2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/bitmap.h>
38 #include <linux/crc32.h>
39 #include <linux/ctype.h>
40 #include <linux/debugfs.h>
41 #include <linux/err.h>
42 #include <linux/etherdevice.h>
43 #include <linux/firmware.h>
45 #include <linux/if_vlan.h>
46 #include <linux/init.h>
47 #include <linux/log2.h>
48 #include <linux/mdio.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/mutex.h>
52 #include <linux/netdevice.h>
53 #include <linux/pci.h>
54 #include <linux/aer.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/sched.h>
57 #include <linux/seq_file.h>
58 #include <linux/sockios.h>
59 #include <linux/vmalloc.h>
60 #include <linux/workqueue.h>
61 #include <net/neighbour.h>
62 #include <net/netevent.h>
63 #include <asm/uaccess.h>
71 #define DRV_VERSION "1.3.0-ko"
72 #define DRV_DESC "Chelsio T4 Network Driver"
75 * Max interrupt hold-off timer value in us. Queues fall back to this value
76 * under extreme memory pressure so it's largish to give the system time to
79 #define MAX_SGE_TIMERVAL 200U
83 * Physical Function provisioning constants.
85 PFRES_NVI = 4, /* # of Virtual Interfaces */
86 PFRES_NETHCTRL = 128, /* # of EQs used for ETH or CTRL Qs */
87 PFRES_NIQFLINT = 128, /* # of ingress Qs/w Free List(s)/intr
89 PFRES_NEQ = 256, /* # of egress queues */
90 PFRES_NIQ = 0, /* # of ingress queues */
91 PFRES_TC = 0, /* PCI-E traffic class */
92 PFRES_NEXACTF = 128, /* # of exact MPS filters */
94 PFRES_R_CAPS = FW_CMD_CAP_PF,
95 PFRES_WX_CAPS = FW_CMD_CAP_PF,
99 * Virtual Function provisioning constants. We need two extra Ingress
100 * Queues with Interrupt capability to serve as the VF's Firmware
101 * Event Queue and Forwarded Interrupt Queue (when using MSI mode) --
102 * neither will have Free Lists associated with them). For each
103 * Ethernet/Control Egress Queue and for each Free List, we need an
106 VFRES_NPORTS = 1, /* # of "ports" per VF */
107 VFRES_NQSETS = 2, /* # of "Queue Sets" per VF */
109 VFRES_NVI = VFRES_NPORTS, /* # of Virtual Interfaces */
110 VFRES_NETHCTRL = VFRES_NQSETS, /* # of EQs used for ETH or CTRL Qs */
111 VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
112 VFRES_NEQ = VFRES_NQSETS*2, /* # of egress queues */
113 VFRES_NIQ = 0, /* # of non-fl/int ingress queues */
114 VFRES_TC = 0, /* PCI-E traffic class */
115 VFRES_NEXACTF = 16, /* # of exact MPS filters */
117 VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
118 VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
123 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
124 * static and likely not to be useful in the long run. We really need to
125 * implement some form of persistent configuration which the firmware
128 static unsigned int pfvfres_pmask(struct adapter *adapter,
129 unsigned int pf, unsigned int vf)
131 unsigned int portn, portvec;
134 * Give PF's access to all of the ports.
137 return FW_PFVF_CMD_PMASK_MASK;
140 * For VFs, we'll assign them access to the ports based purely on the
141 * PF. We assign active ports in order, wrapping around if there are
142 * fewer active ports than PFs: e.g. active port[pf % nports].
143 * Unfortunately the adapter's port_info structs haven't been
144 * initialized yet so we have to compute this.
146 if (adapter->params.nports == 0)
149 portn = pf % adapter->params.nports;
150 portvec = adapter->params.portvec;
153 * Isolate the lowest set bit in the port vector. If we're at
154 * the port number that we want, return that as the pmask.
155 * otherwise mask that bit out of the port vector and
156 * decrement our port number ...
158 unsigned int pmask = portvec ^ (portvec & (portvec-1));
168 MAX_TXQ_ENTRIES = 16384,
169 MAX_CTRL_TXQ_ENTRIES = 1024,
170 MAX_RSPQ_ENTRIES = 16384,
171 MAX_RX_BUFFERS = 16384,
172 MIN_TXQ_ENTRIES = 32,
173 MIN_CTRL_TXQ_ENTRIES = 32,
174 MIN_RSPQ_ENTRIES = 128,
178 /* Host shadow copy of ingress filter entry. This is in host native format
179 * and doesn't match the ordering or bit order, etc. of the hardware of the
180 * firmware command. The use of bit-field structure elements is purely to
181 * remind ourselves of the field size limitations and save memory in the case
182 * where the filter table is large.
184 struct filter_entry {
185 /* Administrative fields for filter.
187 u32 valid:1; /* filter allocated and valid */
188 u32 locked:1; /* filter is administratively locked */
190 u32 pending:1; /* filter action is pending firmware reply */
191 u32 smtidx:8; /* Source MAC Table index for smac */
192 struct l2t_entry *l2t; /* Layer Two Table entry for dmac */
194 /* The filter itself. Most of this is a straight copy of information
195 * provided by the extended ioctl(). Some fields are translated to
196 * internal forms -- for instance the Ingress Queue ID passed in from
197 * the ioctl() is translated into the Absolute Ingress Queue ID.
199 struct ch_filter_specification fs;
202 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
203 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
204 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
206 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
208 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
209 CH_DEVICE(0xa000, 0), /* PE10K */
210 CH_DEVICE(0x4001, -1),
211 CH_DEVICE(0x4002, -1),
212 CH_DEVICE(0x4003, -1),
213 CH_DEVICE(0x4004, -1),
214 CH_DEVICE(0x4005, -1),
215 CH_DEVICE(0x4006, -1),
216 CH_DEVICE(0x4007, -1),
217 CH_DEVICE(0x4008, -1),
218 CH_DEVICE(0x4009, -1),
219 CH_DEVICE(0x400a, -1),
220 CH_DEVICE(0x4401, 4),
221 CH_DEVICE(0x4402, 4),
222 CH_DEVICE(0x4403, 4),
223 CH_DEVICE(0x4404, 4),
224 CH_DEVICE(0x4405, 4),
225 CH_DEVICE(0x4406, 4),
226 CH_DEVICE(0x4407, 4),
227 CH_DEVICE(0x4408, 4),
228 CH_DEVICE(0x4409, 4),
229 CH_DEVICE(0x440a, 4),
230 CH_DEVICE(0x440d, 4),
231 CH_DEVICE(0x440e, 4),
235 #define FW_FNAME "cxgb4/t4fw.bin"
236 #define FW5_FNAME "cxgb4/t5fw.bin"
237 #define FW_CFNAME "cxgb4/t4-config.txt"
238 #define FW5_CFNAME "cxgb4/t5-config.txt"
240 MODULE_DESCRIPTION(DRV_DESC);
241 MODULE_AUTHOR("Chelsio Communications");
242 MODULE_LICENSE("Dual BSD/GPL");
243 MODULE_VERSION(DRV_VERSION);
244 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
245 MODULE_FIRMWARE(FW_FNAME);
246 MODULE_FIRMWARE(FW5_FNAME);
249 * Normally we're willing to become the firmware's Master PF but will be happy
250 * if another PF has already become the Master and initialized the adapter.
251 * Setting "force_init" will cause this driver to forcibly establish itself as
252 * the Master PF and initialize the adapter.
254 static uint force_init;
256 module_param(force_init, uint, 0644);
257 MODULE_PARM_DESC(force_init, "Forcibly become Master PF and initialize adapter");
260 * Normally if the firmware we connect to has Configuration File support, we
261 * use that and only fall back to the old Driver-based initialization if the
262 * Configuration File fails for some reason. If force_old_init is set, then
263 * we'll always use the old Driver-based initialization sequence.
265 static uint force_old_init;
267 module_param(force_old_init, uint, 0644);
268 MODULE_PARM_DESC(force_old_init, "Force old initialization sequence");
270 static int dflt_msg_enable = DFLT_MSG_ENABLE;
272 module_param(dflt_msg_enable, int, 0644);
273 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
276 * The driver uses the best interrupt scheme available on a platform in the
277 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
278 * of these schemes the driver may consider as follows:
280 * msi = 2: choose from among all three options
281 * msi = 1: only consider MSI and INTx interrupts
282 * msi = 0: force INTx interrupts
286 module_param(msi, int, 0644);
287 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
290 * Queue interrupt hold-off timer values. Queues default to the first of these
293 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
295 module_param_array(intr_holdoff, uint, NULL, 0644);
296 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
297 "0..4 in microseconds");
299 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
301 module_param_array(intr_cnt, uint, NULL, 0644);
302 MODULE_PARM_DESC(intr_cnt,
303 "thresholds 1..3 for queue interrupt packet counters");
306 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
307 * offset by 2 bytes in order to have the IP headers line up on 4-byte
308 * boundaries. This is a requirement for many architectures which will throw
309 * a machine check fault if an attempt is made to access one of the 4-byte IP
310 * header fields on a non-4-byte boundary. And it's a major performance issue
311 * even on some architectures which allow it like some implementations of the
312 * x86 ISA. However, some architectures don't mind this and for some very
313 * edge-case performance sensitive applications (like forwarding large volumes
314 * of small packets), setting this DMA offset to 0 will decrease the number of
315 * PCI-E Bus transfers enough to measurably affect performance.
317 static int rx_dma_offset = 2;
321 #ifdef CONFIG_PCI_IOV
322 module_param(vf_acls, bool, 0644);
323 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
325 /* Since T5 has more num of PFs, using NUM_OF_PF_WITH_SRIOV_T5
326 * macro as num_vf array size
328 static unsigned int num_vf[NUM_OF_PF_WITH_SRIOV_T5];
330 module_param_array(num_vf, uint, NULL, 0644);
331 MODULE_PARM_DESC(num_vf,
332 "number of VFs for each of PFs 0-3 for T4 and PFs 0-7 for T5");
336 * The filter TCAM has a fixed portion and a variable portion. The fixed
337 * portion can match on source/destination IP IPv4/IPv6 addresses and TCP/UDP
338 * ports. The variable portion is 36 bits which can include things like Exact
339 * Match MAC Index (9 bits), Ether Type (16 bits), IP Protocol (8 bits),
340 * [Inner] VLAN Tag (17 bits), etc. which, if all were somehow selected, would
341 * far exceed the 36-bit budget for this "compressed" header portion of the
342 * filter. Thus, we have a scarce resource which must be carefully managed.
344 * By default we set this up to mostly match the set of filter matching
345 * capabilities of T3 but with accommodations for some of T4's more
346 * interesting features:
348 * { IP Fragment (1), MPS Match Type (3), IP Protocol (8),
349 * [Inner] VLAN (17), Port (3), FCoE (1) }
352 TP_VLAN_PRI_MAP_DEFAULT = HW_TPL_FR_MT_PR_IV_P_FC,
353 TP_VLAN_PRI_MAP_FIRST = FCOE_SHIFT,
354 TP_VLAN_PRI_MAP_LAST = FRAGMENTATION_SHIFT,
357 static unsigned int tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
359 module_param(tp_vlan_pri_map, uint, 0644);
360 MODULE_PARM_DESC(tp_vlan_pri_map, "global compressed filter configuration");
362 static struct dentry *cxgb4_debugfs_root;
364 static LIST_HEAD(adapter_list);
365 static DEFINE_MUTEX(uld_mutex);
366 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
367 static const char *uld_str[] = { "RDMA", "iSCSI" };
369 static void link_report(struct net_device *dev)
371 if (!netif_carrier_ok(dev))
372 netdev_info(dev, "link down\n");
374 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
376 const char *s = "10Mbps";
377 const struct port_info *p = netdev_priv(dev);
379 switch (p->link_cfg.speed) {
391 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
396 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
398 struct net_device *dev = adapter->port[port_id];
400 /* Skip changes from disabled ports. */
401 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
403 netif_carrier_on(dev);
405 netif_carrier_off(dev);
411 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
413 static const char *mod_str[] = {
414 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
417 const struct net_device *dev = adap->port[port_id];
418 const struct port_info *pi = netdev_priv(dev);
420 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
421 netdev_info(dev, "port module unplugged\n");
422 else if (pi->mod_type < ARRAY_SIZE(mod_str))
423 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
427 * Configure the exact and hash address filters to handle a port's multicast
428 * and secondary unicast MAC addresses.
430 static int set_addr_filters(const struct net_device *dev, bool sleep)
438 const struct netdev_hw_addr *ha;
439 int uc_cnt = netdev_uc_count(dev);
440 int mc_cnt = netdev_mc_count(dev);
441 const struct port_info *pi = netdev_priv(dev);
442 unsigned int mb = pi->adapter->fn;
444 /* first do the secondary unicast addresses */
445 netdev_for_each_uc_addr(ha, dev) {
446 addr[naddr++] = ha->addr;
447 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
448 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
449 naddr, addr, filt_idx, &uhash, sleep);
458 /* next set up the multicast addresses */
459 netdev_for_each_mc_addr(ha, dev) {
460 addr[naddr++] = ha->addr;
461 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
462 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
463 naddr, addr, filt_idx, &mhash, sleep);
472 return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
473 uhash | mhash, sleep);
476 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
477 module_param(dbfifo_int_thresh, int, 0644);
478 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
481 * usecs to sleep while draining the dbfifo
483 static int dbfifo_drain_delay = 1000;
484 module_param(dbfifo_drain_delay, int, 0644);
485 MODULE_PARM_DESC(dbfifo_drain_delay,
486 "usecs to sleep while draining the dbfifo");
489 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
490 * If @mtu is -1 it is left unchanged.
492 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
495 struct port_info *pi = netdev_priv(dev);
497 ret = set_addr_filters(dev, sleep_ok);
499 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
500 (dev->flags & IFF_PROMISC) ? 1 : 0,
501 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
506 static struct workqueue_struct *workq;
509 * link_start - enable a port
510 * @dev: the port to enable
512 * Performs the MAC and PHY actions needed to enable a port.
514 static int link_start(struct net_device *dev)
517 struct port_info *pi = netdev_priv(dev);
518 unsigned int mb = pi->adapter->fn;
521 * We do not set address filters and promiscuity here, the stack does
522 * that step explicitly.
524 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
525 !!(dev->features & NETIF_F_HW_VLAN_RX), true);
527 ret = t4_change_mac(pi->adapter, mb, pi->viid,
528 pi->xact_addr_filt, dev->dev_addr, true,
531 pi->xact_addr_filt = ret;
536 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
539 ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
543 /* Clear a filter and release any of its resources that we own. This also
544 * clears the filter's "pending" status.
546 static void clear_filter(struct adapter *adap, struct filter_entry *f)
548 /* If the new or old filter have loopback rewriteing rules then we'll
549 * need to free any existing Layer Two Table (L2T) entries of the old
550 * filter rule. The firmware will handle freeing up any Source MAC
551 * Table (SMT) entries used for rewriting Source MAC Addresses in
555 cxgb4_l2t_release(f->l2t);
557 /* The zeroing of the filter rule below clears the filter valid,
558 * pending, locked flags, l2t pointer, etc. so it's all we need for
561 memset(f, 0, sizeof(*f));
564 /* Handle a filter write/deletion reply.
566 static void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
568 unsigned int idx = GET_TID(rpl);
569 unsigned int nidx = idx - adap->tids.ftid_base;
571 struct filter_entry *f;
573 if (idx >= adap->tids.ftid_base && nidx <
574 (adap->tids.nftids + adap->tids.nsftids)) {
576 ret = GET_TCB_COOKIE(rpl->cookie);
577 f = &adap->tids.ftid_tab[idx];
579 if (ret == FW_FILTER_WR_FLT_DELETED) {
580 /* Clear the filter when we get confirmation from the
581 * hardware that the filter has been deleted.
