projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'regulator-4.20' into regulator-next
[linux-2.6-block.git] / drivers / net / ethernet / netronome / nfp / nfp_net_common.c
CommitLineData
4c352362 1/*
fdace6c2 2 * Copyright (C) 2015-2017 Netronome Systems, Inc.
4c352362
JK		 3 *
4 * This software is dual licensed under the GNU General License Version 2,
5 * June 1991 as shown in the file COPYING in the top-level directory of this
6 * source tree or the BSD 2-Clause License provided below. You have the
7 * option to license this software under the complete terms of either license.
8 *
9 * The BSD 2-Clause License:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * 1. Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * 2. Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34/*
35 * nfp_net_common.c
36 * Netronome network device driver: Common functions between PF and VF
37 * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
38 * Jason McMullan <jason.mcmullan@netronome.com>
39 * Rolf Neugebauer <rolf.neugebauer@netronome.com>
40 * Brad Petrus <brad.petrus@netronome.com>
41 * Chris Telfer <chris.telfer@netronome.com>
42 */
43
9ff304bf 44#include <linux/bitfield.h>
ecd63a02 45#include <linux/bpf.h>
a67edbf4 46#include <linux/bpf_trace.h>
4c352362
JK		 47#include <linux/module.h>
48#include <linux/kernel.h>
49#include <linux/init.h>
50#include <linux/fs.h>
51#include <linux/netdevice.h>
52#include <linux/etherdevice.h>
53#include <linux/interrupt.h>
54#include <linux/ip.h>
55#include <linux/ipv6.h>
46627170 56#include <linux/mm.h>
5ea14712 57#include <linux/overflow.h>
c0f031bc 58#include <linux/page_ref.h>
4c352362
JK		 59#include <linux/pci.h>
60#include <linux/pci_regs.h>
61#include <linux/msi.h>
62#include <linux/ethtool.h>
63#include <linux/log2.h>
64#include <linux/if_vlan.h>
65#include <linux/random.h>
a7b1ad08 66#include <linux/vmalloc.h>
4c352362
JK		 67#include <linux/ktime.h>
68
8f15df60 69#include <net/switchdev.h>
4c352362
JK		 70#include <net/vxlan.h>
71
ce22f5a2 72#include "nfpcore/nfp_nsp.h"
bb45e51c 73#include "nfp_app.h"
4c352362
JK		 74#include "nfp_net_ctrl.h"
75#include "nfp_net.h"
25528d90 76#include "nfp_net_sriov.h"
eb488c26 77#include "nfp_port.h"
4c352362
JK		 78
79/**
80 * nfp_net_get_fw_version() - Read and parse the FW version
81 * @fw_ver: Output fw_version structure to read to
82 * @ctrl_bar: Mapped address of the control BAR
83 */
84void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
85 void __iomem *ctrl_bar)
86{
87 u32 reg;
88
89 reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
90 put_unaligned_le32(reg, fw_ver);
91}
92
c487e6b1 93static dma_addr_t nfp_net_dma_map_rx(struct nfp_net_dp *dp, void *frag)
c0f031bc 94{
5cd4fbea
JK		 95 return dma_map_single_attrs(dp->dev, frag + NFP_NET_RX_BUF_HEADROOM,
96 dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
97 dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
98}
99
100static void
101nfp_net_dma_sync_dev_rx(const struct nfp_net_dp *dp, dma_addr_t dma_addr)
102{
103 dma_sync_single_for_device(dp->dev, dma_addr,
104 dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
105 dp->rx_dma_dir);
c0f031bc
JK		 106}
107
c487e6b1 108static void nfp_net_dma_unmap_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr)
c0f031bc 109{
5cd4fbea
JK		 110 dma_unmap_single_attrs(dp->dev, dma_addr,
111 dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
112 dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
113}
114
115static void nfp_net_dma_sync_cpu_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr,
116 unsigned int len)
117{
118 dma_sync_single_for_cpu(dp->dev, dma_addr - NFP_NET_RX_BUF_HEADROOM,
119 len, dp->rx_dma_dir);
c0f031bc
JK		 120}
121
3d780b92
JK		 122/* Firmware reconfig
123 *
124 * Firmware reconfig may take a while so we have two versions of it -
125 * synchronous and asynchronous (posted). All synchronous callers are holding
126 * RTNL so we don't have to worry about serializing them.
127 */
128static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
129{
130 nn_writel(nn, NFP_NET_CFG_UPDATE, update);
131 /* ensure update is written before pinging HW */
132 nn_pci_flush(nn);
133 nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
134}
135
136/* Pass 0 as update to run posted reconfigs. */
137static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
138{
139 update |= nn->reconfig_posted;
140 nn->reconfig_posted = 0;
141
142 nfp_net_reconfig_start(nn, update);
143
144 nn->reconfig_timer_active = true;
145 mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
146}
147
148static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
149{
150 u32 reg;
151
152 reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
153 if (reg == 0)
154 return true;
155 if (reg & NFP_NET_CFG_UPDATE_ERR) {
156 nn_err(nn, "Reconfig error: 0x%08x\n", reg);
157 return true;
158 } else if (last_check) {
159 nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
160 return true;
161 }
162
163 return false;
164}
165
166static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
167{
168 bool timed_out = false;
169
170 /* Poll update field, waiting for NFP to ack the config */
171 while (!nfp_net_reconfig_check_done(nn, timed_out)) {
172 msleep(1);
173 timed_out = time_is_before_eq_jiffies(deadline);
174 }
175
176 if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
177 return -EIO;
178
179 return timed_out ? -EIO : 0;
180}
181
3248f77f 182static void nfp_net_reconfig_timer(struct timer_list *t)
3d780b92 183{
3248f77f 184 struct nfp_net *nn = from_timer(nn, t, reconfig_timer);
3d780b92
JK		 185
186 spin_lock_bh(&nn->reconfig_lock);
187
188 nn->reconfig_timer_active = false;
189
190 /* If sync caller is present it will take over from us */
191 if (nn->reconfig_sync_present)
192 goto done;
193
194 /* Read reconfig status and report errors */
195 nfp_net_reconfig_check_done(nn, true);
196
197 if (nn->reconfig_posted)
198 nfp_net_reconfig_start_async(nn, 0);
199done:
200 spin_unlock_bh(&nn->reconfig_lock);
201}
202
203/**
204 * nfp_net_reconfig_post() - Post async reconfig request
205 * @nn: NFP Net device to reconfigure
206 * @update: The value for the update field in the BAR config
207 *
208 * Record FW reconfiguration request. Reconfiguration will be kicked off
209 * whenever reconfiguration machinery is idle. Multiple requests can be
210 * merged together!
211 */
212static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
213{
214 spin_lock_bh(&nn->reconfig_lock);
215
216 /* Sync caller will kick off async reconf when it's done, just post */
217 if (nn->reconfig_sync_present) {
218 nn->reconfig_posted |= update;
219 goto done;
220 }
221
222 /* Opportunistically check if the previous command is done */
223 if (!nn->reconfig_timer_active ||
224 nfp_net_reconfig_check_done(nn, false))
225 nfp_net_reconfig_start_async(nn, update);
226 else
227 nn->reconfig_posted |= update;
228done:
229 spin_unlock_bh(&nn->reconfig_lock);
230}
231
9ad716b9 232static void nfp_net_reconfig_sync_enter(struct nfp_net *nn)
4c352362 233{
3d780b92
JK		 234 bool cancelled_timer = false;
235 u32 pre_posted_requests;
4c352362
JK		 236
237 spin_lock_bh(&nn->reconfig_lock);
238
3d780b92 239 nn->reconfig_sync_present = true;
4c352362 240
3d780b92 241 if (nn->reconfig_timer_active) {
3d780b92
JK		 242 nn->reconfig_timer_active = false;
243 cancelled_timer = true;
244 }
245 pre_posted_requests = nn->reconfig_posted;
246 nn->reconfig_posted = 0;
247
248 spin_unlock_bh(&nn->reconfig_lock);
249
9ad716b9
JK		 250 if (cancelled_timer) {
251 del_timer_sync(&nn->reconfig_timer);
3d780b92 252 nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
9ad716b9 253 }
3d780b92
JK		 254
255 /* Run the posted reconfigs which were issued before we started */
256 if (pre_posted_requests) {
257 nfp_net_reconfig_start(nn, pre_posted_requests);
258 nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
4c352362 259 }
9ad716b9
JK		 260}
261
262static void nfp_net_reconfig_wait_posted(struct nfp_net *nn)
263{
264 nfp_net_reconfig_sync_enter(nn);
265
266 spin_lock_bh(&nn->reconfig_lock);
267 nn->reconfig_sync_present = false;
268 spin_unlock_bh(&nn->reconfig_lock);
269}
270
271/**
272 * nfp_net_reconfig() - Reconfigure the firmware
273 * @nn: NFP Net device to reconfigure
274 * @update: The value for the update field in the BAR config
275 *
276 * Write the update word to the BAR and ping the reconfig queue. The
277 * poll until the firmware has acknowledged the update by zeroing the
278 * update word.
279 *
280 * Return: Negative errno on error, 0 on success
281 */
282int nfp_net_reconfig(struct nfp_net *nn, u32 update)
283{
284 int ret;
285
286 nfp_net_reconfig_sync_enter(nn);
4c352362 287
3d780b92
JK		 288 nfp_net_reconfig_start(nn, update);
289 ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
290
291 spin_lock_bh(&nn->reconfig_lock);
292
293 if (nn->reconfig_posted)
294 nfp_net_reconfig_start_async(nn, 0);
295
296 nn->reconfig_sync_present = false;
297
4c352362 298 spin_unlock_bh(&nn->reconfig_lock);
3d780b92 299
4c352362
JK		 300 return ret;
301}
302
b64052fc
PC		 303/**
304 * nfp_net_reconfig_mbox() - Reconfigure the firmware via the mailbox
305 * @nn: NFP Net device to reconfigure
306 * @mbox_cmd: The value for the mailbox command
307 *
308 * Helper function for mailbox updates
309 *
310 * Return: Negative errno on error, 0 on success
311 */
312static int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd)
313{
527d7d1b 314 u32 mbox = nn->tlv_caps.mbox_off;
b64052fc
PC		 315 int ret;
316
527d7d1b
JK		 317 if (!nfp_net_has_mbox(&nn->tlv_caps)) {
318 nn_err(nn, "no mailbox present, command: %u\n", mbox_cmd);
319 return -EIO;
320 }
321
322 nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
b64052fc
PC		 323
324 ret = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
325 if (ret) {
326 nn_err(nn, "Mailbox update error\n");
327 return ret;
328 }
329
527d7d1b 330 return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
b64052fc
PC		 331}
332
4c352362
JK		 333/* Interrupt configuration and handling
334 */
335
4c352362
JK		 336/**
337 * nfp_net_irq_unmask() - Unmask automasked interrupt
338 * @nn: NFP Network structure
339 * @entry_nr: MSI-X table entry
340 *
416db5c1 341 * Clear the ICR for the IRQ entry.
4c352362
JK		 342 */
343static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
344{
4c352362
JK		 345 nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
346 nn_pci_flush(nn);
347}
348
349/**
fdace6c2
JK		 350 * nfp_net_irqs_alloc() - allocates MSI-X irqs
351 * @pdev: PCI device structure
352 * @irq_entries: Array to be initialized and used to hold the irq entries
353 * @min_irqs: Minimal acceptable number of interrupts
354 * @wanted_irqs: Target number of interrupts to allocate
4c352362 355 *
fdace6c2 356 * Return: Number of irqs obtained or 0 on error.
4c352362 357 */
fdace6c2
JK		 358unsigned int
359nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
360 unsigned int min_irqs, unsigned int wanted_irqs)
4c352362 361{
fdace6c2
JK		 362 unsigned int i;
363 int got_irqs;
4c352362 364
fdace6c2
JK		 365 for (i = 0; i < wanted_irqs; i++)
366 irq_entries[i].entry = i;
4c352362 367
fdace6c2
JK		 368 got_irqs = pci_enable_msix_range(pdev, irq_entries,
369 min_irqs, wanted_irqs);
370 if (got_irqs < 0) {
371 dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
372 min_irqs, wanted_irqs, got_irqs);
4c352362
JK		 373 return 0;
374 }
375
fdace6c2
JK		 376 if (got_irqs < wanted_irqs)
377 dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
378 wanted_irqs, got_irqs);
379
380 return got_irqs;
4c352362
JK		 381}
382
4c352362 383/**
fdace6c2
JK		 384 * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
385 * @nn: NFP Network structure
386 * @irq_entries: Table of allocated interrupts
387 * @n: Size of @irq_entries (number of entries to grab)
4c352362 388 *
fdace6c2
JK		 389 * After interrupts are allocated with nfp_net_irqs_alloc() this function
390 * should be called to assign them to a specific netdev (port).
4c352362 391 */
fdace6c2
JK		 392void
393nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
394 unsigned int n)
4c352362 395{
79c12a75
JK		 396 struct nfp_net_dp *dp = &nn->dp;
397
b33ae997 398 nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
79c12a75 399 dp->num_r_vecs = nn->max_r_vecs;
4c352362 400
fdace6c2 401 memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
4c352362 402
79c12a75
JK		 403 if (dp->num_rx_rings > dp->num_r_vecs ||
404 dp->num_tx_rings > dp->num_r_vecs)
87232d96
JK		 405 dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
406 dp->num_rx_rings, dp->num_tx_rings,
407 dp->num_r_vecs);
fdace6c2 408
79c12a75
JK		 409 dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
410 dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
411 dp->num_stack_tx_rings = dp->num_tx_rings;
4c352362
JK		 412}
413
414/**
415 * nfp_net_irqs_disable() - Disable interrupts
fdace6c2 416 * @pdev: PCI device structure
4c352362
JK		 417 *
418 * Undoes what @nfp_net_irqs_alloc() does.
419 */
fdace6c2 420void nfp_net_irqs_disable(struct pci_dev *pdev)
4c352362 421{
fdace6c2 422 pci_disable_msix(pdev);
4c352362
JK		 423}
424
425/**
426 * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
427 * @irq: Interrupt
428 * @data: Opaque data structure
429 *
430 * Return: Indicate if the interrupt has been handled.
431 */
432static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
433{
434 struct nfp_net_r_vector *r_vec = data;
435
436 napi_schedule_irqoff(&r_vec->napi);
437
438 /* The FW auto-masks any interrupt, either via the MASK bit in
439 * the MSI-X table or via the per entry ICR field. So there
440 * is no need to disable interrupts here.
441 */
442 return IRQ_HANDLED;
443}
444
77ece8d5
JK		 445static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
446{
447 struct nfp_net_r_vector *r_vec = data;
448
449 tasklet_schedule(&r_vec->tasklet);
450
451 return IRQ_HANDLED;
452}
453
4c352362
JK		 454/**
455 * nfp_net_read_link_status() - Reread link status from control BAR
456 * @nn: NFP Network structure
457 */
458static void nfp_net_read_link_status(struct nfp_net *nn)
459{
460 unsigned long flags;
461 bool link_up;
462 u32 sts;
463
464 spin_lock_irqsave(&nn->link_status_lock, flags);
465
466 sts = nn_readl(nn, NFP_NET_CFG_STS);
467 link_up = !!(sts & NFP_NET_CFG_STS_LINK);
468
469 if (nn->link_up == link_up)
470 goto out;
471
472 nn->link_up = link_up;
6d4f8cba
JK		 473 if (nn->port)
474 set_bit(NFP_PORT_CHANGED, &nn->port->flags);
4c352362
JK		 475
476 if (nn->link_up) {
79c12a75
JK		 477 netif_carrier_on(nn->dp.netdev);
478 netdev_info(nn->dp.netdev, "NIC Link is Up\n");
4c352362 479 } else {
79c12a75
JK		 480 netif_carrier_off(nn->dp.netdev);
481 netdev_info(nn->dp.netdev, "NIC Link is Down\n");
4c352362
JK		 482 }
483out:
484 spin_unlock_irqrestore(&nn->link_status_lock, flags);
485}
486
487/**
488 * nfp_net_irq_lsc() - Interrupt service routine for link state changes
489 * @irq: Interrupt
490 * @data: Opaque data structure
491 *
492 * Return: Indicate if the interrupt has been handled.
493 */
494static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
495{
496 struct nfp_net *nn = data;
fdace6c2
JK		 497 struct msix_entry *entry;
498
499 entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
4c352362
JK		 500
501 nfp_net_read_link_status(nn);
502
fdace6c2 503 nfp_net_irq_unmask(nn, entry->entry);
4c352362
JK		 504
505 return IRQ_HANDLED;
506}
507
508/**
509 * nfp_net_irq_exn() - Interrupt service routine for exceptions
510 * @irq: Interrupt
511 * @data: Opaque data structure
512 *
513 * Return: Indicate if the interrupt has been handled.
514 */
515static irqreturn_t nfp_net_irq_exn(int irq, void *data)
516{
517 struct nfp_net *nn = data;
518
519 nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
520 /* XXX TO BE IMPLEMENTED */
521 return IRQ_HANDLED;
522}
523
524/**
525 * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
526 * @tx_ring: TX ring structure
d79737c2
JK		 527 * @r_vec: IRQ vector servicing this ring
528 * @idx: Ring index
92e68195 529 * @is_xdp: Is this an XDP TX ring?
