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 ",
1332 static int get_sset_count(struct net_device *dev, int sset)
1336 return ARRAY_SIZE(stats_strings);
1342 #define T4_REGMAP_SIZE (160 * 1024)
1344 static int get_regs_len(struct net_device *dev)
1346 return T4_REGMAP_SIZE;
1349 static int get_eeprom_len(struct net_device *dev)
1354 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1356 struct adapter *adapter = netdev2adap(dev);
1358 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1359 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1360 strlcpy(info->bus_info, pci_name(adapter->pdev),
1361 sizeof(info->bus_info));
1363 if (adapter->params.fw_vers)
1364 snprintf(info->fw_version, sizeof(info->fw_version),
1365 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1366 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1367 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1368 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1369 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1370 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1371 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1372 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1373 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1376 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1378 if (stringset == ETH_SS_STATS)
1379 memcpy(data, stats_strings, sizeof(stats_strings));
1383 * port stats maintained per queue of the port. They should be in the same
1384 * order as in stats_strings above.
1386 struct queue_port_stats {
1396 static void collect_sge_port_stats(const struct adapter *adap,
1397 const struct port_info *p, struct queue_port_stats *s)
1400 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1401 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1403 memset(s, 0, sizeof(*s));
1404 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1406 s->tx_csum += tx->tx_cso;
1407 s->rx_csum += rx->stats.rx_cso;
1408 s->vlan_ex += rx->stats.vlan_ex;
1409 s->vlan_ins += tx->vlan_ins;
1410 s->gro_pkts += rx->stats.lro_pkts;
1411 s->gro_merged += rx->stats.lro_merged;
1415 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1418 struct port_info *pi = netdev_priv(dev);
1419 struct adapter *adapter = pi->adapter;
1421 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1423 data += sizeof(struct port_stats) / sizeof(u64);
1424 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1428 * Return a version number to identify the type of adapter. The scheme is:
1429 * - bits 0..9: chip version
1430 * - bits 10..15: chip revision
1431 * - bits 16..23: register dump version
1433 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1435 return CHELSIO_CHIP_VERSION(ap->chip) |
1436 (CHELSIO_CHIP_RELEASE(ap->chip) << 10) | (1 << 16);
1439 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1442 u32 *p = buf + start;
1444 for ( ; start <= end; start += sizeof(u32))
1445 *p++ = t4_read_reg(ap, start);
1448 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1451 static const unsigned int reg_ranges[] = {
1672 struct adapter *ap = netdev2adap(dev);
1674 regs->version = mk_adap_vers(ap);
1676 memset(buf, 0, T4_REGMAP_SIZE);
1677 for (i = 0; i < ARRAY_SIZE(reg_ranges); i += 2)
1678 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
1681 static int restart_autoneg(struct net_device *dev)
1683 struct port_info *p = netdev_priv(dev);
1685 if (!netif_running(dev))
1687 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
1689 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
1693 static int identify_port(struct net_device *dev,
1694 enum ethtool_phys_id_state state)
1697 struct adapter *adap = netdev2adap(dev);
1699 if (state == ETHTOOL_ID_ACTIVE)
1701 else if (state == ETHTOOL_ID_INACTIVE)
1706 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
1709 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
1713 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
1714 type == FW_PORT_TYPE_BT_XAUI) {
1716 if (caps & FW_PORT_CAP_SPEED_100M)
1717 v |= SUPPORTED_100baseT_Full;
1718 if (caps & FW_PORT_CAP_SPEED_1G)
1719 v |= SUPPORTED_1000baseT_Full;
1720 if (caps & FW_PORT_CAP_SPEED_10G)
1721 v |= SUPPORTED_10000baseT_Full;
1722 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
1723 v |= SUPPORTED_Backplane;
1724 if (caps & FW_PORT_CAP_SPEED_1G)
1725 v |= SUPPORTED_1000baseKX_Full;
1726 if (caps & FW_PORT_CAP_SPEED_10G)
1727 v |= SUPPORTED_10000baseKX4_Full;
1728 } else if (type == FW_PORT_TYPE_KR)
1729 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
1730 else if (type == FW_PORT_TYPE_BP_AP)
1731 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
1732 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
1733 else if (type == FW_PORT_TYPE_BP4_AP)
1734 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
1735 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
1736 SUPPORTED_10000baseKX4_Full;
1737 else if (type == FW_PORT_TYPE_FIBER_XFI ||
1738 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
1739 v |= SUPPORTED_FIBRE;
1741 if (caps & FW_PORT_CAP_ANEG)
1742 v |= SUPPORTED_Autoneg;
1746 static unsigned int to_fw_linkcaps(unsigned int caps)
1750 if (caps & ADVERTISED_100baseT_Full)
1751 v |= FW_PORT_CAP_SPEED_100M;
1752 if (caps & ADVERTISED_1000baseT_Full)
1753 v |= FW_PORT_CAP_SPEED_1G;
1754 if (caps & ADVERTISED_10000baseT_Full)
1755 v |= FW_PORT_CAP_SPEED_10G;
1759 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1761 const struct port_info *p = netdev_priv(dev);
1763 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
1764 p->port_type == FW_PORT_TYPE_BT_XFI ||
1765 p->port_type == FW_PORT_TYPE_BT_XAUI)
1766 cmd->port = PORT_TP;
1767 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
1768 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
1769 cmd->port = PORT_FIBRE;
1770 else if (p->port_type == FW_PORT_TYPE_SFP) {
1771 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1772 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1773 cmd->port = PORT_DA;
1775 cmd->port = PORT_FIBRE;
1777 cmd->port = PORT_OTHER;
1779 if (p->mdio_addr >= 0) {
1780 cmd->phy_address = p->mdio_addr;
1781 cmd->transceiver = XCVR_EXTERNAL;
1782 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
1783 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
1785 cmd->phy_address = 0; /* not really, but no better option */
1786 cmd->transceiver = XCVR_INTERNAL;
1787 cmd->mdio_support = 0;
1790 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
1791 cmd->advertising = from_fw_linkcaps(p->port_type,
1792 p->link_cfg.advertising);
1793 ethtool_cmd_speed_set(cmd,
1794 netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
1795 cmd->duplex = DUPLEX_FULL;
1796 cmd->autoneg = p->link_cfg.autoneg;
1802 static unsigned int speed_to_caps(int speed)
1804 if (speed == SPEED_100)
1805 return FW_PORT_CAP_SPEED_100M;
1806 if (speed == SPEED_1000)
1807 return FW_PORT_CAP_SPEED_1G;
1808 if (speed == SPEED_10000)
1809 return FW_PORT_CAP_SPEED_10G;
1813 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1816 struct port_info *p = netdev_priv(dev);
1817 struct link_config *lc = &p->link_cfg;
1818 u32 speed = ethtool_cmd_speed(cmd);
1820 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
1823 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
1825 * PHY offers a single speed. See if that's what's
1828 if (cmd->autoneg == AUTONEG_DISABLE &&
1829 (lc->supported & speed_to_caps(speed)))
1834 if (cmd->autoneg == AUTONEG_DISABLE) {
1835 cap = speed_to_caps(speed);
1837 if (!(lc->supported & cap) || (speed == SPEED_1000) ||
1838 (speed == SPEED_10000))
1840 lc->requested_speed = cap;
1841 lc->advertising = 0;
1843 cap = to_fw_linkcaps(cmd->advertising);
1844 if (!(lc->supported & cap))
1846 lc->requested_speed = 0;
1847 lc->advertising = cap | FW_PORT_CAP_ANEG;
1849 lc->autoneg = cmd->autoneg;
1851 if (netif_running(dev))
1852 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1857 static void get_pauseparam(struct net_device *dev,
1858 struct ethtool_pauseparam *epause)
1860 struct port_info *p = netdev_priv(dev);
1862 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1863 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
1864 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
1867 static int set_pauseparam(struct net_device *dev,
1868 struct ethtool_pauseparam *epause)
1870 struct port_info *p = netdev_priv(dev);
1871 struct link_config *lc = &p->link_cfg;
1873 if (epause->autoneg == AUTONEG_DISABLE)
1874 lc->requested_fc = 0;
1875 else if (lc->supported & FW_PORT_CAP_ANEG)
1876 lc->requested_fc = PAUSE_AUTONEG;
1880 if (epause->rx_pause)
1881 lc->requested_fc |= PAUSE_RX;
1882 if (epause->tx_pause)
1883 lc->requested_fc |= PAUSE_TX;
1884 if (netif_running(dev))
1885 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1890 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1892 const struct port_info *pi = netdev_priv(dev);
1893 const struct sge *s = &pi->adapter->sge;
1895 e->rx_max_pending = MAX_RX_BUFFERS;
1896 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1897 e->rx_jumbo_max_pending = 0;
1898 e->tx_max_pending = MAX_TXQ_ENTRIES;
1900 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
1901 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1902 e->rx_jumbo_pending = 0;
1903 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
1906 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1909 const struct port_info *pi = netdev_priv(dev);
1910 struct adapter *adapter = pi->adapter;
1911 struct sge *s = &adapter->sge;
1913 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
1914 e->tx_pending > MAX_TXQ_ENTRIES ||
1915 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1916 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1917 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
1920 if (adapter->flags & FULL_INIT_DONE)
1923 for (i = 0; i < pi->nqsets; ++i) {
1924 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
1925 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
1926 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
1931 static int closest_timer(const struct sge *s, int time)
1933 int i, delta, match = 0, min_delta = INT_MAX;
1935 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1936 delta = time - s->timer_val[i];
1939 if (delta < min_delta) {
1947 static int closest_thres(const struct sge *s, int thres)
1949 int i, delta, match = 0, min_delta = INT_MAX;
1951 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1952 delta = thres - s->counter_val[i];
1955 if (delta < min_delta) {
1964 * Return a queue's interrupt hold-off time in us. 0 means no timer.
1966 static unsigned int qtimer_val(const struct adapter *adap,
1967 const struct sge_rspq *q)
1969 unsigned int idx = q->intr_params >> 1;
1971 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
1975 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1976 * @adap: the adapter
1978 * @us: the hold-off time in us, or 0 to disable timer
1979 * @cnt: the hold-off packet count, or 0 to disable counter
1981 * Sets an Rx queue's interrupt hold-off time and packet count. At least
1982 * one of the two needs to be enabled for the queue to generate interrupts.
1984 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
1985 unsigned int us, unsigned int cnt)
1987 if ((us | cnt) == 0)
1994 new_idx = closest_thres(&adap->sge, cnt);
1995 if (q->desc && q->pktcnt_idx != new_idx) {
1996 /* the queue has already been created, update it */
1997 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1998 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1999 FW_PARAMS_PARAM_YZ(q->cntxt_id);
2000 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
2005 q->pktcnt_idx = new_idx;
2008 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
2009 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
2013 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2015 const struct port_info *pi = netdev_priv(dev);
2016 struct adapter *adap = pi->adapter;
2021 for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++) {
2022 q = &adap->sge.ethrxq[i].rspq;
2023 r = set_rxq_intr_params(adap, q, c->rx_coalesce_usecs,
2024 c->rx_max_coalesced_frames);
2026 dev_err(&dev->dev, "failed to set coalesce %d\n", r);
2033 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2035 const struct port_info *pi = netdev_priv(dev);
2036 const struct adapter *adap = pi->adapter;
2037 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
2039 c->rx_coalesce_usecs = qtimer_val(adap, rq);
2040 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
2041 adap->sge.counter_val[rq->pktcnt_idx] : 0;
2046 * eeprom_ptov - translate a physical EEPROM address to virtual
2047 * @phys_addr: the physical EEPROM address
2048 * @fn: the PCI function number
2049 * @sz: size of function-specific area
2051 * Translate a physical EEPROM address to virtual. The first 1K is
2052 * accessed through virtual addresses starting at 31K, the rest is
2053 * accessed through virtual addresses starting at 0.
