2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/bitmap.h>
38 #include <linux/crc32.h>
39 #include <linux/ctype.h>
40 #include <linux/debugfs.h>
41 #include <linux/err.h>
42 #include <linux/etherdevice.h>
43 #include <linux/firmware.h>
45 #include <linux/if_vlan.h>
46 #include <linux/init.h>
47 #include <linux/log2.h>
48 #include <linux/mdio.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/mutex.h>
52 #include <linux/netdevice.h>
53 #include <linux/pci.h>
54 #include <linux/aer.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/sched.h>
57 #include <linux/seq_file.h>
58 #include <linux/sockios.h>
59 #include <linux/vmalloc.h>
60 #include <linux/workqueue.h>
61 #include <net/neighbour.h>
62 #include <net/netevent.h>
63 #include <asm/uaccess.h>
71 #define DRV_VERSION "1.3.0-ko"
72 #define DRV_DESC "Chelsio T4 Network Driver"
75 * Max interrupt hold-off timer value in us. Queues fall back to this value
76 * under extreme memory pressure so it's largish to give the system time to
79 #define MAX_SGE_TIMERVAL 200U
83 * Physical Function provisioning constants.
85 PFRES_NVI = 4, /* # of Virtual Interfaces */
86 PFRES_NETHCTRL = 128, /* # of EQs used for ETH or CTRL Qs */
87 PFRES_NIQFLINT = 128, /* # of ingress Qs/w Free List(s)/intr
89 PFRES_NEQ = 256, /* # of egress queues */
90 PFRES_NIQ = 0, /* # of ingress queues */
91 PFRES_TC = 0, /* PCI-E traffic class */
92 PFRES_NEXACTF = 128, /* # of exact MPS filters */
94 PFRES_R_CAPS = FW_CMD_CAP_PF,
95 PFRES_WX_CAPS = FW_CMD_CAP_PF,
99 * Virtual Function provisioning constants. We need two extra Ingress
100 * Queues with Interrupt capability to serve as the VF's Firmware
101 * Event Queue and Forwarded Interrupt Queue (when using MSI mode) --
102 * neither will have Free Lists associated with them). For each
103 * Ethernet/Control Egress Queue and for each Free List, we need an
106 VFRES_NPORTS = 1, /* # of "ports" per VF */
107 VFRES_NQSETS = 2, /* # of "Queue Sets" per VF */
109 VFRES_NVI = VFRES_NPORTS, /* # of Virtual Interfaces */
110 VFRES_NETHCTRL = VFRES_NQSETS, /* # of EQs used for ETH or CTRL Qs */
111 VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
112 VFRES_NEQ = VFRES_NQSETS*2, /* # of egress queues */
113 VFRES_NIQ = 0, /* # of non-fl/int ingress queues */
114 VFRES_TC = 0, /* PCI-E traffic class */
115 VFRES_NEXACTF = 16, /* # of exact MPS filters */
117 VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
118 VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
123 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
124 * static and likely not to be useful in the long run. We really need to
125 * implement some form of persistent configuration which the firmware
128 static unsigned int pfvfres_pmask(struct adapter *adapter,
129 unsigned int pf, unsigned int vf)
131 unsigned int portn, portvec;
134 * Give PF's access to all of the ports.
137 return FW_PFVF_CMD_PMASK_MASK;
140 * For VFs, we'll assign them access to the ports based purely on the
141 * PF. We assign active ports in order, wrapping around if there are
142 * fewer active ports than PFs: e.g. active port[pf % nports].
143 * Unfortunately the adapter's port_info structs haven't been
144 * initialized yet so we have to compute this.
146 if (adapter->params.nports == 0)
149 portn = pf % adapter->params.nports;
150 portvec = adapter->params.portvec;
153 * Isolate the lowest set bit in the port vector. If we're at
154 * the port number that we want, return that as the pmask.
155 * otherwise mask that bit out of the port vector and
156 * decrement our port number ...
158 unsigned int pmask = portvec ^ (portvec & (portvec-1));
168 MAX_TXQ_ENTRIES = 16384,
169 MAX_CTRL_TXQ_ENTRIES = 1024,
170 MAX_RSPQ_ENTRIES = 16384,
171 MAX_RX_BUFFERS = 16384,
172 MIN_TXQ_ENTRIES = 32,
173 MIN_CTRL_TXQ_ENTRIES = 32,
174 MIN_RSPQ_ENTRIES = 128,
178 /* Host shadow copy of ingress filter entry. This is in host native format
179 * and doesn't match the ordering or bit order, etc. of the hardware of the
180 * firmware command. The use of bit-field structure elements is purely to
181 * remind ourselves of the field size limitations and save memory in the case
182 * where the filter table is large.
184 struct filter_entry {
185 /* Administrative fields for filter.
187 u32 valid:1; /* filter allocated and valid */
188 u32 locked:1; /* filter is administratively locked */
190 u32 pending:1; /* filter action is pending firmware reply */
191 u32 smtidx:8; /* Source MAC Table index for smac */
192 struct l2t_entry *l2t; /* Layer Two Table entry for dmac */
194 /* The filter itself. Most of this is a straight copy of information
195 * provided by the extended ioctl(). Some fields are translated to
196 * internal forms -- for instance the Ingress Queue ID passed in from
197 * the ioctl() is translated into the Absolute Ingress Queue ID.
199 struct ch_filter_specification fs;
202 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
203 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
204 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
206 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
208 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
209 CH_DEVICE(0xa000, 0), /* PE10K */
210 CH_DEVICE(0x4001, -1),
211 CH_DEVICE(0x4002, -1),
212 CH_DEVICE(0x4003, -1),
213 CH_DEVICE(0x4004, -1),
214 CH_DEVICE(0x4005, -1),
215 CH_DEVICE(0x4006, -1),
216 CH_DEVICE(0x4007, -1),
217 CH_DEVICE(0x4008, -1),
218 CH_DEVICE(0x4009, -1),
219 CH_DEVICE(0x400a, -1),
220 CH_DEVICE(0x4401, 4),
221 CH_DEVICE(0x4402, 4),
222 CH_DEVICE(0x4403, 4),
223 CH_DEVICE(0x4404, 4),
224 CH_DEVICE(0x4405, 4),
225 CH_DEVICE(0x4406, 4),
226 CH_DEVICE(0x4407, 4),
227 CH_DEVICE(0x4408, 4),
228 CH_DEVICE(0x4409, 4),
229 CH_DEVICE(0x440a, 4),
230 CH_DEVICE(0x440d, 4),
231 CH_DEVICE(0x440e, 4),
235 #define FW_FNAME "cxgb4/t4fw.bin"
236 #define FW5_FNAME "cxgb4/t5fw.bin"
237 #define FW_CFNAME "cxgb4/t4-config.txt"
238 #define FW5_CFNAME "cxgb4/t5-config.txt"
240 MODULE_DESCRIPTION(DRV_DESC);
241 MODULE_AUTHOR("Chelsio Communications");
242 MODULE_LICENSE("Dual BSD/GPL");
243 MODULE_VERSION(DRV_VERSION);
244 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
245 MODULE_FIRMWARE(FW_FNAME);
246 MODULE_FIRMWARE(FW5_FNAME);
249 * Normally we're willing to become the firmware's Master PF but will be happy
250 * if another PF has already become the Master and initialized the adapter.
251 * Setting "force_init" will cause this driver to forcibly establish itself as
252 * the Master PF and initialize the adapter.
254 static uint force_init;
256 module_param(force_init, uint, 0644);
257 MODULE_PARM_DESC(force_init, "Forcibly become Master PF and initialize adapter");
260 * Normally if the firmware we connect to has Configuration File support, we
261 * use that and only fall back to the old Driver-based initialization if the
262 * Configuration File fails for some reason. If force_old_init is set, then
263 * we'll always use the old Driver-based initialization sequence.
265 static uint force_old_init;
267 module_param(force_old_init, uint, 0644);
268 MODULE_PARM_DESC(force_old_init, "Force old initialization sequence");
270 static int dflt_msg_enable = DFLT_MSG_ENABLE;
272 module_param(dflt_msg_enable, int, 0644);
273 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
276 * The driver uses the best interrupt scheme available on a platform in the
277 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
278 * of these schemes the driver may consider as follows:
280 * msi = 2: choose from among all three options
281 * msi = 1: only consider MSI and INTx interrupts
282 * msi = 0: force INTx interrupts
286 module_param(msi, int, 0644);
287 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
290 * Queue interrupt hold-off timer values. Queues default to the first of these
293 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
295 module_param_array(intr_holdoff, uint, NULL, 0644);
296 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
297 "0..4 in microseconds");
299 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
301 module_param_array(intr_cnt, uint, NULL, 0644);
302 MODULE_PARM_DESC(intr_cnt,
303 "thresholds 1..3 for queue interrupt packet counters");
306 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
307 * offset by 2 bytes in order to have the IP headers line up on 4-byte
308 * boundaries. This is a requirement for many architectures which will throw
309 * a machine check fault if an attempt is made to access one of the 4-byte IP
310 * header fields on a non-4-byte boundary. And it's a major performance issue
311 * even on some architectures which allow it like some implementations of the
312 * x86 ISA. However, some architectures don't mind this and for some very
313 * edge-case performance sensitive applications (like forwarding large volumes
314 * of small packets), setting this DMA offset to 0 will decrease the number of
315 * PCI-E Bus transfers enough to measurably affect performance.
317 static int rx_dma_offset = 2;
321 #ifdef CONFIG_PCI_IOV
322 module_param(vf_acls, bool, 0644);
323 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
325 /* Since T5 has more num of PFs, using NUM_OF_PF_WITH_SRIOV_T5
326 * macro as num_vf array size
328 static unsigned int num_vf[NUM_OF_PF_WITH_SRIOV_T5];
330 module_param_array(num_vf, uint, NULL, 0644);
331 MODULE_PARM_DESC(num_vf,
332 "number of VFs for each of PFs 0-3 for T4 and PFs 0-7 for T5");
336 * The filter TCAM has a fixed portion and a variable portion. The fixed
337 * portion can match on source/destination IP IPv4/IPv6 addresses and TCP/UDP
338 * ports. The variable portion is 36 bits which can include things like Exact
339 * Match MAC Index (9 bits), Ether Type (16 bits), IP Protocol (8 bits),
340 * [Inner] VLAN Tag (17 bits), etc. which, if all were somehow selected, would
341 * far exceed the 36-bit budget for this "compressed" header portion of the
342 * filter. Thus, we have a scarce resource which must be carefully managed.
344 * By default we set this up to mostly match the set of filter matching
345 * capabilities of T3 but with accommodations for some of T4's more
346 * interesting features:
348 * { IP Fragment (1), MPS Match Type (3), IP Protocol (8),
349 * [Inner] VLAN (17), Port (3), FCoE (1) }
352 TP_VLAN_PRI_MAP_DEFAULT = HW_TPL_FR_MT_PR_IV_P_FC,
353 TP_VLAN_PRI_MAP_FIRST = FCOE_SHIFT,
354 TP_VLAN_PRI_MAP_LAST = FRAGMENTATION_SHIFT,
357 static unsigned int tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
359 module_param(tp_vlan_pri_map, uint, 0644);
360 MODULE_PARM_DESC(tp_vlan_pri_map, "global compressed filter configuration");
362 static struct dentry *cxgb4_debugfs_root;
364 static LIST_HEAD(adapter_list);
365 static DEFINE_MUTEX(uld_mutex);
366 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
367 static const char *uld_str[] = { "RDMA", "iSCSI" };
369 static void link_report(struct net_device *dev)
371 if (!netif_carrier_ok(dev))
372 netdev_info(dev, "link down\n");
374 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
376 const char *s = "10Mbps";
377 const struct port_info *p = netdev_priv(dev);
379 switch (p->link_cfg.speed) {
391 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
396 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
398 struct net_device *dev = adapter->port[port_id];
400 /* Skip changes from disabled ports. */
401 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
403 netif_carrier_on(dev);
405 netif_carrier_off(dev);
411 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
413 static const char *mod_str[] = {
414 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
417 const struct net_device *dev = adap->port[port_id];
418 const struct port_info *pi = netdev_priv(dev);
420 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
421 netdev_info(dev, "port module unplugged\n");
422 else if (pi->mod_type < ARRAY_SIZE(mod_str))
423 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
427 * Configure the exact and hash address filters to handle a port's multicast
428 * and secondary unicast MAC addresses.
430 static int set_addr_filters(const struct net_device *dev, bool sleep)
438 const struct netdev_hw_addr *ha;
439 int uc_cnt = netdev_uc_count(dev);
440 int mc_cnt = netdev_mc_count(dev);
441 const struct port_info *pi = netdev_priv(dev);
442 unsigned int mb = pi->adapter->fn;
444 /* first do the secondary unicast addresses */
445 netdev_for_each_uc_addr(ha, dev) {
446 addr[naddr++] = ha->addr;
447 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
448 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
449 naddr, addr, filt_idx, &uhash, sleep);
458 /* next set up the multicast addresses */
459 netdev_for_each_mc_addr(ha, dev) {
460 addr[naddr++] = ha->addr;
461 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
462 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
463 naddr, addr, filt_idx, &mhash, sleep);
472 return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
473 uhash | mhash, sleep);
476 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
477 module_param(dbfifo_int_thresh, int, 0644);
478 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
481 * usecs to sleep while draining the dbfifo
483 static int dbfifo_drain_delay = 1000;
484 module_param(dbfifo_drain_delay, int, 0644);
485 MODULE_PARM_DESC(dbfifo_drain_delay,
486 "usecs to sleep while draining the dbfifo");
489 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
490 * If @mtu is -1 it is left unchanged.
492 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
495 struct port_info *pi = netdev_priv(dev);
497 ret = set_addr_filters(dev, sleep_ok);
499 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
500 (dev->flags & IFF_PROMISC) ? 1 : 0,
501 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
506 static struct workqueue_struct *workq;
509 * link_start - enable a port
510 * @dev: the port to enable
512 * Performs the MAC and PHY actions needed to enable a port.
514 static int link_start(struct net_device *dev)
517 struct port_info *pi = netdev_priv(dev);
518 unsigned int mb = pi->adapter->fn;
521 * We do not set address filters and promiscuity here, the stack does
522 * that step explicitly.
524 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
525 !!(dev->features & NETIF_F_HW_VLAN_RX), true);
527 ret = t4_change_mac(pi->adapter, mb, pi->viid,
528 pi->xact_addr_filt, dev->dev_addr, true,
531 pi->xact_addr_filt = ret;
536 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
539 ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
543 /* Clear a filter and release any of its resources that we own. This also
544 * clears the filter's "pending" status.
546 static void clear_filter(struct adapter *adap, struct filter_entry *f)
548 /* If the new or old filter have loopback rewriteing rules then we'll
549 * need to free any existing Layer Two Table (L2T) entries of the old
550 * filter rule. The firmware will handle freeing up any Source MAC
551 * Table (SMT) entries used for rewriting Source MAC Addresses in
555 cxgb4_l2t_release(f->l2t);
557 /* The zeroing of the filter rule below clears the filter valid,
558 * pending, locked flags, l2t pointer, etc. so it's all we need for
561 memset(f, 0, sizeof(*f));
564 /* Handle a filter write/deletion reply.
566 static void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
568 unsigned int idx = GET_TID(rpl);
569 unsigned int nidx = idx - adap->tids.ftid_base;
571 struct filter_entry *f;
573 if (idx >= adap->tids.ftid_base && nidx <
574 (adap->tids.nftids + adap->tids.nsftids)) {
576 ret = GET_TCB_COOKIE(rpl->cookie);
577 f = &adap->tids.ftid_tab[idx];
579 if (ret == FW_FILTER_WR_FLT_DELETED) {
580 /* Clear the filter when we get confirmation from the
581 * hardware that the filter has been deleted.
