2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2014 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>
70 #include "cxgb4_dcb.h"
73 #include <../drivers/net/bonding/bonding.h>
78 #define DRV_VERSION "2.0.0-ko"
79 #define DRV_DESC "Chelsio T4/T5 Network Driver"
82 * Max interrupt hold-off timer value in us. Queues fall back to this value
83 * under extreme memory pressure so it's largish to give the system time to
86 #define MAX_SGE_TIMERVAL 200U
90 * Physical Function provisioning constants.
92 PFRES_NVI = 4, /* # of Virtual Interfaces */
93 PFRES_NETHCTRL = 128, /* # of EQs used for ETH or CTRL Qs */
94 PFRES_NIQFLINT = 128, /* # of ingress Qs/w Free List(s)/intr
96 PFRES_NEQ = 256, /* # of egress queues */
97 PFRES_NIQ = 0, /* # of ingress queues */
98 PFRES_TC = 0, /* PCI-E traffic class */
99 PFRES_NEXACTF = 128, /* # of exact MPS filters */
101 PFRES_R_CAPS = FW_CMD_CAP_PF,
102 PFRES_WX_CAPS = FW_CMD_CAP_PF,
104 #ifdef CONFIG_PCI_IOV
106 * Virtual Function provisioning constants. We need two extra Ingress
107 * Queues with Interrupt capability to serve as the VF's Firmware
108 * Event Queue and Forwarded Interrupt Queue (when using MSI mode) --
109 * neither will have Free Lists associated with them). For each
110 * Ethernet/Control Egress Queue and for each Free List, we need an
113 VFRES_NPORTS = 1, /* # of "ports" per VF */
114 VFRES_NQSETS = 2, /* # of "Queue Sets" per VF */
116 VFRES_NVI = VFRES_NPORTS, /* # of Virtual Interfaces */
117 VFRES_NETHCTRL = VFRES_NQSETS, /* # of EQs used for ETH or CTRL Qs */
118 VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
119 VFRES_NEQ = VFRES_NQSETS*2, /* # of egress queues */
120 VFRES_NIQ = 0, /* # of non-fl/int ingress queues */
121 VFRES_TC = 0, /* PCI-E traffic class */
122 VFRES_NEXACTF = 16, /* # of exact MPS filters */
124 VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
125 VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
130 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
131 * static and likely not to be useful in the long run. We really need to
132 * implement some form of persistent configuration which the firmware
135 static unsigned int pfvfres_pmask(struct adapter *adapter,
136 unsigned int pf, unsigned int vf)
138 unsigned int portn, portvec;
141 * Give PF's access to all of the ports.
144 return FW_PFVF_CMD_PMASK_MASK;
147 * For VFs, we'll assign them access to the ports based purely on the
148 * PF. We assign active ports in order, wrapping around if there are
149 * fewer active ports than PFs: e.g. active port[pf % nports].
150 * Unfortunately the adapter's port_info structs haven't been
151 * initialized yet so we have to compute this.
153 if (adapter->params.nports == 0)
156 portn = pf % adapter->params.nports;
157 portvec = adapter->params.portvec;
160 * Isolate the lowest set bit in the port vector. If we're at
161 * the port number that we want, return that as the pmask.
162 * otherwise mask that bit out of the port vector and
163 * decrement our port number ...
165 unsigned int pmask = portvec ^ (portvec & (portvec-1));
175 MAX_TXQ_ENTRIES = 16384,
176 MAX_CTRL_TXQ_ENTRIES = 1024,
177 MAX_RSPQ_ENTRIES = 16384,
178 MAX_RX_BUFFERS = 16384,
179 MIN_TXQ_ENTRIES = 32,
180 MIN_CTRL_TXQ_ENTRIES = 32,
181 MIN_RSPQ_ENTRIES = 128,
185 /* Host shadow copy of ingress filter entry. This is in host native format
186 * and doesn't match the ordering or bit order, etc. of the hardware of the
187 * firmware command. The use of bit-field structure elements is purely to
188 * remind ourselves of the field size limitations and save memory in the case
189 * where the filter table is large.
191 struct filter_entry {
192 /* Administrative fields for filter.
194 u32 valid:1; /* filter allocated and valid */
195 u32 locked:1; /* filter is administratively locked */
197 u32 pending:1; /* filter action is pending firmware reply */
198 u32 smtidx:8; /* Source MAC Table index for smac */
199 struct l2t_entry *l2t; /* Layer Two Table entry for dmac */
201 /* The filter itself. Most of this is a straight copy of information
202 * provided by the extended ioctl(). Some fields are translated to
203 * internal forms -- for instance the Ingress Queue ID passed in from
204 * the ioctl() is translated into the Absolute Ingress Queue ID.
206 struct ch_filter_specification fs;
209 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
210 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
211 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
213 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
215 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
216 CH_DEVICE(0xa000, 0), /* PE10K */
217 CH_DEVICE(0x4001, -1),
218 CH_DEVICE(0x4002, -1),
219 CH_DEVICE(0x4003, -1),
220 CH_DEVICE(0x4004, -1),
221 CH_DEVICE(0x4005, -1),
222 CH_DEVICE(0x4006, -1),
223 CH_DEVICE(0x4007, -1),
224 CH_DEVICE(0x4008, -1),
225 CH_DEVICE(0x4009, -1),
226 CH_DEVICE(0x400a, -1),
227 CH_DEVICE(0x4401, 4),
228 CH_DEVICE(0x4402, 4),
229 CH_DEVICE(0x4403, 4),
230 CH_DEVICE(0x4404, 4),
231 CH_DEVICE(0x4405, 4),
232 CH_DEVICE(0x4406, 4),
233 CH_DEVICE(0x4407, 4),
234 CH_DEVICE(0x4408, 4),
235 CH_DEVICE(0x4409, 4),
236 CH_DEVICE(0x440a, 4),
237 CH_DEVICE(0x440d, 4),
238 CH_DEVICE(0x440e, 4),
239 CH_DEVICE(0x5001, 4),
240 CH_DEVICE(0x5002, 4),
241 CH_DEVICE(0x5003, 4),
242 CH_DEVICE(0x5004, 4),
243 CH_DEVICE(0x5005, 4),
244 CH_DEVICE(0x5006, 4),
245 CH_DEVICE(0x5007, 4),
246 CH_DEVICE(0x5008, 4),
247 CH_DEVICE(0x5009, 4),
248 CH_DEVICE(0x500A, 4),
249 CH_DEVICE(0x500B, 4),
250 CH_DEVICE(0x500C, 4),
251 CH_DEVICE(0x500D, 4),
252 CH_DEVICE(0x500E, 4),
253 CH_DEVICE(0x500F, 4),
254 CH_DEVICE(0x5010, 4),
255 CH_DEVICE(0x5011, 4),
256 CH_DEVICE(0x5012, 4),
257 CH_DEVICE(0x5013, 4),
258 CH_DEVICE(0x5014, 4),
259 CH_DEVICE(0x5015, 4),
260 CH_DEVICE(0x5080, 4),
261 CH_DEVICE(0x5081, 4),
262 CH_DEVICE(0x5082, 4),
263 CH_DEVICE(0x5083, 4),
264 CH_DEVICE(0x5084, 4),
265 CH_DEVICE(0x5085, 4),
266 CH_DEVICE(0x5401, 4),
267 CH_DEVICE(0x5402, 4),
268 CH_DEVICE(0x5403, 4),
269 CH_DEVICE(0x5404, 4),
270 CH_DEVICE(0x5405, 4),
271 CH_DEVICE(0x5406, 4),
272 CH_DEVICE(0x5407, 4),
273 CH_DEVICE(0x5408, 4),
274 CH_DEVICE(0x5409, 4),
275 CH_DEVICE(0x540A, 4),
276 CH_DEVICE(0x540B, 4),
277 CH_DEVICE(0x540C, 4),
278 CH_DEVICE(0x540D, 4),
279 CH_DEVICE(0x540E, 4),
280 CH_DEVICE(0x540F, 4),
281 CH_DEVICE(0x5410, 4),
282 CH_DEVICE(0x5411, 4),
283 CH_DEVICE(0x5412, 4),
284 CH_DEVICE(0x5413, 4),
285 CH_DEVICE(0x5414, 4),
286 CH_DEVICE(0x5415, 4),
287 CH_DEVICE(0x5480, 4),
288 CH_DEVICE(0x5481, 4),
289 CH_DEVICE(0x5482, 4),
290 CH_DEVICE(0x5483, 4),
291 CH_DEVICE(0x5484, 4),
292 CH_DEVICE(0x5485, 4),
296 #define FW4_FNAME "cxgb4/t4fw.bin"
297 #define FW5_FNAME "cxgb4/t5fw.bin"
298 #define FW4_CFNAME "cxgb4/t4-config.txt"
299 #define FW5_CFNAME "cxgb4/t5-config.txt"
301 MODULE_DESCRIPTION(DRV_DESC);
302 MODULE_AUTHOR("Chelsio Communications");
303 MODULE_LICENSE("Dual BSD/GPL");
304 MODULE_VERSION(DRV_VERSION);
305 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
306 MODULE_FIRMWARE(FW4_FNAME);
307 MODULE_FIRMWARE(FW5_FNAME);
310 * Normally we're willing to become the firmware's Master PF but will be happy
311 * if another PF has already become the Master and initialized the adapter.
312 * Setting "force_init" will cause this driver to forcibly establish itself as
313 * the Master PF and initialize the adapter.
315 static uint force_init;
317 module_param(force_init, uint, 0644);
318 MODULE_PARM_DESC(force_init, "Forcibly become Master PF and initialize adapter");
321 * Normally if the firmware we connect to has Configuration File support, we
322 * use that and only fall back to the old Driver-based initialization if the
323 * Configuration File fails for some reason. If force_old_init is set, then
324 * we'll always use the old Driver-based initialization sequence.
326 static uint force_old_init;
328 module_param(force_old_init, uint, 0644);
329 MODULE_PARM_DESC(force_old_init, "Force old initialization sequence");
331 static int dflt_msg_enable = DFLT_MSG_ENABLE;
333 module_param(dflt_msg_enable, int, 0644);
334 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
337 * The driver uses the best interrupt scheme available on a platform in the
338 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
339 * of these schemes the driver may consider as follows:
341 * msi = 2: choose from among all three options
342 * msi = 1: only consider MSI and INTx interrupts
343 * msi = 0: force INTx interrupts
347 module_param(msi, int, 0644);
348 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
351 * Queue interrupt hold-off timer values. Queues default to the first of these
354 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
356 module_param_array(intr_holdoff, uint, NULL, 0644);
357 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
358 "0..4 in microseconds");
360 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
362 module_param_array(intr_cnt, uint, NULL, 0644);
363 MODULE_PARM_DESC(intr_cnt,
364 "thresholds 1..3 for queue interrupt packet counters");
367 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
368 * offset by 2 bytes in order to have the IP headers line up on 4-byte
369 * boundaries. This is a requirement for many architectures which will throw
370 * a machine check fault if an attempt is made to access one of the 4-byte IP
371 * header fields on a non-4-byte boundary. And it's a major performance issue
372 * even on some architectures which allow it like some implementations of the
373 * x86 ISA. However, some architectures don't mind this and for some very
374 * edge-case performance sensitive applications (like forwarding large volumes
375 * of small packets), setting this DMA offset to 0 will decrease the number of
376 * PCI-E Bus transfers enough to measurably affect performance.
378 static int rx_dma_offset = 2;
382 #ifdef CONFIG_PCI_IOV
383 module_param(vf_acls, bool, 0644);
384 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
386 /* Configure the number of PCI-E Virtual Function which are to be instantiated
387 * on SR-IOV Capable Physical Functions.
389 static unsigned int num_vf[NUM_OF_PF_WITH_SRIOV];
391 module_param_array(num_vf, uint, NULL, 0644);
392 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
395 /* TX Queue select used to determine what algorithm to use for selecting TX
396 * queue. Select between the kernel provided function (select_queue=0) or user
397 * cxgb_select_queue function (select_queue=1)
399 * Default: select_queue=0
401 static int select_queue;
402 module_param(select_queue, int, 0644);
403 MODULE_PARM_DESC(select_queue,
404 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
407 * The filter TCAM has a fixed portion and a variable portion. The fixed
408 * portion can match on source/destination IP IPv4/IPv6 addresses and TCP/UDP
409 * ports. The variable portion is 36 bits which can include things like Exact
410 * Match MAC Index (9 bits), Ether Type (16 bits), IP Protocol (8 bits),
411 * [Inner] VLAN Tag (17 bits), etc. which, if all were somehow selected, would
412 * far exceed the 36-bit budget for this "compressed" header portion of the
413 * filter. Thus, we have a scarce resource which must be carefully managed.
415 * By default we set this up to mostly match the set of filter matching
416 * capabilities of T3 but with accommodations for some of T4's more
417 * interesting features:
419 * { IP Fragment (1), MPS Match Type (3), IP Protocol (8),
420 * [Inner] VLAN (17), Port (3), FCoE (1) }
423 TP_VLAN_PRI_MAP_DEFAULT = HW_TPL_FR_MT_PR_IV_P_FC,
424 TP_VLAN_PRI_MAP_FIRST = FCOE_SHIFT,
425 TP_VLAN_PRI_MAP_LAST = FRAGMENTATION_SHIFT,
428 static unsigned int tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
430 module_param(tp_vlan_pri_map, uint, 0644);
431 MODULE_PARM_DESC(tp_vlan_pri_map, "global compressed filter configuration");
433 static struct dentry *cxgb4_debugfs_root;
435 static LIST_HEAD(adapter_list);
436 static DEFINE_MUTEX(uld_mutex);
437 /* Adapter list to be accessed from atomic context */
438 static LIST_HEAD(adap_rcu_list);
439 static DEFINE_SPINLOCK(adap_rcu_lock);
440 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
441 static const char *uld_str[] = { "RDMA", "iSCSI" };
443 static void link_report(struct net_device *dev)
445 if (!netif_carrier_ok(dev))
446 netdev_info(dev, "link down\n");
448 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
450 const char *s = "10Mbps";
451 const struct port_info *p = netdev_priv(dev);
453 switch (p->link_cfg.speed) {
468 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
473 #ifdef CONFIG_CHELSIO_T4_DCB
474 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
475 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
477 struct port_info *pi = netdev_priv(dev);
478 struct adapter *adap = pi->adapter;
479 struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
482 /* We use a simple mapping of Port TX Queue Index to DCB
483 * Priority when we're enabling DCB.
485 for (i = 0; i < pi->nqsets; i++, txq++) {
489 name = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
490 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
491 FW_PARAMS_PARAM_YZ(txq->q.cntxt_id));
492 value = enable ? i : 0xffffffff;
494 /* Since we can be called while atomic (from "interrupt
495 * level") we need to issue the Set Parameters Commannd
496 * without sleeping (timeout < 0).
498 err = t4_set_params_nosleep(adap, adap->mbox, adap->fn, 0, 1,
502 dev_err(adap->pdev_dev,
503 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
504 enable ? "set" : "unset", pi->port_id, i, -err);
507 #endif /* CONFIG_CHELSIO_T4_DCB */
509 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
511 struct net_device *dev = adapter->port[port_id];
513 /* Skip changes from disabled ports. */
514 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
516 netif_carrier_on(dev);
518 #ifdef CONFIG_CHELSIO_T4_DCB
519 cxgb4_dcb_state_init(dev);
520 dcb_tx_queue_prio_enable(dev, false);
521 #endif /* CONFIG_CHELSIO_T4_DCB */
522 netif_carrier_off(dev);
529 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
531 static const char *mod_str[] = {
532 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
535 const struct net_device *dev = adap->port[port_id];
536 const struct port_info *pi = netdev_priv(dev);
538 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
539 netdev_info(dev, "port module unplugged\n");
540 else if (pi->mod_type < ARRAY_SIZE(mod_str))
541 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
545 * Configure the exact and hash address filters to handle a port's multicast
546 * and secondary unicast MAC addresses.
548 static int set_addr_filters(const struct net_device *dev, bool sleep)
556 const struct netdev_hw_addr *ha;
557 int uc_cnt = netdev_uc_count(dev);
558 int mc_cnt = netdev_mc_count(dev);
559 const struct port_info *pi = netdev_priv(dev);
560 unsigned int mb = pi->adapter->fn;
562 /* first do the secondary unicast addresses */
563 netdev_for_each_uc_addr(ha, dev) {
564 addr[naddr++] = ha->addr;
565 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
566 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
567 naddr, addr, filt_idx, &uhash, sleep);
576 /* next set up the multicast addresses */
577 netdev_for_each_mc_addr(ha, dev) {
578 addr[naddr++] = ha->addr;
579 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
580 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
581 naddr, addr, filt_idx, &mhash, sleep);
590 return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
591 uhash | mhash, sleep);
594 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
595 module_param(dbfifo_int_thresh, int, 0644);
596 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
599 * usecs to sleep while draining the dbfifo
601 static int dbfifo_drain_delay = 1000;
602 module_param(dbfifo_drain_delay, int, 0644);
603 MODULE_PARM_DESC(dbfifo_drain_delay,
604 "usecs to sleep while draining the dbfifo");
607 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
608 * If @mtu is -1 it is left unchanged.
610 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
613 struct port_info *pi = netdev_priv(dev);
615 ret = set_addr_filters(dev, sleep_ok);
617 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
618 (dev->flags & IFF_PROMISC) ? 1 : 0,
619 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
624 static struct workqueue_struct *workq;
627 * link_start - enable a port
628 * @dev: the port to enable
630 * Performs the MAC and PHY actions needed to enable a port.
632 static int link_start(struct net_device *dev)
635 struct port_info *pi = netdev_priv(dev);
636 unsigned int mb = pi->adapter->fn;
639 * We do not set address filters and promiscuity here, the stack does
640 * that step explicitly.
642 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
643 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
645 ret = t4_change_mac(pi->adapter, mb, pi->viid,
646 pi->xact_addr_filt, dev->dev_addr, true,
649 pi->xact_addr_filt = ret;
654 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
657 ret = t4_enable_vi_params(pi->adapter, mb, pi->viid, true,
658 true, CXGB4_DCB_ENABLED);
663 int cxgb4_dcb_enabled(const struct net_device *dev)
665 #ifdef CONFIG_CHELSIO_T4_DCB
666 struct port_info *pi = netdev_priv(dev);
668 return pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED;
673 EXPORT_SYMBOL(cxgb4_dcb_enabled);
675 #ifdef CONFIG_CHELSIO_T4_DCB
676 /* Handle a Data Center Bridging update message from the firmware. */
677 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
679 int port = FW_PORT_CMD_PORTID_GET(ntohl(pcmd->op_to_portid));
680 struct net_device *dev = adap->port[port];
681 int old_dcb_enabled = cxgb4_dcb_enabled(dev);
684 cxgb4_dcb_handle_fw_update(adap, pcmd);
685 new_dcb_enabled = cxgb4_dcb_enabled(dev);
687 /* If the DCB has become enabled or disabled on the port then we're
688 * going to need to set up/tear down DCB Priority parameters for the
689 * TX Queues associated with the port.
691 if (new_dcb_enabled != old_dcb_enabled)
692 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
694 #endif /* CONFIG_CHELSIO_T4_DCB */
696 /* Clear a filter and release any of its resources that we own. This also
697 * clears the filter's "pending" status.