4c352362 530 */
d79737c2
JK		 531static void
532nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
92e68195
JK		 533 struct nfp_net_r_vector *r_vec, unsigned int idx,
534 bool is_xdp)
4c352362 535{
4c352362
JK		 536 struct nfp_net *nn = r_vec->nfp_net;
537
d79737c2
JK		 538 tx_ring->idx = idx;
539 tx_ring->r_vec = r_vec;
92e68195 540 tx_ring->is_xdp = is_xdp;
7ceb781a 541 u64_stats_init(&tx_ring->r_vec->tx_sync);
d79737c2 542
4c352362
JK		 543 tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
544 tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
545}
546
547/**
548 * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
549 * @rx_ring: RX ring structure
d79737c2
JK		 550 * @r_vec: IRQ vector servicing this ring
551 * @idx: Ring index
4c352362 552 */
d79737c2
JK		 553static void
554nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
555 struct nfp_net_r_vector *r_vec, unsigned int idx)
4c352362 556{
4c352362
JK		 557 struct nfp_net *nn = r_vec->nfp_net;
558
d79737c2
JK		 559 rx_ring->idx = idx;
560 rx_ring->r_vec = r_vec;
7ceb781a 561 u64_stats_init(&rx_ring->r_vec->rx_sync);
d79737c2 562
4c352362 563 rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
4c352362 564 rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
4c352362
JK		 565}
566
4c352362
JK		 567/**
568 * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
569 * @nn: NFP Network structure
570 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
571 * @format: printf-style format to construct the interrupt name
572 * @name: Pointer to allocated space for interrupt name
573 * @name_sz: Size of space for interrupt name
574 * @vector_idx: Index of MSI-X vector used for this interrupt
575 * @handler: IRQ handler to register for this interrupt
576 */
577static int
578nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
579 const char *format, char *name, size_t name_sz,
580 unsigned int vector_idx, irq_handler_t handler)
581{
582 struct msix_entry *entry;
583 int err;
584
585 entry = &nn->irq_entries[vector_idx];
586
77ece8d5 587 snprintf(name, name_sz, format, nfp_net_name(nn));
4c352362
JK		 588 err = request_irq(entry->vector, handler, 0, name, nn);
589 if (err) {
590 nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
591 entry->vector, err);
592 return err;
593 }
fdace6c2 594 nn_writeb(nn, ctrl_offset, entry->entry);
fc233650 595 nfp_net_irq_unmask(nn, entry->entry);
4c352362
JK		 596
597 return 0;
598}
599
600/**
601 * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
602 * @nn: NFP Network structure
603 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
604 * @vector_idx: Index of MSI-X vector used for this interrupt
605 */
606static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
607 unsigned int vector_idx)
608{
609 nn_writeb(nn, ctrl_offset, 0xff);
fc233650 610 nn_pci_flush(nn);
4c352362
JK		 611 free_irq(nn->irq_entries[vector_idx].vector, nn);
612}
613
614/* Transmit
615 *
616 * One queue controller peripheral queue is used for transmit. The
617 * driver en-queues packets for transmit by advancing the write
618 * pointer. The device indicates that packets have transmitted by
619 * advancing the read pointer. The driver maintains a local copy of
620 * the read and write pointer in @struct nfp_net_tx_ring. The driver
621 * keeps @wr_p in sync with the queue controller write pointer and can
622 * determine how many packets have been transmitted by comparing its
623 * copy of the read pointer @rd_p with the read pointer maintained by
624 * the queue controller peripheral.
625 */
626
627/**
628 * nfp_net_tx_full() - Check if the TX ring is full
629 * @tx_ring: TX ring to check
630 * @dcnt: Number of descriptors that need to be enqueued (must be >= 1)
631 *
632 * This function checks, based on the *host copy* of read/write
633 * pointer if a given TX ring is full. The real TX queue may have
634 * some newly made available slots.
635 *
636 * Return: True if the ring is full.
637 */
fa95f1d2 638static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
4c352362
JK		 639{
640 return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
641}
642
643/* Wrappers for deciding when to stop and restart TX queues */
644static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
645{
646 return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
647}
648
649static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
650{
651 return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
652}
653
654/**
655 * nfp_net_tx_ring_stop() - stop tx ring
656 * @nd_q: netdev queue
657 * @tx_ring: driver tx queue structure
658 *
659 * Safely stop TX ring. Remember that while we are running .start_xmit()
660 * someone else may be cleaning the TX ring completions so we need to be
661 * extra careful here.
662 */
663static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
664 struct nfp_net_tx_ring *tx_ring)
665{
666 netif_tx_stop_queue(nd_q);
667
668 /* We can race with the TX completion out of NAPI so recheck */
669 smp_mb();
670 if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
671 netif_tx_start_queue(nd_q);
672}
673
674/**
675 * nfp_net_tx_tso() - Set up Tx descriptor for LSO
4c352362
JK		 676 * @r_vec: per-ring structure
677 * @txbuf: Pointer to driver soft TX descriptor
678 * @txd: Pointer to HW TX descriptor
679 * @skb: Pointer to SKB
680 *
681 * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
682 * Return error on packet header greater than maximum supported LSO header size.
683 */
79c12a75 684static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
4c352362
JK		 685 struct nfp_net_tx_buf *txbuf,
686 struct nfp_net_tx_desc *txd, struct sk_buff *skb)
687{
688 u32 hdrlen;
689 u16 mss;
690
691 if (!skb_is_gso(skb))
692 return;
693
28063be6
EP		 694 if (!skb->encapsulation) {
695 txd->l3_offset = skb_network_offset(skb);
696 txd->l4_offset = skb_transport_offset(skb);
4c352362 697 hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
28063be6
EP		 698 } else {
699 txd->l3_offset = skb_inner_network_offset(skb);
700 txd->l4_offset = skb_inner_transport_offset(skb);
4c352362
JK		 701 hdrlen = skb_inner_transport_header(skb) - skb->data +
702 inner_tcp_hdrlen(skb);
28063be6 703 }
4c352362
JK		 704
705 txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
706 txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
707
708 mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
e53fc9fa 709 txd->lso_hdrlen = hdrlen;
4c352362
JK		 710 txd->mss = cpu_to_le16(mss);
711 txd->flags |= PCIE_DESC_TX_LSO;
712
713 u64_stats_update_begin(&r_vec->tx_sync);
714 r_vec->tx_lso++;
715 u64_stats_update_end(&r_vec->tx_sync);
716}
717
718/**
719 * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
79c12a75 720 * @dp: NFP Net data path struct
4c352362
JK		 721 * @r_vec: per-ring structure
722 * @txbuf: Pointer to driver soft TX descriptor
723 * @txd: Pointer to TX descriptor
724 * @skb: Pointer to SKB
725 *
726 * This function sets the TX checksum flags in the TX descriptor based
727 * on the configuration and the protocol of the packet to be transmitted.
728 */
79c12a75
JK		 729static void nfp_net_tx_csum(struct nfp_net_dp *dp,
730 struct nfp_net_r_vector *r_vec,
4c352362
JK		 731 struct nfp_net_tx_buf *txbuf,
732 struct nfp_net_tx_desc *txd, struct sk_buff *skb)
733{
734 struct ipv6hdr *ipv6h;
735 struct iphdr *iph;
736 u8 l4_hdr;
737
79c12a75 738 if (!(dp->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
4c352362
JK		 739 return;
740
741 if (skb->ip_summed != CHECKSUM_PARTIAL)
742 return;
743
744 txd->flags |= PCIE_DESC_TX_CSUM;
745 if (skb->encapsulation)
746 txd->flags |= PCIE_DESC_TX_ENCAP;
747
748 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
749 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
750
751 if (iph->version == 4) {
752 txd->flags |= PCIE_DESC_TX_IP4_CSUM;
753 l4_hdr = iph->protocol;
754 } else if (ipv6h->version == 6) {
755 l4_hdr = ipv6h->nexthdr;
756 } else {
79c12a75 757 nn_dp_warn(dp, "partial checksum but ipv=%x!\n", iph->version);
4c352362
JK		 758 return;
759 }
760
761 switch (l4_hdr) {
762 case IPPROTO_TCP:
763 txd->flags |= PCIE_DESC_TX_TCP_CSUM;
764 break;
765 case IPPROTO_UDP:
766 txd->flags |= PCIE_DESC_TX_UDP_CSUM;
767 break;
768 default:
79c12a75 769 nn_dp_warn(dp, "partial checksum but l4 proto=%x!\n", l4_hdr);
4c352362
JK		 770 return;
771 }
772
773 u64_stats_update_begin(&r_vec->tx_sync);
774 if (skb->encapsulation)
775 r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
776 else
777 r_vec->hw_csum_tx += txbuf->pkt_cnt;
778 u64_stats_update_end(&r_vec->tx_sync);
779}
780
ecd63a02
JK		 781static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
782{
783 wmb();
784 nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
785 tx_ring->wr_ptr_add = 0;
786}
787
91bf82ca
SH		 788static int nfp_net_prep_port_id(struct sk_buff *skb)
789{
790 struct metadata_dst *md_dst = skb_metadata_dst(skb);
791 unsigned char *data;
792
793 if (likely(!md_dst))
794 return 0;
795 if (unlikely(md_dst->type != METADATA_HW_PORT_MUX))
796 return 0;
797
798 if (unlikely(skb_cow_head(skb, 8)))
799 return -ENOMEM;
800
801 data = skb_push(skb, 8);
802 put_unaligned_be32(NFP_NET_META_PORTID, data);
803 put_unaligned_be32(md_dst->u.port_info.port_id, data + 4);
804
805 return 8;
806}
807
4c352362
JK		 808/**
809 * nfp_net_tx() - Main transmit entry point
810 * @skb: SKB to transmit
811 * @netdev: netdev structure
812 *
813 * Return: NETDEV_TX_OK on success.
814 */
815static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
816{
817 struct nfp_net *nn = netdev_priv(netdev);
818 const struct skb_frag_struct *frag;
4c352362 819 struct nfp_net_tx_desc *txd, txdg;
91bf82ca 820 int f, nr_frags, wr_idx, md_bytes;
4c352362 821 struct nfp_net_tx_ring *tx_ring;
bef6b1b7
JK		 822 struct nfp_net_r_vector *r_vec;
823 struct nfp_net_tx_buf *txbuf;
4c352362 824 struct netdev_queue *nd_q;
79c12a75 825 struct nfp_net_dp *dp;
4c352362
JK		 826 dma_addr_t dma_addr;
827 unsigned int fsize;
4c352362
JK		 828 u16 qidx;
829
79c12a75 830 dp = &nn->dp;
4c352362 831 qidx = skb_get_queue_mapping(skb);
79c12a75 832 tx_ring = &dp->tx_rings[qidx];
4c352362 833 r_vec = tx_ring->r_vec;
79c12a75 834 nd_q = netdev_get_tx_queue(dp->netdev, qidx);
4c352362
JK		 835
836 nr_frags = skb_shinfo(skb)->nr_frags;
837
838 if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
79c12a75
JK		 839 nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
840 qidx, tx_ring->wr_p, tx_ring->rd_p);
4c352362 841 netif_tx_stop_queue(nd_q);
28b0cfee 842 nfp_net_tx_xmit_more_flush(tx_ring);
4c352362
JK		 843 u64_stats_update_begin(&r_vec->tx_sync);
844 r_vec->tx_busy++;
845 u64_stats_update_end(&r_vec->tx_sync);
846 return NETDEV_TX_BUSY;
847 }
848
91bf82ca
SH		 849 md_bytes = nfp_net_prep_port_id(skb);
850 if (unlikely(md_bytes < 0)) {
851 nfp_net_tx_xmit_more_flush(tx_ring);
852 dev_kfree_skb_any(skb);
853 return NETDEV_TX_OK;
854 }
855
4c352362 856 /* Start with the head skbuf */
79c12a75 857 dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
4c352362 858 DMA_TO_DEVICE);
79c12a75 859 if (dma_mapping_error(dp->dev, dma_addr))
4c352362
JK		 860 goto err_free;
861
4aa3b766 862 wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
4c352362
JK		 863
864 /* Stash the soft descriptor of the head then initialize it */
865 txbuf = &tx_ring->txbufs[wr_idx];
866 txbuf->skb = skb;
867 txbuf->dma_addr = dma_addr;
868 txbuf->fidx = -1;
869 txbuf->pkt_cnt = 1;
870 txbuf->real_len = skb->len;
871
872 /* Build TX descriptor */
873 txd = &tx_ring->txds[wr_idx];
91bf82ca 874 txd->offset_eop = (nr_frags ? 0 : PCIE_DESC_TX_EOP) | md_bytes;
4c352362
JK		 875 txd->dma_len = cpu_to_le16(skb_headlen(skb));
876 nfp_desc_set_dma_addr(txd, dma_addr);
877 txd->data_len = cpu_to_le16(skb->len);
878
879 txd->flags = 0;
880 txd->mss = 0;
e53fc9fa 881 txd->lso_hdrlen = 0;
4c352362 882
28063be6 883 /* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
79c12a75 884 nfp_net_tx_tso(r_vec, txbuf, txd, skb);
79c12a75 885 nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
79c12a75 886 if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
4c352362
JK		 887 txd->flags |= PCIE_DESC_TX_VLAN;
888 txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
889 }
890
891 /* Gather DMA */
892 if (nr_frags > 0) {
893 /* all descs must match except for in addr, length and eop */
894 txdg = *txd;
895
896 for (f = 0; f < nr_frags; f++) {
897 frag = &skb_shinfo(skb)->frags[f];
898 fsize = skb_frag_size(frag);
899
79c12a75 900 dma_addr = skb_frag_dma_map(dp->dev, frag, 0,
4c352362 901 fsize, DMA_TO_DEVICE);
79c12a75 902 if (dma_mapping_error(dp->dev, dma_addr))
4c352362
JK		 903 goto err_unmap;
904
4aa3b766 905 wr_idx = D_IDX(tx_ring, wr_idx + 1);
4c352362
JK		 906 tx_ring->txbufs[wr_idx].skb = skb;
907 tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
908 tx_ring->txbufs[wr_idx].fidx = f;
909
910 txd = &tx_ring->txds[wr_idx];
911 *txd = txdg;
912 txd->dma_len = cpu_to_le16(fsize);
913 nfp_desc_set_dma_addr(txd, dma_addr);
91bf82ca 914 txd->offset_eop |=
4c352362
JK		 915 (f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
916 }
917
918 u64_stats_update_begin(&r_vec->tx_sync);
919 r_vec->tx_gather++;
920 u64_stats_update_end(&r_vec->tx_sync);
921 }
922
923 netdev_tx_sent_queue(nd_q, txbuf->real_len);
924
46f1c52e
JK		 925 skb_tx_timestamp(skb);
926
4c352362
JK		 927 tx_ring->wr_p += nr_frags + 1;
928 if (nfp_net_tx_ring_should_stop(tx_ring))
929 nfp_net_tx_ring_stop(nd_q, tx_ring);
930
931 tx_ring->wr_ptr_add += nr_frags + 1;
ecd63a02
JK		 932 if (!skb->xmit_more || netif_xmit_stopped(nd_q))
933 nfp_net_tx_xmit_more_flush(tx_ring);
4c352362 934
4c352362
JK		 935 return NETDEV_TX_OK;
936
937err_unmap:
eac2c68d 938 while (--f >= 0) {
4c352362 939 frag = &skb_shinfo(skb)->frags[f];
79c12a75 940 dma_unmap_page(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
4c352362
JK		 941 skb_frag_size(frag), DMA_TO_DEVICE);
942 tx_ring->txbufs[wr_idx].skb = NULL;
943 tx_ring->txbufs[wr_idx].dma_addr = 0;
944 tx_ring->txbufs[wr_idx].fidx = -2;
945 wr_idx = wr_idx - 1;
946 if (wr_idx < 0)
947 wr_idx += tx_ring->cnt;
948 }
79c12a75 949 dma_unmap_single(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
4c352362
JK		 950 skb_headlen(skb), DMA_TO_DEVICE);
951 tx_ring->txbufs[wr_idx].skb = NULL;
952 tx_ring->txbufs[wr_idx].dma_addr = 0;
953 tx_ring->txbufs[wr_idx].fidx = -2;
954err_free:
79c12a75 955 nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
28b0cfee 956 nfp_net_tx_xmit_more_flush(tx_ring);
4c352362
JK		 957 u64_stats_update_begin(&r_vec->tx_sync);
958 r_vec->tx_errors++;
959 u64_stats_update_end(&r_vec->tx_sync);
960 dev_kfree_skb_any(skb);
961 return NETDEV_TX_OK;
962}
963
964/**
965 * nfp_net_tx_complete() - Handled completed TX packets
d387b8a1
JK		 966 * @tx_ring: TX ring structure
967 * @budget: NAPI budget (only used as bool to determine if in NAPI context)
4c352362 968 */
d387b8a1 969static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget)
4c352362
JK		 970{
971 struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
79c12a75 972 struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
4c352362
JK		 973 const struct skb_frag_struct *frag;
974 struct netdev_queue *nd_q;
975 u32 done_pkts = 0, done_bytes = 0;
976 struct sk_buff *skb;
977 int todo, nr_frags;
978 u32 qcp_rd_p;
979 int fidx;
980 int idx;
981
d38df0d3
JK		 982 if (tx_ring->wr_p == tx_ring->rd_p)
983 return;
984
4c352362
JK		 985 /* Work out how many descriptors have been transmitted */
986 qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
987
988 if (qcp_rd_p == tx_ring->qcp_rd_p)
989 return;
990
770f0cea 991 todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
4c352362
JK		 992
993 while (todo--) {
4aa3b766 994 idx = D_IDX(tx_ring, tx_ring->rd_p++);
4c352362
JK		 995
996 skb = tx_ring->txbufs[idx].skb;
997 if (!skb)
998 continue;
999
1000 nr_frags = skb_shinfo(skb)->nr_frags;
1001 fidx = tx_ring->txbufs[idx].fidx;
1002
1003 if (fidx == -1) {
1004 /* unmap head */
79c12a75 1005 dma_unmap_single(dp->dev, tx_ring->txbufs[idx].dma_addr,
4c352362
JK		 1006 skb_headlen(skb), DMA_TO_DEVICE);
1007
1008 done_pkts += tx_ring->txbufs[idx].pkt_cnt;
1009 done_bytes += tx_ring->txbufs[idx].real_len;
1010 } else {
1011 /* unmap fragment */
1012 frag = &skb_shinfo(skb)->frags[fidx];
79c12a75 1013 dma_unmap_page(dp->dev, tx_ring->txbufs[idx].dma_addr,
4c352362
JK		 1014 skb_frag_size(frag), DMA_TO_DEVICE);
1015 }
1016
1017 /* check for last gather fragment */
1018 if (fidx == nr_frags - 1)
d387b8a1 1019 napi_consume_skb(skb, budget);
4c352362
JK		 1020
1021 tx_ring->txbufs[idx].dma_addr = 0;
1022 tx_ring->txbufs[idx].skb = NULL;
1023 tx_ring->txbufs[idx].fidx = -2;
1024 }
1025
1026 tx_ring->qcp_rd_p = qcp_rd_p;
1027
1028 u64_stats_update_begin(&r_vec->tx_sync);
1029 r_vec->tx_bytes += done_bytes;
1030 r_vec->tx_pkts += done_pkts;
1031 u64_stats_update_end(&r_vec->tx_sync);
1032
77ece8d5
JK		 1033 if (!dp->netdev)
1034 return;
1035
79c12a75 1036 nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
4c352362
JK		 1037 netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
1038 if (nfp_net_tx_ring_should_wake(tx_ring)) {
1039 /* Make sure TX thread will see updated tx_ring->rd_p */
1040 smp_mb();
1041
1042 if (unlikely(netif_tx_queue_stopped(nd_q)))
1043 netif_tx_wake_queue(nd_q);
1044 }
1045
1046 WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1047 "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1048 tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1049}
1050
abeeec4a 1051static bool nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
ecd63a02
JK		 1052{
1053 struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
ecd63a02 1054 u32 done_pkts = 0, done_bytes = 0;
abeeec4a 1055 bool done_all;
ecd63a02
JK		 1056 int idx, todo;
1057 u32 qcp_rd_p;
1058
1059 /* Work out how many descriptors have been transmitted */
1060 qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1061
1062 if (qcp_rd_p == tx_ring->qcp_rd_p)
abeeec4a 1063 return true;
ecd63a02 1064
770f0cea 1065 todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
ecd63a02 1066
abeeec4a
JK		 1067 done_all = todo <= NFP_NET_XDP_MAX_COMPLETE;
1068 todo = min(todo, NFP_NET_XDP_MAX_COMPLETE);
1069
4aa3b766 1070 tx_ring->qcp_rd_p = D_IDX(tx_ring, tx_ring->qcp_rd_p + todo);
abeeec4a 1071
92e68195 1072 done_pkts = todo;
ecd63a02 1073 while (todo--) {
4aa3b766 1074 idx = D_IDX(tx_ring, tx_ring->rd_p);
ecd63a02
JK		 1075 tx_ring->rd_p++;
1076
ecd63a02 1077 done_bytes += tx_ring->txbufs[idx].real_len;
ecd63a02
JK		 1078 }
1079
ecd63a02
JK		 1080 u64_stats_update_begin(&r_vec->tx_sync);
1081 r_vec->tx_bytes += done_bytes;
1082 r_vec->tx_pkts += done_pkts;
1083 u64_stats_update_end(&r_vec->tx_sync);
1084
1085 WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
abeeec4a 1086 "XDP TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
ecd63a02 1087 tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
abeeec4a
JK		 1088
1089 return done_all;
ecd63a02
JK		 1090}
1091
4c352362 1092/**
827deea9 1093 * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
79c12a75 1094 * @dp: NFP Net data path struct
827deea9 1095 * @tx_ring: TX ring structure
4c352362 1096 *
07300f77 1097 * Assumes that the device is stopped, must be idempotent.
