1 /* SPDX-License-Identifier: BSD-3-Clause */ 2 /* Copyright (c) 2024, Intel Corporation 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are met: 7 * 8 * 1. Redistributions of source code must retain the above copyright notice, 9 * this list of conditions and the following disclaimer. 10 * 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * 3. Neither the name of the Intel Corporation nor the names of its 16 * contributors may be used to endorse or promote products derived from 17 * this software without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 20 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /** 33 * @file if_ice_iflib.c 34 * @brief iflib driver implementation 35 * 36 * Contains the main entry point for the iflib driver implementation. It 37 * implements the various ifdi driver methods, and sets up the module and 38 * driver values to load an iflib driver. 39 */ 40 41 #include "ice_iflib.h" 42 #include "ice_drv_info.h" 43 #include "ice_switch.h" 44 #include "ice_sched.h" 45 #ifdef PCI_IOV 46 #include "ice_iov.h" 47 #endif 48 49 #include <sys/module.h> 50 #include <sys/sockio.h> 51 #include <sys/smp.h> 52 #include <dev/pci/pcivar.h> 53 #include <dev/pci/pcireg.h> 54 55 /* 56 * Device method prototypes 57 */ 58 59 static void *ice_register(device_t); 60 static int ice_if_attach_pre(if_ctx_t); 61 static int ice_attach_pre_recovery_mode(struct ice_softc *sc); 62 static int ice_if_attach_post(if_ctx_t); 63 static void ice_attach_post_recovery_mode(struct ice_softc *sc); 64 static int ice_if_detach(if_ctx_t); 65 static int ice_if_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int ntxqs, int ntxqsets); 66 static int ice_if_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int nqs, int nqsets); 67 static int ice_if_msix_intr_assign(if_ctx_t ctx, int msix); 68 static void ice_if_queues_free(if_ctx_t ctx); 69 static int ice_if_mtu_set(if_ctx_t ctx, uint32_t mtu); 70 static void ice_if_intr_enable(if_ctx_t ctx); 71 static void ice_if_intr_disable(if_ctx_t ctx); 72 static int ice_if_rx_queue_intr_enable(if_ctx_t ctx, uint16_t rxqid); 73 static int ice_if_tx_queue_intr_enable(if_ctx_t ctx, uint16_t txqid); 74 static int ice_if_promisc_set(if_ctx_t ctx, int flags); 75 static void ice_if_media_status(if_ctx_t ctx, struct ifmediareq *ifmr); 76 static int ice_if_media_change(if_ctx_t ctx); 77 static void ice_if_init(if_ctx_t ctx); 78 static void ice_if_timer(if_ctx_t ctx, uint16_t qid); 79 static void ice_if_update_admin_status(if_ctx_t ctx); 80 static void ice_if_multi_set(if_ctx_t ctx); 81 static void ice_if_vlan_register(if_ctx_t ctx, u16 vtag); 82 static void ice_if_vlan_unregister(if_ctx_t ctx, u16 vtag); 83 static void ice_if_stop(if_ctx_t ctx); 84 static uint64_t ice_if_get_counter(if_ctx_t ctx, ift_counter counter); 85 static int ice_if_priv_ioctl(if_ctx_t ctx, u_long command, caddr_t data); 86 static int ice_if_i2c_req(if_ctx_t ctx, struct ifi2creq *req); 87 static int ice_if_suspend(if_ctx_t ctx); 88 static int ice_if_resume(if_ctx_t ctx); 89 static bool ice_if_needs_restart(if_ctx_t ctx, enum iflib_restart_event event); 90 static void ice_init_link(struct ice_softc *sc); 91 #ifdef PCI_IOV 92 static int ice_if_iov_init(if_ctx_t ctx, uint16_t num_vfs, const nvlist_t *params); 93 static void ice_if_iov_uninit(if_ctx_t ctx); 94 static int ice_if_iov_vf_add(if_ctx_t ctx, uint16_t vfnum, const nvlist_t *params); 95 static void ice_if_vflr_handle(if_ctx_t ctx); 96 #endif 97 static int ice_setup_mirror_vsi(struct ice_mirr_if *mif); 98 static int ice_wire_mirror_intrs(struct ice_mirr_if *mif); 99 static void ice_free_irqvs_subif(struct ice_mirr_if *mif); 100 static void *ice_subif_register(device_t); 101 static void ice_subif_setup_scctx(struct ice_mirr_if *mif); 102 static int ice_subif_rebuild(struct ice_softc *sc); 103 static int ice_subif_rebuild_vsi_qmap(struct ice_softc *sc); 104 105 /* Iflib API */ 106 static int ice_subif_if_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, 107 uint64_t *paddrs, int ntxqs, int ntxqsets); 108 static int ice_subif_if_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, 109 uint64_t *paddrs, int nrxqs, int nrxqsets); 110 static int ice_subif_if_rx_queue_intr_enable(if_ctx_t ctx, uint16_t rxqid); 111 static int ice_subif_if_tx_queue_intr_enable(if_ctx_t ctx, uint16_t txqid); 112 static void ice_subif_if_intr_enable(if_ctx_t ctx); 113 static int ice_subif_if_msix_intr_assign(if_ctx_t ctx, int msix); 114 static void ice_subif_if_init(if_ctx_t ctx); 115 static void ice_subif_if_stop(if_ctx_t ctx); 116 static void ice_subif_if_queues_free(if_ctx_t ctx); 117 static int ice_subif_if_attach_pre(if_ctx_t); 118 static int ice_subif_if_attach_post(if_ctx_t); 119 static void ice_subif_if_media_status(if_ctx_t ctx, struct ifmediareq *ifmr); 120 static int ice_subif_if_promisc_set(if_ctx_t ctx, int flags); 121 122 static int ice_msix_que(void *arg); 123 static int ice_msix_admin(void *arg); 124 125 /* 126 * Helper function prototypes 127 */ 128 static int ice_pci_mapping(struct ice_softc *sc); 129 static void ice_free_pci_mapping(struct ice_softc *sc); 130 static void ice_update_link_status(struct ice_softc *sc, bool update_media); 131 static void ice_init_device_features(struct ice_softc *sc); 132 static void ice_init_tx_tracking(struct ice_vsi *vsi); 133 static void ice_handle_reset_event(struct ice_softc *sc); 134 static void ice_handle_pf_reset_request(struct ice_softc *sc); 135 static void ice_prepare_for_reset(struct ice_softc *sc); 136 static int ice_rebuild_pf_vsi_qmap(struct ice_softc *sc); 137 static void ice_rebuild(struct ice_softc *sc); 138 static void ice_rebuild_recovery_mode(struct ice_softc *sc); 139 static void ice_free_irqvs(struct ice_softc *sc); 140 static void ice_update_rx_mbuf_sz(struct ice_softc *sc); 141 static void ice_poll_for_media_avail(struct ice_softc *sc); 142 static void ice_setup_scctx(struct ice_softc *sc); 143 static int ice_allocate_msix(struct ice_softc *sc); 144 static void ice_admin_timer(void *arg); 145 static void ice_transition_recovery_mode(struct ice_softc *sc); 146 static void ice_transition_safe_mode(struct ice_softc *sc); 147 static void ice_set_default_promisc_mask(ice_bitmap_t *promisc_mask); 148 149 /* 150 * Device Interface Declaration 151 */ 152 153 /** 154 * @var ice_methods 155 * @brief ice driver method entry points 156 * 157 * List of device methods implementing the generic device interface used by 158 * the device stack to interact with the ice driver. Since this is an iflib 159 * driver, most of the methods point to the generic iflib implementation. 160 */ 161 static device_method_t ice_methods[] = { 162 /* Device interface */ 163 DEVMETHOD(device_register, ice_register), 164 DEVMETHOD(device_probe, iflib_device_probe_vendor), 165 DEVMETHOD(device_attach, iflib_device_attach), 166 DEVMETHOD(device_detach, iflib_device_detach), 167 DEVMETHOD(device_shutdown, iflib_device_shutdown), 168 DEVMETHOD(device_suspend, iflib_device_suspend), 169 DEVMETHOD(device_resume, iflib_device_resume), 170 #ifdef PCI_IOV 171 DEVMETHOD(pci_iov_init, iflib_device_iov_init), 172 DEVMETHOD(pci_iov_uninit, iflib_device_iov_uninit), 173 DEVMETHOD(pci_iov_add_vf, iflib_device_iov_add_vf), 174 #endif 175 DEVMETHOD_END 176 }; 177 178 /** 179 * @var ice_iflib_methods 180 * @brief iflib method entry points 181 * 182 * List of device methods used by the iflib stack to interact with this 183 * driver. These are the real main entry points used to interact with this 184 * driver. 185 */ 186 static device_method_t ice_iflib_methods[] = { 187 DEVMETHOD(ifdi_attach_pre, ice_if_attach_pre), 188 DEVMETHOD(ifdi_attach_post, ice_if_attach_post), 189 DEVMETHOD(ifdi_detach, ice_if_detach), 190 DEVMETHOD(ifdi_tx_queues_alloc, ice_if_tx_queues_alloc), 191 DEVMETHOD(ifdi_rx_queues_alloc, ice_if_rx_queues_alloc), 192 DEVMETHOD(ifdi_msix_intr_assign, ice_if_msix_intr_assign), 193 DEVMETHOD(ifdi_queues_free, ice_if_queues_free), 194 DEVMETHOD(ifdi_mtu_set, ice_if_mtu_set), 195 DEVMETHOD(ifdi_intr_enable, ice_if_intr_enable), 196 DEVMETHOD(ifdi_intr_disable, ice_if_intr_disable), 197 DEVMETHOD(ifdi_rx_queue_intr_enable, ice_if_rx_queue_intr_enable), 198 DEVMETHOD(ifdi_tx_queue_intr_enable, ice_if_tx_queue_intr_enable), 199 DEVMETHOD(ifdi_promisc_set, ice_if_promisc_set), 200 DEVMETHOD(ifdi_media_status, ice_if_media_status), 201 DEVMETHOD(ifdi_media_change, ice_if_media_change), 202 DEVMETHOD(ifdi_init, ice_if_init), 203 DEVMETHOD(ifdi_stop, ice_if_stop), 204 DEVMETHOD(ifdi_timer, ice_if_timer), 205 DEVMETHOD(ifdi_update_admin_status, ice_if_update_admin_status), 206 DEVMETHOD(ifdi_multi_set, ice_if_multi_set), 207 DEVMETHOD(ifdi_vlan_register, ice_if_vlan_register), 208 DEVMETHOD(ifdi_vlan_unregister, ice_if_vlan_unregister), 209 DEVMETHOD(ifdi_get_counter, ice_if_get_counter), 210 DEVMETHOD(ifdi_priv_ioctl, ice_if_priv_ioctl), 211 DEVMETHOD(ifdi_i2c_req, ice_if_i2c_req), 212 DEVMETHOD(ifdi_suspend, ice_if_suspend), 213 DEVMETHOD(ifdi_resume, ice_if_resume), 214 DEVMETHOD(ifdi_needs_restart, ice_if_needs_restart), 215 #ifdef PCI_IOV 216 DEVMETHOD(ifdi_iov_vf_add, ice_if_iov_vf_add), 217 DEVMETHOD(ifdi_iov_init, ice_if_iov_init), 218 DEVMETHOD(ifdi_iov_uninit, ice_if_iov_uninit), 219 DEVMETHOD(ifdi_vflr_handle, ice_if_vflr_handle), 220 #endif 221 DEVMETHOD_END 222 }; 223 224 /** 225 * @var ice_driver 226 * @brief driver structure for the generic device stack 227 * 228 * driver_t definition used to setup the generic device methods. 229 */ 230 static driver_t ice_driver = { 231 .name = "ice", 232 .methods = ice_methods, 233 .size = sizeof(struct ice_softc), 234 }; 235 236 /** 237 * @var ice_iflib_driver 238 * @brief driver structure for the iflib stack 239 * 240 * driver_t definition used to setup the iflib device methods. 241 */ 242 static driver_t ice_iflib_driver = { 243 .name = "ice", 244 .methods = ice_iflib_methods, 245 .size = sizeof(struct ice_softc), 246 }; 247 248 extern struct if_txrx ice_txrx; 249 extern struct if_txrx ice_recovery_txrx; 250 251 /** 252 * @var ice_sctx 253 * @brief ice driver shared context 254 * 255 * Structure defining shared values (context) that is used by all instances of 256 * the device. Primarily used to setup details about how the iflib stack 257 * should treat this driver. Also defines the default, minimum, and maximum 258 * number of descriptors in each ring. 259 */ 260 static struct if_shared_ctx ice_sctx = { 261 .isc_magic = IFLIB_MAGIC, 262 .isc_q_align = PAGE_SIZE, 263 264 .isc_tx_maxsize = ICE_MAX_FRAME_SIZE, 265 /* We could technically set this as high as ICE_MAX_DMA_SEG_SIZE, but 266 * that doesn't make sense since that would be larger than the maximum 267 * size of a single packet. 268 */ 269 .isc_tx_maxsegsize = ICE_MAX_FRAME_SIZE, 270 271 /* XXX: This is only used by iflib to ensure that 272 * scctx->isc_tx_tso_size_max + the VLAN header is a valid size. 273 */ 274 .isc_tso_maxsize = ICE_TSO_SIZE + sizeof(struct ether_vlan_header), 275 /* XXX: This is used by iflib to set the number of segments in the TSO 276 * DMA tag. However, scctx->isc_tx_tso_segsize_max is used to set the 277 * related ifnet parameter. 278 */ 279 .isc_tso_maxsegsize = ICE_MAX_DMA_SEG_SIZE, 280 281 .isc_rx_maxsize = ICE_MAX_FRAME_SIZE, 282 .isc_rx_nsegments = ICE_MAX_RX_SEGS, 283 .isc_rx_maxsegsize = ICE_MAX_FRAME_SIZE, 284 285 .isc_nfl = 1, 286 .isc_ntxqs = 1, 287 .isc_nrxqs = 1, 288 289 .isc_admin_intrcnt = 1, 290 .isc_vendor_info = ice_vendor_info_array, 291 .isc_driver_version = __DECONST(char *, ice_driver_version), 292 .isc_driver = &ice_iflib_driver, 293 294 /* 295 * IFLIB_NEED_SCRATCH ensures that mbufs have scratch space available 296 * for hardware checksum offload 297 * 298 * IFLIB_TSO_INIT_IP ensures that the TSO packets have zeroed out the 299 * IP sum field, required by our hardware to calculate valid TSO 300 * checksums. 301 * 302 * IFLIB_ADMIN_ALWAYS_RUN ensures that the administrative task runs 303 * even when the interface is down. 304 * 305 * IFLIB_SKIP_MSIX allows the driver to handle allocating MSI-X 306 * vectors manually instead of relying on iflib code to do this. 307 */ 308 .isc_flags = IFLIB_NEED_SCRATCH | IFLIB_TSO_INIT_IP | 309 IFLIB_ADMIN_ALWAYS_RUN | IFLIB_SKIP_MSIX, 310 311 .isc_nrxd_min = {ICE_MIN_DESC_COUNT}, 312 .isc_ntxd_min = {ICE_MIN_DESC_COUNT}, 313 .isc_nrxd_max = {ICE_IFLIB_MAX_DESC_COUNT}, 314 .isc_ntxd_max = {ICE_IFLIB_MAX_DESC_COUNT}, 315 .isc_nrxd_default = {ICE_DEFAULT_DESC_COUNT}, 316 .isc_ntxd_default = {ICE_DEFAULT_DESC_COUNT}, 317 }; 318 319 DRIVER_MODULE(ice, pci, ice_driver, ice_module_event_handler, NULL); 320 321 MODULE_VERSION(ice, 1); 322 MODULE_DEPEND(ice, pci, 1, 1, 1); 323 MODULE_DEPEND(ice, ether, 1, 1, 1); 324 MODULE_DEPEND(ice, iflib, 1, 1, 1); 325 326 IFLIB_PNP_INFO(pci, ice, ice_vendor_info_array); 327 328 /* Static driver-wide sysctls */ 329 #include "ice_iflib_sysctls.h" 330 331 /** 332 * ice_pci_mapping - Map PCI BAR memory 333 * @sc: device private softc 334 * 335 * Map PCI BAR 0 for device operation. 336 */ 337 static int 338 ice_pci_mapping(struct ice_softc *sc) 339 { 340 int rc; 341 342 /* Map BAR0 */ 343 rc = ice_map_bar(sc->dev, &sc->bar0, 0); 344 if (rc) 345 return rc; 346 347 return 0; 348 } 349 350 /** 351 * ice_free_pci_mapping - Release PCI BAR memory 352 * @sc: device private softc 353 * 354 * Release PCI BARs which were previously mapped by ice_pci_mapping(). 355 */ 356 static void 357 ice_free_pci_mapping(struct ice_softc *sc) 358 { 359 /* Free BAR0 */ 360 ice_free_bar(sc->dev, &sc->bar0); 361 } 362 363 /* 364 * Device methods 365 */ 366 367 /** 368 * ice_register - register device method callback 369 * @dev: the device being registered 370 * 371 * Returns a pointer to the shared context structure, which is used by iflib. 372 */ 373 static void * 374 ice_register(device_t dev __unused) 375 { 376 return &ice_sctx; 377 } /* ice_register */ 378 379 /** 380 * ice_setup_scctx - Setup the iflib softc context structure 381 * @sc: the device private structure 382 * 383 * Setup the parameters in if_softc_ctx_t structure used by the iflib stack 384 * when loading. 385 */ 386 static void 387 ice_setup_scctx(struct ice_softc *sc) 388 { 389 if_softc_ctx_t scctx = sc->scctx; 390 struct ice_hw *hw = &sc->hw; 391 device_t dev = sc->dev; 392 bool safe_mode, recovery_mode; 393 394 safe_mode = ice_is_bit_set(sc->feat_en, ICE_FEATURE_SAFE_MODE); 395 recovery_mode = ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE); 396 397 /* 398 * If the driver loads in Safe mode or Recovery mode, limit iflib to 399 * a single queue pair. 400 */ 401 if (safe_mode || recovery_mode) { 402 scctx->isc_ntxqsets = scctx->isc_nrxqsets = 1; 403 scctx->isc_ntxqsets_max = 1; 404 scctx->isc_nrxqsets_max = 1; 405 } else { 406 /* 407 * iflib initially sets the isc_ntxqsets and isc_nrxqsets to 408 * the values of the override sysctls. Cache these initial 409 * values so that the driver can be aware of what the iflib 410 * sysctl value is when setting up MSI-X vectors. 411 */ 412 sc->ifc_sysctl_ntxqs = scctx->isc_ntxqsets; 413 sc->ifc_sysctl_nrxqs = scctx->isc_nrxqsets; 414 415 if (scctx->isc_ntxqsets == 0) 416 scctx->isc_ntxqsets = hw->func_caps.common_cap.rss_table_size; 417 if (scctx->isc_nrxqsets == 0) 418 scctx->isc_nrxqsets = hw->func_caps.common_cap.rss_table_size; 419 420 scctx->isc_ntxqsets_max = hw->func_caps.common_cap.num_txq; 421 scctx->isc_nrxqsets_max = hw->func_caps.common_cap.num_rxq; 422 423 /* 424 * Sanity check that the iflib sysctl values are within the 425 * maximum supported range. 426 */ 427 if (sc->ifc_sysctl_ntxqs > scctx->isc_ntxqsets_max) 428 sc->ifc_sysctl_ntxqs = scctx->isc_ntxqsets_max; 429 if (sc->ifc_sysctl_nrxqs > scctx->isc_nrxqsets_max) 430 sc->ifc_sysctl_nrxqs = scctx->isc_nrxqsets_max; 431 } 432 433 scctx->isc_txqsizes[0] = roundup2(scctx->isc_ntxd[0] 434 * sizeof(struct ice_tx_desc), DBA_ALIGN); 435 scctx->isc_rxqsizes[0] = roundup2(scctx->isc_nrxd[0] 436 * sizeof(union ice_32b_rx_flex_desc), DBA_ALIGN); 437 438 scctx->isc_tx_nsegments = ICE_MAX_TX_SEGS; 439 scctx->isc_tx_tso_segments_max = ICE_MAX_TSO_SEGS; 440 scctx->isc_tx_tso_size_max = ICE_TSO_SIZE; 441 scctx->isc_tx_tso_segsize_max = ICE_MAX_DMA_SEG_SIZE; 442 443 scctx->isc_msix_bar = pci_msix_table_bar(dev); 444 scctx->isc_rss_table_size = hw->func_caps.common_cap.rss_table_size; 445 446 /* 447 * If the driver loads in recovery mode, disable Tx/Rx functionality 448 */ 449 if (recovery_mode) 450 scctx->isc_txrx = &ice_recovery_txrx; 451 else 452 scctx->isc_txrx = &ice_txrx; 453 454 /* 455 * If the driver loads in Safe mode or Recovery mode, disable 456 * advanced features including hardware offloads. 457 */ 458 if (safe_mode || recovery_mode) { 459 scctx->isc_capenable = ICE_SAFE_CAPS; 460 scctx->isc_tx_csum_flags = 0; 461 } else { 462 scctx->isc_capenable = ICE_FULL_CAPS; 463 scctx->isc_tx_csum_flags = ICE_CSUM_OFFLOAD; 464 } 465 466 scctx->isc_capabilities = scctx->isc_capenable; 467 } /* ice_setup_scctx */ 468 469 /** 470 * ice_if_attach_pre - Early device attach logic 471 * @ctx: the iflib context structure 472 * 473 * Called by iflib during the attach process. Earliest main driver entry 474 * point which performs necessary hardware and driver initialization. Called 475 * before the Tx and Rx queues are allocated. 