2055 * The mapping is as follows:
2056 * [0..1K) -> [31K..32K)
2057 * [1K..1K+A) -> [31K-A..31K)
2058 * [1K+A..ES) -> [0..ES-A-1K)
2060 * where A = @fn * @sz, and ES = EEPROM size.
2062 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
2065 if (phys_addr < 1024)
2066 return phys_addr + (31 << 10);
2067 if (phys_addr < 1024 + fn)
2068 return 31744 - fn + phys_addr - 1024;
2069 if (phys_addr < EEPROMSIZE)
2070 return phys_addr - 1024 - fn;
2075 * The next two routines implement eeprom read/write from physical addresses.
2077 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
2079 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2082 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
2083 return vaddr < 0 ? vaddr : 0;
2086 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
2088 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2091 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
2092 return vaddr < 0 ? vaddr : 0;
2095 #define EEPROM_MAGIC 0x38E2F10C
2097 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2101 struct adapter *adapter = netdev2adap(dev);
2103 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2107 e->magic = EEPROM_MAGIC;
2108 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2109 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
2112 memcpy(data, buf + e->offset, e->len);
2117 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2122 u32 aligned_offset, aligned_len, *p;
2123 struct adapter *adapter = netdev2adap(dev);
2125 if (eeprom->magic != EEPROM_MAGIC)
2128 aligned_offset = eeprom->offset & ~3;
2129 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2131 if (adapter->fn > 0) {
2132 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
2134 if (aligned_offset < start ||
2135 aligned_offset + aligned_len > start + EEPROMPFSIZE)
2139 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2141 * RMW possibly needed for first or last words.
2143 buf = kmalloc(aligned_len, GFP_KERNEL);
2146 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
2147 if (!err && aligned_len > 4)
2148 err = eeprom_rd_phys(adapter,
2149 aligned_offset + aligned_len - 4,
2150 (u32 *)&buf[aligned_len - 4]);
2153 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2157 err = t4_seeprom_wp(adapter, false);
2161 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2162 err = eeprom_wr_phys(adapter, aligned_offset, *p);
2163 aligned_offset += 4;
2167 err = t4_seeprom_wp(adapter, true);
2174 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
2177 const struct firmware *fw;
2178 struct adapter *adap = netdev2adap(netdev);
2180 ef->data[sizeof(ef->data) - 1] = '\0';
2181 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
2185 ret = t4_load_fw(adap, fw->data, fw->size);
2186 release_firmware(fw);
2188 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
2192 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2193 #define BCAST_CRC 0xa0ccc1a6
2195 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2197 wol->supported = WAKE_BCAST | WAKE_MAGIC;
2198 wol->wolopts = netdev2adap(dev)->wol;
2199 memset(&wol->sopass, 0, sizeof(wol->sopass));
2202 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2205 struct port_info *pi = netdev_priv(dev);
2207 if (wol->wolopts & ~WOL_SUPPORTED)
2209 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
2210 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
2211 if (wol->wolopts & WAKE_BCAST) {
2212 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
2215 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
2216 ~6ULL, ~0ULL, BCAST_CRC, true);
2218 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
2222 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2224 const struct port_info *pi = netdev_priv(dev);
2225 netdev_features_t changed = dev->features ^ features;
2228 if (!(changed & NETIF_F_HW_VLAN_RX))
2231 err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
2233 !!(features & NETIF_F_HW_VLAN_RX), true);
2235 dev->features = features ^ NETIF_F_HW_VLAN_RX;
2239 static u32 get_rss_table_size(struct net_device *dev)
2241 const struct port_info *pi = netdev_priv(dev);
2243 return pi->rss_size;
2246 static int get_rss_table(struct net_device *dev, u32 *p)
2248 const struct port_info *pi = netdev_priv(dev);
2249 unsigned int n = pi->rss_size;
2256 static int set_rss_table(struct net_device *dev, const u32 *p)
2259 struct port_info *pi = netdev_priv(dev);
2261 for (i = 0; i < pi->rss_size; i++)
2263 if (pi->adapter->flags & FULL_INIT_DONE)
2264 return write_rss(pi, pi->rss);
2268 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
2271 const struct port_info *pi = netdev_priv(dev);
2273 switch (info->cmd) {
2274 case ETHTOOL_GRXFH: {
2275 unsigned int v = pi->rss_mode;
2278 switch (info->flow_type) {
2280 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
2281 info->data = RXH_IP_SRC | RXH_IP_DST |
2282 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2283 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2284 info->data = RXH_IP_SRC | RXH_IP_DST;
2287 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
2288 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2289 info->data = RXH_IP_SRC | RXH_IP_DST |
2290 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2291 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2292 info->data = RXH_IP_SRC | RXH_IP_DST;
2295 case AH_ESP_V4_FLOW:
2297 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2298 info->data = RXH_IP_SRC | RXH_IP_DST;
2301 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
2302 info->data = RXH_IP_SRC | RXH_IP_DST |
2303 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2304 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2305 info->data = RXH_IP_SRC | RXH_IP_DST;
2308 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
2309 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2310 info->data = RXH_IP_SRC | RXH_IP_DST |
2311 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2312 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2313 info->data = RXH_IP_SRC | RXH_IP_DST;
2316 case AH_ESP_V6_FLOW:
2318 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2319 info->data = RXH_IP_SRC | RXH_IP_DST;
2324 case ETHTOOL_GRXRINGS:
2325 info->data = pi->nqsets;
2331 static const struct ethtool_ops cxgb_ethtool_ops = {
2332 .get_settings = get_settings,
2333 .set_settings = set_settings,
2334 .get_drvinfo = get_drvinfo,
2335 .get_msglevel = get_msglevel,
2336 .set_msglevel = set_msglevel,
2337 .get_ringparam = get_sge_param,
2338 .set_ringparam = set_sge_param,
2339 .get_coalesce = get_coalesce,
2340 .set_coalesce = set_coalesce,
2341 .get_eeprom_len = get_eeprom_len,
2342 .get_eeprom = get_eeprom,
2343 .set_eeprom = set_eeprom,
2344 .get_pauseparam = get_pauseparam,
2345 .set_pauseparam = set_pauseparam,
2346 .get_link = ethtool_op_get_link,
2347 .get_strings = get_strings,
2348 .set_phys_id = identify_port,
2349 .nway_reset = restart_autoneg,
2350 .get_sset_count = get_sset_count,
2351 .get_ethtool_stats = get_stats,
2352 .get_regs_len = get_regs_len,
2353 .get_regs = get_regs,
2356 .get_rxnfc = get_rxnfc,
2357 .get_rxfh_indir_size = get_rss_table_size,
2358 .get_rxfh_indir = get_rss_table,
2359 .set_rxfh_indir = set_rss_table,
2360 .flash_device = set_flash,
2366 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2370 loff_t avail = file_inode(file)->i_size;
2371 unsigned int mem = (uintptr_t)file->private_data & 3;
2372 struct adapter *adap = file->private_data - mem;
2378 if (count > avail - pos)
2379 count = avail - pos;
2387 ret = t4_mc_read(adap, pos, data, NULL);
2389 ret = t4_edc_read(adap, mem, pos, data, NULL);
2393 ofst = pos % sizeof(data);
2394 len = min(count, sizeof(data) - ofst);
2395 if (copy_to_user(buf, (u8 *)data + ofst, len))
2402 count = pos - *ppos;
2407 static const struct file_operations mem_debugfs_fops = {
2408 .owner = THIS_MODULE,
2409 .open = simple_open,
2411 .llseek = default_llseek,
2414 static void add_debugfs_mem(struct adapter *adap, const char *name,
2415 unsigned int idx, unsigned int size_mb)
2419 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2420 (void *)adap + idx, &mem_debugfs_fops);
2421 if (de && de->d_inode)
2422 de->d_inode->i_size = size_mb << 20;
2425 static int setup_debugfs(struct adapter *adap)
2429 if (IS_ERR_OR_NULL(adap->debugfs_root))
2432 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2433 if (i & EDRAM0_ENABLE)
2434 add_debugfs_mem(adap, "edc0", MEM_EDC0, 5);
2435 if (i & EDRAM1_ENABLE)
2436 add_debugfs_mem(adap, "edc1", MEM_EDC1, 5);
2437 if (i & EXT_MEM_ENABLE)
2438 add_debugfs_mem(adap, "mc", MEM_MC,
2439 EXT_MEM_SIZE_GET(t4_read_reg(adap, MA_EXT_MEMORY_BAR)));
2441 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2447 * upper-layer driver support
2451 * Allocate an active-open TID and set it to the supplied value.
2453 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2457 spin_lock_bh(&t->atid_lock);
2459 union aopen_entry *p = t->afree;
2461 atid = (p - t->atid_tab) + t->atid_base;
2466 spin_unlock_bh(&t->atid_lock);
2469 EXPORT_SYMBOL(cxgb4_alloc_atid);
2472 * Release an active-open TID.
2474 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2476 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
2478 spin_lock_bh(&t->atid_lock);
2482 spin_unlock_bh(&t->atid_lock);
2484 EXPORT_SYMBOL(cxgb4_free_atid);
2487 * Allocate a server TID and set it to the supplied value.
2489 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
2493 spin_lock_bh(&t->stid_lock);
2494 if (family == PF_INET) {
2495 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
2496 if (stid < t->nstids)
2497 __set_bit(stid, t->stid_bmap);
2501 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
2506 t->stid_tab[stid].data = data;
2507 stid += t->stid_base;
2510 spin_unlock_bh(&t->stid_lock);
2513 EXPORT_SYMBOL(cxgb4_alloc_stid);
2515 /* Allocate a server filter TID and set it to the supplied value.
2517 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
2521 spin_lock_bh(&t->stid_lock);
2522 if (family == PF_INET) {
2523 stid = find_next_zero_bit(t->stid_bmap,
2524 t->nstids + t->nsftids, t->nstids);
2525 if (stid < (t->nstids + t->nsftids))
2526 __set_bit(stid, t->stid_bmap);
2533 t->stid_tab[stid].data = data;
2534 stid += t->stid_base;
2537 spin_unlock_bh(&t->stid_lock);
2540 EXPORT_SYMBOL(cxgb4_alloc_sftid);
2542 /* Release a server TID.