583 clear_filter(adap, f);
584 } else if (ret == FW_FILTER_WR_SMT_TBL_FULL) {
585 dev_err(adap->pdev_dev, "filter %u setup failed due to full SMT\n",
587 clear_filter(adap, f);
588 } else if (ret == FW_FILTER_WR_FLT_ADDED) {
589 f->smtidx = (be64_to_cpu(rpl->oldval) >> 24) & 0xff;
590 f->pending = 0; /* asynchronous setup completed */
593 /* Something went wrong. Issue a warning about the
594 * problem and clear everything out.
596 dev_err(adap->pdev_dev, "filter %u setup failed with error %u\n",
598 clear_filter(adap, f);
603 /* Response queue handler for the FW event queue.
605 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
606 const struct pkt_gl *gl)
608 u8 opcode = ((const struct rss_header *)rsp)->opcode;
610 rsp++; /* skip RSS header */
611 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
612 const struct cpl_sge_egr_update *p = (void *)rsp;
613 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
616 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
618 if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
619 struct sge_eth_txq *eq;
621 eq = container_of(txq, struct sge_eth_txq, q);
622 netif_tx_wake_queue(eq->txq);
624 struct sge_ofld_txq *oq;
626 oq = container_of(txq, struct sge_ofld_txq, q);
627 tasklet_schedule(&oq->qresume_tsk);
629 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
630 const struct cpl_fw6_msg *p = (void *)rsp;
633 t4_handle_fw_rpl(q->adap, p->data);
634 } else if (opcode == CPL_L2T_WRITE_RPL) {
635 const struct cpl_l2t_write_rpl *p = (void *)rsp;
637 do_l2t_write_rpl(q->adap, p);
638 } else if (opcode == CPL_SET_TCB_RPL) {
639 const struct cpl_set_tcb_rpl *p = (void *)rsp;
641 filter_rpl(q->adap, p);
643 dev_err(q->adap->pdev_dev,
644 "unexpected CPL %#x on FW event queue\n", opcode);
649 * uldrx_handler - response queue handler for ULD queues
650 * @q: the response queue that received the packet
651 * @rsp: the response queue descriptor holding the offload message
652 * @gl: the gather list of packet fragments
654 * Deliver an ingress offload packet to a ULD. All processing is done by
655 * the ULD, we just maintain statistics.
657 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
658 const struct pkt_gl *gl)
660 struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
662 if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
668 else if (gl == CXGB4_MSG_AN)
675 static void disable_msi(struct adapter *adapter)
677 if (adapter->flags & USING_MSIX) {
678 pci_disable_msix(adapter->pdev);
679 adapter->flags &= ~USING_MSIX;
680 } else if (adapter->flags & USING_MSI) {
681 pci_disable_msi(adapter->pdev);
682 adapter->flags &= ~USING_MSI;
687 * Interrupt handler for non-data events used with MSI-X.
689 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
691 struct adapter *adap = cookie;
693 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
696 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
698 t4_slow_intr_handler(adap);
703 * Name the MSI-X interrupts.
705 static void name_msix_vecs(struct adapter *adap)
707 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
709 /* non-data interrupts */
710 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
713 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
714 adap->port[0]->name);
716 /* Ethernet queues */
717 for_each_port(adap, j) {
718 struct net_device *d = adap->port[j];
719 const struct port_info *pi = netdev_priv(d);
721 for (i = 0; i < pi->nqsets; i++, msi_idx++)
722 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
727 for_each_ofldrxq(&adap->sge, i)
728 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
729 adap->port[0]->name, i);
731 for_each_rdmarxq(&adap->sge, i)
732 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
733 adap->port[0]->name, i);
736 static int request_msix_queue_irqs(struct adapter *adap)
738 struct sge *s = &adap->sge;
739 int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi_index = 2;
741 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
742 adap->msix_info[1].desc, &s->fw_evtq);
746 for_each_ethrxq(s, ethqidx) {
747 err = request_irq(adap->msix_info[msi_index].vec,
749 adap->msix_info[msi_index].desc,
750 &s->ethrxq[ethqidx].rspq);
755 for_each_ofldrxq(s, ofldqidx) {
756 err = request_irq(adap->msix_info[msi_index].vec,
758 adap->msix_info[msi_index].desc,
759 &s->ofldrxq[ofldqidx].rspq);
764 for_each_rdmarxq(s, rdmaqidx) {
765 err = request_irq(adap->msix_info[msi_index].vec,
767 adap->msix_info[msi_index].desc,
768 &s->rdmarxq[rdmaqidx].rspq);
776 while (--rdmaqidx >= 0)
777 free_irq(adap->msix_info[--msi_index].vec,
778 &s->rdmarxq[rdmaqidx].rspq);
779 while (--ofldqidx >= 0)
780 free_irq(adap->msix_info[--msi_index].vec,
781 &s->ofldrxq[ofldqidx].rspq);
782 while (--ethqidx >= 0)
783 free_irq(adap->msix_info[--msi_index].vec,
784 &s->ethrxq[ethqidx].rspq);
785 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
789 static void free_msix_queue_irqs(struct adapter *adap)
791 int i, msi_index = 2;
792 struct sge *s = &adap->sge;
794 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
795 for_each_ethrxq(s, i)
796 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
797 for_each_ofldrxq(s, i)
798 free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
799 for_each_rdmarxq(s, i)
800 free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
804 * write_rss - write the RSS table for a given port
806 * @queues: array of queue indices for RSS
808 * Sets up the portion of the HW RSS table for the port's VI to distribute
809 * packets to the Rx queues in @queues.
811 static int write_rss(const struct port_info *pi, const u16 *queues)
815 const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
817 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
821 /* map the queue indices to queue ids */
822 for (i = 0; i < pi->rss_size; i++, queues++)
823 rss[i] = q[*queues].rspq.abs_id;
825 err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
826 pi->rss_size, rss, pi->rss_size);
832 * setup_rss - configure RSS
835 * Sets up RSS for each port.
837 static int setup_rss(struct adapter *adap)
841 for_each_port(adap, i) {
842 const struct port_info *pi = adap2pinfo(adap, i);
844 err = write_rss(pi, pi->rss);
852 * Return the channel of the ingress queue with the given qid.
854 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
856 qid -= p->ingr_start;
857 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
861 * Wait until all NAPI handlers are descheduled.
863 static void quiesce_rx(struct adapter *adap)
867 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
868 struct sge_rspq *q = adap->sge.ingr_map[i];
871 napi_disable(&q->napi);
876 * Enable NAPI scheduling and interrupt generation for all Rx queues.
878 static void enable_rx(struct adapter *adap)
882 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
883 struct sge_rspq *q = adap->sge.ingr_map[i];
888 napi_enable(&q->napi);
889 /* 0-increment GTS to start the timer and enable interrupts */
890 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
891 SEINTARM(q->intr_params) |
892 INGRESSQID(q->cntxt_id));
897 * setup_sge_queues - configure SGE Tx/Rx/response queues
900 * Determines how many sets of SGE queues to use and initializes them.
901 * We support multiple queue sets per port if we have MSI-X, otherwise
902 * just one queue set per port.
904 static int setup_sge_queues(struct adapter *adap)
906 int err, msi_idx, i, j;
907 struct sge *s = &adap->sge;
909 bitmap_zero(s->starving_fl, MAX_EGRQ);
910 bitmap_zero(s->txq_maperr, MAX_EGRQ);
912 if (adap->flags & USING_MSIX)
913 msi_idx = 1; /* vector 0 is for non-queue interrupts */
915 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
919 msi_idx = -((int)s->intrq.abs_id + 1);
922 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
923 msi_idx, NULL, fwevtq_handler);
925 freeout: t4_free_sge_resources(adap);
929 for_each_port(adap, i) {
930 struct net_device *dev = adap->port[i];
931 struct port_info *pi = netdev_priv(dev);
932 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
933 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
935 for (j = 0; j < pi->nqsets; j++, q++) {
938 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
944 memset(&q->stats, 0, sizeof(q->stats));
946 for (j = 0; j < pi->nqsets; j++, t++) {
947 err = t4_sge_alloc_eth_txq(adap, t, dev,
948 netdev_get_tx_queue(dev, j),
949 s->fw_evtq.cntxt_id);
955 j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
956 for_each_ofldrxq(s, i) {
957 struct sge_ofld_rxq *q = &s->ofldrxq[i];
958 struct net_device *dev = adap->port[i / j];
962 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
963 &q->fl, uldrx_handler);
966 memset(&q->stats, 0, sizeof(q->stats));
967 s->ofld_rxq[i] = q->rspq.abs_id;
968 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
969 s->fw_evtq.cntxt_id);
974 for_each_rdmarxq(s, i) {
975 struct sge_ofld_rxq *q = &s->rdmarxq[i];
979 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
980 msi_idx, &q->fl, uldrx_handler);
983 memset(&q->stats, 0, sizeof(q->stats));
984 s->rdma_rxq[i] = q->rspq.abs_id;
987 for_each_port(adap, i) {
989 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
990 * have RDMA queues, and that's the right value.
992 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
994 s->rdmarxq[i].rspq.cntxt_id);
999 t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
1000 RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
1001 QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
1006 * Returns 0 if new FW was successfully loaded, a positive errno if a load was
1007 * started but failed, and a negative errno if flash load couldn't start.
1009 static int upgrade_fw(struct adapter *adap)
1012 u32 vers, exp_major;
1013 const struct fw_hdr *hdr;
1014 const struct firmware *fw;
1015 struct device *dev = adap->pdev_dev;
1018 switch (CHELSIO_CHIP_VERSION(adap->chip)) {
1020 fw_file_name = FW_FNAME;
1021 exp_major = FW_VERSION_MAJOR;
1024 fw_file_name = FW5_FNAME;
1025 exp_major = FW_VERSION_MAJOR_T5;
1028 dev_err(dev, "Unsupported chip type, %x\n", adap->chip);
1032 ret = request_firmware(&fw, fw_file_name, dev);
1034 dev_err(dev, "unable to load firmware image %s, error %d\n",
1039 hdr = (const struct fw_hdr *)fw->data;
1040 vers = ntohl(hdr->fw_ver);
1041 if (FW_HDR_FW_VER_MAJOR_GET(vers) != exp_major) {
1042 ret = -EINVAL; /* wrong major version, won't do */
1047 * If the flash FW is unusable or we found something newer, load it.
1049 if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != exp_major ||
1050 vers > adap->params.fw_vers) {
1051 dev_info(dev, "upgrading firmware ...\n");
1052 ret = t4_fw_upgrade(adap, adap->mbox, fw->data, fw->size,
1056 "firmware upgraded to version %pI4 from %s\n",
1057 &hdr->fw_ver, fw_file_name);
1059 dev_err(dev, "firmware upgrade failed! err=%d\n", -ret);
1062 * Tell our caller that we didn't upgrade the firmware.
1067 out: release_firmware(fw);
1072 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1073 * The allocated memory is cleared.
1075 void *t4_alloc_mem(size_t size)
1077 void *p = kzalloc(size, GFP_KERNEL);
1085 * Free memory allocated through alloc_mem().
1087 static void t4_free_mem(void *addr)
1089 if (is_vmalloc_addr(addr))
1095 /* Send a Work Request to write the filter at a specified index. We construct
1096 * a Firmware Filter Work Request to have the work done and put the indicated
1097 * filter into "pending" mode which will prevent any further actions against
1098 * it till we get a reply from the firmware on the completion status of the
1101 static int set_filter_wr(struct adapter *adapter, int fidx)
1103 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1104 struct sk_buff *skb;
1105 struct fw_filter_wr *fwr;
1108 /* If the new filter requires loopback Destination MAC and/or VLAN
1109 * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
1112 if (f->fs.newdmac || f->fs.newvlan) {
1113 /* allocate L2T entry for new filter */
1114 f->l2t = t4_l2t_alloc_switching(adapter->l2t);
1117 if (t4_l2t_set_switching(adapter, f->l2t, f->fs.vlan,
1118 f->fs.eport, f->fs.dmac)) {
1119 cxgb4_l2t_release(f->l2t);
1125 ftid = adapter->tids.ftid_base + fidx;
1127 skb = alloc_skb(sizeof(*fwr), GFP_KERNEL | __GFP_NOFAIL);
1128 fwr = (struct fw_filter_wr *)__skb_put(skb, sizeof(*fwr));
1129 memset(fwr, 0, sizeof(*fwr));
1131 /* It would be nice to put most of the following in t4_hw.c but most
1132 * of the work is translating the cxgbtool ch_filter_specification
1133 * into the Work Request and the definition of that structure is
1134 * currently in cxgbtool.h which isn't appropriate to pull into the
1135 * common code. We may eventually try to come up with a more neutral
1136 * filter specification structure but for now it's easiest to simply
1137 * put this fairly direct code in line ...
1139 fwr->op_pkd = htonl(FW_WR_OP(FW_FILTER_WR));
1140 fwr->len16_pkd = htonl(FW_WR_LEN16(sizeof(*fwr)/16));
1142 htonl(V_FW_FILTER_WR_TID(ftid) |
1143 V_FW_FILTER_WR_RQTYPE(f->fs.type) |
1144 V_FW_FILTER_WR_NOREPLY(0) |
1145 V_FW_FILTER_WR_IQ(f->fs.iq));
1146 fwr->del_filter_to_l2tix =
1147 htonl(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
1148 V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
1149 V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
1150 V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
1151 V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
1152 V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
1153 V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
1154 V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
1155 V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
1156 f->fs.newvlan == VLAN_REWRITE) |
1157 V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
1158 f->fs.newvlan == VLAN_REWRITE) |
1159 V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
1160 V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
1161 V_FW_FILTER_WR_PRIO(f->fs.prio) |
1162 V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
1163 fwr->ethtype = htons(f->fs.val.ethtype);
1164 fwr->ethtypem = htons(f->fs.mask.ethtype);
1165 fwr->frag_to_ovlan_vldm =
1166 (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
1167 V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
1168 V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.ivlan_vld) |
1169 V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.ovlan_vld) |
1170 V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.ivlan_vld) |
1171 V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.ovlan_vld));
1173 fwr->rx_chan_rx_rpl_iq =
1174 htons(V_FW_FILTER_WR_RX_CHAN(0) |
1175 V_FW_FILTER_WR_RX_RPL_IQ(adapter->sge.fw_evtq.abs_id));
1176 fwr->maci_to_matchtypem =
1177 htonl(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
1178 V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
1179 V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
1180 V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
1181 V_FW_FILTER_WR_PORT(f->fs.val.iport) |
1182 V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
1183 V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
1184 V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
1185 fwr->ptcl = f->fs.val.proto;
1186 fwr->ptclm = f->fs.mask.proto;
1187 fwr->ttyp = f->fs.val.tos;
1188 fwr->ttypm = f->fs.mask.tos;
1189 fwr->ivlan = htons(f->fs.val.ivlan);
1190 fwr->ivlanm = htons(f->fs.mask.ivlan);
1191 fwr->ovlan = htons(f->fs.val.ovlan);
1192 fwr->ovlanm = htons(f->fs.mask.ovlan);
1193 memcpy(fwr->lip, f->fs.val.lip, sizeof(fwr->lip));
1194 memcpy(fwr->lipm, f->fs.mask.lip, sizeof(fwr->lipm));
1195 memcpy(fwr->fip, f->fs.val.fip, sizeof(fwr->fip));
1196 memcpy(fwr->fipm, f->fs.mask.fip, sizeof(fwr->fipm));
1197 fwr->lp = htons(f->fs.val.lport);
1198 fwr->lpm = htons(f->fs.mask.lport);
1199 fwr->fp = htons(f->fs.val.fport);
1200 fwr->fpm = htons(f->fs.mask.fport);
1202 memcpy(fwr->sma, f->fs.smac, sizeof(fwr->sma));
1204 /* Mark the filter as "pending" and ship off the Filter Work Request.