476 */ 477 static int 478 ice_if_attach_pre(if_ctx_t ctx) 479 { 480 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 481 enum ice_fw_modes fw_mode; 482 int status; 483 if_softc_ctx_t scctx; 484 struct ice_hw *hw; 485 device_t dev; 486 int err; 487 488 device_printf(iflib_get_dev(ctx), "Loading the iflib ice driver\n"); 489 490 ice_set_state(&sc->state, ICE_STATE_ATTACHING); 491 492 sc->ctx = ctx; 493 sc->media = iflib_get_media(ctx); 494 sc->sctx = iflib_get_sctx(ctx); 495 sc->iflib_ctx_lock = iflib_ctx_lock_get(ctx); 496 sc->ifp = iflib_get_ifp(ctx); 497 498 dev = sc->dev = iflib_get_dev(ctx); 499 scctx = sc->scctx = iflib_get_softc_ctx(ctx); 500 501 hw = &sc->hw; 502 hw->back = sc; 503 504 snprintf(sc->admin_mtx_name, sizeof(sc->admin_mtx_name), 505 "%s:admin", device_get_nameunit(dev)); 506 mtx_init(&sc->admin_mtx, sc->admin_mtx_name, NULL, MTX_DEF); 507 callout_init_mtx(&sc->admin_timer, &sc->admin_mtx, 0); 508 509 ASSERT_CTX_LOCKED(sc); 510 511 if (ice_pci_mapping(sc)) { 512 err = (ENXIO); 513 goto destroy_admin_timer; 514 } 515 516 /* Save off the PCI information */ 517 ice_save_pci_info(hw, dev); 518 519 /* create tunables as early as possible */ 520 ice_add_device_tunables(sc); 521 522 /* Setup ControlQ lengths */ 523 ice_set_ctrlq_len(hw); 524 525 reinit_hw: 526 527 fw_mode = ice_get_fw_mode(hw); 528 if (fw_mode == ICE_FW_MODE_REC) { 529 device_printf(dev, "Firmware recovery mode detected. Limiting functionality. Refer to Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n"); 530 531 err = ice_attach_pre_recovery_mode(sc); 532 if (err) 533 goto free_pci_mapping; 534 535 return (0); 536 } 537 538 /* Initialize the hw data structure */ 539 status = ice_init_hw(hw); 540 if (status) { 541 if (status == ICE_ERR_FW_API_VER) { 542 /* Enter recovery mode, so that the driver remains 543 * loaded. This way, if the system administrator 544 * cannot update the driver, they may still attempt to 545 * downgrade the NVM. 546 */ 547 err = ice_attach_pre_recovery_mode(sc); 548 if (err) 549 goto free_pci_mapping; 550 551 return (0); 552 } else { 553 err = EIO; 554 device_printf(dev, "Unable to initialize hw, err %s aq_err %s\n", 555 ice_status_str(status), 556 ice_aq_str(hw->adminq.sq_last_status)); 557 } 558 goto free_pci_mapping; 559 } 560 561 ice_init_device_features(sc); 562 563 /* Keep flag set by default */ 564 ice_set_state(&sc->state, ICE_STATE_LINK_ACTIVE_ON_DOWN); 565 566 /* Notify firmware of the device driver version */ 567 err = ice_send_version(sc); 568 if (err) 569 goto deinit_hw; 570 571 /* 572 * Success indicates a change was made that requires a reinitialization 573 * of the hardware 574 */ 575 err = ice_load_pkg_file(sc); 576 if (!err) { 577 ice_deinit_hw(hw); 578 goto reinit_hw; 579 } 580 581 err = ice_init_link_events(sc); 582 if (err) { 583 device_printf(dev, "ice_init_link_events failed: %s\n", 584 ice_err_str(err)); 585 goto deinit_hw; 586 } 587 588 /* Initialize VLAN mode in FW; if dual VLAN mode is supported by the package 589 * and firmware, this will force them to use single VLAN mode. 590 */ 591 status = ice_set_vlan_mode(hw); 592 if (status) { 593 err = EIO; 594 device_printf(dev, "Unable to initialize VLAN mode, err %s aq_err %s\n", 595 ice_status_str(status), 596 ice_aq_str(hw->adminq.sq_last_status)); 597 goto deinit_hw; 598 } 599 600 ice_print_nvm_version(sc); 601 602 /* Setup the MAC address */ 603 iflib_set_mac(ctx, hw->port_info->mac.lan_addr); 604 605 /* Setup the iflib softc context structure */ 606 ice_setup_scctx(sc); 607 608 /* Initialize the Tx queue manager */ 609 err = ice_resmgr_init(&sc->tx_qmgr, hw->func_caps.common_cap.num_txq); 610 if (err) { 611 device_printf(dev, "Unable to initialize Tx queue manager: %s\n", 612 ice_err_str(err)); 613 goto deinit_hw; 614 } 615 616 /* Initialize the Rx queue manager */ 617 err = ice_resmgr_init(&sc->rx_qmgr, hw->func_caps.common_cap.num_rxq); 618 if (err) { 619 device_printf(dev, "Unable to initialize Rx queue manager: %s\n", 620 ice_err_str(err)); 621 goto free_tx_qmgr; 622 } 623 624 /* Initialize the PF device interrupt resource manager */ 625 err = ice_alloc_intr_tracking(sc); 626 if (err) 627 /* Errors are already printed */ 628 goto free_rx_qmgr; 629 630 /* Determine maximum number of VSIs we'll prepare for */ 631 sc->num_available_vsi = min(ICE_MAX_VSI_AVAILABLE, 632 hw->func_caps.guar_num_vsi); 633 634 if (!sc->num_available_vsi) { 635 err = EIO; 636 device_printf(dev, "No VSIs allocated to host\n"); 637 goto free_intr_tracking; 638 } 639 640 /* Allocate storage for the VSI pointers */ 641 sc->all_vsi = (struct ice_vsi **) 642 malloc(sizeof(struct ice_vsi *) * sc->num_available_vsi, 643 M_ICE, M_WAITOK | M_ZERO); 644 if (!sc->all_vsi) { 645 err = ENOMEM; 646 device_printf(dev, "Unable to allocate VSI array\n"); 647 goto free_intr_tracking; 648 } 649 650 /* 651 * Prepare the statically allocated primary PF VSI in the softc 652 * structure. Other VSIs will be dynamically allocated as needed. 653 */ 654 ice_setup_pf_vsi(sc); 655 656 ice_alloc_vsi_qmap(&sc->pf_vsi, scctx->isc_ntxqsets_max, 657 scctx->isc_nrxqsets_max); 658 659 /* Allocate MSI-X vectors (due to isc_flags IFLIB_SKIP_MSIX) */ 660 err = ice_allocate_msix(sc); 661 if (err) 662 goto free_main_vsi; 663 664 return 0; 665 666 free_main_vsi: 667 /* ice_release_vsi will free the queue maps if they were allocated */ 668 ice_release_vsi(&sc->pf_vsi); 669 free(sc->all_vsi, M_ICE); 670 sc->all_vsi = NULL; 671 free_intr_tracking: 672 ice_free_intr_tracking(sc); 673 free_rx_qmgr: 674 ice_resmgr_destroy(&sc->rx_qmgr); 675 free_tx_qmgr: 676 ice_resmgr_destroy(&sc->tx_qmgr); 677 deinit_hw: 678 ice_deinit_hw(hw); 679 free_pci_mapping: 680 ice_free_pci_mapping(sc); 681 destroy_admin_timer: 682 mtx_lock(&sc->admin_mtx); 683 callout_stop(&sc->admin_timer); 684 mtx_unlock(&sc->admin_mtx); 685 mtx_destroy(&sc->admin_mtx); 686 return err; 687 } /* ice_if_attach_pre */ 688 689 /** 690 * ice_attach_pre_recovery_mode - Limited driver attach_pre for FW recovery 691 * @sc: the device private softc 692 * 693 * Loads the device driver in limited Firmware Recovery mode, intended to 694 * allow users to update the firmware to attempt to recover the device. 695 * 696 * @remark We may enter recovery mode in case either (a) the firmware is 697 * detected to be in an invalid state and must be re-programmed, or (b) the 698 * driver detects that the loaded firmware has a non-compatible API version 699 * that the driver cannot operate with. 700 */ 701 static int 702 ice_attach_pre_recovery_mode(struct ice_softc *sc) 703 { 704 ice_set_state(&sc->state, ICE_STATE_RECOVERY_MODE); 705 706 /* Setup the iflib softc context */ 707 ice_setup_scctx(sc); 708 709 /* Setup the PF VSI back pointer */ 710 sc->pf_vsi.sc = sc; 711 712 /* 713 * We still need to allocate MSI-X vectors since we need one vector to 714 * run the administrative admin interrupt 715 */ 716 return ice_allocate_msix(sc); 717 } 718 719 /** 720 * ice_update_link_status - notify OS of link state change 721 * @sc: device private softc structure 722 * @update_media: true if we should update media even if link didn't change 723 * 724 * Called to notify iflib core of link status changes. Should be called once 725 * during attach_post, and whenever link status changes during runtime. 726 * 727 * This call only updates the currently supported media types if the link 728 * status changed, or if update_media is set to true. 729 */ 730 static void 731 ice_update_link_status(struct ice_softc *sc, bool update_media) 732 { 733 struct ice_hw *hw = &sc->hw; 734 int status; 735 736 /* Never report link up when in recovery mode */ 737 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 738 return; 739 740 /* Report link status to iflib only once each time it changes */ 741 if (!ice_testandset_state(&sc->state, ICE_STATE_LINK_STATUS_REPORTED)) { 742 if (sc->link_up) { /* link is up */ 743 uint64_t baudrate = ice_aq_speed_to_rate(sc->hw.port_info); 744 745 if (!(hw->port_info->phy.link_info_old.link_info & ICE_AQ_LINK_UP)) 746 ice_set_default_local_lldp_mib(sc); 747 748 iflib_link_state_change(sc->ctx, LINK_STATE_UP, baudrate); 749 ice_rdma_link_change(sc, LINK_STATE_UP, baudrate); 750 751 ice_link_up_msg(sc); 752 } else { /* link is down */ 753 iflib_link_state_change(sc->ctx, LINK_STATE_DOWN, 0); 754 ice_rdma_link_change(sc, LINK_STATE_DOWN, 0); 755 } 756 #ifdef PCI_IOV 757 ice_vc_notify_all_vfs_link_state(sc); 758 #endif 759 update_media = true; 760 } 761 762 /* Update the supported media types */ 763 if (update_media && !ice_test_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET)) { 764 status = ice_add_media_types(sc, sc->media); 765 if (status) 766 device_printf(sc->dev, "Error adding device media types: %s aq_err %s\n", 767 ice_status_str(status), 768 ice_aq_str(hw->adminq.sq_last_status)); 769 } 770 } 771 772 /** 773 * ice_if_attach_post - Late device attach logic 774 * @ctx: the iflib context structure 775 * 776 * Called by iflib to finish up attaching the device. Performs any attach 777 * logic which must wait until after the Tx and Rx queues have been 778 * allocated. 779 */ 780 static int 781 ice_if_attach_post(if_ctx_t ctx) 782 { 783 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 784 if_t ifp = iflib_get_ifp(ctx); 785 int status; 786 int err; 787 788 ASSERT_CTX_LOCKED(sc); 789 790 /* We don't yet support loading if MSI-X is not supported */ 791 if (sc->scctx->isc_intr != IFLIB_INTR_MSIX) { 792 device_printf(sc->dev, "The ice driver does not support loading without MSI-X\n"); 793 return (ENOTSUP); 794 } 795 796 /* The ifnet structure hasn't yet been initialized when the attach_pre 797 * handler is called, so wait until attach_post to setup the 798 * isc_max_frame_size. 799 */ 800 sc->scctx->isc_max_frame_size = if_getmtu(ifp) + 801 ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN; 802 803 /* 804 * If we are in recovery mode, only perform a limited subset of 805 * initialization to support NVM recovery. 806 */ 807 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) { 808 ice_attach_post_recovery_mode(sc); 809 return (0); 810 } 811 812 sc->pf_vsi.max_frame_size = sc->scctx->isc_max_frame_size; 813 814 err = ice_initialize_vsi(&sc->pf_vsi); 815 if (err) { 816 device_printf(sc->dev, "Unable to initialize Main VSI: %s\n", 817 ice_err_str(err)); 818 return err; 819 } 820 821 /* Enable FW health event reporting */ 822 ice_init_health_events(sc); 823 824 /* Configure the main PF VSI for RSS */ 825 err = ice_config_rss(&sc->pf_vsi); 826 if (err) { 827 device_printf(sc->dev, 828 "Unable to configure RSS for the main VSI, err %s\n", 829 ice_err_str(err)); 830 return err; 831 } 832 833 /* Configure switch to drop transmitted LLDP and PAUSE frames */ 834 err = ice_cfg_pf_ethertype_filters(sc); 835 if (err) 836 return err; 837 838 ice_get_and_print_bus_info(sc); 839 840 ice_set_link_management_mode(sc); 841 842 ice_init_saved_phy_cfg(sc); 843 844 ice_cfg_pba_num(sc); 845 846 /* Set a default value for PFC mode on attach since the FW state is unknown 847 * before sysctl tunables are executed and it can't be queried. This fixes an 848 * issue when loading the driver with the FW LLDP agent enabled but the FW 849 * was previously in DSCP PFC mode. 850 */ 851 status = ice_aq_set_pfc_mode(&sc->hw, ICE_AQC_PFC_VLAN_BASED_PFC, NULL); 852 if (status) 853 device_printf(sc->dev, "Setting pfc mode failed, status %s\n", ice_status_str(status)); 854 855 ice_add_device_sysctls(sc); 856 857 #ifdef PCI_IOV 858 if (ice_is_bit_set(sc->feat_cap, ICE_FEATURE_SRIOV)) { 859 err = ice_iov_attach(sc); 860 if (err == ENOMEM) 861 return (err); 862 } 863 #endif /* PCI_IOV */ 864 865 /* Get DCBX/LLDP state and start DCBX agent */ 866 ice_init_dcb_setup(sc); 867 868 /* Setup link, if PHY FW is ready */ 869 ice_init_link(sc); 870 871 /* Configure interrupt causes for the administrative interrupt */ 872 ice_configure_misc_interrupts(sc); 873 874 /* Enable ITR 0 right away, so that we can handle admin interrupts */ 875 ice_enable_intr(&sc->hw, sc->irqvs[0].me); 876 877 err = ice_rdma_pf_attach(sc); 878 if (err) 879 return (err); 880 881 /* Start the admin timer */ 882 mtx_lock(&sc->admin_mtx); 883 callout_reset(&sc->admin_timer, hz/2, ice_admin_timer, sc); 884 mtx_unlock(&sc->admin_mtx); 885 886 if (ice_test_state(&sc->state, ICE_STATE_LINK_ACTIVE_ON_DOWN) && 887 !ice_test_state(&sc->state, ICE_STATE_NO_MEDIA)) 888 ice_set_state(&sc->state, ICE_STATE_FIRST_INIT_LINK); 889 890 ice_clear_state(&sc->state, ICE_STATE_ATTACHING); 891 892 return 0; 893 } /* ice_if_attach_post */ 894 895 /** 896 * ice_attach_post_recovery_mode - Limited driver attach_post for FW recovery 897 * @sc: the device private softc 898 * 899 * Performs minimal work to prepare the driver to recover an NVM in case the 900 * firmware is in recovery mode. 901 */ 902 static void 903 ice_attach_post_recovery_mode(struct ice_softc *sc) 904 { 905 /* Configure interrupt causes for the administrative interrupt */ 906 ice_configure_misc_interrupts(sc); 907 908 /* Enable ITR 0 right away, so that we can handle admin interrupts */ 909 ice_enable_intr(&sc->hw, sc->irqvs[0].me); 910 911 /* Start the admin timer */ 912 mtx_lock(&sc->admin_mtx); 913 callout_reset(&sc->admin_timer, hz/2, ice_admin_timer, sc); 914 mtx_unlock(&sc->admin_mtx); 915 916 ice_clear_state(&sc->state, ICE_STATE_ATTACHING); 917 } 918 919 /** 920 * ice_free_irqvs - Free IRQ vector memory 921 * @sc: the device private softc structure 922 * 923 * Free IRQ vector memory allocated during ice_if_msix_intr_assign. 924 */ 925 static void 926 ice_free_irqvs(struct ice_softc *sc) 927 { 928 struct ice_vsi *vsi = &sc->pf_vsi; 929 if_ctx_t ctx = sc->ctx; 930 int i; 931 932 /* If the irqvs array is NULL, then there are no vectors to free */ 933 if (sc->irqvs == NULL) 934 return; 935 936 /* Free the IRQ vectors */ 937 for (i = 0; i < sc->num_irq_vectors; i++) 938 iflib_irq_free(ctx, &sc->irqvs[i].irq); 939 940 /* Clear the irqv pointers */ 941 for (i = 0; i < vsi->num_rx_queues; i++) 942 vsi->rx_queues[i].irqv = NULL; 943 944 for (i = 0; i < vsi->num_tx_queues; i++) 945 vsi->tx_queues[i].irqv = NULL; 946 947 /* Release the vector array memory */ 948 free(sc->irqvs, M_ICE); 949 sc->irqvs = NULL; 950 sc->num_irq_vectors = 0; 951 } 952 953 /** 954 * ice_if_detach - Device driver detach logic 955 * @ctx: iflib context structure 956 * 957 * Perform device shutdown logic to detach the device driver. 958 * 959 * Note that there is no guarantee of the ordering of ice_if_queues_free() and 960 * ice_if_detach(). It is possible for the functions to be called in either 961 * order, and they must not assume to have a strict ordering. 962 */ 963 static int 964 ice_if_detach(if_ctx_t ctx) 965 { 966 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 967 struct ice_vsi *vsi = &sc->pf_vsi; 968 int status; 969 int i; 970 971 ASSERT_CTX_LOCKED(sc); 972 973 /* Indicate that we're detaching */ 974 ice_set_state(&sc->state, ICE_STATE_DETACHING); 975 976 /* Stop the admin timer */ 977 mtx_lock(&sc->admin_mtx); 978 callout_stop(&sc->admin_timer); 979 mtx_unlock(&sc->admin_mtx); 980 mtx_destroy(&sc->admin_mtx); 981 982 /* Remove additional interfaces if they exist */ 983 if (sc->mirr_if) 984 ice_destroy_mirror_interface(sc); 985 ice_rdma_pf_detach(sc); 986 987 #ifdef PCI_IOV 988 if (ice_is_bit_set(sc->feat_cap, ICE_FEATURE_SRIOV)) 989 ice_iov_detach(sc); 990 #endif /* PCI_IOV */ 991 992 /* Free allocated media types */ 993 ifmedia_removeall(sc->media); 994 995 /* Free the Tx and Rx sysctl contexts, and assign NULL to the node 996 * pointers. Note, the calls here and those in ice_if_queues_free() 997 * are *BOTH* necessary, as we cannot guarantee which path will be 998 * run first 999 */ 1000 ice_vsi_del_txqs_ctx(vsi); 1001 ice_vsi_del_rxqs_ctx(vsi); 1002 1003 /* Release MSI-X resources */ 1004 ice_free_irqvs(sc); 1005 1006 for (i = 0; i < sc->num_available_vsi; i++) { 1007 if (sc->all_vsi[i]) 1008 ice_release_vsi(sc->all_vsi[i]); 1009 } 1010 1011 if (sc->all_vsi) { 1012 free(sc->all_vsi, M_ICE); 1013 sc->all_vsi = NULL; 1014 } 1015 1016 /* Release MSI-X memory */ 1017 pci_release_msi(sc->dev); 1018 1019 if (sc->msix_table != NULL) { 1020 bus_release_resource(sc->dev, SYS_RES_MEMORY, 1021 rman_get_rid(sc->msix_table), 1022 sc->msix_table); 1023 sc->msix_table = NULL; 1024 } 1025 1026 ice_free_intr_tracking(sc); 1027 1028 /* Destroy the queue managers */ 1029 ice_resmgr_destroy(&sc->tx_qmgr); 1030 ice_resmgr_destroy(&sc->rx_qmgr); 1031 1032 if (!ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1033 ice_deinit_hw(&sc->hw); 1034 1035 IFLIB_CTX_UNLOCK(sc); 1036 status = ice_reset(&sc->hw, ICE_RESET_PFR); 1037 IFLIB_CTX_LOCK(sc); 1038 if (status) { 1039 device_printf(sc->dev, "device PF reset failed, err %s\n", 1040 ice_status_str(status)); 1041 } 1042 1043 ice_free_pci_mapping(sc); 1044 1045 return 0; 1046 } /* ice_if_detach */ 1047 1048 /** 1049 * ice_if_tx_queues_alloc - Allocate Tx queue memory 1050 * @ctx: iflib context structure 1051 * @vaddrs: virtual addresses for the queue memory 1052 * @paddrs: physical addresses for the queue memory 1053 * @ntxqs: the number of Tx queues per set (should always be 1) 1054 * @ntxqsets: the number of Tx queue sets to allocate 1055 * 1056 * Called by iflib to allocate Tx queues for the device. Allocates driver 1057 * memory to track each queue, the status arrays used for descriptor 1058 * status reporting, and Tx queue sysctls. 1059 */ 1060 static int 1061 ice_if_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, 1062 int __invariant_only ntxqs, int ntxqsets) 1063 { 1064 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1065 struct ice_vsi *vsi = &sc->pf_vsi; 1066 struct ice_tx_queue *txq; 1067 int err, i, j; 1068 1069 MPASS(ntxqs == 1); 1070 MPASS(sc->scctx->isc_ntxd[0] <= ICE_MAX_DESC_COUNT); 1071 ASSERT_CTX_LOCKED(sc); 1072 1073 /* Do not bother allocating queues if we're in recovery mode */ 1074 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1075 return (0); 1076 1077 /* Allocate queue structure memory */ 1078 if (!(vsi->tx_queues = 1079 (struct ice_tx_queue *) malloc(sizeof(struct ice_tx_queue) * ntxqsets, M_ICE, M_NOWAIT | M_ZERO))) { 1080 device_printf(sc->dev, "Unable to allocate Tx queue memory\n"); 1081 return (ENOMEM); 1082 } 1083 1084 /* Allocate report status arrays */ 1085 for (i = 0, txq = vsi->tx_queues; i < ntxqsets; i++, txq++) { 1086 if (!(txq->tx_rsq = 1087 (uint16_t *) malloc(sizeof(uint16_t) * sc->scctx->isc_ntxd[0], M_ICE, M_NOWAIT))) { 1088 device_printf(sc->dev, "Unable to allocate tx_rsq memory\n"); 1089 err = ENOMEM; 1090 goto free_tx_queues; 1091 } 1092 /* Initialize report status array */ 1093 for (j = 0; j < sc->scctx->isc_ntxd[0]; j++) 1094 txq->tx_rsq[j] = QIDX_INVALID; 1095 } 1096 1097 /* Assign queues from PF space to the main VSI */ 1098 err = ice_resmgr_assign_contiguous(&sc->tx_qmgr, vsi->tx_qmap, ntxqsets); 1099 if (err) { 1100 device_printf(sc->dev, "Unable to assign PF queues: %s\n", 1101 ice_err_str(err)); 1102 goto free_tx_queues; 1103 } 1104 vsi->qmap_type = ICE_RESMGR_ALLOC_CONTIGUOUS; 1105 1106 /* Add Tx queue sysctls context */ 1107 ice_vsi_add_txqs_ctx(vsi); 1108 1109 for (i = 0, txq = vsi->tx_queues; i < ntxqsets; i++, txq++) { 1110 /* q_handle == me when only one TC */ 1111 txq->me = txq->q_handle = i; 1112 txq->vsi = vsi; 1113 1114 /* store the queue size for easier access */ 1115 txq->desc_count = sc->scctx->isc_ntxd[0]; 1116 1117 /* get the virtual and physical address of the hardware queues */ 1118 txq->tail = QTX_COMM_DBELL(vsi->tx_qmap[i]); 1119 txq->tx_base = (struct ice_tx_desc *)vaddrs[i]; 1120 txq->tx_paddr = paddrs[i]; 1121 1122 ice_add_txq_sysctls(txq); 1123 } 1124 1125 vsi->num_tx_queues = ntxqsets; 1126 1127 return (0); 1128 1129 free_tx_queues: 1130 for (i = 0, txq = vsi->tx_queues; i < ntxqsets; i++, txq++) { 1131 if (txq->tx_rsq != NULL) { 1132 free(txq->tx_rsq, M_ICE); 1133 txq->tx_rsq = NULL; 1134 } 1135 } 1136 free(vsi->tx_queues, M_ICE); 1137 vsi->tx_queues = NULL; 1138 return err; 1139 } 1140 1141 /** 1142 * ice_if_rx_queues_alloc - Allocate Rx queue memory 1143 * @ctx: iflib context structure 1144 * @vaddrs: virtual addresses for the queue memory 1145 * @paddrs: physical addresses for the queue memory 1146 * @nrxqs: number of Rx queues per set (should always be 1) 1147 * @nrxqsets: number of Rx queue sets to allocate 1148 * 1149 * Called by iflib to allocate Rx queues for the device. Allocates driver 1150 * memory to track each queue, as well as sets up the Rx queue sysctls. 