699 static void clear_filter(struct adapter *adap, struct filter_entry *f)
701 /* If the new or old filter have loopback rewriteing rules then we'll
702 * need to free any existing Layer Two Table (L2T) entries of the old
703 * filter rule. The firmware will handle freeing up any Source MAC
704 * Table (SMT) entries used for rewriting Source MAC Addresses in
708 cxgb4_l2t_release(f->l2t);
710 /* The zeroing of the filter rule below clears the filter valid,
711 * pending, locked flags, l2t pointer, etc. so it's all we need for
714 memset(f, 0, sizeof(*f));
717 /* Handle a filter write/deletion reply.
719 static void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
721 unsigned int idx = GET_TID(rpl);
722 unsigned int nidx = idx - adap->tids.ftid_base;
724 struct filter_entry *f;
726 if (idx >= adap->tids.ftid_base && nidx <
727 (adap->tids.nftids + adap->tids.nsftids)) {
729 ret = GET_TCB_COOKIE(rpl->cookie);
730 f = &adap->tids.ftid_tab[idx];
732 if (ret == FW_FILTER_WR_FLT_DELETED) {
733 /* Clear the filter when we get confirmation from the
734 * hardware that the filter has been deleted.
736 clear_filter(adap, f);
737 } else if (ret == FW_FILTER_WR_SMT_TBL_FULL) {
738 dev_err(adap->pdev_dev, "filter %u setup failed due to full SMT\n",
740 clear_filter(adap, f);
741 } else if (ret == FW_FILTER_WR_FLT_ADDED) {
742 f->smtidx = (be64_to_cpu(rpl->oldval) >> 24) & 0xff;
743 f->pending = 0; /* asynchronous setup completed */
746 /* Something went wrong. Issue a warning about the
747 * problem and clear everything out.
749 dev_err(adap->pdev_dev, "filter %u setup failed with error %u\n",
751 clear_filter(adap, f);
756 /* Response queue handler for the FW event queue.
758 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
759 const struct pkt_gl *gl)
761 u8 opcode = ((const struct rss_header *)rsp)->opcode;
763 rsp++; /* skip RSS header */
765 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
767 if (unlikely(opcode == CPL_FW4_MSG &&
768 ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
770 opcode = ((const struct rss_header *)rsp)->opcode;
772 if (opcode != CPL_SGE_EGR_UPDATE) {
773 dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
779 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
780 const struct cpl_sge_egr_update *p = (void *)rsp;
781 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
784 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
786 if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
787 struct sge_eth_txq *eq;
789 eq = container_of(txq, struct sge_eth_txq, q);
790 netif_tx_wake_queue(eq->txq);
792 struct sge_ofld_txq *oq;
794 oq = container_of(txq, struct sge_ofld_txq, q);
795 tasklet_schedule(&oq->qresume_tsk);
797 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
798 const struct cpl_fw6_msg *p = (void *)rsp;
800 #ifdef CONFIG_CHELSIO_T4_DCB
801 const struct fw_port_cmd *pcmd = (const void *)p->data;
802 unsigned int cmd = FW_CMD_OP_GET(ntohl(pcmd->op_to_portid));
803 unsigned int action =
804 FW_PORT_CMD_ACTION_GET(ntohl(pcmd->action_to_len16));
806 if (cmd == FW_PORT_CMD &&
807 action == FW_PORT_ACTION_GET_PORT_INFO) {
808 int port = FW_PORT_CMD_PORTID_GET(
809 be32_to_cpu(pcmd->op_to_portid));
810 struct net_device *dev = q->adap->port[port];
811 int state_input = ((pcmd->u.info.dcbxdis_pkd &
813 ? CXGB4_DCB_INPUT_FW_DISABLED
814 : CXGB4_DCB_INPUT_FW_ENABLED);
816 cxgb4_dcb_state_fsm(dev, state_input);
819 if (cmd == FW_PORT_CMD &&
820 action == FW_PORT_ACTION_L2_DCB_CFG)
821 dcb_rpl(q->adap, pcmd);
825 t4_handle_fw_rpl(q->adap, p->data);
826 } else if (opcode == CPL_L2T_WRITE_RPL) {
827 const struct cpl_l2t_write_rpl *p = (void *)rsp;
829 do_l2t_write_rpl(q->adap, p);
830 } else if (opcode == CPL_SET_TCB_RPL) {
831 const struct cpl_set_tcb_rpl *p = (void *)rsp;
833 filter_rpl(q->adap, p);
835 dev_err(q->adap->pdev_dev,
836 "unexpected CPL %#x on FW event queue\n", opcode);
842 * uldrx_handler - response queue handler for ULD queues
843 * @q: the response queue that received the packet
844 * @rsp: the response queue descriptor holding the offload message
845 * @gl: the gather list of packet fragments
847 * Deliver an ingress offload packet to a ULD. All processing is done by
848 * the ULD, we just maintain statistics.
850 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
851 const struct pkt_gl *gl)
853 struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
855 /* FW can send CPLs encapsulated in a CPL_FW4_MSG.
857 if (((const struct rss_header *)rsp)->opcode == CPL_FW4_MSG &&
858 ((const struct cpl_fw4_msg *)(rsp + 1))->type == FW_TYPE_RSSCPL)
861 if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
867 else if (gl == CXGB4_MSG_AN)
874 static void disable_msi(struct adapter *adapter)
876 if (adapter->flags & USING_MSIX) {
877 pci_disable_msix(adapter->pdev);
878 adapter->flags &= ~USING_MSIX;
879 } else if (adapter->flags & USING_MSI) {
880 pci_disable_msi(adapter->pdev);
881 adapter->flags &= ~USING_MSI;
886 * Interrupt handler for non-data events used with MSI-X.
888 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
890 struct adapter *adap = cookie;
892 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
895 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
897 t4_slow_intr_handler(adap);
902 * Name the MSI-X interrupts.
904 static void name_msix_vecs(struct adapter *adap)
906 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
908 /* non-data interrupts */
909 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
912 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
913 adap->port[0]->name);
915 /* Ethernet queues */
916 for_each_port(adap, j) {
917 struct net_device *d = adap->port[j];
918 const struct port_info *pi = netdev_priv(d);
920 for (i = 0; i < pi->nqsets; i++, msi_idx++)
921 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
926 for_each_ofldrxq(&adap->sge, i)
927 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
928 adap->port[0]->name, i);
930 for_each_rdmarxq(&adap->sge, i)
931 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
932 adap->port[0]->name, i);
934 for_each_rdmaciq(&adap->sge, i)
935 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma-ciq%d",
936 adap->port[0]->name, i);
939 static int request_msix_queue_irqs(struct adapter *adap)
941 struct sge *s = &adap->sge;
942 int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, rdmaciqqidx = 0;
945 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
946 adap->msix_info[1].desc, &s->fw_evtq);
950 for_each_ethrxq(s, ethqidx) {
951 err = request_irq(adap->msix_info[msi_index].vec,
953 adap->msix_info[msi_index].desc,
954 &s->ethrxq[ethqidx].rspq);
959 for_each_ofldrxq(s, ofldqidx) {
960 err = request_irq(adap->msix_info[msi_index].vec,
962 adap->msix_info[msi_index].desc,
963 &s->ofldrxq[ofldqidx].rspq);
968 for_each_rdmarxq(s, rdmaqidx) {
969 err = request_irq(adap->msix_info[msi_index].vec,
971 adap->msix_info[msi_index].desc,
972 &s->rdmarxq[rdmaqidx].rspq);
977 for_each_rdmaciq(s, rdmaciqqidx) {
978 err = request_irq(adap->msix_info[msi_index].vec,
980 adap->msix_info[msi_index].desc,
981 &s->rdmaciq[rdmaciqqidx].rspq);
989 while (--rdmaciqqidx >= 0)
990 free_irq(adap->msix_info[--msi_index].vec,
991 &s->rdmaciq[rdmaciqqidx].rspq);
992 while (--rdmaqidx >= 0)
993 free_irq(adap->msix_info[--msi_index].vec,
994 &s->rdmarxq[rdmaqidx].rspq);
995 while (--ofldqidx >= 0)
996 free_irq(adap->msix_info[--msi_index].vec,
997 &s->ofldrxq[ofldqidx].rspq);
998 while (--ethqidx >= 0)
999 free_irq(adap->msix_info[--msi_index].vec,
1000 &s->ethrxq[ethqidx].rspq);
1001 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
1005 static void free_msix_queue_irqs(struct adapter *adap)
1007 int i, msi_index = 2;
1008 struct sge *s = &adap->sge;
1010 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
1011 for_each_ethrxq(s, i)
1012 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
1013 for_each_ofldrxq(s, i)
1014 free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
1015 for_each_rdmarxq(s, i)
1016 free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
1017 for_each_rdmaciq(s, i)
1018 free_irq(adap->msix_info[msi_index++].vec, &s->rdmaciq[i].rspq);
1022 * write_rss - write the RSS table for a given port
1024 * @queues: array of queue indices for RSS
1026 * Sets up the portion of the HW RSS table for the port's VI to distribute
1027 * packets to the Rx queues in @queues.
1029 static int write_rss(const struct port_info *pi, const u16 *queues)
1033 const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
1035 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
1039 /* map the queue indices to queue ids */
1040 for (i = 0; i < pi->rss_size; i++, queues++)
1041 rss[i] = q[*queues].rspq.abs_id;
1043 err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
1044 pi->rss_size, rss, pi->rss_size);
1050 * setup_rss - configure RSS
1051 * @adap: the adapter
1053 * Sets up RSS for each port.
1055 static int setup_rss(struct adapter *adap)
1059 for_each_port(adap, i) {
1060 const struct port_info *pi = adap2pinfo(adap, i);
1062 err = write_rss(pi, pi->rss);
1070 * Return the channel of the ingress queue with the given qid.
1072 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
1074 qid -= p->ingr_start;
1075 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
1079 * Wait until all NAPI handlers are descheduled.
1081 static void quiesce_rx(struct adapter *adap)
1085 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
1086 struct sge_rspq *q = adap->sge.ingr_map[i];
1088 if (q && q->handler)
1089 napi_disable(&q->napi);
1094 * Enable NAPI scheduling and interrupt generation for all Rx queues.
1096 static void enable_rx(struct adapter *adap)
1100 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
1101 struct sge_rspq *q = adap->sge.ingr_map[i];
1106 napi_enable(&q->napi);
1107 /* 0-increment GTS to start the timer and enable interrupts */
1108 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
1109 SEINTARM(q->intr_params) |
1110 INGRESSQID(q->cntxt_id));
1115 * setup_sge_queues - configure SGE Tx/Rx/response queues
1116 * @adap: the adapter
1118 * Determines how many sets of SGE queues to use and initializes them.
1119 * We support multiple queue sets per port if we have MSI-X, otherwise
1120 * just one queue set per port.
1122 static int setup_sge_queues(struct adapter *adap)
1124 int err, msi_idx, i, j;
1125 struct sge *s = &adap->sge;
1127 bitmap_zero(s->starving_fl, MAX_EGRQ);
1128 bitmap_zero(s->txq_maperr, MAX_EGRQ);
1130 if (adap->flags & USING_MSIX)
1131 msi_idx = 1; /* vector 0 is for non-queue interrupts */
1133 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
1137 msi_idx = -((int)s->intrq.abs_id + 1);
1140 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
1141 msi_idx, NULL, fwevtq_handler);
1143 freeout: t4_free_sge_resources(adap);
1147 for_each_port(adap, i) {
1148 struct net_device *dev = adap->port[i];
1149 struct port_info *pi = netdev_priv(dev);
1150 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
1151 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
1153 for (j = 0; j < pi->nqsets; j++, q++) {
1156 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
1162 memset(&q->stats, 0, sizeof(q->stats));
1164 for (j = 0; j < pi->nqsets; j++, t++) {
1165 err = t4_sge_alloc_eth_txq(adap, t, dev,
1166 netdev_get_tx_queue(dev, j),
1167 s->fw_evtq.cntxt_id);
1173 j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
1174 for_each_ofldrxq(s, i) {
1175 struct sge_ofld_rxq *q = &s->ofldrxq[i];
1176 struct net_device *dev = adap->port[i / j];
1180 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
1181 q->fl.size ? &q->fl : NULL,
1185 memset(&q->stats, 0, sizeof(q->stats));
1186 s->ofld_rxq[i] = q->rspq.abs_id;
1187 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
1188 s->fw_evtq.cntxt_id);
1193 for_each_rdmarxq(s, i) {
1194 struct sge_ofld_rxq *q = &s->rdmarxq[i];
1198 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
1199 msi_idx, q->fl.size ? &q->fl : NULL,
1203 memset(&q->stats, 0, sizeof(q->stats));
1204 s->rdma_rxq[i] = q->rspq.abs_id;
1207 for_each_rdmaciq(s, i) {
1208 struct sge_ofld_rxq *q = &s->rdmaciq[i];
1212 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
1213 msi_idx, q->fl.size ? &q->fl : NULL,
1217 memset(&q->stats, 0, sizeof(q->stats));
1218 s->rdma_ciq[i] = q->rspq.abs_id;
1221 for_each_port(adap, i) {
1223 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
1224 * have RDMA queues, and that's the right value.
1226 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
1227 s->fw_evtq.cntxt_id,
1228 s->rdmarxq[i].rspq.cntxt_id);
1233 t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
1234 RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
1235 QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
1240 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1241 * The allocated memory is cleared.
1243 void *t4_alloc_mem(size_t size)
1245 void *p = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1253 * Free memory allocated through alloc_mem().
1255 static void t4_free_mem(void *addr)
1257 if (is_vmalloc_addr(addr))
1263 /* Send a Work Request to write the filter at a specified index. We construct
1264 * a Firmware Filter Work Request to have the work done and put the indicated
1265 * filter into "pending" mode which will prevent any further actions against
1266 * it till we get a reply from the firmware on the completion status of the
1269 static int set_filter_wr(struct adapter *adapter, int fidx)
1271 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1272 struct sk_buff *skb;
1273 struct fw_filter_wr *fwr;
1276 /* If the new filter requires loopback Destination MAC and/or VLAN
1277 * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
1280 if (f->fs.newdmac || f->fs.newvlan) {
1281 /* allocate L2T entry for new filter */
1282 f->l2t = t4_l2t_alloc_switching(adapter->l2t);
1285 if (t4_l2t_set_switching(adapter, f->l2t, f->fs.vlan,
1286 f->fs.eport, f->fs.dmac)) {
1287 cxgb4_l2t_release(f->l2t);
1293 ftid = adapter->tids.ftid_base + fidx;
1295 skb = alloc_skb(sizeof(*fwr), GFP_KERNEL | __GFP_NOFAIL);
1296 fwr = (struct fw_filter_wr *)__skb_put(skb, sizeof(*fwr));
1297 memset(fwr, 0, sizeof(*fwr));
1299 /* It would be nice to put most of the following in t4_hw.c but most
1300 * of the work is translating the cxgbtool ch_filter_specification
1301 * into the Work Request and the definition of that structure is
1302 * currently in cxgbtool.h which isn't appropriate to pull into the
1303 * common code. We may eventually try to come up with a more neutral
1304 * filter specification structure but for now it's easiest to simply
1305 * put this fairly direct code in line ...
1307 fwr->op_pkd = htonl(FW_WR_OP(FW_FILTER_WR));
1308 fwr->len16_pkd = htonl(FW_WR_LEN16(sizeof(*fwr)/16));
1310 htonl(V_FW_FILTER_WR_TID(ftid) |
1311 V_FW_FILTER_WR_RQTYPE(f->fs.type) |
1312 V_FW_FILTER_WR_NOREPLY(0) |
1313 V_FW_FILTER_WR_IQ(f->fs.iq));
1314 fwr->del_filter_to_l2tix =
1315 htonl(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
1316 V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
1317 V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
1318 V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
1319 V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
1320 V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
1321 V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
1322 V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
1323 V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
1324 f->fs.newvlan == VLAN_REWRITE) |
1325 V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
1326 f->fs.newvlan == VLAN_REWRITE) |
1327 V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
1328 V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
1329 V_FW_FILTER_WR_PRIO(f->fs.prio) |
1330 V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
1331 fwr->ethtype = htons(f->fs.val.ethtype);
1332 fwr->ethtypem = htons(f->fs.mask.ethtype);
1333 fwr->frag_to_ovlan_vldm =
1334 (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
1335 V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
1336 V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.ivlan_vld) |
1337 V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.ovlan_vld) |
1338 V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.ivlan_vld) |
1339 V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.ovlan_vld));
1341 fwr->rx_chan_rx_rpl_iq =
1342 htons(V_FW_FILTER_WR_RX_CHAN(0) |
1343 V_FW_FILTER_WR_RX_RPL_IQ(adapter->sge.fw_evtq.abs_id));
1344 fwr->maci_to_matchtypem =
1345 htonl(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
1346 V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
1347 V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
1348 V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
1349 V_FW_FILTER_WR_PORT(f->fs.val.iport) |
1350 V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
1351 V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
1352 V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
1353 fwr->ptcl = f->fs.val.proto;
1354 fwr->ptclm = f->fs.mask.proto;
1355 fwr->ttyp = f->fs.val.tos;
1356 fwr->ttypm = f->fs.mask.tos;
1357 fwr->ivlan = htons(f->fs.val.ivlan);
1358 fwr->ivlanm = htons(f->fs.mask.ivlan);
1359 fwr->ovlan = htons(f->fs.val.ovlan);
1360 fwr->ovlanm = htons(f->fs.mask.ovlan);
1361 memcpy(fwr->lip, f->fs.val.lip, sizeof(fwr->lip));
1362 memcpy(fwr->lipm, f->fs.mask.lip, sizeof(fwr->lipm));
1363 memcpy(fwr->fip, f->fs.val.fip, sizeof(fwr->fip));
1364 memcpy(fwr->fipm, f->fs.mask.fip, sizeof(fwr->fipm));
1365 fwr->lp = htons(f->fs.val.lport);
1366 fwr->lpm = htons(f->fs.mask.lport);
1367 fwr->fp = htons(f->fs.val.fport);
1368 fwr->fpm = htons(f->fs.mask.fport);
1370 memcpy(fwr->sma, f->fs.smac, sizeof(fwr->sma));
1372 /* Mark the filter as "pending" and ship off the Filter Work Request.
1373 * When we get the Work Request Reply we'll clear the pending status.
1376 set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
1377 t4_ofld_send(adapter, skb);
1381 /* Delete the filter at a specified index.
1383 static int del_filter_wr(struct adapter *adapter, int fidx)
1385 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1386 struct sk_buff *skb;
1387 struct fw_filter_wr *fwr;
1388 unsigned int len, ftid;
1391 ftid = adapter->tids.ftid_base + fidx;
1393 skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
1394 fwr = (struct fw_filter_wr *)__skb_put(skb, len);
1395 t4_mk_filtdelwr(ftid, fwr, adapter->sge.fw_evtq.abs_id);
1397 /* Mark the filter as "pending" and ship off the Filter Work Request.
1398 * When we get the Work Request Reply we'll clear the pending status.
1401 t4_mgmt_tx(adapter, skb);
1405 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
1406 void *accel_priv, select_queue_fallback_t fallback)
1410 #ifdef CONFIG_CHELSIO_T4_DCB
1411 /* If a Data Center Bridging has been successfully negotiated on this
1412 * link then we'll use the skb's priority to map it to a TX Queue.
1413 * The skb's priority is determined via the VLAN Tag Priority Code
1416 if (cxgb4_dcb_enabled(dev)) {
1420 err = vlan_get_tag(skb, &vlan_tci);
1421 if (unlikely(err)) {
1422 if (net_ratelimit())
1424 "TX Packet without VLAN Tag on DCB Link\n");
1427 txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1431 #endif /* CONFIG_CHELSIO_T4_DCB */
1434 txq = (skb_rx_queue_recorded(skb)
1435 ? skb_get_rx_queue(skb)
1436 : smp_processor_id());
1438 while (unlikely(txq >= dev->real_num_tx_queues))
1439 txq -= dev->real_num_tx_queues;
1444 return fallback(dev, skb) % dev->real_num_tx_queues;
1447 static inline int is_offload(const struct adapter *adap)
1449 return adap->params.offload;
1453 * Implementation of ethtool operations.
