_tid_release
*req
;
3076 set_wr_txq(skb
, CPL_PRIORITY_SETUP
, chan
);
3077 req
= (struct cpl_tid_release
*)__skb_put(skb
, sizeof(*req
));
3078 INIT_TP_WR(req
, tid
);
3079 OPCODE_TID(req
) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE
, tid
));
3083 * Queue a TID release request and if necessary schedule a work queue to
3086 static void cxgb4_queue_tid_release(struct tid_info
*t
, unsigned int chan
,
3089 void **p
= &t
->tid_tab
[tid
];
3090 struct adapter
*adap
= container_of(t
, struct adapter
, tids
);
3092 spin_lock_bh(&adap
->tid_release_lock
);
3093 *p
= adap
->tid_release_head
;
3094 /* Low 2 bits encode the Tx channel number */
3095 adap
->tid_release_head
= (void **)((uintptr_t)p
| chan
);
3096 if (!adap
->tid_release_task_busy
) {
3097 adap
->tid_release_task_busy
= true;
3098 queue_work(workq
, &adap
->tid_release_task
);
3100 spin_unlock_bh(&adap
->tid_release_lock
);
3104 * Process the list of pending TID release requests.
3106 static void process_tid_release_list(struct work_struct
*work
)
3108 struct sk_buff
*skb
;
3109 struct adapter
*adap
;
3111 adap
= container_of(work
, struct adapter
, tid_release_task
);
3113 spin_lock_bh(&adap
->tid_release_lock
);
3114 while (adap
->tid_release_head
) {
3115 void **p
= adap
->tid_release_head
;
3116 unsigned int chan
= (uintptr_t)p
& 3;
3117 p
= (void *)p
- chan
;
3119 adap
->tid_release_head
= *p
;
3121 spin_unlock_bh(&adap
->tid_release_lock
);
3123 while (!(skb
= alloc_skb(sizeof(struct cpl_tid_release
),
3125 schedule_timeout_uninterruptible(1);
3127 mk_tid_release(skb
, chan
, p
- adap
->tids
.tid_tab
);
3128 t4_ofld_send(adap
, skb
);
3129 spin_lock_bh(&adap
->tid_release_lock
);
3131 adap
->tid_release_task_busy
= false;
3132 spin_unlock_bh(&adap
->tid_release_lock
);
3136 * Release a TID and inform HW. If we are unable to allocate the release
3137 * message we defer to a work queue.
3139 void cxgb4_remove_tid(struct tid_info
*t
, unsigned int chan
, unsigned int tid
)
3142 struct sk_buff
*skb
;
3143 struct adapter
*adap
= container_of(t
, struct adapter
, tids
);
3145 old
= t
->tid_tab
[tid
];
3146 skb
= alloc_skb(sizeof(struct cpl_tid_release
), GFP_ATOMIC
);
3148 t
->tid_tab
[tid
] = NULL
;
3149 mk_tid_release(skb
, chan
, tid
);
3150 t4_ofld_send(adap
, skb
);
3152 cxgb4_queue_tid_release(t
, chan
, tid
);
3154 atomic_dec(&t
->tids_in_use
);
3156 EXPORT_SYMBOL(cxgb4_remove_tid
);
3159 * Allocate and initialize the TID tables. Returns 0 on success.
3161 static int tid_init(struct tid_info
*t
)
3164 unsigned int stid_bmap_size
;
3165 unsigned int natids
= t
->natids
;
3166 struct adapter
*adap
= container_of(t
, struct adapter
, tids
);
3168 stid_bmap_size
= BITS_TO_LONGS(t
->nstids
+ t
->nsftids
);
3169 size
= t
->ntids
* sizeof(*t
->tid_tab
) +
3170 natids
* sizeof(*t
->atid_tab
) +
3171 t
->nstids
* sizeof(*t
->stid_tab
) +
3172 t
->nsftids
* sizeof(*t
->stid_tab
) +
3173 stid_bmap_size
* sizeof(long) +
3174 t
->nftids
* sizeof(*t
->ftid_tab
) +
3175 t
->nsftids
* sizeof(*t
->ftid_tab
);
3177 t
->tid_tab
= t4_alloc_mem(size
);
3181 t
->atid_tab
= (union aopen_entry
*)&t
->tid_tab
[t
->ntids
];
3182 t
->stid_tab
= (struct serv_entry
*)&t
->atid_tab
[natids
];
3183 t
->stid_bmap
= (unsigned long *)&t
->stid_tab
[t
->nstids
+ t
->nsftids
];
3184 t
->ftid_tab
= (struct filter_entry
*)&t
->stid_bmap
[stid_bmap_size
];
3185 spin_lock_init(&t
->stid_lock
);
3186 spin_lock_init(&t
->atid_lock
);
3188 t
->stids_in_use
= 0;
3190 t
->atids_in_use
= 0;
3191 atomic_set(&t
->tids_in_use
, 0);
3193 /* Setup the free list for atid_tab and clear the stid bitmap. */
3196 t
->atid_tab
[natids
- 1].next
= &t
->atid_tab
[natids
];
3197 t
->afree
= t
->atid_tab
;
3199 bitmap_zero(t
->stid_bmap
, t
->nstids
+ t
->nsftids
);
3200 /* Reserve stid 0 for T4/T5 adapters */
3201 if (!t
->stid_base
&&
3202 (is_t4(adap
->params
.chip
) || is_t5(adap
->params
.chip
)))
3203 __set_bit(0, t
->stid_bmap
);
3208 static int cxgb4_clip_get(const struct net_device
*dev
,
3209 const struct in6_addr
*lip
)
3211 struct adapter
*adap
;
3212 struct fw_clip_cmd c
;
3214 adap
= netdev2adap(dev
);
3215 memset(&c
, 0, sizeof(c
));
3216 c
.op_to_write
= htonl(FW_CMD_OP(FW_CLIP_CMD
) |
3217 FW_CMD_REQUEST
| FW_CMD_WRITE
);
3218 c
.alloc_to_len16
= htonl(F_FW_CLIP_CMD_ALLOC
| FW_LEN16(c
));
3219 *(__be64
*)&c
.ip_hi
= *(__be64
*)(lip
->s6_addr
);
3220 *(__be64
*)&c
.ip_lo
= *(__be64
*)(lip
->s6_addr
+ 8);
3221 return t4_wr_mbox_meat(adap
, adap
->mbox
, &c
, sizeof(c
), &c
, false);
3224 static int cxgb4_clip_release(const struct net_device
*dev
,
3225 const struct in6_addr
*lip
)
3227 struct adapter
*adap
;
3228 struct fw_clip_cmd c
;
3230 adap
= netdev2adap(dev
);
3231 memset(&c
, 0, sizeof(c
));
3232 c
.op_to_write
= htonl(FW_CMD_OP(FW_CLIP_CMD
) |
3233 FW_CMD_REQUEST
| FW_CMD_READ
);
3234 c
.alloc_to_len16
= htonl(F_FW_CLIP_CMD_FREE
| FW_LEN16(c
));
3235 *(__be64
*)&c
.ip_hi
= *(__be64
*)(lip
->s6_addr
);
3236 *(__be64
*)&c
.ip_lo
= *(__be64
*)(lip
->s6_addr
+ 8);
3237 return t4_wr_mbox_meat(adap
, adap
->mbox
, &c
, sizeof(c
), &c
, false);
3241 * cxgb4_create_server - create an IP server
3243 * @stid: the server TID
3244 * @sip: local IP address to bind server to
3245 * @sport: the server's TCP port
3246 * @queue: queue to direct messages from this server to
3248 * Create an IP server for the given port and address.
3249 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3251 int cxgb4_create_server(const struct net_device
*dev
, unsigned int stid
,
3252 __be32 sip
, __be16 sport
, __be16 vlan
,
3256 struct sk_buff
*skb
;
3257 struct adapter
*adap
;
3258 struct cpl_pass_open_req
*req
;
3261 skb
= alloc_skb(sizeof(*req
), GFP_KERNEL
);
3265 adap
= netdev2adap(dev
);
3266 req
= (struct cpl_pass_open_req
*)__skb_put(skb
, sizeof(*req
));
3268 OPCODE_TID(req
) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ
, stid
));
3269 req
->local_port
= sport
;
3270 req
->peer_port
= htons(0);
3271 req
->local_ip
= sip
;
3272 req
->peer_ip
= htonl(0);
3273 chan
= rxq_to_chan(&adap
->sge
, queue
);
3274 req
->opt0
= cpu_to_be64(TX_CHAN(chan
));
3275 req
->opt1
= cpu_to_be64(CONN_POLICY_ASK
|
3276 SYN_RSS_ENABLE
| SYN_RSS_QUEUE(queue
));
3277 ret
= t4_mgmt_tx(adap
, skb
);
3278 return net_xmit_eval(ret
);
3280 EXPORT_SYMBOL(cxgb4_create_server
);
3282 /* cxgb4_create_server6 - create an IPv6 server
3284 * @stid: the server TID
3285 * @sip: local IPv6 address to bind server to
3286 * @sport: the server's TCP port
3287 * @queue: queue to direct messages from this server to
3289 * Create an IPv6 server for the given port and address.
3290 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3292 int cxgb4_create_server6(const struct net_device
*dev
, unsigned int stid
,
3293 const struct in6_addr
*sip
, __be16 sport
,
3297 struct sk_buff
*skb
;
3298 struct adapter
*adap
;
3299 struct cpl_pass_open_req6
*req
;
3302 skb
= alloc_skb(sizeof(*req
), GFP_KERNEL
);
3306 adap
= netdev2adap(dev
);
3307 req
= (struct cpl_pass_open_req6
*)__skb_put(skb
, sizeof(*req
));
3309 OPCODE_TID(req
) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6
, stid
));
3310 req
->local_port
= sport
;
3311 req
->peer_port
= htons(0);
3312 req
->local_ip_hi
= *(__be64
*)(sip
->s6_addr
);
3313 req
->local_ip_lo
= *(__be64
*)(sip
->s6_addr
+ 8);
3314 req
->peer_ip_hi
= cpu_to_be64(0);
3315 req
->peer_ip_lo
= cpu_to_be64(0);
3316 chan
= rxq_to_chan(&adap
->sge
, queue
);
3317 req
->opt0
= cpu_to_be64(TX_CHAN(chan
));
3318 req
->opt1
= cpu_to_be64(CONN_POLICY_ASK
|
3319 SYN_RSS_ENABLE
| SYN_RSS_QUEUE(queue
));
3320 ret
= t4_mgmt_tx(adap
, skb
);
3321 return net_xmit_eval(ret
);
3323 EXPORT_SYMBOL(cxgb4_create_server6
);
3325 int cxgb4_remove_server(const struct net_device
*dev
, unsigned int stid
,
3326 unsigned int queue
, bool ipv6
)
3328 struct sk_buff
*skb
;
3329 struct adapter
*adap
;
3330 struct cpl_close_listsvr_req
*req
;
3333 adap
= netdev2adap(dev
);
3335 skb
= alloc_skb(sizeof(*req
), GFP_KERNEL
);
3339 req
= (struct cpl_close_listsvr_req
*)__skb_put(skb
, sizeof(*req
));
3341 OPCODE_TID(req
) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ
, stid
));
3342 req
->reply_ctrl
= htons(NO_REPLY(0) | (ipv6
? LISTSVR_IPV6(1) :
3343 LISTSVR_IPV6(0)) | QUEUENO(queue
));
3344 ret
= t4_mgmt_tx(adap
, skb
);
3345 return net_xmit_eval(ret
);
3347 EXPORT_SYMBOL(cxgb4_remove_server
);
3350 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
3351 * @mtus: the HW MTU table
3352 * @mtu: the target MTU
3353 * @idx: index of selected entry in the MTU table
3355 * Returns the index and the value in the HW MTU table that is closest to
3356 * but does not exceed @mtu, unless @mtu is smaller than any value in the
3357 * table, in which case that smallest available value is selected.
3359 unsigned int cxgb4_best_mtu(const unsigned short *mtus
, unsigned short mtu
,
3364 while (i
< NMTUS
- 1 && mtus
[i
+ 1] <= mtu
)
3370 EXPORT_SYMBOL(cxgb4_best_mtu
);
3373 * cxgb4_port_chan - get the HW channel of a port
3374 * @dev: the net device for the port
3376 * Return the HW Tx channel of the given port.
3378 unsigned int cxgb4_port_chan(const struct net_device
*dev
)
3380 return netdev2pinfo(dev
)->tx_chan
;
3382 EXPORT_SYMBOL(cxgb4_port_chan
);
3384 unsigned int cxgb4_dbfifo_count(const struct net_device
*dev
, int lpfifo
)
3386 struct adapter
*adap
= netdev2adap(dev
);
3387 u32 v1
, v2
, lp_count
, hp_count
;
3389 v1
= t4_read_reg(adap
, A_SGE_DBFIFO_STATUS
);
3390 v2
= t4_read_reg(adap
, SGE_DBFIFO_STATUS2
);
3391 if (is_t4(adap
->params
.chip
)) {
3392 lp_count
= G_LP_COUNT(v1
);
3393 hp_count
= G_HP_COUNT(v1
);
3395 lp_count
= G_LP_COUNT_T5(v1
);
3396 hp_count
= G_HP_COUNT_T5(v2
);
3398 return lpfifo
? lp_count
: hp_count
;
3400 EXPORT_SYMBOL(cxgb4_dbfifo_count
);
3403 * cxgb4_port_viid - get the VI id of a port
3404 * @dev: the net device for the port
3406 * Return the VI id of the given port.
3408 unsigned int cxgb4_port_viid(const struct net_device
*dev
)
3410 return netdev2pinfo(dev
)->viid
;
3412 EXPORT_SYMBOL(cxgb4_port_viid
);
3415 * cxgb4_port_idx - get the index of a port
3416 * @dev: the net device for the port
3418 * Return the index of the given port.