1205 * When we get the Work Request Reply we'll clear the pending status.
1208 set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
1209 t4_ofld_send(adapter, skb);
1213 /* Delete the filter at a specified index.
1215 static int del_filter_wr(struct adapter *adapter, int fidx)
1217 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1218 struct sk_buff *skb;
1219 struct fw_filter_wr *fwr;
1220 unsigned int len, ftid;
1223 ftid = adapter->tids.ftid_base + fidx;
1225 skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
1226 fwr = (struct fw_filter_wr *)__skb_put(skb, len);
1227 t4_mk_filtdelwr(ftid, fwr, adapter->sge.fw_evtq.abs_id);
1229 /* Mark the filter as "pending" and ship off the Filter Work Request.
1230 * When we get the Work Request Reply we'll clear the pending status.
1233 t4_mgmt_tx(adapter, skb);
1237 static inline int is_offload(const struct adapter *adap)
1239 return adap->params.offload;
1243 * Implementation of ethtool operations.
1246 static u32 get_msglevel(struct net_device *dev)
1248 return netdev2adap(dev)->msg_enable;
1251 static void set_msglevel(struct net_device *dev, u32 val)
1253 netdev2adap(dev)->msg_enable = val;
1256 static char stats_strings[][ETH_GSTRING_LEN] = {
1259 "TxBroadcastFrames ",
1260 "TxMulticastFrames ",
1266 "TxFrames128To255 ",
1267 "TxFrames256To511 ",
1268 "TxFrames512To1023 ",
1269 "TxFrames1024To1518 ",
1270 "TxFrames1519ToMax ",
1285 "RxBroadcastFrames ",
1286 "RxMulticastFrames ",
1298 "RxFrames128To255 ",
1299 "RxFrames256To511 ",
1300 "RxFrames512To1023 ",
1301 "RxFrames1024To1518 ",
1302 "RxFrames1519ToMax ",
1314 "RxBG0FramesDropped ",
1315 "RxBG1FramesDropped ",
1316 "RxBG2FramesDropped ",
1317 "RxBG3FramesDropped ",
1318 "RxBG0FramesTrunc ",
1319 "RxBG1FramesTrunc ",
1320 "RxBG2FramesTrunc ",
1321 "RxBG3FramesTrunc ",
1330 "WriteCoalSuccess ",
1334 static int get_sset_count(struct net_device *dev, int sset)
1338 return ARRAY_SIZE(stats_strings);
1344 #define T4_REGMAP_SIZE (160 * 1024)
1345 #define T5_REGMAP_SIZE (332 * 1024)
1347 static int get_regs_len(struct net_device *dev)
1349 struct adapter *adap = netdev2adap(dev);
1350 if (is_t4(adap->chip))
1351 return T4_REGMAP_SIZE;
1353 return T5_REGMAP_SIZE;
1356 static int get_eeprom_len(struct net_device *dev)
1361 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1363 struct adapter *adapter = netdev2adap(dev);
1365 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1366 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1367 strlcpy(info->bus_info, pci_name(adapter->pdev),
1368 sizeof(info->bus_info));
1370 if (adapter->params.fw_vers)
1371 snprintf(info->fw_version, sizeof(info->fw_version),
1372 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1373 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1374 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1375 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1376 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1377 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1378 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1379 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1380 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1383 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1385 if (stringset == ETH_SS_STATS)
1386 memcpy(data, stats_strings, sizeof(stats_strings));
1390 * port stats maintained per queue of the port. They should be in the same
1391 * order as in stats_strings above.
1393 struct queue_port_stats {
1403 static void collect_sge_port_stats(const struct adapter *adap,
1404 const struct port_info *p, struct queue_port_stats *s)
1407 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1408 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1410 memset(s, 0, sizeof(*s));
1411 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1413 s->tx_csum += tx->tx_cso;
1414 s->rx_csum += rx->stats.rx_cso;
1415 s->vlan_ex += rx->stats.vlan_ex;
1416 s->vlan_ins += tx->vlan_ins;
1417 s->gro_pkts += rx->stats.lro_pkts;
1418 s->gro_merged += rx->stats.lro_merged;
1422 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1425 struct port_info *pi = netdev_priv(dev);
1426 struct adapter *adapter = pi->adapter;
1429 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1431 data += sizeof(struct port_stats) / sizeof(u64);
1432 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1433 data += sizeof(struct queue_port_stats) / sizeof(u64);
1434 if (!is_t4(adapter->chip)) {
1435 t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7));
1436 val1 = t4_read_reg(adapter, SGE_STAT_TOTAL);
1437 val2 = t4_read_reg(adapter, SGE_STAT_MATCH);
1438 *data = val1 - val2;
1443 memset(data, 0, 2 * sizeof(u64));
1449 * Return a version number to identify the type of adapter. The scheme is:
1450 * - bits 0..9: chip version
1451 * - bits 10..15: chip revision
1452 * - bits 16..23: register dump version
1454 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1456 return CHELSIO_CHIP_VERSION(ap->chip) |
1457 (CHELSIO_CHIP_RELEASE(ap->chip) << 10) | (1 << 16);
1460 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1463 u32 *p = buf + start;
1465 for ( ; start <= end; start += sizeof(u32))
1466 *p++ = t4_read_reg(ap, start);
1469 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1472 static const unsigned int t4_reg_ranges[] = {
1692 static const unsigned int t5_reg_ranges[] = {
2120 struct adapter *ap = netdev2adap(dev);
2121 static const unsigned int *reg_ranges;
2122 int arr_size = 0, buf_size = 0;
2124 if (is_t4(ap->chip)) {
2125 reg_ranges = &t4_reg_ranges[0];
2126 arr_size = ARRAY_SIZE(t4_reg_ranges);
2127 buf_size = T4_REGMAP_SIZE;
2129 reg_ranges = &t5_reg_ranges[0];
2130 arr_size = ARRAY_SIZE(t5_reg_ranges);
2131 buf_size = T5_REGMAP_SIZE;
2134 regs->version = mk_adap_vers(ap);
2136 memset(buf, 0, buf_size);
2137 for (i = 0; i < arr_size; i += 2)
2138 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
2141 static int restart_autoneg(struct net_device *dev)
2143 struct port_info *p = netdev_priv(dev);
2145 if (!netif_running(dev))
2147 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
2149 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
2153 static int identify_port(struct net_device *dev,
2154 enum ethtool_phys_id_state state)
2157 struct adapter *adap = netdev2adap(dev);
2159 if (state == ETHTOOL_ID_ACTIVE)
2161 else if (state == ETHTOOL_ID_INACTIVE)
2166 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
2169 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
2173 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
2174 type == FW_PORT_TYPE_BT_XAUI) {
2176 if (caps & FW_PORT_CAP_SPEED_100M)
2177 v |= SUPPORTED_100baseT_Full;
2178 if (caps & FW_PORT_CAP_SPEED_1G)
2179 v |= SUPPORTED_1000baseT_Full;
2180 if (caps & FW_PORT_CAP_SPEED_10G)
2181 v |= SUPPORTED_10000baseT_Full;
2182 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
2183 v |= SUPPORTED_Backplane;
2184 if (caps & FW_PORT_CAP_SPEED_1G)
2185 v |= SUPPORTED_1000baseKX_Full;
2186 if (caps & FW_PORT_CAP_SPEED_10G)
2187 v |= SUPPORTED_10000baseKX4_Full;
2188 } else if (type == FW_PORT_TYPE_KR)
2189 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
2190 else if (type == FW_PORT_TYPE_BP_AP)
2191 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2192 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
2193 else if (type == FW_PORT_TYPE_BP4_AP)
2194 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2195 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
2196 SUPPORTED_10000baseKX4_Full;
2197 else if (type == FW_PORT_TYPE_FIBER_XFI ||
2198 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
2199 v |= SUPPORTED_FIBRE;
2201 if (caps & FW_PORT_CAP_ANEG)
2202 v |= SUPPORTED_Autoneg;
2206 static unsigned int to_fw_linkcaps(unsigned int caps)
2210 if (caps & ADVERTISED_100baseT_Full)
2211 v |= FW_PORT_CAP_SPEED_100M;
2212 if (caps & ADVERTISED_1000baseT_Full)
2213 v |= FW_PORT_CAP_SPEED_1G;
2214 if (caps & ADVERTISED_10000baseT_Full)
2215 v |= FW_PORT_CAP_SPEED_10G;
2219 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2221 const struct port_info *p = netdev_priv(dev);
2223 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
2224 p->port_type == FW_PORT_TYPE_BT_XFI ||
2225 p->port_type == FW_PORT_TYPE_BT_XAUI)
2226 cmd->port = PORT_TP;
2227 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
2228 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
2229 cmd->port = PORT_FIBRE;
2230 else if (p->port_type == FW_PORT_TYPE_SFP) {
2231 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
2232 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
2233 cmd->port = PORT_DA;
2235 cmd->port = PORT_FIBRE;
2237 cmd->port = PORT_OTHER;
2239 if (p->mdio_addr >= 0) {
2240 cmd->phy_address = p->mdio_addr;
2241 cmd->transceiver = XCVR_EXTERNAL;
2242 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
2243 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
2245 cmd->phy_address = 0; /* not really, but no better option */
2246 cmd->transceiver = XCVR_INTERNAL;
2247 cmd->mdio_support = 0;
2250 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
2251 cmd->advertising = from_fw_linkcaps(p->port_type,
2252 p->link_cfg.advertising);
2253 ethtool_cmd_speed_set(cmd,
2254 netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
2255 cmd->duplex = DUPLEX_FULL;
2256 cmd->autoneg = p->link_cfg.autoneg;
2262 static unsigned int speed_to_caps(int speed)
2264 if (speed == SPEED_100)
2265 return FW_PORT_CAP_SPEED_100M;
2266 if (speed == SPEED_1000)
2267 return FW_PORT_CAP_SPEED_1G;
2268 if (speed == SPEED_10000)
2269 return FW_PORT_CAP_SPEED_10G;
2273 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2276 struct port_info *p = netdev_priv(dev);
2277 struct link_config *lc = &p->link_cfg;
2278 u32 speed = ethtool_cmd_speed(cmd);
2280 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
2283 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
2285 * PHY offers a single speed. See if that's what's
2288 if (cmd->autoneg == AUTONEG_DISABLE &&
2289 (lc->supported & speed_to_caps(speed)))
2294 if (cmd->autoneg == AUTONEG_DISABLE) {
2295 cap = speed_to_caps(speed);
2297 if (!(lc->supported & cap) || (speed == SPEED_1000) ||
2298 (speed == SPEED_10000))
2300 lc->requested_speed = cap;
2301 lc->advertising = 0;
2303 cap = to_fw_linkcaps(cmd->advertising);
2304 if (!(lc->supported & cap))
2306 lc->requested_speed = 0;
2307 lc->advertising = cap | FW_PORT_CAP_ANEG;
2309 lc->autoneg = cmd->autoneg;
2311 if (netif_running(dev))
2312 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2317 static void get_pauseparam(struct net_device *dev,
2318 struct ethtool_pauseparam *epause)
2320 struct port_info *p = netdev_priv(dev);
2322 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
2323 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
2324 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
2327 static int set_pauseparam(struct net_device *dev,
2328 struct ethtool_pauseparam *epause)
2330 struct port_info *p = netdev_priv(dev);
2331 struct link_config *lc = &p->link_cfg;
2333 if (epause->autoneg == AUTONEG_DISABLE)
2334 lc->requested_fc = 0;
2335 else if (lc->supported & FW_PORT_CAP_ANEG)
2336 lc->requested_fc = PAUSE_AUTONEG;
2340 if (epause->rx_pause)
2341 lc->requested_fc |= PAUSE_RX;
2342 if (epause->tx_pause)
2343 lc->requested_fc |= PAUSE_TX;
2344 if (netif_running(dev))
2345 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2350 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2352 const struct port_info *pi = netdev_priv(dev);
2353 const struct sge *s = &pi->adapter->sge;
2355 e->rx_max_pending = MAX_RX_BUFFERS;
2356 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
2357 e->rx_jumbo_max_pending = 0;
2358 e->tx_max_pending = MAX_TXQ_ENTRIES;
2360 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
2361 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
2362 e->rx_jumbo_pending = 0;
2363 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
2366 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2369 const struct port_info *pi = netdev_priv(dev);
2370 struct adapter *adapter = pi->adapter;
2371 struct sge *s = &adapter->sge;
2373 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
2374 e->tx_pending > MAX_TXQ_ENTRIES ||
2375 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
2376 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
2377 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
2380 if (adapter->flags & FULL_INIT_DONE)
2383 for (i = 0; i < pi->nqsets; ++i) {
2384 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
2385 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
2386 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
2391 static int closest_timer(const struct sge *s, int time)
2393 int i, delta, match = 0, min_delta = INT_MAX;
2395 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
2396 delta = time - s->timer_val[i];
2399 if (delta < min_delta) {
2407 static int closest_thres(const struct sge *s, int thres)
2409 int i, delta, match = 0, min_delta = INT_MAX;
2411 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
2412 delta = thres - s->counter_val[i];
2415 if (delta < min_delta) {
2424 * Return a queue's interrupt hold-off time in us. 0 means no timer.
2426 static unsigned int qtimer_val(const struct adapter *adap,
2427 const struct sge_rspq *q)
2429 unsigned int idx = q->intr_params >> 1;
2431 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
2435 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
2436 * @adap: the adapter
2438 * @us: the hold-off time in us, or 0 to disable timer
2439 * @cnt: the hold-off packet count, or 0 to disable counter
2441 * Sets an Rx queue's interrupt hold-off time and packet count. At least
2442 * one of the two needs to be enabled for the queue to generate interrupts.
2444 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
2445 unsigned int us, unsigned int cnt)
2447 if ((us | cnt) == 0)
2454 new_idx = closest_thres(&adap->sge, cnt);
2455 if (q->desc && q->pktcnt_idx != new_idx) {
2456 /* the queue has already been created, update it */
2457 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
2458 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
2459 FW_PARAMS_PARAM_YZ(q->cntxt_id);
2460 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
2465 q->pktcnt_idx = new_idx;
2468 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
2469 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
2473 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2475 const struct port_info *pi = netdev_priv(dev);
2476 struct adapter *adap = pi->adapter;
2481 for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++) {
2482 q = &adap->sge.ethrxq[i].rspq;
2483 r = set_rxq_intr_params(adap, q, c->rx_coalesce_usecs,
2484 c->rx_max_coalesced_frames);
2486 dev_err(&dev->dev, "failed to set coalesce %d\n", r);
2493 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2495 const struct port_info *pi = netdev_priv(dev);
2496 const struct adapter *adap = pi->adapter;
2497 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
2499 c->rx_coalesce_usecs = qtimer_val(adap, rq);
2500 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
2501 adap->sge.counter_val[rq->pktcnt_idx] : 0;
2506 * eeprom_ptov - translate a physical EEPROM address to virtual
2507 * @phys_addr: the physical EEPROM address
2508 * @fn: the PCI function number
2509 * @sz: size of function-specific area
2511 * Translate a physical EEPROM address to virtual. The first 1K is
2512 * accessed through virtual addresses starting at 31K, the rest is
2513 * accessed through virtual addresses starting at 0.