1151 */ 1152 static int 1153 ice_if_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, 1154 int __invariant_only nrxqs, int nrxqsets) 1155 { 1156 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1157 struct ice_vsi *vsi = &sc->pf_vsi; 1158 struct ice_rx_queue *rxq; 1159 int err, i; 1160 1161 MPASS(nrxqs == 1); 1162 MPASS(sc->scctx->isc_nrxd[0] <= ICE_MAX_DESC_COUNT); 1163 ASSERT_CTX_LOCKED(sc); 1164 1165 /* Do not bother allocating queues if we're in recovery mode */ 1166 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1167 return (0); 1168 1169 /* Allocate queue structure memory */ 1170 if (!(vsi->rx_queues = 1171 (struct ice_rx_queue *) malloc(sizeof(struct ice_rx_queue) * nrxqsets, M_ICE, M_NOWAIT | M_ZERO))) { 1172 device_printf(sc->dev, "Unable to allocate Rx queue memory\n"); 1173 return (ENOMEM); 1174 } 1175 1176 /* Assign queues from PF space to the main VSI */ 1177 err = ice_resmgr_assign_contiguous(&sc->rx_qmgr, vsi->rx_qmap, nrxqsets); 1178 if (err) { 1179 device_printf(sc->dev, "Unable to assign PF queues: %s\n", 1180 ice_err_str(err)); 1181 goto free_rx_queues; 1182 } 1183 vsi->qmap_type = ICE_RESMGR_ALLOC_CONTIGUOUS; 1184 1185 /* Add Rx queue sysctls context */ 1186 ice_vsi_add_rxqs_ctx(vsi); 1187 1188 for (i = 0, rxq = vsi->rx_queues; i < nrxqsets; i++, rxq++) { 1189 rxq->me = i; 1190 rxq->vsi = vsi; 1191 1192 /* store the queue size for easier access */ 1193 rxq->desc_count = sc->scctx->isc_nrxd[0]; 1194 1195 /* get the virtual and physical address of the hardware queues */ 1196 rxq->tail = QRX_TAIL(vsi->rx_qmap[i]); 1197 rxq->rx_base = (union ice_32b_rx_flex_desc *)vaddrs[i]; 1198 rxq->rx_paddr = paddrs[i]; 1199 1200 ice_add_rxq_sysctls(rxq); 1201 } 1202 1203 vsi->num_rx_queues = nrxqsets; 1204 1205 return (0); 1206 1207 free_rx_queues: 1208 free(vsi->rx_queues, M_ICE); 1209 vsi->rx_queues = NULL; 1210 return err; 1211 } 1212 1213 /** 1214 * ice_if_queues_free - Free queue memory 1215 * @ctx: the iflib context structure 1216 * 1217 * Free queue memory allocated by ice_if_tx_queues_alloc() and 1218 * ice_if_rx_queues_alloc(). 1219 * 1220 * There is no guarantee that ice_if_queues_free() and ice_if_detach() will be 1221 * called in the same order. It's possible for ice_if_queues_free() to be 1222 * called prior to ice_if_detach(), and vice versa. 1223 * 1224 * For this reason, the main VSI is a static member of the ice_softc, which is 1225 * not free'd until after iflib finishes calling both of these functions. 1226 * 1227 * Thus, care must be taken in how we manage the memory being freed by this 1228 * function, and in what tasks it can and must perform. 1229 */ 1230 static void 1231 ice_if_queues_free(if_ctx_t ctx) 1232 { 1233 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1234 struct ice_vsi *vsi = &sc->pf_vsi; 1235 struct ice_tx_queue *txq; 1236 int i; 1237 1238 /* Free the Tx and Rx sysctl contexts, and assign NULL to the node 1239 * pointers. Note, the calls here and those in ice_if_detach() 1240 * are *BOTH* necessary, as we cannot guarantee which path will be 1241 * run first 1242 */ 1243 ice_vsi_del_txqs_ctx(vsi); 1244 ice_vsi_del_rxqs_ctx(vsi); 1245 1246 /* Release MSI-X IRQ vectors, if not yet released in ice_if_detach */ 1247 ice_free_irqvs(sc); 1248 1249 if (vsi->tx_queues != NULL) { 1250 /* free the tx_rsq arrays */ 1251 for (i = 0, txq = vsi->tx_queues; i < vsi->num_tx_queues; i++, txq++) { 1252 if (txq->tx_rsq != NULL) { 1253 free(txq->tx_rsq, M_ICE); 1254 txq->tx_rsq = NULL; 1255 } 1256 } 1257 free(vsi->tx_queues, M_ICE); 1258 vsi->tx_queues = NULL; 1259 vsi->num_tx_queues = 0; 1260 } 1261 if (vsi->rx_queues != NULL) { 1262 free(vsi->rx_queues, M_ICE); 1263 vsi->rx_queues = NULL; 1264 vsi->num_rx_queues = 0; 1265 } 1266 } 1267 1268 /** 1269 * ice_msix_que - Fast interrupt handler for MSI-X receive queues 1270 * @arg: The Rx queue memory 1271 * 1272 * Interrupt filter function for iflib MSI-X interrupts. Called by iflib when 1273 * an MSI-X interrupt for a given queue is triggered. Currently this just asks 1274 * iflib to schedule the main Rx thread. 1275 */ 1276 static int 1277 ice_msix_que(void *arg) 1278 { 1279 struct ice_rx_queue __unused *rxq = (struct ice_rx_queue *)arg; 1280 1281 /* TODO: dynamic ITR algorithm?? */ 1282 1283 return (FILTER_SCHEDULE_THREAD); 1284 } 1285 1286 /** 1287 * ice_msix_admin - Fast interrupt handler for MSI-X admin interrupt 1288 * @arg: pointer to device softc memory 1289 * 1290 * Called by iflib when an administrative interrupt occurs. Should perform any 1291 * fast logic for handling the interrupt cause, and then indicate whether the 1292 * admin task needs to be queued. 1293 */ 1294 static int 1295 ice_msix_admin(void *arg) 1296 { 1297 struct ice_softc *sc = (struct ice_softc *)arg; 1298 struct ice_hw *hw = &sc->hw; 1299 device_t dev = sc->dev; 1300 u32 oicr; 1301 1302 /* There is no safe way to modify the enabled miscellaneous causes of 1303 * the OICR vector at runtime, as doing so would be prone to race 1304 * conditions. Reading PFINT_OICR will unmask the associated interrupt 1305 * causes and allow future interrupts to occur. The admin interrupt 1306 * vector will not be re-enabled until after we exit this function, 1307 * but any delayed tasks must be resilient against possible "late 1308 * arrival" interrupts that occur while we're already handling the 1309 * task. This is done by using state bits and serializing these 1310 * delayed tasks via the admin status task function. 1311 */ 1312 oicr = rd32(hw, PFINT_OICR); 1313 1314 /* Processing multiple controlq interrupts on a single vector does not 1315 * provide an indication of which controlq triggered the interrupt. 1316 * We might try reading the INTEVENT bit of the respective PFINT_*_CTL 1317 * registers. However, the INTEVENT bit is not guaranteed to be set as 1318 * it gets automatically cleared when the hardware acknowledges the 1319 * interrupt. 1320 * 1321 * This means we don't really have a good indication of whether or 1322 * which controlq triggered this interrupt. We'll just notify the 1323 * admin task that it should check all the controlqs. 1324 */ 1325 ice_set_state(&sc->state, ICE_STATE_CONTROLQ_EVENT_PENDING); 1326 1327 if (oicr & PFINT_OICR_VFLR_M) { 1328 ice_set_state(&sc->state, ICE_STATE_VFLR_PENDING); 1329 } 1330 1331 if (oicr & PFINT_OICR_MAL_DETECT_M) { 1332 ice_set_state(&sc->state, ICE_STATE_MDD_PENDING); 1333 } 1334 1335 if (oicr & PFINT_OICR_GRST_M) { 1336 u32 reset; 1337 1338 reset = (rd32(hw, GLGEN_RSTAT) & GLGEN_RSTAT_RESET_TYPE_M) >> 1339 GLGEN_RSTAT_RESET_TYPE_S; 1340 1341 if (reset == ICE_RESET_CORER) 1342 sc->soft_stats.corer_count++; 1343 else if (reset == ICE_RESET_GLOBR) 1344 sc->soft_stats.globr_count++; 1345 else 1346 sc->soft_stats.empr_count++; 1347 1348 /* There are a couple of bits at play for handling resets. 1349 * First, the ICE_STATE_RESET_OICR_RECV bit is used to 1350 * indicate that the driver has received an OICR with a reset 1351 * bit active, indicating that a CORER/GLOBR/EMPR is about to 1352 * happen. Second, we set hw->reset_ongoing to indicate that 1353 * the hardware is in reset. We will set this back to false as 1354 * soon as the driver has determined that the hardware is out 1355 * of reset. 1356 * 1357 * If the driver wishes to trigger a request, it can set one of 1358 * the ICE_STATE_RESET_*_REQ bits, which will trigger the 1359 * correct type of reset. 1360 */ 1361 if (!ice_testandset_state(&sc->state, ICE_STATE_RESET_OICR_RECV)) { 1362 hw->reset_ongoing = true; 1363 /* 1364 * During the NVM update process, there is a driver reset and link 1365 * goes down and then up. The below if-statement prevents a second 1366 * link flap from occurring in ice_if_init(). 1367 */ 1368 if (if_getflags(sc->ifp) & IFF_UP) 1369 ice_set_state(&sc->state, ICE_STATE_FIRST_INIT_LINK); 1370 } 1371 } 1372 1373 if (oicr & PFINT_OICR_ECC_ERR_M) { 1374 device_printf(dev, "ECC Error detected!\n"); 1375 ice_set_state(&sc->state, ICE_STATE_RESET_PFR_REQ); 1376 } 1377 1378 if (oicr & (PFINT_OICR_PE_CRITERR_M | PFINT_OICR_HMC_ERR_M)) { 1379 if (oicr & PFINT_OICR_HMC_ERR_M) 1380 /* Log the HMC errors */ 1381 ice_log_hmc_error(hw, dev); 1382 ice_rdma_notify_pe_intr(sc, oicr); 1383 } 1384 1385 if (oicr & PFINT_OICR_PCI_EXCEPTION_M) { 1386 device_printf(dev, "PCI Exception detected!\n"); 1387 ice_set_state(&sc->state, ICE_STATE_RESET_PFR_REQ); 1388 } 1389 1390 return (FILTER_SCHEDULE_THREAD); 1391 } 1392 1393 /** 1394 * ice_allocate_msix - Allocate MSI-X vectors for the interface 1395 * @sc: the device private softc 1396 * 1397 * Map the MSI-X bar, and then request MSI-X vectors in a two-stage process. 1398 * 1399 * First, determine a suitable total number of vectors based on the number 1400 * of CPUs, RSS buckets, the administrative vector, and other demands such as 1401 * RDMA. 1402 * 1403 * Request the desired amount of vectors, and see how many we obtain. If we 1404 * don't obtain as many as desired, reduce the demands by lowering the number 1405 * of requested queues or reducing the demand from other features such as 1406 * RDMA. 1407 * 1408 * @remark This function is required because the driver sets the 1409 * IFLIB_SKIP_MSIX flag indicating that the driver will manage MSI-X vectors 1410 * manually. 1411 * 1412 * @remark This driver will only use MSI-X vectors. If this is not possible, 1413 * neither MSI or legacy interrupts will be tried. 1414 * 1415 * @remark if it exists, os_imgr is initialized here for keeping track of 1416 * the assignments of extra MSIX vectors. 1417 * 1418 * @post on success this function must set the following scctx parameters: 1419 * isc_vectors, isc_nrxqsets, isc_ntxqsets, and isc_intr. 1420 * 1421 * @returns zero on success or an error code on failure. 1422 */ 1423 static int 1424 ice_allocate_msix(struct ice_softc *sc) 1425 { 1426 bool iflib_override_queue_count = false; 1427 if_softc_ctx_t scctx = sc->scctx; 1428 device_t dev = sc->dev; 1429 cpuset_t cpus; 1430 int bar, queues, vectors, requested; 1431 int err = 0; 1432 int rdma; 1433 1434 /* Allocate the MSI-X bar */ 1435 bar = scctx->isc_msix_bar; 1436 sc->msix_table = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &bar, RF_ACTIVE); 1437 if (!sc->msix_table) { 1438 device_printf(dev, "Unable to map MSI-X table\n"); 1439 return (ENOMEM); 1440 } 1441 1442 /* Check if the iflib queue count sysctls have been set */ 1443 if (sc->ifc_sysctl_ntxqs || sc->ifc_sysctl_nrxqs) 1444 iflib_override_queue_count = true; 1445 1446 err = bus_get_cpus(dev, INTR_CPUS, sizeof(cpus), &cpus); 1447 if (err) { 1448 device_printf(dev, "%s: Unable to fetch the CPU list: %s\n", 1449 __func__, ice_err_str(err)); 1450 CPU_COPY(&all_cpus, &cpus); 1451 } 1452 1453 /* Attempt to mimic behavior of iflib_msix_init */ 1454 if (iflib_override_queue_count) { 1455 /* 1456 * If the override sysctls have been set, limit the queues to 1457 * the number of logical CPUs. 1458 */ 1459 queues = mp_ncpus; 1460 } else { 1461 /* 1462 * Otherwise, limit the queue count to the CPUs associated 1463 * with the NUMA node the device is associated with. 1464 */ 1465 queues = CPU_COUNT(&cpus); 1466 } 1467 1468 /* Clamp to the number of RSS buckets */ 1469 queues = imin(queues, rss_getnumbuckets()); 1470 1471 /* 1472 * Clamp the number of queue pairs to the minimum of the requested Tx 1473 * and Rx queues. 1474 */ 1475 queues = imin(queues, sc->ifc_sysctl_ntxqs ?: scctx->isc_ntxqsets); 1476 queues = imin(queues, sc->ifc_sysctl_nrxqs ?: scctx->isc_nrxqsets); 1477 1478 if (ice_is_bit_set(sc->feat_cap, ICE_FEATURE_RDMA)) { 1479 /* 1480 * Choose a number of RDMA vectors based on the number of CPUs 1481 * up to a maximum 1482 */ 1483 rdma = min(CPU_COUNT(&cpus), ICE_RDMA_MAX_MSIX); 1484 1485 /* Further limit by the user configurable tunable */ 1486 rdma = min(rdma, ice_rdma_max_msix); 1487 } else { 1488 rdma = 0; 1489 } 1490 1491 /* 1492 * Determine the number of vectors to request. Note that we also need 1493 * to allocate one vector for administrative tasks. 1494 */ 1495 requested = rdma + queues + 1; 1496 /* Add extra vectors requested by the user for later subinterface 1497 * creation. 1498 */ 1499 if_ctx_t ctx = sc->ctx; 1500 u32 extra_vectors = iflib_get_extra_msix_vectors_sysctl(ctx); 1501 requested += extra_vectors; 1502 1503 vectors = requested; 1504 err = pci_alloc_msix(dev, &vectors); 1505 if (err) { 1506 device_printf(dev, "Failed to allocate %d MSI-X vectors, err %s\n", 1507 vectors, ice_err_str(err)); 1508 goto err_free_msix_table; 1509 } 1510 1511 /* If we don't receive enough vectors, reduce demands */ 1512 if (vectors < requested) { 1513 int diff = requested - vectors; 1514 1515 device_printf(dev, "Requested %d MSI-X vectors, but got only %d\n", 1516 requested, vectors); 1517 1518 diff += extra_vectors; 1519 extra_vectors = 0; 1520 /* 1521 * The OS didn't grant us the requested number of vectors. 1522 * Check to see if we can reduce demands by limiting the 1523 * number of vectors allocated to certain features. 1524 */ 1525 1526 if (rdma >= diff) { 1527 /* Reduce the number of RDMA vectors we reserve */ 1528 rdma -= diff; 1529 diff = 0; 1530 } else { 1531 /* Disable RDMA and reduce the difference */ 1532 ice_clear_bit(ICE_FEATURE_RDMA, sc->feat_cap); 1533 diff -= rdma; 1534 rdma = 0; 1535 } 1536 1537 /* 1538 * If we still have a difference, we need to reduce the number 1539 * of queue pairs. 1540 * 1541 * However, we still need at least one vector for the admin 1542 * interrupt and one queue pair. 1543 */ 1544 if (queues <= diff) { 1545 device_printf(dev, "Unable to allocate sufficient MSI-X vectors\n"); 1546 err = (ERANGE); 1547 goto err_pci_release_msi; 1548 } 1549 1550 queues -= diff; 1551 } 1552 1553 device_printf(dev, "Using %d Tx and Rx queues\n", queues); 1554 if (rdma) 1555 device_printf(dev, "Reserving %d MSI-X interrupts for iRDMA\n", 1556 rdma); 1557 device_printf(dev, "Using MSI-X interrupts with %d vectors\n", 1558 vectors); 1559 1560 /* Split resulting vectors back into requested splits */ 1561 scctx->isc_vectors = vectors; 1562 scctx->isc_nrxqsets = queues; 1563 scctx->isc_ntxqsets = queues; 1564 scctx->isc_intr = IFLIB_INTR_MSIX; 1565 1566 sc->irdma_vectors = rdma; 1567 1568 /* Interrupt allocation tracking isn't required in recovery mode, 1569 * since neither RDMA nor VFs are enabled. 1570 */ 1571 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1572 return (0); 1573 1574 /* Keep track of which interrupt indices are being used for what */ 1575 sc->lan_vectors = vectors - rdma; 1576 sc->lan_vectors -= extra_vectors; 1577 err = ice_resmgr_assign_contiguous(&sc->dev_imgr, sc->pf_imap, sc->lan_vectors); 1578 if (err) { 1579 device_printf(dev, "Unable to assign PF interrupt mapping: %s\n", 1580 ice_err_str(err)); 1581 goto err_pci_release_msi; 1582 } 1583 err = ice_resmgr_assign_contiguous(&sc->dev_imgr, sc->rdma_imap, rdma); 1584 if (err) { 1585 device_printf(dev, "Unable to assign PF RDMA interrupt mapping: %s\n", 1586 ice_err_str(err)); 1587 goto err_release_pf_imap; 1588 } 1589 sc->extra_vectors = extra_vectors; 1590 /* Setup another resource manager to track the assignments of extra OS 1591 * vectors. These OS interrupt allocations don't need to be contiguous, 1592 * unlike the ones that come from the device. 1593 */ 1594 err = ice_resmgr_init(&sc->os_imgr, sc->extra_vectors); 1595 if (err) { 1596 device_printf(dev, "Unable to initialize OS extra interrupt manager: %s\n", 1597 ice_err_str(err)); 1598 ice_resmgr_release_map(&sc->dev_imgr, sc->rdma_imap, 1599 rdma); 1600 goto err_release_pf_imap; 1601 } 1602 return (0); 1603 1604 err_release_pf_imap: 1605 ice_resmgr_release_map(&sc->dev_imgr, sc->pf_imap, 1606 sc->lan_vectors); 1607 err_pci_release_msi: 1608 pci_release_msi(dev); 1609 err_free_msix_table: 1610 if (sc->msix_table != NULL) { 1611 bus_release_resource(sc->dev, SYS_RES_MEMORY, 1612 rman_get_rid(sc->msix_table), 1613 sc->msix_table); 1614 sc->msix_table = NULL; 1615 } 1616 1617 return (err); 1618 } 1619 1620 /** 1621 * ice_if_msix_intr_assign - Assign MSI-X interrupt vectors to queues 1622 * @ctx: the iflib context structure 1623 * @msix: the number of vectors we were assigned 1624 * 1625 * Called by iflib to assign MSI-X vectors to queues. Currently requires that 1626 * we get at least the same number of vectors as we have queues, and that we 1627 * always have the same number of Tx and Rx queues. 1628 * 1629 * Tx queues use a softirq instead of using their own hardware interrupt. 1630 */ 1631 static int 1632 ice_if_msix_intr_assign(if_ctx_t ctx, int msix) 1633 { 1634 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1635 struct ice_vsi *vsi = &sc->pf_vsi; 1636 int err, i, vector; 1637 1638 ASSERT_CTX_LOCKED(sc); 1639 1640 if (vsi->num_rx_queues != vsi->num_tx_queues) { 1641 device_printf(sc->dev, 1642 "iflib requested %d Tx queues, and %d Rx queues, but the driver isn't able to support a differing number of Tx and Rx queues\n", 1643 vsi->num_tx_queues, vsi->num_rx_queues); 1644 return (EOPNOTSUPP); 1645 } 1646 1647 if (msix < (vsi->num_rx_queues + 1)) { 1648 device_printf(sc->dev, 1649 "Not enough MSI-X vectors to assign one vector to each queue pair\n"); 1650 return (EOPNOTSUPP); 1651 } 1652 1653 /* Save the number of vectors for future use */ 1654 sc->num_irq_vectors = vsi->num_rx_queues + 1; 1655 1656 /* Allocate space to store the IRQ vector data */ 1657 if (!