583 clear_filter(adap, f);
584 } else if (ret == FW_FILTER_WR_SMT_TBL_FULL) {
585 dev_err(adap->pdev_dev, "filter %u setup failed due to full SMT\n",
587 clear_filter(adap, f);
588 } else if (ret == FW_FILTER_WR_FLT_ADDED) {
589 f->smtidx = (be64_to_cpu(rpl->oldval) >> 24) & 0xff;
590 f->pending = 0; /* asynchronous setup completed */
593 /* Something went wrong. Issue a warning about the
594 * problem and clear everything out.
596 dev_err(adap->pdev_dev, "filter %u setup failed with error %u\n",
598 clear_filter(adap, f);
603 /* Response queue handler for the FW event queue.
605 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
606 const struct pkt_gl *gl)
608 u8 opcode = ((const struct rss_header *)rsp)->opcode;
610 rsp++; /* skip RSS header */
611 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
612 const struct cpl_sge_egr_update *p = (void *)rsp;
613 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
616 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
618 if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
619 struct sge_eth_txq *eq;
621 eq = container_of(txq, struct sge_eth_txq, q);
622 netif_tx_wake_queue(eq->txq);
624 struct sge_ofld_txq *oq;
626 oq = container_of(txq, struct sge_ofld_txq, q);
627 tasklet_schedule(&oq->qresume_tsk);
629 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
630 const struct cpl_fw6_msg *p = (void *)rsp;
633 t4_handle_fw_rpl(q->adap, p->data);
634 } else if (opcode == CPL_L2T_WRITE_RPL) {
635 const struct cpl_l2t_write_rpl *p = (void *)rsp;
637 do_l2t_write_rpl(q->adap, p);
638 } else if (opcode == CPL_SET_TCB_RPL) {
639 const struct cpl_set_tcb_rpl *p = (void *)rsp;
641 filter_rpl(q->adap, p);
643 dev_err(q->adap->pdev_dev,
644 "unexpected CPL %#x on FW event queue\n", opcode);
649 * uldrx_handler - response queue handler for ULD queues
650 * @q: the response queue that received the packet
651 * @rsp: the response queue descriptor holding the offload message
652 * @gl: the gather list of packet fragments
654 * Deliver an ingress offload packet to a ULD. All processing is done by
655 * the ULD, we just maintain statistics.
657 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
658 const struct pkt_gl *gl)
660 struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
662 if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
668 else if (gl == CXGB4_MSG_AN)
675 static void disable_msi(struct adapter *adapter)
677 if (adapter->flags & USING_MSIX) {
678 pci_disable_msix(adapter->pdev);
679 adapter->flags &= ~USING_MSIX;
680 } else if (adapter->flags & USING_MSI) {
681 pci_disable_msi(adapter->pdev);
682 adapter->flags &= ~USING_MSI;
687 * Interrupt handler for non-data events used with MSI-X.
689 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
691 struct adapter *adap = cookie;
693 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
696 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
698 t4_slow_intr_handler(adap);
703 * Name the MSI-X interrupts.
705 static void name_msix_vecs(struct adapter *adap)
707 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
709 /* non-data interrupts */
710 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
713 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
714 adap->port[0]->name);
716 /* Ethernet queues */
717 for_each_port(adap, j) {
718 struct net_device *d = adap->port[j];
719 const struct port_info *pi = netdev_priv(d);
721 for (i = 0; i < pi->nqsets; i++, msi_idx++)
722 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
727 for_each_ofldrxq(&adap->sge, i)
728 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
729 adap->port[0]->name, i);
731 for_each_rdmarxq(&adap->sge, i)
732 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
733 adap->port[0]->name, i);
736 static int request_msix_queue_irqs(struct adapter *adap)
738 struct sge *s = &adap->sge;
739 int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi_index = 2;
741 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
742 adap->msix_info[1].desc, &s->fw_evtq);
746 for_each_ethrxq(s, ethqidx) {
747 err = request_irq(adap->msix_info[msi_index].vec,
749 adap->msix_info[msi_index].desc,
750 &s->ethrxq[ethqidx].rspq);
755 for_each_ofldrxq(s, ofldqidx) {
756 err = request_irq(adap->msix_info[msi_index].vec,
758 adap->msix_info[msi_index].desc,
759 &s->ofldrxq[ofldqidx].rspq);
764 for_each_rdmarxq(s, rdmaqidx) {
765 err = request_irq(adap->msix_info[msi_index].vec,
767 adap->msix_info[msi_index].desc,
768 &s->rdmarxq[rdmaqidx].rspq);
776 while (--rdmaqidx >= 0)
777 free_irq(adap->msix_info[--msi_index].vec,
778 &s->rdmarxq[rdmaqidx].rspq);
779 while (--ofldqidx >= 0)
780 free_irq(adap->msix_info[--msi_index].vec,
781 &s->ofldrxq[ofldqidx].rspq);
782 while (--ethqidx >= 0)
783 free_irq(adap->msix_info[--msi_index].vec,
784 &s->ethrxq[ethqidx].rspq);
785 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
789 static void free_msix_queue_irqs(struct adapter *adap)
791 int i, msi_index = 2;
792 struct sge *s = &adap->sge;
794 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
795 for_each_ethrxq(s, i)
796 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
797 for_each_ofldrxq(s, i)
798 free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
799 for_each_rdmarxq(s, i)
800 free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
804 * write_rss - write the RSS table for a given port
806 * @queues: array of queue indices for RSS
808 * Sets up the portion of the HW RSS table for the port's VI to distribute
809 * packets to the Rx queues in @queues.
811 static int write_rss(const struct port_info *pi, const u16 *queues)
815 const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
817 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
821 /* map the queue indices to queue ids */
822 for (i = 0; i < pi->rss_size; i++, queues++)
823 rss[i] = q[*queues].rspq.abs_id;
825 err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
826 pi->rss_size, rss, pi->rss_size);
832 * setup_rss - configure RSS
835 * Sets up RSS for each port.
837 static int setup_rss(struct adapter *adap)
841 for_each_port(adap, i) {
842 const struct port_info *pi = adap2pinfo(adap, i);
844 err = write_rss(pi, pi->rss);
852 * Return the channel of the ingress queue with the given qid.
854 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
856 qid -= p->ingr_start;
857 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
861 * Wait until all NAPI handlers are descheduled.
863 static void quiesce_rx(struct adapter *adap)
867 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
868 struct sge_rspq *q = adap->sge.ingr_map[i];
871 napi_disable(&q->napi);
876 * Enable NAPI scheduling and interrupt generation for all Rx queues.
878 static void enable_rx(struct adapter *adap)
882 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
883 struct sge_rspq *q = adap->sge.ingr_map[i];
888 napi_enable(&q->napi);
889 /* 0-increment GTS to start the timer and enable interrupts */
890 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
891 SEINTARM(q->intr_params) |
892 INGRESSQID(q->cntxt_id));
897 * setup_sge_queues - configure SGE Tx/Rx/response queues
900 * Determines how many sets of SGE queues to use and initializes them.
901 * We support multiple queue sets per port if we have MSI-X, otherwise
902 * just one queue set per port.
904 static int setup_sge_queues(struct adapter *adap)
906 int err, msi_idx, i, j;
907 struct sge *s = &adap->sge;
909 bitmap_zero(s->starving_fl, MAX_EGRQ);
910 bitmap_zero(s->txq_maperr, MAX_EGRQ);
912 if (adap->flags & USING_MSIX)
913 msi_idx = 1; /* vector 0 is for non-queue interrupts */
915 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
919 msi_idx = -((int)s->intrq.abs_id + 1);
922 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
923 msi_idx, NULL, fwevtq_handler);
925 freeout: t4_free_sge_resources(adap);
929 for_each_port(adap, i) {
930 struct net_device *dev = adap->port[i];
931 struct port_info *pi = netdev_priv(dev);
932 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
933 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
935 for (j = 0; j < pi->nqsets; j++, q++) {
938 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
944 memset(&q->stats, 0, sizeof(q->stats));
946 for (j = 0; j < pi->nqsets; j++, t++) {
947 err = t4_sge_alloc_eth_txq(adap, t, dev,
948 netdev_get_tx_queue(dev, j),
949 s->fw_evtq.cntxt_id);
955 j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
956 for_each_ofldrxq(s, i) {
957 struct sge_ofld_rxq *q = &s->ofldrxq[i];
958 struct net_device *dev = adap->port[i / j];
962 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
963 &q->fl, uldrx_handler);
966 memset(&q->stats, 0, sizeof(q->stats));
967 s->ofld_rxq[i] = q->rspq.abs_id;
968 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
969 s->fw_evtq.cntxt_id);
974 for_each_rdmarxq(s, i) {
975 struct sge_ofld_rxq *q = &s->rdmarxq[i];
979 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
980 msi_idx, &q->fl, uldrx_handler);
983 memset(&q->stats, 0, sizeof(q->stats));
984 s->rdma_rxq[i] = q->rspq.abs_id;
987 for_each_port(adap, i) {
989 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
990 * have RDMA queues, and that's the right value.
992 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
994 s->rdmarxq[i].rspq.cntxt_id);
999 t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
1000 RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
1001 QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
1006 * Returns 0 if new FW was successfully loaded, a positive errno if a load was
1007 * started but failed, and a negative errno if flash load couldn't start.
1009 static int upgrade_fw(struct adapter *adap)
1012 u32 vers, exp_major;
1013 const struct fw_hdr *hdr;
1014 const struct firmware *fw;
1015 struct device *dev = adap->pdev_dev;
1018 switch (CHELSIO_CHIP_VERSION(adap->chip)) {
1020 fw_file_name = FW_FNAME;
1021 exp_major = FW_VERSION_MAJOR;
1024 fw_file_name = FW5_FNAME;
1025 exp_major = FW_VERSION_MAJOR_T5;
1028 dev_err(dev, "Unsupported chip type, %x\n", adap->chip);
1032 ret = request_firmware(&fw, fw_file_name, dev);
1034 dev_err(dev, "unable to load firmware image %s, error %d\n",
1039 hdr = (const struct fw_hdr *)fw->data;
1040 vers = ntohl(hdr->fw_ver);
1041 if (FW_HDR_FW_VER_MAJOR_GET(vers) != exp_major) {
1042 ret = -EINVAL; /* wrong major version, won't do */
1047 * If the flash FW is unusable or we found something newer, load it.
1049 if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != exp_major ||
1050 vers > adap->params.fw_vers) {
1051 dev_info(dev, "upgrading firmware ...\n");
1052 ret = t4_fw_upgrade(adap, adap->mbox, fw->data, fw->size,
1056 "firmware upgraded to version %pI4 from %s\n",
1057 &hdr->fw_ver, fw_file_name);
1059 dev_err(dev, "firmware upgrade failed! err=%d\n", -ret);
1062 * Tell our caller that we didn't upgrade the firmware.
1067 out: release_firmware(fw);
1072 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1073 * The allocated memory is cleared.
1075 void *t4_alloc_mem(size_t size)
1077 void *p = kzalloc(size, GFP_KERNEL);
1085 * Free memory allocated through alloc_mem().
1087 static void t4_free_mem(void *addr)
1089 if (is_vmalloc_addr(addr))
1095 /* Send a Work Request to write the filter at a specified index. We construct
1096 * a Firmware Filter Work Request to have the work done and put the indicated
1097 * filter into "pending" mode which will prevent any further actions against
1098 * it till we get a reply from the firmware on the completion status of the
1101 static int set_filter_wr(struct adapter *adapter, int fidx)
1103 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1104 struct sk_buff *skb;
1105 struct fw_filter_wr *fwr;
1108 /* If the new filter requires loopback Destination MAC and/or VLAN
1109 * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
1112 if (f->fs.newdmac || f->fs.newvlan) {
1113 /* allocate L2T entry for new filter */
1114 f->l2t = t4_l2t_alloc_switching(adapter->l2t);
1117 if (t4_l2t_set_switching(adapter, f->l2t, f->fs.vlan,
1118 f->fs.eport, f->fs.dmac)) {
1119 cxgb4_l2t_release(f->l2t);
1125 ftid = adapter->tids.ftid_base + fidx;
1127 skb = alloc_skb(sizeof(*fwr), GFP_KERNEL | __GFP_NOFAIL);
1128 fwr = (struct fw_filter_wr *)__skb_put(skb, sizeof(*fwr));
1129 memset(fwr, 0, sizeof(*fwr));
1131 /* It would be nice to put most of the following in t4_hw.c but most
1132 * of the work is translating the cxgbtool ch_filter_specification
1133 * into the Work Request and the definition of that structure is
1134 * currently in cxgbtool.h which isn't appropriate to pull into the
1135 * common code. We may eventually try to come up with a more neutral
1136 * filter specification structure but for now it's easiest to simply
1137 * put this fairly direct code in line ...
1139 fwr->op_pkd = htonl(FW_WR_OP(FW_FILTER_WR));
1140 fwr->len16_pkd = htonl(FW_WR_LEN16(sizeof(*fwr)/16));
1142 htonl(V_FW_FILTER_WR_TID(ftid) |
1143 V_FW_FILTER_WR_RQTYPE(f->fs.type) |
1144 V_FW_FILTER_WR_NOREPLY(0) |
1145 V_FW_FILTER_WR_IQ(f->fs.iq));
1146 fwr->del_filter_to_l2tix =
1147 htonl(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
1148 V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
1149 V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
1150 V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
1151 V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
1152 V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
1153 V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
1154 V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
1155 V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
1156 f->fs.newvlan == VLAN_REWRITE) |
1157 V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
1158 f->fs.newvlan == VLAN_REWRITE) |
1159 V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
1160 V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
1161 V_FW_FILTER_WR_PRIO(f->fs.prio) |
1162 V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
1163 fwr->ethtype = htons(f->fs.val.ethtype);
1164 fwr->ethtypem = htons(f->fs.mask.ethtype);
1165 fwr->frag_to_ovlan_vldm =
1166 (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
1167 V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
1168 V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.ivlan_vld) |
1169 V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.ovlan_vld) |
1170 V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.ivlan_vld) |
1171 V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.ovlan_vld));
1173 fwr->rx_chan_rx_rpl_iq =
1174 htons(V_FW_FILTER_WR_RX_CHAN(0) |
1175 V_FW_FILTER_WR_RX_RPL_IQ(adapter->sge.fw_evtq.abs_id));
1176 fwr->maci_to_matchtypem =
1177 htonl(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
1178 V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
1179 V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
1180 V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
1181 V_FW_FILTER_WR_PORT(f->fs.val.iport) |
1182 V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
1183 V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
1184 V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
1185 fwr->ptcl = f->fs.val.proto;
1186 fwr->ptclm = f->fs.mask.proto;
1187 fwr->ttyp = f->fs.val.tos;
1188 fwr->ttypm = f->fs.mask.tos;
1189 fwr->ivlan = htons(f->fs.val.ivlan);
1190 fwr->ivlanm = htons(f->fs.mask.ivlan);
1191 fwr->ovlan = htons(f->fs.val.ovlan);
1192 fwr->ovlanm = htons(f->fs.mask.ovlan);
1193 memcpy(fwr->lip, f->fs.val.lip, sizeof(fwr->lip));
1194 memcpy(fwr->lipm, f->fs.mask.lip, sizeof(fwr->lipm));
1195 memcpy(fwr->fip, f->fs.val.fip, sizeof(fwr->fip));
1196 memcpy(fwr->fipm, f->fs.mask.fip, sizeof(fwr->fipm));
1197 fwr->lp = htons(f->fs.val.lport);
1198 fwr->lpm = htons(f->fs.mask.lport);
1199 fwr->fp = htons(f->fs.val.fport);
1200 fwr->fpm = htons(f->fs.mask.fport);
1202 memcpy(fwr->sma, f->fs.smac, sizeof(fwr->sma));
1204 /* Mark the filter as "pending" and ship off the Filter Work Request.
1205 * When we get the Work Request Reply we'll clear the pending status.
1208 set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
1209 t4_ofld_send(adapter, skb);
1213 /* Delete the filter at a specified index.