2515 * The mapping is as follows:
2516 * [0..1K) -> [31K..32K)
2517 * [1K..1K+A) -> [31K-A..31K)
2518 * [1K+A..ES) -> [0..ES-A-1K)
2520 * where A = @fn * @sz, and ES = EEPROM size.
2522 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
2525 if (phys_addr < 1024)
2526 return phys_addr + (31 << 10);
2527 if (phys_addr < 1024 + fn)
2528 return 31744 - fn + phys_addr - 1024;
2529 if (phys_addr < EEPROMSIZE)
2530 return phys_addr - 1024 - fn;
2535 * The next two routines implement eeprom read/write from physical addresses.
2537 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
2539 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2542 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
2543 return vaddr < 0 ? vaddr : 0;
2546 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
2548 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2551 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
2552 return vaddr < 0 ? vaddr : 0;
2555 #define EEPROM_MAGIC 0x38E2F10C
2557 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2561 struct adapter *adapter = netdev2adap(dev);
2563 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2567 e->magic = EEPROM_MAGIC;
2568 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2569 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
2572 memcpy(data, buf + e->offset, e->len);
2577 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2582 u32 aligned_offset, aligned_len, *p;
2583 struct adapter *adapter = netdev2adap(dev);
2585 if (eeprom->magic != EEPROM_MAGIC)
2588 aligned_offset = eeprom->offset & ~3;
2589 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2591 if (adapter->fn > 0) {
2592 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
2594 if (aligned_offset < start ||
2595 aligned_offset + aligned_len > start + EEPROMPFSIZE)
2599 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2601 * RMW possibly needed for first or last words.
2603 buf = kmalloc(aligned_len, GFP_KERNEL);
2606 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
2607 if (!err && aligned_len > 4)
2608 err = eeprom_rd_phys(adapter,
2609 aligned_offset + aligned_len - 4,
2610 (u32 *)&buf[aligned_len - 4]);
2613 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2617 err = t4_seeprom_wp(adapter, false);
2621 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2622 err = eeprom_wr_phys(adapter, aligned_offset, *p);
2623 aligned_offset += 4;
2627 err = t4_seeprom_wp(adapter, true);
2634 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
2637 const struct firmware *fw;
2638 struct adapter *adap = netdev2adap(netdev);
2640 ef->data[sizeof(ef->data) - 1] = '\0';
2641 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
2645 ret = t4_load_fw(adap, fw->data, fw->size);
2646 release_firmware(fw);
2648 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
2652 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2653 #define BCAST_CRC 0xa0ccc1a6
2655 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2657 wol->supported = WAKE_BCAST | WAKE_MAGIC;
2658 wol->wolopts = netdev2adap(dev)->wol;
2659 memset(&wol->sopass, 0, sizeof(wol->sopass));
2662 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2665 struct port_info *pi = netdev_priv(dev);
2667 if (wol->wolopts & ~WOL_SUPPORTED)
2669 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
2670 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
2671 if (wol->wolopts & WAKE_BCAST) {
2672 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
2675 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
2676 ~6ULL, ~0ULL, BCAST_CRC, true);
2678 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
2682 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2684 const struct port_info *pi = netdev_priv(dev);
2685 netdev_features_t changed = dev->features ^ features;
2688 if (!(changed & NETIF_F_HW_VLAN_RX))
2691 err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
2693 !!(features & NETIF_F_HW_VLAN_RX), true);
2695 dev->features = features ^ NETIF_F_HW_VLAN_RX;
2699 static u32 get_rss_table_size(struct net_device *dev)
2701 const struct port_info *pi = netdev_priv(dev);
2703 return pi->rss_size;
2706 static int get_rss_table(struct net_device *dev, u32 *p)
2708 const struct port_info *pi = netdev_priv(dev);
2709 unsigned int n = pi->rss_size;
2716 static int set_rss_table(struct net_device *dev, const u32 *p)
2719 struct port_info *pi = netdev_priv(dev);
2721 for (i = 0; i < pi->rss_size; i++)
2723 if (pi->adapter->flags & FULL_INIT_DONE)
2724 return write_rss(pi, pi->rss);
2728 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
2731 const struct port_info *pi = netdev_priv(dev);
2733 switch (info->cmd) {
2734 case ETHTOOL_GRXFH: {
2735 unsigned int v = pi->rss_mode;
2738 switch (info->flow_type) {
2740 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
2741 info->data = RXH_IP_SRC | RXH_IP_DST |
2742 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2743 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2744 info->data = RXH_IP_SRC | RXH_IP_DST;
2747 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
2748 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2749 info->data = RXH_IP_SRC | RXH_IP_DST |
2750 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2751 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2752 info->data = RXH_IP_SRC | RXH_IP_DST;
2755 case AH_ESP_V4_FLOW:
2757 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2758 info->data = RXH_IP_SRC | RXH_IP_DST;
2761 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
2762 info->data = RXH_IP_SRC | RXH_IP_DST |
2763 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2764 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2765 info->data = RXH_IP_SRC | RXH_IP_DST;
2768 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
2769 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2770 info->data = RXH_IP_SRC | RXH_IP_DST |
2771 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2772 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2773 info->data = RXH_IP_SRC | RXH_IP_DST;
2776 case AH_ESP_V6_FLOW:
2778 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2779 info->data = RXH_IP_SRC | RXH_IP_DST;
2784 case ETHTOOL_GRXRINGS:
2785 info->data = pi->nqsets;
2791 static const struct ethtool_ops cxgb_ethtool_ops = {
2792 .get_settings = get_settings,
2793 .set_settings = set_settings,
2794 .get_drvinfo = get_drvinfo,
2795 .get_msglevel = get_msglevel,
2796 .set_msglevel = set_msglevel,
2797 .get_ringparam = get_sge_param,
2798 .set_ringparam = set_sge_param,
2799 .get_coalesce = get_coalesce,
2800 .set_coalesce = set_coalesce,
2801 .get_eeprom_len = get_eeprom_len,
2802 .get_eeprom = get_eeprom,
2803 .set_eeprom = set_eeprom,
2804 .get_pauseparam = get_pauseparam,
2805 .set_pauseparam = set_pauseparam,
2806 .get_link = ethtool_op_get_link,
2807 .get_strings = get_strings,
2808 .set_phys_id = identify_port,
2809 .nway_reset = restart_autoneg,
2810 .get_sset_count = get_sset_count,
2811 .get_ethtool_stats = get_stats,
2812 .get_regs_len = get_regs_len,
2813 .get_regs = get_regs,
2816 .get_rxnfc = get_rxnfc,
2817 .get_rxfh_indir_size = get_rss_table_size,
2818 .get_rxfh_indir = get_rss_table,
2819 .set_rxfh_indir = set_rss_table,
2820 .flash_device = set_flash,
2826 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2830 loff_t avail = file_inode(file)->i_size;
2831 unsigned int mem = (uintptr_t)file->private_data & 3;
2832 struct adapter *adap = file->private_data - mem;
2838 if (count > avail - pos)
2839 count = avail - pos;
2847 ret = t4_mc_read(adap, pos, data, NULL);
2849 ret = t4_edc_read(adap, mem, pos, data, NULL);
2853 ofst = pos % sizeof(data);
2854 len = min(count, sizeof(data) - ofst);
2855 if (copy_to_user(buf, (u8 *)data + ofst, len))
2862 count = pos - *ppos;
2867 static const struct file_operations mem_debugfs_fops = {
2868 .owner = THIS_MODULE,
2869 .open = simple_open,
2871 .llseek = default_llseek,
2874 static void add_debugfs_mem(struct adapter *adap, const char *name,
2875 unsigned int idx, unsigned int size_mb)
2879 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2880 (void *)adap + idx, &mem_debugfs_fops);
2881 if (de && de->d_inode)
2882 de->d_inode->i_size = size_mb << 20;
2885 static int setup_debugfs(struct adapter *adap)
2889 if (IS_ERR_OR_NULL(adap->debugfs_root))
2892 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2893 if (i & EDRAM0_ENABLE)
2894 add_debugfs_mem(adap, "edc0", MEM_EDC0, 5);
2895 if (i & EDRAM1_ENABLE)
2896 add_debugfs_mem(adap, "edc1", MEM_EDC1, 5);
2897 if (i & EXT_MEM_ENABLE)
2898 add_debugfs_mem(adap, "mc", MEM_MC,
2899 EXT_MEM_SIZE_GET(t4_read_reg(adap, MA_EXT_MEMORY_BAR)));
2901 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2907 * upper-layer driver support
2911 * Allocate an active-open TID and set it to the supplied value.
2913 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2917 spin_lock_bh(&t->atid_lock);
2919 union aopen_entry *p = t->afree;
2921 atid = (p - t->atid_tab) + t->atid_base;
2926 spin_unlock_bh(&t->atid_lock);
2929 EXPORT_SYMBOL(cxgb4_alloc_atid);
2932 * Release an active-open TID.
2934 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2936 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
2938 spin_lock_bh(&t->atid_lock);
2942 spin_unlock_bh(&t->atid_lock);
2944 EXPORT_SYMBOL(cxgb4_free_atid);
2947 * Allocate a server TID and set it to the supplied value.
2949 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
2953 spin_lock_bh(&t->stid_lock);
2954 if (family == PF_INET) {
2955 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
2956 if (stid < t->nstids)
2957 __set_bit(stid, t->stid_bmap);
2961 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
2966 t->stid_tab[stid].data = data;
2967 stid += t->stid_base;
2970 spin_unlock_bh(&t->stid_lock);
2973 EXPORT_SYMBOL(cxgb4_alloc_stid);
2975 /* Allocate a server filter TID and set it to the supplied value.
2977 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
2981 spin_lock_bh(&t->stid_lock);
2982 if (family == PF_INET) {
2983 stid = find_next_zero_bit(t->stid_bmap,
2984 t->nstids + t->nsftids, t->nstids);
2985 if (stid < (t->nstids + t->nsftids))
2986 __set_bit(stid, t->stid_bmap);
2993 t->stid_tab[stid].data = data;
2994 stid += t->stid_base;
2997 spin_unlock_bh(&t->stid_lock);
3000 EXPORT_SYMBOL(cxgb4_alloc_sftid);
3002 /* Release a server TID.
3004 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
3006 stid -= t->stid_base;
3007 spin_lock_bh(&t->stid_lock);
3008 if (family == PF_INET)
3009 __clear_bit(stid, t->stid_bmap);
3011 bitmap_release_region(t->stid_bmap, stid, 2);
3012 t->stid_tab[stid].data = NULL;
3014 spin_unlock_bh(&t->stid_lock);
3016 EXPORT_SYMBOL(cxgb4_free_stid);
3019 * Populate a TID_RELEASE WR. Caller must properly size the skb.
3021 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
3024 struct cpl_tid_release *req;
3026 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
3027 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
3028 INIT_TP_WR(req, tid);
3029 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
3033 * Queue a TID release request and if necessary schedule a work queue to
3036 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
3039 void **p = &t->tid_tab[tid];
3040 struct adapter *adap = container_of(t, struct adapter, tids);
3042 spin_lock_bh(&adap->tid_release_lock);
3043 *p = adap->tid_release_head;
3044 /* Low 2 bits encode the Tx channel number */
3045 adap->tid_release_head = (void **)((uintptr_t)p | chan);
3046 if (!adap->tid_release_task_busy) {
3047 adap->tid_release_task_busy = true;
3048 queue_work(workq, &adap->tid_release_task);
3050 spin_unlock_bh(&adap->tid_release_lock);
3054 * Process the list of pending TID release requests.
3056 static void process_tid_release_list(struct work_struct *work)
3058 struct sk_buff *skb;
3059 struct adapter *adap;
3061 adap = container_of(work, struct adapter, tid_release_task);
3063 spin_lock_bh(&adap->tid_release_lock);
3064 while (adap->tid_release_head) {
3065 void **p = adap->tid_release_head;
3066 unsigned int chan = (uintptr_t)p & 3;
3067 p = (void *)p - chan;
3069 adap->tid_release_head = *p;
3071 spin_unlock_bh(&adap->tid_release_lock);
3073 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
3075 schedule_timeout_uninterruptible(1);
3077 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
3078 t4_ofld_send(adap, skb);
3079 spin_lock_bh(&adap->tid_release_lock);
3081 adap->tid_release_task_busy = false;
3082 spin_unlock_bh(&adap->tid_release_lock);
3086 * Release a TID and inform HW. If we are unable to allocate the release
3087 * message we defer to a work queue.
3089 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
3092 struct sk_buff *skb;
3093 struct adapter *adap = container_of(t, struct adapter, tids);
3095 old = t->tid_tab[tid];
3096 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
3098 t->tid_tab[tid] = NULL;
3099 mk_tid_release(skb, chan, tid);
3100 t4_ofld_send(adap, skb);
3102 cxgb4_queue_tid_release(t, chan, tid);
3104 atomic_dec(&t->tids_in_use);
3106 EXPORT_SYMBOL(cxgb4_remove_tid);
3109 * Allocate and initialize the TID tables. Returns 0 on success.
3111 static int tid_init(struct tid_info *t)
3114 unsigned int stid_bmap_size;
3115 unsigned int natids = t->natids;
3117 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
3118 size = t->ntids * sizeof(*t->tid_tab) +
3119 natids * sizeof(*t->atid_tab) +
3120 t->nstids * sizeof(*t->stid_tab) +
3121 t->nsftids * sizeof(*t->stid_tab) +
3122 stid_bmap_size * sizeof(long) +
3123 t->nftids * sizeof(*t->ftid_tab) +
3124 t->nsftids * sizeof(*t->ftid_tab);
3126 t->tid_tab = t4_alloc_mem(size);
3130 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
3131 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
3132 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
3133 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
3134 spin_lock_init(&t->stid_lock);
3135 spin_lock_init(&t->atid_lock);
3137 t->stids_in_use = 0;
3139 t->atids_in_use = 0;
3140 atomic_set(&t->tids_in_use, 0);
3142 /* Setup the free list for atid_tab and clear the stid bitmap. */
3145 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
3146 t->afree = t->atid_tab;
3148 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
3153 * cxgb4_create_server - create an IP server
3155 * @stid: the server TID
3156 * @sip: local IP address to bind server to
3157 * @sport: the server's TCP port
3158 * @queue: queue to direct messages from this server to
3160 * Create an IP server for the given port and address.
3161 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3163 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
3164 __be32 sip, __be16 sport, __be16 vlan,
3168 struct sk_buff *skb;
3169 struct adapter *adap;
3170 struct cpl_pass_open_req *req;
3172 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3176 adap = netdev2adap(dev);
3177 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
3179 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
3180 req->local_port = sport;
3181 req->peer_port = htons(0);
3182 req->local_ip = sip;
3183 req->peer_ip = htonl(0);
3184 chan = rxq_to_chan(&adap->sge, queue);
3185 req->opt0 = cpu_to_be64(TX_CHAN(chan));
3186 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3187 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3188 return t4_mgmt_tx(adap, skb);
3190 EXPORT_SYMBOL(cxgb4_create_server);
3193 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
3194 * @mtus: the HW MTU table
3195 * @mtu: the target MTU
3196 * @idx: index of selected entry in the MTU table
3198 * Returns the index and the value in the HW MTU table that is closest to
3199 * but does not exceed @mtu, unless @mtu is smaller than any value in the
3200 * table, in which case that smallest available value is selected.
3202 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
3207 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
3213 EXPORT_SYMBOL(cxgb4_best_mtu);
3216 * cxgb4_port_chan - get the HW channel of a port
3217 * @dev: the net device for the port
3219 * Return the HW Tx channel of the given port.