1456 static u32 get_msglevel(struct net_device *dev)
1458 return netdev2adap(dev)->msg_enable;
1461 static void set_msglevel(struct net_device *dev, u32 val)
1463 netdev2adap(dev)->msg_enable = val;
1466 static char stats_strings[][ETH_GSTRING_LEN] = {
1469 "TxBroadcastFrames ",
1470 "TxMulticastFrames ",
1476 "TxFrames128To255 ",
1477 "TxFrames256To511 ",
1478 "TxFrames512To1023 ",
1479 "TxFrames1024To1518 ",
1480 "TxFrames1519ToMax ",
1495 "RxBroadcastFrames ",
1496 "RxMulticastFrames ",
1508 "RxFrames128To255 ",
1509 "RxFrames256To511 ",
1510 "RxFrames512To1023 ",
1511 "RxFrames1024To1518 ",
1512 "RxFrames1519ToMax ",
1524 "RxBG0FramesDropped ",
1525 "RxBG1FramesDropped ",
1526 "RxBG2FramesDropped ",
1527 "RxBG3FramesDropped ",
1528 "RxBG0FramesTrunc ",
1529 "RxBG1FramesTrunc ",
1530 "RxBG2FramesTrunc ",
1531 "RxBG3FramesTrunc ",
1540 "WriteCoalSuccess ",
1544 static int get_sset_count(struct net_device *dev, int sset)
1548 return ARRAY_SIZE(stats_strings);
1554 #define T4_REGMAP_SIZE (160 * 1024)
1555 #define T5_REGMAP_SIZE (332 * 1024)
1557 static int get_regs_len(struct net_device *dev)
1559 struct adapter *adap = netdev2adap(dev);
1560 if (is_t4(adap->params.chip))
1561 return T4_REGMAP_SIZE;
1563 return T5_REGMAP_SIZE;
1566 static int get_eeprom_len(struct net_device *dev)
1571 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1573 struct adapter *adapter = netdev2adap(dev);
1575 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1576 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1577 strlcpy(info->bus_info, pci_name(adapter->pdev),
1578 sizeof(info->bus_info));
1580 if (adapter->params.fw_vers)
1581 snprintf(info->fw_version, sizeof(info->fw_version),
1582 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1583 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1584 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1585 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1586 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1587 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1588 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1589 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1590 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1593 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1595 if (stringset == ETH_SS_STATS)
1596 memcpy(data, stats_strings, sizeof(stats_strings));
1600 * port stats maintained per queue of the port. They should be in the same
1601 * order as in stats_strings above.
1603 struct queue_port_stats {
1613 static void collect_sge_port_stats(const struct adapter *adap,
1614 const struct port_info *p, struct queue_port_stats *s)
1617 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1618 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1620 memset(s, 0, sizeof(*s));
1621 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1623 s->tx_csum += tx->tx_cso;
1624 s->rx_csum += rx->stats.rx_cso;
1625 s->vlan_ex += rx->stats.vlan_ex;
1626 s->vlan_ins += tx->vlan_ins;
1627 s->gro_pkts += rx->stats.lro_pkts;
1628 s->gro_merged += rx->stats.lro_merged;
1632 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1635 struct port_info *pi = netdev_priv(dev);
1636 struct adapter *adapter = pi->adapter;
1639 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1641 data += sizeof(struct port_stats) / sizeof(u64);
1642 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1643 data += sizeof(struct queue_port_stats) / sizeof(u64);
1644 if (!is_t4(adapter->params.chip)) {
1645 t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7));
1646 val1 = t4_read_reg(adapter, SGE_STAT_TOTAL);
1647 val2 = t4_read_reg(adapter, SGE_STAT_MATCH);
1648 *data = val1 - val2;
1653 memset(data, 0, 2 * sizeof(u64));
1659 * Return a version number to identify the type of adapter. The scheme is:
1660 * - bits 0..9: chip version
1661 * - bits 10..15: chip revision
1662 * - bits 16..23: register dump version
1664 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1666 return CHELSIO_CHIP_VERSION(ap->params.chip) |
1667 (CHELSIO_CHIP_RELEASE(ap->params.chip) << 10) | (1 << 16);
1670 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1673 u32 *p = buf + start;
1675 for ( ; start <= end; start += sizeof(u32))
1676 *p++ = t4_read_reg(ap, start);
1679 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1682 static const unsigned int t4_reg_ranges[] = {
1902 static const unsigned int t5_reg_ranges[] = {
2330 struct adapter *ap = netdev2adap(dev);
2331 static const unsigned int *reg_ranges;
2332 int arr_size = 0, buf_size = 0;
2334 if (is_t4(ap->params.chip)) {
2335 reg_ranges = &t4_reg_ranges[0];
2336 arr_size = ARRAY_SIZE(t4_reg_ranges);
2337 buf_size = T4_REGMAP_SIZE;
2339 reg_ranges = &t5_reg_ranges[0];
2340 arr_size = ARRAY_SIZE(t5_reg_ranges);
2341 buf_size = T5_REGMAP_SIZE;
2344 regs->version = mk_adap_vers(ap);
2346 memset(buf, 0, buf_size);
2347 for (i = 0; i < arr_size; i += 2)
2348 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
2351 static int restart_autoneg(struct net_device *dev)
2353 struct port_info *p = netdev_priv(dev);
2355 if (!netif_running(dev))
2357 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
2359 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
2363 static int identify_port(struct net_device *dev,
2364 enum ethtool_phys_id_state state)
2367 struct adapter *adap = netdev2adap(dev);
2369 if (state == ETHTOOL_ID_ACTIVE)
2371 else if (state == ETHTOOL_ID_INACTIVE)
2376 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
2379 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
2383 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
2384 type == FW_PORT_TYPE_BT_XAUI) {
2386 if (caps & FW_PORT_CAP_SPEED_100M)
2387 v |= SUPPORTED_100baseT_Full;
2388 if (caps & FW_PORT_CAP_SPEED_1G)
2389 v |= SUPPORTED_1000baseT_Full;
2390 if (caps & FW_PORT_CAP_SPEED_10G)
2391 v |= SUPPORTED_10000baseT_Full;
2392 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
2393 v |= SUPPORTED_Backplane;
2394 if (caps & FW_PORT_CAP_SPEED_1G)
2395 v |= SUPPORTED_1000baseKX_Full;
2396 if (caps & FW_PORT_CAP_SPEED_10G)
2397 v |= SUPPORTED_10000baseKX4_Full;
2398 } else if (type == FW_PORT_TYPE_KR)
2399 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
2400 else if (type == FW_PORT_TYPE_BP_AP)
2401 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2402 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
2403 else if (type == FW_PORT_TYPE_BP4_AP)
2404 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2405 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
2406 SUPPORTED_10000baseKX4_Full;
2407 else if (type == FW_PORT_TYPE_FIBER_XFI ||
2408 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
2409 v |= SUPPORTED_FIBRE;
2410 else if (type == FW_PORT_TYPE_BP40_BA)
2411 v |= SUPPORTED_40000baseSR4_Full;
2413 if (caps & FW_PORT_CAP_ANEG)
2414 v |= SUPPORTED_Autoneg;
2418 static unsigned int to_fw_linkcaps(unsigned int caps)
2422 if (caps & ADVERTISED_100baseT_Full)
2423 v |= FW_PORT_CAP_SPEED_100M;
2424 if (caps & ADVERTISED_1000baseT_Full)
2425 v |= FW_PORT_CAP_SPEED_1G;
2426 if (caps & ADVERTISED_10000baseT_Full)
2427 v |= FW_PORT_CAP_SPEED_10G;
2428 if (caps & ADVERTISED_40000baseSR4_Full)
2429 v |= FW_PORT_CAP_SPEED_40G;
2433 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2435 const struct port_info *p = netdev_priv(dev);
2437 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
2438 p->port_type == FW_PORT_TYPE_BT_XFI ||
2439 p->port_type == FW_PORT_TYPE_BT_XAUI)
2440 cmd->port = PORT_TP;
2441 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
2442 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
2443 cmd->port = PORT_FIBRE;
2444 else if (p->port_type == FW_PORT_TYPE_SFP ||
2445 p->port_type == FW_PORT_TYPE_QSFP_10G ||
2446 p->port_type == FW_PORT_TYPE_QSFP) {
2447 if (p->mod_type == FW_PORT_MOD_TYPE_LR ||
2448 p->mod_type == FW_PORT_MOD_TYPE_SR ||
2449 p->mod_type == FW_PORT_MOD_TYPE_ER ||
2450 p->mod_type == FW_PORT_MOD_TYPE_LRM)
2451 cmd->port = PORT_FIBRE;
2452 else if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
2453 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
2454 cmd->port = PORT_DA;
2456 cmd->port = PORT_OTHER;
2458 cmd->port = PORT_OTHER;
2460 if (p->mdio_addr >= 0) {
2461 cmd->phy_address = p->mdio_addr;
2462 cmd->transceiver = XCVR_EXTERNAL;
2463 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
2464 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
2466 cmd->phy_address = 0; /* not really, but no better option */
2467 cmd->transceiver = XCVR_INTERNAL;
2468 cmd->mdio_support = 0;
2471 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
2472 cmd->advertising = from_fw_linkcaps(p->port_type,
2473 p->link_cfg.advertising);
2474 ethtool_cmd_speed_set(cmd,
2475 netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
2476 cmd->duplex = DUPLEX_FULL;
2477 cmd->autoneg = p->link_cfg.autoneg;
2483 static unsigned int speed_to_caps(int speed)
2486 return FW_PORT_CAP_SPEED_100M;
2488 return FW_PORT_CAP_SPEED_1G;
2490 return FW_PORT_CAP_SPEED_10G;
2492 return FW_PORT_CAP_SPEED_40G;
2496 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2499 struct port_info *p = netdev_priv(dev);
2500 struct link_config *lc = &p->link_cfg;
2501 u32 speed = ethtool_cmd_speed(cmd);
2503 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
2506 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
2508 * PHY offers a single speed. See if that's what's
2511 if (cmd->autoneg == AUTONEG_DISABLE &&
2512 (lc->supported & speed_to_caps(speed)))
2517 if (cmd->autoneg == AUTONEG_DISABLE) {
2518 cap = speed_to_caps(speed);
2520 if (!(lc->supported & cap) ||
2525 lc->requested_speed = cap;
2526 lc->advertising = 0;
2528 cap = to_fw_linkcaps(cmd->advertising);
2529 if (!(lc->supported & cap))
2531 lc->requested_speed = 0;
2532 lc->advertising = cap | FW_PORT_CAP_ANEG;
2534 lc->autoneg = cmd->autoneg;
2536 if (netif_running(dev))
2537 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2542 static void get_pauseparam(struct net_device *dev,
2543 struct ethtool_pauseparam *epause)
2545 struct port_info *p = netdev_priv(dev);
2547 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
2548 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
2549 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
2552 static int set_pauseparam(struct net_device *dev,
2553 struct ethtool_pauseparam *epause)
2555 struct port_info *p = netdev_priv(dev);
2556 struct link_config *lc = &p->link_cfg;
2558 if (epause->autoneg == AUTONEG_DISABLE)
2559 lc->requested_fc = 0;
2560 else if (lc->supported & FW_PORT_CAP_ANEG)
2561 lc->requested_fc = PAUSE_AUTONEG;
2565 if (epause->rx_pause)
2566 lc->requested_fc |= PAUSE_RX;
2567 if (epause->tx_pause)
2568 lc->requested_fc |= PAUSE_TX;
2569 if (netif_running(dev))
2570 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2575 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2577 const struct port_info *pi = netdev_priv(dev);
2578 const struct sge *s = &pi->adapter->sge;
2580 e->rx_max_pending = MAX_RX_BUFFERS;
2581 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
2582 e->rx_jumbo_max_pending = 0;
2583 e->tx_max_pending = MAX_TXQ_ENTRIES;
2585 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
2586 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
2587 e->rx_jumbo_pending = 0;
2588 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
2591 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2594 const struct port_info *pi = netdev_priv(dev);
2595 struct adapter *adapter = pi->adapter;
2596 struct sge *s = &adapter->sge;
2598 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
2599 e->tx_pending > MAX_TXQ_ENTRIES ||
2600 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
2601 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
2602 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
2605 if (adapter->flags & FULL_INIT_DONE)
2608 for (i = 0; i < pi->nqsets; ++i) {
2609 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
2610 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
2611 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
2616 static int closest_timer(const struct sge *s, int time)
2618 int i, delta, match = 0, min_delta = INT_MAX;
2620 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
2621 delta = time - s->timer_val[i];
2624 if (delta < min_delta) {
2632 static int closest_thres(const struct sge *s, int thres)
2634 int i, delta, match = 0, min_delta = INT_MAX;
2636 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
2637 delta = thres - s->counter_val[i];
2640 if (delta < min_delta) {
2649 * Return a queue's interrupt hold-off time in us. 0 means no timer.
2651 static unsigned int qtimer_val(const struct adapter *adap,
2652 const struct sge_rspq *q)
2654 unsigned int idx = q->intr_params >> 1;
2656 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
2660 * set_rspq_intr_params - set a queue's interrupt holdoff parameters
2662 * @us: the hold-off time in us, or 0 to disable timer
2663 * @cnt: the hold-off packet count, or 0 to disable counter
2665 * Sets an Rx queue's interrupt hold-off time and packet count. At least
2666 * one of the two needs to be enabled for the queue to generate interrupts.
2668 static int set_rspq_intr_params(struct sge_rspq *q,
2669 unsigned int us, unsigned int cnt)
2671 struct adapter *adap = q->adap;
2673 if ((us | cnt) == 0)
2680 new_idx = closest_thres(&adap->sge, cnt);
2681 if (q->desc && q->pktcnt_idx != new_idx) {
2682 /* the queue has already been created, update it */
2683 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
2684 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
2685 FW_PARAMS_PARAM_YZ(q->cntxt_id);
2686 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
2691 q->pktcnt_idx = new_idx;
2694 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
2695 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
2700 * set_rx_intr_params - set a net devices's RX interrupt holdoff paramete!
2701 * @dev: the network device
2702 * @us: the hold-off time in us, or 0 to disable timer
2703 * @cnt: the hold-off packet count, or 0 to disable counter
2705 * Set the RX interrupt hold-off parameters for a network device.
2707 static int set_rx_intr_params(struct net_device *dev,
2708 unsigned int us, unsigned int cnt)
2711 struct port_info *pi = netdev_priv(dev);
2712 struct adapter *adap = pi->adapter;
2713 struct sge_eth_rxq *q = &adap->sge.ethrxq[pi->first_qset];
2715 for (i = 0; i < pi->nqsets; i++, q++) {
2716 err = set_rspq_intr_params(&q->rspq, us, cnt);
2723 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2725 return set_rx_intr_params(dev, c->rx_coalesce_usecs,
2726 c->rx_max_coalesced_frames);
2729 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2731 const struct port_info *pi = netdev_priv(dev);
2732 const struct adapter *adap = pi->adapter;
2733 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
2735 c->rx_coalesce_usecs = qtimer_val(adap, rq);
2736 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
2737 adap->sge.counter_val[rq->pktcnt_idx] : 0;
2742 * eeprom_ptov - translate a physical EEPROM address to virtual
2743 * @phys_addr: the physical EEPROM address
2744 * @fn: the PCI function number
2745 * @sz: size of function-specific area
2747 * Translate a physical EEPROM address to virtual. The first 1K is
2748 * accessed through virtual addresses starting at 31K, the rest is
2749 * accessed through virtual addresses starting at 0.
2751 * The mapping is as follows:
2752 * [0..1K) -> [31K..32K)
2753 * [1K..1K+A) -> [31K-A..31K)
2754 * [1K+A..ES) -> [0..ES-A-1K)
2756 * where A = @fn * @sz, and ES = EEPROM size.
2758 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
2761 if (phys_addr < 1024)
2762 return phys_addr + (31 << 10);
2763 if (phys_addr < 1024 + fn)
2764 return 31744 - fn + phys_addr - 1024;
2765 if (phys_addr < EEPROMSIZE)
2766 return phys_addr - 1024 - fn;
2771 * The next two routines implement eeprom read/write from physical addresses.
2773 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
2775 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2778 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
2779 return vaddr < 0 ? vaddr : 0;
2782 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
2784 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2787 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
2788 return vaddr < 0 ? vaddr : 0;
2791 #define EEPROM_MAGIC 0x38E2F10C
2793 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2797 struct adapter *adapter = netdev2adap(dev);
2799 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2803 e->magic = EEPROM_MAGIC;
2804 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2805 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
2808 memcpy(data, buf + e->offset, e->len);
2813 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2818 u32 aligned_offset, aligned_len, *p;
2819 struct adapter *adapter = netdev2adap(dev);
2821 if (eeprom->magic != EEPROM_MAGIC)
2824 aligned_offset = eeprom->offset & ~3;
2825 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2827 if (adapter->fn > 0) {
2828 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
2830 if (aligned_offset < start ||
2831 aligned_offset + aligned_len > start + EEPROMPFSIZE)
2835 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2837 * RMW possibly needed for first or last words.