1215 static int del_filter_wr(struct adapter *adapter, int fidx)
1217 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1218 struct sk_buff *skb;
1219 struct fw_filter_wr *fwr;
1220 unsigned int len, ftid;
1223 ftid = adapter->tids.ftid_base + fidx;
1225 skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
1226 fwr = (struct fw_filter_wr *)__skb_put(skb, len);
1227 t4_mk_filtdelwr(ftid, fwr, adapter->sge.fw_evtq.abs_id);
1229 /* Mark the filter as "pending" and ship off the Filter Work Request.
1230 * When we get the Work Request Reply we'll clear the pending status.
1233 t4_mgmt_tx(adapter, skb);
1237 static inline int is_offload(const struct adapter *adap)
1239 return adap->params.offload;
1243 * Implementation of ethtool operations.
1246 static u32 get_msglevel(struct net_device *dev)
1248 return netdev2adap(dev)->msg_enable;
1251 static void set_msglevel(struct net_device *dev, u32 val)
1253 netdev2adap(dev)->msg_enable = val;
1256 static char stats_strings[][ETH_GSTRING_LEN] = {
1259 "TxBroadcastFrames ",
1260 "TxMulticastFrames ",
1266 "TxFrames128To255 ",
1267 "TxFrames256To511 ",
1268 "TxFrames512To1023 ",
1269 "TxFrames1024To1518 ",
1270 "TxFrames1519ToMax ",
1285 "RxBroadcastFrames ",
1286 "RxMulticastFrames ",
1298 "RxFrames128To255 ",
1299 "RxFrames256To511 ",
1300 "RxFrames512To1023 ",
1301 "RxFrames1024To1518 ",
1302 "RxFrames1519ToMax ",
1314 "RxBG0FramesDropped ",
1315 "RxBG1FramesDropped ",
1316 "RxBG2FramesDropped ",
1317 "RxBG3FramesDropped ",
1318 "RxBG0FramesTrunc ",
1319 "RxBG1FramesTrunc ",
1320 "RxBG2FramesTrunc ",
1321 "RxBG3FramesTrunc ",
1330 "WriteCoalSuccess ",
1334 static int get_sset_count(struct net_device *dev, int sset)
1338 return ARRAY_SIZE(stats_strings);
1344 #define T4_REGMAP_SIZE (160 * 1024)
1345 #define T5_REGMAP_SIZE (332 * 1024)
1347 static int get_regs_len(struct net_device *dev)
1349 struct adapter *adap = netdev2adap(dev);
1350 if (is_t4(adap->chip))
1351 return T4_REGMAP_SIZE;
1353 return T5_REGMAP_SIZE;
1356 static int get_eeprom_len(struct net_device *dev)
1361 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1363 struct adapter *adapter = netdev2adap(dev);
1365 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1366 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1367 strlcpy(info->bus_info, pci_name(adapter->pdev),
1368 sizeof(info->bus_info));
1370 if (adapter->params.fw_vers)
1371 snprintf(info->fw_version, sizeof(info->fw_version),
1372 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1373 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1374 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1375 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1376 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1377 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1378 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1379 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1380 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1383 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1385 if (stringset == ETH_SS_STATS)
1386 memcpy(data, stats_strings, sizeof(stats_strings));
1390 * port stats maintained per queue of the port. They should be in the same
1391 * order as in stats_strings above.
1393 struct queue_port_stats {
1403 static void collect_sge_port_stats(const struct adapter *adap,
1404 const struct port_info *p, struct queue_port_stats *s)
1407 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1408 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1410 memset(s, 0, sizeof(*s));
1411 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1413 s->tx_csum += tx->tx_cso;
1414 s->rx_csum += rx->stats.rx_cso;
1415 s->vlan_ex += rx->stats.vlan_ex;
1416 s->vlan_ins += tx->vlan_ins;
1417 s->gro_pkts += rx->stats.lro_pkts;
1418 s->gro_merged += rx->stats.lro_merged;
1422 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1425 struct port_info *pi = netdev_priv(dev);
1426 struct adapter *adapter = pi->adapter;
1429 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1431 data += sizeof(struct port_stats) / sizeof(u64);
1432 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1433 data += sizeof(struct queue_port_stats) / sizeof(u64);
1434 if (!is_t4(adapter->chip)) {
1435 t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7));
1436 val1 = t4_read_reg(adapter, SGE_STAT_TOTAL);
1437 val2 = t4_read_reg(adapter, SGE_STAT_MATCH);
1438 *data = val1 - val2;
1443 memset(data, 0, 2 * sizeof(u64));
1449 * Return a version number to identify the type of adapter. The scheme is:
1450 * - bits 0..9: chip version
1451 * - bits 10..15: chip revision
1452 * - bits 16..23: register dump version
1454 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1456 return CHELSIO_CHIP_VERSION(ap->chip) |
1457 (CHELSIO_CHIP_RELEASE(ap->chip) << 10) | (1 << 16);
1460 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1463 u32 *p = buf + start;
1465 for ( ; start <= end; start += sizeof(u32))
1466 *p++ = t4_read_reg(ap, start);
1469 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1472 static const unsigned int t4_reg_ranges[] = {
1692 static const unsigned int t5_reg_ranges[] = {
2120 struct adapter *ap = netdev2adap(dev);
2121 static const unsigned int *reg_ranges;
2122 int arr_size = 0, buf_size = 0;
2124 if (is_t4(ap->chip)) {
2125 reg_ranges = &t4_reg_ranges[0];
2126 arr_size = ARRAY_SIZE(t4_reg_ranges);
2127 buf_size = T4_REGMAP_SIZE;
2129 reg_ranges = &t5_reg_ranges[0];
2130 arr_size = ARRAY_SIZE(t5_reg_ranges);
2131 buf_size = T5_REGMAP_SIZE;
2134 regs->version = mk_adap_vers(ap);
2136 memset(buf, 0, buf_size);
2137 for (i = 0; i < arr_size; i += 2)
2138 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
2141 static int restart_autoneg(struct net_device *dev)
2143 struct port_info *p = netdev_priv(dev);
2145 if (!netif_running(dev))
2147 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
2149 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
2153 static int identify_port(struct net_device *dev,
2154 enum ethtool_phys_id_state state)
2157 struct adapter *adap = netdev2adap(dev);
2159 if (state == ETHTOOL_ID_ACTIVE)
2161 else if (state == ETHTOOL_ID_INACTIVE)
2166 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
2169 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
2173 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
2174 type == FW_PORT_TYPE_BT_XAUI) {
2176 if (caps & FW_PORT_CAP_SPEED_100M)
2177 v |= SUPPORTED_100baseT_Full;
2178 if (caps & FW_PORT_CAP_SPEED_1G)
2179 v |= SUPPORTED_1000baseT_Full;
2180 if (caps & FW_PORT_CAP_SPEED_10G)
2181 v |= SUPPORTED_10000baseT_Full;
2182 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
2183 v |= SUPPORTED_Backplane;
2184 if (caps & FW_PORT_CAP_SPEED_1G)
2185 v |= SUPPORTED_1000baseKX_Full;
2186 if (caps & FW_PORT_CAP_SPEED_10G)
2187 v |= SUPPORTED_10000baseKX4_Full;
2188 } else if (type == FW_PORT_TYPE_KR)
2189 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
2190 else if (type == FW_PORT_TYPE_BP_AP)
2191 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2192 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
2193 else if (type == FW_PORT_TYPE_BP4_AP)
2194 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2195 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
2196 SUPPORTED_10000baseKX4_Full;
2197 else if (type == FW_PORT_TYPE_FIBER_XFI ||
2198 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
2199 v |= SUPPORTED_FIBRE;
2201 if (caps & FW_PORT_CAP_ANEG)
2202 v |= SUPPORTED_Autoneg;
2206 static unsigned int to_fw_linkcaps(unsigned int caps)
2210 if (caps & ADVERTISED_100baseT_Full)
2211 v |= FW_PORT_CAP_SPEED_100M;
2212 if (caps & ADVERTISED_1000baseT_Full)
2213 v |= FW_PORT_CAP_SPEED_1G;
2214 if (caps & ADVERTISED_10000baseT_Full)
2215 v |= FW_PORT_CAP_SPEED_10G;
2219 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2221 const struct port_info *p = netdev_priv(dev);
2223 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
2224 p->port_type == FW_PORT_TYPE_BT_XFI ||
2225 p->port_type == FW_PORT_TYPE_BT_XAUI)
2226 cmd->port = PORT_TP;
2227 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
2228 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
2229 cmd->port = PORT_FIBRE;
2230 else if (p->port_type == FW_PORT_TYPE_SFP) {
2231 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
2232 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
2233 cmd->port = PORT_DA;
2235 cmd->port = PORT_FIBRE;
2237 cmd->port = PORT_OTHER;
2239 if (p->mdio_addr >= 0) {
2240 cmd->phy_address = p->mdio_addr;
2241 cmd->transceiver = XCVR_EXTERNAL;
2242 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
2243 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
2245 cmd->phy_address = 0; /* not really, but no better option */
2246 cmd->transceiver = XCVR_INTERNAL;
2247 cmd->mdio_support = 0;
2250 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
2251 cmd->advertising = from_fw_linkcaps(p->port_type,
2252 p->link_cfg.advertising);
2253 ethtool_cmd_speed_set(cmd,
2254 netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
2255 cmd->duplex = DUPLEX_FULL;
2256 cmd->autoneg = p->link_cfg.autoneg;
2262 static unsigned int speed_to_caps(int speed)
2264 if (speed == SPEED_100)
2265 return FW_PORT_CAP_SPEED_100M;
2266 if (speed == SPEED_1000)
2267 return FW_PORT_CAP_SPEED_1G;
2268 if (speed == SPEED_10000)
2269 return FW_PORT_CAP_SPEED_10G;
2273 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2276 struct port_info *p = netdev_priv(dev);
2277 struct link_config *lc = &p->link_cfg;
2278 u32 speed = ethtool_cmd_speed(cmd);
2280 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
2283 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
2285 * PHY offers a single speed. See if that's what's
2288 if (cmd->autoneg == AUTONEG_DISABLE &&
2289 (lc->supported & speed_to_caps(speed)))
2294 if (cmd->autoneg == AUTONEG_DISABLE) {
2295 cap = speed_to_caps(speed);
2297 if (!(lc->supported & cap) || (speed == SPEED_1000) ||
2298 (speed == SPEED_10000))
2300 lc->requested_speed = cap;
2301 lc->advertising = 0;
2303 cap = to_fw_linkcaps(cmd->advertising);
2304 if (!(lc->supported & cap))
2306 lc->requested_speed = 0;
2307 lc->advertising = cap | FW_PORT_CAP_ANEG;
2309 lc->autoneg = cmd->autoneg;
2311 if (netif_running(dev))
2312 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2317 static void get_pauseparam(struct net_device *dev,
2318 struct ethtool_pauseparam *epause)
2320 struct port_info *p = netdev_priv(dev);
2322 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
2323 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
2324 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
2327 static int set_pauseparam(struct net_device *dev,
2328 struct ethtool_pauseparam *epause)
2330 struct port_info *p = netdev_priv(dev);
2331 struct link_config *lc = &p->link_cfg;
2333 if (epause->autoneg == AUTONEG_DISABLE)
2334 lc->requested_fc = 0;
2335 else if (lc->supported & FW_PORT_CAP_ANEG)
2336 lc->requested_fc = PAUSE_AUTONEG;
2340 if (epause->rx_pause)
2341 lc->requested_fc |= PAUSE_RX;
2342 if (epause->tx_pause)
2343 lc->requested_fc |= PAUSE_TX;
2344 if (netif_running(dev))
2345 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2350 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2352 const struct port_info *pi = netdev_priv(dev);
2353 const struct sge *s = &pi->adapter->sge;
2355 e->rx_max_pending = MAX_RX_BUFFERS;
2356 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
2357 e->rx_jumbo_max_pending = 0;
2358 e->tx_max_pending = MAX_TXQ_ENTRIES;
2360 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
2361 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
2362 e->rx_jumbo_pending = 0;
2363 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
2366 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2369 const struct port_info *pi = netdev_priv(dev);
2370 struct adapter *adapter = pi->adapter;
2371 struct sge *s = &adapter->sge;
2373 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
2374 e->tx_pending > MAX_TXQ_ENTRIES ||
2375 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
2376 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
2377 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
2380 if (adapter->flags & FULL_INIT_DONE)
2383 for (i = 0; i < pi->nqsets; ++i) {
2384 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
2385 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
2386 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
2391 static int closest_timer(const struct sge *s, int time)
2393 int i, delta, match = 0, min_delta = INT_MAX;
2395 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
2396 delta = time - s->timer_val[i];
2399 if (delta < min_delta) {
2407 static int closest_thres(const struct sge *s, int thres)
2409 int i, delta, match = 0, min_delta = INT_MAX;
2411 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
2412 delta = thres - s->counter_val[i];
2415 if (delta < min_delta) {
2424 * Return a queue's interrupt hold-off time in us. 0 means no timer.
2426 static unsigned int qtimer_val(const struct adapter *adap,
2427 const struct sge_rspq *q)
2429 unsigned int idx = q->intr_params >> 1;
2431 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
2435 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
2436 * @adap: the adapter
2438 * @us: the hold-off time in us, or 0 to disable timer
2439 * @cnt: the hold-off packet count, or 0 to disable counter
2441 * Sets an Rx queue's interrupt hold-off time and packet count. At least
2442 * one of the two needs to be enabled for the queue to generate interrupts.
2444 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
2445 unsigned int us, unsigned int cnt)
2447 if ((us | cnt) == 0)
2454 new_idx = closest_thres(&adap->sge, cnt);
2455 if (q->desc && q->pktcnt_idx != new_idx) {
2456 /* the queue has already been created, update it */
2457 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
2458 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
2459 FW_PARAMS_PARAM_YZ(q->cntxt_id);
2460 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
2465 q->pktcnt_idx = new_idx;
2468 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
2469 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
2473 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2475 const struct port_info *pi = netdev_priv(dev);
2476 struct adapter *adap = pi->adapter;
2481 for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++) {
2482 q = &adap->sge.ethrxq[i].rspq;
2483 r = set_rxq_intr_params(adap, q, c->rx_coalesce_usecs,
2484 c->rx_max_coalesced_frames);
2486 dev_err(&dev->dev, "failed to set coalesce %d\n", r);
2493 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2495 const struct port_info *pi = netdev_priv(dev);
2496 const struct adapter *adap = pi->adapter;
2497 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
2499 c->rx_coalesce_usecs = qtimer_val(adap, rq);
2500 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
2501 adap->sge.counter_val[rq->pktcnt_idx] : 0;
2506 * eeprom_ptov - translate a physical EEPROM address to virtual
2507 * @phys_addr: the physical EEPROM address
2508 * @fn: the PCI function number
2509 * @sz: size of function-specific area
2511 * Translate a physical EEPROM address to virtual. The first 1K is
2512 * accessed through virtual addresses starting at 31K, the rest is
2513 * accessed through virtual addresses starting at 0.
2515 * The mapping is as follows:
2516 * [0..1K) -> [31K..32K)
2517 * [1K..1K+A) -> [31K-A..31K)
2518 * [1K+A..ES) -> [0..ES-A-1K)
2520 * where A = @fn * @sz, and ES = EEPROM size.
2522 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
2525 if (phys_addr < 1024)
2526 return phys_addr + (31 << 10);
2527 if (phys_addr < 1024 + fn)
2528 return 31744 - fn + phys_addr - 1024;
2529 if (phys_addr < EEPROMSIZE)
2530 return phys_addr - 1024 - fn;
2535 * The next two routines implement eeprom read/write from physical addresses.