3221 unsigned int cxgb4_port_chan(const struct net_device *dev)
3223 return netdev2pinfo(dev)->tx_chan;
3225 EXPORT_SYMBOL(cxgb4_port_chan);
3227 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
3229 struct adapter *adap = netdev2adap(dev);
3232 v = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3233 return lpfifo ? G_LP_COUNT(v) : G_HP_COUNT(v);
3235 EXPORT_SYMBOL(cxgb4_dbfifo_count);
3238 * cxgb4_port_viid - get the VI id of a port
3239 * @dev: the net device for the port
3241 * Return the VI id of the given port.
3243 unsigned int cxgb4_port_viid(const struct net_device *dev)
3245 return netdev2pinfo(dev)->viid;
3247 EXPORT_SYMBOL(cxgb4_port_viid);
3250 * cxgb4_port_idx - get the index of a port
3251 * @dev: the net device for the port
3253 * Return the index of the given port.
3255 unsigned int cxgb4_port_idx(const struct net_device *dev)
3257 return netdev2pinfo(dev)->port_id;
3259 EXPORT_SYMBOL(cxgb4_port_idx);
3261 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
3262 struct tp_tcp_stats *v6)
3264 struct adapter *adap = pci_get_drvdata(pdev);
3266 spin_lock(&adap->stats_lock);
3267 t4_tp_get_tcp_stats(adap, v4, v6);
3268 spin_unlock(&adap->stats_lock);
3270 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
3272 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
3273 const unsigned int *pgsz_order)
3275 struct adapter *adap = netdev2adap(dev);
3277 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
3278 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
3279 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
3280 HPZ3(pgsz_order[3]));
3282 EXPORT_SYMBOL(cxgb4_iscsi_init);
3284 int cxgb4_flush_eq_cache(struct net_device *dev)
3286 struct adapter *adap = netdev2adap(dev);
3289 ret = t4_fwaddrspace_write(adap, adap->mbox,
3290 0xe1000000 + A_SGE_CTXT_CMD, 0x20000000);
3293 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
3295 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
3297 u32 addr = t4_read_reg(adap, A_SGE_DBQ_CTXT_BADDR) + 24 * qid + 8;
3301 ret = t4_mem_win_read_len(adap, addr, (__be32 *)&indices, 8);
3303 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
3304 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
3309 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
3312 struct adapter *adap = netdev2adap(dev);
3313 u16 hw_pidx, hw_cidx;
3316 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
3320 if (pidx != hw_pidx) {
3323 if (pidx >= hw_pidx)
3324 delta = pidx - hw_pidx;
3326 delta = size - hw_pidx + pidx;
3328 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3329 QID(qid) | PIDX(delta));
3334 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
3336 static struct pci_driver cxgb4_driver;
3338 static void check_neigh_update(struct neighbour *neigh)
3340 const struct device *parent;
3341 const struct net_device *netdev = neigh->dev;
3343 if (netdev->priv_flags & IFF_802_1Q_VLAN)
3344 netdev = vlan_dev_real_dev(netdev);
3345 parent = netdev->dev.parent;
3346 if (parent && parent->driver == &cxgb4_driver.driver)
3347 t4_l2t_update(dev_get_drvdata(parent), neigh);
3350 static int netevent_cb(struct notifier_block *nb, unsigned long event,
3354 case NETEVENT_NEIGH_UPDATE:
3355 check_neigh_update(data);
3357 case NETEVENT_REDIRECT:
3364 static bool netevent_registered;
3365 static struct notifier_block cxgb4_netevent_nb = {
3366 .notifier_call = netevent_cb
3369 static void drain_db_fifo(struct adapter *adap, int usecs)
3374 set_current_state(TASK_UNINTERRUPTIBLE);
3375 schedule_timeout(usecs_to_jiffies(usecs));
3376 v = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3377 if (G_LP_COUNT(v) == 0 && G_HP_COUNT(v) == 0)
3382 static void disable_txq_db(struct sge_txq *q)
3384 spin_lock_irq(&q->db_lock);
3386 spin_unlock_irq(&q->db_lock);
3389 static void enable_txq_db(struct sge_txq *q)
3391 spin_lock_irq(&q->db_lock);
3393 spin_unlock_irq(&q->db_lock);
3396 static void disable_dbs(struct adapter *adap)
3400 for_each_ethrxq(&adap->sge, i)
3401 disable_txq_db(&adap->sge.ethtxq[i].q);
3402 for_each_ofldrxq(&adap->sge, i)
3403 disable_txq_db(&adap->sge.ofldtxq[i].q);
3404 for_each_port(adap, i)
3405 disable_txq_db(&adap->sge.ctrlq[i].q);
3408 static void enable_dbs(struct adapter *adap)
3412 for_each_ethrxq(&adap->sge, i)
3413 enable_txq_db(&adap->sge.ethtxq[i].q);
3414 for_each_ofldrxq(&adap->sge, i)
3415 enable_txq_db(&adap->sge.ofldtxq[i].q);
3416 for_each_port(adap, i)
3417 enable_txq_db(&adap->sge.ctrlq[i].q);
3420 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
3422 u16 hw_pidx, hw_cidx;
3425 spin_lock_bh(&q->db_lock);
3426 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
3429 if (q->db_pidx != hw_pidx) {
3432 if (q->db_pidx >= hw_pidx)
3433 delta = q->db_pidx - hw_pidx;
3435 delta = q->size - hw_pidx + q->db_pidx;
3437 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3438 QID(q->cntxt_id) | PIDX(delta));
3442 spin_unlock_bh(&q->db_lock);
3444 CH_WARN(adap, "DB drop recovery failed.\n");
3446 static void recover_all_queues(struct adapter *adap)
3450 for_each_ethrxq(&adap->sge, i)
3451 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
3452 for_each_ofldrxq(&adap->sge, i)
3453 sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
3454 for_each_port(adap, i)
3455 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
3458 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
3460 mutex_lock(&uld_mutex);
3461 if (adap->uld_handle[CXGB4_ULD_RDMA])
3462 ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
3464 mutex_unlock(&uld_mutex);
3467 static void process_db_full(struct work_struct *work)
3469 struct adapter *adap;
3471 adap = container_of(work, struct adapter, db_full_task);
3473 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
3474 drain_db_fifo(adap, dbfifo_drain_delay);
3475 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3476 DBFIFO_HP_INT | DBFIFO_LP_INT,
3477 DBFIFO_HP_INT | DBFIFO_LP_INT);
3478 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3481 static void process_db_drop(struct work_struct *work)
3483 struct adapter *adap;
3485 adap = container_of(work, struct adapter, db_drop_task);
3487 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_DROPPED_DB, 0);
3489 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
3490 drain_db_fifo(adap, 1);
3491 recover_all_queues(adap);
3495 void t4_db_full(struct adapter *adap)
3497 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3498 DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
3499 queue_work(workq, &adap->db_full_task);
3502 void t4_db_dropped(struct adapter *adap)
3504 queue_work(workq, &adap->db_drop_task);
3507 static void uld_attach(struct adapter *adap, unsigned int uld)
3510 struct cxgb4_lld_info lli;
3513 lli.pdev = adap->pdev;
3514 lli.l2t = adap->l2t;
3515 lli.tids = &adap->tids;
3516 lli.ports = adap->port;
3517 lli.vr = &adap->vres;
3518 lli.mtus = adap->params.mtus;
3519 if (uld == CXGB4_ULD_RDMA) {
3520 lli.rxq_ids = adap->sge.rdma_rxq;
3521 lli.nrxq = adap->sge.rdmaqs;
3522 } else if (uld == CXGB4_ULD_ISCSI) {
3523 lli.rxq_ids = adap->sge.ofld_rxq;
3524 lli.nrxq = adap->sge.ofldqsets;
3526 lli.ntxq = adap->sge.ofldqsets;
3527 lli.nchan = adap->params.nports;
3528 lli.nports = adap->params.nports;
3529 lli.wr_cred = adap->params.ofldq_wr_cred;
3530 lli.adapter_type = adap->params.rev;
3531 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
3532 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
3533 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
3535 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
3536 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
3538 lli.filt_mode = adap->filter_mode;
3539 /* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
3540 for (i = 0; i < NCHAN; i++)
3542 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
3543 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
3544 lli.fw_vers = adap->params.fw_vers;
3545 lli.dbfifo_int_thresh = dbfifo_int_thresh;
3546 lli.sge_pktshift = adap->sge.pktshift;
3547 lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
3549 handle = ulds[uld].add(&lli);
3550 if (IS_ERR(handle)) {
3551 dev_warn(adap->pdev_dev,
3552 "could not attach to the %s driver, error %ld\n",
3553 uld_str[uld], PTR_ERR(handle));
3557 adap->uld_handle[uld] = handle;
3559 if (!netevent_registered) {
3560 register_netevent_notifier(&cxgb4_netevent_nb);
3561 netevent_registered = true;
3564 if (adap->flags & FULL_INIT_DONE)
3565 ulds[uld].state_change(handle, CXGB4_STATE_UP);
3568 static void attach_ulds(struct adapter *adap)
3572 mutex_lock(&uld_mutex);
3573 list_add_tail(&adap->list_node, &adapter_list);
3574 for (i = 0; i < CXGB4_ULD_MAX; i++)
3576 uld_attach(adap, i);
3577 mutex_unlock(&uld_mutex);
3580 static void detach_ulds(struct adapter *adap)
3584 mutex_lock(&uld_mutex);
3585 list_del(&adap->list_node);
3586 for (i = 0; i < CXGB4_ULD_MAX; i++)
3587 if (adap->uld_handle[i]) {
3588 ulds[i].state_change(adap->uld_handle[i],
3589 CXGB4_STATE_DETACH);
3590 adap->uld_handle[i] = NULL;
3592 if (netevent_registered && list_empty(&adapter_list)) {
3593 unregister_netevent_notifier(&cxgb4_netevent_nb);
3594 netevent_registered = false;
3596 mutex_unlock(&uld_mutex);
3599 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
3603 mutex_lock(&uld_mutex);
3604 for (i = 0; i < CXGB4_ULD_MAX; i++)
3605 if (adap->uld_handle[i])
3606 ulds[i].state_change(adap->uld_handle[i], new_state);
3607 mutex_unlock(&uld_mutex);
3611 * cxgb4_register_uld - register an upper-layer driver
3612 * @type: the ULD type
3613 * @p: the ULD methods
3615 * Registers an upper-layer driver with this driver and notifies the ULD
3616 * about any presently available devices that support its type. Returns
3617 * %-EBUSY if a ULD of the same type is already registered.
3619 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
3622 struct adapter *adap;
3624 if (type >= CXGB4_ULD_MAX)
3626 mutex_lock(&uld_mutex);
3627 if (ulds[type].add) {
3632 list_for_each_entry(adap, &adapter_list, list_node)
3633 uld_attach(adap, type);
3634 out: mutex_unlock(&uld_mutex);
3637 EXPORT_SYMBOL(cxgb4_register_uld);
3640 * cxgb4_unregister_uld - unregister an upper-layer driver
3641 * @type: the ULD type
3643 * Unregisters an existing upper-layer driver.
3645 int cxgb4_unregister_uld(enum cxgb4_uld type)
3647 struct adapter *adap;
3649 if (type >= CXGB4_ULD_MAX)
3651 mutex_lock(&uld_mutex);
3652 list_for_each_entry(adap, &adapter_list, list_node)
3653 adap->uld_handle[type] = NULL;
3654 ulds[type].add = NULL;
3655 mutex_unlock(&uld_mutex);
3658 EXPORT_SYMBOL(cxgb4_unregister_uld);
3661 * cxgb_up - enable the adapter
3662 * @adap: adapter being enabled
3664 * Called when the first port is enabled, this function performs the
3665 * actions necessary to make an adapter operational, such as completing
3666 * the initialization of HW modules, and enabling interrupts.
3668 * Must be called with the rtnl lock held.
3670 static int cxgb_up(struct adapter *adap)
3674 err = setup_sge_queues(adap);
3677 err = setup_rss(adap);
3681 if (adap->flags & USING_MSIX) {
3682 name_msix_vecs(adap);
3683 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
3684 adap->msix_info[0].desc, adap);
3688 err = request_msix_queue_irqs(adap);
3690 free_irq(adap->msix_info[0].vec, adap);
3694 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
3695 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
3696 adap->port[0]->name, adap);
3702 t4_intr_enable(adap);
3703 adap->flags |= FULL_INIT_DONE;
3704 notify_ulds(adap, CXGB4_STATE_UP);
3708 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
3710 t4_free_sge_resources(adap);
3714 static void cxgb_down(struct adapter *adapter)
3716 t4_intr_disable(adapter);
3717 cancel_work_sync(&adapter->tid_release_task);
3718 cancel_work_sync(&adapter->db_full_task);
3719 cancel_work_sync(&adapter->db_drop_task);
3720 adapter->tid_release_task_busy = false;
3721 adapter->tid_release_head = NULL;
3723 if (adapter->flags & USING_MSIX) {
3724 free_msix_queue_irqs(adapter);
3725 free_irq(adapter->msix_info[0].vec, adapter);
3727 free_irq(adapter->pdev->irq, adapter);
3728 quiesce_rx(adapter);
3729 t4_sge_stop(adapter);
3730 t4_free_sge_resources(adapter);
3731 adapter->flags &= ~FULL_INIT_DONE;
3735 * net_device operations
3737 static int cxgb_open(struct net_device *dev)
3740 struct port_info *pi = netdev_priv(dev);
3741 struct adapter *adapter = pi->adapter;
3743 netif_carrier_off(dev);
3745 if (!(adapter->flags & FULL_INIT_DONE)) {
3746 err = cxgb_up(adapter);
3751 err = link_start(dev);
3753 netif_tx_start_all_queues(dev);
3757 static int cxgb_close(struct net_device *dev)
3759 struct port_info *pi = netdev_priv(dev);
3760 struct adapter *adapter = pi->adapter;
3762 netif_tx_stop_all_queues(dev);
3763 netif_carrier_off(dev);
3764 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
3767 /* Return an error number if the indicated filter isn't writable ...
3769 static int writable_filter(struct filter_entry *f)
3779 /* Delete the filter at the specified index (if valid). The checks for all
3780 * the common problems with doing this like the filter being locked, currently
3781 * pending in another operation, etc.
3783 static int delete_filter(struct adapter *adapter, unsigned int fidx)
3785 struct filter_entry *f;
3788 if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
3791 f = &adapter->tids.ftid_tab[fidx];
3792 ret = writable_filter(f);
3796 return del_filter_wr(adapter, fidx);
3801 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
3802 __be32 sip, __be16 sport, __be16 vlan,
3803 unsigned int queue, unsigned char port, unsigned char mask)
3806 struct filter_entry *f;
3807 struct adapter *adap;
3811 adap = netdev2adap(dev);
3813 /* Adjust stid to correct filter index */
3814 stid -= adap->tids.nstids;
3815 stid += adap->tids.nftids;
3817 /* Check to make sure the filter requested is writable ...
3819 f = &adap->tids.ftid_tab[stid];
3820 ret = writable_filter(f);
3824 /* Clear out any old resources being used by the filter before
3825 * we start constructing the new filter.