2544 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
2546 stid -= t->stid_base;
2547 spin_lock_bh(&t->stid_lock);
2548 if (family == PF_INET)
2549 __clear_bit(stid, t->stid_bmap);
2551 bitmap_release_region(t->stid_bmap, stid, 2);
2552 t->stid_tab[stid].data = NULL;
2554 spin_unlock_bh(&t->stid_lock);
2556 EXPORT_SYMBOL(cxgb4_free_stid);
2559 * Populate a TID_RELEASE WR. Caller must properly size the skb.
2561 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
2564 struct cpl_tid_release *req;
2566 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
2567 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
2568 INIT_TP_WR(req, tid);
2569 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
2573 * Queue a TID release request and if necessary schedule a work queue to
2576 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
2579 void **p = &t->tid_tab[tid];
2580 struct adapter *adap = container_of(t, struct adapter, tids);
2582 spin_lock_bh(&adap->tid_release_lock);
2583 *p = adap->tid_release_head;
2584 /* Low 2 bits encode the Tx channel number */
2585 adap->tid_release_head = (void **)((uintptr_t)p | chan);
2586 if (!adap->tid_release_task_busy) {
2587 adap->tid_release_task_busy = true;
2588 queue_work(workq, &adap->tid_release_task);
2590 spin_unlock_bh(&adap->tid_release_lock);
2594 * Process the list of pending TID release requests.
2596 static void process_tid_release_list(struct work_struct *work)
2598 struct sk_buff *skb;
2599 struct adapter *adap;
2601 adap = container_of(work, struct adapter, tid_release_task);
2603 spin_lock_bh(&adap->tid_release_lock);
2604 while (adap->tid_release_head) {
2605 void **p = adap->tid_release_head;
2606 unsigned int chan = (uintptr_t)p & 3;
2607 p = (void *)p - chan;
2609 adap->tid_release_head = *p;
2611 spin_unlock_bh(&adap->tid_release_lock);
2613 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
2615 schedule_timeout_uninterruptible(1);
2617 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
2618 t4_ofld_send(adap, skb);
2619 spin_lock_bh(&adap->tid_release_lock);
2621 adap->tid_release_task_busy = false;
2622 spin_unlock_bh(&adap->tid_release_lock);
2626 * Release a TID and inform HW. If we are unable to allocate the release
2627 * message we defer to a work queue.
2629 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
2632 struct sk_buff *skb;
2633 struct adapter *adap = container_of(t, struct adapter, tids);
2635 old = t->tid_tab[tid];
2636 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
2638 t->tid_tab[tid] = NULL;
2639 mk_tid_release(skb, chan, tid);
2640 t4_ofld_send(adap, skb);
2642 cxgb4_queue_tid_release(t, chan, tid);
2644 atomic_dec(&t->tids_in_use);
2646 EXPORT_SYMBOL(cxgb4_remove_tid);
2649 * Allocate and initialize the TID tables. Returns 0 on success.
2651 static int tid_init(struct tid_info *t)
2654 unsigned int stid_bmap_size;
2655 unsigned int natids = t->natids;
2657 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
2658 size = t->ntids * sizeof(*t->tid_tab) +
2659 natids * sizeof(*t->atid_tab) +
2660 t->nstids * sizeof(*t->stid_tab) +
2661 t->nsftids * sizeof(*t->stid_tab) +
2662 stid_bmap_size * sizeof(long) +
2663 t->nftids * sizeof(*t->ftid_tab) +
2664 t->nsftids * sizeof(*t->ftid_tab);
2666 t->tid_tab = t4_alloc_mem(size);
2670 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
2671 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
2672 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
2673 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
2674 spin_lock_init(&t->stid_lock);
2675 spin_lock_init(&t->atid_lock);
2677 t->stids_in_use = 0;
2679 t->atids_in_use = 0;
2680 atomic_set(&t->tids_in_use, 0);
2682 /* Setup the free list for atid_tab and clear the stid bitmap. */
2685 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
2686 t->afree = t->atid_tab;
2688 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
2693 * cxgb4_create_server - create an IP server
2695 * @stid: the server TID
2696 * @sip: local IP address to bind server to
2697 * @sport: the server's TCP port
2698 * @queue: queue to direct messages from this server to
2700 * Create an IP server for the given port and address.
2701 * Returns <0 on error and one of the %NET_XMIT_* values on success.
2703 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
2704 __be32 sip, __be16 sport, __be16 vlan,
2708 struct sk_buff *skb;
2709 struct adapter *adap;
2710 struct cpl_pass_open_req *req;
2712 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2716 adap = netdev2adap(dev);
2717 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
2719 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
2720 req->local_port = sport;
2721 req->peer_port = htons(0);
2722 req->local_ip = sip;
2723 req->peer_ip = htonl(0);
2724 chan = rxq_to_chan(&adap->sge, queue);
2725 req->opt0 = cpu_to_be64(TX_CHAN(chan));
2726 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
2727 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
2728 return t4_mgmt_tx(adap, skb);
2730 EXPORT_SYMBOL(cxgb4_create_server);
2733 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
2734 * @mtus: the HW MTU table
2735 * @mtu: the target MTU
2736 * @idx: index of selected entry in the MTU table
2738 * Returns the index and the value in the HW MTU table that is closest to
2739 * but does not exceed @mtu, unless @mtu is smaller than any value in the
2740 * table, in which case that smallest available value is selected.
2742 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
2747 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
2753 EXPORT_SYMBOL(cxgb4_best_mtu);
2756 * cxgb4_port_chan - get the HW channel of a port
2757 * @dev: the net device for the port
2759 * Return the HW Tx channel of the given port.
2761 unsigned int cxgb4_port_chan(const struct net_device *dev)
2763 return netdev2pinfo(dev)->tx_chan;
2765 EXPORT_SYMBOL(cxgb4_port_chan);
2767 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
2769 struct adapter *adap = netdev2adap(dev);
2772 v = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
2773 return lpfifo ? G_LP_COUNT(v) : G_HP_COUNT(v);
2775 EXPORT_SYMBOL(cxgb4_dbfifo_count);
2778 * cxgb4_port_viid - get the VI id of a port
2779 * @dev: the net device for the port
2781 * Return the VI id of the given port.
2783 unsigned int cxgb4_port_viid(const struct net_device *dev)
2785 return netdev2pinfo(dev)->viid;
2787 EXPORT_SYMBOL(cxgb4_port_viid);
2790 * cxgb4_port_idx - get the index of a port
2791 * @dev: the net device for the port
2793 * Return the index of the given port.
2795 unsigned int cxgb4_port_idx(const struct net_device *dev)
2797 return netdev2pinfo(dev)->port_id;
2799 EXPORT_SYMBOL(cxgb4_port_idx);
2801 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
2802 struct tp_tcp_stats *v6)
2804 struct adapter *adap = pci_get_drvdata(pdev);
2806 spin_lock(&adap->stats_lock);
2807 t4_tp_get_tcp_stats(adap, v4, v6);
2808 spin_unlock(&adap->stats_lock);
2810 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
2812 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
2813 const unsigned int *pgsz_order)
2815 struct adapter *adap = netdev2adap(dev);
2817 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
2818 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
2819 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
2820 HPZ3(pgsz_order[3]));
2822 EXPORT_SYMBOL(cxgb4_iscsi_init);
2824 int cxgb4_flush_eq_cache(struct net_device *dev)
2826 struct adapter *adap = netdev2adap(dev);
2829 ret = t4_fwaddrspace_write(adap, adap->mbox,
2830 0xe1000000 + A_SGE_CTXT_CMD, 0x20000000);
2833 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
2835 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
2837 u32 addr = t4_read_reg(adap, A_SGE_DBQ_CTXT_BADDR) + 24 * qid + 8;
2841 ret = t4_mem_win_read_len(adap, addr, (__be32 *)&indices, 8);
2843 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
2844 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
2849 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
2852 struct adapter *adap = netdev2adap(dev);
2853 u16 hw_pidx, hw_cidx;
2856 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
2860 if (pidx != hw_pidx) {
2863 if (pidx >= hw_pidx)
2864 delta = pidx - hw_pidx;
2866 delta = size - hw_pidx + pidx;
2868 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
2869 QID(qid) | PIDX(delta));
2874 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
2876 static struct pci_driver cxgb4_driver;
2878 static void check_neigh_update(struct neighbour *neigh)
2880 const struct device *parent;
2881 const struct net_device *netdev = neigh->dev;
2883 if (netdev->priv_flags & IFF_802_1Q_VLAN)
2884 netdev = vlan_dev_real_dev(netdev);
2885 parent = netdev->dev.parent;
2886 if (parent && parent->driver == &cxgb4_driver.driver)
2887 t4_l2t_update(dev_get_drvdata(parent), neigh);
2890 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2894 case NETEVENT_NEIGH_UPDATE:
2895 check_neigh_update(data);
2897 case NETEVENT_REDIRECT:
2904 static bool netevent_registered;
2905 static struct notifier_block cxgb4_netevent_nb = {
2906 .notifier_call = netevent_cb
2909 static void drain_db_fifo(struct adapter *adap, int usecs)
2914 set_current_state(TASK_UNINTERRUPTIBLE);
2915 schedule_timeout(usecs_to_jiffies(usecs));
2916 v = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
2917 if (G_LP_COUNT(v) == 0 && G_HP_COUNT(v) == 0)
2922 static void disable_txq_db(struct sge_txq *q)
2924 spin_lock_irq(&q->db_lock);
2926 spin_unlock_irq(&q->db_lock);
2929 static void enable_txq_db(struct sge_txq *q)
2931 spin_lock_irq(&q->db_lock);
2933 spin_unlock_irq(&q->db_lock);
2936 static void disable_dbs(struct adapter *adap)
2940 for_each_ethrxq(&adap->sge, i)
2941 disable_txq_db(&adap->sge.ethtxq[i].q);
2942 for_each_ofldrxq(&adap->sge, i)
2943 disable_txq_db(&adap->sge.ofldtxq[i].q);
2944 for_each_port(adap, i)