2537 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
2539 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2542 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
2543 return vaddr < 0 ? vaddr : 0;
2546 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
2548 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2551 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
2552 return vaddr < 0 ? vaddr : 0;
2555 #define EEPROM_MAGIC 0x38E2F10C
2557 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2561 struct adapter *adapter = netdev2adap(dev);
2563 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2567 e->magic = EEPROM_MAGIC;
2568 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2569 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
2572 memcpy(data, buf + e->offset, e->len);
2577 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2582 u32 aligned_offset, aligned_len, *p;
2583 struct adapter *adapter = netdev2adap(dev);
2585 if (eeprom->magic != EEPROM_MAGIC)
2588 aligned_offset = eeprom->offset & ~3;
2589 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2591 if (adapter->fn > 0) {
2592 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
2594 if (aligned_offset < start ||
2595 aligned_offset + aligned_len > start + EEPROMPFSIZE)
2599 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2601 * RMW possibly needed for first or last words.
2603 buf = kmalloc(aligned_len, GFP_KERNEL);
2606 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
2607 if (!err && aligned_len > 4)
2608 err = eeprom_rd_phys(adapter,
2609 aligned_offset + aligned_len - 4,
2610 (u32 *)&buf[aligned_len - 4]);
2613 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2617 err = t4_seeprom_wp(adapter, false);
2621 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2622 err = eeprom_wr_phys(adapter, aligned_offset, *p);
2623 aligned_offset += 4;
2627 err = t4_seeprom_wp(adapter, true);
2634 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
2637 const struct firmware *fw;
2638 struct adapter *adap = netdev2adap(netdev);
2640 ef->data[sizeof(ef->data) - 1] = '\0';
2641 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
2645 ret = t4_load_fw(adap, fw->data, fw->size);
2646 release_firmware(fw);
2648 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
2652 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2653 #define BCAST_CRC 0xa0ccc1a6
2655 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2657 wol->supported = WAKE_BCAST | WAKE_MAGIC;
2658 wol->wolopts = netdev2adap(dev)->wol;
2659 memset(&wol->sopass, 0, sizeof(wol->sopass));
2662 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2665 struct port_info *pi = netdev_priv(dev);
2667 if (wol->wolopts & ~WOL_SUPPORTED)
2669 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
2670 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
2671 if (wol->wolopts & WAKE_BCAST) {
2672 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
2675 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
2676 ~6ULL, ~0ULL, BCAST_CRC, true);
2678 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
2682 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2684 const struct port_info *pi = netdev_priv(dev);
2685 netdev_features_t changed = dev->features ^ features;
2688 if (!(changed & NETIF_F_HW_VLAN_RX))
2691 err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
2693 !!(features & NETIF_F_HW_VLAN_RX), true);
2695 dev->features = features ^ NETIF_F_HW_VLAN_RX;
2699 static u32 get_rss_table_size(struct net_device *dev)
2701 const struct port_info *pi = netdev_priv(dev);
2703 return pi->rss_size;
2706 static int get_rss_table(struct net_device *dev, u32 *p)
2708 const struct port_info *pi = netdev_priv(dev);
2709 unsigned int n = pi->rss_size;
2716 static int set_rss_table(struct net_device *dev, const u32 *p)
2719 struct port_info *pi = netdev_priv(dev);
2721 for (i = 0; i < pi->rss_size; i++)
2723 if (pi->adapter->flags & FULL_INIT_DONE)
2724 return write_rss(pi, pi->rss);
2728 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
2731 const struct port_info *pi = netdev_priv(dev);
2733 switch (info->cmd) {
2734 case ETHTOOL_GRXFH: {
2735 unsigned int v = pi->rss_mode;
2738 switch (info->flow_type) {
2740 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
2741 info->data = RXH_IP_SRC | RXH_IP_DST |
2742 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2743 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2744 info->data = RXH_IP_SRC | RXH_IP_DST;
2747 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
2748 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2749 info->data = RXH_IP_SRC | RXH_IP_DST |
2750 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2751 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2752 info->data = RXH_IP_SRC | RXH_IP_DST;
2755 case AH_ESP_V4_FLOW:
2757 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2758 info->data = RXH_IP_SRC | RXH_IP_DST;
2761 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
2762 info->data = RXH_IP_SRC | RXH_IP_DST |
2763 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2764 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2765 info->data = RXH_IP_SRC | RXH_IP_DST;
2768 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
2769 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2770 info->data = RXH_IP_SRC | RXH_IP_DST |
2771 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2772 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2773 info->data = RXH_IP_SRC | RXH_IP_DST;
2776 case AH_ESP_V6_FLOW:
2778 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2779 info->data = RXH_IP_SRC | RXH_IP_DST;
2784 case ETHTOOL_GRXRINGS:
2785 info->data = pi->nqsets;
2791 static const struct ethtool_ops cxgb_ethtool_ops = {
2792 .get_settings = get_settings,
2793 .set_settings = set_settings,
2794 .get_drvinfo = get_drvinfo,
2795 .get_msglevel = get_msglevel,
2796 .set_msglevel = set_msglevel,
2797 .get_ringparam = get_sge_param,
2798 .set_ringparam = set_sge_param,
2799 .get_coalesce = get_coalesce,
2800 .set_coalesce = set_coalesce,
2801 .get_eeprom_len = get_eeprom_len,
2802 .get_eeprom = get_eeprom,
2803 .set_eeprom = set_eeprom,
2804 .get_pauseparam = get_pauseparam,
2805 .set_pauseparam = set_pauseparam,
2806 .get_link = ethtool_op_get_link,
2807 .get_strings = get_strings,
2808 .set_phys_id = identify_port,
2809 .nway_reset = restart_autoneg,
2810 .get_sset_count = get_sset_count,
2811 .get_ethtool_stats = get_stats,
2812 .get_regs_len = get_regs_len,
2813 .get_regs = get_regs,
2816 .get_rxnfc = get_rxnfc,
2817 .get_rxfh_indir_size = get_rss_table_size,
2818 .get_rxfh_indir = get_rss_table,
2819 .set_rxfh_indir = set_rss_table,
2820 .flash_device = set_flash,
2826 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2830 loff_t avail = file_inode(file)->i_size;
2831 unsigned int mem = (uintptr_t)file->private_data & 3;
2832 struct adapter *adap = file->private_data - mem;
2838 if (count > avail - pos)
2839 count = avail - pos;
2846 if ((mem == MEM_MC) || (mem == MEM_MC1))
2847 ret = t4_mc_read(adap, mem % MEM_MC, pos, data, NULL);
2849 ret = t4_edc_read(adap, mem, pos, data, NULL);
2853 ofst = pos % sizeof(data);
2854 len = min(count, sizeof(data) - ofst);
2855 if (copy_to_user(buf, (u8 *)data + ofst, len))
2862 count = pos - *ppos;
2867 static const struct file_operations mem_debugfs_fops = {
2868 .owner = THIS_MODULE,
2869 .open = simple_open,
2871 .llseek = default_llseek,
2874 static void add_debugfs_mem(struct adapter *adap, const char *name,
2875 unsigned int idx, unsigned int size_mb)
2879 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2880 (void *)adap + idx, &mem_debugfs_fops);
2881 if (de && de->d_inode)
2882 de->d_inode->i_size = size_mb << 20;
2885 static int setup_debugfs(struct adapter *adap)
2890 if (IS_ERR_OR_NULL(adap->debugfs_root))
2893 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2894 if (i & EDRAM0_ENABLE) {
2895 size = t4_read_reg(adap, MA_EDRAM0_BAR);
2896 add_debugfs_mem(adap, "edc0", MEM_EDC0, EDRAM_SIZE_GET(size));
2898 if (i & EDRAM1_ENABLE) {
2899 size = t4_read_reg(adap, MA_EDRAM1_BAR);
2900 add_debugfs_mem(adap, "edc1", MEM_EDC1, EDRAM_SIZE_GET(size));
2902 if (is_t4(adap->chip)) {
2903 size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2904 if (i & EXT_MEM_ENABLE)
2905 add_debugfs_mem(adap, "mc", MEM_MC,
2906 EXT_MEM_SIZE_GET(size));
2908 if (i & EXT_MEM_ENABLE) {
2909 size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2910 add_debugfs_mem(adap, "mc0", MEM_MC0,
2911 EXT_MEM_SIZE_GET(size));
2913 if (i & EXT_MEM1_ENABLE) {
2914 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR);
2915 add_debugfs_mem(adap, "mc1", MEM_MC1,
2916 EXT_MEM_SIZE_GET(size));
2920 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2926 * upper-layer driver support
2930 * Allocate an active-open TID and set it to the supplied value.
2932 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2936 spin_lock_bh(&t->atid_lock);
2938 union aopen_entry *p = t->afree;
2940 atid = (p - t->atid_tab) + t->atid_base;
2945 spin_unlock_bh(&t->atid_lock);
2948 EXPORT_SYMBOL(cxgb4_alloc_atid);
2951 * Release an active-open TID.
2953 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2955 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
2957 spin_lock_bh(&t->atid_lock);
2961 spin_unlock_bh(&t->atid_lock);
2963 EXPORT_SYMBOL(cxgb4_free_atid);
2966 * Allocate a server TID and set it to the supplied value.
2968 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
2972 spin_lock_bh(&t->stid_lock);
2973 if (family == PF_INET) {
2974 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
2975 if (stid < t->nstids)
2976 __set_bit(stid, t->stid_bmap);
2980 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
2985 t->stid_tab[stid].data = data;
2986 stid += t->stid_base;
2989 spin_unlock_bh(&t->stid_lock);
2992 EXPORT_SYMBOL(cxgb4_alloc_stid);
2994 /* Allocate a server filter TID and set it to the supplied value.
2996 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
3000 spin_lock_bh(&t->stid_lock);
3001 if (family == PF_INET) {
3002 stid = find_next_zero_bit(t->stid_bmap,
3003 t->nstids + t->nsftids, t->nstids);
3004 if (stid < (t->nstids + t->nsftids))
3005 __set_bit(stid, t->stid_bmap);
3012 t->stid_tab[stid].data = data;
3013 stid += t->stid_base;
3016 spin_unlock_bh(&t->stid_lock);
3019 EXPORT_SYMBOL(cxgb4_alloc_sftid);
3021 /* Release a server TID.
3023 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
3025 stid -= t->stid_base;
3026 spin_lock_bh(&t->stid_lock);
3027 if (family == PF_INET)
3028 __clear_bit(stid, t->stid_bmap);
3030 bitmap_release_region(t->stid_bmap, stid, 2);
3031 t->stid_tab[stid].data = NULL;
3033 spin_unlock_bh(&t->stid_lock);
3035 EXPORT_SYMBOL(cxgb4_free_stid);
3038 * Populate a TID_RELEASE WR. Caller must properly size the skb.
3040 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
3043 struct cpl_tid_release *req;
3045 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
3046 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
3047 INIT_TP_WR(req, tid);
3048 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
3052 * Queue a TID release request and if necessary schedule a work queue to
3055 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
3058 void **p = &t->tid_tab[tid];
3059 struct adapter *adap = container_of(t, struct adapter, tids);
3061 spin_lock_bh(&adap->tid_release_lock);
3062 *p = adap->tid_release_head;
3063 /* Low 2 bits encode the Tx channel number */
3064 adap->tid_release_head = (void **)((uintptr_t)p | chan);
3065 if (!adap->tid_release_task_busy) {
3066 adap->tid_release_task_busy = true;
3067 queue_work(workq, &adap->tid_release_task);
3069 spin_unlock_bh(&adap->tid_release_lock);
3073 * Process the list of pending TID release requests.
3075 static void process_tid_release_list(struct work_struct *work)
3077 struct sk_buff *skb;
3078 struct adapter *adap;
3080 adap = container_of(work, struct adapter, tid_release_task);
3082 spin_lock_bh(&adap->tid_release_lock);
3083 while (adap->tid_release_head) {
3084 void **p = adap->tid_release_head;
3085 unsigned int chan = (uintptr_t)p & 3;
3086 p = (void *)p - chan;
3088 adap->tid_release_head = *p;
3090 spin_unlock_bh(&adap->tid_release_lock);
3092 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
3094 schedule_timeout_uninterruptible(1);
3096 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
3097 t4_ofld_send(adap, skb);
3098 spin_lock_bh(&adap->tid_release_lock);
3100 adap->tid_release_task_busy = false;
3101 spin_unlock_bh(&adap->tid_release_lock);
3105 * Release a TID and inform HW. If we are unable to allocate the release
3106 * message we defer to a work queue.
3108 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
3111 struct sk_buff *skb;
3112 struct adapter *adap = container_of(t, struct adapter, tids);
3114 old = t->tid_tab[tid];
3115 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
3117 t->tid_tab[tid] = NULL;
3118 mk_tid_release(skb, chan, tid);
3119 t4_ofld_send(adap, skb);
3121 cxgb4_queue_tid_release(t, chan, tid);
3123 atomic_dec(&t->tids_in_use);
3125 EXPORT_SYMBOL(cxgb4_remove_tid);
3128 * Allocate and initialize the TID tables. Returns 0 on success.
3130 static int tid_init(struct tid_info *t)
3133 unsigned int stid_bmap_size;
3134 unsigned int natids = t->natids;
3136 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
3137 size = t->ntids * sizeof(*t->tid_tab) +
3138 natids * sizeof(*t->atid_tab) +
3139 t->nstids * sizeof(*t->stid_tab) +
3140 t->nsftids * sizeof(*t->stid_tab) +
3141 stid_bmap_size * sizeof(long) +
3142 t->nftids * sizeof(*t->ftid_tab) +
3143 t->nsftids * sizeof(*t->ftid_tab);
3145 t->tid_tab = t4_alloc_mem(size);
3149 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
3150 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
3151 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
3152 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
3153 spin_lock_init(&t->stid_lock);
3154 spin_lock_init(&t->atid_lock);
3156 t->stids_in_use = 0;
3158 t->atids_in_use = 0;
3159 atomic_set(&t->tids_in_use, 0);
3161 /* Setup the free list for atid_tab and clear the stid bitmap. */
3164 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
3165 t->afree = t->atid_tab;
3167 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
3172 * cxgb4_create_server - create an IP server
3174 * @stid: the server TID
3175 * @sip: local IP address to bind server to
3176 * @sport: the server's TCP port
3177 * @queue: queue to direct messages from this server to
3179 * Create an IP server for the given port and address.
3180 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3182 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
3183 __be32 sip, __be16 sport, __be16 vlan,
3187 struct sk_buff *skb;
3188 struct adapter *adap;
3189 struct cpl_pass_open_req *req;
3191 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3195 adap = netdev2adap(dev);
3196 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
3198 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
3199 req->local_port = sport;
3200 req->peer_port = htons(0);
3201 req->local_ip = sip;
3202 req->peer_ip = htonl(0);
3203 chan = rxq_to_chan(&adap->sge, queue);
3204 req->opt0 = cpu_to_be64(TX_CHAN(chan));
3205 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3206 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3207 return t4_mgmt_tx(adap, skb);
3209 EXPORT_SYMBOL(cxgb4_create_server);
3212 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
3213 * @mtus: the HW MTU table
3214 * @mtu: the target MTU
3215 * @idx: index of selected entry in the MTU table
3217 * Returns the index and the value in the HW MTU table that is closest to
3218 * but does not exceed @mtu, unless @mtu is smaller than any value in the
3219 * table, in which case that smallest available value is selected.
3221 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
3226 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
3232 EXPORT_SYMBOL(cxgb4_best_mtu);
3235 * cxgb4_port_chan - get the HW channel of a port
3236 * @dev: the net device for the port
3238 * Return the HW Tx channel of the given port.
3240 unsigned int cxgb4_port_chan(const struct net_device *dev)
3242 return netdev2pinfo(dev)->tx_chan;
3244 EXPORT_SYMBOL(cxgb4_port_chan);
3246 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
3248 struct adapter *adap = netdev2adap(dev);
3249 u32 v1, v2, lp_count, hp_count;
3251 v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3252 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3253 if (is_t4(adap->chip)) {
3254 lp_count = G_LP_COUNT(v1);
3255 hp_count = G_HP_COUNT(v1);
3257 lp_count = G_LP_COUNT_T5(v1);
3258 hp_count = G_HP_COUNT_T5(v2);
3260 return lpfifo ? lp_count : hp_count;
3262 EXPORT_SYMBOL(cxgb4_dbfifo_count);
3265 * cxgb4_port_viid - get the VI id of a port
3266 * @dev: the net device for the port
3268 * Return the VI id of the given port.