2839 buf = kmalloc(aligned_len, GFP_KERNEL);
2842 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
2843 if (!err && aligned_len > 4)
2844 err = eeprom_rd_phys(adapter,
2845 aligned_offset + aligned_len - 4,
2846 (u32 *)&buf[aligned_len - 4]);
2849 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2853 err = t4_seeprom_wp(adapter, false);
2857 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2858 err = eeprom_wr_phys(adapter, aligned_offset, *p);
2859 aligned_offset += 4;
2863 err = t4_seeprom_wp(adapter, true);
2870 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
2873 const struct firmware *fw;
2874 struct adapter *adap = netdev2adap(netdev);
2876 ef->data[sizeof(ef->data) - 1] = '\0';
2877 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
2881 ret = t4_load_fw(adap, fw->data, fw->size);
2882 release_firmware(fw);
2884 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
2888 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2889 #define BCAST_CRC 0xa0ccc1a6
2891 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2893 wol->supported = WAKE_BCAST | WAKE_MAGIC;
2894 wol->wolopts = netdev2adap(dev)->wol;
2895 memset(&wol->sopass, 0, sizeof(wol->sopass));
2898 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2901 struct port_info *pi = netdev_priv(dev);
2903 if (wol->wolopts & ~WOL_SUPPORTED)
2905 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
2906 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
2907 if (wol->wolopts & WAKE_BCAST) {
2908 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
2911 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
2912 ~6ULL, ~0ULL, BCAST_CRC, true);
2914 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
2918 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2920 const struct port_info *pi = netdev_priv(dev);
2921 netdev_features_t changed = dev->features ^ features;
2924 if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
2927 err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
2929 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
2931 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
2935 static u32 get_rss_table_size(struct net_device *dev)
2937 const struct port_info *pi = netdev_priv(dev);
2939 return pi->rss_size;
2942 static int get_rss_table(struct net_device *dev, u32 *p, u8 *key)
2944 const struct port_info *pi = netdev_priv(dev);
2945 unsigned int n = pi->rss_size;
2952 static int set_rss_table(struct net_device *dev, const u32 *p, const u8 *key)
2955 struct port_info *pi = netdev_priv(dev);
2957 for (i = 0; i < pi->rss_size; i++)
2959 if (pi->adapter->flags & FULL_INIT_DONE)
2960 return write_rss(pi, pi->rss);
2964 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
2967 const struct port_info *pi = netdev_priv(dev);
2969 switch (info->cmd) {
2970 case ETHTOOL_GRXFH: {
2971 unsigned int v = pi->rss_mode;
2974 switch (info->flow_type) {
2976 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
2977 info->data = RXH_IP_SRC | RXH_IP_DST |
2978 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2979 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2980 info->data = RXH_IP_SRC | RXH_IP_DST;
2983 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
2984 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2985 info->data = RXH_IP_SRC | RXH_IP_DST |
2986 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2987 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2988 info->data = RXH_IP_SRC | RXH_IP_DST;
2991 case AH_ESP_V4_FLOW:
2993 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2994 info->data = RXH_IP_SRC | RXH_IP_DST;
2997 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
2998 info->data = RXH_IP_SRC | RXH_IP_DST |
2999 RXH_L4_B_0_1 | RXH_L4_B_2_3;
3000 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
3001 info->data = RXH_IP_SRC | RXH_IP_DST;
3004 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
3005 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
3006 info->data = RXH_IP_SRC | RXH_IP_DST |
3007 RXH_L4_B_0_1 | RXH_L4_B_2_3;
3008 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
3009 info->data = RXH_IP_SRC | RXH_IP_DST;
3012 case AH_ESP_V6_FLOW:
3014 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
3015 info->data = RXH_IP_SRC | RXH_IP_DST;
3020 case ETHTOOL_GRXRINGS:
3021 info->data = pi->nqsets;
3027 static const struct ethtool_ops cxgb_ethtool_ops = {
3028 .get_settings = get_settings,
3029 .set_settings = set_settings,
3030 .get_drvinfo = get_drvinfo,
3031 .get_msglevel = get_msglevel,
3032 .set_msglevel = set_msglevel,
3033 .get_ringparam = get_sge_param,
3034 .set_ringparam = set_sge_param,
3035 .get_coalesce = get_coalesce,
3036 .set_coalesce = set_coalesce,
3037 .get_eeprom_len = get_eeprom_len,
3038 .get_eeprom = get_eeprom,
3039 .set_eeprom = set_eeprom,
3040 .get_pauseparam = get_pauseparam,
3041 .set_pauseparam = set_pauseparam,
3042 .get_link = ethtool_op_get_link,
3043 .get_strings = get_strings,
3044 .set_phys_id = identify_port,
3045 .nway_reset = restart_autoneg,
3046 .get_sset_count = get_sset_count,
3047 .get_ethtool_stats = get_stats,
3048 .get_regs_len = get_regs_len,
3049 .get_regs = get_regs,
3052 .get_rxnfc = get_rxnfc,
3053 .get_rxfh_indir_size = get_rss_table_size,
3054 .get_rxfh = get_rss_table,
3055 .set_rxfh = set_rss_table,
3056 .flash_device = set_flash,
3062 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
3066 loff_t avail = file_inode(file)->i_size;
3067 unsigned int mem = (uintptr_t)file->private_data & 3;
3068 struct adapter *adap = file->private_data - mem;
3074 if (count > avail - pos)
3075 count = avail - pos;
3082 if ((mem == MEM_MC) || (mem == MEM_MC1))
3083 ret = t4_mc_read(adap, mem % MEM_MC, pos, data, NULL);
3085 ret = t4_edc_read(adap, mem, pos, data, NULL);
3089 ofst = pos % sizeof(data);
3090 len = min(count, sizeof(data) - ofst);
3091 if (copy_to_user(buf, (u8 *)data + ofst, len))
3098 count = pos - *ppos;
3103 static const struct file_operations mem_debugfs_fops = {
3104 .owner = THIS_MODULE,
3105 .open = simple_open,
3107 .llseek = default_llseek,
3110 static void add_debugfs_mem(struct adapter *adap, const char *name,
3111 unsigned int idx, unsigned int size_mb)
3115 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
3116 (void *)adap + idx, &mem_debugfs_fops);
3117 if (de && de->d_inode)
3118 de->d_inode->i_size = size_mb << 20;
3121 static int setup_debugfs(struct adapter *adap)
3126 if (IS_ERR_OR_NULL(adap->debugfs_root))
3129 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
3130 if (i & EDRAM0_ENABLE) {
3131 size = t4_read_reg(adap, MA_EDRAM0_BAR);
3132 add_debugfs_mem(adap, "edc0", MEM_EDC0, EDRAM_SIZE_GET(size));
3134 if (i & EDRAM1_ENABLE) {
3135 size = t4_read_reg(adap, MA_EDRAM1_BAR);
3136 add_debugfs_mem(adap, "edc1", MEM_EDC1, EDRAM_SIZE_GET(size));
3138 if (is_t4(adap->params.chip)) {
3139 size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
3140 if (i & EXT_MEM_ENABLE)
3141 add_debugfs_mem(adap, "mc", MEM_MC,
3142 EXT_MEM_SIZE_GET(size));
3144 if (i & EXT_MEM_ENABLE) {
3145 size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
3146 add_debugfs_mem(adap, "mc0", MEM_MC0,
3147 EXT_MEM_SIZE_GET(size));
3149 if (i & EXT_MEM1_ENABLE) {
3150 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR);
3151 add_debugfs_mem(adap, "mc1", MEM_MC1,
3152 EXT_MEM_SIZE_GET(size));
3156 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
3162 * upper-layer driver support
3166 * Allocate an active-open TID and set it to the supplied value.
3168 int cxgb4_alloc_atid(struct tid_info *t, void *data)
3172 spin_lock_bh(&t->atid_lock);
3174 union aopen_entry *p = t->afree;
3176 atid = (p - t->atid_tab) + t->atid_base;
3181 spin_unlock_bh(&t->atid_lock);
3184 EXPORT_SYMBOL(cxgb4_alloc_atid);
3187 * Release an active-open TID.
3189 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
3191 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
3193 spin_lock_bh(&t->atid_lock);
3197 spin_unlock_bh(&t->atid_lock);
3199 EXPORT_SYMBOL(cxgb4_free_atid);
3202 * Allocate a server TID and set it to the supplied value.
3204 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
3208 spin_lock_bh(&t->stid_lock);
3209 if (family == PF_INET) {
3210 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
3211 if (stid < t->nstids)
3212 __set_bit(stid, t->stid_bmap);
3216 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
3221 t->stid_tab[stid].data = data;
3222 stid += t->stid_base;
3223 /* IPv6 requires max of 520 bits or 16 cells in TCAM
3224 * This is equivalent to 4 TIDs. With CLIP enabled it
3227 if (family == PF_INET)
3230 t->stids_in_use += 4;
3232 spin_unlock_bh(&t->stid_lock);
3235 EXPORT_SYMBOL(cxgb4_alloc_stid);
3237 /* Allocate a server filter TID and set it to the supplied value.
3239 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
3243 spin_lock_bh(&t->stid_lock);
3244 if (family == PF_INET) {
3245 stid = find_next_zero_bit(t->stid_bmap,
3246 t->nstids + t->nsftids, t->nstids);
3247 if (stid < (t->nstids + t->nsftids))
3248 __set_bit(stid, t->stid_bmap);
3255 t->stid_tab[stid].data = data;
3257 stid += t->sftid_base;
3260 spin_unlock_bh(&t->stid_lock);
3263 EXPORT_SYMBOL(cxgb4_alloc_sftid);
3265 /* Release a server TID.
3267 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
3269 /* Is it a server filter TID? */
3270 if (t->nsftids && (stid >= t->sftid_base)) {
3271 stid -= t->sftid_base;
3274 stid -= t->stid_base;
3277 spin_lock_bh(&t->stid_lock);
3278 if (family == PF_INET)
3279 __clear_bit(stid, t->stid_bmap);
3281 bitmap_release_region(t->stid_bmap, stid, 2);
3282 t->stid_tab[stid].data = NULL;
3283 if (family == PF_INET)
3286 t->stids_in_use -= 4;
3287 spin_unlock_bh(&t->stid_lock);
3289 EXPORT_SYMBOL(cxgb4_free_stid);
3292 * Populate a TID_RELEASE WR. Caller must properly size the skb.
3294 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
3297 struct cpl_tid_release *req;
3299 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
3300 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
3301 INIT_TP_WR(req, tid);
3302 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
3306 * Queue a TID release request and if necessary schedule a work queue to
3309 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
3312 void **p = &t->tid_tab[tid];
3313 struct adapter *adap = container_of(t, struct adapter, tids);
3315 spin_lock_bh(&adap->tid_release_lock);
3316 *p = adap->tid_release_head;
3317 /* Low 2 bits encode the Tx channel number */
3318 adap->tid_release_head = (void **)((uintptr_t)p | chan);
3319 if (!adap->tid_release_task_busy) {
3320 adap->tid_release_task_busy = true;
3321 queue_work(workq, &adap->tid_release_task);
3323 spin_unlock_bh(&adap->tid_release_lock);
3327 * Process the list of pending TID release requests.
3329 static void process_tid_release_list(struct work_struct *work)
3331 struct sk_buff *skb;
3332 struct adapter *adap;
3334 adap = container_of(work, struct adapter, tid_release_task);
3336 spin_lock_bh(&adap->tid_release_lock);
3337 while (adap->tid_release_head) {
3338 void **p = adap->tid_release_head;
3339 unsigned int chan = (uintptr_t)p & 3;
3340 p = (void *)p - chan;
3342 adap->tid_release_head = *p;
3344 spin_unlock_bh(&adap->tid_release_lock);
3346 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
3348 schedule_timeout_uninterruptible(1);
3350 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
3351 t4_ofld_send(adap, skb);
3352 spin_lock_bh(&adap->tid_release_lock);
3354 adap->tid_release_task_busy = false;
3355 spin_unlock_bh(&adap->tid_release_lock);
3359 * Release a TID and inform HW. If we are unable to allocate the release
3360 * message we defer to a work queue.
3362 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
3365 struct sk_buff *skb;
3366 struct adapter *adap = container_of(t, struct adapter, tids);
3368 old = t->tid_tab[tid];
3369 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
3371 t->tid_tab[tid] = NULL;
3372 mk_tid_release(skb, chan, tid);
3373 t4_ofld_send(adap, skb);
3375 cxgb4_queue_tid_release(t, chan, tid);
3377 atomic_dec(&t->tids_in_use);
3379 EXPORT_SYMBOL(cxgb4_remove_tid);
3382 * Allocate and initialize the TID tables. Returns 0 on success.
3384 static int tid_init(struct tid_info *t)
3387 unsigned int stid_bmap_size;
3388 unsigned int natids = t->natids;
3389 struct adapter *adap = container_of(t, struct adapter, tids);
3391 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
3392 size = t->ntids * sizeof(*t->tid_tab) +
3393 natids * sizeof(*t->atid_tab) +
3394 t->nstids * sizeof(*t->stid_tab) +
3395 t->nsftids * sizeof(*t->stid_tab) +
3396 stid_bmap_size * sizeof(long) +
3397 t->nftids * sizeof(*t->ftid_tab) +
3398 t->nsftids * sizeof(*t->ftid_tab);
3400 t->tid_tab = t4_alloc_mem(size);
3404 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
3405 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
3406 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
3407 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
3408 spin_lock_init(&t->stid_lock);
3409 spin_lock_init(&t->atid_lock);
3411 t->stids_in_use = 0;
3413 t->atids_in_use = 0;
3414 atomic_set(&t->tids_in_use, 0);
3416 /* Setup the free list for atid_tab and clear the stid bitmap. */
3419 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
3420 t->afree = t->atid_tab;
3422 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
3423 /* Reserve stid 0 for T4/T5 adapters */
3424 if (!t->stid_base &&
3425 (is_t4(adap->params.chip) || is_t5(adap->params.chip)))
3426 __set_bit(0, t->stid_bmap);
3431 static int cxgb4_clip_get(const struct net_device *dev,
3432 const struct in6_addr *lip)
3434 struct adapter *adap;
3435 struct fw_clip_cmd c;
3437 adap = netdev2adap(dev);
3438 memset(&c, 0, sizeof(c));
3439 c.op_to_write = htonl(FW_CMD_OP(FW_CLIP_CMD) |
3440 FW_CMD_REQUEST | FW_CMD_WRITE);
3441 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c));
3442 c.ip_hi = *(__be64 *)(lip->s6_addr);
3443 c.ip_lo = *(__be64 *)(lip->s6_addr + 8);
3444 return t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, false);
3447 static int cxgb4_clip_release(const struct net_device *dev,
3448 const struct in6_addr *lip)
3450 struct adapter *adap;
3451 struct fw_clip_cmd c;
3453 adap = netdev2adap(dev);
3454 memset(&c, 0, sizeof(c));
3455 c.op_to_write = htonl(FW_CMD_OP(FW_CLIP_CMD) |
3456 FW_CMD_REQUEST | FW_CMD_READ);
3457 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c));
3458 c.ip_hi = *(__be64 *)(lip->s6_addr);
3459 c.ip_lo = *(__be64 *)(lip->s6_addr + 8);
3460 return t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, false);
3464 * cxgb4_create_server - create an IP server
3466 * @stid: the server TID
3467 * @sip: local IP address to bind server to
3468 * @sport: the server's TCP port
3469 * @queue: queue to direct messages from this server to
3471 * Create an IP server for the given port and address.
3472 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3474 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
3475 __be32 sip, __be16 sport, __be16 vlan,
3479 struct sk_buff *skb;
3480 struct adapter *adap;
3481 struct cpl_pass_open_req *req;
3484 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3488 adap = netdev2adap(dev);
3489 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
3491 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
3492 req->local_port = sport;
3493 req->peer_port = htons(0);
3494 req->local_ip = sip;
3495 req->peer_ip = htonl(0);
3496 chan = rxq_to_chan(&adap->sge, queue);
3497 req->opt0 = cpu_to_be64(TX_CHAN(chan));
3498 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3499 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3500 ret = t4_mgmt_tx(adap, skb);
3501 return net_xmit_eval(ret);
3503 EXPORT_SYMBOL(cxgb4_create_server);
3505 /* cxgb4_create_server6 - create an IPv6 server
3507 * @stid: the server TID
3508 * @sip: local IPv6 address to bind server to
3509 * @sport: the server's TCP port
3510 * @queue: queue to direct messages from this server to
3512 * Create an IPv6 server for the given port and address.
3513 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3515 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
3516 const struct in6_addr *sip, __be16 sport,
3520 struct sk_buff *skb;
3521 struct adapter *adap;
3522 struct cpl_pass_open_req6 *req;
3525 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3529 adap = netdev2adap(dev);
3530 req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
3532 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
3533 req->local_port = sport;
3534 req->peer_port = htons(0);
3535 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
3536 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
3537 req->peer_ip_hi = cpu_to_be64(0);
3538 req->peer_ip_lo = cpu_to_be64(0);
3539 chan = rxq_to_chan(&adap->sge, queue);
3540 req->opt0 = cpu_to_be64(TX_CHAN(chan));
3541 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3542 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3543 ret = t4_mgmt_tx(adap, skb);
3544 return net_xmit_eval(ret);
3546 EXPORT_SYMBOL(cxgb4_create_server6);
3548 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
3549 unsigned int queue, bool ipv6)
3551 struct sk_buff *skb;
3552 struct adapter *adap;
3553 struct cpl_close_listsvr_req *req;
3556 adap = netdev2adap(dev);
3558 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3562 req = (struct cpl_close_listsvr_req *)__skb_put(skb, sizeof(*req));
3564 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
3565 req->reply_ctrl = htons(NO_REPLY(0) | (ipv6 ? LISTSVR_IPV6(1) :
3566 LISTSVR_IPV6(0)) | QUEUENO(queue));
3567 ret = t4_mgmt_tx(adap, skb);
3568 return net_xmit_eval(ret);
3570 EXPORT_SYMBOL(cxgb4_remove_server);
3573 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
3574 * @mtus: the HW MTU table
3575 * @mtu: the target MTU
3576 * @idx: index of selected entry in the MTU table
3578 * Returns the index and the value in the HW MTU table that is closest to
3579 * but does not exceed @mtu, unless @mtu is smaller than any value in the
3580 * table, in which case that smallest available value is selected.
3582 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
3587 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
3593 EXPORT_SYMBOL(cxgb4_best_mtu);
3596 * cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
3597 * @mtus: the HW MTU table
3598 * @header_size: Header Size
3599 * @data_size_max: maximum Data Segment Size
3600 * @data_size_align: desired Data Segment Size Alignment (2^N)
3601 * @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
3603 * Similar to cxgb4_best_mtu() but instead of searching the Hardware
3604 * MTU Table based solely on a Maximum MTU parameter, we break that
3605 * parameter up into a Header Size and Maximum Data Segment Size, and
3606 * provide a desired Data Segment Size Alignment. If we find an MTU in
3607 * the Hardware MTU Table which will result in a Data Segment Size with
3608 * the requested alignment _and_ that MTU isn't "too far" from the
3609 * closest MTU, then we'll return that rather than the closest MTU.
3611 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
3612 unsigned short header_size,
3613 unsigned short data_size_max,
3614 unsigned short data_size_align,
3615 unsigned int *mtu_idxp)
3617 unsigned short max_mtu = header_size + data_size_max;
3618 unsigned short data_size_align_mask = data_size_align - 1;
3619 int mtu_idx, aligned_mtu_idx;
3621 /* Scan the MTU Table till we find an MTU which is larger than our
3622 * Maximum MTU or we reach the end of the table. Along the way,
3623 * record the last MTU found, if any, which will result in a Data
3624 * Segment Length matching the requested alignment.
3626 for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
3627 unsigned short data_size = mtus[mtu_idx] - header_size;
3629 /* If this MTU minus the Header Size would result in a
3630 * Data Segment Size of the desired alignment, remember it.
3632 if ((data_size & data_size_align_mask) == 0)
3633 aligned_mtu_idx = mtu_idx;
3635 /* If we're not at the end of the Hardware MTU Table and the
3636 * next element is larger than our Maximum MTU, drop out of
3639 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
3643 /* If we fell out of the loop because we ran to the end of the table,
3644 * then we just have to use the last [largest] entry.
3646 if (mtu_idx == NMTUS)
3649 /* If we found an MTU which resulted in the requested Data Segment
3650 * Length alignment and that's "not far" from the largest MTU which is
3651 * less than or equal to the maximum MTU, then use that.
3653 if (aligned_mtu_idx >= 0 &&
3654 mtu_idx - aligned_mtu_idx <= 1)
3655 mtu_idx = aligned_mtu_idx;
3657 /* If the caller has passed in an MTU Index pointer, pass the
3658 * MTU Index back. Return the MTU value.
3661 *mtu_idxp = mtu_idx;
3662 return mtus[mtu_idx];
3664 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
3667 * cxgb4_port_chan - get the HW channel of a port
3668 * @dev: the net device for the port
3670 * Return the HW Tx channel of the given port.
3672 unsigned int cxgb4_port_chan(const struct net_device *dev)
3674 return netdev2pinfo(dev)->tx_chan;
3676 EXPORT_SYMBOL(cxgb4_port_chan);
3678 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
3680 struct adapter *adap = netdev2adap(dev);
3681 u32 v1, v2, lp_count, hp_count;
3683 v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3684 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3685 if (is_t4(adap->params.chip)) {
3686 lp_count = G_LP_COUNT(v1);
3687 hp_count = G_HP_COUNT(v1);
3689 lp_count = G_LP_COUNT_T5(v1);
3690 hp_count = G_HP_COUNT_T5(v2);
3692 return lpfifo ? lp_count : hp_count;
3694 EXPORT_SYMBOL(cxgb4_dbfifo_count);
3697 * cxgb4_port_viid - get the VI id of a port
3698 * @dev: the net device for the port
3700 * Return the VI id of the given port.