(sc->irqvs = 1658 (struct ice_irq_vector *) malloc(sizeof(struct ice_irq_vector) * (sc->num_irq_vectors), 1659 M_ICE, M_NOWAIT))) { 1660 device_printf(sc->dev, 1661 "Unable to allocate irqv memory\n"); 1662 return (ENOMEM); 1663 } 1664 1665 /* Administrative interrupt events will use vector 0 */ 1666 err = iflib_irq_alloc_generic(ctx, &sc->irqvs[0].irq, 1, IFLIB_INTR_ADMIN, 1667 ice_msix_admin, sc, 0, "admin"); 1668 if (err) { 1669 device_printf(sc->dev, 1670 "Failed to register Admin queue handler: %s\n", 1671 ice_err_str(err)); 1672 goto free_irqvs; 1673 } 1674 sc->irqvs[0].me = 0; 1675 1676 /* Do not allocate queue interrupts when in recovery mode */ 1677 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1678 return (0); 1679 1680 int rid; 1681 for (i = 0, vector = 1; i < vsi->num_rx_queues; i++, vector++) { 1682 struct ice_rx_queue *rxq = &vsi->rx_queues[i]; 1683 struct ice_tx_queue *txq = &vsi->tx_queues[i]; 1684 char irq_name[16]; 1685 1686 rid = vector + 1; 1687 1688 snprintf(irq_name, sizeof(irq_name), "rxq%d", i); 1689 err = iflib_irq_alloc_generic(ctx, &sc->irqvs[vector].irq, rid, 1690 IFLIB_INTR_RXTX, ice_msix_que, 1691 rxq, rxq->me, irq_name); 1692 if (err) { 1693 device_printf(sc->dev, 1694 "Failed to allocate q int %d err: %s\n", 1695 i, ice_err_str(err)); 1696 vector--; 1697 i--; 1698 goto fail; 1699 } 1700 sc->irqvs[vector].me = vector; 1701 rxq->irqv = &sc->irqvs[vector]; 1702 1703 bzero(irq_name, sizeof(irq_name)); 1704 1705 snprintf(irq_name, sizeof(irq_name), "txq%d", i); 1706 iflib_softirq_alloc_generic(ctx, &sc->irqvs[vector].irq, 1707 IFLIB_INTR_TX, txq, 1708 txq->me, irq_name); 1709 txq->irqv = &sc->irqvs[vector]; 1710 } 1711 1712 /* For future interrupt assignments */ 1713 sc->last_rid = rid + sc->irdma_vectors; 1714 1715 #ifdef PCI_IOV 1716 /* Create soft IRQ for handling VF resets */ 1717 iflib_softirq_alloc_generic(ctx, NULL, IFLIB_INTR_IOV, sc, 0, "iov"); 1718 #endif 1719 1720 return (0); 1721 fail: 1722 for (; i >= 0; i--, vector--) 1723 iflib_irq_free(ctx, &sc->irqvs[vector].irq); 1724 iflib_irq_free(ctx, &sc->irqvs[0].irq); 1725 free_irqvs: 1726 free(sc->irqvs, M_ICE); 1727 sc->irqvs = NULL; 1728 return err; 1729 } 1730 1731 /** 1732 * ice_if_mtu_set - Set the device MTU 1733 * @ctx: iflib context structure 1734 * @mtu: the MTU requested 1735 * 1736 * Called by iflib to configure the device's Maximum Transmission Unit (MTU). 1737 * 1738 * @pre assumes the caller holds the iflib CTX lock 1739 */ 1740 static int 1741 ice_if_mtu_set(if_ctx_t ctx, uint32_t mtu) 1742 { 1743 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1744 1745 ASSERT_CTX_LOCKED(sc); 1746 1747 /* Do not support configuration when in recovery mode */ 1748 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1749 return (ENOSYS); 1750 1751 if (mtu < ICE_MIN_MTU || mtu > ICE_MAX_MTU) 1752 return (EINVAL); 1753 1754 sc->scctx->isc_max_frame_size = mtu + 1755 ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN; 1756 1757 sc->pf_vsi.max_frame_size = sc->scctx->isc_max_frame_size; 1758 1759 return (0); 1760 } 1761 1762 /** 1763 * ice_if_intr_enable - Enable device interrupts 1764 * @ctx: iflib context structure 1765 * 1766 * Called by iflib to request enabling device interrupts. 1767 */ 1768 static void 1769 ice_if_intr_enable(if_ctx_t ctx) 1770 { 1771 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1772 struct ice_vsi *vsi = &sc->pf_vsi; 1773 struct ice_hw *hw = &sc->hw; 1774 1775 ASSERT_CTX_LOCKED(sc); 1776 1777 /* Enable ITR 0 */ 1778 ice_enable_intr(hw, sc->irqvs[0].me); 1779 1780 /* Do not enable queue interrupts in recovery mode */ 1781 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1782 return; 1783 1784 /* Enable all queue interrupts */ 1785 for (int i = 0; i < vsi->num_rx_queues; i++) 1786 ice_enable_intr(hw, vsi->rx_queues[i].irqv->me); 1787 } 1788 1789 /** 1790 * ice_if_intr_disable - Disable device interrupts 1791 * @ctx: iflib context structure 1792 * 1793 * Called by iflib to request disabling device interrupts. 1794 */ 1795 static void 1796 ice_if_intr_disable(if_ctx_t ctx) 1797 { 1798 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1799 struct ice_hw *hw = &sc->hw; 1800 unsigned int i; 1801 1802 ASSERT_CTX_LOCKED(sc); 1803 1804 /* IFDI_INTR_DISABLE may be called prior to interrupts actually being 1805 * assigned to queues. Instead of assuming that the interrupt 1806 * assignment in the rx_queues structure is valid, just disable all 1807 * possible interrupts 1808 * 1809 * Note that we choose not to disable ITR 0 because this handles the 1810 * AdminQ interrupts, and we want to keep processing these even when 1811 * the interface is offline. 1812 */ 1813 for (i = 1; i < hw->func_caps.common_cap.num_msix_vectors; i++) 1814 ice_disable_intr(hw, i); 1815 } 1816 1817 /** 1818 * ice_if_rx_queue_intr_enable - Enable a specific Rx queue interrupt 1819 * @ctx: iflib context structure 1820 * @rxqid: the Rx queue to enable 1821 * 1822 * Enable a specific Rx queue interrupt. 1823 * 1824 * This function is not protected by the iflib CTX lock. 1825 */ 1826 static int 1827 ice_if_rx_queue_intr_enable(if_ctx_t ctx, uint16_t rxqid) 1828 { 1829 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1830 struct ice_vsi *vsi = &sc->pf_vsi; 1831 struct ice_hw *hw = &sc->hw; 1832 1833 /* Do not enable queue interrupts in recovery mode */ 1834 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1835 return (ENOSYS); 1836 1837 ice_enable_intr(hw, vsi->rx_queues[rxqid].irqv->me); 1838 return (0); 1839 } 1840 1841 /** 1842 * ice_if_tx_queue_intr_enable - Enable a specific Tx queue interrupt 1843 * @ctx: iflib context structure 1844 * @txqid: the Tx queue to enable 1845 * 1846 * Enable a specific Tx queue interrupt. 1847 * 1848 * This function is not protected by the iflib CTX lock. 1849 */ 1850 static int 1851 ice_if_tx_queue_intr_enable(if_ctx_t ctx, uint16_t txqid) 1852 { 1853 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1854 struct ice_vsi *vsi = &sc->pf_vsi; 1855 struct ice_hw *hw = &sc->hw; 1856 1857 /* Do not enable queue interrupts in recovery mode */ 1858 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1859 return (ENOSYS); 1860 1861 ice_enable_intr(hw, vsi->tx_queues[txqid].irqv->me); 1862 return (0); 1863 } 1864 1865 /** 1866 * ice_set_default_promisc_mask - Set default config for promisc settings 1867 * @promisc_mask: bitmask to setup 1868 * 1869 * The ice_(set|clear)_vsi_promisc() function expects a mask of promiscuous 1870 * modes to operate on. The mask used in here is the default one for the 1871 * driver, where promiscuous is enabled/disabled for all types of 1872 * non-VLAN-tagged/VLAN 0 traffic. 1873 */ 1874 static void 1875 ice_set_default_promisc_mask(ice_bitmap_t *promisc_mask) 1876 { 1877 ice_zero_bitmap(promisc_mask, ICE_PROMISC_MAX); 1878 ice_set_bit(ICE_PROMISC_UCAST_TX, promisc_mask); 1879 ice_set_bit(ICE_PROMISC_UCAST_RX, promisc_mask); 1880 ice_set_bit(ICE_PROMISC_MCAST_TX, promisc_mask); 1881 ice_set_bit(ICE_PROMISC_MCAST_RX, promisc_mask); 1882 } 1883 1884 /** 1885 * ice_if_promisc_set - Set device promiscuous mode 1886 * @ctx: iflib context structure 1887 * @flags: promiscuous flags to configure 1888 * 1889 * Called by iflib to configure device promiscuous mode. 1890 * 1891 * @remark Calls to this function will always overwrite the previous setting 1892 */ 1893 static int 1894 ice_if_promisc_set(if_ctx_t ctx, int flags) 1895 { 1896 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1897 struct ice_hw *hw = &sc->hw; 1898 device_t dev = sc->dev; 1899 int status; 1900 bool promisc_enable = flags & IFF_PROMISC; 1901 bool multi_enable = flags & IFF_ALLMULTI; 1902 ice_declare_bitmap(promisc_mask, ICE_PROMISC_MAX); 1903 1904 /* Do not support configuration when in recovery mode */ 1905 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1906 return (ENOSYS); 1907 1908 ice_set_default_promisc_mask(promisc_mask); 1909 1910 if (promisc_enable) { 1911 status = ice_set_vsi_promisc(hw, sc->pf_vsi.idx, 1912 promisc_mask, 0); 1913 if (status && status != ICE_ERR_ALREADY_EXISTS) { 1914 device_printf(dev, 1915 "Failed to enable promiscuous mode for " 1916 "PF VSI, err %s aq_err %s\n", 1917 ice_status_str(status), 1918 ice_aq_str(hw->adminq.sq_last_status)); 1919 return (EIO); 1920 } 1921 } else { 1922 status = ice_clear_vsi_promisc(hw, sc->pf_vsi.idx, 1923 promisc_mask, 0); 1924 if (status) { 1925 device_printf(dev, 1926 "Failed to disable promiscuous mode for" 1927 " PF VSI, err %s aq_err %s\n", 1928 ice_status_str(status), 1929 ice_aq_str(hw->adminq.sq_last_status)); 1930 return (EIO); 1931 } 1932 1933 if (multi_enable) { 1934 ice_clear_bit(ICE_PROMISC_UCAST_TX, promisc_mask); 1935 ice_clear_bit(ICE_PROMISC_UCAST_RX, promisc_mask); 1936 status = ice_set_vsi_promisc(hw, sc->pf_vsi.idx, 1937 promisc_mask, 0); 1938 if (status && status != ICE_ERR_ALREADY_EXISTS) { 1939 device_printf(dev, 1940 "Failed to enable allmulti mode " 1941 "for PF VSI, err %s aq_err %s\n", 1942 ice_status_str(status), 1943 ice_aq_str( 1944 hw->adminq.sq_last_status)); 1945 return (EIO); 1946 } 1947 } 1948 } 1949 1950 return (0); 1951 } 1952 1953 /** 1954 * ice_if_media_change - Change device media 1955 * @ctx: device ctx structure 1956 * 1957 * Called by iflib when a media change is requested. This operation is not 1958 * supported by the hardware, so we just return an error code. 1959 */ 1960 static int 1961 ice_if_media_change(if_ctx_t ctx) 1962 { 1963 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1964 1965 device_printf(sc->dev, "Media change is not supported.\n"); 1966 return (ENODEV); 1967 } 1968 1969 /** 1970 * ice_if_media_status - Report current device media 1971 * @ctx: iflib context structure 1972 * @ifmr: ifmedia request structure to update 1973 * 1974 * Updates the provided ifmr with current device media status, including link 1975 * status and media type. 1976 */ 1977 static void 1978 ice_if_media_status(if_ctx_t ctx, struct ifmediareq *ifmr) 1979 { 1980 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 1981 struct ice_link_status *li = &sc->hw.port_info->phy.link_info; 1982 1983 ifmr->ifm_status = IFM_AVALID; 1984 ifmr->ifm_active = IFM_ETHER; 1985 1986 /* Never report link up or media types when in recovery mode */ 1987 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 1988 return; 1989 1990 if (!sc->link_up) 1991 return; 1992 1993 ifmr->ifm_status |= IFM_ACTIVE; 1994 ifmr->ifm_active |= IFM_FDX; 1995 1996 if (li->phy_type_low) 1997 ifmr->ifm_active |= ice_get_phy_type_low(li->phy_type_low); 1998 else if (li->phy_type_high) 1999 ifmr->ifm_active |= ice_get_phy_type_high(li->phy_type_high); 2000 else 2001 ifmr->ifm_active |= IFM_UNKNOWN; 2002 2003 /* Report flow control status as well */ 2004 if (li->an_info & ICE_AQ_LINK_PAUSE_TX) 2005 ifmr->ifm_active |= IFM_ETH_TXPAUSE; 2006 if (li->an_info & ICE_AQ_LINK_PAUSE_RX) 2007 ifmr->ifm_active |= IFM_ETH_RXPAUSE; 2008 } 2009 2010 /** 2011 * ice_init_tx_tracking - Initialize Tx queue software tracking values 2012 * @vsi: the VSI to initialize 2013 * 2014 * Initialize Tx queue software tracking values, including the Report Status 2015 * queue, and related software tracking values. 2016 */ 2017 static void 2018 ice_init_tx_tracking(struct ice_vsi *vsi) 2019 { 2020 struct ice_tx_queue *txq; 2021 size_t j; 2022 int i; 2023 2024 for (i = 0, txq = vsi->tx_queues; i < vsi->num_tx_queues; i++, txq++) { 2025 2026 txq->tx_rs_cidx = txq->tx_rs_pidx = 0; 2027 2028 /* Initialize the last processed descriptor to be the end of 2029 * the ring, rather than the start, so that we avoid an 2030 * off-by-one error in ice_ift_txd_credits_update for the 2031 * first packet. 2032 */ 2033 txq->tx_cidx_processed = txq->desc_count - 1; 2034 2035 for (j = 0; j < txq->desc_count; j++) 2036 txq->tx_rsq[j] = QIDX_INVALID; 2037 } 2038 } 2039 2040 /** 2041 * ice_update_rx_mbuf_sz - Update the Rx buffer size for all queues 2042 * @sc: the device softc 2043 * 2044 * Called to update the Rx queue mbuf_sz parameter for configuring the receive 2045 * buffer sizes when programming hardware. 2046 */ 2047 static void 2048 ice_update_rx_mbuf_sz(struct ice_softc *sc) 2049 { 2050 uint32_t mbuf_sz = iflib_get_rx_mbuf_sz(sc->ctx); 2051 struct ice_vsi *vsi = &sc->pf_vsi; 2052 2053 MPASS(mbuf_sz <= UINT16_MAX); 2054 vsi->mbuf_sz = mbuf_sz; 2055 } 2056 2057 /** 2058 * ice_if_init - Initialize the device 2059 * @ctx: iflib ctx structure 2060 * 2061 * Called by iflib to bring the device up, i.e. ifconfig ice0 up. Initializes 2062 * device filters and prepares the Tx and Rx engines. 2063 * 2064 * @pre assumes the caller holds the iflib CTX lock 2065 */ 2066 static void 2067 ice_if_init(if_ctx_t ctx) 2068 { 2069 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 2070 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 2071 device_t dev = sc->dev; 2072 int err; 2073 2074 ASSERT_CTX_LOCKED(sc); 2075 2076 /* 2077 * We've seen an issue with 11.3/12.1 where sideband routines are 2078 * called after detach is called. This would call routines after 2079 * if_stop, causing issues with the teardown process. This has 2080 * seemingly been fixed in STABLE snapshots, but it seems like a 2081 * good idea to have this guard here regardless. 2082 */ 2083 if (ice_driver_is_detaching(sc)) 2084 return; 2085 2086 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 2087 return; 2088 2089 if (ice_test_state(&sc->state, ICE_STATE_RESET_FAILED)) { 2090 device_printf(sc->dev, "request to start interface cannot be completed as the device failed to reset\n"); 2091 return; 2092 } 2093 2094 if (ice_test_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET)) { 2095 device_printf(sc->dev, "request to start interface while device is prepared for impending reset\n"); 2096 return; 2097 } 2098 2099 ice_update_rx_mbuf_sz(sc); 2100 2101 /* Update the MAC address... User might use a LAA */ 2102 err = ice_update_laa_mac(sc); 2103 if (err) { 2104 device_printf(dev, 2105 "LAA address change failed, err %s\n", 2106 ice_err_str(err)); 2107 return; 2108 } 2109 2110 /* Initialize software Tx tracking values */ 2111 ice_init_tx_tracking(&sc->pf_vsi); 2112 2113 err = ice_cfg_vsi_for_tx(&sc->pf_vsi); 2114 if (err) { 2115 device_printf(dev, 2116 "Unable to configure the main VSI for Tx: %s\n", 2117 ice_err_str(err)); 2118 return; 2119 } 2120 2121 err = ice_cfg_vsi_for_rx(&sc->pf_vsi); 2122 if (err) { 2123 device_printf(dev, 2124 "Unable to configure the main VSI for Rx: %s\n", 2125 ice_err_str(err)); 2126 goto err_cleanup_tx; 2127 } 2128 2129 err = ice_control_all_rx_queues(&sc->pf_vsi, true); 2130 if (err) { 2131 device_printf(dev, 2132 "Unable to enable Rx rings for transmit: %s\n", 2133 ice_err_str(err)); 2134 goto err_cleanup_tx; 2135 } 2136 2137 err = ice_cfg_pf_default_mac_filters(sc); 2138 if (err) { 2139 device_printf(dev, 2140 "Unable to configure default MAC filters: %s\n", 2141 ice_err_str(err)); 2142 goto err_stop_rx; 2143 } 2144 2145 /* We use software interrupts for Tx, so we only program the hardware 2146 * interrupts for Rx. 2147 */ 2148 ice_configure_all_rxq_interrupts(&sc->pf_vsi); 2149 ice_configure_rx_itr(&sc->pf_vsi); 2150 2151 /* Configure promiscuous mode */ 2152 ice_if_promisc_set(ctx, if_getflags(sc->ifp)); 2153 2154 if (!ice_testandclear_state(&sc->state, ICE_STATE_FIRST_INIT_LINK)) 2155 if (!sc->link_up && ((if_getflags(sc->ifp) & IFF_UP) || 2156 ice_test_state(&sc->state, ICE_STATE_LINK_ACTIVE_ON_DOWN))) 2157 ice_set_link(sc, true); 2158 2159 ice_rdma_pf_init(sc); 2160 2161 ice_set_state(&sc->state, ICE_STATE_DRIVER_INITIALIZED); 2162 2163 if (sc->mirr_if && ice_testandclear_state(&mif->state, ICE_STATE_SUBIF_NEEDS_REINIT)) { 2164 ice_clear_state(&mif->state, ICE_STATE_DRIVER_INITIALIZED); 2165 iflib_request_reset(sc->mirr_if->subctx); 2166 iflib_admin_intr_deferred(sc->mirr_if->subctx); 2167 } 2168 2169 return; 2170 2171 err_stop_rx: 2172 ice_control_all_rx_queues(&sc->pf_vsi, false); 2173 err_cleanup_tx: 2174 ice_vsi_disable_tx(&sc->pf_vsi); 2175 } 2176 2177 /** 2178 * ice_poll_for_media_avail - Re-enable link if media is detected 2179 * @sc: device private structure 2180 * 2181 * Intended to be called from the driver's timer function, this function 2182 * sends the Get Link Status AQ command and re-enables HW link if the 2183 * command says that media is available. 2184 * 2185 * If the driver doesn't have the "NO_MEDIA" state set, then this does nothing, 2186 * since media removal events are supposed to be sent to the driver through 2187 * a link status event. 2188 */ 2189 static void 2190 ice_poll_for_media_avail(struct ice_softc *sc) 2191 { 2192 struct ice_hw *hw = &sc->hw; 2193 struct ice_port_info *pi = hw->port_info; 2194 2195 /* E830 only: There's no interrupt for when the PHY FW has finished loading, 2196 * so poll for the status in the media task here if it's previously 2197 * been detected that it's still loading. 2198 */ 2199 if (ice_is_e830(hw) && 2200 ice_test_state(&sc->state, ICE_STATE_PHY_FW_INIT_PENDING)) { 2201 if (rd32(hw, GL_MNG_FWSM) & GL_MNG_FWSM_FW_LOADING_M) 2202 ice_clear_state(&sc->state, ICE_STATE_PHY_FW_INIT_PENDING); 2203 else 2204 return; 2205 } 2206 2207 if (ice_test_state(&sc->state, ICE_STATE_NO_MEDIA)) { 2208 pi->phy.get_link_info = true; 2209 ice_get_link_status(pi, &sc->link_up); 2210 2211 if (pi->phy.link_info.link_info & ICE_AQ_MEDIA_AVAILABLE) { 2212 int status; 2213 2214 /* Re-enable link and re-apply user link settings */ 2215 if (ice_test_state(&sc->state, ICE_STATE_LINK_ACTIVE_ON_DOWN) || 2216 (if_getflags(sc->ifp) & IFF_UP)) { 2217 ice_apply_saved_phy_cfg(sc, ICE_APPLY_LS_FEC_FC); 2218 2219 /* Update the OS about changes in media capability */ 2220 status = ice_add_media_types(sc, sc->media); 2221 if (status) 2222 device_printf(sc->dev, 2223 "Error adding device media types: %s aq_err %s\n", 2224 ice_status_str(status), 2225 ice_aq_str(hw->adminq.sq_last_status)); 2226 } 2227 2228 ice_clear_state(&sc->state, ICE_STATE_NO_MEDIA); 2229 } 2230 } 2231 } 2232 2233 /** 2234 * ice_if_timer - called by iflib periodically 2235 * @ctx: iflib ctx structure 2236 * @qid: the queue this timer was called for 2237 * 2238 * This callback is triggered by iflib periodically. We use it to update the 2239 * hw statistics. 2240 * 2241 * @remark this function is not protected by the iflib CTX lock. 2242 */ 2243 static void 2244 ice_if_timer(if_ctx_t ctx, uint16_t qid) 2245 { 2246 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 2247 uint64_t prev_link_xoff_rx = sc->stats.cur.link_xoff_rx; 2248 2249 if (qid != 0) 2250 return; 2251 2252 /* Do not attempt to update stats when in recovery mode */ 2253 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 2254 return; 2255 2256 /* Update device statistics */ 2257 ice_update_pf_stats(sc); 2258 2259 /* 2260 * For proper watchdog management, the iflib stack needs to know if 2261 * we've been paused during the last interval. Check if the 2262 * link_xoff_rx stat changed, and set the isc_pause_frames, if so. 2263 */ 2264 if (sc->stats.cur.link_xoff_rx != prev_link_xoff_rx) 2265 sc->scctx->isc_pause_frames = 1; 2266 2267 /* Update the primary VSI stats */ 2268 ice_update_vsi_hw_stats(&sc->pf_vsi); 2269 2270 /* Update mirror VSI stats */ 2271 if (sc->mirr_if && sc->mirr_if->if_attached) 2272 ice_update_vsi_hw_stats(sc->mirr_if->vsi); 2273 } 2274 2275 /** 2276 * ice_admin_timer - called periodically to trigger the admin task 2277 * @arg: callout(9) argument pointing to the device private softc structure 2278 * 2279 * Timer function used as part of a callout(9) timer that will periodically 2280 * trigger the admin task, even when the interface is down. 2281 * 2282 * @remark this function is not called by iflib and is not protected by the 2283 * iflib CTX lock. 2284 * 2285 * @remark because this is a callout function, it cannot sleep and should not 2286 * attempt taking the iflib CTX lock. 2287 */ 2288 static void 2289 ice_admin_timer(void *arg) 2290 { 2291 struct ice_softc *sc = (struct ice_softc *)arg; 2292 2293 /* 2294 * There is a point where callout routines are no longer 2295 * cancelable. So there exists a window of time where the 2296 * driver enters detach() and tries to cancel the callout, but the 2297 * callout routine has passed the cancellation point. The detach() 2298 * routine is unaware of this and tries to free resources that the 2299 * callout routine needs. So we check for the detach state flag to 2300 * at least shrink the window of opportunity. 