3420 unsigned int cxgb4_port_idx(const struct net_device
*dev
)
3422 return netdev2pinfo(dev
)->port_id
;
3424 EXPORT_SYMBOL(cxgb4_port_idx
);
3426 void cxgb4_get_tcp_stats(struct pci_dev
*pdev
, struct tp_tcp_stats
*v4
,
3427 struct tp_tcp_stats
*v6
)
3429 struct adapter
*adap
= pci_get_drvdata(pdev
);
3431 spin_lock(&adap
->stats_lock
);
3432 t4_tp_get_tcp_stats(adap
, v4
, v6
);
3433 spin_unlock(&adap
->stats_lock
);
3435 EXPORT_SYMBOL(cxgb4_get_tcp_stats
);
3437 void cxgb4_iscsi_init(struct net_device
*dev
, unsigned int tag_mask
,
3438 const unsigned int *pgsz_order
)
3440 struct adapter
*adap
= netdev2adap(dev
);
3442 t4_write_reg(adap
, ULP_RX_ISCSI_TAGMASK
, tag_mask
);
3443 t4_write_reg(adap
, ULP_RX_ISCSI_PSZ
, HPZ0(pgsz_order
[0]) |
3444 HPZ1(pgsz_order
[1]) | HPZ2(pgsz_order
[2]) |
3445 HPZ3(pgsz_order
[3]));
3447 EXPORT_SYMBOL(cxgb4_iscsi_init
);
3449 int cxgb4_flush_eq_cache(struct net_device
*dev
)
3451 struct adapter
*adap
= netdev2adap(dev
);
3454 ret
= t4_fwaddrspace_write(adap
, adap
->mbox
,
3455 0xe1000000 + A_SGE_CTXT_CMD
, 0x20000000);
3458 EXPORT_SYMBOL(cxgb4_flush_eq_cache
);
3460 static int read_eq_indices(struct adapter
*adap
, u16 qid
, u16
*pidx
, u16
*cidx
)
3462 u32 addr
= t4_read_reg(adap
, A_SGE_DBQ_CTXT_BADDR
) + 24 * qid
+ 8;
3466 ret
= t4_mem_win_read_len(adap
, addr
, (__be32
*)&indices
, 8);
3468 *cidx
= (be64_to_cpu(indices
) >> 25) & 0xffff;
3469 *pidx
= (be64_to_cpu(indices
) >> 9) & 0xffff;
3474 int cxgb4_sync_txq_pidx(struct net_device
*dev
, u16 qid
, u16 pidx
,
3477 struct adapter
*adap
= netdev2adap(dev
);
3478 u16 hw_pidx
, hw_cidx
;
3481 ret
= read_eq_indices(adap
, qid
, &hw_pidx
, &hw_cidx
);
3485 if (pidx
!= hw_pidx
) {
3488 if (pidx
>= hw_pidx
)
3489 delta
= pidx
- hw_pidx
;
3491 delta
= size
- hw_pidx
+ pidx
;
3493 t4_write_reg(adap
, MYPF_REG(SGE_PF_KDOORBELL
),
3494 QID(qid
) | PIDX(delta
));
3499 EXPORT_SYMBOL(cxgb4_sync_txq_pidx
);
3501 void cxgb4_disable_db_coalescing(struct net_device
*dev
)
3503 struct adapter
*adap
;
3505 adap
= netdev2adap(dev
);
3506 t4_set_reg_field(adap
, A_SGE_DOORBELL_CONTROL
, F_NOCOALESCE
,
3509 EXPORT_SYMBOL(cxgb4_disable_db_coalescing
);
3511 void cxgb4_enable_db_coalescing(struct net_device
*dev
)
3513 struct adapter
*adap
;
3515 adap
= netdev2adap(dev
);
3516 t4_set_reg_field(adap
, A_SGE_DOORBELL_CONTROL
, F_NOCOALESCE
, 0);
3518 EXPORT_SYMBOL(cxgb4_enable_db_coalescing
);
3520 static struct pci_driver cxgb4_driver
;
3522 static void check_neigh_update(struct neighbour
*neigh
)
3524 const struct device
*parent
;
3525 const struct net_device
*netdev
= neigh
->dev
;
3527 if (netdev
->priv_flags
& IFF_802_1Q_VLAN
)
3528 netdev
= vlan_dev_real_dev(netdev
);
3529 parent
= netdev
->dev
.parent
;
3530 if (parent
&& parent
->driver
== &cxgb4_driver
.driver
)
3531 t4_l2t_update(dev_get_drvdata(parent
), neigh
);
3534 static int netevent_cb(struct notifier_block
*nb
, unsigned long event
,
3538 case NETEVENT_NEIGH_UPDATE
:
3539 check_neigh_update(data
);
3541 case NETEVENT_REDIRECT
:
3548 static bool netevent_registered
;
3549 static struct notifier_block cxgb4_netevent_nb
= {
3550 .notifier_call
= netevent_cb
3553 static void drain_db_fifo(struct adapter
*adap
, int usecs
)
3555 u32 v1
, v2
, lp_count
, hp_count
;
3558 v1
= t4_read_reg(adap
, A_SGE_DBFIFO_STATUS
);
3559 v2
= t4_read_reg(adap
, SGE_DBFIFO_STATUS2
);
3560 if (is_t4(adap
->params
.chip
)) {
3561 lp_count
= G_LP_COUNT(v1
);
3562 hp_count
= G_HP_COUNT(v1
);
3564 lp_count
= G_LP_COUNT_T5(v1
);
3565 hp_count
= G_HP_COUNT_T5(v2
);
3568 if (lp_count
== 0 && hp_count
== 0)
3570 set_current_state(TASK_UNINTERRUPTIBLE
);
3571 schedule_timeout(usecs_to_jiffies(usecs
));
3575 static void disable_txq_db(struct sge_txq
*q
)
3577 spin_lock_irq(&q
->db_lock
);
3579 spin_unlock_irq(&q
->db_lock
);
3582 static void enable_txq_db(struct sge_txq
*q
)
3584 spin_lock_irq(&q
->db_lock
);
3586 spin_unlock_irq(&q
->db_lock
);
3589 static void disable_dbs(struct adapter
*adap
)
3593 for_each_ethrxq(&adap
->sge
, i
)
3594 disable_txq_db(&adap
->sge
.ethtxq
[i
].q
);
3595 for_each_ofldrxq(&adap
->sge
, i
)
3596 disable_txq_db(&adap
->sge
.ofldtxq
[i
].q
);
3597 for_each_port(adap
, i
)
3598 disable_txq_db(&adap
->sge
.ctrlq
[i
].q
);
3601 static void enable_dbs(struct adapter
*adap
)
3605 for_each_ethrxq(&adap
->sge
, i
)
3606 enable_txq_db(&adap
->sge
.ethtxq
[i
].q
);
3607 for_each_ofldrxq(&adap
->sge
, i
)
3608 enable_txq_db(&adap
->sge
.ofldtxq
[i
].q
);
3609 for_each_port(adap
, i
)
3610 enable_txq_db(&adap
->sge
.ctrlq
[i
].q
);
3613 static void sync_txq_pidx(struct adapter
*adap
, struct sge_txq
*q
)
3615 u16 hw_pidx
, hw_cidx
;
3618 spin_lock_bh(&q
->db_lock
);
3619 ret
= read_eq_indices(adap
, (u16
)q
->cntxt_id
, &hw_pidx
, &hw_cidx
);
3622 if (q
->db_pidx
!= hw_pidx
) {
3625 if (q
->db_pidx
>= hw_pidx
)
3626 delta
= q
->db_pidx
- hw_pidx
;
3628 delta
= q
->size
- hw_pidx
+ q
->db_pidx
;
3630 t4_write_reg(adap
, MYPF_REG(SGE_PF_KDOORBELL
),
3631 QID(q
->cntxt_id
) | PIDX(delta
));
3635 spin_unlock_bh(&q
->db_lock
);
3637 CH_WARN(adap
, "DB drop recovery failed.\n");
3639 static void recover_all_queues(struct adapter
*adap
)
3643 for_each_ethrxq(&adap
->sge
, i
)
3644 sync_txq_pidx(adap
, &adap
->sge
.ethtxq
[i
].q
);
3645 for_each_ofldrxq(&adap
->sge
, i
)
3646 sync_txq_pidx(adap
, &adap
->sge
.ofldtxq
[i
].q
);
3647 for_each_port(adap
, i
)
3648 sync_txq_pidx(adap
, &adap
->sge
.ctrlq
[i
].q
);
3651 static void notify_rdma_uld(struct adapter
*adap
, enum cxgb4_control cmd
)
3653 mutex_lock(&uld_mutex
);
3654 if (adap
->uld_handle
[CXGB4_ULD_RDMA
])
3655 ulds
[CXGB4_ULD_RDMA
].control(adap
->uld_handle
[CXGB4_ULD_RDMA
],
3657 mutex_unlock(&uld_mutex
);
3660 static void process_db_full(struct work_struct
*work
)
3662 struct adapter
*adap
;
3664 adap
= container_of(work
, struct adapter
, db_full_task
);
3666 notify_rdma_uld(adap
, CXGB4_CONTROL_DB_FULL
);
3667 drain_db_fifo(adap
, dbfifo_drain_delay
);
3668 t4_set_reg_field(adap
, SGE_INT_ENABLE3
,
3669 DBFIFO_HP_INT
| DBFIFO_LP_INT
,
3670 DBFIFO_HP_INT
| DBFIFO_LP_INT
);
3671 notify_rdma_uld(adap
, CXGB4_CONTROL_DB_EMPTY
);
3674 static void process_db_drop(struct work_struct
*work
)
3676 struct adapter
*adap
;
3678 adap
= container_of(work
, struct adapter
, db_drop_task
);
3680 if (is_t4(adap
->params
.chip
)) {
3682 notify_rdma_uld(adap
, CXGB4_CONTROL_DB_DROP
);
3683 drain_db_fifo(adap
, 1);
3684 recover_all_queues(adap
);
3687 u32 dropped_db
= t4_read_reg(adap
, 0x010ac);
3688 u16 qid
= (dropped_db
>> 15) & 0x1ffff;
3689 u16 pidx_inc
= dropped_db
& 0x1fff;
3691 unsigned short udb_density
;
3692 unsigned long qpshift
;
3696 dev_warn(adap
->pdev_dev
,
3697 "Dropped DB 0x%x qid %d bar2 %d coalesce %d pidx %d\n",
3699 (dropped_db
>> 14) & 1,
3700 (dropped_db
>> 13) & 1,
3703 drain_db_fifo(adap
, 1);
3705 s_qpp
= QUEUESPERPAGEPF1
* adap
->fn
;
3706 udb_density
= 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adap
,
3707 SGE_EGRESS_QUEUES_PER_PAGE_PF
) >> s_qpp
);
3708 qpshift
= PAGE_SHIFT
- ilog2(udb_density
);
3709 udb
= qid
<< qpshift
;
3711 page
= udb
/ PAGE_SIZE
;
3712 udb
+= (qid
- (page
* udb_density
)) * 128;
3714 writel(PIDX(pidx_inc
), adap
->bar2
+ udb
+ 8);
3716 /* Re-enable BAR2 WC */
3717 t4_set_reg_field(adap
, 0x10b0, 1<<15, 1<<15);
3720 t4_set_reg_field(adap
, A_SGE_DOORBELL_CONTROL
, F_DROPPED_DB
, 0);
3723 void t4_db_full(struct adapter
*adap
)
3725 if (is_t4(adap
->params
.chip
)) {
3726 t4_set_reg_field(adap
, SGE_INT_ENABLE3
,
3727 DBFIFO_HP_INT
| DBFIFO_LP_INT
, 0);
3728 queue_work(workq
, &adap
->db_full_task
);
3732 void t4_db_dropped(struct adapter
*adap
)
3734 if (is_t4(adap
->params
.chip
))
3735 queue_work(workq
, &adap
->db_drop_task
);
3738 static void uld_attach(struct adapter
*adap
, unsigned int uld
)
3741 struct cxgb4_lld_info lli
;
3744 lli
.pdev
= adap
->pdev
;
3745 lli
.l2t
= adap
->l2t
;
3746 lli
.tids
= &adap
->tids
;
3747 lli
.ports
= adap
->port
;
3748 lli
.vr
= &adap
->vres
;
3749 lli
.mtus
= adap
->params
.mtus
;
3750 if (uld
== CXGB4_ULD_RDMA
) {
3751 lli
.rxq_ids
= adap
->sge
.rdma_rxq
;
3752 lli
.nrxq
= adap
->sge
.rdmaqs
;
3753 } else if (uld
== CXGB4_ULD_ISCSI
) {
3754 lli
.rxq_ids
= adap
->sge
.ofld_rxq
;
3755 lli
.nrxq
= adap
->sge
.ofldqsets
;
3757 lli
.ntxq
= adap
->sge
.ofldqsets
;
3758 lli
.nchan
= adap
->params
.nports
;
3759 lli
.nports
= adap
->params
.nports
;
3760 lli
.wr_cred
= adap
->params
.ofldq_wr_cred
;
3761 lli
.adapter_type
= adap
->params
.chip
;
3762 lli
.iscsi_iolen
= MAXRXDATA_GET(t4_read_reg(adap
, TP_PARA_REG2
));
3763 lli
.udb_density
= 1 << QUEUESPERPAGEPF0_GET(
3764 t4_read_reg(adap
, SGE_EGRESS_QUEUES_PER_PAGE_PF
) >>
3766 lli
.ucq_density
= 1 << QUEUESPERPAGEPF0_GET(
3767 t4_read_reg(adap
, SGE_INGRESS_QUEUES_PER_PAGE_PF
) >>
3769 lli
.filt_mode
= adap
->params
.tp
.vlan_pri_map
;
3770 /* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
3771 for (i
= 0; i
< NCHAN
; i
++)
3773 lli
.gts_reg
= adap
->regs
+ MYPF_REG(SGE_PF_GTS
);
3774 lli
.db_reg
= adap
->regs
+ MYPF_REG(SGE_PF_KDOORBELL
);
3775 lli
.fw_vers
= adap
->params
.fw_vers
;
3776 lli
.dbfifo_int_thresh
= dbfifo_int_thresh
;
3777 lli
.sge_pktshift
= adap
->sge
.pktshift
;
3778 lli
.enable_fw_ofld_conn
= adap
->flags
& FW_OFLD_CONN
;
3780 handle
= ulds
[uld
].add(&lli
);
3781 if (IS_ERR(handle
)) {
3782 dev_warn(adap
->pdev_dev
,
3783 "could not attach to the %s driver, error %ld\n",
3784 uld_str
[uld
], PTR_ERR(handle
));
3788 adap
->uld_handle
[uld
] = handle
;
3790 if (!netevent_registered
) {
3791 register_netevent_notifier(&cxgb4_netevent_nb
);
3792 netevent_registered
= true;
3795 if (adap
->flags
& FULL_INIT_DONE
)
3796 ulds
[uld
].state_change(handle
, CXGB4_STATE_UP
);
3799 static void attach_ulds(struct adapter
*adap
)
3803 spin_lock(&adap_rcu_lock
);
3804 list_add_tail_rcu(&adap
->rcu_node
, &adap_rcu_list
);
3805 spin_unlock(&adap_rcu_lock
);
3807 mutex_lock(&uld_mutex
);
3808 list_add_tail(&adap
->list_node
, &adapter_list
);
3809 for (i
= 0; i
< CXGB4_ULD_MAX
; i
++)
3811 uld_attach(adap
, i
);
3812 mutex_unlock(&uld_mutex
);
3815 static void detach_ulds(struct adapter
*adap
)
3819 mutex_lock(&uld_mutex
);
3820 list_del(&adap
->list_node
);
3821 for (i
= 0; i
< CXGB4_ULD_MAX
; i
++)
3822 if (adap
->uld_handle
[i
]) {
3823 ulds
[i
].state_change(adap
->uld_handle
[i
],
3824 CXGB4_STATE_DETACH
);
3825 adap
->uld_handle
[i
] = NULL
;
3827 if (netevent_registered
&& list_empty(&adapter_list
)) {
3828 unregister_netevent_notifier(&cxgb4_netevent_nb
);
3829 netevent_registered
= false;
3831 mutex_unlock(&uld_mutex
);
3833 spin_lock(&adap_rcu_lock
);
3834 list_del_rcu(&adap
->rcu_node
);
3835 spin_unlock(&adap_rcu_lock
);
3838 static void notify_ulds(struct adapter
*adap
, enum cxgb4_state new_state
)
3842 mutex_lock(&uld_mutex
);
3843 for (i
= 0; i
< CXGB4_ULD_MAX
; i
++)
3844 if (adap
->uld_handle
[i
])
3845 ulds
[i
].state_change(adap
->uld_handle
[i
], new_state
);
3846 mutex_unlock(&uld_mutex
);
3850 * cxgb4_register_uld - register an upper-layer driver
3851 * @type: the ULD type
3852 * @p: the ULD methods
3854 * Registers an upper-layer driver with this driver and notifies the ULD
3855 * about any presently available devices that support its type. Returns
3856 * %-EBUSY if a ULD of the same type is already registered.