3828 clear_filter(adap, f);
3830 /* Clear out filter specifications */
3831 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
3832 f->fs.val.lport = cpu_to_be16(sport);
3833 f->fs.mask.lport = ~0;
3835 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
3836 for (i = 0; i < 4; i++) {
3837 f->fs.val.lip[i] = val[i];
3838 f->fs.mask.lip[i] = ~0;
3840 if (adap->filter_mode & F_PORT) {
3841 f->fs.val.iport = port;
3842 f->fs.mask.iport = mask;
3848 /* Mark filter as locked */
3852 ret = set_filter_wr(adap, stid);
3854 clear_filter(adap, f);
3860 EXPORT_SYMBOL(cxgb4_create_server_filter);
3862 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
3863 unsigned int queue, bool ipv6)
3866 struct filter_entry *f;
3867 struct adapter *adap;
3869 adap = netdev2adap(dev);
3871 /* Adjust stid to correct filter index */
3872 stid -= adap->tids.nstids;
3873 stid += adap->tids.nftids;
3875 f = &adap->tids.ftid_tab[stid];
3876 /* Unlock the filter */
3879 ret = delete_filter(adap, stid);
3885 EXPORT_SYMBOL(cxgb4_remove_server_filter);
3887 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
3888 struct rtnl_link_stats64 *ns)
3890 struct port_stats stats;
3891 struct port_info *p = netdev_priv(dev);
3892 struct adapter *adapter = p->adapter;
3894 spin_lock(&adapter->stats_lock);
3895 t4_get_port_stats(adapter, p->tx_chan, &stats);
3896 spin_unlock(&adapter->stats_lock);
3898 ns->tx_bytes = stats.tx_octets;
3899 ns->tx_packets = stats.tx_frames;
3900 ns->rx_bytes = stats.rx_octets;
3901 ns->rx_packets = stats.rx_frames;
3902 ns->multicast = stats.rx_mcast_frames;
3904 /* detailed rx_errors */
3905 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
3907 ns->rx_over_errors = 0;
3908 ns->rx_crc_errors = stats.rx_fcs_err;
3909 ns->rx_frame_errors = stats.rx_symbol_err;
3910 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
3911 stats.rx_ovflow2 + stats.rx_ovflow3 +
3912 stats.rx_trunc0 + stats.rx_trunc1 +
3913 stats.rx_trunc2 + stats.rx_trunc3;
3914 ns->rx_missed_errors = 0;
3916 /* detailed tx_errors */
3917 ns->tx_aborted_errors = 0;
3918 ns->tx_carrier_errors = 0;
3919 ns->tx_fifo_errors = 0;
3920 ns->tx_heartbeat_errors = 0;
3921 ns->tx_window_errors = 0;
3923 ns->tx_errors = stats.tx_error_frames;
3924 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
3925 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
3929 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
3932 int ret = 0, prtad, devad;
3933 struct port_info *pi = netdev_priv(dev);
3934 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
3938 if (pi->mdio_addr < 0)
3940 data->phy_id = pi->mdio_addr;
3944 if (mdio_phy_id_is_c45(data->phy_id)) {
3945 prtad = mdio_phy_id_prtad(data->phy_id);
3946 devad = mdio_phy_id_devad(data->phy_id);
3947 } else if (data->phy_id < 32) {
3948 prtad = data->phy_id;
3950 data->reg_num &= 0x1f;
3954 mbox = pi->adapter->fn;
3955 if (cmd == SIOCGMIIREG)
3956 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
3957 data->reg_num, &data->val_out);
3959 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
3960 data->reg_num, data->val_in);
3968 static void cxgb_set_rxmode(struct net_device *dev)
3970 /* unfortunately we can't return errors to the stack */
3971 set_rxmode(dev, -1, false);
3974 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
3977 struct port_info *pi = netdev_priv(dev);
3979 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
3981 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
3988 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
3991 struct sockaddr *addr = p;
3992 struct port_info *pi = netdev_priv(dev);
3994 if (!is_valid_ether_addr(addr->sa_data))
3995 return -EADDRNOTAVAIL;
3997 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
3998 pi->xact_addr_filt, addr->sa_data, true, true);
4002 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4003 pi->xact_addr_filt = ret;
4007 #ifdef CONFIG_NET_POLL_CONTROLLER
4008 static void cxgb_netpoll(struct net_device *dev)
4010 struct port_info *pi = netdev_priv(dev);
4011 struct adapter *adap = pi->adapter;
4013 if (adap->flags & USING_MSIX) {
4015 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
4017 for (i = pi->nqsets; i; i--, rx++)
4018 t4_sge_intr_msix(0, &rx->rspq);
4020 t4_intr_handler(adap)(0, adap);
4024 static const struct net_device_ops cxgb4_netdev_ops = {
4025 .ndo_open = cxgb_open,
4026 .ndo_stop = cxgb_close,
4027 .ndo_start_xmit = t4_eth_xmit,
4028 .ndo_get_stats64 = cxgb_get_stats,
4029 .ndo_set_rx_mode = cxgb_set_rxmode,
4030 .ndo_set_mac_address = cxgb_set_mac_addr,
4031 .ndo_set_features = cxgb_set_features,
4032 .ndo_validate_addr = eth_validate_addr,
4033 .ndo_do_ioctl = cxgb_ioctl,
4034 .ndo_change_mtu = cxgb_change_mtu,
4035 #ifdef CONFIG_NET_POLL_CONTROLLER
4036 .ndo_poll_controller = cxgb_netpoll,
4040 void t4_fatal_err(struct adapter *adap)
4042 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
4043 t4_intr_disable(adap);
4044 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
4047 static void setup_memwin(struct adapter *adap)
4051 bar0 = pci_resource_start(adap->pdev, 0); /* truncation intentional */
4052 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
4053 (bar0 + MEMWIN0_BASE) | BIR(0) |
4054 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
4055 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
4056 (bar0 + MEMWIN1_BASE) | BIR(0) |
4057 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
4058 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
4059 (bar0 + MEMWIN2_BASE) | BIR(0) |
4060 WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
4063 static void setup_memwin_rdma(struct adapter *adap)
4065 if (adap->vres.ocq.size) {
4066 unsigned int start, sz_kb;
4068 start = pci_resource_start(adap->pdev, 2) +
4069 OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
4070 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
4072 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
4073 start | BIR(1) | WINDOW(ilog2(sz_kb)));
4075 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
4076 adap->vres.ocq.start);
4078 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
4082 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
4087 /* get device capabilities */
4088 memset(c, 0, sizeof(*c));
4089 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4090 FW_CMD_REQUEST | FW_CMD_READ);
4091 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
4092 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
4096 /* select capabilities we'll be using */
4097 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4099 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4101 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4102 } else if (vf_acls) {
4103 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
4106 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4107 FW_CMD_REQUEST | FW_CMD_WRITE);
4108 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
4112 ret = t4_config_glbl_rss(adap, adap->fn,
4113 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4114 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4115 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
4119 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
4120 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
4126 /* tweak some settings */
4127 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
4128 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
4129 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
4130 v = t4_read_reg(adap, TP_PIO_DATA);
4131 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
4133 /* first 4 Tx modulation queues point to consecutive Tx channels */
4134 adap->params.tp.tx_modq_map = 0xE4;
4135 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
4136 V_TX_MOD_QUEUE_REQ_MAP(adap->params.tp.tx_modq_map));
4138 /* associate each Tx modulation queue with consecutive Tx channels */
4140 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4141 &v, 1, A_TP_TX_SCHED_HDR);
4142 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4143 &v, 1, A_TP_TX_SCHED_FIFO);
4144 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4145 &v, 1, A_TP_TX_SCHED_PCMD);
4147 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4148 if (is_offload(adap)) {
4149 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0,
4150 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4151 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4152 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4153 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4154 t4_write_reg(adap, A_TP_TX_MOD_CHANNEL_WEIGHT,
4155 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4156 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4157 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4158 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4161 /* get basic stuff going */
4162 return t4_early_init(adap, adap->fn);
4166 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
4168 #define MAX_ATIDS 8192U
4171 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4173 * If the firmware we're dealing with has Configuration File support, then
4174 * we use that to perform all configuration
4178 * Tweak configuration based on module parameters, etc. Most of these have
4179 * defaults assigned to them by Firmware Configuration Files (if we're using
4180 * them) but need to be explicitly set if we're using hard-coded
4181 * initialization. But even in the case of using Firmware Configuration
4182 * Files, we'd like to expose the ability to change these via module
4183 * parameters so these are essentially common tweaks/settings for
4184 * Configuration Files and hard-coded initialization ...
4186 static int adap_init0_tweaks(struct adapter *adapter)
4189 * Fix up various Host-Dependent Parameters like Page Size, Cache
4190 * Line Size, etc. The firmware default is for a 4KB Page Size and
4191 * 64B Cache Line Size ...
4193 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4196 * Process module parameters which affect early initialization.
4198 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4199 dev_err(&adapter->pdev->dev,
4200 "Ignoring illegal rx_dma_offset=%d, using 2\n",
4204 t4_set_reg_field(adapter, SGE_CONTROL,
4206 PKTSHIFT(rx_dma_offset));
4209 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4210 * adds the pseudo header itself.
4212 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG,
4213 CSUM_HAS_PSEUDO_HDR, 0);
4219 * Attempt to initialize the adapter via a Firmware Configuration File.
4221 static int adap_init0_config(struct adapter *adapter, int reset)
4223 struct fw_caps_config_cmd caps_cmd;
4224 const struct firmware *cf;
4225 unsigned long mtype = 0, maddr = 0;
4226 u32 finiver, finicsum, cfcsum;
4227 int ret, using_flash;
4228 char *fw_config_file, fw_config_file_path[256];
4231 * Reset device if necessary.
4234 ret = t4_fw_reset(adapter, adapter->mbox,
4235 PIORSTMODE | PIORST);
4241 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4242 * then use that. Otherwise, use the configuration file stored
4243 * in the adapter flash ...
4245 switch (CHELSIO_CHIP_VERSION(adapter->chip)) {
4247 fw_config_file = FW_CFNAME;
4250 fw_config_file = FW5_CFNAME;
4253 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
4254 adapter->pdev->device);
4259 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
4262 mtype = FW_MEMTYPE_CF_FLASH;
4263 maddr = t4_flash_cfg_addr(adapter);
4265 u32 params[7], val[7];
4268 if (cf->size >= FLASH_CFG_MAX_SIZE)
4271 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4272 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4273 ret = t4_query_params(adapter, adapter->mbox,
4274 adapter->fn, 0, 1, params, val);
4277 * For t4_memory_write() below addresses and
4278 * sizes have to be in terms of multiples of 4
4279 * bytes. So, if the Configuration File isn't
4280 * a multiple of 4 bytes in length we'll have
4281 * to write that out separately since we can't
4282 * guarantee that the bytes following the
4283 * residual byte in the buffer returned by
4284 * request_firmware() are zeroed out ...
4286 size_t resid = cf->size & 0x3;
4287 size_t size = cf->size & ~0x3;
4288 __be32 *data = (__be32 *)cf->data;
4290 mtype = FW_PARAMS_PARAM_Y_GET(val[0]);
4291 maddr = FW_PARAMS_PARAM_Z_GET(val[0]) << 16;
4293 ret = t4_memory_write(adapter, mtype, maddr,
4295 if (ret == 0 && resid != 0) {
4302 last.word = data[size >> 2];
4303 for (i = resid; i < 4; i++)
4305 ret = t4_memory_write(adapter, mtype,
4312 release_firmware(cf);
4318 * Issue a Capability Configuration command to the firmware to get it
4319 * to parse the Configuration File. We don't use t4_fw_config_file()
4320 * because we want the ability to modify various features after we've
4321 * processed the configuration file ...
4323 memset(&caps_cmd, 0, sizeof(caps_cmd));
4324 caps_cmd.op_to_write =
4325 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4328 caps_cmd.cfvalid_to_len16 =
4329 htonl(FW_CAPS_CONFIG_CMD_CFVALID |
4330 FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4331 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
4332 FW_LEN16(caps_cmd));
4333 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4338 finiver = ntohl(caps_cmd.finiver);
4339 finicsum = ntohl(caps_cmd.finicsum);
4340 cfcsum = ntohl(caps_cmd.cfcsum);
4341 if (finicsum != cfcsum)
4342 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4343 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4347 * And now tell the firmware to use the configuration we just loaded.
4349 caps_cmd.op_to_write =
4350 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4353 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4354 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4360 * Tweak configuration based on system architecture, module
4363 ret = adap_init0_tweaks(adapter);
4368 * And finally tell the firmware to initialize itself using the
4369 * parameters from the Configuration File.
4371 ret = t4_fw_initialize(adapter, adapter->mbox);
4375 sprintf(fw_config_file_path, "/lib/firmware/%s", fw_config_file);
4377 * Return successfully and note that we're operating with parameters
4378 * not supplied by the driver, rather than from hard-wired
4379 * initialization constants burried in the driver.
4381 adapter->flags |= USING_SOFT_PARAMS;
4382 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4383 "Configuration File %s, version %#x, computed checksum %#x\n",
4386 : fw_config_file_path),
4391 * Something bad happened. Return the error ... (If the "error"
4392 * is that there's no Configuration File on the adapter we don't
4393 * want to issue a warning since this is fairly common.)
4397 dev_warn(adapter->pdev_dev, "Configuration file error %d\n",
4403 * Attempt to initialize the adapter via hard-coded, driver supplied
4406 static int adap_init0_no_config(struct adapter *adapter, int reset)
4408 struct sge *s = &adapter->sge;
4409 struct fw_caps_config_cmd caps_cmd;
4414 * Reset device if necessary
4417 ret = t4_fw_reset(adapter, adapter->mbox,
4418 PIORSTMODE | PIORST);
4424 * Get device capabilities and select which we'll be using.
4426 memset(&caps_cmd, 0, sizeof(caps_cmd));
4427 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4428 FW_CMD_REQUEST | FW_CMD_READ);
4429 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4430 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4435 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4437 caps_cmd.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4439 caps_cmd.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4440 } else if (vf_acls) {
4441 dev_err(adapter->pdev_dev, "virtualization ACLs not supported");
4444 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4445 FW_CMD_REQUEST | FW_CMD_WRITE);
4446 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4452 * Tweak configuration based on system architecture, module
4455 ret = adap_init0_tweaks(adapter);
4460 * Select RSS Global Mode we want to use. We use "Basic Virtual"
4461 * mode which maps each Virtual Interface to its own section of
4462 * the RSS Table and we turn on all map and hash enables ...
4464 adapter->flags |= RSS_TNLALLLOOKUP;
4465 ret = t4_config_glbl_rss(adapter, adapter->mbox,
4466 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4467 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4468 FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ |
4469 ((adapter->flags & RSS_TNLALLLOOKUP) ?
4470 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP : 0));
4475 * Set up our own fundamental resource provisioning ...
4477 ret = t4_cfg_pfvf(adapter, adapter->mbox, adapter->fn, 0,
4478 PFRES_NEQ, PFRES_NETHCTRL,
4479 PFRES_NIQFLINT, PFRES_NIQ,
4480 PFRES_TC, PFRES_NVI,
4481 FW_PFVF_CMD_CMASK_MASK,
4482 pfvfres_pmask(adapter, adapter->fn, 0),
4484 PFRES_R_CAPS, PFRES_WX_CAPS);
4489 * Perform low level SGE initialization. We need to do this before we
4490 * send the firmware the INITIALIZE command because that will cause
4491 * any other PF Drivers which are waiting for the Master
4492 * Initialization to proceed forward.