3270 unsigned int cxgb4_port_viid(const struct net_device *dev)
3272 return netdev2pinfo(dev)->viid;
3274 EXPORT_SYMBOL(cxgb4_port_viid);
3277 * cxgb4_port_idx - get the index of a port
3278 * @dev: the net device for the port
3280 * Return the index of the given port.
3282 unsigned int cxgb4_port_idx(const struct net_device *dev)
3284 return netdev2pinfo(dev)->port_id;
3286 EXPORT_SYMBOL(cxgb4_port_idx);
3288 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
3289 struct tp_tcp_stats *v6)
3291 struct adapter *adap = pci_get_drvdata(pdev);
3293 spin_lock(&adap->stats_lock);
3294 t4_tp_get_tcp_stats(adap, v4, v6);
3295 spin_unlock(&adap->stats_lock);
3297 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
3299 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
3300 const unsigned int *pgsz_order)
3302 struct adapter *adap = netdev2adap(dev);
3304 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
3305 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
3306 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
3307 HPZ3(pgsz_order[3]));
3309 EXPORT_SYMBOL(cxgb4_iscsi_init);
3311 int cxgb4_flush_eq_cache(struct net_device *dev)
3313 struct adapter *adap = netdev2adap(dev);
3316 ret = t4_fwaddrspace_write(adap, adap->mbox,
3317 0xe1000000 + A_SGE_CTXT_CMD, 0x20000000);
3320 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
3322 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
3324 u32 addr = t4_read_reg(adap, A_SGE_DBQ_CTXT_BADDR) + 24 * qid + 8;
3328 ret = t4_mem_win_read_len(adap, addr, (__be32 *)&indices, 8);
3330 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
3331 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
3336 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
3339 struct adapter *adap = netdev2adap(dev);
3340 u16 hw_pidx, hw_cidx;
3343 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
3347 if (pidx != hw_pidx) {
3350 if (pidx >= hw_pidx)
3351 delta = pidx - hw_pidx;
3353 delta = size - hw_pidx + pidx;
3355 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3356 QID(qid) | PIDX(delta));
3361 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
3363 static struct pci_driver cxgb4_driver;
3365 static void check_neigh_update(struct neighbour *neigh)
3367 const struct device *parent;
3368 const struct net_device *netdev = neigh->dev;
3370 if (netdev->priv_flags & IFF_802_1Q_VLAN)
3371 netdev = vlan_dev_real_dev(netdev);
3372 parent = netdev->dev.parent;
3373 if (parent && parent->driver == &cxgb4_driver.driver)
3374 t4_l2t_update(dev_get_drvdata(parent), neigh);
3377 static int netevent_cb(struct notifier_block *nb, unsigned long event,
3381 case NETEVENT_NEIGH_UPDATE:
3382 check_neigh_update(data);
3384 case NETEVENT_REDIRECT:
3391 static bool netevent_registered;
3392 static struct notifier_block cxgb4_netevent_nb = {
3393 .notifier_call = netevent_cb
3396 static void drain_db_fifo(struct adapter *adap, int usecs)
3398 u32 v1, v2, lp_count, hp_count;
3401 v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3402 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3403 if (is_t4(adap->chip)) {
3404 lp_count = G_LP_COUNT(v1);
3405 hp_count = G_HP_COUNT(v1);
3407 lp_count = G_LP_COUNT_T5(v1);
3408 hp_count = G_HP_COUNT_T5(v2);
3411 if (lp_count == 0 && hp_count == 0)
3413 set_current_state(TASK_UNINTERRUPTIBLE);
3414 schedule_timeout(usecs_to_jiffies(usecs));
3418 static void disable_txq_db(struct sge_txq *q)
3420 spin_lock_irq(&q->db_lock);
3422 spin_unlock_irq(&q->db_lock);
3425 static void enable_txq_db(struct sge_txq *q)
3427 spin_lock_irq(&q->db_lock);
3429 spin_unlock_irq(&q->db_lock);
3432 static void disable_dbs(struct adapter *adap)
3436 for_each_ethrxq(&adap->sge, i)
3437 disable_txq_db(&adap->sge.ethtxq[i].q);
3438 for_each_ofldrxq(&adap->sge, i)
3439 disable_txq_db(&adap->sge.ofldtxq[i].q);
3440 for_each_port(adap, i)
3441 disable_txq_db(&adap->sge.ctrlq[i].q);
3444 static void enable_dbs(struct adapter *adap)
3448 for_each_ethrxq(&adap->sge, i)
3449 enable_txq_db(&adap->sge.ethtxq[i].q);
3450 for_each_ofldrxq(&adap->sge, i)
3451 enable_txq_db(&adap->sge.ofldtxq[i].q);
3452 for_each_port(adap, i)
3453 enable_txq_db(&adap->sge.ctrlq[i].q);
3456 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
3458 u16 hw_pidx, hw_cidx;
3461 spin_lock_bh(&q->db_lock);
3462 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
3465 if (q->db_pidx != hw_pidx) {
3468 if (q->db_pidx >= hw_pidx)
3469 delta = q->db_pidx - hw_pidx;
3471 delta = q->size - hw_pidx + q->db_pidx;
3473 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3474 QID(q->cntxt_id) | PIDX(delta));
3478 spin_unlock_bh(&q->db_lock);
3480 CH_WARN(adap, "DB drop recovery failed.\n");
3482 static void recover_all_queues(struct adapter *adap)
3486 for_each_ethrxq(&adap->sge, i)
3487 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
3488 for_each_ofldrxq(&adap->sge, i)
3489 sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
3490 for_each_port(adap, i)
3491 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
3494 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
3496 mutex_lock(&uld_mutex);
3497 if (adap->uld_handle[CXGB4_ULD_RDMA])
3498 ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
3500 mutex_unlock(&uld_mutex);
3503 static void process_db_full(struct work_struct *work)
3505 struct adapter *adap;
3507 adap = container_of(work, struct adapter, db_full_task);
3509 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
3510 drain_db_fifo(adap, dbfifo_drain_delay);
3511 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3512 DBFIFO_HP_INT | DBFIFO_LP_INT,
3513 DBFIFO_HP_INT | DBFIFO_LP_INT);
3514 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3517 static void process_db_drop(struct work_struct *work)
3519 struct adapter *adap;
3521 adap = container_of(work, struct adapter, db_drop_task);
3523 if (is_t4(adap->chip)) {
3525 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
3526 drain_db_fifo(adap, 1);
3527 recover_all_queues(adap);
3530 u32 dropped_db = t4_read_reg(adap, 0x010ac);
3531 u16 qid = (dropped_db >> 15) & 0x1ffff;
3532 u16 pidx_inc = dropped_db & 0x1fff;
3534 unsigned short udb_density;
3535 unsigned long qpshift;
3539 dev_warn(adap->pdev_dev,
3540 "Dropped DB 0x%x qid %d bar2 %d coalesce %d pidx %d\n",
3542 (dropped_db >> 14) & 1,
3543 (dropped_db >> 13) & 1,
3546 drain_db_fifo(adap, 1);
3548 s_qpp = QUEUESPERPAGEPF1 * adap->fn;
3549 udb_density = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adap,
3550 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
3551 qpshift = PAGE_SHIFT - ilog2(udb_density);
3552 udb = qid << qpshift;
3554 page = udb / PAGE_SIZE;
3555 udb += (qid - (page * udb_density)) * 128;
3557 writel(PIDX(pidx_inc), adap->bar2 + udb + 8);
3559 /* Re-enable BAR2 WC */
3560 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
3563 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_DROPPED_DB, 0);
3566 void t4_db_full(struct adapter *adap)
3568 if (is_t4(adap->chip)) {
3569 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3570 DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
3571 queue_work(workq, &adap->db_full_task);
3575 void t4_db_dropped(struct adapter *adap)
3577 if (is_t4(adap->chip))
3578 queue_work(workq, &adap->db_drop_task);
3581 static void uld_attach(struct adapter *adap, unsigned int uld)
3584 struct cxgb4_lld_info lli;
3587 lli.pdev = adap->pdev;
3588 lli.l2t = adap->l2t;
3589 lli.tids = &adap->tids;
3590 lli.ports = adap->port;
3591 lli.vr = &adap->vres;
3592 lli.mtus = adap->params.mtus;
3593 if (uld == CXGB4_ULD_RDMA) {
3594 lli.rxq_ids = adap->sge.rdma_rxq;
3595 lli.nrxq = adap->sge.rdmaqs;
3596 } else if (uld == CXGB4_ULD_ISCSI) {
3597 lli.rxq_ids = adap->sge.ofld_rxq;
3598 lli.nrxq = adap->sge.ofldqsets;
3600 lli.ntxq = adap->sge.ofldqsets;
3601 lli.nchan = adap->params.nports;
3602 lli.nports = adap->params.nports;
3603 lli.wr_cred = adap->params.ofldq_wr_cred;
3604 lli.adapter_type = adap->params.rev;
3605 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
3606 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
3607 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
3609 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
3610 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
3612 lli.filt_mode = adap->filter_mode;
3613 /* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
3614 for (i = 0; i < NCHAN; i++)
3616 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
3617 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
3618 lli.fw_vers = adap->params.fw_vers;
3619 lli.dbfifo_int_thresh = dbfifo_int_thresh;
3620 lli.sge_pktshift = adap->sge.pktshift;
3621 lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
3623 handle = ulds[uld].add(&lli);
3624 if (IS_ERR(handle)) {
3625 dev_warn(adap->pdev_dev,
3626 "could not attach to the %s driver, error %ld\n",
3627 uld_str[uld], PTR_ERR(handle));
3631 adap->uld_handle[uld] = handle;
3633 if (!netevent_registered) {
3634 register_netevent_notifier(&cxgb4_netevent_nb);
3635 netevent_registered = true;
3638 if (adap->flags & FULL_INIT_DONE)
3639 ulds[uld].state_change(handle, CXGB4_STATE_UP);
3642 static void attach_ulds(struct adapter *adap)
3646 mutex_lock(&uld_mutex);
3647 list_add_tail(&adap->list_node, &adapter_list);
3648 for (i = 0; i < CXGB4_ULD_MAX; i++)
3650 uld_attach(adap, i);
3651 mutex_unlock(&uld_mutex);
3654 static void detach_ulds(struct adapter *adap)
3658 mutex_lock(&uld_mutex);
3659 list_del(&adap->list_node);
3660 for (i = 0; i < CXGB4_ULD_MAX; i++)
3661 if (adap->uld_handle[i]) {
3662 ulds[i].state_change(adap->uld_handle[i],
3663 CXGB4_STATE_DETACH);
3664 adap->uld_handle[i] = NULL;
3666 if (netevent_registered && list_empty(&adapter_list)) {
3667 unregister_netevent_notifier(&cxgb4_netevent_nb);
3668 netevent_registered = false;
3670 mutex_unlock(&uld_mutex);
3673 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
3677 mutex_lock(&uld_mutex);
3678 for (i = 0; i < CXGB4_ULD_MAX; i++)
3679 if (adap->uld_handle[i])
3680 ulds[i].state_change(adap->uld_handle[i], new_state);
3681 mutex_unlock(&uld_mutex);
3685 * cxgb4_register_uld - register an upper-layer driver
3686 * @type: the ULD type
3687 * @p: the ULD methods
3689 * Registers an upper-layer driver with this driver and notifies the ULD
3690 * about any presently available devices that support its type. Returns
3691 * %-EBUSY if a ULD of the same type is already registered.
3693 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
3696 struct adapter *adap;
3698 if (type >= CXGB4_ULD_MAX)
3700 mutex_lock(&uld_mutex);
3701 if (ulds[type].add) {
3706 list_for_each_entry(adap, &adapter_list, list_node)
3707 uld_attach(adap, type);
3708 out: mutex_unlock(&uld_mutex);
3711 EXPORT_SYMBOL(cxgb4_register_uld);
3714 * cxgb4_unregister_uld - unregister an upper-layer driver
3715 * @type: the ULD type
3717 * Unregisters an existing upper-layer driver.
3719 int cxgb4_unregister_uld(enum cxgb4_uld type)
3721 struct adapter *adap;
3723 if (type >= CXGB4_ULD_MAX)
3725 mutex_lock(&uld_mutex);
3726 list_for_each_entry(adap, &adapter_list, list_node)
3727 adap->uld_handle[type] = NULL;
3728 ulds[type].add = NULL;
3729 mutex_unlock(&uld_mutex);
3732 EXPORT_SYMBOL(cxgb4_unregister_uld);
3735 * cxgb_up - enable the adapter
3736 * @adap: adapter being enabled
3738 * Called when the first port is enabled, this function performs the
3739 * actions necessary to make an adapter operational, such as completing
3740 * the initialization of HW modules, and enabling interrupts.
3742 * Must be called with the rtnl lock held.
3744 static int cxgb_up(struct adapter *adap)
3748 err = setup_sge_queues(adap);
3751 err = setup_rss(adap);
3755 if (adap->flags & USING_MSIX) {
3756 name_msix_vecs(adap);
3757 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
3758 adap->msix_info[0].desc, adap);
3762 err = request_msix_queue_irqs(adap);
3764 free_irq(adap->msix_info[0].vec, adap);
3768 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
3769 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
3770 adap->port[0]->name, adap);
3776 t4_intr_enable(adap);
3777 adap->flags |= FULL_INIT_DONE;
3778 notify_ulds(adap, CXGB4_STATE_UP);
3782 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
3784 t4_free_sge_resources(adap);
3788 static void cxgb_down(struct adapter *adapter)
3790 t4_intr_disable(adapter);
3791 cancel_work_sync(&adapter->tid_release_task);
3792 cancel_work_sync(&adapter->db_full_task);
3793 cancel_work_sync(&adapter->db_drop_task);
3794 adapter->tid_release_task_busy = false;
3795 adapter->tid_release_head = NULL;
3797 if (adapter->flags & USING_MSIX) {
3798 free_msix_queue_irqs(adapter);
3799 free_irq(adapter->msix_info[0].vec, adapter);
3801 free_irq(adapter->pdev->irq, adapter);
3802 quiesce_rx(adapter);
3803 t4_sge_stop(adapter);
3804 t4_free_sge_resources(adapter);
3805 adapter->flags &= ~FULL_INIT_DONE;
3809 * net_device operations
3811 static int cxgb_open(struct net_device *dev)
3814 struct port_info *pi = netdev_priv(dev);
3815 struct adapter *adapter = pi->adapter;
3817 netif_carrier_off(dev);
3819 if (!(adapter->flags & FULL_INIT_DONE)) {
3820 err = cxgb_up(adapter);
3825 err = link_start(dev);
3827 netif_tx_start_all_queues(dev);
3831 static int cxgb_close(struct net_device *dev)
3833 struct port_info *pi = netdev_priv(dev);
3834 struct adapter *adapter = pi->adapter;
3836 netif_tx_stop_all_queues(dev);
3837 netif_carrier_off(dev);
3838 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
3841 /* Return an error number if the indicated filter isn't writable ...
3843 static int writable_filter(struct filter_entry *f)
3853 /* Delete the filter at the specified index (if valid). The checks for all
3854 * the common problems with doing this like the filter being locked, currently
3855 * pending in another operation, etc.
3857 static int delete_filter(struct adapter *adapter, unsigned int fidx)
3859 struct filter_entry *f;
3862 if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
3865 f = &adapter->tids.ftid_tab[fidx];
3866 ret = writable_filter(f);
3870 return del_filter_wr(adapter, fidx);
3875 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
3876 __be32 sip, __be16 sport, __be16 vlan,
3877 unsigned int queue, unsigned char port, unsigned char mask)
3880 struct filter_entry *f;
3881 struct adapter *adap;
3885 adap = netdev2adap(dev);
3887 /* Adjust stid to correct filter index */
3888 stid -= adap->tids.nstids;
3889 stid += adap->tids.nftids;
3891 /* Check to make sure the filter requested is writable ...