3858 int cxgb4_register_uld(enum cxgb4_uld type
, const struct cxgb4_uld_info
*p
)
3861 struct adapter
*adap
;
3863 if (type
>= CXGB4_ULD_MAX
)
3865 mutex_lock(&uld_mutex
);
3866 if (ulds
[type
].add
) {
3871 list_for_each_entry(adap
, &adapter_list
, list_node
)
3872 uld_attach(adap
, type
);
3873 out
: mutex_unlock(&uld_mutex
);
3876 EXPORT_SYMBOL(cxgb4_register_uld
);
3879 * cxgb4_unregister_uld - unregister an upper-layer driver
3880 * @type: the ULD type
3882 * Unregisters an existing upper-layer driver.
3884 int cxgb4_unregister_uld(enum cxgb4_uld type
)
3886 struct adapter
*adap
;
3888 if (type
>= CXGB4_ULD_MAX
)
3890 mutex_lock(&uld_mutex
);
3891 list_for_each_entry(adap
, &adapter_list
, list_node
)
3892 adap
->uld_handle
[type
] = NULL
;
3893 ulds
[type
].add
= NULL
;
3894 mutex_unlock(&uld_mutex
);
3897 EXPORT_SYMBOL(cxgb4_unregister_uld
);
3899 /* Check if netdev on which event is occured belongs to us or not. Return
3900 * suceess (1) if it belongs otherwise failure (0).
3902 static int cxgb4_netdev(struct net_device
*netdev
)
3904 struct adapter
*adap
;
3907 spin_lock(&adap_rcu_lock
);
3908 list_for_each_entry_rcu(adap
, &adap_rcu_list
, rcu_node
)
3909 for (i
= 0; i
< MAX_NPORTS
; i
++)
3910 if (adap
->port
[i
] == netdev
) {
3911 spin_unlock(&adap_rcu_lock
);
3914 spin_unlock(&adap_rcu_lock
);
3918 static int clip_add(struct net_device
*event_dev
, struct inet6_ifaddr
*ifa
,
3919 unsigned long event
)
3921 int ret
= NOTIFY_DONE
;
3924 if (cxgb4_netdev(event_dev
)) {
3927 ret
= cxgb4_clip_get(event_dev
,
3928 (const struct in6_addr
*)ifa
->addr
.s6_addr
);
3936 cxgb4_clip_release(event_dev
,
3937 (const struct in6_addr
*)ifa
->addr
.s6_addr
);
3948 static int cxgb4_inet6addr_handler(struct notifier_block
*this,
3949 unsigned long event
, void *data
)
3951 struct inet6_ifaddr
*ifa
= data
;
3952 struct net_device
*event_dev
;
3953 int ret
= NOTIFY_DONE
;
3954 struct bonding
*bond
= netdev_priv(ifa
->idev
->dev
);
3955 struct list_head
*iter
;
3956 struct slave
*slave
;
3957 struct pci_dev
*first_pdev
= NULL
;
3959 if (ifa
->idev
->dev
->priv_flags
& IFF_802_1Q_VLAN
) {
3960 event_dev
= vlan_dev_real_dev(ifa
->idev
->dev
);
3961 ret
= clip_add(event_dev
, ifa
, event
);
3962 } else if (ifa
->idev
->dev
->flags
& IFF_MASTER
) {
3963 /* It is possible that two different adapters are bonded in one
3964 * bond. We need to find such different adapters and add clip
3965 * in all of them only once.
3967 read_lock(&bond
->lock
);
3968 bond_for_each_slave(bond
, slave
, iter
) {
3970 ret
= clip_add(slave
->dev
, ifa
, event
);
3971 /* If clip_add is success then only initialize
3972 * first_pdev since it means it is our device
3974 if (ret
== NOTIFY_OK
)
3975 first_pdev
= to_pci_dev(
3976 slave
->dev
->dev
.parent
);
3977 } else if (first_pdev
!=
3978 to_pci_dev(slave
->dev
->dev
.parent
))
3979 ret
= clip_add(slave
->dev
, ifa
, event
);
3981 read_unlock(&bond
->lock
);
3983 ret
= clip_add(ifa
->idev
->dev
, ifa
, event
);
3988 static struct notifier_block cxgb4_inet6addr_notifier
= {
3989 .notifier_call
= cxgb4_inet6addr_handler
3992 /* Retrieves IPv6 addresses from a root device (bond, vlan) associated with
3993 * a physical device.
3994 * The physical device reference is needed to send the actul CLIP command.
3996 static int update_dev_clip(struct net_device
*root_dev
, struct net_device
*dev
)
3998 struct inet6_dev
*idev
= NULL
;
3999 struct inet6_ifaddr
*ifa
;
4002 idev
= __in6_dev_get(root_dev
);
4006 read_lock_bh(&idev
->lock
);
4007 list_for_each_entry(ifa
, &idev
->addr_list
, if_list
) {
4008 ret
= cxgb4_clip_get(dev
,
4009 (const struct in6_addr
*)ifa
->addr
.s6_addr
);
4013 read_unlock_bh(&idev
->lock
);
4018 static int update_root_dev_clip(struct net_device
*dev
)
4020 struct net_device
*root_dev
= NULL
;
4023 /* First populate the real net device's IPv6 addresses */
4024 ret
= update_dev_clip(dev
, dev
);
4028 /* Parse all bond and vlan devices layered on top of the physical dev */
4029 for (i
= 0; i
< VLAN_N_VID
; i
++) {
4030 root_dev
= __vlan_find_dev_deep(dev
, htons(ETH_P_8021Q
), i
);
4034 ret
= update_dev_clip(root_dev
, dev
);
4041 static void update_clip(const struct adapter
*adap
)
4044 struct net_device
*dev
;
4049 for (i
= 0; i
< MAX_NPORTS
; i
++) {
4050 dev
= adap
->port
[i
];
4054 ret
= update_root_dev_clip(dev
);
4063 * cxgb_up - enable the adapter
4064 * @adap: adapter being enabled
4066 * Called when the first port is enabled, this function performs the
4067 * actions necessary to make an adapter operational, such as completing
4068 * the initialization of HW modules, and enabling interrupts.
4070 * Must be called with the rtnl lock held.
4072 static int cxgb_up(struct adapter
*adap
)
4076 err
= setup_sge_queues(adap
);
4079 err
= setup_rss(adap
);
4083 if (adap
->flags
& USING_MSIX
) {
4084 name_msix_vecs(adap
);
4085 err
= request_irq(adap
->msix_info
[0].vec
, t4_nondata_intr
, 0,
4086 adap
->msix_info
[0].desc
, adap
);
4090 err
= request_msix_queue_irqs(adap
);
4092 free_irq(adap
->msix_info
[0].vec
, adap
);
4096 err
= request_irq(adap
->pdev
->irq
, t4_intr_handler(adap
),
4097 (adap
->flags
& USING_MSI
) ? 0 : IRQF_SHARED
,
4098 adap
->port
[0]->name
, adap
);
4104 t4_intr_enable(adap
);
4105 adap
->flags
|= FULL_INIT_DONE
;
4106 notify_ulds(adap
, CXGB4_STATE_UP
);
4111 dev_err(adap
->pdev_dev
, "request_irq failed, err %d\n", err
);
4113 t4_free_sge_resources(adap
);
4117 static void cxgb_down(struct adapter
*adapter
)
4119 t4_intr_disable(adapter
);
4120 cancel_work_sync(&adapter
->tid_release_task
);
4121 cancel_work_sync(&adapter
->db_full_task
);
4122 cancel_work_sync(&adapter
->db_drop_task
);
4123 adapter
->tid_release_task_busy
= false;
4124 adapter
->tid_release_head
= NULL
;
4126 if (adapter
->flags
& USING_MSIX
) {
4127 free_msix_queue_irqs(adapter
);
4128 free_irq(adapter
->msix_info
[0].vec
, adapter
);
4130 free_irq(adapter
->pdev
->irq
, adapter
);
4131 quiesce_rx(adapter
);
4132 t4_sge_stop(adapter
);
4133 t4_free_sge_resources(adapter
);
4134 adapter
->flags
&= ~FULL_INIT_DONE
;
4138 * net_device operations
4140 static int cxgb_open(struct net_device
*dev
)
4143 struct port_info
*pi
= netdev_priv(dev
);
4144 struct adapter
*adapter
= pi
->adapter
;
4146 netif_carrier_off(dev
);
4148 if (!(adapter
->flags
& FULL_INIT_DONE
)) {
4149 err
= cxgb_up(adapter
);
4154 err
= link_start(dev
);
4156 netif_tx_start_all_queues(dev
);
4160 static int cxgb_close(struct net_device
*dev
)
4162 struct port_info
*pi
= netdev_priv(dev
);
4163 struct adapter
*adapter
= pi
->adapter
;
4165 netif_tx_stop_all_queues(dev
);
4166 netif_carrier_off(dev
);
4167 return t4_enable_vi(adapter
, adapter
->fn
, pi
->viid
, false, false);
4170 /* Return an error number if the indicated filter isn't writable ...
4172 static int writable_filter(struct filter_entry
*f
)
4182 /* Delete the filter at the specified index (if valid). The checks for all
4183 * the common problems with doing this like the filter being locked, currently
4184 * pending in another operation, etc.
4186 static int delete_filter(struct adapter
*adapter
, unsigned int fidx
)
4188 struct filter_entry
*f
;
4191 if (fidx
>= adapter
->tids
.nftids
+ adapter
->tids
.nsftids
)
4194 f
= &adapter
->tids
.ftid_tab
[fidx
];
4195 ret
= writable_filter(f
);
4199 return del_filter_wr(adapter
, fidx
);
4204 int cxgb4_create_server_filter(const struct net_device
*dev
, unsigned int stid
,
4205 __be32 sip
, __be16 sport
, __be16 vlan
,
4206 unsigned int queue
, unsigned char port
, unsigned char mask
)
4209 struct filter_entry
*f
;
4210 struct adapter
*adap
;
4214 adap
= netdev2adap(dev
);
4216 /* Adjust stid to correct filter index */
4217 stid
-= adap
->tids
.sftid_base
;
4218 stid
+= adap
->tids
.nftids
;
4220 /* Check to make sure the filter requested is writable ...
4222 f
= &adap
->tids
.ftid_tab
[stid
];
4223 ret
= writable_filter(f
);
4227 /* Clear out any old resources being used by the filter before
4228 * we start constructing the new filter.