2945 disable_txq_db(&adap->sge.ctrlq[i].q);
2948 static void enable_dbs(struct adapter *adap)
2952 for_each_ethrxq(&adap->sge, i)
2953 enable_txq_db(&adap->sge.ethtxq[i].q);
2954 for_each_ofldrxq(&adap->sge, i)
2955 enable_txq_db(&adap->sge.ofldtxq[i].q);
2956 for_each_port(adap, i)
2957 enable_txq_db(&adap->sge.ctrlq[i].q);
2960 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
2962 u16 hw_pidx, hw_cidx;
2965 spin_lock_bh(&q->db_lock);
2966 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
2969 if (q->db_pidx != hw_pidx) {
2972 if (q->db_pidx >= hw_pidx)
2973 delta = q->db_pidx - hw_pidx;
2975 delta = q->size - hw_pidx + q->db_pidx;
2977 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
2978 QID(q->cntxt_id) | PIDX(delta));
2982 spin_unlock_bh(&q->db_lock);
2984 CH_WARN(adap, "DB drop recovery failed.\n");
2986 static void recover_all_queues(struct adapter *adap)
2990 for_each_ethrxq(&adap->sge, i)
2991 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
2992 for_each_ofldrxq(&adap->sge, i)
2993 sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
2994 for_each_port(adap, i)
2995 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2998 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
3000 mutex_lock(&uld_mutex);
3001 if (adap->uld_handle[CXGB4_ULD_RDMA])
3002 ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
3004 mutex_unlock(&uld_mutex);
3007 static void process_db_full(struct work_struct *work)
3009 struct adapter *adap;
3011 adap = container_of(work, struct adapter, db_full_task);
3013 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
3014 drain_db_fifo(adap, dbfifo_drain_delay);
3015 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3016 DBFIFO_HP_INT | DBFIFO_LP_INT,
3017 DBFIFO_HP_INT | DBFIFO_LP_INT);
3018 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3021 static void process_db_drop(struct work_struct *work)
3023 struct adapter *adap;
3025 adap = container_of(work, struct adapter, db_drop_task);
3027 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_DROPPED_DB, 0);
3029 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
3030 drain_db_fifo(adap, 1);
3031 recover_all_queues(adap);
3035 void t4_db_full(struct adapter *adap)
3037 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3038 DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
3039 queue_work(workq, &adap->db_full_task);
3042 void t4_db_dropped(struct adapter *adap)
3044 queue_work(workq, &adap->db_drop_task);
3047 static void uld_attach(struct adapter *adap, unsigned int uld)
3050 struct cxgb4_lld_info lli;
3053 lli.pdev = adap->pdev;
3054 lli.l2t = adap->l2t;
3055 lli.tids = &adap->tids;
3056 lli.ports = adap->port;
3057 lli.vr = &adap->vres;
3058 lli.mtus = adap->params.mtus;
3059 if (uld == CXGB4_ULD_RDMA) {
3060 lli.rxq_ids = adap->sge.rdma_rxq;
3061 lli.nrxq = adap->sge.rdmaqs;
3062 } else if (uld == CXGB4_ULD_ISCSI) {
3063 lli.rxq_ids = adap->sge.ofld_rxq;
3064 lli.nrxq = adap->sge.ofldqsets;
3066 lli.ntxq = adap->sge.ofldqsets;
3067 lli.nchan = adap->params.nports;
3068 lli.nports = adap->params.nports;
3069 lli.wr_cred = adap->params.ofldq_wr_cred;
3070 lli.adapter_type = adap->params.rev;
3071 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
3072 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
3073 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
3075 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
3076 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
3078 lli.filt_mode = adap->filter_mode;
3079 /* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
3080 for (i = 0; i < NCHAN; i++)
3082 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
3083 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
3084 lli.fw_vers = adap->params.fw_vers;
3085 lli.dbfifo_int_thresh = dbfifo_int_thresh;
3086 lli.sge_pktshift = adap->sge.pktshift;
3087 lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
3089 handle = ulds[uld].add(&lli);
3090 if (IS_ERR(handle)) {
3091 dev_warn(adap->pdev_dev,
3092 "could not attach to the %s driver, error %ld\n",
3093 uld_str[uld], PTR_ERR(handle));
3097 adap->uld_handle[uld] = handle;
3099 if (!netevent_registered) {
3100 register_netevent_notifier(&cxgb4_netevent_nb);
3101 netevent_registered = true;
3104 if (adap->flags & FULL_INIT_DONE)
3105 ulds[uld].state_change(handle, CXGB4_STATE_UP);
3108 static void attach_ulds(struct adapter *adap)
3112 mutex_lock(&uld_mutex);
3113 list_add_tail(&adap->list_node, &adapter_list);
3114 for (i = 0; i < CXGB4_ULD_MAX; i++)
3116 uld_attach(adap, i);
3117 mutex_unlock(&uld_mutex);
3120 static void detach_ulds(struct adapter *adap)
3124 mutex_lock(&uld_mutex);
3125 list_del(&adap->list_node);
3126 for (i = 0; i < CXGB4_ULD_MAX; i++)
3127 if (adap->uld_handle[i]) {
3128 ulds[i].state_change(adap->uld_handle[i],
3129 CXGB4_STATE_DETACH);
3130 adap->uld_handle[i] = NULL;
3132 if (netevent_registered && list_empty(&adapter_list)) {
3133 unregister_netevent_notifier(&cxgb4_netevent_nb);
3134 netevent_registered = false;
3136 mutex_unlock(&uld_mutex);
3139 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
3143 mutex_lock(&uld_mutex);
3144 for (i = 0; i < CXGB4_ULD_MAX; i++)
3145 if (adap->uld_handle[i])
3146 ulds[i].state_change(adap->uld_handle[i], new_state);
3147 mutex_unlock(&uld_mutex);
3151 * cxgb4_register_uld - register an upper-layer driver
3152 * @type: the ULD type
3153 * @p: the ULD methods
3155 * Registers an upper-layer driver with this driver and notifies the ULD
3156 * about any presently available devices that support its type. Returns
3157 * %-EBUSY if a ULD of the same type is already registered.
3159 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
3162 struct adapter *adap;
3164 if (type >= CXGB4_ULD_MAX)
3166 mutex_lock(&uld_mutex);
3167 if (ulds[type].add) {
3172 list_for_each_entry(adap, &adapter_list, list_node)
3173 uld_attach(adap, type);
3174 out: mutex_unlock(&uld_mutex);
3177 EXPORT_SYMBOL(cxgb4_register_uld);
3180 * cxgb4_unregister_uld - unregister an upper-layer driver
3181 * @type: the ULD type
3183 * Unregisters an existing upper-layer driver.
3185 int cxgb4_unregister_uld(enum cxgb4_uld type)
3187 struct adapter *adap;
3189 if (type >= CXGB4_ULD_MAX)
3191 mutex_lock(&uld_mutex);
3192 list_for_each_entry(adap, &adapter_list, list_node)
3193 adap->uld_handle[type] = NULL;
3194 ulds[type].add = NULL;
3195 mutex_unlock(&uld_mutex);
3198 EXPORT_SYMBOL(cxgb4_unregister_uld);
3201 * cxgb_up - enable the adapter
3202 * @adap: adapter being enabled
3204 * Called when the first port is enabled, this function performs the
3205 * actions necessary to make an adapter operational, such as completing
3206 * the initialization of HW modules, and enabling interrupts.
3208 * Must be called with the rtnl lock held.
3210 static int cxgb_up(struct adapter *adap)
3214 err = setup_sge_queues(adap);
3217 err = setup_rss(adap);
3221 if (adap->flags & USING_MSIX) {
3222 name_msix_vecs(adap);
3223 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
3224 adap->msix_info[0].desc, adap);
3228 err = request_msix_queue_irqs(adap);
3230 free_irq(adap->msix_info[0].vec, adap);
3234 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
3235 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
3236 adap->port[0]->name, adap);
3242 t4_intr_enable(adap);
3243 adap->flags |= FULL_INIT_DONE;
3244 notify_ulds(adap, CXGB4_STATE_UP);
3248 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
3250 t4_free_sge_resources(adap);
3254 static void cxgb_down(struct adapter *adapter)
3256 t4_intr_disable(adapter);
3257 cancel_work_sync(&adapter->tid_release_task);
3258 cancel_work_sync(&adapter->db_full_task);
3259 cancel_work_sync(&adapter->db_drop_task);
3260 adapter->tid_release_task_busy = false;
3261 adapter->tid_release_head = NULL;
3263 if (adapter->flags & USING_MSIX) {
3264 free_msix_queue_irqs(adapter);
3265 free_irq(adapter->msix_info[0].vec, adapter);
3267 free_irq(adapter->pdev->irq, adapter);
3268 quiesce_rx(adapter);
3269 t4_sge_stop(adapter);
3270 t4_free_sge_resources(adapter);
3271 adapter->flags &= ~FULL_INIT_DONE;
3275 * net_device operations
3277 static int cxgb_open(struct net_device *dev)
3280 struct port_info *pi = netdev_priv(dev);
3281 struct adapter *adapter = pi->adapter;
3283 netif_carrier_off(dev);
3285 if (!(adapter->flags & FULL_INIT_DONE)) {
3286 err = cxgb_up(adapter);
3291 err = link_start(dev);
3293 netif_tx_start_all_queues(dev);
3297 static int cxgb_close(struct net_device *dev)
3299 struct port_info *pi = netdev_priv(dev);
3300 struct adapter *adapter = pi->adapter;
3302 netif_tx_stop_all_queues(dev);
3303 netif_carrier_off(dev);
3304 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
3307 /* Return an error number if the indicated filter isn't writable ...
3309 static int writable_filter(struct filter_entry *f)
3319 /* Delete the filter at the specified index (if valid). The checks for all
3320 * the common problems with doing this like the filter being locked, currently
3321 * pending in another operation, etc.
3323 static int delete_filter(struct adapter *adapter, unsigned int fidx)
3325 struct filter_entry *f;
3328 if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
3331 f = &adapter->tids.ftid_tab[fidx];
3332 ret = writable_filter(f);
3336 return del_filter_wr(adapter, fidx);
3341 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
3342 __be32 sip, __be16 sport, __be16 vlan,
3343 unsigned int queue, unsigned char port, unsigned char mask)
3346 struct filter_entry *f;
3347 struct adapter *adap;
3351 adap = netdev2adap(dev);
3353 /* Adjust stid to correct filter index */
3354 stid -= adap->tids.nstids;
3355 stid += adap->tids.nftids;
3357 /* Check to make sure the filter requested is writable ...
3359 f = &adap->tids.ftid_tab[stid];
3360 ret = writable_filter(f);
3364 /* Clear out any old resources being used by the filter before
3365 * we start constructing the new filter.