3893 f = &adap->tids.ftid_tab[stid];
3894 ret = writable_filter(f);
3898 /* Clear out any old resources being used by the filter before
3899 * we start constructing the new filter.
3902 clear_filter(adap, f);
3904 /* Clear out filter specifications */
3905 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
3906 f->fs.val.lport = cpu_to_be16(sport);
3907 f->fs.mask.lport = ~0;
3909 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
3910 for (i = 0; i < 4; i++) {
3911 f->fs.val.lip[i] = val[i];
3912 f->fs.mask.lip[i] = ~0;
3914 if (adap->filter_mode & F_PORT) {
3915 f->fs.val.iport = port;
3916 f->fs.mask.iport = mask;
3922 /* Mark filter as locked */
3926 ret = set_filter_wr(adap, stid);
3928 clear_filter(adap, f);
3934 EXPORT_SYMBOL(cxgb4_create_server_filter);
3936 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
3937 unsigned int queue, bool ipv6)
3940 struct filter_entry *f;
3941 struct adapter *adap;
3943 adap = netdev2adap(dev);
3945 /* Adjust stid to correct filter index */
3946 stid -= adap->tids.nstids;
3947 stid += adap->tids.nftids;
3949 f = &adap->tids.ftid_tab[stid];
3950 /* Unlock the filter */
3953 ret = delete_filter(adap, stid);
3959 EXPORT_SYMBOL(cxgb4_remove_server_filter);
3961 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
3962 struct rtnl_link_stats64 *ns)
3964 struct port_stats stats;
3965 struct port_info *p = netdev_priv(dev);
3966 struct adapter *adapter = p->adapter;
3968 spin_lock(&adapter->stats_lock);
3969 t4_get_port_stats(adapter, p->tx_chan, &stats);
3970 spin_unlock(&adapter->stats_lock);
3972 ns->tx_bytes = stats.tx_octets;
3973 ns->tx_packets = stats.tx_frames;
3974 ns->rx_bytes = stats.rx_octets;
3975 ns->rx_packets = stats.rx_frames;
3976 ns->multicast = stats.rx_mcast_frames;
3978 /* detailed rx_errors */
3979 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
3981 ns->rx_over_errors = 0;
3982 ns->rx_crc_errors = stats.rx_fcs_err;
3983 ns->rx_frame_errors = stats.rx_symbol_err;
3984 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
3985 stats.rx_ovflow2 + stats.rx_ovflow3 +
3986 stats.rx_trunc0 + stats.rx_trunc1 +
3987 stats.rx_trunc2 + stats.rx_trunc3;
3988 ns->rx_missed_errors = 0;
3990 /* detailed tx_errors */
3991 ns->tx_aborted_errors = 0;
3992 ns->tx_carrier_errors = 0;
3993 ns->tx_fifo_errors = 0;
3994 ns->tx_heartbeat_errors = 0;
3995 ns->tx_window_errors = 0;
3997 ns->tx_errors = stats.tx_error_frames;
3998 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
3999 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
4003 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
4006 int ret = 0, prtad, devad;
4007 struct port_info *pi = netdev_priv(dev);
4008 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
4012 if (pi->mdio_addr < 0)
4014 data->phy_id = pi->mdio_addr;
4018 if (mdio_phy_id_is_c45(data->phy_id)) {
4019 prtad = mdio_phy_id_prtad(data->phy_id);
4020 devad = mdio_phy_id_devad(data->phy_id);
4021 } else if (data->phy_id < 32) {
4022 prtad = data->phy_id;
4024 data->reg_num &= 0x1f;
4028 mbox = pi->adapter->fn;
4029 if (cmd == SIOCGMIIREG)
4030 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
4031 data->reg_num, &data->val_out);
4033 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
4034 data->reg_num, data->val_in);
4042 static void cxgb_set_rxmode(struct net_device *dev)
4044 /* unfortunately we can't return errors to the stack */
4045 set_rxmode(dev, -1, false);
4048 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
4051 struct port_info *pi = netdev_priv(dev);
4053 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
4055 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
4062 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
4065 struct sockaddr *addr = p;
4066 struct port_info *pi = netdev_priv(dev);
4068 if (!is_valid_ether_addr(addr->sa_data))
4069 return -EADDRNOTAVAIL;
4071 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
4072 pi->xact_addr_filt, addr->sa_data, true, true);
4076 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4077 pi->xact_addr_filt = ret;
4081 #ifdef CONFIG_NET_POLL_CONTROLLER
4082 static void cxgb_netpoll(struct net_device *dev)
4084 struct port_info *pi = netdev_priv(dev);
4085 struct adapter *adap = pi->adapter;
4087 if (adap->flags & USING_MSIX) {
4089 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
4091 for (i = pi->nqsets; i; i--, rx++)
4092 t4_sge_intr_msix(0, &rx->rspq);
4094 t4_intr_handler(adap)(0, adap);
4098 static const struct net_device_ops cxgb4_netdev_ops = {
4099 .ndo_open = cxgb_open,
4100 .ndo_stop = cxgb_close,
4101 .ndo_start_xmit = t4_eth_xmit,
4102 .ndo_get_stats64 = cxgb_get_stats,
4103 .ndo_set_rx_mode = cxgb_set_rxmode,
4104 .ndo_set_mac_address = cxgb_set_mac_addr,
4105 .ndo_set_features = cxgb_set_features,
4106 .ndo_validate_addr = eth_validate_addr,
4107 .ndo_do_ioctl = cxgb_ioctl,
4108 .ndo_change_mtu = cxgb_change_mtu,
4109 #ifdef CONFIG_NET_POLL_CONTROLLER
4110 .ndo_poll_controller = cxgb_netpoll,
4114 void t4_fatal_err(struct adapter *adap)
4116 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
4117 t4_intr_disable(adap);
4118 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
4121 static void setup_memwin(struct adapter *adap)
4123 u32 bar0, mem_win0_base, mem_win1_base, mem_win2_base;
4125 bar0 = pci_resource_start(adap->pdev, 0); /* truncation intentional */
4126 if (is_t4(adap->chip)) {
4127 mem_win0_base = bar0 + MEMWIN0_BASE;
4128 mem_win1_base = bar0 + MEMWIN1_BASE;
4129 mem_win2_base = bar0 + MEMWIN2_BASE;
4131 /* For T5, only relative offset inside the PCIe BAR is passed */
4132 mem_win0_base = MEMWIN0_BASE;
4133 mem_win1_base = MEMWIN1_BASE_T5;
4134 mem_win2_base = MEMWIN2_BASE_T5;
4136 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
4137 mem_win0_base | BIR(0) |
4138 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
4139 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
4140 mem_win1_base | BIR(0) |
4141 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
4142 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
4143 mem_win2_base | BIR(0) |
4144 WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
4147 static void setup_memwin_rdma(struct adapter *adap)
4149 if (adap->vres.ocq.size) {
4150 unsigned int start, sz_kb;
4152 start = pci_resource_start(adap->pdev, 2) +
4153 OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
4154 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
4156 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
4157 start | BIR(1) | WINDOW(ilog2(sz_kb)));
4159 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
4160 adap->vres.ocq.start);
4162 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
4166 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
4171 /* get device capabilities */
4172 memset(c, 0, sizeof(*c));
4173 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4174 FW_CMD_REQUEST | FW_CMD_READ);
4175 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
4176 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
4180 /* select capabilities we'll be using */
4181 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4183 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4185 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4186 } else if (vf_acls) {
4187 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
4190 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4191 FW_CMD_REQUEST | FW_CMD_WRITE);
4192 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
4196 ret = t4_config_glbl_rss(adap, adap->fn,
4197 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4198 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4199 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
4203 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
4204 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
4210 /* tweak some settings */
4211 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
4212 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
4213 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
4214 v = t4_read_reg(adap, TP_PIO_DATA);
4215 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
4217 /* first 4 Tx modulation queues point to consecutive Tx channels */
4218 adap->params.tp.tx_modq_map = 0xE4;
4219 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
4220 V_TX_MOD_QUEUE_REQ_MAP(adap->params.tp.tx_modq_map));
4222 /* associate each Tx modulation queue with consecutive Tx channels */
4224 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4225 &v, 1, A_TP_TX_SCHED_HDR);
4226 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4227 &v, 1, A_TP_TX_SCHED_FIFO);
4228 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4229 &v, 1, A_TP_TX_SCHED_PCMD);
4231 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4232 if (is_offload(adap)) {
4233 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0,
4234 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4235 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4236 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4237 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4238 t4_write_reg(adap, A_TP_TX_MOD_CHANNEL_WEIGHT,
4239 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4240 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4241 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4242 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4245 /* get basic stuff going */
4246 return t4_early_init(adap, adap->fn);
4250 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
4252 #define MAX_ATIDS 8192U
4255 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4257 * If the firmware we're dealing with has Configuration File support, then
4258 * we use that to perform all configuration
4262 * Tweak configuration based on module parameters, etc. Most of these have
4263 * defaults assigned to them by Firmware Configuration Files (if we're using
4264 * them) but need to be explicitly set if we're using hard-coded
4265 * initialization. But even in the case of using Firmware Configuration
4266 * Files, we'd like to expose the ability to change these via module
4267 * parameters so these are essentially common tweaks/settings for
4268 * Configuration Files and hard-coded initialization ...
4270 static int adap_init0_tweaks(struct adapter *adapter)
4273 * Fix up various Host-Dependent Parameters like Page Size, Cache
4274 * Line Size, etc. The firmware default is for a 4KB Page Size and
4275 * 64B Cache Line Size ...
4277 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4280 * Process module parameters which affect early initialization.
4282 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4283 dev_err(&adapter->pdev->dev,
4284 "Ignoring illegal rx_dma_offset=%d, using 2\n",
4288 t4_set_reg_field(adapter, SGE_CONTROL,
4290 PKTSHIFT(rx_dma_offset));
4293 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4294 * adds the pseudo header itself.
4296 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG,
4297 CSUM_HAS_PSEUDO_HDR, 0);
4303 * Attempt to initialize the adapter via a Firmware Configuration File.
4305 static int adap_init0_config(struct adapter *adapter, int reset)
4307 struct fw_caps_config_cmd caps_cmd;
4308 const struct firmware *cf;
4309 unsigned long mtype = 0, maddr = 0;
4310 u32 finiver, finicsum, cfcsum;
4311 int ret, using_flash;
4312 char *fw_config_file, fw_config_file_path[256];
4315 * Reset device if necessary.
4318 ret = t4_fw_reset(adapter, adapter->mbox,
4319 PIORSTMODE | PIORST);
4325 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4326 * then use that. Otherwise, use the configuration file stored
4327 * in the adapter flash ...
4329 switch (CHELSIO_CHIP_VERSION(adapter->chip)) {
4331 fw_config_file = FW_CFNAME;
4334 fw_config_file = FW5_CFNAME;
4337 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
4338 adapter->pdev->device);
4343 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
4346 mtype = FW_MEMTYPE_CF_FLASH;
4347 maddr = t4_flash_cfg_addr(adapter);
4349 u32 params[7], val[7];
4352 if (cf->size >= FLASH_CFG_MAX_SIZE)
4355 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4356 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4357 ret = t4_query_params(adapter, adapter->mbox,
4358 adapter->fn, 0, 1, params, val);
4361 * For t4_memory_write() below addresses and
4362 * sizes have to be in terms of multiples of 4
4363 * bytes. So, if the Configuration File isn't
4364 * a multiple of 4 bytes in length we'll have
4365 * to write that out separately since we can't
4366 * guarantee that the bytes following the
4367 * residual byte in the buffer returned by
4368 * request_firmware() are zeroed out ...
4370 size_t resid = cf->size & 0x3;
4371 size_t size = cf->size & ~0x3;
4372 __be32 *data = (__be32 *)cf->data;
4374 mtype = FW_PARAMS_PARAM_Y_GET(val[0]);
4375 maddr = FW_PARAMS_PARAM_Z_GET(val[0]) << 16;
4377 ret = t4_memory_write(adapter, mtype, maddr,
4379 if (ret == 0 && resid != 0) {
4386 last.word = data[size >> 2];
4387 for (i = resid; i < 4; i++)
4389 ret = t4_memory_write(adapter, mtype,
4396 release_firmware(cf);
4402 * Issue a Capability Configuration command to the firmware to get it
4403 * to parse the Configuration File. We don't use t4_fw_config_file()
4404 * because we want the ability to modify various features after we've
4405 * processed the configuration file ...
4407 memset(&caps_cmd, 0, sizeof(caps_cmd));
4408 caps_cmd.op_to_write =
4409 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4412 caps_cmd.cfvalid_to_len16 =
4413 htonl(FW_CAPS_CONFIG_CMD_CFVALID |
4414 FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4415 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
4416 FW_LEN16(caps_cmd));
4417 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4422 finiver = ntohl(caps_cmd.finiver);
4423 finicsum = ntohl(caps_cmd.finicsum);
4424 cfcsum = ntohl(caps_cmd.cfcsum);
4425 if (finicsum != cfcsum)
4426 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4427 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4431 * And now tell the firmware to use the configuration we just loaded.
4433 caps_cmd.op_to_write =
4434 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4437 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4438 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4444 * Tweak configuration based on system architecture, module
4447 ret = adap_init0_tweaks(adapter);
4452 * And finally tell the firmware to initialize itself using the
4453 * parameters from the Configuration File.
4455 ret = t4_fw_initialize(adapter, adapter->mbox);
4459 sprintf(fw_config_file_path, "/lib/firmware/%s", fw_config_file);
4461 * Return successfully and note that we're operating with parameters
4462 * not supplied by the driver, rather than from hard-wired
4463 * initialization constants burried in the driver.
4465 adapter->flags |= USING_SOFT_PARAMS;
4466 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4467 "Configuration File %s, version %#x, computed checksum %#x\n",
4470 : fw_config_file_path),
4475 * Something bad happened. Return the error ... (If the "error"
4476 * is that there's no Configuration File on the adapter we don't
4477 * want to issue a warning since this is fairly common.)
4481 dev_warn(adapter->pdev_dev, "Configuration file error %d\n",
4487 * Attempt to initialize the adapter via hard-coded, driver supplied
4490 static int adap_init0_no_config(struct adapter *adapter, int reset)
4492 struct sge *s = &adapter->sge;
4493 struct fw_caps_config_cmd caps_cmd;
4498 * Reset device if necessary
4501 ret = t4_fw_reset(adapter, adapter->mbox,
4502 PIORSTMODE | PIORST);
4508 * Get device capabilities and select which we'll be using.
4510 memset(&caps_cmd, 0, sizeof(caps_cmd));
4511 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4512 FW_CMD_REQUEST | FW_CMD_READ);
4513 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4514 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4519 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4521 caps_cmd.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4523 caps_cmd.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4524 } else if (vf_acls) {
4525 dev_err(adapter->pdev_dev, "virtualization ACLs not supported");
4528 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4529 FW_CMD_REQUEST | FW_CMD_WRITE);
4530 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4536 * Tweak configuration based on system architecture, module
4539 ret = adap_init0_tweaks(adapter);
4544 * Select RSS Global Mode we want to use. We use "Basic Virtual"
4545 * mode which maps each Virtual Interface to its own section of
4546 * the RSS Table and we turn on all map and hash enables ...
4548 adapter->flags |= RSS_TNLALLLOOKUP;
4549 ret = t4_config_glbl_rss(adapter, adapter->mbox,
4550 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4551 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4552 FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ |
4553 ((adapter->flags & RSS_TNLALLLOOKUP) ?
4554 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP : 0));
4559 * Set up our own fundamental resource provisioning ...
4561 ret = t4_cfg_pfvf(adapter, adapter->mbox, adapter->fn, 0,
4562 PFRES_NEQ, PFRES_NETHCTRL,
4563 PFRES_NIQFLINT, PFRES_NIQ,
4564 PFRES_TC, PFRES_NVI,
4565 FW_PFVF_CMD_CMASK_MASK,
4566 pfvfres_pmask(adapter, adapter->fn, 0),
4568 PFRES_R_CAPS, PFRES_WX_CAPS);
4573 * Perform low level SGE initialization. We need to do this before we
4574 * send the firmware the INITIALIZE command because that will cause
4575 * any other PF Drivers which are waiting for the Master
4576 * Initialization to proceed forward.