4231 clear_filter(adap
, f
);
4233 /* Clear out filter specifications */
4234 memset(&f
->fs
, 0, sizeof(struct ch_filter_specification
));
4235 f
->fs
.val
.lport
= cpu_to_be16(sport
);
4236 f
->fs
.mask
.lport
= ~0;
4238 if ((val
[0] | val
[1] | val
[2] | val
[3]) != 0) {
4239 for (i
= 0; i
< 4; i
++) {
4240 f
->fs
.val
.lip
[i
] = val
[i
];
4241 f
->fs
.mask
.lip
[i
] = ~0;
4243 if (adap
->params
.tp
.vlan_pri_map
& F_PORT
) {
4244 f
->fs
.val
.iport
= port
;
4245 f
->fs
.mask
.iport
= mask
;
4249 if (adap
->params
.tp
.vlan_pri_map
& F_PROTOCOL
) {
4250 f
->fs
.val
.proto
= IPPROTO_TCP
;
4251 f
->fs
.mask
.proto
= ~0;
4256 /* Mark filter as locked */
4260 ret
= set_filter_wr(adap
, stid
);
4262 clear_filter(adap
, f
);
4268 EXPORT_SYMBOL(cxgb4_create_server_filter
);
4270 int cxgb4_remove_server_filter(const struct net_device
*dev
, unsigned int stid
,
4271 unsigned int queue
, bool ipv6
)
4274 struct filter_entry
*f
;
4275 struct adapter
*adap
;
4277 adap
= netdev2adap(dev
);
4279 /* Adjust stid to correct filter index */
4280 stid
-= adap
->tids
.sftid_base
;
4281 stid
+= adap
->tids
.nftids
;
4283 f
= &adap
->tids
.ftid_tab
[stid
];
4284 /* Unlock the filter */
4287 ret
= delete_filter(adap
, stid
);
4293 EXPORT_SYMBOL(cxgb4_remove_server_filter
);
4295 static struct rtnl_link_stats64
*cxgb_get_stats(struct net_device
*dev
,
4296 struct rtnl_link_stats64
*ns
)
4298 struct port_stats stats
;
4299 struct port_info
*p
= netdev_priv(dev
);
4300 struct adapter
*adapter
= p
->adapter
;
4302 /* Block retrieving statistics during EEH error
4303 * recovery. Otherwise, the recovery might fail
4304 * and the PCI device will be removed permanently
4306 spin_lock(&adapter
->stats_lock
);
4307 if (!netif_device_present(dev
)) {
4308 spin_unlock(&adapter
->stats_lock
);
4311 t4_get_port_stats(adapter
, p
->tx_chan
, &stats
);
4312 spin_unlock(&adapter
->stats_lock
);
4314 ns
->tx_bytes
= stats
.tx_octets
;
4315 ns
->tx_packets
= stats
.tx_frames
;
4316 ns
->rx_bytes
= stats
.rx_octets
;
4317 ns
->rx_packets
= stats
.rx_frames
;
4318 ns
->multicast
= stats
.rx_mcast_frames
;
4320 /* detailed rx_errors */
4321 ns
->rx_length_errors
= stats
.rx_jabber
+ stats
.rx_too_long
+
4323 ns
->rx_over_errors
= 0;
4324 ns
->rx_crc_errors
= stats
.rx_fcs_err
;
4325 ns
->rx_frame_errors
= stats
.rx_symbol_err
;
4326 ns
->rx_fifo_errors
= stats
.rx_ovflow0
+ stats
.rx_ovflow1
+
4327 stats
.rx_ovflow2
+ stats
.rx_ovflow3
+
4328 stats
.rx_trunc0
+ stats
.rx_trunc1
+
4329 stats
.rx_trunc2
+ stats
.rx_trunc3
;
4330 ns
->rx_missed_errors
= 0;
4332 /* detailed tx_errors */
4333 ns
->tx_aborted_errors
= 0;
4334 ns
->tx_carrier_errors
= 0;
4335 ns
->tx_fifo_errors
= 0;
4336 ns
->tx_heartbeat_errors
= 0;
4337 ns
->tx_window_errors
= 0;
4339 ns
->tx_errors
= stats
.tx_error_frames
;
4340 ns
->rx_errors
= stats
.rx_symbol_err
+ stats
.rx_fcs_err
+
4341 ns
->rx_length_errors
+ stats
.rx_len_err
+ ns
->rx_fifo_errors
;
4345 static int cxgb_ioctl(struct net_device
*dev
, struct ifreq
*req
, int cmd
)
4348 int ret
= 0, prtad
, devad
;
4349 struct port_info
*pi
= netdev_priv(dev
);
4350 struct mii_ioctl_data
*data
= (struct mii_ioctl_data
*)&req
->ifr_data
;
4354 if (pi
->mdio_addr
< 0)
4356 data
->phy_id
= pi
->mdio_addr
;
4360 if (mdio_phy_id_is_c45(data
->phy_id
)) {
4361 prtad
= mdio_phy_id_prtad(data
->phy_id
);
4362 devad
= mdio_phy_id_devad(data
->phy_id
);
4363 } else if (data
->phy_id
< 32) {
4364 prtad
= data
->phy_id
;
4366 data
->reg_num
&= 0x1f;
4370 mbox
= pi
->adapter
->fn
;
4371 if (cmd
== SIOCGMIIREG
)
4372 ret
= t4_mdio_rd(pi
->adapter
, mbox
, prtad
, devad
,
4373 data
->reg_num
, &data
->val_out
);
4375 ret
= t4_mdio_wr(pi
->adapter
, mbox
, prtad
, devad
,
4376 data
->reg_num
, data
->val_in
);
4384 static void cxgb_set_rxmode(struct net_device
*dev
)
4386 /* unfortunately we can't return errors to the stack */
4387 set_rxmode(dev
, -1, false);
4390 static int cxgb_change_mtu(struct net_device
*dev
, int new_mtu
)
4393 struct port_info
*pi
= netdev_priv(dev
);
4395 if (new_mtu
< 81 || new_mtu
> MAX_MTU
) /* accommodate SACK */
4397 ret
= t4_set_rxmode(pi
->adapter
, pi
->adapter
->fn
, pi
->viid
, new_mtu
, -1,
4404 static int cxgb_set_mac_addr(struct net_device
*dev
, void *p
)
4407 struct sockaddr
*addr
= p
;
4408 struct port_info
*pi
= netdev_priv(dev
);
4410 if (!is_valid_ether_addr(addr
->sa_data
))
4411 return -EADDRNOTAVAIL
;
4413 ret
= t4_change_mac(pi
->adapter
, pi
->adapter
->fn
, pi
->viid
,
4414 pi
->xact_addr_filt
, addr
->sa_data
, true, true);
4418 memcpy(dev
->dev_addr
, addr
->sa_data
, dev
->addr_len
);
4419 pi
->xact_addr_filt
= ret
;
4423 #ifdef CONFIG_NET_POLL_CONTROLLER
4424 static void cxgb_netpoll(struct net_device
*dev
)
4426 struct port_info
*pi
= netdev_priv(dev
);
4427 struct adapter
*adap
= pi
->adapter
;
4429 if (adap
->flags
& USING_MSIX
) {
4431 struct sge_eth_rxq
*rx
= &adap
->sge
.ethrxq
[pi
->first_qset
];
4433 for (i
= pi
->nqsets
; i
; i
--, rx
++)
4434 t4_sge_intr_msix(0, &rx
->rspq
);
4436 t4_intr_handler(adap
)(0, adap
);
4440 static const struct net_device_ops cxgb4_netdev_ops
= {
4441 .ndo_open
= cxgb_open
,
4442 .ndo_stop
= cxgb_close
,
4443 .ndo_start_xmit
= t4_eth_xmit
,
4444 .ndo_get_stats64
= cxgb_get_stats
,
4445 .ndo_set_rx_mode
= cxgb_set_rxmode
,
4446 .ndo_set_mac_address
= cxgb_set_mac_addr
,
4447 .ndo_set_features
= cxgb_set_features
,
4448 .ndo_validate_addr
= eth_validate_addr
,
4449 .ndo_do_ioctl
= cxgb_ioctl
,
4450 .ndo_change_mtu
= cxgb_change_mtu
,
4451 #ifdef CONFIG_NET_POLL_CONTROLLER
4452 .ndo_poll_controller
= cxgb_netpoll
,
4456 void t4_fatal_err(struct adapter
*adap
)
4458 t4_set_reg_field(adap
, SGE_CONTROL
, GLOBALENABLE
, 0);
4459 t4_intr_disable(adap
);
4460 dev_alert(adap
->pdev_dev
, "encountered fatal error, adapter stopped\n");
4463 static void setup_memwin(struct adapter
*adap
)
4465 u32 bar0
, mem_win0_base
, mem_win1_base
, mem_win2_base
;
4467 bar0
= pci_resource_start(adap
->pdev
, 0); /* truncation intentional */
4468 if (is_t4(adap
->params
.chip
)) {
4469 mem_win0_base
= bar0
+ MEMWIN0_BASE
;
4470 mem_win1_base
= bar0
+ MEMWIN1_BASE
;
4471 mem_win2_base
= bar0
+ MEMWIN2_BASE
;
4473 /* For T5, only relative offset inside the PCIe BAR is passed */
4474 mem_win0_base
= MEMWIN0_BASE
;
4475 mem_win1_base
= MEMWIN1_BASE_T5
;
4476 mem_win2_base
= MEMWIN2_BASE_T5
;
4478 t4_write_reg(adap
, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN
, 0),
4479 mem_win0_base
| BIR(0) |
4480 WINDOW(ilog2(MEMWIN0_APERTURE
) - 10));
4481 t4_write_reg(adap
, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN
, 1),
4482 mem_win1_base
| BIR(0) |
4483 WINDOW(ilog2(MEMWIN1_APERTURE
) - 10));
4484 t4_write_reg(adap
, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN
, 2),
4485 mem_win2_base
| BIR(0) |
4486 WINDOW(ilog2(MEMWIN2_APERTURE
) - 10));
4489 static void setup_memwin_rdma(struct adapter
*adap
)
4491 if (adap
->vres
.ocq
.size
) {
4492 unsigned int start
, sz_kb
;
4494 start
= pci_resource_start(adap
->pdev
, 2) +
4495 OCQ_WIN_OFFSET(adap
->pdev
, &adap
->vres
);
4496 sz_kb
= roundup_pow_of_two(adap
->vres
.ocq
.size
) >> 10;
4498 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN
, 3),
4499 start
| BIR(1) | WINDOW(ilog2(sz_kb
)));
4501 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET
, 3),
4502 adap
->vres
.ocq
.start
);
4504 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET
, 3));
4508 static int adap_init1(struct adapter
*adap
, struct fw_caps_config_cmd
*c
)
4513 /* get device capabilities */
4514 memset(c
, 0, sizeof(*c
));
4515 c
->op_to_write
= htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
4516 FW_CMD_REQUEST
| FW_CMD_READ
);
4517 c
->cfvalid_to_len16
= htonl(FW_LEN16(*c
));
4518 ret
= t4_wr_mbox(adap
, adap
->fn
, c
, sizeof(*c
), c
);
4522 /* select capabilities we'll be using */
4523 if (c
->niccaps
& htons(FW_CAPS_CONFIG_NIC_VM
)) {
4525 c
->niccaps
^= htons(FW_CAPS_CONFIG_NIC_VM
);
4527 c
->niccaps
= htons(FW_CAPS_CONFIG_NIC_VM
);
4528 } else if (vf_acls
) {
4529 dev_err(adap
->pdev_dev
, "virtualization ACLs not supported");
4532 c
->op_to_write
= htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
4533 FW_CMD_REQUEST
| FW_CMD_WRITE
);
4534 ret
= t4_wr_mbox(adap
, adap
->fn
, c
, sizeof(*c
), NULL
);
4538 ret
= t4_config_glbl_rss(adap
, adap
->fn
,
4539 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL
,
4540 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN
|
4541 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP
);
4545 ret
= t4_cfg_pfvf(adap
, adap
->fn
, adap
->fn
, 0, MAX_EGRQ
, 64, MAX_INGQ
,
4546 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF
, FW_CMD_CAP_PF
);
4552 /* tweak some settings */
4553 t4_write_reg(adap
, TP_SHIFT_CNT
, 0x64f8849);
4554 t4_write_reg(adap
, ULP_RX_TDDP_PSZ
, HPZ0(PAGE_SHIFT
- 12));
4555 t4_write_reg(adap
, TP_PIO_ADDR
, TP_INGRESS_CONFIG
);
4556 v
= t4_read_reg(adap
, TP_PIO_DATA
);
4557 t4_write_reg(adap
, TP_PIO_DATA
, v
& ~CSUM_HAS_PSEUDO_HDR
);
4559 /* first 4 Tx modulation queues point to consecutive Tx channels */
4560 adap
->params
.tp
.tx_modq_map
= 0xE4;
4561 t4_write_reg(adap
, A_TP_TX_MOD_QUEUE_REQ_MAP
,
4562 V_TX_MOD_QUEUE_REQ_MAP(adap
->params
.tp
.tx_modq_map
));
4564 /* associate each Tx modulation queue with consecutive Tx channels */
4566 t4_write_indirect(adap
, TP_PIO_ADDR
, TP_PIO_DATA
,
4567 &v
, 1, A_TP_TX_SCHED_HDR
);
4568 t4_write_indirect(adap
, TP_PIO_ADDR
, TP_PIO_DATA
,
4569 &v
, 1, A_TP_TX_SCHED_FIFO
);
4570 t4_write_indirect(adap
, TP_PIO_ADDR
, TP_PIO_DATA
,
4571 &v
, 1, A_TP_TX_SCHED_PCMD
);
4573 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4574 if (is_offload(adap
)) {
4575 t4_write_reg(adap
, A_TP_TX_MOD_QUEUE_WEIGHT0
,
4576 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT
) |
4577 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT
) |
4578 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT
) |
4579 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT
));
4580 t4_write_reg(adap
, A_TP_TX_MOD_CHANNEL_WEIGHT
,
4581 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT
) |
4582 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT
) |
4583 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT
) |
4584 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT
));
4587 /* get basic stuff going */
4588 return t4_early_init(adap
, adap
->fn
);
4592 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
4594 #define MAX_ATIDS 8192U
4597 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4599 * If the firmware we're dealing with has Configuration File support, then
4600 * we use that to perform all configuration
4604 * Tweak configuration based on module parameters, etc. Most of these have
4605 * defaults assigned to them by Firmware Configuration Files (if we're using
4606 * them) but need to be explicitly set if we're using hard-coded
4607 * initialization. But even in the case of using Firmware Configuration
4608 * Files, we'd like to expose the ability to change these via module
4609 * parameters so these are essentially common tweaks/settings for
4610 * Configuration Files and hard-coded initialization ...