3368 clear_filter(adap, f);
3370 /* Clear out filter specifications */
3371 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
3372 f->fs.val.lport = cpu_to_be16(sport);
3373 f->fs.mask.lport = ~0;
3375 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
3376 for (i = 0; i < 4; i++) {
3377 f->fs.val.lip[i] = val[i];
3378 f->fs.mask.lip[i] = ~0;
3380 if (adap->filter_mode & F_PORT) {
3381 f->fs.val.iport = port;
3382 f->fs.mask.iport = mask;
3388 /* Mark filter as locked */
3392 ret = set_filter_wr(adap, stid);
3394 clear_filter(adap, f);
3400 EXPORT_SYMBOL(cxgb4_create_server_filter);
3402 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
3403 unsigned int queue, bool ipv6)
3406 struct filter_entry *f;
3407 struct adapter *adap;
3409 adap = netdev2adap(dev);
3411 /* Adjust stid to correct filter index */
3412 stid -= adap->tids.nstids;
3413 stid += adap->tids.nftids;
3415 f = &adap->tids.ftid_tab[stid];
3416 /* Unlock the filter */
3419 ret = delete_filter(adap, stid);
3425 EXPORT_SYMBOL(cxgb4_remove_server_filter);
3427 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
3428 struct rtnl_link_stats64 *ns)
3430 struct port_stats stats;
3431 struct port_info *p = netdev_priv(dev);
3432 struct adapter *adapter = p->adapter;
3434 spin_lock(&adapter->stats_lock);
3435 t4_get_port_stats(adapter, p->tx_chan, &stats);
3436 spin_unlock(&adapter->stats_lock);
3438 ns->tx_bytes = stats.tx_octets;
3439 ns->tx_packets = stats.tx_frames;
3440 ns->rx_bytes = stats.rx_octets;
3441 ns->rx_packets = stats.rx_frames;
3442 ns->multicast = stats.rx_mcast_frames;
3444 /* detailed rx_errors */
3445 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
3447 ns->rx_over_errors = 0;
3448 ns->rx_crc_errors = stats.rx_fcs_err;
3449 ns->rx_frame_errors = stats.rx_symbol_err;
3450 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
3451 stats.rx_ovflow2 + stats.rx_ovflow3 +
3452 stats.rx_trunc0 + stats.rx_trunc1 +
3453 stats.rx_trunc2 + stats.rx_trunc3;
3454 ns->rx_missed_errors = 0;
3456 /* detailed tx_errors */
3457 ns->tx_aborted_errors = 0;
3458 ns->tx_carrier_errors = 0;
3459 ns->tx_fifo_errors = 0;
3460 ns->tx_heartbeat_errors = 0;
3461 ns->tx_window_errors = 0;
3463 ns->tx_errors = stats.tx_error_frames;
3464 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
3465 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
3469 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
3472 int ret = 0, prtad, devad;
3473 struct port_info *pi = netdev_priv(dev);
3474 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
3478 if (pi->mdio_addr < 0)
3480 data->phy_id = pi->mdio_addr;
3484 if (mdio_phy_id_is_c45(data->phy_id)) {
3485 prtad = mdio_phy_id_prtad(data->phy_id);
3486 devad = mdio_phy_id_devad(data->phy_id);
3487 } else if (data->phy_id < 32) {
3488 prtad = data->phy_id;
3490 data->reg_num &= 0x1f;
3494 mbox = pi->adapter->fn;
3495 if (cmd == SIOCGMIIREG)
3496 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
3497 data->reg_num, &data->val_out);
3499 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
3500 data->reg_num, data->val_in);
3508 static void cxgb_set_rxmode(struct net_device *dev)
3510 /* unfortunately we can't return errors to the stack */
3511 set_rxmode(dev, -1, false);
3514 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
3517 struct port_info *pi = netdev_priv(dev);
3519 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
3521 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
3528 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
3531 struct sockaddr *addr = p;
3532 struct port_info *pi = netdev_priv(dev);
3534 if (!is_valid_ether_addr(addr->sa_data))
3535 return -EADDRNOTAVAIL;
3537 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
3538 pi->xact_addr_filt, addr->sa_data, true, true);
3542 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
3543 pi->xact_addr_filt = ret;
3547 #ifdef CONFIG_NET_POLL_CONTROLLER
3548 static void cxgb_netpoll(struct net_device *dev)
3550 struct port_info *pi = netdev_priv(dev);
3551 struct adapter *adap = pi->adapter;
3553 if (adap->flags & USING_MSIX) {
3555 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
3557 for (i = pi->nqsets; i; i--, rx++)
3558 t4_sge_intr_msix(0, &rx->rspq);
3560 t4_intr_handler(adap)(0, adap);
3564 static const struct net_device_ops cxgb4_netdev_ops = {
3565 .ndo_open = cxgb_open,
3566 .ndo_stop = cxgb_close,
3567 .ndo_start_xmit = t4_eth_xmit,
3568 .ndo_get_stats64 = cxgb_get_stats,
3569 .ndo_set_rx_mode = cxgb_set_rxmode,
3570 .ndo_set_mac_address = cxgb_set_mac_addr,
3571 .ndo_set_features = cxgb_set_features,
3572 .ndo_validate_addr = eth_validate_addr,
3573 .ndo_do_ioctl = cxgb_ioctl,
3574 .ndo_change_mtu = cxgb_change_mtu,
3575 #ifdef CONFIG_NET_POLL_CONTROLLER
3576 .ndo_poll_controller = cxgb_netpoll,
3580 void t4_fatal_err(struct adapter *adap)
3582 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
3583 t4_intr_disable(adap);
3584 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3587 static void setup_memwin(struct adapter *adap)
3591 bar0 = pci_resource_start(adap->pdev, 0); /* truncation intentional */
3592 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
3593 (bar0 + MEMWIN0_BASE) | BIR(0) |
3594 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
3595 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
3596 (bar0 + MEMWIN1_BASE) | BIR(0) |
3597 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
3598 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
3599 (bar0 + MEMWIN2_BASE) | BIR(0) |
3600 WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
3603 static void setup_memwin_rdma(struct adapter *adap)
3605 if (adap->vres.ocq.size) {
3606 unsigned int start, sz_kb;
3608 start = pci_resource_start(adap->pdev, 2) +
3609 OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
3610 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3612 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
3613 start | BIR(1) | WINDOW(ilog2(sz_kb)));
3615 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
3616 adap->vres.ocq.start);
3618 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
3622 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
3627 /* get device capabilities */
3628 memset(c, 0, sizeof(*c));
3629 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3630 FW_CMD_REQUEST | FW_CMD_READ);
3631 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
3632 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
3636 /* select capabilities we'll be using */
3637 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
3639 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
3641 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
3642 } else if (vf_acls) {
3643 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
3646 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3647 FW_CMD_REQUEST | FW_CMD_WRITE);
3648 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
3652 ret = t4_config_glbl_rss(adap, adap->fn,
3653 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
3654 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
3655 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
3659 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
3660 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
3666 /* tweak some settings */
3667 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
3668 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
3669 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
3670 v = t4_read_reg(adap, TP_PIO_DATA);
3671 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
3673 /* first 4 Tx modulation queues point to consecutive Tx channels */
3674 adap->params.tp.tx_modq_map = 0xE4;
3675 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
3676 V_TX_MOD_QUEUE_REQ_MAP(adap->params.tp.tx_modq_map));
3678 /* associate each Tx modulation queue with consecutive Tx channels */
3680 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
3681 &v, 1, A_TP_TX_SCHED_HDR);
3682 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
3683 &v, 1, A_TP_TX_SCHED_FIFO);
3684 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
3685 &v, 1, A_TP_TX_SCHED_PCMD);
3687 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3688 if (is_offload(adap)) {
3689 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0,
3690 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3691 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3692 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3693 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3694 t4_write_reg(adap, A_TP_TX_MOD_CHANNEL_WEIGHT,
3695 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3696 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3697 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3698 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3701 /* get basic stuff going */
3702 return t4_early_init(adap, adap->fn);
3706 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
3708 #define MAX_ATIDS 8192U
3711 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3713 * If the firmware we're dealing with has Configuration File support, then
3714 * we use that to perform all configuration
3718 * Tweak configuration based on module parameters, etc. Most of these have
3719 * defaults assigned to them by Firmware Configuration Files (if we're using
3720 * them) but need to be explicitly set if we're using hard-coded
3721 * initialization. But even in the case of using Firmware Configuration
3722 * Files, we'd like to expose the ability to change these via module
3723 * parameters so these are essentially common tweaks/settings for
3724 * Configuration Files and hard-coded initialization ...
3726 static int adap_init0_tweaks(struct adapter *adapter)
3729 * Fix up various Host-Dependent Parameters like Page Size, Cache
3730 * Line Size, etc. The firmware default is for a 4KB Page Size and
3731 * 64B Cache Line Size ...
3733 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
3736 * Process module parameters which affect early initialization.
3738 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
3739 dev_err(&adapter->pdev->dev,
3740 "Ignoring illegal rx_dma_offset=%d, using 2\n",
3744 t4_set_reg_field(adapter, SGE_CONTROL,
3746 PKTSHIFT(rx_dma_offset));
3749 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
3750 * adds the pseudo header itself.
3752 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG,
3753 CSUM_HAS_PSEUDO_HDR, 0);
3759 * Attempt to initialize the adapter via a Firmware Configuration File.
3761 static int adap_init0_config(struct adapter *adapter, int reset)
3763 struct fw_caps_config_cmd caps_cmd;
3764 const struct firmware *cf;
3765 unsigned long mtype = 0, maddr = 0;
3766 u32 finiver, finicsum, cfcsum;
3767 int ret, using_flash;
3768 char *fw_config_file, fw_config_file_path[256];
3771 * Reset device if necessary.
3774 ret = t4_fw_reset(adapter, adapter->mbox,
3775 PIORSTMODE | PIORST);
3781 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3782 * then use that. Otherwise, use the configuration file stored
3783 * in the adapter flash ...
3785 switch (CHELSIO_CHIP_VERSION(adapter->chip)) {
3787 fw_config_file = FW_CFNAME;
3790 fw_config_file = FW5_CFNAME;
3793 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3794 adapter->pdev->device);
3799 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3802 mtype = FW_MEMTYPE_CF_FLASH;
3803 maddr = t4_flash_cfg_addr(adapter);
3805 u32 params[7], val[7];
3808 if (cf->size >= FLASH_CFG_MAX_SIZE)
3811 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3812 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
3813 ret = t4_query_params(adapter, adapter->mbox,
3814 adapter->fn, 0, 1, params, val);
3817 * For t4_memory_write() below addresses and
3818 * sizes have to be in terms of multiples of 4
3819 * bytes. So, if the Configuration File isn't
3820 * a multiple of 4 bytes in length we'll have
3821 * to write that out separately since we can't
3822 * guarantee that the bytes following the
3823 * residual byte in the buffer returned by
3824 * request_firmware() are zeroed out ...
3826 size_t resid = cf->size & 0x3;
3827 size_t size = cf->size & ~0x3;
3828 __be32 *data = (__be32 *)cf->data;
3830 mtype = FW_PARAMS_PARAM_Y_GET(val[0]);
3831 maddr = FW_PARAMS_PARAM_Z_GET(val[0]) << 16;
3833 ret = t4_memory_write(adapter, mtype, maddr,
3835 if (ret == 0 && resid != 0) {
3842 last.word = data[size >> 2];
3843 for (i = resid; i < 4; i++)
3845 ret = t4_memory_write(adapter, mtype,
3852 release_firmware(cf);
3858 * Issue a Capability Configuration command to the firmware to get it
3859 * to parse the Configuration File. We don't use t4_fw_config_file()
3860 * because we want the ability to modify various features after we've
3861 * processed the configuration file ...
3863 memset(&caps_cmd, 0, sizeof(caps_cmd));
3864 caps_cmd.op_to_write =
3865 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3868 caps_cmd.cfvalid_to_len16 =
3869 htonl(FW_CAPS_CONFIG_CMD_CFVALID |
3870 FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3871 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
3872 FW_LEN16(caps_cmd));
3873 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3878 finiver = ntohl(caps_cmd.finiver);
3879 finicsum = ntohl(caps_cmd.finicsum);
3880 cfcsum = ntohl(caps_cmd.cfcsum);
3881 if (finicsum != cfcsum)
3882 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3883 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3887 * And now tell the firmware to use the configuration we just loaded.
3889 caps_cmd.op_to_write =
3890 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3893 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3894 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3900 * Tweak configuration based on system architecture, module
3903 ret = adap_init0_tweaks(adapter);
3908 * And finally tell the firmware to initialize itself using the
3909 * parameters from the Configuration File.
3911 ret = t4_fw_initialize(adapter, adapter->mbox);
3915 sprintf(fw_config_file_path, "/lib/firmware/%s", fw_config_file);
3917 * Return successfully and note that we're operating with parameters
3918 * not supplied by the driver, rather than from hard-wired
3919 * initialization constants burried in the driver.