4578 for (i = 0; i < SGE_NTIMERS - 1; i++)
4579 s->timer_val[i] = min(intr_holdoff[i], MAX_SGE_TIMERVAL);
4580 s->timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
4581 s->counter_val[0] = 1;
4582 for (i = 1; i < SGE_NCOUNTERS; i++)
4583 s->counter_val[i] = min(intr_cnt[i - 1],
4584 THRESHOLD_0_GET(THRESHOLD_0_MASK));
4585 t4_sge_init(adapter);
4587 #ifdef CONFIG_PCI_IOV
4589 * Provision resource limits for Virtual Functions. We currently
4590 * grant them all the same static resource limits except for the Port
4591 * Access Rights Mask which we're assigning based on the PF. All of
4592 * the static provisioning stuff for both the PF and VF really needs
4593 * to be managed in a persistent manner for each device which the
4594 * firmware controls.
4598 int max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
4599 NUM_OF_PF_WITH_SRIOV_T5;
4601 for (pf = 0; pf < max_no_pf; pf++) {
4602 if (num_vf[pf] <= 0)
4605 /* VF numbering starts at 1! */
4606 for (vf = 1; vf <= num_vf[pf]; vf++) {
4607 ret = t4_cfg_pfvf(adapter, adapter->mbox,
4609 VFRES_NEQ, VFRES_NETHCTRL,
4610 VFRES_NIQFLINT, VFRES_NIQ,
4611 VFRES_TC, VFRES_NVI,
4612 FW_PFVF_CMD_CMASK_MASK,
4616 VFRES_R_CAPS, VFRES_WX_CAPS);
4618 dev_warn(adapter->pdev_dev,
4620 "provision pf/vf=%d/%d; "
4621 "err=%d\n", pf, vf, ret);
4628 * Set up the default filter mode. Later we'll want to implement this
4629 * via a firmware command, etc. ... This needs to be done before the
4630 * firmare initialization command ... If the selected set of fields
4631 * isn't equal to the default value, we'll need to make sure that the
4632 * field selections will fit in the 36-bit budget.
4634 if (tp_vlan_pri_map != TP_VLAN_PRI_MAP_DEFAULT) {
4637 for (j = TP_VLAN_PRI_MAP_FIRST; j <= TP_VLAN_PRI_MAP_LAST; j++)
4638 switch (tp_vlan_pri_map & (1 << j)) {
4640 /* compressed filter field not enabled */
4660 case ETHERTYPE_MASK:
4666 case MPSHITTYPE_MASK:
4669 case FRAGMENTATION_MASK:
4675 dev_err(adapter->pdev_dev,
4676 "tp_vlan_pri_map=%#x needs %d bits > 36;"\
4677 " using %#x\n", tp_vlan_pri_map, bits,
4678 TP_VLAN_PRI_MAP_DEFAULT);
4679 tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
4682 v = tp_vlan_pri_map;
4683 t4_write_indirect(adapter, TP_PIO_ADDR, TP_PIO_DATA,
4684 &v, 1, TP_VLAN_PRI_MAP);
4687 * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
4688 * to support any of the compressed filter fields above. Newer
4689 * versions of the firmware do this automatically but it doesn't hurt
4690 * to set it here. Meanwhile, we do _not_ need to set Lookup Every
4691 * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
4692 * since the firmware automatically turns this on and off when we have
4693 * a non-zero number of filters active (since it does have a
4694 * performance impact).
4696 if (tp_vlan_pri_map)
4697 t4_set_reg_field(adapter, TP_GLOBAL_CONFIG,
4698 FIVETUPLELOOKUP_MASK,
4699 FIVETUPLELOOKUP_MASK);
4702 * Tweak some settings.
4704 t4_write_reg(adapter, TP_SHIFT_CNT, SYNSHIFTMAX(6) |
4705 RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
4706 PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
4707 KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
4710 * Get basic stuff going by issuing the Firmware Initialize command.
4711 * Note that this _must_ be after all PFVF commands ...
4713 ret = t4_fw_initialize(adapter, adapter->mbox);
4718 * Return successfully!
4720 dev_info(adapter->pdev_dev, "Successfully configured using built-in "\
4721 "driver parameters\n");
4725 * Something bad happened. Return the error ...
4732 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4734 static int adap_init0(struct adapter *adap)
4738 enum dev_state state;
4739 u32 params[7], val[7];
4740 struct fw_caps_config_cmd caps_cmd;
4744 * Contact FW, advertising Master capability (and potentially forcing
4745 * ourselves as the Master PF if our module parameter force_init is
4748 ret = t4_fw_hello(adap, adap->mbox, adap->fn,
4749 force_init ? MASTER_MUST : MASTER_MAY,
4752 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4756 if (ret == adap->mbox)
4757 adap->flags |= MASTER_PF;
4758 if (force_init && state == DEV_STATE_INIT)
4759 state = DEV_STATE_UNINIT;
4762 * If we're the Master PF Driver and the device is uninitialized,
4763 * then let's consider upgrading the firmware ... (We always want
4764 * to check the firmware version number in order to A. get it for
4765 * later reporting and B. to warn if the currently loaded firmware
4766 * is excessively mismatched relative to the driver.)
4768 ret = t4_check_fw_version(adap);
4769 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
4770 if (ret == -EINVAL || ret > 0) {
4771 if (upgrade_fw(adap) >= 0) {
4773 * Note that the chip was reset as part of the
4774 * firmware upgrade so we don't reset it again
4775 * below and grab the new firmware version.
4778 ret = t4_check_fw_version(adap);
4786 * Grab VPD parameters. This should be done after we establish a
4787 * connection to the firmware since some of the VPD parameters
4788 * (notably the Core Clock frequency) are retrieved via requests to
4789 * the firmware. On the other hand, we need these fairly early on
4790 * so we do this right after getting ahold of the firmware.
4792 ret = get_vpd_params(adap, &adap->params.vpd);
4797 * Find out what ports are available to us. Note that we need to do
4798 * this before calling adap_init0_no_config() since it needs nports
4802 FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4803 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
4804 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1, &v, &port_vec);
4808 adap->params.nports = hweight32(port_vec);
4809 adap->params.portvec = port_vec;
4812 * If the firmware is initialized already (and we're not forcing a
4813 * master initialization), note that we're living with existing
4814 * adapter parameters. Otherwise, it's time to try initializing the
4817 if (state == DEV_STATE_INIT) {
4818 dev_info(adap->pdev_dev, "Coming up as %s: "\
4819 "Adapter already initialized\n",
4820 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
4821 adap->flags |= USING_SOFT_PARAMS;
4823 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4824 "Initializing adapter\n");
4827 * If the firmware doesn't support Configuration
4828 * Files warn user and exit,
4831 dev_warn(adap->pdev_dev, "Firmware doesn't support "
4832 "configuration file.\n");
4834 ret = adap_init0_no_config(adap, reset);
4837 * Find out whether we're dealing with a version of
4838 * the firmware which has configuration file support.
4840 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4841 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4842 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1,
4846 * If the firmware doesn't support Configuration
4847 * Files, use the old Driver-based, hard-wired
4848 * initialization. Otherwise, try using the
4849 * Configuration File support and fall back to the
4850 * Driver-based initialization if there's no
4851 * Configuration File found.
4854 ret = adap_init0_no_config(adap, reset);
4857 * The firmware provides us with a memory
4858 * buffer where we can load a Configuration
4859 * File from the host if we want to override
4860 * the Configuration File in flash.
4863 ret = adap_init0_config(adap, reset);
4864 if (ret == -ENOENT) {
4865 dev_info(adap->pdev_dev,
4866 "No Configuration File present "
4867 "on adapter. Using hard-wired "
4868 "configuration parameters.\n");
4869 ret = adap_init0_no_config(adap, reset);
4874 dev_err(adap->pdev_dev,
4875 "could not initialize adapter, error %d\n",
4882 * If we're living with non-hard-coded parameters (either from a
4883 * Firmware Configuration File or values programmed by a different PF
4884 * Driver), give the SGE code a chance to pull in anything that it
4885 * needs ... Note that this must be called after we retrieve our VPD
4886 * parameters in order to know how to convert core ticks to seconds.
4888 if (adap->flags & USING_SOFT_PARAMS) {
4889 ret = t4_sge_init(adap);
4894 if (is_bypass_device(adap->pdev->device))
4895 adap->params.bypass = 1;
4898 * Grab some of our basic fundamental operating parameters.
4900 #define FW_PARAM_DEV(param) \
4901 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
4902 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
4904 #define FW_PARAM_PFVF(param) \
4905 FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
4906 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)| \
4907 FW_PARAMS_PARAM_Y(0) | \
4908 FW_PARAMS_PARAM_Z(0)
4910 params[0] = FW_PARAM_PFVF(EQ_START);
4911 params[1] = FW_PARAM_PFVF(L2T_START);
4912 params[2] = FW_PARAM_PFVF(L2T_END);
4913 params[3] = FW_PARAM_PFVF(FILTER_START);
4914 params[4] = FW_PARAM_PFVF(FILTER_END);
4915 params[5] = FW_PARAM_PFVF(IQFLINT_START);
4916 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6, params, val);
4919 adap->sge.egr_start = val[0];
4920 adap->l2t_start = val[1];
4921 adap->l2t_end = val[2];
4922 adap->tids.ftid_base = val[3];
4923 adap->tids.nftids = val[4] - val[3] + 1;
4924 adap->sge.ingr_start = val[5];
4926 /* query params related to active filter region */
4927 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4928 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4929 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2, params, val);
4930 /* If Active filter size is set we enable establishing
4931 * offload connection through firmware work request
4933 if ((val[0] != val[1]) && (ret >= 0)) {
4934 adap->flags |= FW_OFLD_CONN;
4935 adap->tids.aftid_base = val[0];
4936 adap->tids.aftid_end = val[1];
4940 * Get device capabilities so we can determine what resources we need
4943 memset(&caps_cmd, 0, sizeof(caps_cmd));
4944 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4945 FW_CMD_REQUEST | FW_CMD_READ);
4946 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4947 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4952 if (caps_cmd.ofldcaps) {
4953 /* query offload-related parameters */
4954 params[0] = FW_PARAM_DEV(NTID);
4955 params[1] = FW_PARAM_PFVF(SERVER_START);
4956 params[2] = FW_PARAM_PFVF(SERVER_END);
4957 params[3] = FW_PARAM_PFVF(TDDP_START);
4958 params[4] = FW_PARAM_PFVF(TDDP_END);
4959 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4960 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
4964 adap->tids.ntids = val[0];
4965 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
4966 adap->tids.stid_base = val[1];
4967 adap->tids.nstids = val[2] - val[1] + 1;
4969 * Setup server filter region. Divide the availble filter
4970 * region into two parts. Regular filters get 1/3rd and server
4971 * filters get 2/3rd part. This is only enabled if workarond
4973 * 1. For regular filters.
4974 * 2. Server filter: This are special filters which are used
4975 * to redirect SYN packets to offload queue.
4977 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
4978 adap->tids.sftid_base = adap->tids.ftid_base +
4979 DIV_ROUND_UP(adap->tids.nftids, 3);
4980 adap->tids.nsftids = adap->tids.nftids -
4981 DIV_ROUND_UP(adap->tids.nftids, 3);
4982 adap->tids.nftids = adap->tids.sftid_base -
4983 adap->tids.ftid_base;
4985 adap->vres.ddp.start = val[3];
4986 adap->vres.ddp.size = val[4] - val[3] + 1;
4987 adap->params.ofldq_wr_cred = val[5];
4989 adap->params.offload = 1;
4991 if (caps_cmd.rdmacaps) {
4992 params[0] = FW_PARAM_PFVF(STAG_START);
4993 params[1] = FW_PARAM_PFVF(STAG_END);
4994 params[2] = FW_PARAM_PFVF(RQ_START);
4995 params[3] = FW_PARAM_PFVF(RQ_END);
4996 params[4] = FW_PARAM_PFVF(PBL_START);
4997 params[5] = FW_PARAM_PFVF(PBL_END);
4998 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5002 adap->vres.stag.start = val[0];
5003 adap->vres.stag.size = val[1] - val[0] + 1;
5004 adap->vres.rq.start = val[2];
5005 adap->vres.rq.size = val[3] - val[2] + 1;
5006 adap->vres.pbl.start = val[4];
5007 adap->vres.pbl.size = val[5] - val[4] + 1;
5009 params[0] = FW_PARAM_PFVF(SQRQ_START);
5010 params[1] = FW_PARAM_PFVF(SQRQ_END);
5011 params[2] = FW_PARAM_PFVF(CQ_START);
5012 params[3] = FW_PARAM_PFVF(CQ_END);
5013 params[4] = FW_PARAM_PFVF(OCQ_START);
5014 params[5] = FW_PARAM_PFVF(OCQ_END);
5015 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
5018 adap->vres.qp.start = val[0];
5019 adap->vres.qp.size = val[1] - val[0] + 1;
5020 adap->vres.cq.start = val[2];
5021 adap->vres.cq.size = val[3] - val[2] + 1;
5022 adap->vres.ocq.start = val[4];
5023 adap->vres.ocq.size = val[5] - val[4] + 1;
5025 if (caps_cmd.iscsicaps) {
5026 params[0] = FW_PARAM_PFVF(ISCSI_START);
5027 params[1] = FW_PARAM_PFVF(ISCSI_END);
5028 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2,
5032 adap->vres.iscsi.start = val[0];
5033 adap->vres.iscsi.size = val[1] - val[0] + 1;
5035 #undef FW_PARAM_PFVF
5039 * These are finalized by FW initialization, load their values now.
5041 v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
5042 adap->params.tp.tre = TIMERRESOLUTION_GET(v);
5043 adap->params.tp.dack_re = DELAYEDACKRESOLUTION_GET(v);
5044 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
5045 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5046 adap->params.b_wnd);
5048 /* MODQ_REQ_MAP defaults to setting queues 0-3 to chan 0-3 */
5049 for (j = 0; j < NCHAN; j++)
5050 adap->params.tp.tx_modq[j] = j;
5052 t4_read_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
5053 &adap->filter_mode, 1,
5056 adap->flags |= FW_OK;
5060 * Something bad happened. If a command timed out or failed with EIO
5061 * FW does not operate within its spec or something catastrophic
5062 * happened to HW/FW, stop issuing commands.
5065 if (ret != -ETIMEDOUT && ret != -EIO)
5066 t4_fw_bye(adap, adap->mbox);
5072 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
5073 pci_channel_state_t state)
5076 struct adapter *adap = pci_get_drvdata(pdev);
5082 adap->flags &= ~FW_OK;
5083 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5084 for_each_port(adap, i) {
5085 struct net_device *dev = adap->port[i];
5087 netif_device_detach(dev);
5088 netif_carrier_off(dev);
5090 if (adap->flags & FULL_INIT_DONE)
5093 pci_disable_device(pdev);
5094 out: return state == pci_channel_io_perm_failure ?