4612 static int adap_init0_tweaks(struct adapter
*adapter
)
4615 * Fix up various Host-Dependent Parameters like Page Size, Cache
4616 * Line Size, etc. The firmware default is for a 4KB Page Size and
4617 * 64B Cache Line Size ...
4619 t4_fixup_host_params(adapter
, PAGE_SIZE
, L1_CACHE_BYTES
);
4622 * Process module parameters which affect early initialization.
4624 if (rx_dma_offset
!= 2 && rx_dma_offset
!= 0) {
4625 dev_err(&adapter
->pdev
->dev
,
4626 "Ignoring illegal rx_dma_offset=%d, using 2\n",
4630 t4_set_reg_field(adapter
, SGE_CONTROL
,
4632 PKTSHIFT(rx_dma_offset
));
4635 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4636 * adds the pseudo header itself.
4638 t4_tp_wr_bits_indirect(adapter
, TP_INGRESS_CONFIG
,
4639 CSUM_HAS_PSEUDO_HDR
, 0);
4645 * Attempt to initialize the adapter via a Firmware Configuration File.
4647 static int adap_init0_config(struct adapter
*adapter
, int reset
)
4649 struct fw_caps_config_cmd caps_cmd
;
4650 const struct firmware
*cf
;
4651 unsigned long mtype
= 0, maddr
= 0;
4652 u32 finiver
, finicsum
, cfcsum
;
4654 int config_issued
= 0;
4655 char *fw_config_file
, fw_config_file_path
[256];
4656 char *config_name
= NULL
;
4659 * Reset device if necessary.
4662 ret
= t4_fw_reset(adapter
, adapter
->mbox
,
4663 PIORSTMODE
| PIORST
);
4669 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4670 * then use that. Otherwise, use the configuration file stored
4671 * in the adapter flash ...
4673 switch (CHELSIO_CHIP_VERSION(adapter
->params
.chip
)) {
4675 fw_config_file
= FW4_CFNAME
;
4678 fw_config_file
= FW5_CFNAME
;
4681 dev_err(adapter
->pdev_dev
, "Device %d is not supported\n",
4682 adapter
->pdev
->device
);
4687 ret
= request_firmware(&cf
, fw_config_file
, adapter
->pdev_dev
);
4689 config_name
= "On FLASH";
4690 mtype
= FW_MEMTYPE_CF_FLASH
;
4691 maddr
= t4_flash_cfg_addr(adapter
);
4693 u32 params
[7], val
[7];
4695 sprintf(fw_config_file_path
,
4696 "/lib/firmware/%s", fw_config_file
);
4697 config_name
= fw_config_file_path
;
4699 if (cf
->size
>= FLASH_CFG_MAX_SIZE
)
4702 params
[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV
) |
4703 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF
));
4704 ret
= t4_query_params(adapter
, adapter
->mbox
,
4705 adapter
->fn
, 0, 1, params
, val
);
4708 * For t4_memory_write() below addresses and
4709 * sizes have to be in terms of multiples of 4
4710 * bytes. So, if the Configuration File isn't
4711 * a multiple of 4 bytes in length we'll have
4712 * to write that out separately since we can't
4713 * guarantee that the bytes following the
4714 * residual byte in the buffer returned by
4715 * request_firmware() are zeroed out ...
4717 size_t resid
= cf
->size
& 0x3;
4718 size_t size
= cf
->size
& ~0x3;
4719 __be32
*data
= (__be32
*)cf
->data
;
4721 mtype
= FW_PARAMS_PARAM_Y_GET(val
[0]);
4722 maddr
= FW_PARAMS_PARAM_Z_GET(val
[0]) << 16;
4724 ret
= t4_memory_write(adapter
, mtype
, maddr
,
4726 if (ret
== 0 && resid
!= 0) {
4733 last
.word
= data
[size
>> 2];
4734 for (i
= resid
; i
< 4; i
++)
4736 ret
= t4_memory_write(adapter
, mtype
,
4743 release_firmware(cf
);
4749 * Issue a Capability Configuration command to the firmware to get it
4750 * to parse the Configuration File. We don't use t4_fw_config_file()
4751 * because we want the ability to modify various features after we've
4752 * processed the configuration file ...
4754 memset(&caps_cmd
, 0, sizeof(caps_cmd
));
4755 caps_cmd
.op_to_write
=
4756 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
4759 caps_cmd
.cfvalid_to_len16
=
4760 htonl(FW_CAPS_CONFIG_CMD_CFVALID
|
4761 FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype
) |
4762 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr
>> 16) |
4763 FW_LEN16(caps_cmd
));
4764 ret
= t4_wr_mbox(adapter
, adapter
->mbox
, &caps_cmd
, sizeof(caps_cmd
),
4767 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
4768 * Configuration File in FLASH), our last gasp effort is to use the
4769 * Firmware Configuration File which is embedded in the firmware. A
4770 * very few early versions of the firmware didn't have one embedded
4771 * but we can ignore those.
4773 if (ret
== -ENOENT
) {
4774 memset(&caps_cmd
, 0, sizeof(caps_cmd
));
4775 caps_cmd
.op_to_write
=
4776 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
4779 caps_cmd
.cfvalid_to_len16
= htonl(FW_LEN16(caps_cmd
));
4780 ret
= t4_wr_mbox(adapter
, adapter
->mbox
, &caps_cmd
,
4781 sizeof(caps_cmd
), &caps_cmd
);
4782 config_name
= "Firmware Default";
4789 finiver
= ntohl(caps_cmd
.finiver
);
4790 finicsum
= ntohl(caps_cmd
.finicsum
);
4791 cfcsum
= ntohl(caps_cmd
.cfcsum
);
4792 if (finicsum
!= cfcsum
)
4793 dev_warn(adapter
->pdev_dev
, "Configuration File checksum "\
4794 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4798 * And now tell the firmware to use the configuration we just loaded.
4800 caps_cmd
.op_to_write
=
4801 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
4804 caps_cmd
.cfvalid_to_len16
= htonl(FW_LEN16(caps_cmd
));
4805 ret
= t4_wr_mbox(adapter
, adapter
->mbox
, &caps_cmd
, sizeof(caps_cmd
),
4811 * Tweak configuration based on system architecture, module
4814 ret
= adap_init0_tweaks(adapter
);
4819 * And finally tell the firmware to initialize itself using the
4820 * parameters from the Configuration File.
4822 ret
= t4_fw_initialize(adapter
, adapter
->mbox
);
4827 * Return successfully and note that we're operating with parameters
4828 * not supplied by the driver, rather than from hard-wired
4829 * initialization constants burried in the driver.
4831 adapter
->flags
|= USING_SOFT_PARAMS
;
4832 dev_info(adapter
->pdev_dev
, "Successfully configured using Firmware "\
4833 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
4834 config_name
, finiver
, cfcsum
);
4838 * Something bad happened. Return the error ... (If the "error"
4839 * is that there's no Configuration File on the adapter we don't
4840 * want to issue a warning since this is fairly common.)
4843 if (config_issued
&& ret
!= -ENOENT
)
4844 dev_warn(adapter
->pdev_dev
, "\"%s\" configuration file error %d\n",
4850 * Attempt to initialize the adapter via hard-coded, driver supplied
4853 static int adap_init0_no_config(struct adapter
*adapter
, int reset
)
4855 struct sge
*s
= &adapter
->sge
;
4856 struct fw_caps_config_cmd caps_cmd
;
4861 * Reset device if necessary
4864 ret
= t4_fw_reset(adapter
, adapter
->mbox
,
4865 PIORSTMODE
| PIORST
);
4871 * Get device capabilities and select which we'll be using.
4873 memset(&caps_cmd
, 0, sizeof(caps_cmd
));
4874 caps_cmd
.op_to_write
= htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
4875 FW_CMD_REQUEST
| FW_CMD_READ
);
4876 caps_cmd
.cfvalid_to_len16
= htonl(FW_LEN16(caps_cmd
));
4877 ret
= t4_wr_mbox(adapter
, adapter
->mbox
, &caps_cmd
, sizeof(caps_cmd
),
4882 if (caps_cmd
.niccaps
& htons(FW_CAPS_CONFIG_NIC_VM
)) {
4884 caps_cmd
.niccaps
^= htons(FW_CAPS_CONFIG_NIC_VM
);
4886 caps_cmd
.niccaps
= htons(FW_CAPS_CONFIG_NIC_VM
);
4887 } else if (vf_acls
) {
4888 dev_err(adapter
->pdev_dev
, "virtualization ACLs not supported");
4891 caps_cmd
.op_to_write
= htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
4892 FW_CMD_REQUEST
| FW_CMD_WRITE
);
4893 ret
= t4_wr_mbox(adapter
, adapter
->mbox
, &caps_cmd
, sizeof(caps_cmd
),
4899 * Tweak configuration based on system architecture, module
4902 ret
= adap_init0_tweaks(adapter
);
4907 * Select RSS Global Mode we want to use. We use "Basic Virtual"
4908 * mode which maps each Virtual Interface to its own section of
4909 * the RSS Table and we turn on all map and hash enables ...
4911 adapter
->flags
|= RSS_TNLALLLOOKUP
;
4912 ret
= t4_config_glbl_rss(adapter
, adapter
->mbox
,
4913 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL
,
4914 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN
|
4915 FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ
|
4916 ((adapter
->flags
& RSS_TNLALLLOOKUP
) ?
4917 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP
: 0));
4922 * Set up our own fundamental resource provisioning ...
4924 ret
= t4_cfg_pfvf(adapter
, adapter
->mbox
, adapter
->fn
, 0,
4925 PFRES_NEQ
, PFRES_NETHCTRL
,
4926 PFRES_NIQFLINT
, PFRES_NIQ
,
4927 PFRES_TC
, PFRES_NVI
,
4928 FW_PFVF_CMD_CMASK_MASK
,
4929 pfvfres_pmask(adapter
, adapter
->fn
, 0),
4931 PFRES_R_CAPS
, PFRES_WX_CAPS
);
4936 * Perform low level SGE initialization. We need to do this before we
4937 * send the firmware the INITIALIZE command because that will cause
4938 * any other PF Drivers which are waiting for the Master
4939 * Initialization to proceed forward.
4941 for (i
= 0; i
< SGE_NTIMERS
- 1; i
++)
4942 s
->timer_val
[i
] = min(intr_holdoff
[i
], MAX_SGE_TIMERVAL
);
4943 s
->timer_val
[SGE_NTIMERS
- 1] = MAX_SGE_TIMERVAL
;
4944 s
->counter_val
[0] = 1;
4945 for (i
= 1; i
< SGE_NCOUNTERS
; i
++)
4946 s
->counter_val
[i
] = min(intr_cnt
[i
- 1],
4947 THRESHOLD_0_GET(THRESHOLD_0_MASK
));
4948 t4_sge_init(adapter
);
4950 #ifdef CONFIG_PCI_IOV
4952 * Provision resource limits for Virtual Functions. We currently
4953 * grant them all the same static resource limits except for the Port
4954 * Access Rights Mask which we're assigning based on the PF. All of
4955 * the static provisioning stuff for both the PF and VF really needs
4956 * to be managed in a persistent manner for each device which the
4957 * firmware controls.
4962 for (pf
= 0; pf
< ARRAY_SIZE(num_vf
); pf
++) {
4963 if (num_vf
[pf
] <= 0)
4966 /* VF numbering starts at 1! */
4967 for (vf
= 1; vf
<= num_vf
[pf
]; vf
++) {
4968 ret
= t4_cfg_pfvf(adapter
, adapter
->mbox
,
4970 VFRES_NEQ
, VFRES_NETHCTRL
,
4971 VFRES_NIQFLINT
, VFRES_NIQ
,
4972 VFRES_TC
, VFRES_NVI
,
4973 FW_PFVF_CMD_CMASK_MASK
,
4977 VFRES_R_CAPS
, VFRES_WX_CAPS
);
4979 dev_warn(adapter
->pdev_dev
,
4981 "provision pf/vf=%d/%d; "
4982 "err=%d\n", pf
, vf
, ret
);
4989 * Set up the default filter mode. Later we'll want to implement this
4990 * via a firmware command, etc. ... This needs to be done before the
4991 * firmare initialization command ... If the selected set of fields
4992 * isn't equal to the default value, we'll need to make sure that the
4993 * field selections will fit in the 36-bit budget.