4494 for (i = 0; i < SGE_NTIMERS - 1; i++)
4495 s->timer_val[i] = min(intr_holdoff[i], MAX_SGE_TIMERVAL);
4496 s->timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
4497 s->counter_val[0] = 1;
4498 for (i = 1; i < SGE_NCOUNTERS; i++)
4499 s->counter_val[i] = min(intr_cnt[i - 1],
4500 THRESHOLD_0_GET(THRESHOLD_0_MASK));
4501 t4_sge_init(adapter);
4503 #ifdef CONFIG_PCI_IOV
4505 * Provision resource limits for Virtual Functions. We currently
4506 * grant them all the same static resource limits except for the Port
4507 * Access Rights Mask which we're assigning based on the PF. All of
4508 * the static provisioning stuff for both the PF and VF really needs
4509 * to be managed in a persistent manner for each device which the
4510 * firmware controls.
4514 int max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
4515 NUM_OF_PF_WITH_SRIOV_T5;
4517 for (pf = 0; pf < max_no_pf; pf++) {
4518 if (num_vf[pf] <= 0)
4521 /* VF numbering starts at 1! */
4522 for (vf = 1; vf <= num_vf[pf]; vf++) {
4523 ret = t4_cfg_pfvf(adapter, adapter->mbox,
4525 VFRES_NEQ, VFRES_NETHCTRL,
4526 VFRES_NIQFLINT, VFRES_NIQ,
4527 VFRES_TC, VFRES_NVI,
4528 FW_PFVF_CMD_CMASK_MASK,
4532 VFRES_R_CAPS, VFRES_WX_CAPS);
4534 dev_warn(adapter->pdev_dev,
4536 "provision pf/vf=%d/%d; "
4537 "err=%d\n", pf, vf, ret);
4544 * Set up the default filter mode. Later we'll want to implement this
4545 * via a firmware command, etc. ... This needs to be done before the
4546 * firmare initialization command ... If the selected set of fields
4547 * isn't equal to the default value, we'll need to make sure that the
4548 * field selections will fit in the 36-bit budget.
4550 if (tp_vlan_pri_map != TP_VLAN_PRI_MAP_DEFAULT) {
4553 for (j = TP_VLAN_PRI_MAP_FIRST; j <= TP_VLAN_PRI_MAP_LAST; j++)
4554 switch (tp_vlan_pri_map & (1 << j)) {
4556 /* compressed filter field not enabled */
4576 case ETHERTYPE_MASK:
4582 case MPSHITTYPE_MASK:
4585 case FRAGMENTATION_MASK:
4591 dev_err(adapter->pdev_dev,
4592 "tp_vlan_pri_map=%#x needs %d bits > 36;"\
4593 " using %#x\n", tp_vlan_pri_map, bits,
4594 TP_VLAN_PRI_MAP_DEFAULT);
4595 tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
4598 v = tp_vlan_pri_map;
4599 t4_write_indirect(adapter, TP_PIO_ADDR, TP_PIO_DATA,
4600 &v, 1, TP_VLAN_PRI_MAP);
4603 * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
4604 * to support any of the compressed filter fields above. Newer
4605 * versions of the firmware do this automatically but it doesn't hurt
4606 * to set it here. Meanwhile, we do _not_ need to set Lookup Every
4607 * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
4608 * since the firmware automatically turns this on and off when we have
4609 * a non-zero number of filters active (since it does have a
4610 * performance impact).
4612 if (tp_vlan_pri_map)
4613 t4_set_reg_field(adapter, TP_GLOBAL_CONFIG,
4614 FIVETUPLELOOKUP_MASK,
4615 FIVETUPLELOOKUP_MASK);
4618 * Tweak some settings.
4620 t4_write_reg(adapter, TP_SHIFT_CNT, SYNSHIFTMAX(6) |
4621 RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
4622 PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
4623 KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
4626 * Get basic stuff going by issuing the Firmware Initialize command.
4627 * Note that this _must_ be after all PFVF commands ...
4629 ret = t4_fw_initialize(adapter, adapter->mbox);
4634 * Return successfully!
4636 dev_info(adapter->pdev_dev, "Successfully configured using built-in "\
4637 "driver parameters\n");
4641 * Something bad happened. Return the error ...
4648 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4650 static int adap_init0(struct adapter *adap)
4654 enum dev_state state;
4655 u32 params[7], val[7];
4656 struct fw_caps_config_cmd caps_cmd;
4660 * Contact FW, advertising Master capability (and potentially forcing
4661 * ourselves as the Master PF if our module parameter force_init is
4664 ret = t4_fw_hello(adap, adap->mbox, adap->fn,
4665 force_init ? MASTER_MUST : MASTER_MAY,
4668 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4672 if (ret == adap->mbox)
4673 adap->flags |= MASTER_PF;
4674 if (force_init && state == DEV_STATE_INIT)
4675 state = DEV_STATE_UNINIT;
4678 * If we're the Master PF Driver and the device is uninitialized,
4679 * then let's consider upgrading the firmware ... (We always want
4680 * to check the firmware version number in order to A. get it for
4681 * later reporting and B. to warn if the currently loaded firmware
4682 * is excessively mismatched relative to the driver.)
4684 ret = t4_check_fw_version(adap);
4685 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
4686 if (ret == -EINVAL || ret > 0) {
4687 if (upgrade_fw(adap) >= 0) {
4689 * Note that the chip was reset as part of the
4690 * firmware upgrade so we don't reset it again
4691 * below and grab the new firmware version.
4694 ret = t4_check_fw_version(adap);
4702 * Grab VPD parameters. This should be done after we establish a
4703 * connection to the firmware since some of the VPD parameters
4704 * (notably the Core Clock frequency) are retrieved via requests to
4705 * the firmware. On the other hand, we need these fairly early on
4706 * so we do this right after getting ahold of the firmware.
4708 ret = get_vpd_params(adap, &adap->params.vpd);
4713 * Find out what ports are available to us. Note that we need to do
4714 * this before calling adap_init0_no_config() since it needs nports
4718 FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4719 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
4720 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1, &v, &port_vec);
4724 adap->params.nports = hweight32(port_vec);
4725 adap->params.portvec = port_vec;
4728 * If the firmware is initialized already (and we're not forcing a
4729 * master initialization), note that we're living with existing
4730 * adapter parameters. Otherwise, it's time to try initializing the
4733 if (state == DEV_STATE_INIT) {
4734 dev_info(adap->pdev_dev, "Coming up as %s: "\
4735 "Adapter already initialized\n",
4736 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
4737 adap->flags |= USING_SOFT_PARAMS;
4739 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4740 "Initializing adapter\n");
4743 * If the firmware doesn't support Configuration
4744 * Files warn user and exit,
4747 dev_warn(adap->pdev_dev, "Firmware doesn't support "
4748 "configuration file.\n");
4750 ret = adap_init0_no_config(adap, reset);
4753 * Find out whether we're dealing with a version of
4754 * the firmware which has configuration file support.
4756 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4757 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4758 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1,
4762 * If the firmware doesn't support Configuration
4763 * Files, use the old Driver-based, hard-wired
4764 * initialization. Otherwise, try using the
4765 * Configuration File support and fall back to the
4766 * Driver-based initialization if there's no
4767 * Configuration File found.
4770 ret = adap_init0_no_config(adap, reset);
4773 * The firmware provides us with a memory
4774 * buffer where we can load a Configuration
4775 * File from the host if we want to override
4776 * the Configuration File in flash.
4779 ret = adap_init0_config(adap, reset);
4780 if (ret == -ENOENT) {
4781 dev_info(adap->pdev_dev,
4782 "No Configuration File present "
4783 "on adapter. Using hard-wired "
4784 "configuration parameters.\n");
4785 ret = adap_init0_no_config(adap, reset);
4790 dev_err(adap->pdev_dev,
4791 "could not initialize adapter, error %d\n",
4798 * If we're living with non-hard-coded parameters (either from a
4799 * Firmware Configuration File or values programmed by a different PF
4800 * Driver), give the SGE code a chance to pull in anything that it
4801 * needs ... Note that this must be called after we retrieve our VPD
4802 * parameters in order to know how to convert core ticks to seconds.
4804 if (adap->flags & USING_SOFT_PARAMS) {
4805 ret = t4_sge_init(adap);
4810 if (is_bypass_device(adap->pdev->device))
4811 adap->params.bypass = 1;
4814 * Grab some of our basic fundamental operating parameters.
4816 #define FW_PARAM_DEV(param) \
4817 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
4818 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
4820 #define FW_PARAM_PFVF(param) \
4821 FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
4822 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)| \
4823 FW_PARAMS_PARAM_Y(0) | \
4824 FW_PARAMS_PARAM_Z(0)
4826 params[0] = FW_PARAM_PFVF(EQ_START);
4827 params[1] = FW_PARAM_PFVF(L2T_START);
4828 params[2] = FW_PARAM_PFVF(L2T_END);
4829 params[3] = FW_PARAM_PFVF(FILTER_START);
4830 params[4] = FW_PARAM_PFVF(FILTER_END);
4831 params[5] = FW_PARAM_PFVF(IQFLINT_START);
4832 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6, params, val);
4835 adap->sge.egr_start = val[0];
4836 adap->l2t_start = val[1];
4837 adap->l2t_end = val[2];
4838 adap->tids.ftid_base = val[3];
4839 adap->tids.nftids = val[4] - val[3] + 1;
4840 adap->sge.ingr_start = val[5];
4842 /* query params related to active filter region */
4843 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4844 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4845 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2, params, val);
4846 /* If Active filter size is set we enable establishing
4847 * offload connection through firmware work request
4849 if ((val[0] != val[1]) && (ret >= 0)) {
4850 adap->flags |= FW_OFLD_CONN;
4851 adap->tids.aftid_base = val[0];
4852 adap->tids.aftid_end = val[1];
4856 * Get device capabilities so we can determine what resources we need
4859 memset(&caps_cmd, 0, sizeof(caps_cmd));
4860 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4861 FW_CMD_REQUEST | FW_CMD_READ);
4862 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4863 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4868 if (caps_cmd.ofldcaps) {
4869 /* query offload-related parameters */
4870 params[0] = FW_PARAM_DEV(NTID);
4871 params[1] = FW_PARAM_PFVF(SERVER_START);
4872 params[2] = FW_PARAM_PFVF(SERVER_END);
4873 params[3] = FW_PARAM_PFVF(TDDP_START);
4874 params[4] = FW_PARAM_PFVF(TDDP_END);
4875 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4876 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
4880 adap->tids.ntids = val[0];
4881 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
4882 adap->tids.stid_base = val[1];
4883 adap->tids.nstids = val[2] - val[1] + 1;
4885 * Setup server filter region. Divide the availble filter
4886 * region into two parts. Regular filters get 1/3rd and server
4887 * filters get 2/3rd part. This is only enabled if workarond
4889 * 1. For regular filters.
4890 * 2. Server filter: This are special filters which are used
4891 * to redirect SYN packets to offload queue.
4893 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
4894 adap->tids.sftid_base = adap->tids.ftid_base +
4895 DIV_ROUND_UP(adap->tids.nftids, 3);
4896 adap->tids.nsftids = adap->tids.nftids -
4897 DIV_ROUND_UP(adap->tids.nftids, 3);
4898 adap->tids.nftids = adap->tids.sftid_base -
4899 adap->tids.ftid_base;
4901 adap->vres.ddp.start = val[3];
4902 adap->vres.ddp.size = val[4] - val[3] + 1;
4903 adap->params.ofldq_wr_cred = val[5];
4905 adap->params.offload = 1;
4907 if (caps_cmd.rdmacaps) {
4908 params[0] = FW_PARAM_PFVF(STAG_START);
4909 params[1] = FW_PARAM_PFVF(STAG_END);
4910 params[2] = FW_PARAM_PFVF(RQ_START);
4911 params[3] = FW_PARAM_PFVF(RQ_END);
4912 params[4] = FW_PARAM_PFVF(PBL_START);
4913 params[5] = FW_PARAM_PFVF(PBL_END);
4914 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
4918 adap->vres.stag.start = val[0];
4919 adap->vres.stag.size = val[1] - val[0] + 1;
4920 adap->vres.rq.start = val[2];
4921 adap->vres.rq.size = val[3] - val[2] + 1;
4922 adap->vres.pbl.start = val[4];
4923 adap->vres.pbl.size = val[5] - val[4] + 1;
4925 params[0] = FW_PARAM_PFVF(SQRQ_START);
4926 params[1] = FW_PARAM_PFVF(SQRQ_END);
4927 params[2] = FW_PARAM_PFVF(CQ_START);
4928 params[3] = FW_PARAM_PFVF(CQ_END);
4929 params[4] = FW_PARAM_PFVF(OCQ_START);
4930 params[5] = FW_PARAM_PFVF(OCQ_END);
4931 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
4934 adap->vres.qp.start = val[0];
4935 adap->vres.qp.size = val[1] - val[0] + 1;
4936 adap->vres.cq.start = val[2];
4937 adap->vres.cq.size = val[3] - val[2] + 1;
4938 adap->vres.ocq.start = val[4];
4939 adap->vres.ocq.size = val[5] - val[4] + 1;
4941 if (caps_cmd.iscsicaps) {
4942 params[0] = FW_PARAM_PFVF(ISCSI_START);
4943 params[1] = FW_PARAM_PFVF(ISCSI_END);
4944 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2,
4948 adap->vres.iscsi.start = val[0];
4949 adap->vres.iscsi.size = val[1] - val[0] + 1;
4951 #undef FW_PARAM_PFVF
4955 * These are finalized by FW initialization, load their values now.
4957 v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
4958 adap->params.tp.tre = TIMERRESOLUTION_GET(v);
4959 adap->params.tp.dack_re = DELAYEDACKRESOLUTION_GET(v);
4960 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
4961 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4962 adap->params.b_wnd);
4964 /* MODQ_REQ_MAP defaults to setting queues 0-3 to chan 0-3 */
4965 for (j = 0; j < NCHAN; j++)
4966 adap->params.tp.tx_modq[j] = j;
4968 t4_read_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4969 &adap->filter_mode, 1,
4972 adap->flags |= FW_OK;
4976 * Something bad happened. If a command timed out or failed with EIO
4977 * FW does not operate within its spec or something catastrophic
4978 * happened to HW/FW, stop issuing commands.
4981 if (ret != -ETIMEDOUT && ret != -EIO)
4982 t4_fw_bye(adap, adap->mbox);
4988 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4989 pci_channel_state_t state)
4992 struct adapter *adap = pci_get_drvdata(pdev);
4998 adap->flags &= ~FW_OK;
4999 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5000 for_each_port(adap, i) {
5001 struct net_device *dev = adap->port[i];
5003 netif_device_detach(dev);
5004 netif_carrier_off(dev);
5006 if (adap->flags & FULL_INIT_DONE)
5009 pci_disable_device(pdev);
5010 out: return state == pci_channel_io_perm_failure ?