5095 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5098 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5101 struct fw_caps_config_cmd c;
5102 struct adapter *adap = pci_get_drvdata(pdev);
5105 pci_restore_state(pdev);
5106 pci_save_state(pdev);
5107 return PCI_ERS_RESULT_RECOVERED;
5110 if (pci_enable_device(pdev)) {
5111 dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
5112 return PCI_ERS_RESULT_DISCONNECT;
5115 pci_set_master(pdev);
5116 pci_restore_state(pdev);
5117 pci_save_state(pdev);
5118 pci_cleanup_aer_uncorrect_error_status(pdev);
5120 if (t4_wait_dev_ready(adap) < 0)
5121 return PCI_ERS_RESULT_DISCONNECT;
5122 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL))
5123 return PCI_ERS_RESULT_DISCONNECT;
5124 adap->flags |= FW_OK;
5125 if (adap_init1(adap, &c))
5126 return PCI_ERS_RESULT_DISCONNECT;
5128 for_each_port(adap, i) {
5129 struct port_info *p = adap2pinfo(adap, i);
5131 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
5134 return PCI_ERS_RESULT_DISCONNECT;
5136 p->xact_addr_filt = -1;
5139 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5140 adap->params.b_wnd);
5143 return PCI_ERS_RESULT_DISCONNECT;
5144 return PCI_ERS_RESULT_RECOVERED;
5147 static void eeh_resume(struct pci_dev *pdev)
5150 struct adapter *adap = pci_get_drvdata(pdev);
5156 for_each_port(adap, i) {
5157 struct net_device *dev = adap->port[i];
5159 if (netif_running(dev)) {
5161 cxgb_set_rxmode(dev);
5163 netif_device_attach(dev);
5168 static const struct pci_error_handlers cxgb4_eeh = {
5169 .error_detected = eeh_err_detected,
5170 .slot_reset = eeh_slot_reset,
5171 .resume = eeh_resume,
5174 static inline bool is_10g_port(const struct link_config *lc)
5176 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
5179 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
5180 unsigned int size, unsigned int iqe_size)
5182 q->intr_params = QINTR_TIMER_IDX(timer_idx) |
5183 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
5184 q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
5185 q->iqe_len = iqe_size;
5190 * Perform default configuration of DMA queues depending on the number and type
5191 * of ports we found and the number of available CPUs. Most settings can be
5192 * modified by the admin prior to actual use.
5194 static void cfg_queues(struct adapter *adap)
5196 struct sge *s = &adap->sge;
5197 int i, q10g = 0, n10g = 0, qidx = 0;
5199 for_each_port(adap, i)
5200 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
5203 * We default to 1 queue per non-10G port and up to # of cores queues
5207 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
5208 if (q10g > netif_get_num_default_rss_queues())
5209 q10g = netif_get_num_default_rss_queues();
5211 for_each_port(adap, i) {
5212 struct port_info *pi = adap2pinfo(adap, i);
5214 pi->first_qset = qidx;
5215 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
5220 s->max_ethqsets = qidx; /* MSI-X may lower it later */
5222 if (is_offload(adap)) {
5224 * For offload we use 1 queue/channel if all ports are up to 1G,
5225 * otherwise we divide all available queues amongst the channels
5226 * capped by the number of available cores.
5229 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
5231 s->ofldqsets = roundup(i, adap->params.nports);
5233 s->ofldqsets = adap->params.nports;
5234 /* For RDMA one Rx queue per channel suffices */
5235 s->rdmaqs = adap->params.nports;
5238 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5239 struct sge_eth_rxq *r = &s->ethrxq[i];
5241 init_rspq(&r->rspq, 0, 0, 1024, 64);
5245 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5246 s->ethtxq[i].q.size = 1024;
5248 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5249 s->ctrlq[i].q.size = 512;
5251 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
5252 s->ofldtxq[i].q.size = 1024;
5254 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
5255 struct sge_ofld_rxq *r = &s->ofldrxq[i];
5257 init_rspq(&r->rspq, 0, 0, 1024, 64);
5258 r->rspq.uld = CXGB4_ULD_ISCSI;
5262 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
5263 struct sge_ofld_rxq *r = &s->rdmarxq[i];
5265 init_rspq(&r->rspq, 0, 0, 511, 64);
5266 r->rspq.uld = CXGB4_ULD_RDMA;
5270 init_rspq(&s->fw_evtq, 6, 0, 512, 64);
5271 init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
5275 * Reduce the number of Ethernet queues across all ports to at most n.
5276 * n provides at least one queue per port.
5278 static void reduce_ethqs(struct adapter *adap, int n)
5281 struct port_info *pi;
5283 while (n < adap->sge.ethqsets)
5284 for_each_port(adap, i) {
5285 pi = adap2pinfo(adap, i);
5286 if (pi->nqsets > 1) {
5288 adap->sge.ethqsets--;
5289 if (adap->sge.ethqsets <= n)
5295 for_each_port(adap, i) {
5296 pi = adap2pinfo(adap, i);
5302 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5303 #define EXTRA_VECS 2
5305 static int enable_msix(struct adapter *adap)
5308 int i, err, want, need;
5309 struct sge *s = &adap->sge;
5310 unsigned int nchan = adap->params.nports;
5311 struct msix_entry entries[MAX_INGQ + 1];
5313 for (i = 0; i < ARRAY_SIZE(entries); ++i)
5314 entries[i].entry = i;
5316 want = s->max_ethqsets + EXTRA_VECS;
5317 if (is_offload(adap)) {
5318 want += s->rdmaqs + s->ofldqsets;
5319 /* need nchan for each possible ULD */
5320 ofld_need = 2 * nchan;
5322 need = adap->params.nports + EXTRA_VECS + ofld_need;
5324 while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
5329 * Distribute available vectors to the various queue groups.
5330 * Every group gets its minimum requirement and NIC gets top
5331 * priority for leftovers.
5333 i = want - EXTRA_VECS - ofld_need;
5334 if (i < s->max_ethqsets) {
5335 s->max_ethqsets = i;
5336 if (i < s->ethqsets)
5337 reduce_ethqs(adap, i);
5339 if (is_offload(adap)) {
5340 i = want - EXTRA_VECS - s->max_ethqsets;
5341 i -= ofld_need - nchan;
5342 s->ofldqsets = (i / nchan) * nchan; /* round down */
5344 for (i = 0; i < want; ++i)
5345 adap->msix_info[i].vec = entries[i].vector;
5347 dev_info(adap->pdev_dev,
5348 "only %d MSI-X vectors left, not using MSI-X\n", err);
5354 static int init_rss(struct adapter *adap)
5358 for_each_port(adap, i) {
5359 struct port_info *pi = adap2pinfo(adap, i);
5361 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5364 for (j = 0; j < pi->rss_size; j++)
5365 pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
5370 static void print_port_info(const struct net_device *dev)
5372 static const char *base[] = {
5373 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
5374 "KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4"
5379 const char *spd = "";
5380 const struct port_info *pi = netdev_priv(dev);
5381 const struct adapter *adap = pi->adapter;
5383 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
5385 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
5388 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
5389 bufp += sprintf(bufp, "100/");
5390 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
5391 bufp += sprintf(bufp, "1000/");
5392 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
5393 bufp += sprintf(bufp, "10G/");
5396 sprintf(bufp, "BASE-%s", base[pi->port_type]);
5398 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
5399 adap->params.vpd.id,
5400 CHELSIO_CHIP_RELEASE(adap->params.rev), buf,
5401 is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
5402 (adap->flags & USING_MSIX) ? " MSI-X" :
5403 (adap->flags & USING_MSI) ? " MSI" : "");
5404 netdev_info(dev, "S/N: %s, E/C: %s\n",
5405 adap->params.vpd.sn, adap->params.vpd.ec);
5408 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
5410 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
5414 * Free the following resources:
5415 * - memory used for tables
5418 * - resources FW is holding for us
5420 static void free_some_resources(struct adapter *adapter)
5424 t4_free_mem(adapter->l2t);
5425 t4_free_mem(adapter->tids.tid_tab);
5426 disable_msi(adapter);
5428 for_each_port(adapter, i)
5429 if (adapter->port[i]) {
5430 kfree(adap2pinfo(adapter, i)->rss);
5431 free_netdev(adapter->port[i]);
5433 if (adapter->flags & FW_OK)
5434 t4_fw_bye(adapter, adapter->fn);
5437 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5438 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5439 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5440 #define SEGMENT_SIZE 128
5442 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5444 int func, i, err, s_qpp, qpp, num_seg;
5445 struct port_info *pi;
5446 bool highdma = false;
5447 struct adapter *adapter = NULL;
5448 #ifdef CONFIG_PCI_IOV
5452 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5454 err = pci_request_regions(pdev, KBUILD_MODNAME);
5456 /* Just info, some other driver may have claimed the device. */
5457 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5461 /* We control everything through one PF */
5462 func = PCI_FUNC(pdev->devfn);
5463 if (func != ent->driver_data) {
5464 pci_save_state(pdev); /* to restore SR-IOV later */
5468 err = pci_enable_device(pdev);
5470 dev_err(&pdev->dev, "cannot enable PCI device\n");
5471 goto out_release_regions;
5474 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5476 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5478 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5479 "coherent allocations\n");
5480 goto out_disable_device;
5483 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5485 dev_err(&pdev->dev, "no usable DMA configuration\n");
5486 goto out_disable_device;
5490 pci_enable_pcie_error_reporting(pdev);
5491 enable_pcie_relaxed_ordering(pdev);
5492 pci_set_master(pdev);
5493 pci_save_state(pdev);
5495 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5498 goto out_disable_device;
5501 adapter->regs = pci_ioremap_bar(pdev, 0);
5502 if (!adapter->regs) {
5503 dev_err(&pdev->dev, "cannot map device registers\n");
5505 goto out_free_adapter;
5508 adapter->pdev = pdev;
5509 adapter->pdev_dev = &pdev->dev;
5510 adapter->mbox = func;
5512 adapter->msg_enable = dflt_msg_enable;
5513 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5515 spin_lock_init(&adapter->stats_lock);
5516 spin_lock_init(&adapter->tid_release_lock);
5518 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5519 INIT_WORK(&adapter->db_full_task, process_db_full);
5520 INIT_WORK(&adapter->db_drop_task, process_db_drop);
5522 err = t4_prep_adapter(adapter);
5524 goto out_unmap_bar0;
5526 if (!is_t4(adapter->chip)) {
5527 s_qpp = QUEUESPERPAGEPF1 * adapter->fn;
5528 qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter,
5529 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
5530 num_seg = PAGE_SIZE / SEGMENT_SIZE;
5532 /* Each segment size is 128B. Write coalescing is enabled only
5533 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5534 * queue is less no of segments that can be accommodated in
5537 if (qpp > num_seg) {
5539 "Incorrect number of egress queues per page\n");
5541 goto out_unmap_bar0;
5543 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
5544 pci_resource_len(pdev, 2));
5545 if (!adapter->bar2) {
5546 dev_err(&pdev->dev, "cannot map device bar2 region\n");
5548 goto out_unmap_bar0;
5552 setup_memwin(adapter);
5553 err = adap_init0(adapter);
5554 setup_memwin_rdma(adapter);
5558 for_each_port(adapter, i) {
5559 struct net_device *netdev;
5561 netdev = alloc_etherdev_mq(sizeof(struct port_info),
5568 SET_NETDEV_DEV(netdev, &pdev->dev);
5570 adapter->port[i] = netdev;
5571 pi = netdev_priv(netdev);
5572 pi->adapter = adapter;
5573 pi->xact_addr_filt = -1;
5575 netdev->irq = pdev->irq;
5577 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5578 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5579 NETIF_F_RXCSUM | NETIF_F_RXHASH |
5580 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
5582 netdev->hw_features |= NETIF_F_HIGHDMA;
5583 netdev->features |= netdev->hw_features;
5584 netdev->vlan_features = netdev->features & VLAN_FEAT;
5586 netdev->priv_flags |= IFF_UNICAST_FLT;
5588 netdev->netdev_ops = &cxgb4_netdev_ops;
5589 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
5592 pci_set_drvdata(pdev, adapter);
5594 if (adapter->flags & FW_OK) {
5595 err = t4_port_init(adapter, func, func, 0);
5601 * Configure queues and allocate tables now, they can be needed as
5602 * soon as the first register_netdev completes.
5604 cfg_queues(adapter);
5606 adapter->l2t = t4_init_l2t();
5607 if (!adapter->l2t) {
5608 /* We tolerate a lack of L2T, giving up some functionality */
5609 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5610 adapter->params.offload = 0;
5613 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
5614 dev_warn(&pdev->dev, "could not allocate TID table, "
5616 adapter->params.offload = 0;
5619 /* See what interrupts we'll be using */
5620 if (msi > 1 && enable_msix(adapter) == 0)
5621 adapter->flags |= USING_MSIX;
5622 else if (msi > 0 && pci_enable_msi(pdev) == 0)
5623 adapter->flags |= USING_MSI;
5625 err = init_rss(adapter);
5630 * The card is now ready to go. If any errors occur during device
5631 * registration we do not fail the whole card but rather proceed only
5632 * with the ports we manage to register successfully. However we must
5633 * register at least one net device.
5635 for_each_port(adapter, i) {
5636 pi = adap2pinfo(adapter, i);
5637 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5638 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5640 err = register_netdev(adapter->port[i]);
5643 adapter->chan_map[pi->tx_chan] = i;
5644 print_port_info(adapter->port[i]);
5647 dev_err(&pdev->dev, "could not register any net devices\n");
5651 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5655 if (cxgb4_debugfs_root) {
5656 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5657 cxgb4_debugfs_root);
5658 setup_debugfs(adapter);
5661 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5662 pdev->needs_freset = 1;
5664 if (is_offload(adapter))
5665 attach_ulds(adapter);
5668 #ifdef CONFIG_PCI_IOV
5669 max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
5670 NUM_OF_PF_WITH_SRIOV_T5;
5672 if (func < max_no_pf && num_vf[func] > 0)
5673 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
5674 dev_info(&pdev->dev,
5675 "instantiated %u virtual functions\n",
5681 free_some_resources(adapter);
5683 if (!is_t4(adapter->chip))
5684 iounmap(adapter->bar2);
5686 iounmap(adapter->regs);
5690 pci_disable_pcie_error_reporting(pdev);
5691 pci_disable_device(pdev);
5692 out_release_regions:
5693 pci_release_regions(pdev);
5694 pci_set_drvdata(pdev, NULL);
5698 static void remove_one(struct pci_dev *pdev)
5700 struct adapter *adapter = pci_get_drvdata(pdev);
5702 #ifdef CONFIG_PCI_IOV
5703 pci_disable_sriov(pdev);
5710 if (is_offload(adapter))
5711 detach_ulds(adapter);
5713 for_each_port(adapter, i)
5714 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5715 unregister_netdev(adapter->port[i]);
5717 if (adapter->debugfs_root)
5718 debugfs_remove_recursive(adapter->debugfs_root);
5720 /* If we allocated filters, free up state associated with any
5723 if (adapter->tids.ftid_tab) {
5724 struct filter_entry *f = &adapter->tids.ftid_tab[0];
5725 for (i = 0; i < (adapter->tids.nftids +
5726 adapter->tids.nsftids); i++, f++)
5728 clear_filter(adapter, f);
5731 if (adapter->flags & FULL_INIT_DONE)
5734 free_some_resources(adapter);
5735 iounmap(adapter->regs);
5736 if (!is_t4(adapter->chip))
5737 iounmap(adapter->bar2);
5739 pci_disable_pcie_error_reporting(pdev);
5740 pci_disable_device(pdev);
5741 pci_release_regions(pdev);
5742 pci_set_drvdata(pdev, NULL);
5744 pci_release_regions(pdev);
5747 static struct pci_driver cxgb4_driver = {
5748 .name = KBUILD_MODNAME,
5749 .id_table = cxgb4_pci_tbl,
5751 .remove = remove_one,
5752 .err_handler = &cxgb4_eeh,
5755 static int __init cxgb4_init_module(void)
5759 workq = create_singlethread_workqueue("cxgb4");
5763 /* Debugfs support is optional, just warn if this fails */
5764 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5765 if (!cxgb4_debugfs_root)
5766 pr_warn("could not create debugfs entry, continuing\n");
5768 ret = pci_register_driver(&cxgb4_driver);
5770 debugfs_remove(cxgb4_debugfs_root);
5774 static void __exit cxgb4_cleanup_module(void)
5776 pci_unregister_driver(&cxgb4_driver);
5777 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
5778 flush_workqueue(workq);
5779 destroy_workqueue(workq);
5782 module_init(cxgb4_init_module);
5783 module_exit(cxgb4_cleanup_module);