4c352362 1098 */
827deea9 1099static void
79c12a75 1100nfp_net_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
4c352362 1101{
4c352362 1102 const struct skb_frag_struct *frag;
ecd63a02 1103 struct netdev_queue *nd_q;
4c352362 1104
92e68195 1105 while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
ecd63a02 1106 struct nfp_net_tx_buf *tx_buf;
92e68195
JK		 1107 struct sk_buff *skb;
1108 int idx, nr_frags;
4c352362 1109
4aa3b766 1110 idx = D_IDX(tx_ring, tx_ring->rd_p);
ecd63a02 1111 tx_buf = &tx_ring->txbufs[idx];
827deea9 1112
92e68195
JK		 1113 skb = tx_ring->txbufs[idx].skb;
1114 nr_frags = skb_shinfo(skb)->nr_frags;
4c352362 1115
92e68195
JK		 1116 if (tx_buf->fidx == -1) {
1117 /* unmap head */
1118 dma_unmap_single(dp->dev, tx_buf->dma_addr,
1119 skb_headlen(skb), DMA_TO_DEVICE);
1120 } else {
1121 /* unmap fragment */
1122 frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1123 dma_unmap_page(dp->dev, tx_buf->dma_addr,
1124 skb_frag_size(frag), DMA_TO_DEVICE);
ecd63a02 1125 }
827deea9 1126
92e68195
JK		 1127 /* check for last gather fragment */
1128 if (tx_buf->fidx == nr_frags - 1)
1129 dev_kfree_skb_any(skb);
1130
ecd63a02
JK		 1131 tx_buf->dma_addr = 0;
1132 tx_buf->skb = NULL;
1133 tx_buf->fidx = -2;
4c352362
JK		 1134
1135 tx_ring->qcp_rd_p++;
1136 tx_ring->rd_p++;
1137 }
1138
5ea14712 1139 memset(tx_ring->txds, 0, tx_ring->size);
827deea9
JK		 1140 tx_ring->wr_p = 0;
1141 tx_ring->rd_p = 0;
1142 tx_ring->qcp_rd_p = 0;
1143 tx_ring->wr_ptr_add = 0;
1144
77ece8d5 1145 if (tx_ring->is_xdp || !dp->netdev)
ecd63a02
JK		 1146 return;
1147
79c12a75 1148 nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
4c352362
JK		 1149 netdev_tx_reset_queue(nd_q);
1150}
1151
1152static void nfp_net_tx_timeout(struct net_device *netdev)
1153{
1154 struct nfp_net *nn = netdev_priv(netdev);
1155 int i;
1156
79c12a75 1157 for (i = 0; i < nn->dp.netdev->real_num_tx_queues; i++) {
4c352362
JK		 1158 if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i)))
1159 continue;
1160 nn_warn(nn, "TX timeout on ring: %d\n", i);
1161 }
1162 nn_warn(nn, "TX watchdog timeout\n");
1163}
1164
1165/* Receive processing
1166 */
bf187ea0 1167static unsigned int
76e1e1a8 1168nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
bf187ea0
JK		 1169{
1170 unsigned int fl_bufsz;
1171
c0f031bc 1172 fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
6fe0c3b4 1173 fl_bufsz += dp->rx_dma_off;
79c12a75 1174 if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
c0f031bc 1175 fl_bufsz += NFP_NET_MAX_PREPEND;
bf187ea0 1176 else
79c12a75 1177 fl_bufsz += dp->rx_offset;
76e1e1a8 1178 fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
bf187ea0 1179
c0f031bc
JK		 1180 fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1181 fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1182
bf187ea0
JK		 1183 return fl_bufsz;
1184}
4c352362 1185
ecd63a02
JK		 1186static void
1187nfp_net_free_frag(void *frag, bool xdp)
1188{
1189 if (!xdp)
1190 skb_free_frag(frag);
1191 else
1192 __free_page(virt_to_page(frag));
1193}
1194
4c352362 1195/**
c0f031bc 1196 * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
783496b0 1197 * @dp: NFP Net data path struct
4c352362
JK		 1198 * @dma_addr: Pointer to storage for DMA address (output param)
1199 *
c0f031bc 1200 * This function will allcate a new page frag, map it for DMA.
4c352362 1201 *
c0f031bc 1202 * Return: allocated page frag or NULL on failure.
4c352362 1203 */
d78005a5 1204static void *nfp_net_rx_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
4c352362 1205{
c0f031bc 1206 void *frag;
4c352362 1207
5f0ca2fb 1208 if (!dp->xdp_prog) {
2195c263 1209 frag = netdev_alloc_frag(dp->fl_bufsz);
5f0ca2fb
JK		 1210 } else {
1211 struct page *page;
1212
453f85d4 1213 page = alloc_page(GFP_KERNEL);
5f0ca2fb
JK		 1214 frag = page ? page_address(page) : NULL;
1215 }
c0f031bc 1216 if (!frag) {
79c12a75 1217 nn_dp_warn(dp, "Failed to alloc receive page frag\n");
4c352362
JK		 1218 return NULL;
1219 }
1220
c487e6b1 1221 *dma_addr = nfp_net_dma_map_rx(dp, frag);
79c12a75 1222 if (dma_mapping_error(dp->dev, *dma_addr)) {
9dc6b116 1223 nfp_net_free_frag(frag, dp->xdp_prog);
79c12a75 1224 nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
4c352362
JK		 1225 return NULL;
1226 }
1227
c0f031bc 1228 return frag;
4c352362
JK		 1229}
1230
c487e6b1 1231static void *nfp_net_napi_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
b64b7bb6
JK		 1232{
1233 void *frag;
1234
5f0ca2fb 1235 if (!dp->xdp_prog) {
79c12a75 1236 frag = napi_alloc_frag(dp->fl_bufsz);
16f50cda
JK		 1237 if (unlikely(!frag))
1238 return NULL;
5f0ca2fb
JK		 1239 } else {
1240 struct page *page;
1241
790a3991 1242 page = dev_alloc_page();
16f50cda
JK		 1243 if (unlikely(!page))
1244 return NULL;
1245 frag = page_address(page);
b64b7bb6
JK		 1246 }
1247
c487e6b1 1248 *dma_addr = nfp_net_dma_map_rx(dp, frag);
79c12a75
JK		 1249 if (dma_mapping_error(dp->dev, *dma_addr)) {
1250 nfp_net_free_frag(frag, dp->xdp_prog);
1251 nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
b64b7bb6
JK		 1252 return NULL;
1253 }
1254
1255 return frag;
1256}
1257
4c352362
JK		 1258/**
1259 * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
6fe0c3b4 1260 * @dp: NFP Net data path struct
4c352362 1261 * @rx_ring: RX ring structure
c0f031bc 1262 * @frag: page fragment buffer
4c352362
JK		 1263 * @dma_addr: DMA address of skb mapping
1264 */
6fe0c3b4
JK		 1265static void nfp_net_rx_give_one(const struct nfp_net_dp *dp,
1266 struct nfp_net_rx_ring *rx_ring,
c0f031bc 1267 void *frag, dma_addr_t dma_addr)
4c352362
JK		 1268{
1269 unsigned int wr_idx;
1270
4aa3b766 1271 wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
4c352362 1272
5cd4fbea
JK		 1273 nfp_net_dma_sync_dev_rx(dp, dma_addr);
1274
4c352362 1275 /* Stash SKB and DMA address away */
c0f031bc 1276 rx_ring->rxbufs[wr_idx].frag = frag;
4c352362
JK		 1277 rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1278
1279 /* Fill freelist descriptor */
1280 rx_ring->rxds[wr_idx].fld.reserved = 0;
1281 rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
6fe0c3b4
JK		 1282 nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld,
1283 dma_addr + dp->rx_dma_off);
4c352362
JK		 1284
1285 rx_ring->wr_p++;
9ed9ea70 1286 if (!(rx_ring->wr_p % NFP_NET_FL_BATCH)) {
4c352362
JK		 1287 /* Update write pointer of the freelist queue. Make
1288 * sure all writes are flushed before telling the hardware.
1289 */
1290 wmb();
9ed9ea70 1291 nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, NFP_NET_FL_BATCH);
4c352362
JK		 1292 }
1293}
1294
1295/**
1934680f
JK		 1296 * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1297 * @rx_ring: RX ring structure
4c352362 1298 *
07300f77 1299 * Assumes that the device is stopped, must be idempotent.
4c352362 1300 */
1934680f 1301static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
4c352362 1302{
1934680f 1303 unsigned int wr_idx, last_idx;
4c352362 1304
07300f77
JK		 1305 /* wr_p == rd_p means ring was never fed FL bufs. RX rings are always
1306 * kept at cnt - 1 FL bufs.
1307 */
1308 if (rx_ring->wr_p == 0 && rx_ring->rd_p == 0)
1309 return;
1310
1934680f 1311 /* Move the empty entry to the end of the list */
4aa3b766 1312 wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1934680f
JK		 1313 last_idx = rx_ring->cnt - 1;
1314 rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
c0f031bc 1315 rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1934680f 1316 rx_ring->rxbufs[last_idx].dma_addr = 0;
c0f031bc 1317 rx_ring->rxbufs[last_idx].frag = NULL;
4c352362 1318
5ea14712 1319 memset(rx_ring->rxds, 0, rx_ring->size);
1934680f
JK		 1320 rx_ring->wr_p = 0;
1321 rx_ring->rd_p = 0;
1934680f 1322}
4c352362 1323
1934680f
JK		 1324/**
1325 * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
79c12a75 1326 * @dp: NFP Net data path struct
1934680f
JK		 1327 * @rx_ring: RX ring to remove buffers from
1328 *
1329 * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1330 * entries. After device is disabled nfp_net_rx_ring_reset() must be called
1331 * to restore required ring geometry.
1332 */
1333static void
79c12a75 1334nfp_net_rx_ring_bufs_free(struct nfp_net_dp *dp,
9dc6b116 1335 struct nfp_net_rx_ring *rx_ring)
1934680f 1336{
1934680f 1337 unsigned int i;
4c352362 1338
1934680f
JK		 1339 for (i = 0; i < rx_ring->cnt - 1; i++) {
1340 /* NULL skb can only happen when initial filling of the ring
1341 * fails to allocate enough buffers and calls here to free
1342 * already allocated ones.
1343 */
c0f031bc 1344 if (!rx_ring->rxbufs[i].frag)
1934680f
JK		 1345 continue;
1346
c487e6b1 1347 nfp_net_dma_unmap_rx(dp, rx_ring->rxbufs[i].dma_addr);
9dc6b116 1348 nfp_net_free_frag(rx_ring->rxbufs[i].frag, dp->xdp_prog);
1934680f 1349 rx_ring->rxbufs[i].dma_addr = 0;
c0f031bc 1350 rx_ring->rxbufs[i].frag = NULL;
4c352362
JK		 1351 }
1352}
1353
1354/**
1934680f 1355 * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
79c12a75 1356 * @dp: NFP Net data path struct
1934680f 1357 * @rx_ring: RX ring to remove buffers from
4c352362 1358 */
1934680f 1359static int
79c12a75 1360nfp_net_rx_ring_bufs_alloc(struct nfp_net_dp *dp,
9dc6b116 1361 struct nfp_net_rx_ring *rx_ring)
4c352362 1362{
1934680f
JK		 1363 struct nfp_net_rx_buf *rxbufs;
1364 unsigned int i;
1365
1366 rxbufs = rx_ring->rxbufs;
4c352362 1367
1934680f 1368 for (i = 0; i < rx_ring->cnt - 1; i++) {
d78005a5 1369 rxbufs[i].frag = nfp_net_rx_alloc_one(dp, &rxbufs[i].dma_addr);
c0f031bc 1370 if (!rxbufs[i].frag) {
9dc6b116 1371 nfp_net_rx_ring_bufs_free(dp, rx_ring);
4c352362
JK		 1372 return -ENOMEM;
1373 }
4c352362
JK		 1374 }
1375
1376 return 0;
1377}
1378
1934680f
JK		 1379/**
1380 * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
6fe0c3b4 1381 * @dp: NFP Net data path struct
1934680f
JK		 1382 * @rx_ring: RX ring to fill
1383 */
6fe0c3b4
JK		 1384static void
1385nfp_net_rx_ring_fill_freelist(struct nfp_net_dp *dp,
1386 struct nfp_net_rx_ring *rx_ring)
1934680f
JK		 1387{
1388 unsigned int i;
1389
1390 for (i = 0; i < rx_ring->cnt - 1; i++)
6fe0c3b4 1391 nfp_net_rx_give_one(dp, rx_ring, rx_ring->rxbufs[i].frag,
1934680f
JK		 1392 rx_ring->rxbufs[i].dma_addr);
1393}
1394
4c352362
JK		 1395/**
1396 * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1397 * @flags: RX descriptor flags field in CPU byte order
1398 */
1399static int nfp_net_rx_csum_has_errors(u16 flags)
1400{
1401 u16 csum_all_checked, csum_all_ok;
1402
1403 csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1404 csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1405
1406 return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1407}
1408
1409/**
1410 * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
79c12a75 1411 * @dp: NFP Net data path struct
4c352362
JK		 1412 * @r_vec: per-ring structure
1413 * @rxd: Pointer to RX descriptor
ddb98d94 1414 * @meta: Parsed metadata prepend
4c352362
JK		 1415 * @skb: Pointer to SKB
1416 */
79c12a75
JK		 1417static void nfp_net_rx_csum(struct nfp_net_dp *dp,
1418 struct nfp_net_r_vector *r_vec,
ddb98d94
JK		 1419 struct nfp_net_rx_desc *rxd,
1420 struct nfp_meta_parsed *meta, struct sk_buff *skb)
4c352362
JK		 1421{
1422 skb_checksum_none_assert(skb);
1423
79c12a75 1424 if (!(dp->netdev->features & NETIF_F_RXCSUM))
4c352362
JK		 1425 return;
1426
ddb98d94
JK		 1427 if (meta->csum_type) {
1428 skb->ip_summed = meta->csum_type;
1429 skb->csum = meta->csum;
1430 u64_stats_update_begin(&r_vec->rx_sync);
0df57e60 1431 r_vec->hw_csum_rx_complete++;
ddb98d94
JK		 1432 u64_stats_update_end(&r_vec->rx_sync);
1433 return;
1434 }
1435
4c352362
JK		 1436 if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1437 u64_stats_update_begin(&r_vec->rx_sync);
1438 r_vec->hw_csum_rx_error++;
1439 u64_stats_update_end(&r_vec->rx_sync);
1440 return;
1441 }
1442
1443 /* Assume that the firmware will never report inner CSUM_OK unless outer
1444 * L4 headers were successfully parsed. FW will always report zero UDP
1445 * checksum as CSUM_OK.
1446 */
1447 if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1448 rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1449 __skb_incr_checksum_unnecessary(skb);
1450 u64_stats_update_begin(&r_vec->rx_sync);
1451 r_vec->hw_csum_rx_ok++;
1452 u64_stats_update_end(&r_vec->rx_sync);
1453 }
1454
1455 if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1456 rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1457 __skb_incr_checksum_unnecessary(skb);
1458 u64_stats_update_begin(&r_vec->rx_sync);
1459 r_vec->hw_csum_rx_inner_ok++;
1460 u64_stats_update_end(&r_vec->rx_sync);
1461 }
1462}
1463
e524a6a9
JK		 1464static void
1465nfp_net_set_hash(struct net_device *netdev, struct nfp_meta_parsed *meta,
1466 unsigned int type, __be32 *hash)
4c352362 1467{
19d0f54e 1468 if (!(netdev->features & NETIF_F_RXHASH))
4c352362
JK		 1469 return;
1470
19d0f54e 1471 switch (type) {
4c352362
JK		 1472 case NFP_NET_RSS_IPV4:
1473 case NFP_NET_RSS_IPV6:
1474 case NFP_NET_RSS_IPV6_EX:
e524a6a9 1475 meta->hash_type = PKT_HASH_TYPE_L3;
4c352362
JK		 1476 break;
1477 default:
e524a6a9 1478 meta->hash_type = PKT_HASH_TYPE_L4;
4c352362
JK		 1479 break;
1480 }
e524a6a9
JK		 1481
1482 meta->hash = get_unaligned_be32(hash);
4c352362
JK		 1483}
1484
19d0f54e 1485static void
e524a6a9 1486nfp_net_set_hash_desc(struct net_device *netdev, struct nfp_meta_parsed *meta,
b92fb77f 1487 void *data, struct nfp_net_rx_desc *rxd)
19d0f54e 1488{
b92fb77f 1489 struct nfp_net_rx_hash *rx_hash = data;
19d0f54e
JK		 1490
1491 if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1492 return;
1493
e524a6a9 1494 nfp_net_set_hash(netdev, meta, get_unaligned_be32(&rx_hash->hash_type),
19d0f54e
JK		 1495 &rx_hash->hash);
1496}
1497
1498static void *
e524a6a9 1499nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
b92fb77f 1500 void *data, int meta_len)
19d0f54e 1501{
19d0f54e
JK		 1502 u32 meta_info;
1503
1504 meta_info = get_unaligned_be32(data);
1505 data += 4;
1506
1507 while (meta_info) {
1508 switch (meta_info & NFP_NET_META_FIELD_MASK) {
1509 case NFP_NET_META_HASH:
1510 meta_info >>= NFP_NET_META_FIELD_SIZE;
e524a6a9 1511 nfp_net_set_hash(netdev, meta,
19d0f54e
JK		 1512 meta_info & NFP_NET_META_FIELD_MASK,
1513 (__be32 *)data);
1514 data += 4;
1515 break;
1516 case NFP_NET_META_MARK:
e524a6a9 1517 meta->mark = get_unaligned_be32(data);
19d0f54e
JK		 1518 data += 4;
1519 break;
91bf82ca
SH		 1520 case NFP_NET_META_PORTID:
1521 meta->portid = get_unaligned_be32(data);
1522 data += 4;
1523 break;
ddb98d94
JK		 1524 case NFP_NET_META_CSUM:
1525 meta->csum_type = CHECKSUM_COMPLETE;
1526 meta->csum =
1527 (__force __wsum)__get_unaligned_cpu32(data);
1528 data += 4;
1529 break;
19d0f54e
JK		 1530 default:
1531 return NULL;
1532 }
1533
1534 meta_info >>= NFP_NET_META_FIELD_SIZE;
1535 }
1536
1537 return data;
1538}
1539
e9949aeb 1540static void
6fe0c3b4
JK		 1541nfp_net_rx_drop(const struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
1542 struct nfp_net_rx_ring *rx_ring, struct nfp_net_rx_buf *rxbuf,
1543 struct sk_buff *skb)
e9949aeb
JK		 1544{
1545 u64_stats_update_begin(&r_vec->rx_sync);
1546 r_vec->rx_drops++;
16f50cda
JK		 1547 /* If we have both skb and rxbuf the replacement buffer allocation
1548 * must have failed, count this as an alloc failure.