3702 unsigned int cxgb4_port_viid(const struct net_device *dev)
3704 return netdev2pinfo(dev)->viid;
3706 EXPORT_SYMBOL(cxgb4_port_viid);
3709 * cxgb4_port_idx - get the index of a port
3710 * @dev: the net device for the port
3712 * Return the index of the given port.
3714 unsigned int cxgb4_port_idx(const struct net_device *dev)
3716 return netdev2pinfo(dev)->port_id;
3718 EXPORT_SYMBOL(cxgb4_port_idx);
3720 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
3721 struct tp_tcp_stats *v6)
3723 struct adapter *adap = pci_get_drvdata(pdev);
3725 spin_lock(&adap->stats_lock);
3726 t4_tp_get_tcp_stats(adap, v4, v6);
3727 spin_unlock(&adap->stats_lock);
3729 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
3731 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
3732 const unsigned int *pgsz_order)
3734 struct adapter *adap = netdev2adap(dev);
3736 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
3737 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
3738 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
3739 HPZ3(pgsz_order[3]));
3741 EXPORT_SYMBOL(cxgb4_iscsi_init);
3743 int cxgb4_flush_eq_cache(struct net_device *dev)
3745 struct adapter *adap = netdev2adap(dev);
3748 ret = t4_fwaddrspace_write(adap, adap->mbox,
3749 0xe1000000 + A_SGE_CTXT_CMD, 0x20000000);
3752 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
3754 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
3756 u32 addr = t4_read_reg(adap, A_SGE_DBQ_CTXT_BADDR) + 24 * qid + 8;
3760 ret = t4_mem_win_read_len(adap, addr, (__be32 *)&indices, 8);
3762 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
3763 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
3768 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
3771 struct adapter *adap = netdev2adap(dev);
3772 u16 hw_pidx, hw_cidx;
3775 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
3779 if (pidx != hw_pidx) {
3782 if (pidx >= hw_pidx)
3783 delta = pidx - hw_pidx;
3785 delta = size - hw_pidx + pidx;
3787 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3788 QID(qid) | PIDX(delta));
3793 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
3795 void cxgb4_disable_db_coalescing(struct net_device *dev)
3797 struct adapter *adap;
3799 adap = netdev2adap(dev);
3800 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_NOCOALESCE,
3803 EXPORT_SYMBOL(cxgb4_disable_db_coalescing);
3805 void cxgb4_enable_db_coalescing(struct net_device *dev)
3807 struct adapter *adap;
3809 adap = netdev2adap(dev);
3810 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_NOCOALESCE, 0);
3812 EXPORT_SYMBOL(cxgb4_enable_db_coalescing);
3814 static struct pci_driver cxgb4_driver;
3816 static void check_neigh_update(struct neighbour *neigh)
3818 const struct device *parent;
3819 const struct net_device *netdev = neigh->dev;
3821 if (netdev->priv_flags & IFF_802_1Q_VLAN)
3822 netdev = vlan_dev_real_dev(netdev);
3823 parent = netdev->dev.parent;
3824 if (parent && parent->driver == &cxgb4_driver.driver)
3825 t4_l2t_update(dev_get_drvdata(parent), neigh);
3828 static int netevent_cb(struct notifier_block *nb, unsigned long event,
3832 case NETEVENT_NEIGH_UPDATE:
3833 check_neigh_update(data);
3835 case NETEVENT_REDIRECT:
3842 static bool netevent_registered;
3843 static struct notifier_block cxgb4_netevent_nb = {
3844 .notifier_call = netevent_cb
3847 static void drain_db_fifo(struct adapter *adap, int usecs)
3849 u32 v1, v2, lp_count, hp_count;
3852 v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3853 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3854 if (is_t4(adap->params.chip)) {
3855 lp_count = G_LP_COUNT(v1);
3856 hp_count = G_HP_COUNT(v1);
3858 lp_count = G_LP_COUNT_T5(v1);
3859 hp_count = G_HP_COUNT_T5(v2);
3862 if (lp_count == 0 && hp_count == 0)
3864 set_current_state(TASK_UNINTERRUPTIBLE);
3865 schedule_timeout(usecs_to_jiffies(usecs));
3869 static void disable_txq_db(struct sge_txq *q)
3871 unsigned long flags;
3873 spin_lock_irqsave(&q->db_lock, flags);
3875 spin_unlock_irqrestore(&q->db_lock, flags);
3878 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
3880 spin_lock_irq(&q->db_lock);
3881 if (q->db_pidx_inc) {
3882 /* Make sure that all writes to the TX descriptors
3883 * are committed before we tell HW about them.
3886 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3887 QID(q->cntxt_id) | PIDX(q->db_pidx_inc));
3891 spin_unlock_irq(&q->db_lock);
3894 static void disable_dbs(struct adapter *adap)
3898 for_each_ethrxq(&adap->sge, i)
3899 disable_txq_db(&adap->sge.ethtxq[i].q);
3900 for_each_ofldrxq(&adap->sge, i)
3901 disable_txq_db(&adap->sge.ofldtxq[i].q);
3902 for_each_port(adap, i)
3903 disable_txq_db(&adap->sge.ctrlq[i].q);
3906 static void enable_dbs(struct adapter *adap)
3910 for_each_ethrxq(&adap->sge, i)
3911 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
3912 for_each_ofldrxq(&adap->sge, i)
3913 enable_txq_db(adap, &adap->sge.ofldtxq[i].q);
3914 for_each_port(adap, i)
3915 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
3918 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
3920 if (adap->uld_handle[CXGB4_ULD_RDMA])
3921 ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
3925 static void process_db_full(struct work_struct *work)
3927 struct adapter *adap;
3929 adap = container_of(work, struct adapter, db_full_task);
3931 drain_db_fifo(adap, dbfifo_drain_delay);
3933 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3934 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3935 DBFIFO_HP_INT | DBFIFO_LP_INT,
3936 DBFIFO_HP_INT | DBFIFO_LP_INT);
3939 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
3941 u16 hw_pidx, hw_cidx;
3944 spin_lock_irq(&q->db_lock);
3945 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
3948 if (q->db_pidx != hw_pidx) {
3951 if (q->db_pidx >= hw_pidx)
3952 delta = q->db_pidx - hw_pidx;
3954 delta = q->size - hw_pidx + q->db_pidx;
3956 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3957 QID(q->cntxt_id) | PIDX(delta));
3962 spin_unlock_irq(&q->db_lock);
3964 CH_WARN(adap, "DB drop recovery failed.\n");
3966 static void recover_all_queues(struct adapter *adap)
3970 for_each_ethrxq(&adap->sge, i)
3971 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
3972 for_each_ofldrxq(&adap->sge, i)
3973 sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
3974 for_each_port(adap, i)
3975 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
3978 static void process_db_drop(struct work_struct *work)
3980 struct adapter *adap;
3982 adap = container_of(work, struct adapter, db_drop_task);
3984 if (is_t4(adap->params.chip)) {
3985 drain_db_fifo(adap, dbfifo_drain_delay);
3986 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
3987 drain_db_fifo(adap, dbfifo_drain_delay);
3988 recover_all_queues(adap);
3989 drain_db_fifo(adap, dbfifo_drain_delay);
3991 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3993 u32 dropped_db = t4_read_reg(adap, 0x010ac);
3994 u16 qid = (dropped_db >> 15) & 0x1ffff;
3995 u16 pidx_inc = dropped_db & 0x1fff;
3997 unsigned short udb_density;
3998 unsigned long qpshift;
4002 dev_warn(adap->pdev_dev,
4003 "Dropped DB 0x%x qid %d bar2 %d coalesce %d pidx %d\n",
4005 (dropped_db >> 14) & 1,
4006 (dropped_db >> 13) & 1,
4009 drain_db_fifo(adap, 1);
4011 s_qpp = QUEUESPERPAGEPF1 * adap->fn;
4012 udb_density = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adap,
4013 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
4014 qpshift = PAGE_SHIFT - ilog2(udb_density);
4015 udb = qid << qpshift;
4017 page = udb / PAGE_SIZE;
4018 udb += (qid - (page * udb_density)) * 128;
4020 writel(PIDX(pidx_inc), adap->bar2 + udb + 8);
4022 /* Re-enable BAR2 WC */
4023 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
4026 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_DROPPED_DB, 0);
4029 void t4_db_full(struct adapter *adap)
4031 if (is_t4(adap->params.chip)) {
4033 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
4034 t4_set_reg_field(adap, SGE_INT_ENABLE3,
4035 DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
4036 queue_work(workq, &adap->db_full_task);
4040 void t4_db_dropped(struct adapter *adap)
4042 if (is_t4(adap->params.chip)) {
4044 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
4046 queue_work(workq, &adap->db_drop_task);
4049 static void uld_attach(struct adapter *adap, unsigned int uld)
4052 struct cxgb4_lld_info lli;
4055 lli.pdev = adap->pdev;
4057 lli.l2t = adap->l2t;
4058 lli.tids = &adap->tids;
4059 lli.ports = adap->port;
4060 lli.vr = &adap->vres;
4061 lli.mtus = adap->params.mtus;
4062 if (uld == CXGB4_ULD_RDMA) {
4063 lli.rxq_ids = adap->sge.rdma_rxq;
4064 lli.ciq_ids = adap->sge.rdma_ciq;
4065 lli.nrxq = adap->sge.rdmaqs;
4066 lli.nciq = adap->sge.rdmaciqs;
4067 } else if (uld == CXGB4_ULD_ISCSI) {
4068 lli.rxq_ids = adap->sge.ofld_rxq;
4069 lli.nrxq = adap->sge.ofldqsets;
4071 lli.ntxq = adap->sge.ofldqsets;
4072 lli.nchan = adap->params.nports;
4073 lli.nports = adap->params.nports;
4074 lli.wr_cred = adap->params.ofldq_wr_cred;
4075 lli.adapter_type = adap->params.chip;
4076 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
4077 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
4078 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
4080 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
4081 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
4083 lli.filt_mode = adap->params.tp.vlan_pri_map;
4084 /* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
4085 for (i = 0; i < NCHAN; i++)
4087 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
4088 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
4089 lli.fw_vers = adap->params.fw_vers;
4090 lli.dbfifo_int_thresh = dbfifo_int_thresh;
4091 lli.sge_pktshift = adap->sge.pktshift;
4092 lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
4093 lli.ulptx_memwrite_dsgl = adap->params.ulptx_memwrite_dsgl;
4095 handle = ulds[uld].add(&lli);
4096 if (IS_ERR(handle)) {
4097 dev_warn(adap->pdev_dev,
4098 "could not attach to the %s driver, error %ld\n",
4099 uld_str[uld], PTR_ERR(handle));
4103 adap->uld_handle[uld] = handle;
4105 if (!netevent_registered) {
4106 register_netevent_notifier(&cxgb4_netevent_nb);
4107 netevent_registered = true;
4110 if (adap->flags & FULL_INIT_DONE)
4111 ulds[uld].state_change(handle, CXGB4_STATE_UP);
4114 static void attach_ulds(struct adapter *adap)
4118 spin_lock(&adap_rcu_lock);
4119 list_add_tail_rcu(&adap->rcu_node, &adap_rcu_list);
4120 spin_unlock(&adap_rcu_lock);
4122 mutex_lock(&uld_mutex);
4123 list_add_tail(&adap->list_node, &adapter_list);
4124 for (i = 0; i < CXGB4_ULD_MAX; i++)
4126 uld_attach(adap, i);
4127 mutex_unlock(&uld_mutex);
4130 static void detach_ulds(struct adapter *adap)
4134 mutex_lock(&uld_mutex);
4135 list_del(&adap->list_node);
4136 for (i = 0; i < CXGB4_ULD_MAX; i++)
4137 if (adap->uld_handle[i]) {
4138 ulds[i].state_change(adap->uld_handle[i],
4139 CXGB4_STATE_DETACH);
4140 adap->uld_handle[i] = NULL;
4142 if (netevent_registered && list_empty(&adapter_list)) {
4143 unregister_netevent_notifier(&cxgb4_netevent_nb);
4144 netevent_registered = false;
4146 mutex_unlock(&uld_mutex);
4148 spin_lock(&adap_rcu_lock);
4149 list_del_rcu(&adap->rcu_node);
4150 spin_unlock(&adap_rcu_lock);
4153 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
4157 mutex_lock(&uld_mutex);
4158 for (i = 0; i < CXGB4_ULD_MAX; i++)
4159 if (adap->uld_handle[i])
4160 ulds[i].state_change(adap->uld_handle[i], new_state);
4161 mutex_unlock(&uld_mutex);
4165 * cxgb4_register_uld - register an upper-layer driver
4166 * @type: the ULD type
4167 * @p: the ULD methods
4169 * Registers an upper-layer driver with this driver and notifies the ULD
4170 * about any presently available devices that support its type. Returns
4171 * %-EBUSY if a ULD of the same type is already registered.
4173 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
4176 struct adapter *adap;
4178 if (type >= CXGB4_ULD_MAX)
4180 mutex_lock(&uld_mutex);
4181 if (ulds[type].add) {
4186 list_for_each_entry(adap, &adapter_list, list_node)
4187 uld_attach(adap, type);
4188 out: mutex_unlock(&uld_mutex);
4191 EXPORT_SYMBOL(cxgb4_register_uld);
4194 * cxgb4_unregister_uld - unregister an upper-layer driver
4195 * @type: the ULD type
4197 * Unregisters an existing upper-layer driver.
4199 int cxgb4_unregister_uld(enum cxgb4_uld type)
4201 struct adapter *adap;
4203 if (type >= CXGB4_ULD_MAX)
4205 mutex_lock(&uld_mutex);
4206 list_for_each_entry(adap, &adapter_list, list_node)
4207 adap->uld_handle[type] = NULL;
4208 ulds[type].add = NULL;
4209 mutex_unlock(&uld_mutex);
4212 EXPORT_SYMBOL(cxgb4_unregister_uld);
4214 /* Check if netdev on which event is occured belongs to us or not. Return
4215 * success (true) if it belongs otherwise failure (false).
4216 * Called with rcu_read_lock() held.
4218 static bool cxgb4_netdev(const struct net_device *netdev)
4220 struct adapter *adap;
4223 list_for_each_entry_rcu(adap, &adap_rcu_list, rcu_node)
4224 for (i = 0; i < MAX_NPORTS; i++)
4225 if (adap->port[i] == netdev)
4230 static int clip_add(struct net_device *event_dev, struct inet6_ifaddr *ifa,
4231 unsigned long event)
4233 int ret = NOTIFY_DONE;
4236 if (cxgb4_netdev(event_dev)) {
4239 ret = cxgb4_clip_get(event_dev,
4240 (const struct in6_addr *)ifa->addr.s6_addr);
4248 cxgb4_clip_release(event_dev,
4249 (const struct in6_addr *)ifa->addr.s6_addr);
4260 static int cxgb4_inet6addr_handler(struct notifier_block *this,
4261 unsigned long event, void *data)
4263 struct inet6_ifaddr *ifa = data;
4264 struct net_device *event_dev;
4265 int ret = NOTIFY_DONE;
4266 struct bonding *bond = netdev_priv(ifa->idev->dev);
4267 struct list_head *iter;
4268 struct slave *slave;
4269 struct pci_dev *first_pdev = NULL;
4271 if (ifa->idev->dev->priv_flags & IFF_802_1Q_VLAN) {
4272 event_dev = vlan_dev_real_dev(ifa->idev->dev);
4273 ret = clip_add(event_dev, ifa, event);
4274 } else if (ifa->idev->dev->flags & IFF_MASTER) {
4275 /* It is possible that two different adapters are bonded in one
4276 * bond. We need to find such different adapters and add clip
4277 * in all of them only once.
4279 read_lock(&bond->lock);
4280 bond_for_each_slave(bond, slave, iter) {
4282 ret = clip_add(slave->dev, ifa, event);
4283 /* If clip_add is success then only initialize
4284 * first_pdev since it means it is our device
4286 if (ret == NOTIFY_OK)
4287 first_pdev = to_pci_dev(
4288 slave->dev->dev.parent);
4289 } else if (first_pdev !=
4290 to_pci_dev(slave->dev->dev.parent))
4291 ret = clip_add(slave->dev, ifa, event);
4293 read_unlock(&bond->lock);
4295 ret = clip_add(ifa->idev->dev, ifa, event);
4300 static struct notifier_block cxgb4_inet6addr_notifier = {
4301 .notifier_call = cxgb4_inet6addr_handler
4304 /* Retrieves IPv6 addresses from a root device (bond, vlan) associated with
4305 * a physical device.
4306 * The physical device reference is needed to send the actul CLIP command.
4308 static int update_dev_clip(struct net_device *root_dev, struct net_device *dev)
4310 struct inet6_dev *idev = NULL;
4311 struct inet6_ifaddr *ifa;
4314 idev = __in6_dev_get(root_dev);
4318 read_lock_bh(&idev->lock);
4319 list_for_each_entry(ifa, &idev->addr_list, if_list) {
4320 ret = cxgb4_clip_get(dev,
4321 (const struct in6_addr *)ifa->addr.s6_addr);
4325 read_unlock_bh(&idev->lock);
4330 static int update_root_dev_clip(struct net_device *dev)
4332 struct net_device *root_dev = NULL;
4335 /* First populate the real net device's IPv6 addresses */
4336 ret = update_dev_clip(dev, dev);
4340 /* Parse all bond and vlan devices layered on top of the physical dev */
4341 for (i = 0; i < VLAN_N_VID; i++) {
4342 root_dev = __vlan_find_dev_deep_rcu(dev, htons(ETH_P_8021Q), i);
4346 ret = update_dev_clip(root_dev, dev);
4353 static void update_clip(const struct adapter *adap)
4356 struct net_device *dev;
4361 for (i = 0; i < MAX_NPORTS; i++) {
4362 dev = adap->port[i];
4366 ret = update_root_dev_clip(dev);
4375 * cxgb_up - enable the adapter
4376 * @adap: adapter being enabled
4378 * Called when the first port is enabled, this function performs the
4379 * actions necessary to make an adapter operational, such as completing
4380 * the initialization of HW modules, and enabling interrupts.
4382 * Must be called with the rtnl lock held.
4384 static int cxgb_up(struct adapter *adap)
4388 err = setup_sge_queues(adap);
4391 err = setup_rss(adap);
4395 if (adap->flags & USING_MSIX) {
4396 name_msix_vecs(adap);
4397 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
4398 adap->msix_info[0].desc, adap);
4402 err = request_msix_queue_irqs(adap);
4404 free_irq(adap->msix_info[0].vec, adap);
4408 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
4409 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
4410 adap->port[0]->name, adap);
4416 t4_intr_enable(adap);
4417 adap->flags |= FULL_INIT_DONE;
4418 notify_ulds(adap, CXGB4_STATE_UP);
4423 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
4425 t4_free_sge_resources(adap);
4429 static void cxgb_down(struct adapter *adapter)
4431 t4_intr_disable(adapter);
4432 cancel_work_sync(&adapter->tid_release_task);
4433 cancel_work_sync(&adapter->db_full_task);
4434 cancel_work_sync(&adapter->db_drop_task);
4435 adapter->tid_release_task_busy = false;
4436 adapter->tid_release_head = NULL;
4438 if (adapter->flags & USING_MSIX) {
4439 free_msix_queue_irqs(adapter);
4440 free_irq(adapter->msix_info[0].vec, adapter);
4442 free_irq(adapter->pdev->irq, adapter);
4443 quiesce_rx(adapter);
4444 t4_sge_stop(adapter);
4445 t4_free_sge_resources(adapter);
4446 adapter->flags &= ~FULL_INIT_DONE;
4450 * net_device operations
4452 static int cxgb_open(struct net_device *dev)
4455 struct port_info *pi = netdev_priv(dev);
4456 struct adapter *adapter = pi->adapter;
4458 netif_carrier_off(dev);
4460 if (!(adapter->flags & FULL_INIT_DONE)) {
4461 err = cxgb_up(adapter);
4466 err = link_start(dev);
4468 netif_tx_start_all_queues(dev);
4472 static int cxgb_close(struct net_device *dev)
4474 struct port_info *pi = netdev_priv(dev);
4475 struct adapter *adapter = pi->adapter;
4477 netif_tx_stop_all_queues(dev);
4478 netif_carrier_off(dev);
4479 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
4482 /* Return an error number if the indicated filter isn't writable ...