3921 adapter->flags |= USING_SOFT_PARAMS;
3922 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
3923 "Configuration File %s, version %#x, computed checksum %#x\n",
3926 : fw_config_file_path),
3931 * Something bad happened. Return the error ... (If the "error"
3932 * is that there's no Configuration File on the adapter we don't
3933 * want to issue a warning since this is fairly common.)
3937 dev_warn(adapter->pdev_dev, "Configuration file error %d\n",
3943 * Attempt to initialize the adapter via hard-coded, driver supplied
3946 static int adap_init0_no_config(struct adapter *adapter, int reset)
3948 struct sge *s = &adapter->sge;
3949 struct fw_caps_config_cmd caps_cmd;
3954 * Reset device if necessary
3957 ret = t4_fw_reset(adapter, adapter->mbox,
3958 PIORSTMODE | PIORST);
3964 * Get device capabilities and select which we'll be using.
3966 memset(&caps_cmd, 0, sizeof(caps_cmd));
3967 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3968 FW_CMD_REQUEST | FW_CMD_READ);
3969 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3970 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3975 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
3977 caps_cmd.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
3979 caps_cmd.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
3980 } else if (vf_acls) {
3981 dev_err(adapter->pdev_dev, "virtualization ACLs not supported");
3984 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3985 FW_CMD_REQUEST | FW_CMD_WRITE);
3986 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3992 * Tweak configuration based on system architecture, module
3995 ret = adap_init0_tweaks(adapter);
4000 * Select RSS Global Mode we want to use. We use "Basic Virtual"
4001 * mode which maps each Virtual Interface to its own section of
4002 * the RSS Table and we turn on all map and hash enables ...
4004 adapter->flags |= RSS_TNLALLLOOKUP;
4005 ret = t4_config_glbl_rss(adapter, adapter->mbox,
4006 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4007 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4008 FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ |
4009 ((adapter->flags & RSS_TNLALLLOOKUP) ?
4010 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP : 0));
4015 * Set up our own fundamental resource provisioning ...
4017 ret = t4_cfg_pfvf(adapter, adapter->mbox, adapter->fn, 0,
4018 PFRES_NEQ, PFRES_NETHCTRL,
4019 PFRES_NIQFLINT, PFRES_NIQ,
4020 PFRES_TC, PFRES_NVI,
4021 FW_PFVF_CMD_CMASK_MASK,
4022 pfvfres_pmask(adapter, adapter->fn, 0),
4024 PFRES_R_CAPS, PFRES_WX_CAPS);
4029 * Perform low level SGE initialization. We need to do this before we
4030 * send the firmware the INITIALIZE command because that will cause
4031 * any other PF Drivers which are waiting for the Master
4032 * Initialization to proceed forward.
4034 for (i = 0; i < SGE_NTIMERS - 1; i++)
4035 s->timer_val[i] = min(intr_holdoff[i], MAX_SGE_TIMERVAL);
4036 s->timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
4037 s->counter_val[0] = 1;
4038 for (i = 1; i < SGE_NCOUNTERS; i++)
4039 s->counter_val[i] = min(intr_cnt[i - 1],
4040 THRESHOLD_0_GET(THRESHOLD_0_MASK));
4041 t4_sge_init(adapter);
4043 #ifdef CONFIG_PCI_IOV
4045 * Provision resource limits for Virtual Functions. We currently
4046 * grant them all the same static resource limits except for the Port
4047 * Access Rights Mask which we're assigning based on the PF. All of
4048 * the static provisioning stuff for both the PF and VF really needs
4049 * to be managed in a persistent manner for each device which the
4050 * firmware controls.
4054 int max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
4055 NUM_OF_PF_WITH_SRIOV_T5;
4057 for (pf = 0; pf < max_no_pf; pf++) {
4058 if (num_vf[pf] <= 0)
4061 /* VF numbering starts at 1! */
4062 for (vf = 1; vf <= num_vf[pf]; vf++) {
4063 ret = t4_cfg_pfvf(adapter, adapter->mbox,
4065 VFRES_NEQ, VFRES_NETHCTRL,
4066 VFRES_NIQFLINT, VFRES_NIQ,
4067 VFRES_TC, VFRES_NVI,
4068 FW_PFVF_CMD_CMASK_MASK,
4072 VFRES_R_CAPS, VFRES_WX_CAPS);
4074 dev_warn(adapter->pdev_dev,
4076 "provision pf/vf=%d/%d; "
4077 "err=%d\n", pf, vf, ret);
4084 * Set up the default filter mode. Later we'll want to implement this
4085 * via a firmware command, etc. ... This needs to be done before the
4086 * firmare initialization command ... If the selected set of fields
4087 * isn't equal to the default value, we'll need to make sure that the
4088 * field selections will fit in the 36-bit budget.
4090 if (tp_vlan_pri_map != TP_VLAN_PRI_MAP_DEFAULT) {
4093 for (j = TP_VLAN_PRI_MAP_FIRST; j <= TP_VLAN_PRI_MAP_LAST; j++)
4094 switch (tp_vlan_pri_map & (1 << j)) {
4096 /* compressed filter field not enabled */
4116 case ETHERTYPE_MASK:
4122 case MPSHITTYPE_MASK:
4125 case FRAGMENTATION_MASK:
4131 dev_err(adapter->pdev_dev,
4132 "tp_vlan_pri_map=%#x needs %d bits > 36;"\
4133 " using %#x\n", tp_vlan_pri_map, bits,
4134 TP_VLAN_PRI_MAP_DEFAULT);
4135 tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
4138 v = tp_vlan_pri_map;
4139 t4_write_indirect(adapter, TP_PIO_ADDR, TP_PIO_DATA,
4140 &v, 1, TP_VLAN_PRI_MAP);
4143 * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
4144 * to support any of the compressed filter fields above. Newer
4145 * versions of the firmware do this automatically but it doesn't hurt
4146 * to set it here. Meanwhile, we do _not_ need to set Lookup Every
4147 * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
4148 * since the firmware automatically turns this on and off when we have
4149 * a non-zero number of filters active (since it does have a
4150 * performance impact).
4152 if (tp_vlan_pri_map)
4153 t4_set_reg_field(adapter, TP_GLOBAL_CONFIG,
4154 FIVETUPLELOOKUP_MASK,
4155 FIVETUPLELOOKUP_MASK);
4158 * Tweak some settings.
4160 t4_write_reg(adapter, TP_SHIFT_CNT, SYNSHIFTMAX(6) |
4161 RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
4162 PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
4163 KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
4166 * Get basic stuff going by issuing the Firmware Initialize command.
4167 * Note that this _must_ be after all PFVF commands ...
4169 ret = t4_fw_initialize(adapter, adapter->mbox);
4174 * Return successfully!
4176 dev_info(adapter->pdev_dev, "Successfully configured using built-in "\
4177 "driver parameters\n");
4181 * Something bad happened. Return the error ...
4188 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4190 static int adap_init0(struct adapter *adap)
4194 enum dev_state state;
4195 u32 params[7], val[7];
4196 struct fw_caps_config_cmd caps_cmd;
4200 * Contact FW, advertising Master capability (and potentially forcing
4201 * ourselves as the Master PF if our module parameter force_init is
4204 ret = t4_fw_hello(adap, adap->mbox, adap->fn,
4205 force_init ? MASTER_MUST : MASTER_MAY,
4208 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4212 if (ret == adap->mbox)
4213 adap->flags |= MASTER_PF;
4214 if (force_init && state == DEV_STATE_INIT)
4215 state = DEV_STATE_UNINIT;
4218 * If we're the Master PF Driver and the device is uninitialized,
4219 * then let's consider upgrading the firmware ... (We always want
4220 * to check the firmware version number in order to A. get it for
4221 * later reporting and B. to warn if the currently loaded firmware
4222 * is excessively mismatched relative to the driver.)
4224 ret = t4_check_fw_version(adap);
4225 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
4226 if (ret == -EINVAL || ret > 0) {
4227 if (upgrade_fw(adap) >= 0) {
4229 * Note that the chip was reset as part of the
4230 * firmware upgrade so we don't reset it again
4231 * below and grab the new firmware version.
4234 ret = t4_check_fw_version(adap);
4242 * Grab VPD parameters. This should be done after we establish a
4243 * connection to the firmware since some of the VPD parameters
4244 * (notably the Core Clock frequency) are retrieved via requests to
4245 * the firmware. On the other hand, we need these fairly early on
4246 * so we do this right after getting ahold of the firmware.
4248 ret = get_vpd_params(adap, &adap->params.vpd);
4253 * Find out what ports are available to us. Note that we need to do
4254 * this before calling adap_init0_no_config() since it needs nports
4258 FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4259 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
4260 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1, &v, &port_vec);
4264 adap->params.nports = hweight32(port_vec);
4265 adap->params.portvec = port_vec;
4268 * If the firmware is initialized already (and we're not forcing a
4269 * master initialization), note that we're living with existing
4270 * adapter parameters. Otherwise, it's time to try initializing the
4273 if (state == DEV_STATE_INIT) {
4274 dev_info(adap->pdev_dev, "Coming up as %s: "\
4275 "Adapter already initialized\n",
4276 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
4277 adap->flags |= USING_SOFT_PARAMS;
4279 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4280 "Initializing adapter\n");
4283 * If the firmware doesn't support Configuration
4284 * Files warn user and exit,
4287 dev_warn(adap->pdev_dev, "Firmware doesn't support "
4288 "configuration file.\n");
4290 ret = adap_init0_no_config(adap, reset);
4293 * Find out whether we're dealing with a version of
4294 * the firmware which has configuration file support.
4296 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4297 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4298 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1,
4302 * If the firmware doesn't support Configuration
4303 * Files, use the old Driver-based, hard-wired
4304 * initialization. Otherwise, try using the
4305 * Configuration File support and fall back to the
4306 * Driver-based initialization if there's no
4307 * Configuration File found.
4310 ret = adap_init0_no_config(adap, reset);
4313 * The firmware provides us with a memory
4314 * buffer where we can load a Configuration
4315 * File from the host if we want to override
4316 * the Configuration File in flash.
4319 ret = adap_init0_config(adap, reset);
4320 if (ret == -ENOENT) {
4321 dev_info(adap->pdev_dev,
4322 "No Configuration File present "
4323 "on adapter. Using hard-wired "
4324 "configuration parameters.\n");
4325 ret = adap_init0_no_config(adap, reset);
4330 dev_err(adap->pdev_dev,
4331 "could not initialize adapter, error %d\n",
4338 * If we're living with non-hard-coded parameters (either from a
4339 * Firmware Configuration File or values programmed by a different PF
4340 * Driver), give the SGE code a chance to pull in anything that it
4341 * needs ... Note that this must be called after we retrieve our VPD
4342 * parameters in order to know how to convert core ticks to seconds.
4344 if (adap->flags & USING_SOFT_PARAMS) {
4345 ret = t4_sge_init(adap);
4350 if (is_bypass_device(adap->pdev->device))
4351 adap->params.bypass = 1;
4354 * Grab some of our basic fundamental operating parameters.