1549 */
1550 if (skb && rxbuf)
1551 r_vec->rx_replace_buf_alloc_fail++;
e9949aeb
JK		 1552 u64_stats_update_end(&r_vec->rx_sync);
1553
c0f031bc
JK		 1554 /* skb is build based on the frag, free_skb() would free the frag
1555 * so to be able to reuse it we need an extra ref.
1556 */
1557 if (skb && rxbuf && skb->head == rxbuf->frag)
1558 page_ref_inc(virt_to_head_page(rxbuf->frag));
e9949aeb 1559 if (rxbuf)
6fe0c3b4 1560 nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag, rxbuf->dma_addr);
e9949aeb
JK		 1561 if (skb)
1562 dev_kfree_skb_any(skb);
1563}
1564
a67edbf4 1565static bool
79c12a75 1566nfp_net_tx_xdp_buf(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring,
ecd63a02 1567 struct nfp_net_tx_ring *tx_ring,
1abae319 1568 struct nfp_net_rx_buf *rxbuf, unsigned int dma_off,
abeeec4a 1569 unsigned int pkt_len, bool *completed)
ecd63a02
JK		 1570{
1571 struct nfp_net_tx_buf *txbuf;
1572 struct nfp_net_tx_desc *txd;
ecd63a02
JK		 1573 int wr_idx;
1574
1575 if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
abeeec4a
JK		 1576 if (!*completed) {
1577 nfp_net_xdp_complete(tx_ring);
1578 *completed = true;
1579 }
1580
1581 if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1582 nfp_net_rx_drop(dp, rx_ring->r_vec, rx_ring, rxbuf,
1583 NULL);
1584 return false;
1585 }
ecd63a02
JK		 1586 }
1587
4aa3b766 1588 wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
ecd63a02
JK		 1589
1590 /* Stash the soft descriptor of the head then initialize it */
1591 txbuf = &tx_ring->txbufs[wr_idx];
92e68195
JK		 1592
1593 nfp_net_rx_give_one(dp, rx_ring, txbuf->frag, txbuf->dma_addr);
1594
ecd63a02
JK		 1595 txbuf->frag = rxbuf->frag;
1596 txbuf->dma_addr = rxbuf->dma_addr;
1597 txbuf->fidx = -1;
1598 txbuf->pkt_cnt = 1;
1599 txbuf->real_len = pkt_len;
1600
1abae319 1601 dma_sync_single_for_device(dp->dev, rxbuf->dma_addr + dma_off,
d58cebb7 1602 pkt_len, DMA_BIDIRECTIONAL);
ecd63a02
JK		 1603
1604 /* Build TX descriptor */
1605 txd = &tx_ring->txds[wr_idx];
1606 txd->offset_eop = PCIE_DESC_TX_EOP;
1607 txd->dma_len = cpu_to_le16(pkt_len);
1abae319 1608 nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + dma_off);
ecd63a02
JK		 1609 txd->data_len = cpu_to_le16(pkt_len);
1610
1611 txd->flags = 0;
1612 txd->mss = 0;
e53fc9fa 1613 txd->lso_hdrlen = 0;
ecd63a02
JK		 1614
1615 tx_ring->wr_p++;
1616 tx_ring->wr_ptr_add++;
a67edbf4 1617 return true;
ecd63a02
JK		 1618}
1619
4c352362
JK		 1620/**
1621 * nfp_net_rx() - receive up to @budget packets on @rx_ring
1622 * @rx_ring: RX ring to receive from
1623 * @budget: NAPI budget
1624 *
1625 * Note, this function is separated out from the napi poll function to
1626 * more cleanly separate packet receive code from other bookkeeping
1627 * functions performed in the napi poll function.
1628 *
4c352362
JK		 1629 * Return: Number of packets received.
1630 */
1631static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1632{
1633 struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
79c12a75 1634 struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
ecd63a02
JK		 1635 struct nfp_net_tx_ring *tx_ring;
1636 struct bpf_prog *xdp_prog;
abeeec4a 1637 bool xdp_tx_cmpl = false;
ecd63a02 1638 unsigned int true_bufsz;
c0f031bc 1639 struct sk_buff *skb;
416db5c1 1640 int pkts_polled = 0;
7f1c684a 1641 struct xdp_buff xdp;
4c352362
JK		 1642 int idx;
1643
ecd63a02 1644 rcu_read_lock();
79c12a75 1645 xdp_prog = READ_ONCE(dp->xdp_prog);
79c12a75 1646 true_bufsz = xdp_prog ? PAGE_SIZE : dp->fl_bufsz;
7f1c684a 1647 xdp.rxq = &rx_ring->xdp_rxq;
ecd63a02
JK		 1648 tx_ring = r_vec->xdp_ring;
1649
416db5c1 1650 while (pkts_polled < budget) {
5cd4fbea 1651 unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
13df4c66
JK		 1652 struct nfp_net_rx_buf *rxbuf;
1653 struct nfp_net_rx_desc *rxd;
e524a6a9 1654 struct nfp_meta_parsed meta;
91bf82ca 1655 struct net_device *netdev;
13df4c66 1656 dma_addr_t new_dma_addr;
65d88fd0 1657 u32 meta_len_xdp = 0;
13df4c66
JK		 1658 void *new_frag;
1659
4aa3b766 1660 idx = D_IDX(rx_ring, rx_ring->rd_p);
4c352362
JK		 1661
1662 rxd = &rx_ring->rxds[idx];
416db5c1 1663 if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
4c352362 1664 break;
416db5c1 1665
4c352362
JK		 1666 /* Memory barrier to ensure that we won't do other reads
1667 * before the DD bit.
1668 */
1669 dma_rmb();
1670
e524a6a9
JK		 1671 memset(&meta, 0, sizeof(meta));
1672
4c352362
JK		 1673 rx_ring->rd_p++;
1674 pkts_polled++;
4c352362 1675
c0f031bc 1676 rxbuf = &rx_ring->rxbufs[idx];
180012dc
JK		 1677 /* < meta_len >
1678 * <-- [rx_offset] -->
1679 * ---------------------------------------------------------
1680 * | [XX] | metadata | packet | XXXX |
1681 * ---------------------------------------------------------
1682 * <---------------- data_len --------------->
1683 *
1684 * The rx_offset is fixed for all packets, the meta_len can vary
1685 * on a packet by packet basis. If rx_offset is set to zero
1686 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1687 * buffer and is immediately followed by the packet (no [XX]).
1688 */
4c352362
JK		 1689 meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1690 data_len = le16_to_cpu(rxd->rxd.data_len);
13df4c66 1691 pkt_len = data_len - meta_len;
4c352362 1692
5cd4fbea 1693 pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
79c12a75 1694 if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
5cd4fbea 1695 pkt_off += meta_len;
180012dc 1696 else
5cd4fbea
JK		 1697 pkt_off += dp->rx_offset;
1698 meta_off = pkt_off - meta_len;
4c352362 1699
4c352362
JK		 1700 /* Stats update */
1701 u64_stats_update_begin(&r_vec->rx_sync);
1702 r_vec->rx_pkts++;
13df4c66 1703 r_vec->rx_bytes += pkt_len;
4c352362
JK		 1704 u64_stats_update_end(&r_vec->rx_sync);
1705
b92fb77f
JK		 1706 if (unlikely(meta_len > NFP_NET_MAX_PREPEND ||
1707 (dp->rx_offset && meta_len > dp->rx_offset))) {
1708 nn_dp_warn(dp, "oversized RX packet metadata %u\n",
1709 meta_len);
6fe0c3b4 1710 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
b92fb77f
JK		 1711 continue;
1712 }
1713
5cd4fbea
JK		 1714 nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,
1715 data_len);
1716
e524a6a9
JK		 1717 if (!dp->chained_metadata_format) {
1718 nfp_net_set_hash_desc(dp->netdev, &meta,
1719 rxbuf->frag + meta_off, rxd);
1720 } else if (meta_len) {
1721 void *end;
1722
1723 end = nfp_net_parse_meta(dp->netdev, &meta,
1724 rxbuf->frag + meta_off,
1725 meta_len);
1726 if (unlikely(end != rxbuf->frag + pkt_off)) {
1727 nn_dp_warn(dp, "invalid RX packet metadata\n");
1728 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1729 NULL);
1730 continue;
1731 }
1732 }
1733
5f428401 1734 if (xdp_prog && !meta.portid) {
65d88fd0 1735 void *orig_data = rxbuf->frag + pkt_off;
1abae319 1736 unsigned int dma_off;
ecd63a02
JK		 1737 int act;
1738
65d88fd0
DB		 1739 xdp.data_hard_start = rxbuf->frag + NFP_NET_RX_BUF_HEADROOM;
1740 xdp.data = orig_data;
1741 xdp.data_meta = orig_data;
1742 xdp.data_end = orig_data + pkt_len;
1743
1744 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1745
5a6a22e3 1746 pkt_len = xdp.data_end - xdp.data;
65d88fd0 1747 pkt_off += xdp.data - orig_data;
6fe0c3b4 1748
ecd63a02
JK		 1749 switch (act) {
1750 case XDP_PASS:
65d88fd0 1751 meta_len_xdp = xdp.data - xdp.data_meta;
ecd63a02
JK		 1752 break;
1753 case XDP_TX:
5cd4fbea 1754 dma_off = pkt_off - NFP_NET_RX_BUF_HEADROOM;
79c12a75 1755 if (unlikely(!nfp_net_tx_xdp_buf(dp, rx_ring,
a67edbf4 1756 tx_ring, rxbuf,
1abae319 1757 dma_off,
abeeec4a
JK		 1758 pkt_len,
1759 &xdp_tx_cmpl)))
79c12a75
JK		 1760 trace_xdp_exception(dp->netdev,
1761 xdp_prog, act);
ecd63a02
JK		 1762 continue;
1763 default:
1764 bpf_warn_invalid_xdp_action(act);
5c914359 1765 /* fall through */
ecd63a02 1766 case XDP_ABORTED:
79c12a75 1767 trace_xdp_exception(dp->netdev, xdp_prog, act);
5c914359 1768 /* fall through */
ecd63a02 1769 case XDP_DROP:
6fe0c3b4 1770 nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
ecd63a02
JK		 1771 rxbuf->dma_addr);
1772 continue;
1773 }
1774 }
1775
78a8b760
JK		 1776 if (likely(!meta.portid)) {
1777 netdev = dp->netdev;
79ca38e8
JK		 1778 } else if (meta.portid == NFP_META_PORT_ID_CTRL) {
1779 struct nfp_net *nn = netdev_priv(dp->netdev);
1780
1781 nfp_app_ctrl_rx_raw(nn->app, rxbuf->frag + pkt_off,
1782 pkt_len);
1783 nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1784 rxbuf->dma_addr);
1785 continue;
78a8b760
JK		 1786 } else {
1787 struct nfp_net *nn;
1788
1789 nn = netdev_priv(dp->netdev);
1790 netdev = nfp_app_repr_get(nn->app, meta.portid);
1791 if (unlikely(!netdev)) {
1792 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1793 NULL);
1794 continue;
1795 }
1796 nfp_repr_inc_rx_stats(netdev, pkt_len);
1797 }
1798
ecd63a02 1799 skb = build_skb(rxbuf->frag, true_bufsz);
13df4c66 1800 if (unlikely(!skb)) {
6fe0c3b4 1801 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
13df4c66
JK		 1802 continue;
1803 }
c487e6b1 1804 new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
13df4c66 1805 if (unlikely(!new_frag)) {
6fe0c3b4 1806 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
13df4c66
JK		 1807 continue;
1808 }
1809
1691a4c0
JK		 1810 nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
1811
1812 nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
1813
5cd4fbea 1814 skb_reserve(skb, pkt_off);
13df4c66
JK		 1815 skb_put(skb, pkt_len);
1816
e524a6a9
JK		 1817 skb->mark = meta.mark;
1818 skb_set_hash(skb, meta.hash, meta.hash_type);
19d0f54e 1819
4c352362 1820 skb_record_rx_queue(skb, rx_ring->idx);
91bf82ca 1821 skb->protocol = eth_type_trans(skb, netdev);
4c352362 1822
ddb98d94 1823 nfp_net_rx_csum(dp, r_vec, rxd, &meta, skb);
4c352362
JK		 1824
1825 if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
1826 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1827 le16_to_cpu(rxd->rxd.vlan));
65d88fd0
DB		 1828 if (meta_len_xdp)
1829 skb_metadata_set(skb, meta_len_xdp);
4c352362
JK		 1830
1831 napi_gro_receive(&rx_ring->r_vec->napi, skb);
1832 }
1833
abeeec4a
JK		 1834 if (xdp_prog) {
1835 if (tx_ring->wr_ptr_add)
1836 nfp_net_tx_xmit_more_flush(tx_ring);
1837 else if (unlikely(tx_ring->wr_p != tx_ring->rd_p) &&
1838 !xdp_tx_cmpl)
1839 if (!nfp_net_xdp_complete(tx_ring))
1840 pkts_polled = budget;
1841 }
ecd63a02
JK		 1842 rcu_read_unlock();
1843
4c352362
JK		 1844 return pkts_polled;
1845}
1846
1847/**
1848 * nfp_net_poll() - napi poll function
1849 * @napi: NAPI structure
1850 * @budget: NAPI budget
1851 *
1852 * Return: number of packets polled.
1853 */
1854static int nfp_net_poll(struct napi_struct *napi, int budget)
1855{
1856 struct nfp_net_r_vector *r_vec =
1857 container_of(napi, struct nfp_net_r_vector, napi);
cbeaf7aa 1858 unsigned int pkts_polled = 0;
4c352362 1859
cbeaf7aa 1860 if (r_vec->tx_ring)
d387b8a1 1861 nfp_net_tx_complete(r_vec->tx_ring, budget);
abeeec4a 1862 if (r_vec->rx_ring)
cbeaf7aa 1863 pkts_polled = nfp_net_rx(r_vec->rx_ring, budget);
4c352362 1864
7de5f115
JK		 1865 if (pkts_polled < budget)
1866 if (napi_complete_done(napi, pkts_polled))
1867 nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
4c352362
JK		 1868
1869 return pkts_polled;
1870}
1871
77ece8d5
JK		 1872/* Control device data path
1873 */
1874
1875static bool
1876nfp_ctrl_tx_one(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1877 struct sk_buff *skb, bool old)
1878{
1879 unsigned int real_len = skb->len, meta_len = 0;
1880 struct nfp_net_tx_ring *tx_ring;
1881 struct nfp_net_tx_buf *txbuf;
1882 struct nfp_net_tx_desc *txd;
1883 struct nfp_net_dp *dp;
1884 dma_addr_t dma_addr;
1885 int wr_idx;
1886
1887 dp = &r_vec->nfp_net->dp;
1888 tx_ring = r_vec->tx_ring;
1889
1890 if (WARN_ON_ONCE(skb_shinfo(skb)->nr_frags)) {
1891 nn_dp_warn(dp, "Driver's CTRL TX does not implement gather\n");
1892 goto err_free;
1893 }
1894
1895 if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1896 u64_stats_update_begin(&r_vec->tx_sync);
1897 r_vec->tx_busy++;
1898 u64_stats_update_end(&r_vec->tx_sync);
1899 if (!old)
1900 __skb_queue_tail(&r_vec->queue, skb);
1901 else
1902 __skb_queue_head(&r_vec->queue, skb);
1903 return true;
1904 }
1905
1906 if (nfp_app_ctrl_has_meta(nn->app)) {
1907 if (unlikely(skb_headroom(skb) < 8)) {
1908 nn_dp_warn(dp, "CTRL TX on skb without headroom\n");
1909 goto err_free;
1910 }
1911 meta_len = 8;
1912 put_unaligned_be32(NFP_META_PORT_ID_CTRL, skb_push(skb, 4));
1913 put_unaligned_be32(NFP_NET_META_PORTID, skb_push(skb, 4));
1914 }
1915
1916 /* Start with the head skbuf */
1917 dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
1918 DMA_TO_DEVICE);
1919 if (dma_mapping_error(dp->dev, dma_addr))
1920 goto err_dma_warn;
1921
1922 wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1923
1924 /* Stash the soft descriptor of the head then initialize it */
1925 txbuf = &tx_ring->txbufs[wr_idx];
1926 txbuf->skb = skb;
1927 txbuf->dma_addr = dma_addr;
1928 txbuf->fidx = -1;
1929 txbuf->pkt_cnt = 1;
1930 txbuf->real_len = real_len;
1931
1932 /* Build TX descriptor */
1933 txd = &tx_ring->txds[wr_idx];
1934 txd->offset_eop = meta_len | PCIE_DESC_TX_EOP;
1935 txd->dma_len = cpu_to_le16(skb_headlen(skb));
1936 nfp_desc_set_dma_addr(txd, dma_addr);
1937 txd->data_len = cpu_to_le16(skb->len);
1938
1939 txd->flags = 0;
1940 txd->mss = 0;
1941 txd->lso_hdrlen = 0;
1942
1943 tx_ring->wr_p++;
1944 tx_ring->wr_ptr_add++;
1945 nfp_net_tx_xmit_more_flush(tx_ring);
1946
1947 return false;
1948
1949err_dma_warn:
1950 nn_dp_warn(dp, "Failed to DMA map TX CTRL buffer\n");
1951err_free:
1952 u64_stats_update_begin(&r_vec->tx_sync);
1953 r_vec->tx_errors++;
1954 u64_stats_update_end(&r_vec->tx_sync);
1955 dev_kfree_skb_any(skb);
1956 return false;
1957}
1958
d48ae231
JK		 1959bool __nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
1960{
1961 struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
1962
1963 return nfp_ctrl_tx_one(nn, r_vec, skb, false);
1964}
1965
77ece8d5
JK		 1966bool nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
1967{
1968 struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
1969 bool ret;
1970
1971 spin_lock_bh(&r_vec->lock);
1972 ret = nfp_ctrl_tx_one(nn, r_vec, skb, false);
1973 spin_unlock_bh(&r_vec->lock);
1974
1975 return ret;
1976}
1977
1978static void __nfp_ctrl_tx_queued(struct nfp_net_r_vector *r_vec)
1979{
1980 struct sk_buff *skb;
1981
1982 while ((skb = __skb_dequeue(&r_vec->queue)))
1983 if (nfp_ctrl_tx_one(r_vec->nfp_net, r_vec, skb, true))
1984 return;
1985}
1986
1987static bool
1988nfp_ctrl_meta_ok(struct nfp_net *nn, void *data, unsigned int meta_len)
1989{
1990 u32 meta_type, meta_tag;
1991
1992 if (!nfp_app_ctrl_has_meta(nn->app))
1993 return !meta_len;
1994
1995 if (meta_len != 8)
1996 return false;
1997
1998 meta_type = get_unaligned_be32(data);
1999 meta_tag = get_unaligned_be32(data + 4);
2000
2001 return (meta_type == NFP_NET_META_PORTID &&
2002 meta_tag == NFP_META_PORT_ID_CTRL);
2003}
2004
2005static bool
2006nfp_ctrl_rx_one(struct nfp_net *nn, struct nfp_net_dp *dp,
2007 struct nfp_net_r_vector *r_vec, struct nfp_net_rx_ring *rx_ring)
2008{
2009 unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
2010 struct nfp_net_rx_buf *rxbuf;
2011 struct nfp_net_rx_desc *rxd;
2012 dma_addr_t new_dma_addr;
2013 struct sk_buff *skb;
2014 void *new_frag;
2015 int idx;
2016
2017 idx = D_IDX(rx_ring, rx_ring->rd_p);
2018
2019 rxd = &rx_ring->rxds[idx];
2020 if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
2021 return false;
2022
2023 /* Memory barrier to ensure that we won't do other reads
2024 * before the DD bit.