5011 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5014 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5017 struct fw_caps_config_cmd c;
5018 struct adapter *adap = pci_get_drvdata(pdev);
5021 pci_restore_state(pdev);
5022 pci_save_state(pdev);
5023 return PCI_ERS_RESULT_RECOVERED;
5026 if (pci_enable_device(pdev)) {
5027 dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
5028 return PCI_ERS_RESULT_DISCONNECT;
5031 pci_set_master(pdev);
5032 pci_restore_state(pdev);
5033 pci_save_state(pdev);
5034 pci_cleanup_aer_uncorrect_error_status(pdev);
5036 if (t4_wait_dev_ready(adap) < 0)
5037 return PCI_ERS_RESULT_DISCONNECT;
5038 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL))
5039 return PCI_ERS_RESULT_DISCONNECT;
5040 adap->flags |= FW_OK;
5041 if (adap_init1(adap, &c))
5042 return PCI_ERS_RESULT_DISCONNECT;
5044 for_each_port(adap, i) {
5045 struct port_info *p = adap2pinfo(adap, i);
5047 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
5050 return PCI_ERS_RESULT_DISCONNECT;
5052 p->xact_addr_filt = -1;
5055 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5056 adap->params.b_wnd);
5059 return PCI_ERS_RESULT_DISCONNECT;
5060 return PCI_ERS_RESULT_RECOVERED;
5063 static void eeh_resume(struct pci_dev *pdev)
5066 struct adapter *adap = pci_get_drvdata(pdev);
5072 for_each_port(adap, i) {
5073 struct net_device *dev = adap->port[i];
5075 if (netif_running(dev)) {
5077 cxgb_set_rxmode(dev);
5079 netif_device_attach(dev);
5084 static const struct pci_error_handlers cxgb4_eeh = {
5085 .error_detected = eeh_err_detected,
5086 .slot_reset = eeh_slot_reset,
5087 .resume = eeh_resume,
5090 static inline bool is_10g_port(const struct link_config *lc)
5092 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
5095 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
5096 unsigned int size, unsigned int iqe_size)
5098 q->intr_params = QINTR_TIMER_IDX(timer_idx) |
5099 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
5100 q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
5101 q->iqe_len = iqe_size;
5106 * Perform default configuration of DMA queues depending on the number and type
5107 * of ports we found and the number of available CPUs. Most settings can be
5108 * modified by the admin prior to actual use.
5110 static void cfg_queues(struct adapter *adap)
5112 struct sge *s = &adap->sge;
5113 int i, q10g = 0, n10g = 0, qidx = 0;
5115 for_each_port(adap, i)
5116 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
5119 * We default to 1 queue per non-10G port and up to # of cores queues
5123 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
5124 if (q10g > netif_get_num_default_rss_queues())
5125 q10g = netif_get_num_default_rss_queues();
5127 for_each_port(adap, i) {
5128 struct port_info *pi = adap2pinfo(adap, i);
5130 pi->first_qset = qidx;
5131 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
5136 s->max_ethqsets = qidx; /* MSI-X may lower it later */
5138 if (is_offload(adap)) {
5140 * For offload we use 1 queue/channel if all ports are up to 1G,
5141 * otherwise we divide all available queues amongst the channels
5142 * capped by the number of available cores.
5145 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
5147 s->ofldqsets = roundup(i, adap->params.nports);
5149 s->ofldqsets = adap->params.nports;
5150 /* For RDMA one Rx queue per channel suffices */
5151 s->rdmaqs = adap->params.nports;
5154 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5155 struct sge_eth_rxq *r = &s->ethrxq[i];
5157 init_rspq(&r->rspq, 0, 0, 1024, 64);
5161 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5162 s->ethtxq[i].q.size = 1024;
5164 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5165 s->ctrlq[i].q.size = 512;
5167 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
5168 s->ofldtxq[i].q.size = 1024;
5170 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
5171 struct sge_ofld_rxq *r = &s->ofldrxq[i];
5173 init_rspq(&r->rspq, 0, 0, 1024, 64);
5174 r->rspq.uld = CXGB4_ULD_ISCSI;
5178 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
5179 struct sge_ofld_rxq *r = &s->rdmarxq[i];
5181 init_rspq(&r->rspq, 0, 0, 511, 64);
5182 r->rspq.uld = CXGB4_ULD_RDMA;
5186 init_rspq(&s->fw_evtq, 6, 0, 512, 64);
5187 init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
5191 * Reduce the number of Ethernet queues across all ports to at most n.
5192 * n provides at least one queue per port.
5194 static void reduce_ethqs(struct adapter *adap, int n)
5197 struct port_info *pi;
5199 while (n < adap->sge.ethqsets)
5200 for_each_port(adap, i) {
5201 pi = adap2pinfo(adap, i);
5202 if (pi->nqsets > 1) {
5204 adap->sge.ethqsets--;
5205 if (adap->sge.ethqsets <= n)
5211 for_each_port(adap, i) {
5212 pi = adap2pinfo(adap, i);
5218 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5219 #define EXTRA_VECS 2
5221 static int enable_msix(struct adapter *adap)
5224 int i, err, want, need;
5225 struct sge *s = &adap->sge;
5226 unsigned int nchan = adap->params.nports;
5227 struct msix_entry entries[MAX_INGQ + 1];
5229 for (i = 0; i < ARRAY_SIZE(entries); ++i)
5230 entries[i].entry = i;
5232 want = s->max_ethqsets + EXTRA_VECS;
5233 if (is_offload(adap)) {
5234 want += s->rdmaqs + s->ofldqsets;
5235 /* need nchan for each possible ULD */
5236 ofld_need = 2 * nchan;
5238 need = adap->params.nports + EXTRA_VECS + ofld_need;
5240 while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
5245 * Distribute available vectors to the various queue groups.
5246 * Every group gets its minimum requirement and NIC gets top
5247 * priority for leftovers.
5249 i = want - EXTRA_VECS - ofld_need;
5250 if (i < s->max_ethqsets) {
5251 s->max_ethqsets = i;
5252 if (i < s->ethqsets)
5253 reduce_ethqs(adap, i);
5255 if (is_offload(adap)) {
5256 i = want - EXTRA_VECS - s->max_ethqsets;
5257 i -= ofld_need - nchan;
5258 s->ofldqsets = (i / nchan) * nchan; /* round down */
5260 for (i = 0; i < want; ++i)
5261 adap->msix_info[i].vec = entries[i].vector;
5263 dev_info(adap->pdev_dev,
5264 "only %d MSI-X vectors left, not using MSI-X\n", err);
5270 static int init_rss(struct adapter *adap)
5274 for_each_port(adap, i) {
5275 struct port_info *pi = adap2pinfo(adap, i);
5277 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5280 for (j = 0; j < pi->rss_size; j++)
5281 pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
5286 static void print_port_info(const struct net_device *dev)
5288 static const char *base[] = {
5289 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
5290 "KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4"
5295 const char *spd = "";
5296 const struct port_info *pi = netdev_priv(dev);
5297 const struct adapter *adap = pi->adapter;
5299 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
5301 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
5304 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
5305 bufp += sprintf(bufp, "100/");
5306 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
5307 bufp += sprintf(bufp, "1000/");
5308 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
5309 bufp += sprintf(bufp, "10G/");
5312 sprintf(bufp, "BASE-%s", base[pi->port_type]);
5314 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
5315 adap->params.vpd.id,
5316 CHELSIO_CHIP_RELEASE(adap->params.rev), buf,
5317 is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
5318 (adap->flags & USING_MSIX) ? " MSI-X" :
5319 (adap->flags & USING_MSI) ? " MSI" : "");
5320 netdev_info(dev, "S/N: %s, E/C: %s\n",
5321 adap->params.vpd.sn, adap->params.vpd.ec);
5324 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
5326 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
5330 * Free the following resources:
5331 * - memory used for tables
5334 * - resources FW is holding for us
5336 static void free_some_resources(struct adapter *adapter)
5340 t4_free_mem(adapter->l2t);
5341 t4_free_mem(adapter->tids.tid_tab);
5342 disable_msi(adapter);
5344 for_each_port(adapter, i)
5345 if (adapter->port[i]) {
5346 kfree(adap2pinfo(adapter, i)->rss);
5347 free_netdev(adapter->port[i]);
5349 if (adapter->flags & FW_OK)
5350 t4_fw_bye(adapter, adapter->fn);
5353 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5354 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5355 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5356 #define SEGMENT_SIZE 128
5358 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5360 int func, i, err, s_qpp, qpp, num_seg;
5361 struct port_info *pi;
5362 bool highdma = false;
5363 struct adapter *adapter = NULL;
5364 #ifdef CONFIG_PCI_IOV
5368 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5370 err = pci_request_regions(pdev, KBUILD_MODNAME);
5372 /* Just info, some other driver may have claimed the device. */
5373 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5377 /* We control everything through one PF */
5378 func = PCI_FUNC(pdev->devfn);
5379 if (func != ent->driver_data) {
5380 pci_save_state(pdev); /* to restore SR-IOV later */
5384 err = pci_enable_device(pdev);
5386 dev_err(&pdev->dev, "cannot enable PCI device\n");
5387 goto out_release_regions;
5390 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5392 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5394 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5395 "coherent allocations\n");
5396 goto out_disable_device;
5399 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5401 dev_err(&pdev->dev, "no usable DMA configuration\n");
5402 goto out_disable_device;
5406 pci_enable_pcie_error_reporting(pdev);
5407 enable_pcie_relaxed_ordering(pdev);
5408 pci_set_master(pdev);
5409 pci_save_state(pdev);
5411 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5414 goto out_disable_device;
5417 adapter->regs = pci_ioremap_bar(pdev, 0);
5418 if (!adapter->regs) {
5419 dev_err(&pdev->dev, "cannot map device registers\n");
5421 goto out_free_adapter;
5424 adapter->pdev = pdev;
5425 adapter->pdev_dev = &pdev->dev;
5426 adapter->mbox = func;
5428 adapter->msg_enable = dflt_msg_enable;
5429 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5431 spin_lock_init(&adapter->stats_lock);
5432 spin_lock_init(&adapter->tid_release_lock);
5434 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5435 INIT_WORK(&adapter->db_full_task, process_db_full);
5436 INIT_WORK(&adapter->db_drop_task, process_db_drop);
5438 err = t4_prep_adapter(adapter);
5440 goto out_unmap_bar0;
5442 if (!is_t4(adapter->chip)) {
5443 s_qpp = QUEUESPERPAGEPF1 * adapter->fn;
5444 qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter,
5445 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
5446 num_seg = PAGE_SIZE / SEGMENT_SIZE;
5448 /* Each segment size is 128B. Write coalescing is enabled only
5449 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5450 * queue is less no of segments that can be accommodated in
5453 if (qpp > num_seg) {
5455 "Incorrect number of egress queues per page\n");
5457 goto out_unmap_bar0;
5459 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
5460 pci_resource_len(pdev, 2));
5461 if (!adapter->bar2) {
5462 dev_err(&pdev->dev, "cannot map device bar2 region\n");
5464 goto out_unmap_bar0;
5468 setup_memwin(adapter);
5469 err = adap_init0(adapter);
5470 setup_memwin_rdma(adapter);
5474 for_each_port(adapter, i) {
5475 struct net_device *netdev;
5477 netdev = alloc_etherdev_mq(sizeof(struct port_info),
5484 SET_NETDEV_DEV(netdev, &pdev->dev);
5486 adapter->port[i] = netdev;
5487 pi = netdev_priv(netdev);
5488 pi->adapter = adapter;
5489 pi->xact_addr_filt = -1;
5491 netdev->irq = pdev->irq;
5493 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5494 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5495 NETIF_F_RXCSUM | NETIF_F_RXHASH |
5496 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
5498 netdev->hw_features |= NETIF_F_HIGHDMA;
5499 netdev->features |= netdev->hw_features;
5500 netdev->vlan_features = netdev->features & VLAN_FEAT;
5502 netdev->priv_flags |= IFF_UNICAST_FLT;
5504 netdev->netdev_ops = &cxgb4_netdev_ops;
5505 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
5508 pci_set_drvdata(pdev, adapter);
5510 if (adapter->flags & FW_OK) {
5511 err = t4_port_init(adapter, func, func, 0);
5517 * Configure queues and allocate tables now, they can be needed as
5518 * soon as the first register_netdev completes.
5520 cfg_queues(adapter);
5522 adapter->l2t = t4_init_l2t();
5523 if (!adapter->l2t) {
5524 /* We tolerate a lack of L2T, giving up some functionality */
5525 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5526 adapter->params.offload = 0;
5529 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
5530 dev_warn(&pdev->dev, "could not allocate TID table, "
5532 adapter->params.offload = 0;
5535 /* See what interrupts we'll be using */
5536 if (msi > 1 && enable_msix(adapter) == 0)
5537 adapter->flags |= USING_MSIX;
5538 else if (msi > 0 && pci_enable_msi(pdev) == 0)
5539 adapter->flags |= USING_MSI;
5541 err = init_rss(adapter);
5546 * The card is now ready to go. If any errors occur during device
5547 * registration we do not fail the whole card but rather proceed only
5548 * with the ports we manage to register successfully. However we must
5549 * register at least one net device.
5551 for_each_port(adapter, i) {
5552 pi = adap2pinfo(adapter, i);
5553 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5554 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5556 err = register_netdev(adapter->port[i]);
5559 adapter->chan_map[pi->tx_chan] = i;
5560 print_port_info(adapter->port[i]);
5563 dev_err(&pdev->dev, "could not register any net devices\n");
5567 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5571 if (cxgb4_debugfs_root) {
5572 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5573 cxgb4_debugfs_root);
5574 setup_debugfs(adapter);
5577 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5578 pdev->needs_freset = 1;
5580 if (is_offload(adapter))
5581 attach_ulds(adapter);
5584 #ifdef CONFIG_PCI_IOV
5585 max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
5586 NUM_OF_PF_WITH_SRIOV_T5;
5588 if (func < max_no_pf && num_vf[func] > 0)
5589 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
5590 dev_info(&pdev->dev,
5591 "instantiated %u virtual functions\n",
5597 free_some_resources(adapter);
5599 if (!is_t4(adapter->chip))
5600 iounmap(adapter->bar2);
5602 iounmap(adapter->regs);
5606 pci_disable_pcie_error_reporting(pdev);
5607 pci_disable_device(pdev);
5608 out_release_regions:
5609 pci_release_regions(pdev);
5610 pci_set_drvdata(pdev, NULL);
5614 static void remove_one(struct pci_dev *pdev)
5616 struct adapter *adapter = pci_get_drvdata(pdev);
5618 #ifdef CONFIG_PCI_IOV
5619 pci_disable_sriov(pdev);
5626 if (is_offload(adapter))
5627 detach_ulds(adapter);
5629 for_each_port(adapter, i)
5630 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5631 unregister_netdev(adapter->port[i]);
5633 if (adapter->debugfs_root)
5634 debugfs_remove_recursive(adapter->debugfs_root);
5636 /* If we allocated filters, free up state associated with any
5639 if (adapter->tids.ftid_tab) {
5640 struct filter_entry *f = &adapter->tids.ftid_tab[0];
5641 for (i = 0; i < (adapter->tids.nftids +
5642 adapter->tids.nsftids); i++, f++)
5644 clear_filter(adapter, f);
5647 if (adapter->flags & FULL_INIT_DONE)
5650 free_some_resources(adapter);
5651 iounmap(adapter->regs);
5652 if (!is_t4(adapter->chip))
5653 iounmap(adapter->bar2);
5655 pci_disable_pcie_error_reporting(pdev);
5656 pci_disable_device(pdev);
5657 pci_release_regions(pdev);
5658 pci_set_drvdata(pdev, NULL);
5660 pci_release_regions(pdev);
5663 static struct pci_driver cxgb4_driver = {
5664 .name = KBUILD_MODNAME,
5665 .id_table = cxgb4_pci_tbl,
5667 .remove = remove_one,
5668 .err_handler = &cxgb4_eeh,
5671 static int __init cxgb4_init_module(void)
5675 workq = create_singlethread_workqueue("cxgb4");
5679 /* Debugfs support is optional, just warn if this fails */
5680 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5681 if (!cxgb4_debugfs_root)
5682 pr_warn("could not create debugfs entry, continuing\n");
5684 ret = pci_register_driver(&cxgb4_driver);
5686 debugfs_remove(cxgb4_debugfs_root);
5690 static void __exit cxgb4_cleanup_module(void)
5692 pci_unregister_driver(&cxgb4_driver);
5693 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
5694 flush_workqueue(workq);
5695 destroy_workqueue(workq);
5698 module_init(cxgb4_init_module);
5699 module_exit(cxgb4_cleanup_module);