4356 #define FW_PARAM_DEV(param) \
4357 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
4358 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
4360 #define FW_PARAM_PFVF(param) \
4361 FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
4362 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)| \
4363 FW_PARAMS_PARAM_Y(0) | \
4364 FW_PARAMS_PARAM_Z(0)
4366 params[0] = FW_PARAM_PFVF(EQ_START);
4367 params[1] = FW_PARAM_PFVF(L2T_START);
4368 params[2] = FW_PARAM_PFVF(L2T_END);
4369 params[3] = FW_PARAM_PFVF(FILTER_START);
4370 params[4] = FW_PARAM_PFVF(FILTER_END);
4371 params[5] = FW_PARAM_PFVF(IQFLINT_START);
4372 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6, params, val);
4375 adap->sge.egr_start = val[0];
4376 adap->l2t_start = val[1];
4377 adap->l2t_end = val[2];
4378 adap->tids.ftid_base = val[3];
4379 adap->tids.nftids = val[4] - val[3] + 1;
4380 adap->sge.ingr_start = val[5];
4382 /* query params related to active filter region */
4383 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4384 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4385 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2, params, val);
4386 /* If Active filter size is set we enable establishing
4387 * offload connection through firmware work request
4389 if ((val[0] != val[1]) && (ret >= 0)) {
4390 adap->flags |= FW_OFLD_CONN;
4391 adap->tids.aftid_base = val[0];
4392 adap->tids.aftid_end = val[1];
4396 * Get device capabilities so we can determine what resources we need
4399 memset(&caps_cmd, 0, sizeof(caps_cmd));
4400 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4401 FW_CMD_REQUEST | FW_CMD_READ);
4402 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4403 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4408 if (caps_cmd.ofldcaps) {
4409 /* query offload-related parameters */
4410 params[0] = FW_PARAM_DEV(NTID);
4411 params[1] = FW_PARAM_PFVF(SERVER_START);
4412 params[2] = FW_PARAM_PFVF(SERVER_END);
4413 params[3] = FW_PARAM_PFVF(TDDP_START);
4414 params[4] = FW_PARAM_PFVF(TDDP_END);
4415 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4416 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
4420 adap->tids.ntids = val[0];
4421 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
4422 adap->tids.stid_base = val[1];
4423 adap->tids.nstids = val[2] - val[1] + 1;
4425 * Setup server filter region. Divide the availble filter
4426 * region into two parts. Regular filters get 1/3rd and server
4427 * filters get 2/3rd part. This is only enabled if workarond
4429 * 1. For regular filters.
4430 * 2. Server filter: This are special filters which are used
4431 * to redirect SYN packets to offload queue.
4433 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
4434 adap->tids.sftid_base = adap->tids.ftid_base +
4435 DIV_ROUND_UP(adap->tids.nftids, 3);
4436 adap->tids.nsftids = adap->tids.nftids -
4437 DIV_ROUND_UP(adap->tids.nftids, 3);
4438 adap->tids.nftids = adap->tids.sftid_base -
4439 adap->tids.ftid_base;
4441 adap->vres.ddp.start = val[3];
4442 adap->vres.ddp.size = val[4] - val[3] + 1;
4443 adap->params.ofldq_wr_cred = val[5];
4445 adap->params.offload = 1;
4447 if (caps_cmd.rdmacaps) {
4448 params[0] = FW_PARAM_PFVF(STAG_START);
4449 params[1] = FW_PARAM_PFVF(STAG_END);
4450 params[2] = FW_PARAM_PFVF(RQ_START);
4451 params[3] = FW_PARAM_PFVF(RQ_END);
4452 params[4] = FW_PARAM_PFVF(PBL_START);
4453 params[5] = FW_PARAM_PFVF(PBL_END);
4454 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
4458 adap->vres.stag.start = val[0];
4459 adap->vres.stag.size = val[1] - val[0] + 1;
4460 adap->vres.rq.start = val[2];
4461 adap->vres.rq.size = val[3] - val[2] + 1;
4462 adap->vres.pbl.start = val[4];
4463 adap->vres.pbl.size = val[5] - val[4] + 1;
4465 params[0] = FW_PARAM_PFVF(SQRQ_START);
4466 params[1] = FW_PARAM_PFVF(SQRQ_END);
4467 params[2] = FW_PARAM_PFVF(CQ_START);
4468 params[3] = FW_PARAM_PFVF(CQ_END);
4469 params[4] = FW_PARAM_PFVF(OCQ_START);
4470 params[5] = FW_PARAM_PFVF(OCQ_END);
4471 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
4474 adap->vres.qp.start = val[0];
4475 adap->vres.qp.size = val[1] - val[0] + 1;
4476 adap->vres.cq.start = val[2];
4477 adap->vres.cq.size = val[3] - val[2] + 1;
4478 adap->vres.ocq.start = val[4];
4479 adap->vres.ocq.size = val[5] - val[4] + 1;
4481 if (caps_cmd.iscsicaps) {
4482 params[0] = FW_PARAM_PFVF(ISCSI_START);
4483 params[1] = FW_PARAM_PFVF(ISCSI_END);
4484 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2,
4488 adap->vres.iscsi.start = val[0];
4489 adap->vres.iscsi.size = val[1] - val[0] + 1;
4491 #undef FW_PARAM_PFVF
4495 * These are finalized by FW initialization, load their values now.
4497 v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
4498 adap->params.tp.tre = TIMERRESOLUTION_GET(v);
4499 adap->params.tp.dack_re = DELAYEDACKRESOLUTION_GET(v);
4500 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
4501 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4502 adap->params.b_wnd);
4504 /* MODQ_REQ_MAP defaults to setting queues 0-3 to chan 0-3 */
4505 for (j = 0; j < NCHAN; j++)
4506 adap->params.tp.tx_modq[j] = j;
4508 t4_read_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4509 &adap->filter_mode, 1,
4512 adap->flags |= FW_OK;
4516 * Something bad happened. If a command timed out or failed with EIO
4517 * FW does not operate within its spec or something catastrophic
4518 * happened to HW/FW, stop issuing commands.
4521 if (ret != -ETIMEDOUT && ret != -EIO)
4522 t4_fw_bye(adap, adap->mbox);
4528 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4529 pci_channel_state_t state)
4532 struct adapter *adap = pci_get_drvdata(pdev);
4538 adap->flags &= ~FW_OK;
4539 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
4540 for_each_port(adap, i) {
4541 struct net_device *dev = adap->port[i];
4543 netif_device_detach(dev);
4544 netif_carrier_off(dev);
4546 if (adap->flags & FULL_INIT_DONE)
4549 pci_disable_device(pdev);
4550 out: return state == pci_channel_io_perm_failure ?
4551 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
4554 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
4557 struct fw_caps_config_cmd c;
4558 struct adapter *adap = pci_get_drvdata(pdev);
4561 pci_restore_state(pdev);
4562 pci_save_state(pdev);
4563 return PCI_ERS_RESULT_RECOVERED;
4566 if (pci_enable_device(pdev)) {
4567 dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
4568 return PCI_ERS_RESULT_DISCONNECT;
4571 pci_set_master(pdev);
4572 pci_restore_state(pdev);
4573 pci_save_state(pdev);
4574 pci_cleanup_aer_uncorrect_error_status(pdev);
4576 if (t4_wait_dev_ready(adap) < 0)
4577 return PCI_ERS_RESULT_DISCONNECT;
4578 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL))
4579 return PCI_ERS_RESULT_DISCONNECT;
4580 adap->flags |= FW_OK;
4581 if (adap_init1(adap, &c))
4582 return PCI_ERS_RESULT_DISCONNECT;
4584 for_each_port(adap, i) {
4585 struct port_info *p = adap2pinfo(adap, i);
4587 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
4590 return PCI_ERS_RESULT_DISCONNECT;
4592 p->xact_addr_filt = -1;
4595 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4596 adap->params.b_wnd);
4599 return PCI_ERS_RESULT_DISCONNECT;
4600 return PCI_ERS_RESULT_RECOVERED;
4603 static void eeh_resume(struct pci_dev *pdev)
4606 struct adapter *adap = pci_get_drvdata(pdev);
4612 for_each_port(adap, i) {
4613 struct net_device *dev = adap->port[i];
4615 if (netif_running(dev)) {
4617 cxgb_set_rxmode(dev);
4619 netif_device_attach(dev);
4624 static const struct pci_error_handlers cxgb4_eeh = {
4625 .error_detected = eeh_err_detected,
4626 .slot_reset = eeh_slot_reset,
4627 .resume = eeh_resume,
4630 static inline bool is_10g_port(const struct link_config *lc)
4632 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
4635 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
4636 unsigned int size, unsigned int iqe_size)
4638 q->intr_params = QINTR_TIMER_IDX(timer_idx) |
4639 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
4640 q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
4641 q->iqe_len = iqe_size;
4646 * Perform default configuration of DMA queues depending on the number and type
4647 * of ports we found and the number of available CPUs. Most settings can be
4648 * modified by the admin prior to actual use.
4650 static void cfg_queues(struct adapter *adap)
4652 struct sge *s = &adap->sge;
4653 int i, q10g = 0, n10g = 0, qidx = 0;
4655 for_each_port(adap, i)
4656 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
4659 * We default to 1 queue per non-10G port and up to # of cores queues
4663 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
4664 if (q10g > netif_get_num_default_rss_queues())
4665 q10g = netif_get_num_default_rss_queues();
4667 for_each_port(adap, i) {
4668 struct port_info *pi = adap2pinfo(adap, i);
4670 pi->first_qset = qidx;
4671 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
4676 s->max_ethqsets = qidx; /* MSI-X may lower it later */
4678 if (is_offload(adap)) {
4680 * For offload we use 1 queue/channel if all ports are up to 1G,
4681 * otherwise we divide all available queues amongst the channels
4682 * capped by the number of available cores.
4685 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
4687 s->ofldqsets = roundup(i, adap->params.nports);
4689 s->ofldqsets = adap->params.nports;
4690 /* For RDMA one Rx queue per channel suffices */
4691 s->rdmaqs = adap->params.nports;
4694 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4695 struct sge_eth_rxq *r = &s->ethrxq[i];
4697 init_rspq(&r->rspq, 0, 0, 1024, 64);
4701 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4702 s->ethtxq[i].q.size = 1024;
4704 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4705 s->ctrlq[i].q.size = 512;
4707 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
4708 s->ofldtxq[i].q.size = 1024;
4710 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
4711 struct sge_ofld_rxq *r = &s->ofldrxq[i];
4713 init_rspq(&r->rspq, 0, 0, 1024, 64);
4714 r->rspq.uld = CXGB4_ULD_ISCSI;
4718 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
4719 struct sge_ofld_rxq *r = &s->rdmarxq[i];
4721 init_rspq(&r->rspq, 0, 0, 511, 64);
4722 r->rspq.uld = CXGB4_ULD_RDMA;
4726 init_rspq(&s->fw_evtq, 6, 0, 512, 64);
4727 init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
4731 * Reduce the number of Ethernet queues across all ports to at most n.
4732 * n provides at least one queue per port.