4995 if (tp_vlan_pri_map
!= TP_VLAN_PRI_MAP_DEFAULT
) {
4998 for (j
= TP_VLAN_PRI_MAP_FIRST
; j
<= TP_VLAN_PRI_MAP_LAST
; j
++)
4999 switch (tp_vlan_pri_map
& (1 << j
)) {
5001 /* compressed filter field not enabled */
5021 case ETHERTYPE_MASK
:
5027 case MPSHITTYPE_MASK
:
5030 case FRAGMENTATION_MASK
:
5036 dev_err(adapter
->pdev_dev
,
5037 "tp_vlan_pri_map=%#x needs %d bits > 36;"\
5038 " using %#x\n", tp_vlan_pri_map
, bits
,
5039 TP_VLAN_PRI_MAP_DEFAULT
);
5040 tp_vlan_pri_map
= TP_VLAN_PRI_MAP_DEFAULT
;
5043 v
= tp_vlan_pri_map
;
5044 t4_write_indirect(adapter
, TP_PIO_ADDR
, TP_PIO_DATA
,
5045 &v
, 1, TP_VLAN_PRI_MAP
);
5048 * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
5049 * to support any of the compressed filter fields above. Newer
5050 * versions of the firmware do this automatically but it doesn't hurt
5051 * to set it here. Meanwhile, we do _not_ need to set Lookup Every
5052 * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
5053 * since the firmware automatically turns this on and off when we have
5054 * a non-zero number of filters active (since it does have a
5055 * performance impact).
5057 if (tp_vlan_pri_map
)
5058 t4_set_reg_field(adapter
, TP_GLOBAL_CONFIG
,
5059 FIVETUPLELOOKUP_MASK
,
5060 FIVETUPLELOOKUP_MASK
);
5063 * Tweak some settings.
5065 t4_write_reg(adapter
, TP_SHIFT_CNT
, SYNSHIFTMAX(6) |
5066 RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
5067 PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
5068 KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
5071 * Get basic stuff going by issuing the Firmware Initialize command.
5072 * Note that this _must_ be after all PFVF commands ...
5074 ret
= t4_fw_initialize(adapter
, adapter
->mbox
);
5079 * Return successfully!
5081 dev_info(adapter
->pdev_dev
, "Successfully configured using built-in "\
5082 "driver parameters\n");
5086 * Something bad happened. Return the error ...
5092 static struct fw_info fw_info_array
[] = {
5095 .fs_name
= FW4_CFNAME
,
5096 .fw_mod_name
= FW4_FNAME
,
5098 .chip
= FW_HDR_CHIP_T4
,
5099 .fw_ver
= __cpu_to_be32(FW_VERSION(T4
)),
5100 .intfver_nic
= FW_INTFVER(T4
, NIC
),
5101 .intfver_vnic
= FW_INTFVER(T4
, VNIC
),
5102 .intfver_ri
= FW_INTFVER(T4
, RI
),
5103 .intfver_iscsi
= FW_INTFVER(T4
, ISCSI
),
5104 .intfver_fcoe
= FW_INTFVER(T4
, FCOE
),
5108 .fs_name
= FW5_CFNAME
,
5109 .fw_mod_name
= FW5_FNAME
,
5111 .chip
= FW_HDR_CHIP_T5
,
5112 .fw_ver
= __cpu_to_be32(FW_VERSION(T5
)),
5113 .intfver_nic
= FW_INTFVER(T5
, NIC
),
5114 .intfver_vnic
= FW_INTFVER(T5
, VNIC
),
5115 .intfver_ri
= FW_INTFVER(T5
, RI
),
5116 .intfver_iscsi
= FW_INTFVER(T5
, ISCSI
),
5117 .intfver_fcoe
= FW_INTFVER(T5
, FCOE
),
5122 static struct fw_info
*find_fw_info(int chip
)
5126 for (i
= 0; i
< ARRAY_SIZE(fw_info_array
); i
++) {
5127 if (fw_info_array
[i
].chip
== chip
)
5128 return &fw_info_array
[i
];
5134 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
5136 static int adap_init0(struct adapter
*adap
)
5140 enum dev_state state
;
5141 u32 params
[7], val
[7];
5142 struct fw_caps_config_cmd caps_cmd
;
5146 * Contact FW, advertising Master capability (and potentially forcing
5147 * ourselves as the Master PF if our module parameter force_init is
5150 ret
= t4_fw_hello(adap
, adap
->mbox
, adap
->fn
,
5151 force_init
? MASTER_MUST
: MASTER_MAY
,
5154 dev_err(adap
->pdev_dev
, "could not connect to FW, error %d\n",
5158 if (ret
== adap
->mbox
)
5159 adap
->flags
|= MASTER_PF
;
5160 if (force_init
&& state
== DEV_STATE_INIT
)
5161 state
= DEV_STATE_UNINIT
;
5164 * If we're the Master PF Driver and the device is uninitialized,
5165 * then let's consider upgrading the firmware ... (We always want
5166 * to check the firmware version number in order to A. get it for
5167 * later reporting and B. to warn if the currently loaded firmware
5168 * is excessively mismatched relative to the driver.)
5170 t4_get_fw_version(adap
, &adap
->params
.fw_vers
);
5171 t4_get_tp_version(adap
, &adap
->params
.tp_vers
);
5172 if ((adap
->flags
& MASTER_PF
) && state
!= DEV_STATE_INIT
) {
5173 struct fw_info
*fw_info
;
5174 struct fw_hdr
*card_fw
;
5175 const struct firmware
*fw
;
5176 const u8
*fw_data
= NULL
;
5177 unsigned int fw_size
= 0;
5179 /* This is the firmware whose headers the driver was compiled
5182 fw_info
= find_fw_info(CHELSIO_CHIP_VERSION(adap
->params
.chip
));
5183 if (fw_info
== NULL
) {
5184 dev_err(adap
->pdev_dev
,
5185 "unable to get firmware info for chip %d.\n",
5186 CHELSIO_CHIP_VERSION(adap
->params
.chip
));
5190 /* allocate memory to read the header of the firmware on the
5193 card_fw
= t4_alloc_mem(sizeof(*card_fw
));
5195 /* Get FW from from /lib/firmware/ */
5196 ret
= request_firmware(&fw
, fw_info
->fw_mod_name
,
5199 dev_err(adap
->pdev_dev
,
5200 "unable to load firmware image %s, error %d\n",
5201 fw_info
->fw_mod_name
, ret
);
5207 /* upgrade FW logic */
5208 ret
= t4_prep_fw(adap
, fw_info
, fw_data
, fw_size
, card_fw
,
5213 release_firmware(fw
);
5214 t4_free_mem(card_fw
);
5221 * Grab VPD parameters. This should be done after we establish a
5222 * connection to the firmware since some of the VPD parameters
5223 * (notably the Core Clock frequency) are retrieved via requests to
5224 * the firmware. On the other hand, we need these fairly early on
5225 * so we do this right after getting ahold of the firmware.
5227 ret
= get_vpd_params(adap
, &adap
->params
.vpd
);
5232 * Find out what ports are available to us. Note that we need to do
5233 * this before calling adap_init0_no_config() since it needs nports
5237 FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV
) |
5238 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC
);
5239 ret
= t4_query_params(adap
, adap
->mbox
, adap
->fn
, 0, 1, &v
, &port_vec
);
5243 adap
->params
.nports
= hweight32(port_vec
);
5244 adap
->params
.portvec
= port_vec
;
5247 * If the firmware is initialized already (and we're not forcing a
5248 * master initialization), note that we're living with existing
5249 * adapter parameters. Otherwise, it's time to try initializing the
5252 if (state
== DEV_STATE_INIT
) {
5253 dev_info(adap
->pdev_dev
, "Coming up as %s: "\
5254 "Adapter already initialized\n",
5255 adap
->flags
& MASTER_PF
? "MASTER" : "SLAVE");
5256 adap
->flags
|= USING_SOFT_PARAMS
;
5258 dev_info(adap
->pdev_dev
, "Coming up as MASTER: "\
5259 "Initializing adapter\n");
5262 * If the firmware doesn't support Configuration
5263 * Files warn user and exit,
5266 dev_warn(adap
->pdev_dev
, "Firmware doesn't support "
5267 "configuration file.\n");
5269 ret
= adap_init0_no_config(adap
, reset
);
5272 * Find out whether we're dealing with a version of
5273 * the firmware which has configuration file support.
5275 params
[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV
) |
5276 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF
));
5277 ret
= t4_query_params(adap
, adap
->mbox
, adap
->fn
, 0, 1,
5281 * If the firmware doesn't support Configuration
5282 * Files, use the old Driver-based, hard-wired
5283 * initialization. Otherwise, try using the
5284 * Configuration File support and fall back to the
5285 * Driver-based initialization if there's no
5286 * Configuration File found.
5289 ret
= adap_init0_no_config(adap
, reset
);
5292 * The firmware provides us with a memory
5293 * buffer where we can load a Configuration
5294 * File from the host if we want to override
5295 * the Configuration File in flash.
5298 ret
= adap_init0_config(adap
, reset
);
5299 if (ret
== -ENOENT
) {
5300 dev_info(adap
->pdev_dev
,
5301 "No Configuration File present "
5302 "on adapter. Using hard-wired "
5303 "configuration parameters.\n");
5304 ret
= adap_init0_no_config(adap
, reset
);
5309 dev_err(adap
->pdev_dev
,
5310 "could not initialize adapter, error %d\n",
5317 * If we're living with non-hard-coded parameters (either from a
5318 * Firmware Configuration File or values programmed by a different PF
5319 * Driver), give the SGE code a chance to pull in anything that it
5320 * needs ... Note that this must be called after we retrieve our VPD
5321 * parameters in order to know how to convert core ticks to seconds.
5323 if (adap
->flags
& USING_SOFT_PARAMS
) {
5324 ret
= t4_sge_init(adap
);
5329 if (is_bypass_device(adap
->pdev
->device
))
5330 adap
->params
.bypass
= 1;
5333 * Grab some of our basic fundamental operating parameters.
5335 #define FW_PARAM_DEV(param) \
5336 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
5337 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
5339 #define FW_PARAM_PFVF(param) \
5340 FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
5341 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)| \
5342 FW_PARAMS_PARAM_Y(0) | \
5343 FW_PARAMS_PARAM_Z(0)
5345 params
[0] = FW_PARAM_PFVF(EQ_START
);
5346 params
[1] = FW_PARAM_PFVF(L2T_START
);
5347 params
[2] = FW_PARAM_PFVF(L2T_END
);
5348 params
[3] = FW_PARAM_PFVF(FILTER_START
);
5349 params
[4] = FW_PARAM_PFVF(FILTER_END
);
5350 params
[5] = FW_PARAM_PFVF(IQFLINT_START
);
5351 ret
= t4_query_params(adap
, adap
->mbox
, adap
->fn
, 0, 6, params
, val
);
5354 adap
->sge
.egr_start
= val
[0];
5355 adap
->l2t_start
= val
[1];
5356 adap
->l2t_end
= val
[2];
5357 adap
->tids
.ftid_base
= val
[3];
5358 adap
->tids
.nftids
= val
[4] - val
[3] + 1;
5359 adap
->sge
.ingr_start
= val
[5];
5361 /* query params related to active filter region */
5362 params
[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START
);
5363 params
[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END
);
5364 ret
= t4_query_params(adap
, adap
->mbox
, adap
->fn
, 0, 2, params
, val
);
5365 /* If Active filter size is set we enable establishing
5366 * offload connection through firmware work request
5368 if ((val
[0] != val
[1]) && (ret
>= 0)) {
5369 adap
->flags
|= FW_OFLD_CONN
;
5370 adap
->tids
.aftid_base
= val
[0];
5371 adap
->tids
.aftid_end
= val
[1];
5374 /* If we're running on newer firmware, let it know that we're
5375 * prepared to deal with encapsulated CPL messages. Older
5376 * firmware won't understand this and we'll just get
5377 * unencapsulated messages ...
5379 params
[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP
);
5381 (void) t4_set_params(adap
, adap
->mbox
, adap
->fn
, 0, 1, params
, val
);
5384 * Get device capabilities so we can determine what resources we need
5387 memset(&caps_cmd
, 0, sizeof(caps_cmd
));
5388 caps_cmd
.op_to_write
= htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD
) |
5389 FW_CMD_REQUEST
| FW_CMD_READ
);
5390 caps_cmd
.cfvalid_to_len16
= htonl(FW_LEN16(caps_cmd
));
5391 ret
= t4_wr_mbox(adap
, adap
->mbox
, &caps_cmd
, sizeof(caps_cmd
),
5396 if (caps_cmd
.ofldcaps
) {
5397 /* query offload-related parameters */
5398 params
[0] = FW_PARAM_DEV(NTID
);
5399 params
[1] = FW_PARAM_PFVF(SERVER_START
);
5400 params
[2] = FW_PARAM_PFVF(SERVER_END
);
5401 params
[3] = FW_PARAM_PFVF(TDDP_START
);
5402 params
[4] = FW_PARAM_PFVF(TDDP_END
);
5403 params
[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ
);
5404 ret
= t4_query_params(adap
, adap
->mbox
, adap
->fn
, 0, 6,
5408 adap
->tids
.ntids
= val
[0];
5409 adap
->tids
.natids
= min(adap
->tids
.ntids
/ 2, MAX_ATIDS
);
5410 adap
->tids
.stid_base
= val
[1];
5411 adap
->tids
.nstids
= val
[2] - val
[1] + 1;
5413 * Setup server filter region. Divide the availble filter
5414 * region into two parts. Regular filters get 1/3rd and server
5415 * filters get 2/3rd part. This is only enabled if workarond
5417 * 1. For regular filters.
5418 * 2. Server filter: This are special filters which are used
5419 * to redirect SYN packets to offload queue.