2301 */ 2302 if (ice_driver_is_detaching(sc)) 2303 return; 2304 2305 /* Fire off the admin task */ 2306 iflib_admin_intr_deferred(sc->ctx); 2307 2308 /* Reschedule the admin timer */ 2309 callout_schedule(&sc->admin_timer, hz/2); 2310 } 2311 2312 /** 2313 * ice_transition_recovery_mode - Transition to recovery mode 2314 * @sc: the device private softc 2315 * 2316 * Called when the driver detects that the firmware has entered recovery mode 2317 * at run time. 2318 */ 2319 static void 2320 ice_transition_recovery_mode(struct ice_softc *sc) 2321 { 2322 struct ice_vsi *vsi = &sc->pf_vsi; 2323 int i; 2324 2325 device_printf(sc->dev, "Firmware recovery mode detected. Limiting functionality. Refer to Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n"); 2326 2327 /* Tell the stack that the link has gone down */ 2328 iflib_link_state_change(sc->ctx, LINK_STATE_DOWN, 0); 2329 2330 /* Request that the device be re-initialized */ 2331 ice_request_stack_reinit(sc); 2332 2333 ice_rdma_pf_detach(sc); 2334 ice_clear_bit(ICE_FEATURE_RDMA, sc->feat_cap); 2335 2336 #ifdef PCI_IOV 2337 if (ice_test_and_clear_bit(ICE_FEATURE_SRIOV, sc->feat_en)) 2338 ice_iov_detach(sc); 2339 #else 2340 ice_clear_bit(ICE_FEATURE_SRIOV, sc->feat_en); 2341 #endif /* PCI_IOV */ 2342 ice_clear_bit(ICE_FEATURE_SRIOV, sc->feat_cap); 2343 2344 ice_vsi_del_txqs_ctx(vsi); 2345 ice_vsi_del_rxqs_ctx(vsi); 2346 2347 for (i = 0; i < sc->num_available_vsi; i++) { 2348 if (sc->all_vsi[i]) 2349 ice_release_vsi(sc->all_vsi[i]); 2350 } 2351 sc->num_available_vsi = 0; 2352 2353 if (sc->all_vsi) { 2354 free(sc->all_vsi, M_ICE); 2355 sc->all_vsi = NULL; 2356 } 2357 2358 /* Destroy the interrupt manager */ 2359 ice_resmgr_destroy(&sc->dev_imgr); 2360 /* Destroy the queue managers */ 2361 ice_resmgr_destroy(&sc->tx_qmgr); 2362 ice_resmgr_destroy(&sc->rx_qmgr); 2363 2364 ice_deinit_hw(&sc->hw); 2365 } 2366 2367 /** 2368 * ice_transition_safe_mode - Transition to safe mode 2369 * @sc: the device private softc 2370 * 2371 * Called when the driver attempts to reload the DDP package during a device 2372 * reset, and the new download fails. If so, we must transition to safe mode 2373 * at run time. 2374 * 2375 * @remark although safe mode normally allocates only a single queue, we can't 2376 * change the number of queues dynamically when using iflib. Due to this, we 2377 * do not attempt to reduce the number of queues. 2378 */ 2379 static void 2380 ice_transition_safe_mode(struct ice_softc *sc) 2381 { 2382 /* Indicate that we are in Safe mode */ 2383 ice_set_bit(ICE_FEATURE_SAFE_MODE, sc->feat_cap); 2384 ice_set_bit(ICE_FEATURE_SAFE_MODE, sc->feat_en); 2385 2386 ice_rdma_pf_detach(sc); 2387 ice_clear_bit(ICE_FEATURE_RDMA, sc->feat_cap); 2388 2389 #ifdef PCI_IOV 2390 if (ice_test_and_clear_bit(ICE_FEATURE_SRIOV, sc->feat_en)) 2391 ice_iov_detach(sc); 2392 #else 2393 ice_clear_bit(ICE_FEATURE_SRIOV, sc->feat_en); 2394 #endif /* PCI_IOV */ 2395 ice_clear_bit(ICE_FEATURE_SRIOV, sc->feat_cap); 2396 2397 ice_clear_bit(ICE_FEATURE_RSS, sc->feat_cap); 2398 ice_clear_bit(ICE_FEATURE_RSS, sc->feat_en); 2399 } 2400 2401 /** 2402 * ice_if_update_admin_status - update admin status 2403 * @ctx: iflib ctx structure 2404 * 2405 * Called by iflib to update the admin status. For our purposes, this means 2406 * check the adminq, and update the link status. It's ultimately triggered by 2407 * our admin interrupt, or by the ice_if_timer periodically. 2408 * 2409 * @pre assumes the caller holds the iflib CTX lock 2410 */ 2411 static void 2412 ice_if_update_admin_status(if_ctx_t ctx) 2413 { 2414 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 2415 enum ice_fw_modes fw_mode; 2416 bool reschedule = false; 2417 u16 pending = 0; 2418 2419 ASSERT_CTX_LOCKED(sc); 2420 2421 /* Check if the firmware entered recovery mode at run time */ 2422 fw_mode = ice_get_fw_mode(&sc->hw); 2423 if (fw_mode == ICE_FW_MODE_REC) { 2424 if (!ice_testandset_state(&sc->state, ICE_STATE_RECOVERY_MODE)) { 2425 /* If we just entered recovery mode, log a warning to 2426 * the system administrator and deinit driver state 2427 * that is no longer functional. 2428 */ 2429 ice_transition_recovery_mode(sc); 2430 } 2431 } else if (fw_mode == ICE_FW_MODE_ROLLBACK) { 2432 if (!ice_testandset_state(&sc->state, ICE_STATE_ROLLBACK_MODE)) { 2433 /* Rollback mode isn't fatal, but we don't want to 2434 * repeatedly post a message about it. 2435 */ 2436 ice_print_rollback_msg(&sc->hw); 2437 } 2438 } 2439 2440 /* Handle global reset events */ 2441 ice_handle_reset_event(sc); 2442 2443 /* Handle PF reset requests */ 2444 ice_handle_pf_reset_request(sc); 2445 2446 /* Handle MDD events */ 2447 ice_handle_mdd_event(sc); 2448 2449 if (ice_test_state(&sc->state, ICE_STATE_RESET_FAILED) || 2450 ice_test_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET) || 2451 ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) { 2452 /* 2453 * If we know the control queues are disabled, skip processing 2454 * the control queues entirely. 2455 */ 2456 ; 2457 } else if (ice_testandclear_state(&sc->state, ICE_STATE_CONTROLQ_EVENT_PENDING)) { 2458 ice_process_ctrlq(sc, ICE_CTL_Q_ADMIN, &pending); 2459 if (pending > 0) 2460 reschedule = true; 2461 2462 if (ice_is_generic_mac(&sc->hw)) { 2463 ice_process_ctrlq(sc, ICE_CTL_Q_SB, &pending); 2464 if (pending > 0) 2465 reschedule = true; 2466 } 2467 2468 ice_process_ctrlq(sc, ICE_CTL_Q_MAILBOX, &pending); 2469 if (pending > 0) 2470 reschedule = true; 2471 } 2472 2473 /* Poll for link up */ 2474 ice_poll_for_media_avail(sc); 2475 2476 /* Check and update link status */ 2477 ice_update_link_status(sc, false); 2478 2479 #ifdef PCI_IOV 2480 /* 2481 * Schedule VFs' reset handler after global resets 2482 * and other events were processed. 2483 */ 2484 if (ice_testandclear_state(&sc->state, ICE_STATE_VFLR_PENDING)) 2485 iflib_iov_intr_deferred(ctx); 2486 #endif 2487 2488 /* 2489 * If there are still messages to process, we need to reschedule 2490 * ourselves. Otherwise, we can just re-enable the interrupt. We'll be 2491 * woken up at the next interrupt or timer event. 2492 */ 2493 if (reschedule) { 2494 ice_set_state(&sc->state, ICE_STATE_CONTROLQ_EVENT_PENDING); 2495 iflib_admin_intr_deferred(ctx); 2496 } else { 2497 ice_enable_intr(&sc->hw, sc->irqvs[0].me); 2498 } 2499 } 2500 2501 /** 2502 * ice_prepare_for_reset - Prepare device for an impending reset 2503 * @sc: The device private softc 2504 * 2505 * Prepare the driver for an impending reset, shutting down VSIs, clearing the 2506 * scheduler setup, and shutting down controlqs. Uses the 2507 * ICE_STATE_PREPARED_FOR_RESET to indicate whether we've already prepared the 2508 * driver for reset or not. 2509 */ 2510 static void 2511 ice_prepare_for_reset(struct ice_softc *sc) 2512 { 2513 struct ice_hw *hw = &sc->hw; 2514 2515 /* If we're already prepared, there's nothing to do */ 2516 if (ice_testandset_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET)) 2517 return; 2518 2519 log(LOG_INFO, "%s: preparing to reset device logic\n", if_name(sc->ifp)); 2520 2521 /* In recovery mode, hardware is not initialized */ 2522 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 2523 return; 2524 2525 /* inform the RDMA client */ 2526 ice_rdma_notify_reset(sc); 2527 /* stop the RDMA client */ 2528 ice_rdma_pf_stop(sc); 2529 2530 /* Release the main PF VSI queue mappings */ 2531 ice_resmgr_release_map(&sc->tx_qmgr, sc->pf_vsi.tx_qmap, 2532 sc->pf_vsi.num_tx_queues); 2533 ice_resmgr_release_map(&sc->rx_qmgr, sc->pf_vsi.rx_qmap, 2534 sc->pf_vsi.num_rx_queues); 2535 if (sc->mirr_if) { 2536 ice_resmgr_release_map(&sc->tx_qmgr, sc->mirr_if->vsi->tx_qmap, 2537 sc->mirr_if->num_irq_vectors); 2538 ice_resmgr_release_map(&sc->rx_qmgr, sc->mirr_if->vsi->rx_qmap, 2539 sc->mirr_if->num_irq_vectors); 2540 } 2541 2542 ice_clear_hw_tbls(hw); 2543 2544 if (hw->port_info) 2545 ice_sched_cleanup_all(hw); 2546 2547 ice_shutdown_all_ctrlq(hw, false); 2548 } 2549 2550 /** 2551 * ice_rebuild_pf_vsi_qmap - Rebuild the main PF VSI queue mapping 2552 * @sc: the device softc pointer 2553 * 2554 * Loops over the Tx and Rx queues for the main PF VSI and reassigns the queue 2555 * mapping after a reset occurred. 2556 */ 2557 static int 2558 ice_rebuild_pf_vsi_qmap(struct ice_softc *sc) 2559 { 2560 struct ice_vsi *vsi = &sc->pf_vsi; 2561 struct ice_tx_queue *txq; 2562 struct ice_rx_queue *rxq; 2563 int err, i; 2564 2565 /* Re-assign Tx queues from PF space to the main VSI */ 2566 err = ice_resmgr_assign_contiguous(&sc->tx_qmgr, vsi->tx_qmap, 2567 vsi->num_tx_queues); 2568 if (err) { 2569 device_printf(sc->dev, "Unable to re-assign PF Tx queues: %s\n", 2570 ice_err_str(err)); 2571 return (err); 2572 } 2573 2574 /* Re-assign Rx queues from PF space to this VSI */ 2575 err = ice_resmgr_assign_contiguous(&sc->rx_qmgr, vsi->rx_qmap, 2576 vsi->num_rx_queues); 2577 if (err) { 2578 device_printf(sc->dev, "Unable to re-assign PF Rx queues: %s\n", 2579 ice_err_str(err)); 2580 goto err_release_tx_queues; 2581 } 2582 2583 vsi->qmap_type = ICE_RESMGR_ALLOC_CONTIGUOUS; 2584 2585 /* Re-assign Tx queue tail pointers */ 2586 for (i = 0, txq = vsi->tx_queues; i < vsi->num_tx_queues; i++, txq++) 2587 txq->tail = QTX_COMM_DBELL(vsi->tx_qmap[i]); 2588 2589 /* Re-assign Rx queue tail pointers */ 2590 for (i = 0, rxq = vsi->rx_queues; i < vsi->num_rx_queues; i++, rxq++) 2591 rxq->tail = QRX_TAIL(vsi->rx_qmap[i]); 2592 2593 return (0); 2594 2595 err_release_tx_queues: 2596 ice_resmgr_release_map(&sc->tx_qmgr, sc->pf_vsi.tx_qmap, 2597 sc->pf_vsi.num_tx_queues); 2598 2599 return (err); 2600 } 2601 2602 /* determine if the iflib context is active */ 2603 #define CTX_ACTIVE(ctx) ((if_getdrvflags(iflib_get_ifp(ctx)) & IFF_DRV_RUNNING)) 2604 2605 /** 2606 * ice_rebuild_recovery_mode - Rebuild driver state while in recovery mode 2607 * @sc: The device private softc 2608 * 2609 * Handle a driver rebuild while in recovery mode. This will only rebuild the 2610 * limited functionality supported while in recovery mode. 2611 */ 2612 static void 2613 ice_rebuild_recovery_mode(struct ice_softc *sc) 2614 { 2615 device_t dev = sc->dev; 2616 2617 /* enable PCIe bus master */ 2618 pci_enable_busmaster(dev); 2619 2620 /* Configure interrupt causes for the administrative interrupt */ 2621 ice_configure_misc_interrupts(sc); 2622 2623 /* Enable ITR 0 right away, so that we can handle admin interrupts */ 2624 ice_enable_intr(&sc->hw, sc->irqvs[0].me); 2625 2626 /* Now that the rebuild is finished, we're no longer prepared to reset */ 2627 ice_clear_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET); 2628 2629 log(LOG_INFO, "%s: device rebuild successful\n", if_name(sc->ifp)); 2630 2631 /* In order to completely restore device functionality, the iflib core 2632 * needs to be reset. We need to request an iflib reset. Additionally, 2633 * because the state of IFC_DO_RESET is cached within task_fn_admin in 2634 * the iflib core, we also want re-run the admin task so that iflib 2635 * resets immediately instead of waiting for the next interrupt. 2636 */ 2637 ice_request_stack_reinit(sc); 2638 2639 return; 2640 } 2641 2642 /** 2643 * ice_rebuild - Rebuild driver state post reset 2644 * @sc: The device private softc 2645 * 2646 * Restore driver state after a reset occurred. Restart the controlqs, setup 2647 * the hardware port, and re-enable the VSIs. 2648 */ 2649 static void 2650 ice_rebuild(struct ice_softc *sc) 2651 { 2652 struct ice_hw *hw = &sc->hw; 2653 device_t dev = sc->dev; 2654 enum ice_ddp_state pkg_state; 2655 int status; 2656 int err; 2657 2658 sc->rebuild_ticks = ticks; 2659 2660 /* If we're rebuilding, then a reset has succeeded. */ 2661 ice_clear_state(&sc->state, ICE_STATE_RESET_FAILED); 2662 2663 /* 2664 * If the firmware is in recovery mode, only restore the limited 2665 * functionality supported by recovery mode. 2666 */ 2667 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) { 2668 ice_rebuild_recovery_mode(sc); 2669 return; 2670 } 2671 2672 /* enable PCIe bus master */ 2673 pci_enable_busmaster(dev); 2674 2675 status = ice_init_all_ctrlq(hw); 2676 if (status) { 2677 device_printf(dev, "failed to re-init controlqs, err %s\n", 2678 ice_status_str(status)); 2679 goto err_shutdown_ctrlq; 2680 } 2681 2682 /* Query the allocated resources for Tx scheduler */ 2683 status = ice_sched_query_res_alloc(hw); 2684 if (status) { 2685 device_printf(dev, 2686 "Failed to query scheduler resources, err %s aq_err %s\n", 2687 ice_status_str(status), 2688 ice_aq_str(hw->adminq.sq_last_status)); 2689 goto err_shutdown_ctrlq; 2690 } 2691 2692 /* Re-enable FW logging. Keep going even if this fails */ 2693 status = ICE_SUCCESS; 2694 if (hw->pf_id == 0) 2695 status = ice_fwlog_set(hw, &hw->fwlog_cfg); 2696 if (!status) { 2697 /* 2698 * We should have the most updated cached copy of the 2699 * configuration, regardless of whether we're rebuilding 2700 * or not. So we'll simply check to see if logging was 2701 * enabled pre-rebuild. 2702 */ 2703 if (hw->fwlog_cfg.options & ICE_FWLOG_OPTION_IS_REGISTERED) { 2704 status = ice_fwlog_register(hw); 2705 if (status) 2706 device_printf(dev, "failed to re-register fw logging, err %s aq_err %s\n", 2707 ice_status_str(status), 2708 ice_aq_str(hw->adminq.sq_last_status)); 2709 } 2710 } else 2711 device_printf(dev, "failed to rebuild fw logging configuration, err %s aq_err %s\n", 2712 ice_status_str(status), 2713 ice_aq_str(hw->adminq.sq_last_status)); 2714 2715 err = ice_send_version(sc); 2716 if (err) 2717 goto err_shutdown_ctrlq; 2718 2719 err = ice_init_link_events(sc); 2720 if (err) { 2721 device_printf(dev, "ice_init_link_events failed: %s\n", 2722 ice_err_str(err)); 2723 goto err_shutdown_ctrlq; 2724 } 2725 2726 status = ice_clear_pf_cfg(hw); 2727 if (status) { 2728 device_printf(dev, "failed to clear PF configuration, err %s\n", 2729 ice_status_str(status)); 2730 goto err_shutdown_ctrlq; 2731 } 2732 2733 ice_clean_all_vsi_rss_cfg(sc); 2734 2735 ice_clear_pxe_mode(hw); 2736 2737 status = ice_get_caps(hw); 2738 if (status) { 2739 device_printf(dev, "failed to get capabilities, err %s\n", 2740 ice_status_str(status)); 2741 goto err_shutdown_ctrlq; 2742 } 2743 2744 status = ice_sched_init_port(hw->port_info); 2745 if (status) { 2746 device_printf(dev, "failed to initialize port, err %s\n", 2747 ice_status_str(status)); 2748 goto err_sched_cleanup; 2749 } 2750 2751 /* If we previously loaded the package, it needs to be reloaded now */ 2752 if (!ice_is_bit_set(sc->feat_en, ICE_FEATURE_SAFE_MODE)) { 2753 pkg_state = ice_init_pkg(hw, hw->pkg_copy, hw->pkg_size); 2754 if (!ice_is_init_pkg_successful(pkg_state)) { 2755 ice_log_pkg_init(sc, pkg_state); 2756 ice_transition_safe_mode(sc); 2757 } 2758 } 2759 2760 ice_reset_pf_stats(sc); 2761 2762 err = ice_rebuild_pf_vsi_qmap(sc); 2763 if (err) { 2764 device_printf(sc->dev, "Unable to re-assign main VSI queues, err %s\n", 2765 ice_err_str(err)); 2766 goto err_sched_cleanup; 2767 } 2768 err = ice_initialize_vsi(&sc->pf_vsi); 2769 if (err) { 2770 device_printf(sc->dev, "Unable to re-initialize Main VSI, err %s\n", 2771 ice_err_str(err)); 2772 goto err_release_queue_allocations; 2773 } 2774 2775 /* Replay all VSI configuration */ 2776 err = ice_replay_all_vsi_cfg(sc); 2777 if (err) 2778 goto err_deinit_pf_vsi; 2779 2780 /* Re-enable FW health event reporting */ 2781 ice_init_health_events(sc); 2782 2783 /* Reconfigure the main PF VSI for RSS */ 2784 err = ice_config_rss(&sc->pf_vsi); 2785 if (err) { 2786 device_printf(sc->dev, 2787 "Unable to reconfigure RSS for the main VSI, err %s\n", 2788 ice_err_str(err)); 2789 goto err_deinit_pf_vsi; 2790 } 2791 2792 if (hw->port_info->qos_cfg.is_sw_lldp) 2793 ice_add_rx_lldp_filter(sc); 2794 2795 /* Apply previous link settings and refresh link status, if PHY 2796 * FW is ready. 2797 */ 2798 ice_clear_state(&sc->state, ICE_STATE_LINK_STATUS_REPORTED); 2799 ice_init_link(sc); 2800 2801 /* RDMA interface will be restarted by the stack re-init */ 2802 2803 /* Configure interrupt causes for the administrative interrupt */ 2804 ice_configure_misc_interrupts(sc); 2805 2806 /* Enable ITR 0 right away, so that we can handle admin interrupts */ 2807 ice_enable_intr(&sc->hw, sc->irqvs[0].me); 2808 2809 /* Now that the rebuild is finished, we're no longer prepared to reset */ 2810 ice_clear_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET); 2811 2812 /* Reconfigure the subinterface */ 2813 if (sc->mirr_if) { 2814 err = ice_subif_rebuild(sc); 2815 if (err) 2816 goto err_deinit_pf_vsi; 2817 } 2818 2819 log(LOG_INFO, "%s: device rebuild successful\n", if_name(sc->ifp)); 2820 2821 /* In order to completely restore device functionality, the iflib core 2822 * needs to be reset. We need to request an iflib reset. Additionally, 2823 * because the state of IFC_DO_RESET is cached within task_fn_admin in 2824 * the iflib core, we also want re-run the admin task so that iflib 2825 * resets immediately instead of waiting for the next interrupt. 2826 * If LLDP is enabled we need to reconfig DCB to properly reinit all TC 2827 * queues, not only 0. It contains ice_request_stack_reinit as well. 2828 */ 2829 if (hw->port_info->qos_cfg.is_sw_lldp) 2830 ice_request_stack_reinit(sc); 2831 else 2832 ice_do_dcb_reconfig(sc, false); 2833 2834 return; 2835 2836 err_deinit_pf_vsi: 2837 ice_deinit_vsi(&sc->pf_vsi); 2838 err_release_queue_allocations: 2839 ice_resmgr_release_map(&sc->tx_qmgr, sc->pf_vsi.tx_qmap, 2840 sc->pf_vsi.num_tx_queues); 2841 ice_resmgr_release_map(&sc->rx_qmgr, sc->pf_vsi.rx_qmap, 2842 sc->pf_vsi.num_rx_queues); 2843 err_sched_cleanup: 2844 ice_sched_cleanup_all(hw); 2845 err_shutdown_ctrlq: 2846 ice_shutdown_all_ctrlq(hw, false); 2847 ice_clear_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET); 2848 ice_set_state(&sc->state, ICE_STATE_RESET_FAILED); 2849 device_printf(dev, "Driver rebuild failed, please reload the device driver\n"); 2850 } 2851 2852 /** 2853 * ice_handle_reset_event - Handle reset events triggered by OICR 2854 * @sc: The device private softc 2855 * 2856 * Handle reset events triggered by an OICR notification. This includes CORER, 2857 * GLOBR, and EMPR resets triggered by software on this or any other PF or by 2858 * firmware. 