4484 static int writable_filter(struct filter_entry *f)
4494 /* Delete the filter at the specified index (if valid). The checks for all
4495 * the common problems with doing this like the filter being locked, currently
4496 * pending in another operation, etc.
4498 static int delete_filter(struct adapter *adapter, unsigned int fidx)
4500 struct filter_entry *f;
4503 if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
4506 f = &adapter->tids.ftid_tab[fidx];
4507 ret = writable_filter(f);
4511 return del_filter_wr(adapter, fidx);
4516 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
4517 __be32 sip, __be16 sport, __be16 vlan,
4518 unsigned int queue, unsigned char port, unsigned char mask)
4521 struct filter_entry *f;
4522 struct adapter *adap;
4526 adap = netdev2adap(dev);
4528 /* Adjust stid to correct filter index */
4529 stid -= adap->tids.sftid_base;
4530 stid += adap->tids.nftids;
4532 /* Check to make sure the filter requested is writable ...
4534 f = &adap->tids.ftid_tab[stid];
4535 ret = writable_filter(f);
4539 /* Clear out any old resources being used by the filter before
4540 * we start constructing the new filter.
4543 clear_filter(adap, f);
4545 /* Clear out filter specifications */
4546 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
4547 f->fs.val.lport = cpu_to_be16(sport);
4548 f->fs.mask.lport = ~0;
4550 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
4551 for (i = 0; i < 4; i++) {
4552 f->fs.val.lip[i] = val[i];
4553 f->fs.mask.lip[i] = ~0;
4555 if (adap->params.tp.vlan_pri_map & F_PORT) {
4556 f->fs.val.iport = port;
4557 f->fs.mask.iport = mask;
4561 if (adap->params.tp.vlan_pri_map & F_PROTOCOL) {
4562 f->fs.val.proto = IPPROTO_TCP;
4563 f->fs.mask.proto = ~0;
4568 /* Mark filter as locked */
4572 ret = set_filter_wr(adap, stid);
4574 clear_filter(adap, f);
4580 EXPORT_SYMBOL(cxgb4_create_server_filter);
4582 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
4583 unsigned int queue, bool ipv6)
4586 struct filter_entry *f;
4587 struct adapter *adap;
4589 adap = netdev2adap(dev);
4591 /* Adjust stid to correct filter index */
4592 stid -= adap->tids.sftid_base;
4593 stid += adap->tids.nftids;
4595 f = &adap->tids.ftid_tab[stid];
4596 /* Unlock the filter */
4599 ret = delete_filter(adap, stid);
4605 EXPORT_SYMBOL(cxgb4_remove_server_filter);
4607 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
4608 struct rtnl_link_stats64 *ns)
4610 struct port_stats stats;
4611 struct port_info *p = netdev_priv(dev);
4612 struct adapter *adapter = p->adapter;
4614 /* Block retrieving statistics during EEH error
4615 * recovery. Otherwise, the recovery might fail
4616 * and the PCI device will be removed permanently
4618 spin_lock(&adapter->stats_lock);
4619 if (!netif_device_present(dev)) {
4620 spin_unlock(&adapter->stats_lock);
4623 t4_get_port_stats(adapter, p->tx_chan, &stats);
4624 spin_unlock(&adapter->stats_lock);
4626 ns->tx_bytes = stats.tx_octets;
4627 ns->tx_packets = stats.tx_frames;
4628 ns->rx_bytes = stats.rx_octets;
4629 ns->rx_packets = stats.rx_frames;
4630 ns->multicast = stats.rx_mcast_frames;
4632 /* detailed rx_errors */
4633 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
4635 ns->rx_over_errors = 0;
4636 ns->rx_crc_errors = stats.rx_fcs_err;
4637 ns->rx_frame_errors = stats.rx_symbol_err;
4638 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
4639 stats.rx_ovflow2 + stats.rx_ovflow3 +
4640 stats.rx_trunc0 + stats.rx_trunc1 +
4641 stats.rx_trunc2 + stats.rx_trunc3;
4642 ns->rx_missed_errors = 0;
4644 /* detailed tx_errors */
4645 ns->tx_aborted_errors = 0;
4646 ns->tx_carrier_errors = 0;
4647 ns->tx_fifo_errors = 0;
4648 ns->tx_heartbeat_errors = 0;
4649 ns->tx_window_errors = 0;
4651 ns->tx_errors = stats.tx_error_frames;
4652 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
4653 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
4657 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
4660 int ret = 0, prtad, devad;
4661 struct port_info *pi = netdev_priv(dev);
4662 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
4666 if (pi->mdio_addr < 0)
4668 data->phy_id = pi->mdio_addr;
4672 if (mdio_phy_id_is_c45(data->phy_id)) {
4673 prtad = mdio_phy_id_prtad(data->phy_id);
4674 devad = mdio_phy_id_devad(data->phy_id);
4675 } else if (data->phy_id < 32) {
4676 prtad = data->phy_id;
4678 data->reg_num &= 0x1f;
4682 mbox = pi->adapter->fn;
4683 if (cmd == SIOCGMIIREG)
4684 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
4685 data->reg_num, &data->val_out);
4687 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
4688 data->reg_num, data->val_in);
4696 static void cxgb_set_rxmode(struct net_device *dev)
4698 /* unfortunately we can't return errors to the stack */
4699 set_rxmode(dev, -1, false);
4702 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
4705 struct port_info *pi = netdev_priv(dev);
4707 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
4709 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
4716 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
4719 struct sockaddr *addr = p;
4720 struct port_info *pi = netdev_priv(dev);
4722 if (!is_valid_ether_addr(addr->sa_data))
4723 return -EADDRNOTAVAIL;
4725 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
4726 pi->xact_addr_filt, addr->sa_data, true, true);
4730 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4731 pi->xact_addr_filt = ret;
4735 #ifdef CONFIG_NET_POLL_CONTROLLER
4736 static void cxgb_netpoll(struct net_device *dev)
4738 struct port_info *pi = netdev_priv(dev);
4739 struct adapter *adap = pi->adapter;
4741 if (adap->flags & USING_MSIX) {
4743 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
4745 for (i = pi->nqsets; i; i--, rx++)
4746 t4_sge_intr_msix(0, &rx->rspq);
4748 t4_intr_handler(adap)(0, adap);
4752 static const struct net_device_ops cxgb4_netdev_ops = {
4753 .ndo_open = cxgb_open,
4754 .ndo_stop = cxgb_close,
4755 .ndo_start_xmit = t4_eth_xmit,
4756 .ndo_select_queue = cxgb_select_queue,
4757 .ndo_get_stats64 = cxgb_get_stats,
4758 .ndo_set_rx_mode = cxgb_set_rxmode,
4759 .ndo_set_mac_address = cxgb_set_mac_addr,
4760 .ndo_set_features = cxgb_set_features,
4761 .ndo_validate_addr = eth_validate_addr,
4762 .ndo_do_ioctl = cxgb_ioctl,
4763 .ndo_change_mtu = cxgb_change_mtu,
4764 #ifdef CONFIG_NET_POLL_CONTROLLER
4765 .ndo_poll_controller = cxgb_netpoll,
4769 void t4_fatal_err(struct adapter *adap)
4771 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
4772 t4_intr_disable(adap);
4773 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
4776 /* Return the specified PCI-E Configuration Space register from our Physical
4777 * Function. We try first via a Firmware LDST Command since we prefer to let
4778 * the firmware own all of these registers, but if that fails we go for it
4779 * directly ourselves.
4781 static u32 t4_read_pcie_cfg4(struct adapter *adap, int reg)
4783 struct fw_ldst_cmd ldst_cmd;
4787 /* Construct and send the Firmware LDST Command to retrieve the
4788 * specified PCI-E Configuration Space register.
4790 memset(&ldst_cmd, 0, sizeof(ldst_cmd));
4791 ldst_cmd.op_to_addrspace =
4792 htonl(FW_CMD_OP(FW_LDST_CMD) |
4795 FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_FUNC_PCIE));
4796 ldst_cmd.cycles_to_len16 = htonl(FW_LEN16(ldst_cmd));
4797 ldst_cmd.u.pcie.select_naccess = FW_LDST_CMD_NACCESS(1);
4798 ldst_cmd.u.pcie.ctrl_to_fn =
4799 (FW_LDST_CMD_LC | FW_LDST_CMD_FN(adap->fn));
4800 ldst_cmd.u.pcie.r = reg;
4801 ret = t4_wr_mbox(adap, adap->mbox, &ldst_cmd, sizeof(ldst_cmd),
4804 /* If the LDST Command suucceeded, exctract the returned register
4805 * value. Otherwise read it directly ourself.
4808 val = ntohl(ldst_cmd.u.pcie.data[0]);
4810 t4_hw_pci_read_cfg4(adap, reg, &val);
4815 static void setup_memwin(struct adapter *adap)
4817 u32 mem_win0_base, mem_win1_base, mem_win2_base, mem_win2_aperture;
4819 if (is_t4(adap->params.chip)) {
4822 /* Truncation intentional: we only read the bottom 32-bits of
4823 * the 64-bit BAR0/BAR1 ... We use the hardware backdoor
4824 * mechanism to read BAR0 instead of using
4825 * pci_resource_start() because we could be operating from
4826 * within a Virtual Machine which is trapping our accesses to
4827 * our Configuration Space and we need to set up the PCI-E
4828 * Memory Window decoders with the actual addresses which will
4829 * be coming across the PCI-E link.
4831 bar0 = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_0);
4832 bar0 &= PCI_BASE_ADDRESS_MEM_MASK;
4833 adap->t4_bar0 = bar0;
4835 mem_win0_base = bar0 + MEMWIN0_BASE;
4836 mem_win1_base = bar0 + MEMWIN1_BASE;
4837 mem_win2_base = bar0 + MEMWIN2_BASE;
4838 mem_win2_aperture = MEMWIN2_APERTURE;
4840 /* For T5, only relative offset inside the PCIe BAR is passed */
4841 mem_win0_base = MEMWIN0_BASE;
4842 mem_win1_base = MEMWIN1_BASE;
4843 mem_win2_base = MEMWIN2_BASE_T5;
4844 mem_win2_aperture = MEMWIN2_APERTURE_T5;
4846 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
4847 mem_win0_base | BIR(0) |
4848 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
4849 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
4850 mem_win1_base | BIR(0) |
4851 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
4852 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
4853 mem_win2_base | BIR(0) |
4854 WINDOW(ilog2(mem_win2_aperture) - 10));
4855 t4_read_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2));
4858 static void setup_memwin_rdma(struct adapter *adap)
4860 if (adap->vres.ocq.size) {
4864 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
4865 start &= PCI_BASE_ADDRESS_MEM_MASK;
4866 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
4867 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
4869 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
4870 start | BIR(1) | WINDOW(ilog2(sz_kb)));
4872 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
4873 adap->vres.ocq.start);
4875 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
4879 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
4884 /* get device capabilities */
4885 memset(c, 0, sizeof(*c));
4886 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4887 FW_CMD_REQUEST | FW_CMD_READ);
4888 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
4889 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
4893 /* select capabilities we'll be using */
4894 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4896 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4898 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4899 } else if (vf_acls) {
4900 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
4903 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4904 FW_CMD_REQUEST | FW_CMD_WRITE);
4905 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
4909 ret = t4_config_glbl_rss(adap, adap->fn,
4910 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4911 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4912 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
4916 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
4917 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
4923 /* tweak some settings */
4924 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
4925 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
4926 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
4927 v = t4_read_reg(adap, TP_PIO_DATA);
4928 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
4930 /* first 4 Tx modulation queues point to consecutive Tx channels */
4931 adap->params.tp.tx_modq_map = 0xE4;
4932 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
4933 V_TX_MOD_QUEUE_REQ_MAP(adap->params.tp.tx_modq_map));
4935 /* associate each Tx modulation queue with consecutive Tx channels */
4937 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4938 &v, 1, A_TP_TX_SCHED_HDR);
4939 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4940 &v, 1, A_TP_TX_SCHED_FIFO);
4941 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4942 &v, 1, A_TP_TX_SCHED_PCMD);
4944 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4945 if (is_offload(adap)) {
4946 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0,
4947 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4948 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4949 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4950 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4951 t4_write_reg(adap, A_TP_TX_MOD_CHANNEL_WEIGHT,
4952 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4953 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4954 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4955 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4958 /* get basic stuff going */
4959 return t4_early_init(adap, adap->fn);
4963 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
4965 #define MAX_ATIDS 8192U
4968 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4970 * If the firmware we're dealing with has Configuration File support, then
4971 * we use that to perform all configuration
4975 * Tweak configuration based on module parameters, etc. Most of these have
4976 * defaults assigned to them by Firmware Configuration Files (if we're using
4977 * them) but need to be explicitly set if we're using hard-coded
4978 * initialization. But even in the case of using Firmware Configuration
4979 * Files, we'd like to expose the ability to change these via module
4980 * parameters so these are essentially common tweaks/settings for
4981 * Configuration Files and hard-coded initialization ...
4983 static int adap_init0_tweaks(struct adapter *adapter)
4986 * Fix up various Host-Dependent Parameters like Page Size, Cache
4987 * Line Size, etc. The firmware default is for a 4KB Page Size and
4988 * 64B Cache Line Size ...
4990 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4993 * Process module parameters which affect early initialization.
4995 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4996 dev_err(&adapter->pdev->dev,
4997 "Ignoring illegal rx_dma_offset=%d, using 2\n",
5001 t4_set_reg_field(adapter, SGE_CONTROL,
5003 PKTSHIFT(rx_dma_offset));
5006 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
5007 * adds the pseudo header itself.
5009 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG,
5010 CSUM_HAS_PSEUDO_HDR, 0);
5016 * Attempt to initialize the adapter via a Firmware Configuration File.
5018 static int adap_init0_config(struct adapter *adapter, int reset)
5020 struct fw_caps_config_cmd caps_cmd;
5021 const struct firmware *cf;
5022 unsigned long mtype = 0, maddr = 0;
5023 u32 finiver, finicsum, cfcsum;
5025 int config_issued = 0;
5026 char *fw_config_file, fw_config_file_path[256];
5027 char *config_name = NULL;
5030 * Reset device if necessary.
5033 ret = t4_fw_reset(adapter, adapter->mbox,
5034 PIORSTMODE | PIORST);
5040 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
5041 * then use that. Otherwise, use the configuration file stored
5042 * in the adapter flash ...
5044 switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
5046 fw_config_file = FW4_CFNAME;
5049 fw_config_file = FW5_CFNAME;
5052 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
5053 adapter->pdev->device);
5058 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
5060 config_name = "On FLASH";
5061 mtype = FW_MEMTYPE_CF_FLASH;
5062 maddr = t4_flash_cfg_addr(adapter);
5064 u32 params[7], val[7];
5066 sprintf(fw_config_file_path,
5067 "/lib/firmware/%s", fw_config_file);
5068 config_name = fw_config_file_path;
5070 if (cf->size >= FLASH_CFG_MAX_SIZE)
5073 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5074 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
5075 ret = t4_query_params(adapter, adapter->mbox,
5076 adapter->fn, 0, 1, params, val);
5079 * For t4_memory_write() below addresses and
5080 * sizes have to be in terms of multiples of 4
5081 * bytes. So, if the Configuration File isn't
5082 * a multiple of 4 bytes in length we'll have
5083 * to write that out separately since we can't
5084 * guarantee that the bytes following the
5085 * residual byte in the buffer returned by
5086 * request_firmware() are zeroed out ...
5088 size_t resid = cf->size & 0x3;
5089 size_t size = cf->size & ~0x3;
5090 __be32 *data = (__be32 *)cf->data;
5092 mtype = FW_PARAMS_PARAM_Y_GET(val[0]);
5093 maddr = FW_PARAMS_PARAM_Z_GET(val[0]) << 16;
5095 ret = t4_memory_write(adapter, mtype, maddr,
5097 if (ret == 0 && resid != 0) {
5104 last.word = data[size >> 2];
5105 for (i = resid; i < 4; i++)
5107 ret = t4_memory_write(adapter, mtype,
5114 release_firmware(cf);
5120 * Issue a Capability Configuration command to the firmware to get it
5121 * to parse the Configuration File. We don't use t4_fw_config_file()
5122 * because we want the ability to modify various features after we've
5123 * processed the configuration file ...
5125 memset(&caps_cmd, 0, sizeof(caps_cmd));
5126 caps_cmd.op_to_write =
5127 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5130 caps_cmd.cfvalid_to_len16 =
5131 htonl(FW_CAPS_CONFIG_CMD_CFVALID |
5132 FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
5133 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
5134 FW_LEN16(caps_cmd));
5135 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
5138 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
5139 * Configuration File in FLASH), our last gasp effort is to use the
5140 * Firmware Configuration File which is embedded in the firmware. A
5141 * very few early versions of the firmware didn't have one embedded
5142 * but we can ignore those.
5144 if (ret == -ENOENT) {
5145 memset(&caps_cmd, 0, sizeof(caps_cmd));
5146 caps_cmd.op_to_write =
5147 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5150 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
5151 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
5152 sizeof(caps_cmd), &caps_cmd);
5153 config_name = "Firmware Default";
5160 finiver = ntohl(caps_cmd.finiver);
5161 finicsum = ntohl(caps_cmd.finicsum);
5162 cfcsum = ntohl(caps_cmd.cfcsum);
5163 if (finicsum != cfcsum)
5164 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
5165 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
5169 * And now tell the firmware to use the configuration we just loaded.
5171 caps_cmd.op_to_write =
5172 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5175 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
5176 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
5182 * Tweak configuration based on system architecture, module
5185 ret = adap_init0_tweaks(adapter);
5190 * And finally tell the firmware to initialize itself using the
5191 * parameters from the Configuration File.
5193 ret = t4_fw_initialize(adapter, adapter->mbox);
5198 * Return successfully and note that we're operating with parameters
5199 * not supplied by the driver, rather than from hard-wired
5200 * initialization constants burried in the driver.
5202 adapter->flags |= USING_SOFT_PARAMS;
5203 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
5204 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
5205 config_name, finiver, cfcsum);
5209 * Something bad happened. Return the error ... (If the "error"
5210 * is that there's no Configuration File on the adapter we don't
5211 * want to issue a warning since this is fairly common.)
5214 if (config_issued && ret != -ENOENT)
5215 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
5221 * Attempt to initialize the adapter via hard-coded, driver supplied
5224 static int adap_init0_no_config(struct adapter *adapter, int reset)
5226 struct sge *s = &adapter->sge;
5227 struct fw_caps_config_cmd caps_cmd;
5232 * Reset device if necessary
5235 ret = t4_fw_reset(adapter, adapter->mbox,
5236 PIORSTMODE | PIORST);
5242 * Get device capabilities and select which we'll be using.