2025 */
2026 dma_rmb();
2027
2028 rx_ring->rd_p++;
2029
2030 rxbuf = &rx_ring->rxbufs[idx];
2031 meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
2032 data_len = le16_to_cpu(rxd->rxd.data_len);
2033 pkt_len = data_len - meta_len;
2034
2035 pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
2036 if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
2037 pkt_off += meta_len;
2038 else
2039 pkt_off += dp->rx_offset;
2040 meta_off = pkt_off - meta_len;
2041
2042 /* Stats update */
2043 u64_stats_update_begin(&r_vec->rx_sync);
2044 r_vec->rx_pkts++;
2045 r_vec->rx_bytes += pkt_len;
2046 u64_stats_update_end(&r_vec->rx_sync);
2047
2048 nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off, data_len);
2049
2050 if (unlikely(!nfp_ctrl_meta_ok(nn, rxbuf->frag + meta_off, meta_len))) {
2051 nn_dp_warn(dp, "incorrect metadata for ctrl packet (%d)\n",
2052 meta_len);
2053 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2054 return true;
2055 }
2056
2057 skb = build_skb(rxbuf->frag, dp->fl_bufsz);
2058 if (unlikely(!skb)) {
2059 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2060 return true;
2061 }
2062 new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
2063 if (unlikely(!new_frag)) {
2064 nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
2065 return true;
2066 }
2067
2068 nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
2069
2070 nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
2071
2072 skb_reserve(skb, pkt_off);
2073 skb_put(skb, pkt_len);
2074
02082701 2075 nfp_app_ctrl_rx(nn->app, skb);
77ece8d5
JK		 2076
2077 return true;
2078}
2079
ff58e2df 2080static bool nfp_ctrl_rx(struct nfp_net_r_vector *r_vec)
77ece8d5
JK		 2081{
2082 struct nfp_net_rx_ring *rx_ring = r_vec->rx_ring;
2083 struct nfp_net *nn = r_vec->nfp_net;
2084 struct nfp_net_dp *dp = &nn->dp;
ff58e2df 2085 unsigned int budget = 512;
77ece8d5 2086
ff58e2df 2087 while (nfp_ctrl_rx_one(nn, dp, r_vec, rx_ring) && budget--)
77ece8d5 2088 continue;
ff58e2df
JK		 2089
2090 return budget;
77ece8d5
JK		 2091}
2092
2093static void nfp_ctrl_poll(unsigned long arg)
2094{
2095 struct nfp_net_r_vector *r_vec = (void *)arg;
2096
2097 spin_lock_bh(&r_vec->lock);
d387b8a1 2098 nfp_net_tx_complete(r_vec->tx_ring, 0);
77ece8d5
JK		 2099 __nfp_ctrl_tx_queued(r_vec);
2100 spin_unlock_bh(&r_vec->lock);
2101
ff58e2df
JK		 2102 if (nfp_ctrl_rx(r_vec)) {
2103 nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2104 } else {
2105 tasklet_schedule(&r_vec->tasklet);
2106 nn_dp_warn(&r_vec->nfp_net->dp,
2107 "control message budget exceeded!\n");
2108 }
77ece8d5
JK		 2109}
2110
4c352362
JK		 2111/* Setup and Configuration
2112 */
2113
cd083ce1
JK		 2114/**
2115 * nfp_net_vecs_init() - Assign IRQs and setup rvecs.
2116 * @nn: NFP Network structure
2117 */
2118static void nfp_net_vecs_init(struct nfp_net *nn)
2119{
2120 struct nfp_net_r_vector *r_vec;
2121 int r;
2122
2123 nn->lsc_handler = nfp_net_irq_lsc;
2124 nn->exn_handler = nfp_net_irq_exn;
2125
2126 for (r = 0; r < nn->max_r_vecs; r++) {
2127 struct msix_entry *entry;
2128
2129 entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r];
2130
2131 r_vec = &nn->r_vecs[r];
2132 r_vec->nfp_net = nn;
cd083ce1
JK		 2133 r_vec->irq_entry = entry->entry;
2134 r_vec->irq_vector = entry->vector;
2135
77ece8d5
JK		 2136 if (nn->dp.netdev) {
2137 r_vec->handler = nfp_net_irq_rxtx;
2138 } else {
2139 r_vec->handler = nfp_ctrl_irq_rxtx;
2140
2141 __skb_queue_head_init(&r_vec->queue);
2142 spin_lock_init(&r_vec->lock);
2143 tasklet_init(&r_vec->tasklet, nfp_ctrl_poll,
2144 (unsigned long)r_vec);
2145 tasklet_disable(&r_vec->tasklet);
2146 }
2147
cd083ce1
JK		 2148 cpumask_set_cpu(r, &r_vec->affinity_mask);
2149 }
2150}
2151
4c352362
JK		 2152/**
2153 * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
2154 * @tx_ring: TX ring to free
2155 */
2156static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
2157{
2158 struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
79c12a75 2159 struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
4c352362 2160
46627170 2161 kvfree(tx_ring->txbufs);
4c352362
JK		 2162
2163 if (tx_ring->txds)
79c12a75 2164 dma_free_coherent(dp->dev, tx_ring->size,
4c352362
JK		 2165 tx_ring->txds, tx_ring->dma);
2166
2167 tx_ring->cnt = 0;
4c352362
JK		 2168 tx_ring->txbufs = NULL;
2169 tx_ring->txds = NULL;
2170 tx_ring->dma = 0;
2171 tx_ring->size = 0;
2172}
2173
2174/**
2175 * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
892a7f70 2176 * @dp: NFP Net data path struct
4c352362
JK		 2177 * @tx_ring: TX Ring structure to allocate
2178 *
2179 * Return: 0 on success, negative errno otherwise.
2180 */
ecd63a02 2181static int
92e68195 2182nfp_net_tx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
4c352362
JK		 2183{
2184 struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
4c352362 2185
892a7f70 2186 tx_ring->cnt = dp->txd_cnt;
4c352362 2187
5ea14712 2188 tx_ring->size = array_size(tx_ring->cnt, sizeof(*tx_ring->txds));
79c12a75 2189 tx_ring->txds = dma_zalloc_coherent(dp->dev, tx_ring->size,
4c352362
JK		 2190 &tx_ring->dma, GFP_KERNEL);
2191 if (!tx_ring->txds)
2192 goto err_alloc;
2193
46627170
JK		 2194 tx_ring->txbufs = kvcalloc(tx_ring->cnt, sizeof(*tx_ring->txbufs),
2195 GFP_KERNEL);
4c352362
JK		 2196 if (!tx_ring->txbufs)
2197 goto err_alloc;
2198
a7b1ad08 2199 if (!tx_ring->is_xdp && dp->netdev)
79c12a75 2200 netif_set_xps_queue(dp->netdev, &r_vec->affinity_mask,
ecd63a02 2201 tx_ring->idx);
4c352362 2202
4c352362
JK		 2203 return 0;
2204
2205err_alloc:
2206 nfp_net_tx_ring_free(tx_ring);
2207 return -ENOMEM;
2208}
2209
92e68195
JK		 2210static void
2211nfp_net_tx_ring_bufs_free(struct nfp_net_dp *dp,
2212 struct nfp_net_tx_ring *tx_ring)
2213{
2214 unsigned int i;
2215
2216 if (!tx_ring->is_xdp)
2217 return;
2218
2219 for (i = 0; i < tx_ring->cnt; i++) {
2220 if (!tx_ring->txbufs[i].frag)
2221 return;
2222
2223 nfp_net_dma_unmap_rx(dp, tx_ring->txbufs[i].dma_addr);
2224 __free_page(virt_to_page(tx_ring->txbufs[i].frag));
2225 }
2226}
2227
2228static int
2229nfp_net_tx_ring_bufs_alloc(struct nfp_net_dp *dp,
2230 struct nfp_net_tx_ring *tx_ring)
2231{
2232 struct nfp_net_tx_buf *txbufs = tx_ring->txbufs;
2233 unsigned int i;
2234
2235 if (!tx_ring->is_xdp)
2236 return 0;
2237
2238 for (i = 0; i < tx_ring->cnt; i++) {
2239 txbufs[i].frag = nfp_net_rx_alloc_one(dp, &txbufs[i].dma_addr);
2240 if (!txbufs[i].frag) {
2241 nfp_net_tx_ring_bufs_free(dp, tx_ring);
2242 return -ENOMEM;
2243 }
2244 }
2245
2246 return 0;
2247}
2248
892a7f70 2249static int nfp_net_tx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
cc7c0333 2250{
cc7c0333
JK		 2251 unsigned int r;
2252
892a7f70
JK		 2253 dp->tx_rings = kcalloc(dp->num_tx_rings, sizeof(*dp->tx_rings),
2254 GFP_KERNEL);
2255 if (!dp->tx_rings)
2256 return -ENOMEM;
cc7c0333 2257
892a7f70 2258 for (r = 0; r < dp->num_tx_rings; r++) {
ecd63a02
JK		 2259 int bias = 0;
2260
512e94dc
JK		 2261 if (r >= dp->num_stack_tx_rings)
2262 bias = dp->num_stack_tx_rings;
cc7c0333 2263
892a7f70 2264 nfp_net_tx_ring_init(&dp->tx_rings[r], &nn->r_vecs[r - bias],
92e68195 2265 r, bias);
ecd63a02 2266
92e68195 2267 if (nfp_net_tx_ring_alloc(dp, &dp->tx_rings[r]))
cc7c0333 2268 goto err_free_prev;
92e68195
JK		 2269
2270 if (nfp_net_tx_ring_bufs_alloc(dp, &dp->tx_rings[r]))
2271 goto err_free_ring;
cc7c0333
JK		 2272 }
2273
892a7f70 2274 return 0;
cc7c0333
JK		 2275
2276err_free_prev:
92e68195
JK		 2277 while (r--) {
2278 nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2279err_free_ring:
892a7f70 2280 nfp_net_tx_ring_free(&dp->tx_rings[r]);
92e68195 2281 }
892a7f70
JK		 2282 kfree(dp->tx_rings);
2283 return -ENOMEM;
cc7c0333
JK		 2284}
2285
892a7f70 2286static void nfp_net_tx_rings_free(struct nfp_net_dp *dp)
cc7c0333
JK		 2287{
2288 unsigned int r;
2289
92e68195
JK		 2290 for (r = 0; r < dp->num_tx_rings; r++) {
2291 nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
892a7f70 2292 nfp_net_tx_ring_free(&dp->tx_rings[r]);
92e68195 2293 }
cc7c0333 2294
892a7f70 2295 kfree(dp->tx_rings);
cc7c0333
JK		 2296}
2297
4c352362
JK		 2298/**
2299 * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
2300 * @rx_ring: RX ring to free
2301 */
2302static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
2303{
2304 struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
79c12a75 2305 struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
4c352362 2306
a9c324be
JK		 2307 if (dp->netdev)
2308 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
46627170 2309 kvfree(rx_ring->rxbufs);
4c352362
JK		 2310
2311 if (rx_ring->rxds)
79c12a75 2312 dma_free_coherent(dp->dev, rx_ring->size,
4c352362
JK		 2313 rx_ring->rxds, rx_ring->dma);
2314
2315 rx_ring->cnt = 0;
4c352362
JK		 2316 rx_ring->rxbufs = NULL;
2317 rx_ring->rxds = NULL;
2318 rx_ring->dma = 0;
2319 rx_ring->size = 0;
2320}
2321
2322/**
2323 * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
2195c263 2324 * @dp: NFP Net data path struct
4c352362
JK		 2325 * @rx_ring: RX ring to allocate
2326 *
2327 * Return: 0 on success, negative errno otherwise.
2328 */
30d21171 2329static int
892a7f70 2330nfp_net_rx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring)
4c352362 2331{
46627170 2332 int err;
7f1c684a 2333
a9c324be
JK		 2334 if (dp->netdev) {
2335 err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, dp->netdev,
2336 rx_ring->idx);
2337 if (err < 0)
2338 return err;
2339 }
4c352362 2340
892a7f70 2341 rx_ring->cnt = dp->rxd_cnt;
5ea14712 2342 rx_ring->size = array_size(rx_ring->cnt, sizeof(*rx_ring->rxds));
79c12a75 2343 rx_ring->rxds = dma_zalloc_coherent(dp->dev, rx_ring->size,
4c352362
JK		 2344 &rx_ring->dma, GFP_KERNEL);
2345 if (!rx_ring->rxds)
2346 goto err_alloc;
2347
46627170
JK		 2348 rx_ring->rxbufs = kvcalloc(rx_ring->cnt, sizeof(*rx_ring->rxbufs),
2349 GFP_KERNEL);
4c352362
JK		 2350 if (!rx_ring->rxbufs)
2351 goto err_alloc;
2352
4c352362
JK		 2353 return 0;
2354
2355err_alloc:
2356 nfp_net_rx_ring_free(rx_ring);
2357 return -ENOMEM;
2358}
2359
892a7f70 2360static int nfp_net_rx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
36a857e4 2361{
36a857e4
JK		 2362 unsigned int r;
2363
892a7f70
JK		 2364 dp->rx_rings = kcalloc(dp->num_rx_rings, sizeof(*dp->rx_rings),
2365 GFP_KERNEL);
2366 if (!dp->rx_rings)
2367 return -ENOMEM;
36a857e4 2368
892a7f70
JK		 2369 for (r = 0; r < dp->num_rx_rings; r++) {
2370 nfp_net_rx_ring_init(&dp->rx_rings[r], &nn->r_vecs[r], r);
36a857e4 2371
892a7f70 2372 if (nfp_net_rx_ring_alloc(dp, &dp->rx_rings[r]))
36a857e4
JK		 2373 goto err_free_prev;
2374
892a7f70 2375 if (nfp_net_rx_ring_bufs_alloc(dp, &dp->rx_rings[r]))
36a857e4
JK		 2376 goto err_free_ring;
2377 }
2378
892a7f70 2379 return 0;
36a857e4
JK		 2380
2381err_free_prev:
2382 while (r--) {
892a7f70 2383 nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
36a857e4 2384err_free_ring:
892a7f70 2385 nfp_net_rx_ring_free(&dp->rx_rings[r]);
36a857e4 2386 }
892a7f70
JK		 2387 kfree(dp->rx_rings);
2388 return -ENOMEM;
36a857e4
JK		 2389}
2390
892a7f70 2391static void nfp_net_rx_rings_free(struct nfp_net_dp *dp)
36a857e4
JK		 2392{
2393 unsigned int r;
2394
892a7f70
JK		 2395 for (r = 0; r < dp->num_rx_rings; r++) {
2396 nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2397 nfp_net_rx_ring_free(&dp->rx_rings[r]);
36a857e4
JK		 2398 }
2399
892a7f70 2400 kfree(dp->rx_rings);
36a857e4
JK		 2401}
2402
e31230f9 2403static void
79c12a75
JK		 2404nfp_net_vector_assign_rings(struct nfp_net_dp *dp,
2405 struct nfp_net_r_vector *r_vec, int idx)
e31230f9 2406{
79c12a75 2407 r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL;
ecd63a02 2408 r_vec->tx_ring =
79c12a75 2409 idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL;
ecd63a02 2410
79c12a75
JK		 2411 r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ?