5421 if (adap
->flags
& FW_OFLD_CONN
&& !is_bypass(adap
)) {
5422 adap
->tids
.sftid_base
= adap
->tids
.ftid_base
+
5423 DIV_ROUND_UP(adap
->tids
.nftids
, 3);
5424 adap
->tids
.nsftids
= adap
->tids
.nftids
-
5425 DIV_ROUND_UP(adap
->tids
.nftids
, 3);
5426 adap
->tids
.nftids
= adap
->tids
.sftid_base
-
5427 adap
->tids
.ftid_base
;
5429 adap
->vres
.ddp
.start
= val
[3];
5430 adap
->vres
.ddp
.size
= val
[4] - val
[3] + 1;
5431 adap
->params
.ofldq_wr_cred
= val
[5];
5433 adap
->params
.offload
= 1;
5435 if (caps_cmd
.rdmacaps
) {
5436 params
[0] = FW_PARAM_PFVF(STAG_START
);
5437 params
[1] = FW_PARAM_PFVF(STAG_END
);
5438 params
[2] = FW_PARAM_PFVF(RQ_START
);
5439 params
[3] = FW_PARAM_PFVF(RQ_END
);
5440 params
[4] = FW_PARAM_PFVF(PBL_START
);
5441 params
[5] = FW_PARAM_PFVF(PBL_END
);
5442 ret
= t4_query_params(adap
, adap
->mbox
, adap
->fn
, 0, 6,
5446 adap
->vres
.stag
.start
= val
[0];
5447 adap
->vres
.stag
.size
= val
[1] - val
[0] + 1;
5448 adap
->vres
.rq
.start
= val
[2];
5449 adap
->vres
.rq
.size
= val
[3] - val
[2] + 1;
5450 adap
->vres
.pbl
.start
= val
[4];
5451 adap
->vres
.pbl
.size
= val
[5] - val
[4] + 1;
5453 params
[0] = FW_PARAM_PFVF(SQRQ_START
);
5454 params
[1] = FW_PARAM_PFVF(SQRQ_END
);
5455 params
[2] = FW_PARAM_PFVF(CQ_START
);
5456 params
[3] = FW_PARAM_PFVF(CQ_END
);
5457 params
[4] = FW_PARAM_PFVF(OCQ_START
);
5458 params
[5] = FW_PARAM_PFVF(OCQ_END
);
5459 ret
= t4_query_params(adap
, 0, 0, 0, 6, params
, val
);
5462 adap
->vres
.qp
.start
= val
[0];
5463 adap
->vres
.qp
.size
= val
[1] - val
[0] + 1;
5464 adap
->vres
.cq
.start
= val
[2];
5465 adap
->vres
.cq
.size
= val
[3] - val
[2] + 1;
5466 adap
->vres
.ocq
.start
= val
[4];
5467 adap
->vres
.ocq
.size
= val
[5] - val
[4] + 1;
5469 if (caps_cmd
.iscsicaps
) {
5470 params
[0] = FW_PARAM_PFVF(ISCSI_START
);
5471 params
[1] = FW_PARAM_PFVF(ISCSI_END
);
5472 ret
= t4_query_params(adap
, adap
->mbox
, adap
->fn
, 0, 2,
5476 adap
->vres
.iscsi
.start
= val
[0];
5477 adap
->vres
.iscsi
.size
= val
[1] - val
[0] + 1;
5479 #undef FW_PARAM_PFVF
5483 * These are finalized by FW initialization, load their values now.
5485 t4_read_mtu_tbl(adap
, adap
->params
.mtus
, NULL
);
5486 t4_load_mtus(adap
, adap
->params
.mtus
, adap
->params
.a_wnd
,
5487 adap
->params
.b_wnd
);
5489 t4_init_tp_params(adap
);
5490 adap
->flags
|= FW_OK
;
5494 * Something bad happened. If a command timed out or failed with EIO
5495 * FW does not operate within its spec or something catastrophic
5496 * happened to HW/FW, stop issuing commands.
5499 if (ret
!= -ETIMEDOUT
&& ret
!= -EIO
)
5500 t4_fw_bye(adap
, adap
->mbox
);
5506 static pci_ers_result_t
eeh_err_detected(struct pci_dev
*pdev
,
5507 pci_channel_state_t state
)
5510 struct adapter
*adap
= pci_get_drvdata(pdev
);
5516 adap
->flags
&= ~FW_OK
;
5517 notify_ulds(adap
, CXGB4_STATE_START_RECOVERY
);
5518 spin_lock(&adap
->stats_lock
);
5519 for_each_port(adap
, i
) {
5520 struct net_device
*dev
= adap
->port
[i
];
5522 netif_device_detach(dev
);
5523 netif_carrier_off(dev
);
5525 spin_unlock(&adap
->stats_lock
);
5526 if (adap
->flags
& FULL_INIT_DONE
)
5529 if ((adap
->flags
& DEV_ENABLED
)) {
5530 pci_disable_device(pdev
);
5531 adap
->flags
&= ~DEV_ENABLED
;
5533 out
: return state
== pci_channel_io_perm_failure
?
5534 PCI_ERS_RESULT_DISCONNECT
: PCI_ERS_RESULT_NEED_RESET
;
5537 static pci_ers_result_t
eeh_slot_reset(struct pci_dev
*pdev
)
5540 struct fw_caps_config_cmd c
;
5541 struct adapter
*adap
= pci_get_drvdata(pdev
);
5544 pci_restore_state(pdev
);
5545 pci_save_state(pdev
);
5546 return PCI_ERS_RESULT_RECOVERED
;
5549 if (!(adap
->flags
& DEV_ENABLED
)) {
5550 if (pci_enable_device(pdev
)) {
5551 dev_err(&pdev
->dev
, "Cannot reenable PCI "
5552 "device after reset\n");
5553 return PCI_ERS_RESULT_DISCONNECT
;
5555 adap
->flags
|= DEV_ENABLED
;
5558 pci_set_master(pdev
);
5559 pci_restore_state(pdev
);
5560 pci_save_state(pdev
);
5561 pci_cleanup_aer_uncorrect_error_status(pdev
);
5563 if (t4_wait_dev_ready(adap
) < 0)
5564 return PCI_ERS_RESULT_DISCONNECT
;
5565 if (t4_fw_hello(adap
, adap
->fn
, adap
->fn
, MASTER_MUST
, NULL
) < 0)
5566 return PCI_ERS_RESULT_DISCONNECT
;
5567 adap
->flags
|= FW_OK
;
5568 if (adap_init1(adap
, &c
))
5569 return PCI_ERS_RESULT_DISCONNECT
;
5571 for_each_port(adap
, i
) {
5572 struct port_info
*p
= adap2pinfo(adap
, i
);
5574 ret
= t4_alloc_vi(adap
, adap
->fn
, p
->tx_chan
, adap
->fn
, 0, 1,
5577 return PCI_ERS_RESULT_DISCONNECT
;
5579 p
->xact_addr_filt
= -1;
5582 t4_load_mtus(adap
, adap
->params
.mtus
, adap
->params
.a_wnd
,
5583 adap
->params
.b_wnd
);
5586 return PCI_ERS_RESULT_DISCONNECT
;
5587 return PCI_ERS_RESULT_RECOVERED
;
5590 static void eeh_resume(struct pci_dev
*pdev
)
5593 struct adapter
*adap
= pci_get_drvdata(pdev
);
5599 for_each_port(adap
, i
) {
5600 struct net_device
*dev
= adap
->port
[i
];
5602 if (netif_running(dev
)) {
5604 cxgb_set_rxmode(dev
);
5606 netif_device_attach(dev
);
5611 static const struct pci_error_handlers cxgb4_eeh
= {
5612 .error_detected
= eeh_err_detected
,
5613 .slot_reset
= eeh_slot_reset
,
5614 .resume
= eeh_resume
,
5617 static inline bool is_10g_port(const struct link_config
*lc
)
5619 return (lc
->supported
& FW_PORT_CAP_SPEED_10G
) != 0;
5622 static inline void init_rspq(struct sge_rspq
*q
, u8 timer_idx
, u8 pkt_cnt_idx
,
5623 unsigned int size
, unsigned int iqe_size
)
5625 q
->intr_params
= QINTR_TIMER_IDX(timer_idx
) |
5626 (pkt_cnt_idx
< SGE_NCOUNTERS
? QINTR_CNT_EN
: 0);
5627 q
->pktcnt_idx
= pkt_cnt_idx
< SGE_NCOUNTERS
? pkt_cnt_idx
: 0;
5628 q
->iqe_len
= iqe_size
;
5633 * Perform default configuration of DMA queues depending on the number and type
5634 * of ports we found and the number of available CPUs. Most settings can be
5635 * modified by the admin prior to actual use.
5637 static void cfg_queues(struct adapter
*adap
)
5639 struct sge
*s
= &adap
->sge
;
5640 int i
, q10g
= 0, n10g
= 0, qidx
= 0;
5642 for_each_port(adap
, i
)
5643 n10g
+= is_10g_port(&adap2pinfo(adap
, i
)->link_cfg
);
5646 * We default to 1 queue per non-10G port and up to # of cores queues
5650 q10g
= (MAX_ETH_QSETS
- (adap
->params
.nports
- n10g
)) / n10g
;
5651 if (q10g
> netif_get_num_default_rss_queues())
5652 q10g
= netif_get_num_default_rss_queues();
5654 for_each_port(adap
, i
) {
5655 struct port_info
*pi
= adap2pinfo(adap
, i
);
5657 pi
->first_qset
= qidx
;
5658 pi
->nqsets
= is_10g_port(&pi
->link_cfg
) ? q10g
: 1;
5663 s
->max_ethqsets
= qidx
; /* MSI-X may lower it later */
5665 if (is_offload(adap
)) {
5667 * For offload we use 1 queue/channel if all ports are up to 1G,
5668 * otherwise we divide all available queues amongst the channels
5669 * capped by the number of available cores.
5672 i
= min_t(int, ARRAY_SIZE(s
->ofldrxq
),
5674 s
->ofldqsets
= roundup(i
, adap
->params
.nports
);
5676 s
->ofldqsets
= adap
->params
.nports
;
5677 /* For RDMA one Rx queue per channel suffices */
5678 s
->rdmaqs
= adap
->params
.nports
;
5681 for (i
= 0; i
< ARRAY_SIZE(s
->ethrxq
); i
++) {
5682 struct sge_eth_rxq
*r
= &s
->ethrxq
[i
];
5684 init_rspq(&r
->rspq
, 0, 0, 1024, 64);
5688 for (i
= 0; i
< ARRAY_SIZE(s
->ethtxq
); i
++)
5689 s
->ethtxq
[i
].q
.size
= 1024;
5691 for (i
= 0; i
< ARRAY_SIZE(s
->ctrlq
); i
++)
5692 s
->ctrlq
[i
].q
.size
= 512;
5694 for (i
= 0; i
< ARRAY_SIZE(s
->ofldtxq
); i
++)
5695 s
->ofldtxq
[i
].q
.size
= 1024;
5697 for (i
= 0; i
< ARRAY_SIZE(s
->ofldrxq
); i
++) {
5698 struct sge_ofld_rxq
*r
= &s
->ofldrxq
[i
];
5700 init_rspq(&r
->rspq
, 0, 0, 1024, 64);
5701 r
->rspq
.uld
= CXGB4_ULD_ISCSI
;
5705 for (i
= 0; i
< ARRAY_SIZE(s
->rdmarxq
); i
++) {
5706 struct sge_ofld_rxq
*r
= &s
->rdmarxq
[i
];
5708 init_rspq(&r
->rspq
, 0, 0, 511, 64);
5709 r
->rspq
.uld
= CXGB4_ULD_RDMA
;
5713 init_rspq(&s
->fw_evtq
, 6, 0, 512, 64);
5714 init_rspq(&s
->intrq
, 6, 0, 2 * MAX_INGQ
, 64);
5718 * Reduce the number of Ethernet queues across all ports to at most n.
5719 * n provides at least one queue per port.
5721 static void reduce_ethqs(struct adapter
*adap
, int n
)
5724 struct port_info
*pi
;
5726 while (n
< adap
->sge
.ethqsets
)
5727 for_each_port(adap
, i
) {
5728 pi
= adap2pinfo(adap
, i
);
5729 if (pi
->nqsets
> 1) {
5731 adap
->sge
.ethqsets
--;
5732 if (adap
->sge
.ethqsets
<= n
)
5738 for_each_port(adap
, i
) {
5739 pi
= adap2pinfo(adap
, i
);
5745 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5746 #define EXTRA_VECS 2
5748 static int enable_msix(struct adapter
*adap
)
5752 struct sge
*s
= &adap
->sge
;
5753 unsigned int nchan
= adap
->params
.nports
;
5754 struct msix_entry entries
[MAX_INGQ
+ 1];
5756 for (i
= 0; i
< ARRAY_SIZE(entries
); ++i
)
5757 entries
[i
].entry
= i
;
5759 want
= s
->max_ethqsets
+ EXTRA_VECS
;
5760 if (is_offload(adap
)) {
5761 want
+= s
->rdmaqs
+ s
->ofldqsets
;
5762 /* need nchan for each possible ULD */
5763 ofld_need
= 2 * nchan
;
5765 need
= adap
->params
.nports
+ EXTRA_VECS
+ ofld_need
;
5767 want
= pci_enable_msix_range(adap
->pdev
, entries
, need
, want
);
5772 * Distribute available vectors to the various queue groups.
5773 * Every group gets its minimum requirement and NIC gets top
5774 * priority for leftovers.