2859 * 2860 * @pre assumes the iflib context lock is held, and will unlock it while 2861 * waiting for the hardware to finish reset. 2862 */ 2863 static void 2864 ice_handle_reset_event(struct ice_softc *sc) 2865 { 2866 struct ice_hw *hw = &sc->hw; 2867 int status; 2868 device_t dev = sc->dev; 2869 2870 /* When a CORER, GLOBR, or EMPR is about to happen, the hardware will 2871 * trigger an OICR interrupt. Our OICR handler will determine when 2872 * this occurs and set the ICE_STATE_RESET_OICR_RECV bit as 2873 * appropriate. 2874 */ 2875 if (!ice_testandclear_state(&sc->state, ICE_STATE_RESET_OICR_RECV)) 2876 return; 2877 2878 ice_prepare_for_reset(sc); 2879 2880 /* 2881 * Release the iflib context lock and wait for the device to finish 2882 * resetting. 2883 */ 2884 IFLIB_CTX_UNLOCK(sc); 2885 2886 #define ICE_EMPR_ADDL_WAIT_MSEC_SLOW 20000 2887 if ((ice_is_e830(hw) || ice_is_e825c(hw)) && 2888 (((rd32(hw, GLGEN_RSTAT) & GLGEN_RSTAT_RESET_TYPE_M) >> 2889 GLGEN_RSTAT_RESET_TYPE_S) == ICE_RESET_EMPR)) 2890 ice_msec_pause(ICE_EMPR_ADDL_WAIT_MSEC_SLOW); 2891 2892 status = ice_check_reset(hw); 2893 IFLIB_CTX_LOCK(sc); 2894 if (status) { 2895 device_printf(dev, "Device never came out of reset, err %s\n", 2896 ice_status_str(status)); 2897 2898 ice_set_state(&sc->state, ICE_STATE_RESET_FAILED); 2899 ice_clear_state(&sc->state, ICE_STATE_RESET_PFR_REQ); 2900 ice_clear_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET); 2901 device_printf(dev, "Reset failed; please reload the device driver\n"); 2902 return; 2903 } 2904 2905 /* We're done with the reset, so we can rebuild driver state */ 2906 sc->hw.reset_ongoing = false; 2907 ice_rebuild(sc); 2908 2909 /* In the unlikely event that a PF reset request occurs at the same 2910 * time as a global reset, clear the request now. This avoids 2911 * resetting a second time right after we reset due to a global event. 2912 */ 2913 if (ice_testandclear_state(&sc->state, ICE_STATE_RESET_PFR_REQ)) 2914 device_printf(dev, "Ignoring PFR request that occurred while a reset was ongoing\n"); 2915 } 2916 2917 /** 2918 * ice_handle_pf_reset_request - Initiate PF reset requested by software 2919 * @sc: The device private softc 2920 * 2921 * Initiate a PF reset requested by software. We handle this in the admin task 2922 * so that only one thread actually handles driver preparation and cleanup, 2923 * rather than having multiple threads possibly attempt to run this code 2924 * simultaneously. 2925 * 2926 * @pre assumes the iflib context lock is held and will unlock it while 2927 * waiting for the PF reset to complete. 2928 */ 2929 static void 2930 ice_handle_pf_reset_request(struct ice_softc *sc) 2931 { 2932 struct ice_hw *hw = &sc->hw; 2933 int status; 2934 2935 /* Check for PF reset requests */ 2936 if (!ice_testandclear_state(&sc->state, ICE_STATE_RESET_PFR_REQ)) 2937 return; 2938 2939 /* Make sure we're prepared for reset */ 2940 ice_prepare_for_reset(sc); 2941 2942 /* 2943 * Release the iflib context lock and wait for the device to finish 2944 * resetting. 2945 */ 2946 IFLIB_CTX_UNLOCK(sc); 2947 status = ice_reset(hw, ICE_RESET_PFR); 2948 IFLIB_CTX_LOCK(sc); 2949 if (status) { 2950 device_printf(sc->dev, "device PF reset failed, err %s\n", 2951 ice_status_str(status)); 2952 ice_set_state(&sc->state, ICE_STATE_RESET_FAILED); 2953 return; 2954 } 2955 2956 sc->soft_stats.pfr_count++; 2957 ice_rebuild(sc); 2958 } 2959 2960 /** 2961 * ice_init_device_features - Init device driver features 2962 * @sc: driver softc structure 2963 * 2964 * @pre assumes that the function capabilities bits have been set up by 2965 * ice_init_hw(). 2966 */ 2967 static void 2968 ice_init_device_features(struct ice_softc *sc) 2969 { 2970 struct ice_hw *hw = &sc->hw; 2971 2972 /* Set capabilities that all devices support */ 2973 ice_set_bit(ICE_FEATURE_SRIOV, sc->feat_cap); 2974 ice_set_bit(ICE_FEATURE_RSS, sc->feat_cap); 2975 ice_set_bit(ICE_FEATURE_RDMA, sc->feat_cap); 2976 ice_set_bit(ICE_FEATURE_LENIENT_LINK_MODE, sc->feat_cap); 2977 ice_set_bit(ICE_FEATURE_LINK_MGMT_VER_1, sc->feat_cap); 2978 ice_set_bit(ICE_FEATURE_LINK_MGMT_VER_2, sc->feat_cap); 2979 ice_set_bit(ICE_FEATURE_HEALTH_STATUS, sc->feat_cap); 2980 ice_set_bit(ICE_FEATURE_FW_LOGGING, sc->feat_cap); 2981 ice_set_bit(ICE_FEATURE_HAS_PBA, sc->feat_cap); 2982 ice_set_bit(ICE_FEATURE_DCB, sc->feat_cap); 2983 ice_set_bit(ICE_FEATURE_TX_BALANCE, sc->feat_cap); 2984 ice_set_bit(ICE_FEATURE_PHY_STATISTICS, sc->feat_cap); 2985 2986 if (ice_is_e810(hw)) 2987 ice_set_bit(ICE_FEATURE_PHY_STATISTICS, sc->feat_en); 2988 2989 if (ice_is_e825c(hw)) 2990 ice_set_bit(ICE_FEATURE_DUAL_NAC, sc->feat_cap); 2991 /* Disable features due to hardware limitations... */ 2992 if (!hw->func_caps.common_cap.rss_table_size) 2993 ice_clear_bit(ICE_FEATURE_RSS, sc->feat_cap); 2994 if (!hw->func_caps.common_cap.iwarp || !ice_enable_irdma) 2995 ice_clear_bit(ICE_FEATURE_RDMA, sc->feat_cap); 2996 if (!hw->func_caps.common_cap.dcb) 2997 ice_clear_bit(ICE_FEATURE_DCB, sc->feat_cap); 2998 /* Disable features due to firmware limitations... */ 2999 if (!ice_is_fw_health_report_supported(hw)) 3000 ice_clear_bit(ICE_FEATURE_HEALTH_STATUS, sc->feat_cap); 3001 if (!ice_fwlog_supported(hw)) 3002 ice_clear_bit(ICE_FEATURE_FW_LOGGING, sc->feat_cap); 3003 if (hw->fwlog_cfg.options & ICE_FWLOG_OPTION_IS_REGISTERED) { 3004 if (ice_is_bit_set(sc->feat_cap, ICE_FEATURE_FW_LOGGING)) 3005 ice_set_bit(ICE_FEATURE_FW_LOGGING, sc->feat_en); 3006 else 3007 ice_fwlog_unregister(hw); 3008 } 3009 3010 /* Disable capabilities not supported by the OS */ 3011 ice_disable_unsupported_features(sc->feat_cap); 3012 3013 /* RSS is always enabled for iflib */ 3014 if (ice_is_bit_set(sc->feat_cap, ICE_FEATURE_RSS)) 3015 ice_set_bit(ICE_FEATURE_RSS, sc->feat_en); 3016 3017 /* Disable features based on sysctl settings */ 3018 if (!ice_tx_balance_en) 3019 ice_clear_bit(ICE_FEATURE_TX_BALANCE, sc->feat_cap); 3020 3021 if (hw->dev_caps.supported_sensors & ICE_SENSOR_SUPPORT_E810_INT_TEMP) { 3022 ice_set_bit(ICE_FEATURE_TEMP_SENSOR, sc->feat_cap); 3023 ice_set_bit(ICE_FEATURE_TEMP_SENSOR, sc->feat_en); 3024 } 3025 3026 if (hw->func_caps.common_cap.next_cluster_id_support || 3027 hw->dev_caps.common_cap.next_cluster_id_support) { 3028 ice_set_bit(ICE_FEATURE_NEXT_CLUSTER_ID, sc->feat_cap); 3029 ice_set_bit(ICE_FEATURE_NEXT_CLUSTER_ID, sc->feat_en); 3030 } 3031 } 3032 3033 /** 3034 * ice_if_multi_set - Callback to update Multicast filters in HW 3035 * @ctx: iflib ctx structure 3036 * 3037 * Called by iflib in response to SIOCDELMULTI and SIOCADDMULTI. Must search 3038 * the if_multiaddrs list and determine which filters have been added or 3039 * removed from the list, and update HW programming to reflect the new list. 3040 * 3041 * @pre assumes the caller holds the iflib CTX lock 3042 */ 3043 static void 3044 ice_if_multi_set(if_ctx_t ctx) 3045 { 3046 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3047 int err; 3048 3049 ASSERT_CTX_LOCKED(sc); 3050 3051 /* Do not handle multicast configuration in recovery mode */ 3052 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 3053 return; 3054 3055 err = ice_sync_multicast_filters(sc); 3056 if (err) { 3057 device_printf(sc->dev, 3058 "Failed to synchronize multicast filter list: %s\n", 3059 ice_err_str(err)); 3060 return; 3061 } 3062 } 3063 3064 /** 3065 * ice_if_vlan_register - Register a VLAN with the hardware 3066 * @ctx: iflib ctx pointer 3067 * @vtag: VLAN to add 3068 * 3069 * Programs the main PF VSI with a hardware filter for the given VLAN. 3070 * 3071 * @pre assumes the caller holds the iflib CTX lock 3072 */ 3073 static void 3074 ice_if_vlan_register(if_ctx_t ctx, u16 vtag) 3075 { 3076 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3077 int status; 3078 3079 ASSERT_CTX_LOCKED(sc); 3080 3081 /* Do not handle VLAN configuration in recovery mode */ 3082 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 3083 return; 3084 3085 status = ice_add_vlan_hw_filter(&sc->pf_vsi, vtag); 3086 if (status) { 3087 device_printf(sc->dev, 3088 "Failure adding VLAN %d to main VSI, err %s aq_err %s\n", 3089 vtag, ice_status_str(status), 3090 ice_aq_str(sc->hw.adminq.sq_last_status)); 3091 } 3092 } 3093 3094 /** 3095 * ice_if_vlan_unregister - Remove a VLAN filter from the hardware 3096 * @ctx: iflib ctx pointer 3097 * @vtag: VLAN to add 3098 * 3099 * Removes the previously programmed VLAN filter from the main PF VSI. 3100 * 3101 * @pre assumes the caller holds the iflib CTX lock 3102 */ 3103 static void 3104 ice_if_vlan_unregister(if_ctx_t ctx, u16 vtag) 3105 { 3106 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3107 int status; 3108 3109 ASSERT_CTX_LOCKED(sc); 3110 3111 /* Do not handle VLAN configuration in recovery mode */ 3112 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 3113 return; 3114 3115 status = ice_remove_vlan_hw_filter(&sc->pf_vsi, vtag); 3116 if (status) { 3117 device_printf(sc->dev, 3118 "Failure removing VLAN %d from main VSI, err %s aq_err %s\n", 3119 vtag, ice_status_str(status), 3120 ice_aq_str(sc->hw.adminq.sq_last_status)); 3121 } 3122 } 3123 3124 /** 3125 * ice_if_stop - Stop the device 3126 * @ctx: iflib context structure 3127 * 3128 * Called by iflib to stop the device and bring it down. (i.e. ifconfig ice0 3129 * down) 3130 * 3131 * @pre assumes the caller holds the iflib CTX lock 3132 */ 3133 static void 3134 ice_if_stop(if_ctx_t ctx) 3135 { 3136 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 3137 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3138 3139 ASSERT_CTX_LOCKED(sc); 3140 3141 /* 3142 * The iflib core may call IFDI_STOP prior to the first call to 3143 * IFDI_INIT. This will cause us to attempt to remove MAC filters we 3144 * don't have, and disable Tx queues which aren't yet configured. 3145 * Although it is likely these extra operations are harmless, they do 3146 * cause spurious warning messages to be displayed, which may confuse 3147 * users. 3148 * 3149 * To avoid these messages, we use a state bit indicating if we've 3150 * been initialized. It will be set when ice_if_init is called, and 3151 * cleared here in ice_if_stop. 3152 */ 3153 if (!ice_testandclear_state(&sc->state, ICE_STATE_DRIVER_INITIALIZED)) 3154 return; 3155 3156 if (ice_test_state(&sc->state, ICE_STATE_RESET_FAILED)) { 3157 device_printf(sc->dev, "request to stop interface cannot be completed as the device failed to reset\n"); 3158 return; 3159 } 3160 3161 if (ice_test_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET)) { 3162 device_printf(sc->dev, "request to stop interface while device is prepared for impending reset\n"); 3163 return; 3164 } 3165 3166 ice_rdma_pf_stop(sc); 3167 3168 /* Remove the MAC filters, stop Tx, and stop Rx. We don't check the 3169 * return of these functions because there's nothing we can really do 3170 * if they fail, and the functions already print error messages. 3171 * Just try to shut down as much as we can. 3172 */ 3173 ice_rm_pf_default_mac_filters(sc); 3174 3175 /* Dissociate the Tx and Rx queues from the interrupts */ 3176 ice_flush_txq_interrupts(&sc->pf_vsi); 3177 ice_flush_rxq_interrupts(&sc->pf_vsi); 3178 3179 /* Disable the Tx and Rx queues */ 3180 ice_vsi_disable_tx(&sc->pf_vsi); 3181 ice_control_all_rx_queues(&sc->pf_vsi, false); 3182 3183 if (!ice_test_state(&sc->state, ICE_STATE_LINK_ACTIVE_ON_DOWN) && 3184 !(if_getflags(sc->ifp) & IFF_UP) && sc->link_up) 3185 ice_set_link(sc, false); 3186 3187 if (sc->mirr_if && ice_test_state(&mif->state, ICE_STATE_SUBIF_NEEDS_REINIT)) { 3188 ice_subif_if_stop(sc->mirr_if->subctx); 3189 device_printf(sc->dev, "The subinterface also comes down and up after reset\n"); 3190 } 3191 } 3192 3193 /** 3194 * ice_if_get_counter - Get current value of an ifnet statistic 3195 * @ctx: iflib context pointer 3196 * @counter: ifnet counter to read 3197 * 3198 * Reads the current value of an ifnet counter for the device. 3199 * 3200 * This function is not protected by the iflib CTX lock. 3201 */ 3202 static uint64_t 3203 ice_if_get_counter(if_ctx_t ctx, ift_counter counter) 3204 { 3205 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3206 3207 /* Return the counter for the main PF VSI */ 3208 return ice_get_ifnet_counter(&sc->pf_vsi, counter); 3209 } 3210 3211 /** 3212 * ice_request_stack_reinit - Request that iflib re-initialize 3213 * @sc: the device private softc 3214 * 3215 * Request that the device be brought down and up, to re-initialize. For 3216 * example, this may be called when a device reset occurs, or when Tx and Rx 3217 * queues need to be re-initialized. 3218 * 3219 * This is required because the iflib state is outside the driver, and must be 3220 * re-initialized if we need to resart Tx and Rx queues. 3221 */ 3222 void 3223 ice_request_stack_reinit(struct ice_softc *sc) 3224 { 3225 if (CTX_ACTIVE(sc->ctx)) { 3226 iflib_request_reset(sc->ctx); 3227 iflib_admin_intr_deferred(sc->ctx); 3228 } 3229 } 3230 3231 /** 3232 * ice_driver_is_detaching - Check if the driver is detaching/unloading 3233 * @sc: device private softc 3234 * 3235 * Returns true if the driver is detaching, false otherwise. 3236 * 3237 * @remark on newer kernels, take advantage of iflib_in_detach in order to 3238 * report detachment correctly as early as possible. 3239 * 3240 * @remark this function is used by various code paths that want to avoid 3241 * running if the driver is about to be removed. This includes sysctls and 3242 * other driver access points. Note that it does not fully resolve 3243 * detach-based race conditions as it is possible for a thread to race with 3244 * iflib_in_detach. 3245 */ 3246 bool 3247 ice_driver_is_detaching(struct ice_softc *sc) 3248 { 3249 return (ice_test_state(&sc->state, ICE_STATE_DETACHING) || 3250 iflib_in_detach(sc->ctx)); 3251 } 3252 3253 /** 3254 * ice_if_priv_ioctl - Device private ioctl handler 3255 * @ctx: iflib context pointer 3256 * @command: The ioctl command issued 3257 * @data: ioctl specific data 3258 * 3259 * iflib callback for handling custom driver specific ioctls. 3260 * 3261 * @pre Assumes that the iflib context lock is held. 3262 */ 3263 static int 3264 ice_if_priv_ioctl(if_ctx_t ctx, u_long command, caddr_t data) 3265 { 3266 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3267 struct ifdrv *ifd; 3268 device_t dev = sc->dev; 3269 3270 if (data == NULL) 3271 return (EINVAL); 3272 3273 ASSERT_CTX_LOCKED(sc); 3274 3275 /* Make sure the command type is valid */ 3276 switch (command) { 3277 case SIOCSDRVSPEC: 3278 case SIOCGDRVSPEC: 3279 /* Accepted commands */ 3280 break; 3281 case SIOCGPRIVATE_0: 3282 /* 3283 * Although we do not support this ioctl command, it's 3284 * expected that iflib will forward it to the IFDI_PRIV_IOCTL 3285 * handler. Do not print a message in this case 3286 */ 3287 return (ENOTSUP); 3288 default: 3289 /* 3290 * If we get a different command for this function, it's 3291 * definitely unexpected, so log a message indicating what 3292 * command we got for debugging purposes. 3293 */ 3294 device_printf(dev, "%s: unexpected ioctl command %08lx\n", 3295 __func__, command); 3296 return (EINVAL); 3297 } 3298 3299 ifd = (struct ifdrv *)data; 3300 3301 switch (ifd->ifd_cmd) { 3302 case ICE_NVM_ACCESS: 3303 return ice_handle_nvm_access_ioctl(sc, ifd); 3304 case ICE_DEBUG_DUMP: 3305 return ice_handle_debug_dump_ioctl(sc, ifd); 3306 default: 3307 return EINVAL; 3308 } 3309 } 3310 3311 /** 3312 * ice_if_i2c_req - I2C request handler for iflib 3313 * @ctx: iflib context pointer 3314 * @req: The I2C parameters to use 3315 * 3316 * Read from the port's I2C eeprom using the parameters from the ioctl. 3317 * 3318 * @remark The iflib-only part is pretty simple. 3319 */ 3320 static int 3321 ice_if_i2c_req(if_ctx_t ctx, struct ifi2creq *req) 3322 { 3323 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3324 3325 return ice_handle_i2c_req(sc, req); 3326 } 3327 3328 /** 3329 * ice_if_suspend - PCI device suspend handler for iflib 3330 * @ctx: iflib context pointer 3331 * 3332 * Deinitializes the driver and clears HW resources in preparation for 3333 * suspend or an FLR. 3334 * 3335 * @returns 0; this return value is ignored 3336 */ 3337 static int 3338 ice_if_suspend(if_ctx_t ctx) 3339 { 3340 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3341 3342 /* At least a PFR is always going to happen after this; 3343 * either via FLR or during the D3->D0 transition. 3344 */ 3345 ice_clear_state(&sc->state, ICE_STATE_RESET_PFR_REQ); 3346 3347 ice_prepare_for_reset(sc); 3348 3349 return (0); 3350 } 3351 3352 /** 3353 * ice_if_resume - PCI device resume handler for iflib 3354 * @ctx: iflib context pointer 3355 * 3356 * Reinitializes the driver and the HW after PCI resume or after 3357 * an FLR. An init is performed by iflib after this function is finished. 3358 * 3359 * @returns 0; this return value is ignored 3360 */ 3361 static int 3362 ice_if_resume(if_ctx_t ctx) 3363 { 3364 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3365 3366 ice_rebuild(sc); 3367 3368 return (0); 3369 } 3370 3371 /** 3372 * ice_if_needs_restart - Tell iflib when the driver needs to be reinitialized 3373 * @ctx: iflib context pointer 3374 * @event: event code to check 3375 * 3376 * Defaults to returning true for unknown events. 3377 * 3378 * @returns true if iflib needs to reinit the interface 3379 */ 3380 static bool 3381 ice_if_needs_restart(if_ctx_t ctx, enum iflib_restart_event event) 3382 { 3383 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3384 3385 switch (event) { 3386 case IFLIB_RESTART_VLAN_CONFIG: 3387 if (!ice_test_state(&sc->state, ICE_STATE_LINK_ACTIVE_ON_DOWN) && 3388 !(if_getflags(sc->ifp) & IFF_UP)) 3389 return false; 3390 default: 3391 return true; 3392 } 3393 } 3394 3395 /** 3396 * ice_init_link - Do link configuration and link status reporting 3397 * @sc: driver private structure 3398 * 3399 * Contains an extra check that skips link config when an E830 device 3400 * does not have the "FW_LOADING"/"PHYBUSY" bit set in GL_MNG_FWSM set. 3401 */ 3402 static void 3403 ice_init_link(struct ice_softc *sc) 3404 { 3405 struct ice_hw *hw = &sc->hw; 3406 device_t dev = sc->dev; 3407 3408 /* Check if FW is ready before setting up link; defer setup to the 3409 * admin task if it isn't. 3410 */ 3411 if (ice_is_e830(hw) && 3412 (rd32(hw, GL_MNG_FWSM) & GL_MNG_FWSM_FW_LOADING_M)) { 3413 ice_set_state(&sc->state, ICE_STATE_PHY_FW_INIT_PENDING); 3414 device_printf(dev, 3415 "Link initialization is blocked by PHY FW initialization.