4734 static void reduce_ethqs(struct adapter *adap, int n)
4737 struct port_info *pi;
4739 while (n < adap->sge.ethqsets)
4740 for_each_port(adap, i) {
4741 pi = adap2pinfo(adap, i);
4742 if (pi->nqsets > 1) {
4744 adap->sge.ethqsets--;
4745 if (adap->sge.ethqsets <= n)
4751 for_each_port(adap, i) {
4752 pi = adap2pinfo(adap, i);
4758 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
4759 #define EXTRA_VECS 2
4761 static int enable_msix(struct adapter *adap)
4764 int i, err, want, need;
4765 struct sge *s = &adap->sge;
4766 unsigned int nchan = adap->params.nports;
4767 struct msix_entry entries[MAX_INGQ + 1];
4769 for (i = 0; i < ARRAY_SIZE(entries); ++i)
4770 entries[i].entry = i;
4772 want = s->max_ethqsets + EXTRA_VECS;
4773 if (is_offload(adap)) {
4774 want += s->rdmaqs + s->ofldqsets;
4775 /* need nchan for each possible ULD */
4776 ofld_need = 2 * nchan;
4778 need = adap->params.nports + EXTRA_VECS + ofld_need;
4780 while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
4785 * Distribute available vectors to the various queue groups.
4786 * Every group gets its minimum requirement and NIC gets top
4787 * priority for leftovers.
4789 i = want - EXTRA_VECS - ofld_need;
4790 if (i < s->max_ethqsets) {
4791 s->max_ethqsets = i;
4792 if (i < s->ethqsets)
4793 reduce_ethqs(adap, i);
4795 if (is_offload(adap)) {
4796 i = want - EXTRA_VECS - s->max_ethqsets;
4797 i -= ofld_need - nchan;
4798 s->ofldqsets = (i / nchan) * nchan; /* round down */
4800 for (i = 0; i < want; ++i)
4801 adap->msix_info[i].vec = entries[i].vector;
4803 dev_info(adap->pdev_dev,
4804 "only %d MSI-X vectors left, not using MSI-X\n", err);
4810 static int init_rss(struct adapter *adap)
4814 for_each_port(adap, i) {
4815 struct port_info *pi = adap2pinfo(adap, i);
4817 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
4820 for (j = 0; j < pi->rss_size; j++)
4821 pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
4826 static void print_port_info(const struct net_device *dev)
4828 static const char *base[] = {
4829 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
4830 "KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4"
4835 const char *spd = "";
4836 const struct port_info *pi = netdev_priv(dev);
4837 const struct adapter *adap = pi->adapter;
4839 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
4841 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
4844 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
4845 bufp += sprintf(bufp, "100/");
4846 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
4847 bufp += sprintf(bufp, "1000/");
4848 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
4849 bufp += sprintf(bufp, "10G/");
4852 sprintf(bufp, "BASE-%s", base[pi->port_type]);
4854 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
4855 adap->params.vpd.id,
4856 CHELSIO_CHIP_RELEASE(adap->params.rev), buf,
4857 is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
4858 (adap->flags & USING_MSIX) ? " MSI-X" :
4859 (adap->flags & USING_MSI) ? " MSI" : "");
4860 netdev_info(dev, "S/N: %s, E/C: %s\n",
4861 adap->params.vpd.sn, adap->params.vpd.ec);
4864 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
4866 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
4870 * Free the following resources:
4871 * - memory used for tables
4874 * - resources FW is holding for us
4876 static void free_some_resources(struct adapter *adapter)
4880 t4_free_mem(adapter->l2t);
4881 t4_free_mem(adapter->tids.tid_tab);
4882 disable_msi(adapter);
4884 for_each_port(adapter, i)
4885 if (adapter->port[i]) {
4886 kfree(adap2pinfo(adapter, i)->rss);
4887 free_netdev(adapter->port[i]);
4889 if (adapter->flags & FW_OK)
4890 t4_fw_bye(adapter, adapter->fn);
4893 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
4894 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
4895 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
4897 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4900 struct port_info *pi;
4901 bool highdma = false;
4902 struct adapter *adapter = NULL;
4903 #ifdef CONFIG_PCI_IOV
4907 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
4909 err = pci_request_regions(pdev, KBUILD_MODNAME);
4911 /* Just info, some other driver may have claimed the device. */
4912 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
4916 /* We control everything through one PF */
4917 func = PCI_FUNC(pdev->devfn);
4918 if (func != ent->driver_data) {
4919 pci_save_state(pdev); /* to restore SR-IOV later */
4923 err = pci_enable_device(pdev);
4925 dev_err(&pdev->dev, "cannot enable PCI device\n");
4926 goto out_release_regions;
4929 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
4931 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4933 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
4934 "coherent allocations\n");
4935 goto out_disable_device;
4938 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4940 dev_err(&pdev->dev, "no usable DMA configuration\n");
4941 goto out_disable_device;
4945 pci_enable_pcie_error_reporting(pdev);
4946 enable_pcie_relaxed_ordering(pdev);
4947 pci_set_master(pdev);
4948 pci_save_state(pdev);
4950 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
4953 goto out_disable_device;
4956 adapter->regs = pci_ioremap_bar(pdev, 0);
4957 if (!adapter->regs) {
4958 dev_err(&pdev->dev, "cannot map device registers\n");
4960 goto out_free_adapter;
4963 adapter->pdev = pdev;
4964 adapter->pdev_dev = &pdev->dev;
4965 adapter->mbox = func;
4967 adapter->msg_enable = dflt_msg_enable;
4968 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
4970 spin_lock_init(&adapter->stats_lock);
4971 spin_lock_init(&adapter->tid_release_lock);
4973 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
4974 INIT_WORK(&adapter->db_full_task, process_db_full);
4975 INIT_WORK(&adapter->db_drop_task, process_db_drop);
4977 err = t4_prep_adapter(adapter);
4980 setup_memwin(adapter);
4981 err = adap_init0(adapter);
4982 setup_memwin_rdma(adapter);
4986 for_each_port(adapter, i) {
4987 struct net_device *netdev;
4989 netdev = alloc_etherdev_mq(sizeof(struct port_info),
4996 SET_NETDEV_DEV(netdev, &pdev->dev);
4998 adapter->port[i] = netdev;
4999 pi = netdev_priv(netdev);
5000 pi->adapter = adapter;
5001 pi->xact_addr_filt = -1;
5003 netdev->irq = pdev->irq;
5005 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5006 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5007 NETIF_F_RXCSUM | NETIF_F_RXHASH |
5008 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
5010 netdev->hw_features |= NETIF_F_HIGHDMA;
5011 netdev->features |= netdev->hw_features;
5012 netdev->vlan_features = netdev->features & VLAN_FEAT;
5014 netdev->priv_flags |= IFF_UNICAST_FLT;
5016 netdev->netdev_ops = &cxgb4_netdev_ops;
5017 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
5020 pci_set_drvdata(pdev, adapter);
5022 if (adapter->flags & FW_OK) {
5023 err = t4_port_init(adapter, func, func, 0);
5029 * Configure queues and allocate tables now, they can be needed as
5030 * soon as the first register_netdev completes.
5032 cfg_queues(adapter);
5034 adapter->l2t = t4_init_l2t();
5035 if (!adapter->l2t) {
5036 /* We tolerate a lack of L2T, giving up some functionality */
5037 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5038 adapter->params.offload = 0;
5041 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
5042 dev_warn(&pdev->dev, "could not allocate TID table, "
5044 adapter->params.offload = 0;
5047 /* See what interrupts we'll be using */
5048 if (msi > 1 && enable_msix(adapter) == 0)
5049 adapter->flags |= USING_MSIX;
5050 else if (msi > 0 && pci_enable_msi(pdev) == 0)
5051 adapter->flags |= USING_MSI;
5053 err = init_rss(adapter);
5058 * The card is now ready to go. If any errors occur during device
5059 * registration we do not fail the whole card but rather proceed only
5060 * with the ports we manage to register successfully. However we must
5061 * register at least one net device.
5063 for_each_port(adapter, i) {
5064 pi = adap2pinfo(adapter, i);
5065 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5066 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5068 err = register_netdev(adapter->port[i]);
5071 adapter->chan_map[pi->tx_chan] = i;
5072 print_port_info(adapter->port[i]);
5075 dev_err(&pdev->dev, "could not register any net devices\n");
5079 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5083 if (cxgb4_debugfs_root) {
5084 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5085 cxgb4_debugfs_root);
5086 setup_debugfs(adapter);
5089 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5090 pdev->needs_freset = 1;
5092 if (is_offload(adapter))
5093 attach_ulds(adapter);
5096 #ifdef CONFIG_PCI_IOV
5097 max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
5098 NUM_OF_PF_WITH_SRIOV_T5;
5100 if (func < max_no_pf && num_vf[func] > 0)
5101 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
5102 dev_info(&pdev->dev,
5103 "instantiated %u virtual functions\n",
5109 free_some_resources(adapter);
5111 iounmap(adapter->regs);
5115 pci_disable_pcie_error_reporting(pdev);
5116 pci_disable_device(pdev);
5117 out_release_regions:
5118 pci_release_regions(pdev);
5119 pci_set_drvdata(pdev, NULL);
5123 static void remove_one(struct pci_dev *pdev)
5125 struct adapter *adapter = pci_get_drvdata(pdev);
5127 #ifdef CONFIG_PCI_IOV
5128 pci_disable_sriov(pdev);
5135 if (is_offload(adapter))
5136 detach_ulds(adapter);
5138 for_each_port(adapter, i)
5139 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5140 unregister_netdev(adapter->port[i]);
5142 if (adapter->debugfs_root)
5143 debugfs_remove_recursive(adapter->debugfs_root);
5145 /* If we allocated filters, free up state associated with any
5148 if (adapter->tids.ftid_tab) {
5149 struct filter_entry *f = &adapter->tids.ftid_tab[0];
5150 for (i = 0; i < (adapter->tids.nftids +
5151 adapter->tids.nsftids); i++, f++)
5153 clear_filter(adapter, f);
5156 if (adapter->flags & FULL_INIT_DONE)
5159 free_some_resources(adapter);
5160 iounmap(adapter->regs);
5162 pci_disable_pcie_error_reporting(pdev);
5163 pci_disable_device(pdev);
5164 pci_release_regions(pdev);
5165 pci_set_drvdata(pdev, NULL);
5167 pci_release_regions(pdev);
5170 static struct pci_driver cxgb4_driver = {
5171 .name = KBUILD_MODNAME,
5172 .id_table = cxgb4_pci_tbl,
5174 .remove = remove_one,
5175 .err_handler = &cxgb4_eeh,
5178 static int __init cxgb4_init_module(void)
5182 workq = create_singlethread_workqueue("cxgb4");
5186 /* Debugfs support is optional, just warn if this fails */
5187 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5188 if (!cxgb4_debugfs_root)
5189 pr_warn("could not create debugfs entry, continuing\n");
5191 ret = pci_register_driver(&cxgb4_driver);
5193 debugfs_remove(cxgb4_debugfs_root);
5197 static void __exit cxgb4_cleanup_module(void)
5199 pci_unregister_driver(&cxgb4_driver);
5200 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
5201 flush_workqueue(workq);
5202 destroy_workqueue(workq);
5205 module_init(cxgb4_init_module);
5206 module_exit(cxgb4_cleanup_module);