2412 &dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL;
e31230f9
JK		 2413}
2414
0afbfb18
JK		 2415static int
2416nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2417 int idx)
4c352362 2418{
0afbfb18 2419 int err;
4c352362 2420
164d1e9e 2421 /* Setup NAPI */
77ece8d5
JK		 2422 if (nn->dp.netdev)
2423 netif_napi_add(nn->dp.netdev, &r_vec->napi,
2424 nfp_net_poll, NAPI_POLL_WEIGHT);
2425 else
2426 tasklet_enable(&r_vec->tasklet);
164d1e9e 2427
0afbfb18 2428 snprintf(r_vec->name, sizeof(r_vec->name),
77ece8d5 2429 "%s-rxtx-%d", nfp_net_name(nn), idx);
fdace6c2
JK		 2430 err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name,
2431 r_vec);
0afbfb18 2432 if (err) {
77ece8d5
JK		 2433 if (nn->dp.netdev)
2434 netif_napi_del(&r_vec->napi);
2435 else
2436 tasklet_disable(&r_vec->tasklet);
2437
fdace6c2 2438 nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector);
0afbfb18
JK		 2439 return err;
2440 }
fdace6c2 2441 disable_irq(r_vec->irq_vector);
4c352362 2442
fdace6c2 2443 irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask);
4c352362 2444
fdace6c2
JK		 2445 nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector,
2446 r_vec->irq_entry);
4c352362 2447
0afbfb18 2448 return 0;
4c352362
JK		 2449}
2450
0afbfb18
JK		 2451static void
2452nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
4c352362 2453{
fdace6c2 2454 irq_set_affinity_hint(r_vec->irq_vector, NULL);
77ece8d5
JK		 2455 if (nn->dp.netdev)
2456 netif_napi_del(&r_vec->napi);
2457 else
2458 tasklet_disable(&r_vec->tasklet);
2459
fdace6c2 2460 free_irq(r_vec->irq_vector, r_vec);
4c352362
JK		 2461}
2462
2463/**
2464 * nfp_net_rss_write_itbl() - Write RSS indirection table to device
2465 * @nn: NFP Net device to reconfigure
2466 */
2467void nfp_net_rss_write_itbl(struct nfp_net *nn)
2468{
2469 int i;
2470
2471 for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
2472 nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
2473 get_unaligned_le32(nn->rss_itbl + i));
2474}
2475
2476/**
2477 * nfp_net_rss_write_key() - Write RSS hash key to device
2478 * @nn: NFP Net device to reconfigure
2479 */
2480void nfp_net_rss_write_key(struct nfp_net *nn)
2481{
2482 int i;
2483
9ff304bf 2484 for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4)
4c352362
JK		 2485 nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
2486 get_unaligned_le32(nn->rss_key + i));
2487}
2488
2489/**
2490 * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
2491 * @nn: NFP Net device to reconfigure
2492 */
2493void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
2494{
2495 u8 i;
2496 u32 factor;
2497 u32 value;
2498
2499 /* Compute factor used to convert coalesce '_usecs' parameters to
2500 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp
2501 * count.
2502 */
ce991ab6 2503 factor = nn->tlv_caps.me_freq_mhz / 16;
4c352362
JK		 2504
2505 /* copy RX interrupt coalesce parameters */
2506 value = (nn->rx_coalesce_max_frames << 16) |
2507 (factor * nn->rx_coalesce_usecs);
79c12a75 2508 for (i = 0; i < nn->dp.num_rx_rings; i++)
4c352362
JK		 2509 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
2510
2511 /* copy TX interrupt coalesce parameters */
2512 value = (nn->tx_coalesce_max_frames << 16) |
2513 (factor * nn->tx_coalesce_usecs);
79c12a75 2514 for (i = 0; i < nn->dp.num_tx_rings; i++)
4c352362
JK		 2515 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
2516}
2517
2518/**
f642963b 2519 * nfp_net_write_mac_addr() - Write mac address to the device control BAR
4c352362 2520 * @nn: NFP Net device to reconfigure
9d372759 2521 * @addr: MAC address to write
4c352362 2522 *
f642963b
JK		 2523 * Writes the MAC address from the netdev to the device control BAR. Does not
2524 * perform the required reconfig. We do a bit of byte swapping dance because
2525 * firmware is LE.
4c352362 2526 */
9d372759 2527static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr)
4c352362 2528{
9d372759
PC		 2529 nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr));
2530 nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4));
4c352362
JK		 2531}
2532
ca40feab
JK		 2533static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
2534{
2535 nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
2536 nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
2537 nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
2538
2539 nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
2540 nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
2541 nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
2542}
2543
4c352362
JK		 2544/**
2545 * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
2546 * @nn: NFP Net device to reconfigure
07300f77
JK		 2547 *
2548 * Warning: must be fully idempotent.
4c352362
JK		 2549 */
2550static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
2551{
2552 u32 new_ctrl, update;
ca40feab 2553 unsigned int r;
4c352362
JK		 2554 int err;
2555
79c12a75 2556 new_ctrl = nn->dp.ctrl;
4c352362
JK		 2557 new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
2558 update = NFP_NET_CFG_UPDATE_GEN;
2559 update |= NFP_NET_CFG_UPDATE_MSIX;
2560 update |= NFP_NET_CFG_UPDATE_RING;
2561
2562 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2563 new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
2564
2565 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2566 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2567
2568 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2569 err = nfp_net_reconfig(nn, update);
aba52df8 2570 if (err)
4c352362 2571 nn_err(nn, "Could not disable device: %d\n", err);
4c352362 2572
79c12a75
JK		 2573 for (r = 0; r < nn->dp.num_rx_rings; r++)
2574 nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]);
2575 for (r = 0; r < nn->dp.num_tx_rings; r++)
2576 nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]);
2577 for (r = 0; r < nn->dp.num_r_vecs; r++)
ca40feab
JK		 2578 nfp_net_vec_clear_ring_data(nn, r);
2579
79c12a75 2580 nn->dp.ctrl = new_ctrl;
4c352362
JK		 2581}
2582
ca40feab 2583static void
cbeaf7aa
JK		 2584nfp_net_rx_ring_hw_cfg_write(struct nfp_net *nn,
2585 struct nfp_net_rx_ring *rx_ring, unsigned int idx)
ca40feab
JK		 2586{
2587 /* Write the DMA address, size and MSI-X info to the device */
cbeaf7aa
JK		 2588 nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), rx_ring->dma);
2589 nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(rx_ring->cnt));
fdace6c2 2590 nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), rx_ring->r_vec->irq_entry);
cbeaf7aa 2591}
ca40feab 2592
cbeaf7aa
JK		 2593static void
2594nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
2595 struct nfp_net_tx_ring *tx_ring, unsigned int idx)
2596{
2597 nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), tx_ring->dma);
2598 nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(tx_ring->cnt));
fdace6c2 2599 nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_entry);
ca40feab
JK		 2600}
2601
ac0488ef
JK		 2602/**
2603 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2604 * @nn: NFP Net device to reconfigure
2605 */
2606static int nfp_net_set_config_and_enable(struct nfp_net *nn)
1cd0cfc4 2607{
ee200a73 2608 u32 bufsz, new_ctrl, update = 0;
1cd0cfc4
JK		 2609 unsigned int r;
2610 int err;
2611
79c12a75 2612 new_ctrl = nn->dp.ctrl;
1cd0cfc4 2613
611bdd49 2614 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) {
1cd0cfc4
JK		 2615 nfp_net_rss_write_key(nn);
2616 nfp_net_rss_write_itbl(nn);
2617 nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
2618 update |= NFP_NET_CFG_UPDATE_RSS;
2619 }
2620
ad50451e 2621 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) {
1cd0cfc4 2622 nfp_net_coalesce_write_cfg(nn);
1cd0cfc4
JK		 2623 update |= NFP_NET_CFG_UPDATE_IRQMOD;
2624 }
2625
79c12a75
JK		 2626 for (r = 0; r < nn->dp.num_tx_rings; r++)
2627 nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r);
2628 for (r = 0; r < nn->dp.num_rx_rings; r++)
2629 nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r);
1cd0cfc4 2630
79c12a75
JK		 2631 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->dp.num_tx_rings == 64 ?
2632 0xffffffffffffffffULL : ((u64)1 << nn->dp.num_tx_rings) - 1);
1cd0cfc4 2633
79c12a75
JK		 2634 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->dp.num_rx_rings == 64 ?
2635 0xffffffffffffffffULL : ((u64)1 << nn->dp.num_rx_rings) - 1);
1cd0cfc4 2636
5c0dbe9e
JK		 2637 if (nn->dp.netdev)
2638 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
1cd0cfc4 2639
5c0dbe9e 2640 nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu);
ee200a73
JK		 2641
2642 bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA;
2643 nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz);
1cd0cfc4
JK		 2644
2645 /* Enable device */
2646 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2647 update |= NFP_NET_CFG_UPDATE_GEN;
2648 update |= NFP_NET_CFG_UPDATE_MSIX;
2649 update |= NFP_NET_CFG_UPDATE_RING;
2650 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2651 new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2652
2653 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2654 err = nfp_net_reconfig(nn, update);
ac0488ef
JK		 2655 if (err) {
2656 nfp_net_clear_config_and_disable(nn);
2657 return err;
2658 }
1cd0cfc4 2659
79c12a75 2660 nn->dp.ctrl = new_ctrl;
1cd0cfc4 2661
79c12a75 2662 for (r = 0; r < nn->dp.num_rx_rings; r++)
6fe0c3b4 2663 nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]);
aba52df8 2664
1cd0cfc4
JK		 2665 /* Since reconfiguration requests while NFP is down are ignored we
2666 * have to wipe the entire VXLAN configuration and reinitialize it.
2667 */
79c12a75 2668 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN) {
1cd0cfc4
JK		 2669 memset(&nn->vxlan_ports, 0, sizeof(nn->vxlan_ports));
2670 memset(&nn->vxlan_usecnt, 0, sizeof(nn->vxlan_usecnt));
79c12a75 2671 udp_tunnel_get_rx_info(nn->dp.netdev);
1cd0cfc4
JK		 2672 }
2673
ac0488ef 2674 return 0;
1cd0cfc4
JK		 2675}
2676
d00ca2f3
JK		 2677/**
2678 * nfp_net_close_stack() - Quiesce the stack (part of close)
2679 * @nn: NFP Net device to reconfigure
2680 */
2681static void nfp_net_close_stack(struct nfp_net *nn)
2682{
2683 unsigned int r;
2684
2685 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2686 netif_carrier_off(nn->dp.netdev);
2687 nn->link_up = false;
2688
2689 for (r = 0; r < nn->dp.num_r_vecs; r++) {
2690 disable_irq(nn->r_vecs[r].irq_vector);
2691 napi_disable(&nn->r_vecs[r].napi);
2692 }
2693
2694 netif_tx_disable(nn->dp.netdev);
2695}
2696
2697/**
2698 * nfp_net_close_free_all() - Free all runtime resources
2699 * @nn: NFP Net device to reconfigure
2700 */
2701static void nfp_net_close_free_all(struct nfp_net *nn)
2702{
2703 unsigned int r;
2704
4621199d
JK		 2705 nfp_net_tx_rings_free(&nn->dp);
2706 nfp_net_rx_rings_free(&nn->dp);
2707
d00ca2f3
JK		 2708 for (r = 0; r < nn->dp.num_r_vecs; r++)
2709 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2710
d00ca2f3
JK		 2711 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2712 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2713}
2714
2715/**
2716 * nfp_net_netdev_close() - Called when the device is downed
2717 * @netdev: netdev structure
2718 */
2719static int nfp_net_netdev_close(struct net_device *netdev)
2720{
2721 struct nfp_net *nn = netdev_priv(netdev);
2722
2723 /* Step 1: Disable RX and TX rings from the Linux kernel perspective
2724 */
2725 nfp_net_close_stack(nn);
2726
2727 /* Step 2: Tell NFP
2728 */
2729 nfp_net_clear_config_and_disable(nn);
447e9ebf 2730 nfp_port_configure(netdev, false);
d00ca2f3
JK		 2731
2732 /* Step 3: Free resources
2733 */
2734 nfp_net_close_free_all(nn);
2735
2736 nn_dbg(nn, "%s down", netdev->name);
2737 return 0;
2738}
2739
77ece8d5
JK		 2740void nfp_ctrl_close(struct nfp_net *nn)
2741{
2742 int r;
2743
2744 rtnl_lock();
2745
2746 for (r = 0; r < nn->dp.num_r_vecs; r++) {
2747 disable_irq(nn->r_vecs[r].irq_vector);
2748 tasklet_disable(&nn->r_vecs[r].tasklet);
2749 }
2750
2751 nfp_net_clear_config_and_disable(nn);
2752
2753 nfp_net_close_free_all(nn);
2754
2755 rtnl_unlock();
2756}
2757
1cd0cfc4
JK		 2758/**
2759 * nfp_net_open_stack() - Start the device from stack's perspective
2760 * @nn: NFP Net device to reconfigure
2761 */
2762static void nfp_net_open_stack(struct nfp_net *nn)
2763{
2764 unsigned int r;
2765
79c12a75 2766 for (r = 0; r < nn->dp.num_r_vecs; r++) {
aba52df8 2767 napi_enable(&nn->r_vecs[r].napi);
fdace6c2 2768 enable_irq(nn->r_vecs[r].irq_vector);
aba52df8 2769 }
1cd0cfc4 2770
79c12a75 2771 netif_tx_wake_all_queues(nn->dp.netdev);
1cd0cfc4 2772
ce449ba7 2773 enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
1cd0cfc4
JK		 2774 nfp_net_read_link_status(nn);
2775}
2776
ee26756d 2777static int nfp_net_open_alloc_all(struct nfp_net *nn)
4c352362 2778{
4c352362 2779 int err, r;
4c352362 2780
4c352362
JK		 2781 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2782 nn->exn_name, sizeof(nn->exn_name),
2783 NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2784 if (err)
2785 return err;
0ba40af9
JK		 2786 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2787 nn->lsc_name, sizeof(nn->lsc_name),
2788 NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2789 if (err)
2790 goto err_free_exn;
ce449ba7 2791 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
4c352362 2792
79c12a75 2793 for (r = 0; r < nn->dp.num_r_vecs; r++) {
0afbfb18
JK		 2794 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2795 if (err)
cbeaf7aa
JK		 2796 goto err_cleanup_vec_p;
2797 }
a10b563d 2798
892a7f70
JK		 2799 err = nfp_net_rx_rings_prepare(nn, &nn->dp);
2800 if (err)
a10b563d 2801 goto err_cleanup_vec;
114bdef0 2802
892a7f70
JK		 2803 err = nfp_net_tx_rings_prepare(nn, &nn->dp);
2804 if (err)
a10b563d 2805 goto err_free_rx_rings;
4c352362 2806
e31230f9 2807 for (r = 0; r < nn->max_r_vecs; r++)
79c12a75 2808 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
e31230f9 2809
ee26756d
JK		 2810 return 0;
2811
2812err_free_rx_rings:
2813 nfp_net_rx_rings_free(&nn->dp);
2814err_cleanup_vec:
2815 r = nn->dp.num_r_vecs;
2816err_cleanup_vec_p:
2817 while (r--)
2818 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2819 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2820err_free_exn:
2821 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2822 return err;
2823}
2824
2825static int nfp_net_netdev_open(struct net_device *netdev)
2826{
2827 struct nfp_net *nn = netdev_priv(netdev);
2828 int err;
2829
2830 /* Step 1: Allocate resources for rings and the like
2831 * - Request interrupts
2832 * - Allocate RX and TX ring resources
2833 * - Setup initial RSS table
2834 */
2835 err = nfp_net_open_alloc_all(nn);
2836 if (err)
2837 return err;
2838
79c12a75 2839 err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
4c352362 2840 if (err)
ee26756d 2841 goto err_free_all;
4c352362 2842
79c12a75 2843 err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
4c352362 2844 if (err)
ee26756d 2845 goto err_free_all;
4c352362 2846
4c352362 2847 /* Step 2: Configure the NFP
447e9ebf 2848 * - Ifup the physical interface if it exists
4c352362
JK		 2849 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2850 * - Write MAC address (in case it changed)
2851 * - Set the MTU
2852 * - Set the Freelist buffer size
2853 * - Enable the FW
2854 */
447e9ebf 2855 err = nfp_port_configure(netdev, true);
4c352362 2856 if (err)
ee26756d 2857 goto err_free_all;
4c352362 2858
447e9ebf
DM		 2859 err = nfp_net_set_config_and_enable(nn);
2860 if (err)
2861 goto err_port_disable;
2862
4c352362
JK		 2863 /* Step 3: Enable for kernel
2864 * - put some freelist descriptors on each RX ring
2865 * - enable NAPI on each ring
2866 * - enable all TX queues
2867 * - set link state
2868 */
1cd0cfc4 2869 nfp_net_open_stack(nn);
4c352362
JK		 2870
2871 return 0;
2872
447e9ebf
DM		 2873err_port_disable:
2874 nfp_port_configure(netdev, false);
ee26756d
JK		 2875err_free_all:
2876 nfp_net_close_free_all(nn);
4c352362
JK		 2877 return err;
2878}
2879
77ece8d5
JK		 2880int nfp_ctrl_open(struct nfp_net *nn)
2881{
2882 int err, r;
2883
2884 /* ring dumping depends on vNICs being opened/closed under rtnl */
2885 rtnl_lock();
2886
2887 err = nfp_net_open_alloc_all(nn);
2888 if (err)
2889 goto err_unlock;
2890
2891 err = nfp_net_set_config_and_enable(nn);
2892 if (err)
2893 goto err_free_all;
2894
2895 for (r = 0; r < nn->dp.num_r_vecs; r++)
2896 enable_irq(nn->r_vecs[r].irq_vector);
2897
2898 rtnl_unlock();
2899
2900 return 0;
2901
2902err_free_all:
2903 nfp_net_close_free_all(nn);
2904err_unlock:
2905 rtnl_unlock();
2906 return err;
2907}
2908
4c352362
JK		 2909static void nfp_net_set_rx_mode(struct net_device *netdev)
2910{
2911 struct nfp_net *nn = netdev_priv(netdev);
2912 u32 new_ctrl;
2913
79c12a75 2914 new_ctrl = nn->dp.ctrl;
4c352362 2915
d0adb51e
JK		 2916 if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI)
2917 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC;
2918 else
2919 new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC;
2920
4c352362
JK		 2921 if (netdev->flags & IFF_PROMISC) {
2922 if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
2923 new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
2924 else
2925 nn_warn(nn, "FW does not support promiscuous mode\n");
2926 } else {
2927 new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
2928 }
2929
79c12a75 2930 if (new_ctrl == nn->dp.ctrl)
4c352362
JK		 2931 return;
2932
2933 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3d780b92 2934 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
4c352362 2935
79c12a75 2936 nn->dp.ctrl = new_ctrl;
4c352362
JK		 2937}
2938
1e9e10d0
JK		 2939static void nfp_net_rss_init_itbl(struct nfp_net *nn)
2940{
2941 int i;
2942
2943 for (i = 0; i < sizeof(nn->rss_itbl); i++)
2944 nn->rss_itbl[i] =
79c12a75 2945 ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
1e9e10d0
JK		 2946}
2947
512e94dc
JK		 2948static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
2949{
2950 struct nfp_net_dp new_dp = *dp;
2951
2952 *dp = nn->dp;
2953 nn->dp = new_dp;
76e1e1a8
JK		 2954
2955 nn->dp.netdev->mtu = new_dp.mtu;
892a7f70
JK		 2956
2957 if (!netif_is_rxfh_configured(nn->dp.netdev))
2958 nfp_net_rss_init_itbl(nn);
512e94dc
JK		 2959}
2960
892a7f70 2961static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
4c352362 2962{
e31230f9 2963 unsigned int r;
164d1e9e 2964 int err;
e31230f9 2965
892a7f70 2966 nfp_net_dp_swap(nn, dp);
164d1e9e 2967
e31230f9 2968 for (r = 0; r < nn->max_r_vecs; r++)
79c12a75 2969 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
e31230f9 2970
79c12a75 2971 err = netif_set_real_num_rx_queues(nn->dp.netdev, nn->dp.num_rx_rings);
08348995
AB		 2972 if (err)
2973 return err;
164d1e9e 2974
79c12a75
JK		 2975 if (nn->dp.netdev->real_num_tx_queues != nn->dp.num_stack_tx_rings) {
2976 err = netif_set_real_num_tx_queues(nn->dp.netdev,
2977 nn->dp.num_stack_tx_rings);
164d1e9e
JK		 2978 if (err)
2979 return err;
2980 }
2981
ac0488ef 2982 return nfp_net_set_config_and_enable(nn);
68453c7a 2983}
36a857e4 2984
783496b0
JK		 2985struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
2986{
2987 struct nfp_net_dp *new;
2988
2989 new = kmalloc(sizeof(*new), GFP_KERNEL);
2990 if (!new)
2991 return NULL;
2992