5776 i
= want
- EXTRA_VECS
- ofld_need
;
5777 if (i
< s
->max_ethqsets
) {
5778 s
->max_ethqsets
= i
;
5779 if (i
< s
->ethqsets
)
5780 reduce_ethqs(adap
, i
);
5782 if (is_offload(adap
)) {
5783 i
= want
- EXTRA_VECS
- s
->max_ethqsets
;
5784 i
-= ofld_need
- nchan
;
5785 s
->ofldqsets
= (i
/ nchan
) * nchan
; /* round down */
5787 for (i
= 0; i
< want
; ++i
)
5788 adap
->msix_info
[i
].vec
= entries
[i
].vector
;
5795 static int init_rss(struct adapter
*adap
)
5799 for_each_port(adap
, i
) {
5800 struct port_info
*pi
= adap2pinfo(adap
, i
);
5802 pi
->rss
= kcalloc(pi
->rss_size
, sizeof(u16
), GFP_KERNEL
);
5805 for (j
= 0; j
< pi
->rss_size
; j
++)
5806 pi
->rss
[j
] = ethtool_rxfh_indir_default(j
, pi
->nqsets
);
5811 static void print_port_info(const struct net_device
*dev
)
5815 const char *spd
= "";
5816 const struct port_info
*pi
= netdev_priv(dev
);
5817 const struct adapter
*adap
= pi
->adapter
;
5819 if (adap
->params
.pci
.speed
== PCI_EXP_LNKSTA_CLS_2_5GB
)
5821 else if (adap
->params
.pci
.speed
== PCI_EXP_LNKSTA_CLS_5_0GB
)
5824 if (pi
->link_cfg
.supported
& FW_PORT_CAP_SPEED_100M
)
5825 bufp
+= sprintf(bufp
, "100/");
5826 if (pi
->link_cfg
.supported
& FW_PORT_CAP_SPEED_1G
)
5827 bufp
+= sprintf(bufp
, "1000/");
5828 if (pi
->link_cfg
.supported
& FW_PORT_CAP_SPEED_10G
)
5829 bufp
+= sprintf(bufp
, "10G/");
5830 if (pi
->link_cfg
.supported
& FW_PORT_CAP_SPEED_40G
)
5831 bufp
+= sprintf(bufp
, "40G/");
5834 sprintf(bufp
, "BASE-%s", t4_get_port_type_description(pi
->port_type
));
5836 netdev_info(dev
, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
5837 adap
->params
.vpd
.id
,
5838 CHELSIO_CHIP_RELEASE(adap
->params
.chip
), buf
,
5839 is_offload(adap
) ? "R" : "", adap
->params
.pci
.width
, spd
,
5840 (adap
->flags
& USING_MSIX
) ? " MSI-X" :
5841 (adap
->flags
& USING_MSI
) ? " MSI" : "");
5842 netdev_info(dev
, "S/N: %s, E/C: %s\n",
5843 adap
->params
.vpd
.sn
, adap
->params
.vpd
.ec
);
5846 static void enable_pcie_relaxed_ordering(struct pci_dev
*dev
)
5848 pcie_capability_set_word(dev
, PCI_EXP_DEVCTL
, PCI_EXP_DEVCTL_RELAX_EN
);
5852 * Free the following resources:
5853 * - memory used for tables
5856 * - resources FW is holding for us
5858 static void free_some_resources(struct adapter
*adapter
)
5862 t4_free_mem(adapter
->l2t
);
5863 t4_free_mem(adapter
->tids
.tid_tab
);
5864 disable_msi(adapter
);
5866 for_each_port(adapter
, i
)
5867 if (adapter
->port
[i
]) {
5868 kfree(adap2pinfo(adapter
, i
)->rss
);
5869 free_netdev(adapter
->port
[i
]);
5871 if (adapter
->flags
& FW_OK
)
5872 t4_fw_bye(adapter
, adapter
->fn
);
5875 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5876 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5877 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5878 #define SEGMENT_SIZE 128
5880 static int init_one(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
5882 int func
, i
, err
, s_qpp
, qpp
, num_seg
;
5883 struct port_info
*pi
;
5884 bool highdma
= false;
5885 struct adapter
*adapter
= NULL
;
5887 printk_once(KERN_INFO
"%s - version %s\n", DRV_DESC
, DRV_VERSION
);
5889 err
= pci_request_regions(pdev
, KBUILD_MODNAME
);
5891 /* Just info, some other driver may have claimed the device. */
5892 dev_info(&pdev
->dev
, "cannot obtain PCI resources\n");
5896 /* We control everything through one PF */
5897 func
= PCI_FUNC(pdev
->devfn
);
5898 if (func
!= ent
->driver_data
) {
5899 pci_save_state(pdev
); /* to restore SR-IOV later */
5903 err
= pci_enable_device(pdev
);
5905 dev_err(&pdev
->dev
, "cannot enable PCI device\n");
5906 goto out_release_regions
;
5909 if (!pci_set_dma_mask(pdev
, DMA_BIT_MASK(64))) {
5911 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
5913 dev_err(&pdev
->dev
, "unable to obtain 64-bit DMA for "
5914 "coherent allocations\n");
5915 goto out_disable_device
;
5918 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
5920 dev_err(&pdev
->dev
, "no usable DMA configuration\n");
5921 goto out_disable_device
;
5925 pci_enable_pcie_error_reporting(pdev
);
5926 enable_pcie_relaxed_ordering(pdev
);
5927 pci_set_master(pdev
);
5928 pci_save_state(pdev
);
5930 adapter
= kzalloc(sizeof(*adapter
), GFP_KERNEL
);
5933 goto out_disable_device
;
5936 /* PCI device has been enabled */
5937 adapter
->flags
|= DEV_ENABLED
;
5939 adapter
->regs
= pci_ioremap_bar(pdev
, 0);
5940 if (!adapter
->regs
) {
5941 dev_err(&pdev
->dev
, "cannot map device registers\n");
5943 goto out_free_adapter
;
5946 adapter
->pdev
= pdev
;
5947 adapter
->pdev_dev
= &pdev
->dev
;
5948 adapter
->mbox
= func
;
5950 adapter
->msg_enable
= dflt_msg_enable
;
5951 memset(adapter
->chan_map
, 0xff, sizeof(adapter
->chan_map
));
5953 spin_lock_init(&adapter
->stats_lock
);
5954 spin_lock_init(&adapter
->tid_release_lock
);
5956 INIT_WORK(&adapter
->tid_release_task
, process_tid_release_list
);
5957 INIT_WORK(&adapter
->db_full_task
, process_db_full
);
5958 INIT_WORK(&adapter
->db_drop_task
, process_db_drop
);
5960 err
= t4_prep_adapter(adapter
);
5962 goto out_unmap_bar0
;
5964 if (!is_t4(adapter
->params
.chip
)) {
5965 s_qpp
= QUEUESPERPAGEPF1
* adapter
->fn
;
5966 qpp
= 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter
,
5967 SGE_EGRESS_QUEUES_PER_PAGE_PF
) >> s_qpp
);
5968 num_seg
= PAGE_SIZE
/ SEGMENT_SIZE
;
5970 /* Each segment size is 128B. Write coalescing is enabled only
5971 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5972 * queue is less no of segments that can be accommodated in
5975 if (qpp
> num_seg
) {
5977 "Incorrect number of egress queues per page\n");
5979 goto out_unmap_bar0
;
5981 adapter
->bar2
= ioremap_wc(pci_resource_start(pdev
, 2),
5982 pci_resource_len(pdev
, 2));
5983 if (!adapter
->bar2
) {
5984 dev_err(&pdev
->dev
, "cannot map device bar2 region\n");
5986 goto out_unmap_bar0
;
5990 setup_memwin(adapter
);
5991 err
= adap_init0(adapter
);
5992 setup_memwin_rdma(adapter
);
5996 for_each_port(adapter
, i
) {
5997 struct net_device
*netdev
;
5999 netdev
= alloc_etherdev_mq(sizeof(struct port_info
),
6006 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
6008 adapter
->port
[i
] = netdev
;
6009 pi
= netdev_priv(netdev
);
6010 pi
->adapter
= adapter
;
6011 pi
->xact_addr_filt
= -1;
6013 netdev
->irq
= pdev
->irq
;
6015 netdev
->hw_features
= NETIF_F_SG
| TSO_FLAGS
|
6016 NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
|
6017 NETIF_F_RXCSUM
| NETIF_F_RXHASH
|
6018 NETIF_F_HW_VLAN_CTAG_TX
| NETIF_F_HW_VLAN_CTAG_RX
;
6020 netdev
->hw_features
|= NETIF_F_HIGHDMA
;
6021 netdev
->features
|= netdev
->hw_features
;
6022 netdev
->vlan_features
= netdev
->features
& VLAN_FEAT
;
6024 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
6026 netdev
->netdev_ops
= &cxgb4_netdev_ops
;
6027 SET_ETHTOOL_OPS(netdev
, &cxgb_ethtool_ops
);
6030 pci_set_drvdata(pdev
, adapter
);
6032 if (adapter
->flags
& FW_OK
) {
6033 err
= t4_port_init(adapter
, func
, func
, 0);
6039 * Configure queues and allocate tables now, they can be needed as
6040 * soon as the first register_netdev completes.
6042 cfg_queues(adapter
);
6044 adapter
->l2t
= t4_init_l2t();
6045 if (!adapter
->l2t
) {
6046 /* We tolerate a lack of L2T, giving up some functionality */
6047 dev_warn(&pdev
->dev
, "could not allocate L2T, continuing\n");
6048 adapter
->params
.offload
= 0;
6051 if (is_offload(adapter
) && tid_init(&adapter
->tids
) < 0) {
6052 dev_warn(&pdev
->dev
, "could not allocate TID table, "
6054 adapter
->params
.offload
= 0;
6057 /* See what interrupts we'll be using */
6058 if (msi
> 1 && enable_msix(adapter
) == 0)
6059 adapter
->flags
|= USING_MSIX
;
6060 else if (msi
> 0 && pci_enable_msi(pdev
) == 0)
6061 adapter
->flags
|= USING_MSI
;
6063 err
= init_rss(adapter
);
6068 * The card is now ready to go. If any errors occur during device
6069 * registration we do not fail the whole card but rather proceed only
6070 * with the ports we manage to register successfully. However we must
6071 * register at least one net device.
6073 for_each_port(adapter
, i
) {
6074 pi
= adap2pinfo(adapter
, i
);
6075 netif_set_real_num_tx_queues(adapter
->port
[i
], pi
->nqsets
);
6076 netif_set_real_num_rx_queues(adapter
->port
[i
], pi
->nqsets
);
6078 err
= register_netdev(adapter
->port
[i
]);
6081 adapter
->chan_map
[pi
->tx_chan
] = i
;
6082 print_port_info(adapter
->port
[i
]);
6085 dev_err(&pdev
->dev
, "could not register any net devices\n");
6089 dev_warn(&pdev
->dev
, "only %d net devices registered\n", i
);
6093 if (cxgb4_debugfs_root
) {
6094 adapter
->debugfs_root
= debugfs_create_dir(pci_name(pdev
),
6095 cxgb4_debugfs_root
);
6096 setup_debugfs(adapter
);
6099 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
6100 pdev
->needs_freset
= 1;
6102 if (is_offload(adapter
))
6103 attach_ulds(adapter
);
6106 #ifdef CONFIG_PCI_IOV
6107 if (func
< ARRAY_SIZE(num_vf
) && num_vf
[func
] > 0)
6108 if (pci_enable_sriov(pdev
, num_vf
[func
]) == 0)
6109 dev_info(&pdev
->dev
,
6110 "instantiated %u virtual functions\n",
6116 free_some_resources(adapter
);
6118 if (!is_t4(adapter
->params
.chip
))
6119 iounmap(adapter
->bar2
);
6121 iounmap(adapter
->regs
);
6125 pci_disable_pcie_error_reporting(pdev
);
6126 pci_disable_device(pdev
);
6127 out_release_regions
:
6128 pci_release_regions(pdev
);
6132 static void remove_one(struct pci_dev
*pdev
)
6134 struct adapter
*adapter
= pci_get_drvdata(pdev
);
6136 #ifdef CONFIG_PCI_IOV
6137 pci_disable_sriov(pdev
);
6144 if (is_offload(adapter
))
6145 detach_ulds(adapter
);
6147 for_each_port(adapter
, i
)
6148 if (adapter
->port
[i
]->reg_state
== NETREG_REGISTERED
)
6149 unregister_netdev(adapter
->port
[i
]);
6151 if (adapter
->debugfs_root
)
6152 debugfs_remove_recursive(adapter
->debugfs_root
);
6154 /* If we allocated filters, free up state associated with any
6157 if (adapter
->tids
.ftid_tab
) {
6158 struct filter_entry
*f
= &adapter
->tids
.ftid_tab
[0];
6159 for (i
= 0; i
< (adapter
->tids
.nftids
+
6160 adapter
->tids
.nsftids
); i
++, f
++)
6162 clear_filter(adapter
, f
);
6165 if (adapter
->flags
& FULL_INIT_DONE
)
6168 free_some_resources(adapter
);
6169 iounmap(adapter
->regs
);
6170 if (!is_t4(adapter
->params
.chip
))
6171 iounmap(adapter
->bar2
);
6172 pci_disable_pcie_error_reporting(pdev
);
6173 if ((adapter
->flags
& DEV_ENABLED
)) {
6174 pci_disable_device(pdev
);
6175 adapter
->flags
&= ~DEV_ENABLED
;
6177 pci_release_regions(pdev
);
6180 pci_release_regions(pdev
);
6183 static struct pci_driver cxgb4_driver
= {
6184 .name
= KBUILD_MODNAME
,
6185 .id_table
= cxgb4_pci_tbl
,
6187 .remove
= remove_one
,
6188 .err_handler
= &cxgb4_eeh
,
6191 static int __init
cxgb4_init_module(void)
6195 workq
= create_singlethread_workqueue("cxgb4");
6199 /* Debugfs support is optional, just warn if this fails */
6200 cxgb4_debugfs_root
= debugfs_create_dir(KBUILD_MODNAME
, NULL
);
6201 if (!cxgb4_debugfs_root
)
6202 pr_warn("could not create debugfs entry, continuing\n");
6204 ret
= pci_register_driver(&cxgb4_driver
);
6206 debugfs_remove(cxgb4_debugfs_root
);
6207 destroy_workqueue(workq
);
6210 register_inet6addr_notifier(&cxgb4_inet6addr_notifier
);
6215 static void __exit
cxgb4_cleanup_module(void)
6217 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier
);
6218 pci_unregister_driver(&cxgb4_driver
);
6219 debugfs_remove(cxgb4_debugfs_root
); /* NULL ok */
6220 flush_workqueue(workq
);
6221 destroy_workqueue(workq
);
6224 module_init(cxgb4_init_module
);
6225 module_exit(cxgb4_cleanup_module
);