\n"); 3416 device_printf(dev, 3417 "Link initialization will continue after PHY FW initialization completes.\n"); 3418 /* Do not access PHY config while PHY FW is busy initializing */ 3419 } else { 3420 ice_clear_state(&sc->state, ICE_STATE_PHY_FW_INIT_PENDING); 3421 ice_init_link_configuration(sc); 3422 ice_update_link_status(sc, true); 3423 } 3424 3425 } 3426 3427 #ifdef PCI_IOV 3428 /** 3429 * ice_if_iov_init - iov init handler for iflib 3430 * @ctx: iflib context pointer 3431 * @num_vfs: number of VFs to create 3432 * @params: configuration parameters for the PF 3433 * 3434 * Configure the driver for SR-IOV mode. Used to setup things like memory 3435 * before any VFs are created. 3436 * 3437 * @remark This is a wrapper for ice_iov_init 3438 */ 3439 static int 3440 ice_if_iov_init(if_ctx_t ctx, uint16_t num_vfs, const nvlist_t *params) 3441 { 3442 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3443 3444 return ice_iov_init(sc, num_vfs, params); 3445 } 3446 3447 /** 3448 * ice_if_iov_uninit - iov uninit handler for iflib 3449 * @ctx: iflib context pointer 3450 * 3451 * Destroys VFs and frees their memory and resources. 3452 * 3453 * @remark This is a wrapper for ice_iov_uninit 3454 */ 3455 static void 3456 ice_if_iov_uninit(if_ctx_t ctx) 3457 { 3458 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3459 3460 ice_iov_uninit(sc); 3461 } 3462 3463 /** 3464 * ice_if_iov_vf_add - iov add vf handler for iflib 3465 * @ctx: iflib context pointer 3466 * @vfnum: index of VF to configure 3467 * @params: configuration parameters for the VF 3468 * 3469 * Sets up the VF given by the vfnum index. This is called by the OS 3470 * for each VF created by the PF driver after it is spawned. 3471 * 3472 * @remark This is a wrapper for ice_iov_vf_add 3473 */ 3474 static int 3475 ice_if_iov_vf_add(if_ctx_t ctx, uint16_t vfnum, const nvlist_t *params) 3476 { 3477 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3478 3479 return ice_iov_add_vf(sc, vfnum, params); 3480 } 3481 3482 /** 3483 * ice_if_vflr_handle - iov VFLR handler 3484 * @ctx: iflib context pointer 3485 * 3486 * Performs the necessar teardown or setup required for a VF after 3487 * a VFLR is initiated. 3488 * 3489 * @remark This is a wrapper for ice_iov_handle_vflr 3490 */ 3491 static void 3492 ice_if_vflr_handle(if_ctx_t ctx) 3493 { 3494 struct ice_softc *sc = (struct ice_softc *)iflib_get_softc(ctx); 3495 ice_iov_handle_vflr(sc); 3496 } 3497 #endif /* PCI_IOV */ 3498 3499 extern struct if_txrx ice_subif_txrx; 3500 3501 /** 3502 * @var ice_subif_methods 3503 * @brief ice driver method entry points 3504 */ 3505 static device_method_t ice_subif_methods[] = { 3506 /* Device interface */ 3507 DEVMETHOD(device_register, ice_subif_register), 3508 DEVMETHOD_END 3509 }; 3510 3511 /** 3512 * @var ice_subif_driver 3513 * @brief driver structure for the device API 3514 */ 3515 static driver_t ice_subif_driver = { 3516 .name = "ice_subif", 3517 .methods = ice_subif_methods, 3518 .size = sizeof(struct ice_mirr_if), 3519 }; 3520 3521 static device_method_t ice_iflib_subif_methods[] = { 3522 DEVMETHOD(ifdi_attach_pre, ice_subif_if_attach_pre), 3523 DEVMETHOD(ifdi_attach_post, ice_subif_if_attach_post), 3524 DEVMETHOD(ifdi_tx_queues_alloc, ice_subif_if_tx_queues_alloc), 3525 DEVMETHOD(ifdi_rx_queues_alloc, ice_subif_if_rx_queues_alloc), 3526 DEVMETHOD(ifdi_msix_intr_assign, ice_subif_if_msix_intr_assign), 3527 DEVMETHOD(ifdi_intr_enable, ice_subif_if_intr_enable), 3528 DEVMETHOD(ifdi_rx_queue_intr_enable, ice_subif_if_rx_queue_intr_enable), 3529 DEVMETHOD(ifdi_tx_queue_intr_enable, ice_subif_if_tx_queue_intr_enable), 3530 DEVMETHOD(ifdi_init, ice_subif_if_init), 3531 DEVMETHOD(ifdi_stop, ice_subif_if_stop), 3532 DEVMETHOD(ifdi_queues_free, ice_subif_if_queues_free), 3533 DEVMETHOD(ifdi_media_status, ice_subif_if_media_status), 3534 DEVMETHOD(ifdi_promisc_set, ice_subif_if_promisc_set), 3535 }; 3536 3537 /** 3538 * @var ice_iflib_subif_driver 3539 * @brief driver structure for the iflib stack 3540 * 3541 * driver_t definition used to setup the iflib device methods. 3542 */ 3543 static driver_t ice_iflib_subif_driver = { 3544 .name = "ice_subif", 3545 .methods = ice_iflib_subif_methods, 3546 .size = sizeof(struct ice_mirr_if), 3547 }; 3548 3549 /** 3550 * @var ice_subif_sctx 3551 * @brief ice driver shared context 3552 * 3553 * Similar to the existing ice_sctx, this structure has these differences: 3554 * - isc_admin_intrcnt is set to 0 3555 * - Uses subif iflib driver methods 3556 * - Flagged as a VF for iflib 3557 */ 3558 static struct if_shared_ctx ice_subif_sctx = { 3559 .isc_magic = IFLIB_MAGIC, 3560 .isc_q_align = PAGE_SIZE, 3561 3562 .isc_tx_maxsize = ICE_MAX_FRAME_SIZE, 3563 .isc_tx_maxsegsize = ICE_MAX_FRAME_SIZE, 3564 .isc_tso_maxsize = ICE_TSO_SIZE + sizeof(struct ether_vlan_header), 3565 .isc_tso_maxsegsize = ICE_MAX_DMA_SEG_SIZE, 3566 3567 .isc_rx_maxsize = ICE_MAX_FRAME_SIZE, 3568 .isc_rx_nsegments = ICE_MAX_RX_SEGS, 3569 .isc_rx_maxsegsize = ICE_MAX_FRAME_SIZE, 3570 3571 .isc_nfl = 1, 3572 .isc_ntxqs = 1, 3573 .isc_nrxqs = 1, 3574 3575 .isc_admin_intrcnt = 0, 3576 .isc_vendor_info = ice_vendor_info_array, 3577 .isc_driver_version = __DECONST(char *, ice_driver_version), 3578 .isc_driver = &ice_iflib_subif_driver, 3579 3580 .isc_flags = IFLIB_NEED_SCRATCH | IFLIB_TSO_INIT_IP | 3581 IFLIB_ADMIN_ALWAYS_RUN | IFLIB_SKIP_MSIX | 3582 IFLIB_IS_VF, 3583 3584 .isc_nrxd_min = {ICE_MIN_DESC_COUNT}, 3585 .isc_ntxd_min = {ICE_MIN_DESC_COUNT}, 3586 .isc_nrxd_max = {ICE_IFLIB_MAX_DESC_COUNT}, 3587 .isc_ntxd_max = {ICE_IFLIB_MAX_DESC_COUNT}, 3588 .isc_nrxd_default = {ICE_DEFAULT_DESC_COUNT}, 3589 .isc_ntxd_default = {ICE_DEFAULT_DESC_COUNT}, 3590 }; 3591 3592 static void * 3593 ice_subif_register(device_t dev __unused) 3594 { 3595 return (&ice_subif_sctx); 3596 } 3597 3598 static void 3599 ice_subif_setup_scctx(struct ice_mirr_if *mif) 3600 { 3601 if_softc_ctx_t scctx = mif->subscctx; 3602 3603 scctx->isc_txrx = &ice_subif_txrx; 3604 3605 scctx->isc_capenable = ICE_FULL_CAPS; 3606 scctx->isc_tx_csum_flags = ICE_CSUM_OFFLOAD; 3607 3608 scctx->isc_ntxqsets = 4; 3609 scctx->isc_nrxqsets = 4; 3610 scctx->isc_vectors = scctx->isc_nrxqsets; 3611 3612 scctx->isc_ntxqsets_max = 256; 3613 scctx->isc_nrxqsets_max = 256; 3614 3615 scctx->isc_txqsizes[0] = roundup2(scctx->isc_ntxd[0] 3616 * sizeof(struct ice_tx_desc), DBA_ALIGN); 3617 scctx->isc_rxqsizes[0] = roundup2(scctx->isc_nrxd[0] 3618 * sizeof(union ice_32b_rx_flex_desc), DBA_ALIGN); 3619 3620 scctx->isc_tx_nsegments = ICE_MAX_TX_SEGS; 3621 scctx->isc_tx_tso_segments_max = ICE_MAX_TSO_SEGS; 3622 scctx->isc_tx_tso_size_max = ICE_TSO_SIZE; 3623 scctx->isc_tx_tso_segsize_max = ICE_MAX_DMA_SEG_SIZE; 3624 } 3625 3626 static int 3627 ice_subif_if_attach_pre(if_ctx_t ctx) 3628 { 3629 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 3630 device_t dev = iflib_get_dev(ctx); 3631 3632 mif->subctx = ctx; 3633 mif->subdev = dev; 3634 mif->subscctx = iflib_get_softc_ctx(ctx); 3635 3636 /* Setup the iflib softc context structure */ 3637 ice_subif_setup_scctx(mif); 3638 3639 return (0); 3640 } 3641 3642 static int 3643 ice_subif_if_attach_post(if_ctx_t ctx __unused) 3644 { 3645 return (0); 3646 } 3647 3648 /** 3649 * ice_destroy_mirror_interface - destroy mirror interface 3650 * @sc: driver private data 3651 * 3652 * Destroys all resources associated with the mirroring interface. 3653 * Will not exit early on failure. 3654 * 3655 * @pre: Mirror interface already exists and is initialized. 3656 */ 3657 void 3658 ice_destroy_mirror_interface(struct ice_softc *sc) 3659 { 3660 struct ice_mirr_if *mif = sc->mirr_if; 3661 struct ice_vsi *vsi = mif->vsi; 3662 bool is_locked = false; 3663 int ret; 3664 3665 is_locked = sx_xlocked(sc->iflib_ctx_lock); 3666 if (is_locked) 3667 IFLIB_CTX_UNLOCK(sc); 3668 3669 if (mif->ifp) { 3670 ret = iflib_device_deregister(mif->subctx); 3671 if (ret) { 3672 device_printf(sc->dev, 3673 "iflib_device_deregister for mirror interface failed: %d\n", 3674 ret); 3675 } 3676 } 3677 3678 bus_topo_lock(); 3679 ret = device_delete_child(sc->dev, mif->subdev); 3680 bus_topo_unlock(); 3681 if (ret) { 3682 device_printf(sc->dev, 3683 "device_delete_child for mirror interface failed: %d\n", 3684 ret); 3685 } 3686 3687 if (is_locked) 3688 IFLIB_CTX_LOCK(sc); 3689 3690 if (mif->if_imap) { 3691 free(mif->if_imap, M_ICE); 3692 mif->if_imap = NULL; 3693 } 3694 if (mif->os_imap) { 3695 free(mif->os_imap, M_ICE); 3696 mif->os_imap = NULL; 3697 } 3698 3699 /* These are freed via ice_subif_queues_free_subif 3700 * vsi: 3701 * - rx_irqvs 3702 * - tx_queues 3703 * - rx_queues 3704 */ 3705 ice_release_vsi(vsi); 3706 3707 free(mif, M_ICE); 3708 sc->mirr_if = NULL; 3709 3710 } 3711 3712 /** 3713 * ice_setup_mirror_vsi - Initialize mirror VSI 3714 * @mif: driver private data for mirror interface 3715 * 3716 * Allocates a VSI for a mirror interface, and sets that VSI up for use as a 3717 * mirror for the main PF VSI. 3718 * 3719 * Returns 0 on success, or a standard error code on failure. 3720 */ 3721 static int 3722 ice_setup_mirror_vsi(struct ice_mirr_if *mif) 3723 { 3724 struct ice_softc *sc = mif->back; 3725 device_t dev = sc->dev; 3726 struct ice_vsi *vsi; 3727 int ret = 0; 3728 3729 /* vsi is for the new mirror vsi, not the PF's main VSI */ 3730 vsi = ice_alloc_vsi(sc, ICE_VSI_VMDQ2); 3731 if (!vsi) { 3732 /* Already prints an error message */ 3733 return (ENOMEM); 3734 } 3735 mif->vsi = vsi; 3736 3737 /* Reserve VSI queue allocation from PF queues */ 3738 ice_alloc_vsi_qmap(vsi, ICE_DEFAULT_VF_QUEUES, ICE_DEFAULT_VF_QUEUES); 3739 vsi->num_tx_queues = vsi->num_rx_queues = ICE_DEFAULT_VF_QUEUES; 3740 3741 /* Assign Tx queues from PF space */ 3742 ret = ice_resmgr_assign_scattered(&sc->tx_qmgr, vsi->tx_qmap, 3743 vsi->num_tx_queues); 3744 if (ret) { 3745 device_printf(dev, "Unable to assign mirror VSI Tx queues: %s\n", 3746 ice_err_str(ret)); 3747 goto release_vsi; 3748 } 3749 /* Assign Rx queues from PF space */ 3750 ret = ice_resmgr_assign_scattered(&sc->rx_qmgr, vsi->rx_qmap, 3751 vsi->num_rx_queues); 3752 if (ret) { 3753 device_printf(dev, "Unable to assign mirror VSI Rx queues: %s\n", 3754 ice_err_str(ret)); 3755 goto release_vsi; 3756 } 3757 vsi->qmap_type = ICE_RESMGR_ALLOC_SCATTERED; 3758 vsi->max_frame_size = ICE_MAX_FRAME_SIZE; 3759 3760 ret = ice_initialize_vsi(vsi); 3761 if (ret) { 3762 device_printf(dev, "%s: Error in ice_initialize_vsi for mirror VSI: %s\n", 3763 __func__, ice_err_str(ret)); 3764 goto release_vsi; 3765 } 3766 3767 /* Setup this VSI for receiving traffic */ 3768 ret = ice_config_rss(vsi); 3769 if (ret) { 3770 device_printf(dev, 3771 "Unable to configure RSS for mirror VSI: %s\n", 3772 ice_err_str(ret)); 3773 goto release_vsi; 3774 } 3775 3776 /* Set HW rules for mirroring traffic */ 3777 vsi->mirror_src_vsi = sc->pf_vsi.idx; 3778 3779 ice_debug(&sc->hw, ICE_DBG_INIT, 3780 "Configuring mirroring from VSI %d to %d\n", 3781 vsi->mirror_src_vsi, vsi->idx); 3782 ice_debug(&sc->hw, ICE_DBG_INIT, "(HW num: VSI %d to %d)\n", 3783 ice_get_hw_vsi_num(&sc->hw, vsi->mirror_src_vsi), 3784 ice_get_hw_vsi_num(&sc->hw, vsi->idx)); 3785 3786 ret = ice_setup_vsi_mirroring(vsi); 3787 if (ret) { 3788 device_printf(dev, 3789 "Unable to configure mirroring for VSI: %s\n", 3790 ice_err_str(ret)); 3791 goto release_vsi; 3792 } 3793 3794 return (0); 3795 3796 release_vsi: 3797 ice_release_vsi(vsi); 3798 mif->vsi = NULL; 3799 return (ret); 3800 } 3801 3802 /** 3803 * ice_create_mirror_interface - Initialize mirror interface 3804 * @sc: driver private data 3805 * 3806 * Creates and sets up a mirror interface that will mirror traffic from 3807 * the main PF interface. Includes a call to iflib_device_register() in order 3808 * to setup necessary iflib structures for this new interface as well. 3809 * 3810 * If it returns successfully, a new interface will be created and will show 3811 * up in the ifconfig interface list. 3812 * 3813 * Returns 0 on success, or a standard error code on failure. 3814 */ 3815 int 3816 ice_create_mirror_interface(struct ice_softc *sc) 3817 { 3818 device_t dev = sc->dev; 3819 struct ice_mirr_if *mif; 3820 struct ifmedia *media; 3821 struct sbuf *sb; 3822 int ret = 0; 3823 3824 mif = (struct ice_mirr_if *)malloc(sizeof(*mif), M_ICE, M_ZERO | M_NOWAIT); 3825 if (!mif) { 3826 device_printf(dev, "malloc() error allocating mirror interface\n"); 3827 return (ENOMEM); 3828 } 3829 3830 /* Set pointers */ 3831 sc->mirr_if = mif; 3832 mif->back = sc; 3833 3834 /* Do early setup because these will be called during iflib_device_register(): 3835 * - ice_subif_if_tx_queues_alloc 3836 * - ice_subif_if_rx_queues_alloc 3837 */ 3838 ret = ice_setup_mirror_vsi(mif); 3839 if (ret) 3840 goto out; 3841 3842 /* Determine name for new interface: 3843 * (base interface name)(modifier name)(modifier unit number) 3844 * e.g. for ice0 with a new mirror interface (modifier m) 3845 * of index 0, this equals "ice0m0" 3846 */ 3847 sb = sbuf_new_auto(); 3848 MPASS(sb != NULL); 3849 sbuf_printf(sb, "%sm", device_get_nameunit(dev)); 3850 sbuf_finish(sb); 3851 3852 bus_topo_lock(); 3853 mif->subdev = device_add_child(dev, sbuf_data(sb), 0); 3854 bus_topo_unlock(); 3855 3856 if (!mif->subdev) { 3857 device_printf(dev, "device_add_child failed for %s0\n", sbuf_data(sb)); 3858 sbuf_delete(sb); 3859 free(mif, M_ICE); 3860 sc->mirr_if = NULL; 3861 return (ENOMEM); 3862 } 3863 sbuf_delete(sb); 3864 3865 device_set_driver(mif->subdev, &ice_subif_driver); 3866 3867 /* Use iflib_device_register() directly because the driver already 3868 * has an initialized softc to pass to iflib 3869 */ 3870 ret = iflib_device_register(mif->subdev, mif, &ice_subif_sctx, &mif->subctx); 3871 if (ret) 3872 goto out; 3873 3874 /* Indicate that created interface will be just for monitoring */ 3875 mif->ifp = iflib_get_ifp(mif->subctx); 3876 if_setflagbits(mif->ifp, IFF_MONITOR, 0); 3877 3878 /* Use autoselect media by default */ 3879 media = iflib_get_media(mif->subctx); 3880 ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL); 3881 ifmedia_set(media, IFM_ETHER | IFM_AUTO); 3882 3883 device_printf(dev, "Created dev %s and ifnet %s for mirroring\n", 3884 device_get_nameunit(mif->subdev), if_name(mif->ifp)); 3885 3886 ice_add_vsi_sysctls(mif->vsi); 3887 3888 ret = ice_wire_mirror_intrs(mif); 3889 if (ret) 3890 goto out; 3891 3892 mif->if_attached = true; 3893 return (0); 3894 3895 out: 3896 ice_destroy_mirror_interface(sc); 3897 return (ret); 3898 } 3899 3900 /** 3901 * ice_wire_mirror_intrs 3902 * @mif: driver private subinterface structure 3903 * 3904 * Helper function that sets up driver interrupt data and calls 3905 * into iflib in order to setup interrupts in its data structures as well. 3906 * 3907 * Like ice_if_msix_intr_assign, currently requires that we get at least the same 3908 * number of vectors as we have queues, and that we always have the same number 3909 * of Tx and Rx queues. Unlike that function, this calls a special 3910 * iflib_irq_alloc_generic_subif() function for RX interrupts because the 3911 * driver needs to get MSI-X resources from the parent device. 3912 * 3913 * Tx queues use a softirq instead of using their own hardware interrupt so that 3914 * remains unchanged. 3915 * 3916 * Returns 0 on success or an error code from iflib_irq_alloc_generic_subctx() 3917 * on failure. 3918 */ 3919 static int 3920 ice_wire_mirror_intrs(struct ice_mirr_if *mif) 3921 { 3922 struct ice_softc *sc = mif->back; 3923 struct ice_hw *hw = &sc->hw; 3924 struct ice_vsi *vsi = mif->vsi; 3925 device_t dev = mif->subdev; 3926 int err, i, rid; 3927 3928 if_ctx_t ctx = mif->subctx; 3929 3930 ice_debug(hw, ICE_DBG_INIT, "%s: Last rid: %d\n", __func__, sc->last_rid); 3931 3932 rid = sc->last_rid + 1; 3933 for (i = 0; i < vsi->num_rx_queues; i++, rid++) { 3934 struct ice_rx_queue *rxq = &vsi->rx_queues[i]; 3935 struct ice_tx_queue *txq = &vsi->tx_queues[i]; 3936 char irq_name[16]; 3937 3938 // TODO: Change to use dynamic interface number 3939 snprintf(irq_name, sizeof(irq_name), "m0rxq%d", i); 3940 /* First arg is parent device (physical port's) iflib ctx */ 3941 err = iflib_irq_alloc_generic_subctx(sc->ctx, ctx, 3942 &mif->rx_irqvs[i].irq, rid, IFLIB_INTR_RXTX, ice_msix_que, 3943 rxq, rxq->me, irq_name); 3944 if (err) { 3945 device_printf(dev, 3946 "Failed to allocate q int %d err: %s\n", 3947 i, ice_err_str(err)); 3948 i--; 3949 goto fail; 3950 } 3951 MPASS(rid - 1 > 0); 3952 /* Set vector number used in interrupt enable/disable functions */ 3953 mif->rx_irqvs[i].me = rid - 1; 3954 rxq->irqv = &mif->rx_irqvs[i]; 3955 3956 bzero(irq_name, sizeof(irq_name)); 3957 snprintf(irq_name, sizeof(irq_name), "m0txq%d", i); 3958 iflib_softirq_alloc_generic(ctx, &mif->rx_irqvs[i].irq, 3959 IFLIB_INTR_TX, txq, txq->me, irq_name); 3960 txq->irqv = &mif->rx_irqvs[i]; 3961 } 3962 3963 sc->last_rid = rid - 1; 3964 3965 ice_debug(hw, ICE_DBG_INIT, "%s: New last rid: %d\n", __func__, 3966 sc->last_rid); 3967 3968 return (0); 3969 3970 fail: 3971 for (; i >= 0; i--) 3972 iflib_irq_free(ctx, &mif->rx_irqvs[i].irq); 3973 return (err); 3974 } 3975 3976 /** 3977 * ice_subif_rebuild - Rebuild subinterface post reset 3978 * @sc: The device private softc 3979 * 3980 * Restore subinterface state after a reset occurred. 3981 * Restart the VSI and enable the mirroring. 3982 */ 3983 static int 3984 ice_subif_rebuild(struct ice_softc *sc) 3985 { 3986 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(sc->ctx); 3987 struct ice_vsi *vsi = sc->mirr_if->vsi; 3988 int err; 3989 3990 err = ice_subif_rebuild_vsi_qmap(sc); 3991 if (err) { 3992 device_printf(sc->dev, "Unable to re-assign mirror VSI queues, err %s\n", 3993 ice_err_str(err)); 3994 return (err); 3995 } 3996 3997 err = ice_initialize_vsi(vsi); 3998 if (err) { 3999 device_printf(sc->dev, "Unable to re-initialize mirror VSI, err %s\n", 4000 ice_err_str(err)); 4001 goto err_release_queue_allocations_subif; 4002 } 4003 4004 err = ice_config_rss(vsi); 4005 if (err) { 4006 device_printf(sc->dev, 4007 "Unable to reconfigure RSS for the mirror VSI, err %s\n", 4008 ice_err_str(err)); 4009 goto err_deinit_subif_vsi; 4010 } 4011 4012 vsi->mirror_src_vsi = sc->pf_vsi.idx; 4013 4014 err = ice_setup_vsi_mirroring(vsi); 4015 if (err) { 4016 device_printf(sc->dev, 4017 "Unable to configure mirroring for VSI: %s\n", 4018 ice_err_str(err)); 4019 goto err_deinit_subif_vsi; 4020 } 4021 4022 ice_set_state(&mif->state, ICE_STATE_SUBIF_NEEDS_REINIT); 4023 4024 return (0); 4025 4026 err_deinit_subif_vsi: 4027 ice_deinit_vsi(vsi); 4028 err_release_queue_allocations_subif: 4029 ice_resmgr_release_map(&sc->tx_qmgr, vsi->tx_qmap, 4030 sc->mirr_if->num_irq_vectors); 4031 ice_resmgr_release_map(&sc->rx_qmgr, vsi->rx_qmap, 4032 sc->mirr_if->num_irq_vectors); 4033 4034 return (err); 4035 } 4036 4037 /** 4038 * ice_subif_rebuild_vsi_qmap - Rebuild the mirror VSI queue mapping 4039 * @sc: the device softc pointer 4040 * 4041 * Loops over the Tx and Rx queues for the mirror VSI and reassigns the queue 4042 * mapping after a reset occurred. 