2993 *new = nn->dp;
2994
2995 /* Clear things which need to be recomputed */
2996 new->fl_bufsz = 0;
2997 new->tx_rings = NULL;
2998 new->rx_rings = NULL;
2999 new->num_r_vecs = 0;
3000 new->num_stack_tx_rings = 0;
3001
3002 return new;
3003}
3004
d957c0f7
JK		 3005static int
3006nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
3007 struct netlink_ext_ack *extack)
ecd63a02
JK		 3008{
3009 /* XDP-enabled tests */
9dc6b116 3010 if (!dp->xdp_prog)
ecd63a02 3011 return 0;
2195c263 3012 if (dp->fl_bufsz > PAGE_SIZE) {
4d463c4d 3013 NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
ecd63a02
JK		 3014 return -EINVAL;
3015 }
892a7f70 3016 if (dp->num_tx_rings > nn->max_tx_rings) {
4d463c4d 3017 NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
ecd63a02
JK		 3018 return -EINVAL;
3019 }
3020
3021 return 0;
3022}
3023
d957c0f7
JK		 3024int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
3025 struct netlink_ext_ack *extack)
cc7c0333 3026{
512e94dc 3027 int r, err;
cc7c0333 3028
76e1e1a8 3029 dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
2195c263 3030
892a7f70 3031 dp->num_stack_tx_rings = dp->num_tx_rings;
9dc6b116 3032 if (dp->xdp_prog)
892a7f70 3033 dp->num_stack_tx_rings -= dp->num_rx_rings;
ecd63a02 3034
892a7f70 3035 dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
ecd63a02 3036
d957c0f7 3037 err = nfp_net_check_config(nn, dp, extack);
ecd63a02 3038 if (err)
783496b0 3039 goto exit_free_dp;
164d1e9e 3040
783496b0 3041 if (!netif_running(dp->netdev)) {
892a7f70 3042 nfp_net_dp_swap(nn, dp);
783496b0
JK		 3043 err = 0;
3044 goto exit_free_dp;
cc7c0333
JK		 3045 }
3046
cc7c0333 3047 /* Prepare new rings */
512e94dc 3048 for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
164d1e9e
JK		 3049 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
3050 if (err) {
512e94dc 3051 dp->num_r_vecs = r;
164d1e9e
JK		 3052 goto err_cleanup_vecs;
3053 }
3054 }
892a7f70
JK		 3055
3056 err = nfp_net_rx_rings_prepare(nn, dp);
3057 if (err)
3058 goto err_cleanup_vecs;
3059
3060 err = nfp_net_tx_rings_prepare(nn, dp);
3061 if (err)
3062 goto err_free_rx;
cc7c0333
JK		 3063
3064 /* Stop device, swap in new rings, try to start the firmware */
3065 nfp_net_close_stack(nn);
3066 nfp_net_clear_config_and_disable(nn);
3067
892a7f70 3068 err = nfp_net_dp_swap_enable(nn, dp);
cc7c0333 3069 if (err) {
68453c7a 3070 int err2;
cc7c0333 3071
68453c7a 3072 nfp_net_clear_config_and_disable(nn);
cc7c0333 3073
68453c7a 3074 /* Try with old configuration and old rings */
892a7f70 3075 err2 = nfp_net_dp_swap_enable(nn, dp);
68453c7a 3076 if (err2)
cc7c0333 3077 nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
68453c7a 3078 err, err2);
cc7c0333 3079 }
512e94dc 3080 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
164d1e9e 3081 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
cc7c0333 3082
892a7f70
JK		 3083 nfp_net_rx_rings_free(dp);
3084 nfp_net_tx_rings_free(dp);
cc7c0333
JK		 3085
3086 nfp_net_open_stack(nn);
783496b0
JK		 3087exit_free_dp:
3088 kfree(dp);
cc7c0333
JK		 3089
3090 return err;
68453c7a
JK		 3091
3092err_free_rx:
892a7f70 3093 nfp_net_rx_rings_free(dp);
164d1e9e 3094err_cleanup_vecs:
512e94dc 3095 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
164d1e9e 3096 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
783496b0 3097 kfree(dp);
68453c7a
JK		 3098 return err;
3099}
3100
3101static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
3102{
3103 struct nfp_net *nn = netdev_priv(netdev);
783496b0 3104 struct nfp_net_dp *dp;
ccbdc596
JK		 3105 int err;
3106
167cebef 3107 err = nfp_app_check_mtu(nn->app, netdev, new_mtu);
ccbdc596
JK		 3108 if (err)
3109 return err;
783496b0
JK		 3110
3111 dp = nfp_net_clone_dp(nn);
3112 if (!dp)
3113 return -ENOMEM;
68453c7a 3114
76e1e1a8
JK		 3115 dp->mtu = new_mtu;
3116
d957c0f7 3117 return nfp_net_ring_reconfig(nn, dp, NULL);
cc7c0333
JK		 3118}
3119
b64052fc
PC		 3120static int
3121nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
3122{
3123 struct nfp_net *nn = netdev_priv(netdev);
3124
3125 /* Priority tagged packets with vlan id 0 are processed by the
3126 * NFP as untagged packets
3127 */
3128 if (!vid)
3129 return 0;
3130
527d7d1b
JK		 3131 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3132 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3133 ETH_P_8021Q);
b64052fc
PC		 3134
3135 return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD);
3136}
3137
3138static int
3139nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
3140{
3141 struct nfp_net *nn = netdev_priv(netdev);
3142
3143 /* Priority tagged packets with vlan id 0 are processed by the
3144 * NFP as untagged packets
3145 */
3146 if (!vid)
3147 return 0;
3148
527d7d1b
JK		 3149 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3150 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3151 ETH_P_8021Q);
b64052fc
PC		 3152
3153 return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL);
3154}
3155
bc1f4470 3156static void nfp_net_stat64(struct net_device *netdev,
3157 struct rtnl_link_stats64 *stats)
4c352362
JK		 3158{
3159 struct nfp_net *nn = netdev_priv(netdev);
3160 int r;
3161
29f534c4 3162 for (r = 0; r < nn->max_r_vecs; r++) {
4c352362
JK		 3163 struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
3164 u64 data[3];
3165 unsigned int start;
3166
3167 do {
3168 start = u64_stats_fetch_begin(&r_vec->rx_sync);
3169 data[0] = r_vec->rx_pkts;
3170 data[1] = r_vec->rx_bytes;
3171 data[2] = r_vec->rx_drops;
3172 } while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
3173 stats->rx_packets += data[0];
3174 stats->rx_bytes += data[1];
3175 stats->rx_dropped += data[2];
3176
3177 do {
3178 start = u64_stats_fetch_begin(&r_vec->tx_sync);
3179 data[0] = r_vec->tx_pkts;
3180 data[1] = r_vec->tx_bytes;
3181 data[2] = r_vec->tx_errors;
3182 } while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
3183 stats->tx_packets += data[0];
3184 stats->tx_bytes += data[1];
3185 stats->tx_errors += data[2];
3186 }
4c352362
JK		 3187}
3188
3189static int nfp_net_set_features(struct net_device *netdev,
3190 netdev_features_t features)
3191{
3192 netdev_features_t changed = netdev->features ^ features;
3193 struct nfp_net *nn = netdev_priv(netdev);
3194 u32 new_ctrl;
3195 int err;
3196
3197 /* Assume this is not called with features we have not advertised */
3198
79c12a75 3199 new_ctrl = nn->dp.ctrl;
4c352362
JK		 3200
3201 if (changed & NETIF_F_RXCSUM) {
3202 if (features & NETIF_F_RXCSUM)
ddb98d94 3203 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
4c352362 3204 else
ddb98d94 3205 new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
4c352362
JK		 3206 }
3207
3208 if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3209 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
3210 new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3211 else
3212 new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
3213 }
3214
3215 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
3216 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
28063be6
EP		 3217 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3218 NFP_NET_CFG_CTRL_LSO;
4c352362 3219 else
28063be6 3220 new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
4c352362
JK		 3221 }
3222
3223 if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
3224 if (features & NETIF_F_HW_VLAN_CTAG_RX)
3225 new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3226 else
3227 new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
3228 }
3229
3230 if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
3231 if (features & NETIF_F_HW_VLAN_CTAG_TX)
3232 new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3233 else
3234 new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
3235 }
3236
b64052fc
PC		 3237 if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
3238 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
3239 new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3240 else
3241 new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
3242 }
3243
4c352362
JK		 3244 if (changed & NETIF_F_SG) {
3245 if (features & NETIF_F_SG)
3246 new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
3247 else
3248 new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
3249 }
3250
d692403e
JK		 3251 err = nfp_port_set_features(netdev, features);
3252 if (err)
3253 return err;
7533fdc0 3254
4c352362
JK		 3255 nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
3256 netdev->features, features, changed);
3257
79c12a75 3258 if (new_ctrl == nn->dp.ctrl)
4c352362
JK		 3259 return 0;
3260
79c12a75 3261 nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
4c352362
JK		 3262 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3263 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
3264 if (err)
3265 return err;
3266
79c12a75 3267 nn->dp.ctrl = new_ctrl;
4c352362
JK		 3268
3269 return 0;
3270}
3271
3272static netdev_features_t
3273nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
3274 netdev_features_t features)
3275{
3276 u8 l4_hdr;
3277
3278 /* We can't do TSO over double tagged packets (802.1AD) */
3279 features &= vlan_features_check(skb, features);
3280
3281 if (!skb->encapsulation)
3282 return features;
3283
3284 /* Ensure that inner L4 header offset fits into TX descriptor field */
3285 if (skb_is_gso(skb)) {
3286 u32 hdrlen;
3287
3288 hdrlen = skb_inner_transport_header(skb) - skb->data +
3289 inner_tcp_hdrlen(skb);
3290
3291 if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
3292 features &= ~NETIF_F_GSO_MASK;
3293 }
3294
3295 /* VXLAN/GRE check */
3296 switch (vlan_get_protocol(skb)) {
3297 case htons(ETH_P_IP):
3298 l4_hdr = ip_hdr(skb)->protocol;
3299 break;
3300 case htons(ETH_P_IPV6):
3301 l4_hdr = ipv6_hdr(skb)->nexthdr;
3302 break;
3303 default:
a188222b 3304 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4c352362
JK		 3305 }
3306
3307 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
3308 skb->inner_protocol != htons(ETH_P_TEB) ||
3309 (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
3310 (l4_hdr == IPPROTO_UDP &&
3311 (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
3312 sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
a188222b 3313 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4c352362
JK		 3314
3315 return features;
3316}
3317
51c1df83
JK		 3318static int
3319nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len)
3320{
3321 struct nfp_net *nn = netdev_priv(netdev);
3322 int n;
3323
3324 if (nn->port)
3325 return nfp_port_get_phys_port_name(netdev, name, len);
3326
fe06a64e 3327 if (nn->dp.is_vf || nn->vnic_no_name)
6fd1cfc0
JK		 3328 return -EOPNOTSUPP;
3329
ca145732 3330 n = snprintf(name, len, "n%d", nn->id);
6fd1cfc0
JK		 3331 if (n >= len)
3332 return -EINVAL;
51c1df83
JK		 3333
3334 return 0;
3335}
3336
4c352362
JK		 3337/**
3338 * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
3339 * @nn: NFP Net device to reconfigure
3340 * @idx: Index into the port table where new port should be written
3341 * @port: UDP port to configure (pass zero to remove VXLAN port)
3342 */
3343static void nfp_net_set_vxlan_port(struct nfp_net *nn, int idx, __be16 port)
3344{
3345 int i;
3346
3347 nn->vxlan_ports[idx] = port;
3348
79c12a75 3349 if (!(nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN))
4c352362
JK		 3350 return;
3351
3352 BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
3353 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2)
3354 nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(port),
3355 be16_to_cpu(nn->vxlan_ports[i + 1]) << 16 |
3356 be16_to_cpu(nn->vxlan_ports[i]));
3357
3d780b92 3358 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_VXLAN);
4c352362
JK		 3359}
3360
3361/**
3362 * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
3363 * @nn: NFP Network structure
3364 * @port: UDP port to look for
3365 *
3366 * Return: if the port is already in the table -- it's position;
3367 * if the port is not in the table -- free position to use;
3368 * if the table is full -- -ENOSPC.
3369 */
3370static int nfp_net_find_vxlan_idx(struct nfp_net *nn, __be16 port)
3371{
3372 int i, free_idx = -ENOSPC;
3373
3374 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i++) {
3375 if (nn->vxlan_ports[i] == port)
3376 return i;
3377 if (!nn->vxlan_usecnt[i])
3378 free_idx = i;
3379 }
3380
3381 return free_idx;
3382}
3383
3384static void nfp_net_add_vxlan_port(struct net_device *netdev,
3ab68837 3385 struct udp_tunnel_info *ti)
4c352362
JK		 3386{
3387 struct nfp_net *nn = netdev_priv(netdev);
3388 int idx;
3389
3ab68837
AD		 3390 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3391 return;
3392
3393 idx = nfp_net_find_vxlan_idx(nn, ti->port);
4c352362
JK		 3394 if (idx == -ENOSPC)
3395 return;
3396
3397 if (!nn->vxlan_usecnt[idx]++)
3ab68837 3398 nfp_net_set_vxlan_port(nn, idx, ti->port);
4c352362
JK		 3399}
3400
3401static void nfp_net_del_vxlan_port(struct net_device *netdev,
3ab68837 3402 struct udp_tunnel_info *ti)
4c352362
JK		 3403{
3404 struct nfp_net *nn = netdev_priv(netdev);
3405 int idx;
3406
3ab68837
AD		 3407 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3408 return;
3409
3410 idx = nfp_net_find_vxlan_idx(nn, ti->port);
f50cef6f 3411 if (idx == -ENOSPC || !nn->vxlan_usecnt[idx])
4c352362
JK		 3412 return;
3413
3414 if (!--nn->vxlan_usecnt[idx])
3415 nfp_net_set_vxlan_port(nn, idx, 0);
3416}
3417
5f428401 3418static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf)
ecd63a02 3419{
5f428401 3420 struct bpf_prog *prog = bpf->prog;
783496b0 3421 struct nfp_net_dp *dp;
5f428401
JK		 3422 int err;
3423
3424 if (!xdp_attachment_flags_ok(&nn->xdp, bpf))
3425 return -EBUSY;
ecd63a02 3426
6a8ef542
JK		 3427 if (!prog == !nn->dp.xdp_prog) {
3428 WRITE_ONCE(nn->dp.xdp_prog, prog);
5f428401 3429 xdp_attachment_setup(&nn->xdp, bpf);
ecd63a02
JK		 3430 return 0;
3431 }
3432
783496b0
JK		 3433 dp = nfp_net_clone_dp(nn);
3434 if (!dp)
3435 return -ENOMEM;
3436
9dc6b116 3437 dp->xdp_prog = prog;
892a7f70 3438 dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
c487e6b1 3439 dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
dbf637ff 3440 dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
ecd63a02
JK		 3441
3442 /* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
5f428401
JK		 3443 err = nfp_net_ring_reconfig(nn, dp, bpf->extack);
3444 if (err)
3445 return err;
3446
3447 xdp_attachment_setup(&nn->xdp, bpf);
3448 return 0;
c443b5ac
JK		 3449}
3450
5f428401 3451static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf)
c443b5ac
JK		 3452{
3453 int err;
3454
5f428401 3455 if (!xdp_attachment_flags_ok(&nn->xdp_hw, bpf))
9f82fca9
JK		 3456 return -EBUSY;
3457
5f428401 3458 err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack);
c443b5ac
JK		 3459 if (err)
3460 return err;
3461
5f428401 3462 xdp_attachment_setup(&nn->xdp_hw, bpf);
ecd63a02
JK		 3463 return 0;
3464}
3465
f4e63525 3466static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
ecd63a02
JK		 3467{
3468 struct nfp_net *nn = netdev_priv(netdev);
3469
3470 switch (xdp->command) {
3471 case XDP_SETUP_PROG:
5f428401 3472 return nfp_net_xdp_setup_drv(nn, xdp);
cafa92ac 3473 case XDP_SETUP_PROG_HW:
5f428401 3474 return nfp_net_xdp_setup_hw(nn, xdp);
ecd63a02 3475 case XDP_QUERY_PROG:
a25717d2
JK		 3476 return xdp_attachment_query(&nn->xdp, xdp);
3477 case XDP_QUERY_PROG_HW:
5f428401 3478 return xdp_attachment_query(&nn->xdp_hw, xdp);
ecd63a02 3479 default:
af93d15a 3480 return nfp_app_bpf(nn->app, nn, xdp);
ecd63a02
JK		 3481 }
3482}
3483
9d372759
PC		 3484static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
3485{
3486 struct nfp_net *nn = netdev_priv(netdev);
3487 struct sockaddr *saddr = addr;
3488 int err;
3489
3490 err = eth_prepare_mac_addr_change(netdev, addr);
3491 if (err)
3492 return err;
3493
3494 nfp_net_write_mac_addr(nn, saddr->sa_data);
3495
3496 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
3497 if (err)
3498 return err;
3499
3500 eth_commit_mac_addr_change(netdev, addr);
3501
3502 return 0;
3503}
3504
eb488c26 3505const struct net_device_ops nfp_net_netdev_ops = {
4612bebf
JK		 3506 .ndo_init = nfp_app_ndo_init,
3507 .ndo_uninit = nfp_app_ndo_uninit,
4c352362
JK		 3508 .ndo_open = nfp_net_netdev_open,
3509 .ndo_stop = nfp_net_netdev_close,
3510 .ndo_start_xmit = nfp_net_tx,
3511 .ndo_get_stats64 = nfp_net_stat64,
b64052fc
PC		 3512 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid,
3513 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid,
25528d90
PC		 3514 .ndo_set_vf_mac = nfp_app_set_vf_mac,
3515 .ndo_set_vf_vlan = nfp_app_set_vf_vlan,
3516 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk,
3517 .ndo_get_vf_config = nfp_app_get_vf_config,
3518 .ndo_set_vf_link_state = nfp_app_set_vf_link_state,
8a276873 3519 .ndo_setup_tc = nfp_port_setup_tc,
4c352362
JK		 3520 .ndo_tx_timeout = nfp_net_tx_timeout,
3521 .ndo_set_rx_mode = nfp_net_set_rx_mode,
3522 .ndo_change_mtu = nfp_net_change_mtu,
9d372759 3523 .ndo_set_mac_address = nfp_net_set_mac_address,
4c352362
JK		 3524 .ndo_set_features = nfp_net_set_features,
3525 .ndo_features_check = nfp_net_features_check,
51c1df83 3526 .ndo_get_phys_port_name = nfp_net_get_phys_port_name,
3ab68837
AD		 3527 .ndo_udp_tunnel_add = nfp_net_add_vxlan_port,
3528 .ndo_udp_tunnel_del = nfp_net_del_vxlan_port,
f4e63525 3529 .ndo_bpf = nfp_net_xdp,
4c352362
JK		 3530};
3531
3532/**
3533 * nfp_net_info() - Print general info about the NIC
3534 * @nn: NFP Net device to reconfigure
3535 */
3536void nfp_net_info(struct nfp_net *nn)
3537{
416db5c1 3538 nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
79c12a75
JK		 3539 nn->dp.is_vf ? "VF " : "",
3540 nn->dp.num_tx_rings, nn->max_tx_rings,
3541 nn->dp.num_rx_rings, nn->max_rx_rings);
4c352362
JK		 3542 nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3543 nn->fw_ver.resv, nn->fw_ver.class,
3544 nn->fw_ver.major, nn->fw_ver.minor,
3545 nn->max_mtu);
b64052fc 3546 nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
4c352362
JK		 3547 nn->cap,
3548 nn->cap & NFP_NET_CFG_CTRL_PROMISC ? "PROMISC " : "",
3549 nn->cap & NFP_NET_CFG_CTRL_L2BC ? "L2BCFILT " : "",
3550 nn->cap & NFP_NET_CFG_CTRL_L2MC ? "L2MCFILT " : "",
3551 nn->cap & NFP_NET_CFG_CTRL_RXCSUM ? "RXCSUM " : "",
3552 nn->cap & NFP_NET_CFG_CTRL_TXCSUM ? "TXCSUM " : "",
3553 nn->cap & NFP_NET_CFG_CTRL_RXVLAN ? "RXVLAN " : "",
3554 nn->cap & NFP_NET_CFG_CTRL_TXVLAN ? "TXVLAN " : "",
3555 nn->cap & NFP_NET_CFG_CTRL_SCATTER ? "SCATTER " : "",
3556 nn->cap & NFP_NET_CFG_CTRL_GATHER ? "GATHER " : "",
28063be6
EP		 3557 nn->cap & NFP_NET_CFG_CTRL_LSO ? "TSO1 " : "",
3558 nn->cap & NFP_NET_CFG_CTRL_LSO2 ? "TSO2 " : "",
611bdd49
EP		 3559 nn->cap & NFP_NET_CFG_CTRL_RSS ? "RSS1 " : "",
3560 nn->cap & NFP_NET_CFG_CTRL_RSS2 ? "RSS2 " : "",
b64052fc 3561 nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "",
4c352362
JK		 3562 nn->cap & NFP_NET_CFG_CTRL_L2SWITCH ? "L2SWITCH " : "",
3563 nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
3564 nn->cap & NFP_NET_CFG_CTRL_IRQMOD ? "IRQMOD " : "",
3565 nn->cap & NFP_NET_CFG_CTRL_VXLAN ? "VXLAN " : "",
7533fdc0 3566 nn->cap & NFP_NET_CFG_CTRL_NVGRE ? "NVGRE " : "",
ddb98d94 3567 nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ?