5244 memset(&caps_cmd, 0, sizeof(caps_cmd));
5245 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5246 FW_CMD_REQUEST | FW_CMD_READ);
5247 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
5248 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
5253 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
5255 caps_cmd.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
5257 caps_cmd.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
5258 } else if (vf_acls) {
5259 dev_err(adapter->pdev_dev, "virtualization ACLs not supported");
5262 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5263 FW_CMD_REQUEST | FW_CMD_WRITE);
5264 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
5270 * Tweak configuration based on system architecture, module
5273 ret = adap_init0_tweaks(adapter);
5278 * Select RSS Global Mode we want to use. We use "Basic Virtual"
5279 * mode which maps each Virtual Interface to its own section of
5280 * the RSS Table and we turn on all map and hash enables ...
5282 adapter->flags |= RSS_TNLALLLOOKUP;
5283 ret = t4_config_glbl_rss(adapter, adapter->mbox,
5284 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
5285 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
5286 FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ |
5287 ((adapter->flags & RSS_TNLALLLOOKUP) ?
5288 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP : 0));
5293 * Set up our own fundamental resource provisioning ...
5295 ret = t4_cfg_pfvf(adapter, adapter->mbox, adapter->fn, 0,
5296 PFRES_NEQ, PFRES_NETHCTRL,
5297 PFRES_NIQFLINT, PFRES_NIQ,
5298 PFRES_TC, PFRES_NVI,
5299 FW_PFVF_CMD_CMASK_MASK,
5300 pfvfres_pmask(adapter, adapter->fn, 0),
5302 PFRES_R_CAPS, PFRES_WX_CAPS);
5307 * Perform low level SGE initialization. We need to do this before we
5308 * send the firmware the INITIALIZE command because that will cause
5309 * any other PF Drivers which are waiting for the Master
5310 * Initialization to proceed forward.
5312 for (i = 0; i < SGE_NTIMERS - 1; i++)
5313 s->timer_val[i] = min(intr_holdoff[i], MAX_SGE_TIMERVAL);
5314 s->timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
5315 s->counter_val[0] = 1;
5316 for (i = 1; i < SGE_NCOUNTERS; i++)
5317 s->counter_val[i] = min(intr_cnt[i - 1],
5318 THRESHOLD_0_GET(THRESHOLD_0_MASK));
5319 t4_sge_init(adapter);
5321 #ifdef CONFIG_PCI_IOV
5323 * Provision resource limits for Virtual Functions. We currently
5324 * grant them all the same static resource limits except for the Port
5325 * Access Rights Mask which we're assigning based on the PF. All of
5326 * the static provisioning stuff for both the PF and VF really needs
5327 * to be managed in a persistent manner for each device which the
5328 * firmware controls.
5333 for (pf = 0; pf < ARRAY_SIZE(num_vf); pf++) {
5334 if (num_vf[pf] <= 0)
5337 /* VF numbering starts at 1! */
5338 for (vf = 1; vf <= num_vf[pf]; vf++) {
5339 ret = t4_cfg_pfvf(adapter, adapter->mbox,
5341 VFRES_NEQ, VFRES_NETHCTRL,
5342 VFRES_NIQFLINT, VFRES_NIQ,
5343 VFRES_TC, VFRES_NVI,
5344 FW_PFVF_CMD_CMASK_MASK,
5348 VFRES_R_CAPS, VFRES_WX_CAPS);
5350 dev_warn(adapter->pdev_dev,
5352 "provision pf/vf=%d/%d; "
5353 "err=%d\n", pf, vf, ret);
5360 * Set up the default filter mode. Later we'll want to implement this
5361 * via a firmware command, etc. ... This needs to be done before the
5362 * firmare initialization command ... If the selected set of fields
5363 * isn't equal to the default value, we'll need to make sure that the
5364 * field selections will fit in the 36-bit budget.
5366 if (tp_vlan_pri_map != TP_VLAN_PRI_MAP_DEFAULT) {
5369 for (j = TP_VLAN_PRI_MAP_FIRST; j <= TP_VLAN_PRI_MAP_LAST; j++)
5370 switch (tp_vlan_pri_map & (1 << j)) {
5372 /* compressed filter field not enabled */
5392 case ETHERTYPE_MASK:
5398 case MPSHITTYPE_MASK:
5401 case FRAGMENTATION_MASK:
5407 dev_err(adapter->pdev_dev,
5408 "tp_vlan_pri_map=%#x needs %d bits > 36;"\
5409 " using %#x\n", tp_vlan_pri_map, bits,
5410 TP_VLAN_PRI_MAP_DEFAULT);
5411 tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
5414 v = tp_vlan_pri_map;
5415 t4_write_indirect(adapter, TP_PIO_ADDR, TP_PIO_DATA,
5416 &v, 1, TP_VLAN_PRI_MAP);
5419 * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
5420 * to support any of the compressed filter fields above. Newer
5421 * versions of the firmware do this automatically but it doesn't hurt
5422 * to set it here. Meanwhile, we do _not_ need to set Lookup Every
5423 * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
5424 * since the firmware automatically turns this on and off when we have
5425 * a non-zero number of filters active (since it does have a
5426 * performance impact).
5428 if (tp_vlan_pri_map)
5429 t4_set_reg_field(adapter, TP_GLOBAL_CONFIG,
5430 FIVETUPLELOOKUP_MASK,
5431 FIVETUPLELOOKUP_MASK);
5434 * Tweak some settings.
5436 t4_write_reg(adapter, TP_SHIFT_CNT, SYNSHIFTMAX(6) |
5437 RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
5438 PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
5439 KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
5442 * Get basic stuff going by issuing the Firmware Initialize command.
5443 * Note that this _must_ be after all PFVF commands ...
5445 ret = t4_fw_initialize(adapter, adapter->mbox);
5450 * Return successfully!
5452 dev_info(adapter->pdev_dev, "Successfully configured using built-in "\
5453 "driver parameters\n");
5457 * Something bad happened. Return the error ...
5463 static struct fw_info fw_info_array[] = {
5466 .fs_name = FW4_CFNAME,
5467 .fw_mod_name = FW4_FNAME,
5469 .chip = FW_HDR_CHIP_T4,
5470 .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
5471 .intfver_nic = FW_INTFVER(T4, NIC),
5472 .intfver_vnic = FW_INTFVER(T4, VNIC),
5473 .intfver_ri = FW_INTFVER(T4, RI),
5474 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
5475 .intfver_fcoe = FW_INTFVER(T4, FCOE),
5479 .fs_name = FW5_CFNAME,
5480 .fw_mod_name = FW5_FNAME,
5482 .chip = FW_HDR_CHIP_T5,
5483 .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
5484 .intfver_nic = FW_INTFVER(T5, NIC),
5485 .intfver_vnic = FW_INTFVER(T5, VNIC),
5486 .intfver_ri = FW_INTFVER(T5, RI),
5487 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
5488 .intfver_fcoe = FW_INTFVER(T5, FCOE),
5493 static struct fw_info *find_fw_info(int chip)
5497 for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
5498 if (fw_info_array[i].chip == chip)
5499 return &fw_info_array[i];
5505 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
5507 static int adap_init0(struct adapter *adap)
5511 enum dev_state state;
5512 u32 params[7], val[7];
5513 struct fw_caps_config_cmd caps_cmd;
5517 * Contact FW, advertising Master capability (and potentially forcing
5518 * ourselves as the Master PF if our module parameter force_init is
5521 ret = t4_fw_hello(adap, adap->mbox, adap->fn,
5522 force_init ? MASTER_MUST : MASTER_MAY,
5525 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
5529 if (ret == adap->mbox)
5530 adap->flags |= MASTER_PF;
5531 if (force_init && state == DEV_STATE_INIT)
5532 state = DEV_STATE_UNINIT;
5535 * If we're the Master PF Driver and the device is uninitialized,
5536 * then let's consider upgrading the firmware ... (We always want
5537 * to check the firmware version number in order to A. get it for
5538 * later reporting and B. to warn if the currently loaded firmware
5539 * is excessively mismatched relative to the driver.)
5541 t4_get_fw_version(adap, &adap->params.fw_vers);
5542 t4_get_tp_version(adap, &adap->params.tp_vers);
5543 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
5544 struct fw_info *fw_info;
5545 struct fw_hdr *card_fw;
5546 const struct firmware *fw;
5547 const u8 *fw_data = NULL;
5548 unsigned int fw_size = 0;
5550 /* This is the firmware whose headers the driver was compiled
5553 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
5554 if (fw_info == NULL) {
5555 dev_err(adap->pdev_dev,
5556 "unable to get firmware info for chip %d.\n",
5557 CHELSIO_CHIP_VERSION(adap->params.chip));
5561 /* allocate memory to read the header of the firmware on the
5564 card_fw = t4_alloc_mem(sizeof(*card_fw));
5566 /* Get FW from from /lib/firmware/ */
5567 ret = request_firmware(&fw, fw_info->fw_mod_name,
5570 dev_err(adap->pdev_dev,
5571 "unable to load firmware image %s, error %d\n",
5572 fw_info->fw_mod_name, ret);
5578 /* upgrade FW logic */
5579 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
5584 release_firmware(fw);
5585 t4_free_mem(card_fw);
5592 * Grab VPD parameters. This should be done after we establish a
5593 * connection to the firmware since some of the VPD parameters
5594 * (notably the Core Clock frequency) are retrieved via requests to
5595 * the firmware. On the other hand, we need these fairly early on
5596 * so we do this right after getting ahold of the firmware.
5598 ret = get_vpd_params(adap, &adap->params.vpd);
5603 * Find out what ports are available to us. Note that we need to do
5604 * this before calling adap_init0_no_config() since it needs nports
5608 FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5609 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
5610 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1, &v, &port_vec);
5614 adap->params.nports = hweight32(port_vec);
5615 adap->params.portvec = port_vec;
5618 * If the firmware is initialized already (and we're not forcing a
5619 * master initialization), note that we're living with existing
5620 * adapter parameters. Otherwise, it's time to try initializing the
5623 if (state == DEV_STATE_INIT) {
5624 dev_info(adap->pdev_dev, "Coming up as %s: "\
5625 "Adapter already initialized\n",
5626 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
5627 adap->flags |= USING_SOFT_PARAMS;
5629 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
5630 "Initializing adapter\n");
5633 * If the firmware doesn't support Configuration
5634 * Files warn user and exit,
5637 dev_warn(adap->pdev_dev, "Firmware doesn't support "
5638 "configuration file.\n");
5640 ret = adap_init0_no_config(adap, reset);
5643 * Find out whether we're dealing with a version of
5644 * the firmware which has configuration file support.
5646 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5647 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
5648 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1,
5652 * If the firmware doesn't support Configuration
5653 * Files, use the old Driver-based, hard-wired
5654 * initialization. Otherwise, try using the
5655 * Configuration File support and fall back to the
5656 * Driver-based initialization if there's no
5657 * Configuration File found.
5660 ret = adap_init0_no_config(adap, reset);
5663 * The firmware provides us with a memory
5664 * buffer where we can load a Configuration
5665 * File from the host if we want to override
5666 * the Configuration File in flash.
5669 ret = adap_init0_config(adap, reset);
5670 if (ret == -ENOENT) {
5671 dev_info(adap->pdev_dev,
5672 "No Configuration File present "
5673 "on adapter. Using hard-wired "
5674 "configuration parameters.\n");
5675 ret = adap_init0_no_config(adap, reset);
5680 dev_err(adap->pdev_dev,
5681 "could not initialize adapter, error %d\n",
5688 * If we're living with non-hard-coded parameters (either from a
5689 * Firmware Configuration File or values programmed by a different PF
5690 * Driver), give the SGE code a chance to pull in anything that it
5691 * needs ... Note that this must be called after we retrieve our VPD
5692 * parameters in order to know how to convert core ticks to seconds.
5694 if (adap->flags & USING_SOFT_PARAMS) {
5695 ret = t4_sge_init(adap);
5700 if (is_bypass_device(adap->pdev->device))
5701 adap->params.bypass = 1;
5704 * Grab some of our basic fundamental operating parameters.
5706 #define FW_PARAM_DEV(param) \
5707 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
5708 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
5710 #define FW_PARAM_PFVF(param) \
5711 FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
5712 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)| \
5713 FW_PARAMS_PARAM_Y(0) | \
5714 FW_PARAMS_PARAM_Z(0)
5716 params[0] = FW_PARAM_PFVF(EQ_START);
5717 params[1] = FW_PARAM_PFVF(L2T_START);
5718 params[2] = FW_PARAM_PFVF(L2T_END);
5719 params[3] = FW_PARAM_PFVF(FILTER_START);
5720 params[4] = FW_PARAM_PFVF(FILTER_END);
5721 params[5] = FW_PARAM_PFVF(IQFLINT_START);
5722 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6, params, val);
5725 adap->sge.egr_start = val[0];
5726 adap->l2t_start = val[1];
5727 adap->l2t_end = val[2];
5728 adap->tids.ftid_base = val[3];
5729 adap->tids.nftids = val[4] - val[3] + 1;
5730 adap->sge.ingr_start = val[5];
5732 /* query params related to active filter region */
5733 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
5734 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
5735 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2, params, val);
5736 /* If Active filter size is set we enable establishing
5737 * offload connection through firmware work request
5739 if ((val[0] != val[1]) && (ret >= 0)) {
5740 adap->flags |= FW_OFLD_CONN;
5741 adap->tids.aftid_base = val[0];
5742 adap->tids.aftid_end = val[1];
5745 /* If we're running on newer firmware, let it know that we're
5746 * prepared to deal with encapsulated CPL messages. Older
5747 * firmware won't understand this and we'll just get
5748 * unencapsulated messages ...
5750 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
5752 (void) t4_set_params(adap, adap->mbox, adap->fn, 0, 1, params, val);
5755 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
5756 * capability. Earlier versions of the firmware didn't have the
5757 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
5758 * permission to use ULPTX MEMWRITE DSGL.
5760 if (is_t4(adap->params.chip)) {
5761 adap->params.ulptx_memwrite_dsgl = false;
5763 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
5764 ret = t4_query_params(adap, adap->mbox, adap->fn, 0,
5766 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
5770 * Get device capabilities so we can determine what resources we need
5773 memset(&caps_cmd, 0, sizeof(caps_cmd));
5774 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5775 FW_CMD_REQUEST | FW_CMD_READ);
5776 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
5777 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
5782 if (caps_cmd.ofldcaps) {
5783 /* query offload-related parameters */
5784 params[0] = FW_PARAM_DEV(NTID);
5785 params[1] = FW_PARAM_PFVF(SERVER_START);
5786 params[2] = FW_PARAM_PFVF(SERVER_END);
5787 params[3] = FW_PARAM_PFVF(TDDP_START);
5788 params[4] = FW_PARAM_PFVF(TDDP_END);
5789 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
5790 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5794 adap->tids.ntids = val[0];
5795 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
5796 adap->tids.stid_base = val[1];
5797 adap->tids.nstids = val[2] - val[1] + 1;
5799 * Setup server filter region. Divide the availble filter
5800 * region into two parts. Regular filters get 1/3rd and server
5801 * filters get 2/3rd part. This is only enabled if workarond
5803 * 1. For regular filters.
5804 * 2. Server filter: This are special filters which are used
5805 * to redirect SYN packets to offload queue.
5807 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
5808 adap->tids.sftid_base = adap->tids.ftid_base +
5809 DIV_ROUND_UP(adap->tids.nftids, 3);
5810 adap->tids.nsftids = adap->tids.nftids -
5811 DIV_ROUND_UP(adap->tids.nftids, 3);
5812 adap->tids.nftids = adap->tids.sftid_base -
5813 adap->tids.ftid_base;
5815 adap->vres.ddp.start = val[3];
5816 adap->vres.ddp.size = val[4] - val[3] + 1;
5817 adap->params.ofldq_wr_cred = val[5];
5819 adap->params.offload = 1;
5821 if (caps_cmd.rdmacaps) {
5822 params[0] = FW_PARAM_PFVF(STAG_START);
5823 params[1] = FW_PARAM_PFVF(STAG_END);
5824 params[2] = FW_PARAM_PFVF(RQ_START);
5825 params[3] = FW_PARAM_PFVF(RQ_END);
5826 params[4] = FW_PARAM_PFVF(PBL_START);
5827 params[5] = FW_PARAM_PFVF(PBL_END);
5828 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5832 adap->vres.stag.start = val[0];
5833 adap->vres.stag.size = val[1] - val[0] + 1;
5834 adap->vres.rq.start = val[2];
5835 adap->vres.rq.size = val[3] - val[2] + 1;
5836 adap->vres.pbl.start = val[4];
5837 adap->vres.pbl.size = val[5] - val[4] + 1;
5839 params[0] = FW_PARAM_PFVF(SQRQ_START);
5840 params[1] = FW_PARAM_PFVF(SQRQ_END);
5841 params[2] = FW_PARAM_PFVF(CQ_START);
5842 params[3] = FW_PARAM_PFVF(CQ_END);
5843 params[4] = FW_PARAM_PFVF(OCQ_START);
5844 params[5] = FW_PARAM_PFVF(OCQ_END);
5845 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
5848 adap->vres.qp.start = val[0];
5849 adap->vres.qp.size = val[1] - val[0] + 1;
5850 adap->vres.cq.start = val[2];
5851 adap->vres.cq.size = val[3] - val[2] + 1;
5852 adap->vres.ocq.start = val[4];
5853 adap->vres.ocq.size = val[5] - val[4] + 1;
5855 if (caps_cmd.iscsicaps) {
5856 params[0] = FW_PARAM_PFVF(ISCSI_START);
5857 params[1] = FW_PARAM_PFVF(ISCSI_END);
5858 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2,
5862 adap->vres.iscsi.start = val[0];
5863 adap->vres.iscsi.size = val[1] - val[0] + 1;
5865 #undef FW_PARAM_PFVF
5868 /* The MTU/MSS Table is initialized by now, so load their values. If
5869 * we're initializing the adapter, then we'll make any modifications
5870 * we want to the MTU/MSS Table and also initialize the congestion
5873 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
5874 if (state != DEV_STATE_INIT) {
5877 /* The default MTU Table contains values 1492 and 1500.
5878 * However, for TCP, it's better to have two values which are
5879 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
5880 * This allows us to have a TCP Data Payload which is a
5881 * multiple of 8 regardless of what combination of TCP Options
5882 * are in use (always a multiple of 4 bytes) which is
5883 * important for performance reasons. For instance, if no
5884 * options are in use, then we have a 20-byte IP header and a
5885 * 20-byte TCP header. In this case, a 1500-byte MSS would
5886 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
5887 * which is not a multiple of 8. So using an MSS of 1488 in
5888 * this case results in a TCP Data Payload of 1448 bytes which
5889 * is a multiple of 8. On the other hand, if 12-byte TCP Time
5890 * Stamps have been negotiated, then an MTU of 1500 bytes
5891 * results in a TCP Data Payload of 1448 bytes which, as
5892 * above, is a multiple of 8 bytes ...
5894 for (i = 0; i < NMTUS; i++)
5895 if (adap->params.mtus[i] == 1492) {
5896 adap->params.mtus[i] = 1488;
5900 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5901 adap->params.b_wnd);
5903 t4_init_tp_params(adap);
5904 adap->flags |= FW_OK;
5908 * Something bad happened. If a command timed out or failed with EIO
5909 * FW does not operate within its spec or something catastrophic
5910 * happened to HW/FW, stop issuing commands.