4043 */ 4044 static int 4045 ice_subif_rebuild_vsi_qmap(struct ice_softc *sc) 4046 { 4047 struct ice_vsi *vsi = sc->mirr_if->vsi; 4048 struct ice_tx_queue *txq; 4049 struct ice_rx_queue *rxq; 4050 int err, i; 4051 4052 err = ice_resmgr_assign_scattered(&sc->tx_qmgr, vsi->tx_qmap, sc->mirr_if->num_irq_vectors); 4053 if (err) { 4054 device_printf(sc->dev, "Unable to assign mirror VSI Tx queues: %s\n", 4055 ice_err_str(err)); 4056 return (err); 4057 } 4058 4059 err = ice_resmgr_assign_scattered(&sc->rx_qmgr, vsi->rx_qmap, sc->mirr_if->num_irq_vectors); 4060 if (err) { 4061 device_printf(sc->dev, "Unable to assign mirror VSI Rx queues: %s\n", 4062 ice_err_str(err)); 4063 goto err_release_tx_queues; 4064 } 4065 4066 vsi->qmap_type = ICE_RESMGR_ALLOC_SCATTERED; 4067 4068 /* Re-assign Tx queue tail pointers */ 4069 for (i = 0, txq = vsi->tx_queues; i < vsi->num_tx_queues; i++, txq++) 4070 txq->tail = QTX_COMM_DBELL(vsi->tx_qmap[i]); 4071 4072 /* Re-assign Rx queue tail pointers */ 4073 for (i = 0, rxq = vsi->rx_queues; i < vsi->num_rx_queues; i++, rxq++) 4074 rxq->tail = QRX_TAIL(vsi->rx_qmap[i]); 4075 4076 return (0); 4077 4078 err_release_tx_queues: 4079 ice_resmgr_release_map(&sc->tx_qmgr, vsi->tx_qmap, vsi->num_tx_queues); 4080 4081 return (err); 4082 } 4083 4084 /** 4085 * ice_subif_if_tx_queues_alloc - Allocate Tx queue memory for subinterfaces 4086 * @ctx: iflib context structure 4087 * @vaddrs: virtual addresses for the queue memory 4088 * @paddrs: physical addresses for the queue memory 4089 * @ntxqs: the number of Tx queues per set (should always be 1) 4090 * @ntxqsets: the number of Tx queue sets to allocate 4091 * 4092 * See ice_if_tx_queues_alloc() description. Similar to that function, but 4093 * for subinterfaces instead. 4094 */ 4095 static int 4096 ice_subif_if_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, 4097 int __invariant_only ntxqs, int ntxqsets) 4098 { 4099 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4100 struct ice_tx_queue *txq; 4101 device_t dev = mif->subdev; 4102 struct ice_vsi *vsi; 4103 int err, i, j; 4104 4105 MPASS(mif != NULL); 4106 MPASS(ntxqs == 1); 4107 MPASS(mif->subscctx->isc_ntxd[0] <= ICE_MAX_DESC_COUNT); 4108 4109 vsi = mif->vsi; 4110 4111 MPASS(vsi->num_tx_queues == ntxqsets); 4112 4113 /* Allocate queue structure memory */ 4114 if (!(vsi->tx_queues = 4115 (struct ice_tx_queue *)malloc(sizeof(struct ice_tx_queue) * ntxqsets, M_ICE, M_NOWAIT | M_ZERO))) { 4116 device_printf(dev, "%s: Unable to allocate Tx queue memory for subfunction\n", 4117 __func__); 4118 return (ENOMEM); 4119 } 4120 4121 /* Allocate report status arrays */ 4122 for (i = 0, txq = vsi->tx_queues; i < ntxqsets; i++, txq++) { 4123 if (!(txq->tx_rsq = 4124 (uint16_t *)malloc(sizeof(uint16_t) * mif->subscctx->isc_ntxd[0], M_ICE, M_NOWAIT))) { 4125 device_printf(dev, 4126 "%s: Unable to allocate tx_rsq memory for subfunction\n", __func__); 4127 err = ENOMEM; 4128 goto free_tx_queues; 4129 } 4130 /* Initialize report status array */ 4131 for (j = 0; j < mif->subscctx->isc_ntxd[0]; j++) 4132 txq->tx_rsq[j] = QIDX_INVALID; 4133 } 4134 4135 /* Add Tx queue sysctls context */ 4136 ice_vsi_add_txqs_ctx(vsi); 4137 4138 for (i = 0, txq = vsi->tx_queues; i < ntxqsets; i++, txq++) { 4139 /* q_handle == me when only one TC */ 4140 txq->me = txq->q_handle = i; 4141 txq->vsi = vsi; 4142 4143 /* store the queue size for easier access */ 4144 txq->desc_count = mif->subscctx->isc_ntxd[0]; 4145 4146 /* get the virtual and physical address of the hardware queues */ 4147 txq->tail = QTX_COMM_DBELL(vsi->tx_qmap[i]); 4148 txq->tx_base = (struct ice_tx_desc *)vaddrs[i]; 4149 txq->tx_paddr = paddrs[i]; 4150 4151 ice_add_txq_sysctls(txq); 4152 } 4153 4154 return (0); 4155 4156 free_tx_queues: 4157 for (i = 0, txq = vsi->tx_queues; i < ntxqsets; i++, txq++) { 4158 if (txq->tx_rsq != NULL) { 4159 free(txq->tx_rsq, M_ICE); 4160 txq->tx_rsq = NULL; 4161 } 4162 } 4163 free(vsi->tx_queues, M_ICE); 4164 vsi->tx_queues = NULL; 4165 return (err); 4166 } 4167 4168 /** 4169 * ice_subif_if_rx_queues_alloc - Allocate Rx queue memory for subinterfaces 4170 * @ctx: iflib context structure 4171 * @vaddrs: virtual addresses for the queue memory 4172 * @paddrs: physical addresses for the queue memory 4173 * @nrxqs: number of Rx queues per set (should always be 1) 4174 * @nrxqsets: number of Rx queue sets to allocate 4175 * 4176 * See ice_if_rx_queues_alloc() for general summary; this is similar to that 4177 * but implemented for subinterfaces. 4178 */ 4179 static int 4180 ice_subif_if_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, 4181 int __invariant_only nrxqs, int nrxqsets) 4182 { 4183 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4184 struct ice_rx_queue *rxq; 4185 device_t dev = mif->subdev; 4186 struct ice_vsi *vsi; 4187 int i; 4188 4189 MPASS(mif != NULL); 4190 MPASS(nrxqs == 1); 4191 MPASS(mif->subscctx->isc_nrxd[0] <= ICE_MAX_DESC_COUNT); 4192 4193 vsi = mif->vsi; 4194 4195 MPASS(vsi->num_rx_queues == nrxqsets); 4196 4197 /* Allocate queue structure memory */ 4198 if (!(vsi->rx_queues = 4199 (struct ice_rx_queue *) malloc(sizeof(struct ice_rx_queue) * nrxqsets, M_ICE, M_NOWAIT | M_ZERO))) { 4200 device_printf(dev, "%s: Unable to allocate Rx queue memory for subfunction\n", 4201 __func__); 4202 return (ENOMEM); 4203 } 4204 4205 /* Add Rx queue sysctls context */ 4206 ice_vsi_add_rxqs_ctx(vsi); 4207 4208 for (i = 0, rxq = vsi->rx_queues; i < nrxqsets; i++, rxq++) { 4209 rxq->me = i; 4210 rxq->vsi = vsi; 4211 4212 /* store the queue size for easier access */ 4213 rxq->desc_count = mif->subscctx->isc_nrxd[0]; 4214 4215 /* get the virtual and physical address of the hardware queues */ 4216 rxq->tail = QRX_TAIL(vsi->rx_qmap[i]); 4217 rxq->rx_base = (union ice_32b_rx_flex_desc *)vaddrs[i]; 4218 rxq->rx_paddr = paddrs[i]; 4219 4220 ice_add_rxq_sysctls(rxq); 4221 } 4222 4223 return (0); 4224 } 4225 4226 /** 4227 * ice_subif_if_msix_intr_assign - Assign MSI-X interrupts to new sub interface 4228 * @ctx: the iflib context structure 4229 * @msix: the number of vectors we were assigned 4230 * 4231 * Allocates and assigns driver private resources for MSI-X interrupt tracking. 4232 * 4233 * @pre OS MSI-X resources have been pre-allocated by parent interface. 4234 */ 4235 static int 4236 ice_subif_if_msix_intr_assign(if_ctx_t ctx, int msix) 4237 { 4238 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4239 struct ice_softc *sc = mif->back; 4240 struct ice_vsi *vsi = mif->vsi; 4241 4242 device_t dev = mif->subdev; 4243 int ret; 4244 4245 if (vsi->num_rx_queues != vsi->num_tx_queues) { 4246 device_printf(dev, 4247 "iflib requested %d Tx queues, and %d Rx queues, but the driver isn't able to support a differing number of Tx and Rx queues\n", 4248 vsi->num_tx_queues, vsi->num_rx_queues); 4249 return (EOPNOTSUPP); 4250 } 4251 4252 if (msix > sc->extra_vectors) { 4253 device_printf(dev, 4254 "%s: Not enough spare (%d) msix vectors for new sub-interface requested (%d)\n", 4255 __func__, sc->extra_vectors, msix); 4256 return (ENOSPC); 4257 } 4258 device_printf(dev, "%s: Using %d vectors for sub-interface\n", __func__, 4259 msix); 4260 4261 /* Allocate space to store the IRQ vector data */ 4262 mif->num_irq_vectors = vsi->num_rx_queues; 4263 mif->rx_irqvs = (struct ice_irq_vector *) 4264 malloc(sizeof(struct ice_irq_vector) * (mif->num_irq_vectors), 4265 M_ICE, M_NOWAIT); 4266 if (!mif->rx_irqvs) { 4267 device_printf(dev, 4268 "Unable to allocate RX irqv memory for mirror's %d vectors\n", 4269 mif->num_irq_vectors); 4270 return (ENOMEM); 4271 } 4272 4273 /* Assign mirror interface interrupts from PF device space */ 4274 if (!(mif->if_imap = 4275 (u16 *)malloc(sizeof(u16) * mif->num_irq_vectors, 4276 M_ICE, M_NOWAIT))) { 4277 device_printf(dev, "Unable to allocate mirror intfc if_imap memory\n"); 4278 ret = ENOMEM; 4279 goto free_irqvs; 4280 } 4281 ret = ice_resmgr_assign_contiguous(&sc->dev_imgr, mif->if_imap, mif->num_irq_vectors); 4282 if (ret) { 4283 device_printf(dev, "Unable to assign mirror intfc PF device interrupt mapping: %s\n", 4284 ice_err_str(ret)); 4285 goto free_if_imap; 4286 } 4287 /* Assign mirror interface interrupts from OS interrupt allocation space */ 4288 if (!(mif->os_imap = 4289 (u16 *)malloc(sizeof(u16) * mif->num_irq_vectors, 4290 M_ICE, M_NOWAIT))) { 4291 device_printf(dev, "Unable to allocate mirror intfc os_imap memory\n"); 4292 ret = ENOMEM; 4293 goto free_if_imap; 4294 } 4295 ret = ice_resmgr_assign_contiguous(&sc->os_imgr, mif->os_imap, mif->num_irq_vectors); 4296 if (ret) { 4297 device_printf(dev, "Unable to assign mirror intfc OS interrupt mapping: %s\n", 4298 ice_err_str(ret)); 4299 goto free_if_imap; 4300 } 4301 4302 return (0); 4303 4304 free_if_imap: 4305 free(mif->if_imap, M_ICE); 4306 mif->if_imap = NULL; 4307 free_irqvs: 4308 free(mif->rx_irqvs, M_ICE); 4309 mif->rx_irqvs = NULL; 4310 return (ret); 4311 } 4312 4313 /** 4314 * ice_subif_if_intr_enable - Enable device interrupts for a subinterface 4315 * @ctx: iflib context structure 4316 * 4317 * Called by iflib to request enabling all interrupts that belong to a 4318 * subinterface. 4319 */ 4320 static void 4321 ice_subif_if_intr_enable(if_ctx_t ctx) 4322 { 4323 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4324 struct ice_softc *sc = mif->back; 4325 struct ice_vsi *vsi = mif->vsi; 4326 struct ice_hw *hw = &sc->hw; 4327 4328 /* Do not enable queue interrupts in recovery mode */ 4329 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 4330 return; 4331 4332 /* Enable all queue interrupts */ 4333 for (int i = 0; i < vsi->num_rx_queues; i++) 4334 ice_enable_intr(hw, vsi->rx_queues[i].irqv->me); 4335 } 4336 4337 /** 4338 * ice_subif_if_rx_queue_intr_enable - Enable a specific Rx queue interrupt 4339 * @ctx: iflib context structure 4340 * @rxqid: the Rx queue to enable 4341 * 4342 * Enable a specific Rx queue interrupt. 4343 * 4344 * This function is not protected by the iflib CTX lock. 4345 */ 4346 static int 4347 ice_subif_if_rx_queue_intr_enable(if_ctx_t ctx, uint16_t rxqid) 4348 { 4349 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4350 struct ice_softc *sc = mif->back; 4351 struct ice_vsi *vsi = mif->vsi; 4352 struct ice_hw *hw = &sc->hw; 4353 4354 /* Do not enable queue interrupts in recovery mode */ 4355 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 4356 return (ENOSYS); 4357 4358 ice_enable_intr(hw, vsi->rx_queues[rxqid].irqv->me); 4359 return (0); 4360 } 4361 4362 /** 4363 * ice_subif_if_tx_queue_intr_enable - Enable a specific Tx queue interrupt 4364 * @ctx: iflib context structure 4365 * @txqid: the Tx queue to enable 4366 * 4367 * Enable a specific Tx queue interrupt. 4368 * 4369 * This function is not protected by the iflib CTX lock. 4370 */ 4371 static int 4372 ice_subif_if_tx_queue_intr_enable(if_ctx_t ctx, uint16_t txqid) 4373 { 4374 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4375 struct ice_softc *sc = mif->back; 4376 struct ice_vsi *vsi = mif->vsi; 4377 struct ice_hw *hw = &sc->hw; 4378 4379 /* Do not enable queue interrupts in recovery mode */ 4380 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 4381 return (ENOSYS); 4382 4383 ice_enable_intr(hw, vsi->tx_queues[txqid].irqv->me); 4384 return (0); 4385 } 4386 4387 /** 4388 * ice_subif_if_init - Initialize the subinterface 4389 * @ctx: iflib ctx structure 4390 * 4391 * Called by iflib to bring the device up, i.e. ifconfig ice0m0 up. 4392 * Prepares the Tx and Rx engines and enables interrupts. 4393 * 4394 * @pre assumes the caller holds the iflib CTX lock 4395 */ 4396 static void 4397 ice_subif_if_init(if_ctx_t ctx) 4398 { 4399 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4400 struct ice_softc *sc = mif->back; 4401 struct ice_vsi *vsi = mif->vsi; 4402 device_t dev = mif->subdev; 4403 int err; 4404 4405 if (ice_driver_is_detaching(sc)) 4406 return; 4407 4408 if (ice_test_state(&sc->state, ICE_STATE_RECOVERY_MODE)) 4409 return; 4410 4411 if (ice_test_state(&sc->state, ICE_STATE_RESET_FAILED)) { 4412 device_printf(dev, 4413 "request to start interface cannot be completed as the parent device %s failed to reset\n", 4414 device_get_nameunit(sc->dev)); 4415 return; 4416 } 4417 4418 if (ice_test_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET)) { 4419 device_printf(dev, 4420 "request to start interface cannot be completed while parent device %s is prepared for impending reset\n", 4421 device_get_nameunit(sc->dev)); 4422 return; 4423 } 4424 4425 /* XXX: Equiv to ice_update_rx_mbuf_sz */ 4426 vsi->mbuf_sz = iflib_get_rx_mbuf_sz(ctx); 4427 4428 /* Initialize software Tx tracking values */ 4429 ice_init_tx_tracking(vsi); 4430 4431 err = ice_cfg_vsi_for_tx(vsi); 4432 if (err) { 4433 device_printf(dev, 4434 "Unable to configure subif VSI for Tx: %s\n", 4435 ice_err_str(err)); 4436 return; 4437 } 4438 4439 err = ice_cfg_vsi_for_rx(vsi); 4440 if (err) { 4441 device_printf(dev, 4442 "Unable to configure subif VSI for Rx: %s\n", 4443 ice_err_str(err)); 4444 goto err_cleanup_tx; 4445 } 4446 4447 err = ice_control_all_rx_queues(vsi, true); 4448 if (err) { 4449 device_printf(dev, 4450 "Unable to enable subif Rx rings for receive: %s\n", 4451 ice_err_str(err)); 4452 goto err_cleanup_tx; 4453 } 4454 4455 ice_configure_all_rxq_interrupts(vsi); 4456 ice_configure_rx_itr(vsi); 4457 4458 ice_set_state(&mif->state, ICE_STATE_DRIVER_INITIALIZED); 4459 return; 4460 4461 err_cleanup_tx: 4462 ice_vsi_disable_tx(vsi); 4463 } 4464 4465 /** 4466 * ice_if_stop_subif - Stop the subinterface 4467 * @ctx: iflib context structure 4468 * @ifs: subinterface context structure 4469 * 4470 * Called by iflib to stop the subinterface and bring it down. 4471 * (e.g. ifconfig ice0m0 down) 4472 * 4473 * @pre assumes the caller holds the iflib CTX lock 4474 */ 4475 static void 4476 ice_subif_if_stop(if_ctx_t ctx) 4477 { 4478 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4479 struct ice_softc *sc = mif->back; 4480 struct ice_vsi *vsi = mif->vsi; 4481 device_t dev = mif->subdev; 4482 4483 if (!ice_testandclear_state(&mif->state, ICE_STATE_DRIVER_INITIALIZED)) 4484 return; 4485 4486 if (ice_test_state(&sc->state, ICE_STATE_RESET_FAILED)) { 4487 device_printf(dev, 4488 "request to stop interface cannot be completed as the parent device %s failed to reset\n", 4489 device_get_nameunit(sc->dev)); 4490 return; 4491 } 4492 4493 if (ice_test_state(&sc->state, ICE_STATE_PREPARED_FOR_RESET)) { 4494 device_printf(dev, 4495 "request to stop interface cannot be completed while parent device %s is prepared for impending reset\n", 4496 device_get_nameunit(sc->dev)); 4497 return; 4498 } 4499 4500 /* Dissociate the Tx and Rx queues from the interrupts */ 4501 ice_flush_txq_interrupts(vsi); 4502 ice_flush_rxq_interrupts(vsi); 4503 4504 /* Disable the Tx and Rx queues */ 4505 ice_vsi_disable_tx(vsi); 4506 ice_control_all_rx_queues(vsi, false); 4507 } 4508 4509 /** 4510 * ice_free_irqvs_subif - Free IRQ vector memory for subinterfaces 4511 * @mif: Mirror interface private structure 4512 * 4513 * Free IRQ vector memory allocated during ice_subif_if_msix_intr_assign. 4514 */ 4515 static void 4516 ice_free_irqvs_subif(struct ice_mirr_if *mif) 4517 { 4518 struct ice_softc *sc = mif->back; 4519 struct ice_vsi *vsi = mif->vsi; 4520 if_ctx_t ctx = sc->ctx; 4521 int i; 4522 4523 /* If the irqvs array is NULL, then there are no vectors to free */ 4524 if (mif->rx_irqvs == NULL) 4525 return; 4526 4527 /* Free the IRQ vectors -- currently subinterfaces have number 4528 * of vectors equal to number of RX queues 4529 * 4530 * XXX: ctx is parent device's ctx, not the subinterface ctx 4531 */ 4532 for (i = 0; i < vsi->num_rx_queues; i++) 4533 iflib_irq_free(ctx, &mif->rx_irqvs[i].irq); 4534 4535 ice_resmgr_release_map(&sc->os_imgr, mif->os_imap, 4536 mif->num_irq_vectors); 4537 ice_resmgr_release_map(&sc->dev_imgr, mif->if_imap, 4538 mif->num_irq_vectors); 4539 4540 sc->last_rid -= vsi->num_rx_queues; 4541 4542 /* Clear the irqv pointers */ 4543 for (i = 0; i < vsi->num_rx_queues; i++) 4544 vsi->rx_queues[i].irqv = NULL; 4545 4546 for (i = 0; i < vsi->num_tx_queues; i++) 4547 vsi->tx_queues[i].irqv = NULL; 4548 4549 /* Release the vector array memory */ 4550 free(mif->rx_irqvs, M_ICE); 4551 mif->rx_irqvs = NULL; 4552 } 4553 4554 /** 4555 * ice_subif_if_queues_free - Free queue memory for subinterfaces 4556 * @ctx: the iflib context structure 4557 * 4558 * Free queue memory allocated by ice_subif_tx_queues_alloc() and 4559 * ice_subif_if_rx_queues_alloc(). 4560 */ 4561 static void 4562 ice_subif_if_queues_free(if_ctx_t ctx) 4563 { 4564 struct ice_mirr_if *mif = (struct ice_mirr_if *)iflib_get_softc(ctx); 4565 struct ice_vsi *vsi = mif->vsi; 4566 struct ice_tx_queue *txq; 4567 int i; 4568 4569 /* Free the Tx and Rx sysctl contexts, and assign NULL to the node 4570 * pointers. 4571 */ 4572 ice_vsi_del_txqs_ctx(vsi); 4573 ice_vsi_del_rxqs_ctx(vsi); 4574 4575 /* Release MSI-X IRQ vectors */ 4576 ice_free_irqvs_subif(mif); 4577 4578 if (vsi->tx_queues != NULL) { 4579 /* free the tx_rsq arrays */ 4580 for (i = 0, txq = vsi->tx_queues; i < vsi->num_tx_queues; i++, txq++) { 4581 if (txq->tx_rsq != NULL) { 4582 free(txq->tx_rsq, M_ICE); 4583 txq->tx_rsq = NULL; 4584 } 4585 } 4586 free(vsi->tx_queues, M_ICE); 4587 vsi->tx_queues = NULL; 4588 } 4589 if (vsi->rx_queues != NULL) { 4590 free(vsi->rx_queues, M_ICE); 4591 vsi->rx_queues = NULL; 4592 } 4593 } 4594 4595 /** 4596 * ice_subif_if_media_status - Report subinterface media 4597 * @ctx: iflib context structure 4598 * @ifmr: ifmedia request structure to update 4599 * 4600 * Updates the provided ifmr with something, in order to prevent a 4601 * "no media types?" message from ifconfig. 4602 * 4603 * Mirror interfaces are always up. 4604 */ 4605 static void 4606 ice_subif_if_media_status(if_ctx_t ctx __unused, struct ifmediareq *ifmr) 4607 { 4608 ifmr->ifm_status = IFM_AVALID | IFM_ACTIVE; 4609 ifmr->ifm_active = IFM_ETHER | IFM_AUTO; 4610 } 4611 4612 /** 4613 * ice_subif_if_promisc_set - Set subinterface promiscuous mode 4614 * @ctx: iflib context structure 4615 * @flags: promiscuous flags to configure 4616 * 4617 * Called by iflib to configure device promiscuous mode. 4618 * 4619 * @remark This does not need to be implemented for now. 4620 */ 4621 static int 4622 ice_subif_if_promisc_set(if_ctx_t ctx __unused, int flags __unused) 4623 { 4624 return (0); 4625 } 4626 4627