9d372759 3568 "RXCSUM_COMPLETE " : "",
bb45e51c
JK		 3569 nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "",
3570 nfp_app_extra_cap(nn->app, nn));
4c352362
JK		 3571}
3572
3573/**
beba69ca 3574 * nfp_net_alloc() - Allocate netdev and related structure
4c352362 3575 * @pdev: PCI device
a7b1ad08 3576 * @needs_netdev: Whether to allocate a netdev for this vNIC
4c352362
JK		 3577 * @max_tx_rings: Maximum number of TX rings supported by device
3578 * @max_rx_rings: Maximum number of RX rings supported by device
3579 *
3580 * This function allocates a netdev device and fills in the initial
a7b1ad08
JK		 3581 * part of the @struct nfp_net structure. In case of control device
3582 * nfp_net structure is allocated without the netdev.
4c352362
JK		 3583 *
3584 * Return: NFP Net device structure, or ERR_PTR on error.
3585 */
a7b1ad08 3586struct nfp_net *nfp_net_alloc(struct pci_dev *pdev, bool needs_netdev,
beba69ca
JK		 3587 unsigned int max_tx_rings,
3588 unsigned int max_rx_rings)
4c352362 3589{
4c352362 3590 struct nfp_net *nn;
4c352362 3591
a7b1ad08
JK		 3592 if (needs_netdev) {
3593 struct net_device *netdev;
4c352362 3594
a7b1ad08
JK		 3595 netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
3596 max_tx_rings, max_rx_rings);
3597 if (!netdev)
3598 return ERR_PTR(-ENOMEM);
3599
3600 SET_NETDEV_DEV(netdev, &pdev->dev);
3601 nn = netdev_priv(netdev);
3602 nn->dp.netdev = netdev;
3603 } else {
3604 nn = vzalloc(sizeof(*nn));
3605 if (!nn)
3606 return ERR_PTR(-ENOMEM);
3607 }
4c352362 3608
79c12a75 3609 nn->dp.dev = &pdev->dev;
4c352362
JK		 3610 nn->pdev = pdev;
3611
3612 nn->max_tx_rings = max_tx_rings;
3613 nn->max_rx_rings = max_rx_rings;
3614
79c12a75
JK		 3615 nn->dp.num_tx_rings = min_t(unsigned int,
3616 max_tx_rings, num_online_cpus());
3617 nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings,
cbeaf7aa 3618 netif_get_num_default_rss_queues());
4c352362 3619
79c12a75
JK		 3620 nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings);
3621 nn->dp.num_r_vecs = min_t(unsigned int,
3622 nn->dp.num_r_vecs, num_online_cpus());
4b27a1eb 3623
79c12a75
JK		 3624 nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
3625 nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
4c352362
JK		 3626
3627 spin_lock_init(&nn->reconfig_lock);
3628 spin_lock_init(&nn->link_status_lock);
3629
3248f77f 3630 timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0);
3d780b92 3631
4c352362
JK		 3632 return nn;
3633}
3634
3635/**
beba69ca 3636 * nfp_net_free() - Undo what @nfp_net_alloc() did
4c352362
JK		 3637 * @nn: NFP Net device to reconfigure
3638 */
beba69ca 3639void nfp_net_free(struct nfp_net *nn)
4c352362 3640{
9ad716b9 3641 WARN_ON(timer_pending(&nn->reconfig_timer) || nn->reconfig_posted);
a7b1ad08
JK		 3642 if (nn->dp.netdev)
3643 free_netdev(nn->dp.netdev);
3644 else
3645 vfree(nn);
4c352362
JK		 3646}
3647
9ff304bf
JK		 3648/**
3649 * nfp_net_rss_key_sz() - Get current size of the RSS key
3650 * @nn: NFP Net device instance
3651 *
3652 * Return: size of the RSS key for currently selected hash function.
3653 */
3654unsigned int nfp_net_rss_key_sz(struct nfp_net *nn)
3655{
3656 switch (nn->rss_hfunc) {
3657 case ETH_RSS_HASH_TOP:
3658 return NFP_NET_CFG_RSS_KEY_SZ;
3659 case ETH_RSS_HASH_XOR:
3660 return 0;
3661 case ETH_RSS_HASH_CRC32:
3662 return 4;
3663 }
3664
3665 nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc);
3666 return 0;
3667}
3668
4c352362
JK		 3669/**
3670 * nfp_net_rss_init() - Set the initial RSS parameters
3671 * @nn: NFP Net device to reconfigure
3672 */
3673static void nfp_net_rss_init(struct nfp_net *nn)
3674{
9ff304bf
JK		 3675 unsigned long func_bit, rss_cap_hfunc;
3676 u32 reg;
3677
3678 /* Read the RSS function capability and select first supported func */
3679 reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP);
3680 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg);
3681 if (!rss_cap_hfunc)
3682 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC,
3683 NFP_NET_CFG_RSS_TOEPLITZ);
3684
3685 func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS);
3686 if (func_bit == NFP_NET_CFG_RSS_HFUNCS) {
79c12a75 3687 dev_warn(nn->dp.dev,
9ff304bf
JK		 3688 "Bad RSS config, defaulting to Toeplitz hash\n");
3689 func_bit = ETH_RSS_HASH_TOP_BIT;
3690 }
3691 nn->rss_hfunc = 1 << func_bit;
3692
3693 netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn));
4c352362 3694
1e9e10d0 3695 nfp_net_rss_init_itbl(nn);
4c352362
JK		 3696
3697 /* Enable IPv4/IPv6 TCP by default */
3698 nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
3699 NFP_NET_CFG_RSS_IPV6_TCP |
9ff304bf 3700 FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) |
4c352362
JK		 3701 NFP_NET_CFG_RSS_MASK;
3702}
3703
3704/**
3705 * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
3706 * @nn: NFP Net device to reconfigure
3707 */
3708static void nfp_net_irqmod_init(struct nfp_net *nn)
3709{
3710 nn->rx_coalesce_usecs = 50;
3711 nn->rx_coalesce_max_frames = 64;
3712 nn->tx_coalesce_usecs = 50;
3713 nn->tx_coalesce_max_frames = 64;
3714}
3715
a7b1ad08 3716static void nfp_net_netdev_init(struct nfp_net *nn)
4c352362 3717{
beba69ca 3718 struct net_device *netdev = nn->dp.netdev;
611bdd49 3719
9d372759 3720 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
4c352362 3721
a7b1ad08 3722 netdev->mtu = nn->dp.mtu;
4c352362
JK		 3723
3724 /* Advertise/enable offloads based on capabilities
3725 *
3726 * Note: netdev->features show the currently enabled features
3727 * and netdev->hw_features advertises which features are
3728 * supported. By default we enable most features.
3729 */
9d372759
PC		 3730 if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
3731 netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3732
4c352362 3733 netdev->hw_features = NETIF_F_HIGHDMA;
ddb98d94 3734 if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) {
4c352362 3735 netdev->hw_features |= NETIF_F_RXCSUM;
ddb98d94 3736 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
4c352362
JK		 3737 }
3738 if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
3739 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
79c12a75 3740 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
4c352362
JK		 3741 }
3742 if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
3743 netdev->hw_features |= NETIF_F_SG;
79c12a75 3744 nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER;
4c352362 3745 }
28063be6
EP		 3746 if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
3747 nn->cap & NFP_NET_CFG_CTRL_LSO2) {
4c352362 3748 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
28063be6
EP		 3749 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3750 NFP_NET_CFG_CTRL_LSO;
4c352362 3751 }
a7b1ad08 3752 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY)
4c352362 3753 netdev->hw_features |= NETIF_F_RXHASH;
4c352362
JK		 3754 if (nn->cap & NFP_NET_CFG_CTRL_VXLAN &&
3755 nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
3756 if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3757 netdev->hw_features |= NETIF_F_GSO_GRE |
3758 NETIF_F_GSO_UDP_TUNNEL;
79c12a75 3759 nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE;
4c352362
JK		 3760
3761 netdev->hw_enc_features = netdev->hw_features;
3762 }
3763
3764 netdev->vlan_features = netdev->hw_features;
3765
3766 if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
3767 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
79c12a75 3768 nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
4c352362
JK		 3769 }
3770 if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
28063be6
EP		 3771 if (nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3772 nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
3773 } else {
3774 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3775 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3776 }
4c352362 3777 }
b64052fc
PC		 3778 if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) {
3779 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3780 nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3781 }
4c352362
JK		 3782
3783 netdev->features = netdev->hw_features;
3784
0b9de4ca 3785 if (nfp_app_has_tc(nn->app) && nn->port)
7533fdc0
JK		 3786 netdev->hw_features |= NETIF_F_HW_TC;
3787
4c352362
JK		 3788 /* Advertise but disable TSO by default. */
3789 netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
28063be6 3790 nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
4c352362 3791
a7b1ad08
JK		 3792 /* Finalise the netdev setup */
3793 netdev->netdev_ops = &nfp_net_netdev_ops;
3794 netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
3795
8f15df60
SH		 3796 SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
3797
a7b1ad08
JK		 3798 /* MTU range: 68 - hw-specific max */
3799 netdev->min_mtu = ETH_MIN_MTU;
3800 netdev->max_mtu = nn->max_mtu;
3801
0d592e52
JK		 3802 netdev->gso_max_segs = NFP_NET_LSO_MAX_SEGS;
3803
a7b1ad08
JK		 3804 netif_carrier_off(netdev);
3805
3806 nfp_net_set_ethtool_ops(netdev);
3807}
3808
545bfa7a 3809static int nfp_net_read_caps(struct nfp_net *nn)
a7b1ad08 3810{
a7b1ad08
JK		 3811 /* Get some of the read-only fields from the BAR */
3812 nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
3813 nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
3814
64a919a9
JK		 3815 /* ABI 4.x and ctrl vNIC always use chained metadata, in other cases
3816 * we allow use of non-chained metadata if RSS(v1) is the only
3817 * advertised capability requiring metadata.
3818 */
a7b1ad08 3819 nn->dp.chained_metadata_format = nn->fw_ver.major == 4 ||
77ece8d5 3820 !nn->dp.netdev ||
64a919a9 3821 !(nn->cap & NFP_NET_CFG_CTRL_RSS) ||
a7b1ad08 3822 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META;
64a919a9
JK		 3823 /* RSS(v1) uses non-chained metadata format, except in ABI 4.x where
3824 * it has the same meaning as RSSv2.
3825 */
a7b1ad08
JK		 3826 if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4)
3827 nn->cap &= ~NFP_NET_CFG_CTRL_RSS;
3828
3829 /* Determine RX packet/metadata boundary offset */
3830 if (nn->fw_ver.major >= 2) {
3831 u32 reg;
3832
3833 reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
3834 if (reg > NFP_NET_MAX_PREPEND) {
3835 nn_err(nn, "Invalid rx offset: %d\n", reg);
3836 return -EINVAL;
3837 }
3838 nn->dp.rx_offset = reg;
3839 } else {
3840 nn->dp.rx_offset = NFP_NET_RX_OFFSET;
3841 }
3842
78a0a65f
JK		 3843 /* For control vNICs mask out the capabilities app doesn't want. */
3844 if (!nn->dp.netdev)
3845 nn->cap &= nn->app->type->ctrl_cap_mask;
3846
545bfa7a
JK		 3847 return 0;
3848}
3849
3850/**
3851 * nfp_net_init() - Initialise/finalise the nfp_net structure
3852 * @nn: NFP Net device structure
3853 *
3854 * Return: 0 on success or negative errno on error.
3855 */
3856int nfp_net_init(struct nfp_net *nn)
3857{
3858 int err;
3859
3860 nn->dp.rx_dma_dir = DMA_FROM_DEVICE;
3861
3862 err = nfp_net_read_caps(nn);
3863 if (err)
3864 return err;
3865
a7b1ad08
JK		 3866 /* Set default MTU and Freelist buffer size */
3867 if (nn->max_mtu < NFP_NET_DEFAULT_MTU)
3868 nn->dp.mtu = nn->max_mtu;
3869 else
3870 nn->dp.mtu = NFP_NET_DEFAULT_MTU;
3871 nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp);
3872
79ca38e8
JK		 3873 if (nfp_app_ctrl_uses_data_vnics(nn->app))
3874 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_CMSG_DATA;
3875
a7b1ad08
JK		 3876 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) {
3877 nfp_net_rss_init(nn);
3878 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?:
3879 NFP_NET_CFG_CTRL_RSS;
3880 }
3881
4c352362
JK		 3882 /* Allow L2 Broadcast and Multicast through by default, if supported */
3883 if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
79c12a75 3884 nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC;
4c352362
JK		 3885
3886 /* Allow IRQ moderation, if supported */
3887 if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
3888 nfp_net_irqmod_init(nn);
79c12a75 3889 nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
4c352362
JK		 3890 }
3891
73a0329b
JK		 3892 err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar,
3893 &nn->tlv_caps);
3894 if (err)
3895 return err;
3896
a7b1ad08
JK		 3897 if (nn->dp.netdev)
3898 nfp_net_netdev_init(nn);
3899
4c352362
JK		 3900 /* Stash the re-configuration queue away. First odd queue in TX Bar */
3901 nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
3902
3903 /* Make sure the FW knows the netdev is supposed to be disabled here */
3904 nn_writel(nn, NFP_NET_CFG_CTRL, 0);
3905 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
3906 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
3907 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
3908 NFP_NET_CFG_UPDATE_GEN);
3909 if (err)
3910 return err;
3911
beba69ca 3912 nfp_net_vecs_init(nn);
4c352362 3913
a7b1ad08
JK		 3914 if (!nn->dp.netdev)
3915 return 0;
3916 return register_netdev(nn->dp.netdev);
4c352362
JK		 3917}
3918
3919/**
beba69ca
JK		 3920 * nfp_net_clean() - Undo what nfp_net_init() did.
3921 * @nn: NFP Net device structure
4c352362 3922 */
beba69ca 3923void nfp_net_clean(struct nfp_net *nn)
4c352362 3924{
a7b1ad08
JK		 3925 if (!nn->dp.netdev)
3926 return;
3927
6f14f443 3928 unregister_netdev(nn->dp.netdev);
9ad716b9 3929 nfp_net_reconfig_wait_posted(nn);
4c352362 3930}
Linux 6.x block layer and io_uring tree(s)