5913 if (ret != -ETIMEDOUT && ret != -EIO)
5914 t4_fw_bye(adap, adap->mbox);
5920 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
5921 pci_channel_state_t state)
5924 struct adapter *adap = pci_get_drvdata(pdev);
5930 adap->flags &= ~FW_OK;
5931 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5932 spin_lock(&adap->stats_lock);
5933 for_each_port(adap, i) {
5934 struct net_device *dev = adap->port[i];
5936 netif_device_detach(dev);
5937 netif_carrier_off(dev);
5939 spin_unlock(&adap->stats_lock);
5940 if (adap->flags & FULL_INIT_DONE)
5943 if ((adap->flags & DEV_ENABLED)) {
5944 pci_disable_device(pdev);
5945 adap->flags &= ~DEV_ENABLED;
5947 out: return state == pci_channel_io_perm_failure ?
5948 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5951 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5954 struct fw_caps_config_cmd c;
5955 struct adapter *adap = pci_get_drvdata(pdev);
5958 pci_restore_state(pdev);
5959 pci_save_state(pdev);
5960 return PCI_ERS_RESULT_RECOVERED;
5963 if (!(adap->flags & DEV_ENABLED)) {
5964 if (pci_enable_device(pdev)) {
5965 dev_err(&pdev->dev, "Cannot reenable PCI "
5966 "device after reset\n");
5967 return PCI_ERS_RESULT_DISCONNECT;
5969 adap->flags |= DEV_ENABLED;
5972 pci_set_master(pdev);
5973 pci_restore_state(pdev);
5974 pci_save_state(pdev);
5975 pci_cleanup_aer_uncorrect_error_status(pdev);
5977 if (t4_wait_dev_ready(adap) < 0)
5978 return PCI_ERS_RESULT_DISCONNECT;
5979 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL) < 0)
5980 return PCI_ERS_RESULT_DISCONNECT;
5981 adap->flags |= FW_OK;
5982 if (adap_init1(adap, &c))
5983 return PCI_ERS_RESULT_DISCONNECT;
5985 for_each_port(adap, i) {
5986 struct port_info *p = adap2pinfo(adap, i);
5988 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
5991 return PCI_ERS_RESULT_DISCONNECT;
5993 p->xact_addr_filt = -1;
5996 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5997 adap->params.b_wnd);
6000 return PCI_ERS_RESULT_DISCONNECT;
6001 return PCI_ERS_RESULT_RECOVERED;
6004 static void eeh_resume(struct pci_dev *pdev)
6007 struct adapter *adap = pci_get_drvdata(pdev);
6013 for_each_port(adap, i) {
6014 struct net_device *dev = adap->port[i];
6016 if (netif_running(dev)) {
6018 cxgb_set_rxmode(dev);
6020 netif_device_attach(dev);
6025 static const struct pci_error_handlers cxgb4_eeh = {
6026 .error_detected = eeh_err_detected,
6027 .slot_reset = eeh_slot_reset,
6028 .resume = eeh_resume,
6031 static inline bool is_x_10g_port(const struct link_config *lc)
6033 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0 ||
6034 (lc->supported & FW_PORT_CAP_SPEED_40G) != 0;
6037 static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
6038 unsigned int us, unsigned int cnt,
6039 unsigned int size, unsigned int iqe_size)
6042 set_rspq_intr_params(q, us, cnt);
6043 q->iqe_len = iqe_size;
6048 * Perform default configuration of DMA queues depending on the number and type
6049 * of ports we found and the number of available CPUs. Most settings can be
6050 * modified by the admin prior to actual use.
6052 static void cfg_queues(struct adapter *adap)
6054 struct sge *s = &adap->sge;
6055 int i, n10g = 0, qidx = 0;
6056 #ifndef CONFIG_CHELSIO_T4_DCB
6061 for_each_port(adap, i)
6062 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
6063 #ifdef CONFIG_CHELSIO_T4_DCB
6064 /* For Data Center Bridging support we need to be able to support up
6065 * to 8 Traffic Priorities; each of which will be assigned to its
6066 * own TX Queue in order to prevent Head-Of-Line Blocking.
6068 if (adap->params.nports * 8 > MAX_ETH_QSETS) {
6069 dev_err(adap->pdev_dev, "MAX_ETH_QSETS=%d < %d!\n",
6070 MAX_ETH_QSETS, adap->params.nports * 8);
6074 for_each_port(adap, i) {
6075 struct port_info *pi = adap2pinfo(adap, i);
6077 pi->first_qset = qidx;
6081 #else /* !CONFIG_CHELSIO_T4_DCB */
6083 * We default to 1 queue per non-10G port and up to # of cores queues
6087 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
6088 if (q10g > netif_get_num_default_rss_queues())
6089 q10g = netif_get_num_default_rss_queues();
6091 for_each_port(adap, i) {
6092 struct port_info *pi = adap2pinfo(adap, i);
6094 pi->first_qset = qidx;
6095 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
6098 #endif /* !CONFIG_CHELSIO_T4_DCB */
6101 s->max_ethqsets = qidx; /* MSI-X may lower it later */
6103 if (is_offload(adap)) {
6105 * For offload we use 1 queue/channel if all ports are up to 1G,
6106 * otherwise we divide all available queues amongst the channels
6107 * capped by the number of available cores.
6110 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
6112 s->ofldqsets = roundup(i, adap->params.nports);
6114 s->ofldqsets = adap->params.nports;
6115 /* For RDMA one Rx queue per channel suffices */
6116 s->rdmaqs = adap->params.nports;
6117 s->rdmaciqs = adap->params.nports;
6120 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
6121 struct sge_eth_rxq *r = &s->ethrxq[i];
6123 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
6127 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
6128 s->ethtxq[i].q.size = 1024;
6130 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
6131 s->ctrlq[i].q.size = 512;
6133 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
6134 s->ofldtxq[i].q.size = 1024;
6136 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
6137 struct sge_ofld_rxq *r = &s->ofldrxq[i];
6139 init_rspq(adap, &r->rspq, 5, 1, 1024, 64);
6140 r->rspq.uld = CXGB4_ULD_ISCSI;
6144 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
6145 struct sge_ofld_rxq *r = &s->rdmarxq[i];
6147 init_rspq(adap, &r->rspq, 5, 1, 511, 64);
6148 r->rspq.uld = CXGB4_ULD_RDMA;
6152 ciq_size = 64 + adap->vres.cq.size + adap->tids.nftids;
6153 if (ciq_size > SGE_MAX_IQ_SIZE) {
6154 CH_WARN(adap, "CIQ size too small for available IQs\n");
6155 ciq_size = SGE_MAX_IQ_SIZE;
6158 for (i = 0; i < ARRAY_SIZE(s->rdmaciq); i++) {
6159 struct sge_ofld_rxq *r = &s->rdmaciq[i];
6161 init_rspq(adap, &r->rspq, 5, 1, ciq_size, 64);
6162 r->rspq.uld = CXGB4_ULD_RDMA;
6165 init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
6166 init_rspq(adap, &s->intrq, 0, 1, 2 * MAX_INGQ, 64);
6170 * Reduce the number of Ethernet queues across all ports to at most n.
6171 * n provides at least one queue per port.
6173 static void reduce_ethqs(struct adapter *adap, int n)
6176 struct port_info *pi;
6178 while (n < adap->sge.ethqsets)
6179 for_each_port(adap, i) {
6180 pi = adap2pinfo(adap, i);
6181 if (pi->nqsets > 1) {
6183 adap->sge.ethqsets--;
6184 if (adap->sge.ethqsets <= n)
6190 for_each_port(adap, i) {
6191 pi = adap2pinfo(adap, i);
6197 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
6198 #define EXTRA_VECS 2
6200 static int enable_msix(struct adapter *adap)
6204 struct sge *s = &adap->sge;
6205 unsigned int nchan = adap->params.nports;
6206 struct msix_entry entries[MAX_INGQ + 1];
6208 for (i = 0; i < ARRAY_SIZE(entries); ++i)
6209 entries[i].entry = i;
6211 want = s->max_ethqsets + EXTRA_VECS;
6212 if (is_offload(adap)) {
6213 want += s->rdmaqs + s->rdmaciqs + s->ofldqsets;
6214 /* need nchan for each possible ULD */
6215 ofld_need = 3 * nchan;
6217 #ifdef CONFIG_CHELSIO_T4_DCB
6218 /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
6221 need = 8 * adap->params.nports + EXTRA_VECS + ofld_need;
6223 need = adap->params.nports + EXTRA_VECS + ofld_need;
6225 want = pci_enable_msix_range(adap->pdev, entries, need, want);
6230 * Distribute available vectors to the various queue groups.
6231 * Every group gets its minimum requirement and NIC gets top
6232 * priority for leftovers.
6234 i = want - EXTRA_VECS - ofld_need;
6235 if (i < s->max_ethqsets) {
6236 s->max_ethqsets = i;
6237 if (i < s->ethqsets)
6238 reduce_ethqs(adap, i);
6240 if (is_offload(adap)) {
6241 i = want - EXTRA_VECS - s->max_ethqsets;
6242 i -= ofld_need - nchan;
6243 s->ofldqsets = (i / nchan) * nchan; /* round down */
6245 for (i = 0; i < want; ++i)
6246 adap->msix_info[i].vec = entries[i].vector;
6253 static int init_rss(struct adapter *adap)
6257 for_each_port(adap, i) {
6258 struct port_info *pi = adap2pinfo(adap, i);
6260 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
6263 for (j = 0; j < pi->rss_size; j++)
6264 pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
6269 static void print_port_info(const struct net_device *dev)
6273 const char *spd = "";
6274 const struct port_info *pi = netdev_priv(dev);
6275 const struct adapter *adap = pi->adapter;
6277 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
6279 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
6281 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_8_0GB)
6284 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
6285 bufp += sprintf(bufp, "100/");
6286 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
6287 bufp += sprintf(bufp, "1000/");
6288 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
6289 bufp += sprintf(bufp, "10G/");
6290 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
6291 bufp += sprintf(bufp, "40G/");
6294 sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
6296 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
6297 adap->params.vpd.id,
6298 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
6299 is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
6300 (adap->flags & USING_MSIX) ? " MSI-X" :
6301 (adap->flags & USING_MSI) ? " MSI" : "");
6302 netdev_info(dev, "S/N: %s, P/N: %s\n",
6303 adap->params.vpd.sn, adap->params.vpd.pn);
6306 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
6308 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
6312 * Free the following resources:
6313 * - memory used for tables
6316 * - resources FW is holding for us
6318 static void free_some_resources(struct adapter *adapter)
6322 t4_free_mem(adapter->l2t);
6323 t4_free_mem(adapter->tids.tid_tab);
6324 disable_msi(adapter);
6326 for_each_port(adapter, i)
6327 if (adapter->port[i]) {
6328 kfree(adap2pinfo(adapter, i)->rss);
6329 free_netdev(adapter->port[i]);
6331 if (adapter->flags & FW_OK)
6332 t4_fw_bye(adapter, adapter->fn);
6335 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
6336 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
6337 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
6338 #define SEGMENT_SIZE 128
6340 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
6342 int func, i, err, s_qpp, qpp, num_seg;
6343 struct port_info *pi;
6344 bool highdma = false;
6345 struct adapter *adapter = NULL;
6347 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
6349 err = pci_request_regions(pdev, KBUILD_MODNAME);
6351 /* Just info, some other driver may have claimed the device. */
6352 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
6356 err = pci_enable_device(pdev);
6358 dev_err(&pdev->dev, "cannot enable PCI device\n");
6359 goto out_release_regions;
6362 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
6364 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
6366 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
6367 "coherent allocations\n");
6368 goto out_disable_device;
6371 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
6373 dev_err(&pdev->dev, "no usable DMA configuration\n");
6374 goto out_disable_device;
6378 pci_enable_pcie_error_reporting(pdev);
6379 enable_pcie_relaxed_ordering(pdev);
6380 pci_set_master(pdev);
6381 pci_save_state(pdev);
6383 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
6386 goto out_disable_device;
6389 /* PCI device has been enabled */
6390 adapter->flags |= DEV_ENABLED;
6392 adapter->regs = pci_ioremap_bar(pdev, 0);
6393 if (!adapter->regs) {
6394 dev_err(&pdev->dev, "cannot map device registers\n");
6396 goto out_free_adapter;
6399 /* We control everything through one PF */
6400 func = SOURCEPF_GET(readl(adapter->regs + PL_WHOAMI));
6401 if ((pdev->device == 0xa000 && func != 0) ||
6402 func != ent->driver_data) {
6403 pci_save_state(pdev); /* to restore SR-IOV later */
6405 goto out_unmap_bar0;
6408 adapter->pdev = pdev;
6409 adapter->pdev_dev = &pdev->dev;
6410 adapter->mbox = func;
6412 adapter->msg_enable = dflt_msg_enable;
6413 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
6415 spin_lock_init(&adapter->stats_lock);
6416 spin_lock_init(&adapter->tid_release_lock);
6418 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
6419 INIT_WORK(&adapter->db_full_task, process_db_full);
6420 INIT_WORK(&adapter->db_drop_task, process_db_drop);
6422 err = t4_prep_adapter(adapter);
6424 goto out_unmap_bar0;
6426 if (!is_t4(adapter->params.chip)) {
6427 s_qpp = QUEUESPERPAGEPF1 * adapter->fn;
6428 qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter,
6429 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
6430 num_seg = PAGE_SIZE / SEGMENT_SIZE;
6432 /* Each segment size is 128B. Write coalescing is enabled only
6433 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
6434 * queue is less no of segments that can be accommodated in
6437 if (qpp > num_seg) {
6439 "Incorrect number of egress queues per page\n");
6441 goto out_unmap_bar0;
6443 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
6444 pci_resource_len(pdev, 2));
6445 if (!adapter->bar2) {
6446 dev_err(&pdev->dev, "cannot map device bar2 region\n");
6448 goto out_unmap_bar0;
6452 setup_memwin(adapter);
6453 err = adap_init0(adapter);
6454 setup_memwin_rdma(adapter);
6458 for_each_port(adapter, i) {
6459 struct net_device *netdev;
6461 netdev = alloc_etherdev_mq(sizeof(struct port_info),
6468 SET_NETDEV_DEV(netdev, &pdev->dev);
6470 adapter->port[i] = netdev;
6471 pi = netdev_priv(netdev);
6472 pi->adapter = adapter;
6473 pi->xact_addr_filt = -1;
6475 netdev->irq = pdev->irq;
6477 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
6478 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
6479 NETIF_F_RXCSUM | NETIF_F_RXHASH |
6480 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
6482 netdev->hw_features |= NETIF_F_HIGHDMA;
6483 netdev->features |= netdev->hw_features;
6484 netdev->vlan_features = netdev->features & VLAN_FEAT;
6486 netdev->priv_flags |= IFF_UNICAST_FLT;
6488 netdev->netdev_ops = &cxgb4_netdev_ops;
6489 #ifdef CONFIG_CHELSIO_T4_DCB
6490 netdev->dcbnl_ops = &cxgb4_dcb_ops;
6491 cxgb4_dcb_state_init(netdev);
6493 netdev->ethtool_ops = &cxgb_ethtool_ops;
6496 pci_set_drvdata(pdev, adapter);
6498 if (adapter->flags & FW_OK) {
6499 err = t4_port_init(adapter, func, func, 0);
6505 * Configure queues and allocate tables now, they can be needed as
6506 * soon as the first register_netdev completes.
6508 cfg_queues(adapter);
6510 adapter->l2t = t4_init_l2t();
6511 if (!adapter->l2t) {
6512 /* We tolerate a lack of L2T, giving up some functionality */
6513 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
6514 adapter->params.offload = 0;
6517 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
6518 dev_warn(&pdev->dev, "could not allocate TID table, "
6520 adapter->params.offload = 0;
6523 /* See what interrupts we'll be using */
6524 if (msi > 1 && enable_msix(adapter) == 0)
6525 adapter->flags |= USING_MSIX;
6526 else if (msi > 0 && pci_enable_msi(pdev) == 0)
6527 adapter->flags |= USING_MSI;
6529 err = init_rss(adapter);
6534 * The card is now ready to go. If any errors occur during device
6535 * registration we do not fail the whole card but rather proceed only
6536 * with the ports we manage to register successfully. However we must
6537 * register at least one net device.
6539 for_each_port(adapter, i) {
6540 pi = adap2pinfo(adapter, i);
6541 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
6542 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
6544 err = register_netdev(adapter->port[i]);
6547 adapter->chan_map[pi->tx_chan] = i;
6548 print_port_info(adapter->port[i]);
6551 dev_err(&pdev->dev, "could not register any net devices\n");
6555 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
6559 if (cxgb4_debugfs_root) {
6560 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
6561 cxgb4_debugfs_root);
6562 setup_debugfs(adapter);
6565 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
6566 pdev->needs_freset = 1;
6568 if (is_offload(adapter))
6569 attach_ulds(adapter);
6571 #ifdef CONFIG_PCI_IOV
6572 if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
6573 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
6574 dev_info(&pdev->dev,
6575 "instantiated %u virtual functions\n",
6581 free_some_resources(adapter);
6583 if (!is_t4(adapter->params.chip))
6584 iounmap(adapter->bar2);
6586 iounmap(adapter->regs);
6590 pci_disable_pcie_error_reporting(pdev);
6591 pci_disable_device(pdev);
6592 out_release_regions:
6593 pci_release_regions(pdev);
6597 static void remove_one(struct pci_dev *pdev)
6599 struct adapter *adapter = pci_get_drvdata(pdev);
6601 #ifdef CONFIG_PCI_IOV
6602 pci_disable_sriov(pdev);
6609 if (is_offload(adapter))
6610 detach_ulds(adapter);
6612 for_each_port(adapter, i)
6613 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
6614 unregister_netdev(adapter->port[i]);
6616 if (adapter->debugfs_root)
6617 debugfs_remove_recursive(adapter->debugfs_root);
6619 /* If we allocated filters, free up state associated with any
6622 if (adapter->tids.ftid_tab) {
6623 struct filter_entry *f = &adapter->tids.ftid_tab[0];
6624 for (i = 0; i < (adapter->tids.nftids +
6625 adapter->tids.nsftids); i++, f++)
6627 clear_filter(adapter, f);
6630 if (adapter->flags & FULL_INIT_DONE)
6633 free_some_resources(adapter);
6634 iounmap(adapter->regs);
6635 if (!is_t4(adapter->params.chip))
6636 iounmap(adapter->bar2);
6637 pci_disable_pcie_error_reporting(pdev);
6638 if ((adapter->flags & DEV_ENABLED)) {
6639 pci_disable_device(pdev);
6640 adapter->flags &= ~DEV_ENABLED;
6642 pci_release_regions(pdev);
6646 pci_release_regions(pdev);
6649 static struct pci_driver cxgb4_driver = {
6650 .name = KBUILD_MODNAME,
6651 .id_table = cxgb4_pci_tbl,
6653 .remove = remove_one,
6654 .shutdown = remove_one,
6655 .err_handler = &cxgb4_eeh,
6658 static int __init cxgb4_init_module(void)
6662 workq = create_singlethread_workqueue("cxgb4");
6666 /* Debugfs support is optional, just warn if this fails */
6667 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
6668 if (!cxgb4_debugfs_root)
6669 pr_warn("could not create debugfs entry, continuing\n");
6671 ret = pci_register_driver(&cxgb4_driver);
6673 debugfs_remove(cxgb4_debugfs_root);
6674 destroy_workqueue(workq);
6677 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6682 static void __exit cxgb4_cleanup_module(void)
6684 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6685 pci_unregister_driver(&cxgb4_driver);
6686 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
6687 flush_workqueue(workq);
6688 destroy_workqueue(workq);
6691 module_init(cxgb4_init_module);
6692 module_exit(cxgb4_cleanup_module);